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
45 #define SCRUB_PAGES_PER_BIO 16 /* 64k per bio */
46 #define SCRUB_BIOS_PER_DEV 16 /* 1 MB per device in flight */
49 u64 flags; /* extent flags */
53 u8 csum[BTRFS_CSUM_SIZE];
58 struct scrub_dev *sdev;
63 struct scrub_page spag[SCRUB_PAGES_PER_BIO];
66 struct btrfs_work work;
70 struct scrub_bio *bios[SCRUB_BIOS_PER_DEV];
71 struct btrfs_device *dev;
77 wait_queue_head_t list_wait;
79 struct list_head csum_list;
85 struct btrfs_scrub_progress stat;
89 struct scrub_fixup_nodatasum {
90 struct scrub_dev *sdev;
92 struct btrfs_root *root;
93 struct btrfs_work work;
97 struct scrub_warning {
98 struct btrfs_path *path;
105 struct btrfs_device *dev;
110 static void scrub_bio_end_io(struct bio *bio, int err);
111 static void scrub_checksum(struct btrfs_work *work);
112 static int scrub_checksum_data(struct scrub_dev *sdev,
113 struct scrub_page *spag, void *buffer);
114 static int scrub_checksum_tree_block(struct scrub_dev *sdev,
115 struct scrub_page *spag, u64 logical,
117 static int scrub_checksum_super(struct scrub_bio *sbio, void *buffer);
118 static int scrub_fixup_check(struct scrub_bio *sbio, int ix);
119 static void scrub_fixup_end_io(struct bio *bio, int err);
120 static int scrub_fixup_io(int rw, struct block_device *bdev, sector_t sector,
122 static void scrub_fixup(struct scrub_bio *sbio, int ix);
125 static void scrub_free_csums(struct scrub_dev *sdev)
127 while (!list_empty(&sdev->csum_list)) {
128 struct btrfs_ordered_sum *sum;
129 sum = list_first_entry(&sdev->csum_list,
130 struct btrfs_ordered_sum, list);
131 list_del(&sum->list);
136 static void scrub_free_bio(struct bio *bio)
139 struct page *last_page = NULL;
144 for (i = 0; i < bio->bi_vcnt; ++i) {
145 if (bio->bi_io_vec[i].bv_page == last_page)
147 last_page = bio->bi_io_vec[i].bv_page;
148 __free_page(last_page);
153 static noinline_for_stack void scrub_free_dev(struct scrub_dev *sdev)
160 for (i = 0; i < SCRUB_BIOS_PER_DEV; ++i) {
161 struct scrub_bio *sbio = sdev->bios[i];
166 scrub_free_bio(sbio->bio);
170 scrub_free_csums(sdev);
174 static noinline_for_stack
175 struct scrub_dev *scrub_setup_dev(struct btrfs_device *dev)
177 struct scrub_dev *sdev;
179 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
181 sdev = kzalloc(sizeof(*sdev), GFP_NOFS);
185 for (i = 0; i < SCRUB_BIOS_PER_DEV; ++i) {
186 struct scrub_bio *sbio;
188 sbio = kzalloc(sizeof(*sbio), GFP_NOFS);
191 sdev->bios[i] = sbio;
196 sbio->work.func = scrub_checksum;
198 if (i != SCRUB_BIOS_PER_DEV-1)
199 sdev->bios[i]->next_free = i + 1;
201 sdev->bios[i]->next_free = -1;
203 sdev->first_free = 0;
205 atomic_set(&sdev->in_flight, 0);
206 atomic_set(&sdev->fixup_cnt, 0);
207 atomic_set(&sdev->cancel_req, 0);
208 sdev->csum_size = btrfs_super_csum_size(fs_info->super_copy);
209 INIT_LIST_HEAD(&sdev->csum_list);
211 spin_lock_init(&sdev->list_lock);
212 spin_lock_init(&sdev->stat_lock);
213 init_waitqueue_head(&sdev->list_wait);
217 scrub_free_dev(sdev);
218 return ERR_PTR(-ENOMEM);
221 static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root, void *ctx)
227 struct extent_buffer *eb;
228 struct btrfs_inode_item *inode_item;
229 struct scrub_warning *swarn = ctx;
230 struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info;
231 struct inode_fs_paths *ipath = NULL;
232 struct btrfs_root *local_root;
233 struct btrfs_key root_key;
235 root_key.objectid = root;
236 root_key.type = BTRFS_ROOT_ITEM_KEY;
237 root_key.offset = (u64)-1;
238 local_root = btrfs_read_fs_root_no_name(fs_info, &root_key);
239 if (IS_ERR(local_root)) {
240 ret = PTR_ERR(local_root);
244 ret = inode_item_info(inum, 0, local_root, swarn->path);
246 btrfs_release_path(swarn->path);
250 eb = swarn->path->nodes[0];
251 inode_item = btrfs_item_ptr(eb, swarn->path->slots[0],
252 struct btrfs_inode_item);
253 isize = btrfs_inode_size(eb, inode_item);
254 nlink = btrfs_inode_nlink(eb, inode_item);
255 btrfs_release_path(swarn->path);
257 ipath = init_ipath(4096, local_root, swarn->path);
259 ret = PTR_ERR(ipath);
263 ret = paths_from_inode(inum, ipath);
269 * we deliberately ignore the bit ipath might have been too small to
270 * hold all of the paths here
272 for (i = 0; i < ipath->fspath->elem_cnt; ++i)
273 printk(KERN_WARNING "btrfs: %s at logical %llu on dev "
274 "%s, sector %llu, root %llu, inode %llu, offset %llu, "
275 "length %llu, links %u (path: %s)\n", swarn->errstr,
276 swarn->logical, swarn->dev->name,
277 (unsigned long long)swarn->sector, root, inum, offset,
278 min(isize - offset, (u64)PAGE_SIZE), nlink,
279 (char *)(unsigned long)ipath->fspath->val[i]);
285 printk(KERN_WARNING "btrfs: %s at logical %llu on dev "
286 "%s, sector %llu, root %llu, inode %llu, offset %llu: path "
287 "resolving failed with ret=%d\n", swarn->errstr,
288 swarn->logical, swarn->dev->name,
289 (unsigned long long)swarn->sector, root, inum, offset, ret);
295 static void scrub_print_warning(const char *errstr, struct scrub_bio *sbio,
298 struct btrfs_device *dev = sbio->sdev->dev;
299 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
300 struct btrfs_path *path;
301 struct btrfs_key found_key;
302 struct extent_buffer *eb;
303 struct btrfs_extent_item *ei;
304 struct scrub_warning swarn;
309 unsigned long ptr = 0;
310 const int bufsize = 4096;
313 path = btrfs_alloc_path();
315 swarn.scratch_buf = kmalloc(bufsize, GFP_NOFS);
316 swarn.msg_buf = kmalloc(bufsize, GFP_NOFS);
317 swarn.sector = (sbio->physical + ix * PAGE_SIZE) >> 9;
318 swarn.logical = sbio->logical + ix * PAGE_SIZE;
319 swarn.errstr = errstr;
321 swarn.msg_bufsize = bufsize;
322 swarn.scratch_bufsize = bufsize;
324 if (!path || !swarn.scratch_buf || !swarn.msg_buf)
327 ret = extent_from_logical(fs_info, swarn.logical, path, &found_key);
331 extent_item_pos = swarn.logical - found_key.objectid;
332 swarn.extent_item_size = found_key.offset;
335 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
336 item_size = btrfs_item_size_nr(eb, path->slots[0]);
337 btrfs_release_path(path);
339 if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
341 ret = tree_backref_for_extent(&ptr, eb, ei, item_size,
342 &ref_root, &ref_level);
344 "btrfs: %s at logical %llu on dev %s, "
345 "sector %llu: metadata %s (level %d) in tree "
346 "%llu\n", errstr, swarn.logical, dev->name,
347 (unsigned long long)swarn.sector,
348 ref_level ? "node" : "leaf",
349 ret < 0 ? -1 : ref_level,
350 ret < 0 ? -1 : ref_root);
354 iterate_extent_inodes(fs_info, path, found_key.objectid,
356 scrub_print_warning_inode, &swarn);
360 btrfs_free_path(path);
361 kfree(swarn.scratch_buf);
362 kfree(swarn.msg_buf);
365 static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *ctx)
367 struct page *page = NULL;
369 struct scrub_fixup_nodatasum *fixup = ctx;
372 struct btrfs_key key;
373 struct inode *inode = NULL;
374 u64 end = offset + PAGE_SIZE - 1;
375 struct btrfs_root *local_root;
378 key.type = BTRFS_ROOT_ITEM_KEY;
379 key.offset = (u64)-1;
380 local_root = btrfs_read_fs_root_no_name(fixup->root->fs_info, &key);
381 if (IS_ERR(local_root))
382 return PTR_ERR(local_root);
384 key.type = BTRFS_INODE_ITEM_KEY;
387 inode = btrfs_iget(fixup->root->fs_info->sb, &key, local_root, NULL);
389 return PTR_ERR(inode);
391 index = offset >> PAGE_CACHE_SHIFT;
393 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
399 if (PageUptodate(page)) {
400 struct btrfs_mapping_tree *map_tree;
401 if (PageDirty(page)) {
403 * we need to write the data to the defect sector. the
404 * data that was in that sector is not in memory,
405 * because the page was modified. we must not write the
406 * modified page to that sector.
408 * TODO: what could be done here: wait for the delalloc
409 * runner to write out that page (might involve
410 * COW) and see whether the sector is still
411 * referenced afterwards.
413 * For the meantime, we'll treat this error
414 * incorrectable, although there is a chance that a
415 * later scrub will find the bad sector again and that
416 * there's no dirty page in memory, then.
421 map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree;
422 ret = repair_io_failure(map_tree, offset, PAGE_SIZE,
423 fixup->logical, page,
429 * we need to get good data first. the general readpage path
430 * will call repair_io_failure for us, we just have to make
431 * sure we read the bad mirror.
433 ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
434 EXTENT_DAMAGED, GFP_NOFS);
436 /* set_extent_bits should give proper error */
443 ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page,
446 wait_on_page_locked(page);
448 corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset,
449 end, EXTENT_DAMAGED, 0, NULL);
451 clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
452 EXTENT_DAMAGED, GFP_NOFS);
464 if (ret == 0 && corrected) {
466 * we only need to call readpage for one of the inodes belonging
467 * to this extent. so make iterate_extent_inodes stop
475 static void scrub_fixup_nodatasum(struct btrfs_work *work)
478 struct scrub_fixup_nodatasum *fixup;
479 struct scrub_dev *sdev;
480 struct btrfs_trans_handle *trans = NULL;
481 struct btrfs_fs_info *fs_info;
482 struct btrfs_path *path;
483 int uncorrectable = 0;
485 fixup = container_of(work, struct scrub_fixup_nodatasum, work);
487 fs_info = fixup->root->fs_info;
489 path = btrfs_alloc_path();
491 spin_lock(&sdev->stat_lock);
492 ++sdev->stat.malloc_errors;
493 spin_unlock(&sdev->stat_lock);
498 trans = btrfs_join_transaction(fixup->root);
505 * the idea is to trigger a regular read through the standard path. we
506 * read a page from the (failed) logical address by specifying the
507 * corresponding copynum of the failed sector. thus, that readpage is
509 * that is the point where on-the-fly error correction will kick in
510 * (once it's finished) and rewrite the failed sector if a good copy
513 ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info,
514 path, scrub_fixup_readpage,
522 spin_lock(&sdev->stat_lock);
523 ++sdev->stat.corrected_errors;
524 spin_unlock(&sdev->stat_lock);
527 if (trans && !IS_ERR(trans))
528 btrfs_end_transaction(trans, fixup->root);
530 spin_lock(&sdev->stat_lock);
531 ++sdev->stat.uncorrectable_errors;
532 spin_unlock(&sdev->stat_lock);
533 printk_ratelimited(KERN_ERR "btrfs: unable to fixup "
534 "(nodatasum) error at logical %llu\n",
538 btrfs_free_path(path);
541 /* see caller why we're pretending to be paused in the scrub counters */
542 mutex_lock(&fs_info->scrub_lock);
543 atomic_dec(&fs_info->scrubs_running);
544 atomic_dec(&fs_info->scrubs_paused);
545 mutex_unlock(&fs_info->scrub_lock);
546 atomic_dec(&sdev->fixup_cnt);
547 wake_up(&fs_info->scrub_pause_wait);
548 wake_up(&sdev->list_wait);
552 * scrub_recheck_error gets called when either verification of the page
553 * failed or the bio failed to read, e.g. with EIO. In the latter case,
554 * recheck_error gets called for every page in the bio, even though only
557 static int scrub_recheck_error(struct scrub_bio *sbio, int ix)
559 struct scrub_dev *sdev = sbio->sdev;
560 u64 sector = (sbio->physical + ix * PAGE_SIZE) >> 9;
561 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
562 DEFAULT_RATELIMIT_BURST);
565 if (scrub_fixup_io(READ, sbio->sdev->dev->bdev, sector,
566 sbio->bio->bi_io_vec[ix].bv_page) == 0) {
567 if (scrub_fixup_check(sbio, ix) == 0)
570 if (__ratelimit(&_rs))
571 scrub_print_warning("i/o error", sbio, ix);
573 if (__ratelimit(&_rs))
574 scrub_print_warning("checksum error", sbio, ix);
577 spin_lock(&sdev->stat_lock);
578 ++sdev->stat.read_errors;
579 spin_unlock(&sdev->stat_lock);
581 scrub_fixup(sbio, ix);
585 static int scrub_fixup_check(struct scrub_bio *sbio, int ix)
590 u64 flags = sbio->spag[ix].flags;
592 page = sbio->bio->bi_io_vec[ix].bv_page;
593 buffer = kmap_atomic(page, KM_USER0);
594 if (flags & BTRFS_EXTENT_FLAG_DATA) {
595 ret = scrub_checksum_data(sbio->sdev,
596 sbio->spag + ix, buffer);
597 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
598 ret = scrub_checksum_tree_block(sbio->sdev,
600 sbio->logical + ix * PAGE_SIZE,
605 kunmap_atomic(buffer, KM_USER0);
610 static void scrub_fixup_end_io(struct bio *bio, int err)
612 complete((struct completion *)bio->bi_private);
615 static void scrub_fixup(struct scrub_bio *sbio, int ix)
617 struct scrub_dev *sdev = sbio->sdev;
618 struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info;
619 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
620 struct btrfs_bio *bbio = NULL;
621 struct scrub_fixup_nodatasum *fixup;
622 u64 logical = sbio->logical + ix * PAGE_SIZE;
626 DECLARE_COMPLETION_ONSTACK(complete);
628 if ((sbio->spag[ix].flags & BTRFS_EXTENT_FLAG_DATA) &&
629 (sbio->spag[ix].have_csum == 0)) {
630 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
634 fixup->logical = logical;
635 fixup->root = fs_info->extent_root;
636 fixup->mirror_num = sbio->spag[ix].mirror_num;
638 * increment scrubs_running to prevent cancel requests from
639 * completing as long as a fixup worker is running. we must also
640 * increment scrubs_paused to prevent deadlocking on pause
641 * requests used for transactions commits (as the worker uses a
642 * transaction context). it is safe to regard the fixup worker
643 * as paused for all matters practical. effectively, we only
644 * avoid cancellation requests from completing.
646 mutex_lock(&fs_info->scrub_lock);
647 atomic_inc(&fs_info->scrubs_running);
648 atomic_inc(&fs_info->scrubs_paused);
649 mutex_unlock(&fs_info->scrub_lock);
650 atomic_inc(&sdev->fixup_cnt);
651 fixup->work.func = scrub_fixup_nodatasum;
652 btrfs_queue_worker(&fs_info->scrub_workers, &fixup->work);
657 ret = btrfs_map_block(map_tree, REQ_WRITE, logical, &length,
659 if (ret || !bbio || length < PAGE_SIZE) {
661 "scrub_fixup: btrfs_map_block failed us for %llu\n",
662 (unsigned long long)logical);
668 if (bbio->num_stripes == 1)
669 /* there aren't any replicas */
673 * first find a good copy
675 for (i = 0; i < bbio->num_stripes; ++i) {
676 if (i + 1 == sbio->spag[ix].mirror_num)
679 if (scrub_fixup_io(READ, bbio->stripes[i].dev->bdev,
680 bbio->stripes[i].physical >> 9,
681 sbio->bio->bi_io_vec[ix].bv_page)) {
682 /* I/O-error, this is not a good copy */
686 if (scrub_fixup_check(sbio, ix) == 0)
689 if (i == bbio->num_stripes)
692 if (!sdev->readonly) {
694 * bi_io_vec[ix].bv_page now contains good data, write it back
696 if (scrub_fixup_io(WRITE, sdev->dev->bdev,
697 (sbio->physical + ix * PAGE_SIZE) >> 9,
698 sbio->bio->bi_io_vec[ix].bv_page)) {
699 /* I/O-error, writeback failed, give up */
705 spin_lock(&sdev->stat_lock);
706 ++sdev->stat.corrected_errors;
707 spin_unlock(&sdev->stat_lock);
709 printk_ratelimited(KERN_ERR "btrfs: fixed up error at logical %llu\n",
710 (unsigned long long)logical);
715 spin_lock(&sdev->stat_lock);
716 ++sdev->stat.uncorrectable_errors;
717 spin_unlock(&sdev->stat_lock);
719 printk_ratelimited(KERN_ERR "btrfs: unable to fixup (regular) error at "
720 "logical %llu\n", (unsigned long long)logical);
723 static int scrub_fixup_io(int rw, struct block_device *bdev, sector_t sector,
726 struct bio *bio = NULL;
728 DECLARE_COMPLETION_ONSTACK(complete);
730 bio = bio_alloc(GFP_NOFS, 1);
732 bio->bi_sector = sector;
733 bio_add_page(bio, page, PAGE_SIZE, 0);
734 bio->bi_end_io = scrub_fixup_end_io;
735 bio->bi_private = &complete;
736 btrfsic_submit_bio(rw, bio);
738 /* this will also unplug the queue */
739 wait_for_completion(&complete);
741 ret = !test_bit(BIO_UPTODATE, &bio->bi_flags);
746 static void scrub_bio_end_io(struct bio *bio, int err)
748 struct scrub_bio *sbio = bio->bi_private;
749 struct scrub_dev *sdev = sbio->sdev;
750 struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info;
755 btrfs_queue_worker(&fs_info->scrub_workers, &sbio->work);
758 static void scrub_checksum(struct btrfs_work *work)
760 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
761 struct scrub_dev *sdev = sbio->sdev;
771 for (i = 0; i < sbio->count; ++i)
772 ret |= scrub_recheck_error(sbio, i);
774 spin_lock(&sdev->stat_lock);
775 ++sdev->stat.unverified_errors;
776 spin_unlock(&sdev->stat_lock);
779 sbio->bio->bi_flags &= ~(BIO_POOL_MASK - 1);
780 sbio->bio->bi_flags |= 1 << BIO_UPTODATE;
781 sbio->bio->bi_phys_segments = 0;
782 sbio->bio->bi_idx = 0;
784 for (i = 0; i < sbio->count; i++) {
786 bi = &sbio->bio->bi_io_vec[i];
788 bi->bv_len = PAGE_SIZE;
792 for (i = 0; i < sbio->count; ++i) {
793 page = sbio->bio->bi_io_vec[i].bv_page;
794 buffer = kmap_atomic(page, KM_USER0);
795 flags = sbio->spag[i].flags;
796 logical = sbio->logical + i * PAGE_SIZE;
798 if (flags & BTRFS_EXTENT_FLAG_DATA) {
799 ret = scrub_checksum_data(sdev, sbio->spag + i, buffer);
800 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
801 ret = scrub_checksum_tree_block(sdev, sbio->spag + i,
803 } else if (flags & BTRFS_EXTENT_FLAG_SUPER) {
805 (void)scrub_checksum_super(sbio, buffer);
809 kunmap_atomic(buffer, KM_USER0);
811 ret = scrub_recheck_error(sbio, i);
813 spin_lock(&sdev->stat_lock);
814 ++sdev->stat.unverified_errors;
815 spin_unlock(&sdev->stat_lock);
821 scrub_free_bio(sbio->bio);
823 spin_lock(&sdev->list_lock);
824 sbio->next_free = sdev->first_free;
825 sdev->first_free = sbio->index;
826 spin_unlock(&sdev->list_lock);
827 atomic_dec(&sdev->in_flight);
828 wake_up(&sdev->list_wait);
831 static int scrub_checksum_data(struct scrub_dev *sdev,
832 struct scrub_page *spag, void *buffer)
834 u8 csum[BTRFS_CSUM_SIZE];
837 struct btrfs_root *root = sdev->dev->dev_root;
839 if (!spag->have_csum)
842 crc = btrfs_csum_data(root, buffer, crc, PAGE_SIZE);
843 btrfs_csum_final(crc, csum);
844 if (memcmp(csum, spag->csum, sdev->csum_size))
847 spin_lock(&sdev->stat_lock);
848 ++sdev->stat.data_extents_scrubbed;
849 sdev->stat.data_bytes_scrubbed += PAGE_SIZE;
851 ++sdev->stat.csum_errors;
852 spin_unlock(&sdev->stat_lock);
857 static int scrub_checksum_tree_block(struct scrub_dev *sdev,
858 struct scrub_page *spag, u64 logical,
861 struct btrfs_header *h;
862 struct btrfs_root *root = sdev->dev->dev_root;
863 struct btrfs_fs_info *fs_info = root->fs_info;
864 u8 csum[BTRFS_CSUM_SIZE];
870 * we don't use the getter functions here, as we
871 * a) don't have an extent buffer and
872 * b) the page is already kmapped
874 h = (struct btrfs_header *)buffer;
876 if (logical != le64_to_cpu(h->bytenr))
879 if (spag->generation != le64_to_cpu(h->generation))
882 if (memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
885 if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
889 crc = btrfs_csum_data(root, buffer + BTRFS_CSUM_SIZE, crc,
890 PAGE_SIZE - BTRFS_CSUM_SIZE);
891 btrfs_csum_final(crc, csum);
892 if (memcmp(csum, h->csum, sdev->csum_size))
895 spin_lock(&sdev->stat_lock);
896 ++sdev->stat.tree_extents_scrubbed;
897 sdev->stat.tree_bytes_scrubbed += PAGE_SIZE;
899 ++sdev->stat.csum_errors;
901 ++sdev->stat.verify_errors;
902 spin_unlock(&sdev->stat_lock);
904 return fail || crc_fail;
907 static int scrub_checksum_super(struct scrub_bio *sbio, void *buffer)
909 struct btrfs_super_block *s;
911 struct scrub_dev *sdev = sbio->sdev;
912 struct btrfs_root *root = sdev->dev->dev_root;
913 struct btrfs_fs_info *fs_info = root->fs_info;
914 u8 csum[BTRFS_CSUM_SIZE];
918 s = (struct btrfs_super_block *)buffer;
919 logical = sbio->logical;
921 if (logical != le64_to_cpu(s->bytenr))
924 if (sbio->spag[0].generation != le64_to_cpu(s->generation))
927 if (memcmp(s->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
930 crc = btrfs_csum_data(root, buffer + BTRFS_CSUM_SIZE, crc,
931 PAGE_SIZE - BTRFS_CSUM_SIZE);
932 btrfs_csum_final(crc, csum);
933 if (memcmp(csum, s->csum, sbio->sdev->csum_size))
938 * if we find an error in a super block, we just report it.
939 * They will get written with the next transaction commit
942 spin_lock(&sdev->stat_lock);
943 ++sdev->stat.super_errors;
944 spin_unlock(&sdev->stat_lock);
950 static void scrub_submit(struct scrub_dev *sdev)
952 struct scrub_bio *sbio;
954 if (sdev->curr == -1)
957 sbio = sdev->bios[sdev->curr];
960 atomic_inc(&sdev->in_flight);
962 btrfsic_submit_bio(READ, sbio->bio);
965 static int scrub_page(struct scrub_dev *sdev, u64 logical, u64 len,
966 u64 physical, u64 flags, u64 gen, int mirror_num,
969 struct scrub_bio *sbio;
975 * grab a fresh bio or wait for one to become available
977 while (sdev->curr == -1) {
978 spin_lock(&sdev->list_lock);
979 sdev->curr = sdev->first_free;
980 if (sdev->curr != -1) {
981 sdev->first_free = sdev->bios[sdev->curr]->next_free;
982 sdev->bios[sdev->curr]->next_free = -1;
983 sdev->bios[sdev->curr]->count = 0;
984 spin_unlock(&sdev->list_lock);
986 spin_unlock(&sdev->list_lock);
987 wait_event(sdev->list_wait, sdev->first_free != -1);
990 sbio = sdev->bios[sdev->curr];
991 if (sbio->count == 0) {
994 sbio->physical = physical;
995 sbio->logical = logical;
996 bio = bio_alloc(GFP_NOFS, SCRUB_PAGES_PER_BIO);
1000 bio->bi_private = sbio;
1001 bio->bi_end_io = scrub_bio_end_io;
1002 bio->bi_bdev = sdev->dev->bdev;
1003 bio->bi_sector = sbio->physical >> 9;
1006 } else if (sbio->physical + sbio->count * PAGE_SIZE != physical ||
1007 sbio->logical + sbio->count * PAGE_SIZE != logical) {
1011 sbio->spag[sbio->count].flags = flags;
1012 sbio->spag[sbio->count].generation = gen;
1013 sbio->spag[sbio->count].have_csum = 0;
1014 sbio->spag[sbio->count].mirror_num = mirror_num;
1016 page = alloc_page(GFP_NOFS);
1020 ret = bio_add_page(sbio->bio, page, PAGE_SIZE, 0);
1028 sbio->spag[sbio->count].have_csum = 1;
1029 memcpy(sbio->spag[sbio->count].csum, csum, sdev->csum_size);
1032 if (sbio->count == SCRUB_PAGES_PER_BIO || force)
1038 static int scrub_find_csum(struct scrub_dev *sdev, u64 logical, u64 len,
1041 struct btrfs_ordered_sum *sum = NULL;
1044 unsigned long num_sectors;
1045 u32 sectorsize = sdev->dev->dev_root->sectorsize;
1047 while (!list_empty(&sdev->csum_list)) {
1048 sum = list_first_entry(&sdev->csum_list,
1049 struct btrfs_ordered_sum, list);
1050 if (sum->bytenr > logical)
1052 if (sum->bytenr + sum->len > logical)
1055 ++sdev->stat.csum_discards;
1056 list_del(&sum->list);
1063 num_sectors = sum->len / sectorsize;
1064 for (i = 0; i < num_sectors; ++i) {
1065 if (sum->sums[i].bytenr == logical) {
1066 memcpy(csum, &sum->sums[i].sum, sdev->csum_size);
1071 if (ret && i == num_sectors - 1) {
1072 list_del(&sum->list);
1078 /* scrub extent tries to collect up to 64 kB for each bio */
1079 static int scrub_extent(struct scrub_dev *sdev, u64 logical, u64 len,
1080 u64 physical, u64 flags, u64 gen, int mirror_num)
1083 u8 csum[BTRFS_CSUM_SIZE];
1086 u64 l = min_t(u64, len, PAGE_SIZE);
1089 if (flags & BTRFS_EXTENT_FLAG_DATA) {
1090 /* push csums to sbio */
1091 have_csum = scrub_find_csum(sdev, logical, l, csum);
1093 ++sdev->stat.no_csum;
1095 ret = scrub_page(sdev, logical, l, physical, flags, gen,
1096 mirror_num, have_csum ? csum : NULL, 0);
1106 static noinline_for_stack int scrub_stripe(struct scrub_dev *sdev,
1107 struct map_lookup *map, int num, u64 base, u64 length)
1109 struct btrfs_path *path;
1110 struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info;
1111 struct btrfs_root *root = fs_info->extent_root;
1112 struct btrfs_root *csum_root = fs_info->csum_root;
1113 struct btrfs_extent_item *extent;
1114 struct blk_plug plug;
1120 struct extent_buffer *l;
1121 struct btrfs_key key;
1126 struct reada_control *reada1;
1127 struct reada_control *reada2;
1128 struct btrfs_key key_start;
1129 struct btrfs_key key_end;
1131 u64 increment = map->stripe_len;
1136 do_div(nstripes, map->stripe_len);
1137 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1138 offset = map->stripe_len * num;
1139 increment = map->stripe_len * map->num_stripes;
1141 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1142 int factor = map->num_stripes / map->sub_stripes;
1143 offset = map->stripe_len * (num / map->sub_stripes);
1144 increment = map->stripe_len * factor;
1145 mirror_num = num % map->sub_stripes + 1;
1146 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1147 increment = map->stripe_len;
1148 mirror_num = num % map->num_stripes + 1;
1149 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1150 increment = map->stripe_len;
1151 mirror_num = num % map->num_stripes + 1;
1153 increment = map->stripe_len;
1157 path = btrfs_alloc_path();
1161 path->search_commit_root = 1;
1162 path->skip_locking = 1;
1165 * trigger the readahead for extent tree csum tree and wait for
1166 * completion. During readahead, the scrub is officially paused
1167 * to not hold off transaction commits
1169 logical = base + offset;
1171 wait_event(sdev->list_wait,
1172 atomic_read(&sdev->in_flight) == 0);
1173 atomic_inc(&fs_info->scrubs_paused);
1174 wake_up(&fs_info->scrub_pause_wait);
1176 /* FIXME it might be better to start readahead at commit root */
1177 key_start.objectid = logical;
1178 key_start.type = BTRFS_EXTENT_ITEM_KEY;
1179 key_start.offset = (u64)0;
1180 key_end.objectid = base + offset + nstripes * increment;
1181 key_end.type = BTRFS_EXTENT_ITEM_KEY;
1182 key_end.offset = (u64)0;
1183 reada1 = btrfs_reada_add(root, &key_start, &key_end);
1185 key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1186 key_start.type = BTRFS_EXTENT_CSUM_KEY;
1187 key_start.offset = logical;
1188 key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1189 key_end.type = BTRFS_EXTENT_CSUM_KEY;
1190 key_end.offset = base + offset + nstripes * increment;
1191 reada2 = btrfs_reada_add(csum_root, &key_start, &key_end);
1193 if (!IS_ERR(reada1))
1194 btrfs_reada_wait(reada1);
1195 if (!IS_ERR(reada2))
1196 btrfs_reada_wait(reada2);
1198 mutex_lock(&fs_info->scrub_lock);
1199 while (atomic_read(&fs_info->scrub_pause_req)) {
1200 mutex_unlock(&fs_info->scrub_lock);
1201 wait_event(fs_info->scrub_pause_wait,
1202 atomic_read(&fs_info->scrub_pause_req) == 0);
1203 mutex_lock(&fs_info->scrub_lock);
1205 atomic_dec(&fs_info->scrubs_paused);
1206 mutex_unlock(&fs_info->scrub_lock);
1207 wake_up(&fs_info->scrub_pause_wait);
1210 * collect all data csums for the stripe to avoid seeking during
1211 * the scrub. This might currently (crc32) end up to be about 1MB
1213 blk_start_plug(&plug);
1216 * now find all extents for each stripe and scrub them
1218 logical = base + offset;
1219 physical = map->stripes[num].physical;
1221 for (i = 0; i < nstripes; ++i) {
1225 if (atomic_read(&fs_info->scrub_cancel_req) ||
1226 atomic_read(&sdev->cancel_req)) {
1231 * check to see if we have to pause
1233 if (atomic_read(&fs_info->scrub_pause_req)) {
1234 /* push queued extents */
1236 wait_event(sdev->list_wait,
1237 atomic_read(&sdev->in_flight) == 0);
1238 atomic_inc(&fs_info->scrubs_paused);
1239 wake_up(&fs_info->scrub_pause_wait);
1240 mutex_lock(&fs_info->scrub_lock);
1241 while (atomic_read(&fs_info->scrub_pause_req)) {
1242 mutex_unlock(&fs_info->scrub_lock);
1243 wait_event(fs_info->scrub_pause_wait,
1244 atomic_read(&fs_info->scrub_pause_req) == 0);
1245 mutex_lock(&fs_info->scrub_lock);
1247 atomic_dec(&fs_info->scrubs_paused);
1248 mutex_unlock(&fs_info->scrub_lock);
1249 wake_up(&fs_info->scrub_pause_wait);
1252 ret = btrfs_lookup_csums_range(csum_root, logical,
1253 logical + map->stripe_len - 1,
1254 &sdev->csum_list, 1);
1258 key.objectid = logical;
1259 key.type = BTRFS_EXTENT_ITEM_KEY;
1260 key.offset = (u64)0;
1262 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1266 ret = btrfs_previous_item(root, path, 0,
1267 BTRFS_EXTENT_ITEM_KEY);
1271 /* there's no smaller item, so stick with the
1273 btrfs_release_path(path);
1274 ret = btrfs_search_slot(NULL, root, &key,
1283 slot = path->slots[0];
1284 if (slot >= btrfs_header_nritems(l)) {
1285 ret = btrfs_next_leaf(root, path);
1293 btrfs_item_key_to_cpu(l, &key, slot);
1295 if (key.objectid + key.offset <= logical)
1298 if (key.objectid >= logical + map->stripe_len)
1301 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY)
1304 extent = btrfs_item_ptr(l, slot,
1305 struct btrfs_extent_item);
1306 flags = btrfs_extent_flags(l, extent);
1307 generation = btrfs_extent_generation(l, extent);
1309 if (key.objectid < logical &&
1310 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
1312 "btrfs scrub: tree block %llu spanning "
1313 "stripes, ignored. logical=%llu\n",
1314 (unsigned long long)key.objectid,
1315 (unsigned long long)logical);
1320 * trim extent to this stripe
1322 if (key.objectid < logical) {
1323 key.offset -= logical - key.objectid;
1324 key.objectid = logical;
1326 if (key.objectid + key.offset >
1327 logical + map->stripe_len) {
1328 key.offset = logical + map->stripe_len -
1332 ret = scrub_extent(sdev, key.objectid, key.offset,
1333 key.objectid - logical + physical,
1334 flags, generation, mirror_num);
1341 btrfs_release_path(path);
1342 logical += increment;
1343 physical += map->stripe_len;
1344 spin_lock(&sdev->stat_lock);
1345 sdev->stat.last_physical = physical;
1346 spin_unlock(&sdev->stat_lock);
1348 /* push queued extents */
1352 blk_finish_plug(&plug);
1353 btrfs_free_path(path);
1354 return ret < 0 ? ret : 0;
1357 static noinline_for_stack int scrub_chunk(struct scrub_dev *sdev,
1358 u64 chunk_tree, u64 chunk_objectid, u64 chunk_offset, u64 length,
1361 struct btrfs_mapping_tree *map_tree =
1362 &sdev->dev->dev_root->fs_info->mapping_tree;
1363 struct map_lookup *map;
1364 struct extent_map *em;
1368 read_lock(&map_tree->map_tree.lock);
1369 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
1370 read_unlock(&map_tree->map_tree.lock);
1375 map = (struct map_lookup *)em->bdev;
1376 if (em->start != chunk_offset)
1379 if (em->len < length)
1382 for (i = 0; i < map->num_stripes; ++i) {
1383 if (map->stripes[i].dev == sdev->dev &&
1384 map->stripes[i].physical == dev_offset) {
1385 ret = scrub_stripe(sdev, map, i, chunk_offset, length);
1391 free_extent_map(em);
1396 static noinline_for_stack
1397 int scrub_enumerate_chunks(struct scrub_dev *sdev, u64 start, u64 end)
1399 struct btrfs_dev_extent *dev_extent = NULL;
1400 struct btrfs_path *path;
1401 struct btrfs_root *root = sdev->dev->dev_root;
1402 struct btrfs_fs_info *fs_info = root->fs_info;
1409 struct extent_buffer *l;
1410 struct btrfs_key key;
1411 struct btrfs_key found_key;
1412 struct btrfs_block_group_cache *cache;
1414 path = btrfs_alloc_path();
1419 path->search_commit_root = 1;
1420 path->skip_locking = 1;
1422 key.objectid = sdev->dev->devid;
1424 key.type = BTRFS_DEV_EXTENT_KEY;
1428 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1432 if (path->slots[0] >=
1433 btrfs_header_nritems(path->nodes[0])) {
1434 ret = btrfs_next_leaf(root, path);
1441 slot = path->slots[0];
1443 btrfs_item_key_to_cpu(l, &found_key, slot);
1445 if (found_key.objectid != sdev->dev->devid)
1448 if (btrfs_key_type(&found_key) != BTRFS_DEV_EXTENT_KEY)
1451 if (found_key.offset >= end)
1454 if (found_key.offset < key.offset)
1457 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1458 length = btrfs_dev_extent_length(l, dev_extent);
1460 if (found_key.offset + length <= start) {
1461 key.offset = found_key.offset + length;
1462 btrfs_release_path(path);
1466 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1467 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1468 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1471 * get a reference on the corresponding block group to prevent
1472 * the chunk from going away while we scrub it
1474 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
1479 ret = scrub_chunk(sdev, chunk_tree, chunk_objectid,
1480 chunk_offset, length, found_key.offset);
1481 btrfs_put_block_group(cache);
1485 key.offset = found_key.offset + length;
1486 btrfs_release_path(path);
1489 btrfs_free_path(path);
1492 * ret can still be 1 from search_slot or next_leaf,
1493 * that's not an error
1495 return ret < 0 ? ret : 0;
1498 static noinline_for_stack int scrub_supers(struct scrub_dev *sdev)
1504 struct btrfs_device *device = sdev->dev;
1505 struct btrfs_root *root = device->dev_root;
1507 gen = root->fs_info->last_trans_committed;
1509 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1510 bytenr = btrfs_sb_offset(i);
1511 if (bytenr + BTRFS_SUPER_INFO_SIZE > device->total_bytes)
1514 ret = scrub_page(sdev, bytenr, PAGE_SIZE, bytenr,
1515 BTRFS_EXTENT_FLAG_SUPER, gen, i, NULL, 1);
1519 wait_event(sdev->list_wait, atomic_read(&sdev->in_flight) == 0);
1525 * get a reference count on fs_info->scrub_workers. start worker if necessary
1527 static noinline_for_stack int scrub_workers_get(struct btrfs_root *root)
1529 struct btrfs_fs_info *fs_info = root->fs_info;
1532 mutex_lock(&fs_info->scrub_lock);
1533 if (fs_info->scrub_workers_refcnt == 0) {
1534 btrfs_init_workers(&fs_info->scrub_workers, "scrub",
1535 fs_info->thread_pool_size, &fs_info->generic_worker);
1536 fs_info->scrub_workers.idle_thresh = 4;
1537 ret = btrfs_start_workers(&fs_info->scrub_workers);
1541 ++fs_info->scrub_workers_refcnt;
1543 mutex_unlock(&fs_info->scrub_lock);
1548 static noinline_for_stack void scrub_workers_put(struct btrfs_root *root)
1550 struct btrfs_fs_info *fs_info = root->fs_info;
1552 mutex_lock(&fs_info->scrub_lock);
1553 if (--fs_info->scrub_workers_refcnt == 0)
1554 btrfs_stop_workers(&fs_info->scrub_workers);
1555 WARN_ON(fs_info->scrub_workers_refcnt < 0);
1556 mutex_unlock(&fs_info->scrub_lock);
1560 int btrfs_scrub_dev(struct btrfs_root *root, u64 devid, u64 start, u64 end,
1561 struct btrfs_scrub_progress *progress, int readonly)
1563 struct scrub_dev *sdev;
1564 struct btrfs_fs_info *fs_info = root->fs_info;
1566 struct btrfs_device *dev;
1568 if (btrfs_fs_closing(root->fs_info))
1572 * check some assumptions
1574 if (root->sectorsize != PAGE_SIZE ||
1575 root->sectorsize != root->leafsize ||
1576 root->sectorsize != root->nodesize) {
1577 printk(KERN_ERR "btrfs_scrub: size assumptions fail\n");
1581 ret = scrub_workers_get(root);
1585 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1586 dev = btrfs_find_device(root, devid, NULL, NULL);
1587 if (!dev || dev->missing) {
1588 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1589 scrub_workers_put(root);
1592 mutex_lock(&fs_info->scrub_lock);
1594 if (!dev->in_fs_metadata) {
1595 mutex_unlock(&fs_info->scrub_lock);
1596 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1597 scrub_workers_put(root);
1601 if (dev->scrub_device) {
1602 mutex_unlock(&fs_info->scrub_lock);
1603 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1604 scrub_workers_put(root);
1605 return -EINPROGRESS;
1607 sdev = scrub_setup_dev(dev);
1609 mutex_unlock(&fs_info->scrub_lock);
1610 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1611 scrub_workers_put(root);
1612 return PTR_ERR(sdev);
1614 sdev->readonly = readonly;
1615 dev->scrub_device = sdev;
1617 atomic_inc(&fs_info->scrubs_running);
1618 mutex_unlock(&fs_info->scrub_lock);
1619 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1621 down_read(&fs_info->scrub_super_lock);
1622 ret = scrub_supers(sdev);
1623 up_read(&fs_info->scrub_super_lock);
1626 ret = scrub_enumerate_chunks(sdev, start, end);
1628 wait_event(sdev->list_wait, atomic_read(&sdev->in_flight) == 0);
1629 atomic_dec(&fs_info->scrubs_running);
1630 wake_up(&fs_info->scrub_pause_wait);
1632 wait_event(sdev->list_wait, atomic_read(&sdev->fixup_cnt) == 0);
1635 memcpy(progress, &sdev->stat, sizeof(*progress));
1637 mutex_lock(&fs_info->scrub_lock);
1638 dev->scrub_device = NULL;
1639 mutex_unlock(&fs_info->scrub_lock);
1641 scrub_free_dev(sdev);
1642 scrub_workers_put(root);
1647 int btrfs_scrub_pause(struct btrfs_root *root)
1649 struct btrfs_fs_info *fs_info = root->fs_info;
1651 mutex_lock(&fs_info->scrub_lock);
1652 atomic_inc(&fs_info->scrub_pause_req);
1653 while (atomic_read(&fs_info->scrubs_paused) !=
1654 atomic_read(&fs_info->scrubs_running)) {
1655 mutex_unlock(&fs_info->scrub_lock);
1656 wait_event(fs_info->scrub_pause_wait,
1657 atomic_read(&fs_info->scrubs_paused) ==
1658 atomic_read(&fs_info->scrubs_running));
1659 mutex_lock(&fs_info->scrub_lock);
1661 mutex_unlock(&fs_info->scrub_lock);
1666 int btrfs_scrub_continue(struct btrfs_root *root)
1668 struct btrfs_fs_info *fs_info = root->fs_info;
1670 atomic_dec(&fs_info->scrub_pause_req);
1671 wake_up(&fs_info->scrub_pause_wait);
1675 int btrfs_scrub_pause_super(struct btrfs_root *root)
1677 down_write(&root->fs_info->scrub_super_lock);
1681 int btrfs_scrub_continue_super(struct btrfs_root *root)
1683 up_write(&root->fs_info->scrub_super_lock);
1687 int btrfs_scrub_cancel(struct btrfs_root *root)
1689 struct btrfs_fs_info *fs_info = root->fs_info;
1691 mutex_lock(&fs_info->scrub_lock);
1692 if (!atomic_read(&fs_info->scrubs_running)) {
1693 mutex_unlock(&fs_info->scrub_lock);
1697 atomic_inc(&fs_info->scrub_cancel_req);
1698 while (atomic_read(&fs_info->scrubs_running)) {
1699 mutex_unlock(&fs_info->scrub_lock);
1700 wait_event(fs_info->scrub_pause_wait,
1701 atomic_read(&fs_info->scrubs_running) == 0);
1702 mutex_lock(&fs_info->scrub_lock);
1704 atomic_dec(&fs_info->scrub_cancel_req);
1705 mutex_unlock(&fs_info->scrub_lock);
1710 int btrfs_scrub_cancel_dev(struct btrfs_root *root, struct btrfs_device *dev)
1712 struct btrfs_fs_info *fs_info = root->fs_info;
1713 struct scrub_dev *sdev;
1715 mutex_lock(&fs_info->scrub_lock);
1716 sdev = dev->scrub_device;
1718 mutex_unlock(&fs_info->scrub_lock);
1721 atomic_inc(&sdev->cancel_req);
1722 while (dev->scrub_device) {
1723 mutex_unlock(&fs_info->scrub_lock);
1724 wait_event(fs_info->scrub_pause_wait,
1725 dev->scrub_device == NULL);
1726 mutex_lock(&fs_info->scrub_lock);
1728 mutex_unlock(&fs_info->scrub_lock);
1733 int btrfs_scrub_cancel_devid(struct btrfs_root *root, u64 devid)
1735 struct btrfs_fs_info *fs_info = root->fs_info;
1736 struct btrfs_device *dev;
1740 * we have to hold the device_list_mutex here so the device
1741 * does not go away in cancel_dev. FIXME: find a better solution
1743 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1744 dev = btrfs_find_device(root, devid, NULL, NULL);
1746 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1749 ret = btrfs_scrub_cancel_dev(root, dev);
1750 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1755 int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
1756 struct btrfs_scrub_progress *progress)
1758 struct btrfs_device *dev;
1759 struct scrub_dev *sdev = NULL;
1761 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1762 dev = btrfs_find_device(root, devid, NULL, NULL);
1764 sdev = dev->scrub_device;
1766 memcpy(progress, &sdev->stat, sizeof(*progress));
1767 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1769 return dev ? (sdev ? 0 : -ENOTCONN) : -ENODEV;