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.
19 #include <linux/version.h>
21 #include <linux/blkdev.h>
22 #include <linux/scatterlist.h>
23 #include <linux/swap.h>
24 #include <linux/radix-tree.h>
25 #include <linux/writeback.h>
26 #include <linux/buffer_head.h> // for block_sync_page
27 #include <linux/workqueue.h>
28 #include <linux/kthread.h>
29 #include <linux/freezer.h>
34 #include "transaction.h"
35 #include "btrfs_inode.h"
37 #include "print-tree.h"
38 #include "async-thread.h"
40 #include "ref-cache.h"
44 static int check_tree_block(struct btrfs_root
*root
, struct extent_buffer
*buf
)
46 if (extent_buffer_blocknr(buf
) != btrfs_header_blocknr(buf
)) {
47 printk(KERN_CRIT
"buf blocknr(buf) is %llu, header is %llu\n",
48 (unsigned long long)extent_buffer_blocknr(buf
),
49 (unsigned long long)btrfs_header_blocknr(buf
));
56 static struct extent_io_ops btree_extent_io_ops
;
57 static void end_workqueue_fn(struct btrfs_work
*work
);
60 * end_io_wq structs are used to do processing in task context when an IO is
61 * complete. This is used during reads to verify checksums, and it is used
62 * by writes to insert metadata for new file extents after IO is complete.
68 struct btrfs_fs_info
*info
;
71 struct list_head list
;
72 struct btrfs_work work
;
76 * async submit bios are used to offload expensive checksumming
77 * onto the worker threads. They checksum file and metadata bios
78 * just before they are sent down the IO stack.
80 struct async_submit_bio
{
83 struct list_head list
;
84 extent_submit_bio_hook_t
*submit_bio_start
;
85 extent_submit_bio_hook_t
*submit_bio_done
;
88 unsigned long bio_flags
;
89 struct btrfs_work work
;
93 * extents on the btree inode are pretty simple, there's one extent
94 * that covers the entire device
96 static struct extent_map
*btree_get_extent(struct inode
*inode
,
97 struct page
*page
, size_t page_offset
, u64 start
, u64 len
,
100 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
101 struct extent_map
*em
;
104 spin_lock(&em_tree
->lock
);
105 em
= lookup_extent_mapping(em_tree
, start
, len
);
108 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
109 spin_unlock(&em_tree
->lock
);
112 spin_unlock(&em_tree
->lock
);
114 em
= alloc_extent_map(GFP_NOFS
);
116 em
= ERR_PTR(-ENOMEM
);
121 em
->block_len
= (u64
)-1;
123 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
125 spin_lock(&em_tree
->lock
);
126 ret
= add_extent_mapping(em_tree
, em
);
127 if (ret
== -EEXIST
) {
128 u64 failed_start
= em
->start
;
129 u64 failed_len
= em
->len
;
131 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
132 em
->start
, em
->len
, em
->block_start
);
134 em
= lookup_extent_mapping(em_tree
, start
, len
);
136 printk("after failing, found %Lu %Lu %Lu\n",
137 em
->start
, em
->len
, em
->block_start
);
140 em
= lookup_extent_mapping(em_tree
, failed_start
,
143 printk("double failure lookup gives us "
144 "%Lu %Lu -> %Lu\n", em
->start
,
145 em
->len
, em
->block_start
);
154 spin_unlock(&em_tree
->lock
);
162 u32
btrfs_csum_data(struct btrfs_root
*root
, char *data
, u32 seed
, size_t len
)
164 return btrfs_crc32c(seed
, data
, len
);
167 void btrfs_csum_final(u32 crc
, char *result
)
169 *(__le32
*)result
= ~cpu_to_le32(crc
);
173 * compute the csum for a btree block, and either verify it or write it
174 * into the csum field of the block.
176 static int csum_tree_block(struct btrfs_root
*root
, struct extent_buffer
*buf
,
180 btrfs_super_csum_size(&root
->fs_info
->super_copy
);
183 unsigned long cur_len
;
184 unsigned long offset
= BTRFS_CSUM_SIZE
;
185 char *map_token
= NULL
;
187 unsigned long map_start
;
188 unsigned long map_len
;
191 unsigned long inline_result
;
193 len
= buf
->len
- offset
;
195 err
= map_private_extent_buffer(buf
, offset
, 32,
197 &map_start
, &map_len
, KM_USER0
);
199 printk("failed to map extent buffer! %lu\n",
203 cur_len
= min(len
, map_len
- (offset
- map_start
));
204 crc
= btrfs_csum_data(root
, kaddr
+ offset
- map_start
,
208 unmap_extent_buffer(buf
, map_token
, KM_USER0
);
210 if (csum_size
> sizeof(inline_result
)) {
211 result
= kzalloc(csum_size
* sizeof(char), GFP_NOFS
);
215 result
= (char *)&inline_result
;
218 btrfs_csum_final(crc
, result
);
221 /* FIXME, this is not good */
222 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
225 memcpy(&found
, result
, csum_size
);
227 read_extent_buffer(buf
, &val
, 0, csum_size
);
228 printk("btrfs: %s checksum verify failed on %llu "
229 "wanted %X found %X level %d\n",
230 root
->fs_info
->sb
->s_id
,
231 buf
->start
, val
, found
, btrfs_header_level(buf
));
232 if (result
!= (char *)&inline_result
)
237 write_extent_buffer(buf
, result
, 0, csum_size
);
239 if (result
!= (char *)&inline_result
)
245 * we can't consider a given block up to date unless the transid of the
246 * block matches the transid in the parent node's pointer. This is how we
247 * detect blocks that either didn't get written at all or got written
248 * in the wrong place.
250 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
251 struct extent_buffer
*eb
, u64 parent_transid
)
255 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
258 lock_extent(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1, GFP_NOFS
);
259 if (extent_buffer_uptodate(io_tree
, eb
) &&
260 btrfs_header_generation(eb
) == parent_transid
) {
264 printk("parent transid verify failed on %llu wanted %llu found %llu\n",
265 (unsigned long long)eb
->start
,
266 (unsigned long long)parent_transid
,
267 (unsigned long long)btrfs_header_generation(eb
));
269 clear_extent_buffer_uptodate(io_tree
, eb
);
271 unlock_extent(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
277 * helper to read a given tree block, doing retries as required when
278 * the checksums don't match and we have alternate mirrors to try.
280 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
281 struct extent_buffer
*eb
,
282 u64 start
, u64 parent_transid
)
284 struct extent_io_tree
*io_tree
;
289 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
291 ret
= read_extent_buffer_pages(io_tree
, eb
, start
, 1,
292 btree_get_extent
, mirror_num
);
294 !verify_parent_transid(io_tree
, eb
, parent_transid
))
296 printk("read extent buffer pages failed with ret %d mirror no %d\n", ret
, mirror_num
);
297 num_copies
= btrfs_num_copies(&root
->fs_info
->mapping_tree
,
303 if (mirror_num
> num_copies
)
310 * checksum a dirty tree block before IO. This has extra checks to make
311 * sure we only fill in the checksum field in the first page of a multi-page block
313 static int csum_dirty_buffer(struct btrfs_root
*root
, struct page
*page
)
315 struct extent_io_tree
*tree
;
316 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
320 struct extent_buffer
*eb
;
323 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
325 if (page
->private == EXTENT_PAGE_PRIVATE
)
329 len
= page
->private >> 2;
333 eb
= alloc_extent_buffer(tree
, start
, len
, page
, GFP_NOFS
);
334 ret
= btree_read_extent_buffer_pages(root
, eb
, start
+ PAGE_CACHE_SIZE
,
335 btrfs_header_generation(eb
));
337 found_start
= btrfs_header_bytenr(eb
);
338 if (found_start
!= start
) {
339 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
340 start
, found_start
, len
);
344 if (eb
->first_page
!= page
) {
345 printk("bad first page %lu %lu\n", eb
->first_page
->index
,
350 if (!PageUptodate(page
)) {
351 printk("csum not up to date page %lu\n", page
->index
);
355 found_level
= btrfs_header_level(eb
);
357 csum_tree_block(root
, eb
, 0);
359 free_extent_buffer(eb
);
364 static int check_tree_block_fsid(struct btrfs_root
*root
,
365 struct extent_buffer
*eb
)
367 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
368 u8 fsid
[BTRFS_UUID_SIZE
];
371 read_extent_buffer(eb
, fsid
, (unsigned long)btrfs_header_fsid(eb
),
374 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
378 fs_devices
= fs_devices
->seed
;
383 static int btree_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
384 struct extent_state
*state
)
386 struct extent_io_tree
*tree
;
390 struct extent_buffer
*eb
;
391 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
394 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
395 if (page
->private == EXTENT_PAGE_PRIVATE
)
399 len
= page
->private >> 2;
403 eb
= alloc_extent_buffer(tree
, start
, len
, page
, GFP_NOFS
);
405 found_start
= btrfs_header_bytenr(eb
);
406 if (found_start
!= start
) {
407 printk("bad tree block start %llu %llu\n",
408 (unsigned long long)found_start
,
409 (unsigned long long)eb
->start
);
413 if (eb
->first_page
!= page
) {
414 printk("bad first page %lu %lu\n", eb
->first_page
->index
,
420 if (check_tree_block_fsid(root
, eb
)) {
421 printk("bad fsid on block %Lu\n", eb
->start
);
425 found_level
= btrfs_header_level(eb
);
427 ret
= csum_tree_block(root
, eb
, 1);
431 end
= min_t(u64
, eb
->len
, PAGE_CACHE_SIZE
);
432 end
= eb
->start
+ end
- 1;
434 free_extent_buffer(eb
);
439 static void end_workqueue_bio(struct bio
*bio
, int err
)
441 struct end_io_wq
*end_io_wq
= bio
->bi_private
;
442 struct btrfs_fs_info
*fs_info
;
444 fs_info
= end_io_wq
->info
;
445 end_io_wq
->error
= err
;
446 end_io_wq
->work
.func
= end_workqueue_fn
;
447 end_io_wq
->work
.flags
= 0;
449 if (bio
->bi_rw
& (1 << BIO_RW
)) {
450 if (end_io_wq
->metadata
)
451 btrfs_queue_worker(&fs_info
->endio_meta_write_workers
,
454 btrfs_queue_worker(&fs_info
->endio_write_workers
,
457 if (end_io_wq
->metadata
)
458 btrfs_queue_worker(&fs_info
->endio_meta_workers
,
461 btrfs_queue_worker(&fs_info
->endio_workers
,
466 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
469 struct end_io_wq
*end_io_wq
;
470 end_io_wq
= kmalloc(sizeof(*end_io_wq
), GFP_NOFS
);
474 end_io_wq
->private = bio
->bi_private
;
475 end_io_wq
->end_io
= bio
->bi_end_io
;
476 end_io_wq
->info
= info
;
477 end_io_wq
->error
= 0;
478 end_io_wq
->bio
= bio
;
479 end_io_wq
->metadata
= metadata
;
481 bio
->bi_private
= end_io_wq
;
482 bio
->bi_end_io
= end_workqueue_bio
;
486 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
488 unsigned long limit
= min_t(unsigned long,
489 info
->workers
.max_workers
,
490 info
->fs_devices
->open_devices
);
494 int btrfs_congested_async(struct btrfs_fs_info
*info
, int iodone
)
496 return atomic_read(&info
->nr_async_bios
) >
497 btrfs_async_submit_limit(info
);
500 static void run_one_async_start(struct btrfs_work
*work
)
502 struct btrfs_fs_info
*fs_info
;
503 struct async_submit_bio
*async
;
505 async
= container_of(work
, struct async_submit_bio
, work
);
506 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
507 async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
508 async
->mirror_num
, async
->bio_flags
);
511 static void run_one_async_done(struct btrfs_work
*work
)
513 struct btrfs_fs_info
*fs_info
;
514 struct async_submit_bio
*async
;
517 async
= container_of(work
, struct async_submit_bio
, work
);
518 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
520 limit
= btrfs_async_submit_limit(fs_info
);
521 limit
= limit
* 2 / 3;
523 atomic_dec(&fs_info
->nr_async_submits
);
525 if (atomic_read(&fs_info
->nr_async_submits
) < limit
&&
526 waitqueue_active(&fs_info
->async_submit_wait
))
527 wake_up(&fs_info
->async_submit_wait
);
529 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
530 async
->mirror_num
, async
->bio_flags
);
533 static void run_one_async_free(struct btrfs_work
*work
)
535 struct async_submit_bio
*async
;
537 async
= container_of(work
, struct async_submit_bio
, work
);
541 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
542 int rw
, struct bio
*bio
, int mirror_num
,
543 unsigned long bio_flags
,
544 extent_submit_bio_hook_t
*submit_bio_start
,
545 extent_submit_bio_hook_t
*submit_bio_done
)
547 struct async_submit_bio
*async
;
549 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
553 async
->inode
= inode
;
556 async
->mirror_num
= mirror_num
;
557 async
->submit_bio_start
= submit_bio_start
;
558 async
->submit_bio_done
= submit_bio_done
;
560 async
->work
.func
= run_one_async_start
;
561 async
->work
.ordered_func
= run_one_async_done
;
562 async
->work
.ordered_free
= run_one_async_free
;
564 async
->work
.flags
= 0;
565 async
->bio_flags
= bio_flags
;
567 atomic_inc(&fs_info
->nr_async_submits
);
568 btrfs_queue_worker(&fs_info
->workers
, &async
->work
);
570 int limit
= btrfs_async_submit_limit(fs_info
);
571 if (atomic_read(&fs_info
->nr_async_submits
) > limit
) {
572 wait_event_timeout(fs_info
->async_submit_wait
,
573 (atomic_read(&fs_info
->nr_async_submits
) < limit
),
576 wait_event_timeout(fs_info
->async_submit_wait
,
577 (atomic_read(&fs_info
->nr_async_bios
) < limit
),
581 while(atomic_read(&fs_info
->async_submit_draining
) &&
582 atomic_read(&fs_info
->nr_async_submits
)) {
583 wait_event(fs_info
->async_submit_wait
,
584 (atomic_read(&fs_info
->nr_async_submits
) == 0));
590 static int btree_csum_one_bio(struct bio
*bio
)
592 struct bio_vec
*bvec
= bio
->bi_io_vec
;
594 struct btrfs_root
*root
;
596 WARN_ON(bio
->bi_vcnt
<= 0);
597 while(bio_index
< bio
->bi_vcnt
) {
598 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
599 csum_dirty_buffer(root
, bvec
->bv_page
);
606 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
607 struct bio
*bio
, int mirror_num
,
608 unsigned long bio_flags
)
611 * when we're called for a write, we're already in the async
612 * submission context. Just jump into btrfs_map_bio
614 btree_csum_one_bio(bio
);
618 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
619 int mirror_num
, unsigned long bio_flags
)
622 * when we're called for a write, we're already in the async
623 * submission context. Just jump into btrfs_map_bio
625 return btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
628 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
629 int mirror_num
, unsigned long bio_flags
)
633 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
637 if (!(rw
& (1 << BIO_RW
))) {
639 * called for a read, do the setup so that checksum validation
640 * can happen in the async kernel threads
642 return btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
646 * kthread helpers are used to submit writes so that checksumming
647 * can happen in parallel across all CPUs
649 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
650 inode
, rw
, bio
, mirror_num
, 0,
651 __btree_submit_bio_start
,
652 __btree_submit_bio_done
);
655 static int btree_writepage(struct page
*page
, struct writeback_control
*wbc
)
657 struct extent_io_tree
*tree
;
658 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
660 if (current
->flags
& PF_MEMALLOC
) {
661 redirty_page_for_writepage(wbc
, page
);
665 return extent_write_full_page(tree
, page
, btree_get_extent
, wbc
);
668 static int btree_writepages(struct address_space
*mapping
,
669 struct writeback_control
*wbc
)
671 struct extent_io_tree
*tree
;
672 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
673 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
676 unsigned long thresh
= 32 * 1024 * 1024;
678 if (wbc
->for_kupdate
)
681 num_dirty
= count_range_bits(tree
, &start
, (u64
)-1,
682 thresh
, EXTENT_DIRTY
);
683 if (num_dirty
< thresh
) {
687 return extent_writepages(tree
, mapping
, btree_get_extent
, wbc
);
690 static int btree_readpage(struct file
*file
, struct page
*page
)
692 struct extent_io_tree
*tree
;
693 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
694 return extent_read_full_page(tree
, page
, btree_get_extent
);
697 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
699 struct extent_io_tree
*tree
;
700 struct extent_map_tree
*map
;
703 if (PageWriteback(page
) || PageDirty(page
))
706 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
707 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
709 ret
= try_release_extent_state(map
, tree
, page
, gfp_flags
);
714 ret
= try_release_extent_buffer(tree
, page
);
716 ClearPagePrivate(page
);
717 set_page_private(page
, 0);
718 page_cache_release(page
);
724 static void btree_invalidatepage(struct page
*page
, unsigned long offset
)
726 struct extent_io_tree
*tree
;
727 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
728 extent_invalidatepage(tree
, page
, offset
);
729 btree_releasepage(page
, GFP_NOFS
);
730 if (PagePrivate(page
)) {
731 printk("warning page private not zero on page %Lu\n",
733 ClearPagePrivate(page
);
734 set_page_private(page
, 0);
735 page_cache_release(page
);
740 static int btree_writepage(struct page
*page
, struct writeback_control
*wbc
)
742 struct buffer_head
*bh
;
743 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
744 struct buffer_head
*head
;
745 if (!page_has_buffers(page
)) {
746 create_empty_buffers(page
, root
->fs_info
->sb
->s_blocksize
,
747 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
749 head
= page_buffers(page
);
752 if (buffer_dirty(bh
))
753 csum_tree_block(root
, bh
, 0);
754 bh
= bh
->b_this_page
;
755 } while (bh
!= head
);
756 return block_write_full_page(page
, btree_get_block
, wbc
);
760 static struct address_space_operations btree_aops
= {
761 .readpage
= btree_readpage
,
762 .writepage
= btree_writepage
,
763 .writepages
= btree_writepages
,
764 .releasepage
= btree_releasepage
,
765 .invalidatepage
= btree_invalidatepage
,
766 .sync_page
= block_sync_page
,
769 int readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
772 struct extent_buffer
*buf
= NULL
;
773 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
776 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
779 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
780 buf
, 0, 0, btree_get_extent
, 0);
781 free_extent_buffer(buf
);
785 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_root
*root
,
786 u64 bytenr
, u32 blocksize
)
788 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
789 struct extent_buffer
*eb
;
790 eb
= find_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
791 bytenr
, blocksize
, GFP_NOFS
);
795 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
796 u64 bytenr
, u32 blocksize
)
798 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
799 struct extent_buffer
*eb
;
801 eb
= alloc_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
802 bytenr
, blocksize
, NULL
, GFP_NOFS
);
807 int btrfs_write_tree_block(struct extent_buffer
*buf
)
809 return btrfs_fdatawrite_range(buf
->first_page
->mapping
, buf
->start
,
810 buf
->start
+ buf
->len
- 1, WB_SYNC_ALL
);
813 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
815 return btrfs_wait_on_page_writeback_range(buf
->first_page
->mapping
,
816 buf
->start
, buf
->start
+ buf
->len
-1);
819 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
820 u32 blocksize
, u64 parent_transid
)
822 struct extent_buffer
*buf
= NULL
;
823 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
824 struct extent_io_tree
*io_tree
;
827 io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
829 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
833 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
836 buf
->flags
|= EXTENT_UPTODATE
;
844 int clean_tree_block(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
845 struct extent_buffer
*buf
)
847 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
848 if (btrfs_header_generation(buf
) ==
849 root
->fs_info
->running_transaction
->transid
) {
850 WARN_ON(!btrfs_tree_locked(buf
));
851 clear_extent_buffer_dirty(&BTRFS_I(btree_inode
)->io_tree
,
857 static int __setup_root(u32 nodesize
, u32 leafsize
, u32 sectorsize
,
858 u32 stripesize
, struct btrfs_root
*root
,
859 struct btrfs_fs_info
*fs_info
,
863 root
->commit_root
= NULL
;
864 root
->ref_tree
= NULL
;
865 root
->sectorsize
= sectorsize
;
866 root
->nodesize
= nodesize
;
867 root
->leafsize
= leafsize
;
868 root
->stripesize
= stripesize
;
870 root
->track_dirty
= 0;
872 root
->fs_info
= fs_info
;
873 root
->objectid
= objectid
;
874 root
->last_trans
= 0;
875 root
->highest_inode
= 0;
876 root
->last_inode_alloc
= 0;
880 INIT_LIST_HEAD(&root
->dirty_list
);
881 INIT_LIST_HEAD(&root
->orphan_list
);
882 INIT_LIST_HEAD(&root
->dead_list
);
883 spin_lock_init(&root
->node_lock
);
884 spin_lock_init(&root
->list_lock
);
885 mutex_init(&root
->objectid_mutex
);
886 mutex_init(&root
->log_mutex
);
887 extent_io_tree_init(&root
->dirty_log_pages
,
888 fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
890 btrfs_leaf_ref_tree_init(&root
->ref_tree_struct
);
891 root
->ref_tree
= &root
->ref_tree_struct
;
893 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
894 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
895 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
896 memset(&root
->root_kobj
, 0, sizeof(root
->root_kobj
));
897 root
->defrag_trans_start
= fs_info
->generation
;
898 init_completion(&root
->kobj_unregister
);
899 root
->defrag_running
= 0;
900 root
->defrag_level
= 0;
901 root
->root_key
.objectid
= objectid
;
902 root
->anon_super
.s_root
= NULL
;
903 root
->anon_super
.s_dev
= 0;
904 INIT_LIST_HEAD(&root
->anon_super
.s_list
);
905 INIT_LIST_HEAD(&root
->anon_super
.s_instances
);
906 init_rwsem(&root
->anon_super
.s_umount
);
911 static int find_and_setup_root(struct btrfs_root
*tree_root
,
912 struct btrfs_fs_info
*fs_info
,
914 struct btrfs_root
*root
)
920 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
921 tree_root
->sectorsize
, tree_root
->stripesize
,
922 root
, fs_info
, objectid
);
923 ret
= btrfs_find_last_root(tree_root
, objectid
,
924 &root
->root_item
, &root
->root_key
);
927 generation
= btrfs_root_generation(&root
->root_item
);
928 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
929 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
930 blocksize
, generation
);
935 int btrfs_free_log_root_tree(struct btrfs_trans_handle
*trans
,
936 struct btrfs_fs_info
*fs_info
)
938 struct extent_buffer
*eb
;
939 struct btrfs_root
*log_root_tree
= fs_info
->log_root_tree
;
948 ret
= find_first_extent_bit(&log_root_tree
->dirty_log_pages
,
949 0, &start
, &end
, EXTENT_DIRTY
);
953 clear_extent_dirty(&log_root_tree
->dirty_log_pages
,
954 start
, end
, GFP_NOFS
);
956 eb
= fs_info
->log_root_tree
->node
;
958 WARN_ON(btrfs_header_level(eb
) != 0);
959 WARN_ON(btrfs_header_nritems(eb
) != 0);
961 ret
= btrfs_free_reserved_extent(fs_info
->tree_root
,
965 free_extent_buffer(eb
);
966 kfree(fs_info
->log_root_tree
);
967 fs_info
->log_root_tree
= NULL
;
971 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
972 struct btrfs_fs_info
*fs_info
)
974 struct btrfs_root
*root
;
975 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
977 root
= kzalloc(sizeof(*root
), GFP_NOFS
);
981 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
982 tree_root
->sectorsize
, tree_root
->stripesize
,
983 root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
985 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
986 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
987 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
990 root
->node
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
,
991 0, BTRFS_TREE_LOG_OBJECTID
,
992 trans
->transid
, 0, 0, 0);
994 btrfs_set_header_nritems(root
->node
, 0);
995 btrfs_set_header_level(root
->node
, 0);
996 btrfs_set_header_bytenr(root
->node
, root
->node
->start
);
997 btrfs_set_header_generation(root
->node
, trans
->transid
);
998 btrfs_set_header_owner(root
->node
, BTRFS_TREE_LOG_OBJECTID
);
1000 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1001 (unsigned long)btrfs_header_fsid(root
->node
),
1003 btrfs_mark_buffer_dirty(root
->node
);
1004 btrfs_tree_unlock(root
->node
);
1005 fs_info
->log_root_tree
= root
;
1009 struct btrfs_root
*btrfs_read_fs_root_no_radix(struct btrfs_root
*tree_root
,
1010 struct btrfs_key
*location
)
1012 struct btrfs_root
*root
;
1013 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1014 struct btrfs_path
*path
;
1015 struct extent_buffer
*l
;
1021 root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1023 return ERR_PTR(-ENOMEM
);
1024 if (location
->offset
== (u64
)-1) {
1025 ret
= find_and_setup_root(tree_root
, fs_info
,
1026 location
->objectid
, root
);
1029 return ERR_PTR(ret
);
1034 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1035 tree_root
->sectorsize
, tree_root
->stripesize
,
1036 root
, fs_info
, location
->objectid
);
1038 path
= btrfs_alloc_path();
1040 ret
= btrfs_search_slot(NULL
, tree_root
, location
, path
, 0, 0);
1047 read_extent_buffer(l
, &root
->root_item
,
1048 btrfs_item_ptr_offset(l
, path
->slots
[0]),
1049 sizeof(root
->root_item
));
1050 memcpy(&root
->root_key
, location
, sizeof(*location
));
1053 btrfs_release_path(root
, path
);
1054 btrfs_free_path(path
);
1057 return ERR_PTR(ret
);
1059 generation
= btrfs_root_generation(&root
->root_item
);
1060 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1061 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1062 blocksize
, generation
);
1063 BUG_ON(!root
->node
);
1065 if (location
->objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1067 ret
= btrfs_find_highest_inode(root
, &highest_inode
);
1069 root
->highest_inode
= highest_inode
;
1070 root
->last_inode_alloc
= highest_inode
;
1076 struct btrfs_root
*btrfs_lookup_fs_root(struct btrfs_fs_info
*fs_info
,
1079 struct btrfs_root
*root
;
1081 if (root_objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1082 return fs_info
->tree_root
;
1083 if (root_objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1084 return fs_info
->extent_root
;
1086 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1087 (unsigned long)root_objectid
);
1091 struct btrfs_root
*btrfs_read_fs_root_no_name(struct btrfs_fs_info
*fs_info
,
1092 struct btrfs_key
*location
)
1094 struct btrfs_root
*root
;
1097 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1098 return fs_info
->tree_root
;
1099 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1100 return fs_info
->extent_root
;
1101 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1102 return fs_info
->chunk_root
;
1103 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1104 return fs_info
->dev_root
;
1105 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1106 return fs_info
->csum_root
;
1108 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1109 (unsigned long)location
->objectid
);
1113 root
= btrfs_read_fs_root_no_radix(fs_info
->tree_root
, location
);
1117 set_anon_super(&root
->anon_super
, NULL
);
1119 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1120 (unsigned long)root
->root_key
.objectid
,
1123 free_extent_buffer(root
->node
);
1125 return ERR_PTR(ret
);
1127 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
1128 ret
= btrfs_find_dead_roots(fs_info
->tree_root
,
1129 root
->root_key
.objectid
, root
);
1131 btrfs_orphan_cleanup(root
);
1136 struct btrfs_root
*btrfs_read_fs_root(struct btrfs_fs_info
*fs_info
,
1137 struct btrfs_key
*location
,
1138 const char *name
, int namelen
)
1140 struct btrfs_root
*root
;
1143 root
= btrfs_read_fs_root_no_name(fs_info
, location
);
1150 ret
= btrfs_set_root_name(root
, name
, namelen
);
1152 free_extent_buffer(root
->node
);
1154 return ERR_PTR(ret
);
1157 ret
= btrfs_sysfs_add_root(root
);
1159 free_extent_buffer(root
->node
);
1162 return ERR_PTR(ret
);
1169 static int add_hasher(struct btrfs_fs_info
*info
, char *type
) {
1170 struct btrfs_hasher
*hasher
;
1172 hasher
= kmalloc(sizeof(*hasher
), GFP_NOFS
);
1175 hasher
->hash_tfm
= crypto_alloc_hash(type
, 0, CRYPTO_ALG_ASYNC
);
1176 if (!hasher
->hash_tfm
) {
1180 spin_lock(&info
->hash_lock
);
1181 list_add(&hasher
->list
, &info
->hashers
);
1182 spin_unlock(&info
->hash_lock
);
1187 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1189 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1191 struct list_head
*cur
;
1192 struct btrfs_device
*device
;
1193 struct backing_dev_info
*bdi
;
1195 if ((bdi_bits
& (1 << BDI_write_congested
)) &&
1196 btrfs_congested_async(info
, 0))
1199 list_for_each(cur
, &info
->fs_devices
->devices
) {
1200 device
= list_entry(cur
, struct btrfs_device
, dev_list
);
1203 bdi
= blk_get_backing_dev_info(device
->bdev
);
1204 if (bdi
&& bdi_congested(bdi
, bdi_bits
)) {
1213 * this unplugs every device on the box, and it is only used when page
1216 static void __unplug_io_fn(struct backing_dev_info
*bdi
, struct page
*page
)
1218 struct list_head
*cur
;
1219 struct btrfs_device
*device
;
1220 struct btrfs_fs_info
*info
;
1222 info
= (struct btrfs_fs_info
*)bdi
->unplug_io_data
;
1223 list_for_each(cur
, &info
->fs_devices
->devices
) {
1224 device
= list_entry(cur
, struct btrfs_device
, dev_list
);
1228 bdi
= blk_get_backing_dev_info(device
->bdev
);
1229 if (bdi
->unplug_io_fn
) {
1230 bdi
->unplug_io_fn(bdi
, page
);
1235 static void btrfs_unplug_io_fn(struct backing_dev_info
*bdi
, struct page
*page
)
1237 struct inode
*inode
;
1238 struct extent_map_tree
*em_tree
;
1239 struct extent_map
*em
;
1240 struct address_space
*mapping
;
1243 /* the generic O_DIRECT read code does this */
1245 __unplug_io_fn(bdi
, page
);
1250 * page->mapping may change at any time. Get a consistent copy
1251 * and use that for everything below
1254 mapping
= page
->mapping
;
1258 inode
= mapping
->host
;
1261 * don't do the expensive searching for a small number of
1264 if (BTRFS_I(inode
)->root
->fs_info
->fs_devices
->open_devices
<= 2) {
1265 __unplug_io_fn(bdi
, page
);
1269 offset
= page_offset(page
);
1271 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1272 spin_lock(&em_tree
->lock
);
1273 em
= lookup_extent_mapping(em_tree
, offset
, PAGE_CACHE_SIZE
);
1274 spin_unlock(&em_tree
->lock
);
1276 __unplug_io_fn(bdi
, page
);
1280 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
1281 free_extent_map(em
);
1282 __unplug_io_fn(bdi
, page
);
1285 offset
= offset
- em
->start
;
1286 btrfs_unplug_page(&BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1287 em
->block_start
+ offset
, page
);
1288 free_extent_map(em
);
1291 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1294 bdi
->ra_pages
= default_backing_dev_info
.ra_pages
;
1296 bdi
->capabilities
= default_backing_dev_info
.capabilities
;
1297 bdi
->unplug_io_fn
= btrfs_unplug_io_fn
;
1298 bdi
->unplug_io_data
= info
;
1299 bdi
->congested_fn
= btrfs_congested_fn
;
1300 bdi
->congested_data
= info
;
1304 static int bio_ready_for_csum(struct bio
*bio
)
1310 struct extent_io_tree
*io_tree
= NULL
;
1311 struct btrfs_fs_info
*info
= NULL
;
1312 struct bio_vec
*bvec
;
1316 bio_for_each_segment(bvec
, bio
, i
) {
1317 page
= bvec
->bv_page
;
1318 if (page
->private == EXTENT_PAGE_PRIVATE
) {
1319 length
+= bvec
->bv_len
;
1322 if (!page
->private) {
1323 length
+= bvec
->bv_len
;
1326 length
= bvec
->bv_len
;
1327 buf_len
= page
->private >> 2;
1328 start
= page_offset(page
) + bvec
->bv_offset
;
1329 io_tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1330 info
= BTRFS_I(page
->mapping
->host
)->root
->fs_info
;
1332 /* are we fully contained in this bio? */
1333 if (buf_len
<= length
)
1336 ret
= extent_range_uptodate(io_tree
, start
+ length
,
1337 start
+ buf_len
- 1);
1344 * called by the kthread helper functions to finally call the bio end_io
1345 * functions. This is where read checksum verification actually happens
1347 static void end_workqueue_fn(struct btrfs_work
*work
)
1350 struct end_io_wq
*end_io_wq
;
1351 struct btrfs_fs_info
*fs_info
;
1354 end_io_wq
= container_of(work
, struct end_io_wq
, work
);
1355 bio
= end_io_wq
->bio
;
1356 fs_info
= end_io_wq
->info
;
1358 /* metadata bio reads are special because the whole tree block must
1359 * be checksummed at once. This makes sure the entire block is in
1360 * ram and up to date before trying to verify things. For
1361 * blocksize <= pagesize, it is basically a noop
1363 if (!(bio
->bi_rw
& (1 << BIO_RW
)) && end_io_wq
->metadata
&&
1364 !bio_ready_for_csum(bio
)) {
1365 btrfs_queue_worker(&fs_info
->endio_meta_workers
,
1369 error
= end_io_wq
->error
;
1370 bio
->bi_private
= end_io_wq
->private;
1371 bio
->bi_end_io
= end_io_wq
->end_io
;
1373 bio_endio(bio
, error
);
1376 static int cleaner_kthread(void *arg
)
1378 struct btrfs_root
*root
= arg
;
1382 if (root
->fs_info
->closing
)
1385 vfs_check_frozen(root
->fs_info
->sb
, SB_FREEZE_WRITE
);
1386 mutex_lock(&root
->fs_info
->cleaner_mutex
);
1387 btrfs_clean_old_snapshots(root
);
1388 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1390 if (freezing(current
)) {
1394 if (root
->fs_info
->closing
)
1396 set_current_state(TASK_INTERRUPTIBLE
);
1398 __set_current_state(TASK_RUNNING
);
1400 } while (!kthread_should_stop());
1404 static int transaction_kthread(void *arg
)
1406 struct btrfs_root
*root
= arg
;
1407 struct btrfs_trans_handle
*trans
;
1408 struct btrfs_transaction
*cur
;
1410 unsigned long delay
;
1415 if (root
->fs_info
->closing
)
1419 vfs_check_frozen(root
->fs_info
->sb
, SB_FREEZE_WRITE
);
1420 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1422 if (root
->fs_info
->total_ref_cache_size
> 20 * 1024 * 1024) {
1423 printk("btrfs: total reference cache size %Lu\n",
1424 root
->fs_info
->total_ref_cache_size
);
1427 mutex_lock(&root
->fs_info
->trans_mutex
);
1428 cur
= root
->fs_info
->running_transaction
;
1430 mutex_unlock(&root
->fs_info
->trans_mutex
);
1434 now
= get_seconds();
1435 if (now
< cur
->start_time
|| now
- cur
->start_time
< 30) {
1436 mutex_unlock(&root
->fs_info
->trans_mutex
);
1440 mutex_unlock(&root
->fs_info
->trans_mutex
);
1441 trans
= btrfs_start_transaction(root
, 1);
1442 ret
= btrfs_commit_transaction(trans
, root
);
1444 wake_up_process(root
->fs_info
->cleaner_kthread
);
1445 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1447 if (freezing(current
)) {
1450 if (root
->fs_info
->closing
)
1452 set_current_state(TASK_INTERRUPTIBLE
);
1453 schedule_timeout(delay
);
1454 __set_current_state(TASK_RUNNING
);
1456 } while (!kthread_should_stop());
1460 struct btrfs_root
*open_ctree(struct super_block
*sb
,
1461 struct btrfs_fs_devices
*fs_devices
,
1471 struct btrfs_key location
;
1472 struct buffer_head
*bh
;
1473 struct btrfs_root
*extent_root
= kzalloc(sizeof(struct btrfs_root
),
1475 struct btrfs_root
*csum_root
= kzalloc(sizeof(struct btrfs_root
),
1477 struct btrfs_root
*tree_root
= kzalloc(sizeof(struct btrfs_root
),
1479 struct btrfs_fs_info
*fs_info
= kzalloc(sizeof(*fs_info
),
1481 struct btrfs_root
*chunk_root
= kzalloc(sizeof(struct btrfs_root
),
1483 struct btrfs_root
*dev_root
= kzalloc(sizeof(struct btrfs_root
),
1485 struct btrfs_root
*log_tree_root
;
1490 struct btrfs_super_block
*disk_super
;
1492 if (!extent_root
|| !tree_root
|| !fs_info
||
1493 !chunk_root
|| !dev_root
|| !csum_root
) {
1497 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_NOFS
);
1498 INIT_LIST_HEAD(&fs_info
->trans_list
);
1499 INIT_LIST_HEAD(&fs_info
->dead_roots
);
1500 INIT_LIST_HEAD(&fs_info
->hashers
);
1501 INIT_LIST_HEAD(&fs_info
->delalloc_inodes
);
1502 spin_lock_init(&fs_info
->hash_lock
);
1503 spin_lock_init(&fs_info
->delalloc_lock
);
1504 spin_lock_init(&fs_info
->new_trans_lock
);
1505 spin_lock_init(&fs_info
->ref_cache_lock
);
1507 init_completion(&fs_info
->kobj_unregister
);
1508 fs_info
->tree_root
= tree_root
;
1509 fs_info
->extent_root
= extent_root
;
1510 fs_info
->csum_root
= csum_root
;
1511 fs_info
->chunk_root
= chunk_root
;
1512 fs_info
->dev_root
= dev_root
;
1513 fs_info
->fs_devices
= fs_devices
;
1514 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
1515 INIT_LIST_HEAD(&fs_info
->space_info
);
1516 btrfs_mapping_init(&fs_info
->mapping_tree
);
1517 atomic_set(&fs_info
->nr_async_submits
, 0);
1518 atomic_set(&fs_info
->async_delalloc_pages
, 0);
1519 atomic_set(&fs_info
->async_submit_draining
, 0);
1520 atomic_set(&fs_info
->nr_async_bios
, 0);
1521 atomic_set(&fs_info
->throttles
, 0);
1522 atomic_set(&fs_info
->throttle_gen
, 0);
1524 fs_info
->max_extent
= (u64
)-1;
1525 fs_info
->max_inline
= 8192 * 1024;
1526 setup_bdi(fs_info
, &fs_info
->bdi
);
1527 fs_info
->btree_inode
= new_inode(sb
);
1528 fs_info
->btree_inode
->i_ino
= 1;
1529 fs_info
->btree_inode
->i_nlink
= 1;
1531 fs_info
->thread_pool_size
= min_t(unsigned long,
1532 num_online_cpus() + 2, 8);
1534 INIT_LIST_HEAD(&fs_info
->ordered_extents
);
1535 spin_lock_init(&fs_info
->ordered_extent_lock
);
1537 sb
->s_blocksize
= 4096;
1538 sb
->s_blocksize_bits
= blksize_bits(4096);
1541 * we set the i_size on the btree inode to the max possible int.
1542 * the real end of the address space is determined by all of
1543 * the devices in the system
1545 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
1546 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
1547 fs_info
->btree_inode
->i_mapping
->backing_dev_info
= &fs_info
->bdi
;
1549 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
1550 fs_info
->btree_inode
->i_mapping
,
1552 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
,
1555 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
1557 spin_lock_init(&fs_info
->block_group_cache_lock
);
1558 fs_info
->block_group_cache_tree
.rb_node
= NULL
;
1560 extent_io_tree_init(&fs_info
->pinned_extents
,
1561 fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
1562 extent_io_tree_init(&fs_info
->pending_del
,
1563 fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
1564 extent_io_tree_init(&fs_info
->extent_ins
,
1565 fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
1566 fs_info
->do_barriers
= 1;
1568 INIT_LIST_HEAD(&fs_info
->dead_reloc_roots
);
1569 btrfs_leaf_ref_tree_init(&fs_info
->reloc_ref_tree
);
1570 btrfs_leaf_ref_tree_init(&fs_info
->shared_ref_tree
);
1572 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
1573 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
1574 sizeof(struct btrfs_key
));
1575 insert_inode_hash(fs_info
->btree_inode
);
1577 mutex_init(&fs_info
->trans_mutex
);
1578 mutex_init(&fs_info
->tree_log_mutex
);
1579 mutex_init(&fs_info
->drop_mutex
);
1580 mutex_init(&fs_info
->extent_ins_mutex
);
1581 mutex_init(&fs_info
->pinned_mutex
);
1582 mutex_init(&fs_info
->chunk_mutex
);
1583 mutex_init(&fs_info
->transaction_kthread_mutex
);
1584 mutex_init(&fs_info
->cleaner_mutex
);
1585 mutex_init(&fs_info
->volume_mutex
);
1586 mutex_init(&fs_info
->tree_reloc_mutex
);
1587 init_waitqueue_head(&fs_info
->transaction_throttle
);
1588 init_waitqueue_head(&fs_info
->transaction_wait
);
1589 init_waitqueue_head(&fs_info
->async_submit_wait
);
1590 init_waitqueue_head(&fs_info
->tree_log_wait
);
1591 atomic_set(&fs_info
->tree_log_commit
, 0);
1592 atomic_set(&fs_info
->tree_log_writers
, 0);
1593 fs_info
->tree_log_transid
= 0;
1596 ret
= add_hasher(fs_info
, "crc32c");
1598 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1603 __setup_root(4096, 4096, 4096, 4096, tree_root
,
1604 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
1607 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
1611 memcpy(&fs_info
->super_copy
, bh
->b_data
, sizeof(fs_info
->super_copy
));
1612 memcpy(&fs_info
->super_for_commit
, &fs_info
->super_copy
,
1613 sizeof(fs_info
->super_for_commit
));
1616 memcpy(fs_info
->fsid
, fs_info
->super_copy
.fsid
, BTRFS_FSID_SIZE
);
1618 disk_super
= &fs_info
->super_copy
;
1619 if (!btrfs_super_root(disk_super
))
1622 ret
= btrfs_parse_options(tree_root
, options
);
1628 features
= btrfs_super_incompat_flags(disk_super
) &
1629 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
1631 printk(KERN_ERR
"BTRFS: couldn't mount because of "
1632 "unsupported optional features (%Lx).\n",
1638 features
= btrfs_super_compat_ro_flags(disk_super
) &
1639 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
1640 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
1641 printk(KERN_ERR
"BTRFS: couldn't mount RDWR because of "
1642 "unsupported option features (%Lx).\n",
1649 * we need to start all the end_io workers up front because the
1650 * queue work function gets called at interrupt time, and so it
1651 * cannot dynamically grow.
1653 btrfs_init_workers(&fs_info
->workers
, "worker",
1654 fs_info
->thread_pool_size
);
1656 btrfs_init_workers(&fs_info
->delalloc_workers
, "delalloc",
1657 fs_info
->thread_pool_size
);
1659 btrfs_init_workers(&fs_info
->submit_workers
, "submit",
1660 min_t(u64
, fs_devices
->num_devices
,
1661 fs_info
->thread_pool_size
));
1663 /* a higher idle thresh on the submit workers makes it much more
1664 * likely that bios will be send down in a sane order to the
1667 fs_info
->submit_workers
.idle_thresh
= 64;
1669 fs_info
->workers
.idle_thresh
= 16;
1670 fs_info
->workers
.ordered
= 1;
1672 fs_info
->delalloc_workers
.idle_thresh
= 2;
1673 fs_info
->delalloc_workers
.ordered
= 1;
1675 btrfs_init_workers(&fs_info
->fixup_workers
, "fixup", 1);
1676 btrfs_init_workers(&fs_info
->endio_workers
, "endio",
1677 fs_info
->thread_pool_size
);
1678 btrfs_init_workers(&fs_info
->endio_meta_workers
, "endio-meta",
1679 fs_info
->thread_pool_size
);
1680 btrfs_init_workers(&fs_info
->endio_meta_write_workers
,
1681 "endio-meta-write", fs_info
->thread_pool_size
);
1682 btrfs_init_workers(&fs_info
->endio_write_workers
, "endio-write",
1683 fs_info
->thread_pool_size
);
1686 * endios are largely parallel and should have a very
1689 fs_info
->endio_workers
.idle_thresh
= 4;
1690 fs_info
->endio_write_workers
.idle_thresh
= 64;
1691 fs_info
->endio_meta_write_workers
.idle_thresh
= 64;
1693 btrfs_start_workers(&fs_info
->workers
, 1);
1694 btrfs_start_workers(&fs_info
->submit_workers
, 1);
1695 btrfs_start_workers(&fs_info
->delalloc_workers
, 1);
1696 btrfs_start_workers(&fs_info
->fixup_workers
, 1);
1697 btrfs_start_workers(&fs_info
->endio_workers
, fs_info
->thread_pool_size
);
1698 btrfs_start_workers(&fs_info
->endio_meta_workers
,
1699 fs_info
->thread_pool_size
);
1700 btrfs_start_workers(&fs_info
->endio_meta_write_workers
,
1701 fs_info
->thread_pool_size
);
1702 btrfs_start_workers(&fs_info
->endio_write_workers
,
1703 fs_info
->thread_pool_size
);
1705 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
1706 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
1707 4 * 1024 * 1024 / PAGE_CACHE_SIZE
);
1709 nodesize
= btrfs_super_nodesize(disk_super
);
1710 leafsize
= btrfs_super_leafsize(disk_super
);
1711 sectorsize
= btrfs_super_sectorsize(disk_super
);
1712 stripesize
= btrfs_super_stripesize(disk_super
);
1713 tree_root
->nodesize
= nodesize
;
1714 tree_root
->leafsize
= leafsize
;
1715 tree_root
->sectorsize
= sectorsize
;
1716 tree_root
->stripesize
= stripesize
;
1718 sb
->s_blocksize
= sectorsize
;
1719 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
1721 if (strncmp((char *)(&disk_super
->magic
), BTRFS_MAGIC
,
1722 sizeof(disk_super
->magic
))) {
1723 printk("btrfs: valid FS not found on %s\n", sb
->s_id
);
1724 goto fail_sb_buffer
;
1727 mutex_lock(&fs_info
->chunk_mutex
);
1728 ret
= btrfs_read_sys_array(tree_root
);
1729 mutex_unlock(&fs_info
->chunk_mutex
);
1731 printk("btrfs: failed to read the system array on %s\n",
1733 goto fail_sys_array
;
1736 blocksize
= btrfs_level_size(tree_root
,
1737 btrfs_super_chunk_root_level(disk_super
));
1738 generation
= btrfs_super_chunk_root_generation(disk_super
);
1740 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
1741 chunk_root
, fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
1743 chunk_root
->node
= read_tree_block(chunk_root
,
1744 btrfs_super_chunk_root(disk_super
),
1745 blocksize
, generation
);
1746 BUG_ON(!chunk_root
->node
);
1748 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
1749 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root
->node
),
1752 mutex_lock(&fs_info
->chunk_mutex
);
1753 ret
= btrfs_read_chunk_tree(chunk_root
);
1754 mutex_unlock(&fs_info
->chunk_mutex
);
1756 printk("btrfs: failed to read chunk tree on %s\n", sb
->s_id
);
1757 goto fail_chunk_root
;
1760 btrfs_close_extra_devices(fs_devices
);
1762 blocksize
= btrfs_level_size(tree_root
,
1763 btrfs_super_root_level(disk_super
));
1764 generation
= btrfs_super_generation(disk_super
);
1766 tree_root
->node
= read_tree_block(tree_root
,
1767 btrfs_super_root(disk_super
),
1768 blocksize
, generation
);
1769 if (!tree_root
->node
)
1770 goto fail_chunk_root
;
1773 ret
= find_and_setup_root(tree_root
, fs_info
,
1774 BTRFS_EXTENT_TREE_OBJECTID
, extent_root
);
1776 goto fail_tree_root
;
1777 extent_root
->track_dirty
= 1;
1779 ret
= find_and_setup_root(tree_root
, fs_info
,
1780 BTRFS_DEV_TREE_OBJECTID
, dev_root
);
1781 dev_root
->track_dirty
= 1;
1784 goto fail_extent_root
;
1786 ret
= find_and_setup_root(tree_root
, fs_info
,
1787 BTRFS_CSUM_TREE_OBJECTID
, csum_root
);
1789 goto fail_extent_root
;
1791 csum_root
->track_dirty
= 1;
1793 btrfs_read_block_groups(extent_root
);
1795 fs_info
->generation
= generation
;
1796 fs_info
->last_trans_committed
= generation
;
1797 fs_info
->data_alloc_profile
= (u64
)-1;
1798 fs_info
->metadata_alloc_profile
= (u64
)-1;
1799 fs_info
->system_alloc_profile
= fs_info
->metadata_alloc_profile
;
1800 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
1802 if (!fs_info
->cleaner_kthread
)
1803 goto fail_csum_root
;
1805 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
1807 "btrfs-transaction");
1808 if (!fs_info
->transaction_kthread
)
1811 if (btrfs_super_log_root(disk_super
) != 0) {
1812 u64 bytenr
= btrfs_super_log_root(disk_super
);
1814 if (fs_devices
->rw_devices
== 0) {
1815 printk("Btrfs log replay required on RO media\n");
1817 goto fail_trans_kthread
;
1820 btrfs_level_size(tree_root
,
1821 btrfs_super_log_root_level(disk_super
));
1823 log_tree_root
= kzalloc(sizeof(struct btrfs_root
),
1826 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
1827 log_tree_root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
1829 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
1832 ret
= btrfs_recover_log_trees(log_tree_root
);
1835 if (sb
->s_flags
& MS_RDONLY
) {
1836 ret
= btrfs_commit_super(tree_root
);
1841 if (!(sb
->s_flags
& MS_RDONLY
)) {
1842 ret
= btrfs_cleanup_reloc_trees(tree_root
);
1846 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
1847 location
.type
= BTRFS_ROOT_ITEM_KEY
;
1848 location
.offset
= (u64
)-1;
1850 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
1851 if (!fs_info
->fs_root
)
1852 goto fail_trans_kthread
;
1856 kthread_stop(fs_info
->transaction_kthread
);
1858 kthread_stop(fs_info
->cleaner_kthread
);
1861 * make sure we're done with the btree inode before we stop our
1864 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
1865 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
1868 free_extent_buffer(csum_root
->node
);
1870 free_extent_buffer(extent_root
->node
);
1872 free_extent_buffer(tree_root
->node
);
1874 free_extent_buffer(chunk_root
->node
);
1876 free_extent_buffer(dev_root
->node
);
1878 btrfs_stop_workers(&fs_info
->fixup_workers
);
1879 btrfs_stop_workers(&fs_info
->delalloc_workers
);
1880 btrfs_stop_workers(&fs_info
->workers
);
1881 btrfs_stop_workers(&fs_info
->endio_workers
);
1882 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
1883 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
1884 btrfs_stop_workers(&fs_info
->endio_write_workers
);
1885 btrfs_stop_workers(&fs_info
->submit_workers
);
1887 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
1888 iput(fs_info
->btree_inode
);
1890 btrfs_close_devices(fs_info
->fs_devices
);
1891 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
1895 bdi_destroy(&fs_info
->bdi
);
1900 return ERR_PTR(err
);
1903 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
1905 char b
[BDEVNAME_SIZE
];
1908 set_buffer_uptodate(bh
);
1910 if (!buffer_eopnotsupp(bh
) && printk_ratelimit()) {
1911 printk(KERN_WARNING
"lost page write due to "
1912 "I/O error on %s\n",
1913 bdevname(bh
->b_bdev
, b
));
1915 /* note, we dont' set_buffer_write_io_error because we have
1916 * our own ways of dealing with the IO errors
1918 clear_buffer_uptodate(bh
);
1924 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
1926 struct buffer_head
*bh
;
1927 struct buffer_head
*latest
= NULL
;
1928 struct btrfs_super_block
*super
;
1933 /* we would like to check all the supers, but that would make
1934 * a btrfs mount succeed after a mkfs from a different FS.
1935 * So, we need to add a special mount option to scan for
1936 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1938 for (i
= 0; i
< 1; i
++) {
1939 bytenr
= btrfs_sb_offset(i
);
1940 if (bytenr
+ 4096 >= i_size_read(bdev
->bd_inode
))
1942 bh
= __bread(bdev
, bytenr
/ 4096, 4096);
1946 super
= (struct btrfs_super_block
*)bh
->b_data
;
1947 if (btrfs_super_bytenr(super
) != bytenr
||
1948 strncmp((char *)(&super
->magic
), BTRFS_MAGIC
,
1949 sizeof(super
->magic
))) {
1954 if (!latest
|| btrfs_super_generation(super
) > transid
) {
1957 transid
= btrfs_super_generation(super
);
1965 static int write_dev_supers(struct btrfs_device
*device
,
1966 struct btrfs_super_block
*sb
,
1967 int do_barriers
, int wait
, int max_mirrors
)
1969 struct buffer_head
*bh
;
1975 int last_barrier
= 0;
1977 if (max_mirrors
== 0)
1978 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
1980 /* make sure only the last submit_bh does a barrier */
1982 for (i
= 0; i
< max_mirrors
; i
++) {
1983 bytenr
= btrfs_sb_offset(i
);
1984 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
1985 device
->total_bytes
)
1991 for (i
= 0; i
< max_mirrors
; i
++) {
1992 bytenr
= btrfs_sb_offset(i
);
1993 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= device
->total_bytes
)
1997 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
1998 BTRFS_SUPER_INFO_SIZE
);
2002 if (buffer_uptodate(bh
)) {
2007 btrfs_set_super_bytenr(sb
, bytenr
);
2010 crc
= btrfs_csum_data(NULL
, (char *)sb
+
2011 BTRFS_CSUM_SIZE
, crc
,
2012 BTRFS_SUPER_INFO_SIZE
-
2014 btrfs_csum_final(crc
, sb
->csum
);
2016 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
2017 BTRFS_SUPER_INFO_SIZE
);
2018 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
2020 set_buffer_uptodate(bh
);
2023 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
2026 if (i
== last_barrier
&& do_barriers
&& device
->barriers
) {
2027 ret
= submit_bh(WRITE_BARRIER
, bh
);
2028 if (ret
== -EOPNOTSUPP
) {
2029 printk("btrfs: disabling barriers on dev %s\n",
2031 set_buffer_uptodate(bh
);
2032 device
->barriers
= 0;
2035 ret
= submit_bh(WRITE
, bh
);
2038 ret
= submit_bh(WRITE
, bh
);
2043 if (!buffer_uptodate(bh
))
2051 return errors
< i
? 0 : -1;
2054 int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
2056 struct list_head
*cur
;
2057 struct list_head
*head
= &root
->fs_info
->fs_devices
->devices
;
2058 struct btrfs_device
*dev
;
2059 struct btrfs_super_block
*sb
;
2060 struct btrfs_dev_item
*dev_item
;
2064 int total_errors
= 0;
2067 max_errors
= btrfs_super_num_devices(&root
->fs_info
->super_copy
) - 1;
2068 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
2070 sb
= &root
->fs_info
->super_for_commit
;
2071 dev_item
= &sb
->dev_item
;
2072 list_for_each(cur
, head
) {
2073 dev
= list_entry(cur
, struct btrfs_device
, dev_list
);
2078 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
2081 btrfs_set_stack_device_generation(dev_item
, 0);
2082 btrfs_set_stack_device_type(dev_item
, dev
->type
);
2083 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
2084 btrfs_set_stack_device_total_bytes(dev_item
, dev
->total_bytes
);
2085 btrfs_set_stack_device_bytes_used(dev_item
, dev
->bytes_used
);
2086 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
2087 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
2088 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
2089 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
2090 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
2092 flags
= btrfs_super_flags(sb
);
2093 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
2095 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
2099 if (total_errors
> max_errors
) {
2100 printk("btrfs: %d errors while writing supers\n", total_errors
);
2105 list_for_each(cur
, head
) {
2106 dev
= list_entry(cur
, struct btrfs_device
, dev_list
);
2109 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
2112 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
2116 if (total_errors
> max_errors
) {
2117 printk("btrfs: %d errors while writing supers\n", total_errors
);
2123 int write_ctree_super(struct btrfs_trans_handle
*trans
,
2124 struct btrfs_root
*root
, int max_mirrors
)
2128 ret
= write_all_supers(root
, max_mirrors
);
2132 int btrfs_free_fs_root(struct btrfs_fs_info
*fs_info
, struct btrfs_root
*root
)
2134 radix_tree_delete(&fs_info
->fs_roots_radix
,
2135 (unsigned long)root
->root_key
.objectid
);
2136 if (root
->anon_super
.s_dev
) {
2137 down_write(&root
->anon_super
.s_umount
);
2138 kill_anon_super(&root
->anon_super
);
2142 btrfs_sysfs_del_root(root
);
2145 free_extent_buffer(root
->node
);
2146 if (root
->commit_root
)
2147 free_extent_buffer(root
->commit_root
);
2154 static int del_fs_roots(struct btrfs_fs_info
*fs_info
)
2157 struct btrfs_root
*gang
[8];
2161 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2166 for (i
= 0; i
< ret
; i
++)
2167 btrfs_free_fs_root(fs_info
, gang
[i
]);
2172 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
2174 u64 root_objectid
= 0;
2175 struct btrfs_root
*gang
[8];
2180 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2181 (void **)gang
, root_objectid
,
2185 for (i
= 0; i
< ret
; i
++) {
2186 root_objectid
= gang
[i
]->root_key
.objectid
;
2187 ret
= btrfs_find_dead_roots(fs_info
->tree_root
,
2188 root_objectid
, gang
[i
]);
2190 btrfs_orphan_cleanup(gang
[i
]);
2197 int btrfs_commit_super(struct btrfs_root
*root
)
2199 struct btrfs_trans_handle
*trans
;
2202 mutex_lock(&root
->fs_info
->cleaner_mutex
);
2203 btrfs_clean_old_snapshots(root
);
2204 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
2205 trans
= btrfs_start_transaction(root
, 1);
2206 ret
= btrfs_commit_transaction(trans
, root
);
2208 /* run commit again to drop the original snapshot */
2209 trans
= btrfs_start_transaction(root
, 1);
2210 btrfs_commit_transaction(trans
, root
);
2211 ret
= btrfs_write_and_wait_transaction(NULL
, root
);
2214 ret
= write_ctree_super(NULL
, root
, 0);
2218 int close_ctree(struct btrfs_root
*root
)
2220 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2223 fs_info
->closing
= 1;
2226 kthread_stop(root
->fs_info
->transaction_kthread
);
2227 kthread_stop(root
->fs_info
->cleaner_kthread
);
2229 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
2230 ret
= btrfs_commit_super(root
);
2232 printk("btrfs: commit super returns %d\n", ret
);
2236 if (fs_info
->delalloc_bytes
) {
2237 printk("btrfs: at unmount delalloc count %Lu\n",
2238 fs_info
->delalloc_bytes
);
2240 if (fs_info
->total_ref_cache_size
) {
2241 printk("btrfs: at umount reference cache size %Lu\n",
2242 fs_info
->total_ref_cache_size
);
2245 if (fs_info
->extent_root
->node
)
2246 free_extent_buffer(fs_info
->extent_root
->node
);
2248 if (fs_info
->tree_root
->node
)
2249 free_extent_buffer(fs_info
->tree_root
->node
);
2251 if (root
->fs_info
->chunk_root
->node
);
2252 free_extent_buffer(root
->fs_info
->chunk_root
->node
);
2254 if (root
->fs_info
->dev_root
->node
);
2255 free_extent_buffer(root
->fs_info
->dev_root
->node
);
2257 if (root
->fs_info
->csum_root
->node
);
2258 free_extent_buffer(root
->fs_info
->csum_root
->node
);
2260 btrfs_free_block_groups(root
->fs_info
);
2262 del_fs_roots(fs_info
);
2264 iput(fs_info
->btree_inode
);
2266 btrfs_stop_workers(&fs_info
->fixup_workers
);
2267 btrfs_stop_workers(&fs_info
->delalloc_workers
);
2268 btrfs_stop_workers(&fs_info
->workers
);
2269 btrfs_stop_workers(&fs_info
->endio_workers
);
2270 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
2271 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
2272 btrfs_stop_workers(&fs_info
->endio_write_workers
);
2273 btrfs_stop_workers(&fs_info
->submit_workers
);
2276 while(!list_empty(&fs_info
->hashers
)) {
2277 struct btrfs_hasher
*hasher
;
2278 hasher
= list_entry(fs_info
->hashers
.next
, struct btrfs_hasher
,
2280 list_del(&hasher
->hashers
);
2281 crypto_free_hash(&fs_info
->hash_tfm
);
2285 btrfs_close_devices(fs_info
->fs_devices
);
2286 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
2288 bdi_destroy(&fs_info
->bdi
);
2290 kfree(fs_info
->extent_root
);
2291 kfree(fs_info
->tree_root
);
2292 kfree(fs_info
->chunk_root
);
2293 kfree(fs_info
->dev_root
);
2294 kfree(fs_info
->csum_root
);
2298 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
)
2301 struct inode
*btree_inode
= buf
->first_page
->mapping
->host
;
2303 ret
= extent_buffer_uptodate(&BTRFS_I(btree_inode
)->io_tree
, buf
);
2307 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
2312 int btrfs_set_buffer_uptodate(struct extent_buffer
*buf
)
2314 struct inode
*btree_inode
= buf
->first_page
->mapping
->host
;
2315 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode
)->io_tree
,
2319 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
2321 struct btrfs_root
*root
= BTRFS_I(buf
->first_page
->mapping
->host
)->root
;
2322 u64 transid
= btrfs_header_generation(buf
);
2323 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
2325 WARN_ON(!btrfs_tree_locked(buf
));
2326 if (transid
!= root
->fs_info
->generation
) {
2327 printk(KERN_CRIT
"transid mismatch buffer %llu, found %Lu running %Lu\n",
2328 (unsigned long long)buf
->start
,
2329 transid
, root
->fs_info
->generation
);
2332 set_extent_buffer_dirty(&BTRFS_I(btree_inode
)->io_tree
, buf
);
2335 void btrfs_btree_balance_dirty(struct btrfs_root
*root
, unsigned long nr
)
2338 * looks as though older kernels can get into trouble with
2339 * this code, they end up stuck in balance_dirty_pages forever
2341 struct extent_io_tree
*tree
;
2344 unsigned long thresh
= 32 * 1024 * 1024;
2345 tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
2347 if (current_is_pdflush() || current
->flags
& PF_MEMALLOC
)
2350 num_dirty
= count_range_bits(tree
, &start
, (u64
)-1,
2351 thresh
, EXTENT_DIRTY
);
2352 if (num_dirty
> thresh
) {
2353 balance_dirty_pages_ratelimited_nr(
2354 root
->fs_info
->btree_inode
->i_mapping
, 1);
2359 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
2361 struct btrfs_root
*root
= BTRFS_I(buf
->first_page
->mapping
->host
)->root
;
2363 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
2365 buf
->flags
|= EXTENT_UPTODATE
;
2370 int btree_lock_page_hook(struct page
*page
)
2372 struct inode
*inode
= page
->mapping
->host
;
2373 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2374 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2375 struct extent_buffer
*eb
;
2377 u64 bytenr
= page_offset(page
);
2379 if (page
->private == EXTENT_PAGE_PRIVATE
)
2382 len
= page
->private >> 2;
2383 eb
= find_extent_buffer(io_tree
, bytenr
, len
, GFP_NOFS
);
2387 btrfs_tree_lock(eb
);
2388 spin_lock(&root
->fs_info
->hash_lock
);
2389 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
2390 spin_unlock(&root
->fs_info
->hash_lock
);
2391 btrfs_tree_unlock(eb
);
2392 free_extent_buffer(eb
);
2398 static struct extent_io_ops btree_extent_io_ops
= {
2399 .write_cache_pages_lock_hook
= btree_lock_page_hook
,
2400 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
2401 .submit_bio_hook
= btree_submit_bio_hook
,
2402 /* note we're sharing with inode.c for the merge bio hook */
2403 .merge_bio_hook
= btrfs_merge_bio_hook
,