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 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,20)
30 # include <linux/freezer.h>
32 # include <linux/sched.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "print-tree.h"
41 #include "async-thread.h"
43 #include "ref-cache.h"
47 static int check_tree_block(struct btrfs_root
*root
, struct extent_buffer
*buf
)
49 if (extent_buffer_blocknr(buf
) != btrfs_header_blocknr(buf
)) {
50 printk(KERN_CRIT
"buf blocknr(buf) is %llu, header is %llu\n",
51 (unsigned long long)extent_buffer_blocknr(buf
),
52 (unsigned long long)btrfs_header_blocknr(buf
));
59 static struct extent_io_ops btree_extent_io_ops
;
60 static void end_workqueue_fn(struct btrfs_work
*work
);
66 struct btrfs_fs_info
*info
;
69 struct list_head list
;
70 struct btrfs_work work
;
73 struct async_submit_bio
{
76 struct list_head list
;
77 extent_submit_bio_hook_t
*submit_bio_hook
;
80 struct btrfs_work work
;
83 struct extent_map
*btree_get_extent(struct inode
*inode
, struct page
*page
,
84 size_t page_offset
, u64 start
, u64 len
,
87 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
88 struct extent_map
*em
;
91 spin_lock(&em_tree
->lock
);
92 em
= lookup_extent_mapping(em_tree
, start
, len
);
95 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
96 spin_unlock(&em_tree
->lock
);
99 spin_unlock(&em_tree
->lock
);
101 em
= alloc_extent_map(GFP_NOFS
);
103 em
= ERR_PTR(-ENOMEM
);
109 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
111 spin_lock(&em_tree
->lock
);
112 ret
= add_extent_mapping(em_tree
, em
);
113 if (ret
== -EEXIST
) {
114 u64 failed_start
= em
->start
;
115 u64 failed_len
= em
->len
;
117 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
118 em
->start
, em
->len
, em
->block_start
);
120 em
= lookup_extent_mapping(em_tree
, start
, len
);
122 printk("after failing, found %Lu %Lu %Lu\n",
123 em
->start
, em
->len
, em
->block_start
);
126 em
= lookup_extent_mapping(em_tree
, failed_start
,
129 printk("double failure lookup gives us "
130 "%Lu %Lu -> %Lu\n", em
->start
,
131 em
->len
, em
->block_start
);
140 spin_unlock(&em_tree
->lock
);
148 u32
btrfs_csum_data(struct btrfs_root
*root
, char *data
, u32 seed
, size_t len
)
150 return btrfs_crc32c(seed
, data
, len
);
153 void btrfs_csum_final(u32 crc
, char *result
)
155 *(__le32
*)result
= ~cpu_to_le32(crc
);
158 static int csum_tree_block(struct btrfs_root
*root
, struct extent_buffer
*buf
,
161 char result
[BTRFS_CRC32_SIZE
];
163 unsigned long cur_len
;
164 unsigned long offset
= BTRFS_CSUM_SIZE
;
165 char *map_token
= NULL
;
167 unsigned long map_start
;
168 unsigned long map_len
;
172 len
= buf
->len
- offset
;
174 err
= map_private_extent_buffer(buf
, offset
, 32,
176 &map_start
, &map_len
, KM_USER0
);
178 printk("failed to map extent buffer! %lu\n",
182 cur_len
= min(len
, map_len
- (offset
- map_start
));
183 crc
= btrfs_csum_data(root
, kaddr
+ offset
- map_start
,
187 unmap_extent_buffer(buf
, map_token
, KM_USER0
);
189 btrfs_csum_final(crc
, result
);
192 /* FIXME, this is not good */
193 if (memcmp_extent_buffer(buf
, result
, 0, BTRFS_CRC32_SIZE
)) {
196 memcpy(&found
, result
, BTRFS_CRC32_SIZE
);
198 read_extent_buffer(buf
, &val
, 0, BTRFS_CRC32_SIZE
);
199 printk("btrfs: %s checksum verify failed on %llu "
200 "wanted %X found %X level %d\n",
201 root
->fs_info
->sb
->s_id
,
202 buf
->start
, val
, found
, btrfs_header_level(buf
));
206 write_extent_buffer(buf
, result
, 0, BTRFS_CRC32_SIZE
);
211 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
212 struct extent_buffer
*eb
, u64 parent_transid
)
216 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
219 lock_extent(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1, GFP_NOFS
);
220 if (extent_buffer_uptodate(io_tree
, eb
) &&
221 btrfs_header_generation(eb
) == parent_transid
) {
225 printk("parent transid verify failed on %llu wanted %llu found %llu\n",
226 (unsigned long long)eb
->start
,
227 (unsigned long long)parent_transid
,
228 (unsigned long long)btrfs_header_generation(eb
));
230 clear_extent_buffer_uptodate(io_tree
, eb
);
232 unlock_extent(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
238 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
239 struct extent_buffer
*eb
,
240 u64 start
, u64 parent_transid
)
242 struct extent_io_tree
*io_tree
;
247 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
249 ret
= read_extent_buffer_pages(io_tree
, eb
, start
, 1,
250 btree_get_extent
, mirror_num
);
252 !verify_parent_transid(io_tree
, eb
, parent_transid
))
254 printk("read extent buffer pages failed with ret %d mirror no %d\n", ret
, mirror_num
);
255 num_copies
= btrfs_num_copies(&root
->fs_info
->mapping_tree
,
261 if (mirror_num
> num_copies
)
267 int csum_dirty_buffer(struct btrfs_root
*root
, struct page
*page
)
269 struct extent_io_tree
*tree
;
270 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
274 struct extent_buffer
*eb
;
277 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
279 if (page
->private == EXTENT_PAGE_PRIVATE
)
283 len
= page
->private >> 2;
287 eb
= alloc_extent_buffer(tree
, start
, len
, page
, GFP_NOFS
);
288 ret
= btree_read_extent_buffer_pages(root
, eb
, start
+ PAGE_CACHE_SIZE
,
289 btrfs_header_generation(eb
));
291 found_start
= btrfs_header_bytenr(eb
);
292 if (found_start
!= start
) {
293 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
294 start
, found_start
, len
);
298 if (eb
->first_page
!= page
) {
299 printk("bad first page %lu %lu\n", eb
->first_page
->index
,
304 if (!PageUptodate(page
)) {
305 printk("csum not up to date page %lu\n", page
->index
);
309 found_level
= btrfs_header_level(eb
);
310 spin_lock(&root
->fs_info
->hash_lock
);
311 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
312 spin_unlock(&root
->fs_info
->hash_lock
);
313 csum_tree_block(root
, eb
, 0);
315 free_extent_buffer(eb
);
320 static int btree_writepage_io_hook(struct page
*page
, u64 start
, u64 end
)
322 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
324 csum_dirty_buffer(root
, page
);
328 int btree_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
329 struct extent_state
*state
)
331 struct extent_io_tree
*tree
;
335 struct extent_buffer
*eb
;
336 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
339 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
340 if (page
->private == EXTENT_PAGE_PRIVATE
)
344 len
= page
->private >> 2;
348 eb
= alloc_extent_buffer(tree
, start
, len
, page
, GFP_NOFS
);
350 found_start
= btrfs_header_bytenr(eb
);
351 if (found_start
!= start
) {
352 printk("bad tree block start %llu %llu\n",
353 (unsigned long long)found_start
,
354 (unsigned long long)eb
->start
);
358 if (eb
->first_page
!= page
) {
359 printk("bad first page %lu %lu\n", eb
->first_page
->index
,
365 if (memcmp_extent_buffer(eb
, root
->fs_info
->fsid
,
366 (unsigned long)btrfs_header_fsid(eb
),
368 printk("bad fsid on block %Lu\n", eb
->start
);
372 found_level
= btrfs_header_level(eb
);
374 ret
= csum_tree_block(root
, eb
, 1);
378 end
= min_t(u64
, eb
->len
, PAGE_CACHE_SIZE
);
379 end
= eb
->start
+ end
- 1;
381 free_extent_buffer(eb
);
386 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
387 static void end_workqueue_bio(struct bio
*bio
, int err
)
389 static int end_workqueue_bio(struct bio
*bio
,
390 unsigned int bytes_done
, int err
)
393 struct end_io_wq
*end_io_wq
= bio
->bi_private
;
394 struct btrfs_fs_info
*fs_info
;
396 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
401 fs_info
= end_io_wq
->info
;
402 end_io_wq
->error
= err
;
403 end_io_wq
->work
.func
= end_workqueue_fn
;
404 end_io_wq
->work
.flags
= 0;
405 if (bio
->bi_rw
& (1 << BIO_RW
))
406 btrfs_queue_worker(&fs_info
->endio_write_workers
,
409 btrfs_queue_worker(&fs_info
->endio_workers
, &end_io_wq
->work
);
411 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
416 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
419 struct end_io_wq
*end_io_wq
;
420 end_io_wq
= kmalloc(sizeof(*end_io_wq
), GFP_NOFS
);
424 end_io_wq
->private = bio
->bi_private
;
425 end_io_wq
->end_io
= bio
->bi_end_io
;
426 end_io_wq
->info
= info
;
427 end_io_wq
->error
= 0;
428 end_io_wq
->bio
= bio
;
429 end_io_wq
->metadata
= metadata
;
431 bio
->bi_private
= end_io_wq
;
432 bio
->bi_end_io
= end_workqueue_bio
;
436 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
438 unsigned long limit
= min_t(unsigned long,
439 info
->workers
.max_workers
,
440 info
->fs_devices
->open_devices
);
444 int btrfs_congested_async(struct btrfs_fs_info
*info
, int iodone
)
446 return atomic_read(&info
->nr_async_bios
) >
447 btrfs_async_submit_limit(info
);
450 static void run_one_async_submit(struct btrfs_work
*work
)
452 struct btrfs_fs_info
*fs_info
;
453 struct async_submit_bio
*async
;
456 async
= container_of(work
, struct async_submit_bio
, work
);
457 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
459 limit
= btrfs_async_submit_limit(fs_info
);
460 limit
= limit
* 2 / 3;
462 atomic_dec(&fs_info
->nr_async_submits
);
464 if (atomic_read(&fs_info
->nr_async_submits
) < limit
&&
465 waitqueue_active(&fs_info
->async_submit_wait
))
466 wake_up(&fs_info
->async_submit_wait
);
468 async
->submit_bio_hook(async
->inode
, async
->rw
, async
->bio
,
473 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
474 int rw
, struct bio
*bio
, int mirror_num
,
475 extent_submit_bio_hook_t
*submit_bio_hook
)
477 struct async_submit_bio
*async
;
478 int limit
= btrfs_async_submit_limit(fs_info
);
480 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
484 async
->inode
= inode
;
487 async
->mirror_num
= mirror_num
;
488 async
->submit_bio_hook
= submit_bio_hook
;
489 async
->work
.func
= run_one_async_submit
;
490 async
->work
.flags
= 0;
491 atomic_inc(&fs_info
->nr_async_submits
);
492 btrfs_queue_worker(&fs_info
->workers
, &async
->work
);
494 if (atomic_read(&fs_info
->nr_async_submits
) > limit
) {
495 wait_event_timeout(fs_info
->async_submit_wait
,
496 (atomic_read(&fs_info
->nr_async_submits
) < limit
),
499 wait_event_timeout(fs_info
->async_submit_wait
,
500 (atomic_read(&fs_info
->nr_async_bios
) < limit
),
506 static int __btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
509 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
513 offset
= bio
->bi_sector
<< 9;
516 * when we're called for a write, we're already in the async
517 * submission context. Just jump into btrfs_map_bio
519 if (rw
& (1 << BIO_RW
)) {
520 return btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
525 * called for a read, do the setup so that checksum validation
526 * can happen in the async kernel threads
528 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 1);
531 return btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
534 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
538 * kthread helpers are used to submit writes so that checksumming
539 * can happen in parallel across all CPUs
541 if (!(rw
& (1 << BIO_RW
))) {
542 return __btree_submit_bio_hook(inode
, rw
, bio
, mirror_num
);
544 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
545 inode
, rw
, bio
, mirror_num
,
546 __btree_submit_bio_hook
);
549 static int btree_writepage(struct page
*page
, struct writeback_control
*wbc
)
551 struct extent_io_tree
*tree
;
552 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
554 if (current
->flags
& PF_MEMALLOC
) {
555 redirty_page_for_writepage(wbc
, page
);
559 return extent_write_full_page(tree
, page
, btree_get_extent
, wbc
);
562 static int btree_writepages(struct address_space
*mapping
,
563 struct writeback_control
*wbc
)
565 struct extent_io_tree
*tree
;
566 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
567 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
570 unsigned long thresh
= 8 * 1024 * 1024;
572 if (wbc
->for_kupdate
)
575 num_dirty
= count_range_bits(tree
, &start
, (u64
)-1,
576 thresh
, EXTENT_DIRTY
);
577 if (num_dirty
< thresh
) {
581 return extent_writepages(tree
, mapping
, btree_get_extent
, wbc
);
584 int btree_readpage(struct file
*file
, struct page
*page
)
586 struct extent_io_tree
*tree
;
587 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
588 return extent_read_full_page(tree
, page
, btree_get_extent
);
591 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
593 struct extent_io_tree
*tree
;
594 struct extent_map_tree
*map
;
597 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
598 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
600 ret
= try_release_extent_state(map
, tree
, page
, gfp_flags
);
605 ret
= try_release_extent_buffer(tree
, page
);
607 ClearPagePrivate(page
);
608 set_page_private(page
, 0);
609 page_cache_release(page
);
615 static void btree_invalidatepage(struct page
*page
, unsigned long offset
)
617 struct extent_io_tree
*tree
;
618 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
619 extent_invalidatepage(tree
, page
, offset
);
620 btree_releasepage(page
, GFP_NOFS
);
621 if (PagePrivate(page
)) {
622 printk("warning page private not zero on page %Lu\n",
624 ClearPagePrivate(page
);
625 set_page_private(page
, 0);
626 page_cache_release(page
);
631 static int btree_writepage(struct page
*page
, struct writeback_control
*wbc
)
633 struct buffer_head
*bh
;
634 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
635 struct buffer_head
*head
;
636 if (!page_has_buffers(page
)) {
637 create_empty_buffers(page
, root
->fs_info
->sb
->s_blocksize
,
638 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
640 head
= page_buffers(page
);
643 if (buffer_dirty(bh
))
644 csum_tree_block(root
, bh
, 0);
645 bh
= bh
->b_this_page
;
646 } while (bh
!= head
);
647 return block_write_full_page(page
, btree_get_block
, wbc
);
651 static struct address_space_operations btree_aops
= {
652 .readpage
= btree_readpage
,
653 .writepage
= btree_writepage
,
654 .writepages
= btree_writepages
,
655 .releasepage
= btree_releasepage
,
656 .invalidatepage
= btree_invalidatepage
,
657 .sync_page
= block_sync_page
,
660 int readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
663 struct extent_buffer
*buf
= NULL
;
664 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
667 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
670 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
671 buf
, 0, 0, btree_get_extent
, 0);
672 free_extent_buffer(buf
);
676 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_root
*root
,
677 u64 bytenr
, u32 blocksize
)
679 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
680 struct extent_buffer
*eb
;
681 eb
= find_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
682 bytenr
, blocksize
, GFP_NOFS
);
686 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
687 u64 bytenr
, u32 blocksize
)
689 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
690 struct extent_buffer
*eb
;
692 eb
= alloc_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
693 bytenr
, blocksize
, NULL
, GFP_NOFS
);
698 int btrfs_write_tree_block(struct extent_buffer
*buf
)
700 return btrfs_fdatawrite_range(buf
->first_page
->mapping
, buf
->start
,
701 buf
->start
+ buf
->len
- 1, WB_SYNC_NONE
);
704 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
706 return btrfs_wait_on_page_writeback_range(buf
->first_page
->mapping
,
707 buf
->start
, buf
->start
+ buf
->len
-1);
710 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
711 u32 blocksize
, u64 parent_transid
)
713 struct extent_buffer
*buf
= NULL
;
714 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
715 struct extent_io_tree
*io_tree
;
718 io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
720 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
724 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
727 buf
->flags
|= EXTENT_UPTODATE
;
735 int clean_tree_block(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
736 struct extent_buffer
*buf
)
738 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
739 if (btrfs_header_generation(buf
) ==
740 root
->fs_info
->running_transaction
->transid
) {
741 WARN_ON(!btrfs_tree_locked(buf
));
742 clear_extent_buffer_dirty(&BTRFS_I(btree_inode
)->io_tree
,
748 static int __setup_root(u32 nodesize
, u32 leafsize
, u32 sectorsize
,
749 u32 stripesize
, struct btrfs_root
*root
,
750 struct btrfs_fs_info
*fs_info
,
755 root
->commit_root
= NULL
;
756 root
->ref_tree
= NULL
;
757 root
->sectorsize
= sectorsize
;
758 root
->nodesize
= nodesize
;
759 root
->leafsize
= leafsize
;
760 root
->stripesize
= stripesize
;
762 root
->track_dirty
= 0;
764 root
->fs_info
= fs_info
;
765 root
->objectid
= objectid
;
766 root
->last_trans
= 0;
767 root
->highest_inode
= 0;
768 root
->last_inode_alloc
= 0;
772 INIT_LIST_HEAD(&root
->dirty_list
);
773 INIT_LIST_HEAD(&root
->orphan_list
);
774 INIT_LIST_HEAD(&root
->dead_list
);
775 spin_lock_init(&root
->node_lock
);
776 spin_lock_init(&root
->list_lock
);
777 mutex_init(&root
->objectid_mutex
);
778 mutex_init(&root
->log_mutex
);
780 btrfs_leaf_ref_tree_init(&root
->ref_tree_struct
);
781 root
->ref_tree
= &root
->ref_tree_struct
;
783 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
784 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
785 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
786 memset(&root
->root_kobj
, 0, sizeof(root
->root_kobj
));
787 root
->defrag_trans_start
= fs_info
->generation
;
788 init_completion(&root
->kobj_unregister
);
789 root
->defrag_running
= 0;
790 root
->defrag_level
= 0;
791 root
->root_key
.objectid
= objectid
;
795 static int find_and_setup_root(struct btrfs_root
*tree_root
,
796 struct btrfs_fs_info
*fs_info
,
798 struct btrfs_root
*root
)
803 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
804 tree_root
->sectorsize
, tree_root
->stripesize
,
805 root
, fs_info
, objectid
);
806 ret
= btrfs_find_last_root(tree_root
, objectid
,
807 &root
->root_item
, &root
->root_key
);
810 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
811 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
817 int btrfs_free_log_root_tree(struct btrfs_trans_handle
*trans
,
818 struct btrfs_fs_info
*fs_info
)
820 struct extent_buffer
*eb
;
823 if (!fs_info
->log_root_tree
)
826 eb
= fs_info
->log_root_tree
->node
;
828 WARN_ON(btrfs_header_level(eb
) != 0);
829 WARN_ON(btrfs_header_nritems(eb
) != 0);
831 ret
= btrfs_free_extent(trans
, fs_info
->tree_root
,
833 BTRFS_TREE_LOG_OBJECTID
, 0, 0, 0, 1);
836 free_extent_buffer(eb
);
837 kfree(fs_info
->log_root_tree
);
838 fs_info
->log_root_tree
= NULL
;
842 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
843 struct btrfs_fs_info
*fs_info
)
845 struct btrfs_root
*root
;
846 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
848 root
= kzalloc(sizeof(*root
), GFP_NOFS
);
852 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
853 tree_root
->sectorsize
, tree_root
->stripesize
,
854 root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
856 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
857 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
858 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
861 root
->node
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
,
862 BTRFS_TREE_LOG_OBJECTID
,
865 btrfs_set_header_nritems(root
->node
, 0);
866 btrfs_set_header_level(root
->node
, 0);
867 btrfs_set_header_bytenr(root
->node
, root
->node
->start
);
868 btrfs_set_header_generation(root
->node
, trans
->transid
);
869 btrfs_set_header_owner(root
->node
, BTRFS_TREE_LOG_OBJECTID
);
871 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
872 (unsigned long)btrfs_header_fsid(root
->node
),
874 btrfs_mark_buffer_dirty(root
->node
);
875 btrfs_tree_unlock(root
->node
);
876 fs_info
->log_root_tree
= root
;
880 struct btrfs_root
*btrfs_read_fs_root_no_radix(struct btrfs_root
*tree_root
,
881 struct btrfs_key
*location
)
883 struct btrfs_root
*root
;
884 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
885 struct btrfs_path
*path
;
886 struct extent_buffer
*l
;
891 root
= kzalloc(sizeof(*root
), GFP_NOFS
);
893 return ERR_PTR(-ENOMEM
);
894 if (location
->offset
== (u64
)-1) {
895 ret
= find_and_setup_root(tree_root
, fs_info
,
896 location
->objectid
, root
);
904 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
905 tree_root
->sectorsize
, tree_root
->stripesize
,
906 root
, fs_info
, location
->objectid
);
908 path
= btrfs_alloc_path();
910 ret
= btrfs_search_slot(NULL
, tree_root
, location
, path
, 0, 0);
917 read_extent_buffer(l
, &root
->root_item
,
918 btrfs_item_ptr_offset(l
, path
->slots
[0]),
919 sizeof(root
->root_item
));
920 memcpy(&root
->root_key
, location
, sizeof(*location
));
923 btrfs_release_path(root
, path
);
924 btrfs_free_path(path
);
929 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
930 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
934 if (location
->objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
936 ret
= btrfs_find_highest_inode(root
, &highest_inode
);
938 root
->highest_inode
= highest_inode
;
939 root
->last_inode_alloc
= highest_inode
;
945 struct btrfs_root
*btrfs_lookup_fs_root(struct btrfs_fs_info
*fs_info
,
948 struct btrfs_root
*root
;
950 if (root_objectid
== BTRFS_ROOT_TREE_OBJECTID
)
951 return fs_info
->tree_root
;
952 if (root_objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
953 return fs_info
->extent_root
;
955 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
956 (unsigned long)root_objectid
);
960 struct btrfs_root
*btrfs_read_fs_root_no_name(struct btrfs_fs_info
*fs_info
,
961 struct btrfs_key
*location
)
963 struct btrfs_root
*root
;
966 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
967 return fs_info
->tree_root
;
968 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
969 return fs_info
->extent_root
;
970 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
971 return fs_info
->chunk_root
;
972 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
973 return fs_info
->dev_root
;
975 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
976 (unsigned long)location
->objectid
);
980 root
= btrfs_read_fs_root_no_radix(fs_info
->tree_root
, location
);
983 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
984 (unsigned long)root
->root_key
.objectid
,
987 free_extent_buffer(root
->node
);
991 ret
= btrfs_find_dead_roots(fs_info
->tree_root
,
992 root
->root_key
.objectid
, root
);
998 struct btrfs_root
*btrfs_read_fs_root(struct btrfs_fs_info
*fs_info
,
999 struct btrfs_key
*location
,
1000 const char *name
, int namelen
)
1002 struct btrfs_root
*root
;
1005 root
= btrfs_read_fs_root_no_name(fs_info
, location
);
1012 ret
= btrfs_set_root_name(root
, name
, namelen
);
1014 free_extent_buffer(root
->node
);
1016 return ERR_PTR(ret
);
1019 ret
= btrfs_sysfs_add_root(root
);
1021 free_extent_buffer(root
->node
);
1024 return ERR_PTR(ret
);
1030 static int add_hasher(struct btrfs_fs_info
*info
, char *type
) {
1031 struct btrfs_hasher
*hasher
;
1033 hasher
= kmalloc(sizeof(*hasher
), GFP_NOFS
);
1036 hasher
->hash_tfm
= crypto_alloc_hash(type
, 0, CRYPTO_ALG_ASYNC
);
1037 if (!hasher
->hash_tfm
) {
1041 spin_lock(&info
->hash_lock
);
1042 list_add(&hasher
->list
, &info
->hashers
);
1043 spin_unlock(&info
->hash_lock
);
1048 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1050 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1052 struct list_head
*cur
;
1053 struct btrfs_device
*device
;
1054 struct backing_dev_info
*bdi
;
1056 if ((bdi_bits
& (1 << BDI_write_congested
)) &&
1057 btrfs_congested_async(info
, 0))
1060 list_for_each(cur
, &info
->fs_devices
->devices
) {
1061 device
= list_entry(cur
, struct btrfs_device
, dev_list
);
1064 bdi
= blk_get_backing_dev_info(device
->bdev
);
1065 if (bdi
&& bdi_congested(bdi
, bdi_bits
)) {
1074 * this unplugs every device on the box, and it is only used when page
1077 static void __unplug_io_fn(struct backing_dev_info
*bdi
, struct page
*page
)
1079 struct list_head
*cur
;
1080 struct btrfs_device
*device
;
1081 struct btrfs_fs_info
*info
;
1083 info
= (struct btrfs_fs_info
*)bdi
->unplug_io_data
;
1084 list_for_each(cur
, &info
->fs_devices
->devices
) {
1085 device
= list_entry(cur
, struct btrfs_device
, dev_list
);
1086 bdi
= blk_get_backing_dev_info(device
->bdev
);
1087 if (bdi
->unplug_io_fn
) {
1088 bdi
->unplug_io_fn(bdi
, page
);
1093 void btrfs_unplug_io_fn(struct backing_dev_info
*bdi
, struct page
*page
)
1095 struct inode
*inode
;
1096 struct extent_map_tree
*em_tree
;
1097 struct extent_map
*em
;
1098 struct address_space
*mapping
;
1101 /* the generic O_DIRECT read code does this */
1103 __unplug_io_fn(bdi
, page
);
1108 * page->mapping may change at any time. Get a consistent copy
1109 * and use that for everything below
1112 mapping
= page
->mapping
;
1116 inode
= mapping
->host
;
1117 offset
= page_offset(page
);
1119 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1120 spin_lock(&em_tree
->lock
);
1121 em
= lookup_extent_mapping(em_tree
, offset
, PAGE_CACHE_SIZE
);
1122 spin_unlock(&em_tree
->lock
);
1124 __unplug_io_fn(bdi
, page
);
1128 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
1129 free_extent_map(em
);
1130 __unplug_io_fn(bdi
, page
);
1133 offset
= offset
- em
->start
;
1134 btrfs_unplug_page(&BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1135 em
->block_start
+ offset
, page
);
1136 free_extent_map(em
);
1139 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1141 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1144 bdi
->ra_pages
= default_backing_dev_info
.ra_pages
;
1146 bdi
->capabilities
= default_backing_dev_info
.capabilities
;
1147 bdi
->unplug_io_fn
= btrfs_unplug_io_fn
;
1148 bdi
->unplug_io_data
= info
;
1149 bdi
->congested_fn
= btrfs_congested_fn
;
1150 bdi
->congested_data
= info
;
1154 static int bio_ready_for_csum(struct bio
*bio
)
1160 struct extent_io_tree
*io_tree
= NULL
;
1161 struct btrfs_fs_info
*info
= NULL
;
1162 struct bio_vec
*bvec
;
1166 bio_for_each_segment(bvec
, bio
, i
) {
1167 page
= bvec
->bv_page
;
1168 if (page
->private == EXTENT_PAGE_PRIVATE
) {
1169 length
+= bvec
->bv_len
;
1172 if (!page
->private) {
1173 length
+= bvec
->bv_len
;
1176 length
= bvec
->bv_len
;
1177 buf_len
= page
->private >> 2;
1178 start
= page_offset(page
) + bvec
->bv_offset
;
1179 io_tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1180 info
= BTRFS_I(page
->mapping
->host
)->root
->fs_info
;
1182 /* are we fully contained in this bio? */
1183 if (buf_len
<= length
)
1186 ret
= extent_range_uptodate(io_tree
, start
+ length
,
1187 start
+ buf_len
- 1);
1194 * called by the kthread helper functions to finally call the bio end_io
1195 * functions. This is where read checksum verification actually happens
1197 static void end_workqueue_fn(struct btrfs_work
*work
)
1200 struct end_io_wq
*end_io_wq
;
1201 struct btrfs_fs_info
*fs_info
;
1204 end_io_wq
= container_of(work
, struct end_io_wq
, work
);
1205 bio
= end_io_wq
->bio
;
1206 fs_info
= end_io_wq
->info
;
1208 /* metadata bios are special because the whole tree block must
1209 * be checksummed at once. This makes sure the entire block is in
1210 * ram and up to date before trying to verify things. For
1211 * blocksize <= pagesize, it is basically a noop
1213 if (end_io_wq
->metadata
&& !bio_ready_for_csum(bio
)) {
1214 btrfs_queue_worker(&fs_info
->endio_workers
,
1218 error
= end_io_wq
->error
;
1219 bio
->bi_private
= end_io_wq
->private;
1220 bio
->bi_end_io
= end_io_wq
->end_io
;
1222 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1223 bio_endio(bio
, bio
->bi_size
, error
);
1225 bio_endio(bio
, error
);
1229 static int cleaner_kthread(void *arg
)
1231 struct btrfs_root
*root
= arg
;
1235 if (root
->fs_info
->closing
)
1238 vfs_check_frozen(root
->fs_info
->sb
, SB_FREEZE_WRITE
);
1239 mutex_lock(&root
->fs_info
->cleaner_mutex
);
1240 btrfs_clean_old_snapshots(root
);
1241 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1243 if (freezing(current
)) {
1247 if (root
->fs_info
->closing
)
1249 set_current_state(TASK_INTERRUPTIBLE
);
1251 __set_current_state(TASK_RUNNING
);
1253 } while (!kthread_should_stop());
1257 static int transaction_kthread(void *arg
)
1259 struct btrfs_root
*root
= arg
;
1260 struct btrfs_trans_handle
*trans
;
1261 struct btrfs_transaction
*cur
;
1263 unsigned long delay
;
1268 if (root
->fs_info
->closing
)
1272 vfs_check_frozen(root
->fs_info
->sb
, SB_FREEZE_WRITE
);
1273 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1275 if (root
->fs_info
->total_ref_cache_size
> 20 * 1024 * 1024) {
1276 printk("btrfs: total reference cache size %Lu\n",
1277 root
->fs_info
->total_ref_cache_size
);
1280 mutex_lock(&root
->fs_info
->trans_mutex
);
1281 cur
= root
->fs_info
->running_transaction
;
1283 mutex_unlock(&root
->fs_info
->trans_mutex
);
1287 now
= get_seconds();
1288 if (now
< cur
->start_time
|| now
- cur
->start_time
< 30) {
1289 mutex_unlock(&root
->fs_info
->trans_mutex
);
1293 mutex_unlock(&root
->fs_info
->trans_mutex
);
1294 trans
= btrfs_start_transaction(root
, 1);
1295 ret
= btrfs_commit_transaction(trans
, root
);
1297 wake_up_process(root
->fs_info
->cleaner_kthread
);
1298 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1300 if (freezing(current
)) {
1303 if (root
->fs_info
->closing
)
1305 set_current_state(TASK_INTERRUPTIBLE
);
1306 schedule_timeout(delay
);
1307 __set_current_state(TASK_RUNNING
);
1309 } while (!kthread_should_stop());
1313 struct btrfs_root
*open_ctree(struct super_block
*sb
,
1314 struct btrfs_fs_devices
*fs_devices
,
1322 struct buffer_head
*bh
;
1323 struct btrfs_root
*extent_root
= kzalloc(sizeof(struct btrfs_root
),
1325 struct btrfs_root
*tree_root
= kzalloc(sizeof(struct btrfs_root
),
1327 struct btrfs_fs_info
*fs_info
= kzalloc(sizeof(*fs_info
),
1329 struct btrfs_root
*chunk_root
= kzalloc(sizeof(struct btrfs_root
),
1331 struct btrfs_root
*dev_root
= kzalloc(sizeof(struct btrfs_root
),
1333 struct btrfs_root
*log_tree_root
;
1338 struct btrfs_super_block
*disk_super
;
1340 if (!extent_root
|| !tree_root
|| !fs_info
) {
1344 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_NOFS
);
1345 INIT_LIST_HEAD(&fs_info
->trans_list
);
1346 INIT_LIST_HEAD(&fs_info
->dead_roots
);
1347 INIT_LIST_HEAD(&fs_info
->hashers
);
1348 INIT_LIST_HEAD(&fs_info
->delalloc_inodes
);
1349 spin_lock_init(&fs_info
->hash_lock
);
1350 spin_lock_init(&fs_info
->delalloc_lock
);
1351 spin_lock_init(&fs_info
->new_trans_lock
);
1352 spin_lock_init(&fs_info
->ref_cache_lock
);
1354 init_completion(&fs_info
->kobj_unregister
);
1355 fs_info
->tree_root
= tree_root
;
1356 fs_info
->extent_root
= extent_root
;
1357 fs_info
->chunk_root
= chunk_root
;
1358 fs_info
->dev_root
= dev_root
;
1359 fs_info
->fs_devices
= fs_devices
;
1360 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
1361 INIT_LIST_HEAD(&fs_info
->space_info
);
1362 btrfs_mapping_init(&fs_info
->mapping_tree
);
1363 atomic_set(&fs_info
->nr_async_submits
, 0);
1364 atomic_set(&fs_info
->nr_async_bios
, 0);
1365 atomic_set(&fs_info
->throttles
, 0);
1366 atomic_set(&fs_info
->throttle_gen
, 0);
1368 fs_info
->max_extent
= (u64
)-1;
1369 fs_info
->max_inline
= 8192 * 1024;
1370 setup_bdi(fs_info
, &fs_info
->bdi
);
1371 fs_info
->btree_inode
= new_inode(sb
);
1372 fs_info
->btree_inode
->i_ino
= 1;
1373 fs_info
->btree_inode
->i_nlink
= 1;
1374 fs_info
->thread_pool_size
= min(num_online_cpus() + 2, 8);
1376 INIT_LIST_HEAD(&fs_info
->ordered_extents
);
1377 spin_lock_init(&fs_info
->ordered_extent_lock
);
1379 sb
->s_blocksize
= 4096;
1380 sb
->s_blocksize_bits
= blksize_bits(4096);
1383 * we set the i_size on the btree inode to the max possible int.
1384 * the real end of the address space is determined by all of
1385 * the devices in the system
1387 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
1388 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
1389 fs_info
->btree_inode
->i_mapping
->backing_dev_info
= &fs_info
->bdi
;
1391 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
1392 fs_info
->btree_inode
->i_mapping
,
1394 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
,
1397 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
1399 extent_io_tree_init(&fs_info
->free_space_cache
,
1400 fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
1401 extent_io_tree_init(&fs_info
->block_group_cache
,
1402 fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
1403 extent_io_tree_init(&fs_info
->pinned_extents
,
1404 fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
1405 extent_io_tree_init(&fs_info
->pending_del
,
1406 fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
1407 extent_io_tree_init(&fs_info
->extent_ins
,
1408 fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
1409 fs_info
->do_barriers
= 1;
1411 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
1412 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
1413 sizeof(struct btrfs_key
));
1414 insert_inode_hash(fs_info
->btree_inode
);
1415 mapping_set_gfp_mask(fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
1417 mutex_init(&fs_info
->trans_mutex
);
1418 mutex_init(&fs_info
->tree_log_mutex
);
1419 mutex_init(&fs_info
->drop_mutex
);
1420 mutex_init(&fs_info
->alloc_mutex
);
1421 mutex_init(&fs_info
->chunk_mutex
);
1422 mutex_init(&fs_info
->transaction_kthread_mutex
);
1423 mutex_init(&fs_info
->cleaner_mutex
);
1424 mutex_init(&fs_info
->volume_mutex
);
1425 init_waitqueue_head(&fs_info
->transaction_throttle
);
1426 init_waitqueue_head(&fs_info
->transaction_wait
);
1427 init_waitqueue_head(&fs_info
->async_submit_wait
);
1428 init_waitqueue_head(&fs_info
->tree_log_wait
);
1429 atomic_set(&fs_info
->tree_log_commit
, 0);
1430 atomic_set(&fs_info
->tree_log_writers
, 0);
1431 fs_info
->tree_log_transid
= 0;
1434 ret
= add_hasher(fs_info
, "crc32c");
1436 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1441 __setup_root(4096, 4096, 4096, 4096, tree_root
,
1442 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
1445 bh
= __bread(fs_devices
->latest_bdev
,
1446 BTRFS_SUPER_INFO_OFFSET
/ 4096, 4096);
1450 memcpy(&fs_info
->super_copy
, bh
->b_data
, sizeof(fs_info
->super_copy
));
1453 memcpy(fs_info
->fsid
, fs_info
->super_copy
.fsid
, BTRFS_FSID_SIZE
);
1455 disk_super
= &fs_info
->super_copy
;
1456 if (!btrfs_super_root(disk_super
))
1457 goto fail_sb_buffer
;
1459 err
= btrfs_parse_options(tree_root
, options
);
1461 goto fail_sb_buffer
;
1464 * we need to start all the end_io workers up front because the
1465 * queue work function gets called at interrupt time, and so it
1466 * cannot dynamically grow.
1468 btrfs_init_workers(&fs_info
->workers
, "worker",
1469 fs_info
->thread_pool_size
);
1470 btrfs_init_workers(&fs_info
->submit_workers
, "submit",
1471 min_t(u64
, fs_devices
->num_devices
,
1472 fs_info
->thread_pool_size
));
1474 /* a higher idle thresh on the submit workers makes it much more
1475 * likely that bios will be send down in a sane order to the
1478 fs_info
->submit_workers
.idle_thresh
= 64;
1480 /* fs_info->workers is responsible for checksumming file data
1481 * blocks and metadata. Using a larger idle thresh allows each
1482 * worker thread to operate on things in roughly the order they
1483 * were sent by the writeback daemons, improving overall locality
1484 * of the IO going down the pipe.
1486 fs_info
->workers
.idle_thresh
= 128;
1488 btrfs_init_workers(&fs_info
->fixup_workers
, "fixup", 1);
1489 btrfs_init_workers(&fs_info
->endio_workers
, "endio",
1490 fs_info
->thread_pool_size
);
1491 btrfs_init_workers(&fs_info
->endio_write_workers
, "endio-write",
1492 fs_info
->thread_pool_size
);
1495 * endios are largely parallel and should have a very
1498 fs_info
->endio_workers
.idle_thresh
= 4;
1499 fs_info
->endio_write_workers
.idle_thresh
= 4;
1501 btrfs_start_workers(&fs_info
->workers
, 1);
1502 btrfs_start_workers(&fs_info
->submit_workers
, 1);
1503 btrfs_start_workers(&fs_info
->fixup_workers
, 1);
1504 btrfs_start_workers(&fs_info
->endio_workers
, fs_info
->thread_pool_size
);
1505 btrfs_start_workers(&fs_info
->endio_write_workers
,
1506 fs_info
->thread_pool_size
);
1509 if (btrfs_super_num_devices(disk_super
) > fs_devices
->open_devices
) {
1510 printk("Btrfs: wanted %llu devices, but found %llu\n",
1511 (unsigned long long)btrfs_super_num_devices(disk_super
),
1512 (unsigned long long)fs_devices
->open_devices
);
1513 if (btrfs_test_opt(tree_root
, DEGRADED
))
1514 printk("continuing in degraded mode\n");
1516 goto fail_sb_buffer
;
1520 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
1522 nodesize
= btrfs_super_nodesize(disk_super
);
1523 leafsize
= btrfs_super_leafsize(disk_super
);
1524 sectorsize
= btrfs_super_sectorsize(disk_super
);
1525 stripesize
= btrfs_super_stripesize(disk_super
);
1526 tree_root
->nodesize
= nodesize
;
1527 tree_root
->leafsize
= leafsize
;
1528 tree_root
->sectorsize
= sectorsize
;
1529 tree_root
->stripesize
= stripesize
;
1531 sb
->s_blocksize
= sectorsize
;
1532 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
1534 if (strncmp((char *)(&disk_super
->magic
), BTRFS_MAGIC
,
1535 sizeof(disk_super
->magic
))) {
1536 printk("btrfs: valid FS not found on %s\n", sb
->s_id
);
1537 goto fail_sb_buffer
;
1540 mutex_lock(&fs_info
->chunk_mutex
);
1541 ret
= btrfs_read_sys_array(tree_root
);
1542 mutex_unlock(&fs_info
->chunk_mutex
);
1544 printk("btrfs: failed to read the system array on %s\n",
1546 goto fail_sys_array
;
1549 blocksize
= btrfs_level_size(tree_root
,
1550 btrfs_super_chunk_root_level(disk_super
));
1552 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
1553 chunk_root
, fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
1555 chunk_root
->node
= read_tree_block(chunk_root
,
1556 btrfs_super_chunk_root(disk_super
),
1558 BUG_ON(!chunk_root
->node
);
1560 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
1561 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root
->node
),
1564 mutex_lock(&fs_info
->chunk_mutex
);
1565 ret
= btrfs_read_chunk_tree(chunk_root
);
1566 mutex_unlock(&fs_info
->chunk_mutex
);
1569 btrfs_close_extra_devices(fs_devices
);
1571 blocksize
= btrfs_level_size(tree_root
,
1572 btrfs_super_root_level(disk_super
));
1575 tree_root
->node
= read_tree_block(tree_root
,
1576 btrfs_super_root(disk_super
),
1578 if (!tree_root
->node
)
1579 goto fail_sb_buffer
;
1582 ret
= find_and_setup_root(tree_root
, fs_info
,
1583 BTRFS_EXTENT_TREE_OBJECTID
, extent_root
);
1585 goto fail_tree_root
;
1586 extent_root
->track_dirty
= 1;
1588 ret
= find_and_setup_root(tree_root
, fs_info
,
1589 BTRFS_DEV_TREE_OBJECTID
, dev_root
);
1590 dev_root
->track_dirty
= 1;
1593 goto fail_extent_root
;
1595 btrfs_read_block_groups(extent_root
);
1597 fs_info
->generation
= btrfs_super_generation(disk_super
) + 1;
1598 fs_info
->data_alloc_profile
= (u64
)-1;
1599 fs_info
->metadata_alloc_profile
= (u64
)-1;
1600 fs_info
->system_alloc_profile
= fs_info
->metadata_alloc_profile
;
1601 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
1603 if (!fs_info
->cleaner_kthread
)
1604 goto fail_extent_root
;
1606 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
1608 "btrfs-transaction");
1609 if (!fs_info
->transaction_kthread
)
1612 if (btrfs_super_log_root(disk_super
) != 0) {
1614 u64 bytenr
= btrfs_super_log_root(disk_super
);
1617 btrfs_level_size(tree_root
,
1618 btrfs_super_log_root_level(disk_super
));
1620 log_tree_root
= kzalloc(sizeof(struct btrfs_root
),
1623 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
1624 log_tree_root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
1626 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
1628 ret
= btrfs_recover_log_trees(log_tree_root
);
1631 fs_info
->last_trans_committed
= btrfs_super_generation(disk_super
);
1635 kthread_stop(fs_info
->cleaner_kthread
);
1637 free_extent_buffer(extent_root
->node
);
1639 free_extent_buffer(tree_root
->node
);
1642 btrfs_stop_workers(&fs_info
->fixup_workers
);
1643 btrfs_stop_workers(&fs_info
->workers
);
1644 btrfs_stop_workers(&fs_info
->endio_workers
);
1645 btrfs_stop_workers(&fs_info
->endio_write_workers
);
1646 btrfs_stop_workers(&fs_info
->submit_workers
);
1648 iput(fs_info
->btree_inode
);
1650 btrfs_close_devices(fs_info
->fs_devices
);
1651 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
1655 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1656 bdi_destroy(&fs_info
->bdi
);
1659 return ERR_PTR(err
);
1662 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
1664 char b
[BDEVNAME_SIZE
];
1667 set_buffer_uptodate(bh
);
1669 if (!buffer_eopnotsupp(bh
) && printk_ratelimit()) {
1670 printk(KERN_WARNING
"lost page write due to "
1671 "I/O error on %s\n",
1672 bdevname(bh
->b_bdev
, b
));
1674 /* note, we dont' set_buffer_write_io_error because we have
1675 * our own ways of dealing with the IO errors
1677 clear_buffer_uptodate(bh
);
1683 int write_all_supers(struct btrfs_root
*root
)
1685 struct list_head
*cur
;
1686 struct list_head
*head
= &root
->fs_info
->fs_devices
->devices
;
1687 struct btrfs_device
*dev
;
1688 struct btrfs_super_block
*sb
;
1689 struct btrfs_dev_item
*dev_item
;
1690 struct buffer_head
*bh
;
1694 int total_errors
= 0;
1698 max_errors
= btrfs_super_num_devices(&root
->fs_info
->super_copy
) - 1;
1699 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
1701 sb
= &root
->fs_info
->super_for_commit
;
1702 dev_item
= &sb
->dev_item
;
1703 list_for_each(cur
, head
) {
1704 dev
= list_entry(cur
, struct btrfs_device
, dev_list
);
1709 if (!dev
->in_fs_metadata
)
1712 btrfs_set_stack_device_type(dev_item
, dev
->type
);
1713 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
1714 btrfs_set_stack_device_total_bytes(dev_item
, dev
->total_bytes
);
1715 btrfs_set_stack_device_bytes_used(dev_item
, dev
->bytes_used
);
1716 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
1717 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
1718 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
1719 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
1720 flags
= btrfs_super_flags(sb
);
1721 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
1725 crc
= btrfs_csum_data(root
, (char *)sb
+ BTRFS_CSUM_SIZE
, crc
,
1726 BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
);
1727 btrfs_csum_final(crc
, sb
->csum
);
1729 bh
= __getblk(dev
->bdev
, BTRFS_SUPER_INFO_OFFSET
/ 4096,
1730 BTRFS_SUPER_INFO_SIZE
);
1732 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
1733 dev
->pending_io
= bh
;
1736 set_buffer_uptodate(bh
);
1738 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
1740 if (do_barriers
&& dev
->barriers
) {
1741 ret
= submit_bh(WRITE_BARRIER
, bh
);
1742 if (ret
== -EOPNOTSUPP
) {
1743 printk("btrfs: disabling barriers on dev %s\n",
1745 set_buffer_uptodate(bh
);
1749 ret
= submit_bh(WRITE
, bh
);
1752 ret
= submit_bh(WRITE
, bh
);
1757 if (total_errors
> max_errors
) {
1758 printk("btrfs: %d errors while writing supers\n", total_errors
);
1763 list_for_each(cur
, head
) {
1764 dev
= list_entry(cur
, struct btrfs_device
, dev_list
);
1767 if (!dev
->in_fs_metadata
)
1770 BUG_ON(!dev
->pending_io
);
1771 bh
= dev
->pending_io
;
1773 if (!buffer_uptodate(dev
->pending_io
)) {
1774 if (do_barriers
&& dev
->barriers
) {
1775 printk("btrfs: disabling barriers on dev %s\n",
1777 set_buffer_uptodate(bh
);
1781 ret
= submit_bh(WRITE
, bh
);
1784 if (!buffer_uptodate(bh
))
1791 dev
->pending_io
= NULL
;
1794 if (total_errors
> max_errors
) {
1795 printk("btrfs: %d errors while writing supers\n", total_errors
);
1801 int write_ctree_super(struct btrfs_trans_handle
*trans
, struct btrfs_root
1806 ret
= write_all_supers(root
);
1810 int btrfs_free_fs_root(struct btrfs_fs_info
*fs_info
, struct btrfs_root
*root
)
1812 radix_tree_delete(&fs_info
->fs_roots_radix
,
1813 (unsigned long)root
->root_key
.objectid
);
1815 btrfs_sysfs_del_root(root
);
1819 free_extent_buffer(root
->node
);
1820 if (root
->commit_root
)
1821 free_extent_buffer(root
->commit_root
);
1828 static int del_fs_roots(struct btrfs_fs_info
*fs_info
)
1831 struct btrfs_root
*gang
[8];
1835 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
1840 for (i
= 0; i
< ret
; i
++)
1841 btrfs_free_fs_root(fs_info
, gang
[i
]);
1846 int close_ctree(struct btrfs_root
*root
)
1849 struct btrfs_trans_handle
*trans
;
1850 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1852 fs_info
->closing
= 1;
1855 kthread_stop(root
->fs_info
->transaction_kthread
);
1856 kthread_stop(root
->fs_info
->cleaner_kthread
);
1858 btrfs_clean_old_snapshots(root
);
1859 trans
= btrfs_start_transaction(root
, 1);
1860 ret
= btrfs_commit_transaction(trans
, root
);
1861 /* run commit again to drop the original snapshot */
1862 trans
= btrfs_start_transaction(root
, 1);
1863 btrfs_commit_transaction(trans
, root
);
1864 ret
= btrfs_write_and_wait_transaction(NULL
, root
);
1867 write_ctree_super(NULL
, root
);
1869 if (fs_info
->delalloc_bytes
) {
1870 printk("btrfs: at unmount delalloc count %Lu\n",
1871 fs_info
->delalloc_bytes
);
1873 if (fs_info
->total_ref_cache_size
) {
1874 printk("btrfs: at umount reference cache size %Lu\n",
1875 fs_info
->total_ref_cache_size
);
1878 if (fs_info
->extent_root
->node
)
1879 free_extent_buffer(fs_info
->extent_root
->node
);
1881 if (fs_info
->tree_root
->node
)
1882 free_extent_buffer(fs_info
->tree_root
->node
);
1884 if (root
->fs_info
->chunk_root
->node
);
1885 free_extent_buffer(root
->fs_info
->chunk_root
->node
);
1887 if (root
->fs_info
->dev_root
->node
);
1888 free_extent_buffer(root
->fs_info
->dev_root
->node
);
1890 btrfs_free_block_groups(root
->fs_info
);
1891 fs_info
->closing
= 2;
1892 del_fs_roots(fs_info
);
1894 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
1896 truncate_inode_pages(fs_info
->btree_inode
->i_mapping
, 0);
1898 btrfs_stop_workers(&fs_info
->fixup_workers
);
1899 btrfs_stop_workers(&fs_info
->workers
);
1900 btrfs_stop_workers(&fs_info
->endio_workers
);
1901 btrfs_stop_workers(&fs_info
->endio_write_workers
);
1902 btrfs_stop_workers(&fs_info
->submit_workers
);
1904 iput(fs_info
->btree_inode
);
1906 while(!list_empty(&fs_info
->hashers
)) {
1907 struct btrfs_hasher
*hasher
;
1908 hasher
= list_entry(fs_info
->hashers
.next
, struct btrfs_hasher
,
1910 list_del(&hasher
->hashers
);
1911 crypto_free_hash(&fs_info
->hash_tfm
);
1915 btrfs_close_devices(fs_info
->fs_devices
);
1916 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
1918 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1919 bdi_destroy(&fs_info
->bdi
);
1922 kfree(fs_info
->extent_root
);
1923 kfree(fs_info
->tree_root
);
1924 kfree(fs_info
->chunk_root
);
1925 kfree(fs_info
->dev_root
);
1929 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
)
1932 struct inode
*btree_inode
= buf
->first_page
->mapping
->host
;
1934 ret
= extent_buffer_uptodate(&BTRFS_I(btree_inode
)->io_tree
, buf
);
1938 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
1943 int btrfs_set_buffer_uptodate(struct extent_buffer
*buf
)
1945 struct inode
*btree_inode
= buf
->first_page
->mapping
->host
;
1946 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode
)->io_tree
,
1950 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
1952 struct btrfs_root
*root
= BTRFS_I(buf
->first_page
->mapping
->host
)->root
;
1953 u64 transid
= btrfs_header_generation(buf
);
1954 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
1956 WARN_ON(!btrfs_tree_locked(buf
));
1957 if (transid
!= root
->fs_info
->generation
) {
1958 printk(KERN_CRIT
"transid mismatch buffer %llu, found %Lu running %Lu\n",
1959 (unsigned long long)buf
->start
,
1960 transid
, root
->fs_info
->generation
);
1963 set_extent_buffer_dirty(&BTRFS_I(btree_inode
)->io_tree
, buf
);
1966 void btrfs_btree_balance_dirty(struct btrfs_root
*root
, unsigned long nr
)
1969 * looks as though older kernels can get into trouble with
1970 * this code, they end up stuck in balance_dirty_pages forever
1972 struct extent_io_tree
*tree
;
1975 unsigned long thresh
= 96 * 1024 * 1024;
1976 tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
1978 if (current_is_pdflush() || current
->flags
& PF_MEMALLOC
)
1981 num_dirty
= count_range_bits(tree
, &start
, (u64
)-1,
1982 thresh
, EXTENT_DIRTY
);
1983 if (num_dirty
> thresh
) {
1984 balance_dirty_pages_ratelimited_nr(
1985 root
->fs_info
->btree_inode
->i_mapping
, 1);
1990 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
1992 struct btrfs_root
*root
= BTRFS_I(buf
->first_page
->mapping
->host
)->root
;
1994 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
1996 buf
->flags
|= EXTENT_UPTODATE
;
2001 static struct extent_io_ops btree_extent_io_ops
= {
2002 .writepage_io_hook
= btree_writepage_io_hook
,
2003 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
2004 .submit_bio_hook
= btree_submit_bio_hook
,
2005 /* note we're sharing with inode.c for the merge bio hook */
2006 .merge_bio_hook
= btrfs_merge_bio_hook
,