2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
34 #include "transaction.h"
35 #include "btrfs_inode.h"
37 #include "print-tree.h"
38 #include "async-thread.h"
41 #include "free-space-cache.h"
43 static struct extent_io_ops btree_extent_io_ops
;
44 static void end_workqueue_fn(struct btrfs_work
*work
);
45 static void free_fs_root(struct btrfs_root
*root
);
47 static atomic_t btrfs_bdi_num
= ATOMIC_INIT(0);
50 * end_io_wq structs are used to do processing in task context when an IO is
51 * complete. This is used during reads to verify checksums, and it is used
52 * by writes to insert metadata for new file extents after IO is complete.
58 struct btrfs_fs_info
*info
;
61 struct list_head list
;
62 struct btrfs_work work
;
66 * async submit bios are used to offload expensive checksumming
67 * onto the worker threads. They checksum file and metadata bios
68 * just before they are sent down the IO stack.
70 struct async_submit_bio
{
73 struct list_head list
;
74 extent_submit_bio_hook_t
*submit_bio_start
;
75 extent_submit_bio_hook_t
*submit_bio_done
;
78 unsigned long bio_flags
;
79 struct btrfs_work work
;
82 /* These are used to set the lockdep class on the extent buffer locks.
83 * The class is set by the readpage_end_io_hook after the buffer has
84 * passed csum validation but before the pages are unlocked.
86 * The lockdep class is also set by btrfs_init_new_buffer on freshly
89 * The class is based on the level in the tree block, which allows lockdep
90 * to know that lower nodes nest inside the locks of higher nodes.
92 * We also add a check to make sure the highest level of the tree is
93 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
94 * code needs update as well.
96 #ifdef CONFIG_DEBUG_LOCK_ALLOC
97 # if BTRFS_MAX_LEVEL != 8
100 static struct lock_class_key btrfs_eb_class
[BTRFS_MAX_LEVEL
+ 1];
101 static const char *btrfs_eb_name
[BTRFS_MAX_LEVEL
+ 1] = {
111 /* highest possible level */
117 * extents on the btree inode are pretty simple, there's one extent
118 * that covers the entire device
120 static struct extent_map
*btree_get_extent(struct inode
*inode
,
121 struct page
*page
, size_t page_offset
, u64 start
, u64 len
,
124 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
125 struct extent_map
*em
;
128 read_lock(&em_tree
->lock
);
129 em
= lookup_extent_mapping(em_tree
, start
, len
);
132 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
133 read_unlock(&em_tree
->lock
);
136 read_unlock(&em_tree
->lock
);
138 em
= alloc_extent_map(GFP_NOFS
);
140 em
= ERR_PTR(-ENOMEM
);
145 em
->block_len
= (u64
)-1;
147 em
->bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
149 write_lock(&em_tree
->lock
);
150 ret
= add_extent_mapping(em_tree
, em
);
151 if (ret
== -EEXIST
) {
152 u64 failed_start
= em
->start
;
153 u64 failed_len
= em
->len
;
156 em
= lookup_extent_mapping(em_tree
, start
, len
);
160 em
= lookup_extent_mapping(em_tree
, failed_start
,
168 write_unlock(&em_tree
->lock
);
176 u32
btrfs_csum_data(struct btrfs_root
*root
, char *data
, u32 seed
, size_t len
)
178 return crc32c(seed
, data
, len
);
181 void btrfs_csum_final(u32 crc
, char *result
)
183 *(__le32
*)result
= ~cpu_to_le32(crc
);
187 * compute the csum for a btree block, and either verify it or write it
188 * into the csum field of the block.
190 static int csum_tree_block(struct btrfs_root
*root
, struct extent_buffer
*buf
,
194 btrfs_super_csum_size(&root
->fs_info
->super_copy
);
197 unsigned long cur_len
;
198 unsigned long offset
= BTRFS_CSUM_SIZE
;
199 char *map_token
= NULL
;
201 unsigned long map_start
;
202 unsigned long map_len
;
205 unsigned long inline_result
;
207 len
= buf
->len
- offset
;
209 err
= map_private_extent_buffer(buf
, offset
, 32,
211 &map_start
, &map_len
, KM_USER0
);
214 cur_len
= min(len
, map_len
- (offset
- map_start
));
215 crc
= btrfs_csum_data(root
, kaddr
+ offset
- map_start
,
219 unmap_extent_buffer(buf
, map_token
, KM_USER0
);
221 if (csum_size
> sizeof(inline_result
)) {
222 result
= kzalloc(csum_size
* sizeof(char), GFP_NOFS
);
226 result
= (char *)&inline_result
;
229 btrfs_csum_final(crc
, result
);
232 if (memcmp_extent_buffer(buf
, result
, 0, csum_size
)) {
235 memcpy(&found
, result
, csum_size
);
237 read_extent_buffer(buf
, &val
, 0, csum_size
);
238 if (printk_ratelimit()) {
239 printk(KERN_INFO
"btrfs: %s checksum verify "
240 "failed on %llu wanted %X found %X "
242 root
->fs_info
->sb
->s_id
,
243 (unsigned long long)buf
->start
, val
, found
,
244 btrfs_header_level(buf
));
246 if (result
!= (char *)&inline_result
)
251 write_extent_buffer(buf
, result
, 0, csum_size
);
253 if (result
!= (char *)&inline_result
)
259 * we can't consider a given block up to date unless the transid of the
260 * block matches the transid in the parent node's pointer. This is how we
261 * detect blocks that either didn't get written at all or got written
262 * in the wrong place.
264 static int verify_parent_transid(struct extent_io_tree
*io_tree
,
265 struct extent_buffer
*eb
, u64 parent_transid
)
267 struct extent_state
*cached_state
= NULL
;
270 if (!parent_transid
|| btrfs_header_generation(eb
) == parent_transid
)
273 lock_extent_bits(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
274 0, &cached_state
, GFP_NOFS
);
275 if (extent_buffer_uptodate(io_tree
, eb
, cached_state
) &&
276 btrfs_header_generation(eb
) == parent_transid
) {
280 if (printk_ratelimit()) {
281 printk("parent transid verify failed on %llu wanted %llu "
283 (unsigned long long)eb
->start
,
284 (unsigned long long)parent_transid
,
285 (unsigned long long)btrfs_header_generation(eb
));
288 clear_extent_buffer_uptodate(io_tree
, eb
, &cached_state
);
290 unlock_extent_cached(io_tree
, eb
->start
, eb
->start
+ eb
->len
- 1,
291 &cached_state
, GFP_NOFS
);
296 * helper to read a given tree block, doing retries as required when
297 * the checksums don't match and we have alternate mirrors to try.
299 static int btree_read_extent_buffer_pages(struct btrfs_root
*root
,
300 struct extent_buffer
*eb
,
301 u64 start
, u64 parent_transid
)
303 struct extent_io_tree
*io_tree
;
308 io_tree
= &BTRFS_I(root
->fs_info
->btree_inode
)->io_tree
;
310 ret
= read_extent_buffer_pages(io_tree
, eb
, start
, 1,
311 btree_get_extent
, mirror_num
);
313 !verify_parent_transid(io_tree
, eb
, parent_transid
))
316 num_copies
= btrfs_num_copies(&root
->fs_info
->mapping_tree
,
322 if (mirror_num
> num_copies
)
329 * checksum a dirty tree block before IO. This has extra checks to make sure
330 * we only fill in the checksum field in the first page of a multi-page block
333 static int csum_dirty_buffer(struct btrfs_root
*root
, struct page
*page
)
335 struct extent_io_tree
*tree
;
336 u64 start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
340 struct extent_buffer
*eb
;
343 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
345 if (page
->private == EXTENT_PAGE_PRIVATE
)
349 len
= page
->private >> 2;
352 eb
= alloc_extent_buffer(tree
, start
, len
, page
, GFP_NOFS
);
353 ret
= btree_read_extent_buffer_pages(root
, eb
, start
+ PAGE_CACHE_SIZE
,
354 btrfs_header_generation(eb
));
356 found_start
= btrfs_header_bytenr(eb
);
357 if (found_start
!= start
) {
361 if (eb
->first_page
!= page
) {
365 if (!PageUptodate(page
)) {
369 found_level
= btrfs_header_level(eb
);
371 csum_tree_block(root
, eb
, 0);
373 free_extent_buffer(eb
);
378 static int check_tree_block_fsid(struct btrfs_root
*root
,
379 struct extent_buffer
*eb
)
381 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
382 u8 fsid
[BTRFS_UUID_SIZE
];
385 read_extent_buffer(eb
, fsid
, (unsigned long)btrfs_header_fsid(eb
),
388 if (!memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
)) {
392 fs_devices
= fs_devices
->seed
;
397 #ifdef CONFIG_DEBUG_LOCK_ALLOC
398 void btrfs_set_buffer_lockdep_class(struct extent_buffer
*eb
, int level
)
400 lockdep_set_class_and_name(&eb
->lock
,
401 &btrfs_eb_class
[level
],
402 btrfs_eb_name
[level
]);
406 static int btree_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
407 struct extent_state
*state
)
409 struct extent_io_tree
*tree
;
413 struct extent_buffer
*eb
;
414 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
417 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
418 if (page
->private == EXTENT_PAGE_PRIVATE
)
423 len
= page
->private >> 2;
426 eb
= alloc_extent_buffer(tree
, start
, len
, page
, GFP_NOFS
);
428 found_start
= btrfs_header_bytenr(eb
);
429 if (found_start
!= start
) {
430 if (printk_ratelimit()) {
431 printk(KERN_INFO
"btrfs bad tree block start "
433 (unsigned long long)found_start
,
434 (unsigned long long)eb
->start
);
439 if (eb
->first_page
!= page
) {
440 printk(KERN_INFO
"btrfs bad first page %lu %lu\n",
441 eb
->first_page
->index
, page
->index
);
446 if (check_tree_block_fsid(root
, eb
)) {
447 if (printk_ratelimit()) {
448 printk(KERN_INFO
"btrfs bad fsid on block %llu\n",
449 (unsigned long long)eb
->start
);
454 found_level
= btrfs_header_level(eb
);
456 btrfs_set_buffer_lockdep_class(eb
, found_level
);
458 ret
= csum_tree_block(root
, eb
, 1);
462 end
= min_t(u64
, eb
->len
, PAGE_CACHE_SIZE
);
463 end
= eb
->start
+ end
- 1;
465 free_extent_buffer(eb
);
470 static void end_workqueue_bio(struct bio
*bio
, int err
)
472 struct end_io_wq
*end_io_wq
= bio
->bi_private
;
473 struct btrfs_fs_info
*fs_info
;
475 fs_info
= end_io_wq
->info
;
476 end_io_wq
->error
= err
;
477 end_io_wq
->work
.func
= end_workqueue_fn
;
478 end_io_wq
->work
.flags
= 0;
480 if (bio
->bi_rw
& (1 << BIO_RW
)) {
481 if (end_io_wq
->metadata
)
482 btrfs_queue_worker(&fs_info
->endio_meta_write_workers
,
485 btrfs_queue_worker(&fs_info
->endio_write_workers
,
488 if (end_io_wq
->metadata
)
489 btrfs_queue_worker(&fs_info
->endio_meta_workers
,
492 btrfs_queue_worker(&fs_info
->endio_workers
,
497 int btrfs_bio_wq_end_io(struct btrfs_fs_info
*info
, struct bio
*bio
,
500 struct end_io_wq
*end_io_wq
;
501 end_io_wq
= kmalloc(sizeof(*end_io_wq
), GFP_NOFS
);
505 end_io_wq
->private = bio
->bi_private
;
506 end_io_wq
->end_io
= bio
->bi_end_io
;
507 end_io_wq
->info
= info
;
508 end_io_wq
->error
= 0;
509 end_io_wq
->bio
= bio
;
510 end_io_wq
->metadata
= metadata
;
512 bio
->bi_private
= end_io_wq
;
513 bio
->bi_end_io
= end_workqueue_bio
;
517 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info
*info
)
519 unsigned long limit
= min_t(unsigned long,
520 info
->workers
.max_workers
,
521 info
->fs_devices
->open_devices
);
525 int btrfs_congested_async(struct btrfs_fs_info
*info
, int iodone
)
527 return atomic_read(&info
->nr_async_bios
) >
528 btrfs_async_submit_limit(info
);
531 static void run_one_async_start(struct btrfs_work
*work
)
533 struct btrfs_fs_info
*fs_info
;
534 struct async_submit_bio
*async
;
536 async
= container_of(work
, struct async_submit_bio
, work
);
537 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
538 async
->submit_bio_start(async
->inode
, async
->rw
, async
->bio
,
539 async
->mirror_num
, async
->bio_flags
);
542 static void run_one_async_done(struct btrfs_work
*work
)
544 struct btrfs_fs_info
*fs_info
;
545 struct async_submit_bio
*async
;
548 async
= container_of(work
, struct async_submit_bio
, work
);
549 fs_info
= BTRFS_I(async
->inode
)->root
->fs_info
;
551 limit
= btrfs_async_submit_limit(fs_info
);
552 limit
= limit
* 2 / 3;
554 atomic_dec(&fs_info
->nr_async_submits
);
556 if (atomic_read(&fs_info
->nr_async_submits
) < limit
&&
557 waitqueue_active(&fs_info
->async_submit_wait
))
558 wake_up(&fs_info
->async_submit_wait
);
560 async
->submit_bio_done(async
->inode
, async
->rw
, async
->bio
,
561 async
->mirror_num
, async
->bio_flags
);
564 static void run_one_async_free(struct btrfs_work
*work
)
566 struct async_submit_bio
*async
;
568 async
= container_of(work
, struct async_submit_bio
, work
);
572 int btrfs_wq_submit_bio(struct btrfs_fs_info
*fs_info
, struct inode
*inode
,
573 int rw
, struct bio
*bio
, int mirror_num
,
574 unsigned long bio_flags
,
575 extent_submit_bio_hook_t
*submit_bio_start
,
576 extent_submit_bio_hook_t
*submit_bio_done
)
578 struct async_submit_bio
*async
;
580 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
584 async
->inode
= inode
;
587 async
->mirror_num
= mirror_num
;
588 async
->submit_bio_start
= submit_bio_start
;
589 async
->submit_bio_done
= submit_bio_done
;
591 async
->work
.func
= run_one_async_start
;
592 async
->work
.ordered_func
= run_one_async_done
;
593 async
->work
.ordered_free
= run_one_async_free
;
595 async
->work
.flags
= 0;
596 async
->bio_flags
= bio_flags
;
598 atomic_inc(&fs_info
->nr_async_submits
);
600 if (rw
& (1 << BIO_RW_SYNCIO
))
601 btrfs_set_work_high_prio(&async
->work
);
603 btrfs_queue_worker(&fs_info
->workers
, &async
->work
);
605 while (atomic_read(&fs_info
->async_submit_draining
) &&
606 atomic_read(&fs_info
->nr_async_submits
)) {
607 wait_event(fs_info
->async_submit_wait
,
608 (atomic_read(&fs_info
->nr_async_submits
) == 0));
614 static int btree_csum_one_bio(struct bio
*bio
)
616 struct bio_vec
*bvec
= bio
->bi_io_vec
;
618 struct btrfs_root
*root
;
620 WARN_ON(bio
->bi_vcnt
<= 0);
621 while (bio_index
< bio
->bi_vcnt
) {
622 root
= BTRFS_I(bvec
->bv_page
->mapping
->host
)->root
;
623 csum_dirty_buffer(root
, bvec
->bv_page
);
630 static int __btree_submit_bio_start(struct inode
*inode
, int rw
,
631 struct bio
*bio
, int mirror_num
,
632 unsigned long bio_flags
)
635 * when we're called for a write, we're already in the async
636 * submission context. Just jump into btrfs_map_bio
638 btree_csum_one_bio(bio
);
642 static int __btree_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
643 int mirror_num
, unsigned long bio_flags
)
646 * when we're called for a write, we're already in the async
647 * submission context. Just jump into btrfs_map_bio
649 return btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
, mirror_num
, 1);
652 static int btree_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
653 int mirror_num
, unsigned long bio_flags
)
657 ret
= btrfs_bio_wq_end_io(BTRFS_I(inode
)->root
->fs_info
,
661 if (!(rw
& (1 << BIO_RW
))) {
663 * called for a read, do the setup so that checksum validation
664 * can happen in the async kernel threads
666 return btrfs_map_bio(BTRFS_I(inode
)->root
, rw
, bio
,
671 * kthread helpers are used to submit writes so that checksumming
672 * can happen in parallel across all CPUs
674 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
675 inode
, rw
, bio
, mirror_num
, 0,
676 __btree_submit_bio_start
,
677 __btree_submit_bio_done
);
680 static int btree_writepage(struct page
*page
, struct writeback_control
*wbc
)
682 struct extent_io_tree
*tree
;
683 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
684 struct extent_buffer
*eb
;
687 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
688 if (!(current
->flags
& PF_MEMALLOC
)) {
689 return extent_write_full_page(tree
, page
,
690 btree_get_extent
, wbc
);
693 redirty_page_for_writepage(wbc
, page
);
694 eb
= btrfs_find_tree_block(root
, page_offset(page
),
698 was_dirty
= test_and_set_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
);
700 spin_lock(&root
->fs_info
->delalloc_lock
);
701 root
->fs_info
->dirty_metadata_bytes
+= PAGE_CACHE_SIZE
;
702 spin_unlock(&root
->fs_info
->delalloc_lock
);
704 free_extent_buffer(eb
);
710 static int btree_writepages(struct address_space
*mapping
,
711 struct writeback_control
*wbc
)
713 struct extent_io_tree
*tree
;
714 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
715 if (wbc
->sync_mode
== WB_SYNC_NONE
) {
716 struct btrfs_root
*root
= BTRFS_I(mapping
->host
)->root
;
718 unsigned long thresh
= 32 * 1024 * 1024;
720 if (wbc
->for_kupdate
)
723 /* this is a bit racy, but that's ok */
724 num_dirty
= root
->fs_info
->dirty_metadata_bytes
;
725 if (num_dirty
< thresh
)
728 return extent_writepages(tree
, mapping
, btree_get_extent
, wbc
);
731 static int btree_readpage(struct file
*file
, struct page
*page
)
733 struct extent_io_tree
*tree
;
734 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
735 return extent_read_full_page(tree
, page
, btree_get_extent
);
738 static int btree_releasepage(struct page
*page
, gfp_t gfp_flags
)
740 struct extent_io_tree
*tree
;
741 struct extent_map_tree
*map
;
744 if (PageWriteback(page
) || PageDirty(page
))
747 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
748 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
750 ret
= try_release_extent_state(map
, tree
, page
, gfp_flags
);
754 ret
= try_release_extent_buffer(tree
, page
);
756 ClearPagePrivate(page
);
757 set_page_private(page
, 0);
758 page_cache_release(page
);
764 static void btree_invalidatepage(struct page
*page
, unsigned long offset
)
766 struct extent_io_tree
*tree
;
767 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
768 extent_invalidatepage(tree
, page
, offset
);
769 btree_releasepage(page
, GFP_NOFS
);
770 if (PagePrivate(page
)) {
771 printk(KERN_WARNING
"btrfs warning page private not zero "
772 "on page %llu\n", (unsigned long long)page_offset(page
));
773 ClearPagePrivate(page
);
774 set_page_private(page
, 0);
775 page_cache_release(page
);
779 static const struct address_space_operations btree_aops
= {
780 .readpage
= btree_readpage
,
781 .writepage
= btree_writepage
,
782 .writepages
= btree_writepages
,
783 .releasepage
= btree_releasepage
,
784 .invalidatepage
= btree_invalidatepage
,
785 .sync_page
= block_sync_page
,
788 int readahead_tree_block(struct btrfs_root
*root
, u64 bytenr
, u32 blocksize
,
791 struct extent_buffer
*buf
= NULL
;
792 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
795 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
798 read_extent_buffer_pages(&BTRFS_I(btree_inode
)->io_tree
,
799 buf
, 0, 0, btree_get_extent
, 0);
800 free_extent_buffer(buf
);
804 struct extent_buffer
*btrfs_find_tree_block(struct btrfs_root
*root
,
805 u64 bytenr
, u32 blocksize
)
807 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
808 struct extent_buffer
*eb
;
809 eb
= find_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
810 bytenr
, blocksize
, GFP_NOFS
);
814 struct extent_buffer
*btrfs_find_create_tree_block(struct btrfs_root
*root
,
815 u64 bytenr
, u32 blocksize
)
817 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
818 struct extent_buffer
*eb
;
820 eb
= alloc_extent_buffer(&BTRFS_I(btree_inode
)->io_tree
,
821 bytenr
, blocksize
, NULL
, GFP_NOFS
);
826 int btrfs_write_tree_block(struct extent_buffer
*buf
)
828 return filemap_fdatawrite_range(buf
->first_page
->mapping
, buf
->start
,
829 buf
->start
+ buf
->len
- 1);
832 int btrfs_wait_tree_block_writeback(struct extent_buffer
*buf
)
834 return filemap_fdatawait_range(buf
->first_page
->mapping
,
835 buf
->start
, buf
->start
+ buf
->len
- 1);
838 struct extent_buffer
*read_tree_block(struct btrfs_root
*root
, u64 bytenr
,
839 u32 blocksize
, u64 parent_transid
)
841 struct extent_buffer
*buf
= NULL
;
842 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
843 struct extent_io_tree
*io_tree
;
846 io_tree
= &BTRFS_I(btree_inode
)->io_tree
;
848 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
852 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
855 set_bit(EXTENT_BUFFER_UPTODATE
, &buf
->bflags
);
860 int clean_tree_block(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
861 struct extent_buffer
*buf
)
863 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
864 if (btrfs_header_generation(buf
) ==
865 root
->fs_info
->running_transaction
->transid
) {
866 btrfs_assert_tree_locked(buf
);
868 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
)) {
869 spin_lock(&root
->fs_info
->delalloc_lock
);
870 if (root
->fs_info
->dirty_metadata_bytes
>= buf
->len
)
871 root
->fs_info
->dirty_metadata_bytes
-= buf
->len
;
874 spin_unlock(&root
->fs_info
->delalloc_lock
);
877 /* ugh, clear_extent_buffer_dirty needs to lock the page */
878 btrfs_set_lock_blocking(buf
);
879 clear_extent_buffer_dirty(&BTRFS_I(btree_inode
)->io_tree
,
885 static int __setup_root(u32 nodesize
, u32 leafsize
, u32 sectorsize
,
886 u32 stripesize
, struct btrfs_root
*root
,
887 struct btrfs_fs_info
*fs_info
,
891 root
->commit_root
= NULL
;
892 root
->sectorsize
= sectorsize
;
893 root
->nodesize
= nodesize
;
894 root
->leafsize
= leafsize
;
895 root
->stripesize
= stripesize
;
897 root
->track_dirty
= 0;
899 root
->clean_orphans
= 0;
901 root
->fs_info
= fs_info
;
902 root
->objectid
= objectid
;
903 root
->last_trans
= 0;
904 root
->highest_objectid
= 0;
907 root
->inode_tree
= RB_ROOT
;
909 INIT_LIST_HEAD(&root
->dirty_list
);
910 INIT_LIST_HEAD(&root
->orphan_list
);
911 INIT_LIST_HEAD(&root
->root_list
);
912 spin_lock_init(&root
->node_lock
);
913 spin_lock_init(&root
->list_lock
);
914 spin_lock_init(&root
->inode_lock
);
915 mutex_init(&root
->objectid_mutex
);
916 mutex_init(&root
->log_mutex
);
917 init_waitqueue_head(&root
->log_writer_wait
);
918 init_waitqueue_head(&root
->log_commit_wait
[0]);
919 init_waitqueue_head(&root
->log_commit_wait
[1]);
920 atomic_set(&root
->log_commit
[0], 0);
921 atomic_set(&root
->log_commit
[1], 0);
922 atomic_set(&root
->log_writers
, 0);
924 root
->log_transid
= 0;
925 root
->last_log_commit
= 0;
926 extent_io_tree_init(&root
->dirty_log_pages
,
927 fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
929 memset(&root
->root_key
, 0, sizeof(root
->root_key
));
930 memset(&root
->root_item
, 0, sizeof(root
->root_item
));
931 memset(&root
->defrag_progress
, 0, sizeof(root
->defrag_progress
));
932 memset(&root
->root_kobj
, 0, sizeof(root
->root_kobj
));
933 root
->defrag_trans_start
= fs_info
->generation
;
934 init_completion(&root
->kobj_unregister
);
935 root
->defrag_running
= 0;
936 root
->root_key
.objectid
= objectid
;
937 root
->anon_super
.s_root
= NULL
;
938 root
->anon_super
.s_dev
= 0;
939 INIT_LIST_HEAD(&root
->anon_super
.s_list
);
940 INIT_LIST_HEAD(&root
->anon_super
.s_instances
);
941 init_rwsem(&root
->anon_super
.s_umount
);
946 static int find_and_setup_root(struct btrfs_root
*tree_root
,
947 struct btrfs_fs_info
*fs_info
,
949 struct btrfs_root
*root
)
955 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
956 tree_root
->sectorsize
, tree_root
->stripesize
,
957 root
, fs_info
, objectid
);
958 ret
= btrfs_find_last_root(tree_root
, objectid
,
959 &root
->root_item
, &root
->root_key
);
964 generation
= btrfs_root_generation(&root
->root_item
);
965 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
966 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
967 blocksize
, generation
);
969 root
->commit_root
= btrfs_root_node(root
);
973 int btrfs_free_log_root_tree(struct btrfs_trans_handle
*trans
,
974 struct btrfs_fs_info
*fs_info
)
976 struct extent_buffer
*eb
;
977 struct btrfs_root
*log_root_tree
= fs_info
->log_root_tree
;
986 ret
= find_first_extent_bit(&log_root_tree
->dirty_log_pages
,
987 0, &start
, &end
, EXTENT_DIRTY
| EXTENT_NEW
);
991 clear_extent_bits(&log_root_tree
->dirty_log_pages
, start
, end
,
992 EXTENT_DIRTY
| EXTENT_NEW
, GFP_NOFS
);
994 eb
= fs_info
->log_root_tree
->node
;
996 WARN_ON(btrfs_header_level(eb
) != 0);
997 WARN_ON(btrfs_header_nritems(eb
) != 0);
999 ret
= btrfs_free_reserved_extent(fs_info
->tree_root
,
1000 eb
->start
, eb
->len
);
1003 free_extent_buffer(eb
);
1004 kfree(fs_info
->log_root_tree
);
1005 fs_info
->log_root_tree
= NULL
;
1009 static struct btrfs_root
*alloc_log_tree(struct btrfs_trans_handle
*trans
,
1010 struct btrfs_fs_info
*fs_info
)
1012 struct btrfs_root
*root
;
1013 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1014 struct extent_buffer
*leaf
;
1016 root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1018 return ERR_PTR(-ENOMEM
);
1020 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1021 tree_root
->sectorsize
, tree_root
->stripesize
,
1022 root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
1024 root
->root_key
.objectid
= BTRFS_TREE_LOG_OBJECTID
;
1025 root
->root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
1026 root
->root_key
.offset
= BTRFS_TREE_LOG_OBJECTID
;
1028 * log trees do not get reference counted because they go away
1029 * before a real commit is actually done. They do store pointers
1030 * to file data extents, and those reference counts still get
1031 * updated (along with back refs to the log tree).
1035 leaf
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
, 0,
1036 BTRFS_TREE_LOG_OBJECTID
, NULL
, 0, 0, 0);
1039 return ERR_CAST(leaf
);
1042 memset_extent_buffer(leaf
, 0, 0, sizeof(struct btrfs_header
));
1043 btrfs_set_header_bytenr(leaf
, leaf
->start
);
1044 btrfs_set_header_generation(leaf
, trans
->transid
);
1045 btrfs_set_header_backref_rev(leaf
, BTRFS_MIXED_BACKREF_REV
);
1046 btrfs_set_header_owner(leaf
, BTRFS_TREE_LOG_OBJECTID
);
1049 write_extent_buffer(root
->node
, root
->fs_info
->fsid
,
1050 (unsigned long)btrfs_header_fsid(root
->node
),
1052 btrfs_mark_buffer_dirty(root
->node
);
1053 btrfs_tree_unlock(root
->node
);
1057 int btrfs_init_log_root_tree(struct btrfs_trans_handle
*trans
,
1058 struct btrfs_fs_info
*fs_info
)
1060 struct btrfs_root
*log_root
;
1062 log_root
= alloc_log_tree(trans
, fs_info
);
1063 if (IS_ERR(log_root
))
1064 return PTR_ERR(log_root
);
1065 WARN_ON(fs_info
->log_root_tree
);
1066 fs_info
->log_root_tree
= log_root
;
1070 int btrfs_add_log_tree(struct btrfs_trans_handle
*trans
,
1071 struct btrfs_root
*root
)
1073 struct btrfs_root
*log_root
;
1074 struct btrfs_inode_item
*inode_item
;
1076 log_root
= alloc_log_tree(trans
, root
->fs_info
);
1077 if (IS_ERR(log_root
))
1078 return PTR_ERR(log_root
);
1080 log_root
->last_trans
= trans
->transid
;
1081 log_root
->root_key
.offset
= root
->root_key
.objectid
;
1083 inode_item
= &log_root
->root_item
.inode
;
1084 inode_item
->generation
= cpu_to_le64(1);
1085 inode_item
->size
= cpu_to_le64(3);
1086 inode_item
->nlink
= cpu_to_le32(1);
1087 inode_item
->nbytes
= cpu_to_le64(root
->leafsize
);
1088 inode_item
->mode
= cpu_to_le32(S_IFDIR
| 0755);
1090 btrfs_set_root_node(&log_root
->root_item
, log_root
->node
);
1092 WARN_ON(root
->log_root
);
1093 root
->log_root
= log_root
;
1094 root
->log_transid
= 0;
1095 root
->last_log_commit
= 0;
1099 struct btrfs_root
*btrfs_read_fs_root_no_radix(struct btrfs_root
*tree_root
,
1100 struct btrfs_key
*location
)
1102 struct btrfs_root
*root
;
1103 struct btrfs_fs_info
*fs_info
= tree_root
->fs_info
;
1104 struct btrfs_path
*path
;
1105 struct extent_buffer
*l
;
1110 root
= kzalloc(sizeof(*root
), GFP_NOFS
);
1112 return ERR_PTR(-ENOMEM
);
1113 if (location
->offset
== (u64
)-1) {
1114 ret
= find_and_setup_root(tree_root
, fs_info
,
1115 location
->objectid
, root
);
1118 return ERR_PTR(ret
);
1123 __setup_root(tree_root
->nodesize
, tree_root
->leafsize
,
1124 tree_root
->sectorsize
, tree_root
->stripesize
,
1125 root
, fs_info
, location
->objectid
);
1127 path
= btrfs_alloc_path();
1129 ret
= btrfs_search_slot(NULL
, tree_root
, location
, path
, 0, 0);
1132 read_extent_buffer(l
, &root
->root_item
,
1133 btrfs_item_ptr_offset(l
, path
->slots
[0]),
1134 sizeof(root
->root_item
));
1135 memcpy(&root
->root_key
, location
, sizeof(*location
));
1137 btrfs_free_path(path
);
1141 return ERR_PTR(ret
);
1144 generation
= btrfs_root_generation(&root
->root_item
);
1145 blocksize
= btrfs_level_size(root
, btrfs_root_level(&root
->root_item
));
1146 root
->node
= read_tree_block(root
, btrfs_root_bytenr(&root
->root_item
),
1147 blocksize
, generation
);
1148 root
->commit_root
= btrfs_root_node(root
);
1149 BUG_ON(!root
->node
);
1151 if (location
->objectid
!= BTRFS_TREE_LOG_OBJECTID
)
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 spin_lock(&fs_info
->fs_roots_radix_lock
);
1190 root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
1191 (unsigned long)location
->objectid
);
1192 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1196 ret
= btrfs_find_orphan_item(fs_info
->tree_root
, location
->objectid
);
1200 return ERR_PTR(ret
);
1202 root
= btrfs_read_fs_root_no_radix(fs_info
->tree_root
, location
);
1206 WARN_ON(btrfs_root_refs(&root
->root_item
) == 0);
1207 set_anon_super(&root
->anon_super
, NULL
);
1209 ret
= radix_tree_preload(GFP_NOFS
& ~__GFP_HIGHMEM
);
1213 spin_lock(&fs_info
->fs_roots_radix_lock
);
1214 ret
= radix_tree_insert(&fs_info
->fs_roots_radix
,
1215 (unsigned long)root
->root_key
.objectid
,
1219 root
->clean_orphans
= 1;
1221 spin_unlock(&fs_info
->fs_roots_radix_lock
);
1222 radix_tree_preload_end();
1224 if (ret
== -EEXIST
) {
1231 ret
= btrfs_find_dead_roots(fs_info
->tree_root
,
1232 root
->root_key
.objectid
);
1237 return ERR_PTR(ret
);
1240 struct btrfs_root
*btrfs_read_fs_root(struct btrfs_fs_info
*fs_info
,
1241 struct btrfs_key
*location
,
1242 const char *name
, int namelen
)
1244 return btrfs_read_fs_root_no_name(fs_info
, location
);
1246 struct btrfs_root
*root
;
1249 root
= btrfs_read_fs_root_no_name(fs_info
, location
);
1256 ret
= btrfs_set_root_name(root
, name
, namelen
);
1258 free_extent_buffer(root
->node
);
1260 return ERR_PTR(ret
);
1263 ret
= btrfs_sysfs_add_root(root
);
1265 free_extent_buffer(root
->node
);
1268 return ERR_PTR(ret
);
1275 static int btrfs_congested_fn(void *congested_data
, int bdi_bits
)
1277 struct btrfs_fs_info
*info
= (struct btrfs_fs_info
*)congested_data
;
1279 struct btrfs_device
*device
;
1280 struct backing_dev_info
*bdi
;
1282 list_for_each_entry(device
, &info
->fs_devices
->devices
, dev_list
) {
1285 bdi
= blk_get_backing_dev_info(device
->bdev
);
1286 if (bdi
&& bdi_congested(bdi
, bdi_bits
)) {
1295 * this unplugs every device on the box, and it is only used when page
1298 static void __unplug_io_fn(struct backing_dev_info
*bdi
, struct page
*page
)
1300 struct btrfs_device
*device
;
1301 struct btrfs_fs_info
*info
;
1303 info
= (struct btrfs_fs_info
*)bdi
->unplug_io_data
;
1304 list_for_each_entry(device
, &info
->fs_devices
->devices
, dev_list
) {
1308 bdi
= blk_get_backing_dev_info(device
->bdev
);
1309 if (bdi
->unplug_io_fn
)
1310 bdi
->unplug_io_fn(bdi
, page
);
1314 static void btrfs_unplug_io_fn(struct backing_dev_info
*bdi
, struct page
*page
)
1316 struct inode
*inode
;
1317 struct extent_map_tree
*em_tree
;
1318 struct extent_map
*em
;
1319 struct address_space
*mapping
;
1322 /* the generic O_DIRECT read code does this */
1324 __unplug_io_fn(bdi
, page
);
1329 * page->mapping may change at any time. Get a consistent copy
1330 * and use that for everything below
1333 mapping
= page
->mapping
;
1337 inode
= mapping
->host
;
1340 * don't do the expensive searching for a small number of
1343 if (BTRFS_I(inode
)->root
->fs_info
->fs_devices
->open_devices
<= 2) {
1344 __unplug_io_fn(bdi
, page
);
1348 offset
= page_offset(page
);
1350 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1351 read_lock(&em_tree
->lock
);
1352 em
= lookup_extent_mapping(em_tree
, offset
, PAGE_CACHE_SIZE
);
1353 read_unlock(&em_tree
->lock
);
1355 __unplug_io_fn(bdi
, page
);
1359 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
1360 free_extent_map(em
);
1361 __unplug_io_fn(bdi
, page
);
1364 offset
= offset
- em
->start
;
1365 btrfs_unplug_page(&BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1366 em
->block_start
+ offset
, page
);
1367 free_extent_map(em
);
1371 * If this fails, caller must call bdi_destroy() to get rid of the
1374 static int setup_bdi(struct btrfs_fs_info
*info
, struct backing_dev_info
*bdi
)
1378 bdi
->name
= "btrfs";
1379 bdi
->capabilities
= BDI_CAP_MAP_COPY
;
1380 err
= bdi_init(bdi
);
1384 err
= bdi_register(bdi
, NULL
, "btrfs-%d",
1385 atomic_inc_return(&btrfs_bdi_num
));
1391 bdi
->ra_pages
= default_backing_dev_info
.ra_pages
;
1392 bdi
->unplug_io_fn
= btrfs_unplug_io_fn
;
1393 bdi
->unplug_io_data
= info
;
1394 bdi
->congested_fn
= btrfs_congested_fn
;
1395 bdi
->congested_data
= info
;
1399 static int bio_ready_for_csum(struct bio
*bio
)
1405 struct extent_io_tree
*io_tree
= NULL
;
1406 struct btrfs_fs_info
*info
= NULL
;
1407 struct bio_vec
*bvec
;
1411 bio_for_each_segment(bvec
, bio
, i
) {
1412 page
= bvec
->bv_page
;
1413 if (page
->private == EXTENT_PAGE_PRIVATE
) {
1414 length
+= bvec
->bv_len
;
1417 if (!page
->private) {
1418 length
+= bvec
->bv_len
;
1421 length
= bvec
->bv_len
;
1422 buf_len
= page
->private >> 2;
1423 start
= page_offset(page
) + bvec
->bv_offset
;
1424 io_tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
1425 info
= BTRFS_I(page
->mapping
->host
)->root
->fs_info
;
1427 /* are we fully contained in this bio? */
1428 if (buf_len
<= length
)
1431 ret
= extent_range_uptodate(io_tree
, start
+ length
,
1432 start
+ buf_len
- 1);
1437 * called by the kthread helper functions to finally call the bio end_io
1438 * functions. This is where read checksum verification actually happens
1440 static void end_workqueue_fn(struct btrfs_work
*work
)
1443 struct end_io_wq
*end_io_wq
;
1444 struct btrfs_fs_info
*fs_info
;
1447 end_io_wq
= container_of(work
, struct end_io_wq
, work
);
1448 bio
= end_io_wq
->bio
;
1449 fs_info
= end_io_wq
->info
;
1451 /* metadata bio reads are special because the whole tree block must
1452 * be checksummed at once. This makes sure the entire block is in
1453 * ram and up to date before trying to verify things. For
1454 * blocksize <= pagesize, it is basically a noop
1456 if (!(bio
->bi_rw
& (1 << BIO_RW
)) && end_io_wq
->metadata
&&
1457 !bio_ready_for_csum(bio
)) {
1458 btrfs_queue_worker(&fs_info
->endio_meta_workers
,
1462 error
= end_io_wq
->error
;
1463 bio
->bi_private
= end_io_wq
->private;
1464 bio
->bi_end_io
= end_io_wq
->end_io
;
1466 bio_endio(bio
, error
);
1469 static int cleaner_kthread(void *arg
)
1471 struct btrfs_root
*root
= arg
;
1475 if (root
->fs_info
->closing
)
1478 vfs_check_frozen(root
->fs_info
->sb
, SB_FREEZE_WRITE
);
1480 if (!(root
->fs_info
->sb
->s_flags
& MS_RDONLY
) &&
1481 mutex_trylock(&root
->fs_info
->cleaner_mutex
)) {
1482 btrfs_run_delayed_iputs(root
);
1483 btrfs_clean_old_snapshots(root
);
1484 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
1487 if (freezing(current
)) {
1491 if (root
->fs_info
->closing
)
1493 set_current_state(TASK_INTERRUPTIBLE
);
1495 __set_current_state(TASK_RUNNING
);
1497 } while (!kthread_should_stop());
1501 static int transaction_kthread(void *arg
)
1503 struct btrfs_root
*root
= arg
;
1504 struct btrfs_trans_handle
*trans
;
1505 struct btrfs_transaction
*cur
;
1507 unsigned long delay
;
1512 if (root
->fs_info
->closing
)
1516 vfs_check_frozen(root
->fs_info
->sb
, SB_FREEZE_WRITE
);
1517 mutex_lock(&root
->fs_info
->transaction_kthread_mutex
);
1519 mutex_lock(&root
->fs_info
->trans_mutex
);
1520 cur
= root
->fs_info
->running_transaction
;
1522 mutex_unlock(&root
->fs_info
->trans_mutex
);
1526 now
= get_seconds();
1527 if (now
< cur
->start_time
|| now
- cur
->start_time
< 30) {
1528 mutex_unlock(&root
->fs_info
->trans_mutex
);
1532 mutex_unlock(&root
->fs_info
->trans_mutex
);
1533 trans
= btrfs_start_transaction(root
, 1);
1534 ret
= btrfs_commit_transaction(trans
, root
);
1537 wake_up_process(root
->fs_info
->cleaner_kthread
);
1538 mutex_unlock(&root
->fs_info
->transaction_kthread_mutex
);
1540 if (freezing(current
)) {
1543 if (root
->fs_info
->closing
)
1545 set_current_state(TASK_INTERRUPTIBLE
);
1546 schedule_timeout(delay
);
1547 __set_current_state(TASK_RUNNING
);
1549 } while (!kthread_should_stop());
1553 struct btrfs_root
*open_ctree(struct super_block
*sb
,
1554 struct btrfs_fs_devices
*fs_devices
,
1564 struct btrfs_key location
;
1565 struct buffer_head
*bh
;
1566 struct btrfs_root
*extent_root
= kzalloc(sizeof(struct btrfs_root
),
1568 struct btrfs_root
*csum_root
= kzalloc(sizeof(struct btrfs_root
),
1570 struct btrfs_root
*tree_root
= kzalloc(sizeof(struct btrfs_root
),
1572 struct btrfs_fs_info
*fs_info
= kzalloc(sizeof(*fs_info
),
1574 struct btrfs_root
*chunk_root
= kzalloc(sizeof(struct btrfs_root
),
1576 struct btrfs_root
*dev_root
= kzalloc(sizeof(struct btrfs_root
),
1578 struct btrfs_root
*log_tree_root
;
1583 struct btrfs_super_block
*disk_super
;
1585 if (!extent_root
|| !tree_root
|| !fs_info
||
1586 !chunk_root
|| !dev_root
|| !csum_root
) {
1591 ret
= init_srcu_struct(&fs_info
->subvol_srcu
);
1597 ret
= setup_bdi(fs_info
, &fs_info
->bdi
);
1603 fs_info
->btree_inode
= new_inode(sb
);
1604 if (!fs_info
->btree_inode
) {
1609 INIT_RADIX_TREE(&fs_info
->fs_roots_radix
, GFP_ATOMIC
);
1610 INIT_LIST_HEAD(&fs_info
->trans_list
);
1611 INIT_LIST_HEAD(&fs_info
->dead_roots
);
1612 INIT_LIST_HEAD(&fs_info
->delayed_iputs
);
1613 INIT_LIST_HEAD(&fs_info
->hashers
);
1614 INIT_LIST_HEAD(&fs_info
->delalloc_inodes
);
1615 INIT_LIST_HEAD(&fs_info
->ordered_operations
);
1616 INIT_LIST_HEAD(&fs_info
->caching_block_groups
);
1617 spin_lock_init(&fs_info
->delalloc_lock
);
1618 spin_lock_init(&fs_info
->new_trans_lock
);
1619 spin_lock_init(&fs_info
->ref_cache_lock
);
1620 spin_lock_init(&fs_info
->fs_roots_radix_lock
);
1621 spin_lock_init(&fs_info
->delayed_iput_lock
);
1623 init_completion(&fs_info
->kobj_unregister
);
1624 fs_info
->tree_root
= tree_root
;
1625 fs_info
->extent_root
= extent_root
;
1626 fs_info
->csum_root
= csum_root
;
1627 fs_info
->chunk_root
= chunk_root
;
1628 fs_info
->dev_root
= dev_root
;
1629 fs_info
->fs_devices
= fs_devices
;
1630 INIT_LIST_HEAD(&fs_info
->dirty_cowonly_roots
);
1631 INIT_LIST_HEAD(&fs_info
->space_info
);
1632 btrfs_mapping_init(&fs_info
->mapping_tree
);
1633 atomic_set(&fs_info
->nr_async_submits
, 0);
1634 atomic_set(&fs_info
->async_delalloc_pages
, 0);
1635 atomic_set(&fs_info
->async_submit_draining
, 0);
1636 atomic_set(&fs_info
->nr_async_bios
, 0);
1638 fs_info
->max_extent
= (u64
)-1;
1639 fs_info
->max_inline
= 8192 * 1024;
1640 fs_info
->metadata_ratio
= 0;
1642 fs_info
->thread_pool_size
= min_t(unsigned long,
1643 num_online_cpus() + 2, 8);
1645 INIT_LIST_HEAD(&fs_info
->ordered_extents
);
1646 spin_lock_init(&fs_info
->ordered_extent_lock
);
1648 sb
->s_blocksize
= 4096;
1649 sb
->s_blocksize_bits
= blksize_bits(4096);
1650 sb
->s_bdi
= &fs_info
->bdi
;
1652 fs_info
->btree_inode
->i_ino
= BTRFS_BTREE_INODE_OBJECTID
;
1653 fs_info
->btree_inode
->i_nlink
= 1;
1655 * we set the i_size on the btree inode to the max possible int.
1656 * the real end of the address space is determined by all of
1657 * the devices in the system
1659 fs_info
->btree_inode
->i_size
= OFFSET_MAX
;
1660 fs_info
->btree_inode
->i_mapping
->a_ops
= &btree_aops
;
1661 fs_info
->btree_inode
->i_mapping
->backing_dev_info
= &fs_info
->bdi
;
1663 RB_CLEAR_NODE(&BTRFS_I(fs_info
->btree_inode
)->rb_node
);
1664 extent_io_tree_init(&BTRFS_I(fs_info
->btree_inode
)->io_tree
,
1665 fs_info
->btree_inode
->i_mapping
,
1667 extent_map_tree_init(&BTRFS_I(fs_info
->btree_inode
)->extent_tree
,
1670 BTRFS_I(fs_info
->btree_inode
)->io_tree
.ops
= &btree_extent_io_ops
;
1672 BTRFS_I(fs_info
->btree_inode
)->root
= tree_root
;
1673 memset(&BTRFS_I(fs_info
->btree_inode
)->location
, 0,
1674 sizeof(struct btrfs_key
));
1675 BTRFS_I(fs_info
->btree_inode
)->dummy_inode
= 1;
1676 insert_inode_hash(fs_info
->btree_inode
);
1678 spin_lock_init(&fs_info
->block_group_cache_lock
);
1679 fs_info
->block_group_cache_tree
= RB_ROOT
;
1681 extent_io_tree_init(&fs_info
->freed_extents
[0],
1682 fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
1683 extent_io_tree_init(&fs_info
->freed_extents
[1],
1684 fs_info
->btree_inode
->i_mapping
, GFP_NOFS
);
1685 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
1686 fs_info
->do_barriers
= 1;
1689 mutex_init(&fs_info
->trans_mutex
);
1690 mutex_init(&fs_info
->ordered_operations_mutex
);
1691 mutex_init(&fs_info
->tree_log_mutex
);
1692 mutex_init(&fs_info
->chunk_mutex
);
1693 mutex_init(&fs_info
->transaction_kthread_mutex
);
1694 mutex_init(&fs_info
->cleaner_mutex
);
1695 mutex_init(&fs_info
->volume_mutex
);
1696 init_rwsem(&fs_info
->extent_commit_sem
);
1697 init_rwsem(&fs_info
->cleanup_work_sem
);
1698 init_rwsem(&fs_info
->subvol_sem
);
1700 btrfs_init_free_cluster(&fs_info
->meta_alloc_cluster
);
1701 btrfs_init_free_cluster(&fs_info
->data_alloc_cluster
);
1703 init_waitqueue_head(&fs_info
->transaction_throttle
);
1704 init_waitqueue_head(&fs_info
->transaction_wait
);
1705 init_waitqueue_head(&fs_info
->async_submit_wait
);
1707 __setup_root(4096, 4096, 4096, 4096, tree_root
,
1708 fs_info
, BTRFS_ROOT_TREE_OBJECTID
);
1711 bh
= btrfs_read_dev_super(fs_devices
->latest_bdev
);
1715 memcpy(&fs_info
->super_copy
, bh
->b_data
, sizeof(fs_info
->super_copy
));
1716 memcpy(&fs_info
->super_for_commit
, &fs_info
->super_copy
,
1717 sizeof(fs_info
->super_for_commit
));
1720 memcpy(fs_info
->fsid
, fs_info
->super_copy
.fsid
, BTRFS_FSID_SIZE
);
1722 disk_super
= &fs_info
->super_copy
;
1723 if (!btrfs_super_root(disk_super
))
1726 ret
= btrfs_parse_options(tree_root
, options
);
1732 features
= btrfs_super_incompat_flags(disk_super
) &
1733 ~BTRFS_FEATURE_INCOMPAT_SUPP
;
1735 printk(KERN_ERR
"BTRFS: couldn't mount because of "
1736 "unsupported optional features (%Lx).\n",
1737 (unsigned long long)features
);
1742 features
= btrfs_super_incompat_flags(disk_super
);
1743 if (!(features
& BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
)) {
1744 features
|= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF
;
1745 btrfs_set_super_incompat_flags(disk_super
, features
);
1748 features
= btrfs_super_compat_ro_flags(disk_super
) &
1749 ~BTRFS_FEATURE_COMPAT_RO_SUPP
;
1750 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
1751 printk(KERN_ERR
"BTRFS: couldn't mount RDWR because of "
1752 "unsupported option features (%Lx).\n",
1753 (unsigned long long)features
);
1758 btrfs_init_workers(&fs_info
->generic_worker
,
1759 "genwork", 1, NULL
);
1761 btrfs_init_workers(&fs_info
->workers
, "worker",
1762 fs_info
->thread_pool_size
,
1763 &fs_info
->generic_worker
);
1765 btrfs_init_workers(&fs_info
->delalloc_workers
, "delalloc",
1766 fs_info
->thread_pool_size
,
1767 &fs_info
->generic_worker
);
1769 btrfs_init_workers(&fs_info
->submit_workers
, "submit",
1770 min_t(u64
, fs_devices
->num_devices
,
1771 fs_info
->thread_pool_size
),
1772 &fs_info
->generic_worker
);
1773 btrfs_init_workers(&fs_info
->enospc_workers
, "enospc",
1774 fs_info
->thread_pool_size
,
1775 &fs_info
->generic_worker
);
1777 /* a higher idle thresh on the submit workers makes it much more
1778 * likely that bios will be send down in a sane order to the
1781 fs_info
->submit_workers
.idle_thresh
= 64;
1783 fs_info
->workers
.idle_thresh
= 16;
1784 fs_info
->workers
.ordered
= 1;
1786 fs_info
->delalloc_workers
.idle_thresh
= 2;
1787 fs_info
->delalloc_workers
.ordered
= 1;
1789 btrfs_init_workers(&fs_info
->fixup_workers
, "fixup", 1,
1790 &fs_info
->generic_worker
);
1791 btrfs_init_workers(&fs_info
->endio_workers
, "endio",
1792 fs_info
->thread_pool_size
,
1793 &fs_info
->generic_worker
);
1794 btrfs_init_workers(&fs_info
->endio_meta_workers
, "endio-meta",
1795 fs_info
->thread_pool_size
,
1796 &fs_info
->generic_worker
);
1797 btrfs_init_workers(&fs_info
->endio_meta_write_workers
,
1798 "endio-meta-write", fs_info
->thread_pool_size
,
1799 &fs_info
->generic_worker
);
1800 btrfs_init_workers(&fs_info
->endio_write_workers
, "endio-write",
1801 fs_info
->thread_pool_size
,
1802 &fs_info
->generic_worker
);
1805 * endios are largely parallel and should have a very
1808 fs_info
->endio_workers
.idle_thresh
= 4;
1809 fs_info
->endio_meta_workers
.idle_thresh
= 4;
1811 fs_info
->endio_write_workers
.idle_thresh
= 2;
1812 fs_info
->endio_meta_write_workers
.idle_thresh
= 2;
1814 btrfs_start_workers(&fs_info
->workers
, 1);
1815 btrfs_start_workers(&fs_info
->generic_worker
, 1);
1816 btrfs_start_workers(&fs_info
->submit_workers
, 1);
1817 btrfs_start_workers(&fs_info
->delalloc_workers
, 1);
1818 btrfs_start_workers(&fs_info
->fixup_workers
, 1);
1819 btrfs_start_workers(&fs_info
->endio_workers
, 1);
1820 btrfs_start_workers(&fs_info
->endio_meta_workers
, 1);
1821 btrfs_start_workers(&fs_info
->endio_meta_write_workers
, 1);
1822 btrfs_start_workers(&fs_info
->endio_write_workers
, 1);
1823 btrfs_start_workers(&fs_info
->enospc_workers
, 1);
1825 fs_info
->bdi
.ra_pages
*= btrfs_super_num_devices(disk_super
);
1826 fs_info
->bdi
.ra_pages
= max(fs_info
->bdi
.ra_pages
,
1827 4 * 1024 * 1024 / PAGE_CACHE_SIZE
);
1829 nodesize
= btrfs_super_nodesize(disk_super
);
1830 leafsize
= btrfs_super_leafsize(disk_super
);
1831 sectorsize
= btrfs_super_sectorsize(disk_super
);
1832 stripesize
= btrfs_super_stripesize(disk_super
);
1833 tree_root
->nodesize
= nodesize
;
1834 tree_root
->leafsize
= leafsize
;
1835 tree_root
->sectorsize
= sectorsize
;
1836 tree_root
->stripesize
= stripesize
;
1838 sb
->s_blocksize
= sectorsize
;
1839 sb
->s_blocksize_bits
= blksize_bits(sectorsize
);
1841 if (strncmp((char *)(&disk_super
->magic
), BTRFS_MAGIC
,
1842 sizeof(disk_super
->magic
))) {
1843 printk(KERN_INFO
"btrfs: valid FS not found on %s\n", sb
->s_id
);
1844 goto fail_sb_buffer
;
1847 mutex_lock(&fs_info
->chunk_mutex
);
1848 ret
= btrfs_read_sys_array(tree_root
);
1849 mutex_unlock(&fs_info
->chunk_mutex
);
1851 printk(KERN_WARNING
"btrfs: failed to read the system "
1852 "array on %s\n", sb
->s_id
);
1853 goto fail_sb_buffer
;
1856 blocksize
= btrfs_level_size(tree_root
,
1857 btrfs_super_chunk_root_level(disk_super
));
1858 generation
= btrfs_super_chunk_root_generation(disk_super
);
1860 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
1861 chunk_root
, fs_info
, BTRFS_CHUNK_TREE_OBJECTID
);
1863 chunk_root
->node
= read_tree_block(chunk_root
,
1864 btrfs_super_chunk_root(disk_super
),
1865 blocksize
, generation
);
1866 BUG_ON(!chunk_root
->node
);
1867 if (!test_bit(EXTENT_BUFFER_UPTODATE
, &chunk_root
->node
->bflags
)) {
1868 printk(KERN_WARNING
"btrfs: failed to read chunk root on %s\n",
1870 goto fail_chunk_root
;
1872 btrfs_set_root_node(&chunk_root
->root_item
, chunk_root
->node
);
1873 chunk_root
->commit_root
= btrfs_root_node(chunk_root
);
1875 read_extent_buffer(chunk_root
->node
, fs_info
->chunk_tree_uuid
,
1876 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root
->node
),
1879 mutex_lock(&fs_info
->chunk_mutex
);
1880 ret
= btrfs_read_chunk_tree(chunk_root
);
1881 mutex_unlock(&fs_info
->chunk_mutex
);
1883 printk(KERN_WARNING
"btrfs: failed to read chunk tree on %s\n",
1885 goto fail_chunk_root
;
1888 btrfs_close_extra_devices(fs_devices
);
1890 blocksize
= btrfs_level_size(tree_root
,
1891 btrfs_super_root_level(disk_super
));
1892 generation
= btrfs_super_generation(disk_super
);
1894 tree_root
->node
= read_tree_block(tree_root
,
1895 btrfs_super_root(disk_super
),
1896 blocksize
, generation
);
1897 if (!tree_root
->node
)
1898 goto fail_chunk_root
;
1899 if (!test_bit(EXTENT_BUFFER_UPTODATE
, &tree_root
->node
->bflags
)) {
1900 printk(KERN_WARNING
"btrfs: failed to read tree root on %s\n",
1902 goto fail_tree_root
;
1904 btrfs_set_root_node(&tree_root
->root_item
, tree_root
->node
);
1905 tree_root
->commit_root
= btrfs_root_node(tree_root
);
1907 ret
= find_and_setup_root(tree_root
, fs_info
,
1908 BTRFS_EXTENT_TREE_OBJECTID
, extent_root
);
1910 goto fail_tree_root
;
1911 extent_root
->track_dirty
= 1;
1913 ret
= find_and_setup_root(tree_root
, fs_info
,
1914 BTRFS_DEV_TREE_OBJECTID
, dev_root
);
1916 goto fail_extent_root
;
1917 dev_root
->track_dirty
= 1;
1919 ret
= find_and_setup_root(tree_root
, fs_info
,
1920 BTRFS_CSUM_TREE_OBJECTID
, csum_root
);
1924 csum_root
->track_dirty
= 1;
1926 btrfs_read_block_groups(extent_root
);
1928 fs_info
->generation
= generation
;
1929 fs_info
->last_trans_committed
= generation
;
1930 fs_info
->data_alloc_profile
= (u64
)-1;
1931 fs_info
->metadata_alloc_profile
= (u64
)-1;
1932 fs_info
->system_alloc_profile
= fs_info
->metadata_alloc_profile
;
1933 fs_info
->cleaner_kthread
= kthread_run(cleaner_kthread
, tree_root
,
1935 if (IS_ERR(fs_info
->cleaner_kthread
))
1936 goto fail_csum_root
;
1938 fs_info
->transaction_kthread
= kthread_run(transaction_kthread
,
1940 "btrfs-transaction");
1941 if (IS_ERR(fs_info
->transaction_kthread
))
1944 if (!btrfs_test_opt(tree_root
, SSD
) &&
1945 !btrfs_test_opt(tree_root
, NOSSD
) &&
1946 !fs_info
->fs_devices
->rotating
) {
1947 printk(KERN_INFO
"Btrfs detected SSD devices, enabling SSD "
1949 btrfs_set_opt(fs_info
->mount_opt
, SSD
);
1952 if (btrfs_super_log_root(disk_super
) != 0) {
1953 u64 bytenr
= btrfs_super_log_root(disk_super
);
1955 if (fs_devices
->rw_devices
== 0) {
1956 printk(KERN_WARNING
"Btrfs log replay required "
1959 goto fail_trans_kthread
;
1962 btrfs_level_size(tree_root
,
1963 btrfs_super_log_root_level(disk_super
));
1965 log_tree_root
= kzalloc(sizeof(struct btrfs_root
),
1968 __setup_root(nodesize
, leafsize
, sectorsize
, stripesize
,
1969 log_tree_root
, fs_info
, BTRFS_TREE_LOG_OBJECTID
);
1971 log_tree_root
->node
= read_tree_block(tree_root
, bytenr
,
1974 ret
= btrfs_recover_log_trees(log_tree_root
);
1977 if (sb
->s_flags
& MS_RDONLY
) {
1978 ret
= btrfs_commit_super(tree_root
);
1983 ret
= btrfs_find_orphan_roots(tree_root
);
1986 if (!(sb
->s_flags
& MS_RDONLY
)) {
1987 ret
= btrfs_recover_relocation(tree_root
);
1990 "btrfs: failed to recover relocation\n");
1992 goto fail_trans_kthread
;
1996 location
.objectid
= BTRFS_FS_TREE_OBJECTID
;
1997 location
.type
= BTRFS_ROOT_ITEM_KEY
;
1998 location
.offset
= (u64
)-1;
2000 fs_info
->fs_root
= btrfs_read_fs_root_no_name(fs_info
, &location
);
2001 if (!fs_info
->fs_root
)
2002 goto fail_trans_kthread
;
2004 if (!(sb
->s_flags
& MS_RDONLY
)) {
2005 down_read(&fs_info
->cleanup_work_sem
);
2006 btrfs_orphan_cleanup(fs_info
->fs_root
);
2007 up_read(&fs_info
->cleanup_work_sem
);
2013 kthread_stop(fs_info
->transaction_kthread
);
2015 kthread_stop(fs_info
->cleaner_kthread
);
2018 * make sure we're done with the btree inode before we stop our
2021 filemap_write_and_wait(fs_info
->btree_inode
->i_mapping
);
2022 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
2025 free_extent_buffer(csum_root
->node
);
2026 free_extent_buffer(csum_root
->commit_root
);
2028 free_extent_buffer(dev_root
->node
);
2029 free_extent_buffer(dev_root
->commit_root
);
2031 free_extent_buffer(extent_root
->node
);
2032 free_extent_buffer(extent_root
->commit_root
);
2034 free_extent_buffer(tree_root
->node
);
2035 free_extent_buffer(tree_root
->commit_root
);
2037 free_extent_buffer(chunk_root
->node
);
2038 free_extent_buffer(chunk_root
->commit_root
);
2040 btrfs_stop_workers(&fs_info
->generic_worker
);
2041 btrfs_stop_workers(&fs_info
->fixup_workers
);
2042 btrfs_stop_workers(&fs_info
->delalloc_workers
);
2043 btrfs_stop_workers(&fs_info
->workers
);
2044 btrfs_stop_workers(&fs_info
->endio_workers
);
2045 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
2046 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
2047 btrfs_stop_workers(&fs_info
->endio_write_workers
);
2048 btrfs_stop_workers(&fs_info
->submit_workers
);
2049 btrfs_stop_workers(&fs_info
->enospc_workers
);
2051 invalidate_inode_pages2(fs_info
->btree_inode
->i_mapping
);
2052 iput(fs_info
->btree_inode
);
2054 btrfs_close_devices(fs_info
->fs_devices
);
2055 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
2057 bdi_destroy(&fs_info
->bdi
);
2059 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
2067 return ERR_PTR(err
);
2070 static void btrfs_end_buffer_write_sync(struct buffer_head
*bh
, int uptodate
)
2072 char b
[BDEVNAME_SIZE
];
2075 set_buffer_uptodate(bh
);
2077 if (!buffer_eopnotsupp(bh
) && printk_ratelimit()) {
2078 printk(KERN_WARNING
"lost page write due to "
2079 "I/O error on %s\n",
2080 bdevname(bh
->b_bdev
, b
));
2082 /* note, we dont' set_buffer_write_io_error because we have
2083 * our own ways of dealing with the IO errors
2085 clear_buffer_uptodate(bh
);
2091 struct buffer_head
*btrfs_read_dev_super(struct block_device
*bdev
)
2093 struct buffer_head
*bh
;
2094 struct buffer_head
*latest
= NULL
;
2095 struct btrfs_super_block
*super
;
2100 /* we would like to check all the supers, but that would make
2101 * a btrfs mount succeed after a mkfs from a different FS.
2102 * So, we need to add a special mount option to scan for
2103 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2105 for (i
= 0; i
< 1; i
++) {
2106 bytenr
= btrfs_sb_offset(i
);
2107 if (bytenr
+ 4096 >= i_size_read(bdev
->bd_inode
))
2109 bh
= __bread(bdev
, bytenr
/ 4096, 4096);
2113 super
= (struct btrfs_super_block
*)bh
->b_data
;
2114 if (btrfs_super_bytenr(super
) != bytenr
||
2115 strncmp((char *)(&super
->magic
), BTRFS_MAGIC
,
2116 sizeof(super
->magic
))) {
2121 if (!latest
|| btrfs_super_generation(super
) > transid
) {
2124 transid
= btrfs_super_generation(super
);
2133 * this should be called twice, once with wait == 0 and
2134 * once with wait == 1. When wait == 0 is done, all the buffer heads
2135 * we write are pinned.
2137 * They are released when wait == 1 is done.
2138 * max_mirrors must be the same for both runs, and it indicates how
2139 * many supers on this one device should be written.
2141 * max_mirrors == 0 means to write them all.
2143 static int write_dev_supers(struct btrfs_device
*device
,
2144 struct btrfs_super_block
*sb
,
2145 int do_barriers
, int wait
, int max_mirrors
)
2147 struct buffer_head
*bh
;
2153 int last_barrier
= 0;
2155 if (max_mirrors
== 0)
2156 max_mirrors
= BTRFS_SUPER_MIRROR_MAX
;
2158 /* make sure only the last submit_bh does a barrier */
2160 for (i
= 0; i
< max_mirrors
; i
++) {
2161 bytenr
= btrfs_sb_offset(i
);
2162 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
2163 device
->total_bytes
)
2169 for (i
= 0; i
< max_mirrors
; i
++) {
2170 bytenr
= btrfs_sb_offset(i
);
2171 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>= device
->total_bytes
)
2175 bh
= __find_get_block(device
->bdev
, bytenr
/ 4096,
2176 BTRFS_SUPER_INFO_SIZE
);
2179 if (!buffer_uptodate(bh
))
2182 /* drop our reference */
2185 /* drop the reference from the wait == 0 run */
2189 btrfs_set_super_bytenr(sb
, bytenr
);
2192 crc
= btrfs_csum_data(NULL
, (char *)sb
+
2193 BTRFS_CSUM_SIZE
, crc
,
2194 BTRFS_SUPER_INFO_SIZE
-
2196 btrfs_csum_final(crc
, sb
->csum
);
2199 * one reference for us, and we leave it for the
2202 bh
= __getblk(device
->bdev
, bytenr
/ 4096,
2203 BTRFS_SUPER_INFO_SIZE
);
2204 memcpy(bh
->b_data
, sb
, BTRFS_SUPER_INFO_SIZE
);
2206 /* one reference for submit_bh */
2209 set_buffer_uptodate(bh
);
2211 bh
->b_end_io
= btrfs_end_buffer_write_sync
;
2214 if (i
== last_barrier
&& do_barriers
&& device
->barriers
) {
2215 ret
= submit_bh(WRITE_BARRIER
, bh
);
2216 if (ret
== -EOPNOTSUPP
) {
2217 printk("btrfs: disabling barriers on dev %s\n",
2219 set_buffer_uptodate(bh
);
2220 device
->barriers
= 0;
2221 /* one reference for submit_bh */
2224 ret
= submit_bh(WRITE_SYNC
, bh
);
2227 ret
= submit_bh(WRITE_SYNC
, bh
);
2233 return errors
< i
? 0 : -1;
2236 int write_all_supers(struct btrfs_root
*root
, int max_mirrors
)
2238 struct list_head
*head
;
2239 struct btrfs_device
*dev
;
2240 struct btrfs_super_block
*sb
;
2241 struct btrfs_dev_item
*dev_item
;
2245 int total_errors
= 0;
2248 max_errors
= btrfs_super_num_devices(&root
->fs_info
->super_copy
) - 1;
2249 do_barriers
= !btrfs_test_opt(root
, NOBARRIER
);
2251 sb
= &root
->fs_info
->super_for_commit
;
2252 dev_item
= &sb
->dev_item
;
2254 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2255 head
= &root
->fs_info
->fs_devices
->devices
;
2256 list_for_each_entry(dev
, head
, dev_list
) {
2261 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
2264 btrfs_set_stack_device_generation(dev_item
, 0);
2265 btrfs_set_stack_device_type(dev_item
, dev
->type
);
2266 btrfs_set_stack_device_id(dev_item
, dev
->devid
);
2267 btrfs_set_stack_device_total_bytes(dev_item
, dev
->total_bytes
);
2268 btrfs_set_stack_device_bytes_used(dev_item
, dev
->bytes_used
);
2269 btrfs_set_stack_device_io_align(dev_item
, dev
->io_align
);
2270 btrfs_set_stack_device_io_width(dev_item
, dev
->io_width
);
2271 btrfs_set_stack_device_sector_size(dev_item
, dev
->sector_size
);
2272 memcpy(dev_item
->uuid
, dev
->uuid
, BTRFS_UUID_SIZE
);
2273 memcpy(dev_item
->fsid
, dev
->fs_devices
->fsid
, BTRFS_UUID_SIZE
);
2275 flags
= btrfs_super_flags(sb
);
2276 btrfs_set_super_flags(sb
, flags
| BTRFS_HEADER_FLAG_WRITTEN
);
2278 ret
= write_dev_supers(dev
, sb
, do_barriers
, 0, max_mirrors
);
2282 if (total_errors
> max_errors
) {
2283 printk(KERN_ERR
"btrfs: %d errors while writing supers\n",
2289 list_for_each_entry(dev
, head
, dev_list
) {
2292 if (!dev
->in_fs_metadata
|| !dev
->writeable
)
2295 ret
= write_dev_supers(dev
, sb
, do_barriers
, 1, max_mirrors
);
2299 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2300 if (total_errors
> max_errors
) {
2301 printk(KERN_ERR
"btrfs: %d errors while writing supers\n",
2308 int write_ctree_super(struct btrfs_trans_handle
*trans
,
2309 struct btrfs_root
*root
, int max_mirrors
)
2313 ret
= write_all_supers(root
, max_mirrors
);
2317 int btrfs_free_fs_root(struct btrfs_fs_info
*fs_info
, struct btrfs_root
*root
)
2319 spin_lock(&fs_info
->fs_roots_radix_lock
);
2320 radix_tree_delete(&fs_info
->fs_roots_radix
,
2321 (unsigned long)root
->root_key
.objectid
);
2322 spin_unlock(&fs_info
->fs_roots_radix_lock
);
2324 if (btrfs_root_refs(&root
->root_item
) == 0)
2325 synchronize_srcu(&fs_info
->subvol_srcu
);
2331 static void free_fs_root(struct btrfs_root
*root
)
2333 WARN_ON(!RB_EMPTY_ROOT(&root
->inode_tree
));
2334 if (root
->anon_super
.s_dev
) {
2335 down_write(&root
->anon_super
.s_umount
);
2336 kill_anon_super(&root
->anon_super
);
2338 free_extent_buffer(root
->node
);
2339 free_extent_buffer(root
->commit_root
);
2344 static int del_fs_roots(struct btrfs_fs_info
*fs_info
)
2347 struct btrfs_root
*gang
[8];
2350 while (!list_empty(&fs_info
->dead_roots
)) {
2351 gang
[0] = list_entry(fs_info
->dead_roots
.next
,
2352 struct btrfs_root
, root_list
);
2353 list_del(&gang
[0]->root_list
);
2355 if (gang
[0]->in_radix
) {
2356 btrfs_free_fs_root(fs_info
, gang
[0]);
2358 free_extent_buffer(gang
[0]->node
);
2359 free_extent_buffer(gang
[0]->commit_root
);
2365 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2370 for (i
= 0; i
< ret
; i
++)
2371 btrfs_free_fs_root(fs_info
, gang
[i
]);
2376 int btrfs_cleanup_fs_roots(struct btrfs_fs_info
*fs_info
)
2378 u64 root_objectid
= 0;
2379 struct btrfs_root
*gang
[8];
2384 ret
= radix_tree_gang_lookup(&fs_info
->fs_roots_radix
,
2385 (void **)gang
, root_objectid
,
2390 root_objectid
= gang
[ret
- 1]->root_key
.objectid
+ 1;
2391 for (i
= 0; i
< ret
; i
++) {
2392 root_objectid
= gang
[i
]->root_key
.objectid
;
2393 btrfs_orphan_cleanup(gang
[i
]);
2400 int btrfs_commit_super(struct btrfs_root
*root
)
2402 struct btrfs_trans_handle
*trans
;
2405 mutex_lock(&root
->fs_info
->cleaner_mutex
);
2406 btrfs_run_delayed_iputs(root
);
2407 btrfs_clean_old_snapshots(root
);
2408 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
2410 /* wait until ongoing cleanup work done */
2411 down_write(&root
->fs_info
->cleanup_work_sem
);
2412 up_write(&root
->fs_info
->cleanup_work_sem
);
2414 trans
= btrfs_start_transaction(root
, 1);
2415 ret
= btrfs_commit_transaction(trans
, root
);
2417 /* run commit again to drop the original snapshot */
2418 trans
= btrfs_start_transaction(root
, 1);
2419 btrfs_commit_transaction(trans
, root
);
2420 ret
= btrfs_write_and_wait_transaction(NULL
, root
);
2423 ret
= write_ctree_super(NULL
, root
, 0);
2427 int close_ctree(struct btrfs_root
*root
)
2429 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2432 fs_info
->closing
= 1;
2435 kthread_stop(root
->fs_info
->transaction_kthread
);
2436 kthread_stop(root
->fs_info
->cleaner_kthread
);
2438 if (!(fs_info
->sb
->s_flags
& MS_RDONLY
)) {
2439 ret
= btrfs_commit_super(root
);
2441 printk(KERN_ERR
"btrfs: commit super ret %d\n", ret
);
2444 fs_info
->closing
= 2;
2447 if (fs_info
->delalloc_bytes
) {
2448 printk(KERN_INFO
"btrfs: at unmount delalloc count %llu\n",
2449 (unsigned long long)fs_info
->delalloc_bytes
);
2451 if (fs_info
->total_ref_cache_size
) {
2452 printk(KERN_INFO
"btrfs: at umount reference cache size %llu\n",
2453 (unsigned long long)fs_info
->total_ref_cache_size
);
2456 free_extent_buffer(fs_info
->extent_root
->node
);
2457 free_extent_buffer(fs_info
->extent_root
->commit_root
);
2458 free_extent_buffer(fs_info
->tree_root
->node
);
2459 free_extent_buffer(fs_info
->tree_root
->commit_root
);
2460 free_extent_buffer(root
->fs_info
->chunk_root
->node
);
2461 free_extent_buffer(root
->fs_info
->chunk_root
->commit_root
);
2462 free_extent_buffer(root
->fs_info
->dev_root
->node
);
2463 free_extent_buffer(root
->fs_info
->dev_root
->commit_root
);
2464 free_extent_buffer(root
->fs_info
->csum_root
->node
);
2465 free_extent_buffer(root
->fs_info
->csum_root
->commit_root
);
2467 btrfs_free_block_groups(root
->fs_info
);
2469 del_fs_roots(fs_info
);
2471 iput(fs_info
->btree_inode
);
2473 btrfs_stop_workers(&fs_info
->generic_worker
);
2474 btrfs_stop_workers(&fs_info
->fixup_workers
);
2475 btrfs_stop_workers(&fs_info
->delalloc_workers
);
2476 btrfs_stop_workers(&fs_info
->workers
);
2477 btrfs_stop_workers(&fs_info
->endio_workers
);
2478 btrfs_stop_workers(&fs_info
->endio_meta_workers
);
2479 btrfs_stop_workers(&fs_info
->endio_meta_write_workers
);
2480 btrfs_stop_workers(&fs_info
->endio_write_workers
);
2481 btrfs_stop_workers(&fs_info
->submit_workers
);
2482 btrfs_stop_workers(&fs_info
->enospc_workers
);
2484 btrfs_close_devices(fs_info
->fs_devices
);
2485 btrfs_mapping_tree_free(&fs_info
->mapping_tree
);
2487 bdi_destroy(&fs_info
->bdi
);
2488 cleanup_srcu_struct(&fs_info
->subvol_srcu
);
2490 kfree(fs_info
->extent_root
);
2491 kfree(fs_info
->tree_root
);
2492 kfree(fs_info
->chunk_root
);
2493 kfree(fs_info
->dev_root
);
2494 kfree(fs_info
->csum_root
);
2498 int btrfs_buffer_uptodate(struct extent_buffer
*buf
, u64 parent_transid
)
2501 struct inode
*btree_inode
= buf
->first_page
->mapping
->host
;
2503 ret
= extent_buffer_uptodate(&BTRFS_I(btree_inode
)->io_tree
, buf
,
2508 ret
= verify_parent_transid(&BTRFS_I(btree_inode
)->io_tree
, buf
,
2513 int btrfs_set_buffer_uptodate(struct extent_buffer
*buf
)
2515 struct inode
*btree_inode
= buf
->first_page
->mapping
->host
;
2516 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode
)->io_tree
,
2520 void btrfs_mark_buffer_dirty(struct extent_buffer
*buf
)
2522 struct btrfs_root
*root
= BTRFS_I(buf
->first_page
->mapping
->host
)->root
;
2523 u64 transid
= btrfs_header_generation(buf
);
2524 struct inode
*btree_inode
= root
->fs_info
->btree_inode
;
2527 btrfs_assert_tree_locked(buf
);
2528 if (transid
!= root
->fs_info
->generation
) {
2529 printk(KERN_CRIT
"btrfs transid mismatch buffer %llu, "
2530 "found %llu running %llu\n",
2531 (unsigned long long)buf
->start
,
2532 (unsigned long long)transid
,
2533 (unsigned long long)root
->fs_info
->generation
);
2536 was_dirty
= set_extent_buffer_dirty(&BTRFS_I(btree_inode
)->io_tree
,
2539 spin_lock(&root
->fs_info
->delalloc_lock
);
2540 root
->fs_info
->dirty_metadata_bytes
+= buf
->len
;
2541 spin_unlock(&root
->fs_info
->delalloc_lock
);
2545 void btrfs_btree_balance_dirty(struct btrfs_root
*root
, unsigned long nr
)
2548 * looks as though older kernels can get into trouble with
2549 * this code, they end up stuck in balance_dirty_pages forever
2552 unsigned long thresh
= 32 * 1024 * 1024;
2554 if (current
->flags
& PF_MEMALLOC
)
2557 num_dirty
= root
->fs_info
->dirty_metadata_bytes
;
2559 if (num_dirty
> thresh
) {
2560 balance_dirty_pages_ratelimited_nr(
2561 root
->fs_info
->btree_inode
->i_mapping
, 1);
2566 int btrfs_read_buffer(struct extent_buffer
*buf
, u64 parent_transid
)
2568 struct btrfs_root
*root
= BTRFS_I(buf
->first_page
->mapping
->host
)->root
;
2570 ret
= btree_read_extent_buffer_pages(root
, buf
, 0, parent_transid
);
2572 set_bit(EXTENT_BUFFER_UPTODATE
, &buf
->bflags
);
2576 int btree_lock_page_hook(struct page
*page
)
2578 struct inode
*inode
= page
->mapping
->host
;
2579 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2580 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2581 struct extent_buffer
*eb
;
2583 u64 bytenr
= page_offset(page
);
2585 if (page
->private == EXTENT_PAGE_PRIVATE
)
2588 len
= page
->private >> 2;
2589 eb
= find_extent_buffer(io_tree
, bytenr
, len
, GFP_NOFS
);
2593 btrfs_tree_lock(eb
);
2594 btrfs_set_header_flag(eb
, BTRFS_HEADER_FLAG_WRITTEN
);
2596 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY
, &eb
->bflags
)) {
2597 spin_lock(&root
->fs_info
->delalloc_lock
);
2598 if (root
->fs_info
->dirty_metadata_bytes
>= eb
->len
)
2599 root
->fs_info
->dirty_metadata_bytes
-= eb
->len
;
2602 spin_unlock(&root
->fs_info
->delalloc_lock
);
2605 btrfs_tree_unlock(eb
);
2606 free_extent_buffer(eb
);
2612 static struct extent_io_ops btree_extent_io_ops
= {
2613 .write_cache_pages_lock_hook
= btree_lock_page_hook
,
2614 .readpage_end_io_hook
= btree_readpage_end_io_hook
,
2615 .submit_bio_hook
= btree_submit_bio_hook
,
2616 /* note we're sharing with inode.c for the merge bio hook */
2617 .merge_bio_hook
= btrfs_merge_bio_hook
,