2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <linux/ratelimit.h>
33 #include <asm/unaligned.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "print-tree.h"
41 #include "async-thread.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
52 #include <asm/cpufeature.h>
55 static struct extent_io_ops btree_extent_io_ops
;
56 static void end_workqueue_fn(struct btrfs_work
*work
);
57 static void free_fs_root(struct btrfs_root
*root
);
58 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
60 static void btrfs_destroy_ordered_operations(struct btrfs_transaction
*t
,
61 struct btrfs_root
*root
);
62 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
);
63 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
64 struct btrfs_root
*root
);
65 static void btrfs_evict_pending_snapshots(struct btrfs_transaction
*t
);
66 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
);
67 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
68 struct extent_io_tree
*dirty_pages
,
70 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
71 struct extent_io_tree
*pinned_extents
);
74 * end_io_wq structs are used to do processing in task context when an IO is
75 * complete. This is used during reads to verify checksums, and it is used
76 * by writes to insert metadata for new file extents after IO is complete.
82 struct btrfs_fs_info
*info
;
85 struct list_head list
;
86 struct btrfs_work work
;
90 * async submit bios are used to offload expensive checksumming
91 * onto the worker threads. They checksum file and metadata bios
92 * just before they are sent down the IO stack.
94 struct async_submit_bio
{
97 struct list_head list
;
98 extent_submit_bio_hook_t
*submit_bio_start
;
99 extent_submit_bio_hook_t
*submit_bio_done
;
102 unsigned long bio_flags
;
104 * bio_offset is optional, can be used if the pages in the bio
105 * can't tell us where in the file the bio should go
108 struct btrfs_work work
;
113 * Lockdep class keys for extent_buffer->lock's in this root. For a given
114 * eb, the lockdep key is determined by the btrfs_root it belongs to and
115 * the level the eb occupies in the tree.
117 * Different roots are used for different purposes and may nest inside each
118 * other and they require separate keysets. As lockdep keys should be
119 * static, assign keysets according to the purpose of the root as indicated
120 * by btrfs_root->objectid. This ensures that all special purpose roots
121 * have separate keysets.
123 * Lock-nesting across peer nodes is always done with the immediate parent
124 * node locked thus preventing deadlock. As lockdep doesn't know this, use
125 * subclass to avoid triggering lockdep warning in such cases.
127 * The key is set by the readpage_end_io_hook after the buffer has passed
128 * csum validation but before the pages are unlocked. It is also set by
129 * btrfs_init_new_buffer on freshly allocated blocks.
131 * We also add a check to make sure the highest level of the tree is the
132 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
133 * needs update as well.
135 #ifdef CONFIG_DEBUG_LOCK_ALLOC
136 # if BTRFS_MAX_LEVEL != 8
140 static struct btrfs_lockdep_keyset
{
141 u64 id
; /* root objectid */
142 const char *name_stem
; /* lock name stem */
143 char names
[BTRFS_MAX_LEVEL
+ 1][20];
144 struct lock_class_key keys
[BTRFS_MAX_LEVEL
+ 1];
145 } btrfs_lockdep_keysets
[] = {
146 { .id
= BTRFS_ROOT_TREE_OBJECTID
, .name_stem
= "root" },
147 { .id
= BTRFS_EXTENT_TREE_OBJECTID
, .name_stem
= "extent" },
148 { .id
= BTRFS_CHUNK_TREE_OBJECTID
, .name_stem
= "chunk" },
149 { .id
= BTRFS_DEV_TREE_OBJECTID
, .name_stem
= "dev" },
150 { .id
= BTRFS_FS_TREE_OBJECTID
, .name_stem
= "fs" },
151 { .id
= BTRFS_CSUM_TREE_OBJECTID
, .name_stem
= "csum" },
152 { .id
= BTRFS_ORPHAN_OBJECTID
, .name_stem
= "orphan" },
153 { .id
= BTRFS_TREE_LOG_OBJECTID
, .name_stem
= "log" },
154 { .id
= BTRFS_TREE_RELOC_OBJECTID
, .name_stem
= "treloc" },
155 { .id
= BTRFS_DATA_RELOC_TREE_OBJECTID
, .name_stem
= "dreloc" },
156 { .id
= 0, .name_stem
= "tree" },
159 void __init
btrfs_init_lockdep(void)
163 /* initialize lockdep class names */
164 for (i
= 0; i
< ARRAY_SIZE(btrfs_lockdep_keysets
); i
++) {
165 struct btrfs_lockdep_keyset
*ks
= &btrfs_lockdep_keysets
[i
];
167 for (j
= 0; j
< ARRAY_SIZE(ks
->names
); j
++)
168 snprintf(ks
->names
[j
], sizeof(ks
->names
[j
]),
169 "btrfs-%s-%02d", ks
->name_stem
, j
);
173 void btrfs_set_buffer_lockdep_class(u64 objectid
, struct extent_buffer
*eb
,
176 struct btrfs_lockdep_keyset
*ks
;
178 BUG_ON(level
>= ARRAY_SIZE(ks
->keys
));
180 /* find the matching keyset, id 0 is the default entry */
181 for (ks
= btrfs_lockdep_keysets
; ks
->id
; ks
++)
182 if (ks
->id
== objectid
)
185 lockdep_set_class_and_name(&eb
->lock
,
186 &ks
->keys
[level
], ks
->names
[level
]);
192 * extents on the btree inode are pretty simple, there's one extent
193 * that covers the entire device
195 static struct extent_map
*btree_get_extent(struct inode
*inode
,
196 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
199 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
200 struct extent_map
*em
;
203 read_lock(&em_tree
->lock
);
204 em
= lookup_extent_mapping(em_tree
, start
, len
);
207 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
208 read_unlock(&em_tree
->lock
);
211 read_unlock(&em_tree
->lock
);
213 em
= alloc_extent_map();
215 em
= ERR_PTR(-ENOMEM
);
220 em
->block_len
= (u64
)-1;
222 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
224 write_lock(&em_tree
->lock
);
225 ret
= add_extent_mapping(em_tree
, em
);
226 if (ret
== -EEXIST
) {
228 em
= lookup_extent_mapping(em_tree
, start
, len
);
235 write_unlock(&em_tree
->lock
);
241 u32
btrfs_csum_data(struct btrfs_root
*root
, char *data
, u32 seed
, size_t len
)
243 return crc32c(seed
, data
, len
);
246 void btrfs_csum_final(u32 crc
, char *result
)
248 put_unaligned_le32(~crc
, result
);
252 * compute the csum for a btree block, and either verify it or write it
253 * into the csum field of the block.
255 static int csum_tree_block(struct btrfs_root
*root
, struct extent_buffer
*buf
,
258 u16 csum_size
= btrfs_super_csum_size(root
->fs_info
->super_copy
);
261 unsigned long cur_len
;
262 unsigned long offset
= BTRFS_CSUM_SIZE
;
264 unsigned long map_start
;
265 unsigned long map_len
;
268 unsigned long inline_result
;
270 len
= buf
->len
- offset
;
272 err
= map_private_extent_buffer(buf
, offset
, 32,
273 &kaddr
, &map_start
, &map_len
);
276 cur_len
= min(len
, map_len
- (offset
- map_start
));
277 crc
= btrfs_csum_data(root
, kaddr
+ offset
- map_start
,
282 if (csum_size
> sizeof(inline_result
)) {
283 result
= kzalloc(csum_size
* sizeof(char), GFP_NOFS
);
287 result
= (char *)&inline_result
;
290 btrfs_csum_final(crc
, result
);
293 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
296 memcpy(&found
, result
, csum_size
);
298 read_extent_buffer(buf
, &val
, 0, csum_size
);
299 printk_ratelimited(KERN_INFO
"btrfs: %s checksum verify "
300 "failed on %llu wanted %X found %X "
302 root
->fs_info
->sb
->s_id
,
303 (unsigned long long)buf
->start
, val
, found
,
304 btrfs_header_level(buf
));
305 if (result
!= (char *)&inline_result
)
310 write_extent_buffer(buf
, result
, 0, csum_size
);
312 if (result
!= (char *)&inline_result
)
318 * we can't consider a given block up to date unless the transid of the
319 * block matches the transid in the parent node's pointer. This is how we
320 * detect blocks that either didn't get written at all or got written
321 * in the wrong place.
323 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
324 struct extent_buffer
*eb
, u64 parent_transid
,
327 struct extent_state
*cached_state
= NULL
;
330 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
336 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
338 if (extent_buffer_uptodate(eb
) &&
339 btrfs_header_generation(eb
) == parent_transid
) {
343 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
345 (unsigned long long)eb
->start
,
346 (unsigned long long)parent_transid
,
347 (unsigned long long)btrfs_header_generation(eb
));
349 clear_extent_buffer_uptodate(eb
);
351 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
352 &cached_state
, GFP_NOFS
);
357 * helper to read a given tree block, doing retries as required when
358 * the checksums don't match and we have alternate mirrors to try.
360 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
361 struct extent_buffer
*eb
,
362 u64 start
, u64 parent_transid
)
364 struct extent_io_tree
*io_tree
;
369 int failed_mirror
= 0;
371 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
372 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
374 ret
= read_extent_buffer_pages(io_tree
, eb
, start
,
376 btree_get_extent
, mirror_num
);
378 if (!verify_parent_transid(io_tree
, eb
,
386 * This buffer's crc is fine, but its contents are corrupted, so
387 * there is no reason to read the other copies, they won't be
390 if (test_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
))
393 num_copies
= btrfs_num_copies(root
->fs_info
,
398 if (!failed_mirror
) {
400 failed_mirror
= eb
->read_mirror
;
404 if (mirror_num
== failed_mirror
)
407 if (mirror_num
> num_copies
)
411 if (failed
&& !ret
&& failed_mirror
)
412 repair_eb_io_failure(root
, eb
, failed_mirror
);
418 * checksum a dirty tree block before IO. This has extra checks to make sure
419 * we only fill in the checksum field in the first page of a multi-page block
422 static int csum_dirty_buffer(struct btrfs_root
*root
, struct page
*page
)
424 struct extent_io_tree
*tree
;
425 u64 start
= page_offset(page
);
427 struct extent_buffer
*eb
;
429 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
431 eb
= (struct extent_buffer
*)page
->private;
432 if (page
!= eb
->pages
[0])
434 found_start
= btrfs_header_bytenr(eb
);
435 if (found_start
!= start
) {
439 if (!PageUptodate(page
)) {
443 csum_tree_block(root
, eb
, 0);
447 static int check_tree_block_fsid(struct btrfs_root
*root
,
448 struct extent_buffer
*eb
)
450 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
451 u8 fsid
[BTRFS_UUID_SIZE
];
454 read_extent_buffer(eb
, fsid
, (unsigned long)btrfs_header_fsid(eb
),
457 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
461 fs_devices
= fs_devices
->seed
;
466 #define CORRUPT(reason, eb, root, slot) \
467 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
468 "root=%llu, slot=%d\n", reason, \
469 (unsigned long long)btrfs_header_bytenr(eb), \
470 (unsigned long long)root->objectid, slot)
472 static noinline
int check_leaf(struct btrfs_root
*root
,
473 struct extent_buffer
*leaf
)
475 struct btrfs_key key
;
476 struct btrfs_key leaf_key
;
477 u32 nritems
= btrfs_header_nritems(leaf
);
483 /* Check the 0 item */
484 if (btrfs_item_offset_nr(leaf
, 0) + btrfs_item_size_nr(leaf
, 0) !=
485 BTRFS_LEAF_DATA_SIZE(root
)) {
486 CORRUPT("invalid item offset size pair", leaf
, root
, 0);
491 * Check to make sure each items keys are in the correct order and their
492 * offsets make sense. We only have to loop through nritems-1 because
493 * we check the current slot against the next slot, which verifies the
494 * next slot's offset+size makes sense and that the current's slot
497 for (slot
= 0; slot
< nritems
- 1; slot
++) {
498 btrfs_item_key_to_cpu(leaf
, &leaf_key
, slot
);
499 btrfs_item_key_to_cpu(leaf
, &key
, slot
+ 1);
501 /* Make sure the keys are in the right order */
502 if (btrfs_comp_cpu_keys(&leaf_key
, &key
) >= 0) {
503 CORRUPT("bad key order", leaf
, root
, slot
);
508 * Make sure the offset and ends are right, remember that the
509 * item data starts at the end of the leaf and grows towards the
512 if (btrfs_item_offset_nr(leaf
, slot
) !=
513 btrfs_item_end_nr(leaf
, slot
+ 1)) {
514 CORRUPT("slot offset bad", leaf
, root
, slot
);
519 * Check to make sure that we don't point outside of the leaf,
520 * just incase all the items are consistent to eachother, but
521 * all point outside of the leaf.
523 if (btrfs_item_end_nr(leaf
, slot
) >
524 BTRFS_LEAF_DATA_SIZE(root
)) {
525 CORRUPT("slot end outside of leaf", leaf
, root
, slot
);
533 struct extent_buffer
*find_eb_for_page(struct extent_io_tree
*tree
,
534 struct page
*page
, int max_walk
)
536 struct extent_buffer
*eb
;
537 u64 start
= page_offset(page
);
541 if (start
< max_walk
)
544 min_start
= start
- max_walk
;
546 while (start
>= min_start
) {
547 eb
= find_extent_buffer(tree
, start
, 0);
550 * we found an extent buffer and it contains our page
553 if (eb
->start
<= target
&&
554 eb
->start
+ eb
->len
> target
)
557 /* we found an extent buffer that wasn't for us */
558 free_extent_buffer(eb
);
563 start
-= PAGE_CACHE_SIZE
;
568 static int btree_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
569 struct extent_state
*state
, int mirror
)
571 struct extent_io_tree
*tree
;
574 struct extent_buffer
*eb
;
575 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
582 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
583 eb
= (struct extent_buffer
*)page
->private;
585 /* the pending IO might have been the only thing that kept this buffer
586 * in memory. Make sure we have a ref for all this other checks
588 extent_buffer_get(eb
);
590 reads_done
= atomic_dec_and_test(&eb
->io_pages
);
594 eb
->read_mirror
= mirror
;
595 if (test_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
)) {
600 found_start
= btrfs_header_bytenr(eb
);
601 if (found_start
!= eb
->start
) {
602 printk_ratelimited(KERN_INFO
"btrfs bad tree block start "
604 (unsigned long long)found_start
,
605 (unsigned long long)eb
->start
);
609 if (check_tree_block_fsid(root
, eb
)) {
610 printk_ratelimited(KERN_INFO
"btrfs bad fsid on block %llu\n",
611 (unsigned long long)eb
->start
);
615 found_level
= btrfs_header_level(eb
);
617 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
620 ret
= csum_tree_block(root
, eb
, 1);
627 * If this is a leaf block and it is corrupt, set the corrupt bit so
628 * that we don't try and read the other copies of this block, just
631 if (found_level
== 0 && check_leaf(root
, eb
)) {
632 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
637 set_extent_buffer_uptodate(eb
);
639 if (test_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
)) {
640 clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
);
641 btree_readahead_hook(root
, eb
, eb
->start
, ret
);
646 * our io error hook is going to dec the io pages
647 * again, we have to make sure it has something
650 atomic_inc(&eb
->io_pages
);
651 clear_extent_buffer_uptodate(eb
);
653 free_extent_buffer(eb
);
658 static int btree_io_failed_hook(struct page
*page
, int failed_mirror
)
660 struct extent_buffer
*eb
;
661 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
663 eb
= (struct extent_buffer
*)page
->private;
664 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
665 eb
->read_mirror
= failed_mirror
;
666 atomic_dec(&eb
->io_pages
);
667 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
668 btree_readahead_hook(root
, eb
, eb
->start
, -EIO
);
669 return -EIO
; /* we fixed nothing */
672 static void end_workqueue_bio(struct bio
*bio
, int err
)
674 struct end_io_wq
*end_io_wq
= bio
->bi_private
;
675 struct btrfs_fs_info
*fs_info
;
677 fs_info
= end_io_wq
->info
;
678 end_io_wq
->error
= err
;
679 end_io_wq
->work
.func
= end_workqueue_fn
;
680 end_io_wq
->work
.flags
= 0;
682 if (bio
->bi_rw
& REQ_WRITE
) {
683 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
)
684 btrfs_queue_worker(&fs_info
->endio_meta_write_workers
,
686 else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
)
687 btrfs_queue_worker(&fs_info
->endio_freespace_worker
,
689 else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
)
690 btrfs_queue_worker(&fs_info
->endio_raid56_workers
,
693 btrfs_queue_worker(&fs_info
->endio_write_workers
,
696 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
)
697 btrfs_queue_worker(&fs_info
->endio_raid56_workers
,
699 else if (end_io_wq
->metadata
)
700 btrfs_queue_worker(&fs_info
->endio_meta_workers
,
703 btrfs_queue_worker(&fs_info
->endio_workers
,
709 * For the metadata arg you want
712 * 1 - if normal metadta
713 * 2 - if writing to the free space cache area
714 * 3 - raid parity work
716 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
719 struct end_io_wq
*end_io_wq
;
720 end_io_wq
= kmalloc(sizeof(*end_io_wq
), GFP_NOFS
);
724 end_io_wq
->private = bio
->bi_private
;
725 end_io_wq
->end_io
= bio
->bi_end_io
;
726 end_io_wq
->info
= info
;
727 end_io_wq
->error
= 0;
728 end_io_wq
->bio
= bio
;
729 end_io_wq
->metadata
= metadata
;
731 bio
->bi_private
= end_io_wq
;
732 bio
->bi_end_io
= end_workqueue_bio
;
736 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
738 unsigned long limit
= min_t(unsigned long,
739 info
->workers
.max_workers
,
740 info
->fs_devices
->open_devices
);
744 static void run_one_async_start(struct btrfs_work
*work
)
746 struct async_submit_bio
*async
;
749 async
= container_of(work
, struct async_submit_bio
, work
);
750 ret
= async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
751 async
->mirror_num
, async
->bio_flags
,
757 static void run_one_async_done(struct btrfs_work
*work
)
759 struct btrfs_fs_info
*fs_info
;
760 struct async_submit_bio
*async
;
763 async
= container_of(work
, struct async_submit_bio
, work
);
764 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
766 limit
= btrfs_async_submit_limit(fs_info
);
767 limit
= limit
* 2 / 3;
769 if (atomic_dec_return(&fs_info
->nr_async_submits
) < limit
&&
770 waitqueue_active(&fs_info
->async_submit_wait
))
771 wake_up(&fs_info
->async_submit_wait
);
773 /* If an error occured we just want to clean up the bio and move on */
775 bio_endio(async
->bio
, async
->error
);
779 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
780 async
->mirror_num
, async
->bio_flags
,
784 static void run_one_async_free(struct btrfs_work
*work
)
786 struct async_submit_bio
*async
;
788 async
= container_of(work
, struct async_submit_bio
, work
);
792 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
793 int rw
, struct bio
*bio
, int mirror_num
,
794 unsigned long bio_flags
,
796 extent_submit_bio_hook_t
*submit_bio_start
,
797 extent_submit_bio_hook_t
*submit_bio_done
)
799 struct async_submit_bio
*async
;
801 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
805 async
->inode
= inode
;
808 async
->mirror_num
= mirror_num
;
809 async
->submit_bio_start
= submit_bio_start
;
810 async
->submit_bio_done
= submit_bio_done
;
812 async
->work
.func
= run_one_async_start
;
813 async
->work
.ordered_func
= run_one_async_done
;
814 async
->work
.ordered_free
= run_one_async_free
;
816 async
->work
.flags
= 0;
817 async
->bio_flags
= bio_flags
;
818 async
->bio_offset
= bio_offset
;
822 atomic_inc(&fs_info
->nr_async_submits
);
825 btrfs_set_work_high_prio(&async
->work
);
827 btrfs_queue_worker(&fs_info
->workers
, &async
->work
);
829 while (atomic_read(&fs_info
->async_submit_draining
) &&
830 atomic_read(&fs_info
->nr_async_submits
)) {
831 wait_event(fs_info
->async_submit_wait
,
832 (atomic_read(&fs_info
->nr_async_submits
) == 0));
838 static int btree_csum_one_bio(struct bio
*bio
)
840 struct bio_vec
*bvec
= bio
->bi_io_vec
;
842 struct btrfs_root
*root
;
845 WARN_ON(bio
->bi_vcnt
<= 0);
846 while (bio_index
< bio
->bi_vcnt
) {
847 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
848 ret
= csum_dirty_buffer(root
, bvec
->bv_page
);
857 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
858 struct bio
*bio
, int mirror_num
,
859 unsigned long bio_flags
,
863 * when we're called for a write, we're already in the async
864 * submission context. Just jump into btrfs_map_bio
866 return btree_csum_one_bio(bio
);
869 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
870 int mirror_num
, unsigned long bio_flags
,
876 * when we're called for a write, we're already in the async
877 * submission context. Just jump into btrfs_map_bio
879 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
885 static int check_async_write(struct inode
*inode
, unsigned long bio_flags
)
887 if (bio_flags
& EXTENT_BIO_TREE_LOG
)
896 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
897 int mirror_num
, unsigned long bio_flags
,
900 int async
= check_async_write(inode
, bio_flags
);
903 if (!(rw
& REQ_WRITE
)) {
905 * called for a read, do the setup so that checksum validation
906 * can happen in the async kernel threads
908 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
912 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
915 ret
= btree_csum_one_bio(bio
);
918 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
922 * kthread helpers are used to submit writes so that
923 * checksumming can happen in parallel across all CPUs
925 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
926 inode
, rw
, bio
, mirror_num
, 0,
928 __btree_submit_bio_start
,
929 __btree_submit_bio_done
);
939 #ifdef CONFIG_MIGRATION
940 static int btree_migratepage(struct address_space
*mapping
,
941 struct page
*newpage
, struct page
*page
,
942 enum migrate_mode mode
)
945 * we can't safely write a btree page from here,
946 * we haven't done the locking hook
951 * Buffers may be managed in a filesystem specific way.
952 * We must have no buffers or drop them.
954 if (page_has_private(page
) &&
955 !try_to_release_page(page
, GFP_KERNEL
))
957 return migrate_page(mapping
, newpage
, page
, mode
);
962 static int btree_writepages(struct address_space
*mapping
,
963 struct writeback_control
*wbc
)
965 struct extent_io_tree
*tree
;
966 struct btrfs_fs_info
*fs_info
;
969 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
970 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
972 if (wbc
->for_kupdate
)
975 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
976 /* this is a bit racy, but that's ok */
977 ret
= percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
978 BTRFS_DIRTY_METADATA_THRESH
);
982 return btree_write_cache_pages(mapping
, wbc
);
985 static int btree_readpage(struct file
*file
, struct page
*page
)
987 struct extent_io_tree
*tree
;
988 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
989 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
992 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
994 if (PageWriteback(page
) || PageDirty(page
))
997 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
998 * slab allocation from alloc_extent_state down the callchain where
999 * it'd hit a BUG_ON as those flags are not allowed.
1001 gfp_flags
&= ~GFP_SLAB_BUG_MASK
;
1003 return try_release_extent_buffer(page
, gfp_flags
);
1006 static void btree_invalidatepage(struct page
*page
, unsigned long offset
)
1008 struct extent_io_tree
*tree
;
1009 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1010 extent_invalidatepage(tree
, page
, offset
);
1011 btree_releasepage(page
, GFP_NOFS
);
1012 if (PagePrivate(page
)) {
1013 printk(KERN_WARNING
"btrfs warning page private not zero "
1014 "on page %llu\n", (unsigned long long)page_offset(page
));
1015 ClearPagePrivate(page
);
1016 set_page_private(page
, 0);
1017 page_cache_release(page
);
1021 static int btree_set_page_dirty(struct page
*page
)
1024 struct extent_buffer
*eb
;
1026 BUG_ON(!PagePrivate(page
));
1027 eb
= (struct extent_buffer
*)page
->private;
1029 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
1030 BUG_ON(!atomic_read(&eb
->refs
));
1031 btrfs_assert_tree_locked(eb
);
1033 return __set_page_dirty_nobuffers(page
);
1036 static const struct address_space_operations btree_aops
= {
1037 .readpage
= btree_readpage
,
1038 .writepages
= btree_writepages
,
1039 .releasepage
= btree_releasepage
,
1040 .invalidatepage
= btree_invalidatepage
,
1041 #ifdef CONFIG_MIGRATION
1042 .migratepage
= btree_migratepage
,
1044 .set_page_dirty
= btree_set_page_dirty
,
1047 int readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
1050 struct extent_buffer
*buf
= NULL
;
1051 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1054 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1057 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1058 buf
, 0, WAIT_NONE
, btree_get_extent
, 0);
1059 free_extent_buffer(buf
);
1063 int reada_tree_block_flagged(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
1064 int mirror_num
, struct extent_buffer
**eb
)
1066 struct extent_buffer
*buf
= NULL
;
1067 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1068 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1071 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1075 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1077 ret
= read_extent_buffer_pages(io_tree
, buf
, 0, WAIT_PAGE_LOCK
,
1078 btree_get_extent
, mirror_num
);
1080 free_extent_buffer(buf
);
1084 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1085 free_extent_buffer(buf
);
1087 } else if (extent_buffer_uptodate(buf
)) {
1090 free_extent_buffer(buf
);
1095 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_root
*root
,
1096 u64 bytenr
, u32 blocksize
)
1098 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1099 struct extent_buffer
*eb
;
1100 eb
= find_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
1105 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
1106 u64 bytenr
, u32 blocksize
)
1108 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1109 struct extent_buffer
*eb
;
1111 eb
= alloc_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
1117 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1119 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1120 buf
->start
+ buf
->len
- 1);
1123 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1125 return filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1126 buf
->start
, buf
->start
+ buf
->len
- 1);
1129 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
1130 u32 blocksize
, u64 parent_transid
)
1132 struct extent_buffer
*buf
= NULL
;
1135 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1139 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
1144 void clean_tree_block(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1145 struct extent_buffer
*buf
)
1147 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1149 if (btrfs_header_generation(buf
) ==
1150 fs_info
->running_transaction
->transid
) {
1151 btrfs_assert_tree_locked(buf
);
1153 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1154 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
1156 fs_info
->dirty_metadata_batch
);
1157 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1158 btrfs_set_lock_blocking(buf
);
1159 clear_extent_buffer_dirty(buf
);
1164 static void __setup_root(u32 nodesize
, u32 leafsize
, u32 sectorsize
,
1165 u32 stripesize
, struct btrfs_root
*root
,
1166 struct btrfs_fs_info
*fs_info
,
1170 root
->commit_root
= NULL
;
1171 root
->sectorsize
= sectorsize
;
1172 root
->nodesize
= nodesize
;
1173 root
->leafsize
= leafsize
;
1174 root
->stripesize
= stripesize
;
1176 root
->track_dirty
= 0;
1178 root
->orphan_item_inserted
= 0;
1179 root
->orphan_cleanup_state
= 0;
1181 root
->objectid
= objectid
;
1182 root
->last_trans
= 0;
1183 root
->highest_objectid
= 0;
1185 root
->inode_tree
= RB_ROOT
;
1186 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1187 root
->block_rsv
= NULL
;
1188 root
->orphan_block_rsv
= NULL
;
1190 INIT_LIST_HEAD(&root
->dirty_list
);
1191 INIT_LIST_HEAD(&root
->root_list
);
1192 INIT_LIST_HEAD(&root
->logged_list
[0]);
1193 INIT_LIST_HEAD(&root
->logged_list
[1]);
1194 spin_lock_init(&root
->orphan_lock
);
1195 spin_lock_init(&root
->inode_lock
);
1196 spin_lock_init(&root
->accounting_lock
);
1197 spin_lock_init(&root
->log_extents_lock
[0]);
1198 spin_lock_init(&root
->log_extents_lock
[1]);
1199 mutex_init(&root
->objectid_mutex
);
1200 mutex_init(&root
->log_mutex
);
1201 init_waitqueue_head(&root
->log_writer_wait
);
1202 init_waitqueue_head(&root
->log_commit_wait
[0]);
1203 init_waitqueue_head(&root
->log_commit_wait
[1]);
1204 atomic_set(&root
->log_commit
[0], 0);
1205 atomic_set(&root
->log_commit
[1], 0);
1206 atomic_set(&root
->log_writers
, 0);
1207 atomic_set(&root
->log_batch
, 0);
1208 atomic_set(&root
->orphan_inodes
, 0);
1209 root
->log_transid
= 0;
1210 root
->last_log_commit
= 0;
1211 extent_io_tree_init(&root
->dirty_log_pages
,
1212 fs_info
->btree_inode
->i_mapping
);
1214 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1215 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1216 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1217 memset(&root
->root_kobj
, 0, sizeof(root
->root_kobj
));
1218 root
->defrag_trans_start
= fs_info
->generation
;
1219 init_completion(&root
->kobj_unregister
);
1220 root
->defrag_running
= 0;
1221 root
->root_key
.objectid
= objectid
;
1224 spin_lock_init(&root
->root_item_lock
);
1227 static int __must_check
find_and_setup_root(struct btrfs_root
*tree_root
,
1228 struct btrfs_fs_info
*fs_info
,
1230 struct btrfs_root
*root
)
1236 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1237 tree_root
->sectorsize
, tree_root
->stripesize
,
1238 root
, fs_info
, objectid
);
1239 ret
= btrfs_find_last_root(tree_root
, objectid
,
1240 &root
->root_item
, &root
->root_key
);
1246 generation
= btrfs_root_generation(&root
->root_item
);
1247 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1248 root
->commit_root
= NULL
;
1249 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1250 blocksize
, generation
);
1251 if (!root
->node
|| !btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1252 free_extent_buffer(root
->node
);
1256 root
->commit_root
= btrfs_root_node(root
);
1260 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
)
1262 struct btrfs_root
*root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1264 root
->fs_info
= fs_info
;
1268 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1269 struct btrfs_fs_info
*fs_info
,
1272 struct extent_buffer
*leaf
;
1273 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1274 struct btrfs_root
*root
;
1275 struct btrfs_key key
;
1279 root
= btrfs_alloc_root(fs_info
);
1281 return ERR_PTR(-ENOMEM
);
1283 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1284 tree_root
->sectorsize
, tree_root
->stripesize
,
1285 root
, fs_info
, objectid
);
1286 root
->root_key
.objectid
= objectid
;
1287 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1288 root
->root_key
.offset
= 0;
1290 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
,
1291 0, objectid
, NULL
, 0, 0, 0);
1293 ret
= PTR_ERR(leaf
);
1298 bytenr
= leaf
->start
;
1299 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1300 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1301 btrfs_set_header_generation(leaf
, trans
->transid
);
1302 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1303 btrfs_set_header_owner(leaf
, objectid
);
1306 write_extent_buffer(leaf
, fs_info
->fsid
,
1307 (unsigned long)btrfs_header_fsid(leaf
),
1309 write_extent_buffer(leaf
, fs_info
->chunk_tree_uuid
,
1310 (unsigned long)btrfs_header_chunk_tree_uuid(leaf
),
1312 btrfs_mark_buffer_dirty(leaf
);
1314 root
->commit_root
= btrfs_root_node(root
);
1315 root
->track_dirty
= 1;
1318 root
->root_item
.flags
= 0;
1319 root
->root_item
.byte_limit
= 0;
1320 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1321 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1322 btrfs_set_root_level(&root
->root_item
, 0);
1323 btrfs_set_root_refs(&root
->root_item
, 1);
1324 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1325 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1326 btrfs_set_root_dirid(&root
->root_item
, 0);
1327 root
->root_item
.drop_level
= 0;
1329 key
.objectid
= objectid
;
1330 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1332 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1336 btrfs_tree_unlock(leaf
);
1342 btrfs_tree_unlock(leaf
);
1343 free_extent_buffer(leaf
);
1347 return ERR_PTR(ret
);
1350 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1351 struct btrfs_fs_info
*fs_info
)
1353 struct btrfs_root
*root
;
1354 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1355 struct extent_buffer
*leaf
;
1357 root
= btrfs_alloc_root(fs_info
);
1359 return ERR_PTR(-ENOMEM
);
1361 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1362 tree_root
->sectorsize
, tree_root
->stripesize
,
1363 root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
1365 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1366 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1367 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1369 * log trees do not get reference counted because they go away
1370 * before a real commit is actually done. They do store pointers
1371 * to file data extents, and those reference counts still get
1372 * updated (along with back refs to the log tree).
1376 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
, 0,
1377 BTRFS_TREE_LOG_OBJECTID
, NULL
,
1381 return ERR_CAST(leaf
);
1384 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1385 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1386 btrfs_set_header_generation(leaf
, trans
->transid
);
1387 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1388 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1391 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1392 (unsigned long)btrfs_header_fsid(root
->node
),
1394 btrfs_mark_buffer_dirty(root
->node
);
1395 btrfs_tree_unlock(root
->node
);
1399 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1400 struct btrfs_fs_info
*fs_info
)
1402 struct btrfs_root
*log_root
;
1404 log_root
= alloc_log_tree(trans
, fs_info
);
1405 if (IS_ERR(log_root
))
1406 return PTR_ERR(log_root
);
1407 WARN_ON(fs_info
->log_root_tree
);
1408 fs_info
->log_root_tree
= log_root
;
1412 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1413 struct btrfs_root
*root
)
1415 struct btrfs_root
*log_root
;
1416 struct btrfs_inode_item
*inode_item
;
1418 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1419 if (IS_ERR(log_root
))
1420 return PTR_ERR(log_root
);
1422 log_root
->last_trans
= trans
->transid
;
1423 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1425 inode_item
= &log_root
->root_item
.inode
;
1426 inode_item
->generation
= cpu_to_le64(1);
1427 inode_item
->size
= cpu_to_le64(3);
1428 inode_item
->nlink
= cpu_to_le32(1);
1429 inode_item
->nbytes
= cpu_to_le64(root
->leafsize
);
1430 inode_item
->mode
= cpu_to_le32(S_IFDIR
| 0755);
1432 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1434 WARN_ON(root
->log_root
);
1435 root
->log_root
= log_root
;
1436 root
->log_transid
= 0;
1437 root
->last_log_commit
= 0;
1441 struct btrfs_root
*btrfs_read_fs_root_no_radix(struct btrfs_root
*tree_root
,
1442 struct btrfs_key
*location
)
1444 struct btrfs_root
*root
;
1445 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1446 struct btrfs_path
*path
;
1447 struct extent_buffer
*l
;
1453 root
= btrfs_alloc_root(fs_info
);
1455 return ERR_PTR(-ENOMEM
);
1456 if (location
->offset
== (u64
)-1) {
1457 ret
= find_and_setup_root(tree_root
, fs_info
,
1458 location
->objectid
, root
);
1461 return ERR_PTR(ret
);
1466 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1467 tree_root
->sectorsize
, tree_root
->stripesize
,
1468 root
, fs_info
, location
->objectid
);
1470 path
= btrfs_alloc_path();
1473 return ERR_PTR(-ENOMEM
);
1475 ret
= btrfs_search_slot(NULL
, tree_root
, location
, path
, 0, 0);
1478 slot
= path
->slots
[0];
1479 btrfs_read_root_item(tree_root
, l
, slot
, &root
->root_item
);
1480 memcpy(&root
->root_key
, location
, sizeof(*location
));
1482 btrfs_free_path(path
);
1487 return ERR_PTR(ret
);
1490 generation
= btrfs_root_generation(&root
->root_item
);
1491 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1492 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1493 blocksize
, generation
);
1494 root
->commit_root
= btrfs_root_node(root
);
1495 BUG_ON(!root
->node
); /* -ENOMEM */
1497 if (location
->objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1499 btrfs_check_and_init_root_item(&root
->root_item
);
1505 struct btrfs_root
*btrfs_read_fs_root_no_name(struct btrfs_fs_info
*fs_info
,
1506 struct btrfs_key
*location
)
1508 struct btrfs_root
*root
;
1511 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1512 return fs_info
->tree_root
;
1513 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1514 return fs_info
->extent_root
;
1515 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1516 return fs_info
->chunk_root
;
1517 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1518 return fs_info
->dev_root
;
1519 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1520 return fs_info
->csum_root
;
1521 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1522 return fs_info
->quota_root
? fs_info
->quota_root
:
1525 spin_lock(&fs_info
->fs_roots_radix_lock
);
1526 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1527 (unsigned long)location
->objectid
);
1528 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1532 root
= btrfs_read_fs_root_no_radix(fs_info
->tree_root
, location
);
1536 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1537 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1539 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1544 btrfs_init_free_ino_ctl(root
);
1545 mutex_init(&root
->fs_commit_mutex
);
1546 spin_lock_init(&root
->cache_lock
);
1547 init_waitqueue_head(&root
->cache_wait
);
1549 ret
= get_anon_bdev(&root
->anon_dev
);
1553 if (btrfs_root_refs(&root
->root_item
) == 0) {
1558 ret
= btrfs_find_orphan_item(fs_info
->tree_root
, location
->objectid
);
1562 root
->orphan_item_inserted
= 1;
1564 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
1568 spin_lock(&fs_info
->fs_roots_radix_lock
);
1569 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1570 (unsigned long)root
->root_key
.objectid
,
1575 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1576 radix_tree_preload_end();
1578 if (ret
== -EEXIST
) {
1585 ret
= btrfs_find_dead_roots(fs_info
->tree_root
,
1586 root
->root_key
.objectid
);
1591 return ERR_PTR(ret
);
1594 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1596 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1598 struct btrfs_device
*device
;
1599 struct backing_dev_info
*bdi
;
1602 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1605 bdi
= blk_get_backing_dev_info(device
->bdev
);
1606 if (bdi
&& bdi_congested(bdi
, bdi_bits
)) {
1616 * If this fails, caller must call bdi_destroy() to get rid of the
1619 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1623 bdi
->capabilities
= BDI_CAP_MAP_COPY
;
1624 err
= bdi_setup_and_register(bdi
, "btrfs", BDI_CAP_MAP_COPY
);
1628 bdi
->ra_pages
= default_backing_dev_info
.ra_pages
;
1629 bdi
->congested_fn
= btrfs_congested_fn
;
1630 bdi
->congested_data
= info
;
1635 * called by the kthread helper functions to finally call the bio end_io
1636 * functions. This is where read checksum verification actually happens
1638 static void end_workqueue_fn(struct btrfs_work
*work
)
1641 struct end_io_wq
*end_io_wq
;
1642 struct btrfs_fs_info
*fs_info
;
1645 end_io_wq
= container_of(work
, struct end_io_wq
, work
);
1646 bio
= end_io_wq
->bio
;
1647 fs_info
= end_io_wq
->info
;
1649 error
= end_io_wq
->error
;
1650 bio
->bi_private
= end_io_wq
->private;
1651 bio
->bi_end_io
= end_io_wq
->end_io
;
1653 bio_endio(bio
, error
);
1656 static int cleaner_kthread(void *arg
)
1658 struct btrfs_root
*root
= arg
;
1661 if (!(root
->fs_info
->sb
->s_flags
& MS_RDONLY
) &&
1662 mutex_trylock(&root
->fs_info
->cleaner_mutex
)) {
1663 btrfs_run_delayed_iputs(root
);
1664 btrfs_clean_old_snapshots(root
);
1665 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1666 btrfs_run_defrag_inodes(root
->fs_info
);
1669 if (!try_to_freeze()) {
1670 set_current_state(TASK_INTERRUPTIBLE
);
1671 if (!kthread_should_stop())
1673 __set_current_state(TASK_RUNNING
);
1675 } while (!kthread_should_stop());
1679 static int transaction_kthread(void *arg
)
1681 struct btrfs_root
*root
= arg
;
1682 struct btrfs_trans_handle
*trans
;
1683 struct btrfs_transaction
*cur
;
1686 unsigned long delay
;
1690 cannot_commit
= false;
1692 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1694 spin_lock(&root
->fs_info
->trans_lock
);
1695 cur
= root
->fs_info
->running_transaction
;
1697 spin_unlock(&root
->fs_info
->trans_lock
);
1701 now
= get_seconds();
1702 if (!cur
->blocked
&&
1703 (now
< cur
->start_time
|| now
- cur
->start_time
< 30)) {
1704 spin_unlock(&root
->fs_info
->trans_lock
);
1708 transid
= cur
->transid
;
1709 spin_unlock(&root
->fs_info
->trans_lock
);
1711 /* If the file system is aborted, this will always fail. */
1712 trans
= btrfs_attach_transaction(root
);
1713 if (IS_ERR(trans
)) {
1714 if (PTR_ERR(trans
) != -ENOENT
)
1715 cannot_commit
= true;
1718 if (transid
== trans
->transid
) {
1719 btrfs_commit_transaction(trans
, root
);
1721 btrfs_end_transaction(trans
, root
);
1724 wake_up_process(root
->fs_info
->cleaner_kthread
);
1725 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1727 if (!try_to_freeze()) {
1728 set_current_state(TASK_INTERRUPTIBLE
);
1729 if (!kthread_should_stop() &&
1730 (!btrfs_transaction_blocked(root
->fs_info
) ||
1732 schedule_timeout(delay
);
1733 __set_current_state(TASK_RUNNING
);
1735 } while (!kthread_should_stop());
1740 * this will find the highest generation in the array of
1741 * root backups. The index of the highest array is returned,
1742 * or -1 if we can't find anything.
1744 * We check to make sure the array is valid by comparing the
1745 * generation of the latest root in the array with the generation
1746 * in the super block. If they don't match we pitch it.
1748 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1751 int newest_index
= -1;
1752 struct btrfs_root_backup
*root_backup
;
1755 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1756 root_backup
= info
->super_copy
->super_roots
+ i
;
1757 cur
= btrfs_backup_tree_root_gen(root_backup
);
1758 if (cur
== newest_gen
)
1762 /* check to see if we actually wrapped around */
1763 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1764 root_backup
= info
->super_copy
->super_roots
;
1765 cur
= btrfs_backup_tree_root_gen(root_backup
);
1766 if (cur
== newest_gen
)
1769 return newest_index
;
1774 * find the oldest backup so we know where to store new entries
1775 * in the backup array. This will set the backup_root_index
1776 * field in the fs_info struct
1778 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1781 int newest_index
= -1;
1783 newest_index
= find_newest_super_backup(info
, newest_gen
);
1784 /* if there was garbage in there, just move along */
1785 if (newest_index
== -1) {
1786 info
->backup_root_index
= 0;
1788 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1793 * copy all the root pointers into the super backup array.
1794 * this will bump the backup pointer by one when it is
1797 static void backup_super_roots(struct btrfs_fs_info
*info
)
1800 struct btrfs_root_backup
*root_backup
;
1803 next_backup
= info
->backup_root_index
;
1804 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1805 BTRFS_NUM_BACKUP_ROOTS
;
1808 * just overwrite the last backup if we're at the same generation
1809 * this happens only at umount
1811 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
1812 if (btrfs_backup_tree_root_gen(root_backup
) ==
1813 btrfs_header_generation(info
->tree_root
->node
))
1814 next_backup
= last_backup
;
1816 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
1819 * make sure all of our padding and empty slots get zero filled
1820 * regardless of which ones we use today
1822 memset(root_backup
, 0, sizeof(*root_backup
));
1824 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1826 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
1827 btrfs_set_backup_tree_root_gen(root_backup
,
1828 btrfs_header_generation(info
->tree_root
->node
));
1830 btrfs_set_backup_tree_root_level(root_backup
,
1831 btrfs_header_level(info
->tree_root
->node
));
1833 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
1834 btrfs_set_backup_chunk_root_gen(root_backup
,
1835 btrfs_header_generation(info
->chunk_root
->node
));
1836 btrfs_set_backup_chunk_root_level(root_backup
,
1837 btrfs_header_level(info
->chunk_root
->node
));
1839 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
1840 btrfs_set_backup_extent_root_gen(root_backup
,
1841 btrfs_header_generation(info
->extent_root
->node
));
1842 btrfs_set_backup_extent_root_level(root_backup
,
1843 btrfs_header_level(info
->extent_root
->node
));
1846 * we might commit during log recovery, which happens before we set
1847 * the fs_root. Make sure it is valid before we fill it in.
1849 if (info
->fs_root
&& info
->fs_root
->node
) {
1850 btrfs_set_backup_fs_root(root_backup
,
1851 info
->fs_root
->node
->start
);
1852 btrfs_set_backup_fs_root_gen(root_backup
,
1853 btrfs_header_generation(info
->fs_root
->node
));
1854 btrfs_set_backup_fs_root_level(root_backup
,
1855 btrfs_header_level(info
->fs_root
->node
));
1858 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
1859 btrfs_set_backup_dev_root_gen(root_backup
,
1860 btrfs_header_generation(info
->dev_root
->node
));
1861 btrfs_set_backup_dev_root_level(root_backup
,
1862 btrfs_header_level(info
->dev_root
->node
));
1864 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
1865 btrfs_set_backup_csum_root_gen(root_backup
,
1866 btrfs_header_generation(info
->csum_root
->node
));
1867 btrfs_set_backup_csum_root_level(root_backup
,
1868 btrfs_header_level(info
->csum_root
->node
));
1870 btrfs_set_backup_total_bytes(root_backup
,
1871 btrfs_super_total_bytes(info
->super_copy
));
1872 btrfs_set_backup_bytes_used(root_backup
,
1873 btrfs_super_bytes_used(info
->super_copy
));
1874 btrfs_set_backup_num_devices(root_backup
,
1875 btrfs_super_num_devices(info
->super_copy
));
1878 * if we don't copy this out to the super_copy, it won't get remembered
1879 * for the next commit
1881 memcpy(&info
->super_copy
->super_roots
,
1882 &info
->super_for_commit
->super_roots
,
1883 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
1887 * this copies info out of the root backup array and back into
1888 * the in-memory super block. It is meant to help iterate through
1889 * the array, so you send it the number of backups you've already
1890 * tried and the last backup index you used.
1892 * this returns -1 when it has tried all the backups
1894 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
1895 struct btrfs_super_block
*super
,
1896 int *num_backups_tried
, int *backup_index
)
1898 struct btrfs_root_backup
*root_backup
;
1899 int newest
= *backup_index
;
1901 if (*num_backups_tried
== 0) {
1902 u64 gen
= btrfs_super_generation(super
);
1904 newest
= find_newest_super_backup(info
, gen
);
1908 *backup_index
= newest
;
1909 *num_backups_tried
= 1;
1910 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
1911 /* we've tried all the backups, all done */
1914 /* jump to the next oldest backup */
1915 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1916 BTRFS_NUM_BACKUP_ROOTS
;
1917 *backup_index
= newest
;
1918 *num_backups_tried
+= 1;
1920 root_backup
= super
->super_roots
+ newest
;
1922 btrfs_set_super_generation(super
,
1923 btrfs_backup_tree_root_gen(root_backup
));
1924 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
1925 btrfs_set_super_root_level(super
,
1926 btrfs_backup_tree_root_level(root_backup
));
1927 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
1930 * fixme: the total bytes and num_devices need to match or we should
1933 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
1934 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
1938 /* helper to cleanup tree roots */
1939 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
1941 free_extent_buffer(info
->tree_root
->node
);
1942 free_extent_buffer(info
->tree_root
->commit_root
);
1943 free_extent_buffer(info
->dev_root
->node
);
1944 free_extent_buffer(info
->dev_root
->commit_root
);
1945 free_extent_buffer(info
->extent_root
->node
);
1946 free_extent_buffer(info
->extent_root
->commit_root
);
1947 free_extent_buffer(info
->csum_root
->node
);
1948 free_extent_buffer(info
->csum_root
->commit_root
);
1949 if (info
->quota_root
) {
1950 free_extent_buffer(info
->quota_root
->node
);
1951 free_extent_buffer(info
->quota_root
->commit_root
);
1954 info
->tree_root
->node
= NULL
;
1955 info
->tree_root
->commit_root
= NULL
;
1956 info
->dev_root
->node
= NULL
;
1957 info
->dev_root
->commit_root
= NULL
;
1958 info
->extent_root
->node
= NULL
;
1959 info
->extent_root
->commit_root
= NULL
;
1960 info
->csum_root
->node
= NULL
;
1961 info
->csum_root
->commit_root
= NULL
;
1962 if (info
->quota_root
) {
1963 info
->quota_root
->node
= NULL
;
1964 info
->quota_root
->commit_root
= NULL
;
1968 free_extent_buffer(info
->chunk_root
->node
);
1969 free_extent_buffer(info
->chunk_root
->commit_root
);
1970 info
->chunk_root
->node
= NULL
;
1971 info
->chunk_root
->commit_root
= NULL
;
1976 int open_ctree(struct super_block
*sb
,
1977 struct btrfs_fs_devices
*fs_devices
,
1987 struct btrfs_key location
;
1988 struct buffer_head
*bh
;
1989 struct btrfs_super_block
*disk_super
;
1990 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
1991 struct btrfs_root
*tree_root
;
1992 struct btrfs_root
*extent_root
;
1993 struct btrfs_root
*csum_root
;
1994 struct btrfs_root
*chunk_root
;
1995 struct btrfs_root
*dev_root
;
1996 struct btrfs_root
*quota_root
;
1997 struct btrfs_root
*log_tree_root
;
2000 int num_backups_tried
= 0;
2001 int backup_index
= 0;
2003 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
);
2004 extent_root
= fs_info
->extent_root
= btrfs_alloc_root(fs_info
);
2005 csum_root
= fs_info
->csum_root
= btrfs_alloc_root(fs_info
);
2006 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
);
2007 dev_root
= fs_info
->dev_root
= btrfs_alloc_root(fs_info
);
2008 quota_root
= fs_info
->quota_root
= btrfs_alloc_root(fs_info
);
2010 if (!tree_root
|| !extent_root
|| !csum_root
||
2011 !chunk_root
|| !dev_root
|| !quota_root
) {
2016 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
2022 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
2028 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0);
2033 fs_info
->dirty_metadata_batch
= PAGE_CACHE_SIZE
*
2034 (1 + ilog2(nr_cpu_ids
));
2036 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0);
2039 goto fail_dirty_metadata_bytes
;
2042 fs_info
->btree_inode
= new_inode(sb
);
2043 if (!fs_info
->btree_inode
) {
2045 goto fail_delalloc_bytes
;
2048 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2050 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2051 INIT_LIST_HEAD(&fs_info
->trans_list
);
2052 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2053 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2054 INIT_LIST_HEAD(&fs_info
->delalloc_inodes
);
2055 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2056 spin_lock_init(&fs_info
->delalloc_lock
);
2057 spin_lock_init(&fs_info
->trans_lock
);
2058 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2059 spin_lock_init(&fs_info
->delayed_iput_lock
);
2060 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2061 spin_lock_init(&fs_info
->free_chunk_lock
);
2062 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2063 rwlock_init(&fs_info
->tree_mod_log_lock
);
2064 mutex_init(&fs_info
->reloc_mutex
);
2065 seqlock_init(&fs_info
->profiles_lock
);
2067 init_completion(&fs_info
->kobj_unregister
);
2068 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2069 INIT_LIST_HEAD(&fs_info
->space_info
);
2070 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2071 btrfs_mapping_init(&fs_info
->mapping_tree
);
2072 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2073 BTRFS_BLOCK_RSV_GLOBAL
);
2074 btrfs_init_block_rsv(&fs_info
->delalloc_block_rsv
,
2075 BTRFS_BLOCK_RSV_DELALLOC
);
2076 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2077 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2078 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2079 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2080 BTRFS_BLOCK_RSV_DELOPS
);
2081 atomic_set(&fs_info
->nr_async_submits
, 0);
2082 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2083 atomic_set(&fs_info
->async_submit_draining
, 0);
2084 atomic_set(&fs_info
->nr_async_bios
, 0);
2085 atomic_set(&fs_info
->defrag_running
, 0);
2086 atomic_set(&fs_info
->tree_mod_seq
, 0);
2088 fs_info
->max_inline
= 8192 * 1024;
2089 fs_info
->metadata_ratio
= 0;
2090 fs_info
->defrag_inodes
= RB_ROOT
;
2091 fs_info
->trans_no_join
= 0;
2092 fs_info
->free_chunk_space
= 0;
2093 fs_info
->tree_mod_log
= RB_ROOT
;
2095 /* readahead state */
2096 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_WAIT
);
2097 spin_lock_init(&fs_info
->reada_lock
);
2099 fs_info
->thread_pool_size
= min_t(unsigned long,
2100 num_online_cpus() + 2, 8);
2102 INIT_LIST_HEAD(&fs_info
->ordered_extents
);
2103 spin_lock_init(&fs_info
->ordered_extent_lock
);
2104 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2106 if (!fs_info
->delayed_root
) {
2110 btrfs_init_delayed_root(fs_info
->delayed_root
);
2112 mutex_init(&fs_info
->scrub_lock
);
2113 atomic_set(&fs_info
->scrubs_running
, 0);
2114 atomic_set(&fs_info
->scrub_pause_req
, 0);
2115 atomic_set(&fs_info
->scrubs_paused
, 0);
2116 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2117 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2118 init_rwsem(&fs_info
->scrub_super_lock
);
2119 fs_info
->scrub_workers_refcnt
= 0;
2120 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2121 fs_info
->check_integrity_print_mask
= 0;
2124 spin_lock_init(&fs_info
->balance_lock
);
2125 mutex_init(&fs_info
->balance_mutex
);
2126 atomic_set(&fs_info
->balance_running
, 0);
2127 atomic_set(&fs_info
->balance_pause_req
, 0);
2128 atomic_set(&fs_info
->balance_cancel_req
, 0);
2129 fs_info
->balance_ctl
= NULL
;
2130 init_waitqueue_head(&fs_info
->balance_wait_q
);
2132 sb
->s_blocksize
= 4096;
2133 sb
->s_blocksize_bits
= blksize_bits(4096);
2134 sb
->s_bdi
= &fs_info
->bdi
;
2136 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2137 set_nlink(fs_info
->btree_inode
, 1);
2139 * we set the i_size on the btree inode to the max possible int.
2140 * the real end of the address space is determined by all of
2141 * the devices in the system
2143 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
2144 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
2145 fs_info
->btree_inode
->i_mapping
->backing_dev_info
= &fs_info
->bdi
;
2147 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
2148 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
2149 fs_info
->btree_inode
->i_mapping
);
2150 BTRFS_I(fs_info
->btree_inode
)->io_tree
.track_uptodate
= 0;
2151 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
);
2153 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2155 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
2156 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
2157 sizeof(struct btrfs_key
));
2158 set_bit(BTRFS_INODE_DUMMY
,
2159 &BTRFS_I(fs_info
->btree_inode
)->runtime_flags
);
2160 insert_inode_hash(fs_info
->btree_inode
);
2162 spin_lock_init(&fs_info
->block_group_cache_lock
);
2163 fs_info
->block_group_cache_tree
= RB_ROOT
;
2164 fs_info
->first_logical_byte
= (u64
)-1;
2166 extent_io_tree_init(&fs_info
->freed_extents
[0],
2167 fs_info
->btree_inode
->i_mapping
);
2168 extent_io_tree_init(&fs_info
->freed_extents
[1],
2169 fs_info
->btree_inode
->i_mapping
);
2170 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2171 fs_info
->do_barriers
= 1;
2174 mutex_init(&fs_info
->ordered_operations_mutex
);
2175 mutex_init(&fs_info
->tree_log_mutex
);
2176 mutex_init(&fs_info
->chunk_mutex
);
2177 mutex_init(&fs_info
->transaction_kthread_mutex
);
2178 mutex_init(&fs_info
->cleaner_mutex
);
2179 mutex_init(&fs_info
->volume_mutex
);
2180 init_rwsem(&fs_info
->extent_commit_sem
);
2181 init_rwsem(&fs_info
->cleanup_work_sem
);
2182 init_rwsem(&fs_info
->subvol_sem
);
2183 fs_info
->dev_replace
.lock_owner
= 0;
2184 atomic_set(&fs_info
->dev_replace
.nesting_level
, 0);
2185 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2186 mutex_init(&fs_info
->dev_replace
.lock_management_lock
);
2187 mutex_init(&fs_info
->dev_replace
.lock
);
2189 spin_lock_init(&fs_info
->qgroup_lock
);
2190 fs_info
->qgroup_tree
= RB_ROOT
;
2191 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2192 fs_info
->qgroup_seq
= 1;
2193 fs_info
->quota_enabled
= 0;
2194 fs_info
->pending_quota_state
= 0;
2196 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2197 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2199 init_waitqueue_head(&fs_info
->transaction_throttle
);
2200 init_waitqueue_head(&fs_info
->transaction_wait
);
2201 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2202 init_waitqueue_head(&fs_info
->async_submit_wait
);
2204 ret
= btrfs_alloc_stripe_hash_table(fs_info
);
2210 __setup_root(4096, 4096, 4096, 4096, tree_root
,
2211 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2213 invalidate_bdev(fs_devices
->latest_bdev
);
2214 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2220 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2221 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2222 sizeof(*fs_info
->super_for_commit
));
2225 memcpy(fs_info
->fsid
, fs_info
->super_copy
->fsid
, BTRFS_FSID_SIZE
);
2227 disk_super
= fs_info
->super_copy
;
2228 if (!btrfs_super_root(disk_super
))
2231 /* check FS state, whether FS is broken. */
2232 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2233 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2235 ret
= btrfs_check_super_valid(fs_info
, sb
->s_flags
& MS_RDONLY
);
2237 printk(KERN_ERR
"btrfs: superblock contains fatal errors\n");
2243 * run through our array of backup supers and setup
2244 * our ring pointer to the oldest one
2246 generation
= btrfs_super_generation(disk_super
);
2247 find_oldest_super_backup(fs_info
, generation
);
2250 * In the long term, we'll store the compression type in the super
2251 * block, and it'll be used for per file compression control.
2253 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2255 ret
= btrfs_parse_options(tree_root
, options
);
2261 features
= btrfs_super_incompat_flags(disk_super
) &
2262 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2264 printk(KERN_ERR
"BTRFS: couldn't mount because of "
2265 "unsupported optional features (%Lx).\n",
2266 (unsigned long long)features
);
2271 if (btrfs_super_leafsize(disk_super
) !=
2272 btrfs_super_nodesize(disk_super
)) {
2273 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2274 "blocksizes don't match. node %d leaf %d\n",
2275 btrfs_super_nodesize(disk_super
),
2276 btrfs_super_leafsize(disk_super
));
2280 if (btrfs_super_leafsize(disk_super
) > BTRFS_MAX_METADATA_BLOCKSIZE
) {
2281 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2282 "blocksize (%d) was too large\n",
2283 btrfs_super_leafsize(disk_super
));
2288 features
= btrfs_super_incompat_flags(disk_super
);
2289 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2290 if (tree_root
->fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2291 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2294 * flag our filesystem as having big metadata blocks if
2295 * they are bigger than the page size
2297 if (btrfs_super_leafsize(disk_super
) > PAGE_CACHE_SIZE
) {
2298 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2299 printk(KERN_INFO
"btrfs flagging fs with big metadata feature\n");
2300 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2303 nodesize
= btrfs_super_nodesize(disk_super
);
2304 leafsize
= btrfs_super_leafsize(disk_super
);
2305 sectorsize
= btrfs_super_sectorsize(disk_super
);
2306 stripesize
= btrfs_super_stripesize(disk_super
);
2307 fs_info
->dirty_metadata_batch
= leafsize
* (1 + ilog2(nr_cpu_ids
));
2308 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
2311 * mixed block groups end up with duplicate but slightly offset
2312 * extent buffers for the same range. It leads to corruptions
2314 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2315 (sectorsize
!= leafsize
)) {
2316 printk(KERN_WARNING
"btrfs: unequal leaf/node/sector sizes "
2317 "are not allowed for mixed block groups on %s\n",
2322 btrfs_set_super_incompat_flags(disk_super
, features
);
2324 features
= btrfs_super_compat_ro_flags(disk_super
) &
2325 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2326 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
2327 printk(KERN_ERR
"BTRFS: couldn't mount RDWR because of "
2328 "unsupported option features (%Lx).\n",
2329 (unsigned long long)features
);
2334 btrfs_init_workers(&fs_info
->generic_worker
,
2335 "genwork", 1, NULL
);
2337 btrfs_init_workers(&fs_info
->workers
, "worker",
2338 fs_info
->thread_pool_size
,
2339 &fs_info
->generic_worker
);
2341 btrfs_init_workers(&fs_info
->delalloc_workers
, "delalloc",
2342 fs_info
->thread_pool_size
,
2343 &fs_info
->generic_worker
);
2345 btrfs_init_workers(&fs_info
->flush_workers
, "flush_delalloc",
2346 fs_info
->thread_pool_size
,
2347 &fs_info
->generic_worker
);
2349 btrfs_init_workers(&fs_info
->submit_workers
, "submit",
2350 min_t(u64
, fs_devices
->num_devices
,
2351 fs_info
->thread_pool_size
),
2352 &fs_info
->generic_worker
);
2354 btrfs_init_workers(&fs_info
->caching_workers
, "cache",
2355 2, &fs_info
->generic_worker
);
2357 /* a higher idle thresh on the submit workers makes it much more
2358 * likely that bios will be send down in a sane order to the
2361 fs_info
->submit_workers
.idle_thresh
= 64;
2363 fs_info
->workers
.idle_thresh
= 16;
2364 fs_info
->workers
.ordered
= 1;
2366 fs_info
->delalloc_workers
.idle_thresh
= 2;
2367 fs_info
->delalloc_workers
.ordered
= 1;
2369 btrfs_init_workers(&fs_info
->fixup_workers
, "fixup", 1,
2370 &fs_info
->generic_worker
);
2371 btrfs_init_workers(&fs_info
->endio_workers
, "endio",
2372 fs_info
->thread_pool_size
,
2373 &fs_info
->generic_worker
);
2374 btrfs_init_workers(&fs_info
->endio_meta_workers
, "endio-meta",
2375 fs_info
->thread_pool_size
,
2376 &fs_info
->generic_worker
);
2377 btrfs_init_workers(&fs_info
->endio_meta_write_workers
,
2378 "endio-meta-write", fs_info
->thread_pool_size
,
2379 &fs_info
->generic_worker
);
2380 btrfs_init_workers(&fs_info
->endio_raid56_workers
,
2381 "endio-raid56", fs_info
->thread_pool_size
,
2382 &fs_info
->generic_worker
);
2383 btrfs_init_workers(&fs_info
->rmw_workers
,
2384 "rmw", fs_info
->thread_pool_size
,
2385 &fs_info
->generic_worker
);
2386 btrfs_init_workers(&fs_info
->endio_write_workers
, "endio-write",
2387 fs_info
->thread_pool_size
,
2388 &fs_info
->generic_worker
);
2389 btrfs_init_workers(&fs_info
->endio_freespace_worker
, "freespace-write",
2390 1, &fs_info
->generic_worker
);
2391 btrfs_init_workers(&fs_info
->delayed_workers
, "delayed-meta",
2392 fs_info
->thread_pool_size
,
2393 &fs_info
->generic_worker
);
2394 btrfs_init_workers(&fs_info
->readahead_workers
, "readahead",
2395 fs_info
->thread_pool_size
,
2396 &fs_info
->generic_worker
);
2399 * endios are largely parallel and should have a very
2402 fs_info
->endio_workers
.idle_thresh
= 4;
2403 fs_info
->endio_meta_workers
.idle_thresh
= 4;
2404 fs_info
->endio_raid56_workers
.idle_thresh
= 4;
2405 fs_info
->rmw_workers
.idle_thresh
= 2;
2407 fs_info
->endio_write_workers
.idle_thresh
= 2;
2408 fs_info
->endio_meta_write_workers
.idle_thresh
= 2;
2409 fs_info
->readahead_workers
.idle_thresh
= 2;
2412 * btrfs_start_workers can really only fail because of ENOMEM so just
2413 * return -ENOMEM if any of these fail.
2415 ret
= btrfs_start_workers(&fs_info
->workers
);
2416 ret
|= btrfs_start_workers(&fs_info
->generic_worker
);
2417 ret
|= btrfs_start_workers(&fs_info
->submit_workers
);
2418 ret
|= btrfs_start_workers(&fs_info
->delalloc_workers
);
2419 ret
|= btrfs_start_workers(&fs_info
->fixup_workers
);
2420 ret
|= btrfs_start_workers(&fs_info
->endio_workers
);
2421 ret
|= btrfs_start_workers(&fs_info
->endio_meta_workers
);
2422 ret
|= btrfs_start_workers(&fs_info
->rmw_workers
);
2423 ret
|= btrfs_start_workers(&fs_info
->endio_raid56_workers
);
2424 ret
|= btrfs_start_workers(&fs_info
->endio_meta_write_workers
);
2425 ret
|= btrfs_start_workers(&fs_info
->endio_write_workers
);
2426 ret
|= btrfs_start_workers(&fs_info
->endio_freespace_worker
);
2427 ret
|= btrfs_start_workers(&fs_info
->delayed_workers
);
2428 ret
|= btrfs_start_workers(&fs_info
->caching_workers
);
2429 ret
|= btrfs_start_workers(&fs_info
->readahead_workers
);
2430 ret
|= btrfs_start_workers(&fs_info
->flush_workers
);
2433 goto fail_sb_buffer
;
2436 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
2437 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
2438 4 * 1024 * 1024 / PAGE_CACHE_SIZE
);
2440 tree_root
->nodesize
= nodesize
;
2441 tree_root
->leafsize
= leafsize
;
2442 tree_root
->sectorsize
= sectorsize
;
2443 tree_root
->stripesize
= stripesize
;
2445 sb
->s_blocksize
= sectorsize
;
2446 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
2448 if (disk_super
->magic
!= cpu_to_le64(BTRFS_MAGIC
)) {
2449 printk(KERN_INFO
"btrfs: valid FS not found on %s\n", sb
->s_id
);
2450 goto fail_sb_buffer
;
2453 if (sectorsize
!= PAGE_SIZE
) {
2454 printk(KERN_WARNING
"btrfs: Incompatible sector size(%lu) "
2455 "found on %s\n", (unsigned long)sectorsize
, sb
->s_id
);
2456 goto fail_sb_buffer
;
2459 mutex_lock(&fs_info
->chunk_mutex
);
2460 ret
= btrfs_read_sys_array(tree_root
);
2461 mutex_unlock(&fs_info
->chunk_mutex
);
2463 printk(KERN_WARNING
"btrfs: failed to read the system "
2464 "array on %s\n", sb
->s_id
);
2465 goto fail_sb_buffer
;
2468 blocksize
= btrfs_level_size(tree_root
,
2469 btrfs_super_chunk_root_level(disk_super
));
2470 generation
= btrfs_super_chunk_root_generation(disk_super
);
2472 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
2473 chunk_root
, fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2475 chunk_root
->node
= read_tree_block(chunk_root
,
2476 btrfs_super_chunk_root(disk_super
),
2477 blocksize
, generation
);
2478 BUG_ON(!chunk_root
->node
); /* -ENOMEM */
2479 if (!test_bit(EXTENT_BUFFER_UPTODATE
, &chunk_root
->node
->bflags
)) {
2480 printk(KERN_WARNING
"btrfs: failed to read chunk root on %s\n",
2482 goto fail_tree_roots
;
2484 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2485 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2487 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2488 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root
->node
),
2491 ret
= btrfs_read_chunk_tree(chunk_root
);
2493 printk(KERN_WARNING
"btrfs: failed to read chunk tree on %s\n",
2495 goto fail_tree_roots
;
2499 * keep the device that is marked to be the target device for the
2500 * dev_replace procedure
2502 btrfs_close_extra_devices(fs_info
, fs_devices
, 0);
2504 if (!fs_devices
->latest_bdev
) {
2505 printk(KERN_CRIT
"btrfs: failed to read devices on %s\n",
2507 goto fail_tree_roots
;
2511 blocksize
= btrfs_level_size(tree_root
,
2512 btrfs_super_root_level(disk_super
));
2513 generation
= btrfs_super_generation(disk_super
);
2515 tree_root
->node
= read_tree_block(tree_root
,
2516 btrfs_super_root(disk_super
),
2517 blocksize
, generation
);
2518 if (!tree_root
->node
||
2519 !test_bit(EXTENT_BUFFER_UPTODATE
, &tree_root
->node
->bflags
)) {
2520 printk(KERN_WARNING
"btrfs: failed to read tree root on %s\n",
2523 goto recovery_tree_root
;
2526 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2527 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2529 ret
= find_and_setup_root(tree_root
, fs_info
,
2530 BTRFS_EXTENT_TREE_OBJECTID
, extent_root
);
2532 goto recovery_tree_root
;
2533 extent_root
->track_dirty
= 1;
2535 ret
= find_and_setup_root(tree_root
, fs_info
,
2536 BTRFS_DEV_TREE_OBJECTID
, dev_root
);
2538 goto recovery_tree_root
;
2539 dev_root
->track_dirty
= 1;
2541 ret
= find_and_setup_root(tree_root
, fs_info
,
2542 BTRFS_CSUM_TREE_OBJECTID
, csum_root
);
2544 goto recovery_tree_root
;
2545 csum_root
->track_dirty
= 1;
2547 ret
= find_and_setup_root(tree_root
, fs_info
,
2548 BTRFS_QUOTA_TREE_OBJECTID
, quota_root
);
2551 quota_root
= fs_info
->quota_root
= NULL
;
2553 quota_root
->track_dirty
= 1;
2554 fs_info
->quota_enabled
= 1;
2555 fs_info
->pending_quota_state
= 1;
2558 fs_info
->generation
= generation
;
2559 fs_info
->last_trans_committed
= generation
;
2561 ret
= btrfs_recover_balance(fs_info
);
2563 printk(KERN_WARNING
"btrfs: failed to recover balance\n");
2564 goto fail_block_groups
;
2567 ret
= btrfs_init_dev_stats(fs_info
);
2569 printk(KERN_ERR
"btrfs: failed to init dev_stats: %d\n",
2571 goto fail_block_groups
;
2574 ret
= btrfs_init_dev_replace(fs_info
);
2576 pr_err("btrfs: failed to init dev_replace: %d\n", ret
);
2577 goto fail_block_groups
;
2580 btrfs_close_extra_devices(fs_info
, fs_devices
, 1);
2582 ret
= btrfs_init_space_info(fs_info
);
2584 printk(KERN_ERR
"Failed to initial space info: %d\n", ret
);
2585 goto fail_block_groups
;
2588 ret
= btrfs_read_block_groups(extent_root
);
2590 printk(KERN_ERR
"Failed to read block groups: %d\n", ret
);
2591 goto fail_block_groups
;
2593 fs_info
->num_tolerated_disk_barrier_failures
=
2594 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2595 if (fs_info
->fs_devices
->missing_devices
>
2596 fs_info
->num_tolerated_disk_barrier_failures
&&
2597 !(sb
->s_flags
& MS_RDONLY
)) {
2599 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2600 goto fail_block_groups
;
2603 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
2605 if (IS_ERR(fs_info
->cleaner_kthread
))
2606 goto fail_block_groups
;
2608 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
2610 "btrfs-transaction");
2611 if (IS_ERR(fs_info
->transaction_kthread
))
2614 if (!btrfs_test_opt(tree_root
, SSD
) &&
2615 !btrfs_test_opt(tree_root
, NOSSD
) &&
2616 !fs_info
->fs_devices
->rotating
) {
2617 printk(KERN_INFO
"Btrfs detected SSD devices, enabling SSD "
2619 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
2622 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2623 if (btrfs_test_opt(tree_root
, CHECK_INTEGRITY
)) {
2624 ret
= btrfsic_mount(tree_root
, fs_devices
,
2625 btrfs_test_opt(tree_root
,
2626 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
2628 fs_info
->check_integrity_print_mask
);
2630 printk(KERN_WARNING
"btrfs: failed to initialize"
2631 " integrity check module %s\n", sb
->s_id
);
2634 ret
= btrfs_read_qgroup_config(fs_info
);
2636 goto fail_trans_kthread
;
2638 /* do not make disk changes in broken FS */
2639 if (btrfs_super_log_root(disk_super
) != 0) {
2640 u64 bytenr
= btrfs_super_log_root(disk_super
);
2642 if (fs_devices
->rw_devices
== 0) {
2643 printk(KERN_WARNING
"Btrfs log replay required "
2649 btrfs_level_size(tree_root
,
2650 btrfs_super_log_root_level(disk_super
));
2652 log_tree_root
= btrfs_alloc_root(fs_info
);
2653 if (!log_tree_root
) {
2658 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
2659 log_tree_root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
2661 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
2664 /* returns with log_tree_root freed on success */
2665 ret
= btrfs_recover_log_trees(log_tree_root
);
2667 btrfs_error(tree_root
->fs_info
, ret
,
2668 "Failed to recover log tree");
2669 free_extent_buffer(log_tree_root
->node
);
2670 kfree(log_tree_root
);
2671 goto fail_trans_kthread
;
2674 if (sb
->s_flags
& MS_RDONLY
) {
2675 ret
= btrfs_commit_super(tree_root
);
2677 goto fail_trans_kthread
;
2681 ret
= btrfs_find_orphan_roots(tree_root
);
2683 goto fail_trans_kthread
;
2685 if (!(sb
->s_flags
& MS_RDONLY
)) {
2686 ret
= btrfs_cleanup_fs_roots(fs_info
);
2688 goto fail_trans_kthread
;
2690 ret
= btrfs_recover_relocation(tree_root
);
2693 "btrfs: failed to recover relocation\n");
2699 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
2700 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2701 location
.offset
= (u64
)-1;
2703 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
2704 if (!fs_info
->fs_root
)
2706 if (IS_ERR(fs_info
->fs_root
)) {
2707 err
= PTR_ERR(fs_info
->fs_root
);
2711 if (sb
->s_flags
& MS_RDONLY
)
2714 down_read(&fs_info
->cleanup_work_sem
);
2715 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
2716 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
2717 up_read(&fs_info
->cleanup_work_sem
);
2718 close_ctree(tree_root
);
2721 up_read(&fs_info
->cleanup_work_sem
);
2723 ret
= btrfs_resume_balance_async(fs_info
);
2725 printk(KERN_WARNING
"btrfs: failed to resume balance\n");
2726 close_ctree(tree_root
);
2730 ret
= btrfs_resume_dev_replace_async(fs_info
);
2732 pr_warn("btrfs: failed to resume dev_replace\n");
2733 close_ctree(tree_root
);
2740 btrfs_free_qgroup_config(fs_info
);
2742 kthread_stop(fs_info
->transaction_kthread
);
2744 kthread_stop(fs_info
->cleaner_kthread
);
2747 * make sure we're done with the btree inode before we stop our
2750 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
2753 btrfs_free_block_groups(fs_info
);
2756 free_root_pointers(fs_info
, 1);
2757 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
2760 btrfs_stop_workers(&fs_info
->generic_worker
);
2761 btrfs_stop_workers(&fs_info
->readahead_workers
);
2762 btrfs_stop_workers(&fs_info
->fixup_workers
);
2763 btrfs_stop_workers(&fs_info
->delalloc_workers
);
2764 btrfs_stop_workers(&fs_info
->workers
);
2765 btrfs_stop_workers(&fs_info
->endio_workers
);
2766 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
2767 btrfs_stop_workers(&fs_info
->endio_raid56_workers
);
2768 btrfs_stop_workers(&fs_info
->rmw_workers
);
2769 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
2770 btrfs_stop_workers(&fs_info
->endio_write_workers
);
2771 btrfs_stop_workers(&fs_info
->endio_freespace_worker
);
2772 btrfs_stop_workers(&fs_info
->submit_workers
);
2773 btrfs_stop_workers(&fs_info
->delayed_workers
);
2774 btrfs_stop_workers(&fs_info
->caching_workers
);
2775 btrfs_stop_workers(&fs_info
->flush_workers
);
2778 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
2780 iput(fs_info
->btree_inode
);
2781 fail_delalloc_bytes
:
2782 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
2783 fail_dirty_metadata_bytes
:
2784 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
2786 bdi_destroy(&fs_info
->bdi
);
2788 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
2790 btrfs_free_stripe_hash_table(fs_info
);
2791 btrfs_close_devices(fs_info
->fs_devices
);
2795 if (!btrfs_test_opt(tree_root
, RECOVERY
))
2796 goto fail_tree_roots
;
2798 free_root_pointers(fs_info
, 0);
2800 /* don't use the log in recovery mode, it won't be valid */
2801 btrfs_set_super_log_root(disk_super
, 0);
2803 /* we can't trust the free space cache either */
2804 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
2806 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
2807 &num_backups_tried
, &backup_index
);
2809 goto fail_block_groups
;
2810 goto retry_root_backup
;
2813 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
2816 set_buffer_uptodate(bh
);
2818 struct btrfs_device
*device
= (struct btrfs_device
*)
2821 printk_ratelimited_in_rcu(KERN_WARNING
"lost page write due to "
2822 "I/O error on %s\n",
2823 rcu_str_deref(device
->name
));
2824 /* note, we dont' set_buffer_write_io_error because we have
2825 * our own ways of dealing with the IO errors
2827 clear_buffer_uptodate(bh
);
2828 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
2834 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
2836 struct buffer_head
*bh
;
2837 struct buffer_head
*latest
= NULL
;
2838 struct btrfs_super_block
*super
;
2843 /* we would like to check all the supers, but that would make
2844 * a btrfs mount succeed after a mkfs from a different FS.
2845 * So, we need to add a special mount option to scan for
2846 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2848 for (i
= 0; i
< 1; i
++) {
2849 bytenr
= btrfs_sb_offset(i
);
2850 if (bytenr
+ 4096 >= i_size_read(bdev
->bd_inode
))
2852 bh
= __bread(bdev
, bytenr
/ 4096, 4096);
2856 super
= (struct btrfs_super_block
*)bh
->b_data
;
2857 if (btrfs_super_bytenr(super
) != bytenr
||
2858 super
->magic
!= cpu_to_le64(BTRFS_MAGIC
)) {
2863 if (!latest
|| btrfs_super_generation(super
) > transid
) {
2866 transid
= btrfs_super_generation(super
);
2875 * this should be called twice, once with wait == 0 and
2876 * once with wait == 1. When wait == 0 is done, all the buffer heads
2877 * we write are pinned.
2879 * They are released when wait == 1 is done.
2880 * max_mirrors must be the same for both runs, and it indicates how
2881 * many supers on this one device should be written.
2883 * max_mirrors == 0 means to write them all.
2885 static int write_dev_supers(struct btrfs_device
*device
,
2886 struct btrfs_super_block
*sb
,
2887 int do_barriers
, int wait
, int max_mirrors
)
2889 struct buffer_head
*bh
;
2896 if (max_mirrors
== 0)
2897 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
2899 for (i
= 0; i
< max_mirrors
; i
++) {
2900 bytenr
= btrfs_sb_offset(i
);
2901 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= device
->total_bytes
)
2905 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
2906 BTRFS_SUPER_INFO_SIZE
);
2909 if (!buffer_uptodate(bh
))
2912 /* drop our reference */
2915 /* drop the reference from the wait == 0 run */
2919 btrfs_set_super_bytenr(sb
, bytenr
);
2922 crc
= btrfs_csum_data(NULL
, (char *)sb
+
2923 BTRFS_CSUM_SIZE
, crc
,
2924 BTRFS_SUPER_INFO_SIZE
-
2926 btrfs_csum_final(crc
, sb
->csum
);
2929 * one reference for us, and we leave it for the
2932 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
2933 BTRFS_SUPER_INFO_SIZE
);
2934 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
2936 /* one reference for submit_bh */
2939 set_buffer_uptodate(bh
);
2941 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
2942 bh
->b_private
= device
;
2946 * we fua the first super. The others we allow
2949 ret
= btrfsic_submit_bh(WRITE_FUA
, bh
);
2953 return errors
< i
? 0 : -1;
2957 * endio for the write_dev_flush, this will wake anyone waiting
2958 * for the barrier when it is done
2960 static void btrfs_end_empty_barrier(struct bio
*bio
, int err
)
2963 if (err
== -EOPNOTSUPP
)
2964 set_bit(BIO_EOPNOTSUPP
, &bio
->bi_flags
);
2965 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2967 if (bio
->bi_private
)
2968 complete(bio
->bi_private
);
2973 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2974 * sent down. With wait == 1, it waits for the previous flush.
2976 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2979 static int write_dev_flush(struct btrfs_device
*device
, int wait
)
2984 if (device
->nobarriers
)
2988 bio
= device
->flush_bio
;
2992 wait_for_completion(&device
->flush_wait
);
2994 if (bio_flagged(bio
, BIO_EOPNOTSUPP
)) {
2995 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
2996 rcu_str_deref(device
->name
));
2997 device
->nobarriers
= 1;
2998 } else if (!bio_flagged(bio
, BIO_UPTODATE
)) {
3000 btrfs_dev_stat_inc_and_print(device
,
3001 BTRFS_DEV_STAT_FLUSH_ERRS
);
3004 /* drop the reference from the wait == 0 run */
3006 device
->flush_bio
= NULL
;
3012 * one reference for us, and we leave it for the
3015 device
->flush_bio
= NULL
;
3016 bio
= bio_alloc(GFP_NOFS
, 0);
3020 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3021 bio
->bi_bdev
= device
->bdev
;
3022 init_completion(&device
->flush_wait
);
3023 bio
->bi_private
= &device
->flush_wait
;
3024 device
->flush_bio
= bio
;
3027 btrfsic_submit_bio(WRITE_FLUSH
, bio
);
3033 * send an empty flush down to each device in parallel,
3034 * then wait for them
3036 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3038 struct list_head
*head
;
3039 struct btrfs_device
*dev
;
3040 int errors_send
= 0;
3041 int errors_wait
= 0;
3044 /* send down all the barriers */
3045 head
= &info
->fs_devices
->devices
;
3046 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3051 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3054 ret
= write_dev_flush(dev
, 0);
3059 /* wait for all the barriers */
3060 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3065 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3068 ret
= write_dev_flush(dev
, 1);
3072 if (errors_send
> info
->num_tolerated_disk_barrier_failures
||
3073 errors_wait
> info
->num_tolerated_disk_barrier_failures
)
3078 int btrfs_calc_num_tolerated_disk_barrier_failures(
3079 struct btrfs_fs_info
*fs_info
)
3081 struct btrfs_ioctl_space_info space
;
3082 struct btrfs_space_info
*sinfo
;
3083 u64 types
[] = {BTRFS_BLOCK_GROUP_DATA
,
3084 BTRFS_BLOCK_GROUP_SYSTEM
,
3085 BTRFS_BLOCK_GROUP_METADATA
,
3086 BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
};
3090 int num_tolerated_disk_barrier_failures
=
3091 (int)fs_info
->fs_devices
->num_devices
;
3093 for (i
= 0; i
< num_types
; i
++) {
3094 struct btrfs_space_info
*tmp
;
3098 list_for_each_entry_rcu(tmp
, &fs_info
->space_info
, list
) {
3099 if (tmp
->flags
== types
[i
]) {
3109 down_read(&sinfo
->groups_sem
);
3110 for (c
= 0; c
< BTRFS_NR_RAID_TYPES
; c
++) {
3111 if (!list_empty(&sinfo
->block_groups
[c
])) {
3114 btrfs_get_block_group_info(
3115 &sinfo
->block_groups
[c
], &space
);
3116 if (space
.total_bytes
== 0 ||
3117 space
.used_bytes
== 0)
3119 flags
= space
.flags
;
3122 * 0: if dup, single or RAID0 is configured for
3123 * any of metadata, system or data, else
3124 * 1: if RAID5 is configured, or if RAID1 or
3125 * RAID10 is configured and only two mirrors
3127 * 2: if RAID6 is configured, else
3128 * num_mirrors - 1: if RAID1 or RAID10 is
3129 * configured and more than
3130 * 2 mirrors are used.
3132 if (num_tolerated_disk_barrier_failures
> 0 &&
3133 ((flags
& (BTRFS_BLOCK_GROUP_DUP
|
3134 BTRFS_BLOCK_GROUP_RAID0
)) ||
3135 ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
)
3137 num_tolerated_disk_barrier_failures
= 0;
3138 else if (num_tolerated_disk_barrier_failures
> 1) {
3139 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3140 BTRFS_BLOCK_GROUP_RAID5
|
3141 BTRFS_BLOCK_GROUP_RAID10
)) {
3142 num_tolerated_disk_barrier_failures
= 1;
3144 BTRFS_BLOCK_GROUP_RAID5
) {
3145 num_tolerated_disk_barrier_failures
= 2;
3150 up_read(&sinfo
->groups_sem
);
3153 return num_tolerated_disk_barrier_failures
;
3156 int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
3158 struct list_head
*head
;
3159 struct btrfs_device
*dev
;
3160 struct btrfs_super_block
*sb
;
3161 struct btrfs_dev_item
*dev_item
;
3165 int total_errors
= 0;
3168 max_errors
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
3169 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
3170 backup_super_roots(root
->fs_info
);
3172 sb
= root
->fs_info
->super_for_commit
;
3173 dev_item
= &sb
->dev_item
;
3175 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3176 head
= &root
->fs_info
->fs_devices
->devices
;
3179 ret
= barrier_all_devices(root
->fs_info
);
3182 &root
->fs_info
->fs_devices
->device_list_mutex
);
3183 btrfs_error(root
->fs_info
, ret
,
3184 "errors while submitting device barriers.");
3189 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3194 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3197 btrfs_set_stack_device_generation(dev_item
, 0);
3198 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3199 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3200 btrfs_set_stack_device_total_bytes(dev_item
, dev
->total_bytes
);
3201 btrfs_set_stack_device_bytes_used(dev_item
, dev
->bytes_used
);
3202 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3203 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3204 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3205 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3206 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
3208 flags
= btrfs_super_flags(sb
);
3209 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3211 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
3215 if (total_errors
> max_errors
) {
3216 printk(KERN_ERR
"btrfs: %d errors while writing supers\n",
3219 /* This shouldn't happen. FUA is masked off if unsupported */
3224 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3227 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3230 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
3234 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3235 if (total_errors
> max_errors
) {
3236 btrfs_error(root
->fs_info
, -EIO
,
3237 "%d errors while writing supers", total_errors
);
3243 int write_ctree_super(struct btrfs_trans_handle
*trans
,
3244 struct btrfs_root
*root
, int max_mirrors
)
3248 ret
= write_all_supers(root
, max_mirrors
);
3252 void btrfs_free_fs_root(struct btrfs_fs_info
*fs_info
, struct btrfs_root
*root
)
3254 spin_lock(&fs_info
->fs_roots_radix_lock
);
3255 radix_tree_delete(&fs_info
->fs_roots_radix
,
3256 (unsigned long)root
->root_key
.objectid
);
3257 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3259 if (btrfs_root_refs(&root
->root_item
) == 0)
3260 synchronize_srcu(&fs_info
->subvol_srcu
);
3262 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
3263 btrfs_free_log(NULL
, root
);
3264 btrfs_free_log_root_tree(NULL
, fs_info
);
3267 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3268 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3272 static void free_fs_root(struct btrfs_root
*root
)
3274 iput(root
->cache_inode
);
3275 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3277 free_anon_bdev(root
->anon_dev
);
3278 free_extent_buffer(root
->node
);
3279 free_extent_buffer(root
->commit_root
);
3280 kfree(root
->free_ino_ctl
);
3281 kfree(root
->free_ino_pinned
);
3286 static void del_fs_roots(struct btrfs_fs_info
*fs_info
)
3289 struct btrfs_root
*gang
[8];
3292 while (!list_empty(&fs_info
->dead_roots
)) {
3293 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
3294 struct btrfs_root
, root_list
);
3295 list_del(&gang
[0]->root_list
);
3297 if (gang
[0]->in_radix
) {
3298 btrfs_free_fs_root(fs_info
, gang
[0]);
3300 free_extent_buffer(gang
[0]->node
);
3301 free_extent_buffer(gang
[0]->commit_root
);
3307 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3312 for (i
= 0; i
< ret
; i
++)
3313 btrfs_free_fs_root(fs_info
, gang
[i
]);
3317 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3319 u64 root_objectid
= 0;
3320 struct btrfs_root
*gang
[8];
3325 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3326 (void **)gang
, root_objectid
,
3331 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3332 for (i
= 0; i
< ret
; i
++) {
3335 root_objectid
= gang
[i
]->root_key
.objectid
;
3336 err
= btrfs_orphan_cleanup(gang
[i
]);
3345 int btrfs_commit_super(struct btrfs_root
*root
)
3347 struct btrfs_trans_handle
*trans
;
3350 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3351 btrfs_run_delayed_iputs(root
);
3352 btrfs_clean_old_snapshots(root
);
3353 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3355 /* wait until ongoing cleanup work done */
3356 down_write(&root
->fs_info
->cleanup_work_sem
);
3357 up_write(&root
->fs_info
->cleanup_work_sem
);
3359 trans
= btrfs_join_transaction(root
);
3361 return PTR_ERR(trans
);
3362 ret
= btrfs_commit_transaction(trans
, root
);
3365 /* run commit again to drop the original snapshot */
3366 trans
= btrfs_join_transaction(root
);
3368 return PTR_ERR(trans
);
3369 ret
= btrfs_commit_transaction(trans
, root
);
3372 ret
= btrfs_write_and_wait_transaction(NULL
, root
);
3374 btrfs_error(root
->fs_info
, ret
,
3375 "Failed to sync btree inode to disk.");
3379 ret
= write_ctree_super(NULL
, root
, 0);
3383 int close_ctree(struct btrfs_root
*root
)
3385 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3388 fs_info
->closing
= 1;
3391 /* pause restriper - we want to resume on mount */
3392 btrfs_pause_balance(fs_info
);
3394 btrfs_dev_replace_suspend_for_unmount(fs_info
);
3396 btrfs_scrub_cancel(fs_info
);
3398 /* wait for any defraggers to finish */
3399 wait_event(fs_info
->transaction_wait
,
3400 (atomic_read(&fs_info
->defrag_running
) == 0));
3402 /* clear out the rbtree of defraggable inodes */
3403 btrfs_cleanup_defrag_inodes(fs_info
);
3405 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
3406 ret
= btrfs_commit_super(root
);
3408 printk(KERN_ERR
"btrfs: commit super ret %d\n", ret
);
3411 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3412 btrfs_error_commit_super(root
);
3414 btrfs_put_block_group_cache(fs_info
);
3416 kthread_stop(fs_info
->transaction_kthread
);
3417 kthread_stop(fs_info
->cleaner_kthread
);
3419 fs_info
->closing
= 2;
3422 btrfs_free_qgroup_config(root
->fs_info
);
3424 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
3425 printk(KERN_INFO
"btrfs: at unmount delalloc count %lld\n",
3426 percpu_counter_sum(&fs_info
->delalloc_bytes
));
3429 free_extent_buffer(fs_info
->extent_root
->node
);
3430 free_extent_buffer(fs_info
->extent_root
->commit_root
);
3431 free_extent_buffer(fs_info
->tree_root
->node
);
3432 free_extent_buffer(fs_info
->tree_root
->commit_root
);
3433 free_extent_buffer(fs_info
->chunk_root
->node
);
3434 free_extent_buffer(fs_info
->chunk_root
->commit_root
);
3435 free_extent_buffer(fs_info
->dev_root
->node
);
3436 free_extent_buffer(fs_info
->dev_root
->commit_root
);
3437 free_extent_buffer(fs_info
->csum_root
->node
);
3438 free_extent_buffer(fs_info
->csum_root
->commit_root
);
3439 if (fs_info
->quota_root
) {
3440 free_extent_buffer(fs_info
->quota_root
->node
);
3441 free_extent_buffer(fs_info
->quota_root
->commit_root
);
3444 btrfs_free_block_groups(fs_info
);
3446 del_fs_roots(fs_info
);
3448 iput(fs_info
->btree_inode
);
3450 btrfs_stop_workers(&fs_info
->generic_worker
);
3451 btrfs_stop_workers(&fs_info
->fixup_workers
);
3452 btrfs_stop_workers(&fs_info
->delalloc_workers
);
3453 btrfs_stop_workers(&fs_info
->workers
);
3454 btrfs_stop_workers(&fs_info
->endio_workers
);
3455 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
3456 btrfs_stop_workers(&fs_info
->endio_raid56_workers
);
3457 btrfs_stop_workers(&fs_info
->rmw_workers
);
3458 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
3459 btrfs_stop_workers(&fs_info
->endio_write_workers
);
3460 btrfs_stop_workers(&fs_info
->endio_freespace_worker
);
3461 btrfs_stop_workers(&fs_info
->submit_workers
);
3462 btrfs_stop_workers(&fs_info
->delayed_workers
);
3463 btrfs_stop_workers(&fs_info
->caching_workers
);
3464 btrfs_stop_workers(&fs_info
->readahead_workers
);
3465 btrfs_stop_workers(&fs_info
->flush_workers
);
3467 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3468 if (btrfs_test_opt(root
, CHECK_INTEGRITY
))
3469 btrfsic_unmount(root
, fs_info
->fs_devices
);
3472 btrfs_close_devices(fs_info
->fs_devices
);
3473 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3475 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3476 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3477 bdi_destroy(&fs_info
->bdi
);
3478 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3480 btrfs_free_stripe_hash_table(fs_info
);
3485 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
3489 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
3491 ret
= extent_buffer_uptodate(buf
);
3495 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
3496 parent_transid
, atomic
);
3502 int btrfs_set_buffer_uptodate(struct extent_buffer
*buf
)
3504 return set_extent_buffer_uptodate(buf
);
3507 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
3509 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3510 u64 transid
= btrfs_header_generation(buf
);
3513 btrfs_assert_tree_locked(buf
);
3514 if (transid
!= root
->fs_info
->generation
)
3515 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, "
3516 "found %llu running %llu\n",
3517 (unsigned long long)buf
->start
,
3518 (unsigned long long)transid
,
3519 (unsigned long long)root
->fs_info
->generation
);
3520 was_dirty
= set_extent_buffer_dirty(buf
);
3522 __percpu_counter_add(&root
->fs_info
->dirty_metadata_bytes
,
3524 root
->fs_info
->dirty_metadata_batch
);
3527 static void __btrfs_btree_balance_dirty(struct btrfs_root
*root
,
3531 * looks as though older kernels can get into trouble with
3532 * this code, they end up stuck in balance_dirty_pages forever
3536 if (current
->flags
& PF_MEMALLOC
)
3540 btrfs_balance_delayed_items(root
);
3542 ret
= percpu_counter_compare(&root
->fs_info
->dirty_metadata_bytes
,
3543 BTRFS_DIRTY_METADATA_THRESH
);
3545 balance_dirty_pages_ratelimited(
3546 root
->fs_info
->btree_inode
->i_mapping
);
3551 void btrfs_btree_balance_dirty(struct btrfs_root
*root
)
3553 __btrfs_btree_balance_dirty(root
, 1);
3556 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root
*root
)
3558 __btrfs_btree_balance_dirty(root
, 0);
3561 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
3563 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3564 return btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
3567 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
3570 if (btrfs_super_csum_type(fs_info
->super_copy
) >= ARRAY_SIZE(btrfs_csum_sizes
)) {
3571 printk(KERN_ERR
"btrfs: unsupported checksum algorithm\n");
3581 void btrfs_error_commit_super(struct btrfs_root
*root
)
3583 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3584 btrfs_run_delayed_iputs(root
);
3585 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3587 down_write(&root
->fs_info
->cleanup_work_sem
);
3588 up_write(&root
->fs_info
->cleanup_work_sem
);
3590 /* cleanup FS via transaction */
3591 btrfs_cleanup_transaction(root
);
3594 static void btrfs_destroy_ordered_operations(struct btrfs_transaction
*t
,
3595 struct btrfs_root
*root
)
3597 struct btrfs_inode
*btrfs_inode
;
3598 struct list_head splice
;
3600 INIT_LIST_HEAD(&splice
);
3602 mutex_lock(&root
->fs_info
->ordered_operations_mutex
);
3603 spin_lock(&root
->fs_info
->ordered_extent_lock
);
3605 list_splice_init(&t
->ordered_operations
, &splice
);
3606 while (!list_empty(&splice
)) {
3607 btrfs_inode
= list_entry(splice
.next
, struct btrfs_inode
,
3608 ordered_operations
);
3610 list_del_init(&btrfs_inode
->ordered_operations
);
3612 btrfs_invalidate_inodes(btrfs_inode
->root
);
3615 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
3616 mutex_unlock(&root
->fs_info
->ordered_operations_mutex
);
3619 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
3621 struct btrfs_ordered_extent
*ordered
;
3623 spin_lock(&root
->fs_info
->ordered_extent_lock
);
3625 * This will just short circuit the ordered completion stuff which will
3626 * make sure the ordered extent gets properly cleaned up.
3628 list_for_each_entry(ordered
, &root
->fs_info
->ordered_extents
,
3630 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
3631 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
3634 int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
3635 struct btrfs_root
*root
)
3637 struct rb_node
*node
;
3638 struct btrfs_delayed_ref_root
*delayed_refs
;
3639 struct btrfs_delayed_ref_node
*ref
;
3642 delayed_refs
= &trans
->delayed_refs
;
3644 spin_lock(&delayed_refs
->lock
);
3645 if (delayed_refs
->num_entries
== 0) {
3646 spin_unlock(&delayed_refs
->lock
);
3647 printk(KERN_INFO
"delayed_refs has NO entry\n");
3651 while ((node
= rb_first(&delayed_refs
->root
)) != NULL
) {
3652 struct btrfs_delayed_ref_head
*head
= NULL
;
3654 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
3655 atomic_set(&ref
->refs
, 1);
3656 if (btrfs_delayed_ref_is_head(ref
)) {
3658 head
= btrfs_delayed_node_to_head(ref
);
3659 if (!mutex_trylock(&head
->mutex
)) {
3660 atomic_inc(&ref
->refs
);
3661 spin_unlock(&delayed_refs
->lock
);
3663 /* Need to wait for the delayed ref to run */
3664 mutex_lock(&head
->mutex
);
3665 mutex_unlock(&head
->mutex
);
3666 btrfs_put_delayed_ref(ref
);
3668 spin_lock(&delayed_refs
->lock
);
3672 btrfs_free_delayed_extent_op(head
->extent_op
);
3673 delayed_refs
->num_heads
--;
3674 if (list_empty(&head
->cluster
))
3675 delayed_refs
->num_heads_ready
--;
3676 list_del_init(&head
->cluster
);
3680 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
3681 delayed_refs
->num_entries
--;
3683 mutex_unlock(&head
->mutex
);
3684 spin_unlock(&delayed_refs
->lock
);
3685 btrfs_put_delayed_ref(ref
);
3688 spin_lock(&delayed_refs
->lock
);
3691 spin_unlock(&delayed_refs
->lock
);
3696 static void btrfs_evict_pending_snapshots(struct btrfs_transaction
*t
)
3698 struct btrfs_pending_snapshot
*snapshot
;
3699 struct list_head splice
;
3701 INIT_LIST_HEAD(&splice
);
3703 list_splice_init(&t
->pending_snapshots
, &splice
);
3705 while (!list_empty(&splice
)) {
3706 snapshot
= list_entry(splice
.next
,
3707 struct btrfs_pending_snapshot
,
3709 snapshot
->error
= -ECANCELED
;
3710 list_del_init(&snapshot
->list
);
3714 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
3716 struct btrfs_inode
*btrfs_inode
;
3717 struct list_head splice
;
3719 INIT_LIST_HEAD(&splice
);
3721 spin_lock(&root
->fs_info
->delalloc_lock
);
3722 list_splice_init(&root
->fs_info
->delalloc_inodes
, &splice
);
3724 while (!list_empty(&splice
)) {
3725 btrfs_inode
= list_entry(splice
.next
, struct btrfs_inode
,
3728 list_del_init(&btrfs_inode
->delalloc_inodes
);
3729 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
3730 &btrfs_inode
->runtime_flags
);
3732 btrfs_invalidate_inodes(btrfs_inode
->root
);
3735 spin_unlock(&root
->fs_info
->delalloc_lock
);
3738 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
3739 struct extent_io_tree
*dirty_pages
,
3744 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
3745 struct extent_buffer
*eb
;
3749 unsigned long index
;
3752 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
3757 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
3758 while (start
<= end
) {
3759 index
= start
>> PAGE_CACHE_SHIFT
;
3760 start
= (u64
)(index
+ 1) << PAGE_CACHE_SHIFT
;
3761 page
= find_get_page(btree_inode
->i_mapping
, index
);
3764 offset
= page_offset(page
);
3766 spin_lock(&dirty_pages
->buffer_lock
);
3767 eb
= radix_tree_lookup(
3768 &(&BTRFS_I(page
->mapping
->host
)->io_tree
)->buffer
,
3769 offset
>> PAGE_CACHE_SHIFT
);
3770 spin_unlock(&dirty_pages
->buffer_lock
);
3772 ret
= test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
3774 if (PageWriteback(page
))
3775 end_page_writeback(page
);
3778 if (PageDirty(page
)) {
3779 clear_page_dirty_for_io(page
);
3780 spin_lock_irq(&page
->mapping
->tree_lock
);
3781 radix_tree_tag_clear(&page
->mapping
->page_tree
,
3783 PAGECACHE_TAG_DIRTY
);
3784 spin_unlock_irq(&page
->mapping
->tree_lock
);
3788 page_cache_release(page
);
3795 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
3796 struct extent_io_tree
*pinned_extents
)
3798 struct extent_io_tree
*unpin
;
3804 unpin
= pinned_extents
;
3807 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
3808 EXTENT_DIRTY
, NULL
);
3813 if (btrfs_test_opt(root
, DISCARD
))
3814 ret
= btrfs_error_discard_extent(root
, start
,
3818 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
3819 btrfs_error_unpin_extent_range(root
, start
, end
);
3824 if (unpin
== &root
->fs_info
->freed_extents
[0])
3825 unpin
= &root
->fs_info
->freed_extents
[1];
3827 unpin
= &root
->fs_info
->freed_extents
[0];
3835 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
3836 struct btrfs_root
*root
)
3838 btrfs_destroy_delayed_refs(cur_trans
, root
);
3839 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
3840 cur_trans
->dirty_pages
.dirty_bytes
);
3842 /* FIXME: cleanup wait for commit */
3843 cur_trans
->in_commit
= 1;
3844 cur_trans
->blocked
= 1;
3845 wake_up(&root
->fs_info
->transaction_blocked_wait
);
3847 btrfs_evict_pending_snapshots(cur_trans
);
3849 cur_trans
->blocked
= 0;
3850 wake_up(&root
->fs_info
->transaction_wait
);
3852 cur_trans
->commit_done
= 1;
3853 wake_up(&cur_trans
->commit_wait
);
3855 btrfs_destroy_delayed_inodes(root
);
3856 btrfs_assert_delayed_root_empty(root
);
3858 btrfs_destroy_marked_extents(root
, &cur_trans
->dirty_pages
,
3860 btrfs_destroy_pinned_extent(root
,
3861 root
->fs_info
->pinned_extents
);
3864 memset(cur_trans, 0, sizeof(*cur_trans));
3865 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3869 int btrfs_cleanup_transaction(struct btrfs_root
*root
)
3871 struct btrfs_transaction
*t
;
3874 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
3876 spin_lock(&root
->fs_info
->trans_lock
);
3877 list_splice_init(&root
->fs_info
->trans_list
, &list
);
3878 root
->fs_info
->trans_no_join
= 1;
3879 spin_unlock(&root
->fs_info
->trans_lock
);
3881 while (!list_empty(&list
)) {
3882 t
= list_entry(list
.next
, struct btrfs_transaction
, list
);
3884 btrfs_destroy_ordered_operations(t
, root
);
3886 btrfs_destroy_ordered_extents(root
);
3888 btrfs_destroy_delayed_refs(t
, root
);
3890 btrfs_block_rsv_release(root
,
3891 &root
->fs_info
->trans_block_rsv
,
3892 t
->dirty_pages
.dirty_bytes
);
3894 /* FIXME: cleanup wait for commit */
3898 if (waitqueue_active(&root
->fs_info
->transaction_blocked_wait
))
3899 wake_up(&root
->fs_info
->transaction_blocked_wait
);
3901 btrfs_evict_pending_snapshots(t
);
3905 if (waitqueue_active(&root
->fs_info
->transaction_wait
))
3906 wake_up(&root
->fs_info
->transaction_wait
);
3910 if (waitqueue_active(&t
->commit_wait
))
3911 wake_up(&t
->commit_wait
);
3913 btrfs_destroy_delayed_inodes(root
);
3914 btrfs_assert_delayed_root_empty(root
);
3916 btrfs_destroy_delalloc_inodes(root
);
3918 spin_lock(&root
->fs_info
->trans_lock
);
3919 root
->fs_info
->running_transaction
= NULL
;
3920 spin_unlock(&root
->fs_info
->trans_lock
);
3922 btrfs_destroy_marked_extents(root
, &t
->dirty_pages
,
3925 btrfs_destroy_pinned_extent(root
,
3926 root
->fs_info
->pinned_extents
);
3928 atomic_set(&t
->use_count
, 0);
3929 list_del_init(&t
->list
);
3930 memset(t
, 0, sizeof(*t
));
3931 kmem_cache_free(btrfs_transaction_cachep
, t
);
3934 spin_lock(&root
->fs_info
->trans_lock
);
3935 root
->fs_info
->trans_no_join
= 0;
3936 spin_unlock(&root
->fs_info
->trans_lock
);
3937 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
3942 static struct extent_io_ops btree_extent_io_ops
= {
3943 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
3944 .readpage_io_failed_hook
= btree_io_failed_hook
,
3945 .submit_bio_hook
= btree_submit_bio_hook
,
3946 /* note we're sharing with inode.c for the merge bio hook */
3947 .merge_bio_hook
= btrfs_merge_bio_hook
,