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/sched.h>
20 #include <linux/pagemap.h>
21 #include <linux/writeback.h>
22 #include <linux/blkdev.h>
23 #include <linux/rbtree.h>
24 #include <linux/slab.h>
25 #include <linux/workqueue.h>
29 #include "ordered-data.h"
32 * This is only the first step towards a full-features scrub. It reads all
33 * extent and super block and verifies the checksums. In case a bad checksum
34 * is found or the extent cannot be read, good data will be written back if
37 * Future enhancements:
38 * - To enhance the performance, better read-ahead strategies for the
39 * extent-tree can be employed.
40 * - In case an unrepairable extent is encountered, track which files are
41 * affected and report them
42 * - In case of a read error on files with nodatasum, map the file and read
43 * the extent to trigger a writeback of the good copy
44 * - track and record media errors, throw out bad devices
45 * - add a mode to also read unallocated space
46 * - make the prefetch cancellable
52 static void scrub_bio_end_io(struct bio *bio, int err);
53 static void scrub_checksum(struct btrfs_work *work);
54 static int scrub_checksum_data(struct scrub_dev *sdev,
55 struct scrub_page *spag, void *buffer);
56 static int scrub_checksum_tree_block(struct scrub_dev *sdev,
57 struct scrub_page *spag, u64 logical,
59 static int scrub_checksum_super(struct scrub_bio *sbio, void *buffer);
60 static int scrub_fixup_check(struct scrub_bio *sbio, int ix);
61 static void scrub_fixup_end_io(struct bio *bio, int err);
62 static int scrub_fixup_io(int rw, struct block_device *bdev, sector_t sector,
64 static void scrub_fixup(struct scrub_bio *sbio, int ix);
66 #define SCRUB_PAGES_PER_BIO 16 /* 64k per bio */
67 #define SCRUB_BIOS_PER_DEV 16 /* 1 MB per device in flight */
70 u64 flags; /* extent flags */
74 u8 csum[BTRFS_CSUM_SIZE];
79 struct scrub_dev *sdev;
84 struct scrub_page spag[SCRUB_PAGES_PER_BIO];
87 struct btrfs_work work;
91 struct scrub_bio *bios[SCRUB_BIOS_PER_DEV];
92 struct btrfs_device *dev;
97 wait_queue_head_t list_wait;
99 struct list_head csum_list;
105 struct btrfs_scrub_progress stat;
106 spinlock_t stat_lock;
109 static void scrub_free_csums(struct scrub_dev *sdev)
111 while (!list_empty(&sdev->csum_list)) {
112 struct btrfs_ordered_sum *sum;
113 sum = list_first_entry(&sdev->csum_list,
114 struct btrfs_ordered_sum, list);
115 list_del(&sum->list);
120 static noinline_for_stack void scrub_free_dev(struct scrub_dev *sdev)
124 struct page *last_page;
129 for (i = 0; i < SCRUB_BIOS_PER_DEV; ++i) {
130 struct scrub_bio *sbio = sdev->bios[i];
139 for (j = 0; j < bio->bi_vcnt; ++j) {
140 if (bio->bi_io_vec[j].bv_page == last_page)
142 last_page = bio->bi_io_vec[j].bv_page;
143 __free_page(last_page);
150 scrub_free_csums(sdev);
154 static noinline_for_stack
155 struct scrub_dev *scrub_setup_dev(struct btrfs_device *dev)
157 struct scrub_dev *sdev;
161 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
163 sdev = kzalloc(sizeof(*sdev), GFP_NOFS);
167 for (i = 0; i < SCRUB_BIOS_PER_DEV; ++i) {
169 struct scrub_bio *sbio;
171 sbio = kzalloc(sizeof(*sbio), GFP_NOFS);
174 sdev->bios[i] = sbio;
176 bio = bio_kmalloc(GFP_NOFS, SCRUB_PAGES_PER_BIO);
184 sbio->work.func = scrub_checksum;
185 bio->bi_private = sdev->bios[i];
186 bio->bi_end_io = scrub_bio_end_io;
188 bio->bi_bdev = dev->bdev;
191 for (j = 0; j < SCRUB_PAGES_PER_BIO; ++j) {
193 page = alloc_page(GFP_NOFS);
197 ret = bio_add_page(bio, page, PAGE_SIZE, 0);
201 WARN_ON(bio->bi_vcnt != SCRUB_PAGES_PER_BIO);
203 if (i != SCRUB_BIOS_PER_DEV-1)
204 sdev->bios[i]->next_free = i + 1;
206 sdev->bios[i]->next_free = -1;
208 sdev->first_free = 0;
210 atomic_set(&sdev->in_flight, 0);
211 atomic_set(&sdev->cancel_req, 0);
212 sdev->csum_size = btrfs_super_csum_size(&fs_info->super_copy);
213 INIT_LIST_HEAD(&sdev->csum_list);
215 spin_lock_init(&sdev->list_lock);
216 spin_lock_init(&sdev->stat_lock);
217 init_waitqueue_head(&sdev->list_wait);
221 scrub_free_dev(sdev);
222 return ERR_PTR(-ENOMEM);
226 * scrub_recheck_error gets called when either verification of the page
227 * failed or the bio failed to read, e.g. with EIO. In the latter case,
228 * recheck_error gets called for every page in the bio, even though only
231 static void scrub_recheck_error(struct scrub_bio *sbio, int ix)
234 if (scrub_fixup_io(READ, sbio->sdev->dev->bdev,
235 (sbio->physical + ix * PAGE_SIZE) >> 9,
236 sbio->bio->bi_io_vec[ix].bv_page) == 0) {
237 if (scrub_fixup_check(sbio, ix) == 0)
242 scrub_fixup(sbio, ix);
245 static int scrub_fixup_check(struct scrub_bio *sbio, int ix)
250 u64 flags = sbio->spag[ix].flags;
252 page = sbio->bio->bi_io_vec[ix].bv_page;
253 buffer = kmap_atomic(page, KM_USER0);
254 if (flags & BTRFS_EXTENT_FLAG_DATA) {
255 ret = scrub_checksum_data(sbio->sdev,
256 sbio->spag + ix, buffer);
257 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
258 ret = scrub_checksum_tree_block(sbio->sdev,
260 sbio->logical + ix * PAGE_SIZE,
265 kunmap_atomic(buffer, KM_USER0);
270 static void scrub_fixup_end_io(struct bio *bio, int err)
272 complete((struct completion *)bio->bi_private);
275 static void scrub_fixup(struct scrub_bio *sbio, int ix)
277 struct scrub_dev *sdev = sbio->sdev;
278 struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info;
279 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
280 struct btrfs_multi_bio *multi = NULL;
281 u64 logical = sbio->logical + ix * PAGE_SIZE;
285 DECLARE_COMPLETION_ONSTACK(complete);
287 if ((sbio->spag[ix].flags & BTRFS_EXTENT_FLAG_DATA) &&
288 (sbio->spag[ix].have_csum == 0)) {
290 * nodatasum, don't try to fix anything
291 * FIXME: we can do better, open the inode and trigger a
298 ret = btrfs_map_block(map_tree, REQ_WRITE, logical, &length,
300 if (ret || !multi || length < PAGE_SIZE) {
302 "scrub_fixup: btrfs_map_block failed us for %llu\n",
303 (unsigned long long)logical);
308 if (multi->num_stripes == 1)
309 /* there aren't any replicas */
313 * first find a good copy
315 for (i = 0; i < multi->num_stripes; ++i) {
316 if (i == sbio->spag[ix].mirror_num)
319 if (scrub_fixup_io(READ, multi->stripes[i].dev->bdev,
320 multi->stripes[i].physical >> 9,
321 sbio->bio->bi_io_vec[ix].bv_page)) {
322 /* I/O-error, this is not a good copy */
326 if (scrub_fixup_check(sbio, ix) == 0)
329 if (i == multi->num_stripes)
332 if (!sdev->readonly) {
334 * bi_io_vec[ix].bv_page now contains good data, write it back
336 if (scrub_fixup_io(WRITE, sdev->dev->bdev,
337 (sbio->physical + ix * PAGE_SIZE) >> 9,
338 sbio->bio->bi_io_vec[ix].bv_page)) {
339 /* I/O-error, writeback failed, give up */
345 spin_lock(&sdev->stat_lock);
346 ++sdev->stat.corrected_errors;
347 spin_unlock(&sdev->stat_lock);
349 if (printk_ratelimit())
350 printk(KERN_ERR "btrfs: fixed up at %llu\n",
351 (unsigned long long)logical);
356 spin_lock(&sdev->stat_lock);
357 ++sdev->stat.uncorrectable_errors;
358 spin_unlock(&sdev->stat_lock);
360 if (printk_ratelimit())
361 printk(KERN_ERR "btrfs: unable to fixup at %llu\n",
362 (unsigned long long)logical);
365 static int scrub_fixup_io(int rw, struct block_device *bdev, sector_t sector,
368 struct bio *bio = NULL;
370 DECLARE_COMPLETION_ONSTACK(complete);
372 /* we are going to wait on this IO */
375 bio = bio_alloc(GFP_NOFS, 1);
377 bio->bi_sector = sector;
378 bio_add_page(bio, page, PAGE_SIZE, 0);
379 bio->bi_end_io = scrub_fixup_end_io;
380 bio->bi_private = &complete;
383 wait_for_completion(&complete);
385 ret = !test_bit(BIO_UPTODATE, &bio->bi_flags);
390 static void scrub_bio_end_io(struct bio *bio, int err)
392 struct scrub_bio *sbio = bio->bi_private;
393 struct scrub_dev *sdev = sbio->sdev;
394 struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info;
398 btrfs_queue_worker(&fs_info->scrub_workers, &sbio->work);
401 static void scrub_checksum(struct btrfs_work *work)
403 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
404 struct scrub_dev *sdev = sbio->sdev;
413 for (i = 0; i < sbio->count; ++i)
414 scrub_recheck_error(sbio, i);
416 sbio->bio->bi_flags &= ~(BIO_POOL_MASK - 1);
417 sbio->bio->bi_flags |= 1 << BIO_UPTODATE;
418 sbio->bio->bi_phys_segments = 0;
419 sbio->bio->bi_idx = 0;
421 for (i = 0; i < sbio->count; i++) {
423 bi = &sbio->bio->bi_io_vec[i];
425 bi->bv_len = PAGE_SIZE;
428 spin_lock(&sdev->stat_lock);
429 ++sdev->stat.read_errors;
430 spin_unlock(&sdev->stat_lock);
433 for (i = 0; i < sbio->count; ++i) {
434 page = sbio->bio->bi_io_vec[i].bv_page;
435 buffer = kmap_atomic(page, KM_USER0);
436 flags = sbio->spag[i].flags;
437 logical = sbio->logical + i * PAGE_SIZE;
439 if (flags & BTRFS_EXTENT_FLAG_DATA) {
440 ret = scrub_checksum_data(sdev, sbio->spag + i, buffer);
441 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
442 ret = scrub_checksum_tree_block(sdev, sbio->spag + i,
444 } else if (flags & BTRFS_EXTENT_FLAG_SUPER) {
446 (void)scrub_checksum_super(sbio, buffer);
450 kunmap_atomic(buffer, KM_USER0);
452 scrub_recheck_error(sbio, i);
456 spin_lock(&sdev->list_lock);
457 sbio->next_free = sdev->first_free;
458 sdev->first_free = sbio->index;
459 spin_unlock(&sdev->list_lock);
460 atomic_dec(&sdev->in_flight);
461 wake_up(&sdev->list_wait);
464 static int scrub_checksum_data(struct scrub_dev *sdev,
465 struct scrub_page *spag, void *buffer)
467 u8 csum[BTRFS_CSUM_SIZE];
470 struct btrfs_root *root = sdev->dev->dev_root;
472 if (!spag->have_csum)
475 crc = btrfs_csum_data(root, buffer, crc, PAGE_SIZE);
476 btrfs_csum_final(crc, csum);
477 if (memcmp(csum, spag->csum, sdev->csum_size))
480 spin_lock(&sdev->stat_lock);
481 ++sdev->stat.data_extents_scrubbed;
482 sdev->stat.data_bytes_scrubbed += PAGE_SIZE;
484 ++sdev->stat.csum_errors;
485 spin_unlock(&sdev->stat_lock);
490 static int scrub_checksum_tree_block(struct scrub_dev *sdev,
491 struct scrub_page *spag, u64 logical,
494 struct btrfs_header *h;
495 struct btrfs_root *root = sdev->dev->dev_root;
496 struct btrfs_fs_info *fs_info = root->fs_info;
497 u8 csum[BTRFS_CSUM_SIZE];
503 * we don't use the getter functions here, as we
504 * a) don't have an extent buffer and
505 * b) the page is already kmapped
507 h = (struct btrfs_header *)buffer;
509 if (logical != le64_to_cpu(h->bytenr))
512 if (spag->generation != le64_to_cpu(h->generation))
515 if (memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
518 if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
522 crc = btrfs_csum_data(root, buffer + BTRFS_CSUM_SIZE, crc,
523 PAGE_SIZE - BTRFS_CSUM_SIZE);
524 btrfs_csum_final(crc, csum);
525 if (memcmp(csum, h->csum, sdev->csum_size))
528 spin_lock(&sdev->stat_lock);
529 ++sdev->stat.tree_extents_scrubbed;
530 sdev->stat.tree_bytes_scrubbed += PAGE_SIZE;
532 ++sdev->stat.csum_errors;
534 ++sdev->stat.verify_errors;
535 spin_unlock(&sdev->stat_lock);
537 return fail || crc_fail;
540 static int scrub_checksum_super(struct scrub_bio *sbio, void *buffer)
542 struct btrfs_super_block *s;
544 struct scrub_dev *sdev = sbio->sdev;
545 struct btrfs_root *root = sdev->dev->dev_root;
546 struct btrfs_fs_info *fs_info = root->fs_info;
547 u8 csum[BTRFS_CSUM_SIZE];
551 s = (struct btrfs_super_block *)buffer;
552 logical = sbio->logical;
554 if (logical != le64_to_cpu(s->bytenr))
557 if (sbio->spag[0].generation != le64_to_cpu(s->generation))
560 if (memcmp(s->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
563 crc = btrfs_csum_data(root, buffer + BTRFS_CSUM_SIZE, crc,
564 PAGE_SIZE - BTRFS_CSUM_SIZE);
565 btrfs_csum_final(crc, csum);
566 if (memcmp(csum, s->csum, sbio->sdev->csum_size))
571 * if we find an error in a super block, we just report it.
572 * They will get written with the next transaction commit
575 spin_lock(&sdev->stat_lock);
576 ++sdev->stat.super_errors;
577 spin_unlock(&sdev->stat_lock);
583 static int scrub_submit(struct scrub_dev *sdev)
585 struct scrub_bio *sbio;
587 if (sdev->curr == -1)
590 sbio = sdev->bios[sdev->curr];
592 sbio->bio->bi_sector = sbio->physical >> 9;
593 sbio->bio->bi_size = sbio->count * PAGE_SIZE;
594 sbio->bio->bi_next = NULL;
595 sbio->bio->bi_flags |= 1 << BIO_UPTODATE;
596 sbio->bio->bi_comp_cpu = -1;
597 sbio->bio->bi_bdev = sdev->dev->bdev;
600 atomic_inc(&sdev->in_flight);
602 submit_bio(0, sbio->bio);
607 static int scrub_page(struct scrub_dev *sdev, u64 logical, u64 len,
608 u64 physical, u64 flags, u64 gen, u64 mirror_num,
611 struct scrub_bio *sbio;
615 * grab a fresh bio or wait for one to become available
617 while (sdev->curr == -1) {
618 spin_lock(&sdev->list_lock);
619 sdev->curr = sdev->first_free;
620 if (sdev->curr != -1) {
621 sdev->first_free = sdev->bios[sdev->curr]->next_free;
622 sdev->bios[sdev->curr]->next_free = -1;
623 sdev->bios[sdev->curr]->count = 0;
624 spin_unlock(&sdev->list_lock);
626 spin_unlock(&sdev->list_lock);
627 wait_event(sdev->list_wait, sdev->first_free != -1);
630 sbio = sdev->bios[sdev->curr];
631 if (sbio->count == 0) {
632 sbio->physical = physical;
633 sbio->logical = logical;
634 } else if (sbio->physical + sbio->count * PAGE_SIZE != physical ||
635 sbio->logical + sbio->count * PAGE_SIZE != logical) {
639 sbio->spag[sbio->count].flags = flags;
640 sbio->spag[sbio->count].generation = gen;
641 sbio->spag[sbio->count].have_csum = 0;
642 sbio->spag[sbio->count].mirror_num = mirror_num;
644 sbio->spag[sbio->count].have_csum = 1;
645 memcpy(sbio->spag[sbio->count].csum, csum, sdev->csum_size);
648 if (sbio->count == SCRUB_PAGES_PER_BIO || force)
654 static int scrub_find_csum(struct scrub_dev *sdev, u64 logical, u64 len,
657 struct btrfs_ordered_sum *sum = NULL;
660 unsigned long num_sectors;
661 u32 sectorsize = sdev->dev->dev_root->sectorsize;
663 while (!list_empty(&sdev->csum_list)) {
664 sum = list_first_entry(&sdev->csum_list,
665 struct btrfs_ordered_sum, list);
666 if (sum->bytenr > logical)
668 if (sum->bytenr + sum->len > logical)
671 ++sdev->stat.csum_discards;
672 list_del(&sum->list);
679 num_sectors = sum->len / sectorsize;
680 for (i = 0; i < num_sectors; ++i) {
681 if (sum->sums[i].bytenr == logical) {
682 memcpy(csum, &sum->sums[i].sum, sdev->csum_size);
687 if (ret && i == num_sectors - 1) {
688 list_del(&sum->list);
694 /* scrub extent tries to collect up to 64 kB for each bio */
695 static int scrub_extent(struct scrub_dev *sdev, u64 logical, u64 len,
696 u64 physical, u64 flags, u64 gen, u64 mirror_num)
699 u8 csum[BTRFS_CSUM_SIZE];
702 u64 l = min_t(u64, len, PAGE_SIZE);
705 if (flags & BTRFS_EXTENT_FLAG_DATA) {
706 /* push csums to sbio */
707 have_csum = scrub_find_csum(sdev, logical, l, csum);
709 ++sdev->stat.no_csum;
711 ret = scrub_page(sdev, logical, l, physical, flags, gen,
712 mirror_num, have_csum ? csum : NULL, 0);
722 static noinline_for_stack int scrub_stripe(struct scrub_dev *sdev,
723 struct map_lookup *map, int num, u64 base, u64 length)
725 struct btrfs_path *path;
726 struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info;
727 struct btrfs_root *root = fs_info->extent_root;
728 struct btrfs_root *csum_root = fs_info->csum_root;
729 struct btrfs_extent_item *extent;
736 struct extent_buffer *l;
737 struct btrfs_key key;
743 u64 increment = map->stripe_len;
748 do_div(nstripes, map->stripe_len);
749 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
750 offset = map->stripe_len * num;
751 increment = map->stripe_len * map->num_stripes;
753 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
754 int factor = map->num_stripes / map->sub_stripes;
755 offset = map->stripe_len * (num / map->sub_stripes);
756 increment = map->stripe_len * factor;
757 mirror_num = num % map->sub_stripes;
758 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
759 increment = map->stripe_len;
760 mirror_num = num % map->num_stripes;
761 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
762 increment = map->stripe_len;
763 mirror_num = num % map->num_stripes;
765 increment = map->stripe_len;
769 path = btrfs_alloc_path();
774 path->search_commit_root = 1;
775 path->skip_locking = 1;
778 * find all extents for each stripe and just read them to get
779 * them into the page cache
780 * FIXME: we can do better. build a more intelligent prefetching
782 logical = base + offset;
783 physical = map->stripes[num].physical;
785 for (i = 0; i < nstripes; ++i) {
786 key.objectid = logical;
787 key.type = BTRFS_EXTENT_ITEM_KEY;
790 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
795 slot = path->slots[0];
796 btrfs_item_key_to_cpu(l, &key, slot);
797 if (key.objectid != logical) {
798 ret = btrfs_previous_item(root, path, 0,
799 BTRFS_EXTENT_ITEM_KEY);
806 slot = path->slots[0];
807 if (slot >= btrfs_header_nritems(l)) {
808 ret = btrfs_next_leaf(root, path);
816 btrfs_item_key_to_cpu(l, &key, slot);
818 if (key.objectid >= logical + map->stripe_len)
823 btrfs_release_path(path);
824 logical += increment;
825 physical += map->stripe_len;
830 * collect all data csums for the stripe to avoid seeking during
831 * the scrub. This might currently (crc32) end up to be about 1MB
835 logical = base + offset + start_stripe * increment;
836 for (i = start_stripe; i < nstripes; ++i) {
837 ret = btrfs_lookup_csums_range(csum_root, logical,
838 logical + map->stripe_len - 1,
839 &sdev->csum_list, 1);
843 logical += increment;
847 * now find all extents for each stripe and scrub them
849 logical = base + offset + start_stripe * increment;
850 physical = map->stripes[num].physical + start_stripe * map->stripe_len;
852 for (i = start_stripe; i < nstripes; ++i) {
856 if (atomic_read(&fs_info->scrub_cancel_req) ||
857 atomic_read(&sdev->cancel_req)) {
862 * check to see if we have to pause
864 if (atomic_read(&fs_info->scrub_pause_req)) {
865 /* push queued extents */
867 wait_event(sdev->list_wait,
868 atomic_read(&sdev->in_flight) == 0);
869 atomic_inc(&fs_info->scrubs_paused);
870 wake_up(&fs_info->scrub_pause_wait);
871 mutex_lock(&fs_info->scrub_lock);
872 while (atomic_read(&fs_info->scrub_pause_req)) {
873 mutex_unlock(&fs_info->scrub_lock);
874 wait_event(fs_info->scrub_pause_wait,
875 atomic_read(&fs_info->scrub_pause_req) == 0);
876 mutex_lock(&fs_info->scrub_lock);
878 atomic_dec(&fs_info->scrubs_paused);
879 mutex_unlock(&fs_info->scrub_lock);
880 wake_up(&fs_info->scrub_pause_wait);
881 scrub_free_csums(sdev);
886 key.objectid = logical;
887 key.type = BTRFS_EXTENT_ITEM_KEY;
890 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
895 slot = path->slots[0];
896 btrfs_item_key_to_cpu(l, &key, slot);
897 if (key.objectid != logical) {
898 ret = btrfs_previous_item(root, path, 0,
899 BTRFS_EXTENT_ITEM_KEY);
906 slot = path->slots[0];
907 if (slot >= btrfs_header_nritems(l)) {
908 ret = btrfs_next_leaf(root, path);
916 btrfs_item_key_to_cpu(l, &key, slot);
918 if (key.objectid + key.offset <= logical)
921 if (key.objectid >= logical + map->stripe_len)
924 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY)
927 extent = btrfs_item_ptr(l, slot,
928 struct btrfs_extent_item);
929 flags = btrfs_extent_flags(l, extent);
930 generation = btrfs_extent_generation(l, extent);
932 if (key.objectid < logical &&
933 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
935 "btrfs scrub: tree block %llu spanning "
936 "stripes, ignored. logical=%llu\n",
937 (unsigned long long)key.objectid,
938 (unsigned long long)logical);
943 * trim extent to this stripe
945 if (key.objectid < logical) {
946 key.offset -= logical - key.objectid;
947 key.objectid = logical;
949 if (key.objectid + key.offset >
950 logical + map->stripe_len) {
951 key.offset = logical + map->stripe_len -
955 ret = scrub_extent(sdev, key.objectid, key.offset,
956 key.objectid - logical + physical,
957 flags, generation, mirror_num);
964 btrfs_release_path(path);
965 logical += increment;
966 physical += map->stripe_len;
967 spin_lock(&sdev->stat_lock);
968 sdev->stat.last_physical = physical;
969 spin_unlock(&sdev->stat_lock);
971 /* push queued extents */
975 btrfs_free_path(path);
976 return ret < 0 ? ret : 0;
979 static noinline_for_stack int scrub_chunk(struct scrub_dev *sdev,
980 u64 chunk_tree, u64 chunk_objectid, u64 chunk_offset, u64 length)
982 struct btrfs_mapping_tree *map_tree =
983 &sdev->dev->dev_root->fs_info->mapping_tree;
984 struct map_lookup *map;
985 struct extent_map *em;
989 read_lock(&map_tree->map_tree.lock);
990 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
991 read_unlock(&map_tree->map_tree.lock);
996 map = (struct map_lookup *)em->bdev;
997 if (em->start != chunk_offset)
1000 if (em->len < length)
1003 for (i = 0; i < map->num_stripes; ++i) {
1004 if (map->stripes[i].dev == sdev->dev) {
1005 ret = scrub_stripe(sdev, map, i, chunk_offset, length);
1011 free_extent_map(em);
1016 static noinline_for_stack
1017 int scrub_enumerate_chunks(struct scrub_dev *sdev, u64 start, u64 end)
1019 struct btrfs_dev_extent *dev_extent = NULL;
1020 struct btrfs_path *path;
1021 struct btrfs_root *root = sdev->dev->dev_root;
1022 struct btrfs_fs_info *fs_info = root->fs_info;
1029 struct extent_buffer *l;
1030 struct btrfs_key key;
1031 struct btrfs_key found_key;
1032 struct btrfs_block_group_cache *cache;
1034 path = btrfs_alloc_path();
1039 path->search_commit_root = 1;
1040 path->skip_locking = 1;
1042 key.objectid = sdev->dev->devid;
1044 key.type = BTRFS_DEV_EXTENT_KEY;
1048 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1054 slot = path->slots[0];
1056 btrfs_item_key_to_cpu(l, &found_key, slot);
1058 if (found_key.objectid != sdev->dev->devid)
1061 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1064 if (found_key.offset >= end)
1067 if (found_key.offset < key.offset)
1070 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1071 length = btrfs_dev_extent_length(l, dev_extent);
1073 if (found_key.offset + length <= start) {
1074 key.offset = found_key.offset + length;
1075 btrfs_release_path(path);
1079 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1080 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1081 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1084 * get a reference on the corresponding block group to prevent
1085 * the chunk from going away while we scrub it
1087 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
1092 ret = scrub_chunk(sdev, chunk_tree, chunk_objectid,
1093 chunk_offset, length);
1094 btrfs_put_block_group(cache);
1098 key.offset = found_key.offset + length;
1099 btrfs_release_path(path);
1103 btrfs_free_path(path);
1107 static noinline_for_stack int scrub_supers(struct scrub_dev *sdev)
1113 struct btrfs_device *device = sdev->dev;
1114 struct btrfs_root *root = device->dev_root;
1116 gen = root->fs_info->last_trans_committed;
1118 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1119 bytenr = btrfs_sb_offset(i);
1120 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
1123 ret = scrub_page(sdev, bytenr, PAGE_SIZE, bytenr,
1124 BTRFS_EXTENT_FLAG_SUPER, gen, i, NULL, 1);
1128 wait_event(sdev->list_wait, atomic_read(&sdev->in_flight) == 0);
1134 * get a reference count on fs_info->scrub_workers. start worker if necessary
1136 static noinline_for_stack int scrub_workers_get(struct btrfs_root *root)
1138 struct btrfs_fs_info *fs_info = root->fs_info;
1140 mutex_lock(&fs_info->scrub_lock);
1141 if (fs_info->scrub_workers_refcnt == 0)
1142 btrfs_start_workers(&fs_info->scrub_workers, 1);
1143 ++fs_info->scrub_workers_refcnt;
1144 mutex_unlock(&fs_info->scrub_lock);
1149 static noinline_for_stack void scrub_workers_put(struct btrfs_root *root)
1151 struct btrfs_fs_info *fs_info = root->fs_info;
1153 mutex_lock(&fs_info->scrub_lock);
1154 if (--fs_info->scrub_workers_refcnt == 0)
1155 btrfs_stop_workers(&fs_info->scrub_workers);
1156 WARN_ON(fs_info->scrub_workers_refcnt < 0);
1157 mutex_unlock(&fs_info->scrub_lock);
1161 int btrfs_scrub_dev(struct btrfs_root *root, u64 devid, u64 start, u64 end,
1162 struct btrfs_scrub_progress *progress, int readonly)
1164 struct scrub_dev *sdev;
1165 struct btrfs_fs_info *fs_info = root->fs_info;
1167 struct btrfs_device *dev;
1169 if (root->fs_info->closing)
1173 * check some assumptions
1175 if (root->sectorsize != PAGE_SIZE ||
1176 root->sectorsize != root->leafsize ||
1177 root->sectorsize != root->nodesize) {
1178 printk(KERN_ERR "btrfs_scrub: size assumptions fail\n");
1182 ret = scrub_workers_get(root);
1186 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1187 dev = btrfs_find_device(root, devid, NULL, NULL);
1188 if (!dev || dev->missing) {
1189 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1190 scrub_workers_put(root);
1193 mutex_lock(&fs_info->scrub_lock);
1195 if (!dev->in_fs_metadata) {
1196 mutex_unlock(&fs_info->scrub_lock);
1197 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1198 scrub_workers_put(root);
1202 if (dev->scrub_device) {
1203 mutex_unlock(&fs_info->scrub_lock);
1204 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1205 scrub_workers_put(root);
1206 return -EINPROGRESS;
1208 sdev = scrub_setup_dev(dev);
1210 mutex_unlock(&fs_info->scrub_lock);
1211 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1212 scrub_workers_put(root);
1213 return PTR_ERR(sdev);
1215 sdev->readonly = readonly;
1216 dev->scrub_device = sdev;
1218 atomic_inc(&fs_info->scrubs_running);
1219 mutex_unlock(&fs_info->scrub_lock);
1220 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1222 down_read(&fs_info->scrub_super_lock);
1223 ret = scrub_supers(sdev);
1224 up_read(&fs_info->scrub_super_lock);
1227 ret = scrub_enumerate_chunks(sdev, start, end);
1229 wait_event(sdev->list_wait, atomic_read(&sdev->in_flight) == 0);
1231 atomic_dec(&fs_info->scrubs_running);
1232 wake_up(&fs_info->scrub_pause_wait);
1235 memcpy(progress, &sdev->stat, sizeof(*progress));
1237 mutex_lock(&fs_info->scrub_lock);
1238 dev->scrub_device = NULL;
1239 mutex_unlock(&fs_info->scrub_lock);
1241 scrub_free_dev(sdev);
1242 scrub_workers_put(root);
1247 int btrfs_scrub_pause(struct btrfs_root *root)
1249 struct btrfs_fs_info *fs_info = root->fs_info;
1251 mutex_lock(&fs_info->scrub_lock);
1252 atomic_inc(&fs_info->scrub_pause_req);
1253 while (atomic_read(&fs_info->scrubs_paused) !=
1254 atomic_read(&fs_info->scrubs_running)) {
1255 mutex_unlock(&fs_info->scrub_lock);
1256 wait_event(fs_info->scrub_pause_wait,
1257 atomic_read(&fs_info->scrubs_paused) ==
1258 atomic_read(&fs_info->scrubs_running));
1259 mutex_lock(&fs_info->scrub_lock);
1261 mutex_unlock(&fs_info->scrub_lock);
1266 int btrfs_scrub_continue(struct btrfs_root *root)
1268 struct btrfs_fs_info *fs_info = root->fs_info;
1270 atomic_dec(&fs_info->scrub_pause_req);
1271 wake_up(&fs_info->scrub_pause_wait);
1275 int btrfs_scrub_pause_super(struct btrfs_root *root)
1277 down_write(&root->fs_info->scrub_super_lock);
1281 int btrfs_scrub_continue_super(struct btrfs_root *root)
1283 up_write(&root->fs_info->scrub_super_lock);
1287 int btrfs_scrub_cancel(struct btrfs_root *root)
1289 struct btrfs_fs_info *fs_info = root->fs_info;
1291 mutex_lock(&fs_info->scrub_lock);
1292 if (!atomic_read(&fs_info->scrubs_running)) {
1293 mutex_unlock(&fs_info->scrub_lock);
1297 atomic_inc(&fs_info->scrub_cancel_req);
1298 while (atomic_read(&fs_info->scrubs_running)) {
1299 mutex_unlock(&fs_info->scrub_lock);
1300 wait_event(fs_info->scrub_pause_wait,
1301 atomic_read(&fs_info->scrubs_running) == 0);
1302 mutex_lock(&fs_info->scrub_lock);
1304 atomic_dec(&fs_info->scrub_cancel_req);
1305 mutex_unlock(&fs_info->scrub_lock);
1310 int btrfs_scrub_cancel_dev(struct btrfs_root *root, struct btrfs_device *dev)
1312 struct btrfs_fs_info *fs_info = root->fs_info;
1313 struct scrub_dev *sdev;
1315 mutex_lock(&fs_info->scrub_lock);
1316 sdev = dev->scrub_device;
1318 mutex_unlock(&fs_info->scrub_lock);
1321 atomic_inc(&sdev->cancel_req);
1322 while (dev->scrub_device) {
1323 mutex_unlock(&fs_info->scrub_lock);
1324 wait_event(fs_info->scrub_pause_wait,
1325 dev->scrub_device == NULL);
1326 mutex_lock(&fs_info->scrub_lock);
1328 mutex_unlock(&fs_info->scrub_lock);
1332 int btrfs_scrub_cancel_devid(struct btrfs_root *root, u64 devid)
1334 struct btrfs_fs_info *fs_info = root->fs_info;
1335 struct btrfs_device *dev;
1339 * we have to hold the device_list_mutex here so the device
1340 * does not go away in cancel_dev. FIXME: find a better solution
1342 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1343 dev = btrfs_find_device(root, devid, NULL, NULL);
1345 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1348 ret = btrfs_scrub_cancel_dev(root, dev);
1349 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1354 int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
1355 struct btrfs_scrub_progress *progress)
1357 struct btrfs_device *dev;
1358 struct scrub_dev *sdev = NULL;
1360 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1361 dev = btrfs_find_device(root, devid, NULL, NULL);
1363 sdev = dev->scrub_device;
1365 memcpy(progress, &sdev->stat, sizeof(*progress));
1366 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1368 return dev ? (sdev ? 0 : -ENOTCONN) : -ENODEV;