2 * Copyright (C) 2011, 2012 STRATO. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/blkdev.h>
20 #include <linux/ratelimit.h>
24 #include "ordered-data.h"
25 #include "transaction.h"
27 #include "extent_io.h"
28 #include "dev-replace.h"
29 #include "check-integrity.h"
30 #include "rcu-string.h"
34 * This is only the first step towards a full-features scrub. It reads all
35 * extent and super block and verifies the checksums. In case a bad checksum
36 * is found or the extent cannot be read, good data will be written back if
39 * Future enhancements:
40 * - In case an unrepairable extent is encountered, track which files are
41 * affected and report them
42 * - track and record media errors, throw out bad devices
43 * - add a mode to also read unallocated space
50 * the following three values only influence the performance.
51 * The last one configures the number of parallel and outstanding I/O
52 * operations. The first two values configure an upper limit for the number
53 * of (dynamically allocated) pages that are added to a bio.
55 #define SCRUB_PAGES_PER_RD_BIO 32 /* 128k per bio */
56 #define SCRUB_PAGES_PER_WR_BIO 32 /* 128k per bio */
57 #define SCRUB_BIOS_PER_SCTX 64 /* 8MB per device in flight */
60 * the following value times PAGE_SIZE needs to be large enough to match the
61 * largest node/leaf/sector size that shall be supported.
62 * Values larger than BTRFS_STRIPE_LEN are not supported.
64 #define SCRUB_MAX_PAGES_PER_BLOCK 16 /* 64k per node/leaf/sector */
66 struct scrub_recover {
68 struct btrfs_bio *bbio;
73 struct scrub_block *sblock;
75 struct btrfs_device *dev;
76 struct list_head list;
77 u64 flags; /* extent flags */
81 u64 physical_for_dev_replace;
84 unsigned int mirror_num:8;
85 unsigned int have_csum:1;
86 unsigned int io_error:1;
88 u8 csum[BTRFS_CSUM_SIZE];
90 struct scrub_recover *recover;
95 struct scrub_ctx *sctx;
96 struct btrfs_device *dev;
101 #if SCRUB_PAGES_PER_WR_BIO >= SCRUB_PAGES_PER_RD_BIO
102 struct scrub_page *pagev[SCRUB_PAGES_PER_WR_BIO];
104 struct scrub_page *pagev[SCRUB_PAGES_PER_RD_BIO];
108 struct btrfs_work work;
112 struct scrub_page *pagev[SCRUB_MAX_PAGES_PER_BLOCK];
114 atomic_t outstanding_pages;
115 atomic_t refs; /* free mem on transition to zero */
116 struct scrub_ctx *sctx;
117 struct scrub_parity *sparity;
119 unsigned int header_error:1;
120 unsigned int checksum_error:1;
121 unsigned int no_io_error_seen:1;
122 unsigned int generation_error:1; /* also sets header_error */
124 /* The following is for the data used to check parity */
125 /* It is for the data with checksum */
126 unsigned int data_corrected:1;
130 /* Used for the chunks with parity stripe such RAID5/6 */
131 struct scrub_parity {
132 struct scrub_ctx *sctx;
134 struct btrfs_device *scrub_dev;
146 struct list_head spages;
148 /* Work of parity check and repair */
149 struct btrfs_work work;
151 /* Mark the parity blocks which have data */
152 unsigned long *dbitmap;
155 * Mark the parity blocks which have data, but errors happen when
156 * read data or check data
158 unsigned long *ebitmap;
160 unsigned long bitmap[0];
163 struct scrub_wr_ctx {
164 struct scrub_bio *wr_curr_bio;
165 struct btrfs_device *tgtdev;
166 int pages_per_wr_bio; /* <= SCRUB_PAGES_PER_WR_BIO */
167 atomic_t flush_all_writes;
168 struct mutex wr_lock;
172 struct scrub_bio *bios[SCRUB_BIOS_PER_SCTX];
173 struct btrfs_root *dev_root;
176 atomic_t bios_in_flight;
177 atomic_t workers_pending;
178 spinlock_t list_lock;
179 wait_queue_head_t list_wait;
181 struct list_head csum_list;
184 int pages_per_rd_bio;
189 struct scrub_wr_ctx wr_ctx;
194 struct btrfs_scrub_progress stat;
195 spinlock_t stat_lock;
198 * Use a ref counter to avoid use-after-free issues. Scrub workers
199 * decrement bios_in_flight and workers_pending and then do a wakeup
200 * on the list_wait wait queue. We must ensure the main scrub task
201 * doesn't free the scrub context before or while the workers are
202 * doing the wakeup() call.
207 struct scrub_fixup_nodatasum {
208 struct scrub_ctx *sctx;
209 struct btrfs_device *dev;
211 struct btrfs_root *root;
212 struct btrfs_work work;
216 struct scrub_nocow_inode {
220 struct list_head list;
223 struct scrub_copy_nocow_ctx {
224 struct scrub_ctx *sctx;
228 u64 physical_for_dev_replace;
229 struct list_head inodes;
230 struct btrfs_work work;
233 struct scrub_warning {
234 struct btrfs_path *path;
235 u64 extent_item_size;
239 struct btrfs_device *dev;
242 static void scrub_pending_bio_inc(struct scrub_ctx *sctx);
243 static void scrub_pending_bio_dec(struct scrub_ctx *sctx);
244 static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx);
245 static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx);
246 static int scrub_handle_errored_block(struct scrub_block *sblock_to_check);
247 static int scrub_setup_recheck_block(struct scrub_block *original_sblock,
248 struct scrub_block *sblocks_for_recheck);
249 static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
250 struct scrub_block *sblock, int is_metadata,
251 int have_csum, u8 *csum, u64 generation,
252 u16 csum_size, int retry_failed_mirror);
253 static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
254 struct scrub_block *sblock,
255 int is_metadata, int have_csum,
256 const u8 *csum, u64 generation,
258 static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
259 struct scrub_block *sblock_good);
260 static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
261 struct scrub_block *sblock_good,
262 int page_num, int force_write);
263 static void scrub_write_block_to_dev_replace(struct scrub_block *sblock);
264 static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
266 static int scrub_checksum_data(struct scrub_block *sblock);
267 static int scrub_checksum_tree_block(struct scrub_block *sblock);
268 static int scrub_checksum_super(struct scrub_block *sblock);
269 static void scrub_block_get(struct scrub_block *sblock);
270 static void scrub_block_put(struct scrub_block *sblock);
271 static void scrub_page_get(struct scrub_page *spage);
272 static void scrub_page_put(struct scrub_page *spage);
273 static void scrub_parity_get(struct scrub_parity *sparity);
274 static void scrub_parity_put(struct scrub_parity *sparity);
275 static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
276 struct scrub_page *spage);
277 static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
278 u64 physical, struct btrfs_device *dev, u64 flags,
279 u64 gen, int mirror_num, u8 *csum, int force,
280 u64 physical_for_dev_replace);
281 static void scrub_bio_end_io(struct bio *bio, int err);
282 static void scrub_bio_end_io_worker(struct btrfs_work *work);
283 static void scrub_block_complete(struct scrub_block *sblock);
284 static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
285 u64 extent_logical, u64 extent_len,
286 u64 *extent_physical,
287 struct btrfs_device **extent_dev,
288 int *extent_mirror_num);
289 static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
290 struct scrub_wr_ctx *wr_ctx,
291 struct btrfs_fs_info *fs_info,
292 struct btrfs_device *dev,
294 static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx);
295 static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
296 struct scrub_page *spage);
297 static void scrub_wr_submit(struct scrub_ctx *sctx);
298 static void scrub_wr_bio_end_io(struct bio *bio, int err);
299 static void scrub_wr_bio_end_io_worker(struct btrfs_work *work);
300 static int write_page_nocow(struct scrub_ctx *sctx,
301 u64 physical_for_dev_replace, struct page *page);
302 static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
303 struct scrub_copy_nocow_ctx *ctx);
304 static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
305 int mirror_num, u64 physical_for_dev_replace);
306 static void copy_nocow_pages_worker(struct btrfs_work *work);
307 static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
308 static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
309 static void scrub_put_ctx(struct scrub_ctx *sctx);
312 static void scrub_pending_bio_inc(struct scrub_ctx *sctx)
314 atomic_inc(&sctx->refs);
315 atomic_inc(&sctx->bios_in_flight);
318 static void scrub_pending_bio_dec(struct scrub_ctx *sctx)
320 atomic_dec(&sctx->bios_in_flight);
321 wake_up(&sctx->list_wait);
325 static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
327 while (atomic_read(&fs_info->scrub_pause_req)) {
328 mutex_unlock(&fs_info->scrub_lock);
329 wait_event(fs_info->scrub_pause_wait,
330 atomic_read(&fs_info->scrub_pause_req) == 0);
331 mutex_lock(&fs_info->scrub_lock);
335 static void scrub_pause_on(struct btrfs_fs_info *fs_info)
337 atomic_inc(&fs_info->scrubs_paused);
338 wake_up(&fs_info->scrub_pause_wait);
341 static void scrub_pause_off(struct btrfs_fs_info *fs_info)
343 mutex_lock(&fs_info->scrub_lock);
344 __scrub_blocked_if_needed(fs_info);
345 atomic_dec(&fs_info->scrubs_paused);
346 mutex_unlock(&fs_info->scrub_lock);
348 wake_up(&fs_info->scrub_pause_wait);
351 static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
353 scrub_pause_on(fs_info);
354 scrub_pause_off(fs_info);
358 * used for workers that require transaction commits (i.e., for the
361 static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx)
363 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
365 atomic_inc(&sctx->refs);
367 * increment scrubs_running to prevent cancel requests from
368 * completing as long as a worker is running. we must also
369 * increment scrubs_paused to prevent deadlocking on pause
370 * requests used for transactions commits (as the worker uses a
371 * transaction context). it is safe to regard the worker
372 * as paused for all matters practical. effectively, we only
373 * avoid cancellation requests from completing.
375 mutex_lock(&fs_info->scrub_lock);
376 atomic_inc(&fs_info->scrubs_running);
377 atomic_inc(&fs_info->scrubs_paused);
378 mutex_unlock(&fs_info->scrub_lock);
381 * check if @scrubs_running=@scrubs_paused condition
382 * inside wait_event() is not an atomic operation.
383 * which means we may inc/dec @scrub_running/paused
384 * at any time. Let's wake up @scrub_pause_wait as
385 * much as we can to let commit transaction blocked less.
387 wake_up(&fs_info->scrub_pause_wait);
389 atomic_inc(&sctx->workers_pending);
392 /* used for workers that require transaction commits */
393 static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx)
395 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
398 * see scrub_pending_trans_workers_inc() why we're pretending
399 * to be paused in the scrub counters
401 mutex_lock(&fs_info->scrub_lock);
402 atomic_dec(&fs_info->scrubs_running);
403 atomic_dec(&fs_info->scrubs_paused);
404 mutex_unlock(&fs_info->scrub_lock);
405 atomic_dec(&sctx->workers_pending);
406 wake_up(&fs_info->scrub_pause_wait);
407 wake_up(&sctx->list_wait);
411 static void scrub_free_csums(struct scrub_ctx *sctx)
413 while (!list_empty(&sctx->csum_list)) {
414 struct btrfs_ordered_sum *sum;
415 sum = list_first_entry(&sctx->csum_list,
416 struct btrfs_ordered_sum, list);
417 list_del(&sum->list);
422 static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx)
429 scrub_free_wr_ctx(&sctx->wr_ctx);
431 /* this can happen when scrub is cancelled */
432 if (sctx->curr != -1) {
433 struct scrub_bio *sbio = sctx->bios[sctx->curr];
435 for (i = 0; i < sbio->page_count; i++) {
436 WARN_ON(!sbio->pagev[i]->page);
437 scrub_block_put(sbio->pagev[i]->sblock);
442 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
443 struct scrub_bio *sbio = sctx->bios[i];
450 scrub_free_csums(sctx);
454 static void scrub_put_ctx(struct scrub_ctx *sctx)
456 if (atomic_dec_and_test(&sctx->refs))
457 scrub_free_ctx(sctx);
460 static noinline_for_stack
461 struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace)
463 struct scrub_ctx *sctx;
465 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
466 int pages_per_rd_bio;
470 * the setting of pages_per_rd_bio is correct for scrub but might
471 * be wrong for the dev_replace code where we might read from
472 * different devices in the initial huge bios. However, that
473 * code is able to correctly handle the case when adding a page
477 pages_per_rd_bio = min_t(int, SCRUB_PAGES_PER_RD_BIO,
478 bio_get_nr_vecs(dev->bdev));
480 pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO;
481 sctx = kzalloc(sizeof(*sctx), GFP_NOFS);
484 atomic_set(&sctx->refs, 1);
485 sctx->is_dev_replace = is_dev_replace;
486 sctx->pages_per_rd_bio = pages_per_rd_bio;
488 sctx->dev_root = dev->dev_root;
489 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
490 struct scrub_bio *sbio;
492 sbio = kzalloc(sizeof(*sbio), GFP_NOFS);
495 sctx->bios[i] = sbio;
499 sbio->page_count = 0;
500 btrfs_init_work(&sbio->work, btrfs_scrub_helper,
501 scrub_bio_end_io_worker, NULL, NULL);
503 if (i != SCRUB_BIOS_PER_SCTX - 1)
504 sctx->bios[i]->next_free = i + 1;
506 sctx->bios[i]->next_free = -1;
508 sctx->first_free = 0;
509 sctx->nodesize = dev->dev_root->nodesize;
510 sctx->sectorsize = dev->dev_root->sectorsize;
511 atomic_set(&sctx->bios_in_flight, 0);
512 atomic_set(&sctx->workers_pending, 0);
513 atomic_set(&sctx->cancel_req, 0);
514 sctx->csum_size = btrfs_super_csum_size(fs_info->super_copy);
515 INIT_LIST_HEAD(&sctx->csum_list);
517 spin_lock_init(&sctx->list_lock);
518 spin_lock_init(&sctx->stat_lock);
519 init_waitqueue_head(&sctx->list_wait);
521 ret = scrub_setup_wr_ctx(sctx, &sctx->wr_ctx, fs_info,
522 fs_info->dev_replace.tgtdev, is_dev_replace);
524 scrub_free_ctx(sctx);
530 scrub_free_ctx(sctx);
531 return ERR_PTR(-ENOMEM);
534 static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root,
541 struct extent_buffer *eb;
542 struct btrfs_inode_item *inode_item;
543 struct scrub_warning *swarn = warn_ctx;
544 struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info;
545 struct inode_fs_paths *ipath = NULL;
546 struct btrfs_root *local_root;
547 struct btrfs_key root_key;
548 struct btrfs_key key;
550 root_key.objectid = root;
551 root_key.type = BTRFS_ROOT_ITEM_KEY;
552 root_key.offset = (u64)-1;
553 local_root = btrfs_read_fs_root_no_name(fs_info, &root_key);
554 if (IS_ERR(local_root)) {
555 ret = PTR_ERR(local_root);
560 * this makes the path point to (inum INODE_ITEM ioff)
563 key.type = BTRFS_INODE_ITEM_KEY;
566 ret = btrfs_search_slot(NULL, local_root, &key, swarn->path, 0, 0);
568 btrfs_release_path(swarn->path);
572 eb = swarn->path->nodes[0];
573 inode_item = btrfs_item_ptr(eb, swarn->path->slots[0],
574 struct btrfs_inode_item);
575 isize = btrfs_inode_size(eb, inode_item);
576 nlink = btrfs_inode_nlink(eb, inode_item);
577 btrfs_release_path(swarn->path);
579 ipath = init_ipath(4096, local_root, swarn->path);
581 ret = PTR_ERR(ipath);
585 ret = paths_from_inode(inum, ipath);
591 * we deliberately ignore the bit ipath might have been too small to
592 * hold all of the paths here
594 for (i = 0; i < ipath->fspath->elem_cnt; ++i)
595 printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
596 "%s, sector %llu, root %llu, inode %llu, offset %llu, "
597 "length %llu, links %u (path: %s)\n", swarn->errstr,
598 swarn->logical, rcu_str_deref(swarn->dev->name),
599 (unsigned long long)swarn->sector, root, inum, offset,
600 min(isize - offset, (u64)PAGE_SIZE), nlink,
601 (char *)(unsigned long)ipath->fspath->val[i]);
607 printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
608 "%s, sector %llu, root %llu, inode %llu, offset %llu: path "
609 "resolving failed with ret=%d\n", swarn->errstr,
610 swarn->logical, rcu_str_deref(swarn->dev->name),
611 (unsigned long long)swarn->sector, root, inum, offset, ret);
617 static void scrub_print_warning(const char *errstr, struct scrub_block *sblock)
619 struct btrfs_device *dev;
620 struct btrfs_fs_info *fs_info;
621 struct btrfs_path *path;
622 struct btrfs_key found_key;
623 struct extent_buffer *eb;
624 struct btrfs_extent_item *ei;
625 struct scrub_warning swarn;
626 unsigned long ptr = 0;
634 WARN_ON(sblock->page_count < 1);
635 dev = sblock->pagev[0]->dev;
636 fs_info = sblock->sctx->dev_root->fs_info;
638 path = btrfs_alloc_path();
642 swarn.sector = (sblock->pagev[0]->physical) >> 9;
643 swarn.logical = sblock->pagev[0]->logical;
644 swarn.errstr = errstr;
647 ret = extent_from_logical(fs_info, swarn.logical, path, &found_key,
652 extent_item_pos = swarn.logical - found_key.objectid;
653 swarn.extent_item_size = found_key.offset;
656 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
657 item_size = btrfs_item_size_nr(eb, path->slots[0]);
659 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
661 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
662 item_size, &ref_root,
664 printk_in_rcu(KERN_WARNING
665 "BTRFS: %s at logical %llu on dev %s, "
666 "sector %llu: metadata %s (level %d) in tree "
667 "%llu\n", errstr, swarn.logical,
668 rcu_str_deref(dev->name),
669 (unsigned long long)swarn.sector,
670 ref_level ? "node" : "leaf",
671 ret < 0 ? -1 : ref_level,
672 ret < 0 ? -1 : ref_root);
674 btrfs_release_path(path);
676 btrfs_release_path(path);
679 iterate_extent_inodes(fs_info, found_key.objectid,
681 scrub_print_warning_inode, &swarn);
685 btrfs_free_path(path);
688 static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx)
690 struct page *page = NULL;
692 struct scrub_fixup_nodatasum *fixup = fixup_ctx;
695 struct btrfs_key key;
696 struct inode *inode = NULL;
697 struct btrfs_fs_info *fs_info;
698 u64 end = offset + PAGE_SIZE - 1;
699 struct btrfs_root *local_root;
703 key.type = BTRFS_ROOT_ITEM_KEY;
704 key.offset = (u64)-1;
706 fs_info = fixup->root->fs_info;
707 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
709 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
710 if (IS_ERR(local_root)) {
711 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
712 return PTR_ERR(local_root);
715 key.type = BTRFS_INODE_ITEM_KEY;
718 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
719 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
721 return PTR_ERR(inode);
723 index = offset >> PAGE_CACHE_SHIFT;
725 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
731 if (PageUptodate(page)) {
732 if (PageDirty(page)) {
734 * we need to write the data to the defect sector. the
735 * data that was in that sector is not in memory,
736 * because the page was modified. we must not write the
737 * modified page to that sector.
739 * TODO: what could be done here: wait for the delalloc
740 * runner to write out that page (might involve
741 * COW) and see whether the sector is still
742 * referenced afterwards.
744 * For the meantime, we'll treat this error
745 * incorrectable, although there is a chance that a
746 * later scrub will find the bad sector again and that
747 * there's no dirty page in memory, then.
752 ret = repair_io_failure(inode, offset, PAGE_SIZE,
753 fixup->logical, page,
754 offset - page_offset(page),
760 * we need to get good data first. the general readpage path
761 * will call repair_io_failure for us, we just have to make
762 * sure we read the bad mirror.
764 ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
765 EXTENT_DAMAGED, GFP_NOFS);
767 /* set_extent_bits should give proper error */
774 ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page,
777 wait_on_page_locked(page);
779 corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset,
780 end, EXTENT_DAMAGED, 0, NULL);
782 clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
783 EXTENT_DAMAGED, GFP_NOFS);
795 if (ret == 0 && corrected) {
797 * we only need to call readpage for one of the inodes belonging
798 * to this extent. so make iterate_extent_inodes stop
806 static void scrub_fixup_nodatasum(struct btrfs_work *work)
809 struct scrub_fixup_nodatasum *fixup;
810 struct scrub_ctx *sctx;
811 struct btrfs_trans_handle *trans = NULL;
812 struct btrfs_path *path;
813 int uncorrectable = 0;
815 fixup = container_of(work, struct scrub_fixup_nodatasum, work);
818 path = btrfs_alloc_path();
820 spin_lock(&sctx->stat_lock);
821 ++sctx->stat.malloc_errors;
822 spin_unlock(&sctx->stat_lock);
827 trans = btrfs_join_transaction(fixup->root);
834 * the idea is to trigger a regular read through the standard path. we
835 * read a page from the (failed) logical address by specifying the
836 * corresponding copynum of the failed sector. thus, that readpage is
838 * that is the point where on-the-fly error correction will kick in
839 * (once it's finished) and rewrite the failed sector if a good copy
842 ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info,
843 path, scrub_fixup_readpage,
851 spin_lock(&sctx->stat_lock);
852 ++sctx->stat.corrected_errors;
853 spin_unlock(&sctx->stat_lock);
856 if (trans && !IS_ERR(trans))
857 btrfs_end_transaction(trans, fixup->root);
859 spin_lock(&sctx->stat_lock);
860 ++sctx->stat.uncorrectable_errors;
861 spin_unlock(&sctx->stat_lock);
862 btrfs_dev_replace_stats_inc(
863 &sctx->dev_root->fs_info->dev_replace.
864 num_uncorrectable_read_errors);
865 printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
866 "unable to fixup (nodatasum) error at logical %llu on dev %s\n",
867 fixup->logical, rcu_str_deref(fixup->dev->name));
870 btrfs_free_path(path);
873 scrub_pending_trans_workers_dec(sctx);
876 static inline void scrub_get_recover(struct scrub_recover *recover)
878 atomic_inc(&recover->refs);
881 static inline void scrub_put_recover(struct scrub_recover *recover)
883 if (atomic_dec_and_test(&recover->refs)) {
884 btrfs_put_bbio(recover->bbio);
890 * scrub_handle_errored_block gets called when either verification of the
891 * pages failed or the bio failed to read, e.g. with EIO. In the latter
892 * case, this function handles all pages in the bio, even though only one
894 * The goal of this function is to repair the errored block by using the
895 * contents of one of the mirrors.
897 static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
899 struct scrub_ctx *sctx = sblock_to_check->sctx;
900 struct btrfs_device *dev;
901 struct btrfs_fs_info *fs_info;
905 unsigned int failed_mirror_index;
906 unsigned int is_metadata;
907 unsigned int have_csum;
909 struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */
910 struct scrub_block *sblock_bad;
915 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
916 DEFAULT_RATELIMIT_BURST);
918 BUG_ON(sblock_to_check->page_count < 1);
919 fs_info = sctx->dev_root->fs_info;
920 if (sblock_to_check->pagev[0]->flags & BTRFS_EXTENT_FLAG_SUPER) {
922 * if we find an error in a super block, we just report it.
923 * They will get written with the next transaction commit
926 spin_lock(&sctx->stat_lock);
927 ++sctx->stat.super_errors;
928 spin_unlock(&sctx->stat_lock);
931 length = sblock_to_check->page_count * PAGE_SIZE;
932 logical = sblock_to_check->pagev[0]->logical;
933 generation = sblock_to_check->pagev[0]->generation;
934 BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1);
935 failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1;
936 is_metadata = !(sblock_to_check->pagev[0]->flags &
937 BTRFS_EXTENT_FLAG_DATA);
938 have_csum = sblock_to_check->pagev[0]->have_csum;
939 csum = sblock_to_check->pagev[0]->csum;
940 dev = sblock_to_check->pagev[0]->dev;
942 if (sctx->is_dev_replace && !is_metadata && !have_csum) {
943 sblocks_for_recheck = NULL;
948 * read all mirrors one after the other. This includes to
949 * re-read the extent or metadata block that failed (that was
950 * the cause that this fixup code is called) another time,
951 * page by page this time in order to know which pages
952 * caused I/O errors and which ones are good (for all mirrors).
953 * It is the goal to handle the situation when more than one
954 * mirror contains I/O errors, but the errors do not
955 * overlap, i.e. the data can be repaired by selecting the
956 * pages from those mirrors without I/O error on the
957 * particular pages. One example (with blocks >= 2 * PAGE_SIZE)
958 * would be that mirror #1 has an I/O error on the first page,
959 * the second page is good, and mirror #2 has an I/O error on
960 * the second page, but the first page is good.
961 * Then the first page of the first mirror can be repaired by
962 * taking the first page of the second mirror, and the
963 * second page of the second mirror can be repaired by
964 * copying the contents of the 2nd page of the 1st mirror.
965 * One more note: if the pages of one mirror contain I/O
966 * errors, the checksum cannot be verified. In order to get
967 * the best data for repairing, the first attempt is to find
968 * a mirror without I/O errors and with a validated checksum.
969 * Only if this is not possible, the pages are picked from
970 * mirrors with I/O errors without considering the checksum.
971 * If the latter is the case, at the end, the checksum of the
972 * repaired area is verified in order to correctly maintain
976 sblocks_for_recheck = kcalloc(BTRFS_MAX_MIRRORS,
977 sizeof(*sblocks_for_recheck), GFP_NOFS);
978 if (!sblocks_for_recheck) {
979 spin_lock(&sctx->stat_lock);
980 sctx->stat.malloc_errors++;
981 sctx->stat.read_errors++;
982 sctx->stat.uncorrectable_errors++;
983 spin_unlock(&sctx->stat_lock);
984 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
988 /* setup the context, map the logical blocks and alloc the pages */
989 ret = scrub_setup_recheck_block(sblock_to_check, sblocks_for_recheck);
991 spin_lock(&sctx->stat_lock);
992 sctx->stat.read_errors++;
993 sctx->stat.uncorrectable_errors++;
994 spin_unlock(&sctx->stat_lock);
995 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
998 BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS);
999 sblock_bad = sblocks_for_recheck + failed_mirror_index;
1001 /* build and submit the bios for the failed mirror, check checksums */
1002 scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum,
1003 csum, generation, sctx->csum_size, 1);
1005 if (!sblock_bad->header_error && !sblock_bad->checksum_error &&
1006 sblock_bad->no_io_error_seen) {
1008 * the error disappeared after reading page by page, or
1009 * the area was part of a huge bio and other parts of the
1010 * bio caused I/O errors, or the block layer merged several
1011 * read requests into one and the error is caused by a
1012 * different bio (usually one of the two latter cases is
1015 spin_lock(&sctx->stat_lock);
1016 sctx->stat.unverified_errors++;
1017 sblock_to_check->data_corrected = 1;
1018 spin_unlock(&sctx->stat_lock);
1020 if (sctx->is_dev_replace)
1021 scrub_write_block_to_dev_replace(sblock_bad);
1025 if (!sblock_bad->no_io_error_seen) {
1026 spin_lock(&sctx->stat_lock);
1027 sctx->stat.read_errors++;
1028 spin_unlock(&sctx->stat_lock);
1029 if (__ratelimit(&_rs))
1030 scrub_print_warning("i/o error", sblock_to_check);
1031 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
1032 } else if (sblock_bad->checksum_error) {
1033 spin_lock(&sctx->stat_lock);
1034 sctx->stat.csum_errors++;
1035 spin_unlock(&sctx->stat_lock);
1036 if (__ratelimit(&_rs))
1037 scrub_print_warning("checksum error", sblock_to_check);
1038 btrfs_dev_stat_inc_and_print(dev,
1039 BTRFS_DEV_STAT_CORRUPTION_ERRS);
1040 } else if (sblock_bad->header_error) {
1041 spin_lock(&sctx->stat_lock);
1042 sctx->stat.verify_errors++;
1043 spin_unlock(&sctx->stat_lock);
1044 if (__ratelimit(&_rs))
1045 scrub_print_warning("checksum/header error",
1047 if (sblock_bad->generation_error)
1048 btrfs_dev_stat_inc_and_print(dev,
1049 BTRFS_DEV_STAT_GENERATION_ERRS);
1051 btrfs_dev_stat_inc_and_print(dev,
1052 BTRFS_DEV_STAT_CORRUPTION_ERRS);
1055 if (sctx->readonly) {
1056 ASSERT(!sctx->is_dev_replace);
1060 if (!is_metadata && !have_csum) {
1061 struct scrub_fixup_nodatasum *fixup_nodatasum;
1063 WARN_ON(sctx->is_dev_replace);
1068 * !is_metadata and !have_csum, this means that the data
1069 * might not be COW'ed, that it might be modified
1070 * concurrently. The general strategy to work on the
1071 * commit root does not help in the case when COW is not
1074 fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS);
1075 if (!fixup_nodatasum)
1076 goto did_not_correct_error;
1077 fixup_nodatasum->sctx = sctx;
1078 fixup_nodatasum->dev = dev;
1079 fixup_nodatasum->logical = logical;
1080 fixup_nodatasum->root = fs_info->extent_root;
1081 fixup_nodatasum->mirror_num = failed_mirror_index + 1;
1082 scrub_pending_trans_workers_inc(sctx);
1083 btrfs_init_work(&fixup_nodatasum->work, btrfs_scrub_helper,
1084 scrub_fixup_nodatasum, NULL, NULL);
1085 btrfs_queue_work(fs_info->scrub_workers,
1086 &fixup_nodatasum->work);
1091 * now build and submit the bios for the other mirrors, check
1093 * First try to pick the mirror which is completely without I/O
1094 * errors and also does not have a checksum error.
1095 * If one is found, and if a checksum is present, the full block
1096 * that is known to contain an error is rewritten. Afterwards
1097 * the block is known to be corrected.
1098 * If a mirror is found which is completely correct, and no
1099 * checksum is present, only those pages are rewritten that had
1100 * an I/O error in the block to be repaired, since it cannot be
1101 * determined, which copy of the other pages is better (and it
1102 * could happen otherwise that a correct page would be
1103 * overwritten by a bad one).
1105 for (mirror_index = 0;
1106 mirror_index < BTRFS_MAX_MIRRORS &&
1107 sblocks_for_recheck[mirror_index].page_count > 0;
1109 struct scrub_block *sblock_other;
1111 if (mirror_index == failed_mirror_index)
1113 sblock_other = sblocks_for_recheck + mirror_index;
1115 /* build and submit the bios, check checksums */
1116 scrub_recheck_block(fs_info, sblock_other, is_metadata,
1117 have_csum, csum, generation,
1118 sctx->csum_size, 0);
1120 if (!sblock_other->header_error &&
1121 !sblock_other->checksum_error &&
1122 sblock_other->no_io_error_seen) {
1123 if (sctx->is_dev_replace) {
1124 scrub_write_block_to_dev_replace(sblock_other);
1125 goto corrected_error;
1127 ret = scrub_repair_block_from_good_copy(
1128 sblock_bad, sblock_other);
1130 goto corrected_error;
1135 if (sblock_bad->no_io_error_seen && !sctx->is_dev_replace)
1136 goto did_not_correct_error;
1139 * In case of I/O errors in the area that is supposed to be
1140 * repaired, continue by picking good copies of those pages.
1141 * Select the good pages from mirrors to rewrite bad pages from
1142 * the area to fix. Afterwards verify the checksum of the block
1143 * that is supposed to be repaired. This verification step is
1144 * only done for the purpose of statistic counting and for the
1145 * final scrub report, whether errors remain.
1146 * A perfect algorithm could make use of the checksum and try
1147 * all possible combinations of pages from the different mirrors
1148 * until the checksum verification succeeds. For example, when
1149 * the 2nd page of mirror #1 faces I/O errors, and the 2nd page
1150 * of mirror #2 is readable but the final checksum test fails,
1151 * then the 2nd page of mirror #3 could be tried, whether now
1152 * the final checksum succeedes. But this would be a rare
1153 * exception and is therefore not implemented. At least it is
1154 * avoided that the good copy is overwritten.
1155 * A more useful improvement would be to pick the sectors
1156 * without I/O error based on sector sizes (512 bytes on legacy
1157 * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one
1158 * mirror could be repaired by taking 512 byte of a different
1159 * mirror, even if other 512 byte sectors in the same PAGE_SIZE
1160 * area are unreadable.
1163 for (page_num = 0; page_num < sblock_bad->page_count;
1165 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1166 struct scrub_block *sblock_other = NULL;
1168 /* skip no-io-error page in scrub */
1169 if (!page_bad->io_error && !sctx->is_dev_replace)
1172 /* try to find no-io-error page in mirrors */
1173 if (page_bad->io_error) {
1174 for (mirror_index = 0;
1175 mirror_index < BTRFS_MAX_MIRRORS &&
1176 sblocks_for_recheck[mirror_index].page_count > 0;
1178 if (!sblocks_for_recheck[mirror_index].
1179 pagev[page_num]->io_error) {
1180 sblock_other = sblocks_for_recheck +
1189 if (sctx->is_dev_replace) {
1191 * did not find a mirror to fetch the page
1192 * from. scrub_write_page_to_dev_replace()
1193 * handles this case (page->io_error), by
1194 * filling the block with zeros before
1195 * submitting the write request
1198 sblock_other = sblock_bad;
1200 if (scrub_write_page_to_dev_replace(sblock_other,
1202 btrfs_dev_replace_stats_inc(
1204 fs_info->dev_replace.
1208 } else if (sblock_other) {
1209 ret = scrub_repair_page_from_good_copy(sblock_bad,
1213 page_bad->io_error = 0;
1219 if (success && !sctx->is_dev_replace) {
1220 if (is_metadata || have_csum) {
1222 * need to verify the checksum now that all
1223 * sectors on disk are repaired (the write
1224 * request for data to be repaired is on its way).
1225 * Just be lazy and use scrub_recheck_block()
1226 * which re-reads the data before the checksum
1227 * is verified, but most likely the data comes out
1228 * of the page cache.
1230 scrub_recheck_block(fs_info, sblock_bad,
1231 is_metadata, have_csum, csum,
1232 generation, sctx->csum_size, 1);
1233 if (!sblock_bad->header_error &&
1234 !sblock_bad->checksum_error &&
1235 sblock_bad->no_io_error_seen)
1236 goto corrected_error;
1238 goto did_not_correct_error;
1241 spin_lock(&sctx->stat_lock);
1242 sctx->stat.corrected_errors++;
1243 sblock_to_check->data_corrected = 1;
1244 spin_unlock(&sctx->stat_lock);
1245 printk_ratelimited_in_rcu(KERN_ERR
1246 "BTRFS: fixed up error at logical %llu on dev %s\n",
1247 logical, rcu_str_deref(dev->name));
1250 did_not_correct_error:
1251 spin_lock(&sctx->stat_lock);
1252 sctx->stat.uncorrectable_errors++;
1253 spin_unlock(&sctx->stat_lock);
1254 printk_ratelimited_in_rcu(KERN_ERR
1255 "BTRFS: unable to fixup (regular) error at logical %llu on dev %s\n",
1256 logical, rcu_str_deref(dev->name));
1260 if (sblocks_for_recheck) {
1261 for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS;
1263 struct scrub_block *sblock = sblocks_for_recheck +
1265 struct scrub_recover *recover;
1268 for (page_index = 0; page_index < sblock->page_count;
1270 sblock->pagev[page_index]->sblock = NULL;
1271 recover = sblock->pagev[page_index]->recover;
1273 scrub_put_recover(recover);
1274 sblock->pagev[page_index]->recover =
1277 scrub_page_put(sblock->pagev[page_index]);
1280 kfree(sblocks_for_recheck);
1286 static inline int scrub_nr_raid_mirrors(struct btrfs_bio *bbio)
1288 if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID5)
1290 else if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID6)
1293 return (int)bbio->num_stripes;
1296 static inline void scrub_stripe_index_and_offset(u64 logical, u64 map_type,
1299 int nstripes, int mirror,
1305 if (map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
1307 for (i = 0; i < nstripes; i++) {
1308 if (raid_map[i] == RAID6_Q_STRIPE ||
1309 raid_map[i] == RAID5_P_STRIPE)
1312 if (logical >= raid_map[i] &&
1313 logical < raid_map[i] + mapped_length)
1318 *stripe_offset = logical - raid_map[i];
1320 /* The other RAID type */
1321 *stripe_index = mirror;
1326 static int scrub_setup_recheck_block(struct scrub_block *original_sblock,
1327 struct scrub_block *sblocks_for_recheck)
1329 struct scrub_ctx *sctx = original_sblock->sctx;
1330 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
1331 u64 length = original_sblock->page_count * PAGE_SIZE;
1332 u64 logical = original_sblock->pagev[0]->logical;
1333 struct scrub_recover *recover;
1334 struct btrfs_bio *bbio;
1345 * note: the two members refs and outstanding_pages
1346 * are not used (and not set) in the blocks that are used for
1347 * the recheck procedure
1350 while (length > 0) {
1351 sublen = min_t(u64, length, PAGE_SIZE);
1352 mapped_length = sublen;
1356 * with a length of PAGE_SIZE, each returned stripe
1357 * represents one mirror
1359 ret = btrfs_map_sblock(fs_info, REQ_GET_READ_MIRRORS, logical,
1360 &mapped_length, &bbio, 0, 1);
1361 if (ret || !bbio || mapped_length < sublen) {
1362 btrfs_put_bbio(bbio);
1366 recover = kzalloc(sizeof(struct scrub_recover), GFP_NOFS);
1368 btrfs_put_bbio(bbio);
1372 atomic_set(&recover->refs, 1);
1373 recover->bbio = bbio;
1374 recover->map_length = mapped_length;
1376 BUG_ON(page_index >= SCRUB_PAGES_PER_RD_BIO);
1378 nmirrors = min(scrub_nr_raid_mirrors(bbio), BTRFS_MAX_MIRRORS);
1380 for (mirror_index = 0; mirror_index < nmirrors;
1382 struct scrub_block *sblock;
1383 struct scrub_page *page;
1385 sblock = sblocks_for_recheck + mirror_index;
1386 sblock->sctx = sctx;
1387 page = kzalloc(sizeof(*page), GFP_NOFS);
1390 spin_lock(&sctx->stat_lock);
1391 sctx->stat.malloc_errors++;
1392 spin_unlock(&sctx->stat_lock);
1393 scrub_put_recover(recover);
1396 scrub_page_get(page);
1397 sblock->pagev[page_index] = page;
1398 page->logical = logical;
1400 scrub_stripe_index_and_offset(logical,
1409 page->physical = bbio->stripes[stripe_index].physical +
1411 page->dev = bbio->stripes[stripe_index].dev;
1413 BUG_ON(page_index >= original_sblock->page_count);
1414 page->physical_for_dev_replace =
1415 original_sblock->pagev[page_index]->
1416 physical_for_dev_replace;
1417 /* for missing devices, dev->bdev is NULL */
1418 page->mirror_num = mirror_index + 1;
1419 sblock->page_count++;
1420 page->page = alloc_page(GFP_NOFS);
1424 scrub_get_recover(recover);
1425 page->recover = recover;
1427 scrub_put_recover(recover);
1436 struct scrub_bio_ret {
1437 struct completion event;
1441 static void scrub_bio_wait_endio(struct bio *bio, int error)
1443 struct scrub_bio_ret *ret = bio->bi_private;
1446 complete(&ret->event);
1449 static inline int scrub_is_page_on_raid56(struct scrub_page *page)
1451 return page->recover &&
1452 (page->recover->bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK);
1455 static int scrub_submit_raid56_bio_wait(struct btrfs_fs_info *fs_info,
1457 struct scrub_page *page)
1459 struct scrub_bio_ret done;
1462 init_completion(&done.event);
1464 bio->bi_iter.bi_sector = page->logical >> 9;
1465 bio->bi_private = &done;
1466 bio->bi_end_io = scrub_bio_wait_endio;
1468 ret = raid56_parity_recover(fs_info->fs_root, bio, page->recover->bbio,
1469 page->recover->map_length,
1470 page->mirror_num, 0);
1474 wait_for_completion(&done.event);
1482 * this function will check the on disk data for checksum errors, header
1483 * errors and read I/O errors. If any I/O errors happen, the exact pages
1484 * which are errored are marked as being bad. The goal is to enable scrub
1485 * to take those pages that are not errored from all the mirrors so that
1486 * the pages that are errored in the just handled mirror can be repaired.
1488 static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
1489 struct scrub_block *sblock, int is_metadata,
1490 int have_csum, u8 *csum, u64 generation,
1491 u16 csum_size, int retry_failed_mirror)
1495 sblock->no_io_error_seen = 1;
1496 sblock->header_error = 0;
1497 sblock->checksum_error = 0;
1499 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1501 struct scrub_page *page = sblock->pagev[page_num];
1503 if (page->dev->bdev == NULL) {
1505 sblock->no_io_error_seen = 0;
1509 WARN_ON(!page->page);
1510 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1513 sblock->no_io_error_seen = 0;
1516 bio->bi_bdev = page->dev->bdev;
1518 bio_add_page(bio, page->page, PAGE_SIZE, 0);
1519 if (!retry_failed_mirror && scrub_is_page_on_raid56(page)) {
1520 if (scrub_submit_raid56_bio_wait(fs_info, bio, page))
1521 sblock->no_io_error_seen = 0;
1523 bio->bi_iter.bi_sector = page->physical >> 9;
1525 if (btrfsic_submit_bio_wait(READ, bio))
1526 sblock->no_io_error_seen = 0;
1532 if (sblock->no_io_error_seen)
1533 scrub_recheck_block_checksum(fs_info, sblock, is_metadata,
1534 have_csum, csum, generation,
1540 static inline int scrub_check_fsid(u8 fsid[],
1541 struct scrub_page *spage)
1543 struct btrfs_fs_devices *fs_devices = spage->dev->fs_devices;
1546 ret = memcmp(fsid, fs_devices->fsid, BTRFS_UUID_SIZE);
1550 static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
1551 struct scrub_block *sblock,
1552 int is_metadata, int have_csum,
1553 const u8 *csum, u64 generation,
1557 u8 calculated_csum[BTRFS_CSUM_SIZE];
1559 void *mapped_buffer;
1561 WARN_ON(!sblock->pagev[0]->page);
1563 struct btrfs_header *h;
1565 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
1566 h = (struct btrfs_header *)mapped_buffer;
1568 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h) ||
1569 !scrub_check_fsid(h->fsid, sblock->pagev[0]) ||
1570 memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
1572 sblock->header_error = 1;
1573 } else if (generation != btrfs_stack_header_generation(h)) {
1574 sblock->header_error = 1;
1575 sblock->generation_error = 1;
1582 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
1585 for (page_num = 0;;) {
1586 if (page_num == 0 && is_metadata)
1587 crc = btrfs_csum_data(
1588 ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE,
1589 crc, PAGE_SIZE - BTRFS_CSUM_SIZE);
1591 crc = btrfs_csum_data(mapped_buffer, crc, PAGE_SIZE);
1593 kunmap_atomic(mapped_buffer);
1595 if (page_num >= sblock->page_count)
1597 WARN_ON(!sblock->pagev[page_num]->page);
1599 mapped_buffer = kmap_atomic(sblock->pagev[page_num]->page);
1602 btrfs_csum_final(crc, calculated_csum);
1603 if (memcmp(calculated_csum, csum, csum_size))
1604 sblock->checksum_error = 1;
1607 static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
1608 struct scrub_block *sblock_good)
1613 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
1616 ret_sub = scrub_repair_page_from_good_copy(sblock_bad,
1626 static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
1627 struct scrub_block *sblock_good,
1628 int page_num, int force_write)
1630 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1631 struct scrub_page *page_good = sblock_good->pagev[page_num];
1633 BUG_ON(page_bad->page == NULL);
1634 BUG_ON(page_good->page == NULL);
1635 if (force_write || sblock_bad->header_error ||
1636 sblock_bad->checksum_error || page_bad->io_error) {
1640 if (!page_bad->dev->bdev) {
1641 printk_ratelimited(KERN_WARNING "BTRFS: "
1642 "scrub_repair_page_from_good_copy(bdev == NULL) "
1643 "is unexpected!\n");
1647 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1650 bio->bi_bdev = page_bad->dev->bdev;
1651 bio->bi_iter.bi_sector = page_bad->physical >> 9;
1653 ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0);
1654 if (PAGE_SIZE != ret) {
1659 if (btrfsic_submit_bio_wait(WRITE, bio)) {
1660 btrfs_dev_stat_inc_and_print(page_bad->dev,
1661 BTRFS_DEV_STAT_WRITE_ERRS);
1662 btrfs_dev_replace_stats_inc(
1663 &sblock_bad->sctx->dev_root->fs_info->
1664 dev_replace.num_write_errors);
1674 static void scrub_write_block_to_dev_replace(struct scrub_block *sblock)
1679 * This block is used for the check of the parity on the source device,
1680 * so the data needn't be written into the destination device.
1682 if (sblock->sparity)
1685 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1688 ret = scrub_write_page_to_dev_replace(sblock, page_num);
1690 btrfs_dev_replace_stats_inc(
1691 &sblock->sctx->dev_root->fs_info->dev_replace.
1696 static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
1699 struct scrub_page *spage = sblock->pagev[page_num];
1701 BUG_ON(spage->page == NULL);
1702 if (spage->io_error) {
1703 void *mapped_buffer = kmap_atomic(spage->page);
1705 memset(mapped_buffer, 0, PAGE_CACHE_SIZE);
1706 flush_dcache_page(spage->page);
1707 kunmap_atomic(mapped_buffer);
1709 return scrub_add_page_to_wr_bio(sblock->sctx, spage);
1712 static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
1713 struct scrub_page *spage)
1715 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1716 struct scrub_bio *sbio;
1719 mutex_lock(&wr_ctx->wr_lock);
1721 if (!wr_ctx->wr_curr_bio) {
1722 wr_ctx->wr_curr_bio = kzalloc(sizeof(*wr_ctx->wr_curr_bio),
1724 if (!wr_ctx->wr_curr_bio) {
1725 mutex_unlock(&wr_ctx->wr_lock);
1728 wr_ctx->wr_curr_bio->sctx = sctx;
1729 wr_ctx->wr_curr_bio->page_count = 0;
1731 sbio = wr_ctx->wr_curr_bio;
1732 if (sbio->page_count == 0) {
1735 sbio->physical = spage->physical_for_dev_replace;
1736 sbio->logical = spage->logical;
1737 sbio->dev = wr_ctx->tgtdev;
1740 bio = btrfs_io_bio_alloc(GFP_NOFS, wr_ctx->pages_per_wr_bio);
1742 mutex_unlock(&wr_ctx->wr_lock);
1748 bio->bi_private = sbio;
1749 bio->bi_end_io = scrub_wr_bio_end_io;
1750 bio->bi_bdev = sbio->dev->bdev;
1751 bio->bi_iter.bi_sector = sbio->physical >> 9;
1753 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
1754 spage->physical_for_dev_replace ||
1755 sbio->logical + sbio->page_count * PAGE_SIZE !=
1757 scrub_wr_submit(sctx);
1761 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
1762 if (ret != PAGE_SIZE) {
1763 if (sbio->page_count < 1) {
1766 mutex_unlock(&wr_ctx->wr_lock);
1769 scrub_wr_submit(sctx);
1773 sbio->pagev[sbio->page_count] = spage;
1774 scrub_page_get(spage);
1776 if (sbio->page_count == wr_ctx->pages_per_wr_bio)
1777 scrub_wr_submit(sctx);
1778 mutex_unlock(&wr_ctx->wr_lock);
1783 static void scrub_wr_submit(struct scrub_ctx *sctx)
1785 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1786 struct scrub_bio *sbio;
1788 if (!wr_ctx->wr_curr_bio)
1791 sbio = wr_ctx->wr_curr_bio;
1792 wr_ctx->wr_curr_bio = NULL;
1793 WARN_ON(!sbio->bio->bi_bdev);
1794 scrub_pending_bio_inc(sctx);
1795 /* process all writes in a single worker thread. Then the block layer
1796 * orders the requests before sending them to the driver which
1797 * doubled the write performance on spinning disks when measured
1799 btrfsic_submit_bio(WRITE, sbio->bio);
1802 static void scrub_wr_bio_end_io(struct bio *bio, int err)
1804 struct scrub_bio *sbio = bio->bi_private;
1805 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
1810 btrfs_init_work(&sbio->work, btrfs_scrubwrc_helper,
1811 scrub_wr_bio_end_io_worker, NULL, NULL);
1812 btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work);
1815 static void scrub_wr_bio_end_io_worker(struct btrfs_work *work)
1817 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
1818 struct scrub_ctx *sctx = sbio->sctx;
1821 WARN_ON(sbio->page_count > SCRUB_PAGES_PER_WR_BIO);
1823 struct btrfs_dev_replace *dev_replace =
1824 &sbio->sctx->dev_root->fs_info->dev_replace;
1826 for (i = 0; i < sbio->page_count; i++) {
1827 struct scrub_page *spage = sbio->pagev[i];
1829 spage->io_error = 1;
1830 btrfs_dev_replace_stats_inc(&dev_replace->
1835 for (i = 0; i < sbio->page_count; i++)
1836 scrub_page_put(sbio->pagev[i]);
1840 scrub_pending_bio_dec(sctx);
1843 static int scrub_checksum(struct scrub_block *sblock)
1848 WARN_ON(sblock->page_count < 1);
1849 flags = sblock->pagev[0]->flags;
1851 if (flags & BTRFS_EXTENT_FLAG_DATA)
1852 ret = scrub_checksum_data(sblock);
1853 else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1854 ret = scrub_checksum_tree_block(sblock);
1855 else if (flags & BTRFS_EXTENT_FLAG_SUPER)
1856 (void)scrub_checksum_super(sblock);
1860 scrub_handle_errored_block(sblock);
1865 static int scrub_checksum_data(struct scrub_block *sblock)
1867 struct scrub_ctx *sctx = sblock->sctx;
1868 u8 csum[BTRFS_CSUM_SIZE];
1877 BUG_ON(sblock->page_count < 1);
1878 if (!sblock->pagev[0]->have_csum)
1881 on_disk_csum = sblock->pagev[0]->csum;
1882 page = sblock->pagev[0]->page;
1883 buffer = kmap_atomic(page);
1885 len = sctx->sectorsize;
1888 u64 l = min_t(u64, len, PAGE_SIZE);
1890 crc = btrfs_csum_data(buffer, crc, l);
1891 kunmap_atomic(buffer);
1896 BUG_ON(index >= sblock->page_count);
1897 BUG_ON(!sblock->pagev[index]->page);
1898 page = sblock->pagev[index]->page;
1899 buffer = kmap_atomic(page);
1902 btrfs_csum_final(crc, csum);
1903 if (memcmp(csum, on_disk_csum, sctx->csum_size))
1909 static int scrub_checksum_tree_block(struct scrub_block *sblock)
1911 struct scrub_ctx *sctx = sblock->sctx;
1912 struct btrfs_header *h;
1913 struct btrfs_root *root = sctx->dev_root;
1914 struct btrfs_fs_info *fs_info = root->fs_info;
1915 u8 calculated_csum[BTRFS_CSUM_SIZE];
1916 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1918 void *mapped_buffer;
1927 BUG_ON(sblock->page_count < 1);
1928 page = sblock->pagev[0]->page;
1929 mapped_buffer = kmap_atomic(page);
1930 h = (struct btrfs_header *)mapped_buffer;
1931 memcpy(on_disk_csum, h->csum, sctx->csum_size);
1934 * we don't use the getter functions here, as we
1935 * a) don't have an extent buffer and
1936 * b) the page is already kmapped
1939 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h))
1942 if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h))
1945 if (!scrub_check_fsid(h->fsid, sblock->pagev[0]))
1948 if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
1952 len = sctx->nodesize - BTRFS_CSUM_SIZE;
1953 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
1954 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
1957 u64 l = min_t(u64, len, mapped_size);
1959 crc = btrfs_csum_data(p, crc, l);
1960 kunmap_atomic(mapped_buffer);
1965 BUG_ON(index >= sblock->page_count);
1966 BUG_ON(!sblock->pagev[index]->page);
1967 page = sblock->pagev[index]->page;
1968 mapped_buffer = kmap_atomic(page);
1969 mapped_size = PAGE_SIZE;
1973 btrfs_csum_final(crc, calculated_csum);
1974 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
1977 return fail || crc_fail;
1980 static int scrub_checksum_super(struct scrub_block *sblock)
1982 struct btrfs_super_block *s;
1983 struct scrub_ctx *sctx = sblock->sctx;
1984 u8 calculated_csum[BTRFS_CSUM_SIZE];
1985 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1987 void *mapped_buffer;
1996 BUG_ON(sblock->page_count < 1);
1997 page = sblock->pagev[0]->page;
1998 mapped_buffer = kmap_atomic(page);
1999 s = (struct btrfs_super_block *)mapped_buffer;
2000 memcpy(on_disk_csum, s->csum, sctx->csum_size);
2002 if (sblock->pagev[0]->logical != btrfs_super_bytenr(s))
2005 if (sblock->pagev[0]->generation != btrfs_super_generation(s))
2008 if (!scrub_check_fsid(s->fsid, sblock->pagev[0]))
2011 len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE;
2012 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
2013 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
2016 u64 l = min_t(u64, len, mapped_size);
2018 crc = btrfs_csum_data(p, crc, l);
2019 kunmap_atomic(mapped_buffer);
2024 BUG_ON(index >= sblock->page_count);
2025 BUG_ON(!sblock->pagev[index]->page);
2026 page = sblock->pagev[index]->page;
2027 mapped_buffer = kmap_atomic(page);
2028 mapped_size = PAGE_SIZE;
2032 btrfs_csum_final(crc, calculated_csum);
2033 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
2036 if (fail_cor + fail_gen) {
2038 * if we find an error in a super block, we just report it.
2039 * They will get written with the next transaction commit
2042 spin_lock(&sctx->stat_lock);
2043 ++sctx->stat.super_errors;
2044 spin_unlock(&sctx->stat_lock);
2046 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
2047 BTRFS_DEV_STAT_CORRUPTION_ERRS);
2049 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
2050 BTRFS_DEV_STAT_GENERATION_ERRS);
2053 return fail_cor + fail_gen;
2056 static void scrub_block_get(struct scrub_block *sblock)
2058 atomic_inc(&sblock->refs);
2061 static void scrub_block_put(struct scrub_block *sblock)
2063 if (atomic_dec_and_test(&sblock->refs)) {
2066 if (sblock->sparity)
2067 scrub_parity_put(sblock->sparity);
2069 for (i = 0; i < sblock->page_count; i++)
2070 scrub_page_put(sblock->pagev[i]);
2075 static void scrub_page_get(struct scrub_page *spage)
2077 atomic_inc(&spage->refs);
2080 static void scrub_page_put(struct scrub_page *spage)
2082 if (atomic_dec_and_test(&spage->refs)) {
2084 __free_page(spage->page);
2089 static void scrub_submit(struct scrub_ctx *sctx)
2091 struct scrub_bio *sbio;
2093 if (sctx->curr == -1)
2096 sbio = sctx->bios[sctx->curr];
2098 scrub_pending_bio_inc(sctx);
2099 btrfsic_submit_bio(READ, sbio->bio);
2102 static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
2103 struct scrub_page *spage)
2105 struct scrub_block *sblock = spage->sblock;
2106 struct scrub_bio *sbio;
2111 * grab a fresh bio or wait for one to become available
2113 while (sctx->curr == -1) {
2114 spin_lock(&sctx->list_lock);
2115 sctx->curr = sctx->first_free;
2116 if (sctx->curr != -1) {
2117 sctx->first_free = sctx->bios[sctx->curr]->next_free;
2118 sctx->bios[sctx->curr]->next_free = -1;
2119 sctx->bios[sctx->curr]->page_count = 0;
2120 spin_unlock(&sctx->list_lock);
2122 spin_unlock(&sctx->list_lock);
2123 wait_event(sctx->list_wait, sctx->first_free != -1);
2126 sbio = sctx->bios[sctx->curr];
2127 if (sbio->page_count == 0) {
2130 sbio->physical = spage->physical;
2131 sbio->logical = spage->logical;
2132 sbio->dev = spage->dev;
2135 bio = btrfs_io_bio_alloc(GFP_NOFS, sctx->pages_per_rd_bio);
2141 bio->bi_private = sbio;
2142 bio->bi_end_io = scrub_bio_end_io;
2143 bio->bi_bdev = sbio->dev->bdev;
2144 bio->bi_iter.bi_sector = sbio->physical >> 9;
2146 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
2148 sbio->logical + sbio->page_count * PAGE_SIZE !=
2150 sbio->dev != spage->dev) {
2155 sbio->pagev[sbio->page_count] = spage;
2156 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
2157 if (ret != PAGE_SIZE) {
2158 if (sbio->page_count < 1) {
2167 scrub_block_get(sblock); /* one for the page added to the bio */
2168 atomic_inc(&sblock->outstanding_pages);
2170 if (sbio->page_count == sctx->pages_per_rd_bio)
2176 static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
2177 u64 physical, struct btrfs_device *dev, u64 flags,
2178 u64 gen, int mirror_num, u8 *csum, int force,
2179 u64 physical_for_dev_replace)
2181 struct scrub_block *sblock;
2184 sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
2186 spin_lock(&sctx->stat_lock);
2187 sctx->stat.malloc_errors++;
2188 spin_unlock(&sctx->stat_lock);
2192 /* one ref inside this function, plus one for each page added to
2194 atomic_set(&sblock->refs, 1);
2195 sblock->sctx = sctx;
2196 sblock->no_io_error_seen = 1;
2198 for (index = 0; len > 0; index++) {
2199 struct scrub_page *spage;
2200 u64 l = min_t(u64, len, PAGE_SIZE);
2202 spage = kzalloc(sizeof(*spage), GFP_NOFS);
2205 spin_lock(&sctx->stat_lock);
2206 sctx->stat.malloc_errors++;
2207 spin_unlock(&sctx->stat_lock);
2208 scrub_block_put(sblock);
2211 BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
2212 scrub_page_get(spage);
2213 sblock->pagev[index] = spage;
2214 spage->sblock = sblock;
2216 spage->flags = flags;
2217 spage->generation = gen;
2218 spage->logical = logical;
2219 spage->physical = physical;
2220 spage->physical_for_dev_replace = physical_for_dev_replace;
2221 spage->mirror_num = mirror_num;
2223 spage->have_csum = 1;
2224 memcpy(spage->csum, csum, sctx->csum_size);
2226 spage->have_csum = 0;
2228 sblock->page_count++;
2229 spage->page = alloc_page(GFP_NOFS);
2235 physical_for_dev_replace += l;
2238 WARN_ON(sblock->page_count == 0);
2239 for (index = 0; index < sblock->page_count; index++) {
2240 struct scrub_page *spage = sblock->pagev[index];
2243 ret = scrub_add_page_to_rd_bio(sctx, spage);
2245 scrub_block_put(sblock);
2253 /* last one frees, either here or in bio completion for last page */
2254 scrub_block_put(sblock);
2258 static void scrub_bio_end_io(struct bio *bio, int err)
2260 struct scrub_bio *sbio = bio->bi_private;
2261 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
2266 btrfs_queue_work(fs_info->scrub_workers, &sbio->work);
2269 static void scrub_bio_end_io_worker(struct btrfs_work *work)
2271 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
2272 struct scrub_ctx *sctx = sbio->sctx;
2275 BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO);
2277 for (i = 0; i < sbio->page_count; i++) {
2278 struct scrub_page *spage = sbio->pagev[i];
2280 spage->io_error = 1;
2281 spage->sblock->no_io_error_seen = 0;
2285 /* now complete the scrub_block items that have all pages completed */
2286 for (i = 0; i < sbio->page_count; i++) {
2287 struct scrub_page *spage = sbio->pagev[i];
2288 struct scrub_block *sblock = spage->sblock;
2290 if (atomic_dec_and_test(&sblock->outstanding_pages))
2291 scrub_block_complete(sblock);
2292 scrub_block_put(sblock);
2297 spin_lock(&sctx->list_lock);
2298 sbio->next_free = sctx->first_free;
2299 sctx->first_free = sbio->index;
2300 spin_unlock(&sctx->list_lock);
2302 if (sctx->is_dev_replace &&
2303 atomic_read(&sctx->wr_ctx.flush_all_writes)) {
2304 mutex_lock(&sctx->wr_ctx.wr_lock);
2305 scrub_wr_submit(sctx);
2306 mutex_unlock(&sctx->wr_ctx.wr_lock);
2309 scrub_pending_bio_dec(sctx);
2312 static inline void __scrub_mark_bitmap(struct scrub_parity *sparity,
2313 unsigned long *bitmap,
2318 int sectorsize = sparity->sctx->dev_root->sectorsize;
2320 if (len >= sparity->stripe_len) {
2321 bitmap_set(bitmap, 0, sparity->nsectors);
2325 start -= sparity->logic_start;
2326 start = div_u64_rem(start, sparity->stripe_len, &offset);
2327 offset /= sectorsize;
2328 nsectors = (int)len / sectorsize;
2330 if (offset + nsectors <= sparity->nsectors) {
2331 bitmap_set(bitmap, offset, nsectors);
2335 bitmap_set(bitmap, offset, sparity->nsectors - offset);
2336 bitmap_set(bitmap, 0, nsectors - (sparity->nsectors - offset));
2339 static inline void scrub_parity_mark_sectors_error(struct scrub_parity *sparity,
2342 __scrub_mark_bitmap(sparity, sparity->ebitmap, start, len);
2345 static inline void scrub_parity_mark_sectors_data(struct scrub_parity *sparity,
2348 __scrub_mark_bitmap(sparity, sparity->dbitmap, start, len);
2351 static void scrub_block_complete(struct scrub_block *sblock)
2355 if (!sblock->no_io_error_seen) {
2357 scrub_handle_errored_block(sblock);
2360 * if has checksum error, write via repair mechanism in
2361 * dev replace case, otherwise write here in dev replace
2364 corrupted = scrub_checksum(sblock);
2365 if (!corrupted && sblock->sctx->is_dev_replace)
2366 scrub_write_block_to_dev_replace(sblock);
2369 if (sblock->sparity && corrupted && !sblock->data_corrected) {
2370 u64 start = sblock->pagev[0]->logical;
2371 u64 end = sblock->pagev[sblock->page_count - 1]->logical +
2374 scrub_parity_mark_sectors_error(sblock->sparity,
2375 start, end - start);
2379 static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len,
2382 struct btrfs_ordered_sum *sum = NULL;
2383 unsigned long index;
2384 unsigned long num_sectors;
2386 while (!list_empty(&sctx->csum_list)) {
2387 sum = list_first_entry(&sctx->csum_list,
2388 struct btrfs_ordered_sum, list);
2389 if (sum->bytenr > logical)
2391 if (sum->bytenr + sum->len > logical)
2394 ++sctx->stat.csum_discards;
2395 list_del(&sum->list);
2402 index = ((u32)(logical - sum->bytenr)) / sctx->sectorsize;
2403 num_sectors = sum->len / sctx->sectorsize;
2404 memcpy(csum, sum->sums + index, sctx->csum_size);
2405 if (index == num_sectors - 1) {
2406 list_del(&sum->list);
2412 /* scrub extent tries to collect up to 64 kB for each bio */
2413 static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len,
2414 u64 physical, struct btrfs_device *dev, u64 flags,
2415 u64 gen, int mirror_num, u64 physical_for_dev_replace)
2418 u8 csum[BTRFS_CSUM_SIZE];
2421 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2422 blocksize = sctx->sectorsize;
2423 spin_lock(&sctx->stat_lock);
2424 sctx->stat.data_extents_scrubbed++;
2425 sctx->stat.data_bytes_scrubbed += len;
2426 spin_unlock(&sctx->stat_lock);
2427 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2428 blocksize = sctx->nodesize;
2429 spin_lock(&sctx->stat_lock);
2430 sctx->stat.tree_extents_scrubbed++;
2431 sctx->stat.tree_bytes_scrubbed += len;
2432 spin_unlock(&sctx->stat_lock);
2434 blocksize = sctx->sectorsize;
2439 u64 l = min_t(u64, len, blocksize);
2442 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2443 /* push csums to sbio */
2444 have_csum = scrub_find_csum(sctx, logical, l, csum);
2446 ++sctx->stat.no_csum;
2447 if (sctx->is_dev_replace && !have_csum) {
2448 ret = copy_nocow_pages(sctx, logical, l,
2450 physical_for_dev_replace);
2451 goto behind_scrub_pages;
2454 ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen,
2455 mirror_num, have_csum ? csum : NULL, 0,
2456 physical_for_dev_replace);
2463 physical_for_dev_replace += l;
2468 static int scrub_pages_for_parity(struct scrub_parity *sparity,
2469 u64 logical, u64 len,
2470 u64 physical, struct btrfs_device *dev,
2471 u64 flags, u64 gen, int mirror_num, u8 *csum)
2473 struct scrub_ctx *sctx = sparity->sctx;
2474 struct scrub_block *sblock;
2477 sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
2479 spin_lock(&sctx->stat_lock);
2480 sctx->stat.malloc_errors++;
2481 spin_unlock(&sctx->stat_lock);
2485 /* one ref inside this function, plus one for each page added to
2487 atomic_set(&sblock->refs, 1);
2488 sblock->sctx = sctx;
2489 sblock->no_io_error_seen = 1;
2490 sblock->sparity = sparity;
2491 scrub_parity_get(sparity);
2493 for (index = 0; len > 0; index++) {
2494 struct scrub_page *spage;
2495 u64 l = min_t(u64, len, PAGE_SIZE);
2497 spage = kzalloc(sizeof(*spage), GFP_NOFS);
2500 spin_lock(&sctx->stat_lock);
2501 sctx->stat.malloc_errors++;
2502 spin_unlock(&sctx->stat_lock);
2503 scrub_block_put(sblock);
2506 BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
2507 /* For scrub block */
2508 scrub_page_get(spage);
2509 sblock->pagev[index] = spage;
2510 /* For scrub parity */
2511 scrub_page_get(spage);
2512 list_add_tail(&spage->list, &sparity->spages);
2513 spage->sblock = sblock;
2515 spage->flags = flags;
2516 spage->generation = gen;
2517 spage->logical = logical;
2518 spage->physical = physical;
2519 spage->mirror_num = mirror_num;
2521 spage->have_csum = 1;
2522 memcpy(spage->csum, csum, sctx->csum_size);
2524 spage->have_csum = 0;
2526 sblock->page_count++;
2527 spage->page = alloc_page(GFP_NOFS);
2535 WARN_ON(sblock->page_count == 0);
2536 for (index = 0; index < sblock->page_count; index++) {
2537 struct scrub_page *spage = sblock->pagev[index];
2540 ret = scrub_add_page_to_rd_bio(sctx, spage);
2542 scrub_block_put(sblock);
2547 /* last one frees, either here or in bio completion for last page */
2548 scrub_block_put(sblock);
2552 static int scrub_extent_for_parity(struct scrub_parity *sparity,
2553 u64 logical, u64 len,
2554 u64 physical, struct btrfs_device *dev,
2555 u64 flags, u64 gen, int mirror_num)
2557 struct scrub_ctx *sctx = sparity->sctx;
2559 u8 csum[BTRFS_CSUM_SIZE];
2562 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2563 blocksize = sctx->sectorsize;
2564 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2565 blocksize = sctx->nodesize;
2567 blocksize = sctx->sectorsize;
2572 u64 l = min_t(u64, len, blocksize);
2575 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2576 /* push csums to sbio */
2577 have_csum = scrub_find_csum(sctx, logical, l, csum);
2581 ret = scrub_pages_for_parity(sparity, logical, l, physical, dev,
2582 flags, gen, mirror_num,
2583 have_csum ? csum : NULL);
2595 * Given a physical address, this will calculate it's
2596 * logical offset. if this is a parity stripe, it will return
2597 * the most left data stripe's logical offset.
2599 * return 0 if it is a data stripe, 1 means parity stripe.
2601 static int get_raid56_logic_offset(u64 physical, int num,
2602 struct map_lookup *map, u64 *offset,
2612 last_offset = (physical - map->stripes[num].physical) *
2613 nr_data_stripes(map);
2615 *stripe_start = last_offset;
2617 *offset = last_offset;
2618 for (i = 0; i < nr_data_stripes(map); i++) {
2619 *offset = last_offset + i * map->stripe_len;
2621 stripe_nr = div_u64(*offset, map->stripe_len);
2622 stripe_nr = div_u64(stripe_nr, nr_data_stripes(map));
2624 /* Work out the disk rotation on this stripe-set */
2625 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, &rot);
2626 /* calculate which stripe this data locates */
2628 stripe_index = rot % map->num_stripes;
2629 if (stripe_index == num)
2631 if (stripe_index < num)
2634 *offset = last_offset + j * map->stripe_len;
2638 static void scrub_free_parity(struct scrub_parity *sparity)
2640 struct scrub_ctx *sctx = sparity->sctx;
2641 struct scrub_page *curr, *next;
2644 nbits = bitmap_weight(sparity->ebitmap, sparity->nsectors);
2646 spin_lock(&sctx->stat_lock);
2647 sctx->stat.read_errors += nbits;
2648 sctx->stat.uncorrectable_errors += nbits;
2649 spin_unlock(&sctx->stat_lock);
2652 list_for_each_entry_safe(curr, next, &sparity->spages, list) {
2653 list_del_init(&curr->list);
2654 scrub_page_put(curr);
2660 static void scrub_parity_bio_endio_worker(struct btrfs_work *work)
2662 struct scrub_parity *sparity = container_of(work, struct scrub_parity,
2664 struct scrub_ctx *sctx = sparity->sctx;
2666 scrub_free_parity(sparity);
2667 scrub_pending_bio_dec(sctx);
2670 static void scrub_parity_bio_endio(struct bio *bio, int error)
2672 struct scrub_parity *sparity = (struct scrub_parity *)bio->bi_private;
2675 bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap,
2680 btrfs_init_work(&sparity->work, btrfs_scrubparity_helper,
2681 scrub_parity_bio_endio_worker, NULL, NULL);
2682 btrfs_queue_work(sparity->sctx->dev_root->fs_info->scrub_parity_workers,
2686 static void scrub_parity_check_and_repair(struct scrub_parity *sparity)
2688 struct scrub_ctx *sctx = sparity->sctx;
2690 struct btrfs_raid_bio *rbio;
2691 struct scrub_page *spage;
2692 struct btrfs_bio *bbio = NULL;
2696 if (!bitmap_andnot(sparity->dbitmap, sparity->dbitmap, sparity->ebitmap,
2700 length = sparity->logic_end - sparity->logic_start;
2701 ret = btrfs_map_sblock(sctx->dev_root->fs_info, WRITE,
2702 sparity->logic_start,
2703 &length, &bbio, 0, 1);
2704 if (ret || !bbio || !bbio->raid_map)
2707 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
2711 bio->bi_iter.bi_sector = sparity->logic_start >> 9;
2712 bio->bi_private = sparity;
2713 bio->bi_end_io = scrub_parity_bio_endio;
2715 rbio = raid56_parity_alloc_scrub_rbio(sctx->dev_root, bio, bbio,
2716 length, sparity->scrub_dev,
2722 list_for_each_entry(spage, &sparity->spages, list)
2723 raid56_add_scrub_pages(rbio, spage->page, spage->logical);
2725 scrub_pending_bio_inc(sctx);
2726 raid56_parity_submit_scrub_rbio(rbio);
2732 btrfs_put_bbio(bbio);
2733 bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap,
2735 spin_lock(&sctx->stat_lock);
2736 sctx->stat.malloc_errors++;
2737 spin_unlock(&sctx->stat_lock);
2739 scrub_free_parity(sparity);
2742 static inline int scrub_calc_parity_bitmap_len(int nsectors)
2744 return DIV_ROUND_UP(nsectors, BITS_PER_LONG) * (BITS_PER_LONG / 8);
2747 static void scrub_parity_get(struct scrub_parity *sparity)
2749 atomic_inc(&sparity->refs);
2752 static void scrub_parity_put(struct scrub_parity *sparity)
2754 if (!atomic_dec_and_test(&sparity->refs))
2757 scrub_parity_check_and_repair(sparity);
2760 static noinline_for_stack int scrub_raid56_parity(struct scrub_ctx *sctx,
2761 struct map_lookup *map,
2762 struct btrfs_device *sdev,
2763 struct btrfs_path *path,
2767 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
2768 struct btrfs_root *root = fs_info->extent_root;
2769 struct btrfs_root *csum_root = fs_info->csum_root;
2770 struct btrfs_extent_item *extent;
2774 struct extent_buffer *l;
2775 struct btrfs_key key;
2778 u64 extent_physical;
2780 struct btrfs_device *extent_dev;
2781 struct scrub_parity *sparity;
2784 int extent_mirror_num;
2787 nsectors = map->stripe_len / root->sectorsize;
2788 bitmap_len = scrub_calc_parity_bitmap_len(nsectors);
2789 sparity = kzalloc(sizeof(struct scrub_parity) + 2 * bitmap_len,
2792 spin_lock(&sctx->stat_lock);
2793 sctx->stat.malloc_errors++;
2794 spin_unlock(&sctx->stat_lock);
2798 sparity->stripe_len = map->stripe_len;
2799 sparity->nsectors = nsectors;
2800 sparity->sctx = sctx;
2801 sparity->scrub_dev = sdev;
2802 sparity->logic_start = logic_start;
2803 sparity->logic_end = logic_end;
2804 atomic_set(&sparity->refs, 1);
2805 INIT_LIST_HEAD(&sparity->spages);
2806 sparity->dbitmap = sparity->bitmap;
2807 sparity->ebitmap = (void *)sparity->bitmap + bitmap_len;
2810 while (logic_start < logic_end) {
2811 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2812 key.type = BTRFS_METADATA_ITEM_KEY;
2814 key.type = BTRFS_EXTENT_ITEM_KEY;
2815 key.objectid = logic_start;
2816 key.offset = (u64)-1;
2818 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2823 ret = btrfs_previous_extent_item(root, path, 0);
2827 btrfs_release_path(path);
2828 ret = btrfs_search_slot(NULL, root, &key,
2840 slot = path->slots[0];
2841 if (slot >= btrfs_header_nritems(l)) {
2842 ret = btrfs_next_leaf(root, path);
2851 btrfs_item_key_to_cpu(l, &key, slot);
2853 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
2854 key.type != BTRFS_METADATA_ITEM_KEY)
2857 if (key.type == BTRFS_METADATA_ITEM_KEY)
2858 bytes = root->nodesize;
2862 if (key.objectid + bytes <= logic_start)
2865 if (key.objectid >= logic_end) {
2870 while (key.objectid >= logic_start + map->stripe_len)
2871 logic_start += map->stripe_len;
2873 extent = btrfs_item_ptr(l, slot,
2874 struct btrfs_extent_item);
2875 flags = btrfs_extent_flags(l, extent);
2876 generation = btrfs_extent_generation(l, extent);
2878 if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) &&
2879 (key.objectid < logic_start ||
2880 key.objectid + bytes >
2881 logic_start + map->stripe_len)) {
2882 btrfs_err(fs_info, "scrub: tree block %llu spanning stripes, ignored. logical=%llu",
2883 key.objectid, logic_start);
2887 extent_logical = key.objectid;
2890 if (extent_logical < logic_start) {
2891 extent_len -= logic_start - extent_logical;
2892 extent_logical = logic_start;
2895 if (extent_logical + extent_len >
2896 logic_start + map->stripe_len)
2897 extent_len = logic_start + map->stripe_len -
2900 scrub_parity_mark_sectors_data(sparity, extent_logical,
2903 scrub_remap_extent(fs_info, extent_logical,
2904 extent_len, &extent_physical,
2906 &extent_mirror_num);
2908 ret = btrfs_lookup_csums_range(csum_root,
2910 extent_logical + extent_len - 1,
2911 &sctx->csum_list, 1);
2915 ret = scrub_extent_for_parity(sparity, extent_logical,
2922 scrub_free_csums(sctx);
2927 if (extent_logical + extent_len <
2928 key.objectid + bytes) {
2929 logic_start += map->stripe_len;
2931 if (logic_start >= logic_end) {
2936 if (logic_start < key.objectid + bytes) {
2945 btrfs_release_path(path);
2950 logic_start += map->stripe_len;
2954 scrub_parity_mark_sectors_error(sparity, logic_start,
2955 logic_end - logic_start);
2956 scrub_parity_put(sparity);
2958 mutex_lock(&sctx->wr_ctx.wr_lock);
2959 scrub_wr_submit(sctx);
2960 mutex_unlock(&sctx->wr_ctx.wr_lock);
2962 btrfs_release_path(path);
2963 return ret < 0 ? ret : 0;
2966 static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
2967 struct map_lookup *map,
2968 struct btrfs_device *scrub_dev,
2969 int num, u64 base, u64 length,
2972 struct btrfs_path *path, *ppath;
2973 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
2974 struct btrfs_root *root = fs_info->extent_root;
2975 struct btrfs_root *csum_root = fs_info->csum_root;
2976 struct btrfs_extent_item *extent;
2977 struct blk_plug plug;
2982 struct extent_buffer *l;
2983 struct btrfs_key key;
2990 struct reada_control *reada1;
2991 struct reada_control *reada2;
2992 struct btrfs_key key_start;
2993 struct btrfs_key key_end;
2994 u64 increment = map->stripe_len;
2997 u64 extent_physical;
3001 struct btrfs_device *extent_dev;
3002 int extent_mirror_num;
3005 physical = map->stripes[num].physical;
3007 nstripes = div_u64(length, map->stripe_len);
3008 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3009 offset = map->stripe_len * num;
3010 increment = map->stripe_len * map->num_stripes;
3012 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3013 int factor = map->num_stripes / map->sub_stripes;
3014 offset = map->stripe_len * (num / map->sub_stripes);
3015 increment = map->stripe_len * factor;
3016 mirror_num = num % map->sub_stripes + 1;
3017 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3018 increment = map->stripe_len;
3019 mirror_num = num % map->num_stripes + 1;
3020 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3021 increment = map->stripe_len;
3022 mirror_num = num % map->num_stripes + 1;
3023 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3024 get_raid56_logic_offset(physical, num, map, &offset, NULL);
3025 increment = map->stripe_len * nr_data_stripes(map);
3028 increment = map->stripe_len;
3032 path = btrfs_alloc_path();
3036 ppath = btrfs_alloc_path();
3038 btrfs_free_path(path);
3043 * work on commit root. The related disk blocks are static as
3044 * long as COW is applied. This means, it is save to rewrite
3045 * them to repair disk errors without any race conditions
3047 path->search_commit_root = 1;
3048 path->skip_locking = 1;
3050 ppath->search_commit_root = 1;
3051 ppath->skip_locking = 1;
3053 * trigger the readahead for extent tree csum tree and wait for
3054 * completion. During readahead, the scrub is officially paused
3055 * to not hold off transaction commits
3057 logical = base + offset;
3058 physical_end = physical + nstripes * map->stripe_len;
3059 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3060 get_raid56_logic_offset(physical_end, num,
3061 map, &logic_end, NULL);
3064 logic_end = logical + increment * nstripes;
3066 wait_event(sctx->list_wait,
3067 atomic_read(&sctx->bios_in_flight) == 0);
3068 scrub_blocked_if_needed(fs_info);
3070 /* FIXME it might be better to start readahead at commit root */
3071 key_start.objectid = logical;
3072 key_start.type = BTRFS_EXTENT_ITEM_KEY;
3073 key_start.offset = (u64)0;
3074 key_end.objectid = logic_end;
3075 key_end.type = BTRFS_METADATA_ITEM_KEY;
3076 key_end.offset = (u64)-1;
3077 reada1 = btrfs_reada_add(root, &key_start, &key_end);
3079 key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
3080 key_start.type = BTRFS_EXTENT_CSUM_KEY;
3081 key_start.offset = logical;
3082 key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
3083 key_end.type = BTRFS_EXTENT_CSUM_KEY;
3084 key_end.offset = logic_end;
3085 reada2 = btrfs_reada_add(csum_root, &key_start, &key_end);
3087 if (!IS_ERR(reada1))
3088 btrfs_reada_wait(reada1);
3089 if (!IS_ERR(reada2))
3090 btrfs_reada_wait(reada2);
3094 * collect all data csums for the stripe to avoid seeking during
3095 * the scrub. This might currently (crc32) end up to be about 1MB
3097 blk_start_plug(&plug);
3100 * now find all extents for each stripe and scrub them
3103 while (physical < physical_end) {
3107 if (atomic_read(&fs_info->scrub_cancel_req) ||
3108 atomic_read(&sctx->cancel_req)) {
3113 * check to see if we have to pause
3115 if (atomic_read(&fs_info->scrub_pause_req)) {
3116 /* push queued extents */
3117 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
3119 mutex_lock(&sctx->wr_ctx.wr_lock);
3120 scrub_wr_submit(sctx);
3121 mutex_unlock(&sctx->wr_ctx.wr_lock);
3122 wait_event(sctx->list_wait,
3123 atomic_read(&sctx->bios_in_flight) == 0);
3124 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
3125 scrub_blocked_if_needed(fs_info);
3128 /* for raid56, we skip parity stripe */
3129 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3130 ret = get_raid56_logic_offset(physical, num, map,
3135 stripe_logical += base;
3136 stripe_end = stripe_logical + increment;
3137 ret = scrub_raid56_parity(sctx, map, scrub_dev,
3138 ppath, stripe_logical,
3146 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
3147 key.type = BTRFS_METADATA_ITEM_KEY;
3149 key.type = BTRFS_EXTENT_ITEM_KEY;
3150 key.objectid = logical;
3151 key.offset = (u64)-1;
3153 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3158 ret = btrfs_previous_extent_item(root, path, 0);
3162 /* there's no smaller item, so stick with the
3164 btrfs_release_path(path);
3165 ret = btrfs_search_slot(NULL, root, &key,
3177 slot = path->slots[0];
3178 if (slot >= btrfs_header_nritems(l)) {
3179 ret = btrfs_next_leaf(root, path);
3188 btrfs_item_key_to_cpu(l, &key, slot);
3190 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3191 key.type != BTRFS_METADATA_ITEM_KEY)
3194 if (key.type == BTRFS_METADATA_ITEM_KEY)
3195 bytes = root->nodesize;
3199 if (key.objectid + bytes <= logical)
3202 if (key.objectid >= logical + map->stripe_len) {
3203 /* out of this device extent */
3204 if (key.objectid >= logic_end)
3209 extent = btrfs_item_ptr(l, slot,
3210 struct btrfs_extent_item);
3211 flags = btrfs_extent_flags(l, extent);
3212 generation = btrfs_extent_generation(l, extent);
3214 if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) &&
3215 (key.objectid < logical ||
3216 key.objectid + bytes >
3217 logical + map->stripe_len)) {
3219 "scrub: tree block %llu spanning "
3220 "stripes, ignored. logical=%llu",
3221 key.objectid, logical);
3226 extent_logical = key.objectid;
3230 * trim extent to this stripe
3232 if (extent_logical < logical) {
3233 extent_len -= logical - extent_logical;
3234 extent_logical = logical;
3236 if (extent_logical + extent_len >
3237 logical + map->stripe_len) {
3238 extent_len = logical + map->stripe_len -
3242 extent_physical = extent_logical - logical + physical;
3243 extent_dev = scrub_dev;
3244 extent_mirror_num = mirror_num;
3246 scrub_remap_extent(fs_info, extent_logical,
3247 extent_len, &extent_physical,
3249 &extent_mirror_num);
3251 ret = btrfs_lookup_csums_range(csum_root,
3255 &sctx->csum_list, 1);
3259 ret = scrub_extent(sctx, extent_logical, extent_len,
3260 extent_physical, extent_dev, flags,
3261 generation, extent_mirror_num,
3262 extent_logical - logical + physical);
3264 scrub_free_csums(sctx);
3269 if (extent_logical + extent_len <
3270 key.objectid + bytes) {
3271 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3273 * loop until we find next data stripe
3274 * or we have finished all stripes.
3277 physical += map->stripe_len;
3278 ret = get_raid56_logic_offset(physical,
3283 if (ret && physical < physical_end) {
3284 stripe_logical += base;
3285 stripe_end = stripe_logical +
3287 ret = scrub_raid56_parity(sctx,
3288 map, scrub_dev, ppath,
3296 physical += map->stripe_len;
3297 logical += increment;
3299 if (logical < key.objectid + bytes) {
3304 if (physical >= physical_end) {
3312 btrfs_release_path(path);
3314 logical += increment;
3315 physical += map->stripe_len;
3316 spin_lock(&sctx->stat_lock);
3318 sctx->stat.last_physical = map->stripes[num].physical +
3321 sctx->stat.last_physical = physical;
3322 spin_unlock(&sctx->stat_lock);
3327 /* push queued extents */
3329 mutex_lock(&sctx->wr_ctx.wr_lock);
3330 scrub_wr_submit(sctx);
3331 mutex_unlock(&sctx->wr_ctx.wr_lock);
3333 blk_finish_plug(&plug);
3334 btrfs_free_path(path);
3335 btrfs_free_path(ppath);
3336 return ret < 0 ? ret : 0;
3339 static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx,
3340 struct btrfs_device *scrub_dev,
3341 u64 chunk_tree, u64 chunk_objectid,
3342 u64 chunk_offset, u64 length,
3343 u64 dev_offset, int is_dev_replace)
3345 struct btrfs_mapping_tree *map_tree =
3346 &sctx->dev_root->fs_info->mapping_tree;
3347 struct map_lookup *map;
3348 struct extent_map *em;
3352 read_lock(&map_tree->map_tree.lock);
3353 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3354 read_unlock(&map_tree->map_tree.lock);
3359 map = (struct map_lookup *)em->bdev;
3360 if (em->start != chunk_offset)
3363 if (em->len < length)
3366 for (i = 0; i < map->num_stripes; ++i) {
3367 if (map->stripes[i].dev->bdev == scrub_dev->bdev &&
3368 map->stripes[i].physical == dev_offset) {
3369 ret = scrub_stripe(sctx, map, scrub_dev, i,
3370 chunk_offset, length,
3377 free_extent_map(em);
3382 static noinline_for_stack
3383 int scrub_enumerate_chunks(struct scrub_ctx *sctx,
3384 struct btrfs_device *scrub_dev, u64 start, u64 end,
3387 struct btrfs_dev_extent *dev_extent = NULL;
3388 struct btrfs_path *path;
3389 struct btrfs_root *root = sctx->dev_root;
3390 struct btrfs_fs_info *fs_info = root->fs_info;
3397 struct extent_buffer *l;
3398 struct btrfs_key key;
3399 struct btrfs_key found_key;
3400 struct btrfs_block_group_cache *cache;
3401 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
3403 path = btrfs_alloc_path();
3408 path->search_commit_root = 1;
3409 path->skip_locking = 1;
3411 key.objectid = scrub_dev->devid;
3413 key.type = BTRFS_DEV_EXTENT_KEY;
3416 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3420 if (path->slots[0] >=
3421 btrfs_header_nritems(path->nodes[0])) {
3422 ret = btrfs_next_leaf(root, path);
3435 slot = path->slots[0];
3437 btrfs_item_key_to_cpu(l, &found_key, slot);
3439 if (found_key.objectid != scrub_dev->devid)
3442 if (found_key.type != BTRFS_DEV_EXTENT_KEY)
3445 if (found_key.offset >= end)
3448 if (found_key.offset < key.offset)
3451 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3452 length = btrfs_dev_extent_length(l, dev_extent);
3454 if (found_key.offset + length <= start)
3457 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3458 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3459 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3462 * get a reference on the corresponding block group to prevent
3463 * the chunk from going away while we scrub it
3465 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3467 /* some chunks are removed but not committed to disk yet,
3468 * continue scrubbing */
3473 * we need call btrfs_inc_block_group_ro() with scrubs_paused,
3474 * to avoid deadlock caused by:
3475 * btrfs_inc_block_group_ro()
3476 * -> btrfs_wait_for_commit()
3477 * -> btrfs_commit_transaction()
3478 * -> btrfs_scrub_pause()
3480 scrub_pause_on(fs_info);
3481 ret = btrfs_inc_block_group_ro(root, cache);
3482 scrub_pause_off(fs_info);
3484 btrfs_put_block_group(cache);
3488 dev_replace->cursor_right = found_key.offset + length;
3489 dev_replace->cursor_left = found_key.offset;
3490 dev_replace->item_needs_writeback = 1;
3491 ret = scrub_chunk(sctx, scrub_dev, chunk_tree, chunk_objectid,
3492 chunk_offset, length, found_key.offset,
3496 * flush, submit all pending read and write bios, afterwards
3498 * Note that in the dev replace case, a read request causes
3499 * write requests that are submitted in the read completion
3500 * worker. Therefore in the current situation, it is required
3501 * that all write requests are flushed, so that all read and
3502 * write requests are really completed when bios_in_flight
3505 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
3507 mutex_lock(&sctx->wr_ctx.wr_lock);
3508 scrub_wr_submit(sctx);
3509 mutex_unlock(&sctx->wr_ctx.wr_lock);
3511 wait_event(sctx->list_wait,
3512 atomic_read(&sctx->bios_in_flight) == 0);
3514 scrub_pause_on(fs_info);
3517 * must be called before we decrease @scrub_paused.
3518 * make sure we don't block transaction commit while
3519 * we are waiting pending workers finished.
3521 wait_event(sctx->list_wait,
3522 atomic_read(&sctx->workers_pending) == 0);
3523 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
3525 scrub_pause_off(fs_info);
3527 btrfs_dec_block_group_ro(root, cache);
3529 btrfs_put_block_group(cache);
3532 if (is_dev_replace &&
3533 atomic64_read(&dev_replace->num_write_errors) > 0) {
3537 if (sctx->stat.malloc_errors > 0) {
3542 dev_replace->cursor_left = dev_replace->cursor_right;
3543 dev_replace->item_needs_writeback = 1;
3545 key.offset = found_key.offset + length;
3546 btrfs_release_path(path);
3549 btrfs_free_path(path);
3554 static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx,
3555 struct btrfs_device *scrub_dev)
3561 struct btrfs_root *root = sctx->dev_root;
3563 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
3566 /* Seed devices of a new filesystem has their own generation. */
3567 if (scrub_dev->fs_devices != root->fs_info->fs_devices)
3568 gen = scrub_dev->generation;
3570 gen = root->fs_info->last_trans_committed;
3572 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
3573 bytenr = btrfs_sb_offset(i);
3574 if (bytenr + BTRFS_SUPER_INFO_SIZE >
3575 scrub_dev->commit_total_bytes)
3578 ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr,
3579 scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i,
3584 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
3590 * get a reference count on fs_info->scrub_workers. start worker if necessary
3592 static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info,
3595 unsigned int flags = WQ_FREEZABLE | WQ_UNBOUND;
3596 int max_active = fs_info->thread_pool_size;
3598 if (fs_info->scrub_workers_refcnt == 0) {
3600 fs_info->scrub_workers =
3601 btrfs_alloc_workqueue("btrfs-scrub", flags,
3604 fs_info->scrub_workers =
3605 btrfs_alloc_workqueue("btrfs-scrub", flags,
3607 if (!fs_info->scrub_workers)
3608 goto fail_scrub_workers;
3610 fs_info->scrub_wr_completion_workers =
3611 btrfs_alloc_workqueue("btrfs-scrubwrc", flags,
3613 if (!fs_info->scrub_wr_completion_workers)
3614 goto fail_scrub_wr_completion_workers;
3616 fs_info->scrub_nocow_workers =
3617 btrfs_alloc_workqueue("btrfs-scrubnc", flags, 1, 0);
3618 if (!fs_info->scrub_nocow_workers)
3619 goto fail_scrub_nocow_workers;
3620 fs_info->scrub_parity_workers =
3621 btrfs_alloc_workqueue("btrfs-scrubparity", flags,
3623 if (!fs_info->scrub_parity_workers)
3624 goto fail_scrub_parity_workers;
3626 ++fs_info->scrub_workers_refcnt;
3629 fail_scrub_parity_workers:
3630 btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
3631 fail_scrub_nocow_workers:
3632 btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
3633 fail_scrub_wr_completion_workers:
3634 btrfs_destroy_workqueue(fs_info->scrub_workers);
3639 static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info)
3641 if (--fs_info->scrub_workers_refcnt == 0) {
3642 btrfs_destroy_workqueue(fs_info->scrub_workers);
3643 btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
3644 btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
3645 btrfs_destroy_workqueue(fs_info->scrub_parity_workers);
3647 WARN_ON(fs_info->scrub_workers_refcnt < 0);
3650 int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start,
3651 u64 end, struct btrfs_scrub_progress *progress,
3652 int readonly, int is_dev_replace)
3654 struct scrub_ctx *sctx;
3656 struct btrfs_device *dev;
3657 struct rcu_string *name;
3659 if (btrfs_fs_closing(fs_info))
3662 if (fs_info->chunk_root->nodesize > BTRFS_STRIPE_LEN) {
3664 * in this case scrub is unable to calculate the checksum
3665 * the way scrub is implemented. Do not handle this
3666 * situation at all because it won't ever happen.
3669 "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails",
3670 fs_info->chunk_root->nodesize, BTRFS_STRIPE_LEN);
3674 if (fs_info->chunk_root->sectorsize != PAGE_SIZE) {
3675 /* not supported for data w/o checksums */
3677 "scrub: size assumption sectorsize != PAGE_SIZE "
3678 "(%d != %lu) fails",
3679 fs_info->chunk_root->sectorsize, PAGE_SIZE);
3683 if (fs_info->chunk_root->nodesize >
3684 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK ||
3685 fs_info->chunk_root->sectorsize >
3686 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) {
3688 * would exhaust the array bounds of pagev member in
3689 * struct scrub_block
3691 btrfs_err(fs_info, "scrub: size assumption nodesize and sectorsize "
3692 "<= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails",
3693 fs_info->chunk_root->nodesize,
3694 SCRUB_MAX_PAGES_PER_BLOCK,
3695 fs_info->chunk_root->sectorsize,
3696 SCRUB_MAX_PAGES_PER_BLOCK);
3701 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3702 dev = btrfs_find_device(fs_info, devid, NULL, NULL);
3703 if (!dev || (dev->missing && !is_dev_replace)) {
3704 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3708 if (!is_dev_replace && !readonly && !dev->writeable) {
3709 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3711 name = rcu_dereference(dev->name);
3712 btrfs_err(fs_info, "scrub: device %s is not writable",
3718 mutex_lock(&fs_info->scrub_lock);
3719 if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) {
3720 mutex_unlock(&fs_info->scrub_lock);
3721 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3725 btrfs_dev_replace_lock(&fs_info->dev_replace);
3726 if (dev->scrub_device ||
3728 btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) {
3729 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3730 mutex_unlock(&fs_info->scrub_lock);
3731 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3732 return -EINPROGRESS;
3734 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3736 ret = scrub_workers_get(fs_info, is_dev_replace);
3738 mutex_unlock(&fs_info->scrub_lock);
3739 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3743 sctx = scrub_setup_ctx(dev, is_dev_replace);
3745 mutex_unlock(&fs_info->scrub_lock);
3746 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3747 scrub_workers_put(fs_info);
3748 return PTR_ERR(sctx);
3750 sctx->readonly = readonly;
3751 dev->scrub_device = sctx;
3752 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3755 * checking @scrub_pause_req here, we can avoid
3756 * race between committing transaction and scrubbing.
3758 __scrub_blocked_if_needed(fs_info);
3759 atomic_inc(&fs_info->scrubs_running);
3760 mutex_unlock(&fs_info->scrub_lock);
3762 if (!is_dev_replace) {
3764 * by holding device list mutex, we can
3765 * kick off writing super in log tree sync.
3767 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3768 ret = scrub_supers(sctx, dev);
3769 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3773 ret = scrub_enumerate_chunks(sctx, dev, start, end,
3776 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
3777 atomic_dec(&fs_info->scrubs_running);
3778 wake_up(&fs_info->scrub_pause_wait);
3780 wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0);
3783 memcpy(progress, &sctx->stat, sizeof(*progress));
3785 mutex_lock(&fs_info->scrub_lock);
3786 dev->scrub_device = NULL;
3787 scrub_workers_put(fs_info);
3788 mutex_unlock(&fs_info->scrub_lock);
3790 scrub_put_ctx(sctx);
3795 void btrfs_scrub_pause(struct btrfs_root *root)
3797 struct btrfs_fs_info *fs_info = root->fs_info;
3799 mutex_lock(&fs_info->scrub_lock);
3800 atomic_inc(&fs_info->scrub_pause_req);
3801 while (atomic_read(&fs_info->scrubs_paused) !=
3802 atomic_read(&fs_info->scrubs_running)) {
3803 mutex_unlock(&fs_info->scrub_lock);
3804 wait_event(fs_info->scrub_pause_wait,
3805 atomic_read(&fs_info->scrubs_paused) ==
3806 atomic_read(&fs_info->scrubs_running));
3807 mutex_lock(&fs_info->scrub_lock);
3809 mutex_unlock(&fs_info->scrub_lock);
3812 void btrfs_scrub_continue(struct btrfs_root *root)
3814 struct btrfs_fs_info *fs_info = root->fs_info;
3816 atomic_dec(&fs_info->scrub_pause_req);
3817 wake_up(&fs_info->scrub_pause_wait);
3820 int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info)
3822 mutex_lock(&fs_info->scrub_lock);
3823 if (!atomic_read(&fs_info->scrubs_running)) {
3824 mutex_unlock(&fs_info->scrub_lock);
3828 atomic_inc(&fs_info->scrub_cancel_req);
3829 while (atomic_read(&fs_info->scrubs_running)) {
3830 mutex_unlock(&fs_info->scrub_lock);
3831 wait_event(fs_info->scrub_pause_wait,
3832 atomic_read(&fs_info->scrubs_running) == 0);
3833 mutex_lock(&fs_info->scrub_lock);
3835 atomic_dec(&fs_info->scrub_cancel_req);
3836 mutex_unlock(&fs_info->scrub_lock);
3841 int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info,
3842 struct btrfs_device *dev)
3844 struct scrub_ctx *sctx;
3846 mutex_lock(&fs_info->scrub_lock);
3847 sctx = dev->scrub_device;
3849 mutex_unlock(&fs_info->scrub_lock);
3852 atomic_inc(&sctx->cancel_req);
3853 while (dev->scrub_device) {
3854 mutex_unlock(&fs_info->scrub_lock);
3855 wait_event(fs_info->scrub_pause_wait,
3856 dev->scrub_device == NULL);
3857 mutex_lock(&fs_info->scrub_lock);
3859 mutex_unlock(&fs_info->scrub_lock);
3864 int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
3865 struct btrfs_scrub_progress *progress)
3867 struct btrfs_device *dev;
3868 struct scrub_ctx *sctx = NULL;
3870 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3871 dev = btrfs_find_device(root->fs_info, devid, NULL, NULL);
3873 sctx = dev->scrub_device;
3875 memcpy(progress, &sctx->stat, sizeof(*progress));
3876 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3878 return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV;
3881 static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
3882 u64 extent_logical, u64 extent_len,
3883 u64 *extent_physical,
3884 struct btrfs_device **extent_dev,
3885 int *extent_mirror_num)
3888 struct btrfs_bio *bbio = NULL;
3891 mapped_length = extent_len;
3892 ret = btrfs_map_block(fs_info, READ, extent_logical,
3893 &mapped_length, &bbio, 0);
3894 if (ret || !bbio || mapped_length < extent_len ||
3895 !bbio->stripes[0].dev->bdev) {
3896 btrfs_put_bbio(bbio);
3900 *extent_physical = bbio->stripes[0].physical;
3901 *extent_mirror_num = bbio->mirror_num;
3902 *extent_dev = bbio->stripes[0].dev;
3903 btrfs_put_bbio(bbio);
3906 static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
3907 struct scrub_wr_ctx *wr_ctx,
3908 struct btrfs_fs_info *fs_info,
3909 struct btrfs_device *dev,
3912 WARN_ON(wr_ctx->wr_curr_bio != NULL);
3914 mutex_init(&wr_ctx->wr_lock);
3915 wr_ctx->wr_curr_bio = NULL;
3916 if (!is_dev_replace)
3919 WARN_ON(!dev->bdev);
3920 wr_ctx->pages_per_wr_bio = min_t(int, SCRUB_PAGES_PER_WR_BIO,
3921 bio_get_nr_vecs(dev->bdev));
3922 wr_ctx->tgtdev = dev;
3923 atomic_set(&wr_ctx->flush_all_writes, 0);
3927 static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx)
3929 mutex_lock(&wr_ctx->wr_lock);
3930 kfree(wr_ctx->wr_curr_bio);
3931 wr_ctx->wr_curr_bio = NULL;
3932 mutex_unlock(&wr_ctx->wr_lock);
3935 static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
3936 int mirror_num, u64 physical_for_dev_replace)
3938 struct scrub_copy_nocow_ctx *nocow_ctx;
3939 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
3941 nocow_ctx = kzalloc(sizeof(*nocow_ctx), GFP_NOFS);
3943 spin_lock(&sctx->stat_lock);
3944 sctx->stat.malloc_errors++;
3945 spin_unlock(&sctx->stat_lock);
3949 scrub_pending_trans_workers_inc(sctx);
3951 nocow_ctx->sctx = sctx;
3952 nocow_ctx->logical = logical;
3953 nocow_ctx->len = len;
3954 nocow_ctx->mirror_num = mirror_num;
3955 nocow_ctx->physical_for_dev_replace = physical_for_dev_replace;
3956 btrfs_init_work(&nocow_ctx->work, btrfs_scrubnc_helper,
3957 copy_nocow_pages_worker, NULL, NULL);
3958 INIT_LIST_HEAD(&nocow_ctx->inodes);
3959 btrfs_queue_work(fs_info->scrub_nocow_workers,
3965 static int record_inode_for_nocow(u64 inum, u64 offset, u64 root, void *ctx)
3967 struct scrub_copy_nocow_ctx *nocow_ctx = ctx;
3968 struct scrub_nocow_inode *nocow_inode;
3970 nocow_inode = kzalloc(sizeof(*nocow_inode), GFP_NOFS);
3973 nocow_inode->inum = inum;
3974 nocow_inode->offset = offset;
3975 nocow_inode->root = root;
3976 list_add_tail(&nocow_inode->list, &nocow_ctx->inodes);
3980 #define COPY_COMPLETE 1
3982 static void copy_nocow_pages_worker(struct btrfs_work *work)
3984 struct scrub_copy_nocow_ctx *nocow_ctx =
3985 container_of(work, struct scrub_copy_nocow_ctx, work);
3986 struct scrub_ctx *sctx = nocow_ctx->sctx;
3987 u64 logical = nocow_ctx->logical;
3988 u64 len = nocow_ctx->len;
3989 int mirror_num = nocow_ctx->mirror_num;
3990 u64 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
3992 struct btrfs_trans_handle *trans = NULL;
3993 struct btrfs_fs_info *fs_info;
3994 struct btrfs_path *path;
3995 struct btrfs_root *root;
3996 int not_written = 0;
3998 fs_info = sctx->dev_root->fs_info;
3999 root = fs_info->extent_root;
4001 path = btrfs_alloc_path();
4003 spin_lock(&sctx->stat_lock);
4004 sctx->stat.malloc_errors++;
4005 spin_unlock(&sctx->stat_lock);
4010 trans = btrfs_join_transaction(root);
4011 if (IS_ERR(trans)) {
4016 ret = iterate_inodes_from_logical(logical, fs_info, path,
4017 record_inode_for_nocow, nocow_ctx);
4018 if (ret != 0 && ret != -ENOENT) {
4019 btrfs_warn(fs_info, "iterate_inodes_from_logical() failed: log %llu, "
4020 "phys %llu, len %llu, mir %u, ret %d",
4021 logical, physical_for_dev_replace, len, mirror_num,
4027 btrfs_end_transaction(trans, root);
4029 while (!list_empty(&nocow_ctx->inodes)) {
4030 struct scrub_nocow_inode *entry;
4031 entry = list_first_entry(&nocow_ctx->inodes,
4032 struct scrub_nocow_inode,
4034 list_del_init(&entry->list);
4035 ret = copy_nocow_pages_for_inode(entry->inum, entry->offset,
4036 entry->root, nocow_ctx);
4038 if (ret == COPY_COMPLETE) {
4046 while (!list_empty(&nocow_ctx->inodes)) {
4047 struct scrub_nocow_inode *entry;
4048 entry = list_first_entry(&nocow_ctx->inodes,
4049 struct scrub_nocow_inode,
4051 list_del_init(&entry->list);
4054 if (trans && !IS_ERR(trans))
4055 btrfs_end_transaction(trans, root);
4057 btrfs_dev_replace_stats_inc(&fs_info->dev_replace.
4058 num_uncorrectable_read_errors);
4060 btrfs_free_path(path);
4063 scrub_pending_trans_workers_dec(sctx);
4066 static int check_extent_to_block(struct inode *inode, u64 start, u64 len,
4069 struct extent_state *cached_state = NULL;
4070 struct btrfs_ordered_extent *ordered;
4071 struct extent_io_tree *io_tree;
4072 struct extent_map *em;
4073 u64 lockstart = start, lockend = start + len - 1;
4076 io_tree = &BTRFS_I(inode)->io_tree;
4078 lock_extent_bits(io_tree, lockstart, lockend, 0, &cached_state);
4079 ordered = btrfs_lookup_ordered_range(inode, lockstart, len);
4081 btrfs_put_ordered_extent(ordered);
4086 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
4093 * This extent does not actually cover the logical extent anymore,
4094 * move on to the next inode.
4096 if (em->block_start > logical ||
4097 em->block_start + em->block_len < logical + len) {
4098 free_extent_map(em);
4102 free_extent_map(em);
4105 unlock_extent_cached(io_tree, lockstart, lockend, &cached_state,
4110 static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
4111 struct scrub_copy_nocow_ctx *nocow_ctx)
4113 struct btrfs_fs_info *fs_info = nocow_ctx->sctx->dev_root->fs_info;
4114 struct btrfs_key key;
4115 struct inode *inode;
4117 struct btrfs_root *local_root;
4118 struct extent_io_tree *io_tree;
4119 u64 physical_for_dev_replace;
4120 u64 nocow_ctx_logical;
4121 u64 len = nocow_ctx->len;
4122 unsigned long index;
4127 key.objectid = root;
4128 key.type = BTRFS_ROOT_ITEM_KEY;
4129 key.offset = (u64)-1;
4131 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
4133 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
4134 if (IS_ERR(local_root)) {
4135 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
4136 return PTR_ERR(local_root);
4139 key.type = BTRFS_INODE_ITEM_KEY;
4140 key.objectid = inum;
4142 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
4143 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
4145 return PTR_ERR(inode);
4147 /* Avoid truncate/dio/punch hole.. */
4148 mutex_lock(&inode->i_mutex);
4149 inode_dio_wait(inode);
4151 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
4152 io_tree = &BTRFS_I(inode)->io_tree;
4153 nocow_ctx_logical = nocow_ctx->logical;
4155 ret = check_extent_to_block(inode, offset, len, nocow_ctx_logical);
4157 ret = ret > 0 ? 0 : ret;
4161 while (len >= PAGE_CACHE_SIZE) {
4162 index = offset >> PAGE_CACHE_SHIFT;
4164 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
4166 btrfs_err(fs_info, "find_or_create_page() failed");
4171 if (PageUptodate(page)) {
4172 if (PageDirty(page))
4175 ClearPageError(page);
4176 err = extent_read_full_page(io_tree, page,
4178 nocow_ctx->mirror_num);
4186 * If the page has been remove from the page cache,
4187 * the data on it is meaningless, because it may be
4188 * old one, the new data may be written into the new
4189 * page in the page cache.
4191 if (page->mapping != inode->i_mapping) {
4193 page_cache_release(page);
4196 if (!PageUptodate(page)) {
4202 ret = check_extent_to_block(inode, offset, len,
4205 ret = ret > 0 ? 0 : ret;
4209 err = write_page_nocow(nocow_ctx->sctx,
4210 physical_for_dev_replace, page);
4215 page_cache_release(page);
4220 offset += PAGE_CACHE_SIZE;
4221 physical_for_dev_replace += PAGE_CACHE_SIZE;
4222 nocow_ctx_logical += PAGE_CACHE_SIZE;
4223 len -= PAGE_CACHE_SIZE;
4225 ret = COPY_COMPLETE;
4227 mutex_unlock(&inode->i_mutex);
4232 static int write_page_nocow(struct scrub_ctx *sctx,
4233 u64 physical_for_dev_replace, struct page *page)
4236 struct btrfs_device *dev;
4239 dev = sctx->wr_ctx.tgtdev;
4243 printk_ratelimited(KERN_WARNING
4244 "BTRFS: scrub write_page_nocow(bdev == NULL) is unexpected!\n");
4247 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
4249 spin_lock(&sctx->stat_lock);
4250 sctx->stat.malloc_errors++;
4251 spin_unlock(&sctx->stat_lock);
4254 bio->bi_iter.bi_size = 0;
4255 bio->bi_iter.bi_sector = physical_for_dev_replace >> 9;
4256 bio->bi_bdev = dev->bdev;
4257 ret = bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
4258 if (ret != PAGE_CACHE_SIZE) {
4261 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
4265 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio))
4266 goto leave_with_eio;