2 * Copyright (C) 2011 STRATO. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/blkdev.h>
20 #include <linux/ratelimit.h>
24 #include "ordered-data.h"
25 #include "transaction.h"
27 #include "extent_io.h"
28 #include "check-integrity.h"
31 * This is only the first step towards a full-features scrub. It reads all
32 * extent and super block and verifies the checksums. In case a bad checksum
33 * is found or the extent cannot be read, good data will be written back if
36 * Future enhancements:
37 * - In case an unrepairable extent is encountered, track which files are
38 * affected and report them
39 * - track and record media errors, throw out bad devices
40 * - add a mode to also read unallocated space
46 #define SCRUB_PAGES_PER_BIO 16 /* 64k per bio */
47 #define SCRUB_BIOS_PER_DEV 16 /* 1 MB per device in flight */
48 #define SCRUB_MAX_PAGES_PER_BLOCK 16 /* 64k per node/leaf/sector */
51 struct scrub_block
*sblock
;
53 struct block_device
*bdev
;
54 u64 flags
; /* extent flags */
59 unsigned int mirror_num
:8;
60 unsigned int have_csum
:1;
61 unsigned int io_error
:1;
63 u8 csum
[BTRFS_CSUM_SIZE
];
68 struct scrub_dev
*sdev
;
73 struct scrub_page
*pagev
[SCRUB_PAGES_PER_BIO
];
76 struct btrfs_work work
;
80 struct scrub_page pagev
[SCRUB_MAX_PAGES_PER_BLOCK
];
82 atomic_t outstanding_pages
;
83 atomic_t ref_count
; /* free mem on transition to zero */
84 struct scrub_dev
*sdev
;
86 unsigned int header_error
:1;
87 unsigned int checksum_error
:1;
88 unsigned int no_io_error_seen
:1;
93 struct scrub_bio
*bios
[SCRUB_BIOS_PER_DEV
];
94 struct btrfs_device
*dev
;
100 wait_queue_head_t list_wait
;
102 struct list_head csum_list
;
105 int pages_per_bio
; /* <= SCRUB_PAGES_PER_BIO */
112 struct btrfs_scrub_progress stat
;
113 spinlock_t stat_lock
;
116 struct scrub_fixup_nodatasum
{
117 struct scrub_dev
*sdev
;
119 struct btrfs_root
*root
;
120 struct btrfs_work work
;
124 struct scrub_warning
{
125 struct btrfs_path
*path
;
126 u64 extent_item_size
;
132 struct btrfs_device
*dev
;
138 static int scrub_handle_errored_block(struct scrub_block
*sblock_to_check
);
139 static int scrub_setup_recheck_block(struct scrub_dev
*sdev
,
140 struct btrfs_mapping_tree
*map_tree
,
141 u64 length
, u64 logical
,
142 struct scrub_block
*sblock
);
143 static int scrub_recheck_block(struct btrfs_fs_info
*fs_info
,
144 struct scrub_block
*sblock
, int is_metadata
,
145 int have_csum
, u8
*csum
, u64 generation
,
147 static void scrub_recheck_block_checksum(struct btrfs_fs_info
*fs_info
,
148 struct scrub_block
*sblock
,
149 int is_metadata
, int have_csum
,
150 const u8
*csum
, u64 generation
,
152 static void scrub_complete_bio_end_io(struct bio
*bio
, int err
);
153 static int scrub_repair_block_from_good_copy(struct scrub_block
*sblock_bad
,
154 struct scrub_block
*sblock_good
,
156 static int scrub_repair_page_from_good_copy(struct scrub_block
*sblock_bad
,
157 struct scrub_block
*sblock_good
,
158 int page_num
, int force_write
);
159 static int scrub_checksum_data(struct scrub_block
*sblock
);
160 static int scrub_checksum_tree_block(struct scrub_block
*sblock
);
161 static int scrub_checksum_super(struct scrub_block
*sblock
);
162 static void scrub_block_get(struct scrub_block
*sblock
);
163 static void scrub_block_put(struct scrub_block
*sblock
);
164 static int scrub_add_page_to_bio(struct scrub_dev
*sdev
,
165 struct scrub_page
*spage
);
166 static int scrub_pages(struct scrub_dev
*sdev
, u64 logical
, u64 len
,
167 u64 physical
, u64 flags
, u64 gen
, int mirror_num
,
168 u8
*csum
, int force
);
169 static void scrub_bio_end_io(struct bio
*bio
, int err
);
170 static void scrub_bio_end_io_worker(struct btrfs_work
*work
);
171 static void scrub_block_complete(struct scrub_block
*sblock
);
174 static void scrub_free_csums(struct scrub_dev
*sdev
)
176 while (!list_empty(&sdev
->csum_list
)) {
177 struct btrfs_ordered_sum
*sum
;
178 sum
= list_first_entry(&sdev
->csum_list
,
179 struct btrfs_ordered_sum
, list
);
180 list_del(&sum
->list
);
185 static noinline_for_stack
void scrub_free_dev(struct scrub_dev
*sdev
)
192 /* this can happen when scrub is cancelled */
193 if (sdev
->curr
!= -1) {
194 struct scrub_bio
*sbio
= sdev
->bios
[sdev
->curr
];
196 for (i
= 0; i
< sbio
->page_count
; i
++) {
197 BUG_ON(!sbio
->pagev
[i
]);
198 BUG_ON(!sbio
->pagev
[i
]->page
);
199 scrub_block_put(sbio
->pagev
[i
]->sblock
);
204 for (i
= 0; i
< SCRUB_BIOS_PER_DEV
; ++i
) {
205 struct scrub_bio
*sbio
= sdev
->bios
[i
];
212 scrub_free_csums(sdev
);
216 static noinline_for_stack
217 struct scrub_dev
*scrub_setup_dev(struct btrfs_device
*dev
)
219 struct scrub_dev
*sdev
;
221 struct btrfs_fs_info
*fs_info
= dev
->dev_root
->fs_info
;
224 pages_per_bio
= min_t(int, SCRUB_PAGES_PER_BIO
,
225 bio_get_nr_vecs(dev
->bdev
));
226 sdev
= kzalloc(sizeof(*sdev
), GFP_NOFS
);
230 sdev
->pages_per_bio
= pages_per_bio
;
232 for (i
= 0; i
< SCRUB_BIOS_PER_DEV
; ++i
) {
233 struct scrub_bio
*sbio
;
235 sbio
= kzalloc(sizeof(*sbio
), GFP_NOFS
);
238 sdev
->bios
[i
] = sbio
;
242 sbio
->page_count
= 0;
243 sbio
->work
.func
= scrub_bio_end_io_worker
;
245 if (i
!= SCRUB_BIOS_PER_DEV
-1)
246 sdev
->bios
[i
]->next_free
= i
+ 1;
248 sdev
->bios
[i
]->next_free
= -1;
250 sdev
->first_free
= 0;
251 sdev
->nodesize
= dev
->dev_root
->nodesize
;
252 sdev
->leafsize
= dev
->dev_root
->leafsize
;
253 sdev
->sectorsize
= dev
->dev_root
->sectorsize
;
254 atomic_set(&sdev
->in_flight
, 0);
255 atomic_set(&sdev
->fixup_cnt
, 0);
256 atomic_set(&sdev
->cancel_req
, 0);
257 sdev
->csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
258 INIT_LIST_HEAD(&sdev
->csum_list
);
260 spin_lock_init(&sdev
->list_lock
);
261 spin_lock_init(&sdev
->stat_lock
);
262 init_waitqueue_head(&sdev
->list_wait
);
266 scrub_free_dev(sdev
);
267 return ERR_PTR(-ENOMEM
);
270 static int scrub_print_warning_inode(u64 inum
, u64 offset
, u64 root
, void *ctx
)
276 struct extent_buffer
*eb
;
277 struct btrfs_inode_item
*inode_item
;
278 struct scrub_warning
*swarn
= ctx
;
279 struct btrfs_fs_info
*fs_info
= swarn
->dev
->dev_root
->fs_info
;
280 struct inode_fs_paths
*ipath
= NULL
;
281 struct btrfs_root
*local_root
;
282 struct btrfs_key root_key
;
284 root_key
.objectid
= root
;
285 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
286 root_key
.offset
= (u64
)-1;
287 local_root
= btrfs_read_fs_root_no_name(fs_info
, &root_key
);
288 if (IS_ERR(local_root
)) {
289 ret
= PTR_ERR(local_root
);
293 ret
= inode_item_info(inum
, 0, local_root
, swarn
->path
);
295 btrfs_release_path(swarn
->path
);
299 eb
= swarn
->path
->nodes
[0];
300 inode_item
= btrfs_item_ptr(eb
, swarn
->path
->slots
[0],
301 struct btrfs_inode_item
);
302 isize
= btrfs_inode_size(eb
, inode_item
);
303 nlink
= btrfs_inode_nlink(eb
, inode_item
);
304 btrfs_release_path(swarn
->path
);
306 ipath
= init_ipath(4096, local_root
, swarn
->path
);
308 ret
= PTR_ERR(ipath
);
312 ret
= paths_from_inode(inum
, ipath
);
318 * we deliberately ignore the bit ipath might have been too small to
319 * hold all of the paths here
321 for (i
= 0; i
< ipath
->fspath
->elem_cnt
; ++i
)
322 printk(KERN_WARNING
"btrfs: %s at logical %llu on dev "
323 "%s, sector %llu, root %llu, inode %llu, offset %llu, "
324 "length %llu, links %u (path: %s)\n", swarn
->errstr
,
325 swarn
->logical
, swarn
->dev
->name
,
326 (unsigned long long)swarn
->sector
, root
, inum
, offset
,
327 min(isize
- offset
, (u64
)PAGE_SIZE
), nlink
,
328 (char *)(unsigned long)ipath
->fspath
->val
[i
]);
334 printk(KERN_WARNING
"btrfs: %s at logical %llu on dev "
335 "%s, sector %llu, root %llu, inode %llu, offset %llu: path "
336 "resolving failed with ret=%d\n", swarn
->errstr
,
337 swarn
->logical
, swarn
->dev
->name
,
338 (unsigned long long)swarn
->sector
, root
, inum
, offset
, ret
);
344 static void scrub_print_warning(const char *errstr
, struct scrub_block
*sblock
)
346 struct btrfs_device
*dev
= sblock
->sdev
->dev
;
347 struct btrfs_fs_info
*fs_info
= dev
->dev_root
->fs_info
;
348 struct btrfs_path
*path
;
349 struct btrfs_key found_key
;
350 struct extent_buffer
*eb
;
351 struct btrfs_extent_item
*ei
;
352 struct scrub_warning swarn
;
357 unsigned long ptr
= 0;
358 const int bufsize
= 4096;
361 path
= btrfs_alloc_path();
363 swarn
.scratch_buf
= kmalloc(bufsize
, GFP_NOFS
);
364 swarn
.msg_buf
= kmalloc(bufsize
, GFP_NOFS
);
365 BUG_ON(sblock
->page_count
< 1);
366 swarn
.sector
= (sblock
->pagev
[0].physical
) >> 9;
367 swarn
.logical
= sblock
->pagev
[0].logical
;
368 swarn
.errstr
= errstr
;
370 swarn
.msg_bufsize
= bufsize
;
371 swarn
.scratch_bufsize
= bufsize
;
373 if (!path
|| !swarn
.scratch_buf
|| !swarn
.msg_buf
)
376 ret
= extent_from_logical(fs_info
, swarn
.logical
, path
, &found_key
);
380 extent_item_pos
= swarn
.logical
- found_key
.objectid
;
381 swarn
.extent_item_size
= found_key
.offset
;
384 ei
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_extent_item
);
385 item_size
= btrfs_item_size_nr(eb
, path
->slots
[0]);
386 btrfs_release_path(path
);
388 if (ret
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
390 ret
= tree_backref_for_extent(&ptr
, eb
, ei
, item_size
,
391 &ref_root
, &ref_level
);
393 "btrfs: %s at logical %llu on dev %s, "
394 "sector %llu: metadata %s (level %d) in tree "
395 "%llu\n", errstr
, swarn
.logical
, dev
->name
,
396 (unsigned long long)swarn
.sector
,
397 ref_level
? "node" : "leaf",
398 ret
< 0 ? -1 : ref_level
,
399 ret
< 0 ? -1 : ref_root
);
403 iterate_extent_inodes(fs_info
, found_key
.objectid
,
405 scrub_print_warning_inode
, &swarn
);
409 btrfs_free_path(path
);
410 kfree(swarn
.scratch_buf
);
411 kfree(swarn
.msg_buf
);
414 static int scrub_fixup_readpage(u64 inum
, u64 offset
, u64 root
, void *ctx
)
416 struct page
*page
= NULL
;
418 struct scrub_fixup_nodatasum
*fixup
= ctx
;
421 struct btrfs_key key
;
422 struct inode
*inode
= NULL
;
423 u64 end
= offset
+ PAGE_SIZE
- 1;
424 struct btrfs_root
*local_root
;
427 key
.type
= BTRFS_ROOT_ITEM_KEY
;
428 key
.offset
= (u64
)-1;
429 local_root
= btrfs_read_fs_root_no_name(fixup
->root
->fs_info
, &key
);
430 if (IS_ERR(local_root
))
431 return PTR_ERR(local_root
);
433 key
.type
= BTRFS_INODE_ITEM_KEY
;
436 inode
= btrfs_iget(fixup
->root
->fs_info
->sb
, &key
, local_root
, NULL
);
438 return PTR_ERR(inode
);
440 index
= offset
>> PAGE_CACHE_SHIFT
;
442 page
= find_or_create_page(inode
->i_mapping
, index
, GFP_NOFS
);
448 if (PageUptodate(page
)) {
449 struct btrfs_mapping_tree
*map_tree
;
450 if (PageDirty(page
)) {
452 * we need to write the data to the defect sector. the
453 * data that was in that sector is not in memory,
454 * because the page was modified. we must not write the
455 * modified page to that sector.
457 * TODO: what could be done here: wait for the delalloc
458 * runner to write out that page (might involve
459 * COW) and see whether the sector is still
460 * referenced afterwards.
462 * For the meantime, we'll treat this error
463 * incorrectable, although there is a chance that a
464 * later scrub will find the bad sector again and that
465 * there's no dirty page in memory, then.
470 map_tree
= &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
;
471 ret
= repair_io_failure(map_tree
, offset
, PAGE_SIZE
,
472 fixup
->logical
, page
,
478 * we need to get good data first. the general readpage path
479 * will call repair_io_failure for us, we just have to make
480 * sure we read the bad mirror.
482 ret
= set_extent_bits(&BTRFS_I(inode
)->io_tree
, offset
, end
,
483 EXTENT_DAMAGED
, GFP_NOFS
);
485 /* set_extent_bits should give proper error */
492 ret
= extent_read_full_page(&BTRFS_I(inode
)->io_tree
, page
,
495 wait_on_page_locked(page
);
497 corrected
= !test_range_bit(&BTRFS_I(inode
)->io_tree
, offset
,
498 end
, EXTENT_DAMAGED
, 0, NULL
);
500 clear_extent_bits(&BTRFS_I(inode
)->io_tree
, offset
, end
,
501 EXTENT_DAMAGED
, GFP_NOFS
);
513 if (ret
== 0 && corrected
) {
515 * we only need to call readpage for one of the inodes belonging
516 * to this extent. so make iterate_extent_inodes stop
524 static void scrub_fixup_nodatasum(struct btrfs_work
*work
)
527 struct scrub_fixup_nodatasum
*fixup
;
528 struct scrub_dev
*sdev
;
529 struct btrfs_trans_handle
*trans
= NULL
;
530 struct btrfs_fs_info
*fs_info
;
531 struct btrfs_path
*path
;
532 int uncorrectable
= 0;
534 fixup
= container_of(work
, struct scrub_fixup_nodatasum
, work
);
536 fs_info
= fixup
->root
->fs_info
;
538 path
= btrfs_alloc_path();
540 spin_lock(&sdev
->stat_lock
);
541 ++sdev
->stat
.malloc_errors
;
542 spin_unlock(&sdev
->stat_lock
);
547 trans
= btrfs_join_transaction(fixup
->root
);
554 * the idea is to trigger a regular read through the standard path. we
555 * read a page from the (failed) logical address by specifying the
556 * corresponding copynum of the failed sector. thus, that readpage is
558 * that is the point where on-the-fly error correction will kick in
559 * (once it's finished) and rewrite the failed sector if a good copy
562 ret
= iterate_inodes_from_logical(fixup
->logical
, fixup
->root
->fs_info
,
563 path
, scrub_fixup_readpage
,
571 spin_lock(&sdev
->stat_lock
);
572 ++sdev
->stat
.corrected_errors
;
573 spin_unlock(&sdev
->stat_lock
);
576 if (trans
&& !IS_ERR(trans
))
577 btrfs_end_transaction(trans
, fixup
->root
);
579 spin_lock(&sdev
->stat_lock
);
580 ++sdev
->stat
.uncorrectable_errors
;
581 spin_unlock(&sdev
->stat_lock
);
582 printk_ratelimited(KERN_ERR
583 "btrfs: unable to fixup (nodatasum) error at logical %llu on dev %s\n",
584 (unsigned long long)fixup
->logical
, sdev
->dev
->name
);
587 btrfs_free_path(path
);
590 /* see caller why we're pretending to be paused in the scrub counters */
591 mutex_lock(&fs_info
->scrub_lock
);
592 atomic_dec(&fs_info
->scrubs_running
);
593 atomic_dec(&fs_info
->scrubs_paused
);
594 mutex_unlock(&fs_info
->scrub_lock
);
595 atomic_dec(&sdev
->fixup_cnt
);
596 wake_up(&fs_info
->scrub_pause_wait
);
597 wake_up(&sdev
->list_wait
);
601 * scrub_handle_errored_block gets called when either verification of the
602 * pages failed or the bio failed to read, e.g. with EIO. In the latter
603 * case, this function handles all pages in the bio, even though only one
605 * The goal of this function is to repair the errored block by using the
606 * contents of one of the mirrors.
608 static int scrub_handle_errored_block(struct scrub_block
*sblock_to_check
)
610 struct scrub_dev
*sdev
= sblock_to_check
->sdev
;
611 struct btrfs_fs_info
*fs_info
;
615 unsigned int failed_mirror_index
;
616 unsigned int is_metadata
;
617 unsigned int have_csum
;
619 struct scrub_block
*sblocks_for_recheck
; /* holds one for each mirror */
620 struct scrub_block
*sblock_bad
;
625 static DEFINE_RATELIMIT_STATE(_rs
, DEFAULT_RATELIMIT_INTERVAL
,
626 DEFAULT_RATELIMIT_BURST
);
628 BUG_ON(sblock_to_check
->page_count
< 1);
629 fs_info
= sdev
->dev
->dev_root
->fs_info
;
630 length
= sblock_to_check
->page_count
* PAGE_SIZE
;
631 logical
= sblock_to_check
->pagev
[0].logical
;
632 generation
= sblock_to_check
->pagev
[0].generation
;
633 BUG_ON(sblock_to_check
->pagev
[0].mirror_num
< 1);
634 failed_mirror_index
= sblock_to_check
->pagev
[0].mirror_num
- 1;
635 is_metadata
= !(sblock_to_check
->pagev
[0].flags
&
636 BTRFS_EXTENT_FLAG_DATA
);
637 have_csum
= sblock_to_check
->pagev
[0].have_csum
;
638 csum
= sblock_to_check
->pagev
[0].csum
;
641 * read all mirrors one after the other. This includes to
642 * re-read the extent or metadata block that failed (that was
643 * the cause that this fixup code is called) another time,
644 * page by page this time in order to know which pages
645 * caused I/O errors and which ones are good (for all mirrors).
646 * It is the goal to handle the situation when more than one
647 * mirror contains I/O errors, but the errors do not
648 * overlap, i.e. the data can be repaired by selecting the
649 * pages from those mirrors without I/O error on the
650 * particular pages. One example (with blocks >= 2 * PAGE_SIZE)
651 * would be that mirror #1 has an I/O error on the first page,
652 * the second page is good, and mirror #2 has an I/O error on
653 * the second page, but the first page is good.
654 * Then the first page of the first mirror can be repaired by
655 * taking the first page of the second mirror, and the
656 * second page of the second mirror can be repaired by
657 * copying the contents of the 2nd page of the 1st mirror.
658 * One more note: if the pages of one mirror contain I/O
659 * errors, the checksum cannot be verified. In order to get
660 * the best data for repairing, the first attempt is to find
661 * a mirror without I/O errors and with a validated checksum.
662 * Only if this is not possible, the pages are picked from
663 * mirrors with I/O errors without considering the checksum.
664 * If the latter is the case, at the end, the checksum of the
665 * repaired area is verified in order to correctly maintain
669 sblocks_for_recheck
= kzalloc(BTRFS_MAX_MIRRORS
*
670 sizeof(*sblocks_for_recheck
),
672 if (!sblocks_for_recheck
) {
673 spin_lock(&sdev
->stat_lock
);
674 sdev
->stat
.malloc_errors
++;
675 sdev
->stat
.read_errors
++;
676 sdev
->stat
.uncorrectable_errors
++;
677 spin_unlock(&sdev
->stat_lock
);
681 /* setup the context, map the logical blocks and alloc the pages */
682 ret
= scrub_setup_recheck_block(sdev
, &fs_info
->mapping_tree
, length
,
683 logical
, sblocks_for_recheck
);
685 spin_lock(&sdev
->stat_lock
);
686 sdev
->stat
.read_errors
++;
687 sdev
->stat
.uncorrectable_errors
++;
688 spin_unlock(&sdev
->stat_lock
);
691 BUG_ON(failed_mirror_index
>= BTRFS_MAX_MIRRORS
);
692 sblock_bad
= sblocks_for_recheck
+ failed_mirror_index
;
694 /* build and submit the bios for the failed mirror, check checksums */
695 ret
= scrub_recheck_block(fs_info
, sblock_bad
, is_metadata
, have_csum
,
696 csum
, generation
, sdev
->csum_size
);
698 spin_lock(&sdev
->stat_lock
);
699 sdev
->stat
.read_errors
++;
700 sdev
->stat
.uncorrectable_errors
++;
701 spin_unlock(&sdev
->stat_lock
);
705 if (!sblock_bad
->header_error
&& !sblock_bad
->checksum_error
&&
706 sblock_bad
->no_io_error_seen
) {
708 * the error disappeared after reading page by page, or
709 * the area was part of a huge bio and other parts of the
710 * bio caused I/O errors, or the block layer merged several
711 * read requests into one and the error is caused by a
712 * different bio (usually one of the two latter cases is
715 spin_lock(&sdev
->stat_lock
);
716 sdev
->stat
.unverified_errors
++;
717 spin_unlock(&sdev
->stat_lock
);
722 if (!sblock_bad
->no_io_error_seen
) {
723 spin_lock(&sdev
->stat_lock
);
724 sdev
->stat
.read_errors
++;
725 spin_unlock(&sdev
->stat_lock
);
726 if (__ratelimit(&_rs
))
727 scrub_print_warning("i/o error", sblock_to_check
);
728 } else if (sblock_bad
->checksum_error
) {
729 spin_lock(&sdev
->stat_lock
);
730 sdev
->stat
.csum_errors
++;
731 spin_unlock(&sdev
->stat_lock
);
732 if (__ratelimit(&_rs
))
733 scrub_print_warning("checksum error", sblock_to_check
);
734 } else if (sblock_bad
->header_error
) {
735 spin_lock(&sdev
->stat_lock
);
736 sdev
->stat
.verify_errors
++;
737 spin_unlock(&sdev
->stat_lock
);
738 if (__ratelimit(&_rs
))
739 scrub_print_warning("checksum/header error",
744 goto did_not_correct_error
;
746 if (!is_metadata
&& !have_csum
) {
747 struct scrub_fixup_nodatasum
*fixup_nodatasum
;
750 * !is_metadata and !have_csum, this means that the data
751 * might not be COW'ed, that it might be modified
752 * concurrently. The general strategy to work on the
753 * commit root does not help in the case when COW is not
756 fixup_nodatasum
= kzalloc(sizeof(*fixup_nodatasum
), GFP_NOFS
);
757 if (!fixup_nodatasum
)
758 goto did_not_correct_error
;
759 fixup_nodatasum
->sdev
= sdev
;
760 fixup_nodatasum
->logical
= logical
;
761 fixup_nodatasum
->root
= fs_info
->extent_root
;
762 fixup_nodatasum
->mirror_num
= failed_mirror_index
+ 1;
764 * increment scrubs_running to prevent cancel requests from
765 * completing as long as a fixup worker is running. we must also
766 * increment scrubs_paused to prevent deadlocking on pause
767 * requests used for transactions commits (as the worker uses a
768 * transaction context). it is safe to regard the fixup worker
769 * as paused for all matters practical. effectively, we only
770 * avoid cancellation requests from completing.
772 mutex_lock(&fs_info
->scrub_lock
);
773 atomic_inc(&fs_info
->scrubs_running
);
774 atomic_inc(&fs_info
->scrubs_paused
);
775 mutex_unlock(&fs_info
->scrub_lock
);
776 atomic_inc(&sdev
->fixup_cnt
);
777 fixup_nodatasum
->work
.func
= scrub_fixup_nodatasum
;
778 btrfs_queue_worker(&fs_info
->scrub_workers
,
779 &fixup_nodatasum
->work
);
784 * now build and submit the bios for the other mirrors, check
787 for (mirror_index
= 0;
788 mirror_index
< BTRFS_MAX_MIRRORS
&&
789 sblocks_for_recheck
[mirror_index
].page_count
> 0;
791 if (mirror_index
== failed_mirror_index
)
794 /* build and submit the bios, check checksums */
795 ret
= scrub_recheck_block(fs_info
,
796 sblocks_for_recheck
+ mirror_index
,
797 is_metadata
, have_csum
, csum
,
798 generation
, sdev
->csum_size
);
800 goto did_not_correct_error
;
804 * first try to pick the mirror which is completely without I/O
805 * errors and also does not have a checksum error.
806 * If one is found, and if a checksum is present, the full block
807 * that is known to contain an error is rewritten. Afterwards
808 * the block is known to be corrected.
809 * If a mirror is found which is completely correct, and no
810 * checksum is present, only those pages are rewritten that had
811 * an I/O error in the block to be repaired, since it cannot be
812 * determined, which copy of the other pages is better (and it
813 * could happen otherwise that a correct page would be
814 * overwritten by a bad one).
816 for (mirror_index
= 0;
817 mirror_index
< BTRFS_MAX_MIRRORS
&&
818 sblocks_for_recheck
[mirror_index
].page_count
> 0;
820 struct scrub_block
*sblock_other
= sblocks_for_recheck
+
823 if (!sblock_other
->header_error
&&
824 !sblock_other
->checksum_error
&&
825 sblock_other
->no_io_error_seen
) {
826 int force_write
= is_metadata
|| have_csum
;
828 ret
= scrub_repair_block_from_good_copy(sblock_bad
,
832 goto corrected_error
;
837 * in case of I/O errors in the area that is supposed to be
838 * repaired, continue by picking good copies of those pages.
839 * Select the good pages from mirrors to rewrite bad pages from
840 * the area to fix. Afterwards verify the checksum of the block
841 * that is supposed to be repaired. This verification step is
842 * only done for the purpose of statistic counting and for the
843 * final scrub report, whether errors remain.
844 * A perfect algorithm could make use of the checksum and try
845 * all possible combinations of pages from the different mirrors
846 * until the checksum verification succeeds. For example, when
847 * the 2nd page of mirror #1 faces I/O errors, and the 2nd page
848 * of mirror #2 is readable but the final checksum test fails,
849 * then the 2nd page of mirror #3 could be tried, whether now
850 * the final checksum succeedes. But this would be a rare
851 * exception and is therefore not implemented. At least it is
852 * avoided that the good copy is overwritten.
853 * A more useful improvement would be to pick the sectors
854 * without I/O error based on sector sizes (512 bytes on legacy
855 * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one
856 * mirror could be repaired by taking 512 byte of a different
857 * mirror, even if other 512 byte sectors in the same PAGE_SIZE
858 * area are unreadable.
861 /* can only fix I/O errors from here on */
862 if (sblock_bad
->no_io_error_seen
)
863 goto did_not_correct_error
;
866 for (page_num
= 0; page_num
< sblock_bad
->page_count
; page_num
++) {
867 struct scrub_page
*page_bad
= sblock_bad
->pagev
+ page_num
;
869 if (!page_bad
->io_error
)
872 for (mirror_index
= 0;
873 mirror_index
< BTRFS_MAX_MIRRORS
&&
874 sblocks_for_recheck
[mirror_index
].page_count
> 0;
876 struct scrub_block
*sblock_other
= sblocks_for_recheck
+
878 struct scrub_page
*page_other
= sblock_other
->pagev
+
881 if (!page_other
->io_error
) {
882 ret
= scrub_repair_page_from_good_copy(
883 sblock_bad
, sblock_other
, page_num
, 0);
885 page_bad
->io_error
= 0;
886 break; /* succeeded for this page */
891 if (page_bad
->io_error
) {
892 /* did not find a mirror to copy the page from */
898 if (is_metadata
|| have_csum
) {
900 * need to verify the checksum now that all
901 * sectors on disk are repaired (the write
902 * request for data to be repaired is on its way).
903 * Just be lazy and use scrub_recheck_block()
904 * which re-reads the data before the checksum
905 * is verified, but most likely the data comes out
908 ret
= scrub_recheck_block(fs_info
, sblock_bad
,
909 is_metadata
, have_csum
, csum
,
910 generation
, sdev
->csum_size
);
911 if (!ret
&& !sblock_bad
->header_error
&&
912 !sblock_bad
->checksum_error
&&
913 sblock_bad
->no_io_error_seen
)
914 goto corrected_error
;
916 goto did_not_correct_error
;
919 spin_lock(&sdev
->stat_lock
);
920 sdev
->stat
.corrected_errors
++;
921 spin_unlock(&sdev
->stat_lock
);
922 printk_ratelimited(KERN_ERR
923 "btrfs: fixed up error at logical %llu on dev %s\n",
924 (unsigned long long)logical
, sdev
->dev
->name
);
927 did_not_correct_error
:
928 spin_lock(&sdev
->stat_lock
);
929 sdev
->stat
.uncorrectable_errors
++;
930 spin_unlock(&sdev
->stat_lock
);
931 printk_ratelimited(KERN_ERR
932 "btrfs: unable to fixup (regular) error at logical %llu on dev %s\n",
933 (unsigned long long)logical
, sdev
->dev
->name
);
937 if (sblocks_for_recheck
) {
938 for (mirror_index
= 0; mirror_index
< BTRFS_MAX_MIRRORS
;
940 struct scrub_block
*sblock
= sblocks_for_recheck
+
944 for (page_index
= 0; page_index
< SCRUB_PAGES_PER_BIO
;
946 if (sblock
->pagev
[page_index
].page
)
948 sblock
->pagev
[page_index
].page
);
950 kfree(sblocks_for_recheck
);
956 static int scrub_setup_recheck_block(struct scrub_dev
*sdev
,
957 struct btrfs_mapping_tree
*map_tree
,
958 u64 length
, u64 logical
,
959 struct scrub_block
*sblocks_for_recheck
)
966 * note: the three members sdev, ref_count and outstanding_pages
967 * are not used (and not set) in the blocks that are used for
968 * the recheck procedure
973 u64 sublen
= min_t(u64
, length
, PAGE_SIZE
);
974 u64 mapped_length
= sublen
;
975 struct btrfs_bio
*bbio
= NULL
;
978 * with a length of PAGE_SIZE, each returned stripe
979 * represents one mirror
981 ret
= btrfs_map_block(map_tree
, WRITE
, logical
, &mapped_length
,
983 if (ret
|| !bbio
|| mapped_length
< sublen
) {
988 BUG_ON(page_index
>= SCRUB_PAGES_PER_BIO
);
989 for (mirror_index
= 0; mirror_index
< (int)bbio
->num_stripes
;
991 struct scrub_block
*sblock
;
992 struct scrub_page
*page
;
994 if (mirror_index
>= BTRFS_MAX_MIRRORS
)
997 sblock
= sblocks_for_recheck
+ mirror_index
;
998 page
= sblock
->pagev
+ page_index
;
999 page
->logical
= logical
;
1000 page
->physical
= bbio
->stripes
[mirror_index
].physical
;
1001 page
->bdev
= bbio
->stripes
[mirror_index
].dev
->bdev
;
1002 page
->mirror_num
= mirror_index
+ 1;
1003 page
->page
= alloc_page(GFP_NOFS
);
1005 spin_lock(&sdev
->stat_lock
);
1006 sdev
->stat
.malloc_errors
++;
1007 spin_unlock(&sdev
->stat_lock
);
1010 sblock
->page_count
++;
1022 * this function will check the on disk data for checksum errors, header
1023 * errors and read I/O errors. If any I/O errors happen, the exact pages
1024 * which are errored are marked as being bad. The goal is to enable scrub
1025 * to take those pages that are not errored from all the mirrors so that
1026 * the pages that are errored in the just handled mirror can be repaired.
1028 static int scrub_recheck_block(struct btrfs_fs_info
*fs_info
,
1029 struct scrub_block
*sblock
, int is_metadata
,
1030 int have_csum
, u8
*csum
, u64 generation
,
1035 sblock
->no_io_error_seen
= 1;
1036 sblock
->header_error
= 0;
1037 sblock
->checksum_error
= 0;
1039 for (page_num
= 0; page_num
< sblock
->page_count
; page_num
++) {
1042 struct scrub_page
*page
= sblock
->pagev
+ page_num
;
1043 DECLARE_COMPLETION_ONSTACK(complete
);
1045 BUG_ON(!page
->page
);
1046 bio
= bio_alloc(GFP_NOFS
, 1);
1047 bio
->bi_bdev
= page
->bdev
;
1048 bio
->bi_sector
= page
->physical
>> 9;
1049 bio
->bi_end_io
= scrub_complete_bio_end_io
;
1050 bio
->bi_private
= &complete
;
1052 ret
= bio_add_page(bio
, page
->page
, PAGE_SIZE
, 0);
1053 if (PAGE_SIZE
!= ret
) {
1057 btrfsic_submit_bio(READ
, bio
);
1059 /* this will also unplug the queue */
1060 wait_for_completion(&complete
);
1062 page
->io_error
= !test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1063 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1064 sblock
->no_io_error_seen
= 0;
1068 if (sblock
->no_io_error_seen
)
1069 scrub_recheck_block_checksum(fs_info
, sblock
, is_metadata
,
1070 have_csum
, csum
, generation
,
1076 static void scrub_recheck_block_checksum(struct btrfs_fs_info
*fs_info
,
1077 struct scrub_block
*sblock
,
1078 int is_metadata
, int have_csum
,
1079 const u8
*csum
, u64 generation
,
1083 u8 calculated_csum
[BTRFS_CSUM_SIZE
];
1085 struct btrfs_root
*root
= fs_info
->extent_root
;
1086 void *mapped_buffer
;
1088 BUG_ON(!sblock
->pagev
[0].page
);
1090 struct btrfs_header
*h
;
1092 mapped_buffer
= kmap_atomic(sblock
->pagev
[0].page
);
1093 h
= (struct btrfs_header
*)mapped_buffer
;
1095 if (sblock
->pagev
[0].logical
!= le64_to_cpu(h
->bytenr
) ||
1096 generation
!= le64_to_cpu(h
->generation
) ||
1097 memcmp(h
->fsid
, fs_info
->fsid
, BTRFS_UUID_SIZE
) ||
1098 memcmp(h
->chunk_tree_uuid
, fs_info
->chunk_tree_uuid
,
1100 sblock
->header_error
= 1;
1106 mapped_buffer
= kmap_atomic(sblock
->pagev
[0].page
);
1109 for (page_num
= 0;;) {
1110 if (page_num
== 0 && is_metadata
)
1111 crc
= btrfs_csum_data(root
,
1112 ((u8
*)mapped_buffer
) + BTRFS_CSUM_SIZE
,
1113 crc
, PAGE_SIZE
- BTRFS_CSUM_SIZE
);
1115 crc
= btrfs_csum_data(root
, mapped_buffer
, crc
,
1118 kunmap_atomic(mapped_buffer
);
1120 if (page_num
>= sblock
->page_count
)
1122 BUG_ON(!sblock
->pagev
[page_num
].page
);
1124 mapped_buffer
= kmap_atomic(sblock
->pagev
[page_num
].page
);
1127 btrfs_csum_final(crc
, calculated_csum
);
1128 if (memcmp(calculated_csum
, csum
, csum_size
))
1129 sblock
->checksum_error
= 1;
1132 static void scrub_complete_bio_end_io(struct bio
*bio
, int err
)
1134 complete((struct completion
*)bio
->bi_private
);
1137 static int scrub_repair_block_from_good_copy(struct scrub_block
*sblock_bad
,
1138 struct scrub_block
*sblock_good
,
1144 for (page_num
= 0; page_num
< sblock_bad
->page_count
; page_num
++) {
1147 ret_sub
= scrub_repair_page_from_good_copy(sblock_bad
,
1158 static int scrub_repair_page_from_good_copy(struct scrub_block
*sblock_bad
,
1159 struct scrub_block
*sblock_good
,
1160 int page_num
, int force_write
)
1162 struct scrub_page
*page_bad
= sblock_bad
->pagev
+ page_num
;
1163 struct scrub_page
*page_good
= sblock_good
->pagev
+ page_num
;
1165 BUG_ON(sblock_bad
->pagev
[page_num
].page
== NULL
);
1166 BUG_ON(sblock_good
->pagev
[page_num
].page
== NULL
);
1167 if (force_write
|| sblock_bad
->header_error
||
1168 sblock_bad
->checksum_error
|| page_bad
->io_error
) {
1171 DECLARE_COMPLETION_ONSTACK(complete
);
1173 bio
= bio_alloc(GFP_NOFS
, 1);
1174 bio
->bi_bdev
= page_bad
->bdev
;
1175 bio
->bi_sector
= page_bad
->physical
>> 9;
1176 bio
->bi_end_io
= scrub_complete_bio_end_io
;
1177 bio
->bi_private
= &complete
;
1179 ret
= bio_add_page(bio
, page_good
->page
, PAGE_SIZE
, 0);
1180 if (PAGE_SIZE
!= ret
) {
1184 btrfsic_submit_bio(WRITE
, bio
);
1186 /* this will also unplug the queue */
1187 wait_for_completion(&complete
);
1194 static void scrub_checksum(struct scrub_block
*sblock
)
1199 BUG_ON(sblock
->page_count
< 1);
1200 flags
= sblock
->pagev
[0].flags
;
1202 if (flags
& BTRFS_EXTENT_FLAG_DATA
)
1203 ret
= scrub_checksum_data(sblock
);
1204 else if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
)
1205 ret
= scrub_checksum_tree_block(sblock
);
1206 else if (flags
& BTRFS_EXTENT_FLAG_SUPER
)
1207 (void)scrub_checksum_super(sblock
);
1211 scrub_handle_errored_block(sblock
);
1214 static int scrub_checksum_data(struct scrub_block
*sblock
)
1216 struct scrub_dev
*sdev
= sblock
->sdev
;
1217 u8 csum
[BTRFS_CSUM_SIZE
];
1223 struct btrfs_root
*root
= sdev
->dev
->dev_root
;
1227 BUG_ON(sblock
->page_count
< 1);
1228 if (!sblock
->pagev
[0].have_csum
)
1231 on_disk_csum
= sblock
->pagev
[0].csum
;
1232 page
= sblock
->pagev
[0].page
;
1233 buffer
= kmap_atomic(page
);
1235 len
= sdev
->sectorsize
;
1238 u64 l
= min_t(u64
, len
, PAGE_SIZE
);
1240 crc
= btrfs_csum_data(root
, buffer
, crc
, l
);
1241 kunmap_atomic(buffer
);
1246 BUG_ON(index
>= sblock
->page_count
);
1247 BUG_ON(!sblock
->pagev
[index
].page
);
1248 page
= sblock
->pagev
[index
].page
;
1249 buffer
= kmap_atomic(page
);
1252 btrfs_csum_final(crc
, csum
);
1253 if (memcmp(csum
, on_disk_csum
, sdev
->csum_size
))
1257 spin_lock(&sdev
->stat_lock
);
1258 ++sdev
->stat
.csum_errors
;
1259 spin_unlock(&sdev
->stat_lock
);
1265 static int scrub_checksum_tree_block(struct scrub_block
*sblock
)
1267 struct scrub_dev
*sdev
= sblock
->sdev
;
1268 struct btrfs_header
*h
;
1269 struct btrfs_root
*root
= sdev
->dev
->dev_root
;
1270 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1271 u8 calculated_csum
[BTRFS_CSUM_SIZE
];
1272 u8 on_disk_csum
[BTRFS_CSUM_SIZE
];
1274 void *mapped_buffer
;
1283 BUG_ON(sblock
->page_count
< 1);
1284 page
= sblock
->pagev
[0].page
;
1285 mapped_buffer
= kmap_atomic(page
);
1286 h
= (struct btrfs_header
*)mapped_buffer
;
1287 memcpy(on_disk_csum
, h
->csum
, sdev
->csum_size
);
1290 * we don't use the getter functions here, as we
1291 * a) don't have an extent buffer and
1292 * b) the page is already kmapped
1295 if (sblock
->pagev
[0].logical
!= le64_to_cpu(h
->bytenr
))
1298 if (sblock
->pagev
[0].generation
!= le64_to_cpu(h
->generation
))
1301 if (memcmp(h
->fsid
, fs_info
->fsid
, BTRFS_UUID_SIZE
))
1304 if (memcmp(h
->chunk_tree_uuid
, fs_info
->chunk_tree_uuid
,
1308 BUG_ON(sdev
->nodesize
!= sdev
->leafsize
);
1309 len
= sdev
->nodesize
- BTRFS_CSUM_SIZE
;
1310 mapped_size
= PAGE_SIZE
- BTRFS_CSUM_SIZE
;
1311 p
= ((u8
*)mapped_buffer
) + BTRFS_CSUM_SIZE
;
1314 u64 l
= min_t(u64
, len
, mapped_size
);
1316 crc
= btrfs_csum_data(root
, p
, crc
, l
);
1317 kunmap_atomic(mapped_buffer
);
1322 BUG_ON(index
>= sblock
->page_count
);
1323 BUG_ON(!sblock
->pagev
[index
].page
);
1324 page
= sblock
->pagev
[index
].page
;
1325 mapped_buffer
= kmap_atomic(page
);
1326 mapped_size
= PAGE_SIZE
;
1330 btrfs_csum_final(crc
, calculated_csum
);
1331 if (memcmp(calculated_csum
, on_disk_csum
, sdev
->csum_size
))
1334 if (crc_fail
|| fail
) {
1335 spin_lock(&sdev
->stat_lock
);
1337 ++sdev
->stat
.csum_errors
;
1339 ++sdev
->stat
.verify_errors
;
1340 spin_unlock(&sdev
->stat_lock
);
1343 return fail
|| crc_fail
;
1346 static int scrub_checksum_super(struct scrub_block
*sblock
)
1348 struct btrfs_super_block
*s
;
1349 struct scrub_dev
*sdev
= sblock
->sdev
;
1350 struct btrfs_root
*root
= sdev
->dev
->dev_root
;
1351 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1352 u8 calculated_csum
[BTRFS_CSUM_SIZE
];
1353 u8 on_disk_csum
[BTRFS_CSUM_SIZE
];
1355 void *mapped_buffer
;
1363 BUG_ON(sblock
->page_count
< 1);
1364 page
= sblock
->pagev
[0].page
;
1365 mapped_buffer
= kmap_atomic(page
);
1366 s
= (struct btrfs_super_block
*)mapped_buffer
;
1367 memcpy(on_disk_csum
, s
->csum
, sdev
->csum_size
);
1369 if (sblock
->pagev
[0].logical
!= le64_to_cpu(s
->bytenr
))
1372 if (sblock
->pagev
[0].generation
!= le64_to_cpu(s
->generation
))
1375 if (memcmp(s
->fsid
, fs_info
->fsid
, BTRFS_UUID_SIZE
))
1378 len
= BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
;
1379 mapped_size
= PAGE_SIZE
- BTRFS_CSUM_SIZE
;
1380 p
= ((u8
*)mapped_buffer
) + BTRFS_CSUM_SIZE
;
1383 u64 l
= min_t(u64
, len
, mapped_size
);
1385 crc
= btrfs_csum_data(root
, p
, crc
, l
);
1386 kunmap_atomic(mapped_buffer
);
1391 BUG_ON(index
>= sblock
->page_count
);
1392 BUG_ON(!sblock
->pagev
[index
].page
);
1393 page
= sblock
->pagev
[index
].page
;
1394 mapped_buffer
= kmap_atomic(page
);
1395 mapped_size
= PAGE_SIZE
;
1399 btrfs_csum_final(crc
, calculated_csum
);
1400 if (memcmp(calculated_csum
, on_disk_csum
, sdev
->csum_size
))
1405 * if we find an error in a super block, we just report it.
1406 * They will get written with the next transaction commit
1409 spin_lock(&sdev
->stat_lock
);
1410 ++sdev
->stat
.super_errors
;
1411 spin_unlock(&sdev
->stat_lock
);
1417 static void scrub_block_get(struct scrub_block
*sblock
)
1419 atomic_inc(&sblock
->ref_count
);
1422 static void scrub_block_put(struct scrub_block
*sblock
)
1424 if (atomic_dec_and_test(&sblock
->ref_count
)) {
1427 for (i
= 0; i
< sblock
->page_count
; i
++)
1428 if (sblock
->pagev
[i
].page
)
1429 __free_page(sblock
->pagev
[i
].page
);
1434 static void scrub_submit(struct scrub_dev
*sdev
)
1436 struct scrub_bio
*sbio
;
1438 if (sdev
->curr
== -1)
1441 sbio
= sdev
->bios
[sdev
->curr
];
1443 atomic_inc(&sdev
->in_flight
);
1445 btrfsic_submit_bio(READ
, sbio
->bio
);
1448 static int scrub_add_page_to_bio(struct scrub_dev
*sdev
,
1449 struct scrub_page
*spage
)
1451 struct scrub_block
*sblock
= spage
->sblock
;
1452 struct scrub_bio
*sbio
;
1457 * grab a fresh bio or wait for one to become available
1459 while (sdev
->curr
== -1) {
1460 spin_lock(&sdev
->list_lock
);
1461 sdev
->curr
= sdev
->first_free
;
1462 if (sdev
->curr
!= -1) {
1463 sdev
->first_free
= sdev
->bios
[sdev
->curr
]->next_free
;
1464 sdev
->bios
[sdev
->curr
]->next_free
= -1;
1465 sdev
->bios
[sdev
->curr
]->page_count
= 0;
1466 spin_unlock(&sdev
->list_lock
);
1468 spin_unlock(&sdev
->list_lock
);
1469 wait_event(sdev
->list_wait
, sdev
->first_free
!= -1);
1472 sbio
= sdev
->bios
[sdev
->curr
];
1473 if (sbio
->page_count
== 0) {
1476 sbio
->physical
= spage
->physical
;
1477 sbio
->logical
= spage
->logical
;
1480 bio
= bio_alloc(GFP_NOFS
, sdev
->pages_per_bio
);
1486 bio
->bi_private
= sbio
;
1487 bio
->bi_end_io
= scrub_bio_end_io
;
1488 bio
->bi_bdev
= sdev
->dev
->bdev
;
1489 bio
->bi_sector
= spage
->physical
>> 9;
1491 } else if (sbio
->physical
+ sbio
->page_count
* PAGE_SIZE
!=
1493 sbio
->logical
+ sbio
->page_count
* PAGE_SIZE
!=
1499 sbio
->pagev
[sbio
->page_count
] = spage
;
1500 ret
= bio_add_page(sbio
->bio
, spage
->page
, PAGE_SIZE
, 0);
1501 if (ret
!= PAGE_SIZE
) {
1502 if (sbio
->page_count
< 1) {
1511 scrub_block_get(sblock
); /* one for the added page */
1512 atomic_inc(&sblock
->outstanding_pages
);
1514 if (sbio
->page_count
== sdev
->pages_per_bio
)
1520 static int scrub_pages(struct scrub_dev
*sdev
, u64 logical
, u64 len
,
1521 u64 physical
, u64 flags
, u64 gen
, int mirror_num
,
1522 u8
*csum
, int force
)
1524 struct scrub_block
*sblock
;
1527 sblock
= kzalloc(sizeof(*sblock
), GFP_NOFS
);
1529 spin_lock(&sdev
->stat_lock
);
1530 sdev
->stat
.malloc_errors
++;
1531 spin_unlock(&sdev
->stat_lock
);
1535 /* one ref inside this function, plus one for each page later on */
1536 atomic_set(&sblock
->ref_count
, 1);
1537 sblock
->sdev
= sdev
;
1538 sblock
->no_io_error_seen
= 1;
1540 for (index
= 0; len
> 0; index
++) {
1541 struct scrub_page
*spage
= sblock
->pagev
+ index
;
1542 u64 l
= min_t(u64
, len
, PAGE_SIZE
);
1544 BUG_ON(index
>= SCRUB_MAX_PAGES_PER_BLOCK
);
1545 spage
->page
= alloc_page(GFP_NOFS
);
1547 spin_lock(&sdev
->stat_lock
);
1548 sdev
->stat
.malloc_errors
++;
1549 spin_unlock(&sdev
->stat_lock
);
1552 __free_page(sblock
->pagev
[index
].page
);
1557 spage
->sblock
= sblock
;
1558 spage
->bdev
= sdev
->dev
->bdev
;
1559 spage
->flags
= flags
;
1560 spage
->generation
= gen
;
1561 spage
->logical
= logical
;
1562 spage
->physical
= physical
;
1563 spage
->mirror_num
= mirror_num
;
1565 spage
->have_csum
= 1;
1566 memcpy(spage
->csum
, csum
, sdev
->csum_size
);
1568 spage
->have_csum
= 0;
1570 sblock
->page_count
++;
1576 BUG_ON(sblock
->page_count
== 0);
1577 for (index
= 0; index
< sblock
->page_count
; index
++) {
1578 struct scrub_page
*spage
= sblock
->pagev
+ index
;
1581 ret
= scrub_add_page_to_bio(sdev
, spage
);
1583 scrub_block_put(sblock
);
1591 /* last one frees, either here or in bio completion for last page */
1592 scrub_block_put(sblock
);
1596 static void scrub_bio_end_io(struct bio
*bio
, int err
)
1598 struct scrub_bio
*sbio
= bio
->bi_private
;
1599 struct scrub_dev
*sdev
= sbio
->sdev
;
1600 struct btrfs_fs_info
*fs_info
= sdev
->dev
->dev_root
->fs_info
;
1605 btrfs_queue_worker(&fs_info
->scrub_workers
, &sbio
->work
);
1608 static void scrub_bio_end_io_worker(struct btrfs_work
*work
)
1610 struct scrub_bio
*sbio
= container_of(work
, struct scrub_bio
, work
);
1611 struct scrub_dev
*sdev
= sbio
->sdev
;
1614 BUG_ON(sbio
->page_count
> SCRUB_PAGES_PER_BIO
);
1616 for (i
= 0; i
< sbio
->page_count
; i
++) {
1617 struct scrub_page
*spage
= sbio
->pagev
[i
];
1619 spage
->io_error
= 1;
1620 spage
->sblock
->no_io_error_seen
= 0;
1624 /* now complete the scrub_block items that have all pages completed */
1625 for (i
= 0; i
< sbio
->page_count
; i
++) {
1626 struct scrub_page
*spage
= sbio
->pagev
[i
];
1627 struct scrub_block
*sblock
= spage
->sblock
;
1629 if (atomic_dec_and_test(&sblock
->outstanding_pages
))
1630 scrub_block_complete(sblock
);
1631 scrub_block_put(sblock
);
1635 /* what is this good for??? */
1636 sbio
->bio
->bi_flags
&= ~(BIO_POOL_MASK
- 1);
1637 sbio
->bio
->bi_flags
|= 1 << BIO_UPTODATE
;
1638 sbio
->bio
->bi_phys_segments
= 0;
1639 sbio
->bio
->bi_idx
= 0;
1641 for (i
= 0; i
< sbio
->page_count
; i
++) {
1643 bi
= &sbio
->bio
->bi_io_vec
[i
];
1645 bi
->bv_len
= PAGE_SIZE
;
1651 spin_lock(&sdev
->list_lock
);
1652 sbio
->next_free
= sdev
->first_free
;
1653 sdev
->first_free
= sbio
->index
;
1654 spin_unlock(&sdev
->list_lock
);
1655 atomic_dec(&sdev
->in_flight
);
1656 wake_up(&sdev
->list_wait
);
1659 static void scrub_block_complete(struct scrub_block
*sblock
)
1661 if (!sblock
->no_io_error_seen
)
1662 scrub_handle_errored_block(sblock
);
1664 scrub_checksum(sblock
);
1667 static int scrub_find_csum(struct scrub_dev
*sdev
, u64 logical
, u64 len
,
1670 struct btrfs_ordered_sum
*sum
= NULL
;
1673 unsigned long num_sectors
;
1675 while (!list_empty(&sdev
->csum_list
)) {
1676 sum
= list_first_entry(&sdev
->csum_list
,
1677 struct btrfs_ordered_sum
, list
);
1678 if (sum
->bytenr
> logical
)
1680 if (sum
->bytenr
+ sum
->len
> logical
)
1683 ++sdev
->stat
.csum_discards
;
1684 list_del(&sum
->list
);
1691 num_sectors
= sum
->len
/ sdev
->sectorsize
;
1692 for (i
= 0; i
< num_sectors
; ++i
) {
1693 if (sum
->sums
[i
].bytenr
== logical
) {
1694 memcpy(csum
, &sum
->sums
[i
].sum
, sdev
->csum_size
);
1699 if (ret
&& i
== num_sectors
- 1) {
1700 list_del(&sum
->list
);
1706 /* scrub extent tries to collect up to 64 kB for each bio */
1707 static int scrub_extent(struct scrub_dev
*sdev
, u64 logical
, u64 len
,
1708 u64 physical
, u64 flags
, u64 gen
, int mirror_num
)
1711 u8 csum
[BTRFS_CSUM_SIZE
];
1714 if (flags
& BTRFS_EXTENT_FLAG_DATA
) {
1715 blocksize
= sdev
->sectorsize
;
1716 spin_lock(&sdev
->stat_lock
);
1717 sdev
->stat
.data_extents_scrubbed
++;
1718 sdev
->stat
.data_bytes_scrubbed
+= len
;
1719 spin_unlock(&sdev
->stat_lock
);
1720 } else if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1721 BUG_ON(sdev
->nodesize
!= sdev
->leafsize
);
1722 blocksize
= sdev
->nodesize
;
1723 spin_lock(&sdev
->stat_lock
);
1724 sdev
->stat
.tree_extents_scrubbed
++;
1725 sdev
->stat
.tree_bytes_scrubbed
+= len
;
1726 spin_unlock(&sdev
->stat_lock
);
1728 blocksize
= sdev
->sectorsize
;
1733 u64 l
= min_t(u64
, len
, blocksize
);
1736 if (flags
& BTRFS_EXTENT_FLAG_DATA
) {
1737 /* push csums to sbio */
1738 have_csum
= scrub_find_csum(sdev
, logical
, l
, csum
);
1740 ++sdev
->stat
.no_csum
;
1742 ret
= scrub_pages(sdev
, logical
, l
, physical
, flags
, gen
,
1743 mirror_num
, have_csum
? csum
: NULL
, 0);
1753 static noinline_for_stack
int scrub_stripe(struct scrub_dev
*sdev
,
1754 struct map_lookup
*map
, int num
, u64 base
, u64 length
)
1756 struct btrfs_path
*path
;
1757 struct btrfs_fs_info
*fs_info
= sdev
->dev
->dev_root
->fs_info
;
1758 struct btrfs_root
*root
= fs_info
->extent_root
;
1759 struct btrfs_root
*csum_root
= fs_info
->csum_root
;
1760 struct btrfs_extent_item
*extent
;
1761 struct blk_plug plug
;
1767 struct extent_buffer
*l
;
1768 struct btrfs_key key
;
1773 struct reada_control
*reada1
;
1774 struct reada_control
*reada2
;
1775 struct btrfs_key key_start
;
1776 struct btrfs_key key_end
;
1778 u64 increment
= map
->stripe_len
;
1783 do_div(nstripes
, map
->stripe_len
);
1784 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
1785 offset
= map
->stripe_len
* num
;
1786 increment
= map
->stripe_len
* map
->num_stripes
;
1788 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
1789 int factor
= map
->num_stripes
/ map
->sub_stripes
;
1790 offset
= map
->stripe_len
* (num
/ map
->sub_stripes
);
1791 increment
= map
->stripe_len
* factor
;
1792 mirror_num
= num
% map
->sub_stripes
+ 1;
1793 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
1794 increment
= map
->stripe_len
;
1795 mirror_num
= num
% map
->num_stripes
+ 1;
1796 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
1797 increment
= map
->stripe_len
;
1798 mirror_num
= num
% map
->num_stripes
+ 1;
1800 increment
= map
->stripe_len
;
1804 path
= btrfs_alloc_path();
1809 * work on commit root. The related disk blocks are static as
1810 * long as COW is applied. This means, it is save to rewrite
1811 * them to repair disk errors without any race conditions
1813 path
->search_commit_root
= 1;
1814 path
->skip_locking
= 1;
1817 * trigger the readahead for extent tree csum tree and wait for
1818 * completion. During readahead, the scrub is officially paused
1819 * to not hold off transaction commits
1821 logical
= base
+ offset
;
1823 wait_event(sdev
->list_wait
,
1824 atomic_read(&sdev
->in_flight
) == 0);
1825 atomic_inc(&fs_info
->scrubs_paused
);
1826 wake_up(&fs_info
->scrub_pause_wait
);
1828 /* FIXME it might be better to start readahead at commit root */
1829 key_start
.objectid
= logical
;
1830 key_start
.type
= BTRFS_EXTENT_ITEM_KEY
;
1831 key_start
.offset
= (u64
)0;
1832 key_end
.objectid
= base
+ offset
+ nstripes
* increment
;
1833 key_end
.type
= BTRFS_EXTENT_ITEM_KEY
;
1834 key_end
.offset
= (u64
)0;
1835 reada1
= btrfs_reada_add(root
, &key_start
, &key_end
);
1837 key_start
.objectid
= BTRFS_EXTENT_CSUM_OBJECTID
;
1838 key_start
.type
= BTRFS_EXTENT_CSUM_KEY
;
1839 key_start
.offset
= logical
;
1840 key_end
.objectid
= BTRFS_EXTENT_CSUM_OBJECTID
;
1841 key_end
.type
= BTRFS_EXTENT_CSUM_KEY
;
1842 key_end
.offset
= base
+ offset
+ nstripes
* increment
;
1843 reada2
= btrfs_reada_add(csum_root
, &key_start
, &key_end
);
1845 if (!IS_ERR(reada1
))
1846 btrfs_reada_wait(reada1
);
1847 if (!IS_ERR(reada2
))
1848 btrfs_reada_wait(reada2
);
1850 mutex_lock(&fs_info
->scrub_lock
);
1851 while (atomic_read(&fs_info
->scrub_pause_req
)) {
1852 mutex_unlock(&fs_info
->scrub_lock
);
1853 wait_event(fs_info
->scrub_pause_wait
,
1854 atomic_read(&fs_info
->scrub_pause_req
) == 0);
1855 mutex_lock(&fs_info
->scrub_lock
);
1857 atomic_dec(&fs_info
->scrubs_paused
);
1858 mutex_unlock(&fs_info
->scrub_lock
);
1859 wake_up(&fs_info
->scrub_pause_wait
);
1862 * collect all data csums for the stripe to avoid seeking during
1863 * the scrub. This might currently (crc32) end up to be about 1MB
1865 blk_start_plug(&plug
);
1868 * now find all extents for each stripe and scrub them
1870 logical
= base
+ offset
;
1871 physical
= map
->stripes
[num
].physical
;
1873 for (i
= 0; i
< nstripes
; ++i
) {
1877 if (atomic_read(&fs_info
->scrub_cancel_req
) ||
1878 atomic_read(&sdev
->cancel_req
)) {
1883 * check to see if we have to pause
1885 if (atomic_read(&fs_info
->scrub_pause_req
)) {
1886 /* push queued extents */
1888 wait_event(sdev
->list_wait
,
1889 atomic_read(&sdev
->in_flight
) == 0);
1890 atomic_inc(&fs_info
->scrubs_paused
);
1891 wake_up(&fs_info
->scrub_pause_wait
);
1892 mutex_lock(&fs_info
->scrub_lock
);
1893 while (atomic_read(&fs_info
->scrub_pause_req
)) {
1894 mutex_unlock(&fs_info
->scrub_lock
);
1895 wait_event(fs_info
->scrub_pause_wait
,
1896 atomic_read(&fs_info
->scrub_pause_req
) == 0);
1897 mutex_lock(&fs_info
->scrub_lock
);
1899 atomic_dec(&fs_info
->scrubs_paused
);
1900 mutex_unlock(&fs_info
->scrub_lock
);
1901 wake_up(&fs_info
->scrub_pause_wait
);
1904 ret
= btrfs_lookup_csums_range(csum_root
, logical
,
1905 logical
+ map
->stripe_len
- 1,
1906 &sdev
->csum_list
, 1);
1910 key
.objectid
= logical
;
1911 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1912 key
.offset
= (u64
)0;
1914 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1918 ret
= btrfs_previous_item(root
, path
, 0,
1919 BTRFS_EXTENT_ITEM_KEY
);
1923 /* there's no smaller item, so stick with the
1925 btrfs_release_path(path
);
1926 ret
= btrfs_search_slot(NULL
, root
, &key
,
1935 slot
= path
->slots
[0];
1936 if (slot
>= btrfs_header_nritems(l
)) {
1937 ret
= btrfs_next_leaf(root
, path
);
1945 btrfs_item_key_to_cpu(l
, &key
, slot
);
1947 if (key
.objectid
+ key
.offset
<= logical
)
1950 if (key
.objectid
>= logical
+ map
->stripe_len
)
1953 if (btrfs_key_type(&key
) != BTRFS_EXTENT_ITEM_KEY
)
1956 extent
= btrfs_item_ptr(l
, slot
,
1957 struct btrfs_extent_item
);
1958 flags
= btrfs_extent_flags(l
, extent
);
1959 generation
= btrfs_extent_generation(l
, extent
);
1961 if (key
.objectid
< logical
&&
1962 (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
)) {
1964 "btrfs scrub: tree block %llu spanning "
1965 "stripes, ignored. logical=%llu\n",
1966 (unsigned long long)key
.objectid
,
1967 (unsigned long long)logical
);
1972 * trim extent to this stripe
1974 if (key
.objectid
< logical
) {
1975 key
.offset
-= logical
- key
.objectid
;
1976 key
.objectid
= logical
;
1978 if (key
.objectid
+ key
.offset
>
1979 logical
+ map
->stripe_len
) {
1980 key
.offset
= logical
+ map
->stripe_len
-
1984 ret
= scrub_extent(sdev
, key
.objectid
, key
.offset
,
1985 key
.objectid
- logical
+ physical
,
1986 flags
, generation
, mirror_num
);
1993 btrfs_release_path(path
);
1994 logical
+= increment
;
1995 physical
+= map
->stripe_len
;
1996 spin_lock(&sdev
->stat_lock
);
1997 sdev
->stat
.last_physical
= physical
;
1998 spin_unlock(&sdev
->stat_lock
);
2000 /* push queued extents */
2004 blk_finish_plug(&plug
);
2005 btrfs_free_path(path
);
2006 return ret
< 0 ? ret
: 0;
2009 static noinline_for_stack
int scrub_chunk(struct scrub_dev
*sdev
,
2010 u64 chunk_tree
, u64 chunk_objectid
, u64 chunk_offset
, u64 length
,
2013 struct btrfs_mapping_tree
*map_tree
=
2014 &sdev
->dev
->dev_root
->fs_info
->mapping_tree
;
2015 struct map_lookup
*map
;
2016 struct extent_map
*em
;
2020 read_lock(&map_tree
->map_tree
.lock
);
2021 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
2022 read_unlock(&map_tree
->map_tree
.lock
);
2027 map
= (struct map_lookup
*)em
->bdev
;
2028 if (em
->start
!= chunk_offset
)
2031 if (em
->len
< length
)
2034 for (i
= 0; i
< map
->num_stripes
; ++i
) {
2035 if (map
->stripes
[i
].dev
== sdev
->dev
&&
2036 map
->stripes
[i
].physical
== dev_offset
) {
2037 ret
= scrub_stripe(sdev
, map
, i
, chunk_offset
, length
);
2043 free_extent_map(em
);
2048 static noinline_for_stack
2049 int scrub_enumerate_chunks(struct scrub_dev
*sdev
, u64 start
, u64 end
)
2051 struct btrfs_dev_extent
*dev_extent
= NULL
;
2052 struct btrfs_path
*path
;
2053 struct btrfs_root
*root
= sdev
->dev
->dev_root
;
2054 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2061 struct extent_buffer
*l
;
2062 struct btrfs_key key
;
2063 struct btrfs_key found_key
;
2064 struct btrfs_block_group_cache
*cache
;
2066 path
= btrfs_alloc_path();
2071 path
->search_commit_root
= 1;
2072 path
->skip_locking
= 1;
2074 key
.objectid
= sdev
->dev
->devid
;
2076 key
.type
= BTRFS_DEV_EXTENT_KEY
;
2080 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2084 if (path
->slots
[0] >=
2085 btrfs_header_nritems(path
->nodes
[0])) {
2086 ret
= btrfs_next_leaf(root
, path
);
2093 slot
= path
->slots
[0];
2095 btrfs_item_key_to_cpu(l
, &found_key
, slot
);
2097 if (found_key
.objectid
!= sdev
->dev
->devid
)
2100 if (btrfs_key_type(&found_key
) != BTRFS_DEV_EXTENT_KEY
)
2103 if (found_key
.offset
>= end
)
2106 if (found_key
.offset
< key
.offset
)
2109 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
2110 length
= btrfs_dev_extent_length(l
, dev_extent
);
2112 if (found_key
.offset
+ length
<= start
) {
2113 key
.offset
= found_key
.offset
+ length
;
2114 btrfs_release_path(path
);
2118 chunk_tree
= btrfs_dev_extent_chunk_tree(l
, dev_extent
);
2119 chunk_objectid
= btrfs_dev_extent_chunk_objectid(l
, dev_extent
);
2120 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
2123 * get a reference on the corresponding block group to prevent
2124 * the chunk from going away while we scrub it
2126 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
2131 ret
= scrub_chunk(sdev
, chunk_tree
, chunk_objectid
,
2132 chunk_offset
, length
, found_key
.offset
);
2133 btrfs_put_block_group(cache
);
2137 key
.offset
= found_key
.offset
+ length
;
2138 btrfs_release_path(path
);
2141 btrfs_free_path(path
);
2144 * ret can still be 1 from search_slot or next_leaf,
2145 * that's not an error
2147 return ret
< 0 ? ret
: 0;
2150 static noinline_for_stack
int scrub_supers(struct scrub_dev
*sdev
)
2156 struct btrfs_device
*device
= sdev
->dev
;
2157 struct btrfs_root
*root
= device
->dev_root
;
2159 if (root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
)
2162 gen
= root
->fs_info
->last_trans_committed
;
2164 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
2165 bytenr
= btrfs_sb_offset(i
);
2166 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
> device
->total_bytes
)
2169 ret
= scrub_pages(sdev
, bytenr
, BTRFS_SUPER_INFO_SIZE
, bytenr
,
2170 BTRFS_EXTENT_FLAG_SUPER
, gen
, i
, NULL
, 1);
2174 wait_event(sdev
->list_wait
, atomic_read(&sdev
->in_flight
) == 0);
2180 * get a reference count on fs_info->scrub_workers. start worker if necessary
2182 static noinline_for_stack
int scrub_workers_get(struct btrfs_root
*root
)
2184 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2187 mutex_lock(&fs_info
->scrub_lock
);
2188 if (fs_info
->scrub_workers_refcnt
== 0) {
2189 btrfs_init_workers(&fs_info
->scrub_workers
, "scrub",
2190 fs_info
->thread_pool_size
, &fs_info
->generic_worker
);
2191 fs_info
->scrub_workers
.idle_thresh
= 4;
2192 ret
= btrfs_start_workers(&fs_info
->scrub_workers
);
2196 ++fs_info
->scrub_workers_refcnt
;
2198 mutex_unlock(&fs_info
->scrub_lock
);
2203 static noinline_for_stack
void scrub_workers_put(struct btrfs_root
*root
)
2205 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2207 mutex_lock(&fs_info
->scrub_lock
);
2208 if (--fs_info
->scrub_workers_refcnt
== 0)
2209 btrfs_stop_workers(&fs_info
->scrub_workers
);
2210 WARN_ON(fs_info
->scrub_workers_refcnt
< 0);
2211 mutex_unlock(&fs_info
->scrub_lock
);
2215 int btrfs_scrub_dev(struct btrfs_root
*root
, u64 devid
, u64 start
, u64 end
,
2216 struct btrfs_scrub_progress
*progress
, int readonly
)
2218 struct scrub_dev
*sdev
;
2219 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2221 struct btrfs_device
*dev
;
2223 if (btrfs_fs_closing(root
->fs_info
))
2227 * check some assumptions
2229 if (root
->nodesize
!= root
->leafsize
) {
2231 "btrfs_scrub: size assumption nodesize == leafsize (%d == %d) fails\n",
2232 root
->nodesize
, root
->leafsize
);
2236 if (root
->nodesize
> BTRFS_STRIPE_LEN
) {
2238 * in this case scrub is unable to calculate the checksum
2239 * the way scrub is implemented. Do not handle this
2240 * situation at all because it won't ever happen.
2243 "btrfs_scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails\n",
2244 root
->nodesize
, BTRFS_STRIPE_LEN
);
2248 if (root
->sectorsize
!= PAGE_SIZE
) {
2249 /* not supported for data w/o checksums */
2251 "btrfs_scrub: size assumption sectorsize != PAGE_SIZE (%d != %lld) fails\n",
2252 root
->sectorsize
, (unsigned long long)PAGE_SIZE
);
2256 ret
= scrub_workers_get(root
);
2260 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2261 dev
= btrfs_find_device(root
, devid
, NULL
, NULL
);
2262 if (!dev
|| dev
->missing
) {
2263 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2264 scrub_workers_put(root
);
2267 mutex_lock(&fs_info
->scrub_lock
);
2269 if (!dev
->in_fs_metadata
) {
2270 mutex_unlock(&fs_info
->scrub_lock
);
2271 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2272 scrub_workers_put(root
);
2276 if (dev
->scrub_device
) {
2277 mutex_unlock(&fs_info
->scrub_lock
);
2278 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2279 scrub_workers_put(root
);
2280 return -EINPROGRESS
;
2282 sdev
= scrub_setup_dev(dev
);
2284 mutex_unlock(&fs_info
->scrub_lock
);
2285 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2286 scrub_workers_put(root
);
2287 return PTR_ERR(sdev
);
2289 sdev
->readonly
= readonly
;
2290 dev
->scrub_device
= sdev
;
2292 atomic_inc(&fs_info
->scrubs_running
);
2293 mutex_unlock(&fs_info
->scrub_lock
);
2294 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2296 down_read(&fs_info
->scrub_super_lock
);
2297 ret
= scrub_supers(sdev
);
2298 up_read(&fs_info
->scrub_super_lock
);
2301 ret
= scrub_enumerate_chunks(sdev
, start
, end
);
2303 wait_event(sdev
->list_wait
, atomic_read(&sdev
->in_flight
) == 0);
2304 atomic_dec(&fs_info
->scrubs_running
);
2305 wake_up(&fs_info
->scrub_pause_wait
);
2307 wait_event(sdev
->list_wait
, atomic_read(&sdev
->fixup_cnt
) == 0);
2310 memcpy(progress
, &sdev
->stat
, sizeof(*progress
));
2312 mutex_lock(&fs_info
->scrub_lock
);
2313 dev
->scrub_device
= NULL
;
2314 mutex_unlock(&fs_info
->scrub_lock
);
2316 scrub_free_dev(sdev
);
2317 scrub_workers_put(root
);
2322 void btrfs_scrub_pause(struct btrfs_root
*root
)
2324 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2326 mutex_lock(&fs_info
->scrub_lock
);
2327 atomic_inc(&fs_info
->scrub_pause_req
);
2328 while (atomic_read(&fs_info
->scrubs_paused
) !=
2329 atomic_read(&fs_info
->scrubs_running
)) {
2330 mutex_unlock(&fs_info
->scrub_lock
);
2331 wait_event(fs_info
->scrub_pause_wait
,
2332 atomic_read(&fs_info
->scrubs_paused
) ==
2333 atomic_read(&fs_info
->scrubs_running
));
2334 mutex_lock(&fs_info
->scrub_lock
);
2336 mutex_unlock(&fs_info
->scrub_lock
);
2339 void btrfs_scrub_continue(struct btrfs_root
*root
)
2341 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2343 atomic_dec(&fs_info
->scrub_pause_req
);
2344 wake_up(&fs_info
->scrub_pause_wait
);
2347 void btrfs_scrub_pause_super(struct btrfs_root
*root
)
2349 down_write(&root
->fs_info
->scrub_super_lock
);
2352 void btrfs_scrub_continue_super(struct btrfs_root
*root
)
2354 up_write(&root
->fs_info
->scrub_super_lock
);
2357 int __btrfs_scrub_cancel(struct btrfs_fs_info
*fs_info
)
2360 mutex_lock(&fs_info
->scrub_lock
);
2361 if (!atomic_read(&fs_info
->scrubs_running
)) {
2362 mutex_unlock(&fs_info
->scrub_lock
);
2366 atomic_inc(&fs_info
->scrub_cancel_req
);
2367 while (atomic_read(&fs_info
->scrubs_running
)) {
2368 mutex_unlock(&fs_info
->scrub_lock
);
2369 wait_event(fs_info
->scrub_pause_wait
,
2370 atomic_read(&fs_info
->scrubs_running
) == 0);
2371 mutex_lock(&fs_info
->scrub_lock
);
2373 atomic_dec(&fs_info
->scrub_cancel_req
);
2374 mutex_unlock(&fs_info
->scrub_lock
);
2379 int btrfs_scrub_cancel(struct btrfs_root
*root
)
2381 return __btrfs_scrub_cancel(root
->fs_info
);
2384 int btrfs_scrub_cancel_dev(struct btrfs_root
*root
, struct btrfs_device
*dev
)
2386 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2387 struct scrub_dev
*sdev
;
2389 mutex_lock(&fs_info
->scrub_lock
);
2390 sdev
= dev
->scrub_device
;
2392 mutex_unlock(&fs_info
->scrub_lock
);
2395 atomic_inc(&sdev
->cancel_req
);
2396 while (dev
->scrub_device
) {
2397 mutex_unlock(&fs_info
->scrub_lock
);
2398 wait_event(fs_info
->scrub_pause_wait
,
2399 dev
->scrub_device
== NULL
);
2400 mutex_lock(&fs_info
->scrub_lock
);
2402 mutex_unlock(&fs_info
->scrub_lock
);
2407 int btrfs_scrub_cancel_devid(struct btrfs_root
*root
, u64 devid
)
2409 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2410 struct btrfs_device
*dev
;
2414 * we have to hold the device_list_mutex here so the device
2415 * does not go away in cancel_dev. FIXME: find a better solution
2417 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2418 dev
= btrfs_find_device(root
, devid
, NULL
, NULL
);
2420 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2423 ret
= btrfs_scrub_cancel_dev(root
, dev
);
2424 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2429 int btrfs_scrub_progress(struct btrfs_root
*root
, u64 devid
,
2430 struct btrfs_scrub_progress
*progress
)
2432 struct btrfs_device
*dev
;
2433 struct scrub_dev
*sdev
= NULL
;
2435 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2436 dev
= btrfs_find_device(root
, devid
, NULL
, NULL
);
2438 sdev
= dev
->scrub_device
;
2440 memcpy(progress
, &sdev
->stat
, sizeof(*progress
));
2441 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2443 return dev
? (sdev
? 0 : -ENOTCONN
) : -ENODEV
;