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/slab.h>
29 #include <linux/migrate.h>
30 #include <linux/ratelimit.h>
31 #include <linux/uuid.h>
32 #include <linux/semaphore.h>
33 #include <asm/unaligned.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "print-tree.h"
43 #include "free-space-cache.h"
44 #include "free-space-tree.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
52 #include "compression.h"
55 #include <asm/cpufeature.h>
58 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
59 BTRFS_HEADER_FLAG_RELOC |\
60 BTRFS_SUPER_FLAG_ERROR |\
61 BTRFS_SUPER_FLAG_SEEDING |\
62 BTRFS_SUPER_FLAG_METADUMP)
64 static const struct extent_io_ops btree_extent_io_ops
;
65 static void end_workqueue_fn(struct btrfs_work
*work
);
66 static void free_fs_root(struct btrfs_root
*root
);
67 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
69 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
);
70 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
71 struct btrfs_root
*root
);
72 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
);
73 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
74 struct extent_io_tree
*dirty_pages
,
76 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
77 struct extent_io_tree
*pinned_extents
);
78 static int btrfs_cleanup_transaction(struct btrfs_root
*root
);
79 static void btrfs_error_commit_super(struct btrfs_root
*root
);
82 * btrfs_end_io_wq structs are used to do processing in task context when an IO
83 * is complete. This is used during reads to verify checksums, and it is used
84 * by writes to insert metadata for new file extents after IO is complete.
86 struct btrfs_end_io_wq
{
90 struct btrfs_fs_info
*info
;
92 enum btrfs_wq_endio_type metadata
;
93 struct list_head list
;
94 struct btrfs_work work
;
97 static struct kmem_cache
*btrfs_end_io_wq_cache
;
99 int __init
btrfs_end_io_wq_init(void)
101 btrfs_end_io_wq_cache
= kmem_cache_create("btrfs_end_io_wq",
102 sizeof(struct btrfs_end_io_wq
),
106 if (!btrfs_end_io_wq_cache
)
111 void btrfs_end_io_wq_exit(void)
113 kmem_cache_destroy(btrfs_end_io_wq_cache
);
117 * async submit bios are used to offload expensive checksumming
118 * onto the worker threads. They checksum file and metadata bios
119 * just before they are sent down the IO stack.
121 struct async_submit_bio
{
124 struct list_head list
;
125 extent_submit_bio_hook_t
*submit_bio_start
;
126 extent_submit_bio_hook_t
*submit_bio_done
;
128 unsigned long bio_flags
;
130 * bio_offset is optional, can be used if the pages in the bio
131 * can't tell us where in the file the bio should go
134 struct btrfs_work work
;
139 * Lockdep class keys for extent_buffer->lock's in this root. For a given
140 * eb, the lockdep key is determined by the btrfs_root it belongs to and
141 * the level the eb occupies in the tree.
143 * Different roots are used for different purposes and may nest inside each
144 * other and they require separate keysets. As lockdep keys should be
145 * static, assign keysets according to the purpose of the root as indicated
146 * by btrfs_root->objectid. This ensures that all special purpose roots
147 * have separate keysets.
149 * Lock-nesting across peer nodes is always done with the immediate parent
150 * node locked thus preventing deadlock. As lockdep doesn't know this, use
151 * subclass to avoid triggering lockdep warning in such cases.
153 * The key is set by the readpage_end_io_hook after the buffer has passed
154 * csum validation but before the pages are unlocked. It is also set by
155 * btrfs_init_new_buffer on freshly allocated blocks.
157 * We also add a check to make sure the highest level of the tree is the
158 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
159 * needs update as well.
161 #ifdef CONFIG_DEBUG_LOCK_ALLOC
162 # if BTRFS_MAX_LEVEL != 8
166 static struct btrfs_lockdep_keyset
{
167 u64 id
; /* root objectid */
168 const char *name_stem
; /* lock name stem */
169 char names
[BTRFS_MAX_LEVEL
+ 1][20];
170 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
171 } btrfs_lockdep_keysets
[] = {
172 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
173 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
174 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
175 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
176 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
177 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
178 { .id
= BTRFS_QUOTA_TREE_OBJECTID
, .name_stem
= "quota" },
179 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
180 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
181 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
182 { .id
= BTRFS_UUID_TREE_OBJECTID
, .name_stem
= "uuid" },
183 { .id
= BTRFS_FREE_SPACE_TREE_OBJECTID
, .name_stem
= "free-space" },
184 { .id
= 0, .name_stem
= "tree" },
187 void __init
btrfs_init_lockdep(void)
191 /* initialize lockdep class names */
192 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
193 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
195 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
196 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
197 "btrfs-%s-%02d", ks
->name_stem
, j
);
201 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
204 struct btrfs_lockdep_keyset
*ks
;
206 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
208 /* find the matching keyset, id 0 is the default entry */
209 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
210 if (ks
->id
== objectid
)
213 lockdep_set_class_and_name(&eb
->lock
,
214 &ks
->keys
[level
], ks
->names
[level
]);
220 * extents on the btree inode are pretty simple, there's one extent
221 * that covers the entire device
223 static struct extent_map
*btree_get_extent(struct inode
*inode
,
224 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
227 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
228 struct extent_map
*em
;
231 read_lock(&em_tree
->lock
);
232 em
= lookup_extent_mapping(em_tree
, start
, len
);
235 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
236 read_unlock(&em_tree
->lock
);
239 read_unlock(&em_tree
->lock
);
241 em
= alloc_extent_map();
243 em
= ERR_PTR(-ENOMEM
);
248 em
->block_len
= (u64
)-1;
250 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
252 write_lock(&em_tree
->lock
);
253 ret
= add_extent_mapping(em_tree
, em
, 0);
254 if (ret
== -EEXIST
) {
256 em
= lookup_extent_mapping(em_tree
, start
, len
);
263 write_unlock(&em_tree
->lock
);
269 u32
btrfs_csum_data(char *data
, u32 seed
, size_t len
)
271 return btrfs_crc32c(seed
, data
, len
);
274 void btrfs_csum_final(u32 crc
, char *result
)
276 put_unaligned_le32(~crc
, result
);
280 * compute the csum for a btree block, and either verify it or write it
281 * into the csum field of the block.
283 static int csum_tree_block(struct btrfs_fs_info
*fs_info
,
284 struct extent_buffer
*buf
,
287 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
290 unsigned long cur_len
;
291 unsigned long offset
= BTRFS_CSUM_SIZE
;
293 unsigned long map_start
;
294 unsigned long map_len
;
297 unsigned long inline_result
;
299 len
= buf
->len
- offset
;
301 err
= map_private_extent_buffer(buf
, offset
, 32,
302 &kaddr
, &map_start
, &map_len
);
305 cur_len
= min(len
, map_len
- (offset
- map_start
));
306 crc
= btrfs_csum_data(kaddr
+ offset
- map_start
,
311 if (csum_size
> sizeof(inline_result
)) {
312 result
= kzalloc(csum_size
, GFP_NOFS
);
316 result
= (char *)&inline_result
;
319 btrfs_csum_final(crc
, result
);
322 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
325 memcpy(&found
, result
, csum_size
);
327 read_extent_buffer(buf
, &val
, 0, csum_size
);
328 btrfs_warn_rl(fs_info
,
329 "%s checksum verify failed on %llu wanted %X found %X level %d",
330 fs_info
->sb
->s_id
, buf
->start
,
331 val
, found
, btrfs_header_level(buf
));
332 if (result
!= (char *)&inline_result
)
337 write_extent_buffer(buf
, result
, 0, csum_size
);
339 if (result
!= (char *)&inline_result
)
345 * we can't consider a given block up to date unless the transid of the
346 * block matches the transid in the parent node's pointer. This is how we
347 * detect blocks that either didn't get written at all or got written
348 * in the wrong place.
350 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
351 struct extent_buffer
*eb
, u64 parent_transid
,
354 struct extent_state
*cached_state
= NULL
;
356 bool need_lock
= (current
->journal_info
== BTRFS_SEND_TRANS_STUB
);
358 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
365 btrfs_tree_read_lock(eb
);
366 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
369 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
371 if (extent_buffer_uptodate(eb
) &&
372 btrfs_header_generation(eb
) == parent_transid
) {
376 btrfs_err_rl(eb
->fs_info
,
377 "parent transid verify failed on %llu wanted %llu found %llu",
379 parent_transid
, btrfs_header_generation(eb
));
383 * Things reading via commit roots that don't have normal protection,
384 * like send, can have a really old block in cache that may point at a
385 * block that has been freed and re-allocated. So don't clear uptodate
386 * if we find an eb that is under IO (dirty/writeback) because we could
387 * end up reading in the stale data and then writing it back out and
388 * making everybody very sad.
390 if (!extent_buffer_under_io(eb
))
391 clear_extent_buffer_uptodate(eb
);
393 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
394 &cached_state
, GFP_NOFS
);
396 btrfs_tree_read_unlock_blocking(eb
);
401 * Return 0 if the superblock checksum type matches the checksum value of that
402 * algorithm. Pass the raw disk superblock data.
404 static int btrfs_check_super_csum(struct btrfs_fs_info
*fs_info
,
407 struct btrfs_super_block
*disk_sb
=
408 (struct btrfs_super_block
*)raw_disk_sb
;
409 u16 csum_type
= btrfs_super_csum_type(disk_sb
);
412 if (csum_type
== BTRFS_CSUM_TYPE_CRC32
) {
414 const int csum_size
= sizeof(crc
);
415 char result
[csum_size
];
418 * The super_block structure does not span the whole
419 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
420 * is filled with zeros and is included in the checksum.
422 crc
= btrfs_csum_data(raw_disk_sb
+ BTRFS_CSUM_SIZE
,
423 crc
, BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
);
424 btrfs_csum_final(crc
, result
);
426 if (memcmp(raw_disk_sb
, result
, csum_size
))
430 if (csum_type
>= ARRAY_SIZE(btrfs_csum_sizes
)) {
431 btrfs_err(fs_info
, "unsupported checksum algorithm %u",
440 * helper to read a given tree block, doing retries as required when
441 * the checksums don't match and we have alternate mirrors to try.
443 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
444 struct extent_buffer
*eb
,
447 struct extent_io_tree
*io_tree
;
452 int failed_mirror
= 0;
454 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
455 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
457 ret
= read_extent_buffer_pages(io_tree
, eb
, WAIT_COMPLETE
,
458 btree_get_extent
, mirror_num
);
460 if (!verify_parent_transid(io_tree
, eb
,
468 * This buffer's crc is fine, but its contents are corrupted, so
469 * there is no reason to read the other copies, they won't be
472 if (test_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
))
475 num_copies
= btrfs_num_copies(root
->fs_info
,
480 if (!failed_mirror
) {
482 failed_mirror
= eb
->read_mirror
;
486 if (mirror_num
== failed_mirror
)
489 if (mirror_num
> num_copies
)
493 if (failed
&& !ret
&& failed_mirror
)
494 repair_eb_io_failure(root
, eb
, failed_mirror
);
500 * checksum a dirty tree block before IO. This has extra checks to make sure
501 * we only fill in the checksum field in the first page of a multi-page block
504 static int csum_dirty_buffer(struct btrfs_fs_info
*fs_info
, struct page
*page
)
506 u64 start
= page_offset(page
);
508 struct extent_buffer
*eb
;
510 eb
= (struct extent_buffer
*)page
->private;
511 if (page
!= eb
->pages
[0])
514 found_start
= btrfs_header_bytenr(eb
);
516 * Please do not consolidate these warnings into a single if.
517 * It is useful to know what went wrong.
519 if (WARN_ON(found_start
!= start
))
521 if (WARN_ON(!PageUptodate(page
)))
524 ASSERT(memcmp_extent_buffer(eb
, fs_info
->fsid
,
525 btrfs_header_fsid(), BTRFS_FSID_SIZE
) == 0);
527 return csum_tree_block(fs_info
, eb
, 0);
530 static int check_tree_block_fsid(struct btrfs_fs_info
*fs_info
,
531 struct extent_buffer
*eb
)
533 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
534 u8 fsid
[BTRFS_UUID_SIZE
];
537 read_extent_buffer(eb
, fsid
, btrfs_header_fsid(), BTRFS_FSID_SIZE
);
539 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
543 fs_devices
= fs_devices
->seed
;
548 #define CORRUPT(reason, eb, root, slot) \
549 btrfs_crit(root->fs_info, "corrupt %s, %s: block=%llu," \
550 " root=%llu, slot=%d", \
551 btrfs_header_level(eb) == 0 ? "leaf" : "node",\
552 reason, btrfs_header_bytenr(eb), root->objectid, slot)
554 static noinline
int check_leaf(struct btrfs_root
*root
,
555 struct extent_buffer
*leaf
)
557 struct btrfs_key key
;
558 struct btrfs_key leaf_key
;
559 u32 nritems
= btrfs_header_nritems(leaf
);
563 struct btrfs_root
*check_root
;
565 key
.objectid
= btrfs_header_owner(leaf
);
566 key
.type
= BTRFS_ROOT_ITEM_KEY
;
567 key
.offset
= (u64
)-1;
569 check_root
= btrfs_get_fs_root(root
->fs_info
, &key
, false);
571 * The only reason we also check NULL here is that during
572 * open_ctree() some roots has not yet been set up.
574 if (!IS_ERR_OR_NULL(check_root
)) {
575 /* if leaf is the root, then it's fine */
577 btrfs_root_bytenr(&check_root
->root_item
)) {
578 CORRUPT("non-root leaf's nritems is 0",
586 /* Check the 0 item */
587 if (btrfs_item_offset_nr(leaf
, 0) + btrfs_item_size_nr(leaf
, 0) !=
588 BTRFS_LEAF_DATA_SIZE(root
)) {
589 CORRUPT("invalid item offset size pair", leaf
, root
, 0);
594 * Check to make sure each items keys are in the correct order and their
595 * offsets make sense. We only have to loop through nritems-1 because
596 * we check the current slot against the next slot, which verifies the
597 * next slot's offset+size makes sense and that the current's slot
600 for (slot
= 0; slot
< nritems
- 1; slot
++) {
601 btrfs_item_key_to_cpu(leaf
, &leaf_key
, slot
);
602 btrfs_item_key_to_cpu(leaf
, &key
, slot
+ 1);
604 /* Make sure the keys are in the right order */
605 if (btrfs_comp_cpu_keys(&leaf_key
, &key
) >= 0) {
606 CORRUPT("bad key order", leaf
, root
, slot
);
611 * Make sure the offset and ends are right, remember that the
612 * item data starts at the end of the leaf and grows towards the
615 if (btrfs_item_offset_nr(leaf
, slot
) !=
616 btrfs_item_end_nr(leaf
, slot
+ 1)) {
617 CORRUPT("slot offset bad", leaf
, root
, slot
);
622 * Check to make sure that we don't point outside of the leaf,
623 * just in case all the items are consistent to each other, but
624 * all point outside of the leaf.
626 if (btrfs_item_end_nr(leaf
, slot
) >
627 BTRFS_LEAF_DATA_SIZE(root
)) {
628 CORRUPT("slot end outside of leaf", leaf
, root
, slot
);
636 static int check_node(struct btrfs_root
*root
, struct extent_buffer
*node
)
638 unsigned long nr
= btrfs_header_nritems(node
);
639 struct btrfs_key key
, next_key
;
644 if (nr
== 0 || nr
> BTRFS_NODEPTRS_PER_BLOCK(root
)) {
645 btrfs_crit(root
->fs_info
,
646 "corrupt node: block %llu root %llu nritems %lu",
647 node
->start
, root
->objectid
, nr
);
651 for (slot
= 0; slot
< nr
- 1; slot
++) {
652 bytenr
= btrfs_node_blockptr(node
, slot
);
653 btrfs_node_key_to_cpu(node
, &key
, slot
);
654 btrfs_node_key_to_cpu(node
, &next_key
, slot
+ 1);
657 CORRUPT("invalid item slot", node
, root
, slot
);
662 if (btrfs_comp_cpu_keys(&key
, &next_key
) >= 0) {
663 CORRUPT("bad key order", node
, root
, slot
);
672 static int btree_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
673 u64 phy_offset
, struct page
*page
,
674 u64 start
, u64 end
, int mirror
)
678 struct extent_buffer
*eb
;
679 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
680 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
687 eb
= (struct extent_buffer
*)page
->private;
689 /* the pending IO might have been the only thing that kept this buffer
690 * in memory. Make sure we have a ref for all this other checks
692 extent_buffer_get(eb
);
694 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
698 eb
->read_mirror
= mirror
;
699 if (test_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
)) {
704 found_start
= btrfs_header_bytenr(eb
);
705 if (found_start
!= eb
->start
) {
706 btrfs_err_rl(fs_info
, "bad tree block start %llu %llu",
707 found_start
, eb
->start
);
711 if (check_tree_block_fsid(fs_info
, eb
)) {
712 btrfs_err_rl(fs_info
, "bad fsid on block %llu",
717 found_level
= btrfs_header_level(eb
);
718 if (found_level
>= BTRFS_MAX_LEVEL
) {
719 btrfs_err(fs_info
, "bad tree block level %d",
720 (int)btrfs_header_level(eb
));
725 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
728 ret
= csum_tree_block(fs_info
, eb
, 1);
733 * If this is a leaf block and it is corrupt, set the corrupt bit so
734 * that we don't try and read the other copies of this block, just
737 if (found_level
== 0 && check_leaf(root
, eb
)) {
738 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
742 if (found_level
> 0 && check_node(root
, eb
))
746 set_extent_buffer_uptodate(eb
);
749 test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
750 btree_readahead_hook(fs_info
, eb
, eb
->start
, ret
);
754 * our io error hook is going to dec the io pages
755 * again, we have to make sure it has something
758 atomic_inc(&eb
->io_pages
);
759 clear_extent_buffer_uptodate(eb
);
761 free_extent_buffer(eb
);
766 static int btree_io_failed_hook(struct page
*page
, int failed_mirror
)
768 struct extent_buffer
*eb
;
770 eb
= (struct extent_buffer
*)page
->private;
771 set_bit(EXTENT_BUFFER_READ_ERR
, &eb
->bflags
);
772 eb
->read_mirror
= failed_mirror
;
773 atomic_dec(&eb
->io_pages
);
774 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
775 btree_readahead_hook(eb
->fs_info
, eb
, eb
->start
, -EIO
);
776 return -EIO
; /* we fixed nothing */
779 static void end_workqueue_bio(struct bio
*bio
)
781 struct btrfs_end_io_wq
*end_io_wq
= bio
->bi_private
;
782 struct btrfs_fs_info
*fs_info
;
783 struct btrfs_workqueue
*wq
;
784 btrfs_work_func_t func
;
786 fs_info
= end_io_wq
->info
;
787 end_io_wq
->error
= bio
->bi_error
;
789 if (bio_op(bio
) == REQ_OP_WRITE
) {
790 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
) {
791 wq
= fs_info
->endio_meta_write_workers
;
792 func
= btrfs_endio_meta_write_helper
;
793 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
) {
794 wq
= fs_info
->endio_freespace_worker
;
795 func
= btrfs_freespace_write_helper
;
796 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
797 wq
= fs_info
->endio_raid56_workers
;
798 func
= btrfs_endio_raid56_helper
;
800 wq
= fs_info
->endio_write_workers
;
801 func
= btrfs_endio_write_helper
;
804 if (unlikely(end_io_wq
->metadata
==
805 BTRFS_WQ_ENDIO_DIO_REPAIR
)) {
806 wq
= fs_info
->endio_repair_workers
;
807 func
= btrfs_endio_repair_helper
;
808 } else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
) {
809 wq
= fs_info
->endio_raid56_workers
;
810 func
= btrfs_endio_raid56_helper
;
811 } else if (end_io_wq
->metadata
) {
812 wq
= fs_info
->endio_meta_workers
;
813 func
= btrfs_endio_meta_helper
;
815 wq
= fs_info
->endio_workers
;
816 func
= btrfs_endio_helper
;
820 btrfs_init_work(&end_io_wq
->work
, func
, end_workqueue_fn
, NULL
, NULL
);
821 btrfs_queue_work(wq
, &end_io_wq
->work
);
824 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
825 enum btrfs_wq_endio_type metadata
)
827 struct btrfs_end_io_wq
*end_io_wq
;
829 end_io_wq
= kmem_cache_alloc(btrfs_end_io_wq_cache
, GFP_NOFS
);
833 end_io_wq
->private = bio
->bi_private
;
834 end_io_wq
->end_io
= bio
->bi_end_io
;
835 end_io_wq
->info
= info
;
836 end_io_wq
->error
= 0;
837 end_io_wq
->bio
= bio
;
838 end_io_wq
->metadata
= metadata
;
840 bio
->bi_private
= end_io_wq
;
841 bio
->bi_end_io
= end_workqueue_bio
;
845 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
847 unsigned long limit
= min_t(unsigned long,
848 info
->thread_pool_size
,
849 info
->fs_devices
->open_devices
);
853 static void run_one_async_start(struct btrfs_work
*work
)
855 struct async_submit_bio
*async
;
858 async
= container_of(work
, struct async_submit_bio
, work
);
859 ret
= async
->submit_bio_start(async
->inode
, async
->bio
,
860 async
->mirror_num
, async
->bio_flags
,
866 static void run_one_async_done(struct btrfs_work
*work
)
868 struct btrfs_fs_info
*fs_info
;
869 struct async_submit_bio
*async
;
872 async
= container_of(work
, struct async_submit_bio
, work
);
873 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
875 limit
= btrfs_async_submit_limit(fs_info
);
876 limit
= limit
* 2 / 3;
879 * atomic_dec_return implies a barrier for waitqueue_active
881 if (atomic_dec_return(&fs_info
->nr_async_submits
) < limit
&&
882 waitqueue_active(&fs_info
->async_submit_wait
))
883 wake_up(&fs_info
->async_submit_wait
);
885 /* If an error occurred we just want to clean up the bio and move on */
887 async
->bio
->bi_error
= async
->error
;
888 bio_endio(async
->bio
);
892 async
->submit_bio_done(async
->inode
, async
->bio
, async
->mirror_num
,
893 async
->bio_flags
, async
->bio_offset
);
896 static void run_one_async_free(struct btrfs_work
*work
)
898 struct async_submit_bio
*async
;
900 async
= container_of(work
, struct async_submit_bio
, work
);
904 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
905 struct bio
*bio
, int mirror_num
,
906 unsigned long bio_flags
,
908 extent_submit_bio_hook_t
*submit_bio_start
,
909 extent_submit_bio_hook_t
*submit_bio_done
)
911 struct async_submit_bio
*async
;
913 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
917 async
->inode
= inode
;
919 async
->mirror_num
= mirror_num
;
920 async
->submit_bio_start
= submit_bio_start
;
921 async
->submit_bio_done
= submit_bio_done
;
923 btrfs_init_work(&async
->work
, btrfs_worker_helper
, run_one_async_start
,
924 run_one_async_done
, run_one_async_free
);
926 async
->bio_flags
= bio_flags
;
927 async
->bio_offset
= bio_offset
;
931 atomic_inc(&fs_info
->nr_async_submits
);
933 if (bio
->bi_opf
& REQ_SYNC
)
934 btrfs_set_work_high_priority(&async
->work
);
936 btrfs_queue_work(fs_info
->workers
, &async
->work
);
938 while (atomic_read(&fs_info
->async_submit_draining
) &&
939 atomic_read(&fs_info
->nr_async_submits
)) {
940 wait_event(fs_info
->async_submit_wait
,
941 (atomic_read(&fs_info
->nr_async_submits
) == 0));
947 static int btree_csum_one_bio(struct bio
*bio
)
949 struct bio_vec
*bvec
;
950 struct btrfs_root
*root
;
953 bio_for_each_segment_all(bvec
, bio
, i
) {
954 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
955 ret
= csum_dirty_buffer(root
->fs_info
, bvec
->bv_page
);
963 static int __btree_submit_bio_start(struct inode
*inode
, struct bio
*bio
,
964 int mirror_num
, unsigned long bio_flags
,
968 * when we're called for a write, we're already in the async
969 * submission context. Just jump into btrfs_map_bio
971 return btree_csum_one_bio(bio
);
974 static int __btree_submit_bio_done(struct inode
*inode
, struct bio
*bio
,
975 int mirror_num
, unsigned long bio_flags
,
981 * when we're called for a write, we're already in the async
982 * submission context. Just jump into btrfs_map_bio
984 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, bio
, mirror_num
, 1);
992 static int check_async_write(struct inode
*inode
, unsigned long bio_flags
)
994 if (bio_flags
& EXTENT_BIO_TREE_LOG
)
997 if (static_cpu_has(X86_FEATURE_XMM4_2
))
1003 static int btree_submit_bio_hook(struct inode
*inode
, struct bio
*bio
,
1004 int mirror_num
, unsigned long bio_flags
,
1007 int async
= check_async_write(inode
, bio_flags
);
1010 if (bio_op(bio
) != REQ_OP_WRITE
) {
1012 * called for a read, do the setup so that checksum validation
1013 * can happen in the async kernel threads
1015 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
1016 bio
, BTRFS_WQ_ENDIO_METADATA
);
1019 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, bio
, mirror_num
, 0);
1020 } else if (!async
) {
1021 ret
= btree_csum_one_bio(bio
);
1024 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, bio
, mirror_num
, 0);
1027 * kthread helpers are used to submit writes so that
1028 * checksumming can happen in parallel across all CPUs
1030 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1031 inode
, bio
, mirror_num
, 0,
1033 __btree_submit_bio_start
,
1034 __btree_submit_bio_done
);
1042 bio
->bi_error
= ret
;
1047 #ifdef CONFIG_MIGRATION
1048 static int btree_migratepage(struct address_space
*mapping
,
1049 struct page
*newpage
, struct page
*page
,
1050 enum migrate_mode mode
)
1053 * we can't safely write a btree page from here,
1054 * we haven't done the locking hook
1056 if (PageDirty(page
))
1059 * Buffers may be managed in a filesystem specific way.
1060 * We must have no buffers or drop them.
1062 if (page_has_private(page
) &&
1063 !try_to_release_page(page
, GFP_KERNEL
))
1065 return migrate_page(mapping
, newpage
, page
, mode
);
1070 static int btree_writepages(struct address_space
*mapping
,
1071 struct writeback_control
*wbc
)
1073 struct btrfs_fs_info
*fs_info
;
1076 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
1078 if (wbc
->for_kupdate
)
1081 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
1082 /* this is a bit racy, but that's ok */
1083 ret
= percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
1084 BTRFS_DIRTY_METADATA_THRESH
);
1088 return btree_write_cache_pages(mapping
, wbc
);
1091 static int btree_readpage(struct file
*file
, struct page
*page
)
1093 struct extent_io_tree
*tree
;
1094 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1095 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
1098 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
1100 if (PageWriteback(page
) || PageDirty(page
))
1103 return try_release_extent_buffer(page
);
1106 static void btree_invalidatepage(struct page
*page
, unsigned int offset
,
1107 unsigned int length
)
1109 struct extent_io_tree
*tree
;
1110 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1111 extent_invalidatepage(tree
, page
, offset
);
1112 btree_releasepage(page
, GFP_NOFS
);
1113 if (PagePrivate(page
)) {
1114 btrfs_warn(BTRFS_I(page
->mapping
->host
)->root
->fs_info
,
1115 "page private not zero on page %llu",
1116 (unsigned long long)page_offset(page
));
1117 ClearPagePrivate(page
);
1118 set_page_private(page
, 0);
1123 static int btree_set_page_dirty(struct page
*page
)
1126 struct extent_buffer
*eb
;
1128 BUG_ON(!PagePrivate(page
));
1129 eb
= (struct extent_buffer
*)page
->private;
1131 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
1132 BUG_ON(!atomic_read(&eb
->refs
));
1133 btrfs_assert_tree_locked(eb
);
1135 return __set_page_dirty_nobuffers(page
);
1138 static const struct address_space_operations btree_aops
= {
1139 .readpage
= btree_readpage
,
1140 .writepages
= btree_writepages
,
1141 .releasepage
= btree_releasepage
,
1142 .invalidatepage
= btree_invalidatepage
,
1143 #ifdef CONFIG_MIGRATION
1144 .migratepage
= btree_migratepage
,
1146 .set_page_dirty
= btree_set_page_dirty
,
1149 void readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
)
1151 struct extent_buffer
*buf
= NULL
;
1152 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1154 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1157 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1158 buf
, WAIT_NONE
, btree_get_extent
, 0);
1159 free_extent_buffer(buf
);
1162 int reada_tree_block_flagged(struct btrfs_root
*root
, u64 bytenr
,
1163 int mirror_num
, struct extent_buffer
**eb
)
1165 struct extent_buffer
*buf
= NULL
;
1166 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1167 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1170 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1174 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1176 ret
= read_extent_buffer_pages(io_tree
, buf
, WAIT_PAGE_LOCK
,
1177 btree_get_extent
, mirror_num
);
1179 free_extent_buffer(buf
);
1183 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1184 free_extent_buffer(buf
);
1186 } else if (extent_buffer_uptodate(buf
)) {
1189 free_extent_buffer(buf
);
1194 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_fs_info
*fs_info
,
1197 return find_extent_buffer(fs_info
, bytenr
);
1200 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
1203 if (btrfs_is_testing(root
->fs_info
))
1204 return alloc_test_extent_buffer(root
->fs_info
, bytenr
,
1206 return alloc_extent_buffer(root
->fs_info
, bytenr
);
1210 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1212 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1213 buf
->start
+ buf
->len
- 1);
1216 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1218 return filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1219 buf
->start
, buf
->start
+ buf
->len
- 1);
1222 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
1225 struct extent_buffer
*buf
= NULL
;
1228 buf
= btrfs_find_create_tree_block(root
, bytenr
);
1232 ret
= btree_read_extent_buffer_pages(root
, buf
, parent_transid
);
1234 free_extent_buffer(buf
);
1235 return ERR_PTR(ret
);
1241 void clean_tree_block(struct btrfs_trans_handle
*trans
,
1242 struct btrfs_fs_info
*fs_info
,
1243 struct extent_buffer
*buf
)
1245 if (btrfs_header_generation(buf
) ==
1246 fs_info
->running_transaction
->transid
) {
1247 btrfs_assert_tree_locked(buf
);
1249 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1250 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
1252 fs_info
->dirty_metadata_batch
);
1253 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1254 btrfs_set_lock_blocking(buf
);
1255 clear_extent_buffer_dirty(buf
);
1260 static struct btrfs_subvolume_writers
*btrfs_alloc_subvolume_writers(void)
1262 struct btrfs_subvolume_writers
*writers
;
1265 writers
= kmalloc(sizeof(*writers
), GFP_NOFS
);
1267 return ERR_PTR(-ENOMEM
);
1269 ret
= percpu_counter_init(&writers
->counter
, 0, GFP_KERNEL
);
1272 return ERR_PTR(ret
);
1275 init_waitqueue_head(&writers
->wait
);
1280 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers
*writers
)
1282 percpu_counter_destroy(&writers
->counter
);
1286 static void __setup_root(u32 nodesize
, u32 sectorsize
, u32 stripesize
,
1287 struct btrfs_root
*root
, struct btrfs_fs_info
*fs_info
,
1290 bool dummy
= test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO
, &fs_info
->fs_state
);
1292 root
->commit_root
= NULL
;
1293 root
->sectorsize
= sectorsize
;
1294 root
->nodesize
= nodesize
;
1295 root
->stripesize
= stripesize
;
1297 root
->orphan_cleanup_state
= 0;
1299 root
->objectid
= objectid
;
1300 root
->last_trans
= 0;
1301 root
->highest_objectid
= 0;
1302 root
->nr_delalloc_inodes
= 0;
1303 root
->nr_ordered_extents
= 0;
1305 root
->inode_tree
= RB_ROOT
;
1306 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1307 root
->block_rsv
= NULL
;
1308 root
->orphan_block_rsv
= NULL
;
1310 INIT_LIST_HEAD(&root
->dirty_list
);
1311 INIT_LIST_HEAD(&root
->root_list
);
1312 INIT_LIST_HEAD(&root
->delalloc_inodes
);
1313 INIT_LIST_HEAD(&root
->delalloc_root
);
1314 INIT_LIST_HEAD(&root
->ordered_extents
);
1315 INIT_LIST_HEAD(&root
->ordered_root
);
1316 INIT_LIST_HEAD(&root
->logged_list
[0]);
1317 INIT_LIST_HEAD(&root
->logged_list
[1]);
1318 spin_lock_init(&root
->orphan_lock
);
1319 spin_lock_init(&root
->inode_lock
);
1320 spin_lock_init(&root
->delalloc_lock
);
1321 spin_lock_init(&root
->ordered_extent_lock
);
1322 spin_lock_init(&root
->accounting_lock
);
1323 spin_lock_init(&root
->log_extents_lock
[0]);
1324 spin_lock_init(&root
->log_extents_lock
[1]);
1325 mutex_init(&root
->objectid_mutex
);
1326 mutex_init(&root
->log_mutex
);
1327 mutex_init(&root
->ordered_extent_mutex
);
1328 mutex_init(&root
->delalloc_mutex
);
1329 init_waitqueue_head(&root
->log_writer_wait
);
1330 init_waitqueue_head(&root
->log_commit_wait
[0]);
1331 init_waitqueue_head(&root
->log_commit_wait
[1]);
1332 INIT_LIST_HEAD(&root
->log_ctxs
[0]);
1333 INIT_LIST_HEAD(&root
->log_ctxs
[1]);
1334 atomic_set(&root
->log_commit
[0], 0);
1335 atomic_set(&root
->log_commit
[1], 0);
1336 atomic_set(&root
->log_writers
, 0);
1337 atomic_set(&root
->log_batch
, 0);
1338 atomic_set(&root
->orphan_inodes
, 0);
1339 atomic_set(&root
->refs
, 1);
1340 atomic_set(&root
->will_be_snapshoted
, 0);
1341 atomic_set(&root
->qgroup_meta_rsv
, 0);
1342 root
->log_transid
= 0;
1343 root
->log_transid_committed
= -1;
1344 root
->last_log_commit
= 0;
1346 extent_io_tree_init(&root
->dirty_log_pages
,
1347 fs_info
->btree_inode
->i_mapping
);
1349 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1350 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1351 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1353 root
->defrag_trans_start
= fs_info
->generation
;
1355 root
->defrag_trans_start
= 0;
1356 root
->root_key
.objectid
= objectid
;
1359 spin_lock_init(&root
->root_item_lock
);
1362 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
,
1365 struct btrfs_root
*root
= kzalloc(sizeof(*root
), flags
);
1367 root
->fs_info
= fs_info
;
1371 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1372 /* Should only be used by the testing infrastructure */
1373 struct btrfs_root
*btrfs_alloc_dummy_root(struct btrfs_fs_info
*fs_info
,
1374 u32 sectorsize
, u32 nodesize
)
1376 struct btrfs_root
*root
;
1379 return ERR_PTR(-EINVAL
);
1381 root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
1383 return ERR_PTR(-ENOMEM
);
1384 /* We don't use the stripesize in selftest, set it as sectorsize */
1385 __setup_root(nodesize
, sectorsize
, sectorsize
, root
, fs_info
,
1386 BTRFS_ROOT_TREE_OBJECTID
);
1387 root
->alloc_bytenr
= 0;
1393 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1394 struct btrfs_fs_info
*fs_info
,
1397 struct extent_buffer
*leaf
;
1398 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1399 struct btrfs_root
*root
;
1400 struct btrfs_key key
;
1404 root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
1406 return ERR_PTR(-ENOMEM
);
1408 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1409 tree_root
->stripesize
, root
, fs_info
, objectid
);
1410 root
->root_key
.objectid
= objectid
;
1411 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1412 root
->root_key
.offset
= 0;
1414 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, objectid
, NULL
, 0, 0, 0);
1416 ret
= PTR_ERR(leaf
);
1421 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1422 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1423 btrfs_set_header_generation(leaf
, trans
->transid
);
1424 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1425 btrfs_set_header_owner(leaf
, objectid
);
1428 write_extent_buffer(leaf
, fs_info
->fsid
, btrfs_header_fsid(),
1430 write_extent_buffer(leaf
, fs_info
->chunk_tree_uuid
,
1431 btrfs_header_chunk_tree_uuid(leaf
),
1433 btrfs_mark_buffer_dirty(leaf
);
1435 root
->commit_root
= btrfs_root_node(root
);
1436 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
1438 root
->root_item
.flags
= 0;
1439 root
->root_item
.byte_limit
= 0;
1440 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1441 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1442 btrfs_set_root_level(&root
->root_item
, 0);
1443 btrfs_set_root_refs(&root
->root_item
, 1);
1444 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1445 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1446 btrfs_set_root_dirid(&root
->root_item
, 0);
1448 memcpy(root
->root_item
.uuid
, uuid
.b
, BTRFS_UUID_SIZE
);
1449 root
->root_item
.drop_level
= 0;
1451 key
.objectid
= objectid
;
1452 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1454 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1458 btrfs_tree_unlock(leaf
);
1464 btrfs_tree_unlock(leaf
);
1465 free_extent_buffer(root
->commit_root
);
1466 free_extent_buffer(leaf
);
1470 return ERR_PTR(ret
);
1473 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1474 struct btrfs_fs_info
*fs_info
)
1476 struct btrfs_root
*root
;
1477 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1478 struct extent_buffer
*leaf
;
1480 root
= btrfs_alloc_root(fs_info
, GFP_NOFS
);
1482 return ERR_PTR(-ENOMEM
);
1484 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1485 tree_root
->stripesize
, root
, fs_info
,
1486 BTRFS_TREE_LOG_OBJECTID
);
1488 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1489 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1490 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1493 * DON'T set REF_COWS for log trees
1495 * log trees do not get reference counted because they go away
1496 * before a real commit is actually done. They do store pointers
1497 * to file data extents, and those reference counts still get
1498 * updated (along with back refs to the log tree).
1501 leaf
= btrfs_alloc_tree_block(trans
, root
, 0, BTRFS_TREE_LOG_OBJECTID
,
1505 return ERR_CAST(leaf
);
1508 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1509 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1510 btrfs_set_header_generation(leaf
, trans
->transid
);
1511 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1512 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1515 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1516 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
1517 btrfs_mark_buffer_dirty(root
->node
);
1518 btrfs_tree_unlock(root
->node
);
1522 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1523 struct btrfs_fs_info
*fs_info
)
1525 struct btrfs_root
*log_root
;
1527 log_root
= alloc_log_tree(trans
, fs_info
);
1528 if (IS_ERR(log_root
))
1529 return PTR_ERR(log_root
);
1530 WARN_ON(fs_info
->log_root_tree
);
1531 fs_info
->log_root_tree
= log_root
;
1535 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1536 struct btrfs_root
*root
)
1538 struct btrfs_root
*log_root
;
1539 struct btrfs_inode_item
*inode_item
;
1541 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1542 if (IS_ERR(log_root
))
1543 return PTR_ERR(log_root
);
1545 log_root
->last_trans
= trans
->transid
;
1546 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1548 inode_item
= &log_root
->root_item
.inode
;
1549 btrfs_set_stack_inode_generation(inode_item
, 1);
1550 btrfs_set_stack_inode_size(inode_item
, 3);
1551 btrfs_set_stack_inode_nlink(inode_item
, 1);
1552 btrfs_set_stack_inode_nbytes(inode_item
, root
->nodesize
);
1553 btrfs_set_stack_inode_mode(inode_item
, S_IFDIR
| 0755);
1555 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1557 WARN_ON(root
->log_root
);
1558 root
->log_root
= log_root
;
1559 root
->log_transid
= 0;
1560 root
->log_transid_committed
= -1;
1561 root
->last_log_commit
= 0;
1565 static struct btrfs_root
*btrfs_read_tree_root(struct btrfs_root
*tree_root
,
1566 struct btrfs_key
*key
)
1568 struct btrfs_root
*root
;
1569 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1570 struct btrfs_path
*path
;
1574 path
= btrfs_alloc_path();
1576 return ERR_PTR(-ENOMEM
);
1578 root
= btrfs_alloc_root(fs_info
, GFP_NOFS
);
1584 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
1585 tree_root
->stripesize
, root
, fs_info
, key
->objectid
);
1587 ret
= btrfs_find_root(tree_root
, key
, path
,
1588 &root
->root_item
, &root
->root_key
);
1595 generation
= btrfs_root_generation(&root
->root_item
);
1596 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1598 if (IS_ERR(root
->node
)) {
1599 ret
= PTR_ERR(root
->node
);
1601 } else if (!btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1603 free_extent_buffer(root
->node
);
1606 root
->commit_root
= btrfs_root_node(root
);
1608 btrfs_free_path(path
);
1614 root
= ERR_PTR(ret
);
1618 struct btrfs_root
*btrfs_read_fs_root(struct btrfs_root
*tree_root
,
1619 struct btrfs_key
*location
)
1621 struct btrfs_root
*root
;
1623 root
= btrfs_read_tree_root(tree_root
, location
);
1627 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1628 set_bit(BTRFS_ROOT_REF_COWS
, &root
->state
);
1629 btrfs_check_and_init_root_item(&root
->root_item
);
1635 int btrfs_init_fs_root(struct btrfs_root
*root
)
1638 struct btrfs_subvolume_writers
*writers
;
1640 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1641 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1643 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1648 writers
= btrfs_alloc_subvolume_writers();
1649 if (IS_ERR(writers
)) {
1650 ret
= PTR_ERR(writers
);
1653 root
->subv_writers
= writers
;
1655 btrfs_init_free_ino_ctl(root
);
1656 spin_lock_init(&root
->ino_cache_lock
);
1657 init_waitqueue_head(&root
->ino_cache_wait
);
1659 ret
= get_anon_bdev(&root
->anon_dev
);
1663 mutex_lock(&root
->objectid_mutex
);
1664 ret
= btrfs_find_highest_objectid(root
,
1665 &root
->highest_objectid
);
1667 mutex_unlock(&root
->objectid_mutex
);
1671 ASSERT(root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
1673 mutex_unlock(&root
->objectid_mutex
);
1677 /* the caller is responsible to call free_fs_root */
1681 struct btrfs_root
*btrfs_lookup_fs_root(struct btrfs_fs_info
*fs_info
,
1684 struct btrfs_root
*root
;
1686 spin_lock(&fs_info
->fs_roots_radix_lock
);
1687 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1688 (unsigned long)root_id
);
1689 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1693 int btrfs_insert_fs_root(struct btrfs_fs_info
*fs_info
,
1694 struct btrfs_root
*root
)
1698 ret
= radix_tree_preload(GFP_NOFS
);
1702 spin_lock(&fs_info
->fs_roots_radix_lock
);
1703 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1704 (unsigned long)root
->root_key
.objectid
,
1707 set_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
);
1708 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1709 radix_tree_preload_end();
1714 struct btrfs_root
*btrfs_get_fs_root(struct btrfs_fs_info
*fs_info
,
1715 struct btrfs_key
*location
,
1718 struct btrfs_root
*root
;
1719 struct btrfs_path
*path
;
1720 struct btrfs_key key
;
1723 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1724 return fs_info
->tree_root
;
1725 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1726 return fs_info
->extent_root
;
1727 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1728 return fs_info
->chunk_root
;
1729 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1730 return fs_info
->dev_root
;
1731 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1732 return fs_info
->csum_root
;
1733 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1734 return fs_info
->quota_root
? fs_info
->quota_root
:
1736 if (location
->objectid
== BTRFS_UUID_TREE_OBJECTID
)
1737 return fs_info
->uuid_root
? fs_info
->uuid_root
:
1739 if (location
->objectid
== BTRFS_FREE_SPACE_TREE_OBJECTID
)
1740 return fs_info
->free_space_root
? fs_info
->free_space_root
:
1743 root
= btrfs_lookup_fs_root(fs_info
, location
->objectid
);
1745 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0)
1746 return ERR_PTR(-ENOENT
);
1750 root
= btrfs_read_fs_root(fs_info
->tree_root
, location
);
1754 if (check_ref
&& btrfs_root_refs(&root
->root_item
) == 0) {
1759 ret
= btrfs_init_fs_root(root
);
1763 path
= btrfs_alloc_path();
1768 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1769 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
1770 key
.offset
= location
->objectid
;
1772 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
1773 btrfs_free_path(path
);
1777 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
);
1779 ret
= btrfs_insert_fs_root(fs_info
, root
);
1781 if (ret
== -EEXIST
) {
1790 return ERR_PTR(ret
);
1793 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1795 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1797 struct btrfs_device
*device
;
1798 struct backing_dev_info
*bdi
;
1801 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1804 bdi
= blk_get_backing_dev_info(device
->bdev
);
1805 if (bdi_congested(bdi
, bdi_bits
)) {
1814 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1818 err
= bdi_setup_and_register(bdi
, "btrfs");
1822 bdi
->ra_pages
= VM_MAX_READAHEAD
* 1024 / PAGE_SIZE
;
1823 bdi
->congested_fn
= btrfs_congested_fn
;
1824 bdi
->congested_data
= info
;
1825 bdi
->capabilities
|= BDI_CAP_CGROUP_WRITEBACK
;
1830 * called by the kthread helper functions to finally call the bio end_io
1831 * functions. This is where read checksum verification actually happens
1833 static void end_workqueue_fn(struct btrfs_work
*work
)
1836 struct btrfs_end_io_wq
*end_io_wq
;
1838 end_io_wq
= container_of(work
, struct btrfs_end_io_wq
, work
);
1839 bio
= end_io_wq
->bio
;
1841 bio
->bi_error
= end_io_wq
->error
;
1842 bio
->bi_private
= end_io_wq
->private;
1843 bio
->bi_end_io
= end_io_wq
->end_io
;
1844 kmem_cache_free(btrfs_end_io_wq_cache
, end_io_wq
);
1848 static int cleaner_kthread(void *arg
)
1850 struct btrfs_root
*root
= arg
;
1852 struct btrfs_trans_handle
*trans
;
1857 /* Make the cleaner go to sleep early. */
1858 if (btrfs_need_cleaner_sleep(root
))
1862 * Do not do anything if we might cause open_ctree() to block
1863 * before we have finished mounting the filesystem.
1865 if (!test_bit(BTRFS_FS_OPEN
, &root
->fs_info
->flags
))
1868 if (!mutex_trylock(&root
->fs_info
->cleaner_mutex
))
1872 * Avoid the problem that we change the status of the fs
1873 * during the above check and trylock.
1875 if (btrfs_need_cleaner_sleep(root
)) {
1876 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1880 mutex_lock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
1881 btrfs_run_delayed_iputs(root
);
1882 mutex_unlock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
1884 again
= btrfs_clean_one_deleted_snapshot(root
);
1885 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1888 * The defragger has dealt with the R/O remount and umount,
1889 * needn't do anything special here.
1891 btrfs_run_defrag_inodes(root
->fs_info
);
1894 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1895 * with relocation (btrfs_relocate_chunk) and relocation
1896 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1897 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1898 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1899 * unused block groups.
1901 btrfs_delete_unused_bgs(root
->fs_info
);
1904 set_current_state(TASK_INTERRUPTIBLE
);
1905 if (!kthread_should_stop())
1907 __set_current_state(TASK_RUNNING
);
1909 } while (!kthread_should_stop());
1912 * Transaction kthread is stopped before us and wakes us up.
1913 * However we might have started a new transaction and COWed some
1914 * tree blocks when deleting unused block groups for example. So
1915 * make sure we commit the transaction we started to have a clean
1916 * shutdown when evicting the btree inode - if it has dirty pages
1917 * when we do the final iput() on it, eviction will trigger a
1918 * writeback for it which will fail with null pointer dereferences
1919 * since work queues and other resources were already released and
1920 * destroyed by the time the iput/eviction/writeback is made.
1922 trans
= btrfs_attach_transaction(root
);
1923 if (IS_ERR(trans
)) {
1924 if (PTR_ERR(trans
) != -ENOENT
)
1925 btrfs_err(root
->fs_info
,
1926 "cleaner transaction attach returned %ld",
1931 ret
= btrfs_commit_transaction(trans
, root
);
1933 btrfs_err(root
->fs_info
,
1934 "cleaner open transaction commit returned %d",
1941 static int transaction_kthread(void *arg
)
1943 struct btrfs_root
*root
= arg
;
1944 struct btrfs_trans_handle
*trans
;
1945 struct btrfs_transaction
*cur
;
1948 unsigned long delay
;
1952 cannot_commit
= false;
1953 delay
= HZ
* root
->fs_info
->commit_interval
;
1954 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1956 spin_lock(&root
->fs_info
->trans_lock
);
1957 cur
= root
->fs_info
->running_transaction
;
1959 spin_unlock(&root
->fs_info
->trans_lock
);
1963 now
= get_seconds();
1964 if (cur
->state
< TRANS_STATE_BLOCKED
&&
1965 (now
< cur
->start_time
||
1966 now
- cur
->start_time
< root
->fs_info
->commit_interval
)) {
1967 spin_unlock(&root
->fs_info
->trans_lock
);
1971 transid
= cur
->transid
;
1972 spin_unlock(&root
->fs_info
->trans_lock
);
1974 /* If the file system is aborted, this will always fail. */
1975 trans
= btrfs_attach_transaction(root
);
1976 if (IS_ERR(trans
)) {
1977 if (PTR_ERR(trans
) != -ENOENT
)
1978 cannot_commit
= true;
1981 if (transid
== trans
->transid
) {
1982 btrfs_commit_transaction(trans
, root
);
1984 btrfs_end_transaction(trans
, root
);
1987 wake_up_process(root
->fs_info
->cleaner_kthread
);
1988 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1990 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR
,
1991 &root
->fs_info
->fs_state
)))
1992 btrfs_cleanup_transaction(root
);
1993 set_current_state(TASK_INTERRUPTIBLE
);
1994 if (!kthread_should_stop() &&
1995 (!btrfs_transaction_blocked(root
->fs_info
) ||
1997 schedule_timeout(delay
);
1998 __set_current_state(TASK_RUNNING
);
1999 } while (!kthread_should_stop());
2004 * this will find the highest generation in the array of
2005 * root backups. The index of the highest array is returned,
2006 * or -1 if we can't find anything.
2008 * We check to make sure the array is valid by comparing the
2009 * generation of the latest root in the array with the generation
2010 * in the super block. If they don't match we pitch it.
2012 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
2015 int newest_index
= -1;
2016 struct btrfs_root_backup
*root_backup
;
2019 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
2020 root_backup
= info
->super_copy
->super_roots
+ i
;
2021 cur
= btrfs_backup_tree_root_gen(root_backup
);
2022 if (cur
== newest_gen
)
2026 /* check to see if we actually wrapped around */
2027 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
2028 root_backup
= info
->super_copy
->super_roots
;
2029 cur
= btrfs_backup_tree_root_gen(root_backup
);
2030 if (cur
== newest_gen
)
2033 return newest_index
;
2038 * find the oldest backup so we know where to store new entries
2039 * in the backup array. This will set the backup_root_index
2040 * field in the fs_info struct
2042 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
2045 int newest_index
= -1;
2047 newest_index
= find_newest_super_backup(info
, newest_gen
);
2048 /* if there was garbage in there, just move along */
2049 if (newest_index
== -1) {
2050 info
->backup_root_index
= 0;
2052 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
2057 * copy all the root pointers into the super backup array.
2058 * this will bump the backup pointer by one when it is
2061 static void backup_super_roots(struct btrfs_fs_info
*info
)
2064 struct btrfs_root_backup
*root_backup
;
2067 next_backup
= info
->backup_root_index
;
2068 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
2069 BTRFS_NUM_BACKUP_ROOTS
;
2072 * just overwrite the last backup if we're at the same generation
2073 * this happens only at umount
2075 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
2076 if (btrfs_backup_tree_root_gen(root_backup
) ==
2077 btrfs_header_generation(info
->tree_root
->node
))
2078 next_backup
= last_backup
;
2080 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
2083 * make sure all of our padding and empty slots get zero filled
2084 * regardless of which ones we use today
2086 memset(root_backup
, 0, sizeof(*root_backup
));
2088 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
2090 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
2091 btrfs_set_backup_tree_root_gen(root_backup
,
2092 btrfs_header_generation(info
->tree_root
->node
));
2094 btrfs_set_backup_tree_root_level(root_backup
,
2095 btrfs_header_level(info
->tree_root
->node
));
2097 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
2098 btrfs_set_backup_chunk_root_gen(root_backup
,
2099 btrfs_header_generation(info
->chunk_root
->node
));
2100 btrfs_set_backup_chunk_root_level(root_backup
,
2101 btrfs_header_level(info
->chunk_root
->node
));
2103 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
2104 btrfs_set_backup_extent_root_gen(root_backup
,
2105 btrfs_header_generation(info
->extent_root
->node
));
2106 btrfs_set_backup_extent_root_level(root_backup
,
2107 btrfs_header_level(info
->extent_root
->node
));
2110 * we might commit during log recovery, which happens before we set
2111 * the fs_root. Make sure it is valid before we fill it in.
2113 if (info
->fs_root
&& info
->fs_root
->node
) {
2114 btrfs_set_backup_fs_root(root_backup
,
2115 info
->fs_root
->node
->start
);
2116 btrfs_set_backup_fs_root_gen(root_backup
,
2117 btrfs_header_generation(info
->fs_root
->node
));
2118 btrfs_set_backup_fs_root_level(root_backup
,
2119 btrfs_header_level(info
->fs_root
->node
));
2122 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
2123 btrfs_set_backup_dev_root_gen(root_backup
,
2124 btrfs_header_generation(info
->dev_root
->node
));
2125 btrfs_set_backup_dev_root_level(root_backup
,
2126 btrfs_header_level(info
->dev_root
->node
));
2128 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
2129 btrfs_set_backup_csum_root_gen(root_backup
,
2130 btrfs_header_generation(info
->csum_root
->node
));
2131 btrfs_set_backup_csum_root_level(root_backup
,
2132 btrfs_header_level(info
->csum_root
->node
));
2134 btrfs_set_backup_total_bytes(root_backup
,
2135 btrfs_super_total_bytes(info
->super_copy
));
2136 btrfs_set_backup_bytes_used(root_backup
,
2137 btrfs_super_bytes_used(info
->super_copy
));
2138 btrfs_set_backup_num_devices(root_backup
,
2139 btrfs_super_num_devices(info
->super_copy
));
2142 * if we don't copy this out to the super_copy, it won't get remembered
2143 * for the next commit
2145 memcpy(&info
->super_copy
->super_roots
,
2146 &info
->super_for_commit
->super_roots
,
2147 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
2151 * this copies info out of the root backup array and back into
2152 * the in-memory super block. It is meant to help iterate through
2153 * the array, so you send it the number of backups you've already
2154 * tried and the last backup index you used.
2156 * this returns -1 when it has tried all the backups
2158 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
2159 struct btrfs_super_block
*super
,
2160 int *num_backups_tried
, int *backup_index
)
2162 struct btrfs_root_backup
*root_backup
;
2163 int newest
= *backup_index
;
2165 if (*num_backups_tried
== 0) {
2166 u64 gen
= btrfs_super_generation(super
);
2168 newest
= find_newest_super_backup(info
, gen
);
2172 *backup_index
= newest
;
2173 *num_backups_tried
= 1;
2174 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
2175 /* we've tried all the backups, all done */
2178 /* jump to the next oldest backup */
2179 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
2180 BTRFS_NUM_BACKUP_ROOTS
;
2181 *backup_index
= newest
;
2182 *num_backups_tried
+= 1;
2184 root_backup
= super
->super_roots
+ newest
;
2186 btrfs_set_super_generation(super
,
2187 btrfs_backup_tree_root_gen(root_backup
));
2188 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
2189 btrfs_set_super_root_level(super
,
2190 btrfs_backup_tree_root_level(root_backup
));
2191 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
2194 * fixme: the total bytes and num_devices need to match or we should
2197 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
2198 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
2202 /* helper to cleanup workers */
2203 static void btrfs_stop_all_workers(struct btrfs_fs_info
*fs_info
)
2205 btrfs_destroy_workqueue(fs_info
->fixup_workers
);
2206 btrfs_destroy_workqueue(fs_info
->delalloc_workers
);
2207 btrfs_destroy_workqueue(fs_info
->workers
);
2208 btrfs_destroy_workqueue(fs_info
->endio_workers
);
2209 btrfs_destroy_workqueue(fs_info
->endio_meta_workers
);
2210 btrfs_destroy_workqueue(fs_info
->endio_raid56_workers
);
2211 btrfs_destroy_workqueue(fs_info
->endio_repair_workers
);
2212 btrfs_destroy_workqueue(fs_info
->rmw_workers
);
2213 btrfs_destroy_workqueue(fs_info
->endio_meta_write_workers
);
2214 btrfs_destroy_workqueue(fs_info
->endio_write_workers
);
2215 btrfs_destroy_workqueue(fs_info
->endio_freespace_worker
);
2216 btrfs_destroy_workqueue(fs_info
->submit_workers
);
2217 btrfs_destroy_workqueue(fs_info
->delayed_workers
);
2218 btrfs_destroy_workqueue(fs_info
->caching_workers
);
2219 btrfs_destroy_workqueue(fs_info
->readahead_workers
);
2220 btrfs_destroy_workqueue(fs_info
->flush_workers
);
2221 btrfs_destroy_workqueue(fs_info
->qgroup_rescan_workers
);
2222 btrfs_destroy_workqueue(fs_info
->extent_workers
);
2225 static void free_root_extent_buffers(struct btrfs_root
*root
)
2228 free_extent_buffer(root
->node
);
2229 free_extent_buffer(root
->commit_root
);
2231 root
->commit_root
= NULL
;
2235 /* helper to cleanup tree roots */
2236 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
2238 free_root_extent_buffers(info
->tree_root
);
2240 free_root_extent_buffers(info
->dev_root
);
2241 free_root_extent_buffers(info
->extent_root
);
2242 free_root_extent_buffers(info
->csum_root
);
2243 free_root_extent_buffers(info
->quota_root
);
2244 free_root_extent_buffers(info
->uuid_root
);
2246 free_root_extent_buffers(info
->chunk_root
);
2247 free_root_extent_buffers(info
->free_space_root
);
2250 void btrfs_free_fs_roots(struct btrfs_fs_info
*fs_info
)
2253 struct btrfs_root
*gang
[8];
2256 while (!list_empty(&fs_info
->dead_roots
)) {
2257 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2258 struct btrfs_root
, root_list
);
2259 list_del(&gang
[0]->root_list
);
2261 if (test_bit(BTRFS_ROOT_IN_RADIX
, &gang
[0]->state
)) {
2262 btrfs_drop_and_free_fs_root(fs_info
, gang
[0]);
2264 free_extent_buffer(gang
[0]->node
);
2265 free_extent_buffer(gang
[0]->commit_root
);
2266 btrfs_put_fs_root(gang
[0]);
2271 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2276 for (i
= 0; i
< ret
; i
++)
2277 btrfs_drop_and_free_fs_root(fs_info
, gang
[i
]);
2280 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
2281 btrfs_free_log_root_tree(NULL
, fs_info
);
2282 btrfs_destroy_pinned_extent(fs_info
->tree_root
,
2283 fs_info
->pinned_extents
);
2287 static void btrfs_init_scrub(struct btrfs_fs_info
*fs_info
)
2289 mutex_init(&fs_info
->scrub_lock
);
2290 atomic_set(&fs_info
->scrubs_running
, 0);
2291 atomic_set(&fs_info
->scrub_pause_req
, 0);
2292 atomic_set(&fs_info
->scrubs_paused
, 0);
2293 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2294 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2295 fs_info
->scrub_workers_refcnt
= 0;
2298 static void btrfs_init_balance(struct btrfs_fs_info
*fs_info
)
2300 spin_lock_init(&fs_info
->balance_lock
);
2301 mutex_init(&fs_info
->balance_mutex
);
2302 atomic_set(&fs_info
->balance_running
, 0);
2303 atomic_set(&fs_info
->balance_pause_req
, 0);
2304 atomic_set(&fs_info
->balance_cancel_req
, 0);
2305 fs_info
->balance_ctl
= NULL
;
2306 init_waitqueue_head(&fs_info
->balance_wait_q
);
2309 static void btrfs_init_btree_inode(struct btrfs_fs_info
*fs_info
,
2310 struct btrfs_root
*tree_root
)
2312 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2313 set_nlink(fs_info
->btree_inode
, 1);
2315 * we set the i_size on the btree inode to the max possible int.
2316 * the real end of the address space is determined by all of
2317 * the devices in the system
2319 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
2320 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
2322 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
2323 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
2324 fs_info
->btree_inode
->i_mapping
);
2325 BTRFS_I(fs_info
->btree_inode
)->io_tree
.track_uptodate
= 0;
2326 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
);
2328 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2330 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
2331 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
2332 sizeof(struct btrfs_key
));
2333 set_bit(BTRFS_INODE_DUMMY
,
2334 &BTRFS_I(fs_info
->btree_inode
)->runtime_flags
);
2335 btrfs_insert_inode_hash(fs_info
->btree_inode
);
2338 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info
*fs_info
)
2340 fs_info
->dev_replace
.lock_owner
= 0;
2341 atomic_set(&fs_info
->dev_replace
.nesting_level
, 0);
2342 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2343 rwlock_init(&fs_info
->dev_replace
.lock
);
2344 atomic_set(&fs_info
->dev_replace
.read_locks
, 0);
2345 atomic_set(&fs_info
->dev_replace
.blocking_readers
, 0);
2346 init_waitqueue_head(&fs_info
->replace_wait
);
2347 init_waitqueue_head(&fs_info
->dev_replace
.read_lock_wq
);
2350 static void btrfs_init_qgroup(struct btrfs_fs_info
*fs_info
)
2352 spin_lock_init(&fs_info
->qgroup_lock
);
2353 mutex_init(&fs_info
->qgroup_ioctl_lock
);
2354 fs_info
->qgroup_tree
= RB_ROOT
;
2355 fs_info
->qgroup_op_tree
= RB_ROOT
;
2356 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2357 fs_info
->qgroup_seq
= 1;
2358 fs_info
->qgroup_ulist
= NULL
;
2359 fs_info
->qgroup_rescan_running
= false;
2360 mutex_init(&fs_info
->qgroup_rescan_lock
);
2363 static int btrfs_init_workqueues(struct btrfs_fs_info
*fs_info
,
2364 struct btrfs_fs_devices
*fs_devices
)
2366 int max_active
= fs_info
->thread_pool_size
;
2367 unsigned int flags
= WQ_MEM_RECLAIM
| WQ_FREEZABLE
| WQ_UNBOUND
;
2370 btrfs_alloc_workqueue(fs_info
, "worker",
2371 flags
| WQ_HIGHPRI
, max_active
, 16);
2373 fs_info
->delalloc_workers
=
2374 btrfs_alloc_workqueue(fs_info
, "delalloc",
2375 flags
, max_active
, 2);
2377 fs_info
->flush_workers
=
2378 btrfs_alloc_workqueue(fs_info
, "flush_delalloc",
2379 flags
, max_active
, 0);
2381 fs_info
->caching_workers
=
2382 btrfs_alloc_workqueue(fs_info
, "cache", flags
, max_active
, 0);
2385 * a higher idle thresh on the submit workers makes it much more
2386 * likely that bios will be send down in a sane order to the
2389 fs_info
->submit_workers
=
2390 btrfs_alloc_workqueue(fs_info
, "submit", flags
,
2391 min_t(u64
, fs_devices
->num_devices
,
2394 fs_info
->fixup_workers
=
2395 btrfs_alloc_workqueue(fs_info
, "fixup", flags
, 1, 0);
2398 * endios are largely parallel and should have a very
2401 fs_info
->endio_workers
=
2402 btrfs_alloc_workqueue(fs_info
, "endio", flags
, max_active
, 4);
2403 fs_info
->endio_meta_workers
=
2404 btrfs_alloc_workqueue(fs_info
, "endio-meta", flags
,
2406 fs_info
->endio_meta_write_workers
=
2407 btrfs_alloc_workqueue(fs_info
, "endio-meta-write", flags
,
2409 fs_info
->endio_raid56_workers
=
2410 btrfs_alloc_workqueue(fs_info
, "endio-raid56", flags
,
2412 fs_info
->endio_repair_workers
=
2413 btrfs_alloc_workqueue(fs_info
, "endio-repair", flags
, 1, 0);
2414 fs_info
->rmw_workers
=
2415 btrfs_alloc_workqueue(fs_info
, "rmw", flags
, max_active
, 2);
2416 fs_info
->endio_write_workers
=
2417 btrfs_alloc_workqueue(fs_info
, "endio-write", flags
,
2419 fs_info
->endio_freespace_worker
=
2420 btrfs_alloc_workqueue(fs_info
, "freespace-write", flags
,
2422 fs_info
->delayed_workers
=
2423 btrfs_alloc_workqueue(fs_info
, "delayed-meta", flags
,
2425 fs_info
->readahead_workers
=
2426 btrfs_alloc_workqueue(fs_info
, "readahead", flags
,
2428 fs_info
->qgroup_rescan_workers
=
2429 btrfs_alloc_workqueue(fs_info
, "qgroup-rescan", flags
, 1, 0);
2430 fs_info
->extent_workers
=
2431 btrfs_alloc_workqueue(fs_info
, "extent-refs", flags
,
2432 min_t(u64
, fs_devices
->num_devices
,
2435 if (!(fs_info
->workers
&& fs_info
->delalloc_workers
&&
2436 fs_info
->submit_workers
&& fs_info
->flush_workers
&&
2437 fs_info
->endio_workers
&& fs_info
->endio_meta_workers
&&
2438 fs_info
->endio_meta_write_workers
&&
2439 fs_info
->endio_repair_workers
&&
2440 fs_info
->endio_write_workers
&& fs_info
->endio_raid56_workers
&&
2441 fs_info
->endio_freespace_worker
&& fs_info
->rmw_workers
&&
2442 fs_info
->caching_workers
&& fs_info
->readahead_workers
&&
2443 fs_info
->fixup_workers
&& fs_info
->delayed_workers
&&
2444 fs_info
->extent_workers
&&
2445 fs_info
->qgroup_rescan_workers
)) {
2452 static int btrfs_replay_log(struct btrfs_fs_info
*fs_info
,
2453 struct btrfs_fs_devices
*fs_devices
)
2456 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
2457 struct btrfs_root
*log_tree_root
;
2458 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2459 u64 bytenr
= btrfs_super_log_root(disk_super
);
2461 if (fs_devices
->rw_devices
== 0) {
2462 btrfs_warn(fs_info
, "log replay required on RO media");
2466 log_tree_root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
2470 __setup_root(tree_root
->nodesize
, tree_root
->sectorsize
,
2471 tree_root
->stripesize
, log_tree_root
, fs_info
,
2472 BTRFS_TREE_LOG_OBJECTID
);
2474 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
2475 fs_info
->generation
+ 1);
2476 if (IS_ERR(log_tree_root
->node
)) {
2477 btrfs_warn(fs_info
, "failed to read log tree");
2478 ret
= PTR_ERR(log_tree_root
->node
);
2479 kfree(log_tree_root
);
2481 } else if (!extent_buffer_uptodate(log_tree_root
->node
)) {
2482 btrfs_err(fs_info
, "failed to read log tree");
2483 free_extent_buffer(log_tree_root
->node
);
2484 kfree(log_tree_root
);
2487 /* returns with log_tree_root freed on success */
2488 ret
= btrfs_recover_log_trees(log_tree_root
);
2490 btrfs_handle_fs_error(tree_root
->fs_info
, ret
,
2491 "Failed to recover log tree");
2492 free_extent_buffer(log_tree_root
->node
);
2493 kfree(log_tree_root
);
2497 if (fs_info
->sb
->s_flags
& MS_RDONLY
) {
2498 ret
= btrfs_commit_super(tree_root
);
2506 static int btrfs_read_roots(struct btrfs_fs_info
*fs_info
,
2507 struct btrfs_root
*tree_root
)
2509 struct btrfs_root
*root
;
2510 struct btrfs_key location
;
2513 location
.objectid
= BTRFS_EXTENT_TREE_OBJECTID
;
2514 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2515 location
.offset
= 0;
2517 root
= btrfs_read_tree_root(tree_root
, &location
);
2519 return PTR_ERR(root
);
2520 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2521 fs_info
->extent_root
= root
;
2523 location
.objectid
= BTRFS_DEV_TREE_OBJECTID
;
2524 root
= btrfs_read_tree_root(tree_root
, &location
);
2526 return PTR_ERR(root
);
2527 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2528 fs_info
->dev_root
= root
;
2529 btrfs_init_devices_late(fs_info
);
2531 location
.objectid
= BTRFS_CSUM_TREE_OBJECTID
;
2532 root
= btrfs_read_tree_root(tree_root
, &location
);
2534 return PTR_ERR(root
);
2535 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2536 fs_info
->csum_root
= root
;
2538 location
.objectid
= BTRFS_QUOTA_TREE_OBJECTID
;
2539 root
= btrfs_read_tree_root(tree_root
, &location
);
2540 if (!IS_ERR(root
)) {
2541 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2542 set_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
);
2543 fs_info
->quota_root
= root
;
2546 location
.objectid
= BTRFS_UUID_TREE_OBJECTID
;
2547 root
= btrfs_read_tree_root(tree_root
, &location
);
2549 ret
= PTR_ERR(root
);
2553 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2554 fs_info
->uuid_root
= root
;
2557 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
2558 location
.objectid
= BTRFS_FREE_SPACE_TREE_OBJECTID
;
2559 root
= btrfs_read_tree_root(tree_root
, &location
);
2561 return PTR_ERR(root
);
2562 set_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
);
2563 fs_info
->free_space_root
= root
;
2569 int open_ctree(struct super_block
*sb
,
2570 struct btrfs_fs_devices
*fs_devices
,
2578 struct btrfs_key location
;
2579 struct buffer_head
*bh
;
2580 struct btrfs_super_block
*disk_super
;
2581 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
2582 struct btrfs_root
*tree_root
;
2583 struct btrfs_root
*chunk_root
;
2586 int num_backups_tried
= 0;
2587 int backup_index
= 0;
2589 int clear_free_space_tree
= 0;
2591 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
2592 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
, GFP_KERNEL
);
2593 if (!tree_root
|| !chunk_root
) {
2598 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
2604 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
2610 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0, GFP_KERNEL
);
2615 fs_info
->dirty_metadata_batch
= PAGE_SIZE
*
2616 (1 + ilog2(nr_cpu_ids
));
2618 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0, GFP_KERNEL
);
2621 goto fail_dirty_metadata_bytes
;
2624 ret
= percpu_counter_init(&fs_info
->bio_counter
, 0, GFP_KERNEL
);
2627 goto fail_delalloc_bytes
;
2630 fs_info
->btree_inode
= new_inode(sb
);
2631 if (!fs_info
->btree_inode
) {
2633 goto fail_bio_counter
;
2636 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2638 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2639 INIT_RADIX_TREE(&fs_info
->buffer_radix
, GFP_ATOMIC
);
2640 INIT_LIST_HEAD(&fs_info
->trans_list
);
2641 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2642 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2643 INIT_LIST_HEAD(&fs_info
->delalloc_roots
);
2644 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2645 spin_lock_init(&fs_info
->delalloc_root_lock
);
2646 spin_lock_init(&fs_info
->trans_lock
);
2647 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2648 spin_lock_init(&fs_info
->delayed_iput_lock
);
2649 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2650 spin_lock_init(&fs_info
->free_chunk_lock
);
2651 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2652 spin_lock_init(&fs_info
->super_lock
);
2653 spin_lock_init(&fs_info
->qgroup_op_lock
);
2654 spin_lock_init(&fs_info
->buffer_lock
);
2655 spin_lock_init(&fs_info
->unused_bgs_lock
);
2656 rwlock_init(&fs_info
->tree_mod_log_lock
);
2657 mutex_init(&fs_info
->unused_bg_unpin_mutex
);
2658 mutex_init(&fs_info
->delete_unused_bgs_mutex
);
2659 mutex_init(&fs_info
->reloc_mutex
);
2660 mutex_init(&fs_info
->delalloc_root_mutex
);
2661 mutex_init(&fs_info
->cleaner_delayed_iput_mutex
);
2662 seqlock_init(&fs_info
->profiles_lock
);
2664 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2665 INIT_LIST_HEAD(&fs_info
->space_info
);
2666 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2667 INIT_LIST_HEAD(&fs_info
->unused_bgs
);
2668 btrfs_mapping_init(&fs_info
->mapping_tree
);
2669 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2670 BTRFS_BLOCK_RSV_GLOBAL
);
2671 btrfs_init_block_rsv(&fs_info
->delalloc_block_rsv
,
2672 BTRFS_BLOCK_RSV_DELALLOC
);
2673 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2674 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2675 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2676 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2677 BTRFS_BLOCK_RSV_DELOPS
);
2678 atomic_set(&fs_info
->nr_async_submits
, 0);
2679 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2680 atomic_set(&fs_info
->async_submit_draining
, 0);
2681 atomic_set(&fs_info
->nr_async_bios
, 0);
2682 atomic_set(&fs_info
->defrag_running
, 0);
2683 atomic_set(&fs_info
->qgroup_op_seq
, 0);
2684 atomic_set(&fs_info
->reada_works_cnt
, 0);
2685 atomic64_set(&fs_info
->tree_mod_seq
, 0);
2686 fs_info
->fs_frozen
= 0;
2688 fs_info
->max_inline
= BTRFS_DEFAULT_MAX_INLINE
;
2689 fs_info
->metadata_ratio
= 0;
2690 fs_info
->defrag_inodes
= RB_ROOT
;
2691 fs_info
->free_chunk_space
= 0;
2692 fs_info
->tree_mod_log
= RB_ROOT
;
2693 fs_info
->commit_interval
= BTRFS_DEFAULT_COMMIT_INTERVAL
;
2694 fs_info
->avg_delayed_ref_runtime
= NSEC_PER_SEC
>> 6; /* div by 64 */
2695 /* readahead state */
2696 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
2697 spin_lock_init(&fs_info
->reada_lock
);
2699 fs_info
->thread_pool_size
= min_t(unsigned long,
2700 num_online_cpus() + 2, 8);
2702 INIT_LIST_HEAD(&fs_info
->ordered_roots
);
2703 spin_lock_init(&fs_info
->ordered_root_lock
);
2704 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2706 if (!fs_info
->delayed_root
) {
2710 btrfs_init_delayed_root(fs_info
->delayed_root
);
2712 btrfs_init_scrub(fs_info
);
2713 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2714 fs_info
->check_integrity_print_mask
= 0;
2716 btrfs_init_balance(fs_info
);
2717 btrfs_init_async_reclaim_work(&fs_info
->async_reclaim_work
);
2719 sb
->s_blocksize
= 4096;
2720 sb
->s_blocksize_bits
= blksize_bits(4096);
2721 sb
->s_bdi
= &fs_info
->bdi
;
2723 btrfs_init_btree_inode(fs_info
, tree_root
);
2725 spin_lock_init(&fs_info
->block_group_cache_lock
);
2726 fs_info
->block_group_cache_tree
= RB_ROOT
;
2727 fs_info
->first_logical_byte
= (u64
)-1;
2729 extent_io_tree_init(&fs_info
->freed_extents
[0],
2730 fs_info
->btree_inode
->i_mapping
);
2731 extent_io_tree_init(&fs_info
->freed_extents
[1],
2732 fs_info
->btree_inode
->i_mapping
);
2733 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2734 set_bit(BTRFS_FS_BARRIER
, &fs_info
->flags
);
2736 mutex_init(&fs_info
->ordered_operations_mutex
);
2737 mutex_init(&fs_info
->tree_log_mutex
);
2738 mutex_init(&fs_info
->chunk_mutex
);
2739 mutex_init(&fs_info
->transaction_kthread_mutex
);
2740 mutex_init(&fs_info
->cleaner_mutex
);
2741 mutex_init(&fs_info
->volume_mutex
);
2742 mutex_init(&fs_info
->ro_block_group_mutex
);
2743 init_rwsem(&fs_info
->commit_root_sem
);
2744 init_rwsem(&fs_info
->cleanup_work_sem
);
2745 init_rwsem(&fs_info
->subvol_sem
);
2746 sema_init(&fs_info
->uuid_tree_rescan_sem
, 1);
2748 btrfs_init_dev_replace_locks(fs_info
);
2749 btrfs_init_qgroup(fs_info
);
2751 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2752 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2754 init_waitqueue_head(&fs_info
->transaction_throttle
);
2755 init_waitqueue_head(&fs_info
->transaction_wait
);
2756 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2757 init_waitqueue_head(&fs_info
->async_submit_wait
);
2759 INIT_LIST_HEAD(&fs_info
->pinned_chunks
);
2761 ret
= btrfs_alloc_stripe_hash_table(fs_info
);
2767 __setup_root(4096, 4096, 4096, tree_root
,
2768 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2770 invalidate_bdev(fs_devices
->latest_bdev
);
2773 * Read super block and check the signature bytes only
2775 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2782 * We want to check superblock checksum, the type is stored inside.
2783 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2785 if (btrfs_check_super_csum(fs_info
, bh
->b_data
)) {
2786 btrfs_err(fs_info
, "superblock checksum mismatch");
2793 * super_copy is zeroed at allocation time and we never touch the
2794 * following bytes up to INFO_SIZE, the checksum is calculated from
2795 * the whole block of INFO_SIZE
2797 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2798 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2799 sizeof(*fs_info
->super_for_commit
));
2802 memcpy(fs_info
->fsid
, fs_info
->super_copy
->fsid
, BTRFS_FSID_SIZE
);
2804 ret
= btrfs_check_super_valid(fs_info
, sb
->s_flags
& MS_RDONLY
);
2806 btrfs_err(fs_info
, "superblock contains fatal errors");
2811 disk_super
= fs_info
->super_copy
;
2812 if (!btrfs_super_root(disk_super
))
2815 /* check FS state, whether FS is broken. */
2816 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2817 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2820 * run through our array of backup supers and setup
2821 * our ring pointer to the oldest one
2823 generation
= btrfs_super_generation(disk_super
);
2824 find_oldest_super_backup(fs_info
, generation
);
2827 * In the long term, we'll store the compression type in the super
2828 * block, and it'll be used for per file compression control.
2830 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2832 ret
= btrfs_parse_options(tree_root
, options
, sb
->s_flags
);
2838 features
= btrfs_super_incompat_flags(disk_super
) &
2839 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2842 "cannot mount because of unsupported optional features (%llx)",
2848 features
= btrfs_super_incompat_flags(disk_super
);
2849 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2850 if (tree_root
->fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2851 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2853 if (features
& BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA
)
2854 btrfs_info(fs_info
, "has skinny extents");
2857 * flag our filesystem as having big metadata blocks if
2858 * they are bigger than the page size
2860 if (btrfs_super_nodesize(disk_super
) > PAGE_SIZE
) {
2861 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2863 "flagging fs with big metadata feature");
2864 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2867 nodesize
= btrfs_super_nodesize(disk_super
);
2868 sectorsize
= btrfs_super_sectorsize(disk_super
);
2869 stripesize
= sectorsize
;
2870 fs_info
->dirty_metadata_batch
= nodesize
* (1 + ilog2(nr_cpu_ids
));
2871 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
2874 * mixed block groups end up with duplicate but slightly offset
2875 * extent buffers for the same range. It leads to corruptions
2877 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2878 (sectorsize
!= nodesize
)) {
2880 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2881 nodesize
, sectorsize
);
2886 * Needn't use the lock because there is no other task which will
2889 btrfs_set_super_incompat_flags(disk_super
, features
);
2891 features
= btrfs_super_compat_ro_flags(disk_super
) &
2892 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2893 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
2895 "cannot mount read-write because of unsupported optional features (%llx)",
2901 max_active
= fs_info
->thread_pool_size
;
2903 ret
= btrfs_init_workqueues(fs_info
, fs_devices
);
2906 goto fail_sb_buffer
;
2909 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
2910 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
2913 tree_root
->nodesize
= nodesize
;
2914 tree_root
->sectorsize
= sectorsize
;
2915 tree_root
->stripesize
= stripesize
;
2917 sb
->s_blocksize
= sectorsize
;
2918 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
2920 mutex_lock(&fs_info
->chunk_mutex
);
2921 ret
= btrfs_read_sys_array(tree_root
);
2922 mutex_unlock(&fs_info
->chunk_mutex
);
2924 btrfs_err(fs_info
, "failed to read the system array: %d", ret
);
2925 goto fail_sb_buffer
;
2928 generation
= btrfs_super_chunk_root_generation(disk_super
);
2930 __setup_root(nodesize
, sectorsize
, stripesize
, chunk_root
,
2931 fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2933 chunk_root
->node
= read_tree_block(chunk_root
,
2934 btrfs_super_chunk_root(disk_super
),
2936 if (IS_ERR(chunk_root
->node
) ||
2937 !extent_buffer_uptodate(chunk_root
->node
)) {
2938 btrfs_err(fs_info
, "failed to read chunk root");
2939 if (!IS_ERR(chunk_root
->node
))
2940 free_extent_buffer(chunk_root
->node
);
2941 chunk_root
->node
= NULL
;
2942 goto fail_tree_roots
;
2944 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2945 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2947 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2948 btrfs_header_chunk_tree_uuid(chunk_root
->node
), BTRFS_UUID_SIZE
);
2950 ret
= btrfs_read_chunk_tree(chunk_root
);
2952 btrfs_err(fs_info
, "failed to read chunk tree: %d", ret
);
2953 goto fail_tree_roots
;
2957 * keep the device that is marked to be the target device for the
2958 * dev_replace procedure
2960 btrfs_close_extra_devices(fs_devices
, 0);
2962 if (!fs_devices
->latest_bdev
) {
2963 btrfs_err(fs_info
, "failed to read devices");
2964 goto fail_tree_roots
;
2968 generation
= btrfs_super_generation(disk_super
);
2970 tree_root
->node
= read_tree_block(tree_root
,
2971 btrfs_super_root(disk_super
),
2973 if (IS_ERR(tree_root
->node
) ||
2974 !extent_buffer_uptodate(tree_root
->node
)) {
2975 btrfs_warn(fs_info
, "failed to read tree root");
2976 if (!IS_ERR(tree_root
->node
))
2977 free_extent_buffer(tree_root
->node
);
2978 tree_root
->node
= NULL
;
2979 goto recovery_tree_root
;
2982 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2983 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2984 btrfs_set_root_refs(&tree_root
->root_item
, 1);
2986 mutex_lock(&tree_root
->objectid_mutex
);
2987 ret
= btrfs_find_highest_objectid(tree_root
,
2988 &tree_root
->highest_objectid
);
2990 mutex_unlock(&tree_root
->objectid_mutex
);
2991 goto recovery_tree_root
;
2994 ASSERT(tree_root
->highest_objectid
<= BTRFS_LAST_FREE_OBJECTID
);
2996 mutex_unlock(&tree_root
->objectid_mutex
);
2998 ret
= btrfs_read_roots(fs_info
, tree_root
);
3000 goto recovery_tree_root
;
3002 fs_info
->generation
= generation
;
3003 fs_info
->last_trans_committed
= generation
;
3005 ret
= btrfs_recover_balance(fs_info
);
3007 btrfs_err(fs_info
, "failed to recover balance: %d", ret
);
3008 goto fail_block_groups
;
3011 ret
= btrfs_init_dev_stats(fs_info
);
3013 btrfs_err(fs_info
, "failed to init dev_stats: %d", ret
);
3014 goto fail_block_groups
;
3017 ret
= btrfs_init_dev_replace(fs_info
);
3019 btrfs_err(fs_info
, "failed to init dev_replace: %d", ret
);
3020 goto fail_block_groups
;
3023 btrfs_close_extra_devices(fs_devices
, 1);
3025 ret
= btrfs_sysfs_add_fsid(fs_devices
, NULL
);
3027 btrfs_err(fs_info
, "failed to init sysfs fsid interface: %d",
3029 goto fail_block_groups
;
3032 ret
= btrfs_sysfs_add_device(fs_devices
);
3034 btrfs_err(fs_info
, "failed to init sysfs device interface: %d",
3036 goto fail_fsdev_sysfs
;
3039 ret
= btrfs_sysfs_add_mounted(fs_info
);
3041 btrfs_err(fs_info
, "failed to init sysfs interface: %d", ret
);
3042 goto fail_fsdev_sysfs
;
3045 ret
= btrfs_init_space_info(fs_info
);
3047 btrfs_err(fs_info
, "failed to initialize space info: %d", ret
);
3051 ret
= btrfs_read_block_groups(fs_info
->extent_root
);
3053 btrfs_err(fs_info
, "failed to read block groups: %d", ret
);
3056 fs_info
->num_tolerated_disk_barrier_failures
=
3057 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3058 if (fs_info
->fs_devices
->missing_devices
>
3059 fs_info
->num_tolerated_disk_barrier_failures
&&
3060 !(sb
->s_flags
& MS_RDONLY
)) {
3062 "missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
3063 fs_info
->fs_devices
->missing_devices
,
3064 fs_info
->num_tolerated_disk_barrier_failures
);
3068 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
3070 if (IS_ERR(fs_info
->cleaner_kthread
))
3073 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
3075 "btrfs-transaction");
3076 if (IS_ERR(fs_info
->transaction_kthread
))
3079 if (!btrfs_test_opt(tree_root
->fs_info
, SSD
) &&
3080 !btrfs_test_opt(tree_root
->fs_info
, NOSSD
) &&
3081 !fs_info
->fs_devices
->rotating
) {
3082 btrfs_info(fs_info
, "detected SSD devices, enabling SSD mode");
3083 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
3087 * Mount does not set all options immediately, we can do it now and do
3088 * not have to wait for transaction commit
3090 btrfs_apply_pending_changes(fs_info
);
3092 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3093 if (btrfs_test_opt(tree_root
->fs_info
, CHECK_INTEGRITY
)) {
3094 ret
= btrfsic_mount(tree_root
, fs_devices
,
3095 btrfs_test_opt(tree_root
->fs_info
,
3096 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
3098 fs_info
->check_integrity_print_mask
);
3101 "failed to initialize integrity check module: %d",
3105 ret
= btrfs_read_qgroup_config(fs_info
);
3107 goto fail_trans_kthread
;
3109 /* do not make disk changes in broken FS or nologreplay is given */
3110 if (btrfs_super_log_root(disk_super
) != 0 &&
3111 !btrfs_test_opt(tree_root
->fs_info
, NOLOGREPLAY
)) {
3112 ret
= btrfs_replay_log(fs_info
, fs_devices
);
3119 ret
= btrfs_find_orphan_roots(tree_root
);
3123 if (!(sb
->s_flags
& MS_RDONLY
)) {
3124 ret
= btrfs_cleanup_fs_roots(fs_info
);
3128 mutex_lock(&fs_info
->cleaner_mutex
);
3129 ret
= btrfs_recover_relocation(tree_root
);
3130 mutex_unlock(&fs_info
->cleaner_mutex
);
3132 btrfs_warn(fs_info
, "failed to recover relocation: %d",
3139 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
3140 location
.type
= BTRFS_ROOT_ITEM_KEY
;
3141 location
.offset
= 0;
3143 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
3144 if (IS_ERR(fs_info
->fs_root
)) {
3145 err
= PTR_ERR(fs_info
->fs_root
);
3149 if (sb
->s_flags
& MS_RDONLY
)
3152 if (btrfs_test_opt(fs_info
, CLEAR_CACHE
) &&
3153 btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
3154 clear_free_space_tree
= 1;
3155 } else if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
) &&
3156 !btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE_VALID
)) {
3157 btrfs_warn(fs_info
, "free space tree is invalid");
3158 clear_free_space_tree
= 1;
3161 if (clear_free_space_tree
) {
3162 btrfs_info(fs_info
, "clearing free space tree");
3163 ret
= btrfs_clear_free_space_tree(fs_info
);
3166 "failed to clear free space tree: %d", ret
);
3167 close_ctree(tree_root
);
3172 if (btrfs_test_opt(tree_root
->fs_info
, FREE_SPACE_TREE
) &&
3173 !btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
)) {
3174 btrfs_info(fs_info
, "creating free space tree");
3175 ret
= btrfs_create_free_space_tree(fs_info
);
3178 "failed to create free space tree: %d", ret
);
3179 close_ctree(tree_root
);
3184 down_read(&fs_info
->cleanup_work_sem
);
3185 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
3186 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
3187 up_read(&fs_info
->cleanup_work_sem
);
3188 close_ctree(tree_root
);
3191 up_read(&fs_info
->cleanup_work_sem
);
3193 ret
= btrfs_resume_balance_async(fs_info
);
3195 btrfs_warn(fs_info
, "failed to resume balance: %d", ret
);
3196 close_ctree(tree_root
);
3200 ret
= btrfs_resume_dev_replace_async(fs_info
);
3202 btrfs_warn(fs_info
, "failed to resume device replace: %d", ret
);
3203 close_ctree(tree_root
);
3207 btrfs_qgroup_rescan_resume(fs_info
);
3209 if (!fs_info
->uuid_root
) {
3210 btrfs_info(fs_info
, "creating UUID tree");
3211 ret
= btrfs_create_uuid_tree(fs_info
);
3214 "failed to create the UUID tree: %d", ret
);
3215 close_ctree(tree_root
);
3218 } else if (btrfs_test_opt(tree_root
->fs_info
, RESCAN_UUID_TREE
) ||
3219 fs_info
->generation
!=
3220 btrfs_super_uuid_tree_generation(disk_super
)) {
3221 btrfs_info(fs_info
, "checking UUID tree");
3222 ret
= btrfs_check_uuid_tree(fs_info
);
3225 "failed to check the UUID tree: %d", ret
);
3226 close_ctree(tree_root
);
3230 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
);
3232 set_bit(BTRFS_FS_OPEN
, &fs_info
->flags
);
3235 * backuproot only affect mount behavior, and if open_ctree succeeded,
3236 * no need to keep the flag
3238 btrfs_clear_opt(fs_info
->mount_opt
, USEBACKUPROOT
);
3243 btrfs_free_qgroup_config(fs_info
);
3245 kthread_stop(fs_info
->transaction_kthread
);
3246 btrfs_cleanup_transaction(fs_info
->tree_root
);
3247 btrfs_free_fs_roots(fs_info
);
3249 kthread_stop(fs_info
->cleaner_kthread
);
3252 * make sure we're done with the btree inode before we stop our
3255 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
3258 btrfs_sysfs_remove_mounted(fs_info
);
3261 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3264 btrfs_put_block_group_cache(fs_info
);
3265 btrfs_free_block_groups(fs_info
);
3268 free_root_pointers(fs_info
, 1);
3269 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3272 btrfs_stop_all_workers(fs_info
);
3275 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3277 iput(fs_info
->btree_inode
);
3279 percpu_counter_destroy(&fs_info
->bio_counter
);
3280 fail_delalloc_bytes
:
3281 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3282 fail_dirty_metadata_bytes
:
3283 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3285 bdi_destroy(&fs_info
->bdi
);
3287 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3289 btrfs_free_stripe_hash_table(fs_info
);
3290 btrfs_close_devices(fs_info
->fs_devices
);
3294 if (!btrfs_test_opt(tree_root
->fs_info
, USEBACKUPROOT
))
3295 goto fail_tree_roots
;
3297 free_root_pointers(fs_info
, 0);
3299 /* don't use the log in recovery mode, it won't be valid */
3300 btrfs_set_super_log_root(disk_super
, 0);
3302 /* we can't trust the free space cache either */
3303 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
3305 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
3306 &num_backups_tried
, &backup_index
);
3308 goto fail_block_groups
;
3309 goto retry_root_backup
;
3312 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
3315 set_buffer_uptodate(bh
);
3317 struct btrfs_device
*device
= (struct btrfs_device
*)
3320 btrfs_warn_rl_in_rcu(device
->dev_root
->fs_info
,
3321 "lost page write due to IO error on %s",
3322 rcu_str_deref(device
->name
));
3323 /* note, we don't set_buffer_write_io_error because we have
3324 * our own ways of dealing with the IO errors
3326 clear_buffer_uptodate(bh
);
3327 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
3333 int btrfs_read_dev_one_super(struct block_device
*bdev
, int copy_num
,
3334 struct buffer_head
**bh_ret
)
3336 struct buffer_head
*bh
;
3337 struct btrfs_super_block
*super
;
3340 bytenr
= btrfs_sb_offset(copy_num
);
3341 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= i_size_read(bdev
->bd_inode
))
3344 bh
= __bread(bdev
, bytenr
/ 4096, BTRFS_SUPER_INFO_SIZE
);
3346 * If we fail to read from the underlying devices, as of now
3347 * the best option we have is to mark it EIO.
3352 super
= (struct btrfs_super_block
*)bh
->b_data
;
3353 if (btrfs_super_bytenr(super
) != bytenr
||
3354 btrfs_super_magic(super
) != BTRFS_MAGIC
) {
3364 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
3366 struct buffer_head
*bh
;
3367 struct buffer_head
*latest
= NULL
;
3368 struct btrfs_super_block
*super
;
3373 /* we would like to check all the supers, but that would make
3374 * a btrfs mount succeed after a mkfs from a different FS.
3375 * So, we need to add a special mount option to scan for
3376 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3378 for (i
= 0; i
< 1; i
++) {
3379 ret
= btrfs_read_dev_one_super(bdev
, i
, &bh
);
3383 super
= (struct btrfs_super_block
*)bh
->b_data
;
3385 if (!latest
|| btrfs_super_generation(super
) > transid
) {
3388 transid
= btrfs_super_generation(super
);
3395 return ERR_PTR(ret
);
3401 * this should be called twice, once with wait == 0 and
3402 * once with wait == 1. When wait == 0 is done, all the buffer heads
3403 * we write are pinned.
3405 * They are released when wait == 1 is done.
3406 * max_mirrors must be the same for both runs, and it indicates how
3407 * many supers on this one device should be written.
3409 * max_mirrors == 0 means to write them all.
3411 static int write_dev_supers(struct btrfs_device
*device
,
3412 struct btrfs_super_block
*sb
,
3413 int do_barriers
, int wait
, int max_mirrors
)
3415 struct buffer_head
*bh
;
3422 if (max_mirrors
== 0)
3423 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
3425 for (i
= 0; i
< max_mirrors
; i
++) {
3426 bytenr
= btrfs_sb_offset(i
);
3427 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
3428 device
->commit_total_bytes
)
3432 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
3433 BTRFS_SUPER_INFO_SIZE
);
3439 if (!buffer_uptodate(bh
))
3442 /* drop our reference */
3445 /* drop the reference from the wait == 0 run */
3449 btrfs_set_super_bytenr(sb
, bytenr
);
3452 crc
= btrfs_csum_data((char *)sb
+
3453 BTRFS_CSUM_SIZE
, crc
,
3454 BTRFS_SUPER_INFO_SIZE
-
3456 btrfs_csum_final(crc
, sb
->csum
);
3459 * one reference for us, and we leave it for the
3462 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
3463 BTRFS_SUPER_INFO_SIZE
);
3465 btrfs_err(device
->dev_root
->fs_info
,
3466 "couldn't get super buffer head for bytenr %llu",
3472 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
3474 /* one reference for submit_bh */
3477 set_buffer_uptodate(bh
);
3479 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
3480 bh
->b_private
= device
;
3484 * we fua the first super. The others we allow
3488 ret
= btrfsic_submit_bh(REQ_OP_WRITE
, WRITE_FUA
, bh
);
3490 ret
= btrfsic_submit_bh(REQ_OP_WRITE
, WRITE_SYNC
, bh
);
3494 return errors
< i
? 0 : -1;
3498 * endio for the write_dev_flush, this will wake anyone waiting
3499 * for the barrier when it is done
3501 static void btrfs_end_empty_barrier(struct bio
*bio
)
3503 if (bio
->bi_private
)
3504 complete(bio
->bi_private
);
3509 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3510 * sent down. With wait == 1, it waits for the previous flush.
3512 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3515 static int write_dev_flush(struct btrfs_device
*device
, int wait
)
3520 if (device
->nobarriers
)
3524 bio
= device
->flush_bio
;
3528 wait_for_completion(&device
->flush_wait
);
3530 if (bio
->bi_error
) {
3531 ret
= bio
->bi_error
;
3532 btrfs_dev_stat_inc_and_print(device
,
3533 BTRFS_DEV_STAT_FLUSH_ERRS
);
3536 /* drop the reference from the wait == 0 run */
3538 device
->flush_bio
= NULL
;
3544 * one reference for us, and we leave it for the
3547 device
->flush_bio
= NULL
;
3548 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 0);
3552 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3553 bio
->bi_bdev
= device
->bdev
;
3554 bio_set_op_attrs(bio
, REQ_OP_WRITE
, WRITE_FLUSH
);
3555 init_completion(&device
->flush_wait
);
3556 bio
->bi_private
= &device
->flush_wait
;
3557 device
->flush_bio
= bio
;
3560 btrfsic_submit_bio(bio
);
3566 * send an empty flush down to each device in parallel,
3567 * then wait for them
3569 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3571 struct list_head
*head
;
3572 struct btrfs_device
*dev
;
3573 int errors_send
= 0;
3574 int errors_wait
= 0;
3577 /* send down all the barriers */
3578 head
= &info
->fs_devices
->devices
;
3579 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3586 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3589 ret
= write_dev_flush(dev
, 0);
3594 /* wait for all the barriers */
3595 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3602 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3605 ret
= write_dev_flush(dev
, 1);
3609 if (errors_send
> info
->num_tolerated_disk_barrier_failures
||
3610 errors_wait
> info
->num_tolerated_disk_barrier_failures
)
3615 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags
)
3618 int min_tolerated
= INT_MAX
;
3620 if ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 ||
3621 (flags
& BTRFS_AVAIL_ALLOC_BIT_SINGLE
))
3622 min_tolerated
= min(min_tolerated
,
3623 btrfs_raid_array
[BTRFS_RAID_SINGLE
].
3624 tolerated_failures
);
3626 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
3627 if (raid_type
== BTRFS_RAID_SINGLE
)
3629 if (!(flags
& btrfs_raid_group
[raid_type
]))
3631 min_tolerated
= min(min_tolerated
,
3632 btrfs_raid_array
[raid_type
].
3633 tolerated_failures
);
3636 if (min_tolerated
== INT_MAX
) {
3637 pr_warn("BTRFS: unknown raid flag: %llu", flags
);
3641 return min_tolerated
;
3644 int btrfs_calc_num_tolerated_disk_barrier_failures(
3645 struct btrfs_fs_info
*fs_info
)
3647 struct btrfs_ioctl_space_info space
;
3648 struct btrfs_space_info
*sinfo
;
3649 u64 types
[] = {BTRFS_BLOCK_GROUP_DATA
,
3650 BTRFS_BLOCK_GROUP_SYSTEM
,
3651 BTRFS_BLOCK_GROUP_METADATA
,
3652 BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
};
3655 int num_tolerated_disk_barrier_failures
=
3656 (int)fs_info
->fs_devices
->num_devices
;
3658 for (i
= 0; i
< ARRAY_SIZE(types
); i
++) {
3659 struct btrfs_space_info
*tmp
;
3663 list_for_each_entry_rcu(tmp
, &fs_info
->space_info
, list
) {
3664 if (tmp
->flags
== types
[i
]) {
3674 down_read(&sinfo
->groups_sem
);
3675 for (c
= 0; c
< BTRFS_NR_RAID_TYPES
; c
++) {
3678 if (list_empty(&sinfo
->block_groups
[c
]))
3681 btrfs_get_block_group_info(&sinfo
->block_groups
[c
],
3683 if (space
.total_bytes
== 0 || space
.used_bytes
== 0)
3685 flags
= space
.flags
;
3687 num_tolerated_disk_barrier_failures
= min(
3688 num_tolerated_disk_barrier_failures
,
3689 btrfs_get_num_tolerated_disk_barrier_failures(
3692 up_read(&sinfo
->groups_sem
);
3695 return num_tolerated_disk_barrier_failures
;
3698 static int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
3700 struct list_head
*head
;
3701 struct btrfs_device
*dev
;
3702 struct btrfs_super_block
*sb
;
3703 struct btrfs_dev_item
*dev_item
;
3707 int total_errors
= 0;
3710 do_barriers
= !btrfs_test_opt(root
->fs_info
, NOBARRIER
);
3711 backup_super_roots(root
->fs_info
);
3713 sb
= root
->fs_info
->super_for_commit
;
3714 dev_item
= &sb
->dev_item
;
3716 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3717 head
= &root
->fs_info
->fs_devices
->devices
;
3718 max_errors
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
3721 ret
= barrier_all_devices(root
->fs_info
);
3724 &root
->fs_info
->fs_devices
->device_list_mutex
);
3725 btrfs_handle_fs_error(root
->fs_info
, ret
,
3726 "errors while submitting device barriers.");
3731 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3736 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3739 btrfs_set_stack_device_generation(dev_item
, 0);
3740 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3741 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3742 btrfs_set_stack_device_total_bytes(dev_item
,
3743 dev
->commit_total_bytes
);
3744 btrfs_set_stack_device_bytes_used(dev_item
,
3745 dev
->commit_bytes_used
);
3746 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3747 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3748 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3749 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3750 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
3752 flags
= btrfs_super_flags(sb
);
3753 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3755 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
3759 if (total_errors
> max_errors
) {
3760 btrfs_err(root
->fs_info
, "%d errors while writing supers",
3762 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3764 /* FUA is masked off if unsupported and can't be the reason */
3765 btrfs_handle_fs_error(root
->fs_info
, -EIO
,
3766 "%d errors while writing supers", total_errors
);
3771 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3774 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3777 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
3781 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3782 if (total_errors
> max_errors
) {
3783 btrfs_handle_fs_error(root
->fs_info
, -EIO
,
3784 "%d errors while writing supers", total_errors
);
3790 int write_ctree_super(struct btrfs_trans_handle
*trans
,
3791 struct btrfs_root
*root
, int max_mirrors
)
3793 return write_all_supers(root
, max_mirrors
);
3796 /* Drop a fs root from the radix tree and free it. */
3797 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info
*fs_info
,
3798 struct btrfs_root
*root
)
3800 spin_lock(&fs_info
->fs_roots_radix_lock
);
3801 radix_tree_delete(&fs_info
->fs_roots_radix
,
3802 (unsigned long)root
->root_key
.objectid
);
3803 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3805 if (btrfs_root_refs(&root
->root_item
) == 0)
3806 synchronize_srcu(&fs_info
->subvol_srcu
);
3808 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
3809 btrfs_free_log(NULL
, root
);
3810 if (root
->reloc_root
) {
3811 free_extent_buffer(root
->reloc_root
->node
);
3812 free_extent_buffer(root
->reloc_root
->commit_root
);
3813 btrfs_put_fs_root(root
->reloc_root
);
3814 root
->reloc_root
= NULL
;
3818 if (root
->free_ino_pinned
)
3819 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3820 if (root
->free_ino_ctl
)
3821 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3825 static void free_fs_root(struct btrfs_root
*root
)
3827 iput(root
->ino_cache_inode
);
3828 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3829 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
3830 root
->orphan_block_rsv
= NULL
;
3832 free_anon_bdev(root
->anon_dev
);
3833 if (root
->subv_writers
)
3834 btrfs_free_subvolume_writers(root
->subv_writers
);
3835 free_extent_buffer(root
->node
);
3836 free_extent_buffer(root
->commit_root
);
3837 kfree(root
->free_ino_ctl
);
3838 kfree(root
->free_ino_pinned
);
3840 btrfs_put_fs_root(root
);
3843 void btrfs_free_fs_root(struct btrfs_root
*root
)
3848 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3850 u64 root_objectid
= 0;
3851 struct btrfs_root
*gang
[8];
3854 unsigned int ret
= 0;
3858 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
3859 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3860 (void **)gang
, root_objectid
,
3863 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3866 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3868 for (i
= 0; i
< ret
; i
++) {
3869 /* Avoid to grab roots in dead_roots */
3870 if (btrfs_root_refs(&gang
[i
]->root_item
) == 0) {
3874 /* grab all the search result for later use */
3875 gang
[i
] = btrfs_grab_fs_root(gang
[i
]);
3877 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
3879 for (i
= 0; i
< ret
; i
++) {
3882 root_objectid
= gang
[i
]->root_key
.objectid
;
3883 err
= btrfs_orphan_cleanup(gang
[i
]);
3886 btrfs_put_fs_root(gang
[i
]);
3891 /* release the uncleaned roots due to error */
3892 for (; i
< ret
; i
++) {
3894 btrfs_put_fs_root(gang
[i
]);
3899 int btrfs_commit_super(struct btrfs_root
*root
)
3901 struct btrfs_trans_handle
*trans
;
3903 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3904 btrfs_run_delayed_iputs(root
);
3905 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3906 wake_up_process(root
->fs_info
->cleaner_kthread
);
3908 /* wait until ongoing cleanup work done */
3909 down_write(&root
->fs_info
->cleanup_work_sem
);
3910 up_write(&root
->fs_info
->cleanup_work_sem
);
3912 trans
= btrfs_join_transaction(root
);
3914 return PTR_ERR(trans
);
3915 return btrfs_commit_transaction(trans
, root
);
3918 void close_ctree(struct btrfs_root
*root
)
3920 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3923 set_bit(BTRFS_FS_CLOSING_START
, &fs_info
->flags
);
3925 /* wait for the qgroup rescan worker to stop */
3926 btrfs_qgroup_wait_for_completion(fs_info
, false);
3928 /* wait for the uuid_scan task to finish */
3929 down(&fs_info
->uuid_tree_rescan_sem
);
3930 /* avoid complains from lockdep et al., set sem back to initial state */
3931 up(&fs_info
->uuid_tree_rescan_sem
);
3933 /* pause restriper - we want to resume on mount */
3934 btrfs_pause_balance(fs_info
);
3936 btrfs_dev_replace_suspend_for_unmount(fs_info
);
3938 btrfs_scrub_cancel(fs_info
);
3940 /* wait for any defraggers to finish */
3941 wait_event(fs_info
->transaction_wait
,
3942 (atomic_read(&fs_info
->defrag_running
) == 0));
3944 /* clear out the rbtree of defraggable inodes */
3945 btrfs_cleanup_defrag_inodes(fs_info
);
3947 cancel_work_sync(&fs_info
->async_reclaim_work
);
3949 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
3951 * If the cleaner thread is stopped and there are
3952 * block groups queued for removal, the deletion will be
3953 * skipped when we quit the cleaner thread.
3955 btrfs_delete_unused_bgs(root
->fs_info
);
3957 ret
= btrfs_commit_super(root
);
3959 btrfs_err(fs_info
, "commit super ret %d", ret
);
3962 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3963 btrfs_error_commit_super(root
);
3965 kthread_stop(fs_info
->transaction_kthread
);
3966 kthread_stop(fs_info
->cleaner_kthread
);
3968 set_bit(BTRFS_FS_CLOSING_DONE
, &fs_info
->flags
);
3970 btrfs_free_qgroup_config(fs_info
);
3972 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
3973 btrfs_info(fs_info
, "at unmount delalloc count %lld",
3974 percpu_counter_sum(&fs_info
->delalloc_bytes
));
3977 btrfs_sysfs_remove_mounted(fs_info
);
3978 btrfs_sysfs_remove_fsid(fs_info
->fs_devices
);
3980 btrfs_free_fs_roots(fs_info
);
3982 btrfs_put_block_group_cache(fs_info
);
3984 btrfs_free_block_groups(fs_info
);
3987 * we must make sure there is not any read request to
3988 * submit after we stopping all workers.
3990 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
3991 btrfs_stop_all_workers(fs_info
);
3993 clear_bit(BTRFS_FS_OPEN
, &fs_info
->flags
);
3994 free_root_pointers(fs_info
, 1);
3996 iput(fs_info
->btree_inode
);
3998 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3999 if (btrfs_test_opt(root
->fs_info
, CHECK_INTEGRITY
))
4000 btrfsic_unmount(root
, fs_info
->fs_devices
);
4003 btrfs_close_devices(fs_info
->fs_devices
);
4004 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
4006 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
4007 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
4008 percpu_counter_destroy(&fs_info
->bio_counter
);
4009 bdi_destroy(&fs_info
->bdi
);
4010 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
4012 btrfs_free_stripe_hash_table(fs_info
);
4014 __btrfs_free_block_rsv(root
->orphan_block_rsv
);
4015 root
->orphan_block_rsv
= NULL
;
4018 while (!list_empty(&fs_info
->pinned_chunks
)) {
4019 struct extent_map
*em
;
4021 em
= list_first_entry(&fs_info
->pinned_chunks
,
4022 struct extent_map
, list
);
4023 list_del_init(&em
->list
);
4024 free_extent_map(em
);
4026 unlock_chunks(root
);
4029 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
4033 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
4035 ret
= extent_buffer_uptodate(buf
);
4039 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
4040 parent_transid
, atomic
);
4046 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
4048 struct btrfs_root
*root
;
4049 u64 transid
= btrfs_header_generation(buf
);
4052 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4054 * This is a fast path so only do this check if we have sanity tests
4055 * enabled. Normal people shouldn't be marking dummy buffers as dirty
4056 * outside of the sanity tests.
4058 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY
, &buf
->bflags
)))
4061 root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
4062 btrfs_assert_tree_locked(buf
);
4063 if (transid
!= root
->fs_info
->generation
)
4064 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4065 buf
->start
, transid
, root
->fs_info
->generation
);
4066 was_dirty
= set_extent_buffer_dirty(buf
);
4068 __percpu_counter_add(&root
->fs_info
->dirty_metadata_bytes
,
4070 root
->fs_info
->dirty_metadata_batch
);
4071 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4072 if (btrfs_header_level(buf
) == 0 && check_leaf(root
, buf
)) {
4073 btrfs_print_leaf(root
, buf
);
4079 static void __btrfs_btree_balance_dirty(struct btrfs_root
*root
,
4083 * looks as though older kernels can get into trouble with
4084 * this code, they end up stuck in balance_dirty_pages forever
4088 if (current
->flags
& PF_MEMALLOC
)
4092 btrfs_balance_delayed_items(root
);
4094 ret
= percpu_counter_compare(&root
->fs_info
->dirty_metadata_bytes
,
4095 BTRFS_DIRTY_METADATA_THRESH
);
4097 balance_dirty_pages_ratelimited(
4098 root
->fs_info
->btree_inode
->i_mapping
);
4102 void btrfs_btree_balance_dirty(struct btrfs_root
*root
)
4104 __btrfs_btree_balance_dirty(root
, 1);
4107 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root
*root
)
4109 __btrfs_btree_balance_dirty(root
, 0);
4112 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
4114 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
4115 return btree_read_extent_buffer_pages(root
, buf
, parent_transid
);
4118 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
4121 struct btrfs_super_block
*sb
= fs_info
->super_copy
;
4122 u64 nodesize
= btrfs_super_nodesize(sb
);
4123 u64 sectorsize
= btrfs_super_sectorsize(sb
);
4126 if (btrfs_super_magic(sb
) != BTRFS_MAGIC
) {
4127 btrfs_err(fs_info
, "no valid FS found");
4130 if (btrfs_super_flags(sb
) & ~BTRFS_SUPER_FLAG_SUPP
)
4131 btrfs_warn(fs_info
, "unrecognized super flag: %llu",
4132 btrfs_super_flags(sb
) & ~BTRFS_SUPER_FLAG_SUPP
);
4133 if (btrfs_super_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
4134 btrfs_err(fs_info
, "tree_root level too big: %d >= %d",
4135 btrfs_super_root_level(sb
), BTRFS_MAX_LEVEL
);
4138 if (btrfs_super_chunk_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
4139 btrfs_err(fs_info
, "chunk_root level too big: %d >= %d",
4140 btrfs_super_chunk_root_level(sb
), BTRFS_MAX_LEVEL
);
4143 if (btrfs_super_log_root_level(sb
) >= BTRFS_MAX_LEVEL
) {
4144 btrfs_err(fs_info
, "log_root level too big: %d >= %d",
4145 btrfs_super_log_root_level(sb
), BTRFS_MAX_LEVEL
);
4150 * Check sectorsize and nodesize first, other check will need it.
4151 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4153 if (!is_power_of_2(sectorsize
) || sectorsize
< 4096 ||
4154 sectorsize
> BTRFS_MAX_METADATA_BLOCKSIZE
) {
4155 btrfs_err(fs_info
, "invalid sectorsize %llu", sectorsize
);
4158 /* Only PAGE SIZE is supported yet */
4159 if (sectorsize
!= PAGE_SIZE
) {
4161 "sectorsize %llu not supported yet, only support %lu",
4162 sectorsize
, PAGE_SIZE
);
4165 if (!is_power_of_2(nodesize
) || nodesize
< sectorsize
||
4166 nodesize
> BTRFS_MAX_METADATA_BLOCKSIZE
) {
4167 btrfs_err(fs_info
, "invalid nodesize %llu", nodesize
);
4170 if (nodesize
!= le32_to_cpu(sb
->__unused_leafsize
)) {
4171 btrfs_err(fs_info
, "invalid leafsize %u, should be %llu",
4172 le32_to_cpu(sb
->__unused_leafsize
), nodesize
);
4176 /* Root alignment check */
4177 if (!IS_ALIGNED(btrfs_super_root(sb
), sectorsize
)) {
4178 btrfs_warn(fs_info
, "tree_root block unaligned: %llu",
4179 btrfs_super_root(sb
));
4182 if (!IS_ALIGNED(btrfs_super_chunk_root(sb
), sectorsize
)) {
4183 btrfs_warn(fs_info
, "chunk_root block unaligned: %llu",
4184 btrfs_super_chunk_root(sb
));
4187 if (!IS_ALIGNED(btrfs_super_log_root(sb
), sectorsize
)) {
4188 btrfs_warn(fs_info
, "log_root block unaligned: %llu",
4189 btrfs_super_log_root(sb
));
4193 if (memcmp(fs_info
->fsid
, sb
->dev_item
.fsid
, BTRFS_UUID_SIZE
) != 0) {
4195 "dev_item UUID does not match fsid: %pU != %pU",
4196 fs_info
->fsid
, sb
->dev_item
.fsid
);
4201 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4204 if (btrfs_super_bytes_used(sb
) < 6 * btrfs_super_nodesize(sb
)) {
4205 btrfs_err(fs_info
, "bytes_used is too small %llu",
4206 btrfs_super_bytes_used(sb
));
4209 if (!is_power_of_2(btrfs_super_stripesize(sb
))) {
4210 btrfs_err(fs_info
, "invalid stripesize %u",
4211 btrfs_super_stripesize(sb
));
4214 if (btrfs_super_num_devices(sb
) > (1UL << 31))
4215 btrfs_warn(fs_info
, "suspicious number of devices: %llu",
4216 btrfs_super_num_devices(sb
));
4217 if (btrfs_super_num_devices(sb
) == 0) {
4218 btrfs_err(fs_info
, "number of devices is 0");
4222 if (btrfs_super_bytenr(sb
) != BTRFS_SUPER_INFO_OFFSET
) {
4223 btrfs_err(fs_info
, "super offset mismatch %llu != %u",
4224 btrfs_super_bytenr(sb
), BTRFS_SUPER_INFO_OFFSET
);
4229 * Obvious sys_chunk_array corruptions, it must hold at least one key
4232 if (btrfs_super_sys_array_size(sb
) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4233 btrfs_err(fs_info
, "system chunk array too big %u > %u",
4234 btrfs_super_sys_array_size(sb
),
4235 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
);
4238 if (btrfs_super_sys_array_size(sb
) < sizeof(struct btrfs_disk_key
)
4239 + sizeof(struct btrfs_chunk
)) {
4240 btrfs_err(fs_info
, "system chunk array too small %u < %zu",
4241 btrfs_super_sys_array_size(sb
),
4242 sizeof(struct btrfs_disk_key
)
4243 + sizeof(struct btrfs_chunk
));
4248 * The generation is a global counter, we'll trust it more than the others
4249 * but it's still possible that it's the one that's wrong.
4251 if (btrfs_super_generation(sb
) < btrfs_super_chunk_root_generation(sb
))
4253 "suspicious: generation < chunk_root_generation: %llu < %llu",
4254 btrfs_super_generation(sb
),
4255 btrfs_super_chunk_root_generation(sb
));
4256 if (btrfs_super_generation(sb
) < btrfs_super_cache_generation(sb
)
4257 && btrfs_super_cache_generation(sb
) != (u64
)-1)
4259 "suspicious: generation < cache_generation: %llu < %llu",
4260 btrfs_super_generation(sb
),
4261 btrfs_super_cache_generation(sb
));
4266 static void btrfs_error_commit_super(struct btrfs_root
*root
)
4268 mutex_lock(&root
->fs_info
->cleaner_mutex
);
4269 btrfs_run_delayed_iputs(root
);
4270 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
4272 down_write(&root
->fs_info
->cleanup_work_sem
);
4273 up_write(&root
->fs_info
->cleanup_work_sem
);
4275 /* cleanup FS via transaction */
4276 btrfs_cleanup_transaction(root
);
4279 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
4281 struct btrfs_ordered_extent
*ordered
;
4283 spin_lock(&root
->ordered_extent_lock
);
4285 * This will just short circuit the ordered completion stuff which will
4286 * make sure the ordered extent gets properly cleaned up.
4288 list_for_each_entry(ordered
, &root
->ordered_extents
,
4290 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
4291 spin_unlock(&root
->ordered_extent_lock
);
4294 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info
*fs_info
)
4296 struct btrfs_root
*root
;
4297 struct list_head splice
;
4299 INIT_LIST_HEAD(&splice
);
4301 spin_lock(&fs_info
->ordered_root_lock
);
4302 list_splice_init(&fs_info
->ordered_roots
, &splice
);
4303 while (!list_empty(&splice
)) {
4304 root
= list_first_entry(&splice
, struct btrfs_root
,
4306 list_move_tail(&root
->ordered_root
,
4307 &fs_info
->ordered_roots
);
4309 spin_unlock(&fs_info
->ordered_root_lock
);
4310 btrfs_destroy_ordered_extents(root
);
4313 spin_lock(&fs_info
->ordered_root_lock
);
4315 spin_unlock(&fs_info
->ordered_root_lock
);
4318 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
4319 struct btrfs_root
*root
)
4321 struct rb_node
*node
;
4322 struct btrfs_delayed_ref_root
*delayed_refs
;
4323 struct btrfs_delayed_ref_node
*ref
;
4326 delayed_refs
= &trans
->delayed_refs
;
4328 spin_lock(&delayed_refs
->lock
);
4329 if (atomic_read(&delayed_refs
->num_entries
) == 0) {
4330 spin_unlock(&delayed_refs
->lock
);
4331 btrfs_info(root
->fs_info
, "delayed_refs has NO entry");
4335 while ((node
= rb_first(&delayed_refs
->href_root
)) != NULL
) {
4336 struct btrfs_delayed_ref_head
*head
;
4337 struct btrfs_delayed_ref_node
*tmp
;
4338 bool pin_bytes
= false;
4340 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
4342 if (!mutex_trylock(&head
->mutex
)) {
4343 atomic_inc(&head
->node
.refs
);
4344 spin_unlock(&delayed_refs
->lock
);
4346 mutex_lock(&head
->mutex
);
4347 mutex_unlock(&head
->mutex
);
4348 btrfs_put_delayed_ref(&head
->node
);
4349 spin_lock(&delayed_refs
->lock
);
4352 spin_lock(&head
->lock
);
4353 list_for_each_entry_safe_reverse(ref
, tmp
, &head
->ref_list
,
4356 list_del(&ref
->list
);
4357 atomic_dec(&delayed_refs
->num_entries
);
4358 btrfs_put_delayed_ref(ref
);
4360 if (head
->must_insert_reserved
)
4362 btrfs_free_delayed_extent_op(head
->extent_op
);
4363 delayed_refs
->num_heads
--;
4364 if (head
->processing
== 0)
4365 delayed_refs
->num_heads_ready
--;
4366 atomic_dec(&delayed_refs
->num_entries
);
4367 head
->node
.in_tree
= 0;
4368 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
4369 spin_unlock(&head
->lock
);
4370 spin_unlock(&delayed_refs
->lock
);
4371 mutex_unlock(&head
->mutex
);
4374 btrfs_pin_extent(root
, head
->node
.bytenr
,
4375 head
->node
.num_bytes
, 1);
4376 btrfs_put_delayed_ref(&head
->node
);
4378 spin_lock(&delayed_refs
->lock
);
4381 spin_unlock(&delayed_refs
->lock
);
4386 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
4388 struct btrfs_inode
*btrfs_inode
;
4389 struct list_head splice
;
4391 INIT_LIST_HEAD(&splice
);
4393 spin_lock(&root
->delalloc_lock
);
4394 list_splice_init(&root
->delalloc_inodes
, &splice
);
4396 while (!list_empty(&splice
)) {
4397 btrfs_inode
= list_first_entry(&splice
, struct btrfs_inode
,
4400 list_del_init(&btrfs_inode
->delalloc_inodes
);
4401 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
4402 &btrfs_inode
->runtime_flags
);
4403 spin_unlock(&root
->delalloc_lock
);
4405 btrfs_invalidate_inodes(btrfs_inode
->root
);
4407 spin_lock(&root
->delalloc_lock
);
4410 spin_unlock(&root
->delalloc_lock
);
4413 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info
*fs_info
)
4415 struct btrfs_root
*root
;
4416 struct list_head splice
;
4418 INIT_LIST_HEAD(&splice
);
4420 spin_lock(&fs_info
->delalloc_root_lock
);
4421 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
4422 while (!list_empty(&splice
)) {
4423 root
= list_first_entry(&splice
, struct btrfs_root
,
4425 list_del_init(&root
->delalloc_root
);
4426 root
= btrfs_grab_fs_root(root
);
4428 spin_unlock(&fs_info
->delalloc_root_lock
);
4430 btrfs_destroy_delalloc_inodes(root
);
4431 btrfs_put_fs_root(root
);
4433 spin_lock(&fs_info
->delalloc_root_lock
);
4435 spin_unlock(&fs_info
->delalloc_root_lock
);
4438 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
4439 struct extent_io_tree
*dirty_pages
,
4443 struct extent_buffer
*eb
;
4448 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
4453 clear_extent_bits(dirty_pages
, start
, end
, mark
);
4454 while (start
<= end
) {
4455 eb
= btrfs_find_tree_block(root
->fs_info
, start
);
4456 start
+= root
->nodesize
;
4459 wait_on_extent_buffer_writeback(eb
);
4461 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
4463 clear_extent_buffer_dirty(eb
);
4464 free_extent_buffer_stale(eb
);
4471 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
4472 struct extent_io_tree
*pinned_extents
)
4474 struct extent_io_tree
*unpin
;
4480 unpin
= pinned_extents
;
4483 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4484 EXTENT_DIRTY
, NULL
);
4488 clear_extent_dirty(unpin
, start
, end
);
4489 btrfs_error_unpin_extent_range(root
, start
, end
);
4494 if (unpin
== &root
->fs_info
->freed_extents
[0])
4495 unpin
= &root
->fs_info
->freed_extents
[1];
4497 unpin
= &root
->fs_info
->freed_extents
[0];
4505 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache
*cache
)
4507 struct inode
*inode
;
4509 inode
= cache
->io_ctl
.inode
;
4511 invalidate_inode_pages2(inode
->i_mapping
);
4512 BTRFS_I(inode
)->generation
= 0;
4513 cache
->io_ctl
.inode
= NULL
;
4516 btrfs_put_block_group(cache
);
4519 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction
*cur_trans
,
4520 struct btrfs_root
*root
)
4522 struct btrfs_block_group_cache
*cache
;
4524 spin_lock(&cur_trans
->dirty_bgs_lock
);
4525 while (!list_empty(&cur_trans
->dirty_bgs
)) {
4526 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
4527 struct btrfs_block_group_cache
,
4530 btrfs_err(root
->fs_info
,
4531 "orphan block group dirty_bgs list");
4532 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4536 if (!list_empty(&cache
->io_list
)) {
4537 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4538 list_del_init(&cache
->io_list
);
4539 btrfs_cleanup_bg_io(cache
);
4540 spin_lock(&cur_trans
->dirty_bgs_lock
);
4543 list_del_init(&cache
->dirty_list
);
4544 spin_lock(&cache
->lock
);
4545 cache
->disk_cache_state
= BTRFS_DC_ERROR
;
4546 spin_unlock(&cache
->lock
);
4548 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4549 btrfs_put_block_group(cache
);
4550 spin_lock(&cur_trans
->dirty_bgs_lock
);
4552 spin_unlock(&cur_trans
->dirty_bgs_lock
);
4554 while (!list_empty(&cur_trans
->io_bgs
)) {
4555 cache
= list_first_entry(&cur_trans
->io_bgs
,
4556 struct btrfs_block_group_cache
,
4559 btrfs_err(root
->fs_info
,
4560 "orphan block group on io_bgs list");
4564 list_del_init(&cache
->io_list
);
4565 spin_lock(&cache
->lock
);
4566 cache
->disk_cache_state
= BTRFS_DC_ERROR
;
4567 spin_unlock(&cache
->lock
);
4568 btrfs_cleanup_bg_io(cache
);
4572 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
4573 struct btrfs_root
*root
)
4575 btrfs_cleanup_dirty_bgs(cur_trans
, root
);
4576 ASSERT(list_empty(&cur_trans
->dirty_bgs
));
4577 ASSERT(list_empty(&cur_trans
->io_bgs
));
4579 btrfs_destroy_delayed_refs(cur_trans
, root
);
4581 cur_trans
->state
= TRANS_STATE_COMMIT_START
;
4582 wake_up(&root
->fs_info
->transaction_blocked_wait
);
4584 cur_trans
->state
= TRANS_STATE_UNBLOCKED
;
4585 wake_up(&root
->fs_info
->transaction_wait
);
4587 btrfs_destroy_delayed_inodes(root
);
4588 btrfs_assert_delayed_root_empty(root
);
4590 btrfs_destroy_marked_extents(root
, &cur_trans
->dirty_pages
,
4592 btrfs_destroy_pinned_extent(root
,
4593 root
->fs_info
->pinned_extents
);
4595 cur_trans
->state
=TRANS_STATE_COMPLETED
;
4596 wake_up(&cur_trans
->commit_wait
);
4599 memset(cur_trans, 0, sizeof(*cur_trans));
4600 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4604 static int btrfs_cleanup_transaction(struct btrfs_root
*root
)
4606 struct btrfs_transaction
*t
;
4608 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
4610 spin_lock(&root
->fs_info
->trans_lock
);
4611 while (!list_empty(&root
->fs_info
->trans_list
)) {
4612 t
= list_first_entry(&root
->fs_info
->trans_list
,
4613 struct btrfs_transaction
, list
);
4614 if (t
->state
>= TRANS_STATE_COMMIT_START
) {
4615 atomic_inc(&t
->use_count
);
4616 spin_unlock(&root
->fs_info
->trans_lock
);
4617 btrfs_wait_for_commit(root
, t
->transid
);
4618 btrfs_put_transaction(t
);
4619 spin_lock(&root
->fs_info
->trans_lock
);
4622 if (t
== root
->fs_info
->running_transaction
) {
4623 t
->state
= TRANS_STATE_COMMIT_DOING
;
4624 spin_unlock(&root
->fs_info
->trans_lock
);
4626 * We wait for 0 num_writers since we don't hold a trans
4627 * handle open currently for this transaction.
4629 wait_event(t
->writer_wait
,
4630 atomic_read(&t
->num_writers
) == 0);
4632 spin_unlock(&root
->fs_info
->trans_lock
);
4634 btrfs_cleanup_one_transaction(t
, root
);
4636 spin_lock(&root
->fs_info
->trans_lock
);
4637 if (t
== root
->fs_info
->running_transaction
)
4638 root
->fs_info
->running_transaction
= NULL
;
4639 list_del_init(&t
->list
);
4640 spin_unlock(&root
->fs_info
->trans_lock
);
4642 btrfs_put_transaction(t
);
4643 trace_btrfs_transaction_commit(root
);
4644 spin_lock(&root
->fs_info
->trans_lock
);
4646 spin_unlock(&root
->fs_info
->trans_lock
);
4647 btrfs_destroy_all_ordered_extents(root
->fs_info
);
4648 btrfs_destroy_delayed_inodes(root
);
4649 btrfs_assert_delayed_root_empty(root
);
4650 btrfs_destroy_pinned_extent(root
, root
->fs_info
->pinned_extents
);
4651 btrfs_destroy_all_delalloc_inodes(root
->fs_info
);
4652 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
4657 static const struct extent_io_ops btree_extent_io_ops
= {
4658 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
4659 .readpage_io_failed_hook
= btree_io_failed_hook
,
4660 .submit_bio_hook
= btree_submit_bio_hook
,
4661 /* note we're sharing with inode.c for the merge bio hook */
4662 .merge_bio_hook
= btrfs_merge_bio_hook
,