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"
47 static struct extent_io_ops btree_extent_io_ops
;
48 static void end_workqueue_fn(struct btrfs_work
*work
);
49 static void free_fs_root(struct btrfs_root
*root
);
50 static void btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
52 static int btrfs_destroy_ordered_operations(struct btrfs_root
*root
);
53 static int btrfs_destroy_ordered_extents(struct btrfs_root
*root
);
54 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
55 struct btrfs_root
*root
);
56 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction
*t
);
57 static int btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
);
58 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
59 struct extent_io_tree
*dirty_pages
,
61 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
62 struct extent_io_tree
*pinned_extents
);
63 static int btrfs_cleanup_transaction(struct btrfs_root
*root
);
66 * end_io_wq structs are used to do processing in task context when an IO is
67 * complete. This is used during reads to verify checksums, and it is used
68 * by writes to insert metadata for new file extents after IO is complete.
74 struct btrfs_fs_info
*info
;
77 struct list_head list
;
78 struct btrfs_work work
;
82 * async submit bios are used to offload expensive checksumming
83 * onto the worker threads. They checksum file and metadata bios
84 * just before they are sent down the IO stack.
86 struct async_submit_bio
{
89 struct list_head list
;
90 extent_submit_bio_hook_t
*submit_bio_start
;
91 extent_submit_bio_hook_t
*submit_bio_done
;
94 unsigned long bio_flags
;
96 * bio_offset is optional, can be used if the pages in the bio
97 * can't tell us where in the file the bio should go
100 struct btrfs_work work
;
103 /* These are used to set the lockdep class on the extent buffer locks.
104 * The class is set by the readpage_end_io_hook after the buffer has
105 * passed csum validation but before the pages are unlocked.
107 * The lockdep class is also set by btrfs_init_new_buffer on freshly
110 * The class is based on the level in the tree block, which allows lockdep
111 * to know that lower nodes nest inside the locks of higher nodes.
113 * We also add a check to make sure the highest level of the tree is
114 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
115 * code needs update as well.
117 #ifdef CONFIG_DEBUG_LOCK_ALLOC
118 # if BTRFS_MAX_LEVEL != 8
121 static struct lock_class_key btrfs_eb_class
[BTRFS_MAX_LEVEL
+ 1];
122 static const char *btrfs_eb_name
[BTRFS_MAX_LEVEL
+ 1] = {
132 /* highest possible level */
138 * extents on the btree inode are pretty simple, there's one extent
139 * that covers the entire device
141 static struct extent_map
*btree_get_extent(struct inode
*inode
,
142 struct page
*page
, size_t pg_offset
, u64 start
, u64 len
,
145 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
146 struct extent_map
*em
;
149 read_lock(&em_tree
->lock
);
150 em
= lookup_extent_mapping(em_tree
, start
, len
);
153 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
154 read_unlock(&em_tree
->lock
);
157 read_unlock(&em_tree
->lock
);
159 em
= alloc_extent_map();
161 em
= ERR_PTR(-ENOMEM
);
166 em
->block_len
= (u64
)-1;
168 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
170 write_lock(&em_tree
->lock
);
171 ret
= add_extent_mapping(em_tree
, em
);
172 if (ret
== -EEXIST
) {
173 u64 failed_start
= em
->start
;
174 u64 failed_len
= em
->len
;
177 em
= lookup_extent_mapping(em_tree
, start
, len
);
181 em
= lookup_extent_mapping(em_tree
, failed_start
,
189 write_unlock(&em_tree
->lock
);
197 u32
btrfs_csum_data(struct btrfs_root
*root
, char *data
, u32 seed
, size_t len
)
199 return crc32c(seed
, data
, len
);
202 void btrfs_csum_final(u32 crc
, char *result
)
204 put_unaligned_le32(~crc
, result
);
208 * compute the csum for a btree block, and either verify it or write it
209 * into the csum field of the block.
211 static int csum_tree_block(struct btrfs_root
*root
, struct extent_buffer
*buf
,
215 btrfs_super_csum_size(&root
->fs_info
->super_copy
);
218 unsigned long cur_len
;
219 unsigned long offset
= BTRFS_CSUM_SIZE
;
220 char *map_token
= NULL
;
222 unsigned long map_start
;
223 unsigned long map_len
;
226 unsigned long inline_result
;
228 len
= buf
->len
- offset
;
230 err
= map_private_extent_buffer(buf
, offset
, 32,
232 &map_start
, &map_len
, KM_USER0
);
235 cur_len
= min(len
, map_len
- (offset
- map_start
));
236 crc
= btrfs_csum_data(root
, kaddr
+ offset
- map_start
,
240 unmap_extent_buffer(buf
, map_token
, KM_USER0
);
242 if (csum_size
> sizeof(inline_result
)) {
243 result
= kzalloc(csum_size
* sizeof(char), GFP_NOFS
);
247 result
= (char *)&inline_result
;
250 btrfs_csum_final(crc
, result
);
253 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
256 memcpy(&found
, result
, csum_size
);
258 read_extent_buffer(buf
, &val
, 0, csum_size
);
259 printk_ratelimited(KERN_INFO
"btrfs: %s checksum verify "
260 "failed on %llu wanted %X found %X "
262 root
->fs_info
->sb
->s_id
,
263 (unsigned long long)buf
->start
, val
, found
,
264 btrfs_header_level(buf
));
265 if (result
!= (char *)&inline_result
)
270 write_extent_buffer(buf
, result
, 0, csum_size
);
272 if (result
!= (char *)&inline_result
)
278 * we can't consider a given block up to date unless the transid of the
279 * block matches the transid in the parent node's pointer. This is how we
280 * detect blocks that either didn't get written at all or got written
281 * in the wrong place.
283 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
284 struct extent_buffer
*eb
, u64 parent_transid
)
286 struct extent_state
*cached_state
= NULL
;
289 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
292 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
293 0, &cached_state
, GFP_NOFS
);
294 if (extent_buffer_uptodate(io_tree
, eb
, cached_state
) &&
295 btrfs_header_generation(eb
) == parent_transid
) {
299 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
301 (unsigned long long)eb
->start
,
302 (unsigned long long)parent_transid
,
303 (unsigned long long)btrfs_header_generation(eb
));
305 clear_extent_buffer_uptodate(io_tree
, eb
, &cached_state
);
307 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
308 &cached_state
, GFP_NOFS
);
313 * helper to read a given tree block, doing retries as required when
314 * the checksums don't match and we have alternate mirrors to try.
316 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
317 struct extent_buffer
*eb
,
318 u64 start
, u64 parent_transid
)
320 struct extent_io_tree
*io_tree
;
325 clear_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
326 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
328 ret
= read_extent_buffer_pages(io_tree
, eb
, start
, 1,
329 btree_get_extent
, mirror_num
);
331 !verify_parent_transid(io_tree
, eb
, parent_transid
))
335 * This buffer's crc is fine, but its contents are corrupted, so
336 * there is no reason to read the other copies, they won't be
339 if (test_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
))
342 num_copies
= btrfs_num_copies(&root
->fs_info
->mapping_tree
,
348 if (mirror_num
> num_copies
)
355 * checksum a dirty tree block before IO. This has extra checks to make sure
356 * we only fill in the checksum field in the first page of a multi-page block
359 static int csum_dirty_buffer(struct btrfs_root
*root
, struct page
*page
)
361 struct extent_io_tree
*tree
;
362 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
365 struct extent_buffer
*eb
;
368 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
370 if (page
->private == EXTENT_PAGE_PRIVATE
) {
374 if (!page
->private) {
378 len
= page
->private >> 2;
381 eb
= alloc_extent_buffer(tree
, start
, len
, page
);
386 ret
= btree_read_extent_buffer_pages(root
, eb
, start
+ PAGE_CACHE_SIZE
,
387 btrfs_header_generation(eb
));
389 WARN_ON(!btrfs_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
));
391 found_start
= btrfs_header_bytenr(eb
);
392 if (found_start
!= start
) {
396 if (eb
->first_page
!= page
) {
400 if (!PageUptodate(page
)) {
404 csum_tree_block(root
, eb
, 0);
406 free_extent_buffer(eb
);
411 static int check_tree_block_fsid(struct btrfs_root
*root
,
412 struct extent_buffer
*eb
)
414 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
415 u8 fsid
[BTRFS_UUID_SIZE
];
418 read_extent_buffer(eb
, fsid
, (unsigned long)btrfs_header_fsid(eb
),
421 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
425 fs_devices
= fs_devices
->seed
;
430 #define CORRUPT(reason, eb, root, slot) \
431 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
432 "root=%llu, slot=%d\n", reason, \
433 (unsigned long long)btrfs_header_bytenr(eb), \
434 (unsigned long long)root->objectid, slot)
436 static noinline
int check_leaf(struct btrfs_root
*root
,
437 struct extent_buffer
*leaf
)
439 struct btrfs_key key
;
440 struct btrfs_key leaf_key
;
441 u32 nritems
= btrfs_header_nritems(leaf
);
447 /* Check the 0 item */
448 if (btrfs_item_offset_nr(leaf
, 0) + btrfs_item_size_nr(leaf
, 0) !=
449 BTRFS_LEAF_DATA_SIZE(root
)) {
450 CORRUPT("invalid item offset size pair", leaf
, root
, 0);
455 * Check to make sure each items keys are in the correct order and their
456 * offsets make sense. We only have to loop through nritems-1 because
457 * we check the current slot against the next slot, which verifies the
458 * next slot's offset+size makes sense and that the current's slot
461 for (slot
= 0; slot
< nritems
- 1; slot
++) {
462 btrfs_item_key_to_cpu(leaf
, &leaf_key
, slot
);
463 btrfs_item_key_to_cpu(leaf
, &key
, slot
+ 1);
465 /* Make sure the keys are in the right order */
466 if (btrfs_comp_cpu_keys(&leaf_key
, &key
) >= 0) {
467 CORRUPT("bad key order", leaf
, root
, slot
);
472 * Make sure the offset and ends are right, remember that the
473 * item data starts at the end of the leaf and grows towards the
476 if (btrfs_item_offset_nr(leaf
, slot
) !=
477 btrfs_item_end_nr(leaf
, slot
+ 1)) {
478 CORRUPT("slot offset bad", leaf
, root
, slot
);
483 * Check to make sure that we don't point outside of the leaf,
484 * just incase all the items are consistent to eachother, but
485 * all point outside of the leaf.
487 if (btrfs_item_end_nr(leaf
, slot
) >
488 BTRFS_LEAF_DATA_SIZE(root
)) {
489 CORRUPT("slot end outside of leaf", leaf
, root
, slot
);
497 #ifdef CONFIG_DEBUG_LOCK_ALLOC
498 void btrfs_set_buffer_lockdep_class(struct extent_buffer
*eb
, int level
)
500 lockdep_set_class_and_name(&eb
->lock
,
501 &btrfs_eb_class
[level
],
502 btrfs_eb_name
[level
]);
506 static int btree_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
507 struct extent_state
*state
)
509 struct extent_io_tree
*tree
;
513 struct extent_buffer
*eb
;
514 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
517 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
518 if (page
->private == EXTENT_PAGE_PRIVATE
)
523 len
= page
->private >> 2;
526 eb
= alloc_extent_buffer(tree
, start
, len
, page
);
532 found_start
= btrfs_header_bytenr(eb
);
533 if (found_start
!= start
) {
534 printk_ratelimited(KERN_INFO
"btrfs bad tree block start "
536 (unsigned long long)found_start
,
537 (unsigned long long)eb
->start
);
541 if (eb
->first_page
!= page
) {
542 printk(KERN_INFO
"btrfs bad first page %lu %lu\n",
543 eb
->first_page
->index
, page
->index
);
548 if (check_tree_block_fsid(root
, eb
)) {
549 printk_ratelimited(KERN_INFO
"btrfs bad fsid on block %llu\n",
550 (unsigned long long)eb
->start
);
554 found_level
= btrfs_header_level(eb
);
556 btrfs_set_buffer_lockdep_class(eb
, found_level
);
558 ret
= csum_tree_block(root
, eb
, 1);
565 * If this is a leaf block and it is corrupt, set the corrupt bit so
566 * that we don't try and read the other copies of this block, just
569 if (found_level
== 0 && check_leaf(root
, eb
)) {
570 set_bit(EXTENT_BUFFER_CORRUPT
, &eb
->bflags
);
574 end
= min_t(u64
, eb
->len
, PAGE_CACHE_SIZE
);
575 end
= eb
->start
+ end
- 1;
577 free_extent_buffer(eb
);
582 static void end_workqueue_bio(struct bio
*bio
, int err
)
584 struct end_io_wq
*end_io_wq
= bio
->bi_private
;
585 struct btrfs_fs_info
*fs_info
;
587 fs_info
= end_io_wq
->info
;
588 end_io_wq
->error
= err
;
589 end_io_wq
->work
.func
= end_workqueue_fn
;
590 end_io_wq
->work
.flags
= 0;
592 if (bio
->bi_rw
& REQ_WRITE
) {
593 if (end_io_wq
->metadata
== 1)
594 btrfs_queue_worker(&fs_info
->endio_meta_write_workers
,
596 else if (end_io_wq
->metadata
== 2)
597 btrfs_queue_worker(&fs_info
->endio_freespace_worker
,
600 btrfs_queue_worker(&fs_info
->endio_write_workers
,
603 if (end_io_wq
->metadata
)
604 btrfs_queue_worker(&fs_info
->endio_meta_workers
,
607 btrfs_queue_worker(&fs_info
->endio_workers
,
613 * For the metadata arg you want
616 * 1 - if normal metadta
617 * 2 - if writing to the free space cache area
619 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
622 struct end_io_wq
*end_io_wq
;
623 end_io_wq
= kmalloc(sizeof(*end_io_wq
), GFP_NOFS
);
627 end_io_wq
->private = bio
->bi_private
;
628 end_io_wq
->end_io
= bio
->bi_end_io
;
629 end_io_wq
->info
= info
;
630 end_io_wq
->error
= 0;
631 end_io_wq
->bio
= bio
;
632 end_io_wq
->metadata
= metadata
;
634 bio
->bi_private
= end_io_wq
;
635 bio
->bi_end_io
= end_workqueue_bio
;
639 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
641 unsigned long limit
= min_t(unsigned long,
642 info
->workers
.max_workers
,
643 info
->fs_devices
->open_devices
);
647 static void run_one_async_start(struct btrfs_work
*work
)
649 struct async_submit_bio
*async
;
651 async
= container_of(work
, struct async_submit_bio
, work
);
652 async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
653 async
->mirror_num
, async
->bio_flags
,
657 static void run_one_async_done(struct btrfs_work
*work
)
659 struct btrfs_fs_info
*fs_info
;
660 struct async_submit_bio
*async
;
663 async
= container_of(work
, struct async_submit_bio
, work
);
664 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
666 limit
= btrfs_async_submit_limit(fs_info
);
667 limit
= limit
* 2 / 3;
669 atomic_dec(&fs_info
->nr_async_submits
);
671 if (atomic_read(&fs_info
->nr_async_submits
) < limit
&&
672 waitqueue_active(&fs_info
->async_submit_wait
))
673 wake_up(&fs_info
->async_submit_wait
);
675 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
676 async
->mirror_num
, async
->bio_flags
,
680 static void run_one_async_free(struct btrfs_work
*work
)
682 struct async_submit_bio
*async
;
684 async
= container_of(work
, struct async_submit_bio
, work
);
688 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
689 int rw
, struct bio
*bio
, int mirror_num
,
690 unsigned long bio_flags
,
692 extent_submit_bio_hook_t
*submit_bio_start
,
693 extent_submit_bio_hook_t
*submit_bio_done
)
695 struct async_submit_bio
*async
;
697 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
701 async
->inode
= inode
;
704 async
->mirror_num
= mirror_num
;
705 async
->submit_bio_start
= submit_bio_start
;
706 async
->submit_bio_done
= submit_bio_done
;
708 async
->work
.func
= run_one_async_start
;
709 async
->work
.ordered_func
= run_one_async_done
;
710 async
->work
.ordered_free
= run_one_async_free
;
712 async
->work
.flags
= 0;
713 async
->bio_flags
= bio_flags
;
714 async
->bio_offset
= bio_offset
;
716 atomic_inc(&fs_info
->nr_async_submits
);
719 btrfs_set_work_high_prio(&async
->work
);
721 btrfs_queue_worker(&fs_info
->workers
, &async
->work
);
723 while (atomic_read(&fs_info
->async_submit_draining
) &&
724 atomic_read(&fs_info
->nr_async_submits
)) {
725 wait_event(fs_info
->async_submit_wait
,
726 (atomic_read(&fs_info
->nr_async_submits
) == 0));
732 static int btree_csum_one_bio(struct bio
*bio
)
734 struct bio_vec
*bvec
= bio
->bi_io_vec
;
736 struct btrfs_root
*root
;
738 WARN_ON(bio
->bi_vcnt
<= 0);
739 while (bio_index
< bio
->bi_vcnt
) {
740 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
741 csum_dirty_buffer(root
, bvec
->bv_page
);
748 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
749 struct bio
*bio
, int mirror_num
,
750 unsigned long bio_flags
,
754 * when we're called for a write, we're already in the async
755 * submission context. Just jump into btrfs_map_bio
757 btree_csum_one_bio(bio
);
761 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
762 int mirror_num
, unsigned long bio_flags
,
766 * when we're called for a write, we're already in the async
767 * submission context. Just jump into btrfs_map_bio
769 return btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
772 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
773 int mirror_num
, unsigned long bio_flags
,
778 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
782 if (!(rw
& REQ_WRITE
)) {
784 * called for a read, do the setup so that checksum validation
785 * can happen in the async kernel threads
787 return btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
792 * kthread helpers are used to submit writes so that checksumming
793 * can happen in parallel across all CPUs
795 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
796 inode
, rw
, bio
, mirror_num
, 0,
798 __btree_submit_bio_start
,
799 __btree_submit_bio_done
);
802 #ifdef CONFIG_MIGRATION
803 static int btree_migratepage(struct address_space
*mapping
,
804 struct page
*newpage
, struct page
*page
)
807 * we can't safely write a btree page from here,
808 * we haven't done the locking hook
813 * Buffers may be managed in a filesystem specific way.
814 * We must have no buffers or drop them.
816 if (page_has_private(page
) &&
817 !try_to_release_page(page
, GFP_KERNEL
))
819 return migrate_page(mapping
, newpage
, page
);
823 static int btree_writepage(struct page
*page
, struct writeback_control
*wbc
)
825 struct extent_io_tree
*tree
;
826 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
827 struct extent_buffer
*eb
;
830 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
831 if (!(current
->flags
& PF_MEMALLOC
)) {
832 return extent_write_full_page(tree
, page
,
833 btree_get_extent
, wbc
);
836 redirty_page_for_writepage(wbc
, page
);
837 eb
= btrfs_find_tree_block(root
, page_offset(page
), PAGE_CACHE_SIZE
);
840 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
842 spin_lock(&root
->fs_info
->delalloc_lock
);
843 root
->fs_info
->dirty_metadata_bytes
+= PAGE_CACHE_SIZE
;
844 spin_unlock(&root
->fs_info
->delalloc_lock
);
846 free_extent_buffer(eb
);
852 static int btree_writepages(struct address_space
*mapping
,
853 struct writeback_control
*wbc
)
855 struct extent_io_tree
*tree
;
856 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
857 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
858 struct btrfs_root
*root
= BTRFS_I(mapping
->host
)->root
;
860 unsigned long thresh
= 32 * 1024 * 1024;
862 if (wbc
->for_kupdate
)
865 /* this is a bit racy, but that's ok */
866 num_dirty
= root
->fs_info
->dirty_metadata_bytes
;
867 if (num_dirty
< thresh
)
870 return extent_writepages(tree
, mapping
, btree_get_extent
, wbc
);
873 static int btree_readpage(struct file
*file
, struct page
*page
)
875 struct extent_io_tree
*tree
;
876 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
877 return extent_read_full_page(tree
, page
, btree_get_extent
);
880 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
882 struct extent_io_tree
*tree
;
883 struct extent_map_tree
*map
;
886 if (PageWriteback(page
) || PageDirty(page
))
889 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
890 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
892 ret
= try_release_extent_state(map
, tree
, page
, gfp_flags
);
896 ret
= try_release_extent_buffer(tree
, page
);
898 ClearPagePrivate(page
);
899 set_page_private(page
, 0);
900 page_cache_release(page
);
906 static void btree_invalidatepage(struct page
*page
, unsigned long offset
)
908 struct extent_io_tree
*tree
;
909 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
910 extent_invalidatepage(tree
, page
, offset
);
911 btree_releasepage(page
, GFP_NOFS
);
912 if (PagePrivate(page
)) {
913 printk(KERN_WARNING
"btrfs warning page private not zero "
914 "on page %llu\n", (unsigned long long)page_offset(page
));
915 ClearPagePrivate(page
);
916 set_page_private(page
, 0);
917 page_cache_release(page
);
921 static const struct address_space_operations btree_aops
= {
922 .readpage
= btree_readpage
,
923 .writepage
= btree_writepage
,
924 .writepages
= btree_writepages
,
925 .releasepage
= btree_releasepage
,
926 .invalidatepage
= btree_invalidatepage
,
927 #ifdef CONFIG_MIGRATION
928 .migratepage
= btree_migratepage
,
932 int readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
935 struct extent_buffer
*buf
= NULL
;
936 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
939 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
942 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
943 buf
, 0, 0, btree_get_extent
, 0);
944 free_extent_buffer(buf
);
948 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_root
*root
,
949 u64 bytenr
, u32 blocksize
)
951 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
952 struct extent_buffer
*eb
;
953 eb
= find_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
958 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
959 u64 bytenr
, u32 blocksize
)
961 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
962 struct extent_buffer
*eb
;
964 eb
= alloc_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
965 bytenr
, blocksize
, NULL
);
970 int btrfs_write_tree_block(struct extent_buffer
*buf
)
972 return filemap_fdatawrite_range(buf
->first_page
->mapping
, buf
->start
,
973 buf
->start
+ buf
->len
- 1);
976 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
978 return filemap_fdatawait_range(buf
->first_page
->mapping
,
979 buf
->start
, buf
->start
+ buf
->len
- 1);
982 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
983 u32 blocksize
, u64 parent_transid
)
985 struct extent_buffer
*buf
= NULL
;
988 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
992 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
995 set_bit(EXTENT_BUFFER_UPTODATE
, &buf
->bflags
);
1000 int clean_tree_block(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1001 struct extent_buffer
*buf
)
1003 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1004 if (btrfs_header_generation(buf
) ==
1005 root
->fs_info
->running_transaction
->transid
) {
1006 btrfs_assert_tree_locked(buf
);
1008 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
1009 spin_lock(&root
->fs_info
->delalloc_lock
);
1010 if (root
->fs_info
->dirty_metadata_bytes
>= buf
->len
)
1011 root
->fs_info
->dirty_metadata_bytes
-= buf
->len
;
1014 spin_unlock(&root
->fs_info
->delalloc_lock
);
1017 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1018 btrfs_set_lock_blocking(buf
);
1019 clear_extent_buffer_dirty(&BTRFS_I(btree_inode
)->io_tree
,
1025 static int __setup_root(u32 nodesize
, u32 leafsize
, u32 sectorsize
,
1026 u32 stripesize
, struct btrfs_root
*root
,
1027 struct btrfs_fs_info
*fs_info
,
1031 root
->commit_root
= NULL
;
1032 root
->sectorsize
= sectorsize
;
1033 root
->nodesize
= nodesize
;
1034 root
->leafsize
= leafsize
;
1035 root
->stripesize
= stripesize
;
1037 root
->track_dirty
= 0;
1039 root
->orphan_item_inserted
= 0;
1040 root
->orphan_cleanup_state
= 0;
1042 root
->fs_info
= fs_info
;
1043 root
->objectid
= objectid
;
1044 root
->last_trans
= 0;
1045 root
->highest_objectid
= 0;
1047 root
->inode_tree
= RB_ROOT
;
1048 INIT_RADIX_TREE(&root
->delayed_nodes_tree
, GFP_ATOMIC
);
1049 root
->block_rsv
= NULL
;
1050 root
->orphan_block_rsv
= NULL
;
1052 INIT_LIST_HEAD(&root
->dirty_list
);
1053 INIT_LIST_HEAD(&root
->orphan_list
);
1054 INIT_LIST_HEAD(&root
->root_list
);
1055 spin_lock_init(&root
->orphan_lock
);
1056 spin_lock_init(&root
->inode_lock
);
1057 spin_lock_init(&root
->accounting_lock
);
1058 mutex_init(&root
->objectid_mutex
);
1059 mutex_init(&root
->log_mutex
);
1060 init_waitqueue_head(&root
->log_writer_wait
);
1061 init_waitqueue_head(&root
->log_commit_wait
[0]);
1062 init_waitqueue_head(&root
->log_commit_wait
[1]);
1063 atomic_set(&root
->log_commit
[0], 0);
1064 atomic_set(&root
->log_commit
[1], 0);
1065 atomic_set(&root
->log_writers
, 0);
1066 root
->log_batch
= 0;
1067 root
->log_transid
= 0;
1068 root
->last_log_commit
= 0;
1069 extent_io_tree_init(&root
->dirty_log_pages
,
1070 fs_info
->btree_inode
->i_mapping
);
1072 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
1073 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
1074 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
1075 memset(&root
->root_kobj
, 0, sizeof(root
->root_kobj
));
1076 root
->defrag_trans_start
= fs_info
->generation
;
1077 init_completion(&root
->kobj_unregister
);
1078 root
->defrag_running
= 0;
1079 root
->root_key
.objectid
= objectid
;
1084 static int find_and_setup_root(struct btrfs_root
*tree_root
,
1085 struct btrfs_fs_info
*fs_info
,
1087 struct btrfs_root
*root
)
1093 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1094 tree_root
->sectorsize
, tree_root
->stripesize
,
1095 root
, fs_info
, objectid
);
1096 ret
= btrfs_find_last_root(tree_root
, objectid
,
1097 &root
->root_item
, &root
->root_key
);
1102 generation
= btrfs_root_generation(&root
->root_item
);
1103 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1104 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1105 blocksize
, generation
);
1106 if (!root
->node
|| !btrfs_buffer_uptodate(root
->node
, generation
)) {
1107 free_extent_buffer(root
->node
);
1110 root
->commit_root
= btrfs_root_node(root
);
1114 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1115 struct btrfs_fs_info
*fs_info
)
1117 struct btrfs_root
*root
;
1118 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1119 struct extent_buffer
*leaf
;
1121 root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1123 return ERR_PTR(-ENOMEM
);
1125 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1126 tree_root
->sectorsize
, tree_root
->stripesize
,
1127 root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
1129 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1130 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1131 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1133 * log trees do not get reference counted because they go away
1134 * before a real commit is actually done. They do store pointers
1135 * to file data extents, and those reference counts still get
1136 * updated (along with back refs to the log tree).
1140 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
, 0,
1141 BTRFS_TREE_LOG_OBJECTID
, NULL
, 0, 0, 0);
1144 return ERR_CAST(leaf
);
1147 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1148 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1149 btrfs_set_header_generation(leaf
, trans
->transid
);
1150 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1151 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1154 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1155 (unsigned long)btrfs_header_fsid(root
->node
),
1157 btrfs_mark_buffer_dirty(root
->node
);
1158 btrfs_tree_unlock(root
->node
);
1162 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1163 struct btrfs_fs_info
*fs_info
)
1165 struct btrfs_root
*log_root
;
1167 log_root
= alloc_log_tree(trans
, fs_info
);
1168 if (IS_ERR(log_root
))
1169 return PTR_ERR(log_root
);
1170 WARN_ON(fs_info
->log_root_tree
);
1171 fs_info
->log_root_tree
= log_root
;
1175 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1176 struct btrfs_root
*root
)
1178 struct btrfs_root
*log_root
;
1179 struct btrfs_inode_item
*inode_item
;
1181 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1182 if (IS_ERR(log_root
))
1183 return PTR_ERR(log_root
);
1185 log_root
->last_trans
= trans
->transid
;
1186 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1188 inode_item
= &log_root
->root_item
.inode
;
1189 inode_item
->generation
= cpu_to_le64(1);
1190 inode_item
->size
= cpu_to_le64(3);
1191 inode_item
->nlink
= cpu_to_le32(1);
1192 inode_item
->nbytes
= cpu_to_le64(root
->leafsize
);
1193 inode_item
->mode
= cpu_to_le32(S_IFDIR
| 0755);
1195 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1197 WARN_ON(root
->log_root
);
1198 root
->log_root
= log_root
;
1199 root
->log_transid
= 0;
1200 root
->last_log_commit
= 0;
1204 struct btrfs_root
*btrfs_read_fs_root_no_radix(struct btrfs_root
*tree_root
,
1205 struct btrfs_key
*location
)
1207 struct btrfs_root
*root
;
1208 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1209 struct btrfs_path
*path
;
1210 struct extent_buffer
*l
;
1215 root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1217 return ERR_PTR(-ENOMEM
);
1218 if (location
->offset
== (u64
)-1) {
1219 ret
= find_and_setup_root(tree_root
, fs_info
,
1220 location
->objectid
, root
);
1223 return ERR_PTR(ret
);
1228 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1229 tree_root
->sectorsize
, tree_root
->stripesize
,
1230 root
, fs_info
, location
->objectid
);
1232 path
= btrfs_alloc_path();
1235 return ERR_PTR(-ENOMEM
);
1237 ret
= btrfs_search_slot(NULL
, tree_root
, location
, path
, 0, 0);
1240 read_extent_buffer(l
, &root
->root_item
,
1241 btrfs_item_ptr_offset(l
, path
->slots
[0]),
1242 sizeof(root
->root_item
));
1243 memcpy(&root
->root_key
, location
, sizeof(*location
));
1245 btrfs_free_path(path
);
1250 return ERR_PTR(ret
);
1253 generation
= btrfs_root_generation(&root
->root_item
);
1254 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1255 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1256 blocksize
, generation
);
1257 root
->commit_root
= btrfs_root_node(root
);
1258 BUG_ON(!root
->node
);
1260 if (location
->objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1262 btrfs_check_and_init_root_item(&root
->root_item
);
1268 struct btrfs_root
*btrfs_read_fs_root_no_name(struct btrfs_fs_info
*fs_info
,
1269 struct btrfs_key
*location
)
1271 struct btrfs_root
*root
;
1274 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1275 return fs_info
->tree_root
;
1276 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1277 return fs_info
->extent_root
;
1278 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1279 return fs_info
->chunk_root
;
1280 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1281 return fs_info
->dev_root
;
1282 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1283 return fs_info
->csum_root
;
1285 spin_lock(&fs_info
->fs_roots_radix_lock
);
1286 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1287 (unsigned long)location
->objectid
);
1288 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1292 root
= btrfs_read_fs_root_no_radix(fs_info
->tree_root
, location
);
1296 root
->free_ino_ctl
= kzalloc(sizeof(*root
->free_ino_ctl
), GFP_NOFS
);
1297 root
->free_ino_pinned
= kzalloc(sizeof(*root
->free_ino_pinned
),
1299 if (!root
->free_ino_pinned
|| !root
->free_ino_ctl
) {
1304 btrfs_init_free_ino_ctl(root
);
1305 mutex_init(&root
->fs_commit_mutex
);
1306 spin_lock_init(&root
->cache_lock
);
1307 init_waitqueue_head(&root
->cache_wait
);
1309 ret
= get_anon_bdev(&root
->anon_dev
);
1313 if (btrfs_root_refs(&root
->root_item
) == 0) {
1318 ret
= btrfs_find_orphan_item(fs_info
->tree_root
, location
->objectid
);
1322 root
->orphan_item_inserted
= 1;
1324 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
1328 spin_lock(&fs_info
->fs_roots_radix_lock
);
1329 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1330 (unsigned long)root
->root_key
.objectid
,
1335 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1336 radix_tree_preload_end();
1338 if (ret
== -EEXIST
) {
1345 ret
= btrfs_find_dead_roots(fs_info
->tree_root
,
1346 root
->root_key
.objectid
);
1351 return ERR_PTR(ret
);
1354 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1356 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1358 struct btrfs_device
*device
;
1359 struct backing_dev_info
*bdi
;
1362 list_for_each_entry_rcu(device
, &info
->fs_devices
->devices
, dev_list
) {
1365 bdi
= blk_get_backing_dev_info(device
->bdev
);
1366 if (bdi
&& bdi_congested(bdi
, bdi_bits
)) {
1376 * If this fails, caller must call bdi_destroy() to get rid of the
1379 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1383 bdi
->capabilities
= BDI_CAP_MAP_COPY
;
1384 err
= bdi_setup_and_register(bdi
, "btrfs", BDI_CAP_MAP_COPY
);
1388 bdi
->ra_pages
= default_backing_dev_info
.ra_pages
;
1389 bdi
->congested_fn
= btrfs_congested_fn
;
1390 bdi
->congested_data
= info
;
1394 static int bio_ready_for_csum(struct bio
*bio
)
1400 struct extent_io_tree
*io_tree
= NULL
;
1401 struct bio_vec
*bvec
;
1405 bio_for_each_segment(bvec
, bio
, i
) {
1406 page
= bvec
->bv_page
;
1407 if (page
->private == EXTENT_PAGE_PRIVATE
) {
1408 length
+= bvec
->bv_len
;
1411 if (!page
->private) {
1412 length
+= bvec
->bv_len
;
1415 length
= bvec
->bv_len
;
1416 buf_len
= page
->private >> 2;
1417 start
= page_offset(page
) + bvec
->bv_offset
;
1418 io_tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1420 /* are we fully contained in this bio? */
1421 if (buf_len
<= length
)
1424 ret
= extent_range_uptodate(io_tree
, start
+ length
,
1425 start
+ buf_len
- 1);
1430 * called by the kthread helper functions to finally call the bio end_io
1431 * functions. This is where read checksum verification actually happens
1433 static void end_workqueue_fn(struct btrfs_work
*work
)
1436 struct end_io_wq
*end_io_wq
;
1437 struct btrfs_fs_info
*fs_info
;
1440 end_io_wq
= container_of(work
, struct end_io_wq
, work
);
1441 bio
= end_io_wq
->bio
;
1442 fs_info
= end_io_wq
->info
;
1444 /* metadata bio reads are special because the whole tree block must
1445 * be checksummed at once. This makes sure the entire block is in
1446 * ram and up to date before trying to verify things. For
1447 * blocksize <= pagesize, it is basically a noop
1449 if (!(bio
->bi_rw
& REQ_WRITE
) && end_io_wq
->metadata
&&
1450 !bio_ready_for_csum(bio
)) {
1451 btrfs_queue_worker(&fs_info
->endio_meta_workers
,
1455 error
= end_io_wq
->error
;
1456 bio
->bi_private
= end_io_wq
->private;
1457 bio
->bi_end_io
= end_io_wq
->end_io
;
1459 bio_endio(bio
, error
);
1462 static int cleaner_kthread(void *arg
)
1464 struct btrfs_root
*root
= arg
;
1467 vfs_check_frozen(root
->fs_info
->sb
, SB_FREEZE_WRITE
);
1469 if (!(root
->fs_info
->sb
->s_flags
& MS_RDONLY
) &&
1470 mutex_trylock(&root
->fs_info
->cleaner_mutex
)) {
1471 btrfs_run_delayed_iputs(root
);
1472 btrfs_clean_old_snapshots(root
);
1473 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1474 btrfs_run_defrag_inodes(root
->fs_info
);
1477 if (freezing(current
)) {
1480 set_current_state(TASK_INTERRUPTIBLE
);
1481 if (!kthread_should_stop())
1483 __set_current_state(TASK_RUNNING
);
1485 } while (!kthread_should_stop());
1489 static int transaction_kthread(void *arg
)
1491 struct btrfs_root
*root
= arg
;
1492 struct btrfs_trans_handle
*trans
;
1493 struct btrfs_transaction
*cur
;
1496 unsigned long delay
;
1501 vfs_check_frozen(root
->fs_info
->sb
, SB_FREEZE_WRITE
);
1502 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1504 spin_lock(&root
->fs_info
->trans_lock
);
1505 cur
= root
->fs_info
->running_transaction
;
1507 spin_unlock(&root
->fs_info
->trans_lock
);
1511 now
= get_seconds();
1512 if (!cur
->blocked
&&
1513 (now
< cur
->start_time
|| now
- cur
->start_time
< 30)) {
1514 spin_unlock(&root
->fs_info
->trans_lock
);
1518 transid
= cur
->transid
;
1519 spin_unlock(&root
->fs_info
->trans_lock
);
1521 trans
= btrfs_join_transaction(root
);
1522 BUG_ON(IS_ERR(trans
));
1523 if (transid
== trans
->transid
) {
1524 ret
= btrfs_commit_transaction(trans
, root
);
1527 btrfs_end_transaction(trans
, root
);
1530 wake_up_process(root
->fs_info
->cleaner_kthread
);
1531 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1533 if (freezing(current
)) {
1536 set_current_state(TASK_INTERRUPTIBLE
);
1537 if (!kthread_should_stop() &&
1538 !btrfs_transaction_blocked(root
->fs_info
))
1539 schedule_timeout(delay
);
1540 __set_current_state(TASK_RUNNING
);
1542 } while (!kthread_should_stop());
1546 struct btrfs_root
*open_ctree(struct super_block
*sb
,
1547 struct btrfs_fs_devices
*fs_devices
,
1557 struct btrfs_key location
;
1558 struct buffer_head
*bh
;
1559 struct btrfs_root
*extent_root
= kzalloc(sizeof(struct btrfs_root
),
1561 struct btrfs_root
*csum_root
= kzalloc(sizeof(struct btrfs_root
),
1563 struct btrfs_root
*tree_root
= btrfs_sb(sb
);
1564 struct btrfs_fs_info
*fs_info
= NULL
;
1565 struct btrfs_root
*chunk_root
= kzalloc(sizeof(struct btrfs_root
),
1567 struct btrfs_root
*dev_root
= kzalloc(sizeof(struct btrfs_root
),
1569 struct btrfs_root
*log_tree_root
;
1574 struct btrfs_super_block
*disk_super
;
1576 if (!extent_root
|| !tree_root
|| !tree_root
->fs_info
||
1577 !chunk_root
|| !dev_root
|| !csum_root
) {
1581 fs_info
= tree_root
->fs_info
;
1583 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
1589 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
1595 fs_info
->btree_inode
= new_inode(sb
);
1596 if (!fs_info
->btree_inode
) {
1601 fs_info
->btree_inode
->i_mapping
->flags
&= ~__GFP_FS
;
1603 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
1604 INIT_LIST_HEAD(&fs_info
->trans_list
);
1605 INIT_LIST_HEAD(&fs_info
->dead_roots
);
1606 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
1607 INIT_LIST_HEAD(&fs_info
->hashers
);
1608 INIT_LIST_HEAD(&fs_info
->delalloc_inodes
);
1609 INIT_LIST_HEAD(&fs_info
->ordered_operations
);
1610 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
1611 spin_lock_init(&fs_info
->delalloc_lock
);
1612 spin_lock_init(&fs_info
->trans_lock
);
1613 spin_lock_init(&fs_info
->ref_cache_lock
);
1614 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
1615 spin_lock_init(&fs_info
->delayed_iput_lock
);
1616 spin_lock_init(&fs_info
->defrag_inodes_lock
);
1617 mutex_init(&fs_info
->reloc_mutex
);
1619 init_completion(&fs_info
->kobj_unregister
);
1620 fs_info
->tree_root
= tree_root
;
1621 fs_info
->extent_root
= extent_root
;
1622 fs_info
->csum_root
= csum_root
;
1623 fs_info
->chunk_root
= chunk_root
;
1624 fs_info
->dev_root
= dev_root
;
1625 fs_info
->fs_devices
= fs_devices
;
1626 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
1627 INIT_LIST_HEAD(&fs_info
->space_info
);
1628 btrfs_mapping_init(&fs_info
->mapping_tree
);
1629 btrfs_init_block_rsv(&fs_info
->global_block_rsv
);
1630 btrfs_init_block_rsv(&fs_info
->delalloc_block_rsv
);
1631 btrfs_init_block_rsv(&fs_info
->trans_block_rsv
);
1632 btrfs_init_block_rsv(&fs_info
->chunk_block_rsv
);
1633 btrfs_init_block_rsv(&fs_info
->empty_block_rsv
);
1634 INIT_LIST_HEAD(&fs_info
->durable_block_rsv_list
);
1635 mutex_init(&fs_info
->durable_block_rsv_mutex
);
1636 atomic_set(&fs_info
->nr_async_submits
, 0);
1637 atomic_set(&fs_info
->async_delalloc_pages
, 0);
1638 atomic_set(&fs_info
->async_submit_draining
, 0);
1639 atomic_set(&fs_info
->nr_async_bios
, 0);
1640 atomic_set(&fs_info
->defrag_running
, 0);
1642 fs_info
->max_inline
= 8192 * 1024;
1643 fs_info
->metadata_ratio
= 0;
1644 fs_info
->defrag_inodes
= RB_ROOT
;
1645 fs_info
->trans_no_join
= 0;
1647 fs_info
->thread_pool_size
= min_t(unsigned long,
1648 num_online_cpus() + 2, 8);
1650 INIT_LIST_HEAD(&fs_info
->ordered_extents
);
1651 spin_lock_init(&fs_info
->ordered_extent_lock
);
1652 fs_info
->delayed_root
= kmalloc(sizeof(struct btrfs_delayed_root
),
1654 if (!fs_info
->delayed_root
) {
1658 btrfs_init_delayed_root(fs_info
->delayed_root
);
1660 mutex_init(&fs_info
->scrub_lock
);
1661 atomic_set(&fs_info
->scrubs_running
, 0);
1662 atomic_set(&fs_info
->scrub_pause_req
, 0);
1663 atomic_set(&fs_info
->scrubs_paused
, 0);
1664 atomic_set(&fs_info
->scrub_cancel_req
, 0);
1665 init_waitqueue_head(&fs_info
->scrub_pause_wait
);
1666 init_rwsem(&fs_info
->scrub_super_lock
);
1667 fs_info
->scrub_workers_refcnt
= 0;
1669 sb
->s_blocksize
= 4096;
1670 sb
->s_blocksize_bits
= blksize_bits(4096);
1671 sb
->s_bdi
= &fs_info
->bdi
;
1673 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
1674 fs_info
->btree_inode
->i_nlink
= 1;
1676 * we set the i_size on the btree inode to the max possible int.
1677 * the real end of the address space is determined by all of
1678 * the devices in the system
1680 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
1681 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
1682 fs_info
->btree_inode
->i_mapping
->backing_dev_info
= &fs_info
->bdi
;
1684 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
1685 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
1686 fs_info
->btree_inode
->i_mapping
);
1687 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
);
1689 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
1691 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
1692 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
1693 sizeof(struct btrfs_key
));
1694 BTRFS_I(fs_info
->btree_inode
)->dummy_inode
= 1;
1695 insert_inode_hash(fs_info
->btree_inode
);
1697 spin_lock_init(&fs_info
->block_group_cache_lock
);
1698 fs_info
->block_group_cache_tree
= RB_ROOT
;
1700 extent_io_tree_init(&fs_info
->freed_extents
[0],
1701 fs_info
->btree_inode
->i_mapping
);
1702 extent_io_tree_init(&fs_info
->freed_extents
[1],
1703 fs_info
->btree_inode
->i_mapping
);
1704 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
1705 fs_info
->do_barriers
= 1;
1708 mutex_init(&fs_info
->ordered_operations_mutex
);
1709 mutex_init(&fs_info
->tree_log_mutex
);
1710 mutex_init(&fs_info
->chunk_mutex
);
1711 mutex_init(&fs_info
->transaction_kthread_mutex
);
1712 mutex_init(&fs_info
->cleaner_mutex
);
1713 mutex_init(&fs_info
->volume_mutex
);
1714 init_rwsem(&fs_info
->extent_commit_sem
);
1715 init_rwsem(&fs_info
->cleanup_work_sem
);
1716 init_rwsem(&fs_info
->subvol_sem
);
1718 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
1719 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
1721 init_waitqueue_head(&fs_info
->transaction_throttle
);
1722 init_waitqueue_head(&fs_info
->transaction_wait
);
1723 init_waitqueue_head(&fs_info
->transaction_blocked_wait
);
1724 init_waitqueue_head(&fs_info
->async_submit_wait
);
1726 __setup_root(4096, 4096, 4096, 4096, tree_root
,
1727 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
1729 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
1735 memcpy(&fs_info
->super_copy
, bh
->b_data
, sizeof(fs_info
->super_copy
));
1736 memcpy(&fs_info
->super_for_commit
, &fs_info
->super_copy
,
1737 sizeof(fs_info
->super_for_commit
));
1740 memcpy(fs_info
->fsid
, fs_info
->super_copy
.fsid
, BTRFS_FSID_SIZE
);
1742 disk_super
= &fs_info
->super_copy
;
1743 if (!btrfs_super_root(disk_super
))
1746 /* check FS state, whether FS is broken. */
1747 fs_info
->fs_state
|= btrfs_super_flags(disk_super
);
1749 btrfs_check_super_valid(fs_info
, sb
->s_flags
& MS_RDONLY
);
1752 * In the long term, we'll store the compression type in the super
1753 * block, and it'll be used for per file compression control.
1755 fs_info
->compress_type
= BTRFS_COMPRESS_ZLIB
;
1757 ret
= btrfs_parse_options(tree_root
, options
);
1763 features
= btrfs_super_incompat_flags(disk_super
) &
1764 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
1766 printk(KERN_ERR
"BTRFS: couldn't mount because of "
1767 "unsupported optional features (%Lx).\n",
1768 (unsigned long long)features
);
1773 features
= btrfs_super_incompat_flags(disk_super
);
1774 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
1775 if (tree_root
->fs_info
->compress_type
& BTRFS_COMPRESS_LZO
)
1776 features
|= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO
;
1777 btrfs_set_super_incompat_flags(disk_super
, features
);
1779 features
= btrfs_super_compat_ro_flags(disk_super
) &
1780 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
1781 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
1782 printk(KERN_ERR
"BTRFS: couldn't mount RDWR because of "
1783 "unsupported option features (%Lx).\n",
1784 (unsigned long long)features
);
1789 btrfs_init_workers(&fs_info
->generic_worker
,
1790 "genwork", 1, NULL
);
1792 btrfs_init_workers(&fs_info
->workers
, "worker",
1793 fs_info
->thread_pool_size
,
1794 &fs_info
->generic_worker
);
1796 btrfs_init_workers(&fs_info
->delalloc_workers
, "delalloc",
1797 fs_info
->thread_pool_size
,
1798 &fs_info
->generic_worker
);
1800 btrfs_init_workers(&fs_info
->submit_workers
, "submit",
1801 min_t(u64
, fs_devices
->num_devices
,
1802 fs_info
->thread_pool_size
),
1803 &fs_info
->generic_worker
);
1805 /* a higher idle thresh on the submit workers makes it much more
1806 * likely that bios will be send down in a sane order to the
1809 fs_info
->submit_workers
.idle_thresh
= 64;
1811 fs_info
->workers
.idle_thresh
= 16;
1812 fs_info
->workers
.ordered
= 1;
1814 fs_info
->delalloc_workers
.idle_thresh
= 2;
1815 fs_info
->delalloc_workers
.ordered
= 1;
1817 btrfs_init_workers(&fs_info
->fixup_workers
, "fixup", 1,
1818 &fs_info
->generic_worker
);
1819 btrfs_init_workers(&fs_info
->endio_workers
, "endio",
1820 fs_info
->thread_pool_size
,
1821 &fs_info
->generic_worker
);
1822 btrfs_init_workers(&fs_info
->endio_meta_workers
, "endio-meta",
1823 fs_info
->thread_pool_size
,
1824 &fs_info
->generic_worker
);
1825 btrfs_init_workers(&fs_info
->endio_meta_write_workers
,
1826 "endio-meta-write", fs_info
->thread_pool_size
,
1827 &fs_info
->generic_worker
);
1828 btrfs_init_workers(&fs_info
->endio_write_workers
, "endio-write",
1829 fs_info
->thread_pool_size
,
1830 &fs_info
->generic_worker
);
1831 btrfs_init_workers(&fs_info
->endio_freespace_worker
, "freespace-write",
1832 1, &fs_info
->generic_worker
);
1833 btrfs_init_workers(&fs_info
->delayed_workers
, "delayed-meta",
1834 fs_info
->thread_pool_size
,
1835 &fs_info
->generic_worker
);
1838 * endios are largely parallel and should have a very
1841 fs_info
->endio_workers
.idle_thresh
= 4;
1842 fs_info
->endio_meta_workers
.idle_thresh
= 4;
1844 fs_info
->endio_write_workers
.idle_thresh
= 2;
1845 fs_info
->endio_meta_write_workers
.idle_thresh
= 2;
1847 btrfs_start_workers(&fs_info
->workers
, 1);
1848 btrfs_start_workers(&fs_info
->generic_worker
, 1);
1849 btrfs_start_workers(&fs_info
->submit_workers
, 1);
1850 btrfs_start_workers(&fs_info
->delalloc_workers
, 1);
1851 btrfs_start_workers(&fs_info
->fixup_workers
, 1);
1852 btrfs_start_workers(&fs_info
->endio_workers
, 1);
1853 btrfs_start_workers(&fs_info
->endio_meta_workers
, 1);
1854 btrfs_start_workers(&fs_info
->endio_meta_write_workers
, 1);
1855 btrfs_start_workers(&fs_info
->endio_write_workers
, 1);
1856 btrfs_start_workers(&fs_info
->endio_freespace_worker
, 1);
1857 btrfs_start_workers(&fs_info
->delayed_workers
, 1);
1859 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
1860 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
1861 4 * 1024 * 1024 / PAGE_CACHE_SIZE
);
1863 nodesize
= btrfs_super_nodesize(disk_super
);
1864 leafsize
= btrfs_super_leafsize(disk_super
);
1865 sectorsize
= btrfs_super_sectorsize(disk_super
);
1866 stripesize
= btrfs_super_stripesize(disk_super
);
1867 tree_root
->nodesize
= nodesize
;
1868 tree_root
->leafsize
= leafsize
;
1869 tree_root
->sectorsize
= sectorsize
;
1870 tree_root
->stripesize
= stripesize
;
1872 sb
->s_blocksize
= sectorsize
;
1873 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
1875 if (strncmp((char *)(&disk_super
->magic
), BTRFS_MAGIC
,
1876 sizeof(disk_super
->magic
))) {
1877 printk(KERN_INFO
"btrfs: valid FS not found on %s\n", sb
->s_id
);
1878 goto fail_sb_buffer
;
1881 mutex_lock(&fs_info
->chunk_mutex
);
1882 ret
= btrfs_read_sys_array(tree_root
);
1883 mutex_unlock(&fs_info
->chunk_mutex
);
1885 printk(KERN_WARNING
"btrfs: failed to read the system "
1886 "array on %s\n", sb
->s_id
);
1887 goto fail_sb_buffer
;
1890 blocksize
= btrfs_level_size(tree_root
,
1891 btrfs_super_chunk_root_level(disk_super
));
1892 generation
= btrfs_super_chunk_root_generation(disk_super
);
1894 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
1895 chunk_root
, fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
1897 chunk_root
->node
= read_tree_block(chunk_root
,
1898 btrfs_super_chunk_root(disk_super
),
1899 blocksize
, generation
);
1900 BUG_ON(!chunk_root
->node
);
1901 if (!test_bit(EXTENT_BUFFER_UPTODATE
, &chunk_root
->node
->bflags
)) {
1902 printk(KERN_WARNING
"btrfs: failed to read chunk root on %s\n",
1904 goto fail_chunk_root
;
1906 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
1907 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
1909 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
1910 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root
->node
),
1913 mutex_lock(&fs_info
->chunk_mutex
);
1914 ret
= btrfs_read_chunk_tree(chunk_root
);
1915 mutex_unlock(&fs_info
->chunk_mutex
);
1917 printk(KERN_WARNING
"btrfs: failed to read chunk tree on %s\n",
1919 goto fail_chunk_root
;
1922 btrfs_close_extra_devices(fs_devices
);
1924 blocksize
= btrfs_level_size(tree_root
,
1925 btrfs_super_root_level(disk_super
));
1926 generation
= btrfs_super_generation(disk_super
);
1928 tree_root
->node
= read_tree_block(tree_root
,
1929 btrfs_super_root(disk_super
),
1930 blocksize
, generation
);
1931 if (!tree_root
->node
)
1932 goto fail_chunk_root
;
1933 if (!test_bit(EXTENT_BUFFER_UPTODATE
, &tree_root
->node
->bflags
)) {
1934 printk(KERN_WARNING
"btrfs: failed to read tree root on %s\n",
1936 goto fail_tree_root
;
1938 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
1939 tree_root
->commit_root
= btrfs_root_node(tree_root
);
1941 ret
= find_and_setup_root(tree_root
, fs_info
,
1942 BTRFS_EXTENT_TREE_OBJECTID
, extent_root
);
1944 goto fail_tree_root
;
1945 extent_root
->track_dirty
= 1;
1947 ret
= find_and_setup_root(tree_root
, fs_info
,
1948 BTRFS_DEV_TREE_OBJECTID
, dev_root
);
1950 goto fail_extent_root
;
1951 dev_root
->track_dirty
= 1;
1953 ret
= find_and_setup_root(tree_root
, fs_info
,
1954 BTRFS_CSUM_TREE_OBJECTID
, csum_root
);
1958 csum_root
->track_dirty
= 1;
1960 fs_info
->generation
= generation
;
1961 fs_info
->last_trans_committed
= generation
;
1962 fs_info
->data_alloc_profile
= (u64
)-1;
1963 fs_info
->metadata_alloc_profile
= (u64
)-1;
1964 fs_info
->system_alloc_profile
= fs_info
->metadata_alloc_profile
;
1966 ret
= btrfs_init_space_info(fs_info
);
1968 printk(KERN_ERR
"Failed to initial space info: %d\n", ret
);
1969 goto fail_block_groups
;
1972 ret
= btrfs_read_block_groups(extent_root
);
1974 printk(KERN_ERR
"Failed to read block groups: %d\n", ret
);
1975 goto fail_block_groups
;
1978 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
1980 if (IS_ERR(fs_info
->cleaner_kthread
))
1981 goto fail_block_groups
;
1983 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
1985 "btrfs-transaction");
1986 if (IS_ERR(fs_info
->transaction_kthread
))
1989 if (!btrfs_test_opt(tree_root
, SSD
) &&
1990 !btrfs_test_opt(tree_root
, NOSSD
) &&
1991 !fs_info
->fs_devices
->rotating
) {
1992 printk(KERN_INFO
"Btrfs detected SSD devices, enabling SSD "
1994 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
1997 /* do not make disk changes in broken FS */
1998 if (btrfs_super_log_root(disk_super
) != 0 &&
1999 !(fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
)) {
2000 u64 bytenr
= btrfs_super_log_root(disk_super
);
2002 if (fs_devices
->rw_devices
== 0) {
2003 printk(KERN_WARNING
"Btrfs log replay required "
2006 goto fail_trans_kthread
;
2009 btrfs_level_size(tree_root
,
2010 btrfs_super_log_root_level(disk_super
));
2012 log_tree_root
= kzalloc(sizeof(struct btrfs_root
), GFP_NOFS
);
2013 if (!log_tree_root
) {
2015 goto fail_trans_kthread
;
2018 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
2019 log_tree_root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
2021 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
2024 ret
= btrfs_recover_log_trees(log_tree_root
);
2027 if (sb
->s_flags
& MS_RDONLY
) {
2028 ret
= btrfs_commit_super(tree_root
);
2033 ret
= btrfs_find_orphan_roots(tree_root
);
2036 if (!(sb
->s_flags
& MS_RDONLY
)) {
2037 ret
= btrfs_cleanup_fs_roots(fs_info
);
2040 ret
= btrfs_recover_relocation(tree_root
);
2043 "btrfs: failed to recover relocation\n");
2045 goto fail_trans_kthread
;
2049 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
2050 location
.type
= BTRFS_ROOT_ITEM_KEY
;
2051 location
.offset
= (u64
)-1;
2053 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
2054 if (!fs_info
->fs_root
)
2055 goto fail_trans_kthread
;
2056 if (IS_ERR(fs_info
->fs_root
)) {
2057 err
= PTR_ERR(fs_info
->fs_root
);
2058 goto fail_trans_kthread
;
2061 if (!(sb
->s_flags
& MS_RDONLY
)) {
2062 down_read(&fs_info
->cleanup_work_sem
);
2063 err
= btrfs_orphan_cleanup(fs_info
->fs_root
);
2065 err
= btrfs_orphan_cleanup(fs_info
->tree_root
);
2066 up_read(&fs_info
->cleanup_work_sem
);
2068 close_ctree(tree_root
);
2069 return ERR_PTR(err
);
2076 kthread_stop(fs_info
->transaction_kthread
);
2078 kthread_stop(fs_info
->cleaner_kthread
);
2081 * make sure we're done with the btree inode before we stop our
2084 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
2085 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
2088 btrfs_free_block_groups(fs_info
);
2089 free_extent_buffer(csum_root
->node
);
2090 free_extent_buffer(csum_root
->commit_root
);
2092 free_extent_buffer(dev_root
->node
);
2093 free_extent_buffer(dev_root
->commit_root
);
2095 free_extent_buffer(extent_root
->node
);
2096 free_extent_buffer(extent_root
->commit_root
);
2098 free_extent_buffer(tree_root
->node
);
2099 free_extent_buffer(tree_root
->commit_root
);
2101 free_extent_buffer(chunk_root
->node
);
2102 free_extent_buffer(chunk_root
->commit_root
);
2104 btrfs_stop_workers(&fs_info
->generic_worker
);
2105 btrfs_stop_workers(&fs_info
->fixup_workers
);
2106 btrfs_stop_workers(&fs_info
->delalloc_workers
);
2107 btrfs_stop_workers(&fs_info
->workers
);
2108 btrfs_stop_workers(&fs_info
->endio_workers
);
2109 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
2110 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
2111 btrfs_stop_workers(&fs_info
->endio_write_workers
);
2112 btrfs_stop_workers(&fs_info
->endio_freespace_worker
);
2113 btrfs_stop_workers(&fs_info
->submit_workers
);
2114 btrfs_stop_workers(&fs_info
->delayed_workers
);
2116 kfree(fs_info
->delayed_root
);
2118 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
2119 iput(fs_info
->btree_inode
);
2121 btrfs_close_devices(fs_info
->fs_devices
);
2122 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
2124 bdi_destroy(&fs_info
->bdi
);
2126 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
2134 return ERR_PTR(err
);
2137 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
2139 char b
[BDEVNAME_SIZE
];
2142 set_buffer_uptodate(bh
);
2144 printk_ratelimited(KERN_WARNING
"lost page write due to "
2145 "I/O error on %s\n",
2146 bdevname(bh
->b_bdev
, b
));
2147 /* note, we dont' set_buffer_write_io_error because we have
2148 * our own ways of dealing with the IO errors
2150 clear_buffer_uptodate(bh
);
2156 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
2158 struct buffer_head
*bh
;
2159 struct buffer_head
*latest
= NULL
;
2160 struct btrfs_super_block
*super
;
2165 /* we would like to check all the supers, but that would make
2166 * a btrfs mount succeed after a mkfs from a different FS.
2167 * So, we need to add a special mount option to scan for
2168 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2170 for (i
= 0; i
< 1; i
++) {
2171 bytenr
= btrfs_sb_offset(i
);
2172 if (bytenr
+ 4096 >= i_size_read(bdev
->bd_inode
))
2174 bh
= __bread(bdev
, bytenr
/ 4096, 4096);
2178 super
= (struct btrfs_super_block
*)bh
->b_data
;
2179 if (btrfs_super_bytenr(super
) != bytenr
||
2180 strncmp((char *)(&super
->magic
), BTRFS_MAGIC
,
2181 sizeof(super
->magic
))) {
2186 if (!latest
|| btrfs_super_generation(super
) > transid
) {
2189 transid
= btrfs_super_generation(super
);
2198 * this should be called twice, once with wait == 0 and
2199 * once with wait == 1. When wait == 0 is done, all the buffer heads
2200 * we write are pinned.
2202 * They are released when wait == 1 is done.
2203 * max_mirrors must be the same for both runs, and it indicates how
2204 * many supers on this one device should be written.
2206 * max_mirrors == 0 means to write them all.
2208 static int write_dev_supers(struct btrfs_device
*device
,
2209 struct btrfs_super_block
*sb
,
2210 int do_barriers
, int wait
, int max_mirrors
)
2212 struct buffer_head
*bh
;
2218 int last_barrier
= 0;
2220 if (max_mirrors
== 0)
2221 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
2223 /* make sure only the last submit_bh does a barrier */
2225 for (i
= 0; i
< max_mirrors
; i
++) {
2226 bytenr
= btrfs_sb_offset(i
);
2227 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
2228 device
->total_bytes
)
2234 for (i
= 0; i
< max_mirrors
; i
++) {
2235 bytenr
= btrfs_sb_offset(i
);
2236 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= device
->total_bytes
)
2240 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
2241 BTRFS_SUPER_INFO_SIZE
);
2244 if (!buffer_uptodate(bh
))
2247 /* drop our reference */
2250 /* drop the reference from the wait == 0 run */
2254 btrfs_set_super_bytenr(sb
, bytenr
);
2257 crc
= btrfs_csum_data(NULL
, (char *)sb
+
2258 BTRFS_CSUM_SIZE
, crc
,
2259 BTRFS_SUPER_INFO_SIZE
-
2261 btrfs_csum_final(crc
, sb
->csum
);
2264 * one reference for us, and we leave it for the
2267 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
2268 BTRFS_SUPER_INFO_SIZE
);
2269 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
2271 /* one reference for submit_bh */
2274 set_buffer_uptodate(bh
);
2276 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
2279 if (i
== last_barrier
&& do_barriers
)
2280 ret
= submit_bh(WRITE_FLUSH_FUA
, bh
);
2282 ret
= submit_bh(WRITE_SYNC
, bh
);
2287 return errors
< i
? 0 : -1;
2290 int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
2292 struct list_head
*head
;
2293 struct btrfs_device
*dev
;
2294 struct btrfs_super_block
*sb
;
2295 struct btrfs_dev_item
*dev_item
;
2299 int total_errors
= 0;
2302 max_errors
= btrfs_super_num_devices(&root
->fs_info
->super_copy
) - 1;
2303 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
2305 sb
= &root
->fs_info
->super_for_commit
;
2306 dev_item
= &sb
->dev_item
;
2308 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2309 head
= &root
->fs_info
->fs_devices
->devices
;
2310 list_for_each_entry_rcu(dev
, head
, dev_list
) {
2315 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
2318 btrfs_set_stack_device_generation(dev_item
, 0);
2319 btrfs_set_stack_device_type(dev_item
, dev
->type
);
2320 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
2321 btrfs_set_stack_device_total_bytes(dev_item
, dev
->total_bytes
);
2322 btrfs_set_stack_device_bytes_used(dev_item
, dev
->bytes_used
);
2323 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
2324 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
2325 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
2326 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
2327 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
2329 flags
= btrfs_super_flags(sb
);
2330 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
2332 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
2336 if (total_errors
> max_errors
) {
2337 printk(KERN_ERR
"btrfs: %d errors while writing supers\n",
2343 list_for_each_entry_rcu(dev
, head
, dev_list
) {
2346 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
2349 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
2353 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2354 if (total_errors
> max_errors
) {
2355 printk(KERN_ERR
"btrfs: %d errors while writing supers\n",
2362 int write_ctree_super(struct btrfs_trans_handle
*trans
,
2363 struct btrfs_root
*root
, int max_mirrors
)
2367 ret
= write_all_supers(root
, max_mirrors
);
2371 int btrfs_free_fs_root(struct btrfs_fs_info
*fs_info
, struct btrfs_root
*root
)
2373 spin_lock(&fs_info
->fs_roots_radix_lock
);
2374 radix_tree_delete(&fs_info
->fs_roots_radix
,
2375 (unsigned long)root
->root_key
.objectid
);
2376 spin_unlock(&fs_info
->fs_roots_radix_lock
);
2378 if (btrfs_root_refs(&root
->root_item
) == 0)
2379 synchronize_srcu(&fs_info
->subvol_srcu
);
2381 __btrfs_remove_free_space_cache(root
->free_ino_pinned
);
2382 __btrfs_remove_free_space_cache(root
->free_ino_ctl
);
2387 static void free_fs_root(struct btrfs_root
*root
)
2389 iput(root
->cache_inode
);
2390 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
2392 free_anon_bdev(root
->anon_dev
);
2393 free_extent_buffer(root
->node
);
2394 free_extent_buffer(root
->commit_root
);
2395 kfree(root
->free_ino_ctl
);
2396 kfree(root
->free_ino_pinned
);
2401 static int del_fs_roots(struct btrfs_fs_info
*fs_info
)
2404 struct btrfs_root
*gang
[8];
2407 while (!list_empty(&fs_info
->dead_roots
)) {
2408 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2409 struct btrfs_root
, root_list
);
2410 list_del(&gang
[0]->root_list
);
2412 if (gang
[0]->in_radix
) {
2413 btrfs_free_fs_root(fs_info
, gang
[0]);
2415 free_extent_buffer(gang
[0]->node
);
2416 free_extent_buffer(gang
[0]->commit_root
);
2422 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2427 for (i
= 0; i
< ret
; i
++)
2428 btrfs_free_fs_root(fs_info
, gang
[i
]);
2433 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
2435 u64 root_objectid
= 0;
2436 struct btrfs_root
*gang
[8];
2441 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2442 (void **)gang
, root_objectid
,
2447 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
2448 for (i
= 0; i
< ret
; i
++) {
2451 root_objectid
= gang
[i
]->root_key
.objectid
;
2452 err
= btrfs_orphan_cleanup(gang
[i
]);
2461 int btrfs_commit_super(struct btrfs_root
*root
)
2463 struct btrfs_trans_handle
*trans
;
2466 mutex_lock(&root
->fs_info
->cleaner_mutex
);
2467 btrfs_run_delayed_iputs(root
);
2468 btrfs_clean_old_snapshots(root
);
2469 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
2471 /* wait until ongoing cleanup work done */
2472 down_write(&root
->fs_info
->cleanup_work_sem
);
2473 up_write(&root
->fs_info
->cleanup_work_sem
);
2475 trans
= btrfs_join_transaction(root
);
2477 return PTR_ERR(trans
);
2478 ret
= btrfs_commit_transaction(trans
, root
);
2480 /* run commit again to drop the original snapshot */
2481 trans
= btrfs_join_transaction(root
);
2483 return PTR_ERR(trans
);
2484 btrfs_commit_transaction(trans
, root
);
2485 ret
= btrfs_write_and_wait_transaction(NULL
, root
);
2488 ret
= write_ctree_super(NULL
, root
, 0);
2492 int close_ctree(struct btrfs_root
*root
)
2494 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2497 fs_info
->closing
= 1;
2500 btrfs_scrub_cancel(root
);
2502 /* wait for any defraggers to finish */
2503 wait_event(fs_info
->transaction_wait
,
2504 (atomic_read(&fs_info
->defrag_running
) == 0));
2506 /* clear out the rbtree of defraggable inodes */
2507 btrfs_run_defrag_inodes(root
->fs_info
);
2509 btrfs_put_block_group_cache(fs_info
);
2512 * Here come 2 situations when btrfs is broken to flip readonly:
2514 * 1. when btrfs flips readonly somewhere else before
2515 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2516 * and btrfs will skip to write sb directly to keep
2517 * ERROR state on disk.
2519 * 2. when btrfs flips readonly just in btrfs_commit_super,
2520 * and in such case, btrfs cannot write sb via btrfs_commit_super,
2521 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2522 * btrfs will cleanup all FS resources first and write sb then.
2524 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
2525 ret
= btrfs_commit_super(root
);
2527 printk(KERN_ERR
"btrfs: commit super ret %d\n", ret
);
2530 if (fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
2531 ret
= btrfs_error_commit_super(root
);
2533 printk(KERN_ERR
"btrfs: commit super ret %d\n", ret
);
2536 kthread_stop(root
->fs_info
->transaction_kthread
);
2537 kthread_stop(root
->fs_info
->cleaner_kthread
);
2539 fs_info
->closing
= 2;
2542 if (fs_info
->delalloc_bytes
) {
2543 printk(KERN_INFO
"btrfs: at unmount delalloc count %llu\n",
2544 (unsigned long long)fs_info
->delalloc_bytes
);
2546 if (fs_info
->total_ref_cache_size
) {
2547 printk(KERN_INFO
"btrfs: at umount reference cache size %llu\n",
2548 (unsigned long long)fs_info
->total_ref_cache_size
);
2551 free_extent_buffer(fs_info
->extent_root
->node
);
2552 free_extent_buffer(fs_info
->extent_root
->commit_root
);
2553 free_extent_buffer(fs_info
->tree_root
->node
);
2554 free_extent_buffer(fs_info
->tree_root
->commit_root
);
2555 free_extent_buffer(root
->fs_info
->chunk_root
->node
);
2556 free_extent_buffer(root
->fs_info
->chunk_root
->commit_root
);
2557 free_extent_buffer(root
->fs_info
->dev_root
->node
);
2558 free_extent_buffer(root
->fs_info
->dev_root
->commit_root
);
2559 free_extent_buffer(root
->fs_info
->csum_root
->node
);
2560 free_extent_buffer(root
->fs_info
->csum_root
->commit_root
);
2562 btrfs_free_block_groups(root
->fs_info
);
2564 del_fs_roots(fs_info
);
2566 iput(fs_info
->btree_inode
);
2567 kfree(fs_info
->delayed_root
);
2569 btrfs_stop_workers(&fs_info
->generic_worker
);
2570 btrfs_stop_workers(&fs_info
->fixup_workers
);
2571 btrfs_stop_workers(&fs_info
->delalloc_workers
);
2572 btrfs_stop_workers(&fs_info
->workers
);
2573 btrfs_stop_workers(&fs_info
->endio_workers
);
2574 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
2575 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
2576 btrfs_stop_workers(&fs_info
->endio_write_workers
);
2577 btrfs_stop_workers(&fs_info
->endio_freespace_worker
);
2578 btrfs_stop_workers(&fs_info
->submit_workers
);
2579 btrfs_stop_workers(&fs_info
->delayed_workers
);
2581 btrfs_close_devices(fs_info
->fs_devices
);
2582 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
2584 bdi_destroy(&fs_info
->bdi
);
2585 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
2587 kfree(fs_info
->extent_root
);
2588 kfree(fs_info
->tree_root
);
2589 kfree(fs_info
->chunk_root
);
2590 kfree(fs_info
->dev_root
);
2591 kfree(fs_info
->csum_root
);
2597 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
)
2600 struct inode
*btree_inode
= buf
->first_page
->mapping
->host
;
2602 ret
= extent_buffer_uptodate(&BTRFS_I(btree_inode
)->io_tree
, buf
,
2607 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
2612 int btrfs_set_buffer_uptodate(struct extent_buffer
*buf
)
2614 struct inode
*btree_inode
= buf
->first_page
->mapping
->host
;
2615 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode
)->io_tree
,
2619 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
2621 struct btrfs_root
*root
= BTRFS_I(buf
->first_page
->mapping
->host
)->root
;
2622 u64 transid
= btrfs_header_generation(buf
);
2623 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
2626 btrfs_assert_tree_locked(buf
);
2627 if (transid
!= root
->fs_info
->generation
) {
2628 printk(KERN_CRIT
"btrfs transid mismatch buffer %llu, "
2629 "found %llu running %llu\n",
2630 (unsigned long long)buf
->start
,
2631 (unsigned long long)transid
,
2632 (unsigned long long)root
->fs_info
->generation
);
2635 was_dirty
= set_extent_buffer_dirty(&BTRFS_I(btree_inode
)->io_tree
,
2638 spin_lock(&root
->fs_info
->delalloc_lock
);
2639 root
->fs_info
->dirty_metadata_bytes
+= buf
->len
;
2640 spin_unlock(&root
->fs_info
->delalloc_lock
);
2644 void btrfs_btree_balance_dirty(struct btrfs_root
*root
, unsigned long nr
)
2647 * looks as though older kernels can get into trouble with
2648 * this code, they end up stuck in balance_dirty_pages forever
2651 unsigned long thresh
= 32 * 1024 * 1024;
2653 if (current
->flags
& PF_MEMALLOC
)
2656 btrfs_balance_delayed_items(root
);
2658 num_dirty
= root
->fs_info
->dirty_metadata_bytes
;
2660 if (num_dirty
> thresh
) {
2661 balance_dirty_pages_ratelimited_nr(
2662 root
->fs_info
->btree_inode
->i_mapping
, 1);
2667 void __btrfs_btree_balance_dirty(struct btrfs_root
*root
, unsigned long nr
)
2670 * looks as though older kernels can get into trouble with
2671 * this code, they end up stuck in balance_dirty_pages forever
2674 unsigned long thresh
= 32 * 1024 * 1024;
2676 if (current
->flags
& PF_MEMALLOC
)
2679 num_dirty
= root
->fs_info
->dirty_metadata_bytes
;
2681 if (num_dirty
> thresh
) {
2682 balance_dirty_pages_ratelimited_nr(
2683 root
->fs_info
->btree_inode
->i_mapping
, 1);
2688 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
2690 struct btrfs_root
*root
= BTRFS_I(buf
->first_page
->mapping
->host
)->root
;
2692 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
2694 set_bit(EXTENT_BUFFER_UPTODATE
, &buf
->bflags
);
2698 int btree_lock_page_hook(struct page
*page
)
2700 struct inode
*inode
= page
->mapping
->host
;
2701 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2702 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2703 struct extent_buffer
*eb
;
2705 u64 bytenr
= page_offset(page
);
2707 if (page
->private == EXTENT_PAGE_PRIVATE
)
2710 len
= page
->private >> 2;
2711 eb
= find_extent_buffer(io_tree
, bytenr
, len
);
2715 btrfs_tree_lock(eb
);
2716 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
2718 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
2719 spin_lock(&root
->fs_info
->delalloc_lock
);
2720 if (root
->fs_info
->dirty_metadata_bytes
>= eb
->len
)
2721 root
->fs_info
->dirty_metadata_bytes
-= eb
->len
;
2724 spin_unlock(&root
->fs_info
->delalloc_lock
);
2727 btrfs_tree_unlock(eb
);
2728 free_extent_buffer(eb
);
2734 static void btrfs_check_super_valid(struct btrfs_fs_info
*fs_info
,
2740 if (fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
)
2741 printk(KERN_WARNING
"warning: mount fs with errors, "
2742 "running btrfsck is recommended\n");
2745 int btrfs_error_commit_super(struct btrfs_root
*root
)
2749 mutex_lock(&root
->fs_info
->cleaner_mutex
);
2750 btrfs_run_delayed_iputs(root
);
2751 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
2753 down_write(&root
->fs_info
->cleanup_work_sem
);
2754 up_write(&root
->fs_info
->cleanup_work_sem
);
2756 /* cleanup FS via transaction */
2757 btrfs_cleanup_transaction(root
);
2759 ret
= write_ctree_super(NULL
, root
, 0);
2764 static int btrfs_destroy_ordered_operations(struct btrfs_root
*root
)
2766 struct btrfs_inode
*btrfs_inode
;
2767 struct list_head splice
;
2769 INIT_LIST_HEAD(&splice
);
2771 mutex_lock(&root
->fs_info
->ordered_operations_mutex
);
2772 spin_lock(&root
->fs_info
->ordered_extent_lock
);
2774 list_splice_init(&root
->fs_info
->ordered_operations
, &splice
);
2775 while (!list_empty(&splice
)) {
2776 btrfs_inode
= list_entry(splice
.next
, struct btrfs_inode
,
2777 ordered_operations
);
2779 list_del_init(&btrfs_inode
->ordered_operations
);
2781 btrfs_invalidate_inodes(btrfs_inode
->root
);
2784 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
2785 mutex_unlock(&root
->fs_info
->ordered_operations_mutex
);
2790 static int btrfs_destroy_ordered_extents(struct btrfs_root
*root
)
2792 struct list_head splice
;
2793 struct btrfs_ordered_extent
*ordered
;
2794 struct inode
*inode
;
2796 INIT_LIST_HEAD(&splice
);
2798 spin_lock(&root
->fs_info
->ordered_extent_lock
);
2800 list_splice_init(&root
->fs_info
->ordered_extents
, &splice
);
2801 while (!list_empty(&splice
)) {
2802 ordered
= list_entry(splice
.next
, struct btrfs_ordered_extent
,
2805 list_del_init(&ordered
->root_extent_list
);
2806 atomic_inc(&ordered
->refs
);
2808 /* the inode may be getting freed (in sys_unlink path). */
2809 inode
= igrab(ordered
->inode
);
2811 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
2815 atomic_set(&ordered
->refs
, 1);
2816 btrfs_put_ordered_extent(ordered
);
2818 spin_lock(&root
->fs_info
->ordered_extent_lock
);
2821 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
2826 static int btrfs_destroy_delayed_refs(struct btrfs_transaction
*trans
,
2827 struct btrfs_root
*root
)
2829 struct rb_node
*node
;
2830 struct btrfs_delayed_ref_root
*delayed_refs
;
2831 struct btrfs_delayed_ref_node
*ref
;
2834 delayed_refs
= &trans
->delayed_refs
;
2836 spin_lock(&delayed_refs
->lock
);
2837 if (delayed_refs
->num_entries
== 0) {
2838 spin_unlock(&delayed_refs
->lock
);
2839 printk(KERN_INFO
"delayed_refs has NO entry\n");
2843 node
= rb_first(&delayed_refs
->root
);
2845 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2846 node
= rb_next(node
);
2849 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
2850 delayed_refs
->num_entries
--;
2852 atomic_set(&ref
->refs
, 1);
2853 if (btrfs_delayed_ref_is_head(ref
)) {
2854 struct btrfs_delayed_ref_head
*head
;
2856 head
= btrfs_delayed_node_to_head(ref
);
2857 mutex_lock(&head
->mutex
);
2858 kfree(head
->extent_op
);
2859 delayed_refs
->num_heads
--;
2860 if (list_empty(&head
->cluster
))
2861 delayed_refs
->num_heads_ready
--;
2862 list_del_init(&head
->cluster
);
2863 mutex_unlock(&head
->mutex
);
2866 spin_unlock(&delayed_refs
->lock
);
2867 btrfs_put_delayed_ref(ref
);
2870 spin_lock(&delayed_refs
->lock
);
2873 spin_unlock(&delayed_refs
->lock
);
2878 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction
*t
)
2880 struct btrfs_pending_snapshot
*snapshot
;
2881 struct list_head splice
;
2883 INIT_LIST_HEAD(&splice
);
2885 list_splice_init(&t
->pending_snapshots
, &splice
);
2887 while (!list_empty(&splice
)) {
2888 snapshot
= list_entry(splice
.next
,
2889 struct btrfs_pending_snapshot
,
2892 list_del_init(&snapshot
->list
);
2900 static int btrfs_destroy_delalloc_inodes(struct btrfs_root
*root
)
2902 struct btrfs_inode
*btrfs_inode
;
2903 struct list_head splice
;
2905 INIT_LIST_HEAD(&splice
);
2907 spin_lock(&root
->fs_info
->delalloc_lock
);
2908 list_splice_init(&root
->fs_info
->delalloc_inodes
, &splice
);
2910 while (!list_empty(&splice
)) {
2911 btrfs_inode
= list_entry(splice
.next
, struct btrfs_inode
,
2914 list_del_init(&btrfs_inode
->delalloc_inodes
);
2916 btrfs_invalidate_inodes(btrfs_inode
->root
);
2919 spin_unlock(&root
->fs_info
->delalloc_lock
);
2924 static int btrfs_destroy_marked_extents(struct btrfs_root
*root
,
2925 struct extent_io_tree
*dirty_pages
,
2930 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
2931 struct extent_buffer
*eb
;
2935 unsigned long index
;
2938 ret
= find_first_extent_bit(dirty_pages
, start
, &start
, &end
,
2943 clear_extent_bits(dirty_pages
, start
, end
, mark
, GFP_NOFS
);
2944 while (start
<= end
) {
2945 index
= start
>> PAGE_CACHE_SHIFT
;
2946 start
= (u64
)(index
+ 1) << PAGE_CACHE_SHIFT
;
2947 page
= find_get_page(btree_inode
->i_mapping
, index
);
2950 offset
= page_offset(page
);
2952 spin_lock(&dirty_pages
->buffer_lock
);
2953 eb
= radix_tree_lookup(
2954 &(&BTRFS_I(page
->mapping
->host
)->io_tree
)->buffer
,
2955 offset
>> PAGE_CACHE_SHIFT
);
2956 spin_unlock(&dirty_pages
->buffer_lock
);
2958 ret
= test_and_clear_bit(EXTENT_BUFFER_DIRTY
,
2960 atomic_set(&eb
->refs
, 1);
2962 if (PageWriteback(page
))
2963 end_page_writeback(page
);
2966 if (PageDirty(page
)) {
2967 clear_page_dirty_for_io(page
);
2968 spin_lock_irq(&page
->mapping
->tree_lock
);
2969 radix_tree_tag_clear(&page
->mapping
->page_tree
,
2971 PAGECACHE_TAG_DIRTY
);
2972 spin_unlock_irq(&page
->mapping
->tree_lock
);
2975 page
->mapping
->a_ops
->invalidatepage(page
, 0);
2983 static int btrfs_destroy_pinned_extent(struct btrfs_root
*root
,
2984 struct extent_io_tree
*pinned_extents
)
2986 struct extent_io_tree
*unpin
;
2991 unpin
= pinned_extents
;
2993 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
2999 if (btrfs_test_opt(root
, DISCARD
))
3000 ret
= btrfs_error_discard_extent(root
, start
,
3004 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
3005 btrfs_error_unpin_extent_range(root
, start
, end
);
3012 static int btrfs_cleanup_transaction(struct btrfs_root
*root
)
3014 struct btrfs_transaction
*t
;
3019 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
3021 spin_lock(&root
->fs_info
->trans_lock
);
3022 list_splice_init(&root
->fs_info
->trans_list
, &list
);
3023 root
->fs_info
->trans_no_join
= 1;
3024 spin_unlock(&root
->fs_info
->trans_lock
);
3026 while (!list_empty(&list
)) {
3027 t
= list_entry(list
.next
, struct btrfs_transaction
, list
);
3031 btrfs_destroy_ordered_operations(root
);
3033 btrfs_destroy_ordered_extents(root
);
3035 btrfs_destroy_delayed_refs(t
, root
);
3037 btrfs_block_rsv_release(root
,
3038 &root
->fs_info
->trans_block_rsv
,
3039 t
->dirty_pages
.dirty_bytes
);
3041 /* FIXME: cleanup wait for commit */
3044 if (waitqueue_active(&root
->fs_info
->transaction_blocked_wait
))
3045 wake_up(&root
->fs_info
->transaction_blocked_wait
);
3048 if (waitqueue_active(&root
->fs_info
->transaction_wait
))
3049 wake_up(&root
->fs_info
->transaction_wait
);
3052 if (waitqueue_active(&t
->commit_wait
))
3053 wake_up(&t
->commit_wait
);
3055 btrfs_destroy_pending_snapshots(t
);
3057 btrfs_destroy_delalloc_inodes(root
);
3059 spin_lock(&root
->fs_info
->trans_lock
);
3060 root
->fs_info
->running_transaction
= NULL
;
3061 spin_unlock(&root
->fs_info
->trans_lock
);
3063 btrfs_destroy_marked_extents(root
, &t
->dirty_pages
,
3066 btrfs_destroy_pinned_extent(root
,
3067 root
->fs_info
->pinned_extents
);
3069 atomic_set(&t
->use_count
, 0);
3070 list_del_init(&t
->list
);
3071 memset(t
, 0, sizeof(*t
));
3072 kmem_cache_free(btrfs_transaction_cachep
, t
);
3075 spin_lock(&root
->fs_info
->trans_lock
);
3076 root
->fs_info
->trans_no_join
= 0;
3077 spin_unlock(&root
->fs_info
->trans_lock
);
3078 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
3083 static struct extent_io_ops btree_extent_io_ops
= {
3084 .write_cache_pages_lock_hook
= btree_lock_page_hook
,
3085 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
3086 .submit_bio_hook
= btree_submit_bio_hook
,
3087 /* note we're sharing with inode.c for the merge bio hook */
3088 .merge_bio_hook
= btrfs_merge_bio_hook
,