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;
74 struct scrub_block *sblock;
76 struct btrfs_device *dev;
77 struct list_head list;
78 u64 flags; /* extent flags */
82 u64 physical_for_dev_replace;
85 unsigned int mirror_num:8;
86 unsigned int have_csum:1;
87 unsigned int io_error:1;
89 u8 csum[BTRFS_CSUM_SIZE];
91 struct scrub_recover *recover;
96 struct scrub_ctx *sctx;
97 struct btrfs_device *dev;
102 #if SCRUB_PAGES_PER_WR_BIO >= SCRUB_PAGES_PER_RD_BIO
103 struct scrub_page *pagev[SCRUB_PAGES_PER_WR_BIO];
105 struct scrub_page *pagev[SCRUB_PAGES_PER_RD_BIO];
109 struct btrfs_work work;
113 struct scrub_page *pagev[SCRUB_MAX_PAGES_PER_BLOCK];
115 atomic_t outstanding_pages;
116 atomic_t ref_count; /* free mem on transition to zero */
117 struct scrub_ctx *sctx;
118 struct scrub_parity *sparity;
120 unsigned int header_error:1;
121 unsigned int checksum_error:1;
122 unsigned int no_io_error_seen:1;
123 unsigned int generation_error:1; /* also sets header_error */
125 /* The following is for the data used to check parity */
126 /* It is for the data with checksum */
127 unsigned int data_corrected:1;
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 struct scrub_fixup_nodatasum {
200 struct scrub_ctx *sctx;
201 struct btrfs_device *dev;
203 struct btrfs_root *root;
204 struct btrfs_work work;
208 struct scrub_nocow_inode {
212 struct list_head list;
215 struct scrub_copy_nocow_ctx {
216 struct scrub_ctx *sctx;
220 u64 physical_for_dev_replace;
221 struct list_head inodes;
222 struct btrfs_work work;
225 struct scrub_warning {
226 struct btrfs_path *path;
227 u64 extent_item_size;
231 struct btrfs_device *dev;
234 static void scrub_pending_bio_inc(struct scrub_ctx *sctx);
235 static void scrub_pending_bio_dec(struct scrub_ctx *sctx);
236 static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx);
237 static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx);
238 static int scrub_handle_errored_block(struct scrub_block *sblock_to_check);
239 static int scrub_setup_recheck_block(struct scrub_ctx *sctx,
240 struct btrfs_fs_info *fs_info,
241 struct scrub_block *original_sblock,
242 u64 length, u64 logical,
243 struct scrub_block *sblocks_for_recheck);
244 static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
245 struct scrub_block *sblock, int is_metadata,
246 int have_csum, u8 *csum, u64 generation,
247 u16 csum_size, int retry_failed_mirror);
248 static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
249 struct scrub_block *sblock,
250 int is_metadata, int have_csum,
251 const u8 *csum, u64 generation,
253 static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
254 struct scrub_block *sblock_good,
256 static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
257 struct scrub_block *sblock_good,
258 int page_num, int force_write);
259 static void scrub_write_block_to_dev_replace(struct scrub_block *sblock);
260 static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
262 static int scrub_checksum_data(struct scrub_block *sblock);
263 static int scrub_checksum_tree_block(struct scrub_block *sblock);
264 static int scrub_checksum_super(struct scrub_block *sblock);
265 static void scrub_block_get(struct scrub_block *sblock);
266 static void scrub_block_put(struct scrub_block *sblock);
267 static void scrub_page_get(struct scrub_page *spage);
268 static void scrub_page_put(struct scrub_page *spage);
269 static void scrub_parity_get(struct scrub_parity *sparity);
270 static void scrub_parity_put(struct scrub_parity *sparity);
271 static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
272 struct scrub_page *spage);
273 static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
274 u64 physical, struct btrfs_device *dev, u64 flags,
275 u64 gen, int mirror_num, u8 *csum, int force,
276 u64 physical_for_dev_replace);
277 static void scrub_bio_end_io(struct bio *bio, int err);
278 static void scrub_bio_end_io_worker(struct btrfs_work *work);
279 static void scrub_block_complete(struct scrub_block *sblock);
280 static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
281 u64 extent_logical, u64 extent_len,
282 u64 *extent_physical,
283 struct btrfs_device **extent_dev,
284 int *extent_mirror_num);
285 static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
286 struct scrub_wr_ctx *wr_ctx,
287 struct btrfs_fs_info *fs_info,
288 struct btrfs_device *dev,
290 static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx);
291 static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
292 struct scrub_page *spage);
293 static void scrub_wr_submit(struct scrub_ctx *sctx);
294 static void scrub_wr_bio_end_io(struct bio *bio, int err);
295 static void scrub_wr_bio_end_io_worker(struct btrfs_work *work);
296 static int write_page_nocow(struct scrub_ctx *sctx,
297 u64 physical_for_dev_replace, struct page *page);
298 static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
299 struct scrub_copy_nocow_ctx *ctx);
300 static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
301 int mirror_num, u64 physical_for_dev_replace);
302 static void copy_nocow_pages_worker(struct btrfs_work *work);
303 static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
304 static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
307 static void scrub_pending_bio_inc(struct scrub_ctx *sctx)
309 atomic_inc(&sctx->bios_in_flight);
312 static void scrub_pending_bio_dec(struct scrub_ctx *sctx)
314 atomic_dec(&sctx->bios_in_flight);
315 wake_up(&sctx->list_wait);
318 static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
320 while (atomic_read(&fs_info->scrub_pause_req)) {
321 mutex_unlock(&fs_info->scrub_lock);
322 wait_event(fs_info->scrub_pause_wait,
323 atomic_read(&fs_info->scrub_pause_req) == 0);
324 mutex_lock(&fs_info->scrub_lock);
328 static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
330 atomic_inc(&fs_info->scrubs_paused);
331 wake_up(&fs_info->scrub_pause_wait);
333 mutex_lock(&fs_info->scrub_lock);
334 __scrub_blocked_if_needed(fs_info);
335 atomic_dec(&fs_info->scrubs_paused);
336 mutex_unlock(&fs_info->scrub_lock);
338 wake_up(&fs_info->scrub_pause_wait);
342 * used for workers that require transaction commits (i.e., for the
345 static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx)
347 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
350 * increment scrubs_running to prevent cancel requests from
351 * completing as long as a worker is running. we must also
352 * increment scrubs_paused to prevent deadlocking on pause
353 * requests used for transactions commits (as the worker uses a
354 * transaction context). it is safe to regard the worker
355 * as paused for all matters practical. effectively, we only
356 * avoid cancellation requests from completing.
358 mutex_lock(&fs_info->scrub_lock);
359 atomic_inc(&fs_info->scrubs_running);
360 atomic_inc(&fs_info->scrubs_paused);
361 mutex_unlock(&fs_info->scrub_lock);
364 * check if @scrubs_running=@scrubs_paused condition
365 * inside wait_event() is not an atomic operation.
366 * which means we may inc/dec @scrub_running/paused
367 * at any time. Let's wake up @scrub_pause_wait as
368 * much as we can to let commit transaction blocked less.
370 wake_up(&fs_info->scrub_pause_wait);
372 atomic_inc(&sctx->workers_pending);
375 /* used for workers that require transaction commits */
376 static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx)
378 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
381 * see scrub_pending_trans_workers_inc() why we're pretending
382 * to be paused in the scrub counters
384 mutex_lock(&fs_info->scrub_lock);
385 atomic_dec(&fs_info->scrubs_running);
386 atomic_dec(&fs_info->scrubs_paused);
387 mutex_unlock(&fs_info->scrub_lock);
388 atomic_dec(&sctx->workers_pending);
389 wake_up(&fs_info->scrub_pause_wait);
390 wake_up(&sctx->list_wait);
393 static void scrub_free_csums(struct scrub_ctx *sctx)
395 while (!list_empty(&sctx->csum_list)) {
396 struct btrfs_ordered_sum *sum;
397 sum = list_first_entry(&sctx->csum_list,
398 struct btrfs_ordered_sum, list);
399 list_del(&sum->list);
404 static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx)
411 scrub_free_wr_ctx(&sctx->wr_ctx);
413 /* this can happen when scrub is cancelled */
414 if (sctx->curr != -1) {
415 struct scrub_bio *sbio = sctx->bios[sctx->curr];
417 for (i = 0; i < sbio->page_count; i++) {
418 WARN_ON(!sbio->pagev[i]->page);
419 scrub_block_put(sbio->pagev[i]->sblock);
424 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
425 struct scrub_bio *sbio = sctx->bios[i];
432 scrub_free_csums(sctx);
436 static noinline_for_stack
437 struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace)
439 struct scrub_ctx *sctx;
441 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
442 int pages_per_rd_bio;
446 * the setting of pages_per_rd_bio is correct for scrub but might
447 * be wrong for the dev_replace code where we might read from
448 * different devices in the initial huge bios. However, that
449 * code is able to correctly handle the case when adding a page
453 pages_per_rd_bio = min_t(int, SCRUB_PAGES_PER_RD_BIO,
454 bio_get_nr_vecs(dev->bdev));
456 pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO;
457 sctx = kzalloc(sizeof(*sctx), GFP_NOFS);
460 sctx->is_dev_replace = is_dev_replace;
461 sctx->pages_per_rd_bio = pages_per_rd_bio;
463 sctx->dev_root = dev->dev_root;
464 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
465 struct scrub_bio *sbio;
467 sbio = kzalloc(sizeof(*sbio), GFP_NOFS);
470 sctx->bios[i] = sbio;
474 sbio->page_count = 0;
475 btrfs_init_work(&sbio->work, btrfs_scrub_helper,
476 scrub_bio_end_io_worker, NULL, NULL);
478 if (i != SCRUB_BIOS_PER_SCTX - 1)
479 sctx->bios[i]->next_free = i + 1;
481 sctx->bios[i]->next_free = -1;
483 sctx->first_free = 0;
484 sctx->nodesize = dev->dev_root->nodesize;
485 sctx->sectorsize = dev->dev_root->sectorsize;
486 atomic_set(&sctx->bios_in_flight, 0);
487 atomic_set(&sctx->workers_pending, 0);
488 atomic_set(&sctx->cancel_req, 0);
489 sctx->csum_size = btrfs_super_csum_size(fs_info->super_copy);
490 INIT_LIST_HEAD(&sctx->csum_list);
492 spin_lock_init(&sctx->list_lock);
493 spin_lock_init(&sctx->stat_lock);
494 init_waitqueue_head(&sctx->list_wait);
496 ret = scrub_setup_wr_ctx(sctx, &sctx->wr_ctx, fs_info,
497 fs_info->dev_replace.tgtdev, is_dev_replace);
499 scrub_free_ctx(sctx);
505 scrub_free_ctx(sctx);
506 return ERR_PTR(-ENOMEM);
509 static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root,
516 struct extent_buffer *eb;
517 struct btrfs_inode_item *inode_item;
518 struct scrub_warning *swarn = warn_ctx;
519 struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info;
520 struct inode_fs_paths *ipath = NULL;
521 struct btrfs_root *local_root;
522 struct btrfs_key root_key;
523 struct btrfs_key key;
525 root_key.objectid = root;
526 root_key.type = BTRFS_ROOT_ITEM_KEY;
527 root_key.offset = (u64)-1;
528 local_root = btrfs_read_fs_root_no_name(fs_info, &root_key);
529 if (IS_ERR(local_root)) {
530 ret = PTR_ERR(local_root);
535 * this makes the path point to (inum INODE_ITEM ioff)
538 key.type = BTRFS_INODE_ITEM_KEY;
541 ret = btrfs_search_slot(NULL, local_root, &key, swarn->path, 0, 0);
543 btrfs_release_path(swarn->path);
547 eb = swarn->path->nodes[0];
548 inode_item = btrfs_item_ptr(eb, swarn->path->slots[0],
549 struct btrfs_inode_item);
550 isize = btrfs_inode_size(eb, inode_item);
551 nlink = btrfs_inode_nlink(eb, inode_item);
552 btrfs_release_path(swarn->path);
554 ipath = init_ipath(4096, local_root, swarn->path);
556 ret = PTR_ERR(ipath);
560 ret = paths_from_inode(inum, ipath);
566 * we deliberately ignore the bit ipath might have been too small to
567 * hold all of the paths here
569 for (i = 0; i < ipath->fspath->elem_cnt; ++i)
570 printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
571 "%s, sector %llu, root %llu, inode %llu, offset %llu, "
572 "length %llu, links %u (path: %s)\n", swarn->errstr,
573 swarn->logical, rcu_str_deref(swarn->dev->name),
574 (unsigned long long)swarn->sector, root, inum, offset,
575 min(isize - offset, (u64)PAGE_SIZE), nlink,
576 (char *)(unsigned long)ipath->fspath->val[i]);
582 printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
583 "%s, sector %llu, root %llu, inode %llu, offset %llu: path "
584 "resolving failed with ret=%d\n", swarn->errstr,
585 swarn->logical, rcu_str_deref(swarn->dev->name),
586 (unsigned long long)swarn->sector, root, inum, offset, ret);
592 static void scrub_print_warning(const char *errstr, struct scrub_block *sblock)
594 struct btrfs_device *dev;
595 struct btrfs_fs_info *fs_info;
596 struct btrfs_path *path;
597 struct btrfs_key found_key;
598 struct extent_buffer *eb;
599 struct btrfs_extent_item *ei;
600 struct scrub_warning swarn;
601 unsigned long ptr = 0;
609 WARN_ON(sblock->page_count < 1);
610 dev = sblock->pagev[0]->dev;
611 fs_info = sblock->sctx->dev_root->fs_info;
613 path = btrfs_alloc_path();
617 swarn.sector = (sblock->pagev[0]->physical) >> 9;
618 swarn.logical = sblock->pagev[0]->logical;
619 swarn.errstr = errstr;
622 ret = extent_from_logical(fs_info, swarn.logical, path, &found_key,
627 extent_item_pos = swarn.logical - found_key.objectid;
628 swarn.extent_item_size = found_key.offset;
631 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
632 item_size = btrfs_item_size_nr(eb, path->slots[0]);
634 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
636 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
637 item_size, &ref_root,
639 printk_in_rcu(KERN_WARNING
640 "BTRFS: %s at logical %llu on dev %s, "
641 "sector %llu: metadata %s (level %d) in tree "
642 "%llu\n", errstr, swarn.logical,
643 rcu_str_deref(dev->name),
644 (unsigned long long)swarn.sector,
645 ref_level ? "node" : "leaf",
646 ret < 0 ? -1 : ref_level,
647 ret < 0 ? -1 : ref_root);
649 btrfs_release_path(path);
651 btrfs_release_path(path);
654 iterate_extent_inodes(fs_info, found_key.objectid,
656 scrub_print_warning_inode, &swarn);
660 btrfs_free_path(path);
663 static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx)
665 struct page *page = NULL;
667 struct scrub_fixup_nodatasum *fixup = fixup_ctx;
670 struct btrfs_key key;
671 struct inode *inode = NULL;
672 struct btrfs_fs_info *fs_info;
673 u64 end = offset + PAGE_SIZE - 1;
674 struct btrfs_root *local_root;
678 key.type = BTRFS_ROOT_ITEM_KEY;
679 key.offset = (u64)-1;
681 fs_info = fixup->root->fs_info;
682 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
684 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
685 if (IS_ERR(local_root)) {
686 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
687 return PTR_ERR(local_root);
690 key.type = BTRFS_INODE_ITEM_KEY;
693 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
694 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
696 return PTR_ERR(inode);
698 index = offset >> PAGE_CACHE_SHIFT;
700 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
706 if (PageUptodate(page)) {
707 if (PageDirty(page)) {
709 * we need to write the data to the defect sector. the
710 * data that was in that sector is not in memory,
711 * because the page was modified. we must not write the
712 * modified page to that sector.
714 * TODO: what could be done here: wait for the delalloc
715 * runner to write out that page (might involve
716 * COW) and see whether the sector is still
717 * referenced afterwards.
719 * For the meantime, we'll treat this error
720 * incorrectable, although there is a chance that a
721 * later scrub will find the bad sector again and that
722 * there's no dirty page in memory, then.
727 ret = repair_io_failure(inode, offset, PAGE_SIZE,
728 fixup->logical, page,
729 offset - page_offset(page),
735 * we need to get good data first. the general readpage path
736 * will call repair_io_failure for us, we just have to make
737 * sure we read the bad mirror.
739 ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
740 EXTENT_DAMAGED, GFP_NOFS);
742 /* set_extent_bits should give proper error */
749 ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page,
752 wait_on_page_locked(page);
754 corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset,
755 end, EXTENT_DAMAGED, 0, NULL);
757 clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
758 EXTENT_DAMAGED, GFP_NOFS);
770 if (ret == 0 && corrected) {
772 * we only need to call readpage for one of the inodes belonging
773 * to this extent. so make iterate_extent_inodes stop
781 static void scrub_fixup_nodatasum(struct btrfs_work *work)
784 struct scrub_fixup_nodatasum *fixup;
785 struct scrub_ctx *sctx;
786 struct btrfs_trans_handle *trans = NULL;
787 struct btrfs_path *path;
788 int uncorrectable = 0;
790 fixup = container_of(work, struct scrub_fixup_nodatasum, work);
793 path = btrfs_alloc_path();
795 spin_lock(&sctx->stat_lock);
796 ++sctx->stat.malloc_errors;
797 spin_unlock(&sctx->stat_lock);
802 trans = btrfs_join_transaction(fixup->root);
809 * the idea is to trigger a regular read through the standard path. we
810 * read a page from the (failed) logical address by specifying the
811 * corresponding copynum of the failed sector. thus, that readpage is
813 * that is the point where on-the-fly error correction will kick in
814 * (once it's finished) and rewrite the failed sector if a good copy
817 ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info,
818 path, scrub_fixup_readpage,
826 spin_lock(&sctx->stat_lock);
827 ++sctx->stat.corrected_errors;
828 spin_unlock(&sctx->stat_lock);
831 if (trans && !IS_ERR(trans))
832 btrfs_end_transaction(trans, fixup->root);
834 spin_lock(&sctx->stat_lock);
835 ++sctx->stat.uncorrectable_errors;
836 spin_unlock(&sctx->stat_lock);
837 btrfs_dev_replace_stats_inc(
838 &sctx->dev_root->fs_info->dev_replace.
839 num_uncorrectable_read_errors);
840 printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
841 "unable to fixup (nodatasum) error at logical %llu on dev %s\n",
842 fixup->logical, rcu_str_deref(fixup->dev->name));
845 btrfs_free_path(path);
848 scrub_pending_trans_workers_dec(sctx);
851 static inline void scrub_get_recover(struct scrub_recover *recover)
853 atomic_inc(&recover->refs);
856 static inline void scrub_put_recover(struct scrub_recover *recover)
858 if (atomic_dec_and_test(&recover->refs)) {
859 kfree(recover->bbio);
860 kfree(recover->raid_map);
866 * scrub_handle_errored_block gets called when either verification of the
867 * pages failed or the bio failed to read, e.g. with EIO. In the latter
868 * case, this function handles all pages in the bio, even though only one
870 * The goal of this function is to repair the errored block by using the
871 * contents of one of the mirrors.
873 static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
875 struct scrub_ctx *sctx = sblock_to_check->sctx;
876 struct btrfs_device *dev;
877 struct btrfs_fs_info *fs_info;
881 unsigned int failed_mirror_index;
882 unsigned int is_metadata;
883 unsigned int have_csum;
885 struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */
886 struct scrub_block *sblock_bad;
891 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
892 DEFAULT_RATELIMIT_BURST);
894 BUG_ON(sblock_to_check->page_count < 1);
895 fs_info = sctx->dev_root->fs_info;
896 if (sblock_to_check->pagev[0]->flags & BTRFS_EXTENT_FLAG_SUPER) {
898 * if we find an error in a super block, we just report it.
899 * They will get written with the next transaction commit
902 spin_lock(&sctx->stat_lock);
903 ++sctx->stat.super_errors;
904 spin_unlock(&sctx->stat_lock);
907 length = sblock_to_check->page_count * PAGE_SIZE;
908 logical = sblock_to_check->pagev[0]->logical;
909 generation = sblock_to_check->pagev[0]->generation;
910 BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1);
911 failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1;
912 is_metadata = !(sblock_to_check->pagev[0]->flags &
913 BTRFS_EXTENT_FLAG_DATA);
914 have_csum = sblock_to_check->pagev[0]->have_csum;
915 csum = sblock_to_check->pagev[0]->csum;
916 dev = sblock_to_check->pagev[0]->dev;
918 if (sctx->is_dev_replace && !is_metadata && !have_csum) {
919 sblocks_for_recheck = NULL;
924 * read all mirrors one after the other. This includes to
925 * re-read the extent or metadata block that failed (that was
926 * the cause that this fixup code is called) another time,
927 * page by page this time in order to know which pages
928 * caused I/O errors and which ones are good (for all mirrors).
929 * It is the goal to handle the situation when more than one
930 * mirror contains I/O errors, but the errors do not
931 * overlap, i.e. the data can be repaired by selecting the
932 * pages from those mirrors without I/O error on the
933 * particular pages. One example (with blocks >= 2 * PAGE_SIZE)
934 * would be that mirror #1 has an I/O error on the first page,
935 * the second page is good, and mirror #2 has an I/O error on
936 * the second page, but the first page is good.
937 * Then the first page of the first mirror can be repaired by
938 * taking the first page of the second mirror, and the
939 * second page of the second mirror can be repaired by
940 * copying the contents of the 2nd page of the 1st mirror.
941 * One more note: if the pages of one mirror contain I/O
942 * errors, the checksum cannot be verified. In order to get
943 * the best data for repairing, the first attempt is to find
944 * a mirror without I/O errors and with a validated checksum.
945 * Only if this is not possible, the pages are picked from
946 * mirrors with I/O errors without considering the checksum.
947 * If the latter is the case, at the end, the checksum of the
948 * repaired area is verified in order to correctly maintain
952 sblocks_for_recheck = kzalloc(BTRFS_MAX_MIRRORS *
953 sizeof(*sblocks_for_recheck),
955 if (!sblocks_for_recheck) {
956 spin_lock(&sctx->stat_lock);
957 sctx->stat.malloc_errors++;
958 sctx->stat.read_errors++;
959 sctx->stat.uncorrectable_errors++;
960 spin_unlock(&sctx->stat_lock);
961 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
965 /* setup the context, map the logical blocks and alloc the pages */
966 ret = scrub_setup_recheck_block(sctx, fs_info, sblock_to_check, length,
967 logical, sblocks_for_recheck);
969 spin_lock(&sctx->stat_lock);
970 sctx->stat.read_errors++;
971 sctx->stat.uncorrectable_errors++;
972 spin_unlock(&sctx->stat_lock);
973 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
976 BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS);
977 sblock_bad = sblocks_for_recheck + failed_mirror_index;
979 /* build and submit the bios for the failed mirror, check checksums */
980 scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum,
981 csum, generation, sctx->csum_size, 1);
983 if (!sblock_bad->header_error && !sblock_bad->checksum_error &&
984 sblock_bad->no_io_error_seen) {
986 * the error disappeared after reading page by page, or
987 * the area was part of a huge bio and other parts of the
988 * bio caused I/O errors, or the block layer merged several
989 * read requests into one and the error is caused by a
990 * different bio (usually one of the two latter cases is
993 spin_lock(&sctx->stat_lock);
994 sctx->stat.unverified_errors++;
995 sblock_to_check->data_corrected = 1;
996 spin_unlock(&sctx->stat_lock);
998 if (sctx->is_dev_replace)
999 scrub_write_block_to_dev_replace(sblock_bad);
1003 if (!sblock_bad->no_io_error_seen) {
1004 spin_lock(&sctx->stat_lock);
1005 sctx->stat.read_errors++;
1006 spin_unlock(&sctx->stat_lock);
1007 if (__ratelimit(&_rs))
1008 scrub_print_warning("i/o error", sblock_to_check);
1009 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
1010 } else if (sblock_bad->checksum_error) {
1011 spin_lock(&sctx->stat_lock);
1012 sctx->stat.csum_errors++;
1013 spin_unlock(&sctx->stat_lock);
1014 if (__ratelimit(&_rs))
1015 scrub_print_warning("checksum error", sblock_to_check);
1016 btrfs_dev_stat_inc_and_print(dev,
1017 BTRFS_DEV_STAT_CORRUPTION_ERRS);
1018 } else if (sblock_bad->header_error) {
1019 spin_lock(&sctx->stat_lock);
1020 sctx->stat.verify_errors++;
1021 spin_unlock(&sctx->stat_lock);
1022 if (__ratelimit(&_rs))
1023 scrub_print_warning("checksum/header error",
1025 if (sblock_bad->generation_error)
1026 btrfs_dev_stat_inc_and_print(dev,
1027 BTRFS_DEV_STAT_GENERATION_ERRS);
1029 btrfs_dev_stat_inc_and_print(dev,
1030 BTRFS_DEV_STAT_CORRUPTION_ERRS);
1033 if (sctx->readonly) {
1034 ASSERT(!sctx->is_dev_replace);
1038 if (!is_metadata && !have_csum) {
1039 struct scrub_fixup_nodatasum *fixup_nodatasum;
1042 WARN_ON(sctx->is_dev_replace);
1045 * !is_metadata and !have_csum, this means that the data
1046 * might not be COW'ed, that it might be modified
1047 * concurrently. The general strategy to work on the
1048 * commit root does not help in the case when COW is not
1051 fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS);
1052 if (!fixup_nodatasum)
1053 goto did_not_correct_error;
1054 fixup_nodatasum->sctx = sctx;
1055 fixup_nodatasum->dev = dev;
1056 fixup_nodatasum->logical = logical;
1057 fixup_nodatasum->root = fs_info->extent_root;
1058 fixup_nodatasum->mirror_num = failed_mirror_index + 1;
1059 scrub_pending_trans_workers_inc(sctx);
1060 btrfs_init_work(&fixup_nodatasum->work, btrfs_scrub_helper,
1061 scrub_fixup_nodatasum, NULL, NULL);
1062 btrfs_queue_work(fs_info->scrub_workers,
1063 &fixup_nodatasum->work);
1068 * now build and submit the bios for the other mirrors, check
1070 * First try to pick the mirror which is completely without I/O
1071 * errors and also does not have a checksum error.
1072 * If one is found, and if a checksum is present, the full block
1073 * that is known to contain an error is rewritten. Afterwards
1074 * the block is known to be corrected.
1075 * If a mirror is found which is completely correct, and no
1076 * checksum is present, only those pages are rewritten that had
1077 * an I/O error in the block to be repaired, since it cannot be
1078 * determined, which copy of the other pages is better (and it
1079 * could happen otherwise that a correct page would be
1080 * overwritten by a bad one).
1082 for (mirror_index = 0;
1083 mirror_index < BTRFS_MAX_MIRRORS &&
1084 sblocks_for_recheck[mirror_index].page_count > 0;
1086 struct scrub_block *sblock_other;
1088 if (mirror_index == failed_mirror_index)
1090 sblock_other = sblocks_for_recheck + mirror_index;
1092 /* build and submit the bios, check checksums */
1093 scrub_recheck_block(fs_info, sblock_other, is_metadata,
1094 have_csum, csum, generation,
1095 sctx->csum_size, 0);
1097 if (!sblock_other->header_error &&
1098 !sblock_other->checksum_error &&
1099 sblock_other->no_io_error_seen) {
1100 if (sctx->is_dev_replace) {
1101 scrub_write_block_to_dev_replace(sblock_other);
1103 int force_write = is_metadata || have_csum;
1105 ret = scrub_repair_block_from_good_copy(
1106 sblock_bad, sblock_other,
1110 goto corrected_error;
1115 * for dev_replace, pick good pages and write to the target device.
1117 if (sctx->is_dev_replace) {
1119 for (page_num = 0; page_num < sblock_bad->page_count;
1124 for (mirror_index = 0;
1125 mirror_index < BTRFS_MAX_MIRRORS &&
1126 sblocks_for_recheck[mirror_index].page_count > 0;
1128 struct scrub_block *sblock_other =
1129 sblocks_for_recheck + mirror_index;
1130 struct scrub_page *page_other =
1131 sblock_other->pagev[page_num];
1133 if (!page_other->io_error) {
1134 ret = scrub_write_page_to_dev_replace(
1135 sblock_other, page_num);
1137 /* succeeded for this page */
1141 btrfs_dev_replace_stats_inc(
1143 fs_info->dev_replace.
1151 * did not find a mirror to fetch the page
1152 * from. scrub_write_page_to_dev_replace()
1153 * handles this case (page->io_error), by
1154 * filling the block with zeros before
1155 * submitting the write request
1158 ret = scrub_write_page_to_dev_replace(
1159 sblock_bad, page_num);
1161 btrfs_dev_replace_stats_inc(
1162 &sctx->dev_root->fs_info->
1163 dev_replace.num_write_errors);
1171 * for regular scrub, repair those pages that are errored.
1172 * In case of I/O errors in the area that is supposed to be
1173 * repaired, continue by picking good copies of those pages.
1174 * Select the good pages from mirrors to rewrite bad pages from
1175 * the area to fix. Afterwards verify the checksum of the block
1176 * that is supposed to be repaired. This verification step is
1177 * only done for the purpose of statistic counting and for the
1178 * final scrub report, whether errors remain.
1179 * A perfect algorithm could make use of the checksum and try
1180 * all possible combinations of pages from the different mirrors
1181 * until the checksum verification succeeds. For example, when
1182 * the 2nd page of mirror #1 faces I/O errors, and the 2nd page
1183 * of mirror #2 is readable but the final checksum test fails,
1184 * then the 2nd page of mirror #3 could be tried, whether now
1185 * the final checksum succeedes. But this would be a rare
1186 * exception and is therefore not implemented. At least it is
1187 * avoided that the good copy is overwritten.
1188 * A more useful improvement would be to pick the sectors
1189 * without I/O error based on sector sizes (512 bytes on legacy
1190 * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one
1191 * mirror could be repaired by taking 512 byte of a different
1192 * mirror, even if other 512 byte sectors in the same PAGE_SIZE
1193 * area are unreadable.
1196 /* can only fix I/O errors from here on */
1197 if (sblock_bad->no_io_error_seen)
1198 goto did_not_correct_error;
1201 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
1202 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1204 if (!page_bad->io_error)
1207 for (mirror_index = 0;
1208 mirror_index < BTRFS_MAX_MIRRORS &&
1209 sblocks_for_recheck[mirror_index].page_count > 0;
1211 struct scrub_block *sblock_other = sblocks_for_recheck +
1213 struct scrub_page *page_other = sblock_other->pagev[
1216 if (!page_other->io_error) {
1217 ret = scrub_repair_page_from_good_copy(
1218 sblock_bad, sblock_other, page_num, 0);
1220 page_bad->io_error = 0;
1221 break; /* succeeded for this page */
1226 if (page_bad->io_error) {
1227 /* did not find a mirror to copy the page from */
1233 if (is_metadata || have_csum) {
1235 * need to verify the checksum now that all
1236 * sectors on disk are repaired (the write
1237 * request for data to be repaired is on its way).
1238 * Just be lazy and use scrub_recheck_block()
1239 * which re-reads the data before the checksum
1240 * is verified, but most likely the data comes out
1241 * of the page cache.
1243 scrub_recheck_block(fs_info, sblock_bad,
1244 is_metadata, have_csum, csum,
1245 generation, sctx->csum_size, 1);
1246 if (!sblock_bad->header_error &&
1247 !sblock_bad->checksum_error &&
1248 sblock_bad->no_io_error_seen)
1249 goto corrected_error;
1251 goto did_not_correct_error;
1254 spin_lock(&sctx->stat_lock);
1255 sctx->stat.corrected_errors++;
1256 sblock_to_check->data_corrected = 1;
1257 spin_unlock(&sctx->stat_lock);
1258 printk_ratelimited_in_rcu(KERN_ERR
1259 "BTRFS: fixed up error at logical %llu on dev %s\n",
1260 logical, rcu_str_deref(dev->name));
1263 did_not_correct_error:
1264 spin_lock(&sctx->stat_lock);
1265 sctx->stat.uncorrectable_errors++;
1266 spin_unlock(&sctx->stat_lock);
1267 printk_ratelimited_in_rcu(KERN_ERR
1268 "BTRFS: unable to fixup (regular) error at logical %llu on dev %s\n",
1269 logical, rcu_str_deref(dev->name));
1273 if (sblocks_for_recheck) {
1274 for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS;
1276 struct scrub_block *sblock = sblocks_for_recheck +
1278 struct scrub_recover *recover;
1281 for (page_index = 0; page_index < sblock->page_count;
1283 sblock->pagev[page_index]->sblock = NULL;
1284 recover = sblock->pagev[page_index]->recover;
1286 scrub_put_recover(recover);
1287 sblock->pagev[page_index]->recover =
1290 scrub_page_put(sblock->pagev[page_index]);
1293 kfree(sblocks_for_recheck);
1299 static inline int scrub_nr_raid_mirrors(struct btrfs_bio *bbio, u64 *raid_map)
1302 if (raid_map[bbio->num_stripes - 1] == RAID6_Q_STRIPE)
1307 return (int)bbio->num_stripes;
1311 static inline void scrub_stripe_index_and_offset(u64 logical, u64 *raid_map,
1313 int nstripes, int mirror,
1321 for (i = 0; i < nstripes; i++) {
1322 if (raid_map[i] == RAID6_Q_STRIPE ||
1323 raid_map[i] == RAID5_P_STRIPE)
1326 if (logical >= raid_map[i] &&
1327 logical < raid_map[i] + mapped_length)
1332 *stripe_offset = logical - raid_map[i];
1334 /* The other RAID type */
1335 *stripe_index = mirror;
1340 static int scrub_setup_recheck_block(struct scrub_ctx *sctx,
1341 struct btrfs_fs_info *fs_info,
1342 struct scrub_block *original_sblock,
1343 u64 length, u64 logical,
1344 struct scrub_block *sblocks_for_recheck)
1346 struct scrub_recover *recover;
1347 struct btrfs_bio *bbio;
1359 * note: the two members ref_count and outstanding_pages
1360 * are not used (and not set) in the blocks that are used for
1361 * the recheck procedure
1365 while (length > 0) {
1366 sublen = min_t(u64, length, PAGE_SIZE);
1367 mapped_length = sublen;
1372 * with a length of PAGE_SIZE, each returned stripe
1373 * represents one mirror
1375 ret = btrfs_map_sblock(fs_info, REQ_GET_READ_MIRRORS, logical,
1376 &mapped_length, &bbio, 0, &raid_map);
1377 if (ret || !bbio || mapped_length < sublen) {
1383 recover = kzalloc(sizeof(struct scrub_recover), GFP_NOFS);
1390 atomic_set(&recover->refs, 1);
1391 recover->bbio = bbio;
1392 recover->raid_map = raid_map;
1393 recover->map_length = mapped_length;
1395 BUG_ON(page_index >= SCRUB_PAGES_PER_RD_BIO);
1397 nmirrors = scrub_nr_raid_mirrors(bbio, raid_map);
1398 for (mirror_index = 0; mirror_index < nmirrors;
1400 struct scrub_block *sblock;
1401 struct scrub_page *page;
1403 if (mirror_index >= BTRFS_MAX_MIRRORS)
1406 sblock = sblocks_for_recheck + mirror_index;
1407 sblock->sctx = sctx;
1408 page = kzalloc(sizeof(*page), GFP_NOFS);
1411 spin_lock(&sctx->stat_lock);
1412 sctx->stat.malloc_errors++;
1413 spin_unlock(&sctx->stat_lock);
1414 scrub_put_recover(recover);
1417 scrub_page_get(page);
1418 sblock->pagev[page_index] = page;
1419 page->logical = logical;
1421 scrub_stripe_index_and_offset(logical, raid_map,
1427 page->physical = bbio->stripes[stripe_index].physical +
1429 page->dev = bbio->stripes[stripe_index].dev;
1431 BUG_ON(page_index >= original_sblock->page_count);
1432 page->physical_for_dev_replace =
1433 original_sblock->pagev[page_index]->
1434 physical_for_dev_replace;
1435 /* for missing devices, dev->bdev is NULL */
1436 page->mirror_num = mirror_index + 1;
1437 sblock->page_count++;
1438 page->page = alloc_page(GFP_NOFS);
1442 scrub_get_recover(recover);
1443 page->recover = recover;
1445 scrub_put_recover(recover);
1454 struct scrub_bio_ret {
1455 struct completion event;
1459 static void scrub_bio_wait_endio(struct bio *bio, int error)
1461 struct scrub_bio_ret *ret = bio->bi_private;
1464 complete(&ret->event);
1467 static inline int scrub_is_page_on_raid56(struct scrub_page *page)
1469 return page->recover && page->recover->raid_map;
1472 static int scrub_submit_raid56_bio_wait(struct btrfs_fs_info *fs_info,
1474 struct scrub_page *page)
1476 struct scrub_bio_ret done;
1479 init_completion(&done.event);
1481 bio->bi_iter.bi_sector = page->logical >> 9;
1482 bio->bi_private = &done;
1483 bio->bi_end_io = scrub_bio_wait_endio;
1485 ret = raid56_parity_recover(fs_info->fs_root, bio, page->recover->bbio,
1486 page->recover->raid_map,
1487 page->recover->map_length,
1488 page->mirror_num, 0);
1492 wait_for_completion(&done.event);
1500 * this function will check the on disk data for checksum errors, header
1501 * errors and read I/O errors. If any I/O errors happen, the exact pages
1502 * which are errored are marked as being bad. The goal is to enable scrub
1503 * to take those pages that are not errored from all the mirrors so that
1504 * the pages that are errored in the just handled mirror can be repaired.
1506 static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
1507 struct scrub_block *sblock, int is_metadata,
1508 int have_csum, u8 *csum, u64 generation,
1509 u16 csum_size, int retry_failed_mirror)
1513 sblock->no_io_error_seen = 1;
1514 sblock->header_error = 0;
1515 sblock->checksum_error = 0;
1517 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1519 struct scrub_page *page = sblock->pagev[page_num];
1521 if (page->dev->bdev == NULL) {
1523 sblock->no_io_error_seen = 0;
1527 WARN_ON(!page->page);
1528 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1531 sblock->no_io_error_seen = 0;
1534 bio->bi_bdev = page->dev->bdev;
1536 bio_add_page(bio, page->page, PAGE_SIZE, 0);
1537 if (!retry_failed_mirror && scrub_is_page_on_raid56(page)) {
1538 if (scrub_submit_raid56_bio_wait(fs_info, bio, page))
1539 sblock->no_io_error_seen = 0;
1541 bio->bi_iter.bi_sector = page->physical >> 9;
1543 if (btrfsic_submit_bio_wait(READ, bio))
1544 sblock->no_io_error_seen = 0;
1550 if (sblock->no_io_error_seen)
1551 scrub_recheck_block_checksum(fs_info, sblock, is_metadata,
1552 have_csum, csum, generation,
1558 static inline int scrub_check_fsid(u8 fsid[],
1559 struct scrub_page *spage)
1561 struct btrfs_fs_devices *fs_devices = spage->dev->fs_devices;
1564 ret = memcmp(fsid, fs_devices->fsid, BTRFS_UUID_SIZE);
1568 static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
1569 struct scrub_block *sblock,
1570 int is_metadata, int have_csum,
1571 const u8 *csum, u64 generation,
1575 u8 calculated_csum[BTRFS_CSUM_SIZE];
1577 void *mapped_buffer;
1579 WARN_ON(!sblock->pagev[0]->page);
1581 struct btrfs_header *h;
1583 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
1584 h = (struct btrfs_header *)mapped_buffer;
1586 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h) ||
1587 !scrub_check_fsid(h->fsid, sblock->pagev[0]) ||
1588 memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
1590 sblock->header_error = 1;
1591 } else if (generation != btrfs_stack_header_generation(h)) {
1592 sblock->header_error = 1;
1593 sblock->generation_error = 1;
1600 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
1603 for (page_num = 0;;) {
1604 if (page_num == 0 && is_metadata)
1605 crc = btrfs_csum_data(
1606 ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE,
1607 crc, PAGE_SIZE - BTRFS_CSUM_SIZE);
1609 crc = btrfs_csum_data(mapped_buffer, crc, PAGE_SIZE);
1611 kunmap_atomic(mapped_buffer);
1613 if (page_num >= sblock->page_count)
1615 WARN_ON(!sblock->pagev[page_num]->page);
1617 mapped_buffer = kmap_atomic(sblock->pagev[page_num]->page);
1620 btrfs_csum_final(crc, calculated_csum);
1621 if (memcmp(calculated_csum, csum, csum_size))
1622 sblock->checksum_error = 1;
1625 static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
1626 struct scrub_block *sblock_good,
1632 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
1635 ret_sub = scrub_repair_page_from_good_copy(sblock_bad,
1646 static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
1647 struct scrub_block *sblock_good,
1648 int page_num, int force_write)
1650 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1651 struct scrub_page *page_good = sblock_good->pagev[page_num];
1653 BUG_ON(page_bad->page == NULL);
1654 BUG_ON(page_good->page == NULL);
1655 if (force_write || sblock_bad->header_error ||
1656 sblock_bad->checksum_error || page_bad->io_error) {
1660 if (!page_bad->dev->bdev) {
1661 printk_ratelimited(KERN_WARNING "BTRFS: "
1662 "scrub_repair_page_from_good_copy(bdev == NULL) "
1663 "is unexpected!\n");
1667 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1670 bio->bi_bdev = page_bad->dev->bdev;
1671 bio->bi_iter.bi_sector = page_bad->physical >> 9;
1673 ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0);
1674 if (PAGE_SIZE != ret) {
1679 if (btrfsic_submit_bio_wait(WRITE, bio)) {
1680 btrfs_dev_stat_inc_and_print(page_bad->dev,
1681 BTRFS_DEV_STAT_WRITE_ERRS);
1682 btrfs_dev_replace_stats_inc(
1683 &sblock_bad->sctx->dev_root->fs_info->
1684 dev_replace.num_write_errors);
1694 static void scrub_write_block_to_dev_replace(struct scrub_block *sblock)
1699 * This block is used for the check of the parity on the source device,
1700 * so the data needn't be written into the destination device.
1702 if (sblock->sparity)
1705 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1708 ret = scrub_write_page_to_dev_replace(sblock, page_num);
1710 btrfs_dev_replace_stats_inc(
1711 &sblock->sctx->dev_root->fs_info->dev_replace.
1716 static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
1719 struct scrub_page *spage = sblock->pagev[page_num];
1721 BUG_ON(spage->page == NULL);
1722 if (spage->io_error) {
1723 void *mapped_buffer = kmap_atomic(spage->page);
1725 memset(mapped_buffer, 0, PAGE_CACHE_SIZE);
1726 flush_dcache_page(spage->page);
1727 kunmap_atomic(mapped_buffer);
1729 return scrub_add_page_to_wr_bio(sblock->sctx, spage);
1732 static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
1733 struct scrub_page *spage)
1735 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1736 struct scrub_bio *sbio;
1739 mutex_lock(&wr_ctx->wr_lock);
1741 if (!wr_ctx->wr_curr_bio) {
1742 wr_ctx->wr_curr_bio = kzalloc(sizeof(*wr_ctx->wr_curr_bio),
1744 if (!wr_ctx->wr_curr_bio) {
1745 mutex_unlock(&wr_ctx->wr_lock);
1748 wr_ctx->wr_curr_bio->sctx = sctx;
1749 wr_ctx->wr_curr_bio->page_count = 0;
1751 sbio = wr_ctx->wr_curr_bio;
1752 if (sbio->page_count == 0) {
1755 sbio->physical = spage->physical_for_dev_replace;
1756 sbio->logical = spage->logical;
1757 sbio->dev = wr_ctx->tgtdev;
1760 bio = btrfs_io_bio_alloc(GFP_NOFS, wr_ctx->pages_per_wr_bio);
1762 mutex_unlock(&wr_ctx->wr_lock);
1768 bio->bi_private = sbio;
1769 bio->bi_end_io = scrub_wr_bio_end_io;
1770 bio->bi_bdev = sbio->dev->bdev;
1771 bio->bi_iter.bi_sector = sbio->physical >> 9;
1773 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
1774 spage->physical_for_dev_replace ||
1775 sbio->logical + sbio->page_count * PAGE_SIZE !=
1777 scrub_wr_submit(sctx);
1781 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
1782 if (ret != PAGE_SIZE) {
1783 if (sbio->page_count < 1) {
1786 mutex_unlock(&wr_ctx->wr_lock);
1789 scrub_wr_submit(sctx);
1793 sbio->pagev[sbio->page_count] = spage;
1794 scrub_page_get(spage);
1796 if (sbio->page_count == wr_ctx->pages_per_wr_bio)
1797 scrub_wr_submit(sctx);
1798 mutex_unlock(&wr_ctx->wr_lock);
1803 static void scrub_wr_submit(struct scrub_ctx *sctx)
1805 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1806 struct scrub_bio *sbio;
1808 if (!wr_ctx->wr_curr_bio)
1811 sbio = wr_ctx->wr_curr_bio;
1812 wr_ctx->wr_curr_bio = NULL;
1813 WARN_ON(!sbio->bio->bi_bdev);
1814 scrub_pending_bio_inc(sctx);
1815 /* process all writes in a single worker thread. Then the block layer
1816 * orders the requests before sending them to the driver which
1817 * doubled the write performance on spinning disks when measured
1819 btrfsic_submit_bio(WRITE, sbio->bio);
1822 static void scrub_wr_bio_end_io(struct bio *bio, int err)
1824 struct scrub_bio *sbio = bio->bi_private;
1825 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
1830 btrfs_init_work(&sbio->work, btrfs_scrubwrc_helper,
1831 scrub_wr_bio_end_io_worker, NULL, NULL);
1832 btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work);
1835 static void scrub_wr_bio_end_io_worker(struct btrfs_work *work)
1837 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
1838 struct scrub_ctx *sctx = sbio->sctx;
1841 WARN_ON(sbio->page_count > SCRUB_PAGES_PER_WR_BIO);
1843 struct btrfs_dev_replace *dev_replace =
1844 &sbio->sctx->dev_root->fs_info->dev_replace;
1846 for (i = 0; i < sbio->page_count; i++) {
1847 struct scrub_page *spage = sbio->pagev[i];
1849 spage->io_error = 1;
1850 btrfs_dev_replace_stats_inc(&dev_replace->
1855 for (i = 0; i < sbio->page_count; i++)
1856 scrub_page_put(sbio->pagev[i]);
1860 scrub_pending_bio_dec(sctx);
1863 static int scrub_checksum(struct scrub_block *sblock)
1868 WARN_ON(sblock->page_count < 1);
1869 flags = sblock->pagev[0]->flags;
1871 if (flags & BTRFS_EXTENT_FLAG_DATA)
1872 ret = scrub_checksum_data(sblock);
1873 else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1874 ret = scrub_checksum_tree_block(sblock);
1875 else if (flags & BTRFS_EXTENT_FLAG_SUPER)
1876 (void)scrub_checksum_super(sblock);
1880 scrub_handle_errored_block(sblock);
1885 static int scrub_checksum_data(struct scrub_block *sblock)
1887 struct scrub_ctx *sctx = sblock->sctx;
1888 u8 csum[BTRFS_CSUM_SIZE];
1897 BUG_ON(sblock->page_count < 1);
1898 if (!sblock->pagev[0]->have_csum)
1901 on_disk_csum = sblock->pagev[0]->csum;
1902 page = sblock->pagev[0]->page;
1903 buffer = kmap_atomic(page);
1905 len = sctx->sectorsize;
1908 u64 l = min_t(u64, len, PAGE_SIZE);
1910 crc = btrfs_csum_data(buffer, crc, l);
1911 kunmap_atomic(buffer);
1916 BUG_ON(index >= sblock->page_count);
1917 BUG_ON(!sblock->pagev[index]->page);
1918 page = sblock->pagev[index]->page;
1919 buffer = kmap_atomic(page);
1922 btrfs_csum_final(crc, csum);
1923 if (memcmp(csum, on_disk_csum, sctx->csum_size))
1929 static int scrub_checksum_tree_block(struct scrub_block *sblock)
1931 struct scrub_ctx *sctx = sblock->sctx;
1932 struct btrfs_header *h;
1933 struct btrfs_root *root = sctx->dev_root;
1934 struct btrfs_fs_info *fs_info = root->fs_info;
1935 u8 calculated_csum[BTRFS_CSUM_SIZE];
1936 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1938 void *mapped_buffer;
1947 BUG_ON(sblock->page_count < 1);
1948 page = sblock->pagev[0]->page;
1949 mapped_buffer = kmap_atomic(page);
1950 h = (struct btrfs_header *)mapped_buffer;
1951 memcpy(on_disk_csum, h->csum, sctx->csum_size);
1954 * we don't use the getter functions here, as we
1955 * a) don't have an extent buffer and
1956 * b) the page is already kmapped
1959 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h))
1962 if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h))
1965 if (!scrub_check_fsid(h->fsid, sblock->pagev[0]))
1968 if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
1972 len = sctx->nodesize - BTRFS_CSUM_SIZE;
1973 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
1974 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
1977 u64 l = min_t(u64, len, mapped_size);
1979 crc = btrfs_csum_data(p, crc, l);
1980 kunmap_atomic(mapped_buffer);
1985 BUG_ON(index >= sblock->page_count);
1986 BUG_ON(!sblock->pagev[index]->page);
1987 page = sblock->pagev[index]->page;
1988 mapped_buffer = kmap_atomic(page);
1989 mapped_size = PAGE_SIZE;
1993 btrfs_csum_final(crc, calculated_csum);
1994 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
1997 return fail || crc_fail;
2000 static int scrub_checksum_super(struct scrub_block *sblock)
2002 struct btrfs_super_block *s;
2003 struct scrub_ctx *sctx = sblock->sctx;
2004 u8 calculated_csum[BTRFS_CSUM_SIZE];
2005 u8 on_disk_csum[BTRFS_CSUM_SIZE];
2007 void *mapped_buffer;
2016 BUG_ON(sblock->page_count < 1);
2017 page = sblock->pagev[0]->page;
2018 mapped_buffer = kmap_atomic(page);
2019 s = (struct btrfs_super_block *)mapped_buffer;
2020 memcpy(on_disk_csum, s->csum, sctx->csum_size);
2022 if (sblock->pagev[0]->logical != btrfs_super_bytenr(s))
2025 if (sblock->pagev[0]->generation != btrfs_super_generation(s))
2028 if (!scrub_check_fsid(s->fsid, sblock->pagev[0]))
2031 len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE;
2032 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
2033 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
2036 u64 l = min_t(u64, len, mapped_size);
2038 crc = btrfs_csum_data(p, crc, l);
2039 kunmap_atomic(mapped_buffer);
2044 BUG_ON(index >= sblock->page_count);
2045 BUG_ON(!sblock->pagev[index]->page);
2046 page = sblock->pagev[index]->page;
2047 mapped_buffer = kmap_atomic(page);
2048 mapped_size = PAGE_SIZE;
2052 btrfs_csum_final(crc, calculated_csum);
2053 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
2056 if (fail_cor + fail_gen) {
2058 * if we find an error in a super block, we just report it.
2059 * They will get written with the next transaction commit
2062 spin_lock(&sctx->stat_lock);
2063 ++sctx->stat.super_errors;
2064 spin_unlock(&sctx->stat_lock);
2066 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
2067 BTRFS_DEV_STAT_CORRUPTION_ERRS);
2069 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
2070 BTRFS_DEV_STAT_GENERATION_ERRS);
2073 return fail_cor + fail_gen;
2076 static void scrub_block_get(struct scrub_block *sblock)
2078 atomic_inc(&sblock->ref_count);
2081 static void scrub_block_put(struct scrub_block *sblock)
2083 if (atomic_dec_and_test(&sblock->ref_count)) {
2086 if (sblock->sparity)
2087 scrub_parity_put(sblock->sparity);
2089 for (i = 0; i < sblock->page_count; i++)
2090 scrub_page_put(sblock->pagev[i]);
2095 static void scrub_page_get(struct scrub_page *spage)
2097 atomic_inc(&spage->ref_count);
2100 static void scrub_page_put(struct scrub_page *spage)
2102 if (atomic_dec_and_test(&spage->ref_count)) {
2104 __free_page(spage->page);
2109 static void scrub_submit(struct scrub_ctx *sctx)
2111 struct scrub_bio *sbio;
2113 if (sctx->curr == -1)
2116 sbio = sctx->bios[sctx->curr];
2118 scrub_pending_bio_inc(sctx);
2120 if (!sbio->bio->bi_bdev) {
2122 * this case should not happen. If btrfs_map_block() is
2123 * wrong, it could happen for dev-replace operations on
2124 * missing devices when no mirrors are available, but in
2125 * this case it should already fail the mount.
2126 * This case is handled correctly (but _very_ slowly).
2128 printk_ratelimited(KERN_WARNING
2129 "BTRFS: scrub_submit(bio bdev == NULL) is unexpected!\n");
2130 bio_endio(sbio->bio, -EIO);
2132 btrfsic_submit_bio(READ, sbio->bio);
2136 static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
2137 struct scrub_page *spage)
2139 struct scrub_block *sblock = spage->sblock;
2140 struct scrub_bio *sbio;
2145 * grab a fresh bio or wait for one to become available
2147 while (sctx->curr == -1) {
2148 spin_lock(&sctx->list_lock);
2149 sctx->curr = sctx->first_free;
2150 if (sctx->curr != -1) {
2151 sctx->first_free = sctx->bios[sctx->curr]->next_free;
2152 sctx->bios[sctx->curr]->next_free = -1;
2153 sctx->bios[sctx->curr]->page_count = 0;
2154 spin_unlock(&sctx->list_lock);
2156 spin_unlock(&sctx->list_lock);
2157 wait_event(sctx->list_wait, sctx->first_free != -1);
2160 sbio = sctx->bios[sctx->curr];
2161 if (sbio->page_count == 0) {
2164 sbio->physical = spage->physical;
2165 sbio->logical = spage->logical;
2166 sbio->dev = spage->dev;
2169 bio = btrfs_io_bio_alloc(GFP_NOFS, sctx->pages_per_rd_bio);
2175 bio->bi_private = sbio;
2176 bio->bi_end_io = scrub_bio_end_io;
2177 bio->bi_bdev = sbio->dev->bdev;
2178 bio->bi_iter.bi_sector = sbio->physical >> 9;
2180 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
2182 sbio->logical + sbio->page_count * PAGE_SIZE !=
2184 sbio->dev != spage->dev) {
2189 sbio->pagev[sbio->page_count] = spage;
2190 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
2191 if (ret != PAGE_SIZE) {
2192 if (sbio->page_count < 1) {
2201 scrub_block_get(sblock); /* one for the page added to the bio */
2202 atomic_inc(&sblock->outstanding_pages);
2204 if (sbio->page_count == sctx->pages_per_rd_bio)
2210 static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
2211 u64 physical, struct btrfs_device *dev, u64 flags,
2212 u64 gen, int mirror_num, u8 *csum, int force,
2213 u64 physical_for_dev_replace)
2215 struct scrub_block *sblock;
2218 sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
2220 spin_lock(&sctx->stat_lock);
2221 sctx->stat.malloc_errors++;
2222 spin_unlock(&sctx->stat_lock);
2226 /* one ref inside this function, plus one for each page added to
2228 atomic_set(&sblock->ref_count, 1);
2229 sblock->sctx = sctx;
2230 sblock->no_io_error_seen = 1;
2232 for (index = 0; len > 0; index++) {
2233 struct scrub_page *spage;
2234 u64 l = min_t(u64, len, PAGE_SIZE);
2236 spage = kzalloc(sizeof(*spage), GFP_NOFS);
2239 spin_lock(&sctx->stat_lock);
2240 sctx->stat.malloc_errors++;
2241 spin_unlock(&sctx->stat_lock);
2242 scrub_block_put(sblock);
2245 BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
2246 scrub_page_get(spage);
2247 sblock->pagev[index] = spage;
2248 spage->sblock = sblock;
2250 spage->flags = flags;
2251 spage->generation = gen;
2252 spage->logical = logical;
2253 spage->physical = physical;
2254 spage->physical_for_dev_replace = physical_for_dev_replace;
2255 spage->mirror_num = mirror_num;
2257 spage->have_csum = 1;
2258 memcpy(spage->csum, csum, sctx->csum_size);
2260 spage->have_csum = 0;
2262 sblock->page_count++;
2263 spage->page = alloc_page(GFP_NOFS);
2269 physical_for_dev_replace += l;
2272 WARN_ON(sblock->page_count == 0);
2273 for (index = 0; index < sblock->page_count; index++) {
2274 struct scrub_page *spage = sblock->pagev[index];
2277 ret = scrub_add_page_to_rd_bio(sctx, spage);
2279 scrub_block_put(sblock);
2287 /* last one frees, either here or in bio completion for last page */
2288 scrub_block_put(sblock);
2292 static void scrub_bio_end_io(struct bio *bio, int err)
2294 struct scrub_bio *sbio = bio->bi_private;
2295 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
2300 btrfs_queue_work(fs_info->scrub_workers, &sbio->work);
2303 static void scrub_bio_end_io_worker(struct btrfs_work *work)
2305 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
2306 struct scrub_ctx *sctx = sbio->sctx;
2309 BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO);
2311 for (i = 0; i < sbio->page_count; i++) {
2312 struct scrub_page *spage = sbio->pagev[i];
2314 spage->io_error = 1;
2315 spage->sblock->no_io_error_seen = 0;
2319 /* now complete the scrub_block items that have all pages completed */
2320 for (i = 0; i < sbio->page_count; i++) {
2321 struct scrub_page *spage = sbio->pagev[i];
2322 struct scrub_block *sblock = spage->sblock;
2324 if (atomic_dec_and_test(&sblock->outstanding_pages))
2325 scrub_block_complete(sblock);
2326 scrub_block_put(sblock);
2331 spin_lock(&sctx->list_lock);
2332 sbio->next_free = sctx->first_free;
2333 sctx->first_free = sbio->index;
2334 spin_unlock(&sctx->list_lock);
2336 if (sctx->is_dev_replace &&
2337 atomic_read(&sctx->wr_ctx.flush_all_writes)) {
2338 mutex_lock(&sctx->wr_ctx.wr_lock);
2339 scrub_wr_submit(sctx);
2340 mutex_unlock(&sctx->wr_ctx.wr_lock);
2343 scrub_pending_bio_dec(sctx);
2346 static inline void __scrub_mark_bitmap(struct scrub_parity *sparity,
2347 unsigned long *bitmap,
2352 int sectorsize = sparity->sctx->dev_root->sectorsize;
2354 if (len >= sparity->stripe_len) {
2355 bitmap_set(bitmap, 0, sparity->nsectors);
2359 start -= sparity->logic_start;
2360 offset = (int)do_div(start, sparity->stripe_len);
2361 offset /= sectorsize;
2362 nsectors = (int)len / sectorsize;
2364 if (offset + nsectors <= sparity->nsectors) {
2365 bitmap_set(bitmap, offset, nsectors);
2369 bitmap_set(bitmap, offset, sparity->nsectors - offset);
2370 bitmap_set(bitmap, 0, nsectors - (sparity->nsectors - offset));
2373 static inline void scrub_parity_mark_sectors_error(struct scrub_parity *sparity,
2376 __scrub_mark_bitmap(sparity, sparity->ebitmap, start, len);
2379 static inline void scrub_parity_mark_sectors_data(struct scrub_parity *sparity,
2382 __scrub_mark_bitmap(sparity, sparity->dbitmap, start, len);
2385 static void scrub_block_complete(struct scrub_block *sblock)
2389 if (!sblock->no_io_error_seen) {
2391 scrub_handle_errored_block(sblock);
2394 * if has checksum error, write via repair mechanism in
2395 * dev replace case, otherwise write here in dev replace
2398 corrupted = scrub_checksum(sblock);
2399 if (!corrupted && sblock->sctx->is_dev_replace)
2400 scrub_write_block_to_dev_replace(sblock);
2403 if (sblock->sparity && corrupted && !sblock->data_corrected) {
2404 u64 start = sblock->pagev[0]->logical;
2405 u64 end = sblock->pagev[sblock->page_count - 1]->logical +
2408 scrub_parity_mark_sectors_error(sblock->sparity,
2409 start, end - start);
2413 static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len,
2416 struct btrfs_ordered_sum *sum = NULL;
2417 unsigned long index;
2418 unsigned long num_sectors;
2420 while (!list_empty(&sctx->csum_list)) {
2421 sum = list_first_entry(&sctx->csum_list,
2422 struct btrfs_ordered_sum, list);
2423 if (sum->bytenr > logical)
2425 if (sum->bytenr + sum->len > logical)
2428 ++sctx->stat.csum_discards;
2429 list_del(&sum->list);
2436 index = ((u32)(logical - sum->bytenr)) / sctx->sectorsize;
2437 num_sectors = sum->len / sctx->sectorsize;
2438 memcpy(csum, sum->sums + index, sctx->csum_size);
2439 if (index == num_sectors - 1) {
2440 list_del(&sum->list);
2446 /* scrub extent tries to collect up to 64 kB for each bio */
2447 static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len,
2448 u64 physical, struct btrfs_device *dev, u64 flags,
2449 u64 gen, int mirror_num, u64 physical_for_dev_replace)
2452 u8 csum[BTRFS_CSUM_SIZE];
2455 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2456 blocksize = sctx->sectorsize;
2457 spin_lock(&sctx->stat_lock);
2458 sctx->stat.data_extents_scrubbed++;
2459 sctx->stat.data_bytes_scrubbed += len;
2460 spin_unlock(&sctx->stat_lock);
2461 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2462 blocksize = sctx->nodesize;
2463 spin_lock(&sctx->stat_lock);
2464 sctx->stat.tree_extents_scrubbed++;
2465 sctx->stat.tree_bytes_scrubbed += len;
2466 spin_unlock(&sctx->stat_lock);
2468 blocksize = sctx->sectorsize;
2473 u64 l = min_t(u64, len, blocksize);
2476 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2477 /* push csums to sbio */
2478 have_csum = scrub_find_csum(sctx, logical, l, csum);
2480 ++sctx->stat.no_csum;
2481 if (sctx->is_dev_replace && !have_csum) {
2482 ret = copy_nocow_pages(sctx, logical, l,
2484 physical_for_dev_replace);
2485 goto behind_scrub_pages;
2488 ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen,
2489 mirror_num, have_csum ? csum : NULL, 0,
2490 physical_for_dev_replace);
2497 physical_for_dev_replace += l;
2502 static int scrub_pages_for_parity(struct scrub_parity *sparity,
2503 u64 logical, u64 len,
2504 u64 physical, struct btrfs_device *dev,
2505 u64 flags, u64 gen, int mirror_num, u8 *csum)
2507 struct scrub_ctx *sctx = sparity->sctx;
2508 struct scrub_block *sblock;
2511 sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
2513 spin_lock(&sctx->stat_lock);
2514 sctx->stat.malloc_errors++;
2515 spin_unlock(&sctx->stat_lock);
2519 /* one ref inside this function, plus one for each page added to
2521 atomic_set(&sblock->ref_count, 1);
2522 sblock->sctx = sctx;
2523 sblock->no_io_error_seen = 1;
2524 sblock->sparity = sparity;
2525 scrub_parity_get(sparity);
2527 for (index = 0; len > 0; index++) {
2528 struct scrub_page *spage;
2529 u64 l = min_t(u64, len, PAGE_SIZE);
2531 spage = kzalloc(sizeof(*spage), GFP_NOFS);
2534 spin_lock(&sctx->stat_lock);
2535 sctx->stat.malloc_errors++;
2536 spin_unlock(&sctx->stat_lock);
2537 scrub_block_put(sblock);
2540 BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
2541 /* For scrub block */
2542 scrub_page_get(spage);
2543 sblock->pagev[index] = spage;
2544 /* For scrub parity */
2545 scrub_page_get(spage);
2546 list_add_tail(&spage->list, &sparity->spages);
2547 spage->sblock = sblock;
2549 spage->flags = flags;
2550 spage->generation = gen;
2551 spage->logical = logical;
2552 spage->physical = physical;
2553 spage->mirror_num = mirror_num;
2555 spage->have_csum = 1;
2556 memcpy(spage->csum, csum, sctx->csum_size);
2558 spage->have_csum = 0;
2560 sblock->page_count++;
2561 spage->page = alloc_page(GFP_NOFS);
2569 WARN_ON(sblock->page_count == 0);
2570 for (index = 0; index < sblock->page_count; index++) {
2571 struct scrub_page *spage = sblock->pagev[index];
2574 ret = scrub_add_page_to_rd_bio(sctx, spage);
2576 scrub_block_put(sblock);
2581 /* last one frees, either here or in bio completion for last page */
2582 scrub_block_put(sblock);
2586 static int scrub_extent_for_parity(struct scrub_parity *sparity,
2587 u64 logical, u64 len,
2588 u64 physical, struct btrfs_device *dev,
2589 u64 flags, u64 gen, int mirror_num)
2591 struct scrub_ctx *sctx = sparity->sctx;
2593 u8 csum[BTRFS_CSUM_SIZE];
2596 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2597 blocksize = sctx->sectorsize;
2598 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2599 blocksize = sctx->nodesize;
2601 blocksize = sctx->sectorsize;
2606 u64 l = min_t(u64, len, blocksize);
2609 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2610 /* push csums to sbio */
2611 have_csum = scrub_find_csum(sctx, logical, l, csum);
2615 ret = scrub_pages_for_parity(sparity, logical, l, physical, dev,
2616 flags, gen, mirror_num,
2617 have_csum ? csum : NULL);
2629 * Given a physical address, this will calculate it's
2630 * logical offset. if this is a parity stripe, it will return
2631 * the most left data stripe's logical offset.
2633 * return 0 if it is a data stripe, 1 means parity stripe.
2635 static int get_raid56_logic_offset(u64 physical, int num,
2636 struct map_lookup *map, u64 *offset,
2646 last_offset = (physical - map->stripes[num].physical) *
2647 nr_data_stripes(map);
2649 *stripe_start = last_offset;
2651 *offset = last_offset;
2652 for (i = 0; i < nr_data_stripes(map); i++) {
2653 *offset = last_offset + i * map->stripe_len;
2655 stripe_nr = *offset;
2656 do_div(stripe_nr, map->stripe_len);
2657 do_div(stripe_nr, nr_data_stripes(map));
2659 /* Work out the disk rotation on this stripe-set */
2660 rot = do_div(stripe_nr, map->num_stripes);
2661 /* calculate which stripe this data locates */
2663 stripe_index = rot % map->num_stripes;
2664 if (stripe_index == num)
2666 if (stripe_index < num)
2669 *offset = last_offset + j * map->stripe_len;
2673 static void scrub_free_parity(struct scrub_parity *sparity)
2675 struct scrub_ctx *sctx = sparity->sctx;
2676 struct scrub_page *curr, *next;
2679 nbits = bitmap_weight(sparity->ebitmap, sparity->nsectors);
2681 spin_lock(&sctx->stat_lock);
2682 sctx->stat.read_errors += nbits;
2683 sctx->stat.uncorrectable_errors += nbits;
2684 spin_unlock(&sctx->stat_lock);
2687 list_for_each_entry_safe(curr, next, &sparity->spages, list) {
2688 list_del_init(&curr->list);
2689 scrub_page_put(curr);
2695 static void scrub_parity_bio_endio(struct bio *bio, int error)
2697 struct scrub_parity *sparity = (struct scrub_parity *)bio->bi_private;
2698 struct scrub_ctx *sctx = sparity->sctx;
2701 bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap,
2704 scrub_free_parity(sparity);
2705 scrub_pending_bio_dec(sctx);
2709 static void scrub_parity_check_and_repair(struct scrub_parity *sparity)
2711 struct scrub_ctx *sctx = sparity->sctx;
2713 struct btrfs_raid_bio *rbio;
2714 struct scrub_page *spage;
2715 struct btrfs_bio *bbio = NULL;
2716 u64 *raid_map = NULL;
2720 if (!bitmap_andnot(sparity->dbitmap, sparity->dbitmap, sparity->ebitmap,
2724 length = sparity->logic_end - sparity->logic_start + 1;
2725 ret = btrfs_map_sblock(sctx->dev_root->fs_info, WRITE,
2726 sparity->logic_start,
2727 &length, &bbio, 0, &raid_map);
2728 if (ret || !bbio || !raid_map)
2731 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
2735 bio->bi_iter.bi_sector = sparity->logic_start >> 9;
2736 bio->bi_private = sparity;
2737 bio->bi_end_io = scrub_parity_bio_endio;
2739 rbio = raid56_parity_alloc_scrub_rbio(sctx->dev_root, bio, bbio,
2747 list_for_each_entry(spage, &sparity->spages, list)
2748 raid56_parity_add_scrub_pages(rbio, spage->page,
2751 scrub_pending_bio_inc(sctx);
2752 raid56_parity_submit_scrub_rbio(rbio);
2760 bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap,
2762 spin_lock(&sctx->stat_lock);
2763 sctx->stat.malloc_errors++;
2764 spin_unlock(&sctx->stat_lock);
2766 scrub_free_parity(sparity);
2769 static inline int scrub_calc_parity_bitmap_len(int nsectors)
2771 return DIV_ROUND_UP(nsectors, BITS_PER_LONG) * (BITS_PER_LONG / 8);
2774 static void scrub_parity_get(struct scrub_parity *sparity)
2776 atomic_inc(&sparity->ref_count);
2779 static void scrub_parity_put(struct scrub_parity *sparity)
2781 if (!atomic_dec_and_test(&sparity->ref_count))
2784 scrub_parity_check_and_repair(sparity);
2787 static noinline_for_stack int scrub_raid56_parity(struct scrub_ctx *sctx,
2788 struct map_lookup *map,
2789 struct btrfs_device *sdev,
2790 struct btrfs_path *path,
2794 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
2795 struct btrfs_root *root = fs_info->extent_root;
2796 struct btrfs_root *csum_root = fs_info->csum_root;
2797 struct btrfs_extent_item *extent;
2801 struct extent_buffer *l;
2802 struct btrfs_key key;
2805 u64 extent_physical;
2807 struct btrfs_device *extent_dev;
2808 struct scrub_parity *sparity;
2811 int extent_mirror_num;
2814 nsectors = map->stripe_len / root->sectorsize;
2815 bitmap_len = scrub_calc_parity_bitmap_len(nsectors);
2816 sparity = kzalloc(sizeof(struct scrub_parity) + 2 * bitmap_len,
2819 spin_lock(&sctx->stat_lock);
2820 sctx->stat.malloc_errors++;
2821 spin_unlock(&sctx->stat_lock);
2825 sparity->stripe_len = map->stripe_len;
2826 sparity->nsectors = nsectors;
2827 sparity->sctx = sctx;
2828 sparity->scrub_dev = sdev;
2829 sparity->logic_start = logic_start;
2830 sparity->logic_end = logic_end;
2831 atomic_set(&sparity->ref_count, 1);
2832 INIT_LIST_HEAD(&sparity->spages);
2833 sparity->dbitmap = sparity->bitmap;
2834 sparity->ebitmap = (void *)sparity->bitmap + bitmap_len;
2837 while (logic_start < logic_end) {
2838 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2839 key.type = BTRFS_METADATA_ITEM_KEY;
2841 key.type = BTRFS_EXTENT_ITEM_KEY;
2842 key.objectid = logic_start;
2843 key.offset = (u64)-1;
2845 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2850 ret = btrfs_previous_extent_item(root, path, 0);
2854 btrfs_release_path(path);
2855 ret = btrfs_search_slot(NULL, root, &key,
2867 slot = path->slots[0];
2868 if (slot >= btrfs_header_nritems(l)) {
2869 ret = btrfs_next_leaf(root, path);
2878 btrfs_item_key_to_cpu(l, &key, slot);
2880 if (key.type == BTRFS_METADATA_ITEM_KEY)
2881 bytes = root->nodesize;
2885 if (key.objectid + bytes <= logic_start)
2888 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
2889 key.type != BTRFS_METADATA_ITEM_KEY)
2892 if (key.objectid > logic_end) {
2897 while (key.objectid >= logic_start + map->stripe_len)
2898 logic_start += map->stripe_len;
2900 extent = btrfs_item_ptr(l, slot,
2901 struct btrfs_extent_item);
2902 flags = btrfs_extent_flags(l, extent);
2903 generation = btrfs_extent_generation(l, extent);
2905 if (key.objectid < logic_start &&
2906 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
2908 "scrub: tree block %llu spanning stripes, ignored. logical=%llu",
2909 key.objectid, logic_start);
2913 extent_logical = key.objectid;
2916 if (extent_logical < logic_start) {
2917 extent_len -= logic_start - extent_logical;
2918 extent_logical = logic_start;
2921 if (extent_logical + extent_len >
2922 logic_start + map->stripe_len)
2923 extent_len = logic_start + map->stripe_len -
2926 scrub_parity_mark_sectors_data(sparity, extent_logical,
2929 scrub_remap_extent(fs_info, extent_logical,
2930 extent_len, &extent_physical,
2932 &extent_mirror_num);
2934 ret = btrfs_lookup_csums_range(csum_root,
2936 extent_logical + extent_len - 1,
2937 &sctx->csum_list, 1);
2941 ret = scrub_extent_for_parity(sparity, extent_logical,
2950 scrub_free_csums(sctx);
2951 if (extent_logical + extent_len <
2952 key.objectid + bytes) {
2953 logic_start += map->stripe_len;
2955 if (logic_start >= logic_end) {
2960 if (logic_start < key.objectid + bytes) {
2969 btrfs_release_path(path);
2974 logic_start += map->stripe_len;
2978 scrub_parity_mark_sectors_error(sparity, logic_start,
2979 logic_end - logic_start + 1);
2980 scrub_parity_put(sparity);
2982 mutex_lock(&sctx->wr_ctx.wr_lock);
2983 scrub_wr_submit(sctx);
2984 mutex_unlock(&sctx->wr_ctx.wr_lock);
2986 btrfs_release_path(path);
2987 return ret < 0 ? ret : 0;
2990 static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
2991 struct map_lookup *map,
2992 struct btrfs_device *scrub_dev,
2993 int num, u64 base, u64 length,
2996 struct btrfs_path *path, *ppath;
2997 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
2998 struct btrfs_root *root = fs_info->extent_root;
2999 struct btrfs_root *csum_root = fs_info->csum_root;
3000 struct btrfs_extent_item *extent;
3001 struct blk_plug plug;
3006 struct extent_buffer *l;
3007 struct btrfs_key key;
3014 struct reada_control *reada1;
3015 struct reada_control *reada2;
3016 struct btrfs_key key_start;
3017 struct btrfs_key key_end;
3018 u64 increment = map->stripe_len;
3021 u64 extent_physical;
3025 struct btrfs_device *extent_dev;
3026 int extent_mirror_num;
3030 physical = map->stripes[num].physical;
3032 do_div(nstripes, map->stripe_len);
3033 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3034 offset = map->stripe_len * num;
3035 increment = map->stripe_len * map->num_stripes;
3037 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3038 int factor = map->num_stripes / map->sub_stripes;
3039 offset = map->stripe_len * (num / map->sub_stripes);
3040 increment = map->stripe_len * factor;
3041 mirror_num = num % map->sub_stripes + 1;
3042 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3043 increment = map->stripe_len;
3044 mirror_num = num % map->num_stripes + 1;
3045 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3046 increment = map->stripe_len;
3047 mirror_num = num % map->num_stripes + 1;
3048 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
3049 BTRFS_BLOCK_GROUP_RAID6)) {
3050 get_raid56_logic_offset(physical, num, map, &offset, NULL);
3051 increment = map->stripe_len * nr_data_stripes(map);
3054 increment = map->stripe_len;
3058 path = btrfs_alloc_path();
3062 ppath = btrfs_alloc_path();
3064 btrfs_free_path(ppath);
3069 * work on commit root. The related disk blocks are static as
3070 * long as COW is applied. This means, it is save to rewrite
3071 * them to repair disk errors without any race conditions
3073 path->search_commit_root = 1;
3074 path->skip_locking = 1;
3077 * trigger the readahead for extent tree csum tree and wait for
3078 * completion. During readahead, the scrub is officially paused
3079 * to not hold off transaction commits
3081 logical = base + offset;
3082 physical_end = physical + nstripes * map->stripe_len;
3083 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
3084 BTRFS_BLOCK_GROUP_RAID6)) {
3085 get_raid56_logic_offset(physical_end, num,
3086 map, &logic_end, NULL);
3089 logic_end = logical + increment * nstripes;
3091 wait_event(sctx->list_wait,
3092 atomic_read(&sctx->bios_in_flight) == 0);
3093 scrub_blocked_if_needed(fs_info);
3095 /* FIXME it might be better to start readahead at commit root */
3096 key_start.objectid = logical;
3097 key_start.type = BTRFS_EXTENT_ITEM_KEY;
3098 key_start.offset = (u64)0;
3099 key_end.objectid = logic_end;
3100 key_end.type = BTRFS_METADATA_ITEM_KEY;
3101 key_end.offset = (u64)-1;
3102 reada1 = btrfs_reada_add(root, &key_start, &key_end);
3104 key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
3105 key_start.type = BTRFS_EXTENT_CSUM_KEY;
3106 key_start.offset = logical;
3107 key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
3108 key_end.type = BTRFS_EXTENT_CSUM_KEY;
3109 key_end.offset = logic_end;
3110 reada2 = btrfs_reada_add(csum_root, &key_start, &key_end);
3112 if (!IS_ERR(reada1))
3113 btrfs_reada_wait(reada1);
3114 if (!IS_ERR(reada2))
3115 btrfs_reada_wait(reada2);
3119 * collect all data csums for the stripe to avoid seeking during
3120 * the scrub. This might currently (crc32) end up to be about 1MB
3122 blk_start_plug(&plug);
3125 * now find all extents for each stripe and scrub them
3128 while (physical < physical_end) {
3129 /* for raid56, we skip parity stripe */
3130 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
3131 BTRFS_BLOCK_GROUP_RAID6)) {
3132 ret = get_raid56_logic_offset(physical, num,
3133 map, &logical, &stripe_logical);
3136 stripe_logical += base;
3137 stripe_end = stripe_logical + increment - 1;
3138 ret = scrub_raid56_parity(sctx, map, scrub_dev,
3139 ppath, stripe_logical,
3149 if (atomic_read(&fs_info->scrub_cancel_req) ||
3150 atomic_read(&sctx->cancel_req)) {
3155 * check to see if we have to pause
3157 if (atomic_read(&fs_info->scrub_pause_req)) {
3158 /* push queued extents */
3159 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
3161 mutex_lock(&sctx->wr_ctx.wr_lock);
3162 scrub_wr_submit(sctx);
3163 mutex_unlock(&sctx->wr_ctx.wr_lock);
3164 wait_event(sctx->list_wait,
3165 atomic_read(&sctx->bios_in_flight) == 0);
3166 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
3167 scrub_blocked_if_needed(fs_info);
3170 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
3171 key.type = BTRFS_METADATA_ITEM_KEY;
3173 key.type = BTRFS_EXTENT_ITEM_KEY;
3174 key.objectid = logical;
3175 key.offset = (u64)-1;
3177 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3182 ret = btrfs_previous_extent_item(root, path, 0);
3186 /* there's no smaller item, so stick with the
3188 btrfs_release_path(path);
3189 ret = btrfs_search_slot(NULL, root, &key,
3201 slot = path->slots[0];
3202 if (slot >= btrfs_header_nritems(l)) {
3203 ret = btrfs_next_leaf(root, path);
3212 btrfs_item_key_to_cpu(l, &key, slot);
3214 if (key.type == BTRFS_METADATA_ITEM_KEY)
3215 bytes = root->nodesize;
3219 if (key.objectid + bytes <= logical)
3222 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3223 key.type != BTRFS_METADATA_ITEM_KEY)
3226 if (key.objectid >= logical + map->stripe_len) {
3227 /* out of this device extent */
3228 if (key.objectid >= logic_end)
3233 extent = btrfs_item_ptr(l, slot,
3234 struct btrfs_extent_item);
3235 flags = btrfs_extent_flags(l, extent);
3236 generation = btrfs_extent_generation(l, extent);
3238 if (key.objectid < logical &&
3239 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
3241 "scrub: tree block %llu spanning "
3242 "stripes, ignored. logical=%llu",
3243 key.objectid, logical);
3248 extent_logical = key.objectid;
3252 * trim extent to this stripe
3254 if (extent_logical < logical) {
3255 extent_len -= logical - extent_logical;
3256 extent_logical = logical;
3258 if (extent_logical + extent_len >
3259 logical + map->stripe_len) {
3260 extent_len = logical + map->stripe_len -
3264 extent_physical = extent_logical - logical + physical;
3265 extent_dev = scrub_dev;
3266 extent_mirror_num = mirror_num;
3268 scrub_remap_extent(fs_info, extent_logical,
3269 extent_len, &extent_physical,
3271 &extent_mirror_num);
3273 ret = btrfs_lookup_csums_range(csum_root, logical,
3274 logical + map->stripe_len - 1,
3275 &sctx->csum_list, 1);
3279 ret = scrub_extent(sctx, extent_logical, extent_len,
3280 extent_physical, extent_dev, flags,
3281 generation, extent_mirror_num,
3282 extent_logical - logical + physical);
3286 scrub_free_csums(sctx);
3287 if (extent_logical + extent_len <
3288 key.objectid + bytes) {
3289 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
3290 BTRFS_BLOCK_GROUP_RAID6)) {
3292 * loop until we find next data stripe
3293 * or we have finished all stripes.
3296 physical += map->stripe_len;
3297 ret = get_raid56_logic_offset(physical,
3302 if (ret && physical < physical_end) {
3303 stripe_logical += base;
3304 stripe_end = stripe_logical +
3306 ret = scrub_raid56_parity(sctx,
3307 map, scrub_dev, ppath,
3315 physical += map->stripe_len;
3316 logical += increment;
3318 if (logical < key.objectid + bytes) {
3323 if (physical >= physical_end) {
3331 btrfs_release_path(path);
3333 logical += increment;
3334 physical += map->stripe_len;
3335 spin_lock(&sctx->stat_lock);
3337 sctx->stat.last_physical = map->stripes[num].physical +
3340 sctx->stat.last_physical = physical;
3341 spin_unlock(&sctx->stat_lock);
3346 /* push queued extents */
3348 mutex_lock(&sctx->wr_ctx.wr_lock);
3349 scrub_wr_submit(sctx);
3350 mutex_unlock(&sctx->wr_ctx.wr_lock);
3352 blk_finish_plug(&plug);
3353 btrfs_free_path(path);
3354 btrfs_free_path(ppath);
3355 return ret < 0 ? ret : 0;
3358 static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx,
3359 struct btrfs_device *scrub_dev,
3360 u64 chunk_tree, u64 chunk_objectid,
3361 u64 chunk_offset, u64 length,
3362 u64 dev_offset, int is_dev_replace)
3364 struct btrfs_mapping_tree *map_tree =
3365 &sctx->dev_root->fs_info->mapping_tree;
3366 struct map_lookup *map;
3367 struct extent_map *em;
3371 read_lock(&map_tree->map_tree.lock);
3372 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3373 read_unlock(&map_tree->map_tree.lock);
3378 map = (struct map_lookup *)em->bdev;
3379 if (em->start != chunk_offset)
3382 if (em->len < length)
3385 for (i = 0; i < map->num_stripes; ++i) {
3386 if (map->stripes[i].dev->bdev == scrub_dev->bdev &&
3387 map->stripes[i].physical == dev_offset) {
3388 ret = scrub_stripe(sctx, map, scrub_dev, i,
3389 chunk_offset, length,
3396 free_extent_map(em);
3401 static noinline_for_stack
3402 int scrub_enumerate_chunks(struct scrub_ctx *sctx,
3403 struct btrfs_device *scrub_dev, u64 start, u64 end,
3406 struct btrfs_dev_extent *dev_extent = NULL;
3407 struct btrfs_path *path;
3408 struct btrfs_root *root = sctx->dev_root;
3409 struct btrfs_fs_info *fs_info = root->fs_info;
3416 struct extent_buffer *l;
3417 struct btrfs_key key;
3418 struct btrfs_key found_key;
3419 struct btrfs_block_group_cache *cache;
3420 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
3422 path = btrfs_alloc_path();
3427 path->search_commit_root = 1;
3428 path->skip_locking = 1;
3430 key.objectid = scrub_dev->devid;
3432 key.type = BTRFS_DEV_EXTENT_KEY;
3435 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3439 if (path->slots[0] >=
3440 btrfs_header_nritems(path->nodes[0])) {
3441 ret = btrfs_next_leaf(root, path);
3448 slot = path->slots[0];
3450 btrfs_item_key_to_cpu(l, &found_key, slot);
3452 if (found_key.objectid != scrub_dev->devid)
3455 if (found_key.type != BTRFS_DEV_EXTENT_KEY)
3458 if (found_key.offset >= end)
3461 if (found_key.offset < key.offset)
3464 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3465 length = btrfs_dev_extent_length(l, dev_extent);
3467 if (found_key.offset + length <= start)
3470 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3471 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3472 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3475 * get a reference on the corresponding block group to prevent
3476 * the chunk from going away while we scrub it
3478 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3480 /* some chunks are removed but not committed to disk yet,
3481 * continue scrubbing */
3485 dev_replace->cursor_right = found_key.offset + length;
3486 dev_replace->cursor_left = found_key.offset;
3487 dev_replace->item_needs_writeback = 1;
3488 ret = scrub_chunk(sctx, scrub_dev, chunk_tree, chunk_objectid,
3489 chunk_offset, length, found_key.offset,
3493 * flush, submit all pending read and write bios, afterwards
3495 * Note that in the dev replace case, a read request causes
3496 * write requests that are submitted in the read completion
3497 * worker. Therefore in the current situation, it is required
3498 * that all write requests are flushed, so that all read and
3499 * write requests are really completed when bios_in_flight
3502 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
3504 mutex_lock(&sctx->wr_ctx.wr_lock);
3505 scrub_wr_submit(sctx);
3506 mutex_unlock(&sctx->wr_ctx.wr_lock);
3508 wait_event(sctx->list_wait,
3509 atomic_read(&sctx->bios_in_flight) == 0);
3510 atomic_inc(&fs_info->scrubs_paused);
3511 wake_up(&fs_info->scrub_pause_wait);
3514 * must be called before we decrease @scrub_paused.
3515 * make sure we don't block transaction commit while
3516 * we are waiting pending workers finished.
3518 wait_event(sctx->list_wait,
3519 atomic_read(&sctx->workers_pending) == 0);
3520 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
3522 mutex_lock(&fs_info->scrub_lock);
3523 __scrub_blocked_if_needed(fs_info);
3524 atomic_dec(&fs_info->scrubs_paused);
3525 mutex_unlock(&fs_info->scrub_lock);
3526 wake_up(&fs_info->scrub_pause_wait);
3528 btrfs_put_block_group(cache);
3531 if (is_dev_replace &&
3532 atomic64_read(&dev_replace->num_write_errors) > 0) {
3536 if (sctx->stat.malloc_errors > 0) {
3541 dev_replace->cursor_left = dev_replace->cursor_right;
3542 dev_replace->item_needs_writeback = 1;
3544 key.offset = found_key.offset + length;
3545 btrfs_release_path(path);
3548 btrfs_free_path(path);
3551 * ret can still be 1 from search_slot or next_leaf,
3552 * that's not an error
3554 return ret < 0 ? ret : 0;
3557 static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx,
3558 struct btrfs_device *scrub_dev)
3564 struct btrfs_root *root = sctx->dev_root;
3566 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
3569 /* Seed devices of a new filesystem has their own generation. */
3570 if (scrub_dev->fs_devices != root->fs_info->fs_devices)
3571 gen = scrub_dev->generation;
3573 gen = root->fs_info->last_trans_committed;
3575 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
3576 bytenr = btrfs_sb_offset(i);
3577 if (bytenr + BTRFS_SUPER_INFO_SIZE >
3578 scrub_dev->commit_total_bytes)
3581 ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr,
3582 scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i,
3587 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
3593 * get a reference count on fs_info->scrub_workers. start worker if necessary
3595 static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info,
3599 int flags = WQ_FREEZABLE | WQ_UNBOUND;
3600 int max_active = fs_info->thread_pool_size;
3602 if (fs_info->scrub_workers_refcnt == 0) {
3604 fs_info->scrub_workers =
3605 btrfs_alloc_workqueue("btrfs-scrub", flags,
3608 fs_info->scrub_workers =
3609 btrfs_alloc_workqueue("btrfs-scrub", flags,
3611 if (!fs_info->scrub_workers) {
3615 fs_info->scrub_wr_completion_workers =
3616 btrfs_alloc_workqueue("btrfs-scrubwrc", flags,
3618 if (!fs_info->scrub_wr_completion_workers) {
3622 fs_info->scrub_nocow_workers =
3623 btrfs_alloc_workqueue("btrfs-scrubnc", flags, 1, 0);
3624 if (!fs_info->scrub_nocow_workers) {
3629 ++fs_info->scrub_workers_refcnt;
3634 static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info)
3636 if (--fs_info->scrub_workers_refcnt == 0) {
3637 btrfs_destroy_workqueue(fs_info->scrub_workers);
3638 btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
3639 btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
3641 WARN_ON(fs_info->scrub_workers_refcnt < 0);
3644 int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start,
3645 u64 end, struct btrfs_scrub_progress *progress,
3646 int readonly, int is_dev_replace)
3648 struct scrub_ctx *sctx;
3650 struct btrfs_device *dev;
3651 struct rcu_string *name;
3653 if (btrfs_fs_closing(fs_info))
3656 if (fs_info->chunk_root->nodesize > BTRFS_STRIPE_LEN) {
3658 * in this case scrub is unable to calculate the checksum
3659 * the way scrub is implemented. Do not handle this
3660 * situation at all because it won't ever happen.
3663 "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails",
3664 fs_info->chunk_root->nodesize, BTRFS_STRIPE_LEN);
3668 if (fs_info->chunk_root->sectorsize != PAGE_SIZE) {
3669 /* not supported for data w/o checksums */
3671 "scrub: size assumption sectorsize != PAGE_SIZE "
3672 "(%d != %lu) fails",
3673 fs_info->chunk_root->sectorsize, PAGE_SIZE);
3677 if (fs_info->chunk_root->nodesize >
3678 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK ||
3679 fs_info->chunk_root->sectorsize >
3680 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) {
3682 * would exhaust the array bounds of pagev member in
3683 * struct scrub_block
3685 btrfs_err(fs_info, "scrub: size assumption nodesize and sectorsize "
3686 "<= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails",
3687 fs_info->chunk_root->nodesize,
3688 SCRUB_MAX_PAGES_PER_BLOCK,
3689 fs_info->chunk_root->sectorsize,
3690 SCRUB_MAX_PAGES_PER_BLOCK);
3695 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3696 dev = btrfs_find_device(fs_info, devid, NULL, NULL);
3697 if (!dev || (dev->missing && !is_dev_replace)) {
3698 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3702 if (!is_dev_replace && !readonly && !dev->writeable) {
3703 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3705 name = rcu_dereference(dev->name);
3706 btrfs_err(fs_info, "scrub: device %s is not writable",
3712 mutex_lock(&fs_info->scrub_lock);
3713 if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) {
3714 mutex_unlock(&fs_info->scrub_lock);
3715 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3719 btrfs_dev_replace_lock(&fs_info->dev_replace);
3720 if (dev->scrub_device ||
3722 btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) {
3723 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3724 mutex_unlock(&fs_info->scrub_lock);
3725 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3726 return -EINPROGRESS;
3728 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3730 ret = scrub_workers_get(fs_info, is_dev_replace);
3732 mutex_unlock(&fs_info->scrub_lock);
3733 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3737 sctx = scrub_setup_ctx(dev, is_dev_replace);
3739 mutex_unlock(&fs_info->scrub_lock);
3740 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3741 scrub_workers_put(fs_info);
3742 return PTR_ERR(sctx);
3744 sctx->readonly = readonly;
3745 dev->scrub_device = sctx;
3746 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3749 * checking @scrub_pause_req here, we can avoid
3750 * race between committing transaction and scrubbing.
3752 __scrub_blocked_if_needed(fs_info);
3753 atomic_inc(&fs_info->scrubs_running);
3754 mutex_unlock(&fs_info->scrub_lock);
3756 if (!is_dev_replace) {
3758 * by holding device list mutex, we can
3759 * kick off writing super in log tree sync.
3761 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3762 ret = scrub_supers(sctx, dev);
3763 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3767 ret = scrub_enumerate_chunks(sctx, dev, start, end,
3770 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
3771 atomic_dec(&fs_info->scrubs_running);
3772 wake_up(&fs_info->scrub_pause_wait);
3774 wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0);
3777 memcpy(progress, &sctx->stat, sizeof(*progress));
3779 mutex_lock(&fs_info->scrub_lock);
3780 dev->scrub_device = NULL;
3781 scrub_workers_put(fs_info);
3782 mutex_unlock(&fs_info->scrub_lock);
3784 scrub_free_ctx(sctx);
3789 void btrfs_scrub_pause(struct btrfs_root *root)
3791 struct btrfs_fs_info *fs_info = root->fs_info;
3793 mutex_lock(&fs_info->scrub_lock);
3794 atomic_inc(&fs_info->scrub_pause_req);
3795 while (atomic_read(&fs_info->scrubs_paused) !=
3796 atomic_read(&fs_info->scrubs_running)) {
3797 mutex_unlock(&fs_info->scrub_lock);
3798 wait_event(fs_info->scrub_pause_wait,
3799 atomic_read(&fs_info->scrubs_paused) ==
3800 atomic_read(&fs_info->scrubs_running));
3801 mutex_lock(&fs_info->scrub_lock);
3803 mutex_unlock(&fs_info->scrub_lock);
3806 void btrfs_scrub_continue(struct btrfs_root *root)
3808 struct btrfs_fs_info *fs_info = root->fs_info;
3810 atomic_dec(&fs_info->scrub_pause_req);
3811 wake_up(&fs_info->scrub_pause_wait);
3814 int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info)
3816 mutex_lock(&fs_info->scrub_lock);
3817 if (!atomic_read(&fs_info->scrubs_running)) {
3818 mutex_unlock(&fs_info->scrub_lock);
3822 atomic_inc(&fs_info->scrub_cancel_req);
3823 while (atomic_read(&fs_info->scrubs_running)) {
3824 mutex_unlock(&fs_info->scrub_lock);
3825 wait_event(fs_info->scrub_pause_wait,
3826 atomic_read(&fs_info->scrubs_running) == 0);
3827 mutex_lock(&fs_info->scrub_lock);
3829 atomic_dec(&fs_info->scrub_cancel_req);
3830 mutex_unlock(&fs_info->scrub_lock);
3835 int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info,
3836 struct btrfs_device *dev)
3838 struct scrub_ctx *sctx;
3840 mutex_lock(&fs_info->scrub_lock);
3841 sctx = dev->scrub_device;
3843 mutex_unlock(&fs_info->scrub_lock);
3846 atomic_inc(&sctx->cancel_req);
3847 while (dev->scrub_device) {
3848 mutex_unlock(&fs_info->scrub_lock);
3849 wait_event(fs_info->scrub_pause_wait,
3850 dev->scrub_device == NULL);
3851 mutex_lock(&fs_info->scrub_lock);
3853 mutex_unlock(&fs_info->scrub_lock);
3858 int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
3859 struct btrfs_scrub_progress *progress)
3861 struct btrfs_device *dev;
3862 struct scrub_ctx *sctx = NULL;
3864 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3865 dev = btrfs_find_device(root->fs_info, devid, NULL, NULL);
3867 sctx = dev->scrub_device;
3869 memcpy(progress, &sctx->stat, sizeof(*progress));
3870 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3872 return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV;
3875 static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
3876 u64 extent_logical, u64 extent_len,
3877 u64 *extent_physical,
3878 struct btrfs_device **extent_dev,
3879 int *extent_mirror_num)
3882 struct btrfs_bio *bbio = NULL;
3885 mapped_length = extent_len;
3886 ret = btrfs_map_block(fs_info, READ, extent_logical,
3887 &mapped_length, &bbio, 0);
3888 if (ret || !bbio || mapped_length < extent_len ||
3889 !bbio->stripes[0].dev->bdev) {
3894 *extent_physical = bbio->stripes[0].physical;
3895 *extent_mirror_num = bbio->mirror_num;
3896 *extent_dev = bbio->stripes[0].dev;
3900 static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
3901 struct scrub_wr_ctx *wr_ctx,
3902 struct btrfs_fs_info *fs_info,
3903 struct btrfs_device *dev,
3906 WARN_ON(wr_ctx->wr_curr_bio != NULL);
3908 mutex_init(&wr_ctx->wr_lock);
3909 wr_ctx->wr_curr_bio = NULL;
3910 if (!is_dev_replace)
3913 WARN_ON(!dev->bdev);
3914 wr_ctx->pages_per_wr_bio = min_t(int, SCRUB_PAGES_PER_WR_BIO,
3915 bio_get_nr_vecs(dev->bdev));
3916 wr_ctx->tgtdev = dev;
3917 atomic_set(&wr_ctx->flush_all_writes, 0);
3921 static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx)
3923 mutex_lock(&wr_ctx->wr_lock);
3924 kfree(wr_ctx->wr_curr_bio);
3925 wr_ctx->wr_curr_bio = NULL;
3926 mutex_unlock(&wr_ctx->wr_lock);
3929 static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
3930 int mirror_num, u64 physical_for_dev_replace)
3932 struct scrub_copy_nocow_ctx *nocow_ctx;
3933 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
3935 nocow_ctx = kzalloc(sizeof(*nocow_ctx), GFP_NOFS);
3937 spin_lock(&sctx->stat_lock);
3938 sctx->stat.malloc_errors++;
3939 spin_unlock(&sctx->stat_lock);
3943 scrub_pending_trans_workers_inc(sctx);
3945 nocow_ctx->sctx = sctx;
3946 nocow_ctx->logical = logical;
3947 nocow_ctx->len = len;
3948 nocow_ctx->mirror_num = mirror_num;
3949 nocow_ctx->physical_for_dev_replace = physical_for_dev_replace;
3950 btrfs_init_work(&nocow_ctx->work, btrfs_scrubnc_helper,
3951 copy_nocow_pages_worker, NULL, NULL);
3952 INIT_LIST_HEAD(&nocow_ctx->inodes);
3953 btrfs_queue_work(fs_info->scrub_nocow_workers,
3959 static int record_inode_for_nocow(u64 inum, u64 offset, u64 root, void *ctx)
3961 struct scrub_copy_nocow_ctx *nocow_ctx = ctx;
3962 struct scrub_nocow_inode *nocow_inode;
3964 nocow_inode = kzalloc(sizeof(*nocow_inode), GFP_NOFS);
3967 nocow_inode->inum = inum;
3968 nocow_inode->offset = offset;
3969 nocow_inode->root = root;
3970 list_add_tail(&nocow_inode->list, &nocow_ctx->inodes);
3974 #define COPY_COMPLETE 1
3976 static void copy_nocow_pages_worker(struct btrfs_work *work)
3978 struct scrub_copy_nocow_ctx *nocow_ctx =
3979 container_of(work, struct scrub_copy_nocow_ctx, work);
3980 struct scrub_ctx *sctx = nocow_ctx->sctx;
3981 u64 logical = nocow_ctx->logical;
3982 u64 len = nocow_ctx->len;
3983 int mirror_num = nocow_ctx->mirror_num;
3984 u64 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
3986 struct btrfs_trans_handle *trans = NULL;
3987 struct btrfs_fs_info *fs_info;
3988 struct btrfs_path *path;
3989 struct btrfs_root *root;
3990 int not_written = 0;
3992 fs_info = sctx->dev_root->fs_info;
3993 root = fs_info->extent_root;
3995 path = btrfs_alloc_path();
3997 spin_lock(&sctx->stat_lock);
3998 sctx->stat.malloc_errors++;
3999 spin_unlock(&sctx->stat_lock);
4004 trans = btrfs_join_transaction(root);
4005 if (IS_ERR(trans)) {
4010 ret = iterate_inodes_from_logical(logical, fs_info, path,
4011 record_inode_for_nocow, nocow_ctx);
4012 if (ret != 0 && ret != -ENOENT) {
4013 btrfs_warn(fs_info, "iterate_inodes_from_logical() failed: log %llu, "
4014 "phys %llu, len %llu, mir %u, ret %d",
4015 logical, physical_for_dev_replace, len, mirror_num,
4021 btrfs_end_transaction(trans, root);
4023 while (!list_empty(&nocow_ctx->inodes)) {
4024 struct scrub_nocow_inode *entry;
4025 entry = list_first_entry(&nocow_ctx->inodes,
4026 struct scrub_nocow_inode,
4028 list_del_init(&entry->list);
4029 ret = copy_nocow_pages_for_inode(entry->inum, entry->offset,
4030 entry->root, nocow_ctx);
4032 if (ret == COPY_COMPLETE) {
4040 while (!list_empty(&nocow_ctx->inodes)) {
4041 struct scrub_nocow_inode *entry;
4042 entry = list_first_entry(&nocow_ctx->inodes,
4043 struct scrub_nocow_inode,
4045 list_del_init(&entry->list);
4048 if (trans && !IS_ERR(trans))
4049 btrfs_end_transaction(trans, root);
4051 btrfs_dev_replace_stats_inc(&fs_info->dev_replace.
4052 num_uncorrectable_read_errors);
4054 btrfs_free_path(path);
4057 scrub_pending_trans_workers_dec(sctx);
4060 static int check_extent_to_block(struct inode *inode, u64 start, u64 len,
4063 struct extent_state *cached_state = NULL;
4064 struct btrfs_ordered_extent *ordered;
4065 struct extent_io_tree *io_tree;
4066 struct extent_map *em;
4067 u64 lockstart = start, lockend = start + len - 1;
4070 io_tree = &BTRFS_I(inode)->io_tree;
4072 lock_extent_bits(io_tree, lockstart, lockend, 0, &cached_state);
4073 ordered = btrfs_lookup_ordered_range(inode, lockstart, len);
4075 btrfs_put_ordered_extent(ordered);
4080 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
4087 * This extent does not actually cover the logical extent anymore,
4088 * move on to the next inode.
4090 if (em->block_start > logical ||
4091 em->block_start + em->block_len < logical + len) {
4092 free_extent_map(em);
4096 free_extent_map(em);
4099 unlock_extent_cached(io_tree, lockstart, lockend, &cached_state,
4104 static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
4105 struct scrub_copy_nocow_ctx *nocow_ctx)
4107 struct btrfs_fs_info *fs_info = nocow_ctx->sctx->dev_root->fs_info;
4108 struct btrfs_key key;
4109 struct inode *inode;
4111 struct btrfs_root *local_root;
4112 struct extent_io_tree *io_tree;
4113 u64 physical_for_dev_replace;
4114 u64 nocow_ctx_logical;
4115 u64 len = nocow_ctx->len;
4116 unsigned long index;
4121 key.objectid = root;
4122 key.type = BTRFS_ROOT_ITEM_KEY;
4123 key.offset = (u64)-1;
4125 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
4127 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
4128 if (IS_ERR(local_root)) {
4129 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
4130 return PTR_ERR(local_root);
4133 key.type = BTRFS_INODE_ITEM_KEY;
4134 key.objectid = inum;
4136 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
4137 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
4139 return PTR_ERR(inode);
4141 /* Avoid truncate/dio/punch hole.. */
4142 mutex_lock(&inode->i_mutex);
4143 inode_dio_wait(inode);
4145 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
4146 io_tree = &BTRFS_I(inode)->io_tree;
4147 nocow_ctx_logical = nocow_ctx->logical;
4149 ret = check_extent_to_block(inode, offset, len, nocow_ctx_logical);
4151 ret = ret > 0 ? 0 : ret;
4155 while (len >= PAGE_CACHE_SIZE) {
4156 index = offset >> PAGE_CACHE_SHIFT;
4158 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
4160 btrfs_err(fs_info, "find_or_create_page() failed");
4165 if (PageUptodate(page)) {
4166 if (PageDirty(page))
4169 ClearPageError(page);
4170 err = extent_read_full_page(io_tree, page,
4172 nocow_ctx->mirror_num);
4180 * If the page has been remove from the page cache,
4181 * the data on it is meaningless, because it may be
4182 * old one, the new data may be written into the new
4183 * page in the page cache.
4185 if (page->mapping != inode->i_mapping) {
4187 page_cache_release(page);
4190 if (!PageUptodate(page)) {
4196 ret = check_extent_to_block(inode, offset, len,
4199 ret = ret > 0 ? 0 : ret;
4203 err = write_page_nocow(nocow_ctx->sctx,
4204 physical_for_dev_replace, page);
4209 page_cache_release(page);
4214 offset += PAGE_CACHE_SIZE;
4215 physical_for_dev_replace += PAGE_CACHE_SIZE;
4216 nocow_ctx_logical += PAGE_CACHE_SIZE;
4217 len -= PAGE_CACHE_SIZE;
4219 ret = COPY_COMPLETE;
4221 mutex_unlock(&inode->i_mutex);
4226 static int write_page_nocow(struct scrub_ctx *sctx,
4227 u64 physical_for_dev_replace, struct page *page)
4230 struct btrfs_device *dev;
4233 dev = sctx->wr_ctx.tgtdev;
4237 printk_ratelimited(KERN_WARNING
4238 "BTRFS: scrub write_page_nocow(bdev == NULL) is unexpected!\n");
4241 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
4243 spin_lock(&sctx->stat_lock);
4244 sctx->stat.malloc_errors++;
4245 spin_unlock(&sctx->stat_lock);
4248 bio->bi_iter.bi_size = 0;
4249 bio->bi_iter.bi_sector = physical_for_dev_replace >> 9;
4250 bio->bi_bdev = dev->bdev;
4251 ret = bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
4252 if (ret != PAGE_CACHE_SIZE) {
4255 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
4259 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio))
4260 goto leave_with_eio;