1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2011, 2012 STRATO. All rights reserved.
6 #include <linux/blkdev.h>
7 #include <linux/ratelimit.h>
8 #include <linux/sched/mm.h>
9 #include <crypto/hash.h>
14 #include "ordered-data.h"
15 #include "transaction.h"
17 #include "extent_io.h"
18 #include "dev-replace.h"
20 #include "block-group.h"
23 #include "accessors.h"
24 #include "file-item.h"
26 #include "raid-stripe-tree.h"
29 * This is only the first step towards a full-features scrub. It reads all
30 * extent and super block and verifies the checksums. In case a bad checksum
31 * is found or the extent cannot be read, good data will be written back if
34 * Future enhancements:
35 * - In case an unrepairable extent is encountered, track which files are
36 * affected and report them
37 * - track and record media errors, throw out bad devices
38 * - add a mode to also read unallocated space
44 * The following value only influences the performance.
46 * This determines how many stripes would be submitted in one go,
47 * which is 512KiB (BTRFS_STRIPE_LEN * SCRUB_STRIPES_PER_GROUP).
49 #define SCRUB_STRIPES_PER_GROUP 8
52 * How many groups we have for each sctx.
54 * This would be 8M per device, the same value as the old scrub in-flight bios
57 #define SCRUB_GROUPS_PER_SCTX 16
59 #define SCRUB_TOTAL_STRIPES (SCRUB_GROUPS_PER_SCTX * SCRUB_STRIPES_PER_GROUP)
62 * The following value times PAGE_SIZE needs to be large enough to match the
63 * largest node/leaf/sector size that shall be supported.
65 #define SCRUB_MAX_SECTORS_PER_BLOCK (BTRFS_MAX_METADATA_BLOCKSIZE / SZ_4K)
67 /* Represent one sector and its needed info to verify the content. */
68 struct scrub_sector_verification {
73 * Csum pointer for data csum verification. Should point to a
74 * sector csum inside scrub_stripe::csums.
76 * NULL if this data sector has no csum.
81 * Extra info for metadata verification. All sectors inside a
82 * tree block share the same generation.
88 enum scrub_stripe_flags {
89 /* Set when @mirror_num, @dev, @physical and @logical are set. */
90 SCRUB_STRIPE_FLAG_INITIALIZED,
92 /* Set when the read-repair is finished. */
93 SCRUB_STRIPE_FLAG_REPAIR_DONE,
96 * Set for data stripes if it's triggered from P/Q stripe.
97 * During such scrub, we should not report errors in data stripes, nor
98 * update the accounting.
100 SCRUB_STRIPE_FLAG_NO_REPORT,
103 #define SCRUB_STRIPE_PAGES (BTRFS_STRIPE_LEN / PAGE_SIZE)
106 * Represent one contiguous range with a length of BTRFS_STRIPE_LEN.
108 struct scrub_stripe {
109 struct scrub_ctx *sctx;
110 struct btrfs_block_group *bg;
112 struct page *pages[SCRUB_STRIPE_PAGES];
113 struct scrub_sector_verification *sectors;
115 struct btrfs_device *dev;
121 /* Should be BTRFS_STRIPE_LEN / sectorsize. */
125 * How many data/meta extents are in this stripe. Only for scrub status
126 * reporting purposes.
132 wait_queue_head_t io_wait;
133 wait_queue_head_t repair_wait;
136 * Indicate the states of the stripe. Bits are defined in
137 * scrub_stripe_flags enum.
141 /* Indicate which sectors are covered by extent items. */
142 unsigned long extent_sector_bitmap;
145 * The errors hit during the initial read of the stripe.
147 * Would be utilized for error reporting and repair.
149 * The remaining init_nr_* records the number of errors hit, only used
150 * by error reporting.
152 unsigned long init_error_bitmap;
153 unsigned int init_nr_io_errors;
154 unsigned int init_nr_csum_errors;
155 unsigned int init_nr_meta_errors;
158 * The following error bitmaps are all for the current status.
159 * Every time we submit a new read, these bitmaps may be updated.
161 * error_bitmap = io_error_bitmap | csum_error_bitmap | meta_error_bitmap;
163 * IO and csum errors can happen for both metadata and data.
165 unsigned long error_bitmap;
166 unsigned long io_error_bitmap;
167 unsigned long csum_error_bitmap;
168 unsigned long meta_error_bitmap;
170 /* For writeback (repair or replace) error reporting. */
171 unsigned long write_error_bitmap;
173 /* Writeback can be concurrent, thus we need to protect the bitmap. */
174 spinlock_t write_error_lock;
177 * Checksum for the whole stripe if this stripe is inside a data block
182 struct work_struct work;
186 struct scrub_stripe stripes[SCRUB_TOTAL_STRIPES];
187 struct scrub_stripe *raid56_data_stripes;
188 struct btrfs_fs_info *fs_info;
189 struct btrfs_path extent_path;
190 struct btrfs_path csum_path;
196 /* State of IO submission throttling affecting the associated device */
197 ktime_t throttle_deadline;
203 struct mutex wr_lock;
204 struct btrfs_device *wr_tgtdev;
209 struct btrfs_scrub_progress stat;
210 spinlock_t stat_lock;
213 * Use a ref counter to avoid use-after-free issues. Scrub workers
214 * decrement bios_in_flight and workers_pending and then do a wakeup
215 * on the list_wait wait queue. We must ensure the main scrub task
216 * doesn't free the scrub context before or while the workers are
217 * doing the wakeup() call.
222 struct scrub_warning {
223 struct btrfs_path *path;
224 u64 extent_item_size;
228 struct btrfs_device *dev;
231 static void release_scrub_stripe(struct scrub_stripe *stripe)
236 for (int i = 0; i < SCRUB_STRIPE_PAGES; i++) {
237 if (stripe->pages[i])
238 __free_page(stripe->pages[i]);
239 stripe->pages[i] = NULL;
241 kfree(stripe->sectors);
242 kfree(stripe->csums);
243 stripe->sectors = NULL;
244 stripe->csums = NULL;
249 static int init_scrub_stripe(struct btrfs_fs_info *fs_info,
250 struct scrub_stripe *stripe)
254 memset(stripe, 0, sizeof(*stripe));
256 stripe->nr_sectors = BTRFS_STRIPE_LEN >> fs_info->sectorsize_bits;
259 init_waitqueue_head(&stripe->io_wait);
260 init_waitqueue_head(&stripe->repair_wait);
261 atomic_set(&stripe->pending_io, 0);
262 spin_lock_init(&stripe->write_error_lock);
264 ret = btrfs_alloc_page_array(SCRUB_STRIPE_PAGES, stripe->pages, 0);
268 stripe->sectors = kcalloc(stripe->nr_sectors,
269 sizeof(struct scrub_sector_verification),
271 if (!stripe->sectors)
274 stripe->csums = kcalloc(BTRFS_STRIPE_LEN >> fs_info->sectorsize_bits,
275 fs_info->csum_size, GFP_KERNEL);
280 release_scrub_stripe(stripe);
284 static void wait_scrub_stripe_io(struct scrub_stripe *stripe)
286 wait_event(stripe->io_wait, atomic_read(&stripe->pending_io) == 0);
289 static void scrub_put_ctx(struct scrub_ctx *sctx);
291 static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
293 while (atomic_read(&fs_info->scrub_pause_req)) {
294 mutex_unlock(&fs_info->scrub_lock);
295 wait_event(fs_info->scrub_pause_wait,
296 atomic_read(&fs_info->scrub_pause_req) == 0);
297 mutex_lock(&fs_info->scrub_lock);
301 static void scrub_pause_on(struct btrfs_fs_info *fs_info)
303 atomic_inc(&fs_info->scrubs_paused);
304 wake_up(&fs_info->scrub_pause_wait);
307 static void scrub_pause_off(struct btrfs_fs_info *fs_info)
309 mutex_lock(&fs_info->scrub_lock);
310 __scrub_blocked_if_needed(fs_info);
311 atomic_dec(&fs_info->scrubs_paused);
312 mutex_unlock(&fs_info->scrub_lock);
314 wake_up(&fs_info->scrub_pause_wait);
317 static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
319 scrub_pause_on(fs_info);
320 scrub_pause_off(fs_info);
323 static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx)
330 for (i = 0; i < SCRUB_TOTAL_STRIPES; i++)
331 release_scrub_stripe(&sctx->stripes[i]);
336 static void scrub_put_ctx(struct scrub_ctx *sctx)
338 if (refcount_dec_and_test(&sctx->refs))
339 scrub_free_ctx(sctx);
342 static noinline_for_stack struct scrub_ctx *scrub_setup_ctx(
343 struct btrfs_fs_info *fs_info, int is_dev_replace)
345 struct scrub_ctx *sctx;
348 /* Since sctx has inline 128 stripes, it can go beyond 64K easily. Use
351 sctx = kvzalloc(sizeof(*sctx), GFP_KERNEL);
354 refcount_set(&sctx->refs, 1);
355 sctx->is_dev_replace = is_dev_replace;
356 sctx->fs_info = fs_info;
357 sctx->extent_path.search_commit_root = 1;
358 sctx->extent_path.skip_locking = 1;
359 sctx->csum_path.search_commit_root = 1;
360 sctx->csum_path.skip_locking = 1;
361 for (i = 0; i < SCRUB_TOTAL_STRIPES; i++) {
364 ret = init_scrub_stripe(fs_info, &sctx->stripes[i]);
367 sctx->stripes[i].sctx = sctx;
369 sctx->first_free = 0;
370 atomic_set(&sctx->cancel_req, 0);
372 spin_lock_init(&sctx->stat_lock);
373 sctx->throttle_deadline = 0;
375 mutex_init(&sctx->wr_lock);
376 if (is_dev_replace) {
377 WARN_ON(!fs_info->dev_replace.tgtdev);
378 sctx->wr_tgtdev = fs_info->dev_replace.tgtdev;
384 scrub_free_ctx(sctx);
385 return ERR_PTR(-ENOMEM);
388 static int scrub_print_warning_inode(u64 inum, u64 offset, u64 num_bytes,
389 u64 root, void *warn_ctx)
395 struct extent_buffer *eb;
396 struct btrfs_inode_item *inode_item;
397 struct scrub_warning *swarn = warn_ctx;
398 struct btrfs_fs_info *fs_info = swarn->dev->fs_info;
399 struct inode_fs_paths *ipath = NULL;
400 struct btrfs_root *local_root;
401 struct btrfs_key key;
403 local_root = btrfs_get_fs_root(fs_info, root, true);
404 if (IS_ERR(local_root)) {
405 ret = PTR_ERR(local_root);
410 * this makes the path point to (inum INODE_ITEM ioff)
413 key.type = BTRFS_INODE_ITEM_KEY;
416 ret = btrfs_search_slot(NULL, local_root, &key, swarn->path, 0, 0);
418 btrfs_put_root(local_root);
419 btrfs_release_path(swarn->path);
423 eb = swarn->path->nodes[0];
424 inode_item = btrfs_item_ptr(eb, swarn->path->slots[0],
425 struct btrfs_inode_item);
426 nlink = btrfs_inode_nlink(eb, inode_item);
427 btrfs_release_path(swarn->path);
430 * init_path might indirectly call vmalloc, or use GFP_KERNEL. Scrub
431 * uses GFP_NOFS in this context, so we keep it consistent but it does
432 * not seem to be strictly necessary.
434 nofs_flag = memalloc_nofs_save();
435 ipath = init_ipath(4096, local_root, swarn->path);
436 memalloc_nofs_restore(nofs_flag);
438 btrfs_put_root(local_root);
439 ret = PTR_ERR(ipath);
443 ret = paths_from_inode(inum, ipath);
449 * we deliberately ignore the bit ipath might have been too small to
450 * hold all of the paths here
452 for (i = 0; i < ipath->fspath->elem_cnt; ++i)
453 btrfs_warn_in_rcu(fs_info,
454 "%s at logical %llu on dev %s, physical %llu, root %llu, inode %llu, offset %llu, length %u, links %u (path: %s)",
455 swarn->errstr, swarn->logical,
456 btrfs_dev_name(swarn->dev),
459 fs_info->sectorsize, nlink,
460 (char *)(unsigned long)ipath->fspath->val[i]);
462 btrfs_put_root(local_root);
467 btrfs_warn_in_rcu(fs_info,
468 "%s at logical %llu on dev %s, physical %llu, root %llu, inode %llu, offset %llu: path resolving failed with ret=%d",
469 swarn->errstr, swarn->logical,
470 btrfs_dev_name(swarn->dev),
472 root, inum, offset, ret);
478 static void scrub_print_common_warning(const char *errstr, struct btrfs_device *dev,
479 bool is_super, u64 logical, u64 physical)
481 struct btrfs_fs_info *fs_info = dev->fs_info;
482 struct btrfs_path *path;
483 struct btrfs_key found_key;
484 struct extent_buffer *eb;
485 struct btrfs_extent_item *ei;
486 struct scrub_warning swarn;
491 /* Super block error, no need to search extent tree. */
493 btrfs_warn_in_rcu(fs_info, "%s on device %s, physical %llu",
494 errstr, btrfs_dev_name(dev), physical);
497 path = btrfs_alloc_path();
501 swarn.physical = physical;
502 swarn.logical = logical;
503 swarn.errstr = errstr;
506 ret = extent_from_logical(fs_info, swarn.logical, path, &found_key,
511 swarn.extent_item_size = found_key.offset;
514 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
515 item_size = btrfs_item_size(eb, path->slots[0]);
517 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
518 unsigned long ptr = 0;
523 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
524 item_size, &ref_root,
528 "failed to resolve tree backref for logical %llu: %d",
534 btrfs_warn_in_rcu(fs_info,
535 "%s at logical %llu on dev %s, physical %llu: metadata %s (level %d) in tree %llu",
536 errstr, swarn.logical, btrfs_dev_name(dev),
537 swarn.physical, (ref_level ? "node" : "leaf"),
538 ref_level, ref_root);
540 btrfs_release_path(path);
542 struct btrfs_backref_walk_ctx ctx = { 0 };
544 btrfs_release_path(path);
546 ctx.bytenr = found_key.objectid;
547 ctx.extent_item_pos = swarn.logical - found_key.objectid;
548 ctx.fs_info = fs_info;
553 iterate_extent_inodes(&ctx, true, scrub_print_warning_inode, &swarn);
557 btrfs_free_path(path);
560 static int fill_writer_pointer_gap(struct scrub_ctx *sctx, u64 physical)
565 if (!btrfs_is_zoned(sctx->fs_info))
568 if (!btrfs_dev_is_sequential(sctx->wr_tgtdev, physical))
571 if (sctx->write_pointer < physical) {
572 length = physical - sctx->write_pointer;
574 ret = btrfs_zoned_issue_zeroout(sctx->wr_tgtdev,
575 sctx->write_pointer, length);
577 sctx->write_pointer = physical;
582 static struct page *scrub_stripe_get_page(struct scrub_stripe *stripe, int sector_nr)
584 struct btrfs_fs_info *fs_info = stripe->bg->fs_info;
585 int page_index = (sector_nr << fs_info->sectorsize_bits) >> PAGE_SHIFT;
587 return stripe->pages[page_index];
590 static unsigned int scrub_stripe_get_page_offset(struct scrub_stripe *stripe,
593 struct btrfs_fs_info *fs_info = stripe->bg->fs_info;
595 return offset_in_page(sector_nr << fs_info->sectorsize_bits);
598 static void scrub_verify_one_metadata(struct scrub_stripe *stripe, int sector_nr)
600 struct btrfs_fs_info *fs_info = stripe->bg->fs_info;
601 const u32 sectors_per_tree = fs_info->nodesize >> fs_info->sectorsize_bits;
602 const u64 logical = stripe->logical + (sector_nr << fs_info->sectorsize_bits);
603 const struct page *first_page = scrub_stripe_get_page(stripe, sector_nr);
604 const unsigned int first_off = scrub_stripe_get_page_offset(stripe, sector_nr);
605 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
606 u8 on_disk_csum[BTRFS_CSUM_SIZE];
607 u8 calculated_csum[BTRFS_CSUM_SIZE];
608 struct btrfs_header *header;
611 * Here we don't have a good way to attach the pages (and subpages)
612 * to a dummy extent buffer, thus we have to directly grab the members
615 header = (struct btrfs_header *)(page_address(first_page) + first_off);
616 memcpy(on_disk_csum, header->csum, fs_info->csum_size);
618 if (logical != btrfs_stack_header_bytenr(header)) {
619 bitmap_set(&stripe->csum_error_bitmap, sector_nr, sectors_per_tree);
620 bitmap_set(&stripe->error_bitmap, sector_nr, sectors_per_tree);
621 btrfs_warn_rl(fs_info,
622 "tree block %llu mirror %u has bad bytenr, has %llu want %llu",
623 logical, stripe->mirror_num,
624 btrfs_stack_header_bytenr(header), logical);
627 if (memcmp(header->fsid, fs_info->fs_devices->metadata_uuid,
628 BTRFS_FSID_SIZE) != 0) {
629 bitmap_set(&stripe->meta_error_bitmap, sector_nr, sectors_per_tree);
630 bitmap_set(&stripe->error_bitmap, sector_nr, sectors_per_tree);
631 btrfs_warn_rl(fs_info,
632 "tree block %llu mirror %u has bad fsid, has %pU want %pU",
633 logical, stripe->mirror_num,
634 header->fsid, fs_info->fs_devices->fsid);
637 if (memcmp(header->chunk_tree_uuid, fs_info->chunk_tree_uuid,
638 BTRFS_UUID_SIZE) != 0) {
639 bitmap_set(&stripe->meta_error_bitmap, sector_nr, sectors_per_tree);
640 bitmap_set(&stripe->error_bitmap, sector_nr, sectors_per_tree);
641 btrfs_warn_rl(fs_info,
642 "tree block %llu mirror %u has bad chunk tree uuid, has %pU want %pU",
643 logical, stripe->mirror_num,
644 header->chunk_tree_uuid, fs_info->chunk_tree_uuid);
648 /* Now check tree block csum. */
649 shash->tfm = fs_info->csum_shash;
650 crypto_shash_init(shash);
651 crypto_shash_update(shash, page_address(first_page) + first_off +
652 BTRFS_CSUM_SIZE, fs_info->sectorsize - BTRFS_CSUM_SIZE);
654 for (int i = sector_nr + 1; i < sector_nr + sectors_per_tree; i++) {
655 struct page *page = scrub_stripe_get_page(stripe, i);
656 unsigned int page_off = scrub_stripe_get_page_offset(stripe, i);
658 crypto_shash_update(shash, page_address(page) + page_off,
659 fs_info->sectorsize);
662 crypto_shash_final(shash, calculated_csum);
663 if (memcmp(calculated_csum, on_disk_csum, fs_info->csum_size) != 0) {
664 bitmap_set(&stripe->meta_error_bitmap, sector_nr, sectors_per_tree);
665 bitmap_set(&stripe->error_bitmap, sector_nr, sectors_per_tree);
666 btrfs_warn_rl(fs_info,
667 "tree block %llu mirror %u has bad csum, has " CSUM_FMT " want " CSUM_FMT,
668 logical, stripe->mirror_num,
669 CSUM_FMT_VALUE(fs_info->csum_size, on_disk_csum),
670 CSUM_FMT_VALUE(fs_info->csum_size, calculated_csum));
673 if (stripe->sectors[sector_nr].generation !=
674 btrfs_stack_header_generation(header)) {
675 bitmap_set(&stripe->meta_error_bitmap, sector_nr, sectors_per_tree);
676 bitmap_set(&stripe->error_bitmap, sector_nr, sectors_per_tree);
677 btrfs_warn_rl(fs_info,
678 "tree block %llu mirror %u has bad generation, has %llu want %llu",
679 logical, stripe->mirror_num,
680 btrfs_stack_header_generation(header),
681 stripe->sectors[sector_nr].generation);
684 bitmap_clear(&stripe->error_bitmap, sector_nr, sectors_per_tree);
685 bitmap_clear(&stripe->csum_error_bitmap, sector_nr, sectors_per_tree);
686 bitmap_clear(&stripe->meta_error_bitmap, sector_nr, sectors_per_tree);
689 static void scrub_verify_one_sector(struct scrub_stripe *stripe, int sector_nr)
691 struct btrfs_fs_info *fs_info = stripe->bg->fs_info;
692 struct scrub_sector_verification *sector = &stripe->sectors[sector_nr];
693 const u32 sectors_per_tree = fs_info->nodesize >> fs_info->sectorsize_bits;
694 struct page *page = scrub_stripe_get_page(stripe, sector_nr);
695 unsigned int pgoff = scrub_stripe_get_page_offset(stripe, sector_nr);
696 u8 csum_buf[BTRFS_CSUM_SIZE];
699 ASSERT(sector_nr >= 0 && sector_nr < stripe->nr_sectors);
701 /* Sector not utilized, skip it. */
702 if (!test_bit(sector_nr, &stripe->extent_sector_bitmap))
705 /* IO error, no need to check. */
706 if (test_bit(sector_nr, &stripe->io_error_bitmap))
709 /* Metadata, verify the full tree block. */
710 if (sector->is_metadata) {
712 * Check if the tree block crosses the stripe boundary. If
713 * crossed the boundary, we cannot verify it but only give a
716 * This can only happen on a very old filesystem where chunks
717 * are not ensured to be stripe aligned.
719 if (unlikely(sector_nr + sectors_per_tree > stripe->nr_sectors)) {
720 btrfs_warn_rl(fs_info,
721 "tree block at %llu crosses stripe boundary %llu",
723 (sector_nr << fs_info->sectorsize_bits),
727 scrub_verify_one_metadata(stripe, sector_nr);
732 * Data is easier, we just verify the data csum (if we have it). For
733 * cases without csum, we have no other choice but to trust it.
736 clear_bit(sector_nr, &stripe->error_bitmap);
740 ret = btrfs_check_sector_csum(fs_info, page, pgoff, csum_buf, sector->csum);
742 set_bit(sector_nr, &stripe->csum_error_bitmap);
743 set_bit(sector_nr, &stripe->error_bitmap);
745 clear_bit(sector_nr, &stripe->csum_error_bitmap);
746 clear_bit(sector_nr, &stripe->error_bitmap);
750 /* Verify specified sectors of a stripe. */
751 static void scrub_verify_one_stripe(struct scrub_stripe *stripe, unsigned long bitmap)
753 struct btrfs_fs_info *fs_info = stripe->bg->fs_info;
754 const u32 sectors_per_tree = fs_info->nodesize >> fs_info->sectorsize_bits;
757 for_each_set_bit(sector_nr, &bitmap, stripe->nr_sectors) {
758 scrub_verify_one_sector(stripe, sector_nr);
759 if (stripe->sectors[sector_nr].is_metadata)
760 sector_nr += sectors_per_tree - 1;
764 static int calc_sector_number(struct scrub_stripe *stripe, struct bio_vec *first_bvec)
768 for (i = 0; i < stripe->nr_sectors; i++) {
769 if (scrub_stripe_get_page(stripe, i) == first_bvec->bv_page &&
770 scrub_stripe_get_page_offset(stripe, i) == first_bvec->bv_offset)
773 ASSERT(i < stripe->nr_sectors);
778 * Repair read is different to the regular read:
780 * - Only reads the failed sectors
781 * - May have extra blocksize limits
783 static void scrub_repair_read_endio(struct btrfs_bio *bbio)
785 struct scrub_stripe *stripe = bbio->private;
786 struct btrfs_fs_info *fs_info = stripe->bg->fs_info;
787 struct bio_vec *bvec;
788 int sector_nr = calc_sector_number(stripe, bio_first_bvec_all(&bbio->bio));
792 ASSERT(sector_nr < stripe->nr_sectors);
794 bio_for_each_bvec_all(bvec, &bbio->bio, i)
795 bio_size += bvec->bv_len;
797 if (bbio->bio.bi_status) {
798 bitmap_set(&stripe->io_error_bitmap, sector_nr,
799 bio_size >> fs_info->sectorsize_bits);
800 bitmap_set(&stripe->error_bitmap, sector_nr,
801 bio_size >> fs_info->sectorsize_bits);
803 bitmap_clear(&stripe->io_error_bitmap, sector_nr,
804 bio_size >> fs_info->sectorsize_bits);
807 if (atomic_dec_and_test(&stripe->pending_io))
808 wake_up(&stripe->io_wait);
811 static int calc_next_mirror(int mirror, int num_copies)
813 ASSERT(mirror <= num_copies);
814 return (mirror + 1 > num_copies) ? 1 : mirror + 1;
817 static void scrub_stripe_submit_repair_read(struct scrub_stripe *stripe,
818 int mirror, int blocksize, bool wait)
820 struct btrfs_fs_info *fs_info = stripe->bg->fs_info;
821 struct btrfs_bio *bbio = NULL;
822 const unsigned long old_error_bitmap = stripe->error_bitmap;
825 ASSERT(stripe->mirror_num >= 1);
826 ASSERT(atomic_read(&stripe->pending_io) == 0);
828 for_each_set_bit(i, &old_error_bitmap, stripe->nr_sectors) {
833 page = scrub_stripe_get_page(stripe, i);
834 pgoff = scrub_stripe_get_page_offset(stripe, i);
836 /* The current sector cannot be merged, submit the bio. */
837 if (bbio && ((i > 0 && !test_bit(i - 1, &stripe->error_bitmap)) ||
838 bbio->bio.bi_iter.bi_size >= blocksize)) {
839 ASSERT(bbio->bio.bi_iter.bi_size);
840 atomic_inc(&stripe->pending_io);
841 btrfs_submit_bio(bbio, mirror);
843 wait_scrub_stripe_io(stripe);
848 bbio = btrfs_bio_alloc(stripe->nr_sectors, REQ_OP_READ,
849 fs_info, scrub_repair_read_endio, stripe);
850 bbio->bio.bi_iter.bi_sector = (stripe->logical +
851 (i << fs_info->sectorsize_bits)) >> SECTOR_SHIFT;
854 ret = bio_add_page(&bbio->bio, page, fs_info->sectorsize, pgoff);
855 ASSERT(ret == fs_info->sectorsize);
858 ASSERT(bbio->bio.bi_iter.bi_size);
859 atomic_inc(&stripe->pending_io);
860 btrfs_submit_bio(bbio, mirror);
862 wait_scrub_stripe_io(stripe);
866 static void scrub_stripe_report_errors(struct scrub_ctx *sctx,
867 struct scrub_stripe *stripe)
869 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
870 DEFAULT_RATELIMIT_BURST);
871 struct btrfs_fs_info *fs_info = sctx->fs_info;
872 struct btrfs_device *dev = NULL;
874 int nr_data_sectors = 0;
875 int nr_meta_sectors = 0;
876 int nr_nodatacsum_sectors = 0;
877 int nr_repaired_sectors = 0;
880 if (test_bit(SCRUB_STRIPE_FLAG_NO_REPORT, &stripe->state))
884 * Init needed infos for error reporting.
886 * Although our scrub_stripe infrastructure is mostly based on btrfs_submit_bio()
887 * thus no need for dev/physical, error reporting still needs dev and physical.
889 if (!bitmap_empty(&stripe->init_error_bitmap, stripe->nr_sectors)) {
890 u64 mapped_len = fs_info->sectorsize;
891 struct btrfs_io_context *bioc = NULL;
892 int stripe_index = stripe->mirror_num - 1;
895 /* For scrub, our mirror_num should always start at 1. */
896 ASSERT(stripe->mirror_num >= 1);
897 ret = btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
898 stripe->logical, &mapped_len, &bioc,
901 * If we failed, dev will be NULL, and later detailed reports
902 * will just be skipped.
906 physical = bioc->stripes[stripe_index].physical;
907 dev = bioc->stripes[stripe_index].dev;
908 btrfs_put_bioc(bioc);
912 for_each_set_bit(sector_nr, &stripe->extent_sector_bitmap, stripe->nr_sectors) {
913 bool repaired = false;
915 if (stripe->sectors[sector_nr].is_metadata) {
919 if (!stripe->sectors[sector_nr].csum)
920 nr_nodatacsum_sectors++;
923 if (test_bit(sector_nr, &stripe->init_error_bitmap) &&
924 !test_bit(sector_nr, &stripe->error_bitmap)) {
925 nr_repaired_sectors++;
929 /* Good sector from the beginning, nothing need to be done. */
930 if (!test_bit(sector_nr, &stripe->init_error_bitmap))
934 * Report error for the corrupted sectors. If repaired, just
935 * output the message of repaired message.
939 btrfs_err_rl_in_rcu(fs_info,
940 "fixed up error at logical %llu on dev %s physical %llu",
941 stripe->logical, btrfs_dev_name(dev),
944 btrfs_err_rl_in_rcu(fs_info,
945 "fixed up error at logical %llu on mirror %u",
946 stripe->logical, stripe->mirror_num);
951 /* The remaining are all for unrepaired. */
953 btrfs_err_rl_in_rcu(fs_info,
954 "unable to fixup (regular) error at logical %llu on dev %s physical %llu",
955 stripe->logical, btrfs_dev_name(dev),
958 btrfs_err_rl_in_rcu(fs_info,
959 "unable to fixup (regular) error at logical %llu on mirror %u",
960 stripe->logical, stripe->mirror_num);
963 if (test_bit(sector_nr, &stripe->io_error_bitmap))
964 if (__ratelimit(&rs) && dev)
965 scrub_print_common_warning("i/o error", dev, false,
966 stripe->logical, physical);
967 if (test_bit(sector_nr, &stripe->csum_error_bitmap))
968 if (__ratelimit(&rs) && dev)
969 scrub_print_common_warning("checksum error", dev, false,
970 stripe->logical, physical);
971 if (test_bit(sector_nr, &stripe->meta_error_bitmap))
972 if (__ratelimit(&rs) && dev)
973 scrub_print_common_warning("header error", dev, false,
974 stripe->logical, physical);
977 spin_lock(&sctx->stat_lock);
978 sctx->stat.data_extents_scrubbed += stripe->nr_data_extents;
979 sctx->stat.tree_extents_scrubbed += stripe->nr_meta_extents;
980 sctx->stat.data_bytes_scrubbed += nr_data_sectors << fs_info->sectorsize_bits;
981 sctx->stat.tree_bytes_scrubbed += nr_meta_sectors << fs_info->sectorsize_bits;
982 sctx->stat.no_csum += nr_nodatacsum_sectors;
983 sctx->stat.read_errors += stripe->init_nr_io_errors;
984 sctx->stat.csum_errors += stripe->init_nr_csum_errors;
985 sctx->stat.verify_errors += stripe->init_nr_meta_errors;
986 sctx->stat.uncorrectable_errors +=
987 bitmap_weight(&stripe->error_bitmap, stripe->nr_sectors);
988 sctx->stat.corrected_errors += nr_repaired_sectors;
989 spin_unlock(&sctx->stat_lock);
992 static void scrub_write_sectors(struct scrub_ctx *sctx, struct scrub_stripe *stripe,
993 unsigned long write_bitmap, bool dev_replace);
996 * The main entrance for all read related scrub work, including:
998 * - Wait for the initial read to finish
999 * - Verify and locate any bad sectors
1000 * - Go through the remaining mirrors and try to read as large blocksize as
1002 * - Go through all mirrors (including the failed mirror) sector-by-sector
1003 * - Submit writeback for repaired sectors
1005 * Writeback for dev-replace does not happen here, it needs extra
1006 * synchronization for zoned devices.
1008 static void scrub_stripe_read_repair_worker(struct work_struct *work)
1010 struct scrub_stripe *stripe = container_of(work, struct scrub_stripe, work);
1011 struct scrub_ctx *sctx = stripe->sctx;
1012 struct btrfs_fs_info *fs_info = sctx->fs_info;
1013 int num_copies = btrfs_num_copies(fs_info, stripe->bg->start,
1014 stripe->bg->length);
1015 unsigned long repaired;
1019 ASSERT(stripe->mirror_num > 0);
1021 wait_scrub_stripe_io(stripe);
1022 scrub_verify_one_stripe(stripe, stripe->extent_sector_bitmap);
1023 /* Save the initial failed bitmap for later repair and report usage. */
1024 stripe->init_error_bitmap = stripe->error_bitmap;
1025 stripe->init_nr_io_errors = bitmap_weight(&stripe->io_error_bitmap,
1026 stripe->nr_sectors);
1027 stripe->init_nr_csum_errors = bitmap_weight(&stripe->csum_error_bitmap,
1028 stripe->nr_sectors);
1029 stripe->init_nr_meta_errors = bitmap_weight(&stripe->meta_error_bitmap,
1030 stripe->nr_sectors);
1032 if (bitmap_empty(&stripe->init_error_bitmap, stripe->nr_sectors))
1036 * Try all remaining mirrors.
1038 * Here we still try to read as large block as possible, as this is
1039 * faster and we have extra safety nets to rely on.
1041 for (mirror = calc_next_mirror(stripe->mirror_num, num_copies);
1042 mirror != stripe->mirror_num;
1043 mirror = calc_next_mirror(mirror, num_copies)) {
1044 const unsigned long old_error_bitmap = stripe->error_bitmap;
1046 scrub_stripe_submit_repair_read(stripe, mirror,
1047 BTRFS_STRIPE_LEN, false);
1048 wait_scrub_stripe_io(stripe);
1049 scrub_verify_one_stripe(stripe, old_error_bitmap);
1050 if (bitmap_empty(&stripe->error_bitmap, stripe->nr_sectors))
1055 * Last safety net, try re-checking all mirrors, including the failed
1056 * one, sector-by-sector.
1058 * As if one sector failed the drive's internal csum, the whole read
1059 * containing the offending sector would be marked as error.
1060 * Thus here we do sector-by-sector read.
1062 * This can be slow, thus we only try it as the last resort.
1065 for (i = 0, mirror = stripe->mirror_num;
1067 i++, mirror = calc_next_mirror(mirror, num_copies)) {
1068 const unsigned long old_error_bitmap = stripe->error_bitmap;
1070 scrub_stripe_submit_repair_read(stripe, mirror,
1071 fs_info->sectorsize, true);
1072 wait_scrub_stripe_io(stripe);
1073 scrub_verify_one_stripe(stripe, old_error_bitmap);
1074 if (bitmap_empty(&stripe->error_bitmap, stripe->nr_sectors))
1079 * Submit the repaired sectors. For zoned case, we cannot do repair
1080 * in-place, but queue the bg to be relocated.
1082 bitmap_andnot(&repaired, &stripe->init_error_bitmap, &stripe->error_bitmap,
1083 stripe->nr_sectors);
1084 if (!sctx->readonly && !bitmap_empty(&repaired, stripe->nr_sectors)) {
1085 if (btrfs_is_zoned(fs_info)) {
1086 btrfs_repair_one_zone(fs_info, sctx->stripes[0].bg->start);
1088 scrub_write_sectors(sctx, stripe, repaired, false);
1089 wait_scrub_stripe_io(stripe);
1093 scrub_stripe_report_errors(sctx, stripe);
1094 set_bit(SCRUB_STRIPE_FLAG_REPAIR_DONE, &stripe->state);
1095 wake_up(&stripe->repair_wait);
1098 static void scrub_read_endio(struct btrfs_bio *bbio)
1100 struct scrub_stripe *stripe = bbio->private;
1101 struct bio_vec *bvec;
1102 int sector_nr = calc_sector_number(stripe, bio_first_bvec_all(&bbio->bio));
1107 ASSERT(sector_nr < stripe->nr_sectors);
1108 bio_for_each_bvec_all(bvec, &bbio->bio, i)
1109 bio_size += bvec->bv_len;
1110 num_sectors = bio_size >> stripe->bg->fs_info->sectorsize_bits;
1112 if (bbio->bio.bi_status) {
1113 bitmap_set(&stripe->io_error_bitmap, sector_nr, num_sectors);
1114 bitmap_set(&stripe->error_bitmap, sector_nr, num_sectors);
1116 bitmap_clear(&stripe->io_error_bitmap, sector_nr, num_sectors);
1118 bio_put(&bbio->bio);
1119 if (atomic_dec_and_test(&stripe->pending_io)) {
1120 wake_up(&stripe->io_wait);
1121 INIT_WORK(&stripe->work, scrub_stripe_read_repair_worker);
1122 queue_work(stripe->bg->fs_info->scrub_workers, &stripe->work);
1126 static void scrub_write_endio(struct btrfs_bio *bbio)
1128 struct scrub_stripe *stripe = bbio->private;
1129 struct btrfs_fs_info *fs_info = stripe->bg->fs_info;
1130 struct bio_vec *bvec;
1131 int sector_nr = calc_sector_number(stripe, bio_first_bvec_all(&bbio->bio));
1135 bio_for_each_bvec_all(bvec, &bbio->bio, i)
1136 bio_size += bvec->bv_len;
1138 if (bbio->bio.bi_status) {
1139 unsigned long flags;
1141 spin_lock_irqsave(&stripe->write_error_lock, flags);
1142 bitmap_set(&stripe->write_error_bitmap, sector_nr,
1143 bio_size >> fs_info->sectorsize_bits);
1144 spin_unlock_irqrestore(&stripe->write_error_lock, flags);
1146 bio_put(&bbio->bio);
1148 if (atomic_dec_and_test(&stripe->pending_io))
1149 wake_up(&stripe->io_wait);
1152 static void scrub_submit_write_bio(struct scrub_ctx *sctx,
1153 struct scrub_stripe *stripe,
1154 struct btrfs_bio *bbio, bool dev_replace)
1156 struct btrfs_fs_info *fs_info = sctx->fs_info;
1157 u32 bio_len = bbio->bio.bi_iter.bi_size;
1158 u32 bio_off = (bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT) -
1161 fill_writer_pointer_gap(sctx, stripe->physical + bio_off);
1162 atomic_inc(&stripe->pending_io);
1163 btrfs_submit_repair_write(bbio, stripe->mirror_num, dev_replace);
1164 if (!btrfs_is_zoned(fs_info))
1167 * For zoned writeback, queue depth must be 1, thus we must wait for
1168 * the write to finish before the next write.
1170 wait_scrub_stripe_io(stripe);
1173 * And also need to update the write pointer if write finished
1176 if (!test_bit(bio_off >> fs_info->sectorsize_bits,
1177 &stripe->write_error_bitmap))
1178 sctx->write_pointer += bio_len;
1182 * Submit the write bio(s) for the sectors specified by @write_bitmap.
1184 * Here we utilize btrfs_submit_repair_write(), which has some extra benefits:
1186 * - Only needs logical bytenr and mirror_num
1187 * Just like the scrub read path
1189 * - Would only result in writes to the specified mirror
1190 * Unlike the regular writeback path, which would write back to all stripes
1192 * - Handle dev-replace and read-repair writeback differently
1194 static void scrub_write_sectors(struct scrub_ctx *sctx, struct scrub_stripe *stripe,
1195 unsigned long write_bitmap, bool dev_replace)
1197 struct btrfs_fs_info *fs_info = stripe->bg->fs_info;
1198 struct btrfs_bio *bbio = NULL;
1201 for_each_set_bit(sector_nr, &write_bitmap, stripe->nr_sectors) {
1202 struct page *page = scrub_stripe_get_page(stripe, sector_nr);
1203 unsigned int pgoff = scrub_stripe_get_page_offset(stripe, sector_nr);
1206 /* We should only writeback sectors covered by an extent. */
1207 ASSERT(test_bit(sector_nr, &stripe->extent_sector_bitmap));
1209 /* Cannot merge with previous sector, submit the current one. */
1210 if (bbio && sector_nr && !test_bit(sector_nr - 1, &write_bitmap)) {
1211 scrub_submit_write_bio(sctx, stripe, bbio, dev_replace);
1215 bbio = btrfs_bio_alloc(stripe->nr_sectors, REQ_OP_WRITE,
1216 fs_info, scrub_write_endio, stripe);
1217 bbio->bio.bi_iter.bi_sector = (stripe->logical +
1218 (sector_nr << fs_info->sectorsize_bits)) >>
1221 ret = bio_add_page(&bbio->bio, page, fs_info->sectorsize, pgoff);
1222 ASSERT(ret == fs_info->sectorsize);
1225 scrub_submit_write_bio(sctx, stripe, bbio, dev_replace);
1229 * Throttling of IO submission, bandwidth-limit based, the timeslice is 1
1230 * second. Limit can be set via /sys/fs/UUID/devinfo/devid/scrub_speed_max.
1232 static void scrub_throttle_dev_io(struct scrub_ctx *sctx, struct btrfs_device *device,
1233 unsigned int bio_size)
1235 const int time_slice = 1000;
1241 bwlimit = READ_ONCE(device->scrub_speed_max);
1246 * Slice is divided into intervals when the IO is submitted, adjust by
1247 * bwlimit and maximum of 64 intervals.
1249 div = max_t(u32, 1, (u32)(bwlimit / (16 * 1024 * 1024)));
1250 div = min_t(u32, 64, div);
1252 /* Start new epoch, set deadline */
1254 if (sctx->throttle_deadline == 0) {
1255 sctx->throttle_deadline = ktime_add_ms(now, time_slice / div);
1256 sctx->throttle_sent = 0;
1259 /* Still in the time to send? */
1260 if (ktime_before(now, sctx->throttle_deadline)) {
1261 /* If current bio is within the limit, send it */
1262 sctx->throttle_sent += bio_size;
1263 if (sctx->throttle_sent <= div_u64(bwlimit, div))
1266 /* We're over the limit, sleep until the rest of the slice */
1267 delta = ktime_ms_delta(sctx->throttle_deadline, now);
1269 /* New request after deadline, start new epoch */
1276 timeout = div_u64(delta * HZ, 1000);
1277 schedule_timeout_interruptible(timeout);
1280 /* Next call will start the deadline period */
1281 sctx->throttle_deadline = 0;
1285 * Given a physical address, this will calculate it's
1286 * logical offset. if this is a parity stripe, it will return
1287 * the most left data stripe's logical offset.
1289 * return 0 if it is a data stripe, 1 means parity stripe.
1291 static int get_raid56_logic_offset(u64 physical, int num,
1292 struct btrfs_chunk_map *map, u64 *offset,
1298 const int data_stripes = nr_data_stripes(map);
1300 last_offset = (physical - map->stripes[num].physical) * data_stripes;
1302 *stripe_start = last_offset;
1304 *offset = last_offset;
1305 for (i = 0; i < data_stripes; i++) {
1310 *offset = last_offset + btrfs_stripe_nr_to_offset(i);
1312 stripe_nr = (u32)(*offset >> BTRFS_STRIPE_LEN_SHIFT) / data_stripes;
1314 /* Work out the disk rotation on this stripe-set */
1315 rot = stripe_nr % map->num_stripes;
1316 /* calculate which stripe this data locates */
1318 stripe_index = rot % map->num_stripes;
1319 if (stripe_index == num)
1321 if (stripe_index < num)
1324 *offset = last_offset + btrfs_stripe_nr_to_offset(j);
1329 * Return 0 if the extent item range covers any byte of the range.
1330 * Return <0 if the extent item is before @search_start.
1331 * Return >0 if the extent item is after @start_start + @search_len.
1333 static int compare_extent_item_range(struct btrfs_path *path,
1334 u64 search_start, u64 search_len)
1336 struct btrfs_fs_info *fs_info = path->nodes[0]->fs_info;
1338 struct btrfs_key key;
1340 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1341 ASSERT(key.type == BTRFS_EXTENT_ITEM_KEY ||
1342 key.type == BTRFS_METADATA_ITEM_KEY);
1343 if (key.type == BTRFS_METADATA_ITEM_KEY)
1344 len = fs_info->nodesize;
1348 if (key.objectid + len <= search_start)
1350 if (key.objectid >= search_start + search_len)
1356 * Locate one extent item which covers any byte in range
1357 * [@search_start, @search_start + @search_length)
1359 * If the path is not initialized, we will initialize the search by doing
1360 * a btrfs_search_slot().
1361 * If the path is already initialized, we will use the path as the initial
1362 * slot, to avoid duplicated btrfs_search_slot() calls.
1364 * NOTE: If an extent item starts before @search_start, we will still
1365 * return the extent item. This is for data extent crossing stripe boundary.
1367 * Return 0 if we found such extent item, and @path will point to the extent item.
1368 * Return >0 if no such extent item can be found, and @path will be released.
1369 * Return <0 if hit fatal error, and @path will be released.
1371 static int find_first_extent_item(struct btrfs_root *extent_root,
1372 struct btrfs_path *path,
1373 u64 search_start, u64 search_len)
1375 struct btrfs_fs_info *fs_info = extent_root->fs_info;
1376 struct btrfs_key key;
1379 /* Continue using the existing path */
1381 goto search_forward;
1383 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1384 key.type = BTRFS_METADATA_ITEM_KEY;
1386 key.type = BTRFS_EXTENT_ITEM_KEY;
1387 key.objectid = search_start;
1388 key.offset = (u64)-1;
1390 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1396 * Here we intentionally pass 0 as @min_objectid, as there could be
1397 * an extent item starting before @search_start.
1399 ret = btrfs_previous_extent_item(extent_root, path, 0);
1403 * No matter whether we have found an extent item, the next loop will
1404 * properly do every check on the key.
1408 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1409 if (key.objectid >= search_start + search_len)
1411 if (key.type != BTRFS_METADATA_ITEM_KEY &&
1412 key.type != BTRFS_EXTENT_ITEM_KEY)
1415 ret = compare_extent_item_range(path, search_start, search_len);
1421 ret = btrfs_next_item(extent_root, path);
1423 /* Either no more items or a fatal error. */
1424 btrfs_release_path(path);
1428 btrfs_release_path(path);
1432 static void get_extent_info(struct btrfs_path *path, u64 *extent_start_ret,
1433 u64 *size_ret, u64 *flags_ret, u64 *generation_ret)
1435 struct btrfs_key key;
1436 struct btrfs_extent_item *ei;
1438 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1439 ASSERT(key.type == BTRFS_METADATA_ITEM_KEY ||
1440 key.type == BTRFS_EXTENT_ITEM_KEY);
1441 *extent_start_ret = key.objectid;
1442 if (key.type == BTRFS_METADATA_ITEM_KEY)
1443 *size_ret = path->nodes[0]->fs_info->nodesize;
1445 *size_ret = key.offset;
1446 ei = btrfs_item_ptr(path->nodes[0], path->slots[0], struct btrfs_extent_item);
1447 *flags_ret = btrfs_extent_flags(path->nodes[0], ei);
1448 *generation_ret = btrfs_extent_generation(path->nodes[0], ei);
1451 static int sync_write_pointer_for_zoned(struct scrub_ctx *sctx, u64 logical,
1452 u64 physical, u64 physical_end)
1454 struct btrfs_fs_info *fs_info = sctx->fs_info;
1457 if (!btrfs_is_zoned(fs_info))
1460 mutex_lock(&sctx->wr_lock);
1461 if (sctx->write_pointer < physical_end) {
1462 ret = btrfs_sync_zone_write_pointer(sctx->wr_tgtdev, logical,
1464 sctx->write_pointer);
1467 "zoned: failed to recover write pointer");
1469 mutex_unlock(&sctx->wr_lock);
1470 btrfs_dev_clear_zone_empty(sctx->wr_tgtdev, physical);
1475 static void fill_one_extent_info(struct btrfs_fs_info *fs_info,
1476 struct scrub_stripe *stripe,
1477 u64 extent_start, u64 extent_len,
1478 u64 extent_flags, u64 extent_gen)
1480 for (u64 cur_logical = max(stripe->logical, extent_start);
1481 cur_logical < min(stripe->logical + BTRFS_STRIPE_LEN,
1482 extent_start + extent_len);
1483 cur_logical += fs_info->sectorsize) {
1484 const int nr_sector = (cur_logical - stripe->logical) >>
1485 fs_info->sectorsize_bits;
1486 struct scrub_sector_verification *sector =
1487 &stripe->sectors[nr_sector];
1489 set_bit(nr_sector, &stripe->extent_sector_bitmap);
1490 if (extent_flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1491 sector->is_metadata = true;
1492 sector->generation = extent_gen;
1497 static void scrub_stripe_reset_bitmaps(struct scrub_stripe *stripe)
1499 stripe->extent_sector_bitmap = 0;
1500 stripe->init_error_bitmap = 0;
1501 stripe->init_nr_io_errors = 0;
1502 stripe->init_nr_csum_errors = 0;
1503 stripe->init_nr_meta_errors = 0;
1504 stripe->error_bitmap = 0;
1505 stripe->io_error_bitmap = 0;
1506 stripe->csum_error_bitmap = 0;
1507 stripe->meta_error_bitmap = 0;
1511 * Locate one stripe which has at least one extent in its range.
1513 * Return 0 if found such stripe, and store its info into @stripe.
1514 * Return >0 if there is no such stripe in the specified range.
1515 * Return <0 for error.
1517 static int scrub_find_fill_first_stripe(struct btrfs_block_group *bg,
1518 struct btrfs_path *extent_path,
1519 struct btrfs_path *csum_path,
1520 struct btrfs_device *dev, u64 physical,
1521 int mirror_num, u64 logical_start,
1523 struct scrub_stripe *stripe)
1525 struct btrfs_fs_info *fs_info = bg->fs_info;
1526 struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bg->start);
1527 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bg->start);
1528 const u64 logical_end = logical_start + logical_len;
1529 u64 cur_logical = logical_start;
1537 memset(stripe->sectors, 0, sizeof(struct scrub_sector_verification) *
1538 stripe->nr_sectors);
1539 scrub_stripe_reset_bitmaps(stripe);
1541 /* The range must be inside the bg. */
1542 ASSERT(logical_start >= bg->start && logical_end <= bg->start + bg->length);
1544 ret = find_first_extent_item(extent_root, extent_path, logical_start,
1546 /* Either error or not found. */
1549 get_extent_info(extent_path, &extent_start, &extent_len, &extent_flags,
1551 if (extent_flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1552 stripe->nr_meta_extents++;
1553 if (extent_flags & BTRFS_EXTENT_FLAG_DATA)
1554 stripe->nr_data_extents++;
1555 cur_logical = max(extent_start, cur_logical);
1558 * Round down to stripe boundary.
1560 * The extra calculation against bg->start is to handle block groups
1561 * whose logical bytenr is not BTRFS_STRIPE_LEN aligned.
1563 stripe->logical = round_down(cur_logical - bg->start, BTRFS_STRIPE_LEN) +
1565 stripe->physical = physical + stripe->logical - logical_start;
1568 stripe->mirror_num = mirror_num;
1569 stripe_end = stripe->logical + BTRFS_STRIPE_LEN - 1;
1571 /* Fill the first extent info into stripe->sectors[] array. */
1572 fill_one_extent_info(fs_info, stripe, extent_start, extent_len,
1573 extent_flags, extent_gen);
1574 cur_logical = extent_start + extent_len;
1576 /* Fill the extent info for the remaining sectors. */
1577 while (cur_logical <= stripe_end) {
1578 ret = find_first_extent_item(extent_root, extent_path, cur_logical,
1579 stripe_end - cur_logical + 1);
1586 get_extent_info(extent_path, &extent_start, &extent_len,
1587 &extent_flags, &extent_gen);
1588 if (extent_flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1589 stripe->nr_meta_extents++;
1590 if (extent_flags & BTRFS_EXTENT_FLAG_DATA)
1591 stripe->nr_data_extents++;
1592 fill_one_extent_info(fs_info, stripe, extent_start, extent_len,
1593 extent_flags, extent_gen);
1594 cur_logical = extent_start + extent_len;
1597 /* Now fill the data csum. */
1598 if (bg->flags & BTRFS_BLOCK_GROUP_DATA) {
1600 unsigned long csum_bitmap = 0;
1602 /* Csum space should have already been allocated. */
1603 ASSERT(stripe->csums);
1606 * Our csum bitmap should be large enough, as BTRFS_STRIPE_LEN
1607 * should contain at most 16 sectors.
1609 ASSERT(BITS_PER_LONG >= BTRFS_STRIPE_LEN >> fs_info->sectorsize_bits);
1611 ret = btrfs_lookup_csums_bitmap(csum_root, csum_path,
1612 stripe->logical, stripe_end,
1613 stripe->csums, &csum_bitmap);
1619 for_each_set_bit(sector_nr, &csum_bitmap, stripe->nr_sectors) {
1620 stripe->sectors[sector_nr].csum = stripe->csums +
1621 sector_nr * fs_info->csum_size;
1624 set_bit(SCRUB_STRIPE_FLAG_INITIALIZED, &stripe->state);
1629 static void scrub_reset_stripe(struct scrub_stripe *stripe)
1631 scrub_stripe_reset_bitmaps(stripe);
1633 stripe->nr_meta_extents = 0;
1634 stripe->nr_data_extents = 0;
1637 for (int i = 0; i < stripe->nr_sectors; i++) {
1638 stripe->sectors[i].is_metadata = false;
1639 stripe->sectors[i].csum = NULL;
1640 stripe->sectors[i].generation = 0;
1644 static void scrub_submit_extent_sector_read(struct scrub_ctx *sctx,
1645 struct scrub_stripe *stripe)
1647 struct btrfs_fs_info *fs_info = stripe->bg->fs_info;
1648 struct btrfs_bio *bbio = NULL;
1649 unsigned int nr_sectors = min(BTRFS_STRIPE_LEN, stripe->bg->start +
1650 stripe->bg->length - stripe->logical) >>
1651 fs_info->sectorsize_bits;
1652 u64 stripe_len = BTRFS_STRIPE_LEN;
1653 int mirror = stripe->mirror_num;
1656 atomic_inc(&stripe->pending_io);
1658 for_each_set_bit(i, &stripe->extent_sector_bitmap, stripe->nr_sectors) {
1659 struct page *page = scrub_stripe_get_page(stripe, i);
1660 unsigned int pgoff = scrub_stripe_get_page_offset(stripe, i);
1662 /* We're beyond the chunk boundary, no need to read anymore. */
1663 if (i >= nr_sectors)
1666 /* The current sector cannot be merged, submit the bio. */
1669 !test_bit(i - 1, &stripe->extent_sector_bitmap)) ||
1670 bbio->bio.bi_iter.bi_size >= stripe_len)) {
1671 ASSERT(bbio->bio.bi_iter.bi_size);
1672 atomic_inc(&stripe->pending_io);
1673 btrfs_submit_bio(bbio, mirror);
1678 struct btrfs_io_stripe io_stripe = {};
1679 struct btrfs_io_context *bioc = NULL;
1680 const u64 logical = stripe->logical +
1681 (i << fs_info->sectorsize_bits);
1684 bbio = btrfs_bio_alloc(stripe->nr_sectors, REQ_OP_READ,
1685 fs_info, scrub_read_endio, stripe);
1686 bbio->bio.bi_iter.bi_sector = logical >> SECTOR_SHIFT;
1688 io_stripe.is_scrub = true;
1689 err = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
1690 &stripe_len, &bioc, &io_stripe,
1692 btrfs_put_bioc(bioc);
1694 btrfs_bio_end_io(bbio,
1695 errno_to_blk_status(err));
1700 __bio_add_page(&bbio->bio, page, fs_info->sectorsize, pgoff);
1704 ASSERT(bbio->bio.bi_iter.bi_size);
1705 atomic_inc(&stripe->pending_io);
1706 btrfs_submit_bio(bbio, mirror);
1709 if (atomic_dec_and_test(&stripe->pending_io)) {
1710 wake_up(&stripe->io_wait);
1711 INIT_WORK(&stripe->work, scrub_stripe_read_repair_worker);
1712 queue_work(stripe->bg->fs_info->scrub_workers, &stripe->work);
1716 static void scrub_submit_initial_read(struct scrub_ctx *sctx,
1717 struct scrub_stripe *stripe)
1719 struct btrfs_fs_info *fs_info = sctx->fs_info;
1720 struct btrfs_bio *bbio;
1721 unsigned int nr_sectors = min(BTRFS_STRIPE_LEN, stripe->bg->start +
1722 stripe->bg->length - stripe->logical) >>
1723 fs_info->sectorsize_bits;
1724 int mirror = stripe->mirror_num;
1727 ASSERT(stripe->mirror_num > 0);
1728 ASSERT(test_bit(SCRUB_STRIPE_FLAG_INITIALIZED, &stripe->state));
1730 if (btrfs_need_stripe_tree_update(fs_info, stripe->bg->flags)) {
1731 scrub_submit_extent_sector_read(sctx, stripe);
1735 bbio = btrfs_bio_alloc(SCRUB_STRIPE_PAGES, REQ_OP_READ, fs_info,
1736 scrub_read_endio, stripe);
1738 bbio->bio.bi_iter.bi_sector = stripe->logical >> SECTOR_SHIFT;
1739 /* Read the whole range inside the chunk boundary. */
1740 for (unsigned int cur = 0; cur < nr_sectors; cur++) {
1741 struct page *page = scrub_stripe_get_page(stripe, cur);
1742 unsigned int pgoff = scrub_stripe_get_page_offset(stripe, cur);
1745 ret = bio_add_page(&bbio->bio, page, fs_info->sectorsize, pgoff);
1746 /* We should have allocated enough bio vectors. */
1747 ASSERT(ret == fs_info->sectorsize);
1749 atomic_inc(&stripe->pending_io);
1752 * For dev-replace, either user asks to avoid the source dev, or
1753 * the device is missing, we try the next mirror instead.
1755 if (sctx->is_dev_replace &&
1756 (fs_info->dev_replace.cont_reading_from_srcdev_mode ==
1757 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID ||
1758 !stripe->dev->bdev)) {
1759 int num_copies = btrfs_num_copies(fs_info, stripe->bg->start,
1760 stripe->bg->length);
1762 mirror = calc_next_mirror(mirror, num_copies);
1764 btrfs_submit_bio(bbio, mirror);
1767 static bool stripe_has_metadata_error(struct scrub_stripe *stripe)
1771 for_each_set_bit(i, &stripe->error_bitmap, stripe->nr_sectors) {
1772 if (stripe->sectors[i].is_metadata) {
1773 struct btrfs_fs_info *fs_info = stripe->bg->fs_info;
1776 "stripe %llu has unrepaired metadata sector at %llu",
1778 stripe->logical + (i << fs_info->sectorsize_bits));
1785 static void submit_initial_group_read(struct scrub_ctx *sctx,
1786 unsigned int first_slot,
1787 unsigned int nr_stripes)
1789 struct blk_plug plug;
1791 ASSERT(first_slot < SCRUB_TOTAL_STRIPES);
1792 ASSERT(first_slot + nr_stripes <= SCRUB_TOTAL_STRIPES);
1794 scrub_throttle_dev_io(sctx, sctx->stripes[0].dev,
1795 btrfs_stripe_nr_to_offset(nr_stripes));
1796 blk_start_plug(&plug);
1797 for (int i = 0; i < nr_stripes; i++) {
1798 struct scrub_stripe *stripe = &sctx->stripes[first_slot + i];
1800 /* Those stripes should be initialized. */
1801 ASSERT(test_bit(SCRUB_STRIPE_FLAG_INITIALIZED, &stripe->state));
1802 scrub_submit_initial_read(sctx, stripe);
1804 blk_finish_plug(&plug);
1807 static int flush_scrub_stripes(struct scrub_ctx *sctx)
1809 struct btrfs_fs_info *fs_info = sctx->fs_info;
1810 struct scrub_stripe *stripe;
1811 const int nr_stripes = sctx->cur_stripe;
1817 ASSERT(test_bit(SCRUB_STRIPE_FLAG_INITIALIZED, &sctx->stripes[0].state));
1819 /* Submit the stripes which are populated but not submitted. */
1820 if (nr_stripes % SCRUB_STRIPES_PER_GROUP) {
1821 const int first_slot = round_down(nr_stripes, SCRUB_STRIPES_PER_GROUP);
1823 submit_initial_group_read(sctx, first_slot, nr_stripes - first_slot);
1826 for (int i = 0; i < nr_stripes; i++) {
1827 stripe = &sctx->stripes[i];
1829 wait_event(stripe->repair_wait,
1830 test_bit(SCRUB_STRIPE_FLAG_REPAIR_DONE, &stripe->state));
1833 /* Submit for dev-replace. */
1834 if (sctx->is_dev_replace) {
1836 * For dev-replace, if we know there is something wrong with
1837 * metadata, we should immediately abort.
1839 for (int i = 0; i < nr_stripes; i++) {
1840 if (stripe_has_metadata_error(&sctx->stripes[i])) {
1845 for (int i = 0; i < nr_stripes; i++) {
1848 stripe = &sctx->stripes[i];
1850 ASSERT(stripe->dev == fs_info->dev_replace.srcdev);
1852 bitmap_andnot(&good, &stripe->extent_sector_bitmap,
1853 &stripe->error_bitmap, stripe->nr_sectors);
1854 scrub_write_sectors(sctx, stripe, good, true);
1858 /* Wait for the above writebacks to finish. */
1859 for (int i = 0; i < nr_stripes; i++) {
1860 stripe = &sctx->stripes[i];
1862 wait_scrub_stripe_io(stripe);
1863 scrub_reset_stripe(stripe);
1866 sctx->cur_stripe = 0;
1870 static void raid56_scrub_wait_endio(struct bio *bio)
1872 complete(bio->bi_private);
1875 static int queue_scrub_stripe(struct scrub_ctx *sctx, struct btrfs_block_group *bg,
1876 struct btrfs_device *dev, int mirror_num,
1877 u64 logical, u32 length, u64 physical,
1878 u64 *found_logical_ret)
1880 struct scrub_stripe *stripe;
1884 * There should always be one slot left, as caller filling the last
1885 * slot should flush them all.
1887 ASSERT(sctx->cur_stripe < SCRUB_TOTAL_STRIPES);
1889 /* @found_logical_ret must be specified. */
1890 ASSERT(found_logical_ret);
1892 stripe = &sctx->stripes[sctx->cur_stripe];
1893 scrub_reset_stripe(stripe);
1894 ret = scrub_find_fill_first_stripe(bg, &sctx->extent_path,
1895 &sctx->csum_path, dev, physical,
1896 mirror_num, logical, length, stripe);
1897 /* Either >0 as no more extents or <0 for error. */
1900 *found_logical_ret = stripe->logical;
1903 /* We filled one group, submit it. */
1904 if (sctx->cur_stripe % SCRUB_STRIPES_PER_GROUP == 0) {
1905 const int first_slot = sctx->cur_stripe - SCRUB_STRIPES_PER_GROUP;
1907 submit_initial_group_read(sctx, first_slot, SCRUB_STRIPES_PER_GROUP);
1910 /* Last slot used, flush them all. */
1911 if (sctx->cur_stripe == SCRUB_TOTAL_STRIPES)
1912 return flush_scrub_stripes(sctx);
1916 static int scrub_raid56_parity_stripe(struct scrub_ctx *sctx,
1917 struct btrfs_device *scrub_dev,
1918 struct btrfs_block_group *bg,
1919 struct btrfs_chunk_map *map,
1920 u64 full_stripe_start)
1922 DECLARE_COMPLETION_ONSTACK(io_done);
1923 struct btrfs_fs_info *fs_info = sctx->fs_info;
1924 struct btrfs_raid_bio *rbio;
1925 struct btrfs_io_context *bioc = NULL;
1926 struct btrfs_path extent_path = { 0 };
1927 struct btrfs_path csum_path = { 0 };
1929 struct scrub_stripe *stripe;
1930 bool all_empty = true;
1931 const int data_stripes = nr_data_stripes(map);
1932 unsigned long extent_bitmap = 0;
1933 u64 length = btrfs_stripe_nr_to_offset(data_stripes);
1936 ASSERT(sctx->raid56_data_stripes);
1939 * For data stripe search, we cannot re-use the same extent/csum paths,
1940 * as the data stripe bytenr may be smaller than previous extent. Thus
1941 * we have to use our own extent/csum paths.
1943 extent_path.search_commit_root = 1;
1944 extent_path.skip_locking = 1;
1945 csum_path.search_commit_root = 1;
1946 csum_path.skip_locking = 1;
1948 for (int i = 0; i < data_stripes; i++) {
1953 stripe = &sctx->raid56_data_stripes[i];
1954 rot = div_u64(full_stripe_start - bg->start,
1955 data_stripes) >> BTRFS_STRIPE_LEN_SHIFT;
1956 stripe_index = (i + rot) % map->num_stripes;
1957 physical = map->stripes[stripe_index].physical +
1958 btrfs_stripe_nr_to_offset(rot);
1960 scrub_reset_stripe(stripe);
1961 set_bit(SCRUB_STRIPE_FLAG_NO_REPORT, &stripe->state);
1962 ret = scrub_find_fill_first_stripe(bg, &extent_path, &csum_path,
1963 map->stripes[stripe_index].dev, physical, 1,
1964 full_stripe_start + btrfs_stripe_nr_to_offset(i),
1965 BTRFS_STRIPE_LEN, stripe);
1969 * No extent in this data stripe, need to manually mark them
1970 * initialized to make later read submission happy.
1973 stripe->logical = full_stripe_start +
1974 btrfs_stripe_nr_to_offset(i);
1975 stripe->dev = map->stripes[stripe_index].dev;
1976 stripe->mirror_num = 1;
1977 set_bit(SCRUB_STRIPE_FLAG_INITIALIZED, &stripe->state);
1981 /* Check if all data stripes are empty. */
1982 for (int i = 0; i < data_stripes; i++) {
1983 stripe = &sctx->raid56_data_stripes[i];
1984 if (!bitmap_empty(&stripe->extent_sector_bitmap, stripe->nr_sectors)) {
1994 for (int i = 0; i < data_stripes; i++) {
1995 stripe = &sctx->raid56_data_stripes[i];
1996 scrub_submit_initial_read(sctx, stripe);
1998 for (int i = 0; i < data_stripes; i++) {
1999 stripe = &sctx->raid56_data_stripes[i];
2001 wait_event(stripe->repair_wait,
2002 test_bit(SCRUB_STRIPE_FLAG_REPAIR_DONE, &stripe->state));
2004 /* For now, no zoned support for RAID56. */
2005 ASSERT(!btrfs_is_zoned(sctx->fs_info));
2008 * Now all data stripes are properly verified. Check if we have any
2009 * unrepaired, if so abort immediately or we could further corrupt the
2012 * During the loop, also populate extent_bitmap.
2014 for (int i = 0; i < data_stripes; i++) {
2015 unsigned long error;
2017 stripe = &sctx->raid56_data_stripes[i];
2020 * We should only check the errors where there is an extent.
2021 * As we may hit an empty data stripe while it's missing.
2023 bitmap_and(&error, &stripe->error_bitmap,
2024 &stripe->extent_sector_bitmap, stripe->nr_sectors);
2025 if (!bitmap_empty(&error, stripe->nr_sectors)) {
2027 "unrepaired sectors detected, full stripe %llu data stripe %u errors %*pbl",
2028 full_stripe_start, i, stripe->nr_sectors,
2033 bitmap_or(&extent_bitmap, &extent_bitmap,
2034 &stripe->extent_sector_bitmap, stripe->nr_sectors);
2037 /* Now we can check and regenerate the P/Q stripe. */
2038 bio = bio_alloc(NULL, 1, REQ_OP_READ, GFP_NOFS);
2039 bio->bi_iter.bi_sector = full_stripe_start >> SECTOR_SHIFT;
2040 bio->bi_private = &io_done;
2041 bio->bi_end_io = raid56_scrub_wait_endio;
2043 btrfs_bio_counter_inc_blocked(fs_info);
2044 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, full_stripe_start,
2045 &length, &bioc, NULL, NULL);
2047 btrfs_put_bioc(bioc);
2048 btrfs_bio_counter_dec(fs_info);
2051 rbio = raid56_parity_alloc_scrub_rbio(bio, bioc, scrub_dev, &extent_bitmap,
2052 BTRFS_STRIPE_LEN >> fs_info->sectorsize_bits);
2053 btrfs_put_bioc(bioc);
2056 btrfs_bio_counter_dec(fs_info);
2059 /* Use the recovered stripes as cache to avoid read them from disk again. */
2060 for (int i = 0; i < data_stripes; i++) {
2061 stripe = &sctx->raid56_data_stripes[i];
2063 raid56_parity_cache_data_pages(rbio, stripe->pages,
2064 full_stripe_start + (i << BTRFS_STRIPE_LEN_SHIFT));
2066 raid56_parity_submit_scrub_rbio(rbio);
2067 wait_for_completion_io(&io_done);
2068 ret = blk_status_to_errno(bio->bi_status);
2070 btrfs_bio_counter_dec(fs_info);
2072 btrfs_release_path(&extent_path);
2073 btrfs_release_path(&csum_path);
2079 * Scrub one range which can only has simple mirror based profile.
2080 * (Including all range in SINGLE/DUP/RAID1/RAID1C*, and each stripe in
2083 * Since we may need to handle a subset of block group, we need @logical_start
2084 * and @logical_length parameter.
2086 static int scrub_simple_mirror(struct scrub_ctx *sctx,
2087 struct btrfs_block_group *bg,
2088 struct btrfs_chunk_map *map,
2089 u64 logical_start, u64 logical_length,
2090 struct btrfs_device *device,
2091 u64 physical, int mirror_num)
2093 struct btrfs_fs_info *fs_info = sctx->fs_info;
2094 const u64 logical_end = logical_start + logical_length;
2095 u64 cur_logical = logical_start;
2098 /* The range must be inside the bg */
2099 ASSERT(logical_start >= bg->start && logical_end <= bg->start + bg->length);
2101 /* Go through each extent items inside the logical range */
2102 while (cur_logical < logical_end) {
2103 u64 found_logical = U64_MAX;
2104 u64 cur_physical = physical + cur_logical - logical_start;
2107 if (atomic_read(&fs_info->scrub_cancel_req) ||
2108 atomic_read(&sctx->cancel_req)) {
2113 if (atomic_read(&fs_info->scrub_pause_req)) {
2114 /* Push queued extents */
2115 scrub_blocked_if_needed(fs_info);
2117 /* Block group removed? */
2118 spin_lock(&bg->lock);
2119 if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &bg->runtime_flags)) {
2120 spin_unlock(&bg->lock);
2124 spin_unlock(&bg->lock);
2126 ret = queue_scrub_stripe(sctx, bg, device, mirror_num,
2127 cur_logical, logical_end - cur_logical,
2128 cur_physical, &found_logical);
2130 /* No more extent, just update the accounting */
2131 sctx->stat.last_physical = physical + logical_length;
2138 /* queue_scrub_stripe() returned 0, @found_logical must be updated. */
2139 ASSERT(found_logical != U64_MAX);
2140 cur_logical = found_logical + BTRFS_STRIPE_LEN;
2142 /* Don't hold CPU for too long time */
2148 /* Calculate the full stripe length for simple stripe based profiles */
2149 static u64 simple_stripe_full_stripe_len(const struct btrfs_chunk_map *map)
2151 ASSERT(map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2152 BTRFS_BLOCK_GROUP_RAID10));
2154 return btrfs_stripe_nr_to_offset(map->num_stripes / map->sub_stripes);
2157 /* Get the logical bytenr for the stripe */
2158 static u64 simple_stripe_get_logical(struct btrfs_chunk_map *map,
2159 struct btrfs_block_group *bg,
2162 ASSERT(map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2163 BTRFS_BLOCK_GROUP_RAID10));
2164 ASSERT(stripe_index < map->num_stripes);
2167 * (stripe_index / sub_stripes) gives how many data stripes we need to
2170 return btrfs_stripe_nr_to_offset(stripe_index / map->sub_stripes) +
2174 /* Get the mirror number for the stripe */
2175 static int simple_stripe_mirror_num(struct btrfs_chunk_map *map, int stripe_index)
2177 ASSERT(map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2178 BTRFS_BLOCK_GROUP_RAID10));
2179 ASSERT(stripe_index < map->num_stripes);
2181 /* For RAID0, it's fixed to 1, for RAID10 it's 0,1,0,1... */
2182 return stripe_index % map->sub_stripes + 1;
2185 static int scrub_simple_stripe(struct scrub_ctx *sctx,
2186 struct btrfs_block_group *bg,
2187 struct btrfs_chunk_map *map,
2188 struct btrfs_device *device,
2191 const u64 logical_increment = simple_stripe_full_stripe_len(map);
2192 const u64 orig_logical = simple_stripe_get_logical(map, bg, stripe_index);
2193 const u64 orig_physical = map->stripes[stripe_index].physical;
2194 const int mirror_num = simple_stripe_mirror_num(map, stripe_index);
2195 u64 cur_logical = orig_logical;
2196 u64 cur_physical = orig_physical;
2199 while (cur_logical < bg->start + bg->length) {
2201 * Inside each stripe, RAID0 is just SINGLE, and RAID10 is
2202 * just RAID1, so we can reuse scrub_simple_mirror() to scrub
2205 ret = scrub_simple_mirror(sctx, bg, map, cur_logical,
2206 BTRFS_STRIPE_LEN, device, cur_physical,
2210 /* Skip to next stripe which belongs to the target device */
2211 cur_logical += logical_increment;
2212 /* For physical offset, we just go to next stripe */
2213 cur_physical += BTRFS_STRIPE_LEN;
2218 static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
2219 struct btrfs_block_group *bg,
2220 struct btrfs_chunk_map *map,
2221 struct btrfs_device *scrub_dev,
2224 struct btrfs_fs_info *fs_info = sctx->fs_info;
2225 const u64 profile = map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK;
2226 const u64 chunk_logical = bg->start;
2229 u64 physical = map->stripes[stripe_index].physical;
2230 const u64 dev_stripe_len = btrfs_calc_stripe_length(map);
2231 const u64 physical_end = physical + dev_stripe_len;
2234 /* The logical increment after finishing one stripe */
2236 /* Offset inside the chunk */
2241 /* Extent_path should be released by now. */
2242 ASSERT(sctx->extent_path.nodes[0] == NULL);
2244 scrub_blocked_if_needed(fs_info);
2246 if (sctx->is_dev_replace &&
2247 btrfs_dev_is_sequential(sctx->wr_tgtdev, physical)) {
2248 mutex_lock(&sctx->wr_lock);
2249 sctx->write_pointer = physical;
2250 mutex_unlock(&sctx->wr_lock);
2253 /* Prepare the extra data stripes used by RAID56. */
2254 if (profile & BTRFS_BLOCK_GROUP_RAID56_MASK) {
2255 ASSERT(sctx->raid56_data_stripes == NULL);
2257 sctx->raid56_data_stripes = kcalloc(nr_data_stripes(map),
2258 sizeof(struct scrub_stripe),
2260 if (!sctx->raid56_data_stripes) {
2264 for (int i = 0; i < nr_data_stripes(map); i++) {
2265 ret = init_scrub_stripe(fs_info,
2266 &sctx->raid56_data_stripes[i]);
2269 sctx->raid56_data_stripes[i].bg = bg;
2270 sctx->raid56_data_stripes[i].sctx = sctx;
2274 * There used to be a big double loop to handle all profiles using the
2275 * same routine, which grows larger and more gross over time.
2277 * So here we handle each profile differently, so simpler profiles
2278 * have simpler scrubbing function.
2280 if (!(profile & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10 |
2281 BTRFS_BLOCK_GROUP_RAID56_MASK))) {
2283 * Above check rules out all complex profile, the remaining
2284 * profiles are SINGLE|DUP|RAID1|RAID1C*, which is simple
2285 * mirrored duplication without stripe.
2287 * Only @physical and @mirror_num needs to calculated using
2290 ret = scrub_simple_mirror(sctx, bg, map, bg->start, bg->length,
2291 scrub_dev, map->stripes[stripe_index].physical,
2296 if (profile & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
2297 ret = scrub_simple_stripe(sctx, bg, map, scrub_dev, stripe_index);
2298 offset = btrfs_stripe_nr_to_offset(stripe_index / map->sub_stripes);
2302 /* Only RAID56 goes through the old code */
2303 ASSERT(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK);
2306 /* Calculate the logical end of the stripe */
2307 get_raid56_logic_offset(physical_end, stripe_index,
2308 map, &logic_end, NULL);
2309 logic_end += chunk_logical;
2311 /* Initialize @offset in case we need to go to out: label */
2312 get_raid56_logic_offset(physical, stripe_index, map, &offset, NULL);
2313 increment = btrfs_stripe_nr_to_offset(nr_data_stripes(map));
2316 * Due to the rotation, for RAID56 it's better to iterate each stripe
2317 * using their physical offset.
2319 while (physical < physical_end) {
2320 ret = get_raid56_logic_offset(physical, stripe_index, map,
2321 &logical, &stripe_logical);
2322 logical += chunk_logical;
2324 /* it is parity strip */
2325 stripe_logical += chunk_logical;
2326 ret = scrub_raid56_parity_stripe(sctx, scrub_dev, bg,
2327 map, stripe_logical);
2334 * Now we're at a data stripe, scrub each extents in the range.
2336 * At this stage, if we ignore the repair part, inside each data
2337 * stripe it is no different than SINGLE profile.
2338 * We can reuse scrub_simple_mirror() here, as the repair part
2339 * is still based on @mirror_num.
2341 ret = scrub_simple_mirror(sctx, bg, map, logical, BTRFS_STRIPE_LEN,
2342 scrub_dev, physical, 1);
2346 logical += increment;
2347 physical += BTRFS_STRIPE_LEN;
2348 spin_lock(&sctx->stat_lock);
2350 sctx->stat.last_physical =
2351 map->stripes[stripe_index].physical + dev_stripe_len;
2353 sctx->stat.last_physical = physical;
2354 spin_unlock(&sctx->stat_lock);
2359 ret2 = flush_scrub_stripes(sctx);
2362 btrfs_release_path(&sctx->extent_path);
2363 btrfs_release_path(&sctx->csum_path);
2365 if (sctx->raid56_data_stripes) {
2366 for (int i = 0; i < nr_data_stripes(map); i++)
2367 release_scrub_stripe(&sctx->raid56_data_stripes[i]);
2368 kfree(sctx->raid56_data_stripes);
2369 sctx->raid56_data_stripes = NULL;
2372 if (sctx->is_dev_replace && ret >= 0) {
2375 ret2 = sync_write_pointer_for_zoned(sctx,
2376 chunk_logical + offset,
2377 map->stripes[stripe_index].physical,
2383 return ret < 0 ? ret : 0;
2386 static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx,
2387 struct btrfs_block_group *bg,
2388 struct btrfs_device *scrub_dev,
2392 struct btrfs_fs_info *fs_info = sctx->fs_info;
2393 struct btrfs_chunk_map *map;
2397 map = btrfs_find_chunk_map(fs_info, bg->start, bg->length);
2400 * Might have been an unused block group deleted by the cleaner
2401 * kthread or relocation.
2403 spin_lock(&bg->lock);
2404 if (!test_bit(BLOCK_GROUP_FLAG_REMOVED, &bg->runtime_flags))
2406 spin_unlock(&bg->lock);
2410 if (map->start != bg->start)
2412 if (map->chunk_len < dev_extent_len)
2415 for (i = 0; i < map->num_stripes; ++i) {
2416 if (map->stripes[i].dev->bdev == scrub_dev->bdev &&
2417 map->stripes[i].physical == dev_offset) {
2418 ret = scrub_stripe(sctx, bg, map, scrub_dev, i);
2424 btrfs_free_chunk_map(map);
2429 static int finish_extent_writes_for_zoned(struct btrfs_root *root,
2430 struct btrfs_block_group *cache)
2432 struct btrfs_fs_info *fs_info = cache->fs_info;
2433 struct btrfs_trans_handle *trans;
2435 if (!btrfs_is_zoned(fs_info))
2438 btrfs_wait_block_group_reservations(cache);
2439 btrfs_wait_nocow_writers(cache);
2440 btrfs_wait_ordered_roots(fs_info, U64_MAX, cache->start, cache->length);
2442 trans = btrfs_join_transaction(root);
2444 return PTR_ERR(trans);
2445 return btrfs_commit_transaction(trans);
2448 static noinline_for_stack
2449 int scrub_enumerate_chunks(struct scrub_ctx *sctx,
2450 struct btrfs_device *scrub_dev, u64 start, u64 end)
2452 struct btrfs_dev_extent *dev_extent = NULL;
2453 struct btrfs_path *path;
2454 struct btrfs_fs_info *fs_info = sctx->fs_info;
2455 struct btrfs_root *root = fs_info->dev_root;
2460 struct extent_buffer *l;
2461 struct btrfs_key key;
2462 struct btrfs_key found_key;
2463 struct btrfs_block_group *cache;
2464 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
2466 path = btrfs_alloc_path();
2470 path->reada = READA_FORWARD;
2471 path->search_commit_root = 1;
2472 path->skip_locking = 1;
2474 key.objectid = scrub_dev->devid;
2476 key.type = BTRFS_DEV_EXTENT_KEY;
2481 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2485 if (path->slots[0] >=
2486 btrfs_header_nritems(path->nodes[0])) {
2487 ret = btrfs_next_leaf(root, path);
2500 slot = path->slots[0];
2502 btrfs_item_key_to_cpu(l, &found_key, slot);
2504 if (found_key.objectid != scrub_dev->devid)
2507 if (found_key.type != BTRFS_DEV_EXTENT_KEY)
2510 if (found_key.offset >= end)
2513 if (found_key.offset < key.offset)
2516 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2517 dev_extent_len = btrfs_dev_extent_length(l, dev_extent);
2519 if (found_key.offset + dev_extent_len <= start)
2522 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2525 * get a reference on the corresponding block group to prevent
2526 * the chunk from going away while we scrub it
2528 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2530 /* some chunks are removed but not committed to disk yet,
2531 * continue scrubbing */
2535 ASSERT(cache->start <= chunk_offset);
2537 * We are using the commit root to search for device extents, so
2538 * that means we could have found a device extent item from a
2539 * block group that was deleted in the current transaction. The
2540 * logical start offset of the deleted block group, stored at
2541 * @chunk_offset, might be part of the logical address range of
2542 * a new block group (which uses different physical extents).
2543 * In this case btrfs_lookup_block_group() has returned the new
2544 * block group, and its start address is less than @chunk_offset.
2546 * We skip such new block groups, because it's pointless to
2547 * process them, as we won't find their extents because we search
2548 * for them using the commit root of the extent tree. For a device
2549 * replace it's also fine to skip it, we won't miss copying them
2550 * to the target device because we have the write duplication
2551 * setup through the regular write path (by btrfs_map_block()),
2552 * and we have committed a transaction when we started the device
2553 * replace, right after setting up the device replace state.
2555 if (cache->start < chunk_offset) {
2556 btrfs_put_block_group(cache);
2560 if (sctx->is_dev_replace && btrfs_is_zoned(fs_info)) {
2561 if (!test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags)) {
2562 btrfs_put_block_group(cache);
2568 * Make sure that while we are scrubbing the corresponding block
2569 * group doesn't get its logical address and its device extents
2570 * reused for another block group, which can possibly be of a
2571 * different type and different profile. We do this to prevent
2572 * false error detections and crashes due to bogus attempts to
2575 spin_lock(&cache->lock);
2576 if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &cache->runtime_flags)) {
2577 spin_unlock(&cache->lock);
2578 btrfs_put_block_group(cache);
2581 btrfs_freeze_block_group(cache);
2582 spin_unlock(&cache->lock);
2585 * we need call btrfs_inc_block_group_ro() with scrubs_paused,
2586 * to avoid deadlock caused by:
2587 * btrfs_inc_block_group_ro()
2588 * -> btrfs_wait_for_commit()
2589 * -> btrfs_commit_transaction()
2590 * -> btrfs_scrub_pause()
2592 scrub_pause_on(fs_info);
2595 * Don't do chunk preallocation for scrub.
2597 * This is especially important for SYSTEM bgs, or we can hit
2598 * -EFBIG from btrfs_finish_chunk_alloc() like:
2599 * 1. The only SYSTEM bg is marked RO.
2600 * Since SYSTEM bg is small, that's pretty common.
2601 * 2. New SYSTEM bg will be allocated
2602 * Due to regular version will allocate new chunk.
2603 * 3. New SYSTEM bg is empty and will get cleaned up
2604 * Before cleanup really happens, it's marked RO again.
2605 * 4. Empty SYSTEM bg get scrubbed
2608 * This can easily boost the amount of SYSTEM chunks if cleaner
2609 * thread can't be triggered fast enough, and use up all space
2610 * of btrfs_super_block::sys_chunk_array
2612 * While for dev replace, we need to try our best to mark block
2613 * group RO, to prevent race between:
2614 * - Write duplication
2615 * Contains latest data
2617 * Contains data from commit tree
2619 * If target block group is not marked RO, nocow writes can
2620 * be overwritten by scrub copy, causing data corruption.
2621 * So for dev-replace, it's not allowed to continue if a block
2624 ret = btrfs_inc_block_group_ro(cache, sctx->is_dev_replace);
2625 if (!ret && sctx->is_dev_replace) {
2626 ret = finish_extent_writes_for_zoned(root, cache);
2628 btrfs_dec_block_group_ro(cache);
2629 scrub_pause_off(fs_info);
2630 btrfs_put_block_group(cache);
2637 } else if (ret == -ENOSPC && !sctx->is_dev_replace &&
2638 !(cache->flags & BTRFS_BLOCK_GROUP_RAID56_MASK)) {
2640 * btrfs_inc_block_group_ro return -ENOSPC when it
2641 * failed in creating new chunk for metadata.
2642 * It is not a problem for scrub, because
2643 * metadata are always cowed, and our scrub paused
2644 * commit_transactions.
2646 * For RAID56 chunks, we have to mark them read-only
2647 * for scrub, as later we would use our own cache
2648 * out of RAID56 realm.
2649 * Thus we want the RAID56 bg to be marked RO to
2650 * prevent RMW from screwing up out cache.
2653 } else if (ret == -ETXTBSY) {
2655 "skipping scrub of block group %llu due to active swapfile",
2657 scrub_pause_off(fs_info);
2662 "failed setting block group ro: %d", ret);
2663 btrfs_unfreeze_block_group(cache);
2664 btrfs_put_block_group(cache);
2665 scrub_pause_off(fs_info);
2670 * Now the target block is marked RO, wait for nocow writes to
2671 * finish before dev-replace.
2672 * COW is fine, as COW never overwrites extents in commit tree.
2674 if (sctx->is_dev_replace) {
2675 btrfs_wait_nocow_writers(cache);
2676 btrfs_wait_ordered_roots(fs_info, U64_MAX, cache->start,
2680 scrub_pause_off(fs_info);
2681 down_write(&dev_replace->rwsem);
2682 dev_replace->cursor_right = found_key.offset + dev_extent_len;
2683 dev_replace->cursor_left = found_key.offset;
2684 dev_replace->item_needs_writeback = 1;
2685 up_write(&dev_replace->rwsem);
2687 ret = scrub_chunk(sctx, cache, scrub_dev, found_key.offset,
2689 if (sctx->is_dev_replace &&
2690 !btrfs_finish_block_group_to_copy(dev_replace->srcdev,
2691 cache, found_key.offset))
2694 down_write(&dev_replace->rwsem);
2695 dev_replace->cursor_left = dev_replace->cursor_right;
2696 dev_replace->item_needs_writeback = 1;
2697 up_write(&dev_replace->rwsem);
2700 btrfs_dec_block_group_ro(cache);
2703 * We might have prevented the cleaner kthread from deleting
2704 * this block group if it was already unused because we raced
2705 * and set it to RO mode first. So add it back to the unused
2706 * list, otherwise it might not ever be deleted unless a manual
2707 * balance is triggered or it becomes used and unused again.
2709 spin_lock(&cache->lock);
2710 if (!test_bit(BLOCK_GROUP_FLAG_REMOVED, &cache->runtime_flags) &&
2711 !cache->ro && cache->reserved == 0 && cache->used == 0) {
2712 spin_unlock(&cache->lock);
2713 if (btrfs_test_opt(fs_info, DISCARD_ASYNC))
2714 btrfs_discard_queue_work(&fs_info->discard_ctl,
2717 btrfs_mark_bg_unused(cache);
2719 spin_unlock(&cache->lock);
2722 btrfs_unfreeze_block_group(cache);
2723 btrfs_put_block_group(cache);
2726 if (sctx->is_dev_replace &&
2727 atomic64_read(&dev_replace->num_write_errors) > 0) {
2731 if (sctx->stat.malloc_errors > 0) {
2736 key.offset = found_key.offset + dev_extent_len;
2737 btrfs_release_path(path);
2740 btrfs_free_path(path);
2745 static int scrub_one_super(struct scrub_ctx *sctx, struct btrfs_device *dev,
2746 struct page *page, u64 physical, u64 generation)
2748 struct btrfs_fs_info *fs_info = sctx->fs_info;
2749 struct bio_vec bvec;
2751 struct btrfs_super_block *sb = page_address(page);
2754 bio_init(&bio, dev->bdev, &bvec, 1, REQ_OP_READ);
2755 bio.bi_iter.bi_sector = physical >> SECTOR_SHIFT;
2756 __bio_add_page(&bio, page, BTRFS_SUPER_INFO_SIZE, 0);
2757 ret = submit_bio_wait(&bio);
2762 ret = btrfs_check_super_csum(fs_info, sb);
2764 btrfs_err_rl(fs_info,
2765 "super block at physical %llu devid %llu has bad csum",
2766 physical, dev->devid);
2769 if (btrfs_super_generation(sb) != generation) {
2770 btrfs_err_rl(fs_info,
2771 "super block at physical %llu devid %llu has bad generation %llu expect %llu",
2772 physical, dev->devid,
2773 btrfs_super_generation(sb), generation);
2777 return btrfs_validate_super(fs_info, sb, -1);
2780 static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx,
2781 struct btrfs_device *scrub_dev)
2788 struct btrfs_fs_info *fs_info = sctx->fs_info;
2790 if (BTRFS_FS_ERROR(fs_info))
2793 page = alloc_page(GFP_KERNEL);
2795 spin_lock(&sctx->stat_lock);
2796 sctx->stat.malloc_errors++;
2797 spin_unlock(&sctx->stat_lock);
2801 /* Seed devices of a new filesystem has their own generation. */
2802 if (scrub_dev->fs_devices != fs_info->fs_devices)
2803 gen = scrub_dev->generation;
2805 gen = btrfs_get_last_trans_committed(fs_info);
2807 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2808 ret = btrfs_sb_log_location(scrub_dev, i, 0, &bytenr);
2813 spin_lock(&sctx->stat_lock);
2814 sctx->stat.super_errors++;
2815 spin_unlock(&sctx->stat_lock);
2819 if (bytenr + BTRFS_SUPER_INFO_SIZE >
2820 scrub_dev->commit_total_bytes)
2822 if (!btrfs_check_super_location(scrub_dev, bytenr))
2825 ret = scrub_one_super(sctx, scrub_dev, page, bytenr, gen);
2827 spin_lock(&sctx->stat_lock);
2828 sctx->stat.super_errors++;
2829 spin_unlock(&sctx->stat_lock);
2836 static void scrub_workers_put(struct btrfs_fs_info *fs_info)
2838 if (refcount_dec_and_mutex_lock(&fs_info->scrub_workers_refcnt,
2839 &fs_info->scrub_lock)) {
2840 struct workqueue_struct *scrub_workers = fs_info->scrub_workers;
2842 fs_info->scrub_workers = NULL;
2843 mutex_unlock(&fs_info->scrub_lock);
2846 destroy_workqueue(scrub_workers);
2851 * get a reference count on fs_info->scrub_workers. start worker if necessary
2853 static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info)
2855 struct workqueue_struct *scrub_workers = NULL;
2856 unsigned int flags = WQ_FREEZABLE | WQ_UNBOUND;
2857 int max_active = fs_info->thread_pool_size;
2860 if (refcount_inc_not_zero(&fs_info->scrub_workers_refcnt))
2863 scrub_workers = alloc_workqueue("btrfs-scrub", flags, max_active);
2867 mutex_lock(&fs_info->scrub_lock);
2868 if (refcount_read(&fs_info->scrub_workers_refcnt) == 0) {
2869 ASSERT(fs_info->scrub_workers == NULL);
2870 fs_info->scrub_workers = scrub_workers;
2871 refcount_set(&fs_info->scrub_workers_refcnt, 1);
2872 mutex_unlock(&fs_info->scrub_lock);
2875 /* Other thread raced in and created the workers for us */
2876 refcount_inc(&fs_info->scrub_workers_refcnt);
2877 mutex_unlock(&fs_info->scrub_lock);
2881 destroy_workqueue(scrub_workers);
2885 int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start,
2886 u64 end, struct btrfs_scrub_progress *progress,
2887 int readonly, int is_dev_replace)
2889 struct btrfs_dev_lookup_args args = { .devid = devid };
2890 struct scrub_ctx *sctx;
2892 struct btrfs_device *dev;
2893 unsigned int nofs_flag;
2894 bool need_commit = false;
2896 if (btrfs_fs_closing(fs_info))
2899 /* At mount time we have ensured nodesize is in the range of [4K, 64K]. */
2900 ASSERT(fs_info->nodesize <= BTRFS_STRIPE_LEN);
2903 * SCRUB_MAX_SECTORS_PER_BLOCK is calculated using the largest possible
2904 * value (max nodesize / min sectorsize), thus nodesize should always
2907 ASSERT(fs_info->nodesize <=
2908 SCRUB_MAX_SECTORS_PER_BLOCK << fs_info->sectorsize_bits);
2910 /* Allocate outside of device_list_mutex */
2911 sctx = scrub_setup_ctx(fs_info, is_dev_replace);
2913 return PTR_ERR(sctx);
2915 ret = scrub_workers_get(fs_info);
2919 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2920 dev = btrfs_find_device(fs_info->fs_devices, &args);
2921 if (!dev || (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) &&
2923 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2928 if (!is_dev_replace && !readonly &&
2929 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
2930 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2931 btrfs_err_in_rcu(fs_info,
2932 "scrub on devid %llu: filesystem on %s is not writable",
2933 devid, btrfs_dev_name(dev));
2938 mutex_lock(&fs_info->scrub_lock);
2939 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
2940 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &dev->dev_state)) {
2941 mutex_unlock(&fs_info->scrub_lock);
2942 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2947 down_read(&fs_info->dev_replace.rwsem);
2948 if (dev->scrub_ctx ||
2950 btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) {
2951 up_read(&fs_info->dev_replace.rwsem);
2952 mutex_unlock(&fs_info->scrub_lock);
2953 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2957 up_read(&fs_info->dev_replace.rwsem);
2959 sctx->readonly = readonly;
2960 dev->scrub_ctx = sctx;
2961 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2964 * checking @scrub_pause_req here, we can avoid
2965 * race between committing transaction and scrubbing.
2967 __scrub_blocked_if_needed(fs_info);
2968 atomic_inc(&fs_info->scrubs_running);
2969 mutex_unlock(&fs_info->scrub_lock);
2972 * In order to avoid deadlock with reclaim when there is a transaction
2973 * trying to pause scrub, make sure we use GFP_NOFS for all the
2974 * allocations done at btrfs_scrub_sectors() and scrub_sectors_for_parity()
2975 * invoked by our callees. The pausing request is done when the
2976 * transaction commit starts, and it blocks the transaction until scrub
2977 * is paused (done at specific points at scrub_stripe() or right above
2978 * before incrementing fs_info->scrubs_running).
2980 nofs_flag = memalloc_nofs_save();
2981 if (!is_dev_replace) {
2982 u64 old_super_errors;
2984 spin_lock(&sctx->stat_lock);
2985 old_super_errors = sctx->stat.super_errors;
2986 spin_unlock(&sctx->stat_lock);
2988 btrfs_info(fs_info, "scrub: started on devid %llu", devid);
2990 * by holding device list mutex, we can
2991 * kick off writing super in log tree sync.
2993 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2994 ret = scrub_supers(sctx, dev);
2995 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2997 spin_lock(&sctx->stat_lock);
2999 * Super block errors found, but we can not commit transaction
3000 * at current context, since btrfs_commit_transaction() needs
3001 * to pause the current running scrub (hold by ourselves).
3003 if (sctx->stat.super_errors > old_super_errors && !sctx->readonly)
3005 spin_unlock(&sctx->stat_lock);
3009 ret = scrub_enumerate_chunks(sctx, dev, start, end);
3010 memalloc_nofs_restore(nofs_flag);
3012 atomic_dec(&fs_info->scrubs_running);
3013 wake_up(&fs_info->scrub_pause_wait);
3016 memcpy(progress, &sctx->stat, sizeof(*progress));
3018 if (!is_dev_replace)
3019 btrfs_info(fs_info, "scrub: %s on devid %llu with status: %d",
3020 ret ? "not finished" : "finished", devid, ret);
3022 mutex_lock(&fs_info->scrub_lock);
3023 dev->scrub_ctx = NULL;
3024 mutex_unlock(&fs_info->scrub_lock);
3026 scrub_workers_put(fs_info);
3027 scrub_put_ctx(sctx);
3030 * We found some super block errors before, now try to force a
3031 * transaction commit, as scrub has finished.
3034 struct btrfs_trans_handle *trans;
3036 trans = btrfs_start_transaction(fs_info->tree_root, 0);
3037 if (IS_ERR(trans)) {
3038 ret = PTR_ERR(trans);
3040 "scrub: failed to start transaction to fix super block errors: %d", ret);
3043 ret = btrfs_commit_transaction(trans);
3046 "scrub: failed to commit transaction to fix super block errors: %d", ret);
3050 scrub_workers_put(fs_info);
3052 scrub_free_ctx(sctx);
3057 void btrfs_scrub_pause(struct btrfs_fs_info *fs_info)
3059 mutex_lock(&fs_info->scrub_lock);
3060 atomic_inc(&fs_info->scrub_pause_req);
3061 while (atomic_read(&fs_info->scrubs_paused) !=
3062 atomic_read(&fs_info->scrubs_running)) {
3063 mutex_unlock(&fs_info->scrub_lock);
3064 wait_event(fs_info->scrub_pause_wait,
3065 atomic_read(&fs_info->scrubs_paused) ==
3066 atomic_read(&fs_info->scrubs_running));
3067 mutex_lock(&fs_info->scrub_lock);
3069 mutex_unlock(&fs_info->scrub_lock);
3072 void btrfs_scrub_continue(struct btrfs_fs_info *fs_info)
3074 atomic_dec(&fs_info->scrub_pause_req);
3075 wake_up(&fs_info->scrub_pause_wait);
3078 int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info)
3080 mutex_lock(&fs_info->scrub_lock);
3081 if (!atomic_read(&fs_info->scrubs_running)) {
3082 mutex_unlock(&fs_info->scrub_lock);
3086 atomic_inc(&fs_info->scrub_cancel_req);
3087 while (atomic_read(&fs_info->scrubs_running)) {
3088 mutex_unlock(&fs_info->scrub_lock);
3089 wait_event(fs_info->scrub_pause_wait,
3090 atomic_read(&fs_info->scrubs_running) == 0);
3091 mutex_lock(&fs_info->scrub_lock);
3093 atomic_dec(&fs_info->scrub_cancel_req);
3094 mutex_unlock(&fs_info->scrub_lock);
3099 int btrfs_scrub_cancel_dev(struct btrfs_device *dev)
3101 struct btrfs_fs_info *fs_info = dev->fs_info;
3102 struct scrub_ctx *sctx;
3104 mutex_lock(&fs_info->scrub_lock);
3105 sctx = dev->scrub_ctx;
3107 mutex_unlock(&fs_info->scrub_lock);
3110 atomic_inc(&sctx->cancel_req);
3111 while (dev->scrub_ctx) {
3112 mutex_unlock(&fs_info->scrub_lock);
3113 wait_event(fs_info->scrub_pause_wait,
3114 dev->scrub_ctx == NULL);
3115 mutex_lock(&fs_info->scrub_lock);
3117 mutex_unlock(&fs_info->scrub_lock);
3122 int btrfs_scrub_progress(struct btrfs_fs_info *fs_info, u64 devid,
3123 struct btrfs_scrub_progress *progress)
3125 struct btrfs_dev_lookup_args args = { .devid = devid };
3126 struct btrfs_device *dev;
3127 struct scrub_ctx *sctx = NULL;
3129 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3130 dev = btrfs_find_device(fs_info->fs_devices, &args);
3132 sctx = dev->scrub_ctx;
3134 memcpy(progress, &sctx->stat, sizeof(*progress));
3135 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3137 return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV;