aboutsummaryrefslogtreecommitdiffstats
path: root/fs/btrfs/scrub.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/btrfs/scrub.c')
-rw-r--r--fs/btrfs/scrub.c1353
1 files changed, 1013 insertions, 340 deletions
diff --git a/fs/btrfs/scrub.c b/fs/btrfs/scrub.c
index e68bab4ffcd4..5221e072bb65 100644
--- a/fs/btrfs/scrub.c
+++ b/fs/btrfs/scrub.c
@@ -40,16 +40,26 @@
40 * - add a mode to also read unallocated space 40 * - add a mode to also read unallocated space
41 */ 41 */
42 42
43struct scrub_block;
43struct scrub_dev; 44struct scrub_dev;
44 45
45#define SCRUB_PAGES_PER_BIO 16 /* 64k per bio */ 46#define SCRUB_PAGES_PER_BIO 16 /* 64k per bio */
46#define SCRUB_BIOS_PER_DEV 16 /* 1 MB per device in flight */ 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 */
47 49
48struct scrub_page { 50struct scrub_page {
51 struct scrub_block *sblock;
52 struct page *page;
53 struct block_device *bdev;
49 u64 flags; /* extent flags */ 54 u64 flags; /* extent flags */
50 u64 generation; 55 u64 generation;
51 int mirror_num; 56 u64 logical;
52 int have_csum; 57 u64 physical;
58 struct {
59 unsigned int mirror_num:8;
60 unsigned int have_csum:1;
61 unsigned int io_error:1;
62 };
53 u8 csum[BTRFS_CSUM_SIZE]; 63 u8 csum[BTRFS_CSUM_SIZE];
54}; 64};
55 65
@@ -60,12 +70,25 @@ struct scrub_bio {
60 int err; 70 int err;
61 u64 logical; 71 u64 logical;
62 u64 physical; 72 u64 physical;
63 struct scrub_page spag[SCRUB_PAGES_PER_BIO]; 73 struct scrub_page *pagev[SCRUB_PAGES_PER_BIO];
64 u64 count; 74 int page_count;
65 int next_free; 75 int next_free;
66 struct btrfs_work work; 76 struct btrfs_work work;
67}; 77};
68 78
79struct scrub_block {
80 struct scrub_page pagev[SCRUB_MAX_PAGES_PER_BLOCK];
81 int page_count;
82 atomic_t outstanding_pages;
83 atomic_t ref_count; /* free mem on transition to zero */
84 struct scrub_dev *sdev;
85 struct {
86 unsigned int header_error:1;
87 unsigned int checksum_error:1;
88 unsigned int no_io_error_seen:1;
89 };
90};
91
69struct scrub_dev { 92struct scrub_dev {
70 struct scrub_bio *bios[SCRUB_BIOS_PER_DEV]; 93 struct scrub_bio *bios[SCRUB_BIOS_PER_DEV];
71 struct btrfs_device *dev; 94 struct btrfs_device *dev;
@@ -79,6 +102,10 @@ struct scrub_dev {
79 struct list_head csum_list; 102 struct list_head csum_list;
80 atomic_t cancel_req; 103 atomic_t cancel_req;
81 int readonly; 104 int readonly;
105 int pages_per_bio; /* <= SCRUB_PAGES_PER_BIO */
106 u32 sectorsize;
107 u32 nodesize;
108 u32 leafsize;
82 /* 109 /*
83 * statistics 110 * statistics
84 */ 111 */
@@ -107,19 +134,41 @@ struct scrub_warning {
107 int scratch_bufsize; 134 int scratch_bufsize;
108}; 135};
109 136
137
138static int scrub_handle_errored_block(struct scrub_block *sblock_to_check);
139static 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);
143static 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,
146 u16 csum_size);
147static 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,
151 u16 csum_size);
152static void scrub_complete_bio_end_io(struct bio *bio, int err);
153static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
154 struct scrub_block *sblock_good,
155 int force_write);
156static 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);
159static int scrub_checksum_data(struct scrub_block *sblock);
160static int scrub_checksum_tree_block(struct scrub_block *sblock);
161static int scrub_checksum_super(struct scrub_block *sblock);
162static void scrub_block_get(struct scrub_block *sblock);
163static void scrub_block_put(struct scrub_block *sblock);
164static int scrub_add_page_to_bio(struct scrub_dev *sdev,
165 struct scrub_page *spage);
166static 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);
110static void scrub_bio_end_io(struct bio *bio, int err); 169static void scrub_bio_end_io(struct bio *bio, int err);
111static void scrub_checksum(struct btrfs_work *work); 170static void scrub_bio_end_io_worker(struct btrfs_work *work);
112static int scrub_checksum_data(struct scrub_dev *sdev, 171static void scrub_block_complete(struct scrub_block *sblock);
113 struct scrub_page *spag, void *buffer);
114static int scrub_checksum_tree_block(struct scrub_dev *sdev,
115 struct scrub_page *spag, u64 logical,
116 void *buffer);
117static int scrub_checksum_super(struct scrub_bio *sbio, void *buffer);
118static int scrub_fixup_check(struct scrub_bio *sbio, int ix);
119static void scrub_fixup_end_io(struct bio *bio, int err);
120static int scrub_fixup_io(int rw, struct block_device *bdev, sector_t sector,
121 struct page *page);
122static void scrub_fixup(struct scrub_bio *sbio, int ix);
123 172
124 173
125static void scrub_free_csums(struct scrub_dev *sdev) 174static void scrub_free_csums(struct scrub_dev *sdev)
@@ -133,23 +182,6 @@ static void scrub_free_csums(struct scrub_dev *sdev)
133 } 182 }
134} 183}
135 184
136static void scrub_free_bio(struct bio *bio)
137{
138 int i;
139 struct page *last_page = NULL;
140
141 if (!bio)
142 return;
143
144 for (i = 0; i < bio->bi_vcnt; ++i) {
145 if (bio->bi_io_vec[i].bv_page == last_page)
146 continue;
147 last_page = bio->bi_io_vec[i].bv_page;
148 __free_page(last_page);
149 }
150 bio_put(bio);
151}
152
153static noinline_for_stack void scrub_free_dev(struct scrub_dev *sdev) 185static noinline_for_stack void scrub_free_dev(struct scrub_dev *sdev)
154{ 186{
155 int i; 187 int i;
@@ -157,13 +189,23 @@ static noinline_for_stack void scrub_free_dev(struct scrub_dev *sdev)
157 if (!sdev) 189 if (!sdev)
158 return; 190 return;
159 191
192 /* this can happen when scrub is cancelled */
193 if (sdev->curr != -1) {
194 struct scrub_bio *sbio = sdev->bios[sdev->curr];
195
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);
200 }
201 bio_put(sbio->bio);
202 }
203
160 for (i = 0; i < SCRUB_BIOS_PER_DEV; ++i) { 204 for (i = 0; i < SCRUB_BIOS_PER_DEV; ++i) {
161 struct scrub_bio *sbio = sdev->bios[i]; 205 struct scrub_bio *sbio = sdev->bios[i];
162 206
163 if (!sbio) 207 if (!sbio)
164 break; 208 break;
165
166 scrub_free_bio(sbio->bio);
167 kfree(sbio); 209 kfree(sbio);
168 } 210 }
169 211
@@ -177,11 +219,16 @@ struct scrub_dev *scrub_setup_dev(struct btrfs_device *dev)
177 struct scrub_dev *sdev; 219 struct scrub_dev *sdev;
178 int i; 220 int i;
179 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info; 221 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
222 int pages_per_bio;
180 223
224 pages_per_bio = min_t(int, SCRUB_PAGES_PER_BIO,
225 bio_get_nr_vecs(dev->bdev));
181 sdev = kzalloc(sizeof(*sdev), GFP_NOFS); 226 sdev = kzalloc(sizeof(*sdev), GFP_NOFS);
182 if (!sdev) 227 if (!sdev)
183 goto nomem; 228 goto nomem;
184 sdev->dev = dev; 229 sdev->dev = dev;
230 sdev->pages_per_bio = pages_per_bio;
231 sdev->curr = -1;
185 for (i = 0; i < SCRUB_BIOS_PER_DEV; ++i) { 232 for (i = 0; i < SCRUB_BIOS_PER_DEV; ++i) {
186 struct scrub_bio *sbio; 233 struct scrub_bio *sbio;
187 234
@@ -192,8 +239,8 @@ struct scrub_dev *scrub_setup_dev(struct btrfs_device *dev)
192 239
193 sbio->index = i; 240 sbio->index = i;
194 sbio->sdev = sdev; 241 sbio->sdev = sdev;
195 sbio->count = 0; 242 sbio->page_count = 0;
196 sbio->work.func = scrub_checksum; 243 sbio->work.func = scrub_bio_end_io_worker;
197 244
198 if (i != SCRUB_BIOS_PER_DEV-1) 245 if (i != SCRUB_BIOS_PER_DEV-1)
199 sdev->bios[i]->next_free = i + 1; 246 sdev->bios[i]->next_free = i + 1;
@@ -201,7 +248,9 @@ struct scrub_dev *scrub_setup_dev(struct btrfs_device *dev)
201 sdev->bios[i]->next_free = -1; 248 sdev->bios[i]->next_free = -1;
202 } 249 }
203 sdev->first_free = 0; 250 sdev->first_free = 0;
204 sdev->curr = -1; 251 sdev->nodesize = dev->dev_root->nodesize;
252 sdev->leafsize = dev->dev_root->leafsize;
253 sdev->sectorsize = dev->dev_root->sectorsize;
205 atomic_set(&sdev->in_flight, 0); 254 atomic_set(&sdev->in_flight, 0);
206 atomic_set(&sdev->fixup_cnt, 0); 255 atomic_set(&sdev->fixup_cnt, 0);
207 atomic_set(&sdev->cancel_req, 0); 256 atomic_set(&sdev->cancel_req, 0);
@@ -292,10 +341,9 @@ err:
292 return 0; 341 return 0;
293} 342}
294 343
295static void scrub_print_warning(const char *errstr, struct scrub_bio *sbio, 344static void scrub_print_warning(const char *errstr, struct scrub_block *sblock)
296 int ix)
297{ 345{
298 struct btrfs_device *dev = sbio->sdev->dev; 346 struct btrfs_device *dev = sblock->sdev->dev;
299 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info; 347 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
300 struct btrfs_path *path; 348 struct btrfs_path *path;
301 struct btrfs_key found_key; 349 struct btrfs_key found_key;
@@ -314,8 +362,9 @@ static void scrub_print_warning(const char *errstr, struct scrub_bio *sbio,
314 362
315 swarn.scratch_buf = kmalloc(bufsize, GFP_NOFS); 363 swarn.scratch_buf = kmalloc(bufsize, GFP_NOFS);
316 swarn.msg_buf = kmalloc(bufsize, GFP_NOFS); 364 swarn.msg_buf = kmalloc(bufsize, GFP_NOFS);
317 swarn.sector = (sbio->physical + ix * PAGE_SIZE) >> 9; 365 BUG_ON(sblock->page_count < 1);
318 swarn.logical = sbio->logical + ix * PAGE_SIZE; 366 swarn.sector = (sblock->pagev[0].physical) >> 9;
367 swarn.logical = sblock->pagev[0].logical;
319 swarn.errstr = errstr; 368 swarn.errstr = errstr;
320 swarn.dev = dev; 369 swarn.dev = dev;
321 swarn.msg_bufsize = bufsize; 370 swarn.msg_bufsize = bufsize;
@@ -530,9 +579,9 @@ out:
530 spin_lock(&sdev->stat_lock); 579 spin_lock(&sdev->stat_lock);
531 ++sdev->stat.uncorrectable_errors; 580 ++sdev->stat.uncorrectable_errors;
532 spin_unlock(&sdev->stat_lock); 581 spin_unlock(&sdev->stat_lock);
533 printk_ratelimited(KERN_ERR "btrfs: unable to fixup " 582 printk_ratelimited(KERN_ERR
534 "(nodatasum) error at logical %llu\n", 583 "btrfs: unable to fixup (nodatasum) error at logical %llu on dev %s\n",
535 fixup->logical); 584 (unsigned long long)fixup->logical, sdev->dev->name);
536 } 585 }
537 586
538 btrfs_free_path(path); 587 btrfs_free_path(path);
@@ -549,91 +598,168 @@ out:
549} 598}
550 599
551/* 600/*
552 * scrub_recheck_error gets called when either verification of the page 601 * scrub_handle_errored_block gets called when either verification of the
553 * failed or the bio failed to read, e.g. with EIO. In the latter case, 602 * pages failed or the bio failed to read, e.g. with EIO. In the latter
554 * recheck_error gets called for every page in the bio, even though only 603 * case, this function handles all pages in the bio, even though only one
555 * one may be bad 604 * may be bad.
605 * The goal of this function is to repair the errored block by using the
606 * contents of one of the mirrors.
556 */ 607 */
557static int scrub_recheck_error(struct scrub_bio *sbio, int ix) 608static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
558{ 609{
559 struct scrub_dev *sdev = sbio->sdev; 610 struct scrub_dev *sdev = sblock_to_check->sdev;
560 u64 sector = (sbio->physical + ix * PAGE_SIZE) >> 9; 611 struct btrfs_fs_info *fs_info;
612 u64 length;
613 u64 logical;
614 u64 generation;
615 unsigned int failed_mirror_index;
616 unsigned int is_metadata;
617 unsigned int have_csum;
618 u8 *csum;
619 struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */
620 struct scrub_block *sblock_bad;
621 int ret;
622 int mirror_index;
623 int page_num;
624 int success;
561 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL, 625 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
562 DEFAULT_RATELIMIT_BURST); 626 DEFAULT_RATELIMIT_BURST);
627
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;
563 639
564 if (sbio->err) { 640 /*
565 if (scrub_fixup_io(READ, sbio->sdev->dev->bdev, sector, 641 * read all mirrors one after the other. This includes to
566 sbio->bio->bi_io_vec[ix].bv_page) == 0) { 642 * re-read the extent or metadata block that failed (that was
567 if (scrub_fixup_check(sbio, ix) == 0) 643 * the cause that this fixup code is called) another time,
568 return 0; 644 * page by page this time in order to know which pages
569 } 645 * caused I/O errors and which ones are good (for all mirrors).
570 if (__ratelimit(&_rs)) 646 * It is the goal to handle the situation when more than one
571 scrub_print_warning("i/o error", sbio, ix); 647 * mirror contains I/O errors, but the errors do not
572 } else { 648 * overlap, i.e. the data can be repaired by selecting the
573 if (__ratelimit(&_rs)) 649 * pages from those mirrors without I/O error on the
574 scrub_print_warning("checksum error", sbio, ix); 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
666 * the statistics.
667 */
668
669 sblocks_for_recheck = kzalloc(BTRFS_MAX_MIRRORS *
670 sizeof(*sblocks_for_recheck),
671 GFP_NOFS);
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);
678 goto out;
575 } 679 }
576 680
577 spin_lock(&sdev->stat_lock); 681 /* setup the context, map the logical blocks and alloc the pages */
578 ++sdev->stat.read_errors; 682 ret = scrub_setup_recheck_block(sdev, &fs_info->mapping_tree, length,
579 spin_unlock(&sdev->stat_lock); 683 logical, sblocks_for_recheck);
684 if (ret) {
685 spin_lock(&sdev->stat_lock);
686 sdev->stat.read_errors++;
687 sdev->stat.uncorrectable_errors++;
688 spin_unlock(&sdev->stat_lock);
689 goto out;
690 }
691 BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS);
692 sblock_bad = sblocks_for_recheck + failed_mirror_index;
580 693
581 scrub_fixup(sbio, ix); 694 /* build and submit the bios for the failed mirror, check checksums */
582 return 1; 695 ret = scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum,
583} 696 csum, generation, sdev->csum_size);
697 if (ret) {
698 spin_lock(&sdev->stat_lock);
699 sdev->stat.read_errors++;
700 sdev->stat.uncorrectable_errors++;
701 spin_unlock(&sdev->stat_lock);
702 goto out;
703 }
584 704
585static int scrub_fixup_check(struct scrub_bio *sbio, int ix) 705 if (!sblock_bad->header_error && !sblock_bad->checksum_error &&
586{ 706 sblock_bad->no_io_error_seen) {
587 int ret = 1; 707 /*
588 struct page *page; 708 * the error disappeared after reading page by page, or
589 void *buffer; 709 * the area was part of a huge bio and other parts of the
590 u64 flags = sbio->spag[ix].flags; 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
713 * the cause)
714 */
715 spin_lock(&sdev->stat_lock);
716 sdev->stat.unverified_errors++;
717 spin_unlock(&sdev->stat_lock);
591 718
592 page = sbio->bio->bi_io_vec[ix].bv_page; 719 goto out;
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,
599 sbio->spag + ix,
600 sbio->logical + ix * PAGE_SIZE,
601 buffer);
602 } else {
603 WARN_ON(1);
604 } 720 }
605 kunmap_atomic(buffer, KM_USER0);
606 721
607 return ret; 722 if (!sblock_bad->no_io_error_seen) {
608} 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",
740 sblock_to_check);
741 }
609 742
610static void scrub_fixup_end_io(struct bio *bio, int err) 743 if (sdev->readonly)
611{ 744 goto did_not_correct_error;
612 complete((struct completion *)bio->bi_private); 745
613} 746 if (!is_metadata && !have_csum) {
747 struct scrub_fixup_nodatasum *fixup_nodatasum;
614 748
615static void scrub_fixup(struct scrub_bio *sbio, int ix) 749 /*
616{ 750 * !is_metadata and !have_csum, this means that the data
617 struct scrub_dev *sdev = sbio->sdev; 751 * might not be COW'ed, that it might be modified
618 struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info; 752 * concurrently. The general strategy to work on the
619 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; 753 * commit root does not help in the case when COW is not
620 struct btrfs_bio *bbio = NULL; 754 * used.
621 struct scrub_fixup_nodatasum *fixup; 755 */
622 u64 logical = sbio->logical + ix * PAGE_SIZE; 756 fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS);
623 u64 length; 757 if (!fixup_nodatasum)
624 int i; 758 goto did_not_correct_error;
625 int ret; 759 fixup_nodatasum->sdev = sdev;
626 DECLARE_COMPLETION_ONSTACK(complete); 760 fixup_nodatasum->logical = logical;
627 761 fixup_nodatasum->root = fs_info->extent_root;
628 if ((sbio->spag[ix].flags & BTRFS_EXTENT_FLAG_DATA) && 762 fixup_nodatasum->mirror_num = failed_mirror_index + 1;
629 (sbio->spag[ix].have_csum == 0)) {
630 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
631 if (!fixup)
632 goto uncorrectable;
633 fixup->sdev = sdev;
634 fixup->logical = logical;
635 fixup->root = fs_info->extent_root;
636 fixup->mirror_num = sbio->spag[ix].mirror_num;
637 /* 763 /*
638 * increment scrubs_running to prevent cancel requests from 764 * increment scrubs_running to prevent cancel requests from
639 * completing as long as a fixup worker is running. we must also 765 * completing as long as a fixup worker is running. we must also
@@ -648,235 +774,529 @@ static void scrub_fixup(struct scrub_bio *sbio, int ix)
648 atomic_inc(&fs_info->scrubs_paused); 774 atomic_inc(&fs_info->scrubs_paused);
649 mutex_unlock(&fs_info->scrub_lock); 775 mutex_unlock(&fs_info->scrub_lock);
650 atomic_inc(&sdev->fixup_cnt); 776 atomic_inc(&sdev->fixup_cnt);
651 fixup->work.func = scrub_fixup_nodatasum; 777 fixup_nodatasum->work.func = scrub_fixup_nodatasum;
652 btrfs_queue_worker(&fs_info->scrub_workers, &fixup->work); 778 btrfs_queue_worker(&fs_info->scrub_workers,
653 return; 779 &fixup_nodatasum->work);
780 goto out;
654 } 781 }
655 782
656 length = PAGE_SIZE; 783 /*
657 ret = btrfs_map_block(map_tree, REQ_WRITE, logical, &length, 784 * now build and submit the bios for the other mirrors, check
658 &bbio, 0); 785 * checksums
659 if (ret || !bbio || length < PAGE_SIZE) { 786 */
660 printk(KERN_ERR 787 for (mirror_index = 0;
661 "scrub_fixup: btrfs_map_block failed us for %llu\n", 788 mirror_index < BTRFS_MAX_MIRRORS &&
662 (unsigned long long)logical); 789 sblocks_for_recheck[mirror_index].page_count > 0;
663 WARN_ON(1); 790 mirror_index++) {
664 kfree(bbio); 791 if (mirror_index == failed_mirror_index)
665 return; 792 continue;
793
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);
799 if (ret)
800 goto did_not_correct_error;
666 } 801 }
667 802
668 if (bbio->num_stripes == 1) 803 /*
669 /* there aren't any replicas */ 804 * first try to pick the mirror which is completely without I/O
670 goto uncorrectable; 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).
815 */
816 for (mirror_index = 0;
817 mirror_index < BTRFS_MAX_MIRRORS &&
818 sblocks_for_recheck[mirror_index].page_count > 0;
819 mirror_index++) {
820 struct scrub_block *sblock_other = sblocks_for_recheck +
821 mirror_index;
822
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;
827
828 ret = scrub_repair_block_from_good_copy(sblock_bad,
829 sblock_other,
830 force_write);
831 if (0 == ret)
832 goto corrected_error;
833 }
834 }
671 835
672 /* 836 /*
673 * first find a good copy 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.
674 */ 859 */
675 for (i = 0; i < bbio->num_stripes; ++i) {
676 if (i + 1 == sbio->spag[ix].mirror_num)
677 continue;
678 860
679 if (scrub_fixup_io(READ, bbio->stripes[i].dev->bdev, 861 /* can only fix I/O errors from here on */
680 bbio->stripes[i].physical >> 9, 862 if (sblock_bad->no_io_error_seen)
681 sbio->bio->bi_io_vec[ix].bv_page)) { 863 goto did_not_correct_error;
682 /* I/O-error, this is not a good copy */ 864
865 success = 1;
866 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
867 struct scrub_page *page_bad = sblock_bad->pagev + page_num;
868
869 if (!page_bad->io_error)
683 continue; 870 continue;
871
872 for (mirror_index = 0;
873 mirror_index < BTRFS_MAX_MIRRORS &&
874 sblocks_for_recheck[mirror_index].page_count > 0;
875 mirror_index++) {
876 struct scrub_block *sblock_other = sblocks_for_recheck +
877 mirror_index;
878 struct scrub_page *page_other = sblock_other->pagev +
879 page_num;
880
881 if (!page_other->io_error) {
882 ret = scrub_repair_page_from_good_copy(
883 sblock_bad, sblock_other, page_num, 0);
884 if (0 == ret) {
885 page_bad->io_error = 0;
886 break; /* succeeded for this page */
887 }
888 }
684 } 889 }
685 890
686 if (scrub_fixup_check(sbio, ix) == 0) 891 if (page_bad->io_error) {
687 break; 892 /* did not find a mirror to copy the page from */
893 success = 0;
894 }
688 } 895 }
689 if (i == bbio->num_stripes)
690 goto uncorrectable;
691 896
692 if (!sdev->readonly) { 897 if (success) {
693 /* 898 if (is_metadata || have_csum) {
694 * bi_io_vec[ix].bv_page now contains good data, write it back 899 /*
695 */ 900 * need to verify the checksum now that all
696 if (scrub_fixup_io(WRITE, sdev->dev->bdev, 901 * sectors on disk are repaired (the write
697 (sbio->physical + ix * PAGE_SIZE) >> 9, 902 * request for data to be repaired is on its way).
698 sbio->bio->bi_io_vec[ix].bv_page)) { 903 * Just be lazy and use scrub_recheck_block()
699 /* I/O-error, writeback failed, give up */ 904 * which re-reads the data before the checksum
700 goto uncorrectable; 905 * is verified, but most likely the data comes out
906 * of the page cache.
907 */
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;
915 else
916 goto did_not_correct_error;
917 } else {
918corrected_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);
701 } 925 }
926 } else {
927did_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);
702 } 934 }
703 935
704 kfree(bbio); 936out:
705 spin_lock(&sdev->stat_lock); 937 if (sblocks_for_recheck) {
706 ++sdev->stat.corrected_errors; 938 for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS;
707 spin_unlock(&sdev->stat_lock); 939 mirror_index++) {
940 struct scrub_block *sblock = sblocks_for_recheck +
941 mirror_index;
942 int page_index;
943
944 for (page_index = 0; page_index < SCRUB_PAGES_PER_BIO;
945 page_index++)
946 if (sblock->pagev[page_index].page)
947 __free_page(
948 sblock->pagev[page_index].page);
949 }
950 kfree(sblocks_for_recheck);
951 }
708 952
709 printk_ratelimited(KERN_ERR "btrfs: fixed up error at logical %llu\n", 953 return 0;
710 (unsigned long long)logical); 954}
711 return;
712 955
713uncorrectable: 956static int scrub_setup_recheck_block(struct scrub_dev *sdev,
714 kfree(bbio); 957 struct btrfs_mapping_tree *map_tree,
715 spin_lock(&sdev->stat_lock); 958 u64 length, u64 logical,
716 ++sdev->stat.uncorrectable_errors; 959 struct scrub_block *sblocks_for_recheck)
717 spin_unlock(&sdev->stat_lock); 960{
961 int page_index;
962 int mirror_index;
963 int ret;
964
965 /*
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
969 */
970
971 page_index = 0;
972 while (length > 0) {
973 u64 sublen = min_t(u64, length, PAGE_SIZE);
974 u64 mapped_length = sublen;
975 struct btrfs_bio *bbio = NULL;
718 976
719 printk_ratelimited(KERN_ERR "btrfs: unable to fixup (regular) error at " 977 /*
720 "logical %llu\n", (unsigned long long)logical); 978 * with a length of PAGE_SIZE, each returned stripe
979 * represents one mirror
980 */
981 ret = btrfs_map_block(map_tree, WRITE, logical, &mapped_length,
982 &bbio, 0);
983 if (ret || !bbio || mapped_length < sublen) {
984 kfree(bbio);
985 return -EIO;
986 }
987
988 BUG_ON(page_index >= SCRUB_PAGES_PER_BIO);
989 for (mirror_index = 0; mirror_index < (int)bbio->num_stripes;
990 mirror_index++) {
991 struct scrub_block *sblock;
992 struct scrub_page *page;
993
994 if (mirror_index >= BTRFS_MAX_MIRRORS)
995 continue;
996
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);
1004 if (!page->page) {
1005 spin_lock(&sdev->stat_lock);
1006 sdev->stat.malloc_errors++;
1007 spin_unlock(&sdev->stat_lock);
1008 return -ENOMEM;
1009 }
1010 sblock->page_count++;
1011 }
1012 kfree(bbio);
1013 length -= sublen;
1014 logical += sublen;
1015 page_index++;
1016 }
1017
1018 return 0;
721} 1019}
722 1020
723static int scrub_fixup_io(int rw, struct block_device *bdev, sector_t sector, 1021/*
724 struct page *page) 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.
1027 */
1028static 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,
1031 u16 csum_size)
725{ 1032{
726 struct bio *bio = NULL; 1033 int page_num;
727 int ret;
728 DECLARE_COMPLETION_ONSTACK(complete);
729 1034
730 bio = bio_alloc(GFP_NOFS, 1); 1035 sblock->no_io_error_seen = 1;
731 bio->bi_bdev = bdev; 1036 sblock->header_error = 0;
732 bio->bi_sector = sector; 1037 sblock->checksum_error = 0;
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);
737 1038
738 /* this will also unplug the queue */ 1039 for (page_num = 0; page_num < sblock->page_count; page_num++) {
739 wait_for_completion(&complete); 1040 struct bio *bio;
1041 int ret;
1042 struct scrub_page *page = sblock->pagev + page_num;
1043 DECLARE_COMPLETION_ONSTACK(complete);
1044
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;
1051
1052 ret = bio_add_page(bio, page->page, PAGE_SIZE, 0);
1053 if (PAGE_SIZE != ret) {
1054 bio_put(bio);
1055 return -EIO;
1056 }
1057 btrfsic_submit_bio(READ, bio);
740 1058
741 ret = !test_bit(BIO_UPTODATE, &bio->bi_flags); 1059 /* this will also unplug the queue */
742 bio_put(bio); 1060 wait_for_completion(&complete);
743 return ret; 1061
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;
1065 bio_put(bio);
1066 }
1067
1068 if (sblock->no_io_error_seen)
1069 scrub_recheck_block_checksum(fs_info, sblock, is_metadata,
1070 have_csum, csum, generation,
1071 csum_size);
1072
1073 return 0;
744} 1074}
745 1075
746static void scrub_bio_end_io(struct bio *bio, int err) 1076static 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,
1080 u16 csum_size)
747{ 1081{
748 struct scrub_bio *sbio = bio->bi_private; 1082 int page_num;
749 struct scrub_dev *sdev = sbio->sdev; 1083 u8 calculated_csum[BTRFS_CSUM_SIZE];
750 struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info; 1084 u32 crc = ~(u32)0;
1085 struct btrfs_root *root = fs_info->extent_root;
1086 void *mapped_buffer;
1087
1088 BUG_ON(!sblock->pagev[0].page);
1089 if (is_metadata) {
1090 struct btrfs_header *h;
1091
1092 mapped_buffer = kmap_atomic(sblock->pagev[0].page, KM_USER0);
1093 h = (struct btrfs_header *)mapped_buffer;
1094
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,
1099 BTRFS_UUID_SIZE))
1100 sblock->header_error = 1;
1101 csum = h->csum;
1102 } else {
1103 if (!have_csum)
1104 return;
751 1105
752 sbio->err = err; 1106 mapped_buffer = kmap_atomic(sblock->pagev[0].page, KM_USER0);
753 sbio->bio = bio; 1107 }
754 1108
755 btrfs_queue_worker(&fs_info->scrub_workers, &sbio->work); 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);
1114 else
1115 crc = btrfs_csum_data(root, mapped_buffer, crc,
1116 PAGE_SIZE);
1117
1118 kunmap_atomic(mapped_buffer, KM_USER0);
1119 page_num++;
1120 if (page_num >= sblock->page_count)
1121 break;
1122 BUG_ON(!sblock->pagev[page_num].page);
1123
1124 mapped_buffer = kmap_atomic(sblock->pagev[page_num].page,
1125 KM_USER0);
1126 }
1127
1128 btrfs_csum_final(crc, calculated_csum);
1129 if (memcmp(calculated_csum, csum, csum_size))
1130 sblock->checksum_error = 1;
756} 1131}
757 1132
758static void scrub_checksum(struct btrfs_work *work) 1133static void scrub_complete_bio_end_io(struct bio *bio, int err)
759{ 1134{
760 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work); 1135 complete((struct completion *)bio->bi_private);
761 struct scrub_dev *sdev = sbio->sdev; 1136}
762 struct page *page;
763 void *buffer;
764 int i;
765 u64 flags;
766 u64 logical;
767 int ret;
768 1137
769 if (sbio->err) { 1138static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
770 ret = 0; 1139 struct scrub_block *sblock_good,
771 for (i = 0; i < sbio->count; ++i) 1140 int force_write)
772 ret |= scrub_recheck_error(sbio, i); 1141{
773 if (!ret) { 1142 int page_num;
774 spin_lock(&sdev->stat_lock); 1143 int ret = 0;
775 ++sdev->stat.unverified_errors;
776 spin_unlock(&sdev->stat_lock);
777 }
778 1144
779 sbio->bio->bi_flags &= ~(BIO_POOL_MASK - 1); 1145 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
780 sbio->bio->bi_flags |= 1 << BIO_UPTODATE; 1146 int ret_sub;
781 sbio->bio->bi_phys_segments = 0;
782 sbio->bio->bi_idx = 0;
783 1147
784 for (i = 0; i < sbio->count; i++) { 1148 ret_sub = scrub_repair_page_from_good_copy(sblock_bad,
785 struct bio_vec *bi; 1149 sblock_good,
786 bi = &sbio->bio->bi_io_vec[i]; 1150 page_num,
787 bi->bv_offset = 0; 1151 force_write);
788 bi->bv_len = PAGE_SIZE; 1152 if (ret_sub)
789 } 1153 ret = ret_sub;
790 goto out;
791 } 1154 }
792 for (i = 0; i < sbio->count; ++i) { 1155
793 page = sbio->bio->bi_io_vec[i].bv_page; 1156 return ret;
794 buffer = kmap_atomic(page, KM_USER0); 1157}
795 flags = sbio->spag[i].flags; 1158
796 logical = sbio->logical + i * PAGE_SIZE; 1159static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
797 ret = 0; 1160 struct scrub_block *sblock_good,
798 if (flags & BTRFS_EXTENT_FLAG_DATA) { 1161 int page_num, int force_write)
799 ret = scrub_checksum_data(sdev, sbio->spag + i, buffer); 1162{
800 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 1163 struct scrub_page *page_bad = sblock_bad->pagev + page_num;
801 ret = scrub_checksum_tree_block(sdev, sbio->spag + i, 1164 struct scrub_page *page_good = sblock_good->pagev + page_num;
802 logical, buffer); 1165
803 } else if (flags & BTRFS_EXTENT_FLAG_SUPER) { 1166 BUG_ON(sblock_bad->pagev[page_num].page == NULL);
804 BUG_ON(i); 1167 BUG_ON(sblock_good->pagev[page_num].page == NULL);
805 (void)scrub_checksum_super(sbio, buffer); 1168 if (force_write || sblock_bad->header_error ||
806 } else { 1169 sblock_bad->checksum_error || page_bad->io_error) {
807 WARN_ON(1); 1170 struct bio *bio;
808 } 1171 int ret;
809 kunmap_atomic(buffer, KM_USER0); 1172 DECLARE_COMPLETION_ONSTACK(complete);
810 if (ret) { 1173
811 ret = scrub_recheck_error(sbio, i); 1174 bio = bio_alloc(GFP_NOFS, 1);
812 if (!ret) { 1175 bio->bi_bdev = page_bad->bdev;
813 spin_lock(&sdev->stat_lock); 1176 bio->bi_sector = page_bad->physical >> 9;
814 ++sdev->stat.unverified_errors; 1177 bio->bi_end_io = scrub_complete_bio_end_io;
815 spin_unlock(&sdev->stat_lock); 1178 bio->bi_private = &complete;
816 } 1179
1180 ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0);
1181 if (PAGE_SIZE != ret) {
1182 bio_put(bio);
1183 return -EIO;
817 } 1184 }
1185 btrfsic_submit_bio(WRITE, bio);
1186
1187 /* this will also unplug the queue */
1188 wait_for_completion(&complete);
1189 bio_put(bio);
818 } 1190 }
819 1191
820out: 1192 return 0;
821 scrub_free_bio(sbio->bio); 1193}
822 sbio->bio = NULL; 1194
823 spin_lock(&sdev->list_lock); 1195static void scrub_checksum(struct scrub_block *sblock)
824 sbio->next_free = sdev->first_free; 1196{
825 sdev->first_free = sbio->index; 1197 u64 flags;
826 spin_unlock(&sdev->list_lock); 1198 int ret;
827 atomic_dec(&sdev->in_flight); 1199
828 wake_up(&sdev->list_wait); 1200 BUG_ON(sblock->page_count < 1);
1201 flags = sblock->pagev[0].flags;
1202 ret = 0;
1203 if (flags & BTRFS_EXTENT_FLAG_DATA)
1204 ret = scrub_checksum_data(sblock);
1205 else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1206 ret = scrub_checksum_tree_block(sblock);
1207 else if (flags & BTRFS_EXTENT_FLAG_SUPER)
1208 (void)scrub_checksum_super(sblock);
1209 else
1210 WARN_ON(1);
1211 if (ret)
1212 scrub_handle_errored_block(sblock);
829} 1213}
830 1214
831static int scrub_checksum_data(struct scrub_dev *sdev, 1215static int scrub_checksum_data(struct scrub_block *sblock)
832 struct scrub_page *spag, void *buffer)
833{ 1216{
1217 struct scrub_dev *sdev = sblock->sdev;
834 u8 csum[BTRFS_CSUM_SIZE]; 1218 u8 csum[BTRFS_CSUM_SIZE];
1219 u8 *on_disk_csum;
1220 struct page *page;
1221 void *buffer;
835 u32 crc = ~(u32)0; 1222 u32 crc = ~(u32)0;
836 int fail = 0; 1223 int fail = 0;
837 struct btrfs_root *root = sdev->dev->dev_root; 1224 struct btrfs_root *root = sdev->dev->dev_root;
1225 u64 len;
1226 int index;
838 1227
839 if (!spag->have_csum) 1228 BUG_ON(sblock->page_count < 1);
1229 if (!sblock->pagev[0].have_csum)
840 return 0; 1230 return 0;
841 1231
842 crc = btrfs_csum_data(root, buffer, crc, PAGE_SIZE); 1232 on_disk_csum = sblock->pagev[0].csum;
1233 page = sblock->pagev[0].page;
1234 buffer = kmap_atomic(page, KM_USER0);
1235
1236 len = sdev->sectorsize;
1237 index = 0;
1238 for (;;) {
1239 u64 l = min_t(u64, len, PAGE_SIZE);
1240
1241 crc = btrfs_csum_data(root, buffer, crc, l);
1242 kunmap_atomic(buffer, KM_USER0);
1243 len -= l;
1244 if (len == 0)
1245 break;
1246 index++;
1247 BUG_ON(index >= sblock->page_count);
1248 BUG_ON(!sblock->pagev[index].page);
1249 page = sblock->pagev[index].page;
1250 buffer = kmap_atomic(page, KM_USER0);
1251 }
1252
843 btrfs_csum_final(crc, csum); 1253 btrfs_csum_final(crc, csum);
844 if (memcmp(csum, spag->csum, sdev->csum_size)) 1254 if (memcmp(csum, on_disk_csum, sdev->csum_size))
845 fail = 1; 1255 fail = 1;
846 1256
847 spin_lock(&sdev->stat_lock); 1257 if (fail) {
848 ++sdev->stat.data_extents_scrubbed; 1258 spin_lock(&sdev->stat_lock);
849 sdev->stat.data_bytes_scrubbed += PAGE_SIZE;
850 if (fail)
851 ++sdev->stat.csum_errors; 1259 ++sdev->stat.csum_errors;
852 spin_unlock(&sdev->stat_lock); 1260 spin_unlock(&sdev->stat_lock);
1261 }
853 1262
854 return fail; 1263 return fail;
855} 1264}
856 1265
857static int scrub_checksum_tree_block(struct scrub_dev *sdev, 1266static int scrub_checksum_tree_block(struct scrub_block *sblock)
858 struct scrub_page *spag, u64 logical,
859 void *buffer)
860{ 1267{
1268 struct scrub_dev *sdev = sblock->sdev;
861 struct btrfs_header *h; 1269 struct btrfs_header *h;
862 struct btrfs_root *root = sdev->dev->dev_root; 1270 struct btrfs_root *root = sdev->dev->dev_root;
863 struct btrfs_fs_info *fs_info = root->fs_info; 1271 struct btrfs_fs_info *fs_info = root->fs_info;
864 u8 csum[BTRFS_CSUM_SIZE]; 1272 u8 calculated_csum[BTRFS_CSUM_SIZE];
1273 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1274 struct page *page;
1275 void *mapped_buffer;
1276 u64 mapped_size;
1277 void *p;
865 u32 crc = ~(u32)0; 1278 u32 crc = ~(u32)0;
866 int fail = 0; 1279 int fail = 0;
867 int crc_fail = 0; 1280 int crc_fail = 0;
1281 u64 len;
1282 int index;
1283
1284 BUG_ON(sblock->page_count < 1);
1285 page = sblock->pagev[0].page;
1286 mapped_buffer = kmap_atomic(page, KM_USER0);
1287 h = (struct btrfs_header *)mapped_buffer;
1288 memcpy(on_disk_csum, h->csum, sdev->csum_size);
868 1289
869 /* 1290 /*
870 * we don't use the getter functions here, as we 1291 * we don't use the getter functions here, as we
871 * a) don't have an extent buffer and 1292 * a) don't have an extent buffer and
872 * b) the page is already kmapped 1293 * b) the page is already kmapped
873 */ 1294 */
874 h = (struct btrfs_header *)buffer;
875 1295
876 if (logical != le64_to_cpu(h->bytenr)) 1296 if (sblock->pagev[0].logical != le64_to_cpu(h->bytenr))
877 ++fail; 1297 ++fail;
878 1298
879 if (spag->generation != le64_to_cpu(h->generation)) 1299 if (sblock->pagev[0].generation != le64_to_cpu(h->generation))
880 ++fail; 1300 ++fail;
881 1301
882 if (memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE)) 1302 if (memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
@@ -886,51 +1306,99 @@ static int scrub_checksum_tree_block(struct scrub_dev *sdev,
886 BTRFS_UUID_SIZE)) 1306 BTRFS_UUID_SIZE))
887 ++fail; 1307 ++fail;
888 1308
889 crc = btrfs_csum_data(root, buffer + BTRFS_CSUM_SIZE, crc, 1309 BUG_ON(sdev->nodesize != sdev->leafsize);
890 PAGE_SIZE - BTRFS_CSUM_SIZE); 1310 len = sdev->nodesize - BTRFS_CSUM_SIZE;
891 btrfs_csum_final(crc, csum); 1311 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
892 if (memcmp(csum, h->csum, sdev->csum_size)) 1312 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
1313 index = 0;
1314 for (;;) {
1315 u64 l = min_t(u64, len, mapped_size);
1316
1317 crc = btrfs_csum_data(root, p, crc, l);
1318 kunmap_atomic(mapped_buffer, KM_USER0);
1319 len -= l;
1320 if (len == 0)
1321 break;
1322 index++;
1323 BUG_ON(index >= sblock->page_count);
1324 BUG_ON(!sblock->pagev[index].page);
1325 page = sblock->pagev[index].page;
1326 mapped_buffer = kmap_atomic(page, KM_USER0);
1327 mapped_size = PAGE_SIZE;
1328 p = mapped_buffer;
1329 }
1330
1331 btrfs_csum_final(crc, calculated_csum);
1332 if (memcmp(calculated_csum, on_disk_csum, sdev->csum_size))
893 ++crc_fail; 1333 ++crc_fail;
894 1334
895 spin_lock(&sdev->stat_lock); 1335 if (crc_fail || fail) {
896 ++sdev->stat.tree_extents_scrubbed; 1336 spin_lock(&sdev->stat_lock);
897 sdev->stat.tree_bytes_scrubbed += PAGE_SIZE; 1337 if (crc_fail)
898 if (crc_fail) 1338 ++sdev->stat.csum_errors;
899 ++sdev->stat.csum_errors; 1339 if (fail)
900 if (fail) 1340 ++sdev->stat.verify_errors;
901 ++sdev->stat.verify_errors; 1341 spin_unlock(&sdev->stat_lock);
902 spin_unlock(&sdev->stat_lock); 1342 }
903 1343
904 return fail || crc_fail; 1344 return fail || crc_fail;
905} 1345}
906 1346
907static int scrub_checksum_super(struct scrub_bio *sbio, void *buffer) 1347static int scrub_checksum_super(struct scrub_block *sblock)
908{ 1348{
909 struct btrfs_super_block *s; 1349 struct btrfs_super_block *s;
910 u64 logical; 1350 struct scrub_dev *sdev = sblock->sdev;
911 struct scrub_dev *sdev = sbio->sdev;
912 struct btrfs_root *root = sdev->dev->dev_root; 1351 struct btrfs_root *root = sdev->dev->dev_root;
913 struct btrfs_fs_info *fs_info = root->fs_info; 1352 struct btrfs_fs_info *fs_info = root->fs_info;
914 u8 csum[BTRFS_CSUM_SIZE]; 1353 u8 calculated_csum[BTRFS_CSUM_SIZE];
1354 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1355 struct page *page;
1356 void *mapped_buffer;
1357 u64 mapped_size;
1358 void *p;
915 u32 crc = ~(u32)0; 1359 u32 crc = ~(u32)0;
916 int fail = 0; 1360 int fail = 0;
1361 u64 len;
1362 int index;
917 1363
918 s = (struct btrfs_super_block *)buffer; 1364 BUG_ON(sblock->page_count < 1);
919 logical = sbio->logical; 1365 page = sblock->pagev[0].page;
1366 mapped_buffer = kmap_atomic(page, KM_USER0);
1367 s = (struct btrfs_super_block *)mapped_buffer;
1368 memcpy(on_disk_csum, s->csum, sdev->csum_size);
920 1369
921 if (logical != le64_to_cpu(s->bytenr)) 1370 if (sblock->pagev[0].logical != le64_to_cpu(s->bytenr))
922 ++fail; 1371 ++fail;
923 1372
924 if (sbio->spag[0].generation != le64_to_cpu(s->generation)) 1373 if (sblock->pagev[0].generation != le64_to_cpu(s->generation))
925 ++fail; 1374 ++fail;
926 1375
927 if (memcmp(s->fsid, fs_info->fsid, BTRFS_UUID_SIZE)) 1376 if (memcmp(s->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
928 ++fail; 1377 ++fail;
929 1378
930 crc = btrfs_csum_data(root, buffer + BTRFS_CSUM_SIZE, crc, 1379 len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE;
931 PAGE_SIZE - BTRFS_CSUM_SIZE); 1380 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
932 btrfs_csum_final(crc, csum); 1381 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
933 if (memcmp(csum, s->csum, sbio->sdev->csum_size)) 1382 index = 0;
1383 for (;;) {
1384 u64 l = min_t(u64, len, mapped_size);
1385
1386 crc = btrfs_csum_data(root, p, crc, l);
1387 kunmap_atomic(mapped_buffer, KM_USER0);
1388 len -= l;
1389 if (len == 0)
1390 break;
1391 index++;
1392 BUG_ON(index >= sblock->page_count);
1393 BUG_ON(!sblock->pagev[index].page);
1394 page = sblock->pagev[index].page;
1395 mapped_buffer = kmap_atomic(page, KM_USER0);
1396 mapped_size = PAGE_SIZE;
1397 p = mapped_buffer;
1398 }
1399
1400 btrfs_csum_final(crc, calculated_csum);
1401 if (memcmp(calculated_csum, on_disk_csum, sdev->csum_size))
934 ++fail; 1402 ++fail;
935 1403
936 if (fail) { 1404 if (fail) {
@@ -947,6 +1415,23 @@ static int scrub_checksum_super(struct scrub_bio *sbio, void *buffer)
947 return fail; 1415 return fail;
948} 1416}
949 1417
1418static void scrub_block_get(struct scrub_block *sblock)
1419{
1420 atomic_inc(&sblock->ref_count);
1421}
1422
1423static void scrub_block_put(struct scrub_block *sblock)
1424{
1425 if (atomic_dec_and_test(&sblock->ref_count)) {
1426 int i;
1427
1428 for (i = 0; i < sblock->page_count; i++)
1429 if (sblock->pagev[i].page)
1430 __free_page(sblock->pagev[i].page);
1431 kfree(sblock);
1432 }
1433}
1434
950static void scrub_submit(struct scrub_dev *sdev) 1435static void scrub_submit(struct scrub_dev *sdev)
951{ 1436{
952 struct scrub_bio *sbio; 1437 struct scrub_bio *sbio;
@@ -955,19 +1440,17 @@ static void scrub_submit(struct scrub_dev *sdev)
955 return; 1440 return;
956 1441
957 sbio = sdev->bios[sdev->curr]; 1442 sbio = sdev->bios[sdev->curr];
958 sbio->err = 0;
959 sdev->curr = -1; 1443 sdev->curr = -1;
960 atomic_inc(&sdev->in_flight); 1444 atomic_inc(&sdev->in_flight);
961 1445
962 btrfsic_submit_bio(READ, sbio->bio); 1446 btrfsic_submit_bio(READ, sbio->bio);
963} 1447}
964 1448
965static int scrub_page(struct scrub_dev *sdev, u64 logical, u64 len, 1449static int scrub_add_page_to_bio(struct scrub_dev *sdev,
966 u64 physical, u64 flags, u64 gen, int mirror_num, 1450 struct scrub_page *spage)
967 u8 *csum, int force)
968{ 1451{
1452 struct scrub_block *sblock = spage->sblock;
969 struct scrub_bio *sbio; 1453 struct scrub_bio *sbio;
970 struct page *page;
971 int ret; 1454 int ret;
972 1455
973again: 1456again:
@@ -980,7 +1463,7 @@ again:
980 if (sdev->curr != -1) { 1463 if (sdev->curr != -1) {
981 sdev->first_free = sdev->bios[sdev->curr]->next_free; 1464 sdev->first_free = sdev->bios[sdev->curr]->next_free;
982 sdev->bios[sdev->curr]->next_free = -1; 1465 sdev->bios[sdev->curr]->next_free = -1;
983 sdev->bios[sdev->curr]->count = 0; 1466 sdev->bios[sdev->curr]->page_count = 0;
984 spin_unlock(&sdev->list_lock); 1467 spin_unlock(&sdev->list_lock);
985 } else { 1468 } else {
986 spin_unlock(&sdev->list_lock); 1469 spin_unlock(&sdev->list_lock);
@@ -988,53 +1471,200 @@ again:
988 } 1471 }
989 } 1472 }
990 sbio = sdev->bios[sdev->curr]; 1473 sbio = sdev->bios[sdev->curr];
991 if (sbio->count == 0) { 1474 if (sbio->page_count == 0) {
992 struct bio *bio; 1475 struct bio *bio;
993 1476
994 sbio->physical = physical; 1477 sbio->physical = spage->physical;
995 sbio->logical = logical; 1478 sbio->logical = spage->logical;
996 bio = bio_alloc(GFP_NOFS, SCRUB_PAGES_PER_BIO); 1479 bio = sbio->bio;
997 if (!bio) 1480 if (!bio) {
998 return -ENOMEM; 1481 bio = bio_alloc(GFP_NOFS, sdev->pages_per_bio);
1482 if (!bio)
1483 return -ENOMEM;
1484 sbio->bio = bio;
1485 }
999 1486
1000 bio->bi_private = sbio; 1487 bio->bi_private = sbio;
1001 bio->bi_end_io = scrub_bio_end_io; 1488 bio->bi_end_io = scrub_bio_end_io;
1002 bio->bi_bdev = sdev->dev->bdev; 1489 bio->bi_bdev = sdev->dev->bdev;
1003 bio->bi_sector = sbio->physical >> 9; 1490 bio->bi_sector = spage->physical >> 9;
1004 sbio->err = 0; 1491 sbio->err = 0;
1005 sbio->bio = bio; 1492 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
1006 } else if (sbio->physical + sbio->count * PAGE_SIZE != physical || 1493 spage->physical ||
1007 sbio->logical + sbio->count * PAGE_SIZE != logical) { 1494 sbio->logical + sbio->page_count * PAGE_SIZE !=
1495 spage->logical) {
1008 scrub_submit(sdev); 1496 scrub_submit(sdev);
1009 goto again; 1497 goto again;
1010 } 1498 }
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;
1015
1016 page = alloc_page(GFP_NOFS);
1017 if (!page)
1018 return -ENOMEM;
1019 1499
1020 ret = bio_add_page(sbio->bio, page, PAGE_SIZE, 0); 1500 sbio->pagev[sbio->page_count] = spage;
1021 if (!ret) { 1501 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
1022 __free_page(page); 1502 if (ret != PAGE_SIZE) {
1503 if (sbio->page_count < 1) {
1504 bio_put(sbio->bio);
1505 sbio->bio = NULL;
1506 return -EIO;
1507 }
1023 scrub_submit(sdev); 1508 scrub_submit(sdev);
1024 goto again; 1509 goto again;
1025 } 1510 }
1026 1511
1027 if (csum) { 1512 scrub_block_get(sblock); /* one for the added page */
1028 sbio->spag[sbio->count].have_csum = 1; 1513 atomic_inc(&sblock->outstanding_pages);
1029 memcpy(sbio->spag[sbio->count].csum, csum, sdev->csum_size); 1514 sbio->page_count++;
1515 if (sbio->page_count == sdev->pages_per_bio)
1516 scrub_submit(sdev);
1517
1518 return 0;
1519}
1520
1521static int scrub_pages(struct scrub_dev *sdev, u64 logical, u64 len,
1522 u64 physical, u64 flags, u64 gen, int mirror_num,
1523 u8 *csum, int force)
1524{
1525 struct scrub_block *sblock;
1526 int index;
1527
1528 sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
1529 if (!sblock) {
1530 spin_lock(&sdev->stat_lock);
1531 sdev->stat.malloc_errors++;
1532 spin_unlock(&sdev->stat_lock);
1533 return -ENOMEM;
1534 }
1535
1536 /* one ref inside this function, plus one for each page later on */
1537 atomic_set(&sblock->ref_count, 1);
1538 sblock->sdev = sdev;
1539 sblock->no_io_error_seen = 1;
1540
1541 for (index = 0; len > 0; index++) {
1542 struct scrub_page *spage = sblock->pagev + index;
1543 u64 l = min_t(u64, len, PAGE_SIZE);
1544
1545 BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
1546 spage->page = alloc_page(GFP_NOFS);
1547 if (!spage->page) {
1548 spin_lock(&sdev->stat_lock);
1549 sdev->stat.malloc_errors++;
1550 spin_unlock(&sdev->stat_lock);
1551 while (index > 0) {
1552 index--;
1553 __free_page(sblock->pagev[index].page);
1554 }
1555 kfree(sblock);
1556 return -ENOMEM;
1557 }
1558 spage->sblock = sblock;
1559 spage->bdev = sdev->dev->bdev;
1560 spage->flags = flags;
1561 spage->generation = gen;
1562 spage->logical = logical;
1563 spage->physical = physical;
1564 spage->mirror_num = mirror_num;
1565 if (csum) {
1566 spage->have_csum = 1;
1567 memcpy(spage->csum, csum, sdev->csum_size);
1568 } else {
1569 spage->have_csum = 0;
1570 }
1571 sblock->page_count++;
1572 len -= l;
1573 logical += l;
1574 physical += l;
1575 }
1576
1577 BUG_ON(sblock->page_count == 0);
1578 for (index = 0; index < sblock->page_count; index++) {
1579 struct scrub_page *spage = sblock->pagev + index;
1580 int ret;
1581
1582 ret = scrub_add_page_to_bio(sdev, spage);
1583 if (ret) {
1584 scrub_block_put(sblock);
1585 return ret;
1586 }
1030 } 1587 }
1031 ++sbio->count; 1588
1032 if (sbio->count == SCRUB_PAGES_PER_BIO || force) 1589 if (force)
1033 scrub_submit(sdev); 1590 scrub_submit(sdev);
1034 1591
1592 /* last one frees, either here or in bio completion for last page */
1593 scrub_block_put(sblock);
1035 return 0; 1594 return 0;
1036} 1595}
1037 1596
1597static void scrub_bio_end_io(struct bio *bio, int err)
1598{
1599 struct scrub_bio *sbio = bio->bi_private;
1600 struct scrub_dev *sdev = sbio->sdev;
1601 struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info;
1602
1603 sbio->err = err;
1604 sbio->bio = bio;
1605
1606 btrfs_queue_worker(&fs_info->scrub_workers, &sbio->work);
1607}
1608
1609static void scrub_bio_end_io_worker(struct btrfs_work *work)
1610{
1611 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
1612 struct scrub_dev *sdev = sbio->sdev;
1613 int i;
1614
1615 BUG_ON(sbio->page_count > SCRUB_PAGES_PER_BIO);
1616 if (sbio->err) {
1617 for (i = 0; i < sbio->page_count; i++) {
1618 struct scrub_page *spage = sbio->pagev[i];
1619
1620 spage->io_error = 1;
1621 spage->sblock->no_io_error_seen = 0;
1622 }
1623 }
1624
1625 /* now complete the scrub_block items that have all pages completed */
1626 for (i = 0; i < sbio->page_count; i++) {
1627 struct scrub_page *spage = sbio->pagev[i];
1628 struct scrub_block *sblock = spage->sblock;
1629
1630 if (atomic_dec_and_test(&sblock->outstanding_pages))
1631 scrub_block_complete(sblock);
1632 scrub_block_put(sblock);
1633 }
1634
1635 if (sbio->err) {
1636 /* what is this good for??? */
1637 sbio->bio->bi_flags &= ~(BIO_POOL_MASK - 1);
1638 sbio->bio->bi_flags |= 1 << BIO_UPTODATE;
1639 sbio->bio->bi_phys_segments = 0;
1640 sbio->bio->bi_idx = 0;
1641
1642 for (i = 0; i < sbio->page_count; i++) {
1643 struct bio_vec *bi;
1644 bi = &sbio->bio->bi_io_vec[i];
1645 bi->bv_offset = 0;
1646 bi->bv_len = PAGE_SIZE;
1647 }
1648 }
1649
1650 bio_put(sbio->bio);
1651 sbio->bio = NULL;
1652 spin_lock(&sdev->list_lock);
1653 sbio->next_free = sdev->first_free;
1654 sdev->first_free = sbio->index;
1655 spin_unlock(&sdev->list_lock);
1656 atomic_dec(&sdev->in_flight);
1657 wake_up(&sdev->list_wait);
1658}
1659
1660static void scrub_block_complete(struct scrub_block *sblock)
1661{
1662 if (!sblock->no_io_error_seen)
1663 scrub_handle_errored_block(sblock);
1664 else
1665 scrub_checksum(sblock);
1666}
1667
1038static int scrub_find_csum(struct scrub_dev *sdev, u64 logical, u64 len, 1668static int scrub_find_csum(struct scrub_dev *sdev, u64 logical, u64 len,
1039 u8 *csum) 1669 u8 *csum)
1040{ 1670{
@@ -1042,7 +1672,6 @@ static int scrub_find_csum(struct scrub_dev *sdev, u64 logical, u64 len,
1042 int ret = 0; 1672 int ret = 0;
1043 unsigned long i; 1673 unsigned long i;
1044 unsigned long num_sectors; 1674 unsigned long num_sectors;
1045 u32 sectorsize = sdev->dev->dev_root->sectorsize;
1046 1675
1047 while (!list_empty(&sdev->csum_list)) { 1676 while (!list_empty(&sdev->csum_list)) {
1048 sum = list_first_entry(&sdev->csum_list, 1677 sum = list_first_entry(&sdev->csum_list,
@@ -1060,7 +1689,7 @@ static int scrub_find_csum(struct scrub_dev *sdev, u64 logical, u64 len,
1060 if (!sum) 1689 if (!sum)
1061 return 0; 1690 return 0;
1062 1691
1063 num_sectors = sum->len / sectorsize; 1692 num_sectors = sum->len / sdev->sectorsize;
1064 for (i = 0; i < num_sectors; ++i) { 1693 for (i = 0; i < num_sectors; ++i) {
1065 if (sum->sums[i].bytenr == logical) { 1694 if (sum->sums[i].bytenr == logical) {
1066 memcpy(csum, &sum->sums[i].sum, sdev->csum_size); 1695 memcpy(csum, &sum->sums[i].sum, sdev->csum_size);
@@ -1081,9 +1710,28 @@ static int scrub_extent(struct scrub_dev *sdev, u64 logical, u64 len,
1081{ 1710{
1082 int ret; 1711 int ret;
1083 u8 csum[BTRFS_CSUM_SIZE]; 1712 u8 csum[BTRFS_CSUM_SIZE];
1713 u32 blocksize;
1714
1715 if (flags & BTRFS_EXTENT_FLAG_DATA) {
1716 blocksize = sdev->sectorsize;
1717 spin_lock(&sdev->stat_lock);
1718 sdev->stat.data_extents_scrubbed++;
1719 sdev->stat.data_bytes_scrubbed += len;
1720 spin_unlock(&sdev->stat_lock);
1721 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1722 BUG_ON(sdev->nodesize != sdev->leafsize);
1723 blocksize = sdev->nodesize;
1724 spin_lock(&sdev->stat_lock);
1725 sdev->stat.tree_extents_scrubbed++;
1726 sdev->stat.tree_bytes_scrubbed += len;
1727 spin_unlock(&sdev->stat_lock);
1728 } else {
1729 blocksize = sdev->sectorsize;
1730 BUG_ON(1);
1731 }
1084 1732
1085 while (len) { 1733 while (len) {
1086 u64 l = min_t(u64, len, PAGE_SIZE); 1734 u64 l = min_t(u64, len, blocksize);
1087 int have_csum = 0; 1735 int have_csum = 0;
1088 1736
1089 if (flags & BTRFS_EXTENT_FLAG_DATA) { 1737 if (flags & BTRFS_EXTENT_FLAG_DATA) {
@@ -1092,8 +1740,8 @@ static int scrub_extent(struct scrub_dev *sdev, u64 logical, u64 len,
1092 if (have_csum == 0) 1740 if (have_csum == 0)
1093 ++sdev->stat.no_csum; 1741 ++sdev->stat.no_csum;
1094 } 1742 }
1095 ret = scrub_page(sdev, logical, l, physical, flags, gen, 1743 ret = scrub_pages(sdev, logical, l, physical, flags, gen,
1096 mirror_num, have_csum ? csum : NULL, 0); 1744 mirror_num, have_csum ? csum : NULL, 0);
1097 if (ret) 1745 if (ret)
1098 return ret; 1746 return ret;
1099 len -= l; 1747 len -= l;
@@ -1158,6 +1806,11 @@ static noinline_for_stack int scrub_stripe(struct scrub_dev *sdev,
1158 if (!path) 1806 if (!path)
1159 return -ENOMEM; 1807 return -ENOMEM;
1160 1808
1809 /*
1810 * work on commit root. The related disk blocks are static as
1811 * long as COW is applied. This means, it is save to rewrite
1812 * them to repair disk errors without any race conditions
1813 */
1161 path->search_commit_root = 1; 1814 path->search_commit_root = 1;
1162 path->skip_locking = 1; 1815 path->skip_locking = 1;
1163 1816
@@ -1511,8 +2164,8 @@ static noinline_for_stack int scrub_supers(struct scrub_dev *sdev)
1511 if (bytenr + BTRFS_SUPER_INFO_SIZE > device->total_bytes) 2164 if (bytenr + BTRFS_SUPER_INFO_SIZE > device->total_bytes)
1512 break; 2165 break;
1513 2166
1514 ret = scrub_page(sdev, bytenr, PAGE_SIZE, bytenr, 2167 ret = scrub_pages(sdev, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr,
1515 BTRFS_EXTENT_FLAG_SUPER, gen, i, NULL, 1); 2168 BTRFS_EXTENT_FLAG_SUPER, gen, i, NULL, 1);
1516 if (ret) 2169 if (ret)
1517 return ret; 2170 return ret;
1518 } 2171 }
@@ -1571,10 +2224,30 @@ int btrfs_scrub_dev(struct btrfs_root *root, u64 devid, u64 start, u64 end,
1571 /* 2224 /*
1572 * check some assumptions 2225 * check some assumptions
1573 */ 2226 */
1574 if (root->sectorsize != PAGE_SIZE || 2227 if (root->nodesize != root->leafsize) {
1575 root->sectorsize != root->leafsize || 2228 printk(KERN_ERR
1576 root->sectorsize != root->nodesize) { 2229 "btrfs_scrub: size assumption nodesize == leafsize (%d == %d) fails\n",
1577 printk(KERN_ERR "btrfs_scrub: size assumptions fail\n"); 2230 root->nodesize, root->leafsize);
2231 return -EINVAL;
2232 }
2233
2234 if (root->nodesize > BTRFS_STRIPE_LEN) {
2235 /*
2236 * in this case scrub is unable to calculate the checksum
2237 * the way scrub is implemented. Do not handle this
2238 * situation at all because it won't ever happen.
2239 */
2240 printk(KERN_ERR
2241 "btrfs_scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails\n",
2242 root->nodesize, BTRFS_STRIPE_LEN);
2243 return -EINVAL;
2244 }
2245
2246 if (root->sectorsize != PAGE_SIZE) {
2247 /* not supported for data w/o checksums */
2248 printk(KERN_ERR
2249 "btrfs_scrub: size assumption sectorsize != PAGE_SIZE (%d != %lld) fails\n",
2250 root->sectorsize, (unsigned long long)PAGE_SIZE);
1578 return -EINVAL; 2251 return -EINVAL;
1579 } 2252 }
1580 2253