2 * Copyright (C) 2011, 2012 STRATO. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/blkdev.h>
20 #include <linux/ratelimit.h>
24 #include "ordered-data.h"
25 #include "transaction.h"
27 #include "extent_io.h"
28 #include "dev-replace.h"
29 #include "check-integrity.h"
30 #include "rcu-string.h"
34 * This is only the first step towards a full-features scrub. It reads all
35 * extent and super block and verifies the checksums. In case a bad checksum
36 * is found or the extent cannot be read, good data will be written back if
39 * Future enhancements:
40 * - In case an unrepairable extent is encountered, track which files are
41 * affected and report them
42 * - track and record media errors, throw out bad devices
43 * - add a mode to also read unallocated space
50 * the following three values only influence the performance.
51 * The last one configures the number of parallel and outstanding I/O
52 * operations. The first two values configure an upper limit for the number
53 * of (dynamically allocated) pages that are added to a bio.
55 #define SCRUB_PAGES_PER_RD_BIO 32 /* 128k per bio */
56 #define SCRUB_PAGES_PER_WR_BIO 32 /* 128k per bio */
57 #define SCRUB_BIOS_PER_SCTX 64 /* 8MB per device in flight */
60 * the following value times PAGE_SIZE needs to be large enough to match the
61 * largest node/leaf/sector size that shall be supported.
62 * Values larger than BTRFS_STRIPE_LEN are not supported.
64 #define SCRUB_MAX_PAGES_PER_BLOCK 16 /* 64k per node/leaf/sector */
66 struct scrub_recover {
68 struct btrfs_bio *bbio;
73 struct scrub_block *sblock;
75 struct btrfs_device *dev;
76 struct list_head list;
77 u64 flags; /* extent flags */
81 u64 physical_for_dev_replace;
84 unsigned int mirror_num:8;
85 unsigned int have_csum:1;
86 unsigned int io_error:1;
88 u8 csum[BTRFS_CSUM_SIZE];
90 struct scrub_recover *recover;
95 struct scrub_ctx *sctx;
96 struct btrfs_device *dev;
101 #if SCRUB_PAGES_PER_WR_BIO >= SCRUB_PAGES_PER_RD_BIO
102 struct scrub_page *pagev[SCRUB_PAGES_PER_WR_BIO];
104 struct scrub_page *pagev[SCRUB_PAGES_PER_RD_BIO];
108 struct btrfs_work work;
112 struct scrub_page *pagev[SCRUB_MAX_PAGES_PER_BLOCK];
114 atomic_t outstanding_pages;
115 atomic_t refs; /* free mem on transition to zero */
116 struct scrub_ctx *sctx;
117 struct scrub_parity *sparity;
119 unsigned int header_error:1;
120 unsigned int checksum_error:1;
121 unsigned int no_io_error_seen:1;
122 unsigned int generation_error:1; /* also sets header_error */
124 /* The following is for the data used to check parity */
125 /* It is for the data with checksum */
126 unsigned int data_corrected:1;
128 struct btrfs_work work;
131 /* Used for the chunks with parity stripe such RAID5/6 */
132 struct scrub_parity {
133 struct scrub_ctx *sctx;
135 struct btrfs_device *scrub_dev;
147 struct list_head spages;
149 /* Work of parity check and repair */
150 struct btrfs_work work;
152 /* Mark the parity blocks which have data */
153 unsigned long *dbitmap;
156 * Mark the parity blocks which have data, but errors happen when
157 * read data or check data
159 unsigned long *ebitmap;
161 unsigned long bitmap[0];
164 struct scrub_wr_ctx {
165 struct scrub_bio *wr_curr_bio;
166 struct btrfs_device *tgtdev;
167 int pages_per_wr_bio; /* <= SCRUB_PAGES_PER_WR_BIO */
168 atomic_t flush_all_writes;
169 struct mutex wr_lock;
173 struct scrub_bio *bios[SCRUB_BIOS_PER_SCTX];
174 struct btrfs_root *dev_root;
177 atomic_t bios_in_flight;
178 atomic_t workers_pending;
179 spinlock_t list_lock;
180 wait_queue_head_t list_wait;
182 struct list_head csum_list;
185 int pages_per_rd_bio;
190 struct scrub_wr_ctx wr_ctx;
195 struct btrfs_scrub_progress stat;
196 spinlock_t stat_lock;
199 * Use a ref counter to avoid use-after-free issues. Scrub workers
200 * decrement bios_in_flight and workers_pending and then do a wakeup
201 * on the list_wait wait queue. We must ensure the main scrub task
202 * doesn't free the scrub context before or while the workers are
203 * doing the wakeup() call.
208 struct scrub_fixup_nodatasum {
209 struct scrub_ctx *sctx;
210 struct btrfs_device *dev;
212 struct btrfs_root *root;
213 struct btrfs_work work;
217 struct scrub_nocow_inode {
221 struct list_head list;
224 struct scrub_copy_nocow_ctx {
225 struct scrub_ctx *sctx;
229 u64 physical_for_dev_replace;
230 struct list_head inodes;
231 struct btrfs_work work;
234 struct scrub_warning {
235 struct btrfs_path *path;
236 u64 extent_item_size;
240 struct btrfs_device *dev;
243 static void scrub_pending_bio_inc(struct scrub_ctx *sctx);
244 static void scrub_pending_bio_dec(struct scrub_ctx *sctx);
245 static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx);
246 static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx);
247 static int scrub_handle_errored_block(struct scrub_block *sblock_to_check);
248 static int scrub_setup_recheck_block(struct scrub_block *original_sblock,
249 struct scrub_block *sblocks_for_recheck);
250 static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
251 struct scrub_block *sblock, int is_metadata,
252 int have_csum, u8 *csum, u64 generation,
253 u16 csum_size, int retry_failed_mirror);
254 static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
255 struct scrub_block *sblock,
256 int is_metadata, int have_csum,
257 const u8 *csum, u64 generation,
259 static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
260 struct scrub_block *sblock_good);
261 static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
262 struct scrub_block *sblock_good,
263 int page_num, int force_write);
264 static void scrub_write_block_to_dev_replace(struct scrub_block *sblock);
265 static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
267 static int scrub_checksum_data(struct scrub_block *sblock);
268 static int scrub_checksum_tree_block(struct scrub_block *sblock);
269 static int scrub_checksum_super(struct scrub_block *sblock);
270 static void scrub_block_get(struct scrub_block *sblock);
271 static void scrub_block_put(struct scrub_block *sblock);
272 static void scrub_page_get(struct scrub_page *spage);
273 static void scrub_page_put(struct scrub_page *spage);
274 static void scrub_parity_get(struct scrub_parity *sparity);
275 static void scrub_parity_put(struct scrub_parity *sparity);
276 static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
277 struct scrub_page *spage);
278 static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
279 u64 physical, struct btrfs_device *dev, u64 flags,
280 u64 gen, int mirror_num, u8 *csum, int force,
281 u64 physical_for_dev_replace);
282 static void scrub_bio_end_io(struct bio *bio);
283 static void scrub_bio_end_io_worker(struct btrfs_work *work);
284 static void scrub_block_complete(struct scrub_block *sblock);
285 static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
286 u64 extent_logical, u64 extent_len,
287 u64 *extent_physical,
288 struct btrfs_device **extent_dev,
289 int *extent_mirror_num);
290 static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
291 struct scrub_wr_ctx *wr_ctx,
292 struct btrfs_fs_info *fs_info,
293 struct btrfs_device *dev,
295 static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx);
296 static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
297 struct scrub_page *spage);
298 static void scrub_wr_submit(struct scrub_ctx *sctx);
299 static void scrub_wr_bio_end_io(struct bio *bio);
300 static void scrub_wr_bio_end_io_worker(struct btrfs_work *work);
301 static int write_page_nocow(struct scrub_ctx *sctx,
302 u64 physical_for_dev_replace, struct page *page);
303 static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
304 struct scrub_copy_nocow_ctx *ctx);
305 static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
306 int mirror_num, u64 physical_for_dev_replace);
307 static void copy_nocow_pages_worker(struct btrfs_work *work);
308 static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
309 static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
310 static void scrub_put_ctx(struct scrub_ctx *sctx);
313 static void scrub_pending_bio_inc(struct scrub_ctx *sctx)
315 atomic_inc(&sctx->refs);
316 atomic_inc(&sctx->bios_in_flight);
319 static void scrub_pending_bio_dec(struct scrub_ctx *sctx)
321 atomic_dec(&sctx->bios_in_flight);
322 wake_up(&sctx->list_wait);
326 static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
328 while (atomic_read(&fs_info->scrub_pause_req)) {
329 mutex_unlock(&fs_info->scrub_lock);
330 wait_event(fs_info->scrub_pause_wait,
331 atomic_read(&fs_info->scrub_pause_req) == 0);
332 mutex_lock(&fs_info->scrub_lock);
336 static void scrub_pause_on(struct btrfs_fs_info *fs_info)
338 atomic_inc(&fs_info->scrubs_paused);
339 wake_up(&fs_info->scrub_pause_wait);
342 static void scrub_pause_off(struct btrfs_fs_info *fs_info)
344 mutex_lock(&fs_info->scrub_lock);
345 __scrub_blocked_if_needed(fs_info);
346 atomic_dec(&fs_info->scrubs_paused);
347 mutex_unlock(&fs_info->scrub_lock);
349 wake_up(&fs_info->scrub_pause_wait);
352 static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
354 scrub_pause_on(fs_info);
355 scrub_pause_off(fs_info);
359 * used for workers that require transaction commits (i.e., for the
362 static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx)
364 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
366 atomic_inc(&sctx->refs);
368 * increment scrubs_running to prevent cancel requests from
369 * completing as long as a worker is running. we must also
370 * increment scrubs_paused to prevent deadlocking on pause
371 * requests used for transactions commits (as the worker uses a
372 * transaction context). it is safe to regard the worker
373 * as paused for all matters practical. effectively, we only
374 * avoid cancellation requests from completing.
376 mutex_lock(&fs_info->scrub_lock);
377 atomic_inc(&fs_info->scrubs_running);
378 atomic_inc(&fs_info->scrubs_paused);
379 mutex_unlock(&fs_info->scrub_lock);
382 * check if @scrubs_running=@scrubs_paused condition
383 * inside wait_event() is not an atomic operation.
384 * which means we may inc/dec @scrub_running/paused
385 * at any time. Let's wake up @scrub_pause_wait as
386 * much as we can to let commit transaction blocked less.
388 wake_up(&fs_info->scrub_pause_wait);
390 atomic_inc(&sctx->workers_pending);
393 /* used for workers that require transaction commits */
394 static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx)
396 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
399 * see scrub_pending_trans_workers_inc() why we're pretending
400 * to be paused in the scrub counters
402 mutex_lock(&fs_info->scrub_lock);
403 atomic_dec(&fs_info->scrubs_running);
404 atomic_dec(&fs_info->scrubs_paused);
405 mutex_unlock(&fs_info->scrub_lock);
406 atomic_dec(&sctx->workers_pending);
407 wake_up(&fs_info->scrub_pause_wait);
408 wake_up(&sctx->list_wait);
412 static void scrub_free_csums(struct scrub_ctx *sctx)
414 while (!list_empty(&sctx->csum_list)) {
415 struct btrfs_ordered_sum *sum;
416 sum = list_first_entry(&sctx->csum_list,
417 struct btrfs_ordered_sum, list);
418 list_del(&sum->list);
423 static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx)
430 scrub_free_wr_ctx(&sctx->wr_ctx);
432 /* this can happen when scrub is cancelled */
433 if (sctx->curr != -1) {
434 struct scrub_bio *sbio = sctx->bios[sctx->curr];
436 for (i = 0; i < sbio->page_count; i++) {
437 WARN_ON(!sbio->pagev[i]->page);
438 scrub_block_put(sbio->pagev[i]->sblock);
443 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
444 struct scrub_bio *sbio = sctx->bios[i];
451 scrub_free_csums(sctx);
455 static void scrub_put_ctx(struct scrub_ctx *sctx)
457 if (atomic_dec_and_test(&sctx->refs))
458 scrub_free_ctx(sctx);
461 static noinline_for_stack
462 struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace)
464 struct scrub_ctx *sctx;
466 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
469 sctx = kzalloc(sizeof(*sctx), GFP_NOFS);
472 atomic_set(&sctx->refs, 1);
473 sctx->is_dev_replace = is_dev_replace;
474 sctx->pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO;
476 sctx->dev_root = dev->dev_root;
477 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
478 struct scrub_bio *sbio;
480 sbio = kzalloc(sizeof(*sbio), GFP_NOFS);
483 sctx->bios[i] = sbio;
487 sbio->page_count = 0;
488 btrfs_init_work(&sbio->work, btrfs_scrub_helper,
489 scrub_bio_end_io_worker, NULL, NULL);
491 if (i != SCRUB_BIOS_PER_SCTX - 1)
492 sctx->bios[i]->next_free = i + 1;
494 sctx->bios[i]->next_free = -1;
496 sctx->first_free = 0;
497 sctx->nodesize = dev->dev_root->nodesize;
498 sctx->sectorsize = dev->dev_root->sectorsize;
499 atomic_set(&sctx->bios_in_flight, 0);
500 atomic_set(&sctx->workers_pending, 0);
501 atomic_set(&sctx->cancel_req, 0);
502 sctx->csum_size = btrfs_super_csum_size(fs_info->super_copy);
503 INIT_LIST_HEAD(&sctx->csum_list);
505 spin_lock_init(&sctx->list_lock);
506 spin_lock_init(&sctx->stat_lock);
507 init_waitqueue_head(&sctx->list_wait);
509 ret = scrub_setup_wr_ctx(sctx, &sctx->wr_ctx, fs_info,
510 fs_info->dev_replace.tgtdev, is_dev_replace);
512 scrub_free_ctx(sctx);
518 scrub_free_ctx(sctx);
519 return ERR_PTR(-ENOMEM);
522 static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root,
529 struct extent_buffer *eb;
530 struct btrfs_inode_item *inode_item;
531 struct scrub_warning *swarn = warn_ctx;
532 struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info;
533 struct inode_fs_paths *ipath = NULL;
534 struct btrfs_root *local_root;
535 struct btrfs_key root_key;
536 struct btrfs_key key;
538 root_key.objectid = root;
539 root_key.type = BTRFS_ROOT_ITEM_KEY;
540 root_key.offset = (u64)-1;
541 local_root = btrfs_read_fs_root_no_name(fs_info, &root_key);
542 if (IS_ERR(local_root)) {
543 ret = PTR_ERR(local_root);
548 * this makes the path point to (inum INODE_ITEM ioff)
551 key.type = BTRFS_INODE_ITEM_KEY;
554 ret = btrfs_search_slot(NULL, local_root, &key, swarn->path, 0, 0);
556 btrfs_release_path(swarn->path);
560 eb = swarn->path->nodes[0];
561 inode_item = btrfs_item_ptr(eb, swarn->path->slots[0],
562 struct btrfs_inode_item);
563 isize = btrfs_inode_size(eb, inode_item);
564 nlink = btrfs_inode_nlink(eb, inode_item);
565 btrfs_release_path(swarn->path);
567 ipath = init_ipath(4096, local_root, swarn->path);
569 ret = PTR_ERR(ipath);
573 ret = paths_from_inode(inum, ipath);
579 * we deliberately ignore the bit ipath might have been too small to
580 * hold all of the paths here
582 for (i = 0; i < ipath->fspath->elem_cnt; ++i)
583 btrfs_warn_in_rcu(fs_info, "%s at logical %llu on dev "
584 "%s, sector %llu, root %llu, inode %llu, offset %llu, "
585 "length %llu, links %u (path: %s)", swarn->errstr,
586 swarn->logical, rcu_str_deref(swarn->dev->name),
587 (unsigned long long)swarn->sector, root, inum, offset,
588 min(isize - offset, (u64)PAGE_SIZE), nlink,
589 (char *)(unsigned long)ipath->fspath->val[i]);
595 btrfs_warn_in_rcu(fs_info, "%s at logical %llu on dev "
596 "%s, sector %llu, root %llu, inode %llu, offset %llu: path "
597 "resolving failed with ret=%d", swarn->errstr,
598 swarn->logical, rcu_str_deref(swarn->dev->name),
599 (unsigned long long)swarn->sector, root, inum, offset, ret);
605 static void scrub_print_warning(const char *errstr, struct scrub_block *sblock)
607 struct btrfs_device *dev;
608 struct btrfs_fs_info *fs_info;
609 struct btrfs_path *path;
610 struct btrfs_key found_key;
611 struct extent_buffer *eb;
612 struct btrfs_extent_item *ei;
613 struct scrub_warning swarn;
614 unsigned long ptr = 0;
622 WARN_ON(sblock->page_count < 1);
623 dev = sblock->pagev[0]->dev;
624 fs_info = sblock->sctx->dev_root->fs_info;
626 path = btrfs_alloc_path();
630 swarn.sector = (sblock->pagev[0]->physical) >> 9;
631 swarn.logical = sblock->pagev[0]->logical;
632 swarn.errstr = errstr;
635 ret = extent_from_logical(fs_info, swarn.logical, path, &found_key,
640 extent_item_pos = swarn.logical - found_key.objectid;
641 swarn.extent_item_size = found_key.offset;
644 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
645 item_size = btrfs_item_size_nr(eb, path->slots[0]);
647 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
649 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
650 item_size, &ref_root,
652 btrfs_warn_in_rcu(fs_info,
653 "%s at logical %llu on dev %s, "
654 "sector %llu: metadata %s (level %d) in tree "
655 "%llu", errstr, swarn.logical,
656 rcu_str_deref(dev->name),
657 (unsigned long long)swarn.sector,
658 ref_level ? "node" : "leaf",
659 ret < 0 ? -1 : ref_level,
660 ret < 0 ? -1 : ref_root);
662 btrfs_release_path(path);
664 btrfs_release_path(path);
667 iterate_extent_inodes(fs_info, found_key.objectid,
669 scrub_print_warning_inode, &swarn);
673 btrfs_free_path(path);
676 static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx)
678 struct page *page = NULL;
680 struct scrub_fixup_nodatasum *fixup = fixup_ctx;
683 struct btrfs_key key;
684 struct inode *inode = NULL;
685 struct btrfs_fs_info *fs_info;
686 u64 end = offset + PAGE_SIZE - 1;
687 struct btrfs_root *local_root;
691 key.type = BTRFS_ROOT_ITEM_KEY;
692 key.offset = (u64)-1;
694 fs_info = fixup->root->fs_info;
695 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
697 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
698 if (IS_ERR(local_root)) {
699 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
700 return PTR_ERR(local_root);
703 key.type = BTRFS_INODE_ITEM_KEY;
706 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
707 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
709 return PTR_ERR(inode);
711 index = offset >> PAGE_CACHE_SHIFT;
713 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
719 if (PageUptodate(page)) {
720 if (PageDirty(page)) {
722 * we need to write the data to the defect sector. the
723 * data that was in that sector is not in memory,
724 * because the page was modified. we must not write the
725 * modified page to that sector.
727 * TODO: what could be done here: wait for the delalloc
728 * runner to write out that page (might involve
729 * COW) and see whether the sector is still
730 * referenced afterwards.
732 * For the meantime, we'll treat this error
733 * incorrectable, although there is a chance that a
734 * later scrub will find the bad sector again and that
735 * there's no dirty page in memory, then.
740 ret = repair_io_failure(inode, offset, PAGE_SIZE,
741 fixup->logical, page,
742 offset - page_offset(page),
748 * we need to get good data first. the general readpage path
749 * will call repair_io_failure for us, we just have to make
750 * sure we read the bad mirror.
752 ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
753 EXTENT_DAMAGED, GFP_NOFS);
755 /* set_extent_bits should give proper error */
762 ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page,
765 wait_on_page_locked(page);
767 corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset,
768 end, EXTENT_DAMAGED, 0, NULL);
770 clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
771 EXTENT_DAMAGED, GFP_NOFS);
783 if (ret == 0 && corrected) {
785 * we only need to call readpage for one of the inodes belonging
786 * to this extent. so make iterate_extent_inodes stop
794 static void scrub_fixup_nodatasum(struct btrfs_work *work)
797 struct scrub_fixup_nodatasum *fixup;
798 struct scrub_ctx *sctx;
799 struct btrfs_trans_handle *trans = NULL;
800 struct btrfs_path *path;
801 int uncorrectable = 0;
803 fixup = container_of(work, struct scrub_fixup_nodatasum, work);
806 path = btrfs_alloc_path();
808 spin_lock(&sctx->stat_lock);
809 ++sctx->stat.malloc_errors;
810 spin_unlock(&sctx->stat_lock);
815 trans = btrfs_join_transaction(fixup->root);
822 * the idea is to trigger a regular read through the standard path. we
823 * read a page from the (failed) logical address by specifying the
824 * corresponding copynum of the failed sector. thus, that readpage is
826 * that is the point where on-the-fly error correction will kick in
827 * (once it's finished) and rewrite the failed sector if a good copy
830 ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info,
831 path, scrub_fixup_readpage,
839 spin_lock(&sctx->stat_lock);
840 ++sctx->stat.corrected_errors;
841 spin_unlock(&sctx->stat_lock);
844 if (trans && !IS_ERR(trans))
845 btrfs_end_transaction(trans, fixup->root);
847 spin_lock(&sctx->stat_lock);
848 ++sctx->stat.uncorrectable_errors;
849 spin_unlock(&sctx->stat_lock);
850 btrfs_dev_replace_stats_inc(
851 &sctx->dev_root->fs_info->dev_replace.
852 num_uncorrectable_read_errors);
853 btrfs_err_rl_in_rcu(sctx->dev_root->fs_info,
854 "unable to fixup (nodatasum) error at logical %llu on dev %s",
855 fixup->logical, rcu_str_deref(fixup->dev->name));
858 btrfs_free_path(path);
861 scrub_pending_trans_workers_dec(sctx);
864 static inline void scrub_get_recover(struct scrub_recover *recover)
866 atomic_inc(&recover->refs);
869 static inline void scrub_put_recover(struct scrub_recover *recover)
871 if (atomic_dec_and_test(&recover->refs)) {
872 btrfs_put_bbio(recover->bbio);
878 * scrub_handle_errored_block gets called when either verification of the
879 * pages failed or the bio failed to read, e.g. with EIO. In the latter
880 * case, this function handles all pages in the bio, even though only one
882 * The goal of this function is to repair the errored block by using the
883 * contents of one of the mirrors.
885 static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
887 struct scrub_ctx *sctx = sblock_to_check->sctx;
888 struct btrfs_device *dev;
889 struct btrfs_fs_info *fs_info;
893 unsigned int failed_mirror_index;
894 unsigned int is_metadata;
895 unsigned int have_csum;
897 struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */
898 struct scrub_block *sblock_bad;
903 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
904 DEFAULT_RATELIMIT_BURST);
906 BUG_ON(sblock_to_check->page_count < 1);
907 fs_info = sctx->dev_root->fs_info;
908 if (sblock_to_check->pagev[0]->flags & BTRFS_EXTENT_FLAG_SUPER) {
910 * if we find an error in a super block, we just report it.
911 * They will get written with the next transaction commit
914 spin_lock(&sctx->stat_lock);
915 ++sctx->stat.super_errors;
916 spin_unlock(&sctx->stat_lock);
919 length = sblock_to_check->page_count * PAGE_SIZE;
920 logical = sblock_to_check->pagev[0]->logical;
921 generation = sblock_to_check->pagev[0]->generation;
922 BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1);
923 failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1;
924 is_metadata = !(sblock_to_check->pagev[0]->flags &
925 BTRFS_EXTENT_FLAG_DATA);
926 have_csum = sblock_to_check->pagev[0]->have_csum;
927 csum = sblock_to_check->pagev[0]->csum;
928 dev = sblock_to_check->pagev[0]->dev;
930 if (sctx->is_dev_replace && !is_metadata && !have_csum) {
931 sblocks_for_recheck = NULL;
936 * read all mirrors one after the other. This includes to
937 * re-read the extent or metadata block that failed (that was
938 * the cause that this fixup code is called) another time,
939 * page by page this time in order to know which pages
940 * caused I/O errors and which ones are good (for all mirrors).
941 * It is the goal to handle the situation when more than one
942 * mirror contains I/O errors, but the errors do not
943 * overlap, i.e. the data can be repaired by selecting the
944 * pages from those mirrors without I/O error on the
945 * particular pages. One example (with blocks >= 2 * PAGE_SIZE)
946 * would be that mirror #1 has an I/O error on the first page,
947 * the second page is good, and mirror #2 has an I/O error on
948 * the second page, but the first page is good.
949 * Then the first page of the first mirror can be repaired by
950 * taking the first page of the second mirror, and the
951 * second page of the second mirror can be repaired by
952 * copying the contents of the 2nd page of the 1st mirror.
953 * One more note: if the pages of one mirror contain I/O
954 * errors, the checksum cannot be verified. In order to get
955 * the best data for repairing, the first attempt is to find
956 * a mirror without I/O errors and with a validated checksum.
957 * Only if this is not possible, the pages are picked from
958 * mirrors with I/O errors without considering the checksum.
959 * If the latter is the case, at the end, the checksum of the
960 * repaired area is verified in order to correctly maintain
964 sblocks_for_recheck = kcalloc(BTRFS_MAX_MIRRORS,
965 sizeof(*sblocks_for_recheck), GFP_NOFS);
966 if (!sblocks_for_recheck) {
967 spin_lock(&sctx->stat_lock);
968 sctx->stat.malloc_errors++;
969 sctx->stat.read_errors++;
970 sctx->stat.uncorrectable_errors++;
971 spin_unlock(&sctx->stat_lock);
972 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
976 /* setup the context, map the logical blocks and alloc the pages */
977 ret = scrub_setup_recheck_block(sblock_to_check, sblocks_for_recheck);
979 spin_lock(&sctx->stat_lock);
980 sctx->stat.read_errors++;
981 sctx->stat.uncorrectable_errors++;
982 spin_unlock(&sctx->stat_lock);
983 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
986 BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS);
987 sblock_bad = sblocks_for_recheck + failed_mirror_index;
989 /* build and submit the bios for the failed mirror, check checksums */
990 scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum,
991 csum, generation, sctx->csum_size, 1);
993 if (!sblock_bad->header_error && !sblock_bad->checksum_error &&
994 sblock_bad->no_io_error_seen) {
996 * the error disappeared after reading page by page, or
997 * the area was part of a huge bio and other parts of the
998 * bio caused I/O errors, or the block layer merged several
999 * read requests into one and the error is caused by a
1000 * different bio (usually one of the two latter cases is
1003 spin_lock(&sctx->stat_lock);
1004 sctx->stat.unverified_errors++;
1005 sblock_to_check->data_corrected = 1;
1006 spin_unlock(&sctx->stat_lock);
1008 if (sctx->is_dev_replace)
1009 scrub_write_block_to_dev_replace(sblock_bad);
1013 if (!sblock_bad->no_io_error_seen) {
1014 spin_lock(&sctx->stat_lock);
1015 sctx->stat.read_errors++;
1016 spin_unlock(&sctx->stat_lock);
1017 if (__ratelimit(&_rs))
1018 scrub_print_warning("i/o error", sblock_to_check);
1019 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
1020 } else if (sblock_bad->checksum_error) {
1021 spin_lock(&sctx->stat_lock);
1022 sctx->stat.csum_errors++;
1023 spin_unlock(&sctx->stat_lock);
1024 if (__ratelimit(&_rs))
1025 scrub_print_warning("checksum error", sblock_to_check);
1026 btrfs_dev_stat_inc_and_print(dev,
1027 BTRFS_DEV_STAT_CORRUPTION_ERRS);
1028 } else if (sblock_bad->header_error) {
1029 spin_lock(&sctx->stat_lock);
1030 sctx->stat.verify_errors++;
1031 spin_unlock(&sctx->stat_lock);
1032 if (__ratelimit(&_rs))
1033 scrub_print_warning("checksum/header error",
1035 if (sblock_bad->generation_error)
1036 btrfs_dev_stat_inc_and_print(dev,
1037 BTRFS_DEV_STAT_GENERATION_ERRS);
1039 btrfs_dev_stat_inc_and_print(dev,
1040 BTRFS_DEV_STAT_CORRUPTION_ERRS);
1043 if (sctx->readonly) {
1044 ASSERT(!sctx->is_dev_replace);
1048 if (!is_metadata && !have_csum) {
1049 struct scrub_fixup_nodatasum *fixup_nodatasum;
1051 WARN_ON(sctx->is_dev_replace);
1056 * !is_metadata and !have_csum, this means that the data
1057 * might not be COW'ed, that it might be modified
1058 * concurrently. The general strategy to work on the
1059 * commit root does not help in the case when COW is not
1062 fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS);
1063 if (!fixup_nodatasum)
1064 goto did_not_correct_error;
1065 fixup_nodatasum->sctx = sctx;
1066 fixup_nodatasum->dev = dev;
1067 fixup_nodatasum->logical = logical;
1068 fixup_nodatasum->root = fs_info->extent_root;
1069 fixup_nodatasum->mirror_num = failed_mirror_index + 1;
1070 scrub_pending_trans_workers_inc(sctx);
1071 btrfs_init_work(&fixup_nodatasum->work, btrfs_scrub_helper,
1072 scrub_fixup_nodatasum, NULL, NULL);
1073 btrfs_queue_work(fs_info->scrub_workers,
1074 &fixup_nodatasum->work);
1079 * now build and submit the bios for the other mirrors, check
1081 * First try to pick the mirror which is completely without I/O
1082 * errors and also does not have a checksum error.
1083 * If one is found, and if a checksum is present, the full block
1084 * that is known to contain an error is rewritten. Afterwards
1085 * the block is known to be corrected.
1086 * If a mirror is found which is completely correct, and no
1087 * checksum is present, only those pages are rewritten that had
1088 * an I/O error in the block to be repaired, since it cannot be
1089 * determined, which copy of the other pages is better (and it
1090 * could happen otherwise that a correct page would be
1091 * overwritten by a bad one).
1093 for (mirror_index = 0;
1094 mirror_index < BTRFS_MAX_MIRRORS &&
1095 sblocks_for_recheck[mirror_index].page_count > 0;
1097 struct scrub_block *sblock_other;
1099 if (mirror_index == failed_mirror_index)
1101 sblock_other = sblocks_for_recheck + mirror_index;
1103 /* build and submit the bios, check checksums */
1104 scrub_recheck_block(fs_info, sblock_other, is_metadata,
1105 have_csum, csum, generation,
1106 sctx->csum_size, 0);
1108 if (!sblock_other->header_error &&
1109 !sblock_other->checksum_error &&
1110 sblock_other->no_io_error_seen) {
1111 if (sctx->is_dev_replace) {
1112 scrub_write_block_to_dev_replace(sblock_other);
1113 goto corrected_error;
1115 ret = scrub_repair_block_from_good_copy(
1116 sblock_bad, sblock_other);
1118 goto corrected_error;
1123 if (sblock_bad->no_io_error_seen && !sctx->is_dev_replace)
1124 goto did_not_correct_error;
1127 * In case of I/O errors in the area that is supposed to be
1128 * repaired, continue by picking good copies of those pages.
1129 * Select the good pages from mirrors to rewrite bad pages from
1130 * the area to fix. Afterwards verify the checksum of the block
1131 * that is supposed to be repaired. This verification step is
1132 * only done for the purpose of statistic counting and for the
1133 * final scrub report, whether errors remain.
1134 * A perfect algorithm could make use of the checksum and try
1135 * all possible combinations of pages from the different mirrors
1136 * until the checksum verification succeeds. For example, when
1137 * the 2nd page of mirror #1 faces I/O errors, and the 2nd page
1138 * of mirror #2 is readable but the final checksum test fails,
1139 * then the 2nd page of mirror #3 could be tried, whether now
1140 * the final checksum succeedes. But this would be a rare
1141 * exception and is therefore not implemented. At least it is
1142 * avoided that the good copy is overwritten.
1143 * A more useful improvement would be to pick the sectors
1144 * without I/O error based on sector sizes (512 bytes on legacy
1145 * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one
1146 * mirror could be repaired by taking 512 byte of a different
1147 * mirror, even if other 512 byte sectors in the same PAGE_SIZE
1148 * area are unreadable.
1151 for (page_num = 0; page_num < sblock_bad->page_count;
1153 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1154 struct scrub_block *sblock_other = NULL;
1156 /* skip no-io-error page in scrub */
1157 if (!page_bad->io_error && !sctx->is_dev_replace)
1160 /* try to find no-io-error page in mirrors */
1161 if (page_bad->io_error) {
1162 for (mirror_index = 0;
1163 mirror_index < BTRFS_MAX_MIRRORS &&
1164 sblocks_for_recheck[mirror_index].page_count > 0;
1166 if (!sblocks_for_recheck[mirror_index].
1167 pagev[page_num]->io_error) {
1168 sblock_other = sblocks_for_recheck +
1177 if (sctx->is_dev_replace) {
1179 * did not find a mirror to fetch the page
1180 * from. scrub_write_page_to_dev_replace()
1181 * handles this case (page->io_error), by
1182 * filling the block with zeros before
1183 * submitting the write request
1186 sblock_other = sblock_bad;
1188 if (scrub_write_page_to_dev_replace(sblock_other,
1190 btrfs_dev_replace_stats_inc(
1192 fs_info->dev_replace.
1196 } else if (sblock_other) {
1197 ret = scrub_repair_page_from_good_copy(sblock_bad,
1201 page_bad->io_error = 0;
1207 if (success && !sctx->is_dev_replace) {
1208 if (is_metadata || have_csum) {
1210 * need to verify the checksum now that all
1211 * sectors on disk are repaired (the write
1212 * request for data to be repaired is on its way).
1213 * Just be lazy and use scrub_recheck_block()
1214 * which re-reads the data before the checksum
1215 * is verified, but most likely the data comes out
1216 * of the page cache.
1218 scrub_recheck_block(fs_info, sblock_bad,
1219 is_metadata, have_csum, csum,
1220 generation, sctx->csum_size, 1);
1221 if (!sblock_bad->header_error &&
1222 !sblock_bad->checksum_error &&
1223 sblock_bad->no_io_error_seen)
1224 goto corrected_error;
1226 goto did_not_correct_error;
1229 spin_lock(&sctx->stat_lock);
1230 sctx->stat.corrected_errors++;
1231 sblock_to_check->data_corrected = 1;
1232 spin_unlock(&sctx->stat_lock);
1233 btrfs_err_rl_in_rcu(fs_info,
1234 "fixed up error at logical %llu on dev %s",
1235 logical, rcu_str_deref(dev->name));
1238 did_not_correct_error:
1239 spin_lock(&sctx->stat_lock);
1240 sctx->stat.uncorrectable_errors++;
1241 spin_unlock(&sctx->stat_lock);
1242 btrfs_err_rl_in_rcu(fs_info,
1243 "unable to fixup (regular) error at logical %llu on dev %s",
1244 logical, rcu_str_deref(dev->name));
1248 if (sblocks_for_recheck) {
1249 for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS;
1251 struct scrub_block *sblock = sblocks_for_recheck +
1253 struct scrub_recover *recover;
1256 for (page_index = 0; page_index < sblock->page_count;
1258 sblock->pagev[page_index]->sblock = NULL;
1259 recover = sblock->pagev[page_index]->recover;
1261 scrub_put_recover(recover);
1262 sblock->pagev[page_index]->recover =
1265 scrub_page_put(sblock->pagev[page_index]);
1268 kfree(sblocks_for_recheck);
1274 static inline int scrub_nr_raid_mirrors(struct btrfs_bio *bbio)
1276 if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID5)
1278 else if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID6)
1281 return (int)bbio->num_stripes;
1284 static inline void scrub_stripe_index_and_offset(u64 logical, u64 map_type,
1287 int nstripes, int mirror,
1293 if (map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
1295 for (i = 0; i < nstripes; i++) {
1296 if (raid_map[i] == RAID6_Q_STRIPE ||
1297 raid_map[i] == RAID5_P_STRIPE)
1300 if (logical >= raid_map[i] &&
1301 logical < raid_map[i] + mapped_length)
1306 *stripe_offset = logical - raid_map[i];
1308 /* The other RAID type */
1309 *stripe_index = mirror;
1314 static int scrub_setup_recheck_block(struct scrub_block *original_sblock,
1315 struct scrub_block *sblocks_for_recheck)
1317 struct scrub_ctx *sctx = original_sblock->sctx;
1318 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
1319 u64 length = original_sblock->page_count * PAGE_SIZE;
1320 u64 logical = original_sblock->pagev[0]->logical;
1321 u64 generation = original_sblock->pagev[0]->generation;
1322 u64 flags = original_sblock->pagev[0]->flags;
1323 u64 have_csum = original_sblock->pagev[0]->have_csum;
1324 struct scrub_recover *recover;
1325 struct btrfs_bio *bbio;
1336 * note: the two members refs and outstanding_pages
1337 * are not used (and not set) in the blocks that are used for
1338 * the recheck procedure
1341 while (length > 0) {
1342 sublen = min_t(u64, length, PAGE_SIZE);
1343 mapped_length = sublen;
1347 * with a length of PAGE_SIZE, each returned stripe
1348 * represents one mirror
1350 ret = btrfs_map_sblock(fs_info, REQ_GET_READ_MIRRORS, logical,
1351 &mapped_length, &bbio, 0, 1);
1352 if (ret || !bbio || mapped_length < sublen) {
1353 btrfs_put_bbio(bbio);
1357 recover = kzalloc(sizeof(struct scrub_recover), GFP_NOFS);
1359 btrfs_put_bbio(bbio);
1363 atomic_set(&recover->refs, 1);
1364 recover->bbio = bbio;
1365 recover->map_length = mapped_length;
1367 BUG_ON(page_index >= SCRUB_PAGES_PER_RD_BIO);
1369 nmirrors = min(scrub_nr_raid_mirrors(bbio), BTRFS_MAX_MIRRORS);
1371 for (mirror_index = 0; mirror_index < nmirrors;
1373 struct scrub_block *sblock;
1374 struct scrub_page *page;
1376 sblock = sblocks_for_recheck + mirror_index;
1377 sblock->sctx = sctx;
1379 page = kzalloc(sizeof(*page), GFP_NOFS);
1382 spin_lock(&sctx->stat_lock);
1383 sctx->stat.malloc_errors++;
1384 spin_unlock(&sctx->stat_lock);
1385 scrub_put_recover(recover);
1388 scrub_page_get(page);
1389 sblock->pagev[page_index] = page;
1390 page->sblock = sblock;
1391 page->flags = flags;
1392 page->generation = generation;
1393 page->logical = logical;
1394 page->have_csum = have_csum;
1397 original_sblock->pagev[0]->csum,
1400 scrub_stripe_index_and_offset(logical,
1409 page->physical = bbio->stripes[stripe_index].physical +
1411 page->dev = bbio->stripes[stripe_index].dev;
1413 BUG_ON(page_index >= original_sblock->page_count);
1414 page->physical_for_dev_replace =
1415 original_sblock->pagev[page_index]->
1416 physical_for_dev_replace;
1417 /* for missing devices, dev->bdev is NULL */
1418 page->mirror_num = mirror_index + 1;
1419 sblock->page_count++;
1420 page->page = alloc_page(GFP_NOFS);
1424 scrub_get_recover(recover);
1425 page->recover = recover;
1427 scrub_put_recover(recover);
1436 struct scrub_bio_ret {
1437 struct completion event;
1441 static void scrub_bio_wait_endio(struct bio *bio)
1443 struct scrub_bio_ret *ret = bio->bi_private;
1445 ret->error = bio->bi_error;
1446 complete(&ret->event);
1449 static inline int scrub_is_page_on_raid56(struct scrub_page *page)
1451 return page->recover &&
1452 (page->recover->bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK);
1455 static int scrub_submit_raid56_bio_wait(struct btrfs_fs_info *fs_info,
1457 struct scrub_page *page)
1459 struct scrub_bio_ret done;
1462 init_completion(&done.event);
1464 bio->bi_iter.bi_sector = page->logical >> 9;
1465 bio->bi_private = &done;
1466 bio->bi_end_io = scrub_bio_wait_endio;
1468 ret = raid56_parity_recover(fs_info->fs_root, bio, page->recover->bbio,
1469 page->recover->map_length,
1470 page->mirror_num, 0);
1474 wait_for_completion(&done.event);
1482 * this function will check the on disk data for checksum errors, header
1483 * errors and read I/O errors. If any I/O errors happen, the exact pages
1484 * which are errored are marked as being bad. The goal is to enable scrub
1485 * to take those pages that are not errored from all the mirrors so that
1486 * the pages that are errored in the just handled mirror can be repaired.
1488 static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
1489 struct scrub_block *sblock, int is_metadata,
1490 int have_csum, u8 *csum, u64 generation,
1491 u16 csum_size, int retry_failed_mirror)
1495 sblock->no_io_error_seen = 1;
1496 sblock->header_error = 0;
1497 sblock->checksum_error = 0;
1499 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1501 struct scrub_page *page = sblock->pagev[page_num];
1503 if (page->dev->bdev == NULL) {
1505 sblock->no_io_error_seen = 0;
1509 WARN_ON(!page->page);
1510 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1513 sblock->no_io_error_seen = 0;
1516 bio->bi_bdev = page->dev->bdev;
1518 bio_add_page(bio, page->page, PAGE_SIZE, 0);
1519 if (!retry_failed_mirror && scrub_is_page_on_raid56(page)) {
1520 if (scrub_submit_raid56_bio_wait(fs_info, bio, page))
1521 sblock->no_io_error_seen = 0;
1523 bio->bi_iter.bi_sector = page->physical >> 9;
1525 if (btrfsic_submit_bio_wait(READ, bio))
1526 sblock->no_io_error_seen = 0;
1532 if (sblock->no_io_error_seen)
1533 scrub_recheck_block_checksum(fs_info, sblock, is_metadata,
1534 have_csum, csum, generation,
1540 static inline int scrub_check_fsid(u8 fsid[],
1541 struct scrub_page *spage)
1543 struct btrfs_fs_devices *fs_devices = spage->dev->fs_devices;
1546 ret = memcmp(fsid, fs_devices->fsid, BTRFS_UUID_SIZE);
1550 static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
1551 struct scrub_block *sblock,
1552 int is_metadata, int have_csum,
1553 const u8 *csum, u64 generation,
1557 u8 calculated_csum[BTRFS_CSUM_SIZE];
1559 void *mapped_buffer;
1561 WARN_ON(!sblock->pagev[0]->page);
1563 struct btrfs_header *h;
1565 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
1566 h = (struct btrfs_header *)mapped_buffer;
1568 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h) ||
1569 !scrub_check_fsid(h->fsid, sblock->pagev[0]) ||
1570 memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
1572 sblock->header_error = 1;
1573 } else if (generation != btrfs_stack_header_generation(h)) {
1574 sblock->header_error = 1;
1575 sblock->generation_error = 1;
1582 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
1585 for (page_num = 0;;) {
1586 if (page_num == 0 && is_metadata)
1587 crc = btrfs_csum_data(
1588 ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE,
1589 crc, PAGE_SIZE - BTRFS_CSUM_SIZE);
1591 crc = btrfs_csum_data(mapped_buffer, crc, PAGE_SIZE);
1593 kunmap_atomic(mapped_buffer);
1595 if (page_num >= sblock->page_count)
1597 WARN_ON(!sblock->pagev[page_num]->page);
1599 mapped_buffer = kmap_atomic(sblock->pagev[page_num]->page);
1602 btrfs_csum_final(crc, calculated_csum);
1603 if (memcmp(calculated_csum, csum, csum_size))
1604 sblock->checksum_error = 1;
1607 static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
1608 struct scrub_block *sblock_good)
1613 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
1616 ret_sub = scrub_repair_page_from_good_copy(sblock_bad,
1626 static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
1627 struct scrub_block *sblock_good,
1628 int page_num, int force_write)
1630 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1631 struct scrub_page *page_good = sblock_good->pagev[page_num];
1633 BUG_ON(page_bad->page == NULL);
1634 BUG_ON(page_good->page == NULL);
1635 if (force_write || sblock_bad->header_error ||
1636 sblock_bad->checksum_error || page_bad->io_error) {
1640 if (!page_bad->dev->bdev) {
1641 btrfs_warn_rl(sblock_bad->sctx->dev_root->fs_info,
1642 "scrub_repair_page_from_good_copy(bdev == NULL) "
1647 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1650 bio->bi_bdev = page_bad->dev->bdev;
1651 bio->bi_iter.bi_sector = page_bad->physical >> 9;
1653 ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0);
1654 if (PAGE_SIZE != ret) {
1659 if (btrfsic_submit_bio_wait(WRITE, bio)) {
1660 btrfs_dev_stat_inc_and_print(page_bad->dev,
1661 BTRFS_DEV_STAT_WRITE_ERRS);
1662 btrfs_dev_replace_stats_inc(
1663 &sblock_bad->sctx->dev_root->fs_info->
1664 dev_replace.num_write_errors);
1674 static void scrub_write_block_to_dev_replace(struct scrub_block *sblock)
1679 * This block is used for the check of the parity on the source device,
1680 * so the data needn't be written into the destination device.
1682 if (sblock->sparity)
1685 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1688 ret = scrub_write_page_to_dev_replace(sblock, page_num);
1690 btrfs_dev_replace_stats_inc(
1691 &sblock->sctx->dev_root->fs_info->dev_replace.
1696 static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
1699 struct scrub_page *spage = sblock->pagev[page_num];
1701 BUG_ON(spage->page == NULL);
1702 if (spage->io_error) {
1703 void *mapped_buffer = kmap_atomic(spage->page);
1705 memset(mapped_buffer, 0, PAGE_CACHE_SIZE);
1706 flush_dcache_page(spage->page);
1707 kunmap_atomic(mapped_buffer);
1709 return scrub_add_page_to_wr_bio(sblock->sctx, spage);
1712 static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
1713 struct scrub_page *spage)
1715 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1716 struct scrub_bio *sbio;
1719 mutex_lock(&wr_ctx->wr_lock);
1721 if (!wr_ctx->wr_curr_bio) {
1722 wr_ctx->wr_curr_bio = kzalloc(sizeof(*wr_ctx->wr_curr_bio),
1724 if (!wr_ctx->wr_curr_bio) {
1725 mutex_unlock(&wr_ctx->wr_lock);
1728 wr_ctx->wr_curr_bio->sctx = sctx;
1729 wr_ctx->wr_curr_bio->page_count = 0;
1731 sbio = wr_ctx->wr_curr_bio;
1732 if (sbio->page_count == 0) {
1735 sbio->physical = spage->physical_for_dev_replace;
1736 sbio->logical = spage->logical;
1737 sbio->dev = wr_ctx->tgtdev;
1740 bio = btrfs_io_bio_alloc(GFP_NOFS, wr_ctx->pages_per_wr_bio);
1742 mutex_unlock(&wr_ctx->wr_lock);
1748 bio->bi_private = sbio;
1749 bio->bi_end_io = scrub_wr_bio_end_io;
1750 bio->bi_bdev = sbio->dev->bdev;
1751 bio->bi_iter.bi_sector = sbio->physical >> 9;
1753 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
1754 spage->physical_for_dev_replace ||
1755 sbio->logical + sbio->page_count * PAGE_SIZE !=
1757 scrub_wr_submit(sctx);
1761 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
1762 if (ret != PAGE_SIZE) {
1763 if (sbio->page_count < 1) {
1766 mutex_unlock(&wr_ctx->wr_lock);
1769 scrub_wr_submit(sctx);
1773 sbio->pagev[sbio->page_count] = spage;
1774 scrub_page_get(spage);
1776 if (sbio->page_count == wr_ctx->pages_per_wr_bio)
1777 scrub_wr_submit(sctx);
1778 mutex_unlock(&wr_ctx->wr_lock);
1783 static void scrub_wr_submit(struct scrub_ctx *sctx)
1785 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1786 struct scrub_bio *sbio;
1788 if (!wr_ctx->wr_curr_bio)
1791 sbio = wr_ctx->wr_curr_bio;
1792 wr_ctx->wr_curr_bio = NULL;
1793 WARN_ON(!sbio->bio->bi_bdev);
1794 scrub_pending_bio_inc(sctx);
1795 /* process all writes in a single worker thread. Then the block layer
1796 * orders the requests before sending them to the driver which
1797 * doubled the write performance on spinning disks when measured
1799 btrfsic_submit_bio(WRITE, sbio->bio);
1802 static void scrub_wr_bio_end_io(struct bio *bio)
1804 struct scrub_bio *sbio = bio->bi_private;
1805 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
1807 sbio->err = bio->bi_error;
1810 btrfs_init_work(&sbio->work, btrfs_scrubwrc_helper,
1811 scrub_wr_bio_end_io_worker, NULL, NULL);
1812 btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work);
1815 static void scrub_wr_bio_end_io_worker(struct btrfs_work *work)
1817 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
1818 struct scrub_ctx *sctx = sbio->sctx;
1821 WARN_ON(sbio->page_count > SCRUB_PAGES_PER_WR_BIO);
1823 struct btrfs_dev_replace *dev_replace =
1824 &sbio->sctx->dev_root->fs_info->dev_replace;
1826 for (i = 0; i < sbio->page_count; i++) {
1827 struct scrub_page *spage = sbio->pagev[i];
1829 spage->io_error = 1;
1830 btrfs_dev_replace_stats_inc(&dev_replace->
1835 for (i = 0; i < sbio->page_count; i++)
1836 scrub_page_put(sbio->pagev[i]);
1840 scrub_pending_bio_dec(sctx);
1843 static int scrub_checksum(struct scrub_block *sblock)
1848 WARN_ON(sblock->page_count < 1);
1849 flags = sblock->pagev[0]->flags;
1851 if (flags & BTRFS_EXTENT_FLAG_DATA)
1852 ret = scrub_checksum_data(sblock);
1853 else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1854 ret = scrub_checksum_tree_block(sblock);
1855 else if (flags & BTRFS_EXTENT_FLAG_SUPER)
1856 (void)scrub_checksum_super(sblock);
1860 scrub_handle_errored_block(sblock);
1865 static int scrub_checksum_data(struct scrub_block *sblock)
1867 struct scrub_ctx *sctx = sblock->sctx;
1868 u8 csum[BTRFS_CSUM_SIZE];
1877 BUG_ON(sblock->page_count < 1);
1878 if (!sblock->pagev[0]->have_csum)
1881 on_disk_csum = sblock->pagev[0]->csum;
1882 page = sblock->pagev[0]->page;
1883 buffer = kmap_atomic(page);
1885 len = sctx->sectorsize;
1888 u64 l = min_t(u64, len, PAGE_SIZE);
1890 crc = btrfs_csum_data(buffer, crc, l);
1891 kunmap_atomic(buffer);
1896 BUG_ON(index >= sblock->page_count);
1897 BUG_ON(!sblock->pagev[index]->page);
1898 page = sblock->pagev[index]->page;
1899 buffer = kmap_atomic(page);
1902 btrfs_csum_final(crc, csum);
1903 if (memcmp(csum, on_disk_csum, sctx->csum_size))
1909 static int scrub_checksum_tree_block(struct scrub_block *sblock)
1911 struct scrub_ctx *sctx = sblock->sctx;
1912 struct btrfs_header *h;
1913 struct btrfs_root *root = sctx->dev_root;
1914 struct btrfs_fs_info *fs_info = root->fs_info;
1915 u8 calculated_csum[BTRFS_CSUM_SIZE];
1916 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1918 void *mapped_buffer;
1927 BUG_ON(sblock->page_count < 1);
1928 page = sblock->pagev[0]->page;
1929 mapped_buffer = kmap_atomic(page);
1930 h = (struct btrfs_header *)mapped_buffer;
1931 memcpy(on_disk_csum, h->csum, sctx->csum_size);
1934 * we don't use the getter functions here, as we
1935 * a) don't have an extent buffer and
1936 * b) the page is already kmapped
1939 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h))
1942 if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h))
1945 if (!scrub_check_fsid(h->fsid, sblock->pagev[0]))
1948 if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
1952 len = sctx->nodesize - BTRFS_CSUM_SIZE;
1953 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
1954 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
1957 u64 l = min_t(u64, len, mapped_size);
1959 crc = btrfs_csum_data(p, crc, l);
1960 kunmap_atomic(mapped_buffer);
1965 BUG_ON(index >= sblock->page_count);
1966 BUG_ON(!sblock->pagev[index]->page);
1967 page = sblock->pagev[index]->page;
1968 mapped_buffer = kmap_atomic(page);
1969 mapped_size = PAGE_SIZE;
1973 btrfs_csum_final(crc, calculated_csum);
1974 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
1977 return fail || crc_fail;
1980 static int scrub_checksum_super(struct scrub_block *sblock)
1982 struct btrfs_super_block *s;
1983 struct scrub_ctx *sctx = sblock->sctx;
1984 u8 calculated_csum[BTRFS_CSUM_SIZE];
1985 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1987 void *mapped_buffer;
1996 BUG_ON(sblock->page_count < 1);
1997 page = sblock->pagev[0]->page;
1998 mapped_buffer = kmap_atomic(page);
1999 s = (struct btrfs_super_block *)mapped_buffer;
2000 memcpy(on_disk_csum, s->csum, sctx->csum_size);
2002 if (sblock->pagev[0]->logical != btrfs_super_bytenr(s))
2005 if (sblock->pagev[0]->generation != btrfs_super_generation(s))
2008 if (!scrub_check_fsid(s->fsid, sblock->pagev[0]))
2011 len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE;
2012 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
2013 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
2016 u64 l = min_t(u64, len, mapped_size);
2018 crc = btrfs_csum_data(p, crc, l);
2019 kunmap_atomic(mapped_buffer);
2024 BUG_ON(index >= sblock->page_count);
2025 BUG_ON(!sblock->pagev[index]->page);
2026 page = sblock->pagev[index]->page;
2027 mapped_buffer = kmap_atomic(page);
2028 mapped_size = PAGE_SIZE;
2032 btrfs_csum_final(crc, calculated_csum);
2033 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
2036 if (fail_cor + fail_gen) {
2038 * if we find an error in a super block, we just report it.
2039 * They will get written with the next transaction commit
2042 spin_lock(&sctx->stat_lock);
2043 ++sctx->stat.super_errors;
2044 spin_unlock(&sctx->stat_lock);
2046 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
2047 BTRFS_DEV_STAT_CORRUPTION_ERRS);
2049 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
2050 BTRFS_DEV_STAT_GENERATION_ERRS);
2053 return fail_cor + fail_gen;
2056 static void scrub_block_get(struct scrub_block *sblock)
2058 atomic_inc(&sblock->refs);
2061 static void scrub_block_put(struct scrub_block *sblock)
2063 if (atomic_dec_and_test(&sblock->refs)) {
2066 if (sblock->sparity)
2067 scrub_parity_put(sblock->sparity);
2069 for (i = 0; i < sblock->page_count; i++)
2070 scrub_page_put(sblock->pagev[i]);
2075 static void scrub_page_get(struct scrub_page *spage)
2077 atomic_inc(&spage->refs);
2080 static void scrub_page_put(struct scrub_page *spage)
2082 if (atomic_dec_and_test(&spage->refs)) {
2084 __free_page(spage->page);
2089 static void scrub_submit(struct scrub_ctx *sctx)
2091 struct scrub_bio *sbio;
2093 if (sctx->curr == -1)
2096 sbio = sctx->bios[sctx->curr];
2098 scrub_pending_bio_inc(sctx);
2099 btrfsic_submit_bio(READ, sbio->bio);
2102 static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
2103 struct scrub_page *spage)
2105 struct scrub_block *sblock = spage->sblock;
2106 struct scrub_bio *sbio;
2111 * grab a fresh bio or wait for one to become available
2113 while (sctx->curr == -1) {
2114 spin_lock(&sctx->list_lock);
2115 sctx->curr = sctx->first_free;
2116 if (sctx->curr != -1) {
2117 sctx->first_free = sctx->bios[sctx->curr]->next_free;
2118 sctx->bios[sctx->curr]->next_free = -1;
2119 sctx->bios[sctx->curr]->page_count = 0;
2120 spin_unlock(&sctx->list_lock);
2122 spin_unlock(&sctx->list_lock);
2123 wait_event(sctx->list_wait, sctx->first_free != -1);
2126 sbio = sctx->bios[sctx->curr];
2127 if (sbio->page_count == 0) {
2130 sbio->physical = spage->physical;
2131 sbio->logical = spage->logical;
2132 sbio->dev = spage->dev;
2135 bio = btrfs_io_bio_alloc(GFP_NOFS, sctx->pages_per_rd_bio);
2141 bio->bi_private = sbio;
2142 bio->bi_end_io = scrub_bio_end_io;
2143 bio->bi_bdev = sbio->dev->bdev;
2144 bio->bi_iter.bi_sector = sbio->physical >> 9;
2146 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
2148 sbio->logical + sbio->page_count * PAGE_SIZE !=
2150 sbio->dev != spage->dev) {
2155 sbio->pagev[sbio->page_count] = spage;
2156 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
2157 if (ret != PAGE_SIZE) {
2158 if (sbio->page_count < 1) {
2167 scrub_block_get(sblock); /* one for the page added to the bio */
2168 atomic_inc(&sblock->outstanding_pages);
2170 if (sbio->page_count == sctx->pages_per_rd_bio)
2176 static void scrub_missing_raid56_end_io(struct bio *bio)
2178 struct scrub_block *sblock = bio->bi_private;
2179 struct btrfs_fs_info *fs_info = sblock->sctx->dev_root->fs_info;
2182 sblock->no_io_error_seen = 0;
2184 btrfs_queue_work(fs_info->scrub_workers, &sblock->work);
2187 static void scrub_missing_raid56_worker(struct btrfs_work *work)
2189 struct scrub_block *sblock = container_of(work, struct scrub_block, work);
2190 struct scrub_ctx *sctx = sblock->sctx;
2191 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
2192 unsigned int is_metadata;
2193 unsigned int have_csum;
2197 struct btrfs_device *dev;
2199 is_metadata = !(sblock->pagev[0]->flags & BTRFS_EXTENT_FLAG_DATA);
2200 have_csum = sblock->pagev[0]->have_csum;
2201 csum = sblock->pagev[0]->csum;
2202 generation = sblock->pagev[0]->generation;
2203 logical = sblock->pagev[0]->logical;
2204 dev = sblock->pagev[0]->dev;
2206 if (sblock->no_io_error_seen) {
2207 scrub_recheck_block_checksum(fs_info, sblock, is_metadata,
2208 have_csum, csum, generation,
2212 if (!sblock->no_io_error_seen) {
2213 spin_lock(&sctx->stat_lock);
2214 sctx->stat.read_errors++;
2215 spin_unlock(&sctx->stat_lock);
2216 btrfs_err_rl_in_rcu(fs_info,
2217 "IO error rebuilding logical %llu for dev %s",
2218 logical, rcu_str_deref(dev->name));
2219 } else if (sblock->header_error || sblock->checksum_error) {
2220 spin_lock(&sctx->stat_lock);
2221 sctx->stat.uncorrectable_errors++;
2222 spin_unlock(&sctx->stat_lock);
2223 btrfs_err_rl_in_rcu(fs_info,
2224 "failed to rebuild valid logical %llu for dev %s",
2225 logical, rcu_str_deref(dev->name));
2227 scrub_write_block_to_dev_replace(sblock);
2230 scrub_block_put(sblock);
2232 if (sctx->is_dev_replace &&
2233 atomic_read(&sctx->wr_ctx.flush_all_writes)) {
2234 mutex_lock(&sctx->wr_ctx.wr_lock);
2235 scrub_wr_submit(sctx);
2236 mutex_unlock(&sctx->wr_ctx.wr_lock);
2239 scrub_pending_bio_dec(sctx);
2242 static void scrub_missing_raid56_pages(struct scrub_block *sblock)
2244 struct scrub_ctx *sctx = sblock->sctx;
2245 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
2246 u64 length = sblock->page_count * PAGE_SIZE;
2247 u64 logical = sblock->pagev[0]->logical;
2248 struct btrfs_bio *bbio;
2250 struct btrfs_raid_bio *rbio;
2254 ret = btrfs_map_sblock(fs_info, REQ_GET_READ_MIRRORS, logical, &length,
2256 if (ret || !bbio || !bbio->raid_map)
2259 if (WARN_ON(!sctx->is_dev_replace ||
2260 !(bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK))) {
2262 * We shouldn't be scrubbing a missing device. Even for dev
2263 * replace, we should only get here for RAID 5/6. We either
2264 * managed to mount something with no mirrors remaining or
2265 * there's a bug in scrub_remap_extent()/btrfs_map_block().
2270 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
2274 bio->bi_iter.bi_sector = logical >> 9;
2275 bio->bi_private = sblock;
2276 bio->bi_end_io = scrub_missing_raid56_end_io;
2278 rbio = raid56_alloc_missing_rbio(sctx->dev_root, bio, bbio, length);
2282 for (i = 0; i < sblock->page_count; i++) {
2283 struct scrub_page *spage = sblock->pagev[i];
2285 raid56_add_scrub_pages(rbio, spage->page, spage->logical);
2288 btrfs_init_work(&sblock->work, btrfs_scrub_helper,
2289 scrub_missing_raid56_worker, NULL, NULL);
2290 scrub_block_get(sblock);
2291 scrub_pending_bio_inc(sctx);
2292 raid56_submit_missing_rbio(rbio);
2298 btrfs_put_bbio(bbio);
2299 spin_lock(&sctx->stat_lock);
2300 sctx->stat.malloc_errors++;
2301 spin_unlock(&sctx->stat_lock);
2304 static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
2305 u64 physical, struct btrfs_device *dev, u64 flags,
2306 u64 gen, int mirror_num, u8 *csum, int force,
2307 u64 physical_for_dev_replace)
2309 struct scrub_block *sblock;
2312 sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
2314 spin_lock(&sctx->stat_lock);
2315 sctx->stat.malloc_errors++;
2316 spin_unlock(&sctx->stat_lock);
2320 /* one ref inside this function, plus one for each page added to
2322 atomic_set(&sblock->refs, 1);
2323 sblock->sctx = sctx;
2324 sblock->no_io_error_seen = 1;
2326 for (index = 0; len > 0; index++) {
2327 struct scrub_page *spage;
2328 u64 l = min_t(u64, len, PAGE_SIZE);
2330 spage = kzalloc(sizeof(*spage), GFP_NOFS);
2333 spin_lock(&sctx->stat_lock);
2334 sctx->stat.malloc_errors++;
2335 spin_unlock(&sctx->stat_lock);
2336 scrub_block_put(sblock);
2339 BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
2340 scrub_page_get(spage);
2341 sblock->pagev[index] = spage;
2342 spage->sblock = sblock;
2344 spage->flags = flags;
2345 spage->generation = gen;
2346 spage->logical = logical;
2347 spage->physical = physical;
2348 spage->physical_for_dev_replace = physical_for_dev_replace;
2349 spage->mirror_num = mirror_num;
2351 spage->have_csum = 1;
2352 memcpy(spage->csum, csum, sctx->csum_size);
2354 spage->have_csum = 0;
2356 sblock->page_count++;
2357 spage->page = alloc_page(GFP_NOFS);
2363 physical_for_dev_replace += l;
2366 WARN_ON(sblock->page_count == 0);
2369 * This case should only be hit for RAID 5/6 device replace. See
2370 * the comment in scrub_missing_raid56_pages() for details.
2372 scrub_missing_raid56_pages(sblock);
2374 for (index = 0; index < sblock->page_count; index++) {
2375 struct scrub_page *spage = sblock->pagev[index];
2378 ret = scrub_add_page_to_rd_bio(sctx, spage);
2380 scrub_block_put(sblock);
2389 /* last one frees, either here or in bio completion for last page */
2390 scrub_block_put(sblock);
2394 static void scrub_bio_end_io(struct bio *bio)
2396 struct scrub_bio *sbio = bio->bi_private;
2397 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
2399 sbio->err = bio->bi_error;
2402 btrfs_queue_work(fs_info->scrub_workers, &sbio->work);
2405 static void scrub_bio_end_io_worker(struct btrfs_work *work)
2407 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
2408 struct scrub_ctx *sctx = sbio->sctx;
2411 BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO);
2413 for (i = 0; i < sbio->page_count; i++) {
2414 struct scrub_page *spage = sbio->pagev[i];
2416 spage->io_error = 1;
2417 spage->sblock->no_io_error_seen = 0;
2421 /* now complete the scrub_block items that have all pages completed */
2422 for (i = 0; i < sbio->page_count; i++) {
2423 struct scrub_page *spage = sbio->pagev[i];
2424 struct scrub_block *sblock = spage->sblock;
2426 if (atomic_dec_and_test(&sblock->outstanding_pages))
2427 scrub_block_complete(sblock);
2428 scrub_block_put(sblock);
2433 spin_lock(&sctx->list_lock);
2434 sbio->next_free = sctx->first_free;
2435 sctx->first_free = sbio->index;
2436 spin_unlock(&sctx->list_lock);
2438 if (sctx->is_dev_replace &&
2439 atomic_read(&sctx->wr_ctx.flush_all_writes)) {
2440 mutex_lock(&sctx->wr_ctx.wr_lock);
2441 scrub_wr_submit(sctx);
2442 mutex_unlock(&sctx->wr_ctx.wr_lock);
2445 scrub_pending_bio_dec(sctx);
2448 static inline void __scrub_mark_bitmap(struct scrub_parity *sparity,
2449 unsigned long *bitmap,
2454 int sectorsize = sparity->sctx->dev_root->sectorsize;
2456 if (len >= sparity->stripe_len) {
2457 bitmap_set(bitmap, 0, sparity->nsectors);
2461 start -= sparity->logic_start;
2462 start = div_u64_rem(start, sparity->stripe_len, &offset);
2463 offset /= sectorsize;
2464 nsectors = (int)len / sectorsize;
2466 if (offset + nsectors <= sparity->nsectors) {
2467 bitmap_set(bitmap, offset, nsectors);
2471 bitmap_set(bitmap, offset, sparity->nsectors - offset);
2472 bitmap_set(bitmap, 0, nsectors - (sparity->nsectors - offset));
2475 static inline void scrub_parity_mark_sectors_error(struct scrub_parity *sparity,
2478 __scrub_mark_bitmap(sparity, sparity->ebitmap, start, len);
2481 static inline void scrub_parity_mark_sectors_data(struct scrub_parity *sparity,
2484 __scrub_mark_bitmap(sparity, sparity->dbitmap, start, len);
2487 static void scrub_block_complete(struct scrub_block *sblock)
2491 if (!sblock->no_io_error_seen) {
2493 scrub_handle_errored_block(sblock);
2496 * if has checksum error, write via repair mechanism in
2497 * dev replace case, otherwise write here in dev replace
2500 corrupted = scrub_checksum(sblock);
2501 if (!corrupted && sblock->sctx->is_dev_replace)
2502 scrub_write_block_to_dev_replace(sblock);
2505 if (sblock->sparity && corrupted && !sblock->data_corrected) {
2506 u64 start = sblock->pagev[0]->logical;
2507 u64 end = sblock->pagev[sblock->page_count - 1]->logical +
2510 scrub_parity_mark_sectors_error(sblock->sparity,
2511 start, end - start);
2515 static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len,
2518 struct btrfs_ordered_sum *sum = NULL;
2519 unsigned long index;
2520 unsigned long num_sectors;
2522 while (!list_empty(&sctx->csum_list)) {
2523 sum = list_first_entry(&sctx->csum_list,
2524 struct btrfs_ordered_sum, list);
2525 if (sum->bytenr > logical)
2527 if (sum->bytenr + sum->len > logical)
2530 ++sctx->stat.csum_discards;
2531 list_del(&sum->list);
2538 index = ((u32)(logical - sum->bytenr)) / sctx->sectorsize;
2539 num_sectors = sum->len / sctx->sectorsize;
2540 memcpy(csum, sum->sums + index, sctx->csum_size);
2541 if (index == num_sectors - 1) {
2542 list_del(&sum->list);
2548 /* scrub extent tries to collect up to 64 kB for each bio */
2549 static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len,
2550 u64 physical, struct btrfs_device *dev, u64 flags,
2551 u64 gen, int mirror_num, u64 physical_for_dev_replace)
2554 u8 csum[BTRFS_CSUM_SIZE];
2557 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2558 blocksize = sctx->sectorsize;
2559 spin_lock(&sctx->stat_lock);
2560 sctx->stat.data_extents_scrubbed++;
2561 sctx->stat.data_bytes_scrubbed += len;
2562 spin_unlock(&sctx->stat_lock);
2563 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2564 blocksize = sctx->nodesize;
2565 spin_lock(&sctx->stat_lock);
2566 sctx->stat.tree_extents_scrubbed++;
2567 sctx->stat.tree_bytes_scrubbed += len;
2568 spin_unlock(&sctx->stat_lock);
2570 blocksize = sctx->sectorsize;
2575 u64 l = min_t(u64, len, blocksize);
2578 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2579 /* push csums to sbio */
2580 have_csum = scrub_find_csum(sctx, logical, l, csum);
2582 ++sctx->stat.no_csum;
2583 if (sctx->is_dev_replace && !have_csum) {
2584 ret = copy_nocow_pages(sctx, logical, l,
2586 physical_for_dev_replace);
2587 goto behind_scrub_pages;
2590 ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen,
2591 mirror_num, have_csum ? csum : NULL, 0,
2592 physical_for_dev_replace);
2599 physical_for_dev_replace += l;
2604 static int scrub_pages_for_parity(struct scrub_parity *sparity,
2605 u64 logical, u64 len,
2606 u64 physical, struct btrfs_device *dev,
2607 u64 flags, u64 gen, int mirror_num, u8 *csum)
2609 struct scrub_ctx *sctx = sparity->sctx;
2610 struct scrub_block *sblock;
2613 sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
2615 spin_lock(&sctx->stat_lock);
2616 sctx->stat.malloc_errors++;
2617 spin_unlock(&sctx->stat_lock);
2621 /* one ref inside this function, plus one for each page added to
2623 atomic_set(&sblock->refs, 1);
2624 sblock->sctx = sctx;
2625 sblock->no_io_error_seen = 1;
2626 sblock->sparity = sparity;
2627 scrub_parity_get(sparity);
2629 for (index = 0; len > 0; index++) {
2630 struct scrub_page *spage;
2631 u64 l = min_t(u64, len, PAGE_SIZE);
2633 spage = kzalloc(sizeof(*spage), GFP_NOFS);
2636 spin_lock(&sctx->stat_lock);
2637 sctx->stat.malloc_errors++;
2638 spin_unlock(&sctx->stat_lock);
2639 scrub_block_put(sblock);
2642 BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
2643 /* For scrub block */
2644 scrub_page_get(spage);
2645 sblock->pagev[index] = spage;
2646 /* For scrub parity */
2647 scrub_page_get(spage);
2648 list_add_tail(&spage->list, &sparity->spages);
2649 spage->sblock = sblock;
2651 spage->flags = flags;
2652 spage->generation = gen;
2653 spage->logical = logical;
2654 spage->physical = physical;
2655 spage->mirror_num = mirror_num;
2657 spage->have_csum = 1;
2658 memcpy(spage->csum, csum, sctx->csum_size);
2660 spage->have_csum = 0;
2662 sblock->page_count++;
2663 spage->page = alloc_page(GFP_NOFS);
2671 WARN_ON(sblock->page_count == 0);
2672 for (index = 0; index < sblock->page_count; index++) {
2673 struct scrub_page *spage = sblock->pagev[index];
2676 ret = scrub_add_page_to_rd_bio(sctx, spage);
2678 scrub_block_put(sblock);
2683 /* last one frees, either here or in bio completion for last page */
2684 scrub_block_put(sblock);
2688 static int scrub_extent_for_parity(struct scrub_parity *sparity,
2689 u64 logical, u64 len,
2690 u64 physical, struct btrfs_device *dev,
2691 u64 flags, u64 gen, int mirror_num)
2693 struct scrub_ctx *sctx = sparity->sctx;
2695 u8 csum[BTRFS_CSUM_SIZE];
2699 scrub_parity_mark_sectors_error(sparity, logical, len);
2703 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2704 blocksize = sctx->sectorsize;
2705 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2706 blocksize = sctx->nodesize;
2708 blocksize = sctx->sectorsize;
2713 u64 l = min_t(u64, len, blocksize);
2716 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2717 /* push csums to sbio */
2718 have_csum = scrub_find_csum(sctx, logical, l, csum);
2722 ret = scrub_pages_for_parity(sparity, logical, l, physical, dev,
2723 flags, gen, mirror_num,
2724 have_csum ? csum : NULL);
2736 * Given a physical address, this will calculate it's
2737 * logical offset. if this is a parity stripe, it will return
2738 * the most left data stripe's logical offset.
2740 * return 0 if it is a data stripe, 1 means parity stripe.
2742 static int get_raid56_logic_offset(u64 physical, int num,
2743 struct map_lookup *map, u64 *offset,
2753 last_offset = (physical - map->stripes[num].physical) *
2754 nr_data_stripes(map);
2756 *stripe_start = last_offset;
2758 *offset = last_offset;
2759 for (i = 0; i < nr_data_stripes(map); i++) {
2760 *offset = last_offset + i * map->stripe_len;
2762 stripe_nr = div_u64(*offset, map->stripe_len);
2763 stripe_nr = div_u64(stripe_nr, nr_data_stripes(map));
2765 /* Work out the disk rotation on this stripe-set */
2766 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, &rot);
2767 /* calculate which stripe this data locates */
2769 stripe_index = rot % map->num_stripes;
2770 if (stripe_index == num)
2772 if (stripe_index < num)
2775 *offset = last_offset + j * map->stripe_len;
2779 static void scrub_free_parity(struct scrub_parity *sparity)
2781 struct scrub_ctx *sctx = sparity->sctx;
2782 struct scrub_page *curr, *next;
2785 nbits = bitmap_weight(sparity->ebitmap, sparity->nsectors);
2787 spin_lock(&sctx->stat_lock);
2788 sctx->stat.read_errors += nbits;
2789 sctx->stat.uncorrectable_errors += nbits;
2790 spin_unlock(&sctx->stat_lock);
2793 list_for_each_entry_safe(curr, next, &sparity->spages, list) {
2794 list_del_init(&curr->list);
2795 scrub_page_put(curr);
2801 static void scrub_parity_bio_endio_worker(struct btrfs_work *work)
2803 struct scrub_parity *sparity = container_of(work, struct scrub_parity,
2805 struct scrub_ctx *sctx = sparity->sctx;
2807 scrub_free_parity(sparity);
2808 scrub_pending_bio_dec(sctx);
2811 static void scrub_parity_bio_endio(struct bio *bio)
2813 struct scrub_parity *sparity = (struct scrub_parity *)bio->bi_private;
2816 bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap,
2821 btrfs_init_work(&sparity->work, btrfs_scrubparity_helper,
2822 scrub_parity_bio_endio_worker, NULL, NULL);
2823 btrfs_queue_work(sparity->sctx->dev_root->fs_info->scrub_parity_workers,
2827 static void scrub_parity_check_and_repair(struct scrub_parity *sparity)
2829 struct scrub_ctx *sctx = sparity->sctx;
2831 struct btrfs_raid_bio *rbio;
2832 struct scrub_page *spage;
2833 struct btrfs_bio *bbio = NULL;
2837 if (!bitmap_andnot(sparity->dbitmap, sparity->dbitmap, sparity->ebitmap,
2841 length = sparity->logic_end - sparity->logic_start;
2842 ret = btrfs_map_sblock(sctx->dev_root->fs_info, WRITE,
2843 sparity->logic_start,
2844 &length, &bbio, 0, 1);
2845 if (ret || !bbio || !bbio->raid_map)
2848 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
2852 bio->bi_iter.bi_sector = sparity->logic_start >> 9;
2853 bio->bi_private = sparity;
2854 bio->bi_end_io = scrub_parity_bio_endio;
2856 rbio = raid56_parity_alloc_scrub_rbio(sctx->dev_root, bio, bbio,
2857 length, sparity->scrub_dev,
2863 list_for_each_entry(spage, &sparity->spages, list)
2864 raid56_add_scrub_pages(rbio, spage->page, spage->logical);
2866 scrub_pending_bio_inc(sctx);
2867 raid56_parity_submit_scrub_rbio(rbio);
2873 btrfs_put_bbio(bbio);
2874 bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap,
2876 spin_lock(&sctx->stat_lock);
2877 sctx->stat.malloc_errors++;
2878 spin_unlock(&sctx->stat_lock);
2880 scrub_free_parity(sparity);
2883 static inline int scrub_calc_parity_bitmap_len(int nsectors)
2885 return DIV_ROUND_UP(nsectors, BITS_PER_LONG) * (BITS_PER_LONG / 8);
2888 static void scrub_parity_get(struct scrub_parity *sparity)
2890 atomic_inc(&sparity->refs);
2893 static void scrub_parity_put(struct scrub_parity *sparity)
2895 if (!atomic_dec_and_test(&sparity->refs))
2898 scrub_parity_check_and_repair(sparity);
2901 static noinline_for_stack int scrub_raid56_parity(struct scrub_ctx *sctx,
2902 struct map_lookup *map,
2903 struct btrfs_device *sdev,
2904 struct btrfs_path *path,
2908 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
2909 struct btrfs_root *root = fs_info->extent_root;
2910 struct btrfs_root *csum_root = fs_info->csum_root;
2911 struct btrfs_extent_item *extent;
2912 struct btrfs_bio *bbio = NULL;
2916 struct extent_buffer *l;
2917 struct btrfs_key key;
2920 u64 extent_physical;
2923 struct btrfs_device *extent_dev;
2924 struct scrub_parity *sparity;
2927 int extent_mirror_num;
2930 nsectors = map->stripe_len / root->sectorsize;
2931 bitmap_len = scrub_calc_parity_bitmap_len(nsectors);
2932 sparity = kzalloc(sizeof(struct scrub_parity) + 2 * bitmap_len,
2935 spin_lock(&sctx->stat_lock);
2936 sctx->stat.malloc_errors++;
2937 spin_unlock(&sctx->stat_lock);
2941 sparity->stripe_len = map->stripe_len;
2942 sparity->nsectors = nsectors;
2943 sparity->sctx = sctx;
2944 sparity->scrub_dev = sdev;
2945 sparity->logic_start = logic_start;
2946 sparity->logic_end = logic_end;
2947 atomic_set(&sparity->refs, 1);
2948 INIT_LIST_HEAD(&sparity->spages);
2949 sparity->dbitmap = sparity->bitmap;
2950 sparity->ebitmap = (void *)sparity->bitmap + bitmap_len;
2953 while (logic_start < logic_end) {
2954 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2955 key.type = BTRFS_METADATA_ITEM_KEY;
2957 key.type = BTRFS_EXTENT_ITEM_KEY;
2958 key.objectid = logic_start;
2959 key.offset = (u64)-1;
2961 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2966 ret = btrfs_previous_extent_item(root, path, 0);
2970 btrfs_release_path(path);
2971 ret = btrfs_search_slot(NULL, root, &key,
2983 slot = path->slots[0];
2984 if (slot >= btrfs_header_nritems(l)) {
2985 ret = btrfs_next_leaf(root, path);
2994 btrfs_item_key_to_cpu(l, &key, slot);
2996 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
2997 key.type != BTRFS_METADATA_ITEM_KEY)
3000 if (key.type == BTRFS_METADATA_ITEM_KEY)
3001 bytes = root->nodesize;
3005 if (key.objectid + bytes <= logic_start)
3008 if (key.objectid >= logic_end) {
3013 while (key.objectid >= logic_start + map->stripe_len)
3014 logic_start += map->stripe_len;
3016 extent = btrfs_item_ptr(l, slot,
3017 struct btrfs_extent_item);
3018 flags = btrfs_extent_flags(l, extent);
3019 generation = btrfs_extent_generation(l, extent);
3021 if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) &&
3022 (key.objectid < logic_start ||
3023 key.objectid + bytes >
3024 logic_start + map->stripe_len)) {
3025 btrfs_err(fs_info, "scrub: tree block %llu spanning stripes, ignored. logical=%llu",
3026 key.objectid, logic_start);
3027 spin_lock(&sctx->stat_lock);
3028 sctx->stat.uncorrectable_errors++;
3029 spin_unlock(&sctx->stat_lock);
3033 extent_logical = key.objectid;
3036 if (extent_logical < logic_start) {
3037 extent_len -= logic_start - extent_logical;
3038 extent_logical = logic_start;
3041 if (extent_logical + extent_len >
3042 logic_start + map->stripe_len)
3043 extent_len = logic_start + map->stripe_len -
3046 scrub_parity_mark_sectors_data(sparity, extent_logical,
3049 mapped_length = extent_len;
3050 ret = btrfs_map_block(fs_info, READ, extent_logical,
3051 &mapped_length, &bbio, 0);
3053 if (!bbio || mapped_length < extent_len)
3057 btrfs_put_bbio(bbio);
3060 extent_physical = bbio->stripes[0].physical;
3061 extent_mirror_num = bbio->mirror_num;
3062 extent_dev = bbio->stripes[0].dev;
3063 btrfs_put_bbio(bbio);
3065 ret = btrfs_lookup_csums_range(csum_root,
3067 extent_logical + extent_len - 1,
3068 &sctx->csum_list, 1);
3072 ret = scrub_extent_for_parity(sparity, extent_logical,
3079 scrub_free_csums(sctx);
3084 if (extent_logical + extent_len <
3085 key.objectid + bytes) {
3086 logic_start += map->stripe_len;
3088 if (logic_start >= logic_end) {
3093 if (logic_start < key.objectid + bytes) {
3102 btrfs_release_path(path);
3107 logic_start += map->stripe_len;
3111 scrub_parity_mark_sectors_error(sparity, logic_start,
3112 logic_end - logic_start);
3113 scrub_parity_put(sparity);
3115 mutex_lock(&sctx->wr_ctx.wr_lock);
3116 scrub_wr_submit(sctx);
3117 mutex_unlock(&sctx->wr_ctx.wr_lock);
3119 btrfs_release_path(path);
3120 return ret < 0 ? ret : 0;
3123 static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
3124 struct map_lookup *map,
3125 struct btrfs_device *scrub_dev,
3126 int num, u64 base, u64 length,
3129 struct btrfs_path *path, *ppath;
3130 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
3131 struct btrfs_root *root = fs_info->extent_root;
3132 struct btrfs_root *csum_root = fs_info->csum_root;
3133 struct btrfs_extent_item *extent;
3134 struct blk_plug plug;
3139 struct extent_buffer *l;
3140 struct btrfs_key key;
3147 struct reada_control *reada1;
3148 struct reada_control *reada2;
3149 struct btrfs_key key_start;
3150 struct btrfs_key key_end;
3151 u64 increment = map->stripe_len;
3154 u64 extent_physical;
3158 struct btrfs_device *extent_dev;
3159 int extent_mirror_num;
3162 physical = map->stripes[num].physical;
3164 nstripes = div_u64(length, map->stripe_len);
3165 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3166 offset = map->stripe_len * num;
3167 increment = map->stripe_len * map->num_stripes;
3169 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3170 int factor = map->num_stripes / map->sub_stripes;
3171 offset = map->stripe_len * (num / map->sub_stripes);
3172 increment = map->stripe_len * factor;
3173 mirror_num = num % map->sub_stripes + 1;
3174 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3175 increment = map->stripe_len;
3176 mirror_num = num % map->num_stripes + 1;
3177 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3178 increment = map->stripe_len;
3179 mirror_num = num % map->num_stripes + 1;
3180 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3181 get_raid56_logic_offset(physical, num, map, &offset, NULL);
3182 increment = map->stripe_len * nr_data_stripes(map);
3185 increment = map->stripe_len;
3189 path = btrfs_alloc_path();
3193 ppath = btrfs_alloc_path();
3195 btrfs_free_path(path);
3200 * work on commit root. The related disk blocks are static as
3201 * long as COW is applied. This means, it is save to rewrite
3202 * them to repair disk errors without any race conditions
3204 path->search_commit_root = 1;
3205 path->skip_locking = 1;
3207 ppath->search_commit_root = 1;
3208 ppath->skip_locking = 1;
3210 * trigger the readahead for extent tree csum tree and wait for
3211 * completion. During readahead, the scrub is officially paused
3212 * to not hold off transaction commits
3214 logical = base + offset;
3215 physical_end = physical + nstripes * map->stripe_len;
3216 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3217 get_raid56_logic_offset(physical_end, num,
3218 map, &logic_end, NULL);
3221 logic_end = logical + increment * nstripes;
3223 wait_event(sctx->list_wait,
3224 atomic_read(&sctx->bios_in_flight) == 0);
3225 scrub_blocked_if_needed(fs_info);
3227 /* FIXME it might be better to start readahead at commit root */
3228 key_start.objectid = logical;
3229 key_start.type = BTRFS_EXTENT_ITEM_KEY;
3230 key_start.offset = (u64)0;
3231 key_end.objectid = logic_end;
3232 key_end.type = BTRFS_METADATA_ITEM_KEY;
3233 key_end.offset = (u64)-1;
3234 reada1 = btrfs_reada_add(root, &key_start, &key_end);
3236 key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
3237 key_start.type = BTRFS_EXTENT_CSUM_KEY;
3238 key_start.offset = logical;
3239 key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
3240 key_end.type = BTRFS_EXTENT_CSUM_KEY;
3241 key_end.offset = logic_end;
3242 reada2 = btrfs_reada_add(csum_root, &key_start, &key_end);
3244 if (!IS_ERR(reada1))
3245 btrfs_reada_wait(reada1);
3246 if (!IS_ERR(reada2))
3247 btrfs_reada_wait(reada2);
3251 * collect all data csums for the stripe to avoid seeking during
3252 * the scrub. This might currently (crc32) end up to be about 1MB
3254 blk_start_plug(&plug);
3257 * now find all extents for each stripe and scrub them
3260 while (physical < physical_end) {
3264 if (atomic_read(&fs_info->scrub_cancel_req) ||
3265 atomic_read(&sctx->cancel_req)) {
3270 * check to see if we have to pause
3272 if (atomic_read(&fs_info->scrub_pause_req)) {
3273 /* push queued extents */
3274 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
3276 mutex_lock(&sctx->wr_ctx.wr_lock);
3277 scrub_wr_submit(sctx);
3278 mutex_unlock(&sctx->wr_ctx.wr_lock);
3279 wait_event(sctx->list_wait,
3280 atomic_read(&sctx->bios_in_flight) == 0);
3281 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
3282 scrub_blocked_if_needed(fs_info);
3285 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3286 ret = get_raid56_logic_offset(physical, num, map,
3291 /* it is parity strip */
3292 stripe_logical += base;
3293 stripe_end = stripe_logical + increment;
3294 ret = scrub_raid56_parity(sctx, map, scrub_dev,
3295 ppath, stripe_logical,
3303 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
3304 key.type = BTRFS_METADATA_ITEM_KEY;
3306 key.type = BTRFS_EXTENT_ITEM_KEY;
3307 key.objectid = logical;
3308 key.offset = (u64)-1;
3310 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3315 ret = btrfs_previous_extent_item(root, path, 0);
3319 /* there's no smaller item, so stick with the
3321 btrfs_release_path(path);
3322 ret = btrfs_search_slot(NULL, root, &key,
3334 slot = path->slots[0];
3335 if (slot >= btrfs_header_nritems(l)) {
3336 ret = btrfs_next_leaf(root, path);
3345 btrfs_item_key_to_cpu(l, &key, slot);
3347 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3348 key.type != BTRFS_METADATA_ITEM_KEY)
3351 if (key.type == BTRFS_METADATA_ITEM_KEY)
3352 bytes = root->nodesize;
3356 if (key.objectid + bytes <= logical)
3359 if (key.objectid >= logical + map->stripe_len) {
3360 /* out of this device extent */
3361 if (key.objectid >= logic_end)
3366 extent = btrfs_item_ptr(l, slot,
3367 struct btrfs_extent_item);
3368 flags = btrfs_extent_flags(l, extent);
3369 generation = btrfs_extent_generation(l, extent);
3371 if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) &&
3372 (key.objectid < logical ||
3373 key.objectid + bytes >
3374 logical + map->stripe_len)) {
3376 "scrub: tree block %llu spanning "
3377 "stripes, ignored. logical=%llu",
3378 key.objectid, logical);
3379 spin_lock(&sctx->stat_lock);
3380 sctx->stat.uncorrectable_errors++;
3381 spin_unlock(&sctx->stat_lock);
3386 extent_logical = key.objectid;
3390 * trim extent to this stripe
3392 if (extent_logical < logical) {
3393 extent_len -= logical - extent_logical;
3394 extent_logical = logical;
3396 if (extent_logical + extent_len >
3397 logical + map->stripe_len) {
3398 extent_len = logical + map->stripe_len -
3402 extent_physical = extent_logical - logical + physical;
3403 extent_dev = scrub_dev;
3404 extent_mirror_num = mirror_num;
3406 scrub_remap_extent(fs_info, extent_logical,
3407 extent_len, &extent_physical,
3409 &extent_mirror_num);
3411 ret = btrfs_lookup_csums_range(csum_root,
3415 &sctx->csum_list, 1);
3419 ret = scrub_extent(sctx, extent_logical, extent_len,
3420 extent_physical, extent_dev, flags,
3421 generation, extent_mirror_num,
3422 extent_logical - logical + physical);
3424 scrub_free_csums(sctx);
3429 if (extent_logical + extent_len <
3430 key.objectid + bytes) {
3431 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3433 * loop until we find next data stripe
3434 * or we have finished all stripes.
3437 physical += map->stripe_len;
3438 ret = get_raid56_logic_offset(physical,
3443 if (ret && physical < physical_end) {
3444 stripe_logical += base;
3445 stripe_end = stripe_logical +
3447 ret = scrub_raid56_parity(sctx,
3448 map, scrub_dev, ppath,
3456 physical += map->stripe_len;
3457 logical += increment;
3459 if (logical < key.objectid + bytes) {
3464 if (physical >= physical_end) {
3472 btrfs_release_path(path);
3474 logical += increment;
3475 physical += map->stripe_len;
3476 spin_lock(&sctx->stat_lock);
3478 sctx->stat.last_physical = map->stripes[num].physical +
3481 sctx->stat.last_physical = physical;
3482 spin_unlock(&sctx->stat_lock);
3487 /* push queued extents */
3489 mutex_lock(&sctx->wr_ctx.wr_lock);
3490 scrub_wr_submit(sctx);
3491 mutex_unlock(&sctx->wr_ctx.wr_lock);
3493 blk_finish_plug(&plug);
3494 btrfs_free_path(path);
3495 btrfs_free_path(ppath);
3496 return ret < 0 ? ret : 0;
3499 static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx,
3500 struct btrfs_device *scrub_dev,
3501 u64 chunk_offset, u64 length,
3502 u64 dev_offset, int is_dev_replace)
3504 struct btrfs_mapping_tree *map_tree =
3505 &sctx->dev_root->fs_info->mapping_tree;
3506 struct map_lookup *map;
3507 struct extent_map *em;
3511 read_lock(&map_tree->map_tree.lock);
3512 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3513 read_unlock(&map_tree->map_tree.lock);
3518 map = (struct map_lookup *)em->bdev;
3519 if (em->start != chunk_offset)
3522 if (em->len < length)
3525 for (i = 0; i < map->num_stripes; ++i) {
3526 if (map->stripes[i].dev->bdev == scrub_dev->bdev &&
3527 map->stripes[i].physical == dev_offset) {
3528 ret = scrub_stripe(sctx, map, scrub_dev, i,
3529 chunk_offset, length,
3536 free_extent_map(em);
3541 static noinline_for_stack
3542 int scrub_enumerate_chunks(struct scrub_ctx *sctx,
3543 struct btrfs_device *scrub_dev, u64 start, u64 end,
3546 struct btrfs_dev_extent *dev_extent = NULL;
3547 struct btrfs_path *path;
3548 struct btrfs_root *root = sctx->dev_root;
3549 struct btrfs_fs_info *fs_info = root->fs_info;
3554 struct extent_buffer *l;
3555 struct btrfs_key key;
3556 struct btrfs_key found_key;
3557 struct btrfs_block_group_cache *cache;
3558 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
3560 path = btrfs_alloc_path();
3565 path->search_commit_root = 1;
3566 path->skip_locking = 1;
3568 key.objectid = scrub_dev->devid;
3570 key.type = BTRFS_DEV_EXTENT_KEY;
3573 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3577 if (path->slots[0] >=
3578 btrfs_header_nritems(path->nodes[0])) {
3579 ret = btrfs_next_leaf(root, path);
3592 slot = path->slots[0];
3594 btrfs_item_key_to_cpu(l, &found_key, slot);
3596 if (found_key.objectid != scrub_dev->devid)
3599 if (found_key.type != BTRFS_DEV_EXTENT_KEY)
3602 if (found_key.offset >= end)
3605 if (found_key.offset < key.offset)
3608 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3609 length = btrfs_dev_extent_length(l, dev_extent);
3611 if (found_key.offset + length <= start)
3614 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3617 * get a reference on the corresponding block group to prevent
3618 * the chunk from going away while we scrub it
3620 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3622 /* some chunks are removed but not committed to disk yet,
3623 * continue scrubbing */
3628 * we need call btrfs_inc_block_group_ro() with scrubs_paused,
3629 * to avoid deadlock caused by:
3630 * btrfs_inc_block_group_ro()
3631 * -> btrfs_wait_for_commit()
3632 * -> btrfs_commit_transaction()
3633 * -> btrfs_scrub_pause()
3635 scrub_pause_on(fs_info);
3636 ret = btrfs_inc_block_group_ro(root, cache);
3637 scrub_pause_off(fs_info);
3639 btrfs_put_block_group(cache);
3643 dev_replace->cursor_right = found_key.offset + length;
3644 dev_replace->cursor_left = found_key.offset;
3645 dev_replace->item_needs_writeback = 1;
3646 ret = scrub_chunk(sctx, scrub_dev, chunk_offset, length,
3647 found_key.offset, is_dev_replace);
3650 * flush, submit all pending read and write bios, afterwards
3652 * Note that in the dev replace case, a read request causes
3653 * write requests that are submitted in the read completion
3654 * worker. Therefore in the current situation, it is required
3655 * that all write requests are flushed, so that all read and
3656 * write requests are really completed when bios_in_flight
3659 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
3661 mutex_lock(&sctx->wr_ctx.wr_lock);
3662 scrub_wr_submit(sctx);
3663 mutex_unlock(&sctx->wr_ctx.wr_lock);
3665 wait_event(sctx->list_wait,
3666 atomic_read(&sctx->bios_in_flight) == 0);
3668 scrub_pause_on(fs_info);
3671 * must be called before we decrease @scrub_paused.
3672 * make sure we don't block transaction commit while
3673 * we are waiting pending workers finished.
3675 wait_event(sctx->list_wait,
3676 atomic_read(&sctx->workers_pending) == 0);
3677 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
3679 scrub_pause_off(fs_info);
3681 btrfs_dec_block_group_ro(root, cache);
3683 btrfs_put_block_group(cache);
3686 if (is_dev_replace &&
3687 atomic64_read(&dev_replace->num_write_errors) > 0) {
3691 if (sctx->stat.malloc_errors > 0) {
3696 dev_replace->cursor_left = dev_replace->cursor_right;
3697 dev_replace->item_needs_writeback = 1;
3699 key.offset = found_key.offset + length;
3700 btrfs_release_path(path);
3703 btrfs_free_path(path);
3708 static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx,
3709 struct btrfs_device *scrub_dev)
3715 struct btrfs_root *root = sctx->dev_root;
3717 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
3720 /* Seed devices of a new filesystem has their own generation. */
3721 if (scrub_dev->fs_devices != root->fs_info->fs_devices)
3722 gen = scrub_dev->generation;
3724 gen = root->fs_info->last_trans_committed;
3726 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
3727 bytenr = btrfs_sb_offset(i);
3728 if (bytenr + BTRFS_SUPER_INFO_SIZE >
3729 scrub_dev->commit_total_bytes)
3732 ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr,
3733 scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i,
3738 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
3744 * get a reference count on fs_info->scrub_workers. start worker if necessary
3746 static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info,
3749 unsigned int flags = WQ_FREEZABLE | WQ_UNBOUND;
3750 int max_active = fs_info->thread_pool_size;
3752 if (fs_info->scrub_workers_refcnt == 0) {
3754 fs_info->scrub_workers =
3755 btrfs_alloc_workqueue("btrfs-scrub", flags,
3758 fs_info->scrub_workers =
3759 btrfs_alloc_workqueue("btrfs-scrub", flags,
3761 if (!fs_info->scrub_workers)
3762 goto fail_scrub_workers;
3764 fs_info->scrub_wr_completion_workers =
3765 btrfs_alloc_workqueue("btrfs-scrubwrc", flags,
3767 if (!fs_info->scrub_wr_completion_workers)
3768 goto fail_scrub_wr_completion_workers;
3770 fs_info->scrub_nocow_workers =
3771 btrfs_alloc_workqueue("btrfs-scrubnc", flags, 1, 0);
3772 if (!fs_info->scrub_nocow_workers)
3773 goto fail_scrub_nocow_workers;
3774 fs_info->scrub_parity_workers =
3775 btrfs_alloc_workqueue("btrfs-scrubparity", flags,
3777 if (!fs_info->scrub_parity_workers)
3778 goto fail_scrub_parity_workers;
3780 ++fs_info->scrub_workers_refcnt;
3783 fail_scrub_parity_workers:
3784 btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
3785 fail_scrub_nocow_workers:
3786 btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
3787 fail_scrub_wr_completion_workers:
3788 btrfs_destroy_workqueue(fs_info->scrub_workers);
3793 static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info)
3795 if (--fs_info->scrub_workers_refcnt == 0) {
3796 btrfs_destroy_workqueue(fs_info->scrub_workers);
3797 btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
3798 btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
3799 btrfs_destroy_workqueue(fs_info->scrub_parity_workers);
3801 WARN_ON(fs_info->scrub_workers_refcnt < 0);
3804 int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start,
3805 u64 end, struct btrfs_scrub_progress *progress,
3806 int readonly, int is_dev_replace)
3808 struct scrub_ctx *sctx;
3810 struct btrfs_device *dev;
3811 struct rcu_string *name;
3813 if (btrfs_fs_closing(fs_info))
3816 if (fs_info->chunk_root->nodesize > BTRFS_STRIPE_LEN) {
3818 * in this case scrub is unable to calculate the checksum
3819 * the way scrub is implemented. Do not handle this
3820 * situation at all because it won't ever happen.
3823 "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails",
3824 fs_info->chunk_root->nodesize, BTRFS_STRIPE_LEN);
3828 if (fs_info->chunk_root->sectorsize != PAGE_SIZE) {
3829 /* not supported for data w/o checksums */
3831 "scrub: size assumption sectorsize != PAGE_SIZE "
3832 "(%d != %lu) fails",
3833 fs_info->chunk_root->sectorsize, PAGE_SIZE);
3837 if (fs_info->chunk_root->nodesize >
3838 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK ||
3839 fs_info->chunk_root->sectorsize >
3840 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) {
3842 * would exhaust the array bounds of pagev member in
3843 * struct scrub_block
3845 btrfs_err(fs_info, "scrub: size assumption nodesize and sectorsize "
3846 "<= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails",
3847 fs_info->chunk_root->nodesize,
3848 SCRUB_MAX_PAGES_PER_BLOCK,
3849 fs_info->chunk_root->sectorsize,
3850 SCRUB_MAX_PAGES_PER_BLOCK);
3855 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3856 dev = btrfs_find_device(fs_info, devid, NULL, NULL);
3857 if (!dev || (dev->missing && !is_dev_replace)) {
3858 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3862 if (!is_dev_replace && !readonly && !dev->writeable) {
3863 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3865 name = rcu_dereference(dev->name);
3866 btrfs_err(fs_info, "scrub: device %s is not writable",
3872 mutex_lock(&fs_info->scrub_lock);
3873 if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) {
3874 mutex_unlock(&fs_info->scrub_lock);
3875 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3879 btrfs_dev_replace_lock(&fs_info->dev_replace);
3880 if (dev->scrub_device ||
3882 btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) {
3883 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3884 mutex_unlock(&fs_info->scrub_lock);
3885 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3886 return -EINPROGRESS;
3888 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3890 ret = scrub_workers_get(fs_info, is_dev_replace);
3892 mutex_unlock(&fs_info->scrub_lock);
3893 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3897 sctx = scrub_setup_ctx(dev, is_dev_replace);
3899 mutex_unlock(&fs_info->scrub_lock);
3900 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3901 scrub_workers_put(fs_info);
3902 return PTR_ERR(sctx);
3904 sctx->readonly = readonly;
3905 dev->scrub_device = sctx;
3906 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3909 * checking @scrub_pause_req here, we can avoid
3910 * race between committing transaction and scrubbing.
3912 __scrub_blocked_if_needed(fs_info);
3913 atomic_inc(&fs_info->scrubs_running);
3914 mutex_unlock(&fs_info->scrub_lock);
3916 if (!is_dev_replace) {
3918 * by holding device list mutex, we can
3919 * kick off writing super in log tree sync.
3921 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3922 ret = scrub_supers(sctx, dev);
3923 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3927 ret = scrub_enumerate_chunks(sctx, dev, start, end,
3930 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
3931 atomic_dec(&fs_info->scrubs_running);
3932 wake_up(&fs_info->scrub_pause_wait);
3934 wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0);
3937 memcpy(progress, &sctx->stat, sizeof(*progress));
3939 mutex_lock(&fs_info->scrub_lock);
3940 dev->scrub_device = NULL;
3941 scrub_workers_put(fs_info);
3942 mutex_unlock(&fs_info->scrub_lock);
3944 scrub_put_ctx(sctx);
3949 void btrfs_scrub_pause(struct btrfs_root *root)
3951 struct btrfs_fs_info *fs_info = root->fs_info;
3953 mutex_lock(&fs_info->scrub_lock);
3954 atomic_inc(&fs_info->scrub_pause_req);
3955 while (atomic_read(&fs_info->scrubs_paused) !=
3956 atomic_read(&fs_info->scrubs_running)) {
3957 mutex_unlock(&fs_info->scrub_lock);
3958 wait_event(fs_info->scrub_pause_wait,
3959 atomic_read(&fs_info->scrubs_paused) ==
3960 atomic_read(&fs_info->scrubs_running));
3961 mutex_lock(&fs_info->scrub_lock);
3963 mutex_unlock(&fs_info->scrub_lock);
3966 void btrfs_scrub_continue(struct btrfs_root *root)
3968 struct btrfs_fs_info *fs_info = root->fs_info;
3970 atomic_dec(&fs_info->scrub_pause_req);
3971 wake_up(&fs_info->scrub_pause_wait);
3974 int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info)
3976 mutex_lock(&fs_info->scrub_lock);
3977 if (!atomic_read(&fs_info->scrubs_running)) {
3978 mutex_unlock(&fs_info->scrub_lock);
3982 atomic_inc(&fs_info->scrub_cancel_req);
3983 while (atomic_read(&fs_info->scrubs_running)) {
3984 mutex_unlock(&fs_info->scrub_lock);
3985 wait_event(fs_info->scrub_pause_wait,
3986 atomic_read(&fs_info->scrubs_running) == 0);
3987 mutex_lock(&fs_info->scrub_lock);
3989 atomic_dec(&fs_info->scrub_cancel_req);
3990 mutex_unlock(&fs_info->scrub_lock);
3995 int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info,
3996 struct btrfs_device *dev)
3998 struct scrub_ctx *sctx;
4000 mutex_lock(&fs_info->scrub_lock);
4001 sctx = dev->scrub_device;
4003 mutex_unlock(&fs_info->scrub_lock);
4006 atomic_inc(&sctx->cancel_req);
4007 while (dev->scrub_device) {
4008 mutex_unlock(&fs_info->scrub_lock);
4009 wait_event(fs_info->scrub_pause_wait,
4010 dev->scrub_device == NULL);
4011 mutex_lock(&fs_info->scrub_lock);
4013 mutex_unlock(&fs_info->scrub_lock);
4018 int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
4019 struct btrfs_scrub_progress *progress)
4021 struct btrfs_device *dev;
4022 struct scrub_ctx *sctx = NULL;
4024 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
4025 dev = btrfs_find_device(root->fs_info, devid, NULL, NULL);
4027 sctx = dev->scrub_device;
4029 memcpy(progress, &sctx->stat, sizeof(*progress));
4030 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
4032 return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV;
4035 static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
4036 u64 extent_logical, u64 extent_len,
4037 u64 *extent_physical,
4038 struct btrfs_device **extent_dev,
4039 int *extent_mirror_num)
4042 struct btrfs_bio *bbio = NULL;
4045 mapped_length = extent_len;
4046 ret = btrfs_map_block(fs_info, READ, extent_logical,
4047 &mapped_length, &bbio, 0);
4048 if (ret || !bbio || mapped_length < extent_len ||
4049 !bbio->stripes[0].dev->bdev) {
4050 btrfs_put_bbio(bbio);
4054 *extent_physical = bbio->stripes[0].physical;
4055 *extent_mirror_num = bbio->mirror_num;
4056 *extent_dev = bbio->stripes[0].dev;
4057 btrfs_put_bbio(bbio);
4060 static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
4061 struct scrub_wr_ctx *wr_ctx,
4062 struct btrfs_fs_info *fs_info,
4063 struct btrfs_device *dev,
4066 WARN_ON(wr_ctx->wr_curr_bio != NULL);
4068 mutex_init(&wr_ctx->wr_lock);
4069 wr_ctx->wr_curr_bio = NULL;
4070 if (!is_dev_replace)
4073 WARN_ON(!dev->bdev);
4074 wr_ctx->pages_per_wr_bio = SCRUB_PAGES_PER_WR_BIO;
4075 wr_ctx->tgtdev = dev;
4076 atomic_set(&wr_ctx->flush_all_writes, 0);
4080 static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx)
4082 mutex_lock(&wr_ctx->wr_lock);
4083 kfree(wr_ctx->wr_curr_bio);
4084 wr_ctx->wr_curr_bio = NULL;
4085 mutex_unlock(&wr_ctx->wr_lock);
4088 static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
4089 int mirror_num, u64 physical_for_dev_replace)
4091 struct scrub_copy_nocow_ctx *nocow_ctx;
4092 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
4094 nocow_ctx = kzalloc(sizeof(*nocow_ctx), GFP_NOFS);
4096 spin_lock(&sctx->stat_lock);
4097 sctx->stat.malloc_errors++;
4098 spin_unlock(&sctx->stat_lock);
4102 scrub_pending_trans_workers_inc(sctx);
4104 nocow_ctx->sctx = sctx;
4105 nocow_ctx->logical = logical;
4106 nocow_ctx->len = len;
4107 nocow_ctx->mirror_num = mirror_num;
4108 nocow_ctx->physical_for_dev_replace = physical_for_dev_replace;
4109 btrfs_init_work(&nocow_ctx->work, btrfs_scrubnc_helper,
4110 copy_nocow_pages_worker, NULL, NULL);
4111 INIT_LIST_HEAD(&nocow_ctx->inodes);
4112 btrfs_queue_work(fs_info->scrub_nocow_workers,
4118 static int record_inode_for_nocow(u64 inum, u64 offset, u64 root, void *ctx)
4120 struct scrub_copy_nocow_ctx *nocow_ctx = ctx;
4121 struct scrub_nocow_inode *nocow_inode;
4123 nocow_inode = kzalloc(sizeof(*nocow_inode), GFP_NOFS);
4126 nocow_inode->inum = inum;
4127 nocow_inode->offset = offset;
4128 nocow_inode->root = root;
4129 list_add_tail(&nocow_inode->list, &nocow_ctx->inodes);
4133 #define COPY_COMPLETE 1
4135 static void copy_nocow_pages_worker(struct btrfs_work *work)
4137 struct scrub_copy_nocow_ctx *nocow_ctx =
4138 container_of(work, struct scrub_copy_nocow_ctx, work);
4139 struct scrub_ctx *sctx = nocow_ctx->sctx;
4140 u64 logical = nocow_ctx->logical;
4141 u64 len = nocow_ctx->len;
4142 int mirror_num = nocow_ctx->mirror_num;
4143 u64 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
4145 struct btrfs_trans_handle *trans = NULL;
4146 struct btrfs_fs_info *fs_info;
4147 struct btrfs_path *path;
4148 struct btrfs_root *root;
4149 int not_written = 0;
4151 fs_info = sctx->dev_root->fs_info;
4152 root = fs_info->extent_root;
4154 path = btrfs_alloc_path();
4156 spin_lock(&sctx->stat_lock);
4157 sctx->stat.malloc_errors++;
4158 spin_unlock(&sctx->stat_lock);
4163 trans = btrfs_join_transaction(root);
4164 if (IS_ERR(trans)) {
4169 ret = iterate_inodes_from_logical(logical, fs_info, path,
4170 record_inode_for_nocow, nocow_ctx);
4171 if (ret != 0 && ret != -ENOENT) {
4172 btrfs_warn(fs_info, "iterate_inodes_from_logical() failed: log %llu, "
4173 "phys %llu, len %llu, mir %u, ret %d",
4174 logical, physical_for_dev_replace, len, mirror_num,
4180 btrfs_end_transaction(trans, root);
4182 while (!list_empty(&nocow_ctx->inodes)) {
4183 struct scrub_nocow_inode *entry;
4184 entry = list_first_entry(&nocow_ctx->inodes,
4185 struct scrub_nocow_inode,
4187 list_del_init(&entry->list);
4188 ret = copy_nocow_pages_for_inode(entry->inum, entry->offset,
4189 entry->root, nocow_ctx);
4191 if (ret == COPY_COMPLETE) {
4199 while (!list_empty(&nocow_ctx->inodes)) {
4200 struct scrub_nocow_inode *entry;
4201 entry = list_first_entry(&nocow_ctx->inodes,
4202 struct scrub_nocow_inode,
4204 list_del_init(&entry->list);
4207 if (trans && !IS_ERR(trans))
4208 btrfs_end_transaction(trans, root);
4210 btrfs_dev_replace_stats_inc(&fs_info->dev_replace.
4211 num_uncorrectable_read_errors);
4213 btrfs_free_path(path);
4216 scrub_pending_trans_workers_dec(sctx);
4219 static int check_extent_to_block(struct inode *inode, u64 start, u64 len,
4222 struct extent_state *cached_state = NULL;
4223 struct btrfs_ordered_extent *ordered;
4224 struct extent_io_tree *io_tree;
4225 struct extent_map *em;
4226 u64 lockstart = start, lockend = start + len - 1;
4229 io_tree = &BTRFS_I(inode)->io_tree;
4231 lock_extent_bits(io_tree, lockstart, lockend, 0, &cached_state);
4232 ordered = btrfs_lookup_ordered_range(inode, lockstart, len);
4234 btrfs_put_ordered_extent(ordered);
4239 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
4246 * This extent does not actually cover the logical extent anymore,
4247 * move on to the next inode.
4249 if (em->block_start > logical ||
4250 em->block_start + em->block_len < logical + len) {
4251 free_extent_map(em);
4255 free_extent_map(em);
4258 unlock_extent_cached(io_tree, lockstart, lockend, &cached_state,
4263 static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
4264 struct scrub_copy_nocow_ctx *nocow_ctx)
4266 struct btrfs_fs_info *fs_info = nocow_ctx->sctx->dev_root->fs_info;
4267 struct btrfs_key key;
4268 struct inode *inode;
4270 struct btrfs_root *local_root;
4271 struct extent_io_tree *io_tree;
4272 u64 physical_for_dev_replace;
4273 u64 nocow_ctx_logical;
4274 u64 len = nocow_ctx->len;
4275 unsigned long index;
4280 key.objectid = root;
4281 key.type = BTRFS_ROOT_ITEM_KEY;
4282 key.offset = (u64)-1;
4284 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
4286 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
4287 if (IS_ERR(local_root)) {
4288 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
4289 return PTR_ERR(local_root);
4292 key.type = BTRFS_INODE_ITEM_KEY;
4293 key.objectid = inum;
4295 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
4296 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
4298 return PTR_ERR(inode);
4300 /* Avoid truncate/dio/punch hole.. */
4301 mutex_lock(&inode->i_mutex);
4302 inode_dio_wait(inode);
4304 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
4305 io_tree = &BTRFS_I(inode)->io_tree;
4306 nocow_ctx_logical = nocow_ctx->logical;
4308 ret = check_extent_to_block(inode, offset, len, nocow_ctx_logical);
4310 ret = ret > 0 ? 0 : ret;
4314 while (len >= PAGE_CACHE_SIZE) {
4315 index = offset >> PAGE_CACHE_SHIFT;
4317 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
4319 btrfs_err(fs_info, "find_or_create_page() failed");
4324 if (PageUptodate(page)) {
4325 if (PageDirty(page))
4328 ClearPageError(page);
4329 err = extent_read_full_page(io_tree, page,
4331 nocow_ctx->mirror_num);
4339 * If the page has been remove from the page cache,
4340 * the data on it is meaningless, because it may be
4341 * old one, the new data may be written into the new
4342 * page in the page cache.
4344 if (page->mapping != inode->i_mapping) {
4346 page_cache_release(page);
4349 if (!PageUptodate(page)) {
4355 ret = check_extent_to_block(inode, offset, len,
4358 ret = ret > 0 ? 0 : ret;
4362 err = write_page_nocow(nocow_ctx->sctx,
4363 physical_for_dev_replace, page);
4368 page_cache_release(page);
4373 offset += PAGE_CACHE_SIZE;
4374 physical_for_dev_replace += PAGE_CACHE_SIZE;
4375 nocow_ctx_logical += PAGE_CACHE_SIZE;
4376 len -= PAGE_CACHE_SIZE;
4378 ret = COPY_COMPLETE;
4380 mutex_unlock(&inode->i_mutex);
4385 static int write_page_nocow(struct scrub_ctx *sctx,
4386 u64 physical_for_dev_replace, struct page *page)
4389 struct btrfs_device *dev;
4392 dev = sctx->wr_ctx.tgtdev;
4396 btrfs_warn_rl(dev->dev_root->fs_info,
4397 "scrub write_page_nocow(bdev == NULL) is unexpected");
4400 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
4402 spin_lock(&sctx->stat_lock);
4403 sctx->stat.malloc_errors++;
4404 spin_unlock(&sctx->stat_lock);
4407 bio->bi_iter.bi_size = 0;
4408 bio->bi_iter.bi_sector = physical_for_dev_replace >> 9;
4409 bio->bi_bdev = dev->bdev;
4410 ret = bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
4411 if (ret != PAGE_CACHE_SIZE) {
4414 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
4418 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio))
4419 goto leave_with_eio;