2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
33 #include "transaction.h"
34 #include "btrfs_inode.h"
36 #include "print-tree.h"
37 #include "async-thread.h"
40 #include "free-space-cache.h"
42 static struct extent_io_ops btree_extent_io_ops
;
43 static void end_workqueue_fn(struct btrfs_work
*work
);
46 * end_io_wq structs are used to do processing in task context when an IO is
47 * complete. This is used during reads to verify checksums, and it is used
48 * by writes to insert metadata for new file extents after IO is complete.
54 struct btrfs_fs_info
*info
;
57 struct list_head list
;
58 struct btrfs_work work
;
62 * async submit bios are used to offload expensive checksumming
63 * onto the worker threads. They checksum file and metadata bios
64 * just before they are sent down the IO stack.
66 struct async_submit_bio
{
69 struct list_head list
;
70 extent_submit_bio_hook_t
*submit_bio_start
;
71 extent_submit_bio_hook_t
*submit_bio_done
;
74 unsigned long bio_flags
;
75 struct btrfs_work work
;
78 /* These are used to set the lockdep class on the extent buffer locks.
79 * The class is set by the readpage_end_io_hook after the buffer has
80 * passed csum validation but before the pages are unlocked.
82 * The lockdep class is also set by btrfs_init_new_buffer on freshly
85 * The class is based on the level in the tree block, which allows lockdep
86 * to know that lower nodes nest inside the locks of higher nodes.
88 * We also add a check to make sure the highest level of the tree is
89 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
90 * code needs update as well.
92 #ifdef CONFIG_DEBUG_LOCK_ALLOC
93 # if BTRFS_MAX_LEVEL != 8
96 static struct lock_class_key btrfs_eb_class
[BTRFS_MAX_LEVEL
+ 1];
97 static const char *btrfs_eb_name
[BTRFS_MAX_LEVEL
+ 1] = {
107 /* highest possible level */
113 * extents on the btree inode are pretty simple, there's one extent
114 * that covers the entire device
116 static struct extent_map
*btree_get_extent(struct inode
*inode
,
117 struct page
*page
, size_t page_offset
, u64 start
, u64 len
,
120 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
121 struct extent_map
*em
;
124 spin_lock(&em_tree
->lock
);
125 em
= lookup_extent_mapping(em_tree
, start
, len
);
128 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
129 spin_unlock(&em_tree
->lock
);
132 spin_unlock(&em_tree
->lock
);
134 em
= alloc_extent_map(GFP_NOFS
);
136 em
= ERR_PTR(-ENOMEM
);
141 em
->block_len
= (u64
)-1;
143 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
145 spin_lock(&em_tree
->lock
);
146 ret
= add_extent_mapping(em_tree
, em
);
147 if (ret
== -EEXIST
) {
148 u64 failed_start
= em
->start
;
149 u64 failed_len
= em
->len
;
152 em
= lookup_extent_mapping(em_tree
, start
, len
);
156 em
= lookup_extent_mapping(em_tree
, failed_start
,
164 spin_unlock(&em_tree
->lock
);
172 u32
btrfs_csum_data(struct btrfs_root
*root
, char *data
, u32 seed
, size_t len
)
174 return crc32c(seed
, data
, len
);
177 void btrfs_csum_final(u32 crc
, char *result
)
179 *(__le32
*)result
= ~cpu_to_le32(crc
);
183 * compute the csum for a btree block, and either verify it or write it
184 * into the csum field of the block.
186 static int csum_tree_block(struct btrfs_root
*root
, struct extent_buffer
*buf
,
190 btrfs_super_csum_size(&root
->fs_info
->super_copy
);
193 unsigned long cur_len
;
194 unsigned long offset
= BTRFS_CSUM_SIZE
;
195 char *map_token
= NULL
;
197 unsigned long map_start
;
198 unsigned long map_len
;
201 unsigned long inline_result
;
203 len
= buf
->len
- offset
;
205 err
= map_private_extent_buffer(buf
, offset
, 32,
207 &map_start
, &map_len
, KM_USER0
);
210 cur_len
= min(len
, map_len
- (offset
- map_start
));
211 crc
= btrfs_csum_data(root
, kaddr
+ offset
- map_start
,
215 unmap_extent_buffer(buf
, map_token
, KM_USER0
);
217 if (csum_size
> sizeof(inline_result
)) {
218 result
= kzalloc(csum_size
* sizeof(char), GFP_NOFS
);
222 result
= (char *)&inline_result
;
225 btrfs_csum_final(crc
, result
);
228 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
231 memcpy(&found
, result
, csum_size
);
233 read_extent_buffer(buf
, &val
, 0, csum_size
);
234 if (printk_ratelimit()) {
235 printk(KERN_INFO
"btrfs: %s checksum verify "
236 "failed on %llu wanted %X found %X "
238 root
->fs_info
->sb
->s_id
,
239 (unsigned long long)buf
->start
, val
, found
,
240 btrfs_header_level(buf
));
242 if (result
!= (char *)&inline_result
)
247 write_extent_buffer(buf
, result
, 0, csum_size
);
249 if (result
!= (char *)&inline_result
)
255 * we can't consider a given block up to date unless the transid of the
256 * block matches the transid in the parent node's pointer. This is how we
257 * detect blocks that either didn't get written at all or got written
258 * in the wrong place.
260 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
261 struct extent_buffer
*eb
, u64 parent_transid
)
265 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
268 lock_extent(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1, GFP_NOFS
);
269 if (extent_buffer_uptodate(io_tree
, eb
) &&
270 btrfs_header_generation(eb
) == parent_transid
) {
274 if (printk_ratelimit()) {
275 printk("parent transid verify failed on %llu wanted %llu "
277 (unsigned long long)eb
->start
,
278 (unsigned long long)parent_transid
,
279 (unsigned long long)btrfs_header_generation(eb
));
282 clear_extent_buffer_uptodate(io_tree
, eb
);
284 unlock_extent(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
290 * helper to read a given tree block, doing retries as required when
291 * the checksums don't match and we have alternate mirrors to try.
293 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
294 struct extent_buffer
*eb
,
295 u64 start
, u64 parent_transid
)
297 struct extent_io_tree
*io_tree
;
302 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
304 ret
= read_extent_buffer_pages(io_tree
, eb
, start
, 1,
305 btree_get_extent
, mirror_num
);
307 !verify_parent_transid(io_tree
, eb
, parent_transid
))
310 num_copies
= btrfs_num_copies(&root
->fs_info
->mapping_tree
,
316 if (mirror_num
> num_copies
)
323 * checksum a dirty tree block before IO. This has extra checks to make sure
324 * we only fill in the checksum field in the first page of a multi-page block
327 static int csum_dirty_buffer(struct btrfs_root
*root
, struct page
*page
)
329 struct extent_io_tree
*tree
;
330 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
334 struct extent_buffer
*eb
;
337 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
339 if (page
->private == EXTENT_PAGE_PRIVATE
)
343 len
= page
->private >> 2;
346 eb
= alloc_extent_buffer(tree
, start
, len
, page
, GFP_NOFS
);
347 ret
= btree_read_extent_buffer_pages(root
, eb
, start
+ PAGE_CACHE_SIZE
,
348 btrfs_header_generation(eb
));
350 found_start
= btrfs_header_bytenr(eb
);
351 if (found_start
!= start
) {
355 if (eb
->first_page
!= page
) {
359 if (!PageUptodate(page
)) {
363 found_level
= btrfs_header_level(eb
);
365 csum_tree_block(root
, eb
, 0);
367 free_extent_buffer(eb
);
372 static int check_tree_block_fsid(struct btrfs_root
*root
,
373 struct extent_buffer
*eb
)
375 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
376 u8 fsid
[BTRFS_UUID_SIZE
];
379 read_extent_buffer(eb
, fsid
, (unsigned long)btrfs_header_fsid(eb
),
382 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
386 fs_devices
= fs_devices
->seed
;
391 #ifdef CONFIG_DEBUG_LOCK_ALLOC
392 void btrfs_set_buffer_lockdep_class(struct extent_buffer
*eb
, int level
)
394 lockdep_set_class_and_name(&eb
->lock
,
395 &btrfs_eb_class
[level
],
396 btrfs_eb_name
[level
]);
400 static int btree_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
401 struct extent_state
*state
)
403 struct extent_io_tree
*tree
;
407 struct extent_buffer
*eb
;
408 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
411 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
412 if (page
->private == EXTENT_PAGE_PRIVATE
)
417 len
= page
->private >> 2;
420 eb
= alloc_extent_buffer(tree
, start
, len
, page
, GFP_NOFS
);
422 found_start
= btrfs_header_bytenr(eb
);
423 if (found_start
!= start
) {
424 if (printk_ratelimit()) {
425 printk(KERN_INFO
"btrfs bad tree block start "
427 (unsigned long long)found_start
,
428 (unsigned long long)eb
->start
);
433 if (eb
->first_page
!= page
) {
434 printk(KERN_INFO
"btrfs bad first page %lu %lu\n",
435 eb
->first_page
->index
, page
->index
);
440 if (check_tree_block_fsid(root
, eb
)) {
441 if (printk_ratelimit()) {
442 printk(KERN_INFO
"btrfs bad fsid on block %llu\n",
443 (unsigned long long)eb
->start
);
448 found_level
= btrfs_header_level(eb
);
450 btrfs_set_buffer_lockdep_class(eb
, found_level
);
452 ret
= csum_tree_block(root
, eb
, 1);
456 end
= min_t(u64
, eb
->len
, PAGE_CACHE_SIZE
);
457 end
= eb
->start
+ end
- 1;
459 free_extent_buffer(eb
);
464 static void end_workqueue_bio(struct bio
*bio
, int err
)
466 struct end_io_wq
*end_io_wq
= bio
->bi_private
;
467 struct btrfs_fs_info
*fs_info
;
469 fs_info
= end_io_wq
->info
;
470 end_io_wq
->error
= err
;
471 end_io_wq
->work
.func
= end_workqueue_fn
;
472 end_io_wq
->work
.flags
= 0;
474 if (bio
->bi_rw
& (1 << BIO_RW
)) {
475 if (end_io_wq
->metadata
)
476 btrfs_queue_worker(&fs_info
->endio_meta_write_workers
,
479 btrfs_queue_worker(&fs_info
->endio_write_workers
,
482 if (end_io_wq
->metadata
)
483 btrfs_queue_worker(&fs_info
->endio_meta_workers
,
486 btrfs_queue_worker(&fs_info
->endio_workers
,
491 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
494 struct end_io_wq
*end_io_wq
;
495 end_io_wq
= kmalloc(sizeof(*end_io_wq
), GFP_NOFS
);
499 end_io_wq
->private = bio
->bi_private
;
500 end_io_wq
->end_io
= bio
->bi_end_io
;
501 end_io_wq
->info
= info
;
502 end_io_wq
->error
= 0;
503 end_io_wq
->bio
= bio
;
504 end_io_wq
->metadata
= metadata
;
506 bio
->bi_private
= end_io_wq
;
507 bio
->bi_end_io
= end_workqueue_bio
;
511 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
513 unsigned long limit
= min_t(unsigned long,
514 info
->workers
.max_workers
,
515 info
->fs_devices
->open_devices
);
519 int btrfs_congested_async(struct btrfs_fs_info
*info
, int iodone
)
521 return atomic_read(&info
->nr_async_bios
) >
522 btrfs_async_submit_limit(info
);
525 static void run_one_async_start(struct btrfs_work
*work
)
527 struct btrfs_fs_info
*fs_info
;
528 struct async_submit_bio
*async
;
530 async
= container_of(work
, struct async_submit_bio
, work
);
531 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
532 async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
533 async
->mirror_num
, async
->bio_flags
);
536 static void run_one_async_done(struct btrfs_work
*work
)
538 struct btrfs_fs_info
*fs_info
;
539 struct async_submit_bio
*async
;
542 async
= container_of(work
, struct async_submit_bio
, work
);
543 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
545 limit
= btrfs_async_submit_limit(fs_info
);
546 limit
= limit
* 2 / 3;
548 atomic_dec(&fs_info
->nr_async_submits
);
550 if (atomic_read(&fs_info
->nr_async_submits
) < limit
&&
551 waitqueue_active(&fs_info
->async_submit_wait
))
552 wake_up(&fs_info
->async_submit_wait
);
554 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
555 async
->mirror_num
, async
->bio_flags
);
558 static void run_one_async_free(struct btrfs_work
*work
)
560 struct async_submit_bio
*async
;
562 async
= container_of(work
, struct async_submit_bio
, work
);
566 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
567 int rw
, struct bio
*bio
, int mirror_num
,
568 unsigned long bio_flags
,
569 extent_submit_bio_hook_t
*submit_bio_start
,
570 extent_submit_bio_hook_t
*submit_bio_done
)
572 struct async_submit_bio
*async
;
574 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
578 async
->inode
= inode
;
581 async
->mirror_num
= mirror_num
;
582 async
->submit_bio_start
= submit_bio_start
;
583 async
->submit_bio_done
= submit_bio_done
;
585 async
->work
.func
= run_one_async_start
;
586 async
->work
.ordered_func
= run_one_async_done
;
587 async
->work
.ordered_free
= run_one_async_free
;
589 async
->work
.flags
= 0;
590 async
->bio_flags
= bio_flags
;
592 atomic_inc(&fs_info
->nr_async_submits
);
594 if (rw
& (1 << BIO_RW_SYNCIO
))
595 btrfs_set_work_high_prio(&async
->work
);
597 btrfs_queue_worker(&fs_info
->workers
, &async
->work
);
599 while (atomic_read(&fs_info
->async_submit_draining
) &&
600 atomic_read(&fs_info
->nr_async_submits
)) {
601 wait_event(fs_info
->async_submit_wait
,
602 (atomic_read(&fs_info
->nr_async_submits
) == 0));
608 static int btree_csum_one_bio(struct bio
*bio
)
610 struct bio_vec
*bvec
= bio
->bi_io_vec
;
612 struct btrfs_root
*root
;
614 WARN_ON(bio
->bi_vcnt
<= 0);
615 while (bio_index
< bio
->bi_vcnt
) {
616 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
617 csum_dirty_buffer(root
, bvec
->bv_page
);
624 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
625 struct bio
*bio
, int mirror_num
,
626 unsigned long bio_flags
)
629 * when we're called for a write, we're already in the async
630 * submission context. Just jump into btrfs_map_bio
632 btree_csum_one_bio(bio
);
636 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
637 int mirror_num
, unsigned long bio_flags
)
640 * when we're called for a write, we're already in the async
641 * submission context. Just jump into btrfs_map_bio
643 return btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
646 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
647 int mirror_num
, unsigned long bio_flags
)
651 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
655 if (!(rw
& (1 << BIO_RW
))) {
657 * called for a read, do the setup so that checksum validation
658 * can happen in the async kernel threads
660 return btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
665 * kthread helpers are used to submit writes so that checksumming
666 * can happen in parallel across all CPUs
668 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
669 inode
, rw
, bio
, mirror_num
, 0,
670 __btree_submit_bio_start
,
671 __btree_submit_bio_done
);
674 static int btree_writepage(struct page
*page
, struct writeback_control
*wbc
)
676 struct extent_io_tree
*tree
;
677 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
678 struct extent_buffer
*eb
;
681 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
682 if (!(current
->flags
& PF_MEMALLOC
)) {
683 return extent_write_full_page(tree
, page
,
684 btree_get_extent
, wbc
);
687 redirty_page_for_writepage(wbc
, page
);
688 eb
= btrfs_find_tree_block(root
, page_offset(page
),
692 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
694 spin_lock(&root
->fs_info
->delalloc_lock
);
695 root
->fs_info
->dirty_metadata_bytes
+= PAGE_CACHE_SIZE
;
696 spin_unlock(&root
->fs_info
->delalloc_lock
);
698 free_extent_buffer(eb
);
704 static int btree_writepages(struct address_space
*mapping
,
705 struct writeback_control
*wbc
)
707 struct extent_io_tree
*tree
;
708 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
709 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
710 struct btrfs_root
*root
= BTRFS_I(mapping
->host
)->root
;
712 unsigned long thresh
= 32 * 1024 * 1024;
714 if (wbc
->for_kupdate
)
717 /* this is a bit racy, but that's ok */
718 num_dirty
= root
->fs_info
->dirty_metadata_bytes
;
719 if (num_dirty
< thresh
)
722 return extent_writepages(tree
, mapping
, btree_get_extent
, wbc
);
725 static int btree_readpage(struct file
*file
, struct page
*page
)
727 struct extent_io_tree
*tree
;
728 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
729 return extent_read_full_page(tree
, page
, btree_get_extent
);
732 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
734 struct extent_io_tree
*tree
;
735 struct extent_map_tree
*map
;
738 if (PageWriteback(page
) || PageDirty(page
))
741 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
742 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
744 ret
= try_release_extent_state(map
, tree
, page
, gfp_flags
);
748 ret
= try_release_extent_buffer(tree
, page
);
750 ClearPagePrivate(page
);
751 set_page_private(page
, 0);
752 page_cache_release(page
);
758 static void btree_invalidatepage(struct page
*page
, unsigned long offset
)
760 struct extent_io_tree
*tree
;
761 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
762 extent_invalidatepage(tree
, page
, offset
);
763 btree_releasepage(page
, GFP_NOFS
);
764 if (PagePrivate(page
)) {
765 printk(KERN_WARNING
"btrfs warning page private not zero "
766 "on page %llu\n", (unsigned long long)page_offset(page
));
767 ClearPagePrivate(page
);
768 set_page_private(page
, 0);
769 page_cache_release(page
);
773 static struct address_space_operations btree_aops
= {
774 .readpage
= btree_readpage
,
775 .writepage
= btree_writepage
,
776 .writepages
= btree_writepages
,
777 .releasepage
= btree_releasepage
,
778 .invalidatepage
= btree_invalidatepage
,
779 .sync_page
= block_sync_page
,
782 int readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
785 struct extent_buffer
*buf
= NULL
;
786 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
789 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
792 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
793 buf
, 0, 0, btree_get_extent
, 0);
794 free_extent_buffer(buf
);
798 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_root
*root
,
799 u64 bytenr
, u32 blocksize
)
801 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
802 struct extent_buffer
*eb
;
803 eb
= find_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
804 bytenr
, blocksize
, GFP_NOFS
);
808 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
809 u64 bytenr
, u32 blocksize
)
811 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
812 struct extent_buffer
*eb
;
814 eb
= alloc_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
815 bytenr
, blocksize
, NULL
, GFP_NOFS
);
820 int btrfs_write_tree_block(struct extent_buffer
*buf
)
822 return btrfs_fdatawrite_range(buf
->first_page
->mapping
, buf
->start
,
823 buf
->start
+ buf
->len
- 1, WB_SYNC_ALL
);
826 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
828 return btrfs_wait_on_page_writeback_range(buf
->first_page
->mapping
,
829 buf
->start
, buf
->start
+ buf
->len
- 1);
832 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
833 u32 blocksize
, u64 parent_transid
)
835 struct extent_buffer
*buf
= NULL
;
836 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
837 struct extent_io_tree
*io_tree
;
840 io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
842 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
846 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
849 set_bit(EXTENT_BUFFER_UPTODATE
, &buf
->bflags
);
854 int clean_tree_block(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
855 struct extent_buffer
*buf
)
857 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
858 if (btrfs_header_generation(buf
) ==
859 root
->fs_info
->running_transaction
->transid
) {
860 btrfs_assert_tree_locked(buf
);
862 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
863 spin_lock(&root
->fs_info
->delalloc_lock
);
864 if (root
->fs_info
->dirty_metadata_bytes
>= buf
->len
)
865 root
->fs_info
->dirty_metadata_bytes
-= buf
->len
;
868 spin_unlock(&root
->fs_info
->delalloc_lock
);
871 /* ugh, clear_extent_buffer_dirty needs to lock the page */
872 btrfs_set_lock_blocking(buf
);
873 clear_extent_buffer_dirty(&BTRFS_I(btree_inode
)->io_tree
,
879 static int __setup_root(u32 nodesize
, u32 leafsize
, u32 sectorsize
,
880 u32 stripesize
, struct btrfs_root
*root
,
881 struct btrfs_fs_info
*fs_info
,
885 root
->commit_root
= NULL
;
886 root
->sectorsize
= sectorsize
;
887 root
->nodesize
= nodesize
;
888 root
->leafsize
= leafsize
;
889 root
->stripesize
= stripesize
;
891 root
->track_dirty
= 0;
893 root
->fs_info
= fs_info
;
894 root
->objectid
= objectid
;
895 root
->last_trans
= 0;
896 root
->highest_inode
= 0;
897 root
->last_inode_alloc
= 0;
900 root
->inode_tree
.rb_node
= NULL
;
902 INIT_LIST_HEAD(&root
->dirty_list
);
903 INIT_LIST_HEAD(&root
->orphan_list
);
904 INIT_LIST_HEAD(&root
->root_list
);
905 spin_lock_init(&root
->node_lock
);
906 spin_lock_init(&root
->list_lock
);
907 spin_lock_init(&root
->inode_lock
);
908 mutex_init(&root
->objectid_mutex
);
909 mutex_init(&root
->log_mutex
);
910 init_waitqueue_head(&root
->log_writer_wait
);
911 init_waitqueue_head(&root
->log_commit_wait
[0]);
912 init_waitqueue_head(&root
->log_commit_wait
[1]);
913 atomic_set(&root
->log_commit
[0], 0);
914 atomic_set(&root
->log_commit
[1], 0);
915 atomic_set(&root
->log_writers
, 0);
917 root
->log_transid
= 0;
918 extent_io_tree_init(&root
->dirty_log_pages
,
919 fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
921 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
922 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
923 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
924 memset(&root
->root_kobj
, 0, sizeof(root
->root_kobj
));
925 root
->defrag_trans_start
= fs_info
->generation
;
926 init_completion(&root
->kobj_unregister
);
927 root
->defrag_running
= 0;
928 root
->defrag_level
= 0;
929 root
->root_key
.objectid
= objectid
;
930 root
->anon_super
.s_root
= NULL
;
931 root
->anon_super
.s_dev
= 0;
932 INIT_LIST_HEAD(&root
->anon_super
.s_list
);
933 INIT_LIST_HEAD(&root
->anon_super
.s_instances
);
934 init_rwsem(&root
->anon_super
.s_umount
);
939 static int find_and_setup_root(struct btrfs_root
*tree_root
,
940 struct btrfs_fs_info
*fs_info
,
942 struct btrfs_root
*root
)
948 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
949 tree_root
->sectorsize
, tree_root
->stripesize
,
950 root
, fs_info
, objectid
);
951 ret
= btrfs_find_last_root(tree_root
, objectid
,
952 &root
->root_item
, &root
->root_key
);
955 generation
= btrfs_root_generation(&root
->root_item
);
956 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
957 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
958 blocksize
, generation
);
959 root
->commit_root
= btrfs_root_node(root
);
964 int btrfs_free_log_root_tree(struct btrfs_trans_handle
*trans
,
965 struct btrfs_fs_info
*fs_info
)
967 struct extent_buffer
*eb
;
968 struct btrfs_root
*log_root_tree
= fs_info
->log_root_tree
;
977 ret
= find_first_extent_bit(&log_root_tree
->dirty_log_pages
,
978 0, &start
, &end
, EXTENT_DIRTY
);
982 clear_extent_dirty(&log_root_tree
->dirty_log_pages
,
983 start
, end
, GFP_NOFS
);
985 eb
= fs_info
->log_root_tree
->node
;
987 WARN_ON(btrfs_header_level(eb
) != 0);
988 WARN_ON(btrfs_header_nritems(eb
) != 0);
990 ret
= btrfs_free_reserved_extent(fs_info
->tree_root
,
994 free_extent_buffer(eb
);
995 kfree(fs_info
->log_root_tree
);
996 fs_info
->log_root_tree
= NULL
;
1000 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1001 struct btrfs_fs_info
*fs_info
)
1003 struct btrfs_root
*root
;
1004 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1005 struct extent_buffer
*leaf
;
1007 root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1009 return ERR_PTR(-ENOMEM
);
1011 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1012 tree_root
->sectorsize
, tree_root
->stripesize
,
1013 root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
1015 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1016 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1017 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1019 * log trees do not get reference counted because they go away
1020 * before a real commit is actually done. They do store pointers
1021 * to file data extents, and those reference counts still get
1022 * updated (along with back refs to the log tree).
1026 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
, 0,
1027 BTRFS_TREE_LOG_OBJECTID
, NULL
, 0, 0, 0);
1030 return ERR_CAST(leaf
);
1033 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1034 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1035 btrfs_set_header_generation(leaf
, trans
->transid
);
1036 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1037 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1040 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1041 (unsigned long)btrfs_header_fsid(root
->node
),
1043 btrfs_mark_buffer_dirty(root
->node
);
1044 btrfs_tree_unlock(root
->node
);
1048 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1049 struct btrfs_fs_info
*fs_info
)
1051 struct btrfs_root
*log_root
;
1053 log_root
= alloc_log_tree(trans
, fs_info
);
1054 if (IS_ERR(log_root
))
1055 return PTR_ERR(log_root
);
1056 WARN_ON(fs_info
->log_root_tree
);
1057 fs_info
->log_root_tree
= log_root
;
1061 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1062 struct btrfs_root
*root
)
1064 struct btrfs_root
*log_root
;
1065 struct btrfs_inode_item
*inode_item
;
1067 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1068 if (IS_ERR(log_root
))
1069 return PTR_ERR(log_root
);
1071 log_root
->last_trans
= trans
->transid
;
1072 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1074 inode_item
= &log_root
->root_item
.inode
;
1075 inode_item
->generation
= cpu_to_le64(1);
1076 inode_item
->size
= cpu_to_le64(3);
1077 inode_item
->nlink
= cpu_to_le32(1);
1078 inode_item
->nbytes
= cpu_to_le64(root
->leafsize
);
1079 inode_item
->mode
= cpu_to_le32(S_IFDIR
| 0755);
1081 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1083 WARN_ON(root
->log_root
);
1084 root
->log_root
= log_root
;
1085 root
->log_transid
= 0;
1089 struct btrfs_root
*btrfs_read_fs_root_no_radix(struct btrfs_root
*tree_root
,
1090 struct btrfs_key
*location
)
1092 struct btrfs_root
*root
;
1093 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1094 struct btrfs_path
*path
;
1095 struct extent_buffer
*l
;
1101 root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1103 return ERR_PTR(-ENOMEM
);
1104 if (location
->offset
== (u64
)-1) {
1105 ret
= find_and_setup_root(tree_root
, fs_info
,
1106 location
->objectid
, root
);
1109 return ERR_PTR(ret
);
1114 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1115 tree_root
->sectorsize
, tree_root
->stripesize
,
1116 root
, fs_info
, location
->objectid
);
1118 path
= btrfs_alloc_path();
1120 ret
= btrfs_search_slot(NULL
, tree_root
, location
, path
, 0, 0);
1127 read_extent_buffer(l
, &root
->root_item
,
1128 btrfs_item_ptr_offset(l
, path
->slots
[0]),
1129 sizeof(root
->root_item
));
1130 memcpy(&root
->root_key
, location
, sizeof(*location
));
1133 btrfs_release_path(root
, path
);
1134 btrfs_free_path(path
);
1137 return ERR_PTR(ret
);
1139 generation
= btrfs_root_generation(&root
->root_item
);
1140 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1141 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1142 blocksize
, generation
);
1143 root
->commit_root
= btrfs_root_node(root
);
1144 BUG_ON(!root
->node
);
1146 if (location
->objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
1148 ret
= btrfs_find_highest_inode(root
, &highest_inode
);
1150 root
->highest_inode
= highest_inode
;
1151 root
->last_inode_alloc
= highest_inode
;
1157 struct btrfs_root
*btrfs_lookup_fs_root(struct btrfs_fs_info
*fs_info
,
1160 struct btrfs_root
*root
;
1162 if (root_objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1163 return fs_info
->tree_root
;
1164 if (root_objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1165 return fs_info
->extent_root
;
1167 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1168 (unsigned long)root_objectid
);
1172 struct btrfs_root
*btrfs_read_fs_root_no_name(struct btrfs_fs_info
*fs_info
,
1173 struct btrfs_key
*location
)
1175 struct btrfs_root
*root
;
1178 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
1179 return fs_info
->tree_root
;
1180 if (location
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
1181 return fs_info
->extent_root
;
1182 if (location
->objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
1183 return fs_info
->chunk_root
;
1184 if (location
->objectid
== BTRFS_DEV_TREE_OBJECTID
)
1185 return fs_info
->dev_root
;
1186 if (location
->objectid
== BTRFS_CSUM_TREE_OBJECTID
)
1187 return fs_info
->csum_root
;
1189 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1190 (unsigned long)location
->objectid
);
1194 root
= btrfs_read_fs_root_no_radix(fs_info
->tree_root
, location
);
1198 set_anon_super(&root
->anon_super
, NULL
);
1200 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1201 (unsigned long)root
->root_key
.objectid
,
1204 free_extent_buffer(root
->node
);
1206 return ERR_PTR(ret
);
1208 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
1209 ret
= btrfs_find_dead_roots(fs_info
->tree_root
,
1210 root
->root_key
.objectid
);
1212 btrfs_orphan_cleanup(root
);
1217 struct btrfs_root
*btrfs_read_fs_root(struct btrfs_fs_info
*fs_info
,
1218 struct btrfs_key
*location
,
1219 const char *name
, int namelen
)
1221 struct btrfs_root
*root
;
1224 root
= btrfs_read_fs_root_no_name(fs_info
, location
);
1231 ret
= btrfs_set_root_name(root
, name
, namelen
);
1233 free_extent_buffer(root
->node
);
1235 return ERR_PTR(ret
);
1238 ret
= btrfs_sysfs_add_root(root
);
1240 free_extent_buffer(root
->node
);
1243 return ERR_PTR(ret
);
1250 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1252 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1254 struct btrfs_device
*device
;
1255 struct backing_dev_info
*bdi
;
1257 list_for_each_entry(device
, &info
->fs_devices
->devices
, dev_list
) {
1260 bdi
= blk_get_backing_dev_info(device
->bdev
);
1261 if (bdi
&& bdi_congested(bdi
, bdi_bits
)) {
1270 * this unplugs every device on the box, and it is only used when page
1273 static void __unplug_io_fn(struct backing_dev_info
*bdi
, struct page
*page
)
1275 struct btrfs_device
*device
;
1276 struct btrfs_fs_info
*info
;
1278 info
= (struct btrfs_fs_info
*)bdi
->unplug_io_data
;
1279 list_for_each_entry(device
, &info
->fs_devices
->devices
, dev_list
) {
1283 bdi
= blk_get_backing_dev_info(device
->bdev
);
1284 if (bdi
->unplug_io_fn
)
1285 bdi
->unplug_io_fn(bdi
, page
);
1289 static void btrfs_unplug_io_fn(struct backing_dev_info
*bdi
, struct page
*page
)
1291 struct inode
*inode
;
1292 struct extent_map_tree
*em_tree
;
1293 struct extent_map
*em
;
1294 struct address_space
*mapping
;
1297 /* the generic O_DIRECT read code does this */
1299 __unplug_io_fn(bdi
, page
);
1304 * page->mapping may change at any time. Get a consistent copy
1305 * and use that for everything below
1308 mapping
= page
->mapping
;
1312 inode
= mapping
->host
;
1315 * don't do the expensive searching for a small number of
1318 if (BTRFS_I(inode
)->root
->fs_info
->fs_devices
->open_devices
<= 2) {
1319 __unplug_io_fn(bdi
, page
);
1323 offset
= page_offset(page
);
1325 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1326 spin_lock(&em_tree
->lock
);
1327 em
= lookup_extent_mapping(em_tree
, offset
, PAGE_CACHE_SIZE
);
1328 spin_unlock(&em_tree
->lock
);
1330 __unplug_io_fn(bdi
, page
);
1334 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
1335 free_extent_map(em
);
1336 __unplug_io_fn(bdi
, page
);
1339 offset
= offset
- em
->start
;
1340 btrfs_unplug_page(&BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1341 em
->block_start
+ offset
, page
);
1342 free_extent_map(em
);
1345 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1348 bdi
->ra_pages
= default_backing_dev_info
.ra_pages
;
1350 bdi
->capabilities
= default_backing_dev_info
.capabilities
;
1351 bdi
->unplug_io_fn
= btrfs_unplug_io_fn
;
1352 bdi
->unplug_io_data
= info
;
1353 bdi
->congested_fn
= btrfs_congested_fn
;
1354 bdi
->congested_data
= info
;
1358 static int bio_ready_for_csum(struct bio
*bio
)
1364 struct extent_io_tree
*io_tree
= NULL
;
1365 struct btrfs_fs_info
*info
= NULL
;
1366 struct bio_vec
*bvec
;
1370 bio_for_each_segment(bvec
, bio
, i
) {
1371 page
= bvec
->bv_page
;
1372 if (page
->private == EXTENT_PAGE_PRIVATE
) {
1373 length
+= bvec
->bv_len
;
1376 if (!page
->private) {
1377 length
+= bvec
->bv_len
;
1380 length
= bvec
->bv_len
;
1381 buf_len
= page
->private >> 2;
1382 start
= page_offset(page
) + bvec
->bv_offset
;
1383 io_tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1384 info
= BTRFS_I(page
->mapping
->host
)->root
->fs_info
;
1386 /* are we fully contained in this bio? */
1387 if (buf_len
<= length
)
1390 ret
= extent_range_uptodate(io_tree
, start
+ length
,
1391 start
+ buf_len
- 1);
1396 * called by the kthread helper functions to finally call the bio end_io
1397 * functions. This is where read checksum verification actually happens
1399 static void end_workqueue_fn(struct btrfs_work
*work
)
1402 struct end_io_wq
*end_io_wq
;
1403 struct btrfs_fs_info
*fs_info
;
1406 end_io_wq
= container_of(work
, struct end_io_wq
, work
);
1407 bio
= end_io_wq
->bio
;
1408 fs_info
= end_io_wq
->info
;
1410 /* metadata bio reads are special because the whole tree block must
1411 * be checksummed at once. This makes sure the entire block is in
1412 * ram and up to date before trying to verify things. For
1413 * blocksize <= pagesize, it is basically a noop
1415 if (!(bio
->bi_rw
& (1 << BIO_RW
)) && end_io_wq
->metadata
&&
1416 !bio_ready_for_csum(bio
)) {
1417 btrfs_queue_worker(&fs_info
->endio_meta_workers
,
1421 error
= end_io_wq
->error
;
1422 bio
->bi_private
= end_io_wq
->private;
1423 bio
->bi_end_io
= end_io_wq
->end_io
;
1425 bio_endio(bio
, error
);
1428 static int cleaner_kthread(void *arg
)
1430 struct btrfs_root
*root
= arg
;
1434 if (root
->fs_info
->closing
)
1437 vfs_check_frozen(root
->fs_info
->sb
, SB_FREEZE_WRITE
);
1438 mutex_lock(&root
->fs_info
->cleaner_mutex
);
1439 btrfs_clean_old_snapshots(root
);
1440 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1442 if (freezing(current
)) {
1446 if (root
->fs_info
->closing
)
1448 set_current_state(TASK_INTERRUPTIBLE
);
1450 __set_current_state(TASK_RUNNING
);
1452 } while (!kthread_should_stop());
1456 static int transaction_kthread(void *arg
)
1458 struct btrfs_root
*root
= arg
;
1459 struct btrfs_trans_handle
*trans
;
1460 struct btrfs_transaction
*cur
;
1462 unsigned long delay
;
1467 if (root
->fs_info
->closing
)
1471 vfs_check_frozen(root
->fs_info
->sb
, SB_FREEZE_WRITE
);
1472 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1474 mutex_lock(&root
->fs_info
->trans_mutex
);
1475 cur
= root
->fs_info
->running_transaction
;
1477 mutex_unlock(&root
->fs_info
->trans_mutex
);
1481 now
= get_seconds();
1482 if (now
< cur
->start_time
|| now
- cur
->start_time
< 30) {
1483 mutex_unlock(&root
->fs_info
->trans_mutex
);
1487 mutex_unlock(&root
->fs_info
->trans_mutex
);
1488 trans
= btrfs_start_transaction(root
, 1);
1489 ret
= btrfs_commit_transaction(trans
, root
);
1492 wake_up_process(root
->fs_info
->cleaner_kthread
);
1493 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1495 if (freezing(current
)) {
1498 if (root
->fs_info
->closing
)
1500 set_current_state(TASK_INTERRUPTIBLE
);
1501 schedule_timeout(delay
);
1502 __set_current_state(TASK_RUNNING
);
1504 } while (!kthread_should_stop());
1508 struct btrfs_root
*open_ctree(struct super_block
*sb
,
1509 struct btrfs_fs_devices
*fs_devices
,
1519 struct btrfs_key location
;
1520 struct buffer_head
*bh
;
1521 struct btrfs_root
*extent_root
= kzalloc(sizeof(struct btrfs_root
),
1523 struct btrfs_root
*csum_root
= kzalloc(sizeof(struct btrfs_root
),
1525 struct btrfs_root
*tree_root
= kzalloc(sizeof(struct btrfs_root
),
1527 struct btrfs_fs_info
*fs_info
= kzalloc(sizeof(*fs_info
),
1529 struct btrfs_root
*chunk_root
= kzalloc(sizeof(struct btrfs_root
),
1531 struct btrfs_root
*dev_root
= kzalloc(sizeof(struct btrfs_root
),
1533 struct btrfs_root
*log_tree_root
;
1538 struct btrfs_super_block
*disk_super
;
1540 if (!extent_root
|| !tree_root
|| !fs_info
||
1541 !chunk_root
|| !dev_root
|| !csum_root
) {
1545 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_NOFS
);
1546 INIT_LIST_HEAD(&fs_info
->trans_list
);
1547 INIT_LIST_HEAD(&fs_info
->dead_roots
);
1548 INIT_LIST_HEAD(&fs_info
->hashers
);
1549 INIT_LIST_HEAD(&fs_info
->delalloc_inodes
);
1550 INIT_LIST_HEAD(&fs_info
->ordered_operations
);
1551 spin_lock_init(&fs_info
->delalloc_lock
);
1552 spin_lock_init(&fs_info
->new_trans_lock
);
1553 spin_lock_init(&fs_info
->ref_cache_lock
);
1555 init_completion(&fs_info
->kobj_unregister
);
1556 fs_info
->tree_root
= tree_root
;
1557 fs_info
->extent_root
= extent_root
;
1558 fs_info
->csum_root
= csum_root
;
1559 fs_info
->chunk_root
= chunk_root
;
1560 fs_info
->dev_root
= dev_root
;
1561 fs_info
->fs_devices
= fs_devices
;
1562 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
1563 INIT_LIST_HEAD(&fs_info
->space_info
);
1564 btrfs_mapping_init(&fs_info
->mapping_tree
);
1565 atomic_set(&fs_info
->nr_async_submits
, 0);
1566 atomic_set(&fs_info
->async_delalloc_pages
, 0);
1567 atomic_set(&fs_info
->async_submit_draining
, 0);
1568 atomic_set(&fs_info
->nr_async_bios
, 0);
1570 fs_info
->max_extent
= (u64
)-1;
1571 fs_info
->max_inline
= 8192 * 1024;
1572 setup_bdi(fs_info
, &fs_info
->bdi
);
1573 fs_info
->btree_inode
= new_inode(sb
);
1574 fs_info
->btree_inode
->i_ino
= 1;
1575 fs_info
->btree_inode
->i_nlink
= 1;
1576 fs_info
->metadata_ratio
= 8;
1578 fs_info
->thread_pool_size
= min_t(unsigned long,
1579 num_online_cpus() + 2, 8);
1581 INIT_LIST_HEAD(&fs_info
->ordered_extents
);
1582 spin_lock_init(&fs_info
->ordered_extent_lock
);
1584 sb
->s_blocksize
= 4096;
1585 sb
->s_blocksize_bits
= blksize_bits(4096);
1588 * we set the i_size on the btree inode to the max possible int.
1589 * the real end of the address space is determined by all of
1590 * the devices in the system
1592 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
1593 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
1594 fs_info
->btree_inode
->i_mapping
->backing_dev_info
= &fs_info
->bdi
;
1596 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
1597 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
1598 fs_info
->btree_inode
->i_mapping
,
1600 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
,
1603 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
1605 spin_lock_init(&fs_info
->block_group_cache_lock
);
1606 fs_info
->block_group_cache_tree
.rb_node
= NULL
;
1608 extent_io_tree_init(&fs_info
->pinned_extents
,
1609 fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
1610 fs_info
->do_barriers
= 1;
1612 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
1613 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
1614 sizeof(struct btrfs_key
));
1615 insert_inode_hash(fs_info
->btree_inode
);
1617 mutex_init(&fs_info
->trans_mutex
);
1618 mutex_init(&fs_info
->ordered_operations_mutex
);
1619 mutex_init(&fs_info
->tree_log_mutex
);
1620 mutex_init(&fs_info
->drop_mutex
);
1621 mutex_init(&fs_info
->chunk_mutex
);
1622 mutex_init(&fs_info
->transaction_kthread_mutex
);
1623 mutex_init(&fs_info
->cleaner_mutex
);
1624 mutex_init(&fs_info
->volume_mutex
);
1625 mutex_init(&fs_info
->tree_reloc_mutex
);
1627 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
1628 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
1630 init_waitqueue_head(&fs_info
->transaction_throttle
);
1631 init_waitqueue_head(&fs_info
->transaction_wait
);
1632 init_waitqueue_head(&fs_info
->async_submit_wait
);
1634 __setup_root(4096, 4096, 4096, 4096, tree_root
,
1635 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
1638 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
1642 memcpy(&fs_info
->super_copy
, bh
->b_data
, sizeof(fs_info
->super_copy
));
1643 memcpy(&fs_info
->super_for_commit
, &fs_info
->super_copy
,
1644 sizeof(fs_info
->super_for_commit
));
1647 memcpy(fs_info
->fsid
, fs_info
->super_copy
.fsid
, BTRFS_FSID_SIZE
);
1649 disk_super
= &fs_info
->super_copy
;
1650 if (!btrfs_super_root(disk_super
))
1653 ret
= btrfs_parse_options(tree_root
, options
);
1659 features
= btrfs_super_incompat_flags(disk_super
) &
1660 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
1662 printk(KERN_ERR
"BTRFS: couldn't mount because of "
1663 "unsupported optional features (%Lx).\n",
1664 (unsigned long long)features
);
1669 features
= btrfs_super_incompat_flags(disk_super
);
1670 if (!(features
& BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
)) {
1671 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
1672 btrfs_set_super_incompat_flags(disk_super
, features
);
1675 features
= btrfs_super_compat_ro_flags(disk_super
) &
1676 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
1677 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
1678 printk(KERN_ERR
"BTRFS: couldn't mount RDWR because of "
1679 "unsupported option features (%Lx).\n",
1680 (unsigned long long)features
);
1686 * we need to start all the end_io workers up front because the
1687 * queue work function gets called at interrupt time, and so it
1688 * cannot dynamically grow.
1690 btrfs_init_workers(&fs_info
->workers
, "worker",
1691 fs_info
->thread_pool_size
);
1693 btrfs_init_workers(&fs_info
->delalloc_workers
, "delalloc",
1694 fs_info
->thread_pool_size
);
1696 btrfs_init_workers(&fs_info
->submit_workers
, "submit",
1697 min_t(u64
, fs_devices
->num_devices
,
1698 fs_info
->thread_pool_size
));
1700 /* a higher idle thresh on the submit workers makes it much more
1701 * likely that bios will be send down in a sane order to the
1704 fs_info
->submit_workers
.idle_thresh
= 64;
1706 fs_info
->workers
.idle_thresh
= 16;
1707 fs_info
->workers
.ordered
= 1;
1709 fs_info
->delalloc_workers
.idle_thresh
= 2;
1710 fs_info
->delalloc_workers
.ordered
= 1;
1712 btrfs_init_workers(&fs_info
->fixup_workers
, "fixup", 1);
1713 btrfs_init_workers(&fs_info
->endio_workers
, "endio",
1714 fs_info
->thread_pool_size
);
1715 btrfs_init_workers(&fs_info
->endio_meta_workers
, "endio-meta",
1716 fs_info
->thread_pool_size
);
1717 btrfs_init_workers(&fs_info
->endio_meta_write_workers
,
1718 "endio-meta-write", fs_info
->thread_pool_size
);
1719 btrfs_init_workers(&fs_info
->endio_write_workers
, "endio-write",
1720 fs_info
->thread_pool_size
);
1723 * endios are largely parallel and should have a very
1726 fs_info
->endio_workers
.idle_thresh
= 4;
1727 fs_info
->endio_meta_workers
.idle_thresh
= 4;
1729 fs_info
->endio_write_workers
.idle_thresh
= 64;
1730 fs_info
->endio_meta_write_workers
.idle_thresh
= 64;
1732 btrfs_start_workers(&fs_info
->workers
, 1);
1733 btrfs_start_workers(&fs_info
->submit_workers
, 1);
1734 btrfs_start_workers(&fs_info
->delalloc_workers
, 1);
1735 btrfs_start_workers(&fs_info
->fixup_workers
, 1);
1736 btrfs_start_workers(&fs_info
->endio_workers
, fs_info
->thread_pool_size
);
1737 btrfs_start_workers(&fs_info
->endio_meta_workers
,
1738 fs_info
->thread_pool_size
);
1739 btrfs_start_workers(&fs_info
->endio_meta_write_workers
,
1740 fs_info
->thread_pool_size
);
1741 btrfs_start_workers(&fs_info
->endio_write_workers
,
1742 fs_info
->thread_pool_size
);
1744 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
1745 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
1746 4 * 1024 * 1024 / PAGE_CACHE_SIZE
);
1748 nodesize
= btrfs_super_nodesize(disk_super
);
1749 leafsize
= btrfs_super_leafsize(disk_super
);
1750 sectorsize
= btrfs_super_sectorsize(disk_super
);
1751 stripesize
= btrfs_super_stripesize(disk_super
);
1752 tree_root
->nodesize
= nodesize
;
1753 tree_root
->leafsize
= leafsize
;
1754 tree_root
->sectorsize
= sectorsize
;
1755 tree_root
->stripesize
= stripesize
;
1757 sb
->s_blocksize
= sectorsize
;
1758 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
1760 if (strncmp((char *)(&disk_super
->magic
), BTRFS_MAGIC
,
1761 sizeof(disk_super
->magic
))) {
1762 printk(KERN_INFO
"btrfs: valid FS not found on %s\n", sb
->s_id
);
1763 goto fail_sb_buffer
;
1766 mutex_lock(&fs_info
->chunk_mutex
);
1767 ret
= btrfs_read_sys_array(tree_root
);
1768 mutex_unlock(&fs_info
->chunk_mutex
);
1770 printk(KERN_WARNING
"btrfs: failed to read the system "
1771 "array on %s\n", sb
->s_id
);
1772 goto fail_sb_buffer
;
1775 blocksize
= btrfs_level_size(tree_root
,
1776 btrfs_super_chunk_root_level(disk_super
));
1777 generation
= btrfs_super_chunk_root_generation(disk_super
);
1779 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
1780 chunk_root
, fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
1782 chunk_root
->node
= read_tree_block(chunk_root
,
1783 btrfs_super_chunk_root(disk_super
),
1784 blocksize
, generation
);
1785 BUG_ON(!chunk_root
->node
);
1786 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
1787 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
1789 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
1790 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root
->node
),
1793 mutex_lock(&fs_info
->chunk_mutex
);
1794 ret
= btrfs_read_chunk_tree(chunk_root
);
1795 mutex_unlock(&fs_info
->chunk_mutex
);
1797 printk(KERN_WARNING
"btrfs: failed to read chunk tree on %s\n",
1799 goto fail_chunk_root
;
1802 btrfs_close_extra_devices(fs_devices
);
1804 blocksize
= btrfs_level_size(tree_root
,
1805 btrfs_super_root_level(disk_super
));
1806 generation
= btrfs_super_generation(disk_super
);
1808 tree_root
->node
= read_tree_block(tree_root
,
1809 btrfs_super_root(disk_super
),
1810 blocksize
, generation
);
1811 if (!tree_root
->node
)
1812 goto fail_chunk_root
;
1813 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
1814 tree_root
->commit_root
= btrfs_root_node(tree_root
);
1816 ret
= find_and_setup_root(tree_root
, fs_info
,
1817 BTRFS_EXTENT_TREE_OBJECTID
, extent_root
);
1819 goto fail_tree_root
;
1820 extent_root
->track_dirty
= 1;
1822 ret
= find_and_setup_root(tree_root
, fs_info
,
1823 BTRFS_DEV_TREE_OBJECTID
, dev_root
);
1825 goto fail_extent_root
;
1826 dev_root
->track_dirty
= 1;
1828 ret
= find_and_setup_root(tree_root
, fs_info
,
1829 BTRFS_CSUM_TREE_OBJECTID
, csum_root
);
1833 csum_root
->track_dirty
= 1;
1835 btrfs_read_block_groups(extent_root
);
1837 fs_info
->generation
= generation
;
1838 fs_info
->last_trans_committed
= generation
;
1839 fs_info
->data_alloc_profile
= (u64
)-1;
1840 fs_info
->metadata_alloc_profile
= (u64
)-1;
1841 fs_info
->system_alloc_profile
= fs_info
->metadata_alloc_profile
;
1842 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
1844 if (IS_ERR(fs_info
->cleaner_kthread
))
1845 goto fail_csum_root
;
1847 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
1849 "btrfs-transaction");
1850 if (IS_ERR(fs_info
->transaction_kthread
))
1853 if (!btrfs_test_opt(tree_root
, SSD
) &&
1854 !btrfs_test_opt(tree_root
, NOSSD
) &&
1855 !fs_info
->fs_devices
->rotating
) {
1856 printk(KERN_INFO
"Btrfs detected SSD devices, enabling SSD "
1858 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
1861 if (btrfs_super_log_root(disk_super
) != 0) {
1862 u64 bytenr
= btrfs_super_log_root(disk_super
);
1864 if (fs_devices
->rw_devices
== 0) {
1865 printk(KERN_WARNING
"Btrfs log replay required "
1868 goto fail_trans_kthread
;
1871 btrfs_level_size(tree_root
,
1872 btrfs_super_log_root_level(disk_super
));
1874 log_tree_root
= kzalloc(sizeof(struct btrfs_root
),
1877 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
1878 log_tree_root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
1880 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
1883 ret
= btrfs_recover_log_trees(log_tree_root
);
1886 if (sb
->s_flags
& MS_RDONLY
) {
1887 ret
= btrfs_commit_super(tree_root
);
1892 if (!(sb
->s_flags
& MS_RDONLY
)) {
1893 ret
= btrfs_recover_relocation(tree_root
);
1897 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
1898 location
.type
= BTRFS_ROOT_ITEM_KEY
;
1899 location
.offset
= (u64
)-1;
1901 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
1902 if (!fs_info
->fs_root
)
1903 goto fail_trans_kthread
;
1908 kthread_stop(fs_info
->transaction_kthread
);
1910 kthread_stop(fs_info
->cleaner_kthread
);
1913 * make sure we're done with the btree inode before we stop our
1916 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
1917 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
1920 free_extent_buffer(csum_root
->node
);
1921 free_extent_buffer(csum_root
->commit_root
);
1923 free_extent_buffer(dev_root
->node
);
1924 free_extent_buffer(dev_root
->commit_root
);
1926 free_extent_buffer(extent_root
->node
);
1927 free_extent_buffer(extent_root
->commit_root
);
1929 free_extent_buffer(tree_root
->node
);
1930 free_extent_buffer(tree_root
->commit_root
);
1932 free_extent_buffer(chunk_root
->node
);
1933 free_extent_buffer(chunk_root
->commit_root
);
1935 btrfs_stop_workers(&fs_info
->fixup_workers
);
1936 btrfs_stop_workers(&fs_info
->delalloc_workers
);
1937 btrfs_stop_workers(&fs_info
->workers
);
1938 btrfs_stop_workers(&fs_info
->endio_workers
);
1939 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
1940 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
1941 btrfs_stop_workers(&fs_info
->endio_write_workers
);
1942 btrfs_stop_workers(&fs_info
->submit_workers
);
1944 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
1945 iput(fs_info
->btree_inode
);
1947 btrfs_close_devices(fs_info
->fs_devices
);
1948 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
1949 bdi_destroy(&fs_info
->bdi
);
1958 return ERR_PTR(err
);
1961 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
1963 char b
[BDEVNAME_SIZE
];
1966 set_buffer_uptodate(bh
);
1968 if (!buffer_eopnotsupp(bh
) && printk_ratelimit()) {
1969 printk(KERN_WARNING
"lost page write due to "
1970 "I/O error on %s\n",
1971 bdevname(bh
->b_bdev
, b
));
1973 /* note, we dont' set_buffer_write_io_error because we have
1974 * our own ways of dealing with the IO errors
1976 clear_buffer_uptodate(bh
);
1982 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
1984 struct buffer_head
*bh
;
1985 struct buffer_head
*latest
= NULL
;
1986 struct btrfs_super_block
*super
;
1991 /* we would like to check all the supers, but that would make
1992 * a btrfs mount succeed after a mkfs from a different FS.
1993 * So, we need to add a special mount option to scan for
1994 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1996 for (i
= 0; i
< 1; i
++) {
1997 bytenr
= btrfs_sb_offset(i
);
1998 if (bytenr
+ 4096 >= i_size_read(bdev
->bd_inode
))
2000 bh
= __bread(bdev
, bytenr
/ 4096, 4096);
2004 super
= (struct btrfs_super_block
*)bh
->b_data
;
2005 if (btrfs_super_bytenr(super
) != bytenr
||
2006 strncmp((char *)(&super
->magic
), BTRFS_MAGIC
,
2007 sizeof(super
->magic
))) {
2012 if (!latest
|| btrfs_super_generation(super
) > transid
) {
2015 transid
= btrfs_super_generation(super
);
2024 * this should be called twice, once with wait == 0 and
2025 * once with wait == 1. When wait == 0 is done, all the buffer heads
2026 * we write are pinned.
2028 * They are released when wait == 1 is done.
2029 * max_mirrors must be the same for both runs, and it indicates how
2030 * many supers on this one device should be written.
2032 * max_mirrors == 0 means to write them all.
2034 static int write_dev_supers(struct btrfs_device
*device
,
2035 struct btrfs_super_block
*sb
,
2036 int do_barriers
, int wait
, int max_mirrors
)
2038 struct buffer_head
*bh
;
2044 int last_barrier
= 0;
2046 if (max_mirrors
== 0)
2047 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
2049 /* make sure only the last submit_bh does a barrier */
2051 for (i
= 0; i
< max_mirrors
; i
++) {
2052 bytenr
= btrfs_sb_offset(i
);
2053 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
2054 device
->total_bytes
)
2060 for (i
= 0; i
< max_mirrors
; i
++) {
2061 bytenr
= btrfs_sb_offset(i
);
2062 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= device
->total_bytes
)
2066 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
2067 BTRFS_SUPER_INFO_SIZE
);
2070 if (!buffer_uptodate(bh
))
2073 /* drop our reference */
2076 /* drop the reference from the wait == 0 run */
2080 btrfs_set_super_bytenr(sb
, bytenr
);
2083 crc
= btrfs_csum_data(NULL
, (char *)sb
+
2084 BTRFS_CSUM_SIZE
, crc
,
2085 BTRFS_SUPER_INFO_SIZE
-
2087 btrfs_csum_final(crc
, sb
->csum
);
2090 * one reference for us, and we leave it for the
2093 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
2094 BTRFS_SUPER_INFO_SIZE
);
2095 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
2097 /* one reference for submit_bh */
2100 set_buffer_uptodate(bh
);
2102 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
2105 if (i
== last_barrier
&& do_barriers
&& device
->barriers
) {
2106 ret
= submit_bh(WRITE_BARRIER
, bh
);
2107 if (ret
== -EOPNOTSUPP
) {
2108 printk("btrfs: disabling barriers on dev %s\n",
2110 set_buffer_uptodate(bh
);
2111 device
->barriers
= 0;
2112 /* one reference for submit_bh */
2115 ret
= submit_bh(WRITE_SYNC
, bh
);
2118 ret
= submit_bh(WRITE_SYNC
, bh
);
2124 return errors
< i
? 0 : -1;
2127 int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
2129 struct list_head
*head
;
2130 struct btrfs_device
*dev
;
2131 struct btrfs_super_block
*sb
;
2132 struct btrfs_dev_item
*dev_item
;
2136 int total_errors
= 0;
2139 max_errors
= btrfs_super_num_devices(&root
->fs_info
->super_copy
) - 1;
2140 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
2142 sb
= &root
->fs_info
->super_for_commit
;
2143 dev_item
= &sb
->dev_item
;
2145 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2146 head
= &root
->fs_info
->fs_devices
->devices
;
2147 list_for_each_entry(dev
, head
, dev_list
) {
2152 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
2155 btrfs_set_stack_device_generation(dev_item
, 0);
2156 btrfs_set_stack_device_type(dev_item
, dev
->type
);
2157 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
2158 btrfs_set_stack_device_total_bytes(dev_item
, dev
->total_bytes
);
2159 btrfs_set_stack_device_bytes_used(dev_item
, dev
->bytes_used
);
2160 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
2161 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
2162 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
2163 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
2164 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
2166 flags
= btrfs_super_flags(sb
);
2167 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
2169 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
2173 if (total_errors
> max_errors
) {
2174 printk(KERN_ERR
"btrfs: %d errors while writing supers\n",
2180 list_for_each_entry(dev
, head
, dev_list
) {
2183 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
2186 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
2190 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2191 if (total_errors
> max_errors
) {
2192 printk(KERN_ERR
"btrfs: %d errors while writing supers\n",
2199 int write_ctree_super(struct btrfs_trans_handle
*trans
,
2200 struct btrfs_root
*root
, int max_mirrors
)
2204 ret
= write_all_supers(root
, max_mirrors
);
2208 int btrfs_free_fs_root(struct btrfs_fs_info
*fs_info
, struct btrfs_root
*root
)
2210 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
2211 radix_tree_delete(&fs_info
->fs_roots_radix
,
2212 (unsigned long)root
->root_key
.objectid
);
2213 if (root
->anon_super
.s_dev
) {
2214 down_write(&root
->anon_super
.s_umount
);
2215 kill_anon_super(&root
->anon_super
);
2218 free_extent_buffer(root
->node
);
2219 if (root
->commit_root
)
2220 free_extent_buffer(root
->commit_root
);
2226 static int del_fs_roots(struct btrfs_fs_info
*fs_info
)
2229 struct btrfs_root
*gang
[8];
2233 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2238 for (i
= 0; i
< ret
; i
++)
2239 btrfs_free_fs_root(fs_info
, gang
[i
]);
2244 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
2246 u64 root_objectid
= 0;
2247 struct btrfs_root
*gang
[8];
2252 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2253 (void **)gang
, root_objectid
,
2258 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
2259 for (i
= 0; i
< ret
; i
++) {
2260 root_objectid
= gang
[i
]->root_key
.objectid
;
2261 ret
= btrfs_find_dead_roots(fs_info
->tree_root
,
2264 btrfs_orphan_cleanup(gang
[i
]);
2271 int btrfs_commit_super(struct btrfs_root
*root
)
2273 struct btrfs_trans_handle
*trans
;
2276 mutex_lock(&root
->fs_info
->cleaner_mutex
);
2277 btrfs_clean_old_snapshots(root
);
2278 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
2279 trans
= btrfs_start_transaction(root
, 1);
2280 ret
= btrfs_commit_transaction(trans
, root
);
2282 /* run commit again to drop the original snapshot */
2283 trans
= btrfs_start_transaction(root
, 1);
2284 btrfs_commit_transaction(trans
, root
);
2285 ret
= btrfs_write_and_wait_transaction(NULL
, root
);
2288 ret
= write_ctree_super(NULL
, root
, 0);
2292 int close_ctree(struct btrfs_root
*root
)
2294 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2297 fs_info
->closing
= 1;
2300 kthread_stop(root
->fs_info
->transaction_kthread
);
2301 kthread_stop(root
->fs_info
->cleaner_kthread
);
2303 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
2304 ret
= btrfs_commit_super(root
);
2306 printk(KERN_ERR
"btrfs: commit super ret %d\n", ret
);
2309 if (fs_info
->delalloc_bytes
) {
2310 printk(KERN_INFO
"btrfs: at unmount delalloc count %llu\n",
2311 (unsigned long long)fs_info
->delalloc_bytes
);
2313 if (fs_info
->total_ref_cache_size
) {
2314 printk(KERN_INFO
"btrfs: at umount reference cache size %llu\n",
2315 (unsigned long long)fs_info
->total_ref_cache_size
);
2318 free_extent_buffer(fs_info
->extent_root
->node
);
2319 free_extent_buffer(fs_info
->extent_root
->commit_root
);
2320 free_extent_buffer(fs_info
->tree_root
->node
);
2321 free_extent_buffer(fs_info
->tree_root
->commit_root
);
2322 free_extent_buffer(root
->fs_info
->chunk_root
->node
);
2323 free_extent_buffer(root
->fs_info
->chunk_root
->commit_root
);
2324 free_extent_buffer(root
->fs_info
->dev_root
->node
);
2325 free_extent_buffer(root
->fs_info
->dev_root
->commit_root
);
2326 free_extent_buffer(root
->fs_info
->csum_root
->node
);
2327 free_extent_buffer(root
->fs_info
->csum_root
->commit_root
);
2329 btrfs_free_block_groups(root
->fs_info
);
2331 del_fs_roots(fs_info
);
2333 iput(fs_info
->btree_inode
);
2335 btrfs_stop_workers(&fs_info
->fixup_workers
);
2336 btrfs_stop_workers(&fs_info
->delalloc_workers
);
2337 btrfs_stop_workers(&fs_info
->workers
);
2338 btrfs_stop_workers(&fs_info
->endio_workers
);
2339 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
2340 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
2341 btrfs_stop_workers(&fs_info
->endio_write_workers
);
2342 btrfs_stop_workers(&fs_info
->submit_workers
);
2344 btrfs_close_devices(fs_info
->fs_devices
);
2345 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
2347 bdi_destroy(&fs_info
->bdi
);
2349 kfree(fs_info
->extent_root
);
2350 kfree(fs_info
->tree_root
);
2351 kfree(fs_info
->chunk_root
);
2352 kfree(fs_info
->dev_root
);
2353 kfree(fs_info
->csum_root
);
2357 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
)
2360 struct inode
*btree_inode
= buf
->first_page
->mapping
->host
;
2362 ret
= extent_buffer_uptodate(&BTRFS_I(btree_inode
)->io_tree
, buf
);
2366 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
2371 int btrfs_set_buffer_uptodate(struct extent_buffer
*buf
)
2373 struct inode
*btree_inode
= buf
->first_page
->mapping
->host
;
2374 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode
)->io_tree
,
2378 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
2380 struct btrfs_root
*root
= BTRFS_I(buf
->first_page
->mapping
->host
)->root
;
2381 u64 transid
= btrfs_header_generation(buf
);
2382 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
2385 btrfs_assert_tree_locked(buf
);
2386 if (transid
!= root
->fs_info
->generation
) {
2387 printk(KERN_CRIT
"btrfs transid mismatch buffer %llu, "
2388 "found %llu running %llu\n",
2389 (unsigned long long)buf
->start
,
2390 (unsigned long long)transid
,
2391 (unsigned long long)root
->fs_info
->generation
);
2394 was_dirty
= set_extent_buffer_dirty(&BTRFS_I(btree_inode
)->io_tree
,
2397 spin_lock(&root
->fs_info
->delalloc_lock
);
2398 root
->fs_info
->dirty_metadata_bytes
+= buf
->len
;
2399 spin_unlock(&root
->fs_info
->delalloc_lock
);
2403 void btrfs_btree_balance_dirty(struct btrfs_root
*root
, unsigned long nr
)
2406 * looks as though older kernels can get into trouble with
2407 * this code, they end up stuck in balance_dirty_pages forever
2410 unsigned long thresh
= 32 * 1024 * 1024;
2412 if (current
->flags
& PF_MEMALLOC
)
2415 num_dirty
= root
->fs_info
->dirty_metadata_bytes
;
2417 if (num_dirty
> thresh
) {
2418 balance_dirty_pages_ratelimited_nr(
2419 root
->fs_info
->btree_inode
->i_mapping
, 1);
2424 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
2426 struct btrfs_root
*root
= BTRFS_I(buf
->first_page
->mapping
->host
)->root
;
2428 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
2430 set_bit(EXTENT_BUFFER_UPTODATE
, &buf
->bflags
);
2434 int btree_lock_page_hook(struct page
*page
)
2436 struct inode
*inode
= page
->mapping
->host
;
2437 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2438 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2439 struct extent_buffer
*eb
;
2441 u64 bytenr
= page_offset(page
);
2443 if (page
->private == EXTENT_PAGE_PRIVATE
)
2446 len
= page
->private >> 2;
2447 eb
= find_extent_buffer(io_tree
, bytenr
, len
, GFP_NOFS
);
2451 btrfs_tree_lock(eb
);
2452 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
2454 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
2455 spin_lock(&root
->fs_info
->delalloc_lock
);
2456 if (root
->fs_info
->dirty_metadata_bytes
>= eb
->len
)
2457 root
->fs_info
->dirty_metadata_bytes
-= eb
->len
;
2460 spin_unlock(&root
->fs_info
->delalloc_lock
);
2463 btrfs_tree_unlock(eb
);
2464 free_extent_buffer(eb
);
2470 static struct extent_io_ops btree_extent_io_ops
= {
2471 .write_cache_pages_lock_hook
= btree_lock_page_hook
,
2472 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
2473 .submit_bio_hook
= btree_submit_bio_hook
,
2474 /* note we're sharing with inode.c for the merge bio hook */
2475 .merge_bio_hook
= btrfs_merge_bio_hook
,