btrfs scrub: added unverified_errors
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / btrfs / scrub.c
blob35099fa97d564433b17ec5e7f435f3ca87d38dd3
1 /*
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 "ctree.h"
21 #include "volumes.h"
22 #include "disk-io.h"
23 #include "ordered-data.h"
26 * This is only the first step towards a full-features scrub. It reads all
27 * extent and super block and verifies the checksums. In case a bad checksum
28 * is found or the extent cannot be read, good data will be written back if
29 * any can be found.
31 * Future enhancements:
32 * - To enhance the performance, better read-ahead strategies for the
33 * extent-tree can be employed.
34 * - In case an unrepairable extent is encountered, track which files are
35 * affected and report them
36 * - In case of a read error on files with nodatasum, map the file and read
37 * the extent to trigger a writeback of the good copy
38 * - track and record media errors, throw out bad devices
39 * - add a mode to also read unallocated space
40 * - make the prefetch cancellable
43 struct scrub_bio;
44 struct scrub_page;
45 struct scrub_dev;
46 static void scrub_bio_end_io(struct bio *bio, int err);
47 static void scrub_checksum(struct btrfs_work *work);
48 static int scrub_checksum_data(struct scrub_dev *sdev,
49 struct scrub_page *spag, void *buffer);
50 static int scrub_checksum_tree_block(struct scrub_dev *sdev,
51 struct scrub_page *spag, u64 logical,
52 void *buffer);
53 static int scrub_checksum_super(struct scrub_bio *sbio, void *buffer);
54 static int scrub_fixup_check(struct scrub_bio *sbio, int ix);
55 static void scrub_fixup_end_io(struct bio *bio, int err);
56 static int scrub_fixup_io(int rw, struct block_device *bdev, sector_t sector,
57 struct page *page);
58 static void scrub_fixup(struct scrub_bio *sbio, int ix);
60 #define SCRUB_PAGES_PER_BIO 16 /* 64k per bio */
61 #define SCRUB_BIOS_PER_DEV 16 /* 1 MB per device in flight */
63 struct scrub_page {
64 u64 flags; /* extent flags */
65 u64 generation;
66 u64 mirror_num;
67 int have_csum;
68 u8 csum[BTRFS_CSUM_SIZE];
71 struct scrub_bio {
72 int index;
73 struct scrub_dev *sdev;
74 struct bio *bio;
75 int err;
76 u64 logical;
77 u64 physical;
78 struct scrub_page spag[SCRUB_PAGES_PER_BIO];
79 u64 count;
80 int next_free;
81 struct btrfs_work work;
84 struct scrub_dev {
85 struct scrub_bio *bios[SCRUB_BIOS_PER_DEV];
86 struct btrfs_device *dev;
87 int first_free;
88 int curr;
89 atomic_t in_flight;
90 spinlock_t list_lock;
91 wait_queue_head_t list_wait;
92 u16 csum_size;
93 struct list_head csum_list;
94 atomic_t cancel_req;
95 int readonly;
97 * statistics
99 struct btrfs_scrub_progress stat;
100 spinlock_t stat_lock;
103 static void scrub_free_csums(struct scrub_dev *sdev)
105 while (!list_empty(&sdev->csum_list)) {
106 struct btrfs_ordered_sum *sum;
107 sum = list_first_entry(&sdev->csum_list,
108 struct btrfs_ordered_sum, list);
109 list_del(&sum->list);
110 kfree(sum);
114 static void scrub_free_bio(struct bio *bio)
116 int i;
117 struct page *last_page = NULL;
119 if (!bio)
120 return;
122 for (i = 0; i < bio->bi_vcnt; ++i) {
123 if (bio->bi_io_vec[i].bv_page == last_page)
124 continue;
125 last_page = bio->bi_io_vec[i].bv_page;
126 __free_page(last_page);
128 bio_put(bio);
131 static noinline_for_stack void scrub_free_dev(struct scrub_dev *sdev)
133 int i;
135 if (!sdev)
136 return;
138 for (i = 0; i < SCRUB_BIOS_PER_DEV; ++i) {
139 struct scrub_bio *sbio = sdev->bios[i];
141 if (!sbio)
142 break;
144 scrub_free_bio(sbio->bio);
145 kfree(sbio);
148 scrub_free_csums(sdev);
149 kfree(sdev);
152 static noinline_for_stack
153 struct scrub_dev *scrub_setup_dev(struct btrfs_device *dev)
155 struct scrub_dev *sdev;
156 int i;
157 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
159 sdev = kzalloc(sizeof(*sdev), GFP_NOFS);
160 if (!sdev)
161 goto nomem;
162 sdev->dev = dev;
163 for (i = 0; i < SCRUB_BIOS_PER_DEV; ++i) {
164 struct scrub_bio *sbio;
166 sbio = kzalloc(sizeof(*sbio), GFP_NOFS);
167 if (!sbio)
168 goto nomem;
169 sdev->bios[i] = sbio;
171 sbio->index = i;
172 sbio->sdev = sdev;
173 sbio->count = 0;
174 sbio->work.func = scrub_checksum;
176 if (i != SCRUB_BIOS_PER_DEV-1)
177 sdev->bios[i]->next_free = i + 1;
178 else
179 sdev->bios[i]->next_free = -1;
181 sdev->first_free = 0;
182 sdev->curr = -1;
183 atomic_set(&sdev->in_flight, 0);
184 atomic_set(&sdev->cancel_req, 0);
185 sdev->csum_size = btrfs_super_csum_size(&fs_info->super_copy);
186 INIT_LIST_HEAD(&sdev->csum_list);
188 spin_lock_init(&sdev->list_lock);
189 spin_lock_init(&sdev->stat_lock);
190 init_waitqueue_head(&sdev->list_wait);
191 return sdev;
193 nomem:
194 scrub_free_dev(sdev);
195 return ERR_PTR(-ENOMEM);
199 * scrub_recheck_error gets called when either verification of the page
200 * failed or the bio failed to read, e.g. with EIO. In the latter case,
201 * recheck_error gets called for every page in the bio, even though only
202 * one may be bad
204 static int scrub_recheck_error(struct scrub_bio *sbio, int ix)
206 struct scrub_dev *sdev = sbio->sdev;
207 u64 sector = (sbio->physical + ix * PAGE_SIZE) >> 9;
209 if (sbio->err) {
210 if (scrub_fixup_io(READ, sbio->sdev->dev->bdev, sector,
211 sbio->bio->bi_io_vec[ix].bv_page) == 0) {
212 if (scrub_fixup_check(sbio, ix) == 0)
213 return 0;
217 spin_lock(&sdev->stat_lock);
218 ++sdev->stat.read_errors;
219 spin_unlock(&sdev->stat_lock);
221 scrub_fixup(sbio, ix);
222 return 1;
225 static int scrub_fixup_check(struct scrub_bio *sbio, int ix)
227 int ret = 1;
228 struct page *page;
229 void *buffer;
230 u64 flags = sbio->spag[ix].flags;
232 page = sbio->bio->bi_io_vec[ix].bv_page;
233 buffer = kmap_atomic(page, KM_USER0);
234 if (flags & BTRFS_EXTENT_FLAG_DATA) {
235 ret = scrub_checksum_data(sbio->sdev,
236 sbio->spag + ix, buffer);
237 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
238 ret = scrub_checksum_tree_block(sbio->sdev,
239 sbio->spag + ix,
240 sbio->logical + ix * PAGE_SIZE,
241 buffer);
242 } else {
243 WARN_ON(1);
245 kunmap_atomic(buffer, KM_USER0);
247 return ret;
250 static void scrub_fixup_end_io(struct bio *bio, int err)
252 complete((struct completion *)bio->bi_private);
255 static void scrub_fixup(struct scrub_bio *sbio, int ix)
257 struct scrub_dev *sdev = sbio->sdev;
258 struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info;
259 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
260 struct btrfs_multi_bio *multi = NULL;
261 u64 logical = sbio->logical + ix * PAGE_SIZE;
262 u64 length;
263 int i;
264 int ret;
265 DECLARE_COMPLETION_ONSTACK(complete);
267 if ((sbio->spag[ix].flags & BTRFS_EXTENT_FLAG_DATA) &&
268 (sbio->spag[ix].have_csum == 0)) {
270 * nodatasum, don't try to fix anything
271 * FIXME: we can do better, open the inode and trigger a
272 * writeback
274 goto uncorrectable;
277 length = PAGE_SIZE;
278 ret = btrfs_map_block(map_tree, REQ_WRITE, logical, &length,
279 &multi, 0);
280 if (ret || !multi || length < PAGE_SIZE) {
281 printk(KERN_ERR
282 "scrub_fixup: btrfs_map_block failed us for %llu\n",
283 (unsigned long long)logical);
284 WARN_ON(1);
285 return;
288 if (multi->num_stripes == 1)
289 /* there aren't any replicas */
290 goto uncorrectable;
293 * first find a good copy
295 for (i = 0; i < multi->num_stripes; ++i) {
296 if (i == sbio->spag[ix].mirror_num)
297 continue;
299 if (scrub_fixup_io(READ, multi->stripes[i].dev->bdev,
300 multi->stripes[i].physical >> 9,
301 sbio->bio->bi_io_vec[ix].bv_page)) {
302 /* I/O-error, this is not a good copy */
303 continue;
306 if (scrub_fixup_check(sbio, ix) == 0)
307 break;
309 if (i == multi->num_stripes)
310 goto uncorrectable;
312 if (!sdev->readonly) {
314 * bi_io_vec[ix].bv_page now contains good data, write it back
316 if (scrub_fixup_io(WRITE, sdev->dev->bdev,
317 (sbio->physical + ix * PAGE_SIZE) >> 9,
318 sbio->bio->bi_io_vec[ix].bv_page)) {
319 /* I/O-error, writeback failed, give up */
320 goto uncorrectable;
324 kfree(multi);
325 spin_lock(&sdev->stat_lock);
326 ++sdev->stat.corrected_errors;
327 spin_unlock(&sdev->stat_lock);
329 if (printk_ratelimit())
330 printk(KERN_ERR "btrfs: fixed up at %llu\n",
331 (unsigned long long)logical);
332 return;
334 uncorrectable:
335 kfree(multi);
336 spin_lock(&sdev->stat_lock);
337 ++sdev->stat.uncorrectable_errors;
338 spin_unlock(&sdev->stat_lock);
340 if (printk_ratelimit())
341 printk(KERN_ERR "btrfs: unable to fixup at %llu\n",
342 (unsigned long long)logical);
345 static int scrub_fixup_io(int rw, struct block_device *bdev, sector_t sector,
346 struct page *page)
348 struct bio *bio = NULL;
349 int ret;
350 DECLARE_COMPLETION_ONSTACK(complete);
352 bio = bio_alloc(GFP_NOFS, 1);
353 bio->bi_bdev = bdev;
354 bio->bi_sector = sector;
355 bio_add_page(bio, page, PAGE_SIZE, 0);
356 bio->bi_end_io = scrub_fixup_end_io;
357 bio->bi_private = &complete;
358 submit_bio(rw, bio);
360 /* this will also unplug the queue */
361 wait_for_completion(&complete);
363 ret = !test_bit(BIO_UPTODATE, &bio->bi_flags);
364 bio_put(bio);
365 return ret;
368 static void scrub_bio_end_io(struct bio *bio, int err)
370 struct scrub_bio *sbio = bio->bi_private;
371 struct scrub_dev *sdev = sbio->sdev;
372 struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info;
374 sbio->err = err;
375 sbio->bio = bio;
377 btrfs_queue_worker(&fs_info->scrub_workers, &sbio->work);
380 static void scrub_checksum(struct btrfs_work *work)
382 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
383 struct scrub_dev *sdev = sbio->sdev;
384 struct page *page;
385 void *buffer;
386 int i;
387 u64 flags;
388 u64 logical;
389 int ret;
391 if (sbio->err) {
392 ret = 0;
393 for (i = 0; i < sbio->count; ++i)
394 ret |= scrub_recheck_error(sbio, i);
395 if (!ret) {
396 spin_lock(&sdev->stat_lock);
397 ++sdev->stat.unverified_errors;
398 spin_unlock(&sdev->stat_lock);
401 sbio->bio->bi_flags &= ~(BIO_POOL_MASK - 1);
402 sbio->bio->bi_flags |= 1 << BIO_UPTODATE;
403 sbio->bio->bi_phys_segments = 0;
404 sbio->bio->bi_idx = 0;
406 for (i = 0; i < sbio->count; i++) {
407 struct bio_vec *bi;
408 bi = &sbio->bio->bi_io_vec[i];
409 bi->bv_offset = 0;
410 bi->bv_len = PAGE_SIZE;
412 goto out;
414 for (i = 0; i < sbio->count; ++i) {
415 page = sbio->bio->bi_io_vec[i].bv_page;
416 buffer = kmap_atomic(page, KM_USER0);
417 flags = sbio->spag[i].flags;
418 logical = sbio->logical + i * PAGE_SIZE;
419 ret = 0;
420 if (flags & BTRFS_EXTENT_FLAG_DATA) {
421 ret = scrub_checksum_data(sdev, sbio->spag + i, buffer);
422 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
423 ret = scrub_checksum_tree_block(sdev, sbio->spag + i,
424 logical, buffer);
425 } else if (flags & BTRFS_EXTENT_FLAG_SUPER) {
426 BUG_ON(i);
427 (void)scrub_checksum_super(sbio, buffer);
428 } else {
429 WARN_ON(1);
431 kunmap_atomic(buffer, KM_USER0);
432 if (ret) {
433 ret = scrub_recheck_error(sbio, i);
434 if (!ret) {
435 spin_lock(&sdev->stat_lock);
436 ++sdev->stat.unverified_errors;
437 spin_unlock(&sdev->stat_lock);
442 out:
443 scrub_free_bio(sbio->bio);
444 sbio->bio = NULL;
445 spin_lock(&sdev->list_lock);
446 sbio->next_free = sdev->first_free;
447 sdev->first_free = sbio->index;
448 spin_unlock(&sdev->list_lock);
449 atomic_dec(&sdev->in_flight);
450 wake_up(&sdev->list_wait);
453 static int scrub_checksum_data(struct scrub_dev *sdev,
454 struct scrub_page *spag, void *buffer)
456 u8 csum[BTRFS_CSUM_SIZE];
457 u32 crc = ~(u32)0;
458 int fail = 0;
459 struct btrfs_root *root = sdev->dev->dev_root;
461 if (!spag->have_csum)
462 return 0;
464 crc = btrfs_csum_data(root, buffer, crc, PAGE_SIZE);
465 btrfs_csum_final(crc, csum);
466 if (memcmp(csum, spag->csum, sdev->csum_size))
467 fail = 1;
469 spin_lock(&sdev->stat_lock);
470 ++sdev->stat.data_extents_scrubbed;
471 sdev->stat.data_bytes_scrubbed += PAGE_SIZE;
472 if (fail)
473 ++sdev->stat.csum_errors;
474 spin_unlock(&sdev->stat_lock);
476 return fail;
479 static int scrub_checksum_tree_block(struct scrub_dev *sdev,
480 struct scrub_page *spag, u64 logical,
481 void *buffer)
483 struct btrfs_header *h;
484 struct btrfs_root *root = sdev->dev->dev_root;
485 struct btrfs_fs_info *fs_info = root->fs_info;
486 u8 csum[BTRFS_CSUM_SIZE];
487 u32 crc = ~(u32)0;
488 int fail = 0;
489 int crc_fail = 0;
492 * we don't use the getter functions here, as we
493 * a) don't have an extent buffer and
494 * b) the page is already kmapped
496 h = (struct btrfs_header *)buffer;
498 if (logical != le64_to_cpu(h->bytenr))
499 ++fail;
501 if (spag->generation != le64_to_cpu(h->generation))
502 ++fail;
504 if (memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
505 ++fail;
507 if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
508 BTRFS_UUID_SIZE))
509 ++fail;
511 crc = btrfs_csum_data(root, buffer + BTRFS_CSUM_SIZE, crc,
512 PAGE_SIZE - BTRFS_CSUM_SIZE);
513 btrfs_csum_final(crc, csum);
514 if (memcmp(csum, h->csum, sdev->csum_size))
515 ++crc_fail;
517 spin_lock(&sdev->stat_lock);
518 ++sdev->stat.tree_extents_scrubbed;
519 sdev->stat.tree_bytes_scrubbed += PAGE_SIZE;
520 if (crc_fail)
521 ++sdev->stat.csum_errors;
522 if (fail)
523 ++sdev->stat.verify_errors;
524 spin_unlock(&sdev->stat_lock);
526 return fail || crc_fail;
529 static int scrub_checksum_super(struct scrub_bio *sbio, void *buffer)
531 struct btrfs_super_block *s;
532 u64 logical;
533 struct scrub_dev *sdev = sbio->sdev;
534 struct btrfs_root *root = sdev->dev->dev_root;
535 struct btrfs_fs_info *fs_info = root->fs_info;
536 u8 csum[BTRFS_CSUM_SIZE];
537 u32 crc = ~(u32)0;
538 int fail = 0;
540 s = (struct btrfs_super_block *)buffer;
541 logical = sbio->logical;
543 if (logical != le64_to_cpu(s->bytenr))
544 ++fail;
546 if (sbio->spag[0].generation != le64_to_cpu(s->generation))
547 ++fail;
549 if (memcmp(s->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
550 ++fail;
552 crc = btrfs_csum_data(root, buffer + BTRFS_CSUM_SIZE, crc,
553 PAGE_SIZE - BTRFS_CSUM_SIZE);
554 btrfs_csum_final(crc, csum);
555 if (memcmp(csum, s->csum, sbio->sdev->csum_size))
556 ++fail;
558 if (fail) {
560 * if we find an error in a super block, we just report it.
561 * They will get written with the next transaction commit
562 * anyway
564 spin_lock(&sdev->stat_lock);
565 ++sdev->stat.super_errors;
566 spin_unlock(&sdev->stat_lock);
569 return fail;
572 static int scrub_submit(struct scrub_dev *sdev)
574 struct scrub_bio *sbio;
575 struct bio *bio;
576 int i;
578 if (sdev->curr == -1)
579 return 0;
581 sbio = sdev->bios[sdev->curr];
583 bio = bio_alloc(GFP_NOFS, sbio->count);
584 if (!bio)
585 goto nomem;
587 bio->bi_private = sbio;
588 bio->bi_end_io = scrub_bio_end_io;
589 bio->bi_bdev = sdev->dev->bdev;
590 bio->bi_sector = sbio->physical >> 9;
592 for (i = 0; i < sbio->count; ++i) {
593 struct page *page;
594 int ret;
596 page = alloc_page(GFP_NOFS);
597 if (!page)
598 goto nomem;
600 ret = bio_add_page(bio, page, PAGE_SIZE, 0);
601 if (!ret) {
602 __free_page(page);
603 goto nomem;
607 sbio->err = 0;
608 sdev->curr = -1;
609 atomic_inc(&sdev->in_flight);
611 submit_bio(READ, bio);
613 return 0;
615 nomem:
616 scrub_free_bio(bio);
618 return -ENOMEM;
621 static int scrub_page(struct scrub_dev *sdev, u64 logical, u64 len,
622 u64 physical, u64 flags, u64 gen, u64 mirror_num,
623 u8 *csum, int force)
625 struct scrub_bio *sbio;
627 again:
629 * grab a fresh bio or wait for one to become available
631 while (sdev->curr == -1) {
632 spin_lock(&sdev->list_lock);
633 sdev->curr = sdev->first_free;
634 if (sdev->curr != -1) {
635 sdev->first_free = sdev->bios[sdev->curr]->next_free;
636 sdev->bios[sdev->curr]->next_free = -1;
637 sdev->bios[sdev->curr]->count = 0;
638 spin_unlock(&sdev->list_lock);
639 } else {
640 spin_unlock(&sdev->list_lock);
641 wait_event(sdev->list_wait, sdev->first_free != -1);
644 sbio = sdev->bios[sdev->curr];
645 if (sbio->count == 0) {
646 sbio->physical = physical;
647 sbio->logical = logical;
648 } else if (sbio->physical + sbio->count * PAGE_SIZE != physical ||
649 sbio->logical + sbio->count * PAGE_SIZE != logical) {
650 int ret;
652 ret = scrub_submit(sdev);
653 if (ret)
654 return ret;
655 goto again;
657 sbio->spag[sbio->count].flags = flags;
658 sbio->spag[sbio->count].generation = gen;
659 sbio->spag[sbio->count].have_csum = 0;
660 sbio->spag[sbio->count].mirror_num = mirror_num;
661 if (csum) {
662 sbio->spag[sbio->count].have_csum = 1;
663 memcpy(sbio->spag[sbio->count].csum, csum, sdev->csum_size);
665 ++sbio->count;
666 if (sbio->count == SCRUB_PAGES_PER_BIO || force) {
667 int ret;
669 ret = scrub_submit(sdev);
670 if (ret)
671 return ret;
674 return 0;
677 static int scrub_find_csum(struct scrub_dev *sdev, u64 logical, u64 len,
678 u8 *csum)
680 struct btrfs_ordered_sum *sum = NULL;
681 int ret = 0;
682 unsigned long i;
683 unsigned long num_sectors;
684 u32 sectorsize = sdev->dev->dev_root->sectorsize;
686 while (!list_empty(&sdev->csum_list)) {
687 sum = list_first_entry(&sdev->csum_list,
688 struct btrfs_ordered_sum, list);
689 if (sum->bytenr > logical)
690 return 0;
691 if (sum->bytenr + sum->len > logical)
692 break;
694 ++sdev->stat.csum_discards;
695 list_del(&sum->list);
696 kfree(sum);
697 sum = NULL;
699 if (!sum)
700 return 0;
702 num_sectors = sum->len / sectorsize;
703 for (i = 0; i < num_sectors; ++i) {
704 if (sum->sums[i].bytenr == logical) {
705 memcpy(csum, &sum->sums[i].sum, sdev->csum_size);
706 ret = 1;
707 break;
710 if (ret && i == num_sectors - 1) {
711 list_del(&sum->list);
712 kfree(sum);
714 return ret;
717 /* scrub extent tries to collect up to 64 kB for each bio */
718 static int scrub_extent(struct scrub_dev *sdev, u64 logical, u64 len,
719 u64 physical, u64 flags, u64 gen, u64 mirror_num)
721 int ret;
722 u8 csum[BTRFS_CSUM_SIZE];
724 while (len) {
725 u64 l = min_t(u64, len, PAGE_SIZE);
726 int have_csum = 0;
728 if (flags & BTRFS_EXTENT_FLAG_DATA) {
729 /* push csums to sbio */
730 have_csum = scrub_find_csum(sdev, logical, l, csum);
731 if (have_csum == 0)
732 ++sdev->stat.no_csum;
734 ret = scrub_page(sdev, logical, l, physical, flags, gen,
735 mirror_num, have_csum ? csum : NULL, 0);
736 if (ret)
737 return ret;
738 len -= l;
739 logical += l;
740 physical += l;
742 return 0;
745 static noinline_for_stack int scrub_stripe(struct scrub_dev *sdev,
746 struct map_lookup *map, int num, u64 base, u64 length)
748 struct btrfs_path *path;
749 struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info;
750 struct btrfs_root *root = fs_info->extent_root;
751 struct btrfs_root *csum_root = fs_info->csum_root;
752 struct btrfs_extent_item *extent;
753 struct blk_plug plug;
754 u64 flags;
755 int ret;
756 int slot;
757 int i;
758 u64 nstripes;
759 int start_stripe;
760 struct extent_buffer *l;
761 struct btrfs_key key;
762 u64 physical;
763 u64 logical;
764 u64 generation;
765 u64 mirror_num;
767 u64 increment = map->stripe_len;
768 u64 offset;
770 nstripes = length;
771 offset = 0;
772 do_div(nstripes, map->stripe_len);
773 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
774 offset = map->stripe_len * num;
775 increment = map->stripe_len * map->num_stripes;
776 mirror_num = 0;
777 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
778 int factor = map->num_stripes / map->sub_stripes;
779 offset = map->stripe_len * (num / map->sub_stripes);
780 increment = map->stripe_len * factor;
781 mirror_num = num % map->sub_stripes;
782 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
783 increment = map->stripe_len;
784 mirror_num = num % map->num_stripes;
785 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
786 increment = map->stripe_len;
787 mirror_num = num % map->num_stripes;
788 } else {
789 increment = map->stripe_len;
790 mirror_num = 0;
793 path = btrfs_alloc_path();
794 if (!path)
795 return -ENOMEM;
797 path->reada = 2;
798 path->search_commit_root = 1;
799 path->skip_locking = 1;
802 * find all extents for each stripe and just read them to get
803 * them into the page cache
804 * FIXME: we can do better. build a more intelligent prefetching
806 logical = base + offset;
807 physical = map->stripes[num].physical;
808 ret = 0;
809 for (i = 0; i < nstripes; ++i) {
810 key.objectid = logical;
811 key.type = BTRFS_EXTENT_ITEM_KEY;
812 key.offset = (u64)0;
814 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
815 if (ret < 0)
816 goto out_noplug;
819 * we might miss half an extent here, but that doesn't matter,
820 * as it's only the prefetch
822 while (1) {
823 l = path->nodes[0];
824 slot = path->slots[0];
825 if (slot >= btrfs_header_nritems(l)) {
826 ret = btrfs_next_leaf(root, path);
827 if (ret == 0)
828 continue;
829 if (ret < 0)
830 goto out_noplug;
832 break;
834 btrfs_item_key_to_cpu(l, &key, slot);
836 if (key.objectid >= logical + map->stripe_len)
837 break;
839 path->slots[0]++;
841 btrfs_release_path(path);
842 logical += increment;
843 physical += map->stripe_len;
844 cond_resched();
848 * collect all data csums for the stripe to avoid seeking during
849 * the scrub. This might currently (crc32) end up to be about 1MB
851 start_stripe = 0;
852 blk_start_plug(&plug);
853 again:
854 logical = base + offset + start_stripe * increment;
855 for (i = start_stripe; i < nstripes; ++i) {
856 ret = btrfs_lookup_csums_range(csum_root, logical,
857 logical + map->stripe_len - 1,
858 &sdev->csum_list, 1);
859 if (ret)
860 goto out;
862 logical += increment;
863 cond_resched();
866 * now find all extents for each stripe and scrub them
868 logical = base + offset + start_stripe * increment;
869 physical = map->stripes[num].physical + start_stripe * map->stripe_len;
870 ret = 0;
871 for (i = start_stripe; i < nstripes; ++i) {
873 * canceled?
875 if (atomic_read(&fs_info->scrub_cancel_req) ||
876 atomic_read(&sdev->cancel_req)) {
877 ret = -ECANCELED;
878 goto out;
881 * check to see if we have to pause
883 if (atomic_read(&fs_info->scrub_pause_req)) {
884 /* push queued extents */
885 scrub_submit(sdev);
886 wait_event(sdev->list_wait,
887 atomic_read(&sdev->in_flight) == 0);
888 atomic_inc(&fs_info->scrubs_paused);
889 wake_up(&fs_info->scrub_pause_wait);
890 mutex_lock(&fs_info->scrub_lock);
891 while (atomic_read(&fs_info->scrub_pause_req)) {
892 mutex_unlock(&fs_info->scrub_lock);
893 wait_event(fs_info->scrub_pause_wait,
894 atomic_read(&fs_info->scrub_pause_req) == 0);
895 mutex_lock(&fs_info->scrub_lock);
897 atomic_dec(&fs_info->scrubs_paused);
898 mutex_unlock(&fs_info->scrub_lock);
899 wake_up(&fs_info->scrub_pause_wait);
900 scrub_free_csums(sdev);
901 start_stripe = i;
902 goto again;
905 key.objectid = logical;
906 key.type = BTRFS_EXTENT_ITEM_KEY;
907 key.offset = (u64)0;
909 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
910 if (ret < 0)
911 goto out;
912 if (ret > 0) {
913 ret = btrfs_previous_item(root, path, 0,
914 BTRFS_EXTENT_ITEM_KEY);
915 if (ret < 0)
916 goto out;
917 if (ret > 0) {
918 /* there's no smaller item, so stick with the
919 * larger one */
920 btrfs_release_path(path);
921 ret = btrfs_search_slot(NULL, root, &key,
922 path, 0, 0);
923 if (ret < 0)
924 goto out;
928 while (1) {
929 l = path->nodes[0];
930 slot = path->slots[0];
931 if (slot >= btrfs_header_nritems(l)) {
932 ret = btrfs_next_leaf(root, path);
933 if (ret == 0)
934 continue;
935 if (ret < 0)
936 goto out;
938 break;
940 btrfs_item_key_to_cpu(l, &key, slot);
942 if (key.objectid + key.offset <= logical)
943 goto next;
945 if (key.objectid >= logical + map->stripe_len)
946 break;
948 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY)
949 goto next;
951 extent = btrfs_item_ptr(l, slot,
952 struct btrfs_extent_item);
953 flags = btrfs_extent_flags(l, extent);
954 generation = btrfs_extent_generation(l, extent);
956 if (key.objectid < logical &&
957 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
958 printk(KERN_ERR
959 "btrfs scrub: tree block %llu spanning "
960 "stripes, ignored. logical=%llu\n",
961 (unsigned long long)key.objectid,
962 (unsigned long long)logical);
963 goto next;
967 * trim extent to this stripe
969 if (key.objectid < logical) {
970 key.offset -= logical - key.objectid;
971 key.objectid = logical;
973 if (key.objectid + key.offset >
974 logical + map->stripe_len) {
975 key.offset = logical + map->stripe_len -
976 key.objectid;
979 ret = scrub_extent(sdev, key.objectid, key.offset,
980 key.objectid - logical + physical,
981 flags, generation, mirror_num);
982 if (ret)
983 goto out;
985 next:
986 path->slots[0]++;
988 btrfs_release_path(path);
989 logical += increment;
990 physical += map->stripe_len;
991 spin_lock(&sdev->stat_lock);
992 sdev->stat.last_physical = physical;
993 spin_unlock(&sdev->stat_lock);
995 /* push queued extents */
996 scrub_submit(sdev);
998 out:
999 blk_finish_plug(&plug);
1000 out_noplug:
1001 btrfs_free_path(path);
1002 return ret < 0 ? ret : 0;
1005 static noinline_for_stack int scrub_chunk(struct scrub_dev *sdev,
1006 u64 chunk_tree, u64 chunk_objectid, u64 chunk_offset, u64 length)
1008 struct btrfs_mapping_tree *map_tree =
1009 &sdev->dev->dev_root->fs_info->mapping_tree;
1010 struct map_lookup *map;
1011 struct extent_map *em;
1012 int i;
1013 int ret = -EINVAL;
1015 read_lock(&map_tree->map_tree.lock);
1016 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
1017 read_unlock(&map_tree->map_tree.lock);
1019 if (!em)
1020 return -EINVAL;
1022 map = (struct map_lookup *)em->bdev;
1023 if (em->start != chunk_offset)
1024 goto out;
1026 if (em->len < length)
1027 goto out;
1029 for (i = 0; i < map->num_stripes; ++i) {
1030 if (map->stripes[i].dev == sdev->dev) {
1031 ret = scrub_stripe(sdev, map, i, chunk_offset, length);
1032 if (ret)
1033 goto out;
1036 out:
1037 free_extent_map(em);
1039 return ret;
1042 static noinline_for_stack
1043 int scrub_enumerate_chunks(struct scrub_dev *sdev, u64 start, u64 end)
1045 struct btrfs_dev_extent *dev_extent = NULL;
1046 struct btrfs_path *path;
1047 struct btrfs_root *root = sdev->dev->dev_root;
1048 struct btrfs_fs_info *fs_info = root->fs_info;
1049 u64 length;
1050 u64 chunk_tree;
1051 u64 chunk_objectid;
1052 u64 chunk_offset;
1053 int ret;
1054 int slot;
1055 struct extent_buffer *l;
1056 struct btrfs_key key;
1057 struct btrfs_key found_key;
1058 struct btrfs_block_group_cache *cache;
1060 path = btrfs_alloc_path();
1061 if (!path)
1062 return -ENOMEM;
1064 path->reada = 2;
1065 path->search_commit_root = 1;
1066 path->skip_locking = 1;
1068 key.objectid = sdev->dev->devid;
1069 key.offset = 0ull;
1070 key.type = BTRFS_DEV_EXTENT_KEY;
1073 while (1) {
1074 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1075 if (ret < 0)
1076 break;
1077 if (ret > 0) {
1078 if (path->slots[0] >=
1079 btrfs_header_nritems(path->nodes[0])) {
1080 ret = btrfs_next_leaf(root, path);
1081 if (ret)
1082 break;
1086 l = path->nodes[0];
1087 slot = path->slots[0];
1089 btrfs_item_key_to_cpu(l, &found_key, slot);
1091 if (found_key.objectid != sdev->dev->devid)
1092 break;
1094 if (btrfs_key_type(&found_key) != BTRFS_DEV_EXTENT_KEY)
1095 break;
1097 if (found_key.offset >= end)
1098 break;
1100 if (found_key.offset < key.offset)
1101 break;
1103 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1104 length = btrfs_dev_extent_length(l, dev_extent);
1106 if (found_key.offset + length <= start) {
1107 key.offset = found_key.offset + length;
1108 btrfs_release_path(path);
1109 continue;
1112 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1113 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1114 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1117 * get a reference on the corresponding block group to prevent
1118 * the chunk from going away while we scrub it
1120 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
1121 if (!cache) {
1122 ret = -ENOENT;
1123 break;
1125 ret = scrub_chunk(sdev, chunk_tree, chunk_objectid,
1126 chunk_offset, length);
1127 btrfs_put_block_group(cache);
1128 if (ret)
1129 break;
1131 key.offset = found_key.offset + length;
1132 btrfs_release_path(path);
1135 btrfs_free_path(path);
1138 * ret can still be 1 from search_slot or next_leaf,
1139 * that's not an error
1141 return ret < 0 ? ret : 0;
1144 static noinline_for_stack int scrub_supers(struct scrub_dev *sdev)
1146 int i;
1147 u64 bytenr;
1148 u64 gen;
1149 int ret;
1150 struct btrfs_device *device = sdev->dev;
1151 struct btrfs_root *root = device->dev_root;
1153 gen = root->fs_info->last_trans_committed;
1155 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1156 bytenr = btrfs_sb_offset(i);
1157 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
1158 break;
1160 ret = scrub_page(sdev, bytenr, PAGE_SIZE, bytenr,
1161 BTRFS_EXTENT_FLAG_SUPER, gen, i, NULL, 1);
1162 if (ret)
1163 return ret;
1165 wait_event(sdev->list_wait, atomic_read(&sdev->in_flight) == 0);
1167 return 0;
1171 * get a reference count on fs_info->scrub_workers. start worker if necessary
1173 static noinline_for_stack int scrub_workers_get(struct btrfs_root *root)
1175 struct btrfs_fs_info *fs_info = root->fs_info;
1177 mutex_lock(&fs_info->scrub_lock);
1178 if (fs_info->scrub_workers_refcnt == 0) {
1179 btrfs_init_workers(&fs_info->scrub_workers, "scrub",
1180 fs_info->thread_pool_size, &fs_info->generic_worker);
1181 fs_info->scrub_workers.idle_thresh = 4;
1182 btrfs_start_workers(&fs_info->scrub_workers, 1);
1184 ++fs_info->scrub_workers_refcnt;
1185 mutex_unlock(&fs_info->scrub_lock);
1187 return 0;
1190 static noinline_for_stack void scrub_workers_put(struct btrfs_root *root)
1192 struct btrfs_fs_info *fs_info = root->fs_info;
1194 mutex_lock(&fs_info->scrub_lock);
1195 if (--fs_info->scrub_workers_refcnt == 0)
1196 btrfs_stop_workers(&fs_info->scrub_workers);
1197 WARN_ON(fs_info->scrub_workers_refcnt < 0);
1198 mutex_unlock(&fs_info->scrub_lock);
1202 int btrfs_scrub_dev(struct btrfs_root *root, u64 devid, u64 start, u64 end,
1203 struct btrfs_scrub_progress *progress, int readonly)
1205 struct scrub_dev *sdev;
1206 struct btrfs_fs_info *fs_info = root->fs_info;
1207 int ret;
1208 struct btrfs_device *dev;
1210 if (btrfs_fs_closing(root->fs_info))
1211 return -EINVAL;
1214 * check some assumptions
1216 if (root->sectorsize != PAGE_SIZE ||
1217 root->sectorsize != root->leafsize ||
1218 root->sectorsize != root->nodesize) {
1219 printk(KERN_ERR "btrfs_scrub: size assumptions fail\n");
1220 return -EINVAL;
1223 ret = scrub_workers_get(root);
1224 if (ret)
1225 return ret;
1227 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1228 dev = btrfs_find_device(root, devid, NULL, NULL);
1229 if (!dev || dev->missing) {
1230 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1231 scrub_workers_put(root);
1232 return -ENODEV;
1234 mutex_lock(&fs_info->scrub_lock);
1236 if (!dev->in_fs_metadata) {
1237 mutex_unlock(&fs_info->scrub_lock);
1238 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1239 scrub_workers_put(root);
1240 return -ENODEV;
1243 if (dev->scrub_device) {
1244 mutex_unlock(&fs_info->scrub_lock);
1245 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1246 scrub_workers_put(root);
1247 return -EINPROGRESS;
1249 sdev = scrub_setup_dev(dev);
1250 if (IS_ERR(sdev)) {
1251 mutex_unlock(&fs_info->scrub_lock);
1252 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1253 scrub_workers_put(root);
1254 return PTR_ERR(sdev);
1256 sdev->readonly = readonly;
1257 dev->scrub_device = sdev;
1259 atomic_inc(&fs_info->scrubs_running);
1260 mutex_unlock(&fs_info->scrub_lock);
1261 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1263 down_read(&fs_info->scrub_super_lock);
1264 ret = scrub_supers(sdev);
1265 up_read(&fs_info->scrub_super_lock);
1267 if (!ret)
1268 ret = scrub_enumerate_chunks(sdev, start, end);
1270 wait_event(sdev->list_wait, atomic_read(&sdev->in_flight) == 0);
1272 atomic_dec(&fs_info->scrubs_running);
1273 wake_up(&fs_info->scrub_pause_wait);
1275 if (progress)
1276 memcpy(progress, &sdev->stat, sizeof(*progress));
1278 mutex_lock(&fs_info->scrub_lock);
1279 dev->scrub_device = NULL;
1280 mutex_unlock(&fs_info->scrub_lock);
1282 scrub_free_dev(sdev);
1283 scrub_workers_put(root);
1285 return ret;
1288 int btrfs_scrub_pause(struct btrfs_root *root)
1290 struct btrfs_fs_info *fs_info = root->fs_info;
1292 mutex_lock(&fs_info->scrub_lock);
1293 atomic_inc(&fs_info->scrub_pause_req);
1294 while (atomic_read(&fs_info->scrubs_paused) !=
1295 atomic_read(&fs_info->scrubs_running)) {
1296 mutex_unlock(&fs_info->scrub_lock);
1297 wait_event(fs_info->scrub_pause_wait,
1298 atomic_read(&fs_info->scrubs_paused) ==
1299 atomic_read(&fs_info->scrubs_running));
1300 mutex_lock(&fs_info->scrub_lock);
1302 mutex_unlock(&fs_info->scrub_lock);
1304 return 0;
1307 int btrfs_scrub_continue(struct btrfs_root *root)
1309 struct btrfs_fs_info *fs_info = root->fs_info;
1311 atomic_dec(&fs_info->scrub_pause_req);
1312 wake_up(&fs_info->scrub_pause_wait);
1313 return 0;
1316 int btrfs_scrub_pause_super(struct btrfs_root *root)
1318 down_write(&root->fs_info->scrub_super_lock);
1319 return 0;
1322 int btrfs_scrub_continue_super(struct btrfs_root *root)
1324 up_write(&root->fs_info->scrub_super_lock);
1325 return 0;
1328 int btrfs_scrub_cancel(struct btrfs_root *root)
1330 struct btrfs_fs_info *fs_info = root->fs_info;
1332 mutex_lock(&fs_info->scrub_lock);
1333 if (!atomic_read(&fs_info->scrubs_running)) {
1334 mutex_unlock(&fs_info->scrub_lock);
1335 return -ENOTCONN;
1338 atomic_inc(&fs_info->scrub_cancel_req);
1339 while (atomic_read(&fs_info->scrubs_running)) {
1340 mutex_unlock(&fs_info->scrub_lock);
1341 wait_event(fs_info->scrub_pause_wait,
1342 atomic_read(&fs_info->scrubs_running) == 0);
1343 mutex_lock(&fs_info->scrub_lock);
1345 atomic_dec(&fs_info->scrub_cancel_req);
1346 mutex_unlock(&fs_info->scrub_lock);
1348 return 0;
1351 int btrfs_scrub_cancel_dev(struct btrfs_root *root, struct btrfs_device *dev)
1353 struct btrfs_fs_info *fs_info = root->fs_info;
1354 struct scrub_dev *sdev;
1356 mutex_lock(&fs_info->scrub_lock);
1357 sdev = dev->scrub_device;
1358 if (!sdev) {
1359 mutex_unlock(&fs_info->scrub_lock);
1360 return -ENOTCONN;
1362 atomic_inc(&sdev->cancel_req);
1363 while (dev->scrub_device) {
1364 mutex_unlock(&fs_info->scrub_lock);
1365 wait_event(fs_info->scrub_pause_wait,
1366 dev->scrub_device == NULL);
1367 mutex_lock(&fs_info->scrub_lock);
1369 mutex_unlock(&fs_info->scrub_lock);
1371 return 0;
1373 int btrfs_scrub_cancel_devid(struct btrfs_root *root, u64 devid)
1375 struct btrfs_fs_info *fs_info = root->fs_info;
1376 struct btrfs_device *dev;
1377 int ret;
1380 * we have to hold the device_list_mutex here so the device
1381 * does not go away in cancel_dev. FIXME: find a better solution
1383 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1384 dev = btrfs_find_device(root, devid, NULL, NULL);
1385 if (!dev) {
1386 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1387 return -ENODEV;
1389 ret = btrfs_scrub_cancel_dev(root, dev);
1390 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1392 return ret;
1395 int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
1396 struct btrfs_scrub_progress *progress)
1398 struct btrfs_device *dev;
1399 struct scrub_dev *sdev = NULL;
1401 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1402 dev = btrfs_find_device(root, devid, NULL, NULL);
1403 if (dev)
1404 sdev = dev->scrub_device;
1405 if (sdev)
1406 memcpy(progress, &sdev->stat, sizeof(*progress));
1407 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1409 return dev ? (sdev ? 0 : -ENOTCONN) : -ENODEV;