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
, 0);
226 if (ret
== -EEXIST
) {
228 em
= lookup_extent_mapping(em_tree
, start
, len
);
235 write_unlock(&em_tree
->lock
);
241 u32
btrfs_csum_data(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(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
);
616 if (found_level
>= BTRFS_MAX_LEVEL
) {
617 btrfs_info(root
->fs_info
, "bad tree block level %d\n",
618 (int)btrfs_header_level(eb
));
623 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
),
626 ret
= csum_tree_block(root
, eb
, 1);
633 * If this is a leaf block and it is corrupt, set the corrupt bit so
634 * that we don't try and read the other copies of this block, just
637 if (found_level
== 0 && check_leaf(root
, eb
)) {
638 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
643 set_extent_buffer_uptodate(eb
);
646 test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
647 btree_readahead_hook(root
, eb
, eb
->start
, ret
);
651 * our io error hook is going to dec the io pages
652 * again, we have to make sure it has something
655 atomic_inc(&eb
->io_pages
);
656 clear_extent_buffer_uptodate(eb
);
658 free_extent_buffer(eb
);
663 static int btree_io_failed_hook(struct page
*page
, int failed_mirror
)
665 struct extent_buffer
*eb
;
666 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
668 eb
= (struct extent_buffer
*)page
->private;
669 set_bit(EXTENT_BUFFER_IOERR
, &eb
->bflags
);
670 eb
->read_mirror
= failed_mirror
;
671 atomic_dec(&eb
->io_pages
);
672 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD
, &eb
->bflags
))
673 btree_readahead_hook(root
, eb
, eb
->start
, -EIO
);
674 return -EIO
; /* we fixed nothing */
677 static void end_workqueue_bio(struct bio
*bio
, int err
)
679 struct end_io_wq
*end_io_wq
= bio
->bi_private
;
680 struct btrfs_fs_info
*fs_info
;
682 fs_info
= end_io_wq
->info
;
683 end_io_wq
->error
= err
;
684 end_io_wq
->work
.func
= end_workqueue_fn
;
685 end_io_wq
->work
.flags
= 0;
687 if (bio
->bi_rw
& REQ_WRITE
) {
688 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_METADATA
)
689 btrfs_queue_worker(&fs_info
->endio_meta_write_workers
,
691 else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_FREE_SPACE
)
692 btrfs_queue_worker(&fs_info
->endio_freespace_worker
,
694 else if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
)
695 btrfs_queue_worker(&fs_info
->endio_raid56_workers
,
698 btrfs_queue_worker(&fs_info
->endio_write_workers
,
701 if (end_io_wq
->metadata
== BTRFS_WQ_ENDIO_RAID56
)
702 btrfs_queue_worker(&fs_info
->endio_raid56_workers
,
704 else if (end_io_wq
->metadata
)
705 btrfs_queue_worker(&fs_info
->endio_meta_workers
,
708 btrfs_queue_worker(&fs_info
->endio_workers
,
714 * For the metadata arg you want
717 * 1 - if normal metadta
718 * 2 - if writing to the free space cache area
719 * 3 - raid parity work
721 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
724 struct end_io_wq
*end_io_wq
;
725 end_io_wq
= kmalloc(sizeof(*end_io_wq
), GFP_NOFS
);
729 end_io_wq
->private = bio
->bi_private
;
730 end_io_wq
->end_io
= bio
->bi_end_io
;
731 end_io_wq
->info
= info
;
732 end_io_wq
->error
= 0;
733 end_io_wq
->bio
= bio
;
734 end_io_wq
->metadata
= metadata
;
736 bio
->bi_private
= end_io_wq
;
737 bio
->bi_end_io
= end_workqueue_bio
;
741 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
743 unsigned long limit
= min_t(unsigned long,
744 info
->workers
.max_workers
,
745 info
->fs_devices
->open_devices
);
749 static void run_one_async_start(struct btrfs_work
*work
)
751 struct async_submit_bio
*async
;
754 async
= container_of(work
, struct async_submit_bio
, work
);
755 ret
= async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
756 async
->mirror_num
, async
->bio_flags
,
762 static void run_one_async_done(struct btrfs_work
*work
)
764 struct btrfs_fs_info
*fs_info
;
765 struct async_submit_bio
*async
;
768 async
= container_of(work
, struct async_submit_bio
, work
);
769 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
771 limit
= btrfs_async_submit_limit(fs_info
);
772 limit
= limit
* 2 / 3;
774 if (atomic_dec_return(&fs_info
->nr_async_submits
) < limit
&&
775 waitqueue_active(&fs_info
->async_submit_wait
))
776 wake_up(&fs_info
->async_submit_wait
);
778 /* If an error occured we just want to clean up the bio and move on */
780 bio_endio(async
->bio
, async
->error
);
784 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
785 async
->mirror_num
, async
->bio_flags
,
789 static void run_one_async_free(struct btrfs_work
*work
)
791 struct async_submit_bio
*async
;
793 async
= container_of(work
, struct async_submit_bio
, work
);
797 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
798 int rw
, struct bio
*bio
, int mirror_num
,
799 unsigned long bio_flags
,
801 extent_submit_bio_hook_t
*submit_bio_start
,
802 extent_submit_bio_hook_t
*submit_bio_done
)
804 struct async_submit_bio
*async
;
806 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
810 async
->inode
= inode
;
813 async
->mirror_num
= mirror_num
;
814 async
->submit_bio_start
= submit_bio_start
;
815 async
->submit_bio_done
= submit_bio_done
;
817 async
->work
.func
= run_one_async_start
;
818 async
->work
.ordered_func
= run_one_async_done
;
819 async
->work
.ordered_free
= run_one_async_free
;
821 async
->work
.flags
= 0;
822 async
->bio_flags
= bio_flags
;
823 async
->bio_offset
= bio_offset
;
827 atomic_inc(&fs_info
->nr_async_submits
);
830 btrfs_set_work_high_prio(&async
->work
);
832 btrfs_queue_worker(&fs_info
->workers
, &async
->work
);
834 while (atomic_read(&fs_info
->async_submit_draining
) &&
835 atomic_read(&fs_info
->nr_async_submits
)) {
836 wait_event(fs_info
->async_submit_wait
,
837 (atomic_read(&fs_info
->nr_async_submits
) == 0));
843 static int btree_csum_one_bio(struct bio
*bio
)
845 struct bio_vec
*bvec
= bio
->bi_io_vec
;
847 struct btrfs_root
*root
;
850 WARN_ON(bio
->bi_vcnt
<= 0);
851 while (bio_index
< bio
->bi_vcnt
) {
852 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
853 ret
= csum_dirty_buffer(root
, bvec
->bv_page
);
862 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
863 struct bio
*bio
, int mirror_num
,
864 unsigned long bio_flags
,
868 * when we're called for a write, we're already in the async
869 * submission context. Just jump into btrfs_map_bio
871 return btree_csum_one_bio(bio
);
874 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
875 int mirror_num
, unsigned long bio_flags
,
881 * when we're called for a write, we're already in the async
882 * submission context. Just jump into btrfs_map_bio
884 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
890 static int check_async_write(struct inode
*inode
, unsigned long bio_flags
)
892 if (bio_flags
& EXTENT_BIO_TREE_LOG
)
901 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
902 int mirror_num
, unsigned long bio_flags
,
905 int async
= check_async_write(inode
, bio_flags
);
908 if (!(rw
& REQ_WRITE
)) {
910 * called for a read, do the setup so that checksum validation
911 * can happen in the async kernel threads
913 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
917 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
920 ret
= btree_csum_one_bio(bio
);
923 ret
= btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
927 * kthread helpers are used to submit writes so that
928 * checksumming can happen in parallel across all CPUs
930 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
931 inode
, rw
, bio
, mirror_num
, 0,
933 __btree_submit_bio_start
,
934 __btree_submit_bio_done
);
944 #ifdef CONFIG_MIGRATION
945 static int btree_migratepage(struct address_space
*mapping
,
946 struct page
*newpage
, struct page
*page
,
947 enum migrate_mode mode
)
950 * we can't safely write a btree page from here,
951 * we haven't done the locking hook
956 * Buffers may be managed in a filesystem specific way.
957 * We must have no buffers or drop them.
959 if (page_has_private(page
) &&
960 !try_to_release_page(page
, GFP_KERNEL
))
962 return migrate_page(mapping
, newpage
, page
, mode
);
967 static int btree_writepages(struct address_space
*mapping
,
968 struct writeback_control
*wbc
)
970 struct extent_io_tree
*tree
;
971 struct btrfs_fs_info
*fs_info
;
974 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
975 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
977 if (wbc
->for_kupdate
)
980 fs_info
= BTRFS_I(mapping
->host
)->root
->fs_info
;
981 /* this is a bit racy, but that's ok */
982 ret
= percpu_counter_compare(&fs_info
->dirty_metadata_bytes
,
983 BTRFS_DIRTY_METADATA_THRESH
);
987 return btree_write_cache_pages(mapping
, wbc
);
990 static int btree_readpage(struct file
*file
, struct page
*page
)
992 struct extent_io_tree
*tree
;
993 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
994 return extent_read_full_page(tree
, page
, btree_get_extent
, 0);
997 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
999 if (PageWriteback(page
) || PageDirty(page
))
1002 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
1003 * slab allocation from alloc_extent_state down the callchain where
1004 * it'd hit a BUG_ON as those flags are not allowed.
1006 gfp_flags
&= ~GFP_SLAB_BUG_MASK
;
1008 return try_release_extent_buffer(page
, gfp_flags
);
1011 static void btree_invalidatepage(struct page
*page
, unsigned long offset
)
1013 struct extent_io_tree
*tree
;
1014 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1015 extent_invalidatepage(tree
, page
, offset
);
1016 btree_releasepage(page
, GFP_NOFS
);
1017 if (PagePrivate(page
)) {
1018 printk(KERN_WARNING
"btrfs warning page private not zero "
1019 "on page %llu\n", (unsigned long long)page_offset(page
));
1020 ClearPagePrivate(page
);
1021 set_page_private(page
, 0);
1022 page_cache_release(page
);
1026 static int btree_set_page_dirty(struct page
*page
)
1029 struct extent_buffer
*eb
;
1031 BUG_ON(!PagePrivate(page
));
1032 eb
= (struct extent_buffer
*)page
->private;
1034 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
));
1035 BUG_ON(!atomic_read(&eb
->refs
));
1036 btrfs_assert_tree_locked(eb
);
1038 return __set_page_dirty_nobuffers(page
);
1041 static const struct address_space_operations btree_aops
= {
1042 .readpage
= btree_readpage
,
1043 .writepages
= btree_writepages
,
1044 .releasepage
= btree_releasepage
,
1045 .invalidatepage
= btree_invalidatepage
,
1046 #ifdef CONFIG_MIGRATION
1047 .migratepage
= btree_migratepage
,
1049 .set_page_dirty
= btree_set_page_dirty
,
1052 int readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
1055 struct extent_buffer
*buf
= NULL
;
1056 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1059 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1062 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
1063 buf
, 0, WAIT_NONE
, btree_get_extent
, 0);
1064 free_extent_buffer(buf
);
1068 int reada_tree_block_flagged(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
1069 int mirror_num
, struct extent_buffer
**eb
)
1071 struct extent_buffer
*buf
= NULL
;
1072 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1073 struct extent_io_tree
*io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
1076 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1080 set_bit(EXTENT_BUFFER_READAHEAD
, &buf
->bflags
);
1082 ret
= read_extent_buffer_pages(io_tree
, buf
, 0, WAIT_PAGE_LOCK
,
1083 btree_get_extent
, mirror_num
);
1085 free_extent_buffer(buf
);
1089 if (test_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
)) {
1090 free_extent_buffer(buf
);
1092 } else if (extent_buffer_uptodate(buf
)) {
1095 free_extent_buffer(buf
);
1100 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_root
*root
,
1101 u64 bytenr
, u32 blocksize
)
1103 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1104 struct extent_buffer
*eb
;
1105 eb
= find_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
1110 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
1111 u64 bytenr
, u32 blocksize
)
1113 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1114 struct extent_buffer
*eb
;
1116 eb
= alloc_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
1122 int btrfs_write_tree_block(struct extent_buffer
*buf
)
1124 return filemap_fdatawrite_range(buf
->pages
[0]->mapping
, buf
->start
,
1125 buf
->start
+ buf
->len
- 1);
1128 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
1130 return filemap_fdatawait_range(buf
->pages
[0]->mapping
,
1131 buf
->start
, buf
->start
+ buf
->len
- 1);
1134 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
1135 u32 blocksize
, u64 parent_transid
)
1137 struct extent_buffer
*buf
= NULL
;
1140 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
1144 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
1149 void clean_tree_block(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1150 struct extent_buffer
*buf
)
1152 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1154 if (btrfs_header_generation(buf
) ==
1155 fs_info
->running_transaction
->transid
) {
1156 btrfs_assert_tree_locked(buf
);
1158 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1159 __percpu_counter_add(&fs_info
->dirty_metadata_bytes
,
1161 fs_info
->dirty_metadata_batch
);
1162 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1163 btrfs_set_lock_blocking(buf
);
1164 clear_extent_buffer_dirty(buf
);
1169 static void __setup_root(u32 nodesize
, u32 leafsize
, u32 sectorsize
,
1170 u32 stripesize
, struct btrfs_root
*root
,
1171 struct btrfs_fs_info
*fs_info
,
1175 root
->commit_root
= NULL
;
1176 root
->sectorsize
= sectorsize
;
1177 root
->nodesize
= nodesize
;
1178 root
->leafsize
= leafsize
;
1179 root
->stripesize
= stripesize
;
1181 root
->track_dirty
= 0;
1183 root
->orphan_item_inserted
= 0;
1184 root
->orphan_cleanup_state
= 0;
1186 root
->objectid
= objectid
;
1187 root
->last_trans
= 0;
1188 root
->highest_objectid
= 0;
1190 root
->inode_tree
= RB_ROOT
;
1191 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1192 root
->block_rsv
= NULL
;
1193 root
->orphan_block_rsv
= NULL
;
1195 INIT_LIST_HEAD(&root
->dirty_list
);
1196 INIT_LIST_HEAD(&root
->root_list
);
1197 INIT_LIST_HEAD(&root
->logged_list
[0]);
1198 INIT_LIST_HEAD(&root
->logged_list
[1]);
1199 spin_lock_init(&root
->orphan_lock
);
1200 spin_lock_init(&root
->inode_lock
);
1201 spin_lock_init(&root
->accounting_lock
);
1202 spin_lock_init(&root
->log_extents_lock
[0]);
1203 spin_lock_init(&root
->log_extents_lock
[1]);
1204 mutex_init(&root
->objectid_mutex
);
1205 mutex_init(&root
->log_mutex
);
1206 init_waitqueue_head(&root
->log_writer_wait
);
1207 init_waitqueue_head(&root
->log_commit_wait
[0]);
1208 init_waitqueue_head(&root
->log_commit_wait
[1]);
1209 atomic_set(&root
->log_commit
[0], 0);
1210 atomic_set(&root
->log_commit
[1], 0);
1211 atomic_set(&root
->log_writers
, 0);
1212 atomic_set(&root
->log_batch
, 0);
1213 atomic_set(&root
->orphan_inodes
, 0);
1214 root
->log_transid
= 0;
1215 root
->last_log_commit
= 0;
1216 extent_io_tree_init(&root
->dirty_log_pages
,
1217 fs_info
->btree_inode
->i_mapping
);
1219 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1220 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1221 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1222 memset(&root
->root_kobj
, 0, sizeof(root
->root_kobj
));
1223 root
->defrag_trans_start
= fs_info
->generation
;
1224 init_completion(&root
->kobj_unregister
);
1225 root
->defrag_running
= 0;
1226 root
->root_key
.objectid
= objectid
;
1229 spin_lock_init(&root
->root_item_lock
);
1232 static int __must_check
find_and_setup_root(struct btrfs_root
*tree_root
,
1233 struct btrfs_fs_info
*fs_info
,
1235 struct btrfs_root
*root
)
1241 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1242 tree_root
->sectorsize
, tree_root
->stripesize
,
1243 root
, fs_info
, objectid
);
1244 ret
= btrfs_find_last_root(tree_root
, objectid
,
1245 &root
->root_item
, &root
->root_key
);
1251 generation
= btrfs_root_generation(&root
->root_item
);
1252 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1253 root
->commit_root
= NULL
;
1254 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1255 blocksize
, generation
);
1256 if (!root
->node
|| !btrfs_buffer_uptodate(root
->node
, generation
, 0)) {
1257 free_extent_buffer(root
->node
);
1261 root
->commit_root
= btrfs_root_node(root
);
1265 static struct btrfs_root
*btrfs_alloc_root(struct btrfs_fs_info
*fs_info
)
1267 struct btrfs_root
*root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1269 root
->fs_info
= fs_info
;
1273 struct btrfs_root
*btrfs_create_tree(struct btrfs_trans_handle
*trans
,
1274 struct btrfs_fs_info
*fs_info
,
1277 struct extent_buffer
*leaf
;
1278 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1279 struct btrfs_root
*root
;
1280 struct btrfs_key key
;
1284 root
= btrfs_alloc_root(fs_info
);
1286 return ERR_PTR(-ENOMEM
);
1288 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1289 tree_root
->sectorsize
, tree_root
->stripesize
,
1290 root
, fs_info
, objectid
);
1291 root
->root_key
.objectid
= objectid
;
1292 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1293 root
->root_key
.offset
= 0;
1295 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
,
1296 0, objectid
, NULL
, 0, 0, 0);
1298 ret
= PTR_ERR(leaf
);
1303 bytenr
= leaf
->start
;
1304 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1305 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1306 btrfs_set_header_generation(leaf
, trans
->transid
);
1307 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1308 btrfs_set_header_owner(leaf
, objectid
);
1311 write_extent_buffer(leaf
, fs_info
->fsid
,
1312 (unsigned long)btrfs_header_fsid(leaf
),
1314 write_extent_buffer(leaf
, fs_info
->chunk_tree_uuid
,
1315 (unsigned long)btrfs_header_chunk_tree_uuid(leaf
),
1317 btrfs_mark_buffer_dirty(leaf
);
1319 root
->commit_root
= btrfs_root_node(root
);
1320 root
->track_dirty
= 1;
1323 root
->root_item
.flags
= 0;
1324 root
->root_item
.byte_limit
= 0;
1325 btrfs_set_root_bytenr(&root
->root_item
, leaf
->start
);
1326 btrfs_set_root_generation(&root
->root_item
, trans
->transid
);
1327 btrfs_set_root_level(&root
->root_item
, 0);
1328 btrfs_set_root_refs(&root
->root_item
, 1);
1329 btrfs_set_root_used(&root
->root_item
, leaf
->len
);
1330 btrfs_set_root_last_snapshot(&root
->root_item
, 0);
1331 btrfs_set_root_dirid(&root
->root_item
, 0);
1332 root
->root_item
.drop_level
= 0;
1334 key
.objectid
= objectid
;
1335 key
.type
= BTRFS_ROOT_ITEM_KEY
;
1337 ret
= btrfs_insert_root(trans
, tree_root
, &key
, &root
->root_item
);
1341 btrfs_tree_unlock(leaf
);
1347 btrfs_tree_unlock(leaf
);
1348 free_extent_buffer(leaf
);
1352 return ERR_PTR(ret
);
1355 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1356 struct btrfs_fs_info
*fs_info
)
1358 struct btrfs_root
*root
;
1359 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1360 struct extent_buffer
*leaf
;
1362 root
= btrfs_alloc_root(fs_info
);
1364 return ERR_PTR(-ENOMEM
);
1366 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1367 tree_root
->sectorsize
, tree_root
->stripesize
,
1368 root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
1370 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1371 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1372 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1374 * log trees do not get reference counted because they go away
1375 * before a real commit is actually done. They do store pointers
1376 * to file data extents, and those reference counts still get
1377 * updated (along with back refs to the log tree).
1381 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
, 0,
1382 BTRFS_TREE_LOG_OBJECTID
, NULL
,
1386 return ERR_CAST(leaf
);
1389 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1390 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1391 btrfs_set_header_generation(leaf
, trans
->transid
);
1392 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1393 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1396 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1397 (unsigned long)btrfs_header_fsid(root
->node
),
1399 btrfs_mark_buffer_dirty(root
->node
);
1400 btrfs_tree_unlock(root
->node
);
1404 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1405 struct btrfs_fs_info
*fs_info
)
1407 struct btrfs_root
*log_root
;
1409 log_root
= alloc_log_tree(trans
, fs_info
);
1410 if (IS_ERR(log_root
))
1411 return PTR_ERR(log_root
);
1412 WARN_ON(fs_info
->log_root_tree
);
1413 fs_info
->log_root_tree
= log_root
;
1417 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1418 struct btrfs_root
*root
)
1420 struct btrfs_root
*log_root
;
1421 struct btrfs_inode_item
*inode_item
;
1423 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1424 if (IS_ERR(log_root
))
1425 return PTR_ERR(log_root
);
1427 log_root
->last_trans
= trans
->transid
;
1428 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1430 inode_item
= &log_root
->root_item
.inode
;
1431 inode_item
->generation
= cpu_to_le64(1);
1432 inode_item
->size
= cpu_to_le64(3);
1433 inode_item
->nlink
= cpu_to_le32(1);
1434 inode_item
->nbytes
= cpu_to_le64(root
->leafsize
);
1435 inode_item
->mode
= cpu_to_le32(S_IFDIR
| 0755);
1437 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1439 WARN_ON(root
->log_root
);
1440 root
->log_root
= log_root
;
1441 root
->log_transid
= 0;
1442 root
->last_log_commit
= 0;
1446 struct btrfs_root
*btrfs_read_fs_root_no_radix(struct btrfs_root
*tree_root
,
1447 struct btrfs_key
*location
)
1449 struct btrfs_root
*root
;
1450 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1451 struct btrfs_path
*path
;
1452 struct extent_buffer
*l
;
1458 root
= btrfs_alloc_root(fs_info
);
1460 return ERR_PTR(-ENOMEM
);
1461 if (location
->offset
== (u64
)-1) {
1462 ret
= find_and_setup_root(tree_root
, fs_info
,
1463 location
->objectid
, root
);
1466 return ERR_PTR(ret
);
1471 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1472 tree_root
->sectorsize
, tree_root
->stripesize
,
1473 root
, fs_info
, location
->objectid
);
1475 path
= btrfs_alloc_path();
1478 return ERR_PTR(-ENOMEM
);
1480 ret
= btrfs_search_slot(NULL
, tree_root
, location
, path
, 0, 0);
1483 slot
= path
->slots
[0];
1484 btrfs_read_root_item(tree_root
, l
, slot
, &root
->root_item
);
1485 memcpy(&root
->root_key
, location
, sizeof(*location
));
1487 btrfs_free_path(path
);
1492 return ERR_PTR(ret
);
1495 generation
= btrfs_root_generation(&root
->root_item
);
1496 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1497 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1498 blocksize
, generation
);
1499 if (!root
->node
|| !extent_buffer_uptodate(root
->node
)) {
1500 ret
= (!root
->node
) ? -ENOMEM
: -EIO
;
1502 free_extent_buffer(root
->node
);
1504 return ERR_PTR(ret
);
1507 root
->commit_root
= btrfs_root_node(root
);
1508 BUG_ON(!root
->node
); /* -ENOMEM */
1510 if (location
->objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1512 btrfs_check_and_init_root_item(&root
->root_item
);
1518 struct btrfs_root
*btrfs_read_fs_root_no_name(struct btrfs_fs_info
*fs_info
,
1519 struct btrfs_key
*location
)
1521 struct btrfs_root
*root
;
1524 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1525 return fs_info
->tree_root
;
1526 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1527 return fs_info
->extent_root
;
1528 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1529 return fs_info
->chunk_root
;
1530 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1531 return fs_info
->dev_root
;
1532 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1533 return fs_info
->csum_root
;
1534 if (location
->objectid
== BTRFS_QUOTA_TREE_OBJECTID
)
1535 return fs_info
->quota_root
? fs_info
->quota_root
:
1538 spin_lock(&fs_info
->fs_roots_radix_lock
);
1539 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1540 (unsigned long)location
->objectid
);
1541 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1545 root
= btrfs_read_fs_root_no_radix(fs_info
->tree_root
, location
);
1549 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1550 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1552 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1557 btrfs_init_free_ino_ctl(root
);
1558 mutex_init(&root
->fs_commit_mutex
);
1559 spin_lock_init(&root
->cache_lock
);
1560 init_waitqueue_head(&root
->cache_wait
);
1562 ret
= get_anon_bdev(&root
->anon_dev
);
1566 if (btrfs_root_refs(&root
->root_item
) == 0) {
1571 ret
= btrfs_find_orphan_item(fs_info
->tree_root
, location
->objectid
);
1575 root
->orphan_item_inserted
= 1;
1577 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
1581 spin_lock(&fs_info
->fs_roots_radix_lock
);
1582 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1583 (unsigned long)root
->root_key
.objectid
,
1588 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1589 radix_tree_preload_end();
1591 if (ret
== -EEXIST
) {
1598 ret
= btrfs_find_dead_roots(fs_info
->tree_root
,
1599 root
->root_key
.objectid
);
1604 return ERR_PTR(ret
);
1607 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1609 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1611 struct btrfs_device
*device
;
1612 struct backing_dev_info
*bdi
;
1615 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1618 bdi
= blk_get_backing_dev_info(device
->bdev
);
1619 if (bdi
&& bdi_congested(bdi
, bdi_bits
)) {
1629 * If this fails, caller must call bdi_destroy() to get rid of the
1632 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1636 bdi
->capabilities
= BDI_CAP_MAP_COPY
;
1637 err
= bdi_setup_and_register(bdi
, "btrfs", BDI_CAP_MAP_COPY
);
1641 bdi
->ra_pages
= default_backing_dev_info
.ra_pages
;
1642 bdi
->congested_fn
= btrfs_congested_fn
;
1643 bdi
->congested_data
= info
;
1648 * called by the kthread helper functions to finally call the bio end_io
1649 * functions. This is where read checksum verification actually happens
1651 static void end_workqueue_fn(struct btrfs_work
*work
)
1654 struct end_io_wq
*end_io_wq
;
1655 struct btrfs_fs_info
*fs_info
;
1658 end_io_wq
= container_of(work
, struct end_io_wq
, work
);
1659 bio
= end_io_wq
->bio
;
1660 fs_info
= end_io_wq
->info
;
1662 error
= end_io_wq
->error
;
1663 bio
->bi_private
= end_io_wq
->private;
1664 bio
->bi_end_io
= end_io_wq
->end_io
;
1666 bio_endio(bio
, error
);
1669 static int cleaner_kthread(void *arg
)
1671 struct btrfs_root
*root
= arg
;
1676 if (!(root
->fs_info
->sb
->s_flags
& MS_RDONLY
) &&
1677 down_read_trylock(&root
->fs_info
->sb
->s_umount
)) {
1678 if (mutex_trylock(&root
->fs_info
->cleaner_mutex
)) {
1679 btrfs_run_delayed_iputs(root
);
1680 again
= btrfs_clean_one_deleted_snapshot(root
);
1681 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1683 btrfs_run_defrag_inodes(root
->fs_info
);
1684 up_read(&root
->fs_info
->sb
->s_umount
);
1687 if (!try_to_freeze() && !again
) {
1688 set_current_state(TASK_INTERRUPTIBLE
);
1689 if (!kthread_should_stop())
1691 __set_current_state(TASK_RUNNING
);
1693 } while (!kthread_should_stop());
1697 static int transaction_kthread(void *arg
)
1699 struct btrfs_root
*root
= arg
;
1700 struct btrfs_trans_handle
*trans
;
1701 struct btrfs_transaction
*cur
;
1704 unsigned long delay
;
1708 cannot_commit
= false;
1710 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1712 spin_lock(&root
->fs_info
->trans_lock
);
1713 cur
= root
->fs_info
->running_transaction
;
1715 spin_unlock(&root
->fs_info
->trans_lock
);
1719 now
= get_seconds();
1720 if (!cur
->blocked
&&
1721 (now
< cur
->start_time
|| now
- cur
->start_time
< 30)) {
1722 spin_unlock(&root
->fs_info
->trans_lock
);
1726 transid
= cur
->transid
;
1727 spin_unlock(&root
->fs_info
->trans_lock
);
1729 /* If the file system is aborted, this will always fail. */
1730 trans
= btrfs_attach_transaction(root
);
1731 if (IS_ERR(trans
)) {
1732 if (PTR_ERR(trans
) != -ENOENT
)
1733 cannot_commit
= true;
1736 if (transid
== trans
->transid
) {
1737 btrfs_commit_transaction(trans
, root
);
1739 btrfs_end_transaction(trans
, root
);
1742 wake_up_process(root
->fs_info
->cleaner_kthread
);
1743 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1745 if (!try_to_freeze()) {
1746 set_current_state(TASK_INTERRUPTIBLE
);
1747 if (!kthread_should_stop() &&
1748 (!btrfs_transaction_blocked(root
->fs_info
) ||
1750 schedule_timeout(delay
);
1751 __set_current_state(TASK_RUNNING
);
1753 } while (!kthread_should_stop());
1758 * this will find the highest generation in the array of
1759 * root backups. The index of the highest array is returned,
1760 * or -1 if we can't find anything.
1762 * We check to make sure the array is valid by comparing the
1763 * generation of the latest root in the array with the generation
1764 * in the super block. If they don't match we pitch it.
1766 static int find_newest_super_backup(struct btrfs_fs_info
*info
, u64 newest_gen
)
1769 int newest_index
= -1;
1770 struct btrfs_root_backup
*root_backup
;
1773 for (i
= 0; i
< BTRFS_NUM_BACKUP_ROOTS
; i
++) {
1774 root_backup
= info
->super_copy
->super_roots
+ i
;
1775 cur
= btrfs_backup_tree_root_gen(root_backup
);
1776 if (cur
== newest_gen
)
1780 /* check to see if we actually wrapped around */
1781 if (newest_index
== BTRFS_NUM_BACKUP_ROOTS
- 1) {
1782 root_backup
= info
->super_copy
->super_roots
;
1783 cur
= btrfs_backup_tree_root_gen(root_backup
);
1784 if (cur
== newest_gen
)
1787 return newest_index
;
1792 * find the oldest backup so we know where to store new entries
1793 * in the backup array. This will set the backup_root_index
1794 * field in the fs_info struct
1796 static void find_oldest_super_backup(struct btrfs_fs_info
*info
,
1799 int newest_index
= -1;
1801 newest_index
= find_newest_super_backup(info
, newest_gen
);
1802 /* if there was garbage in there, just move along */
1803 if (newest_index
== -1) {
1804 info
->backup_root_index
= 0;
1806 info
->backup_root_index
= (newest_index
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1811 * copy all the root pointers into the super backup array.
1812 * this will bump the backup pointer by one when it is
1815 static void backup_super_roots(struct btrfs_fs_info
*info
)
1818 struct btrfs_root_backup
*root_backup
;
1821 next_backup
= info
->backup_root_index
;
1822 last_backup
= (next_backup
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1823 BTRFS_NUM_BACKUP_ROOTS
;
1826 * just overwrite the last backup if we're at the same generation
1827 * this happens only at umount
1829 root_backup
= info
->super_for_commit
->super_roots
+ last_backup
;
1830 if (btrfs_backup_tree_root_gen(root_backup
) ==
1831 btrfs_header_generation(info
->tree_root
->node
))
1832 next_backup
= last_backup
;
1834 root_backup
= info
->super_for_commit
->super_roots
+ next_backup
;
1837 * make sure all of our padding and empty slots get zero filled
1838 * regardless of which ones we use today
1840 memset(root_backup
, 0, sizeof(*root_backup
));
1842 info
->backup_root_index
= (next_backup
+ 1) % BTRFS_NUM_BACKUP_ROOTS
;
1844 btrfs_set_backup_tree_root(root_backup
, info
->tree_root
->node
->start
);
1845 btrfs_set_backup_tree_root_gen(root_backup
,
1846 btrfs_header_generation(info
->tree_root
->node
));
1848 btrfs_set_backup_tree_root_level(root_backup
,
1849 btrfs_header_level(info
->tree_root
->node
));
1851 btrfs_set_backup_chunk_root(root_backup
, info
->chunk_root
->node
->start
);
1852 btrfs_set_backup_chunk_root_gen(root_backup
,
1853 btrfs_header_generation(info
->chunk_root
->node
));
1854 btrfs_set_backup_chunk_root_level(root_backup
,
1855 btrfs_header_level(info
->chunk_root
->node
));
1857 btrfs_set_backup_extent_root(root_backup
, info
->extent_root
->node
->start
);
1858 btrfs_set_backup_extent_root_gen(root_backup
,
1859 btrfs_header_generation(info
->extent_root
->node
));
1860 btrfs_set_backup_extent_root_level(root_backup
,
1861 btrfs_header_level(info
->extent_root
->node
));
1864 * we might commit during log recovery, which happens before we set
1865 * the fs_root. Make sure it is valid before we fill it in.
1867 if (info
->fs_root
&& info
->fs_root
->node
) {
1868 btrfs_set_backup_fs_root(root_backup
,
1869 info
->fs_root
->node
->start
);
1870 btrfs_set_backup_fs_root_gen(root_backup
,
1871 btrfs_header_generation(info
->fs_root
->node
));
1872 btrfs_set_backup_fs_root_level(root_backup
,
1873 btrfs_header_level(info
->fs_root
->node
));
1876 btrfs_set_backup_dev_root(root_backup
, info
->dev_root
->node
->start
);
1877 btrfs_set_backup_dev_root_gen(root_backup
,
1878 btrfs_header_generation(info
->dev_root
->node
));
1879 btrfs_set_backup_dev_root_level(root_backup
,
1880 btrfs_header_level(info
->dev_root
->node
));
1882 btrfs_set_backup_csum_root(root_backup
, info
->csum_root
->node
->start
);
1883 btrfs_set_backup_csum_root_gen(root_backup
,
1884 btrfs_header_generation(info
->csum_root
->node
));
1885 btrfs_set_backup_csum_root_level(root_backup
,
1886 btrfs_header_level(info
->csum_root
->node
));
1888 btrfs_set_backup_total_bytes(root_backup
,
1889 btrfs_super_total_bytes(info
->super_copy
));
1890 btrfs_set_backup_bytes_used(root_backup
,
1891 btrfs_super_bytes_used(info
->super_copy
));
1892 btrfs_set_backup_num_devices(root_backup
,
1893 btrfs_super_num_devices(info
->super_copy
));
1896 * if we don't copy this out to the super_copy, it won't get remembered
1897 * for the next commit
1899 memcpy(&info
->super_copy
->super_roots
,
1900 &info
->super_for_commit
->super_roots
,
1901 sizeof(*root_backup
) * BTRFS_NUM_BACKUP_ROOTS
);
1905 * this copies info out of the root backup array and back into
1906 * the in-memory super block. It is meant to help iterate through
1907 * the array, so you send it the number of backups you've already
1908 * tried and the last backup index you used.
1910 * this returns -1 when it has tried all the backups
1912 static noinline
int next_root_backup(struct btrfs_fs_info
*info
,
1913 struct btrfs_super_block
*super
,
1914 int *num_backups_tried
, int *backup_index
)
1916 struct btrfs_root_backup
*root_backup
;
1917 int newest
= *backup_index
;
1919 if (*num_backups_tried
== 0) {
1920 u64 gen
= btrfs_super_generation(super
);
1922 newest
= find_newest_super_backup(info
, gen
);
1926 *backup_index
= newest
;
1927 *num_backups_tried
= 1;
1928 } else if (*num_backups_tried
== BTRFS_NUM_BACKUP_ROOTS
) {
1929 /* we've tried all the backups, all done */
1932 /* jump to the next oldest backup */
1933 newest
= (*backup_index
+ BTRFS_NUM_BACKUP_ROOTS
- 1) %
1934 BTRFS_NUM_BACKUP_ROOTS
;
1935 *backup_index
= newest
;
1936 *num_backups_tried
+= 1;
1938 root_backup
= super
->super_roots
+ newest
;
1940 btrfs_set_super_generation(super
,
1941 btrfs_backup_tree_root_gen(root_backup
));
1942 btrfs_set_super_root(super
, btrfs_backup_tree_root(root_backup
));
1943 btrfs_set_super_root_level(super
,
1944 btrfs_backup_tree_root_level(root_backup
));
1945 btrfs_set_super_bytes_used(super
, btrfs_backup_bytes_used(root_backup
));
1948 * fixme: the total bytes and num_devices need to match or we should
1951 btrfs_set_super_total_bytes(super
, btrfs_backup_total_bytes(root_backup
));
1952 btrfs_set_super_num_devices(super
, btrfs_backup_num_devices(root_backup
));
1956 /* helper to cleanup workers */
1957 static void btrfs_stop_all_workers(struct btrfs_fs_info
*fs_info
)
1959 btrfs_stop_workers(&fs_info
->generic_worker
);
1960 btrfs_stop_workers(&fs_info
->fixup_workers
);
1961 btrfs_stop_workers(&fs_info
->delalloc_workers
);
1962 btrfs_stop_workers(&fs_info
->workers
);
1963 btrfs_stop_workers(&fs_info
->endio_workers
);
1964 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
1965 btrfs_stop_workers(&fs_info
->endio_raid56_workers
);
1966 btrfs_stop_workers(&fs_info
->rmw_workers
);
1967 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
1968 btrfs_stop_workers(&fs_info
->endio_write_workers
);
1969 btrfs_stop_workers(&fs_info
->endio_freespace_worker
);
1970 btrfs_stop_workers(&fs_info
->submit_workers
);
1971 btrfs_stop_workers(&fs_info
->delayed_workers
);
1972 btrfs_stop_workers(&fs_info
->caching_workers
);
1973 btrfs_stop_workers(&fs_info
->readahead_workers
);
1974 btrfs_stop_workers(&fs_info
->flush_workers
);
1977 /* helper to cleanup tree roots */
1978 static void free_root_pointers(struct btrfs_fs_info
*info
, int chunk_root
)
1980 free_extent_buffer(info
->tree_root
->node
);
1981 free_extent_buffer(info
->tree_root
->commit_root
);
1982 free_extent_buffer(info
->dev_root
->node
);
1983 free_extent_buffer(info
->dev_root
->commit_root
);
1984 free_extent_buffer(info
->extent_root
->node
);
1985 free_extent_buffer(info
->extent_root
->commit_root
);
1986 free_extent_buffer(info
->csum_root
->node
);
1987 free_extent_buffer(info
->csum_root
->commit_root
);
1988 if (info
->quota_root
) {
1989 free_extent_buffer(info
->quota_root
->node
);
1990 free_extent_buffer(info
->quota_root
->commit_root
);
1993 info
->tree_root
->node
= NULL
;
1994 info
->tree_root
->commit_root
= NULL
;
1995 info
->dev_root
->node
= NULL
;
1996 info
->dev_root
->commit_root
= NULL
;
1997 info
->extent_root
->node
= NULL
;
1998 info
->extent_root
->commit_root
= NULL
;
1999 info
->csum_root
->node
= NULL
;
2000 info
->csum_root
->commit_root
= NULL
;
2001 if (info
->quota_root
) {
2002 info
->quota_root
->node
= NULL
;
2003 info
->quota_root
->commit_root
= NULL
;
2007 free_extent_buffer(info
->chunk_root
->node
);
2008 free_extent_buffer(info
->chunk_root
->commit_root
);
2009 info
->chunk_root
->node
= NULL
;
2010 info
->chunk_root
->commit_root
= NULL
;
2014 static void del_fs_roots(struct btrfs_fs_info
*fs_info
)
2017 struct btrfs_root
*gang
[8];
2020 while (!list_empty(&fs_info
->dead_roots
)) {
2021 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2022 struct btrfs_root
, root_list
);
2023 list_del(&gang
[0]->root_list
);
2025 if (gang
[0]->in_radix
) {
2026 btrfs_free_fs_root(fs_info
, gang
[0]);
2028 free_extent_buffer(gang
[0]->node
);
2029 free_extent_buffer(gang
[0]->commit_root
);
2035 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2040 for (i
= 0; i
< ret
; i
++)
2041 btrfs_free_fs_root(fs_info
, gang
[i
]);
2045 int open_ctree(struct super_block
*sb
,
2046 struct btrfs_fs_devices
*fs_devices
,
2056 struct btrfs_key location
;
2057 struct buffer_head
*bh
;
2058 struct btrfs_super_block
*disk_super
;
2059 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
2060 struct btrfs_root
*tree_root
;
2061 struct btrfs_root
*extent_root
;
2062 struct btrfs_root
*csum_root
;
2063 struct btrfs_root
*chunk_root
;
2064 struct btrfs_root
*dev_root
;
2065 struct btrfs_root
*quota_root
;
2066 struct btrfs_root
*log_tree_root
;
2069 int num_backups_tried
= 0;
2070 int backup_index
= 0;
2072 tree_root
= fs_info
->tree_root
= btrfs_alloc_root(fs_info
);
2073 extent_root
= fs_info
->extent_root
= btrfs_alloc_root(fs_info
);
2074 csum_root
= fs_info
->csum_root
= btrfs_alloc_root(fs_info
);
2075 chunk_root
= fs_info
->chunk_root
= btrfs_alloc_root(fs_info
);
2076 dev_root
= fs_info
->dev_root
= btrfs_alloc_root(fs_info
);
2077 quota_root
= fs_info
->quota_root
= btrfs_alloc_root(fs_info
);
2079 if (!tree_root
|| !extent_root
|| !csum_root
||
2080 !chunk_root
|| !dev_root
|| !quota_root
) {
2085 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
2091 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
2097 ret
= percpu_counter_init(&fs_info
->dirty_metadata_bytes
, 0);
2102 fs_info
->dirty_metadata_batch
= PAGE_CACHE_SIZE
*
2103 (1 + ilog2(nr_cpu_ids
));
2105 ret
= percpu_counter_init(&fs_info
->delalloc_bytes
, 0);
2108 goto fail_dirty_metadata_bytes
;
2111 fs_info
->btree_inode
= new_inode(sb
);
2112 if (!fs_info
->btree_inode
) {
2114 goto fail_delalloc_bytes
;
2117 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
2119 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
2120 INIT_LIST_HEAD(&fs_info
->trans_list
);
2121 INIT_LIST_HEAD(&fs_info
->dead_roots
);
2122 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
2123 INIT_LIST_HEAD(&fs_info
->delalloc_inodes
);
2124 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
2125 spin_lock_init(&fs_info
->delalloc_lock
);
2126 spin_lock_init(&fs_info
->trans_lock
);
2127 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
2128 spin_lock_init(&fs_info
->delayed_iput_lock
);
2129 spin_lock_init(&fs_info
->defrag_inodes_lock
);
2130 spin_lock_init(&fs_info
->free_chunk_lock
);
2131 spin_lock_init(&fs_info
->tree_mod_seq_lock
);
2132 spin_lock_init(&fs_info
->super_lock
);
2133 rwlock_init(&fs_info
->tree_mod_log_lock
);
2134 mutex_init(&fs_info
->reloc_mutex
);
2135 seqlock_init(&fs_info
->profiles_lock
);
2137 init_completion(&fs_info
->kobj_unregister
);
2138 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
2139 INIT_LIST_HEAD(&fs_info
->space_info
);
2140 INIT_LIST_HEAD(&fs_info
->tree_mod_seq_list
);
2141 btrfs_mapping_init(&fs_info
->mapping_tree
);
2142 btrfs_init_block_rsv(&fs_info
->global_block_rsv
,
2143 BTRFS_BLOCK_RSV_GLOBAL
);
2144 btrfs_init_block_rsv(&fs_info
->delalloc_block_rsv
,
2145 BTRFS_BLOCK_RSV_DELALLOC
);
2146 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
, BTRFS_BLOCK_RSV_TRANS
);
2147 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
, BTRFS_BLOCK_RSV_CHUNK
);
2148 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
, BTRFS_BLOCK_RSV_EMPTY
);
2149 btrfs_init_block_rsv(&fs_info
->delayed_block_rsv
,
2150 BTRFS_BLOCK_RSV_DELOPS
);
2151 atomic_set(&fs_info
->nr_async_submits
, 0);
2152 atomic_set(&fs_info
->async_delalloc_pages
, 0);
2153 atomic_set(&fs_info
->async_submit_draining
, 0);
2154 atomic_set(&fs_info
->nr_async_bios
, 0);
2155 atomic_set(&fs_info
->defrag_running
, 0);
2156 atomic_set(&fs_info
->tree_mod_seq
, 0);
2158 fs_info
->max_inline
= 8192 * 1024;
2159 fs_info
->metadata_ratio
= 0;
2160 fs_info
->defrag_inodes
= RB_ROOT
;
2161 fs_info
->trans_no_join
= 0;
2162 fs_info
->free_chunk_space
= 0;
2163 fs_info
->tree_mod_log
= RB_ROOT
;
2165 /* readahead state */
2166 INIT_RADIX_TREE(&fs_info
->reada_tree
, GFP_NOFS
& ~__GFP_WAIT
);
2167 spin_lock_init(&fs_info
->reada_lock
);
2169 fs_info
->thread_pool_size
= min_t(unsigned long,
2170 num_online_cpus() + 2, 8);
2172 INIT_LIST_HEAD(&fs_info
->ordered_extents
);
2173 spin_lock_init(&fs_info
->ordered_extent_lock
);
2174 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
2176 if (!fs_info
->delayed_root
) {
2180 btrfs_init_delayed_root(fs_info
->delayed_root
);
2182 mutex_init(&fs_info
->scrub_lock
);
2183 atomic_set(&fs_info
->scrubs_running
, 0);
2184 atomic_set(&fs_info
->scrub_pause_req
, 0);
2185 atomic_set(&fs_info
->scrubs_paused
, 0);
2186 atomic_set(&fs_info
->scrub_cancel_req
, 0);
2187 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
2188 init_rwsem(&fs_info
->scrub_super_lock
);
2189 fs_info
->scrub_workers_refcnt
= 0;
2190 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2191 fs_info
->check_integrity_print_mask
= 0;
2194 spin_lock_init(&fs_info
->balance_lock
);
2195 mutex_init(&fs_info
->balance_mutex
);
2196 atomic_set(&fs_info
->balance_running
, 0);
2197 atomic_set(&fs_info
->balance_pause_req
, 0);
2198 atomic_set(&fs_info
->balance_cancel_req
, 0);
2199 fs_info
->balance_ctl
= NULL
;
2200 init_waitqueue_head(&fs_info
->balance_wait_q
);
2202 sb
->s_blocksize
= 4096;
2203 sb
->s_blocksize_bits
= blksize_bits(4096);
2204 sb
->s_bdi
= &fs_info
->bdi
;
2206 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
2207 set_nlink(fs_info
->btree_inode
, 1);
2209 * we set the i_size on the btree inode to the max possible int.
2210 * the real end of the address space is determined by all of
2211 * the devices in the system
2213 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
2214 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
2215 fs_info
->btree_inode
->i_mapping
->backing_dev_info
= &fs_info
->bdi
;
2217 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
2218 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
2219 fs_info
->btree_inode
->i_mapping
);
2220 BTRFS_I(fs_info
->btree_inode
)->io_tree
.track_uptodate
= 0;
2221 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
);
2223 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
2225 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
2226 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
2227 sizeof(struct btrfs_key
));
2228 set_bit(BTRFS_INODE_DUMMY
,
2229 &BTRFS_I(fs_info
->btree_inode
)->runtime_flags
);
2230 insert_inode_hash(fs_info
->btree_inode
);
2232 spin_lock_init(&fs_info
->block_group_cache_lock
);
2233 fs_info
->block_group_cache_tree
= RB_ROOT
;
2234 fs_info
->first_logical_byte
= (u64
)-1;
2236 extent_io_tree_init(&fs_info
->freed_extents
[0],
2237 fs_info
->btree_inode
->i_mapping
);
2238 extent_io_tree_init(&fs_info
->freed_extents
[1],
2239 fs_info
->btree_inode
->i_mapping
);
2240 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
2241 fs_info
->do_barriers
= 1;
2244 mutex_init(&fs_info
->ordered_operations_mutex
);
2245 mutex_init(&fs_info
->tree_log_mutex
);
2246 mutex_init(&fs_info
->chunk_mutex
);
2247 mutex_init(&fs_info
->transaction_kthread_mutex
);
2248 mutex_init(&fs_info
->cleaner_mutex
);
2249 mutex_init(&fs_info
->volume_mutex
);
2250 init_rwsem(&fs_info
->extent_commit_sem
);
2251 init_rwsem(&fs_info
->cleanup_work_sem
);
2252 init_rwsem(&fs_info
->subvol_sem
);
2253 fs_info
->dev_replace
.lock_owner
= 0;
2254 atomic_set(&fs_info
->dev_replace
.nesting_level
, 0);
2255 mutex_init(&fs_info
->dev_replace
.lock_finishing_cancel_unmount
);
2256 mutex_init(&fs_info
->dev_replace
.lock_management_lock
);
2257 mutex_init(&fs_info
->dev_replace
.lock
);
2259 spin_lock_init(&fs_info
->qgroup_lock
);
2260 mutex_init(&fs_info
->qgroup_ioctl_lock
);
2261 fs_info
->qgroup_tree
= RB_ROOT
;
2262 INIT_LIST_HEAD(&fs_info
->dirty_qgroups
);
2263 fs_info
->qgroup_seq
= 1;
2264 fs_info
->quota_enabled
= 0;
2265 fs_info
->pending_quota_state
= 0;
2267 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
2268 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
2270 init_waitqueue_head(&fs_info
->transaction_throttle
);
2271 init_waitqueue_head(&fs_info
->transaction_wait
);
2272 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
2273 init_waitqueue_head(&fs_info
->async_submit_wait
);
2275 ret
= btrfs_alloc_stripe_hash_table(fs_info
);
2281 __setup_root(4096, 4096, 4096, 4096, tree_root
,
2282 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
2284 invalidate_bdev(fs_devices
->latest_bdev
);
2285 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
2291 memcpy(fs_info
->super_copy
, bh
->b_data
, sizeof(*fs_info
->super_copy
));
2292 memcpy(fs_info
->super_for_commit
, fs_info
->super_copy
,
2293 sizeof(*fs_info
->super_for_commit
));
2296 memcpy(fs_info
->fsid
, fs_info
->super_copy
->fsid
, BTRFS_FSID_SIZE
);
2298 disk_super
= fs_info
->super_copy
;
2299 if (!btrfs_super_root(disk_super
))
2302 /* check FS state, whether FS is broken. */
2303 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_ERROR
)
2304 set_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
);
2306 ret
= btrfs_check_super_valid(fs_info
, sb
->s_flags
& MS_RDONLY
);
2308 printk(KERN_ERR
"btrfs: superblock contains fatal errors\n");
2314 * run through our array of backup supers and setup
2315 * our ring pointer to the oldest one
2317 generation
= btrfs_super_generation(disk_super
);
2318 find_oldest_super_backup(fs_info
, generation
);
2321 * In the long term, we'll store the compression type in the super
2322 * block, and it'll be used for per file compression control.
2324 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
2326 ret
= btrfs_parse_options(tree_root
, options
);
2332 features
= btrfs_super_incompat_flags(disk_super
) &
2333 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
2335 printk(KERN_ERR
"BTRFS: couldn't mount because of "
2336 "unsupported optional features (%Lx).\n",
2337 (unsigned long long)features
);
2342 if (btrfs_super_leafsize(disk_super
) !=
2343 btrfs_super_nodesize(disk_super
)) {
2344 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2345 "blocksizes don't match. node %d leaf %d\n",
2346 btrfs_super_nodesize(disk_super
),
2347 btrfs_super_leafsize(disk_super
));
2351 if (btrfs_super_leafsize(disk_super
) > BTRFS_MAX_METADATA_BLOCKSIZE
) {
2352 printk(KERN_ERR
"BTRFS: couldn't mount because metadata "
2353 "blocksize (%d) was too large\n",
2354 btrfs_super_leafsize(disk_super
));
2359 features
= btrfs_super_incompat_flags(disk_super
);
2360 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
2361 if (tree_root
->fs_info
->compress_type
== BTRFS_COMPRESS_LZO
)
2362 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
2364 if (features
& BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA
)
2365 printk(KERN_ERR
"btrfs: has skinny extents\n");
2368 * flag our filesystem as having big metadata blocks if
2369 * they are bigger than the page size
2371 if (btrfs_super_leafsize(disk_super
) > PAGE_CACHE_SIZE
) {
2372 if (!(features
& BTRFS_FEATURE_INCOMPAT_BIG_METADATA
))
2373 printk(KERN_INFO
"btrfs flagging fs with big metadata feature\n");
2374 features
|= BTRFS_FEATURE_INCOMPAT_BIG_METADATA
;
2377 nodesize
= btrfs_super_nodesize(disk_super
);
2378 leafsize
= btrfs_super_leafsize(disk_super
);
2379 sectorsize
= btrfs_super_sectorsize(disk_super
);
2380 stripesize
= btrfs_super_stripesize(disk_super
);
2381 fs_info
->dirty_metadata_batch
= leafsize
* (1 + ilog2(nr_cpu_ids
));
2382 fs_info
->delalloc_batch
= sectorsize
* 512 * (1 + ilog2(nr_cpu_ids
));
2385 * mixed block groups end up with duplicate but slightly offset
2386 * extent buffers for the same range. It leads to corruptions
2388 if ((features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2389 (sectorsize
!= leafsize
)) {
2390 printk(KERN_WARNING
"btrfs: unequal leaf/node/sector sizes "
2391 "are not allowed for mixed block groups on %s\n",
2397 * Needn't use the lock because there is no other task which will
2400 btrfs_set_super_incompat_flags(disk_super
, features
);
2402 features
= btrfs_super_compat_ro_flags(disk_super
) &
2403 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
2404 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
2405 printk(KERN_ERR
"BTRFS: couldn't mount RDWR because of "
2406 "unsupported option features (%Lx).\n",
2407 (unsigned long long)features
);
2412 btrfs_init_workers(&fs_info
->generic_worker
,
2413 "genwork", 1, NULL
);
2415 btrfs_init_workers(&fs_info
->workers
, "worker",
2416 fs_info
->thread_pool_size
,
2417 &fs_info
->generic_worker
);
2419 btrfs_init_workers(&fs_info
->delalloc_workers
, "delalloc",
2420 fs_info
->thread_pool_size
,
2421 &fs_info
->generic_worker
);
2423 btrfs_init_workers(&fs_info
->flush_workers
, "flush_delalloc",
2424 fs_info
->thread_pool_size
,
2425 &fs_info
->generic_worker
);
2427 btrfs_init_workers(&fs_info
->submit_workers
, "submit",
2428 min_t(u64
, fs_devices
->num_devices
,
2429 fs_info
->thread_pool_size
),
2430 &fs_info
->generic_worker
);
2432 btrfs_init_workers(&fs_info
->caching_workers
, "cache",
2433 2, &fs_info
->generic_worker
);
2435 /* a higher idle thresh on the submit workers makes it much more
2436 * likely that bios will be send down in a sane order to the
2439 fs_info
->submit_workers
.idle_thresh
= 64;
2441 fs_info
->workers
.idle_thresh
= 16;
2442 fs_info
->workers
.ordered
= 1;
2444 fs_info
->delalloc_workers
.idle_thresh
= 2;
2445 fs_info
->delalloc_workers
.ordered
= 1;
2447 btrfs_init_workers(&fs_info
->fixup_workers
, "fixup", 1,
2448 &fs_info
->generic_worker
);
2449 btrfs_init_workers(&fs_info
->endio_workers
, "endio",
2450 fs_info
->thread_pool_size
,
2451 &fs_info
->generic_worker
);
2452 btrfs_init_workers(&fs_info
->endio_meta_workers
, "endio-meta",
2453 fs_info
->thread_pool_size
,
2454 &fs_info
->generic_worker
);
2455 btrfs_init_workers(&fs_info
->endio_meta_write_workers
,
2456 "endio-meta-write", fs_info
->thread_pool_size
,
2457 &fs_info
->generic_worker
);
2458 btrfs_init_workers(&fs_info
->endio_raid56_workers
,
2459 "endio-raid56", fs_info
->thread_pool_size
,
2460 &fs_info
->generic_worker
);
2461 btrfs_init_workers(&fs_info
->rmw_workers
,
2462 "rmw", fs_info
->thread_pool_size
,
2463 &fs_info
->generic_worker
);
2464 btrfs_init_workers(&fs_info
->endio_write_workers
, "endio-write",
2465 fs_info
->thread_pool_size
,
2466 &fs_info
->generic_worker
);
2467 btrfs_init_workers(&fs_info
->endio_freespace_worker
, "freespace-write",
2468 1, &fs_info
->generic_worker
);
2469 btrfs_init_workers(&fs_info
->delayed_workers
, "delayed-meta",
2470 fs_info
->thread_pool_size
,
2471 &fs_info
->generic_worker
);
2472 btrfs_init_workers(&fs_info
->readahead_workers
, "readahead",
2473 fs_info
->thread_pool_size
,
2474 &fs_info
->generic_worker
);
2477 * endios are largely parallel and should have a very
2480 fs_info
->endio_workers
.idle_thresh
= 4;
2481 fs_info
->endio_meta_workers
.idle_thresh
= 4;
2482 fs_info
->endio_raid56_workers
.idle_thresh
= 4;
2483 fs_info
->rmw_workers
.idle_thresh
= 2;
2485 fs_info
->endio_write_workers
.idle_thresh
= 2;
2486 fs_info
->endio_meta_write_workers
.idle_thresh
= 2;
2487 fs_info
->readahead_workers
.idle_thresh
= 2;
2490 * btrfs_start_workers can really only fail because of ENOMEM so just
2491 * return -ENOMEM if any of these fail.
2493 ret
= btrfs_start_workers(&fs_info
->workers
);
2494 ret
|= btrfs_start_workers(&fs_info
->generic_worker
);
2495 ret
|= btrfs_start_workers(&fs_info
->submit_workers
);
2496 ret
|= btrfs_start_workers(&fs_info
->delalloc_workers
);
2497 ret
|= btrfs_start_workers(&fs_info
->fixup_workers
);
2498 ret
|= btrfs_start_workers(&fs_info
->endio_workers
);
2499 ret
|= btrfs_start_workers(&fs_info
->endio_meta_workers
);
2500 ret
|= btrfs_start_workers(&fs_info
->rmw_workers
);
2501 ret
|= btrfs_start_workers(&fs_info
->endio_raid56_workers
);
2502 ret
|= btrfs_start_workers(&fs_info
->endio_meta_write_workers
);
2503 ret
|= btrfs_start_workers(&fs_info
->endio_write_workers
);
2504 ret
|= btrfs_start_workers(&fs_info
->endio_freespace_worker
);
2505 ret
|= btrfs_start_workers(&fs_info
->delayed_workers
);
2506 ret
|= btrfs_start_workers(&fs_info
->caching_workers
);
2507 ret
|= btrfs_start_workers(&fs_info
->readahead_workers
);
2508 ret
|= btrfs_start_workers(&fs_info
->flush_workers
);
2511 goto fail_sb_buffer
;
2514 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
2515 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
2516 4 * 1024 * 1024 / PAGE_CACHE_SIZE
);
2518 tree_root
->nodesize
= nodesize
;
2519 tree_root
->leafsize
= leafsize
;
2520 tree_root
->sectorsize
= sectorsize
;
2521 tree_root
->stripesize
= stripesize
;
2523 sb
->s_blocksize
= sectorsize
;
2524 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
2526 if (disk_super
->magic
!= cpu_to_le64(BTRFS_MAGIC
)) {
2527 printk(KERN_INFO
"btrfs: valid FS not found on %s\n", sb
->s_id
);
2528 goto fail_sb_buffer
;
2531 if (sectorsize
!= PAGE_SIZE
) {
2532 printk(KERN_WARNING
"btrfs: Incompatible sector size(%lu) "
2533 "found on %s\n", (unsigned long)sectorsize
, sb
->s_id
);
2534 goto fail_sb_buffer
;
2537 mutex_lock(&fs_info
->chunk_mutex
);
2538 ret
= btrfs_read_sys_array(tree_root
);
2539 mutex_unlock(&fs_info
->chunk_mutex
);
2541 printk(KERN_WARNING
"btrfs: failed to read the system "
2542 "array on %s\n", sb
->s_id
);
2543 goto fail_sb_buffer
;
2546 blocksize
= btrfs_level_size(tree_root
,
2547 btrfs_super_chunk_root_level(disk_super
));
2548 generation
= btrfs_super_chunk_root_generation(disk_super
);
2550 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
2551 chunk_root
, fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
2553 chunk_root
->node
= read_tree_block(chunk_root
,
2554 btrfs_super_chunk_root(disk_super
),
2555 blocksize
, generation
);
2556 if (!chunk_root
->node
||
2557 !test_bit(EXTENT_BUFFER_UPTODATE
, &chunk_root
->node
->bflags
)) {
2558 printk(KERN_WARNING
"btrfs: failed to read chunk root on %s\n",
2560 goto fail_tree_roots
;
2562 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
2563 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
2565 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
2566 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root
->node
),
2569 ret
= btrfs_read_chunk_tree(chunk_root
);
2571 printk(KERN_WARNING
"btrfs: failed to read chunk tree on %s\n",
2573 goto fail_tree_roots
;
2577 * keep the device that is marked to be the target device for the
2578 * dev_replace procedure
2580 btrfs_close_extra_devices(fs_info
, fs_devices
, 0);
2582 if (!fs_devices
->latest_bdev
) {
2583 printk(KERN_CRIT
"btrfs: failed to read devices on %s\n",
2585 goto fail_tree_roots
;
2589 blocksize
= btrfs_level_size(tree_root
,
2590 btrfs_super_root_level(disk_super
));
2591 generation
= btrfs_super_generation(disk_super
);
2593 tree_root
->node
= read_tree_block(tree_root
,
2594 btrfs_super_root(disk_super
),
2595 blocksize
, generation
);
2596 if (!tree_root
->node
||
2597 !test_bit(EXTENT_BUFFER_UPTODATE
, &tree_root
->node
->bflags
)) {
2598 printk(KERN_WARNING
"btrfs: failed to read tree root on %s\n",
2601 goto recovery_tree_root
;
2604 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
2605 tree_root
->commit_root
= btrfs_root_node(tree_root
);
2607 ret
= find_and_setup_root(tree_root
, fs_info
,
2608 BTRFS_EXTENT_TREE_OBJECTID
, extent_root
);
2610 goto recovery_tree_root
;
2611 extent_root
->track_dirty
= 1;
2613 ret
= find_and_setup_root(tree_root
, fs_info
,
2614 BTRFS_DEV_TREE_OBJECTID
, dev_root
);
2616 goto recovery_tree_root
;
2617 dev_root
->track_dirty
= 1;
2619 ret
= find_and_setup_root(tree_root
, fs_info
,
2620 BTRFS_CSUM_TREE_OBJECTID
, csum_root
);
2622 goto recovery_tree_root
;
2623 csum_root
->track_dirty
= 1;
2625 ret
= find_and_setup_root(tree_root
, fs_info
,
2626 BTRFS_QUOTA_TREE_OBJECTID
, quota_root
);
2629 quota_root
= fs_info
->quota_root
= NULL
;
2631 quota_root
->track_dirty
= 1;
2632 fs_info
->quota_enabled
= 1;
2633 fs_info
->pending_quota_state
= 1;
2636 fs_info
->generation
= generation
;
2637 fs_info
->last_trans_committed
= generation
;
2639 ret
= btrfs_recover_balance(fs_info
);
2641 printk(KERN_WARNING
"btrfs: failed to recover balance\n");
2642 goto fail_block_groups
;
2645 ret
= btrfs_init_dev_stats(fs_info
);
2647 printk(KERN_ERR
"btrfs: failed to init dev_stats: %d\n",
2649 goto fail_block_groups
;
2652 ret
= btrfs_init_dev_replace(fs_info
);
2654 pr_err("btrfs: failed to init dev_replace: %d\n", ret
);
2655 goto fail_block_groups
;
2658 btrfs_close_extra_devices(fs_info
, fs_devices
, 1);
2660 ret
= btrfs_init_space_info(fs_info
);
2662 printk(KERN_ERR
"Failed to initial space info: %d\n", ret
);
2663 goto fail_block_groups
;
2666 ret
= btrfs_read_block_groups(extent_root
);
2668 printk(KERN_ERR
"Failed to read block groups: %d\n", ret
);
2669 goto fail_block_groups
;
2671 fs_info
->num_tolerated_disk_barrier_failures
=
2672 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2673 if (fs_info
->fs_devices
->missing_devices
>
2674 fs_info
->num_tolerated_disk_barrier_failures
&&
2675 !(sb
->s_flags
& MS_RDONLY
)) {
2677 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2678 goto fail_block_groups
;
2681 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
2683 if (IS_ERR(fs_info
->cleaner_kthread
))
2684 goto fail_block_groups
;
2686 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
2688 "btrfs-transaction");
2689 if (IS_ERR(fs_info
->transaction_kthread
))
2692 if (!btrfs_test_opt(tree_root
, SSD
) &&
2693 !btrfs_test_opt(tree_root
, NOSSD
) &&
2694 !fs_info
->fs_devices
->rotating
) {
2695 printk(KERN_INFO
"Btrfs detected SSD devices, enabling SSD "
2697 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
2700 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2701 if (btrfs_test_opt(tree_root
, CHECK_INTEGRITY
)) {
2702 ret
= btrfsic_mount(tree_root
, fs_devices
,
2703 btrfs_test_opt(tree_root
,
2704 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA
) ?
2706 fs_info
->check_integrity_print_mask
);
2708 printk(KERN_WARNING
"btrfs: failed to initialize"
2709 " integrity check module %s\n", sb
->s_id
);
2712 ret
= btrfs_read_qgroup_config(fs_info
);
2714 goto fail_trans_kthread
;
2716 /* do not make disk changes in broken FS */
2717 if (btrfs_super_log_root(disk_super
) != 0) {
2718 u64 bytenr
= btrfs_super_log_root(disk_super
);
2720 if (fs_devices
->rw_devices
== 0) {
2721 printk(KERN_WARNING
"Btrfs log replay required "
2727 btrfs_level_size(tree_root
,
2728 btrfs_super_log_root_level(disk_super
));
2730 log_tree_root
= btrfs_alloc_root(fs_info
);
2731 if (!log_tree_root
) {
2736 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
2737 log_tree_root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
2739 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
2742 if (!log_tree_root
->node
||
2743 !extent_buffer_uptodate(log_tree_root
->node
)) {
2744 printk(KERN_ERR
"btrfs: failed to read log tree\n");
2745 free_extent_buffer(log_tree_root
->node
);
2746 kfree(log_tree_root
);
2747 goto fail_trans_kthread
;
2749 /* returns with log_tree_root freed on success */
2750 ret
= btrfs_recover_log_trees(log_tree_root
);
2752 btrfs_error(tree_root
->fs_info
, ret
,
2753 "Failed to recover log tree");
2754 free_extent_buffer(log_tree_root
->node
);
2755 kfree(log_tree_root
);
2756 goto fail_trans_kthread
;
2759 if (sb
->s_flags
& MS_RDONLY
) {
2760 ret
= btrfs_commit_super(tree_root
);
2762 goto fail_trans_kthread
;
2766 ret
= btrfs_find_orphan_roots(tree_root
);
2768 goto fail_trans_kthread
;
2770 if (!(sb
->s_flags
& MS_RDONLY
)) {
2771 ret
= btrfs_cleanup_fs_roots(fs_info
);
2773 goto fail_trans_kthread
;
2775 ret
= btrfs_recover_relocation(tree_root
);
2778 "btrfs: failed to recover relocation\n");
2784 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
2785 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2786 location
.offset
= (u64
)-1;
2788 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
2789 if (!fs_info
->fs_root
)
2791 if (IS_ERR(fs_info
->fs_root
)) {
2792 err
= PTR_ERR(fs_info
->fs_root
);
2796 if (sb
->s_flags
& MS_RDONLY
)
2799 down_read(&fs_info
->cleanup_work_sem
);
2800 if ((ret
= btrfs_orphan_cleanup(fs_info
->fs_root
)) ||
2801 (ret
= btrfs_orphan_cleanup(fs_info
->tree_root
))) {
2802 up_read(&fs_info
->cleanup_work_sem
);
2803 close_ctree(tree_root
);
2806 up_read(&fs_info
->cleanup_work_sem
);
2808 ret
= btrfs_resume_balance_async(fs_info
);
2810 printk(KERN_WARNING
"btrfs: failed to resume balance\n");
2811 close_ctree(tree_root
);
2815 ret
= btrfs_resume_dev_replace_async(fs_info
);
2817 pr_warn("btrfs: failed to resume dev_replace\n");
2818 close_ctree(tree_root
);
2825 btrfs_free_qgroup_config(fs_info
);
2827 kthread_stop(fs_info
->transaction_kthread
);
2828 del_fs_roots(fs_info
);
2830 kthread_stop(fs_info
->cleaner_kthread
);
2833 * make sure we're done with the btree inode before we stop our
2836 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
2839 btrfs_free_block_groups(fs_info
);
2842 free_root_pointers(fs_info
, 1);
2843 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
2846 btrfs_stop_all_workers(fs_info
);
2849 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
2851 iput(fs_info
->btree_inode
);
2852 fail_delalloc_bytes
:
2853 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
2854 fail_dirty_metadata_bytes
:
2855 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
2857 bdi_destroy(&fs_info
->bdi
);
2859 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
2861 btrfs_free_stripe_hash_table(fs_info
);
2862 btrfs_close_devices(fs_info
->fs_devices
);
2866 if (!btrfs_test_opt(tree_root
, RECOVERY
))
2867 goto fail_tree_roots
;
2869 free_root_pointers(fs_info
, 0);
2871 /* don't use the log in recovery mode, it won't be valid */
2872 btrfs_set_super_log_root(disk_super
, 0);
2874 /* we can't trust the free space cache either */
2875 btrfs_set_opt(fs_info
->mount_opt
, CLEAR_CACHE
);
2877 ret
= next_root_backup(fs_info
, fs_info
->super_copy
,
2878 &num_backups_tried
, &backup_index
);
2880 goto fail_block_groups
;
2881 goto retry_root_backup
;
2884 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
2887 set_buffer_uptodate(bh
);
2889 struct btrfs_device
*device
= (struct btrfs_device
*)
2892 printk_ratelimited_in_rcu(KERN_WARNING
"lost page write due to "
2893 "I/O error on %s\n",
2894 rcu_str_deref(device
->name
));
2895 /* note, we dont' set_buffer_write_io_error because we have
2896 * our own ways of dealing with the IO errors
2898 clear_buffer_uptodate(bh
);
2899 btrfs_dev_stat_inc_and_print(device
, BTRFS_DEV_STAT_WRITE_ERRS
);
2905 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
2907 struct buffer_head
*bh
;
2908 struct buffer_head
*latest
= NULL
;
2909 struct btrfs_super_block
*super
;
2914 /* we would like to check all the supers, but that would make
2915 * a btrfs mount succeed after a mkfs from a different FS.
2916 * So, we need to add a special mount option to scan for
2917 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2919 for (i
= 0; i
< 1; i
++) {
2920 bytenr
= btrfs_sb_offset(i
);
2921 if (bytenr
+ 4096 >= i_size_read(bdev
->bd_inode
))
2923 bh
= __bread(bdev
, bytenr
/ 4096, 4096);
2927 super
= (struct btrfs_super_block
*)bh
->b_data
;
2928 if (btrfs_super_bytenr(super
) != bytenr
||
2929 super
->magic
!= cpu_to_le64(BTRFS_MAGIC
)) {
2934 if (!latest
|| btrfs_super_generation(super
) > transid
) {
2937 transid
= btrfs_super_generation(super
);
2946 * this should be called twice, once with wait == 0 and
2947 * once with wait == 1. When wait == 0 is done, all the buffer heads
2948 * we write are pinned.
2950 * They are released when wait == 1 is done.
2951 * max_mirrors must be the same for both runs, and it indicates how
2952 * many supers on this one device should be written.
2954 * max_mirrors == 0 means to write them all.
2956 static int write_dev_supers(struct btrfs_device
*device
,
2957 struct btrfs_super_block
*sb
,
2958 int do_barriers
, int wait
, int max_mirrors
)
2960 struct buffer_head
*bh
;
2967 if (max_mirrors
== 0)
2968 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
2970 for (i
= 0; i
< max_mirrors
; i
++) {
2971 bytenr
= btrfs_sb_offset(i
);
2972 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= device
->total_bytes
)
2976 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
2977 BTRFS_SUPER_INFO_SIZE
);
2980 if (!buffer_uptodate(bh
))
2983 /* drop our reference */
2986 /* drop the reference from the wait == 0 run */
2990 btrfs_set_super_bytenr(sb
, bytenr
);
2993 crc
= btrfs_csum_data((char *)sb
+
2994 BTRFS_CSUM_SIZE
, crc
,
2995 BTRFS_SUPER_INFO_SIZE
-
2997 btrfs_csum_final(crc
, sb
->csum
);
3000 * one reference for us, and we leave it for the
3003 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
3004 BTRFS_SUPER_INFO_SIZE
);
3005 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
3007 /* one reference for submit_bh */
3010 set_buffer_uptodate(bh
);
3012 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
3013 bh
->b_private
= device
;
3017 * we fua the first super. The others we allow
3020 ret
= btrfsic_submit_bh(WRITE_FUA
, bh
);
3024 return errors
< i
? 0 : -1;
3028 * endio for the write_dev_flush, this will wake anyone waiting
3029 * for the barrier when it is done
3031 static void btrfs_end_empty_barrier(struct bio
*bio
, int err
)
3034 if (err
== -EOPNOTSUPP
)
3035 set_bit(BIO_EOPNOTSUPP
, &bio
->bi_flags
);
3036 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
3038 if (bio
->bi_private
)
3039 complete(bio
->bi_private
);
3044 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3045 * sent down. With wait == 1, it waits for the previous flush.
3047 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3050 static int write_dev_flush(struct btrfs_device
*device
, int wait
)
3055 if (device
->nobarriers
)
3059 bio
= device
->flush_bio
;
3063 wait_for_completion(&device
->flush_wait
);
3065 if (bio_flagged(bio
, BIO_EOPNOTSUPP
)) {
3066 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3067 rcu_str_deref(device
->name
));
3068 device
->nobarriers
= 1;
3069 } else if (!bio_flagged(bio
, BIO_UPTODATE
)) {
3071 btrfs_dev_stat_inc_and_print(device
,
3072 BTRFS_DEV_STAT_FLUSH_ERRS
);
3075 /* drop the reference from the wait == 0 run */
3077 device
->flush_bio
= NULL
;
3083 * one reference for us, and we leave it for the
3086 device
->flush_bio
= NULL
;
3087 bio
= bio_alloc(GFP_NOFS
, 0);
3091 bio
->bi_end_io
= btrfs_end_empty_barrier
;
3092 bio
->bi_bdev
= device
->bdev
;
3093 init_completion(&device
->flush_wait
);
3094 bio
->bi_private
= &device
->flush_wait
;
3095 device
->flush_bio
= bio
;
3098 btrfsic_submit_bio(WRITE_FLUSH
, bio
);
3104 * send an empty flush down to each device in parallel,
3105 * then wait for them
3107 static int barrier_all_devices(struct btrfs_fs_info
*info
)
3109 struct list_head
*head
;
3110 struct btrfs_device
*dev
;
3111 int errors_send
= 0;
3112 int errors_wait
= 0;
3115 /* send down all the barriers */
3116 head
= &info
->fs_devices
->devices
;
3117 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3122 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3125 ret
= write_dev_flush(dev
, 0);
3130 /* wait for all the barriers */
3131 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3136 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3139 ret
= write_dev_flush(dev
, 1);
3143 if (errors_send
> info
->num_tolerated_disk_barrier_failures
||
3144 errors_wait
> info
->num_tolerated_disk_barrier_failures
)
3149 int btrfs_calc_num_tolerated_disk_barrier_failures(
3150 struct btrfs_fs_info
*fs_info
)
3152 struct btrfs_ioctl_space_info space
;
3153 struct btrfs_space_info
*sinfo
;
3154 u64 types
[] = {BTRFS_BLOCK_GROUP_DATA
,
3155 BTRFS_BLOCK_GROUP_SYSTEM
,
3156 BTRFS_BLOCK_GROUP_METADATA
,
3157 BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
};
3161 int num_tolerated_disk_barrier_failures
=
3162 (int)fs_info
->fs_devices
->num_devices
;
3164 for (i
= 0; i
< num_types
; i
++) {
3165 struct btrfs_space_info
*tmp
;
3169 list_for_each_entry_rcu(tmp
, &fs_info
->space_info
, list
) {
3170 if (tmp
->flags
== types
[i
]) {
3180 down_read(&sinfo
->groups_sem
);
3181 for (c
= 0; c
< BTRFS_NR_RAID_TYPES
; c
++) {
3182 if (!list_empty(&sinfo
->block_groups
[c
])) {
3185 btrfs_get_block_group_info(
3186 &sinfo
->block_groups
[c
], &space
);
3187 if (space
.total_bytes
== 0 ||
3188 space
.used_bytes
== 0)
3190 flags
= space
.flags
;
3193 * 0: if dup, single or RAID0 is configured for
3194 * any of metadata, system or data, else
3195 * 1: if RAID5 is configured, or if RAID1 or
3196 * RAID10 is configured and only two mirrors
3198 * 2: if RAID6 is configured, else
3199 * num_mirrors - 1: if RAID1 or RAID10 is
3200 * configured and more than
3201 * 2 mirrors are used.
3203 if (num_tolerated_disk_barrier_failures
> 0 &&
3204 ((flags
& (BTRFS_BLOCK_GROUP_DUP
|
3205 BTRFS_BLOCK_GROUP_RAID0
)) ||
3206 ((flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
)
3208 num_tolerated_disk_barrier_failures
= 0;
3209 else if (num_tolerated_disk_barrier_failures
> 1) {
3210 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3211 BTRFS_BLOCK_GROUP_RAID5
|
3212 BTRFS_BLOCK_GROUP_RAID10
)) {
3213 num_tolerated_disk_barrier_failures
= 1;
3215 BTRFS_BLOCK_GROUP_RAID5
) {
3216 num_tolerated_disk_barrier_failures
= 2;
3221 up_read(&sinfo
->groups_sem
);
3224 return num_tolerated_disk_barrier_failures
;
3227 int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
3229 struct list_head
*head
;
3230 struct btrfs_device
*dev
;
3231 struct btrfs_super_block
*sb
;
3232 struct btrfs_dev_item
*dev_item
;
3236 int total_errors
= 0;
3239 max_errors
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
3240 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
3241 backup_super_roots(root
->fs_info
);
3243 sb
= root
->fs_info
->super_for_commit
;
3244 dev_item
= &sb
->dev_item
;
3246 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3247 head
= &root
->fs_info
->fs_devices
->devices
;
3250 ret
= barrier_all_devices(root
->fs_info
);
3253 &root
->fs_info
->fs_devices
->device_list_mutex
);
3254 btrfs_error(root
->fs_info
, ret
,
3255 "errors while submitting device barriers.");
3260 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3265 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3268 btrfs_set_stack_device_generation(dev_item
, 0);
3269 btrfs_set_stack_device_type(dev_item
, dev
->type
);
3270 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
3271 btrfs_set_stack_device_total_bytes(dev_item
, dev
->total_bytes
);
3272 btrfs_set_stack_device_bytes_used(dev_item
, dev
->bytes_used
);
3273 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
3274 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
3275 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
3276 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
3277 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
3279 flags
= btrfs_super_flags(sb
);
3280 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
3282 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
3286 if (total_errors
> max_errors
) {
3287 printk(KERN_ERR
"btrfs: %d errors while writing supers\n",
3290 /* This shouldn't happen. FUA is masked off if unsupported */
3295 list_for_each_entry_rcu(dev
, head
, dev_list
) {
3298 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
3301 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
3305 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3306 if (total_errors
> max_errors
) {
3307 btrfs_error(root
->fs_info
, -EIO
,
3308 "%d errors while writing supers", total_errors
);
3314 int write_ctree_super(struct btrfs_trans_handle
*trans
,
3315 struct btrfs_root
*root
, int max_mirrors
)
3319 ret
= write_all_supers(root
, max_mirrors
);
3323 void btrfs_free_fs_root(struct btrfs_fs_info
*fs_info
, struct btrfs_root
*root
)
3325 spin_lock(&fs_info
->fs_roots_radix_lock
);
3326 radix_tree_delete(&fs_info
->fs_roots_radix
,
3327 (unsigned long)root
->root_key
.objectid
);
3328 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3330 if (btrfs_root_refs(&root
->root_item
) == 0)
3331 synchronize_srcu(&fs_info
->subvol_srcu
);
3333 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
3334 btrfs_free_log(NULL
, root
);
3335 btrfs_free_log_root_tree(NULL
, fs_info
);
3338 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
3339 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
3343 static void free_fs_root(struct btrfs_root
*root
)
3345 iput(root
->cache_inode
);
3346 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
3348 free_anon_bdev(root
->anon_dev
);
3349 free_extent_buffer(root
->node
);
3350 free_extent_buffer(root
->commit_root
);
3351 kfree(root
->free_ino_ctl
);
3352 kfree(root
->free_ino_pinned
);
3357 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
3359 u64 root_objectid
= 0;
3360 struct btrfs_root
*gang
[8];
3365 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
3366 (void **)gang
, root_objectid
,
3371 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
3372 for (i
= 0; i
< ret
; i
++) {
3375 root_objectid
= gang
[i
]->root_key
.objectid
;
3376 err
= btrfs_orphan_cleanup(gang
[i
]);
3385 int btrfs_commit_super(struct btrfs_root
*root
)
3387 struct btrfs_trans_handle
*trans
;
3390 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3391 btrfs_run_delayed_iputs(root
);
3392 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3393 wake_up_process(root
->fs_info
->cleaner_kthread
);
3395 /* wait until ongoing cleanup work done */
3396 down_write(&root
->fs_info
->cleanup_work_sem
);
3397 up_write(&root
->fs_info
->cleanup_work_sem
);
3399 trans
= btrfs_join_transaction(root
);
3401 return PTR_ERR(trans
);
3402 ret
= btrfs_commit_transaction(trans
, root
);
3405 /* run commit again to drop the original snapshot */
3406 trans
= btrfs_join_transaction(root
);
3408 return PTR_ERR(trans
);
3409 ret
= btrfs_commit_transaction(trans
, root
);
3412 ret
= btrfs_write_and_wait_transaction(NULL
, root
);
3414 btrfs_error(root
->fs_info
, ret
,
3415 "Failed to sync btree inode to disk.");
3419 ret
= write_ctree_super(NULL
, root
, 0);
3423 int close_ctree(struct btrfs_root
*root
)
3425 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3428 fs_info
->closing
= 1;
3431 /* pause restriper - we want to resume on mount */
3432 btrfs_pause_balance(fs_info
);
3434 btrfs_dev_replace_suspend_for_unmount(fs_info
);
3436 btrfs_scrub_cancel(fs_info
);
3438 /* wait for any defraggers to finish */
3439 wait_event(fs_info
->transaction_wait
,
3440 (atomic_read(&fs_info
->defrag_running
) == 0));
3442 /* clear out the rbtree of defraggable inodes */
3443 btrfs_cleanup_defrag_inodes(fs_info
);
3445 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
3446 ret
= btrfs_commit_super(root
);
3448 printk(KERN_ERR
"btrfs: commit super ret %d\n", ret
);
3451 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
3452 btrfs_error_commit_super(root
);
3454 btrfs_put_block_group_cache(fs_info
);
3456 kthread_stop(fs_info
->transaction_kthread
);
3457 kthread_stop(fs_info
->cleaner_kthread
);
3459 fs_info
->closing
= 2;
3462 btrfs_free_qgroup_config(root
->fs_info
);
3464 if (percpu_counter_sum(&fs_info
->delalloc_bytes
)) {
3465 printk(KERN_INFO
"btrfs: at unmount delalloc count %lld\n",
3466 percpu_counter_sum(&fs_info
->delalloc_bytes
));
3469 free_root_pointers(fs_info
, 1);
3471 btrfs_free_block_groups(fs_info
);
3473 del_fs_roots(fs_info
);
3475 iput(fs_info
->btree_inode
);
3477 btrfs_stop_all_workers(fs_info
);
3479 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3480 if (btrfs_test_opt(root
, CHECK_INTEGRITY
))
3481 btrfsic_unmount(root
, fs_info
->fs_devices
);
3484 btrfs_close_devices(fs_info
->fs_devices
);
3485 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
3487 percpu_counter_destroy(&fs_info
->dirty_metadata_bytes
);
3488 percpu_counter_destroy(&fs_info
->delalloc_bytes
);
3489 bdi_destroy(&fs_info
->bdi
);
3490 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
3492 btrfs_free_stripe_hash_table(fs_info
);
3497 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
,
3501 struct inode
*btree_inode
= buf
->pages
[0]->mapping
->host
;
3503 ret
= extent_buffer_uptodate(buf
);
3507 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
3508 parent_transid
, atomic
);
3514 int btrfs_set_buffer_uptodate(struct extent_buffer
*buf
)
3516 return set_extent_buffer_uptodate(buf
);
3519 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
3521 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3522 u64 transid
= btrfs_header_generation(buf
);
3525 btrfs_assert_tree_locked(buf
);
3526 if (transid
!= root
->fs_info
->generation
)
3527 WARN(1, KERN_CRIT
"btrfs transid mismatch buffer %llu, "
3528 "found %llu running %llu\n",
3529 (unsigned long long)buf
->start
,
3530 (unsigned long long)transid
,
3531 (unsigned long long)root
->fs_info
->generation
);
3532 was_dirty
= set_extent_buffer_dirty(buf
);
3534 __percpu_counter_add(&root
->fs_info
->dirty_metadata_bytes
,
3536 root
->fs_info
->dirty_metadata_batch
);
3539 static void __btrfs_btree_balance_dirty(struct btrfs_root
*root
,
3543 * looks as though older kernels can get into trouble with
3544 * this code, they end up stuck in balance_dirty_pages forever
3548 if (current
->flags
& PF_MEMALLOC
)
3552 btrfs_balance_delayed_items(root
);
3554 ret
= percpu_counter_compare(&root
->fs_info
->dirty_metadata_bytes
,
3555 BTRFS_DIRTY_METADATA_THRESH
);
3557 balance_dirty_pages_ratelimited(
3558 root
->fs_info
->btree_inode
->i_mapping
);
3563 void btrfs_btree_balance_dirty(struct btrfs_root
*root
)
3565 __btrfs_btree_balance_dirty(root
, 1);
3568 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root
*root
)
3570 __btrfs_btree_balance_dirty(root
, 0);
3573 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
3575 struct btrfs_root
*root
= BTRFS_I(buf
->pages
[0]->mapping
->host
)->root
;
3576 return btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
3579 static int btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
3582 if (btrfs_super_csum_type(fs_info
->super_copy
) >= ARRAY_SIZE(btrfs_csum_sizes
)) {
3583 printk(KERN_ERR
"btrfs: unsupported checksum algorithm\n");
3593 void btrfs_error_commit_super(struct btrfs_root
*root
)
3595 mutex_lock(&root
->fs_info
->cleaner_mutex
);
3596 btrfs_run_delayed_iputs(root
);
3597 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
3599 down_write(&root
->fs_info
->cleanup_work_sem
);
3600 up_write(&root
->fs_info
->cleanup_work_sem
);
3602 /* cleanup FS via transaction */
3603 btrfs_cleanup_transaction(root
);
3606 static void btrfs_destroy_ordered_operations(struct btrfs_transaction
*t
,
3607 struct btrfs_root
*root
)
3609 struct btrfs_inode
*btrfs_inode
;
3610 struct list_head splice
;
3612 INIT_LIST_HEAD(&splice
);
3614 mutex_lock(&root
->fs_info
->ordered_operations_mutex
);
3615 spin_lock(&root
->fs_info
->ordered_extent_lock
);
3617 list_splice_init(&t
->ordered_operations
, &splice
);
3618 while (!list_empty(&splice
)) {
3619 btrfs_inode
= list_entry(splice
.next
, struct btrfs_inode
,
3620 ordered_operations
);
3622 list_del_init(&btrfs_inode
->ordered_operations
);
3624 btrfs_invalidate_inodes(btrfs_inode
->root
);
3627 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
3628 mutex_unlock(&root
->fs_info
->ordered_operations_mutex
);
3631 static void btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
3633 struct btrfs_ordered_extent
*ordered
;
3635 spin_lock(&root
->fs_info
->ordered_extent_lock
);
3637 * This will just short circuit the ordered completion stuff which will
3638 * make sure the ordered extent gets properly cleaned up.
3640 list_for_each_entry(ordered
, &root
->fs_info
->ordered_extents
,
3642 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
3643 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
3646 int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
3647 struct btrfs_root
*root
)
3649 struct rb_node
*node
;
3650 struct btrfs_delayed_ref_root
*delayed_refs
;
3651 struct btrfs_delayed_ref_node
*ref
;
3654 delayed_refs
= &trans
->delayed_refs
;
3656 spin_lock(&delayed_refs
->lock
);
3657 if (delayed_refs
->num_entries
== 0) {
3658 spin_unlock(&delayed_refs
->lock
);
3659 printk(KERN_INFO
"delayed_refs has NO entry\n");
3663 while ((node
= rb_first(&delayed_refs
->root
)) != NULL
) {
3664 struct btrfs_delayed_ref_head
*head
= NULL
;
3666 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
3667 atomic_set(&ref
->refs
, 1);
3668 if (btrfs_delayed_ref_is_head(ref
)) {
3670 head
= btrfs_delayed_node_to_head(ref
);
3671 if (!mutex_trylock(&head
->mutex
)) {
3672 atomic_inc(&ref
->refs
);
3673 spin_unlock(&delayed_refs
->lock
);
3675 /* Need to wait for the delayed ref to run */
3676 mutex_lock(&head
->mutex
);
3677 mutex_unlock(&head
->mutex
);
3678 btrfs_put_delayed_ref(ref
);
3680 spin_lock(&delayed_refs
->lock
);
3684 btrfs_free_delayed_extent_op(head
->extent_op
);
3685 delayed_refs
->num_heads
--;
3686 if (list_empty(&head
->cluster
))
3687 delayed_refs
->num_heads_ready
--;
3688 list_del_init(&head
->cluster
);
3692 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
3693 delayed_refs
->num_entries
--;
3695 mutex_unlock(&head
->mutex
);
3696 spin_unlock(&delayed_refs
->lock
);
3697 btrfs_put_delayed_ref(ref
);
3700 spin_lock(&delayed_refs
->lock
);
3703 spin_unlock(&delayed_refs
->lock
);
3708 static void btrfs_evict_pending_snapshots(struct btrfs_transaction
*t
)
3710 struct btrfs_pending_snapshot
*snapshot
;
3711 struct list_head splice
;
3713 INIT_LIST_HEAD(&splice
);
3715 list_splice_init(&t
->pending_snapshots
, &splice
);
3717 while (!list_empty(&splice
)) {
3718 snapshot
= list_entry(splice
.next
,
3719 struct btrfs_pending_snapshot
,
3721 snapshot
->error
= -ECANCELED
;
3722 list_del_init(&snapshot
->list
);
3726 static void btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
3728 struct btrfs_inode
*btrfs_inode
;
3729 struct list_head splice
;
3731 INIT_LIST_HEAD(&splice
);
3733 spin_lock(&root
->fs_info
->delalloc_lock
);
3734 list_splice_init(&root
->fs_info
->delalloc_inodes
, &splice
);
3736 while (!list_empty(&splice
)) {
3737 btrfs_inode
= list_entry(splice
.next
, struct btrfs_inode
,
3740 list_del_init(&btrfs_inode
->delalloc_inodes
);
3741 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
3742 &btrfs_inode
->runtime_flags
);
3744 btrfs_invalidate_inodes(btrfs_inode
->root
);
3747 spin_unlock(&root
->fs_info
->delalloc_lock
);
3750 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
3751 struct extent_io_tree
*dirty_pages
,
3755 struct extent_buffer
*eb
;
3760 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
3765 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
3766 while (start
<= end
) {
3767 eb
= btrfs_find_tree_block(root
, start
,
3772 wait_on_extent_buffer_writeback(eb
);
3774 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
3776 clear_extent_buffer_dirty(eb
);
3777 free_extent_buffer_stale(eb
);
3784 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
3785 struct extent_io_tree
*pinned_extents
)
3787 struct extent_io_tree
*unpin
;
3793 unpin
= pinned_extents
;
3796 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
3797 EXTENT_DIRTY
, NULL
);
3802 if (btrfs_test_opt(root
, DISCARD
))
3803 ret
= btrfs_error_discard_extent(root
, start
,
3807 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
3808 btrfs_error_unpin_extent_range(root
, start
, end
);
3813 if (unpin
== &root
->fs_info
->freed_extents
[0])
3814 unpin
= &root
->fs_info
->freed_extents
[1];
3816 unpin
= &root
->fs_info
->freed_extents
[0];
3824 void btrfs_cleanup_one_transaction(struct btrfs_transaction
*cur_trans
,
3825 struct btrfs_root
*root
)
3827 btrfs_destroy_delayed_refs(cur_trans
, root
);
3828 btrfs_block_rsv_release(root
, &root
->fs_info
->trans_block_rsv
,
3829 cur_trans
->dirty_pages
.dirty_bytes
);
3831 /* FIXME: cleanup wait for commit */
3832 cur_trans
->in_commit
= 1;
3833 cur_trans
->blocked
= 1;
3834 wake_up(&root
->fs_info
->transaction_blocked_wait
);
3836 btrfs_evict_pending_snapshots(cur_trans
);
3838 cur_trans
->blocked
= 0;
3839 wake_up(&root
->fs_info
->transaction_wait
);
3841 cur_trans
->commit_done
= 1;
3842 wake_up(&cur_trans
->commit_wait
);
3844 btrfs_destroy_delayed_inodes(root
);
3845 btrfs_assert_delayed_root_empty(root
);
3847 btrfs_destroy_marked_extents(root
, &cur_trans
->dirty_pages
,
3849 btrfs_destroy_pinned_extent(root
,
3850 root
->fs_info
->pinned_extents
);
3853 memset(cur_trans, 0, sizeof(*cur_trans));
3854 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3858 int btrfs_cleanup_transaction(struct btrfs_root
*root
)
3860 struct btrfs_transaction
*t
;
3863 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
3865 spin_lock(&root
->fs_info
->trans_lock
);
3866 list_splice_init(&root
->fs_info
->trans_list
, &list
);
3867 root
->fs_info
->trans_no_join
= 1;
3868 spin_unlock(&root
->fs_info
->trans_lock
);
3870 while (!list_empty(&list
)) {
3871 t
= list_entry(list
.next
, struct btrfs_transaction
, list
);
3873 btrfs_destroy_ordered_operations(t
, root
);
3875 btrfs_destroy_ordered_extents(root
);
3877 btrfs_destroy_delayed_refs(t
, root
);
3879 btrfs_block_rsv_release(root
,
3880 &root
->fs_info
->trans_block_rsv
,
3881 t
->dirty_pages
.dirty_bytes
);
3883 /* FIXME: cleanup wait for commit */
3887 if (waitqueue_active(&root
->fs_info
->transaction_blocked_wait
))
3888 wake_up(&root
->fs_info
->transaction_blocked_wait
);
3890 btrfs_evict_pending_snapshots(t
);
3894 if (waitqueue_active(&root
->fs_info
->transaction_wait
))
3895 wake_up(&root
->fs_info
->transaction_wait
);
3899 if (waitqueue_active(&t
->commit_wait
))
3900 wake_up(&t
->commit_wait
);
3902 btrfs_destroy_delayed_inodes(root
);
3903 btrfs_assert_delayed_root_empty(root
);
3905 btrfs_destroy_delalloc_inodes(root
);
3907 spin_lock(&root
->fs_info
->trans_lock
);
3908 root
->fs_info
->running_transaction
= NULL
;
3909 spin_unlock(&root
->fs_info
->trans_lock
);
3911 btrfs_destroy_marked_extents(root
, &t
->dirty_pages
,
3914 btrfs_destroy_pinned_extent(root
,
3915 root
->fs_info
->pinned_extents
);
3917 atomic_set(&t
->use_count
, 0);
3918 list_del_init(&t
->list
);
3919 memset(t
, 0, sizeof(*t
));
3920 kmem_cache_free(btrfs_transaction_cachep
, t
);
3923 spin_lock(&root
->fs_info
->trans_lock
);
3924 root
->fs_info
->trans_no_join
= 0;
3925 spin_unlock(&root
->fs_info
->trans_lock
);
3926 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
3931 static struct extent_io_ops btree_extent_io_ops
= {
3932 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
3933 .readpage_io_failed_hook
= btree_io_failed_hook
,
3934 .submit_bio_hook
= btree_submit_bio_hook
,
3935 /* note we're sharing with inode.c for the merge bio hook */
3936 .merge_bio_hook
= btrfs_merge_bio_hook
,