2 * Copyright (C) 2011, 2012 STRATO. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/blkdev.h>
20 #include <linux/ratelimit.h>
24 #include "ordered-data.h"
25 #include "transaction.h"
27 #include "extent_io.h"
28 #include "dev-replace.h"
29 #include "check-integrity.h"
30 #include "rcu-string.h"
34 * This is only the first step towards a full-features scrub. It reads all
35 * extent and super block and verifies the checksums. In case a bad checksum
36 * is found or the extent cannot be read, good data will be written back if
39 * Future enhancements:
40 * - In case an unrepairable extent is encountered, track which files are
41 * affected and report them
42 * - track and record media errors, throw out bad devices
43 * - add a mode to also read unallocated space
50 * the following three values only influence the performance.
51 * The last one configures the number of parallel and outstanding I/O
52 * operations. The first two values configure an upper limit for the number
53 * of (dynamically allocated) pages that are added to a bio.
55 #define SCRUB_PAGES_PER_RD_BIO 32 /* 128k per bio */
56 #define SCRUB_PAGES_PER_WR_BIO 32 /* 128k per bio */
57 #define SCRUB_BIOS_PER_SCTX 64 /* 8MB per device in flight */
60 * the following value times PAGE_SIZE needs to be large enough to match the
61 * largest node/leaf/sector size that shall be supported.
62 * Values larger than BTRFS_STRIPE_LEN are not supported.
64 #define SCRUB_MAX_PAGES_PER_BLOCK 16 /* 64k per node/leaf/sector */
66 struct scrub_recover {
68 struct btrfs_bio *bbio;
73 struct scrub_block *sblock;
75 struct btrfs_device *dev;
76 struct list_head list;
77 u64 flags; /* extent flags */
81 u64 physical_for_dev_replace;
84 unsigned int mirror_num:8;
85 unsigned int have_csum:1;
86 unsigned int io_error:1;
88 u8 csum[BTRFS_CSUM_SIZE];
90 struct scrub_recover *recover;
95 struct scrub_ctx *sctx;
96 struct btrfs_device *dev;
101 #if SCRUB_PAGES_PER_WR_BIO >= SCRUB_PAGES_PER_RD_BIO
102 struct scrub_page *pagev[SCRUB_PAGES_PER_WR_BIO];
104 struct scrub_page *pagev[SCRUB_PAGES_PER_RD_BIO];
108 struct btrfs_work work;
112 struct scrub_page *pagev[SCRUB_MAX_PAGES_PER_BLOCK];
114 atomic_t outstanding_pages;
115 atomic_t refs; /* free mem on transition to zero */
116 struct scrub_ctx *sctx;
117 struct scrub_parity *sparity;
119 unsigned int header_error:1;
120 unsigned int checksum_error:1;
121 unsigned int no_io_error_seen:1;
122 unsigned int generation_error:1; /* also sets header_error */
124 /* The following is for the data used to check parity */
125 /* It is for the data with checksum */
126 unsigned int data_corrected:1;
128 struct btrfs_work work;
131 /* Used for the chunks with parity stripe such RAID5/6 */
132 struct scrub_parity {
133 struct scrub_ctx *sctx;
135 struct btrfs_device *scrub_dev;
147 struct list_head spages;
149 /* Work of parity check and repair */
150 struct btrfs_work work;
152 /* Mark the parity blocks which have data */
153 unsigned long *dbitmap;
156 * Mark the parity blocks which have data, but errors happen when
157 * read data or check data
159 unsigned long *ebitmap;
161 unsigned long bitmap[0];
164 struct scrub_wr_ctx {
165 struct scrub_bio *wr_curr_bio;
166 struct btrfs_device *tgtdev;
167 int pages_per_wr_bio; /* <= SCRUB_PAGES_PER_WR_BIO */
168 atomic_t flush_all_writes;
169 struct mutex wr_lock;
173 struct scrub_bio *bios[SCRUB_BIOS_PER_SCTX];
174 struct btrfs_root *dev_root;
177 atomic_t bios_in_flight;
178 atomic_t workers_pending;
179 spinlock_t list_lock;
180 wait_queue_head_t list_wait;
182 struct list_head csum_list;
185 int pages_per_rd_bio;
190 struct scrub_wr_ctx wr_ctx;
195 struct btrfs_scrub_progress stat;
196 spinlock_t stat_lock;
199 * Use a ref counter to avoid use-after-free issues. Scrub workers
200 * decrement bios_in_flight and workers_pending and then do a wakeup
201 * on the list_wait wait queue. We must ensure the main scrub task
202 * doesn't free the scrub context before or while the workers are
203 * doing the wakeup() call.
208 struct scrub_fixup_nodatasum {
209 struct scrub_ctx *sctx;
210 struct btrfs_device *dev;
212 struct btrfs_root *root;
213 struct btrfs_work work;
217 struct scrub_nocow_inode {
221 struct list_head list;
224 struct scrub_copy_nocow_ctx {
225 struct scrub_ctx *sctx;
229 u64 physical_for_dev_replace;
230 struct list_head inodes;
231 struct btrfs_work work;
234 struct scrub_warning {
235 struct btrfs_path *path;
236 u64 extent_item_size;
240 struct btrfs_device *dev;
243 static void scrub_pending_bio_inc(struct scrub_ctx *sctx);
244 static void scrub_pending_bio_dec(struct scrub_ctx *sctx);
245 static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx);
246 static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx);
247 static int scrub_handle_errored_block(struct scrub_block *sblock_to_check);
248 static int scrub_setup_recheck_block(struct scrub_block *original_sblock,
249 struct scrub_block *sblocks_for_recheck);
250 static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
251 struct scrub_block *sblock, int is_metadata,
252 int have_csum, u8 *csum, u64 generation,
253 u16 csum_size, int retry_failed_mirror);
254 static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
255 struct scrub_block *sblock,
256 int is_metadata, int have_csum,
257 const u8 *csum, u64 generation,
259 static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
260 struct scrub_block *sblock_good);
261 static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
262 struct scrub_block *sblock_good,
263 int page_num, int force_write);
264 static void scrub_write_block_to_dev_replace(struct scrub_block *sblock);
265 static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
267 static int scrub_checksum_data(struct scrub_block *sblock);
268 static int scrub_checksum_tree_block(struct scrub_block *sblock);
269 static int scrub_checksum_super(struct scrub_block *sblock);
270 static void scrub_block_get(struct scrub_block *sblock);
271 static void scrub_block_put(struct scrub_block *sblock);
272 static void scrub_page_get(struct scrub_page *spage);
273 static void scrub_page_put(struct scrub_page *spage);
274 static void scrub_parity_get(struct scrub_parity *sparity);
275 static void scrub_parity_put(struct scrub_parity *sparity);
276 static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
277 struct scrub_page *spage);
278 static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
279 u64 physical, struct btrfs_device *dev, u64 flags,
280 u64 gen, int mirror_num, u8 *csum, int force,
281 u64 physical_for_dev_replace);
282 static void scrub_bio_end_io(struct bio *bio);
283 static void scrub_bio_end_io_worker(struct btrfs_work *work);
284 static void scrub_block_complete(struct scrub_block *sblock);
285 static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
286 u64 extent_logical, u64 extent_len,
287 u64 *extent_physical,
288 struct btrfs_device **extent_dev,
289 int *extent_mirror_num);
290 static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
291 struct scrub_wr_ctx *wr_ctx,
292 struct btrfs_fs_info *fs_info,
293 struct btrfs_device *dev,
295 static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx);
296 static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
297 struct scrub_page *spage);
298 static void scrub_wr_submit(struct scrub_ctx *sctx);
299 static void scrub_wr_bio_end_io(struct bio *bio);
300 static void scrub_wr_bio_end_io_worker(struct btrfs_work *work);
301 static int write_page_nocow(struct scrub_ctx *sctx,
302 u64 physical_for_dev_replace, struct page *page);
303 static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
304 struct scrub_copy_nocow_ctx *ctx);
305 static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
306 int mirror_num, u64 physical_for_dev_replace);
307 static void copy_nocow_pages_worker(struct btrfs_work *work);
308 static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
309 static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
310 static void scrub_put_ctx(struct scrub_ctx *sctx);
313 static void scrub_pending_bio_inc(struct scrub_ctx *sctx)
315 atomic_inc(&sctx->refs);
316 atomic_inc(&sctx->bios_in_flight);
319 static void scrub_pending_bio_dec(struct scrub_ctx *sctx)
321 atomic_dec(&sctx->bios_in_flight);
322 wake_up(&sctx->list_wait);
326 static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
328 while (atomic_read(&fs_info->scrub_pause_req)) {
329 mutex_unlock(&fs_info->scrub_lock);
330 wait_event(fs_info->scrub_pause_wait,
331 atomic_read(&fs_info->scrub_pause_req) == 0);
332 mutex_lock(&fs_info->scrub_lock);
336 static void scrub_pause_on(struct btrfs_fs_info *fs_info)
338 atomic_inc(&fs_info->scrubs_paused);
339 wake_up(&fs_info->scrub_pause_wait);
342 static void scrub_pause_off(struct btrfs_fs_info *fs_info)
344 mutex_lock(&fs_info->scrub_lock);
345 __scrub_blocked_if_needed(fs_info);
346 atomic_dec(&fs_info->scrubs_paused);
347 mutex_unlock(&fs_info->scrub_lock);
349 wake_up(&fs_info->scrub_pause_wait);
352 static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
354 scrub_pause_on(fs_info);
355 scrub_pause_off(fs_info);
359 * used for workers that require transaction commits (i.e., for the
362 static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx)
364 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
366 atomic_inc(&sctx->refs);
368 * increment scrubs_running to prevent cancel requests from
369 * completing as long as a worker is running. we must also
370 * increment scrubs_paused to prevent deadlocking on pause
371 * requests used for transactions commits (as the worker uses a
372 * transaction context). it is safe to regard the worker
373 * as paused for all matters practical. effectively, we only
374 * avoid cancellation requests from completing.
376 mutex_lock(&fs_info->scrub_lock);
377 atomic_inc(&fs_info->scrubs_running);
378 atomic_inc(&fs_info->scrubs_paused);
379 mutex_unlock(&fs_info->scrub_lock);
382 * check if @scrubs_running=@scrubs_paused condition
383 * inside wait_event() is not an atomic operation.
384 * which means we may inc/dec @scrub_running/paused
385 * at any time. Let's wake up @scrub_pause_wait as
386 * much as we can to let commit transaction blocked less.
388 wake_up(&fs_info->scrub_pause_wait);
390 atomic_inc(&sctx->workers_pending);
393 /* used for workers that require transaction commits */
394 static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx)
396 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
399 * see scrub_pending_trans_workers_inc() why we're pretending
400 * to be paused in the scrub counters
402 mutex_lock(&fs_info->scrub_lock);
403 atomic_dec(&fs_info->scrubs_running);
404 atomic_dec(&fs_info->scrubs_paused);
405 mutex_unlock(&fs_info->scrub_lock);
406 atomic_dec(&sctx->workers_pending);
407 wake_up(&fs_info->scrub_pause_wait);
408 wake_up(&sctx->list_wait);
412 static void scrub_free_csums(struct scrub_ctx *sctx)
414 while (!list_empty(&sctx->csum_list)) {
415 struct btrfs_ordered_sum *sum;
416 sum = list_first_entry(&sctx->csum_list,
417 struct btrfs_ordered_sum, list);
418 list_del(&sum->list);
423 static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx)
430 scrub_free_wr_ctx(&sctx->wr_ctx);
432 /* this can happen when scrub is cancelled */
433 if (sctx->curr != -1) {
434 struct scrub_bio *sbio = sctx->bios[sctx->curr];
436 for (i = 0; i < sbio->page_count; i++) {
437 WARN_ON(!sbio->pagev[i]->page);
438 scrub_block_put(sbio->pagev[i]->sblock);
443 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
444 struct scrub_bio *sbio = sctx->bios[i];
451 scrub_free_csums(sctx);
455 static void scrub_put_ctx(struct scrub_ctx *sctx)
457 if (atomic_dec_and_test(&sctx->refs))
458 scrub_free_ctx(sctx);
461 static noinline_for_stack
462 struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace)
464 struct scrub_ctx *sctx;
466 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
469 sctx = kzalloc(sizeof(*sctx), GFP_NOFS);
472 atomic_set(&sctx->refs, 1);
473 sctx->is_dev_replace = is_dev_replace;
474 sctx->pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO;
476 sctx->dev_root = dev->dev_root;
477 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
478 struct scrub_bio *sbio;
480 sbio = kzalloc(sizeof(*sbio), GFP_NOFS);
483 sctx->bios[i] = sbio;
487 sbio->page_count = 0;
488 btrfs_init_work(&sbio->work, btrfs_scrub_helper,
489 scrub_bio_end_io_worker, NULL, NULL);
491 if (i != SCRUB_BIOS_PER_SCTX - 1)
492 sctx->bios[i]->next_free = i + 1;
494 sctx->bios[i]->next_free = -1;
496 sctx->first_free = 0;
497 sctx->nodesize = dev->dev_root->nodesize;
498 sctx->sectorsize = dev->dev_root->sectorsize;
499 atomic_set(&sctx->bios_in_flight, 0);
500 atomic_set(&sctx->workers_pending, 0);
501 atomic_set(&sctx->cancel_req, 0);
502 sctx->csum_size = btrfs_super_csum_size(fs_info->super_copy);
503 INIT_LIST_HEAD(&sctx->csum_list);
505 spin_lock_init(&sctx->list_lock);
506 spin_lock_init(&sctx->stat_lock);
507 init_waitqueue_head(&sctx->list_wait);
509 ret = scrub_setup_wr_ctx(sctx, &sctx->wr_ctx, fs_info,
510 fs_info->dev_replace.tgtdev, is_dev_replace);
512 scrub_free_ctx(sctx);
518 scrub_free_ctx(sctx);
519 return ERR_PTR(-ENOMEM);
522 static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root,
529 struct extent_buffer *eb;
530 struct btrfs_inode_item *inode_item;
531 struct scrub_warning *swarn = warn_ctx;
532 struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info;
533 struct inode_fs_paths *ipath = NULL;
534 struct btrfs_root *local_root;
535 struct btrfs_key root_key;
536 struct btrfs_key key;
538 root_key.objectid = root;
539 root_key.type = BTRFS_ROOT_ITEM_KEY;
540 root_key.offset = (u64)-1;
541 local_root = btrfs_read_fs_root_no_name(fs_info, &root_key);
542 if (IS_ERR(local_root)) {
543 ret = PTR_ERR(local_root);
548 * this makes the path point to (inum INODE_ITEM ioff)
551 key.type = BTRFS_INODE_ITEM_KEY;
554 ret = btrfs_search_slot(NULL, local_root, &key, swarn->path, 0, 0);
556 btrfs_release_path(swarn->path);
560 eb = swarn->path->nodes[0];
561 inode_item = btrfs_item_ptr(eb, swarn->path->slots[0],
562 struct btrfs_inode_item);
563 isize = btrfs_inode_size(eb, inode_item);
564 nlink = btrfs_inode_nlink(eb, inode_item);
565 btrfs_release_path(swarn->path);
567 ipath = init_ipath(4096, local_root, swarn->path);
569 ret = PTR_ERR(ipath);
573 ret = paths_from_inode(inum, ipath);
579 * we deliberately ignore the bit ipath might have been too small to
580 * hold all of the paths here
582 for (i = 0; i < ipath->fspath->elem_cnt; ++i)
583 printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
584 "%s, sector %llu, root %llu, inode %llu, offset %llu, "
585 "length %llu, links %u (path: %s)\n", swarn->errstr,
586 swarn->logical, rcu_str_deref(swarn->dev->name),
587 (unsigned long long)swarn->sector, root, inum, offset,
588 min(isize - offset, (u64)PAGE_SIZE), nlink,
589 (char *)(unsigned long)ipath->fspath->val[i]);
595 printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
596 "%s, sector %llu, root %llu, inode %llu, offset %llu: path "
597 "resolving failed with ret=%d\n", swarn->errstr,
598 swarn->logical, rcu_str_deref(swarn->dev->name),
599 (unsigned long long)swarn->sector, root, inum, offset, ret);
605 static void scrub_print_warning(const char *errstr, struct scrub_block *sblock)
607 struct btrfs_device *dev;
608 struct btrfs_fs_info *fs_info;
609 struct btrfs_path *path;
610 struct btrfs_key found_key;
611 struct extent_buffer *eb;
612 struct btrfs_extent_item *ei;
613 struct scrub_warning swarn;
614 unsigned long ptr = 0;
622 WARN_ON(sblock->page_count < 1);
623 dev = sblock->pagev[0]->dev;
624 fs_info = sblock->sctx->dev_root->fs_info;
626 path = btrfs_alloc_path();
630 swarn.sector = (sblock->pagev[0]->physical) >> 9;
631 swarn.logical = sblock->pagev[0]->logical;
632 swarn.errstr = errstr;
635 ret = extent_from_logical(fs_info, swarn.logical, path, &found_key,
640 extent_item_pos = swarn.logical - found_key.objectid;
641 swarn.extent_item_size = found_key.offset;
644 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
645 item_size = btrfs_item_size_nr(eb, path->slots[0]);
647 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
649 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
650 item_size, &ref_root,
652 printk_in_rcu(KERN_WARNING
653 "BTRFS: %s at logical %llu on dev %s, "
654 "sector %llu: metadata %s (level %d) in tree "
655 "%llu\n", errstr, swarn.logical,
656 rcu_str_deref(dev->name),
657 (unsigned long long)swarn.sector,
658 ref_level ? "node" : "leaf",
659 ret < 0 ? -1 : ref_level,
660 ret < 0 ? -1 : ref_root);
662 btrfs_release_path(path);
664 btrfs_release_path(path);
667 iterate_extent_inodes(fs_info, found_key.objectid,
669 scrub_print_warning_inode, &swarn);
673 btrfs_free_path(path);
676 static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx)
678 struct page *page = NULL;
680 struct scrub_fixup_nodatasum *fixup = fixup_ctx;
683 struct btrfs_key key;
684 struct inode *inode = NULL;
685 struct btrfs_fs_info *fs_info;
686 u64 end = offset + PAGE_SIZE - 1;
687 struct btrfs_root *local_root;
691 key.type = BTRFS_ROOT_ITEM_KEY;
692 key.offset = (u64)-1;
694 fs_info = fixup->root->fs_info;
695 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
697 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
698 if (IS_ERR(local_root)) {
699 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
700 return PTR_ERR(local_root);
703 key.type = BTRFS_INODE_ITEM_KEY;
706 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
707 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
709 return PTR_ERR(inode);
711 index = offset >> PAGE_CACHE_SHIFT;
713 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
719 if (PageUptodate(page)) {
720 if (PageDirty(page)) {
722 * we need to write the data to the defect sector. the
723 * data that was in that sector is not in memory,
724 * because the page was modified. we must not write the
725 * modified page to that sector.
727 * TODO: what could be done here: wait for the delalloc
728 * runner to write out that page (might involve
729 * COW) and see whether the sector is still
730 * referenced afterwards.
732 * For the meantime, we'll treat this error
733 * incorrectable, although there is a chance that a
734 * later scrub will find the bad sector again and that
735 * there's no dirty page in memory, then.
740 ret = repair_io_failure(inode, offset, PAGE_SIZE,
741 fixup->logical, page,
742 offset - page_offset(page),
748 * we need to get good data first. the general readpage path
749 * will call repair_io_failure for us, we just have to make
750 * sure we read the bad mirror.
752 ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
753 EXTENT_DAMAGED, GFP_NOFS);
755 /* set_extent_bits should give proper error */
762 ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page,
765 wait_on_page_locked(page);
767 corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset,
768 end, EXTENT_DAMAGED, 0, NULL);
770 clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
771 EXTENT_DAMAGED, GFP_NOFS);
783 if (ret == 0 && corrected) {
785 * we only need to call readpage for one of the inodes belonging
786 * to this extent. so make iterate_extent_inodes stop
794 static void scrub_fixup_nodatasum(struct btrfs_work *work)
797 struct scrub_fixup_nodatasum *fixup;
798 struct scrub_ctx *sctx;
799 struct btrfs_trans_handle *trans = NULL;
800 struct btrfs_path *path;
801 int uncorrectable = 0;
803 fixup = container_of(work, struct scrub_fixup_nodatasum, work);
806 path = btrfs_alloc_path();
808 spin_lock(&sctx->stat_lock);
809 ++sctx->stat.malloc_errors;
810 spin_unlock(&sctx->stat_lock);
815 trans = btrfs_join_transaction(fixup->root);
822 * the idea is to trigger a regular read through the standard path. we
823 * read a page from the (failed) logical address by specifying the
824 * corresponding copynum of the failed sector. thus, that readpage is
826 * that is the point where on-the-fly error correction will kick in
827 * (once it's finished) and rewrite the failed sector if a good copy
830 ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info,
831 path, scrub_fixup_readpage,
839 spin_lock(&sctx->stat_lock);
840 ++sctx->stat.corrected_errors;
841 spin_unlock(&sctx->stat_lock);
844 if (trans && !IS_ERR(trans))
845 btrfs_end_transaction(trans, fixup->root);
847 spin_lock(&sctx->stat_lock);
848 ++sctx->stat.uncorrectable_errors;
849 spin_unlock(&sctx->stat_lock);
850 btrfs_dev_replace_stats_inc(
851 &sctx->dev_root->fs_info->dev_replace.
852 num_uncorrectable_read_errors);
853 printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
854 "unable to fixup (nodatasum) error at logical %llu on dev %s\n",
855 fixup->logical, rcu_str_deref(fixup->dev->name));
858 btrfs_free_path(path);
861 scrub_pending_trans_workers_dec(sctx);
864 static inline void scrub_get_recover(struct scrub_recover *recover)
866 atomic_inc(&recover->refs);
869 static inline void scrub_put_recover(struct scrub_recover *recover)
871 if (atomic_dec_and_test(&recover->refs)) {
872 btrfs_put_bbio(recover->bbio);
878 * scrub_handle_errored_block gets called when either verification of the
879 * pages failed or the bio failed to read, e.g. with EIO. In the latter
880 * case, this function handles all pages in the bio, even though only one
882 * The goal of this function is to repair the errored block by using the
883 * contents of one of the mirrors.
885 static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
887 struct scrub_ctx *sctx = sblock_to_check->sctx;
888 struct btrfs_device *dev;
889 struct btrfs_fs_info *fs_info;
893 unsigned int failed_mirror_index;
894 unsigned int is_metadata;
895 unsigned int have_csum;
897 struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */
898 struct scrub_block *sblock_bad;
903 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
904 DEFAULT_RATELIMIT_BURST);
906 BUG_ON(sblock_to_check->page_count < 1);
907 fs_info = sctx->dev_root->fs_info;
908 if (sblock_to_check->pagev[0]->flags & BTRFS_EXTENT_FLAG_SUPER) {
910 * if we find an error in a super block, we just report it.
911 * They will get written with the next transaction commit
914 spin_lock(&sctx->stat_lock);
915 ++sctx->stat.super_errors;
916 spin_unlock(&sctx->stat_lock);
919 length = sblock_to_check->page_count * PAGE_SIZE;
920 logical = sblock_to_check->pagev[0]->logical;
921 generation = sblock_to_check->pagev[0]->generation;
922 BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1);
923 failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1;
924 is_metadata = !(sblock_to_check->pagev[0]->flags &
925 BTRFS_EXTENT_FLAG_DATA);
926 have_csum = sblock_to_check->pagev[0]->have_csum;
927 csum = sblock_to_check->pagev[0]->csum;
928 dev = sblock_to_check->pagev[0]->dev;
930 if (sctx->is_dev_replace && !is_metadata && !have_csum) {
931 sblocks_for_recheck = NULL;
936 * read all mirrors one after the other. This includes to
937 * re-read the extent or metadata block that failed (that was
938 * the cause that this fixup code is called) another time,
939 * page by page this time in order to know which pages
940 * caused I/O errors and which ones are good (for all mirrors).
941 * It is the goal to handle the situation when more than one
942 * mirror contains I/O errors, but the errors do not
943 * overlap, i.e. the data can be repaired by selecting the
944 * pages from those mirrors without I/O error on the
945 * particular pages. One example (with blocks >= 2 * PAGE_SIZE)
946 * would be that mirror #1 has an I/O error on the first page,
947 * the second page is good, and mirror #2 has an I/O error on
948 * the second page, but the first page is good.
949 * Then the first page of the first mirror can be repaired by
950 * taking the first page of the second mirror, and the
951 * second page of the second mirror can be repaired by
952 * copying the contents of the 2nd page of the 1st mirror.
953 * One more note: if the pages of one mirror contain I/O
954 * errors, the checksum cannot be verified. In order to get
955 * the best data for repairing, the first attempt is to find
956 * a mirror without I/O errors and with a validated checksum.
957 * Only if this is not possible, the pages are picked from
958 * mirrors with I/O errors without considering the checksum.
959 * If the latter is the case, at the end, the checksum of the
960 * repaired area is verified in order to correctly maintain
964 sblocks_for_recheck = kcalloc(BTRFS_MAX_MIRRORS,
965 sizeof(*sblocks_for_recheck), GFP_NOFS);
966 if (!sblocks_for_recheck) {
967 spin_lock(&sctx->stat_lock);
968 sctx->stat.malloc_errors++;
969 sctx->stat.read_errors++;
970 sctx->stat.uncorrectable_errors++;
971 spin_unlock(&sctx->stat_lock);
972 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
976 /* setup the context, map the logical blocks and alloc the pages */
977 ret = scrub_setup_recheck_block(sblock_to_check, sblocks_for_recheck);
979 spin_lock(&sctx->stat_lock);
980 sctx->stat.read_errors++;
981 sctx->stat.uncorrectable_errors++;
982 spin_unlock(&sctx->stat_lock);
983 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
986 BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS);
987 sblock_bad = sblocks_for_recheck + failed_mirror_index;
989 /* build and submit the bios for the failed mirror, check checksums */
990 scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum,
991 csum, generation, sctx->csum_size, 1);
993 if (!sblock_bad->header_error && !sblock_bad->checksum_error &&
994 sblock_bad->no_io_error_seen) {
996 * the error disappeared after reading page by page, or
997 * the area was part of a huge bio and other parts of the
998 * bio caused I/O errors, or the block layer merged several
999 * read requests into one and the error is caused by a
1000 * different bio (usually one of the two latter cases is
1003 spin_lock(&sctx->stat_lock);
1004 sctx->stat.unverified_errors++;
1005 sblock_to_check->data_corrected = 1;
1006 spin_unlock(&sctx->stat_lock);
1008 if (sctx->is_dev_replace)
1009 scrub_write_block_to_dev_replace(sblock_bad);
1013 if (!sblock_bad->no_io_error_seen) {
1014 spin_lock(&sctx->stat_lock);
1015 sctx->stat.read_errors++;
1016 spin_unlock(&sctx->stat_lock);
1017 if (__ratelimit(&_rs))
1018 scrub_print_warning("i/o error", sblock_to_check);
1019 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
1020 } else if (sblock_bad->checksum_error) {
1021 spin_lock(&sctx->stat_lock);
1022 sctx->stat.csum_errors++;
1023 spin_unlock(&sctx->stat_lock);
1024 if (__ratelimit(&_rs))
1025 scrub_print_warning("checksum error", sblock_to_check);
1026 btrfs_dev_stat_inc_and_print(dev,
1027 BTRFS_DEV_STAT_CORRUPTION_ERRS);
1028 } else if (sblock_bad->header_error) {
1029 spin_lock(&sctx->stat_lock);
1030 sctx->stat.verify_errors++;
1031 spin_unlock(&sctx->stat_lock);
1032 if (__ratelimit(&_rs))
1033 scrub_print_warning("checksum/header error",
1035 if (sblock_bad->generation_error)
1036 btrfs_dev_stat_inc_and_print(dev,
1037 BTRFS_DEV_STAT_GENERATION_ERRS);
1039 btrfs_dev_stat_inc_and_print(dev,
1040 BTRFS_DEV_STAT_CORRUPTION_ERRS);
1043 if (sctx->readonly) {
1044 ASSERT(!sctx->is_dev_replace);
1048 if (!is_metadata && !have_csum) {
1049 struct scrub_fixup_nodatasum *fixup_nodatasum;
1051 WARN_ON(sctx->is_dev_replace);
1056 * !is_metadata and !have_csum, this means that the data
1057 * might not be COW'ed, that it might be modified
1058 * concurrently. The general strategy to work on the
1059 * commit root does not help in the case when COW is not
1062 fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS);
1063 if (!fixup_nodatasum)
1064 goto did_not_correct_error;
1065 fixup_nodatasum->sctx = sctx;
1066 fixup_nodatasum->dev = dev;
1067 fixup_nodatasum->logical = logical;
1068 fixup_nodatasum->root = fs_info->extent_root;
1069 fixup_nodatasum->mirror_num = failed_mirror_index + 1;
1070 scrub_pending_trans_workers_inc(sctx);
1071 btrfs_init_work(&fixup_nodatasum->work, btrfs_scrub_helper,
1072 scrub_fixup_nodatasum, NULL, NULL);
1073 btrfs_queue_work(fs_info->scrub_workers,
1074 &fixup_nodatasum->work);
1079 * now build and submit the bios for the other mirrors, check
1081 * First try to pick the mirror which is completely without I/O
1082 * errors and also does not have a checksum error.
1083 * If one is found, and if a checksum is present, the full block
1084 * that is known to contain an error is rewritten. Afterwards
1085 * the block is known to be corrected.
1086 * If a mirror is found which is completely correct, and no
1087 * checksum is present, only those pages are rewritten that had
1088 * an I/O error in the block to be repaired, since it cannot be
1089 * determined, which copy of the other pages is better (and it
1090 * could happen otherwise that a correct page would be
1091 * overwritten by a bad one).
1093 for (mirror_index = 0;
1094 mirror_index < BTRFS_MAX_MIRRORS &&
1095 sblocks_for_recheck[mirror_index].page_count > 0;
1097 struct scrub_block *sblock_other;
1099 if (mirror_index == failed_mirror_index)
1101 sblock_other = sblocks_for_recheck + mirror_index;
1103 /* build and submit the bios, check checksums */
1104 scrub_recheck_block(fs_info, sblock_other, is_metadata,
1105 have_csum, csum, generation,
1106 sctx->csum_size, 0);
1108 if (!sblock_other->header_error &&
1109 !sblock_other->checksum_error &&
1110 sblock_other->no_io_error_seen) {
1111 if (sctx->is_dev_replace) {
1112 scrub_write_block_to_dev_replace(sblock_other);
1113 goto corrected_error;
1115 ret = scrub_repair_block_from_good_copy(
1116 sblock_bad, sblock_other);
1118 goto corrected_error;
1123 if (sblock_bad->no_io_error_seen && !sctx->is_dev_replace)
1124 goto did_not_correct_error;
1127 * In case of I/O errors in the area that is supposed to be
1128 * repaired, continue by picking good copies of those pages.
1129 * Select the good pages from mirrors to rewrite bad pages from
1130 * the area to fix. Afterwards verify the checksum of the block
1131 * that is supposed to be repaired. This verification step is
1132 * only done for the purpose of statistic counting and for the
1133 * final scrub report, whether errors remain.
1134 * A perfect algorithm could make use of the checksum and try
1135 * all possible combinations of pages from the different mirrors
1136 * until the checksum verification succeeds. For example, when
1137 * the 2nd page of mirror #1 faces I/O errors, and the 2nd page
1138 * of mirror #2 is readable but the final checksum test fails,
1139 * then the 2nd page of mirror #3 could be tried, whether now
1140 * the final checksum succeedes. But this would be a rare
1141 * exception and is therefore not implemented. At least it is
1142 * avoided that the good copy is overwritten.
1143 * A more useful improvement would be to pick the sectors
1144 * without I/O error based on sector sizes (512 bytes on legacy
1145 * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one
1146 * mirror could be repaired by taking 512 byte of a different
1147 * mirror, even if other 512 byte sectors in the same PAGE_SIZE
1148 * area are unreadable.
1151 for (page_num = 0; page_num < sblock_bad->page_count;
1153 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1154 struct scrub_block *sblock_other = NULL;
1156 /* skip no-io-error page in scrub */
1157 if (!page_bad->io_error && !sctx->is_dev_replace)
1160 /* try to find no-io-error page in mirrors */
1161 if (page_bad->io_error) {
1162 for (mirror_index = 0;
1163 mirror_index < BTRFS_MAX_MIRRORS &&
1164 sblocks_for_recheck[mirror_index].page_count > 0;
1166 if (!sblocks_for_recheck[mirror_index].
1167 pagev[page_num]->io_error) {
1168 sblock_other = sblocks_for_recheck +
1177 if (sctx->is_dev_replace) {
1179 * did not find a mirror to fetch the page
1180 * from. scrub_write_page_to_dev_replace()
1181 * handles this case (page->io_error), by
1182 * filling the block with zeros before
1183 * submitting the write request
1186 sblock_other = sblock_bad;
1188 if (scrub_write_page_to_dev_replace(sblock_other,
1190 btrfs_dev_replace_stats_inc(
1192 fs_info->dev_replace.
1196 } else if (sblock_other) {
1197 ret = scrub_repair_page_from_good_copy(sblock_bad,
1201 page_bad->io_error = 0;
1207 if (success && !sctx->is_dev_replace) {
1208 if (is_metadata || have_csum) {
1210 * need to verify the checksum now that all
1211 * sectors on disk are repaired (the write
1212 * request for data to be repaired is on its way).
1213 * Just be lazy and use scrub_recheck_block()
1214 * which re-reads the data before the checksum
1215 * is verified, but most likely the data comes out
1216 * of the page cache.
1218 scrub_recheck_block(fs_info, sblock_bad,
1219 is_metadata, have_csum, csum,
1220 generation, sctx->csum_size, 1);
1221 if (!sblock_bad->header_error &&
1222 !sblock_bad->checksum_error &&
1223 sblock_bad->no_io_error_seen)
1224 goto corrected_error;
1226 goto did_not_correct_error;
1229 spin_lock(&sctx->stat_lock);
1230 sctx->stat.corrected_errors++;
1231 sblock_to_check->data_corrected = 1;
1232 spin_unlock(&sctx->stat_lock);
1233 printk_ratelimited_in_rcu(KERN_ERR
1234 "BTRFS: fixed up error at logical %llu on dev %s\n",
1235 logical, rcu_str_deref(dev->name));
1238 did_not_correct_error:
1239 spin_lock(&sctx->stat_lock);
1240 sctx->stat.uncorrectable_errors++;
1241 spin_unlock(&sctx->stat_lock);
1242 printk_ratelimited_in_rcu(KERN_ERR
1243 "BTRFS: unable to fixup (regular) error at logical %llu on dev %s\n",
1244 logical, rcu_str_deref(dev->name));
1248 if (sblocks_for_recheck) {
1249 for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS;
1251 struct scrub_block *sblock = sblocks_for_recheck +
1253 struct scrub_recover *recover;
1256 for (page_index = 0; page_index < sblock->page_count;
1258 sblock->pagev[page_index]->sblock = NULL;
1259 recover = sblock->pagev[page_index]->recover;
1261 scrub_put_recover(recover);
1262 sblock->pagev[page_index]->recover =
1265 scrub_page_put(sblock->pagev[page_index]);
1268 kfree(sblocks_for_recheck);
1274 static inline int scrub_nr_raid_mirrors(struct btrfs_bio *bbio)
1276 if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID5)
1278 else if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID6)
1281 return (int)bbio->num_stripes;
1284 static inline void scrub_stripe_index_and_offset(u64 logical, u64 map_type,
1287 int nstripes, int mirror,
1293 if (map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
1295 for (i = 0; i < nstripes; i++) {
1296 if (raid_map[i] == RAID6_Q_STRIPE ||
1297 raid_map[i] == RAID5_P_STRIPE)
1300 if (logical >= raid_map[i] &&
1301 logical < raid_map[i] + mapped_length)
1306 *stripe_offset = logical - raid_map[i];
1308 /* The other RAID type */
1309 *stripe_index = mirror;
1314 static int scrub_setup_recheck_block(struct scrub_block *original_sblock,
1315 struct scrub_block *sblocks_for_recheck)
1317 struct scrub_ctx *sctx = original_sblock->sctx;
1318 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
1319 u64 length = original_sblock->page_count * PAGE_SIZE;
1320 u64 logical = original_sblock->pagev[0]->logical;
1321 struct scrub_recover *recover;
1322 struct btrfs_bio *bbio;
1333 * note: the two members refs and outstanding_pages
1334 * are not used (and not set) in the blocks that are used for
1335 * the recheck procedure
1338 while (length > 0) {
1339 sublen = min_t(u64, length, PAGE_SIZE);
1340 mapped_length = sublen;
1344 * with a length of PAGE_SIZE, each returned stripe
1345 * represents one mirror
1347 ret = btrfs_map_sblock(fs_info, REQ_GET_READ_MIRRORS, logical,
1348 &mapped_length, &bbio, 0, 1);
1349 if (ret || !bbio || mapped_length < sublen) {
1350 btrfs_put_bbio(bbio);
1354 recover = kzalloc(sizeof(struct scrub_recover), GFP_NOFS);
1356 btrfs_put_bbio(bbio);
1360 atomic_set(&recover->refs, 1);
1361 recover->bbio = bbio;
1362 recover->map_length = mapped_length;
1364 BUG_ON(page_index >= SCRUB_PAGES_PER_RD_BIO);
1366 nmirrors = min(scrub_nr_raid_mirrors(bbio), BTRFS_MAX_MIRRORS);
1368 for (mirror_index = 0; mirror_index < nmirrors;
1370 struct scrub_block *sblock;
1371 struct scrub_page *page;
1373 sblock = sblocks_for_recheck + mirror_index;
1374 sblock->sctx = sctx;
1375 page = kzalloc(sizeof(*page), GFP_NOFS);
1378 spin_lock(&sctx->stat_lock);
1379 sctx->stat.malloc_errors++;
1380 spin_unlock(&sctx->stat_lock);
1381 scrub_put_recover(recover);
1384 scrub_page_get(page);
1385 sblock->pagev[page_index] = page;
1386 page->logical = logical;
1388 scrub_stripe_index_and_offset(logical,
1397 page->physical = bbio->stripes[stripe_index].physical +
1399 page->dev = bbio->stripes[stripe_index].dev;
1401 BUG_ON(page_index >= original_sblock->page_count);
1402 page->physical_for_dev_replace =
1403 original_sblock->pagev[page_index]->
1404 physical_for_dev_replace;
1405 /* for missing devices, dev->bdev is NULL */
1406 page->mirror_num = mirror_index + 1;
1407 sblock->page_count++;
1408 page->page = alloc_page(GFP_NOFS);
1412 scrub_get_recover(recover);
1413 page->recover = recover;
1415 scrub_put_recover(recover);
1424 struct scrub_bio_ret {
1425 struct completion event;
1429 static void scrub_bio_wait_endio(struct bio *bio)
1431 struct scrub_bio_ret *ret = bio->bi_private;
1433 ret->error = bio->bi_error;
1434 complete(&ret->event);
1437 static inline int scrub_is_page_on_raid56(struct scrub_page *page)
1439 return page->recover &&
1440 (page->recover->bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK);
1443 static int scrub_submit_raid56_bio_wait(struct btrfs_fs_info *fs_info,
1445 struct scrub_page *page)
1447 struct scrub_bio_ret done;
1450 init_completion(&done.event);
1452 bio->bi_iter.bi_sector = page->logical >> 9;
1453 bio->bi_private = &done;
1454 bio->bi_end_io = scrub_bio_wait_endio;
1456 ret = raid56_parity_recover(fs_info->fs_root, bio, page->recover->bbio,
1457 page->recover->map_length,
1458 page->mirror_num, 0);
1462 wait_for_completion(&done.event);
1470 * this function will check the on disk data for checksum errors, header
1471 * errors and read I/O errors. If any I/O errors happen, the exact pages
1472 * which are errored are marked as being bad. The goal is to enable scrub
1473 * to take those pages that are not errored from all the mirrors so that
1474 * the pages that are errored in the just handled mirror can be repaired.
1476 static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
1477 struct scrub_block *sblock, int is_metadata,
1478 int have_csum, u8 *csum, u64 generation,
1479 u16 csum_size, int retry_failed_mirror)
1483 sblock->no_io_error_seen = 1;
1484 sblock->header_error = 0;
1485 sblock->checksum_error = 0;
1487 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1489 struct scrub_page *page = sblock->pagev[page_num];
1491 if (page->dev->bdev == NULL) {
1493 sblock->no_io_error_seen = 0;
1497 WARN_ON(!page->page);
1498 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1501 sblock->no_io_error_seen = 0;
1504 bio->bi_bdev = page->dev->bdev;
1506 bio_add_page(bio, page->page, PAGE_SIZE, 0);
1507 if (!retry_failed_mirror && scrub_is_page_on_raid56(page)) {
1508 if (scrub_submit_raid56_bio_wait(fs_info, bio, page))
1509 sblock->no_io_error_seen = 0;
1511 bio->bi_iter.bi_sector = page->physical >> 9;
1513 if (btrfsic_submit_bio_wait(READ, bio))
1514 sblock->no_io_error_seen = 0;
1520 if (sblock->no_io_error_seen)
1521 scrub_recheck_block_checksum(fs_info, sblock, is_metadata,
1522 have_csum, csum, generation,
1528 static inline int scrub_check_fsid(u8 fsid[],
1529 struct scrub_page *spage)
1531 struct btrfs_fs_devices *fs_devices = spage->dev->fs_devices;
1534 ret = memcmp(fsid, fs_devices->fsid, BTRFS_UUID_SIZE);
1538 static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
1539 struct scrub_block *sblock,
1540 int is_metadata, int have_csum,
1541 const u8 *csum, u64 generation,
1545 u8 calculated_csum[BTRFS_CSUM_SIZE];
1547 void *mapped_buffer;
1549 WARN_ON(!sblock->pagev[0]->page);
1551 struct btrfs_header *h;
1553 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
1554 h = (struct btrfs_header *)mapped_buffer;
1556 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h) ||
1557 !scrub_check_fsid(h->fsid, sblock->pagev[0]) ||
1558 memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
1560 sblock->header_error = 1;
1561 } else if (generation != btrfs_stack_header_generation(h)) {
1562 sblock->header_error = 1;
1563 sblock->generation_error = 1;
1570 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
1573 for (page_num = 0;;) {
1574 if (page_num == 0 && is_metadata)
1575 crc = btrfs_csum_data(
1576 ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE,
1577 crc, PAGE_SIZE - BTRFS_CSUM_SIZE);
1579 crc = btrfs_csum_data(mapped_buffer, crc, PAGE_SIZE);
1581 kunmap_atomic(mapped_buffer);
1583 if (page_num >= sblock->page_count)
1585 WARN_ON(!sblock->pagev[page_num]->page);
1587 mapped_buffer = kmap_atomic(sblock->pagev[page_num]->page);
1590 btrfs_csum_final(crc, calculated_csum);
1591 if (memcmp(calculated_csum, csum, csum_size))
1592 sblock->checksum_error = 1;
1595 static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
1596 struct scrub_block *sblock_good)
1601 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
1604 ret_sub = scrub_repair_page_from_good_copy(sblock_bad,
1614 static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
1615 struct scrub_block *sblock_good,
1616 int page_num, int force_write)
1618 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1619 struct scrub_page *page_good = sblock_good->pagev[page_num];
1621 BUG_ON(page_bad->page == NULL);
1622 BUG_ON(page_good->page == NULL);
1623 if (force_write || sblock_bad->header_error ||
1624 sblock_bad->checksum_error || page_bad->io_error) {
1628 if (!page_bad->dev->bdev) {
1629 printk_ratelimited(KERN_WARNING "BTRFS: "
1630 "scrub_repair_page_from_good_copy(bdev == NULL) "
1631 "is unexpected!\n");
1635 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1638 bio->bi_bdev = page_bad->dev->bdev;
1639 bio->bi_iter.bi_sector = page_bad->physical >> 9;
1641 ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0);
1642 if (PAGE_SIZE != ret) {
1647 if (btrfsic_submit_bio_wait(WRITE, bio)) {
1648 btrfs_dev_stat_inc_and_print(page_bad->dev,
1649 BTRFS_DEV_STAT_WRITE_ERRS);
1650 btrfs_dev_replace_stats_inc(
1651 &sblock_bad->sctx->dev_root->fs_info->
1652 dev_replace.num_write_errors);
1662 static void scrub_write_block_to_dev_replace(struct scrub_block *sblock)
1667 * This block is used for the check of the parity on the source device,
1668 * so the data needn't be written into the destination device.
1670 if (sblock->sparity)
1673 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1676 ret = scrub_write_page_to_dev_replace(sblock, page_num);
1678 btrfs_dev_replace_stats_inc(
1679 &sblock->sctx->dev_root->fs_info->dev_replace.
1684 static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
1687 struct scrub_page *spage = sblock->pagev[page_num];
1689 BUG_ON(spage->page == NULL);
1690 if (spage->io_error) {
1691 void *mapped_buffer = kmap_atomic(spage->page);
1693 memset(mapped_buffer, 0, PAGE_CACHE_SIZE);
1694 flush_dcache_page(spage->page);
1695 kunmap_atomic(mapped_buffer);
1697 return scrub_add_page_to_wr_bio(sblock->sctx, spage);
1700 static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
1701 struct scrub_page *spage)
1703 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1704 struct scrub_bio *sbio;
1707 mutex_lock(&wr_ctx->wr_lock);
1709 if (!wr_ctx->wr_curr_bio) {
1710 wr_ctx->wr_curr_bio = kzalloc(sizeof(*wr_ctx->wr_curr_bio),
1712 if (!wr_ctx->wr_curr_bio) {
1713 mutex_unlock(&wr_ctx->wr_lock);
1716 wr_ctx->wr_curr_bio->sctx = sctx;
1717 wr_ctx->wr_curr_bio->page_count = 0;
1719 sbio = wr_ctx->wr_curr_bio;
1720 if (sbio->page_count == 0) {
1723 sbio->physical = spage->physical_for_dev_replace;
1724 sbio->logical = spage->logical;
1725 sbio->dev = wr_ctx->tgtdev;
1728 bio = btrfs_io_bio_alloc(GFP_NOFS, wr_ctx->pages_per_wr_bio);
1730 mutex_unlock(&wr_ctx->wr_lock);
1736 bio->bi_private = sbio;
1737 bio->bi_end_io = scrub_wr_bio_end_io;
1738 bio->bi_bdev = sbio->dev->bdev;
1739 bio->bi_iter.bi_sector = sbio->physical >> 9;
1741 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
1742 spage->physical_for_dev_replace ||
1743 sbio->logical + sbio->page_count * PAGE_SIZE !=
1745 scrub_wr_submit(sctx);
1749 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
1750 if (ret != PAGE_SIZE) {
1751 if (sbio->page_count < 1) {
1754 mutex_unlock(&wr_ctx->wr_lock);
1757 scrub_wr_submit(sctx);
1761 sbio->pagev[sbio->page_count] = spage;
1762 scrub_page_get(spage);
1764 if (sbio->page_count == wr_ctx->pages_per_wr_bio)
1765 scrub_wr_submit(sctx);
1766 mutex_unlock(&wr_ctx->wr_lock);
1771 static void scrub_wr_submit(struct scrub_ctx *sctx)
1773 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1774 struct scrub_bio *sbio;
1776 if (!wr_ctx->wr_curr_bio)
1779 sbio = wr_ctx->wr_curr_bio;
1780 wr_ctx->wr_curr_bio = NULL;
1781 WARN_ON(!sbio->bio->bi_bdev);
1782 scrub_pending_bio_inc(sctx);
1783 /* process all writes in a single worker thread. Then the block layer
1784 * orders the requests before sending them to the driver which
1785 * doubled the write performance on spinning disks when measured
1787 btrfsic_submit_bio(WRITE, sbio->bio);
1790 static void scrub_wr_bio_end_io(struct bio *bio)
1792 struct scrub_bio *sbio = bio->bi_private;
1793 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
1795 sbio->err = bio->bi_error;
1798 btrfs_init_work(&sbio->work, btrfs_scrubwrc_helper,
1799 scrub_wr_bio_end_io_worker, NULL, NULL);
1800 btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work);
1803 static void scrub_wr_bio_end_io_worker(struct btrfs_work *work)
1805 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
1806 struct scrub_ctx *sctx = sbio->sctx;
1809 WARN_ON(sbio->page_count > SCRUB_PAGES_PER_WR_BIO);
1811 struct btrfs_dev_replace *dev_replace =
1812 &sbio->sctx->dev_root->fs_info->dev_replace;
1814 for (i = 0; i < sbio->page_count; i++) {
1815 struct scrub_page *spage = sbio->pagev[i];
1817 spage->io_error = 1;
1818 btrfs_dev_replace_stats_inc(&dev_replace->
1823 for (i = 0; i < sbio->page_count; i++)
1824 scrub_page_put(sbio->pagev[i]);
1828 scrub_pending_bio_dec(sctx);
1831 static int scrub_checksum(struct scrub_block *sblock)
1836 WARN_ON(sblock->page_count < 1);
1837 flags = sblock->pagev[0]->flags;
1839 if (flags & BTRFS_EXTENT_FLAG_DATA)
1840 ret = scrub_checksum_data(sblock);
1841 else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1842 ret = scrub_checksum_tree_block(sblock);
1843 else if (flags & BTRFS_EXTENT_FLAG_SUPER)
1844 (void)scrub_checksum_super(sblock);
1848 scrub_handle_errored_block(sblock);
1853 static int scrub_checksum_data(struct scrub_block *sblock)
1855 struct scrub_ctx *sctx = sblock->sctx;
1856 u8 csum[BTRFS_CSUM_SIZE];
1865 BUG_ON(sblock->page_count < 1);
1866 if (!sblock->pagev[0]->have_csum)
1869 on_disk_csum = sblock->pagev[0]->csum;
1870 page = sblock->pagev[0]->page;
1871 buffer = kmap_atomic(page);
1873 len = sctx->sectorsize;
1876 u64 l = min_t(u64, len, PAGE_SIZE);
1878 crc = btrfs_csum_data(buffer, crc, l);
1879 kunmap_atomic(buffer);
1884 BUG_ON(index >= sblock->page_count);
1885 BUG_ON(!sblock->pagev[index]->page);
1886 page = sblock->pagev[index]->page;
1887 buffer = kmap_atomic(page);
1890 btrfs_csum_final(crc, csum);
1891 if (memcmp(csum, on_disk_csum, sctx->csum_size))
1897 static int scrub_checksum_tree_block(struct scrub_block *sblock)
1899 struct scrub_ctx *sctx = sblock->sctx;
1900 struct btrfs_header *h;
1901 struct btrfs_root *root = sctx->dev_root;
1902 struct btrfs_fs_info *fs_info = root->fs_info;
1903 u8 calculated_csum[BTRFS_CSUM_SIZE];
1904 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1906 void *mapped_buffer;
1915 BUG_ON(sblock->page_count < 1);
1916 page = sblock->pagev[0]->page;
1917 mapped_buffer = kmap_atomic(page);
1918 h = (struct btrfs_header *)mapped_buffer;
1919 memcpy(on_disk_csum, h->csum, sctx->csum_size);
1922 * we don't use the getter functions here, as we
1923 * a) don't have an extent buffer and
1924 * b) the page is already kmapped
1927 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h))
1930 if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h))
1933 if (!scrub_check_fsid(h->fsid, sblock->pagev[0]))
1936 if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
1940 len = sctx->nodesize - BTRFS_CSUM_SIZE;
1941 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
1942 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
1945 u64 l = min_t(u64, len, mapped_size);
1947 crc = btrfs_csum_data(p, crc, l);
1948 kunmap_atomic(mapped_buffer);
1953 BUG_ON(index >= sblock->page_count);
1954 BUG_ON(!sblock->pagev[index]->page);
1955 page = sblock->pagev[index]->page;
1956 mapped_buffer = kmap_atomic(page);
1957 mapped_size = PAGE_SIZE;
1961 btrfs_csum_final(crc, calculated_csum);
1962 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
1965 return fail || crc_fail;
1968 static int scrub_checksum_super(struct scrub_block *sblock)
1970 struct btrfs_super_block *s;
1971 struct scrub_ctx *sctx = sblock->sctx;
1972 u8 calculated_csum[BTRFS_CSUM_SIZE];
1973 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1975 void *mapped_buffer;
1984 BUG_ON(sblock->page_count < 1);
1985 page = sblock->pagev[0]->page;
1986 mapped_buffer = kmap_atomic(page);
1987 s = (struct btrfs_super_block *)mapped_buffer;
1988 memcpy(on_disk_csum, s->csum, sctx->csum_size);
1990 if (sblock->pagev[0]->logical != btrfs_super_bytenr(s))
1993 if (sblock->pagev[0]->generation != btrfs_super_generation(s))
1996 if (!scrub_check_fsid(s->fsid, sblock->pagev[0]))
1999 len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE;
2000 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
2001 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
2004 u64 l = min_t(u64, len, mapped_size);
2006 crc = btrfs_csum_data(p, crc, l);
2007 kunmap_atomic(mapped_buffer);
2012 BUG_ON(index >= sblock->page_count);
2013 BUG_ON(!sblock->pagev[index]->page);
2014 page = sblock->pagev[index]->page;
2015 mapped_buffer = kmap_atomic(page);
2016 mapped_size = PAGE_SIZE;
2020 btrfs_csum_final(crc, calculated_csum);
2021 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
2024 if (fail_cor + fail_gen) {
2026 * if we find an error in a super block, we just report it.
2027 * They will get written with the next transaction commit
2030 spin_lock(&sctx->stat_lock);
2031 ++sctx->stat.super_errors;
2032 spin_unlock(&sctx->stat_lock);
2034 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
2035 BTRFS_DEV_STAT_CORRUPTION_ERRS);
2037 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
2038 BTRFS_DEV_STAT_GENERATION_ERRS);
2041 return fail_cor + fail_gen;
2044 static void scrub_block_get(struct scrub_block *sblock)
2046 atomic_inc(&sblock->refs);
2049 static void scrub_block_put(struct scrub_block *sblock)
2051 if (atomic_dec_and_test(&sblock->refs)) {
2054 if (sblock->sparity)
2055 scrub_parity_put(sblock->sparity);
2057 for (i = 0; i < sblock->page_count; i++)
2058 scrub_page_put(sblock->pagev[i]);
2063 static void scrub_page_get(struct scrub_page *spage)
2065 atomic_inc(&spage->refs);
2068 static void scrub_page_put(struct scrub_page *spage)
2070 if (atomic_dec_and_test(&spage->refs)) {
2072 __free_page(spage->page);
2077 static void scrub_submit(struct scrub_ctx *sctx)
2079 struct scrub_bio *sbio;
2081 if (sctx->curr == -1)
2084 sbio = sctx->bios[sctx->curr];
2086 scrub_pending_bio_inc(sctx);
2087 btrfsic_submit_bio(READ, sbio->bio);
2090 static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
2091 struct scrub_page *spage)
2093 struct scrub_block *sblock = spage->sblock;
2094 struct scrub_bio *sbio;
2099 * grab a fresh bio or wait for one to become available
2101 while (sctx->curr == -1) {
2102 spin_lock(&sctx->list_lock);
2103 sctx->curr = sctx->first_free;
2104 if (sctx->curr != -1) {
2105 sctx->first_free = sctx->bios[sctx->curr]->next_free;
2106 sctx->bios[sctx->curr]->next_free = -1;
2107 sctx->bios[sctx->curr]->page_count = 0;
2108 spin_unlock(&sctx->list_lock);
2110 spin_unlock(&sctx->list_lock);
2111 wait_event(sctx->list_wait, sctx->first_free != -1);
2114 sbio = sctx->bios[sctx->curr];
2115 if (sbio->page_count == 0) {
2118 sbio->physical = spage->physical;
2119 sbio->logical = spage->logical;
2120 sbio->dev = spage->dev;
2123 bio = btrfs_io_bio_alloc(GFP_NOFS, sctx->pages_per_rd_bio);
2129 bio->bi_private = sbio;
2130 bio->bi_end_io = scrub_bio_end_io;
2131 bio->bi_bdev = sbio->dev->bdev;
2132 bio->bi_iter.bi_sector = sbio->physical >> 9;
2134 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
2136 sbio->logical + sbio->page_count * PAGE_SIZE !=
2138 sbio->dev != spage->dev) {
2143 sbio->pagev[sbio->page_count] = spage;
2144 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
2145 if (ret != PAGE_SIZE) {
2146 if (sbio->page_count < 1) {
2155 scrub_block_get(sblock); /* one for the page added to the bio */
2156 atomic_inc(&sblock->outstanding_pages);
2158 if (sbio->page_count == sctx->pages_per_rd_bio)
2164 static void scrub_missing_raid56_end_io(struct bio *bio)
2166 struct scrub_block *sblock = bio->bi_private;
2167 struct btrfs_fs_info *fs_info = sblock->sctx->dev_root->fs_info;
2170 sblock->no_io_error_seen = 0;
2172 btrfs_queue_work(fs_info->scrub_workers, &sblock->work);
2175 static void scrub_missing_raid56_worker(struct btrfs_work *work)
2177 struct scrub_block *sblock = container_of(work, struct scrub_block, work);
2178 struct scrub_ctx *sctx = sblock->sctx;
2179 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
2180 unsigned int is_metadata;
2181 unsigned int have_csum;
2185 struct btrfs_device *dev;
2187 is_metadata = !(sblock->pagev[0]->flags & BTRFS_EXTENT_FLAG_DATA);
2188 have_csum = sblock->pagev[0]->have_csum;
2189 csum = sblock->pagev[0]->csum;
2190 generation = sblock->pagev[0]->generation;
2191 logical = sblock->pagev[0]->logical;
2192 dev = sblock->pagev[0]->dev;
2194 if (sblock->no_io_error_seen) {
2195 scrub_recheck_block_checksum(fs_info, sblock, is_metadata,
2196 have_csum, csum, generation,
2200 if (!sblock->no_io_error_seen) {
2201 spin_lock(&sctx->stat_lock);
2202 sctx->stat.read_errors++;
2203 spin_unlock(&sctx->stat_lock);
2204 printk_ratelimited_in_rcu(KERN_ERR
2205 "BTRFS: I/O error rebulding logical %llu for dev %s\n",
2206 logical, rcu_str_deref(dev->name));
2207 } else if (sblock->header_error || sblock->checksum_error) {
2208 spin_lock(&sctx->stat_lock);
2209 sctx->stat.uncorrectable_errors++;
2210 spin_unlock(&sctx->stat_lock);
2211 printk_ratelimited_in_rcu(KERN_ERR
2212 "BTRFS: failed to rebuild valid logical %llu for dev %s\n",
2213 logical, rcu_str_deref(dev->name));
2215 scrub_write_block_to_dev_replace(sblock);
2218 scrub_block_put(sblock);
2220 if (sctx->is_dev_replace &&
2221 atomic_read(&sctx->wr_ctx.flush_all_writes)) {
2222 mutex_lock(&sctx->wr_ctx.wr_lock);
2223 scrub_wr_submit(sctx);
2224 mutex_unlock(&sctx->wr_ctx.wr_lock);
2227 scrub_pending_bio_dec(sctx);
2230 static void scrub_missing_raid56_pages(struct scrub_block *sblock)
2232 struct scrub_ctx *sctx = sblock->sctx;
2233 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
2234 u64 length = sblock->page_count * PAGE_SIZE;
2235 u64 logical = sblock->pagev[0]->logical;
2236 struct btrfs_bio *bbio;
2238 struct btrfs_raid_bio *rbio;
2242 ret = btrfs_map_sblock(fs_info, REQ_GET_READ_MIRRORS, logical, &length,
2244 if (ret || !bbio || !bbio->raid_map)
2247 if (WARN_ON(!sctx->is_dev_replace ||
2248 !(bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK))) {
2250 * We shouldn't be scrubbing a missing device. Even for dev
2251 * replace, we should only get here for RAID 5/6. We either
2252 * managed to mount something with no mirrors remaining or
2253 * there's a bug in scrub_remap_extent()/btrfs_map_block().
2258 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
2262 bio->bi_iter.bi_sector = logical >> 9;
2263 bio->bi_private = sblock;
2264 bio->bi_end_io = scrub_missing_raid56_end_io;
2266 rbio = raid56_alloc_missing_rbio(sctx->dev_root, bio, bbio, length);
2270 for (i = 0; i < sblock->page_count; i++) {
2271 struct scrub_page *spage = sblock->pagev[i];
2273 raid56_add_scrub_pages(rbio, spage->page, spage->logical);
2276 btrfs_init_work(&sblock->work, btrfs_scrub_helper,
2277 scrub_missing_raid56_worker, NULL, NULL);
2278 scrub_block_get(sblock);
2279 scrub_pending_bio_inc(sctx);
2280 raid56_submit_missing_rbio(rbio);
2286 btrfs_put_bbio(bbio);
2287 spin_lock(&sctx->stat_lock);
2288 sctx->stat.malloc_errors++;
2289 spin_unlock(&sctx->stat_lock);
2292 static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
2293 u64 physical, struct btrfs_device *dev, u64 flags,
2294 u64 gen, int mirror_num, u8 *csum, int force,
2295 u64 physical_for_dev_replace)
2297 struct scrub_block *sblock;
2300 sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
2302 spin_lock(&sctx->stat_lock);
2303 sctx->stat.malloc_errors++;
2304 spin_unlock(&sctx->stat_lock);
2308 /* one ref inside this function, plus one for each page added to
2310 atomic_set(&sblock->refs, 1);
2311 sblock->sctx = sctx;
2312 sblock->no_io_error_seen = 1;
2314 for (index = 0; len > 0; index++) {
2315 struct scrub_page *spage;
2316 u64 l = min_t(u64, len, PAGE_SIZE);
2318 spage = kzalloc(sizeof(*spage), GFP_NOFS);
2321 spin_lock(&sctx->stat_lock);
2322 sctx->stat.malloc_errors++;
2323 spin_unlock(&sctx->stat_lock);
2324 scrub_block_put(sblock);
2327 BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
2328 scrub_page_get(spage);
2329 sblock->pagev[index] = spage;
2330 spage->sblock = sblock;
2332 spage->flags = flags;
2333 spage->generation = gen;
2334 spage->logical = logical;
2335 spage->physical = physical;
2336 spage->physical_for_dev_replace = physical_for_dev_replace;
2337 spage->mirror_num = mirror_num;
2339 spage->have_csum = 1;
2340 memcpy(spage->csum, csum, sctx->csum_size);
2342 spage->have_csum = 0;
2344 sblock->page_count++;
2345 spage->page = alloc_page(GFP_NOFS);
2351 physical_for_dev_replace += l;
2354 WARN_ON(sblock->page_count == 0);
2357 * This case should only be hit for RAID 5/6 device replace. See
2358 * the comment in scrub_missing_raid56_pages() for details.
2360 scrub_missing_raid56_pages(sblock);
2362 for (index = 0; index < sblock->page_count; index++) {
2363 struct scrub_page *spage = sblock->pagev[index];
2366 ret = scrub_add_page_to_rd_bio(sctx, spage);
2368 scrub_block_put(sblock);
2377 /* last one frees, either here or in bio completion for last page */
2378 scrub_block_put(sblock);
2382 static void scrub_bio_end_io(struct bio *bio)
2384 struct scrub_bio *sbio = bio->bi_private;
2385 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
2387 sbio->err = bio->bi_error;
2390 btrfs_queue_work(fs_info->scrub_workers, &sbio->work);
2393 static void scrub_bio_end_io_worker(struct btrfs_work *work)
2395 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
2396 struct scrub_ctx *sctx = sbio->sctx;
2399 BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO);
2401 for (i = 0; i < sbio->page_count; i++) {
2402 struct scrub_page *spage = sbio->pagev[i];
2404 spage->io_error = 1;
2405 spage->sblock->no_io_error_seen = 0;
2409 /* now complete the scrub_block items that have all pages completed */
2410 for (i = 0; i < sbio->page_count; i++) {
2411 struct scrub_page *spage = sbio->pagev[i];
2412 struct scrub_block *sblock = spage->sblock;
2414 if (atomic_dec_and_test(&sblock->outstanding_pages))
2415 scrub_block_complete(sblock);
2416 scrub_block_put(sblock);
2421 spin_lock(&sctx->list_lock);
2422 sbio->next_free = sctx->first_free;
2423 sctx->first_free = sbio->index;
2424 spin_unlock(&sctx->list_lock);
2426 if (sctx->is_dev_replace &&
2427 atomic_read(&sctx->wr_ctx.flush_all_writes)) {
2428 mutex_lock(&sctx->wr_ctx.wr_lock);
2429 scrub_wr_submit(sctx);
2430 mutex_unlock(&sctx->wr_ctx.wr_lock);
2433 scrub_pending_bio_dec(sctx);
2436 static inline void __scrub_mark_bitmap(struct scrub_parity *sparity,
2437 unsigned long *bitmap,
2442 int sectorsize = sparity->sctx->dev_root->sectorsize;
2444 if (len >= sparity->stripe_len) {
2445 bitmap_set(bitmap, 0, sparity->nsectors);
2449 start -= sparity->logic_start;
2450 start = div_u64_rem(start, sparity->stripe_len, &offset);
2451 offset /= sectorsize;
2452 nsectors = (int)len / sectorsize;
2454 if (offset + nsectors <= sparity->nsectors) {
2455 bitmap_set(bitmap, offset, nsectors);
2459 bitmap_set(bitmap, offset, sparity->nsectors - offset);
2460 bitmap_set(bitmap, 0, nsectors - (sparity->nsectors - offset));
2463 static inline void scrub_parity_mark_sectors_error(struct scrub_parity *sparity,
2466 __scrub_mark_bitmap(sparity, sparity->ebitmap, start, len);
2469 static inline void scrub_parity_mark_sectors_data(struct scrub_parity *sparity,
2472 __scrub_mark_bitmap(sparity, sparity->dbitmap, start, len);
2475 static void scrub_block_complete(struct scrub_block *sblock)
2479 if (!sblock->no_io_error_seen) {
2481 scrub_handle_errored_block(sblock);
2484 * if has checksum error, write via repair mechanism in
2485 * dev replace case, otherwise write here in dev replace
2488 corrupted = scrub_checksum(sblock);
2489 if (!corrupted && sblock->sctx->is_dev_replace)
2490 scrub_write_block_to_dev_replace(sblock);
2493 if (sblock->sparity && corrupted && !sblock->data_corrected) {
2494 u64 start = sblock->pagev[0]->logical;
2495 u64 end = sblock->pagev[sblock->page_count - 1]->logical +
2498 scrub_parity_mark_sectors_error(sblock->sparity,
2499 start, end - start);
2503 static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len,
2506 struct btrfs_ordered_sum *sum = NULL;
2507 unsigned long index;
2508 unsigned long num_sectors;
2510 while (!list_empty(&sctx->csum_list)) {
2511 sum = list_first_entry(&sctx->csum_list,
2512 struct btrfs_ordered_sum, list);
2513 if (sum->bytenr > logical)
2515 if (sum->bytenr + sum->len > logical)
2518 ++sctx->stat.csum_discards;
2519 list_del(&sum->list);
2526 index = ((u32)(logical - sum->bytenr)) / sctx->sectorsize;
2527 num_sectors = sum->len / sctx->sectorsize;
2528 memcpy(csum, sum->sums + index, sctx->csum_size);
2529 if (index == num_sectors - 1) {
2530 list_del(&sum->list);
2536 /* scrub extent tries to collect up to 64 kB for each bio */
2537 static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len,
2538 u64 physical, struct btrfs_device *dev, u64 flags,
2539 u64 gen, int mirror_num, u64 physical_for_dev_replace)
2542 u8 csum[BTRFS_CSUM_SIZE];
2545 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2546 blocksize = sctx->sectorsize;
2547 spin_lock(&sctx->stat_lock);
2548 sctx->stat.data_extents_scrubbed++;
2549 sctx->stat.data_bytes_scrubbed += len;
2550 spin_unlock(&sctx->stat_lock);
2551 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2552 blocksize = sctx->nodesize;
2553 spin_lock(&sctx->stat_lock);
2554 sctx->stat.tree_extents_scrubbed++;
2555 sctx->stat.tree_bytes_scrubbed += len;
2556 spin_unlock(&sctx->stat_lock);
2558 blocksize = sctx->sectorsize;
2563 u64 l = min_t(u64, len, blocksize);
2566 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2567 /* push csums to sbio */
2568 have_csum = scrub_find_csum(sctx, logical, l, csum);
2570 ++sctx->stat.no_csum;
2571 if (sctx->is_dev_replace && !have_csum) {
2572 ret = copy_nocow_pages(sctx, logical, l,
2574 physical_for_dev_replace);
2575 goto behind_scrub_pages;
2578 ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen,
2579 mirror_num, have_csum ? csum : NULL, 0,
2580 physical_for_dev_replace);
2587 physical_for_dev_replace += l;
2592 static int scrub_pages_for_parity(struct scrub_parity *sparity,
2593 u64 logical, u64 len,
2594 u64 physical, struct btrfs_device *dev,
2595 u64 flags, u64 gen, int mirror_num, u8 *csum)
2597 struct scrub_ctx *sctx = sparity->sctx;
2598 struct scrub_block *sblock;
2601 sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
2603 spin_lock(&sctx->stat_lock);
2604 sctx->stat.malloc_errors++;
2605 spin_unlock(&sctx->stat_lock);
2609 /* one ref inside this function, plus one for each page added to
2611 atomic_set(&sblock->refs, 1);
2612 sblock->sctx = sctx;
2613 sblock->no_io_error_seen = 1;
2614 sblock->sparity = sparity;
2615 scrub_parity_get(sparity);
2617 for (index = 0; len > 0; index++) {
2618 struct scrub_page *spage;
2619 u64 l = min_t(u64, len, PAGE_SIZE);
2621 spage = kzalloc(sizeof(*spage), GFP_NOFS);
2624 spin_lock(&sctx->stat_lock);
2625 sctx->stat.malloc_errors++;
2626 spin_unlock(&sctx->stat_lock);
2627 scrub_block_put(sblock);
2630 BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
2631 /* For scrub block */
2632 scrub_page_get(spage);
2633 sblock->pagev[index] = spage;
2634 /* For scrub parity */
2635 scrub_page_get(spage);
2636 list_add_tail(&spage->list, &sparity->spages);
2637 spage->sblock = sblock;
2639 spage->flags = flags;
2640 spage->generation = gen;
2641 spage->logical = logical;
2642 spage->physical = physical;
2643 spage->mirror_num = mirror_num;
2645 spage->have_csum = 1;
2646 memcpy(spage->csum, csum, sctx->csum_size);
2648 spage->have_csum = 0;
2650 sblock->page_count++;
2651 spage->page = alloc_page(GFP_NOFS);
2659 WARN_ON(sblock->page_count == 0);
2660 for (index = 0; index < sblock->page_count; index++) {
2661 struct scrub_page *spage = sblock->pagev[index];
2664 ret = scrub_add_page_to_rd_bio(sctx, spage);
2666 scrub_block_put(sblock);
2671 /* last one frees, either here or in bio completion for last page */
2672 scrub_block_put(sblock);
2676 static int scrub_extent_for_parity(struct scrub_parity *sparity,
2677 u64 logical, u64 len,
2678 u64 physical, struct btrfs_device *dev,
2679 u64 flags, u64 gen, int mirror_num)
2681 struct scrub_ctx *sctx = sparity->sctx;
2683 u8 csum[BTRFS_CSUM_SIZE];
2687 scrub_parity_mark_sectors_error(sparity, logical, len);
2691 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2692 blocksize = sctx->sectorsize;
2693 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2694 blocksize = sctx->nodesize;
2696 blocksize = sctx->sectorsize;
2701 u64 l = min_t(u64, len, blocksize);
2704 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2705 /* push csums to sbio */
2706 have_csum = scrub_find_csum(sctx, logical, l, csum);
2710 ret = scrub_pages_for_parity(sparity, logical, l, physical, dev,
2711 flags, gen, mirror_num,
2712 have_csum ? csum : NULL);
2724 * Given a physical address, this will calculate it's
2725 * logical offset. if this is a parity stripe, it will return
2726 * the most left data stripe's logical offset.
2728 * return 0 if it is a data stripe, 1 means parity stripe.
2730 static int get_raid56_logic_offset(u64 physical, int num,
2731 struct map_lookup *map, u64 *offset,
2741 last_offset = (physical - map->stripes[num].physical) *
2742 nr_data_stripes(map);
2744 *stripe_start = last_offset;
2746 *offset = last_offset;
2747 for (i = 0; i < nr_data_stripes(map); i++) {
2748 *offset = last_offset + i * map->stripe_len;
2750 stripe_nr = div_u64(*offset, map->stripe_len);
2751 stripe_nr = div_u64(stripe_nr, nr_data_stripes(map));
2753 /* Work out the disk rotation on this stripe-set */
2754 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, &rot);
2755 /* calculate which stripe this data locates */
2757 stripe_index = rot % map->num_stripes;
2758 if (stripe_index == num)
2760 if (stripe_index < num)
2763 *offset = last_offset + j * map->stripe_len;
2767 static void scrub_free_parity(struct scrub_parity *sparity)
2769 struct scrub_ctx *sctx = sparity->sctx;
2770 struct scrub_page *curr, *next;
2773 nbits = bitmap_weight(sparity->ebitmap, sparity->nsectors);
2775 spin_lock(&sctx->stat_lock);
2776 sctx->stat.read_errors += nbits;
2777 sctx->stat.uncorrectable_errors += nbits;
2778 spin_unlock(&sctx->stat_lock);
2781 list_for_each_entry_safe(curr, next, &sparity->spages, list) {
2782 list_del_init(&curr->list);
2783 scrub_page_put(curr);
2789 static void scrub_parity_bio_endio_worker(struct btrfs_work *work)
2791 struct scrub_parity *sparity = container_of(work, struct scrub_parity,
2793 struct scrub_ctx *sctx = sparity->sctx;
2795 scrub_free_parity(sparity);
2796 scrub_pending_bio_dec(sctx);
2799 static void scrub_parity_bio_endio(struct bio *bio)
2801 struct scrub_parity *sparity = (struct scrub_parity *)bio->bi_private;
2804 bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap,
2809 btrfs_init_work(&sparity->work, btrfs_scrubparity_helper,
2810 scrub_parity_bio_endio_worker, NULL, NULL);
2811 btrfs_queue_work(sparity->sctx->dev_root->fs_info->scrub_parity_workers,
2815 static void scrub_parity_check_and_repair(struct scrub_parity *sparity)
2817 struct scrub_ctx *sctx = sparity->sctx;
2819 struct btrfs_raid_bio *rbio;
2820 struct scrub_page *spage;
2821 struct btrfs_bio *bbio = NULL;
2825 if (!bitmap_andnot(sparity->dbitmap, sparity->dbitmap, sparity->ebitmap,
2829 length = sparity->logic_end - sparity->logic_start;
2830 ret = btrfs_map_sblock(sctx->dev_root->fs_info, WRITE,
2831 sparity->logic_start,
2832 &length, &bbio, 0, 1);
2833 if (ret || !bbio || !bbio->raid_map)
2836 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
2840 bio->bi_iter.bi_sector = sparity->logic_start >> 9;
2841 bio->bi_private = sparity;
2842 bio->bi_end_io = scrub_parity_bio_endio;
2844 rbio = raid56_parity_alloc_scrub_rbio(sctx->dev_root, bio, bbio,
2845 length, sparity->scrub_dev,
2851 list_for_each_entry(spage, &sparity->spages, list)
2852 raid56_add_scrub_pages(rbio, spage->page, spage->logical);
2854 scrub_pending_bio_inc(sctx);
2855 raid56_parity_submit_scrub_rbio(rbio);
2861 btrfs_put_bbio(bbio);
2862 bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap,
2864 spin_lock(&sctx->stat_lock);
2865 sctx->stat.malloc_errors++;
2866 spin_unlock(&sctx->stat_lock);
2868 scrub_free_parity(sparity);
2871 static inline int scrub_calc_parity_bitmap_len(int nsectors)
2873 return DIV_ROUND_UP(nsectors, BITS_PER_LONG) * (BITS_PER_LONG / 8);
2876 static void scrub_parity_get(struct scrub_parity *sparity)
2878 atomic_inc(&sparity->refs);
2881 static void scrub_parity_put(struct scrub_parity *sparity)
2883 if (!atomic_dec_and_test(&sparity->refs))
2886 scrub_parity_check_and_repair(sparity);
2889 static noinline_for_stack int scrub_raid56_parity(struct scrub_ctx *sctx,
2890 struct map_lookup *map,
2891 struct btrfs_device *sdev,
2892 struct btrfs_path *path,
2896 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
2897 struct btrfs_root *root = fs_info->extent_root;
2898 struct btrfs_root *csum_root = fs_info->csum_root;
2899 struct btrfs_extent_item *extent;
2900 struct btrfs_bio *bbio = NULL;
2904 struct extent_buffer *l;
2905 struct btrfs_key key;
2908 u64 extent_physical;
2911 struct btrfs_device *extent_dev;
2912 struct scrub_parity *sparity;
2915 int extent_mirror_num;
2918 nsectors = map->stripe_len / root->sectorsize;
2919 bitmap_len = scrub_calc_parity_bitmap_len(nsectors);
2920 sparity = kzalloc(sizeof(struct scrub_parity) + 2 * bitmap_len,
2923 spin_lock(&sctx->stat_lock);
2924 sctx->stat.malloc_errors++;
2925 spin_unlock(&sctx->stat_lock);
2929 sparity->stripe_len = map->stripe_len;
2930 sparity->nsectors = nsectors;
2931 sparity->sctx = sctx;
2932 sparity->scrub_dev = sdev;
2933 sparity->logic_start = logic_start;
2934 sparity->logic_end = logic_end;
2935 atomic_set(&sparity->refs, 1);
2936 INIT_LIST_HEAD(&sparity->spages);
2937 sparity->dbitmap = sparity->bitmap;
2938 sparity->ebitmap = (void *)sparity->bitmap + bitmap_len;
2941 while (logic_start < logic_end) {
2942 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2943 key.type = BTRFS_METADATA_ITEM_KEY;
2945 key.type = BTRFS_EXTENT_ITEM_KEY;
2946 key.objectid = logic_start;
2947 key.offset = (u64)-1;
2949 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2954 ret = btrfs_previous_extent_item(root, path, 0);
2958 btrfs_release_path(path);
2959 ret = btrfs_search_slot(NULL, root, &key,
2971 slot = path->slots[0];
2972 if (slot >= btrfs_header_nritems(l)) {
2973 ret = btrfs_next_leaf(root, path);
2982 btrfs_item_key_to_cpu(l, &key, slot);
2984 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
2985 key.type != BTRFS_METADATA_ITEM_KEY)
2988 if (key.type == BTRFS_METADATA_ITEM_KEY)
2989 bytes = root->nodesize;
2993 if (key.objectid + bytes <= logic_start)
2996 if (key.objectid >= logic_end) {
3001 while (key.objectid >= logic_start + map->stripe_len)
3002 logic_start += map->stripe_len;
3004 extent = btrfs_item_ptr(l, slot,
3005 struct btrfs_extent_item);
3006 flags = btrfs_extent_flags(l, extent);
3007 generation = btrfs_extent_generation(l, extent);
3009 if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) &&
3010 (key.objectid < logic_start ||
3011 key.objectid + bytes >
3012 logic_start + map->stripe_len)) {
3013 btrfs_err(fs_info, "scrub: tree block %llu spanning stripes, ignored. logical=%llu",
3014 key.objectid, logic_start);
3018 extent_logical = key.objectid;
3021 if (extent_logical < logic_start) {
3022 extent_len -= logic_start - extent_logical;
3023 extent_logical = logic_start;
3026 if (extent_logical + extent_len >
3027 logic_start + map->stripe_len)
3028 extent_len = logic_start + map->stripe_len -
3031 scrub_parity_mark_sectors_data(sparity, extent_logical,
3034 mapped_length = extent_len;
3035 ret = btrfs_map_block(fs_info, READ, extent_logical,
3036 &mapped_length, &bbio, 0);
3038 if (!bbio || mapped_length < extent_len)
3042 btrfs_put_bbio(bbio);
3045 extent_physical = bbio->stripes[0].physical;
3046 extent_mirror_num = bbio->mirror_num;
3047 extent_dev = bbio->stripes[0].dev;
3048 btrfs_put_bbio(bbio);
3050 ret = btrfs_lookup_csums_range(csum_root,
3052 extent_logical + extent_len - 1,
3053 &sctx->csum_list, 1);
3057 ret = scrub_extent_for_parity(sparity, extent_logical,
3064 scrub_free_csums(sctx);
3069 if (extent_logical + extent_len <
3070 key.objectid + bytes) {
3071 logic_start += map->stripe_len;
3073 if (logic_start >= logic_end) {
3078 if (logic_start < key.objectid + bytes) {
3087 btrfs_release_path(path);
3092 logic_start += map->stripe_len;
3096 scrub_parity_mark_sectors_error(sparity, logic_start,
3097 logic_end - logic_start);
3098 scrub_parity_put(sparity);
3100 mutex_lock(&sctx->wr_ctx.wr_lock);
3101 scrub_wr_submit(sctx);
3102 mutex_unlock(&sctx->wr_ctx.wr_lock);
3104 btrfs_release_path(path);
3105 return ret < 0 ? ret : 0;
3108 static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
3109 struct map_lookup *map,
3110 struct btrfs_device *scrub_dev,
3111 int num, u64 base, u64 length,
3114 struct btrfs_path *path, *ppath;
3115 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
3116 struct btrfs_root *root = fs_info->extent_root;
3117 struct btrfs_root *csum_root = fs_info->csum_root;
3118 struct btrfs_extent_item *extent;
3119 struct blk_plug plug;
3124 struct extent_buffer *l;
3125 struct btrfs_key key;
3132 struct reada_control *reada1;
3133 struct reada_control *reada2;
3134 struct btrfs_key key_start;
3135 struct btrfs_key key_end;
3136 u64 increment = map->stripe_len;
3139 u64 extent_physical;
3143 struct btrfs_device *extent_dev;
3144 int extent_mirror_num;
3147 physical = map->stripes[num].physical;
3149 nstripes = div_u64(length, map->stripe_len);
3150 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3151 offset = map->stripe_len * num;
3152 increment = map->stripe_len * map->num_stripes;
3154 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3155 int factor = map->num_stripes / map->sub_stripes;
3156 offset = map->stripe_len * (num / map->sub_stripes);
3157 increment = map->stripe_len * factor;
3158 mirror_num = num % map->sub_stripes + 1;
3159 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3160 increment = map->stripe_len;
3161 mirror_num = num % map->num_stripes + 1;
3162 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3163 increment = map->stripe_len;
3164 mirror_num = num % map->num_stripes + 1;
3165 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3166 get_raid56_logic_offset(physical, num, map, &offset, NULL);
3167 increment = map->stripe_len * nr_data_stripes(map);
3170 increment = map->stripe_len;
3174 path = btrfs_alloc_path();
3178 ppath = btrfs_alloc_path();
3180 btrfs_free_path(path);
3185 * work on commit root. The related disk blocks are static as
3186 * long as COW is applied. This means, it is save to rewrite
3187 * them to repair disk errors without any race conditions
3189 path->search_commit_root = 1;
3190 path->skip_locking = 1;
3192 ppath->search_commit_root = 1;
3193 ppath->skip_locking = 1;
3195 * trigger the readahead for extent tree csum tree and wait for
3196 * completion. During readahead, the scrub is officially paused
3197 * to not hold off transaction commits
3199 logical = base + offset;
3200 physical_end = physical + nstripes * map->stripe_len;
3201 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3202 get_raid56_logic_offset(physical_end, num,
3203 map, &logic_end, NULL);
3206 logic_end = logical + increment * nstripes;
3208 wait_event(sctx->list_wait,
3209 atomic_read(&sctx->bios_in_flight) == 0);
3210 scrub_blocked_if_needed(fs_info);
3212 /* FIXME it might be better to start readahead at commit root */
3213 key_start.objectid = logical;
3214 key_start.type = BTRFS_EXTENT_ITEM_KEY;
3215 key_start.offset = (u64)0;
3216 key_end.objectid = logic_end;
3217 key_end.type = BTRFS_METADATA_ITEM_KEY;
3218 key_end.offset = (u64)-1;
3219 reada1 = btrfs_reada_add(root, &key_start, &key_end);
3221 key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
3222 key_start.type = BTRFS_EXTENT_CSUM_KEY;
3223 key_start.offset = logical;
3224 key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
3225 key_end.type = BTRFS_EXTENT_CSUM_KEY;
3226 key_end.offset = logic_end;
3227 reada2 = btrfs_reada_add(csum_root, &key_start, &key_end);
3229 if (!IS_ERR(reada1))
3230 btrfs_reada_wait(reada1);
3231 if (!IS_ERR(reada2))
3232 btrfs_reada_wait(reada2);
3236 * collect all data csums for the stripe to avoid seeking during
3237 * the scrub. This might currently (crc32) end up to be about 1MB
3239 blk_start_plug(&plug);
3242 * now find all extents for each stripe and scrub them
3245 while (physical < physical_end) {
3249 if (atomic_read(&fs_info->scrub_cancel_req) ||
3250 atomic_read(&sctx->cancel_req)) {
3255 * check to see if we have to pause
3257 if (atomic_read(&fs_info->scrub_pause_req)) {
3258 /* push queued extents */
3259 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
3261 mutex_lock(&sctx->wr_ctx.wr_lock);
3262 scrub_wr_submit(sctx);
3263 mutex_unlock(&sctx->wr_ctx.wr_lock);
3264 wait_event(sctx->list_wait,
3265 atomic_read(&sctx->bios_in_flight) == 0);
3266 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
3267 scrub_blocked_if_needed(fs_info);
3270 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3271 ret = get_raid56_logic_offset(physical, num, map,
3276 /* it is parity strip */
3277 stripe_logical += base;
3278 stripe_end = stripe_logical + increment;
3279 ret = scrub_raid56_parity(sctx, map, scrub_dev,
3280 ppath, stripe_logical,
3288 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
3289 key.type = BTRFS_METADATA_ITEM_KEY;
3291 key.type = BTRFS_EXTENT_ITEM_KEY;
3292 key.objectid = logical;
3293 key.offset = (u64)-1;
3295 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3300 ret = btrfs_previous_extent_item(root, path, 0);
3304 /* there's no smaller item, so stick with the
3306 btrfs_release_path(path);
3307 ret = btrfs_search_slot(NULL, root, &key,
3319 slot = path->slots[0];
3320 if (slot >= btrfs_header_nritems(l)) {
3321 ret = btrfs_next_leaf(root, path);
3330 btrfs_item_key_to_cpu(l, &key, slot);
3332 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3333 key.type != BTRFS_METADATA_ITEM_KEY)
3336 if (key.type == BTRFS_METADATA_ITEM_KEY)
3337 bytes = root->nodesize;
3341 if (key.objectid + bytes <= logical)
3344 if (key.objectid >= logical + map->stripe_len) {
3345 /* out of this device extent */
3346 if (key.objectid >= logic_end)
3351 extent = btrfs_item_ptr(l, slot,
3352 struct btrfs_extent_item);
3353 flags = btrfs_extent_flags(l, extent);
3354 generation = btrfs_extent_generation(l, extent);
3356 if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) &&
3357 (key.objectid < logical ||
3358 key.objectid + bytes >
3359 logical + map->stripe_len)) {
3361 "scrub: tree block %llu spanning "
3362 "stripes, ignored. logical=%llu",
3363 key.objectid, logical);
3368 extent_logical = key.objectid;
3372 * trim extent to this stripe
3374 if (extent_logical < logical) {
3375 extent_len -= logical - extent_logical;
3376 extent_logical = logical;
3378 if (extent_logical + extent_len >
3379 logical + map->stripe_len) {
3380 extent_len = logical + map->stripe_len -
3384 extent_physical = extent_logical - logical + physical;
3385 extent_dev = scrub_dev;
3386 extent_mirror_num = mirror_num;
3388 scrub_remap_extent(fs_info, extent_logical,
3389 extent_len, &extent_physical,
3391 &extent_mirror_num);
3393 ret = btrfs_lookup_csums_range(csum_root,
3397 &sctx->csum_list, 1);
3401 ret = scrub_extent(sctx, extent_logical, extent_len,
3402 extent_physical, extent_dev, flags,
3403 generation, extent_mirror_num,
3404 extent_logical - logical + physical);
3406 scrub_free_csums(sctx);
3411 if (extent_logical + extent_len <
3412 key.objectid + bytes) {
3413 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3415 * loop until we find next data stripe
3416 * or we have finished all stripes.
3419 physical += map->stripe_len;
3420 ret = get_raid56_logic_offset(physical,
3425 if (ret && physical < physical_end) {
3426 stripe_logical += base;
3427 stripe_end = stripe_logical +
3429 ret = scrub_raid56_parity(sctx,
3430 map, scrub_dev, ppath,
3438 physical += map->stripe_len;
3439 logical += increment;
3441 if (logical < key.objectid + bytes) {
3446 if (physical >= physical_end) {
3454 btrfs_release_path(path);
3456 logical += increment;
3457 physical += map->stripe_len;
3458 spin_lock(&sctx->stat_lock);
3460 sctx->stat.last_physical = map->stripes[num].physical +
3463 sctx->stat.last_physical = physical;
3464 spin_unlock(&sctx->stat_lock);
3469 /* push queued extents */
3471 mutex_lock(&sctx->wr_ctx.wr_lock);
3472 scrub_wr_submit(sctx);
3473 mutex_unlock(&sctx->wr_ctx.wr_lock);
3475 blk_finish_plug(&plug);
3476 btrfs_free_path(path);
3477 btrfs_free_path(ppath);
3478 return ret < 0 ? ret : 0;
3481 static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx,
3482 struct btrfs_device *scrub_dev,
3483 u64 chunk_offset, u64 length,
3484 u64 dev_offset, int is_dev_replace)
3486 struct btrfs_mapping_tree *map_tree =
3487 &sctx->dev_root->fs_info->mapping_tree;
3488 struct map_lookup *map;
3489 struct extent_map *em;
3493 read_lock(&map_tree->map_tree.lock);
3494 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3495 read_unlock(&map_tree->map_tree.lock);
3500 map = (struct map_lookup *)em->bdev;
3501 if (em->start != chunk_offset)
3504 if (em->len < length)
3507 for (i = 0; i < map->num_stripes; ++i) {
3508 if (map->stripes[i].dev->bdev == scrub_dev->bdev &&
3509 map->stripes[i].physical == dev_offset) {
3510 ret = scrub_stripe(sctx, map, scrub_dev, i,
3511 chunk_offset, length,
3518 free_extent_map(em);
3523 static noinline_for_stack
3524 int scrub_enumerate_chunks(struct scrub_ctx *sctx,
3525 struct btrfs_device *scrub_dev, u64 start, u64 end,
3528 struct btrfs_dev_extent *dev_extent = NULL;
3529 struct btrfs_path *path;
3530 struct btrfs_root *root = sctx->dev_root;
3531 struct btrfs_fs_info *fs_info = root->fs_info;
3536 struct extent_buffer *l;
3537 struct btrfs_key key;
3538 struct btrfs_key found_key;
3539 struct btrfs_block_group_cache *cache;
3540 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
3542 path = btrfs_alloc_path();
3547 path->search_commit_root = 1;
3548 path->skip_locking = 1;
3550 key.objectid = scrub_dev->devid;
3552 key.type = BTRFS_DEV_EXTENT_KEY;
3555 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3559 if (path->slots[0] >=
3560 btrfs_header_nritems(path->nodes[0])) {
3561 ret = btrfs_next_leaf(root, path);
3574 slot = path->slots[0];
3576 btrfs_item_key_to_cpu(l, &found_key, slot);
3578 if (found_key.objectid != scrub_dev->devid)
3581 if (found_key.type != BTRFS_DEV_EXTENT_KEY)
3584 if (found_key.offset >= end)
3587 if (found_key.offset < key.offset)
3590 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3591 length = btrfs_dev_extent_length(l, dev_extent);
3593 if (found_key.offset + length <= start)
3596 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3599 * get a reference on the corresponding block group to prevent
3600 * the chunk from going away while we scrub it
3602 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3604 /* some chunks are removed but not committed to disk yet,
3605 * continue scrubbing */
3610 * we need call btrfs_inc_block_group_ro() with scrubs_paused,
3611 * to avoid deadlock caused by:
3612 * btrfs_inc_block_group_ro()
3613 * -> btrfs_wait_for_commit()
3614 * -> btrfs_commit_transaction()
3615 * -> btrfs_scrub_pause()
3617 scrub_pause_on(fs_info);
3618 ret = btrfs_inc_block_group_ro(root, cache);
3619 scrub_pause_off(fs_info);
3621 btrfs_put_block_group(cache);
3625 dev_replace->cursor_right = found_key.offset + length;
3626 dev_replace->cursor_left = found_key.offset;
3627 dev_replace->item_needs_writeback = 1;
3628 ret = scrub_chunk(sctx, scrub_dev, chunk_offset, length,
3629 found_key.offset, is_dev_replace);
3632 * flush, submit all pending read and write bios, afterwards
3634 * Note that in the dev replace case, a read request causes
3635 * write requests that are submitted in the read completion
3636 * worker. Therefore in the current situation, it is required
3637 * that all write requests are flushed, so that all read and
3638 * write requests are really completed when bios_in_flight
3641 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
3643 mutex_lock(&sctx->wr_ctx.wr_lock);
3644 scrub_wr_submit(sctx);
3645 mutex_unlock(&sctx->wr_ctx.wr_lock);
3647 wait_event(sctx->list_wait,
3648 atomic_read(&sctx->bios_in_flight) == 0);
3650 scrub_pause_on(fs_info);
3653 * must be called before we decrease @scrub_paused.
3654 * make sure we don't block transaction commit while
3655 * we are waiting pending workers finished.
3657 wait_event(sctx->list_wait,
3658 atomic_read(&sctx->workers_pending) == 0);
3659 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
3661 scrub_pause_off(fs_info);
3663 btrfs_dec_block_group_ro(root, cache);
3665 btrfs_put_block_group(cache);
3668 if (is_dev_replace &&
3669 atomic64_read(&dev_replace->num_write_errors) > 0) {
3673 if (sctx->stat.malloc_errors > 0) {
3678 dev_replace->cursor_left = dev_replace->cursor_right;
3679 dev_replace->item_needs_writeback = 1;
3681 key.offset = found_key.offset + length;
3682 btrfs_release_path(path);
3685 btrfs_free_path(path);
3690 static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx,
3691 struct btrfs_device *scrub_dev)
3697 struct btrfs_root *root = sctx->dev_root;
3699 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
3702 /* Seed devices of a new filesystem has their own generation. */
3703 if (scrub_dev->fs_devices != root->fs_info->fs_devices)
3704 gen = scrub_dev->generation;
3706 gen = root->fs_info->last_trans_committed;
3708 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
3709 bytenr = btrfs_sb_offset(i);
3710 if (bytenr + BTRFS_SUPER_INFO_SIZE >
3711 scrub_dev->commit_total_bytes)
3714 ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr,
3715 scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i,
3720 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
3726 * get a reference count on fs_info->scrub_workers. start worker if necessary
3728 static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info,
3731 unsigned int flags = WQ_FREEZABLE | WQ_UNBOUND;
3732 int max_active = fs_info->thread_pool_size;
3734 if (fs_info->scrub_workers_refcnt == 0) {
3736 fs_info->scrub_workers =
3737 btrfs_alloc_workqueue("btrfs-scrub", flags,
3740 fs_info->scrub_workers =
3741 btrfs_alloc_workqueue("btrfs-scrub", flags,
3743 if (!fs_info->scrub_workers)
3744 goto fail_scrub_workers;
3746 fs_info->scrub_wr_completion_workers =
3747 btrfs_alloc_workqueue("btrfs-scrubwrc", flags,
3749 if (!fs_info->scrub_wr_completion_workers)
3750 goto fail_scrub_wr_completion_workers;
3752 fs_info->scrub_nocow_workers =
3753 btrfs_alloc_workqueue("btrfs-scrubnc", flags, 1, 0);
3754 if (!fs_info->scrub_nocow_workers)
3755 goto fail_scrub_nocow_workers;
3756 fs_info->scrub_parity_workers =
3757 btrfs_alloc_workqueue("btrfs-scrubparity", flags,
3759 if (!fs_info->scrub_parity_workers)
3760 goto fail_scrub_parity_workers;
3762 ++fs_info->scrub_workers_refcnt;
3765 fail_scrub_parity_workers:
3766 btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
3767 fail_scrub_nocow_workers:
3768 btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
3769 fail_scrub_wr_completion_workers:
3770 btrfs_destroy_workqueue(fs_info->scrub_workers);
3775 static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info)
3777 if (--fs_info->scrub_workers_refcnt == 0) {
3778 btrfs_destroy_workqueue(fs_info->scrub_workers);
3779 btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
3780 btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
3781 btrfs_destroy_workqueue(fs_info->scrub_parity_workers);
3783 WARN_ON(fs_info->scrub_workers_refcnt < 0);
3786 int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start,
3787 u64 end, struct btrfs_scrub_progress *progress,
3788 int readonly, int is_dev_replace)
3790 struct scrub_ctx *sctx;
3792 struct btrfs_device *dev;
3793 struct rcu_string *name;
3795 if (btrfs_fs_closing(fs_info))
3798 if (fs_info->chunk_root->nodesize > BTRFS_STRIPE_LEN) {
3800 * in this case scrub is unable to calculate the checksum
3801 * the way scrub is implemented. Do not handle this
3802 * situation at all because it won't ever happen.
3805 "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails",
3806 fs_info->chunk_root->nodesize, BTRFS_STRIPE_LEN);
3810 if (fs_info->chunk_root->sectorsize != PAGE_SIZE) {
3811 /* not supported for data w/o checksums */
3813 "scrub: size assumption sectorsize != PAGE_SIZE "
3814 "(%d != %lu) fails",
3815 fs_info->chunk_root->sectorsize, PAGE_SIZE);
3819 if (fs_info->chunk_root->nodesize >
3820 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK ||
3821 fs_info->chunk_root->sectorsize >
3822 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) {
3824 * would exhaust the array bounds of pagev member in
3825 * struct scrub_block
3827 btrfs_err(fs_info, "scrub: size assumption nodesize and sectorsize "
3828 "<= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails",
3829 fs_info->chunk_root->nodesize,
3830 SCRUB_MAX_PAGES_PER_BLOCK,
3831 fs_info->chunk_root->sectorsize,
3832 SCRUB_MAX_PAGES_PER_BLOCK);
3837 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3838 dev = btrfs_find_device(fs_info, devid, NULL, NULL);
3839 if (!dev || (dev->missing && !is_dev_replace)) {
3840 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3844 if (!is_dev_replace && !readonly && !dev->writeable) {
3845 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3847 name = rcu_dereference(dev->name);
3848 btrfs_err(fs_info, "scrub: device %s is not writable",
3854 mutex_lock(&fs_info->scrub_lock);
3855 if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) {
3856 mutex_unlock(&fs_info->scrub_lock);
3857 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3861 btrfs_dev_replace_lock(&fs_info->dev_replace);
3862 if (dev->scrub_device ||
3864 btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) {
3865 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3866 mutex_unlock(&fs_info->scrub_lock);
3867 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3868 return -EINPROGRESS;
3870 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3872 ret = scrub_workers_get(fs_info, is_dev_replace);
3874 mutex_unlock(&fs_info->scrub_lock);
3875 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3879 sctx = scrub_setup_ctx(dev, is_dev_replace);
3881 mutex_unlock(&fs_info->scrub_lock);
3882 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3883 scrub_workers_put(fs_info);
3884 return PTR_ERR(sctx);
3886 sctx->readonly = readonly;
3887 dev->scrub_device = sctx;
3888 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3891 * checking @scrub_pause_req here, we can avoid
3892 * race between committing transaction and scrubbing.
3894 __scrub_blocked_if_needed(fs_info);
3895 atomic_inc(&fs_info->scrubs_running);
3896 mutex_unlock(&fs_info->scrub_lock);
3898 if (!is_dev_replace) {
3900 * by holding device list mutex, we can
3901 * kick off writing super in log tree sync.
3903 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3904 ret = scrub_supers(sctx, dev);
3905 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3909 ret = scrub_enumerate_chunks(sctx, dev, start, end,
3912 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
3913 atomic_dec(&fs_info->scrubs_running);
3914 wake_up(&fs_info->scrub_pause_wait);
3916 wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0);
3919 memcpy(progress, &sctx->stat, sizeof(*progress));
3921 mutex_lock(&fs_info->scrub_lock);
3922 dev->scrub_device = NULL;
3923 scrub_workers_put(fs_info);
3924 mutex_unlock(&fs_info->scrub_lock);
3926 scrub_put_ctx(sctx);
3931 void btrfs_scrub_pause(struct btrfs_root *root)
3933 struct btrfs_fs_info *fs_info = root->fs_info;
3935 mutex_lock(&fs_info->scrub_lock);
3936 atomic_inc(&fs_info->scrub_pause_req);
3937 while (atomic_read(&fs_info->scrubs_paused) !=
3938 atomic_read(&fs_info->scrubs_running)) {
3939 mutex_unlock(&fs_info->scrub_lock);
3940 wait_event(fs_info->scrub_pause_wait,
3941 atomic_read(&fs_info->scrubs_paused) ==
3942 atomic_read(&fs_info->scrubs_running));
3943 mutex_lock(&fs_info->scrub_lock);
3945 mutex_unlock(&fs_info->scrub_lock);
3948 void btrfs_scrub_continue(struct btrfs_root *root)
3950 struct btrfs_fs_info *fs_info = root->fs_info;
3952 atomic_dec(&fs_info->scrub_pause_req);
3953 wake_up(&fs_info->scrub_pause_wait);
3956 int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info)
3958 mutex_lock(&fs_info->scrub_lock);
3959 if (!atomic_read(&fs_info->scrubs_running)) {
3960 mutex_unlock(&fs_info->scrub_lock);
3964 atomic_inc(&fs_info->scrub_cancel_req);
3965 while (atomic_read(&fs_info->scrubs_running)) {
3966 mutex_unlock(&fs_info->scrub_lock);
3967 wait_event(fs_info->scrub_pause_wait,
3968 atomic_read(&fs_info->scrubs_running) == 0);
3969 mutex_lock(&fs_info->scrub_lock);
3971 atomic_dec(&fs_info->scrub_cancel_req);
3972 mutex_unlock(&fs_info->scrub_lock);
3977 int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info,
3978 struct btrfs_device *dev)
3980 struct scrub_ctx *sctx;
3982 mutex_lock(&fs_info->scrub_lock);
3983 sctx = dev->scrub_device;
3985 mutex_unlock(&fs_info->scrub_lock);
3988 atomic_inc(&sctx->cancel_req);
3989 while (dev->scrub_device) {
3990 mutex_unlock(&fs_info->scrub_lock);
3991 wait_event(fs_info->scrub_pause_wait,
3992 dev->scrub_device == NULL);
3993 mutex_lock(&fs_info->scrub_lock);
3995 mutex_unlock(&fs_info->scrub_lock);
4000 int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
4001 struct btrfs_scrub_progress *progress)
4003 struct btrfs_device *dev;
4004 struct scrub_ctx *sctx = NULL;
4006 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
4007 dev = btrfs_find_device(root->fs_info, devid, NULL, NULL);
4009 sctx = dev->scrub_device;
4011 memcpy(progress, &sctx->stat, sizeof(*progress));
4012 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
4014 return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV;
4017 static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
4018 u64 extent_logical, u64 extent_len,
4019 u64 *extent_physical,
4020 struct btrfs_device **extent_dev,
4021 int *extent_mirror_num)
4024 struct btrfs_bio *bbio = NULL;
4027 mapped_length = extent_len;
4028 ret = btrfs_map_block(fs_info, READ, extent_logical,
4029 &mapped_length, &bbio, 0);
4030 if (ret || !bbio || mapped_length < extent_len ||
4031 !bbio->stripes[0].dev->bdev) {
4032 btrfs_put_bbio(bbio);
4036 *extent_physical = bbio->stripes[0].physical;
4037 *extent_mirror_num = bbio->mirror_num;
4038 *extent_dev = bbio->stripes[0].dev;
4039 btrfs_put_bbio(bbio);
4042 static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
4043 struct scrub_wr_ctx *wr_ctx,
4044 struct btrfs_fs_info *fs_info,
4045 struct btrfs_device *dev,
4048 WARN_ON(wr_ctx->wr_curr_bio != NULL);
4050 mutex_init(&wr_ctx->wr_lock);
4051 wr_ctx->wr_curr_bio = NULL;
4052 if (!is_dev_replace)
4055 WARN_ON(!dev->bdev);
4056 wr_ctx->pages_per_wr_bio = SCRUB_PAGES_PER_WR_BIO;
4057 wr_ctx->tgtdev = dev;
4058 atomic_set(&wr_ctx->flush_all_writes, 0);
4062 static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx)
4064 mutex_lock(&wr_ctx->wr_lock);
4065 kfree(wr_ctx->wr_curr_bio);
4066 wr_ctx->wr_curr_bio = NULL;
4067 mutex_unlock(&wr_ctx->wr_lock);
4070 static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
4071 int mirror_num, u64 physical_for_dev_replace)
4073 struct scrub_copy_nocow_ctx *nocow_ctx;
4074 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
4076 nocow_ctx = kzalloc(sizeof(*nocow_ctx), GFP_NOFS);
4078 spin_lock(&sctx->stat_lock);
4079 sctx->stat.malloc_errors++;
4080 spin_unlock(&sctx->stat_lock);
4084 scrub_pending_trans_workers_inc(sctx);
4086 nocow_ctx->sctx = sctx;
4087 nocow_ctx->logical = logical;
4088 nocow_ctx->len = len;
4089 nocow_ctx->mirror_num = mirror_num;
4090 nocow_ctx->physical_for_dev_replace = physical_for_dev_replace;
4091 btrfs_init_work(&nocow_ctx->work, btrfs_scrubnc_helper,
4092 copy_nocow_pages_worker, NULL, NULL);
4093 INIT_LIST_HEAD(&nocow_ctx->inodes);
4094 btrfs_queue_work(fs_info->scrub_nocow_workers,
4100 static int record_inode_for_nocow(u64 inum, u64 offset, u64 root, void *ctx)
4102 struct scrub_copy_nocow_ctx *nocow_ctx = ctx;
4103 struct scrub_nocow_inode *nocow_inode;
4105 nocow_inode = kzalloc(sizeof(*nocow_inode), GFP_NOFS);
4108 nocow_inode->inum = inum;
4109 nocow_inode->offset = offset;
4110 nocow_inode->root = root;
4111 list_add_tail(&nocow_inode->list, &nocow_ctx->inodes);
4115 #define COPY_COMPLETE 1
4117 static void copy_nocow_pages_worker(struct btrfs_work *work)
4119 struct scrub_copy_nocow_ctx *nocow_ctx =
4120 container_of(work, struct scrub_copy_nocow_ctx, work);
4121 struct scrub_ctx *sctx = nocow_ctx->sctx;
4122 u64 logical = nocow_ctx->logical;
4123 u64 len = nocow_ctx->len;
4124 int mirror_num = nocow_ctx->mirror_num;
4125 u64 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
4127 struct btrfs_trans_handle *trans = NULL;
4128 struct btrfs_fs_info *fs_info;
4129 struct btrfs_path *path;
4130 struct btrfs_root *root;
4131 int not_written = 0;
4133 fs_info = sctx->dev_root->fs_info;
4134 root = fs_info->extent_root;
4136 path = btrfs_alloc_path();
4138 spin_lock(&sctx->stat_lock);
4139 sctx->stat.malloc_errors++;
4140 spin_unlock(&sctx->stat_lock);
4145 trans = btrfs_join_transaction(root);
4146 if (IS_ERR(trans)) {
4151 ret = iterate_inodes_from_logical(logical, fs_info, path,
4152 record_inode_for_nocow, nocow_ctx);
4153 if (ret != 0 && ret != -ENOENT) {
4154 btrfs_warn(fs_info, "iterate_inodes_from_logical() failed: log %llu, "
4155 "phys %llu, len %llu, mir %u, ret %d",
4156 logical, physical_for_dev_replace, len, mirror_num,
4162 btrfs_end_transaction(trans, root);
4164 while (!list_empty(&nocow_ctx->inodes)) {
4165 struct scrub_nocow_inode *entry;
4166 entry = list_first_entry(&nocow_ctx->inodes,
4167 struct scrub_nocow_inode,
4169 list_del_init(&entry->list);
4170 ret = copy_nocow_pages_for_inode(entry->inum, entry->offset,
4171 entry->root, nocow_ctx);
4173 if (ret == COPY_COMPLETE) {
4181 while (!list_empty(&nocow_ctx->inodes)) {
4182 struct scrub_nocow_inode *entry;
4183 entry = list_first_entry(&nocow_ctx->inodes,
4184 struct scrub_nocow_inode,
4186 list_del_init(&entry->list);
4189 if (trans && !IS_ERR(trans))
4190 btrfs_end_transaction(trans, root);
4192 btrfs_dev_replace_stats_inc(&fs_info->dev_replace.
4193 num_uncorrectable_read_errors);
4195 btrfs_free_path(path);
4198 scrub_pending_trans_workers_dec(sctx);
4201 static int check_extent_to_block(struct inode *inode, u64 start, u64 len,
4204 struct extent_state *cached_state = NULL;
4205 struct btrfs_ordered_extent *ordered;
4206 struct extent_io_tree *io_tree;
4207 struct extent_map *em;
4208 u64 lockstart = start, lockend = start + len - 1;
4211 io_tree = &BTRFS_I(inode)->io_tree;
4213 lock_extent_bits(io_tree, lockstart, lockend, 0, &cached_state);
4214 ordered = btrfs_lookup_ordered_range(inode, lockstart, len);
4216 btrfs_put_ordered_extent(ordered);
4221 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
4228 * This extent does not actually cover the logical extent anymore,
4229 * move on to the next inode.
4231 if (em->block_start > logical ||
4232 em->block_start + em->block_len < logical + len) {
4233 free_extent_map(em);
4237 free_extent_map(em);
4240 unlock_extent_cached(io_tree, lockstart, lockend, &cached_state,
4245 static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
4246 struct scrub_copy_nocow_ctx *nocow_ctx)
4248 struct btrfs_fs_info *fs_info = nocow_ctx->sctx->dev_root->fs_info;
4249 struct btrfs_key key;
4250 struct inode *inode;
4252 struct btrfs_root *local_root;
4253 struct extent_io_tree *io_tree;
4254 u64 physical_for_dev_replace;
4255 u64 nocow_ctx_logical;
4256 u64 len = nocow_ctx->len;
4257 unsigned long index;
4262 key.objectid = root;
4263 key.type = BTRFS_ROOT_ITEM_KEY;
4264 key.offset = (u64)-1;
4266 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
4268 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
4269 if (IS_ERR(local_root)) {
4270 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
4271 return PTR_ERR(local_root);
4274 key.type = BTRFS_INODE_ITEM_KEY;
4275 key.objectid = inum;
4277 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
4278 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
4280 return PTR_ERR(inode);
4282 /* Avoid truncate/dio/punch hole.. */
4283 mutex_lock(&inode->i_mutex);
4284 inode_dio_wait(inode);
4286 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
4287 io_tree = &BTRFS_I(inode)->io_tree;
4288 nocow_ctx_logical = nocow_ctx->logical;
4290 ret = check_extent_to_block(inode, offset, len, nocow_ctx_logical);
4292 ret = ret > 0 ? 0 : ret;
4296 while (len >= PAGE_CACHE_SIZE) {
4297 index = offset >> PAGE_CACHE_SHIFT;
4299 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
4301 btrfs_err(fs_info, "find_or_create_page() failed");
4306 if (PageUptodate(page)) {
4307 if (PageDirty(page))
4310 ClearPageError(page);
4311 err = extent_read_full_page(io_tree, page,
4313 nocow_ctx->mirror_num);
4321 * If the page has been remove from the page cache,
4322 * the data on it is meaningless, because it may be
4323 * old one, the new data may be written into the new
4324 * page in the page cache.
4326 if (page->mapping != inode->i_mapping) {
4328 page_cache_release(page);
4331 if (!PageUptodate(page)) {
4337 ret = check_extent_to_block(inode, offset, len,
4340 ret = ret > 0 ? 0 : ret;
4344 err = write_page_nocow(nocow_ctx->sctx,
4345 physical_for_dev_replace, page);
4350 page_cache_release(page);
4355 offset += PAGE_CACHE_SIZE;
4356 physical_for_dev_replace += PAGE_CACHE_SIZE;
4357 nocow_ctx_logical += PAGE_CACHE_SIZE;
4358 len -= PAGE_CACHE_SIZE;
4360 ret = COPY_COMPLETE;
4362 mutex_unlock(&inode->i_mutex);
4367 static int write_page_nocow(struct scrub_ctx *sctx,
4368 u64 physical_for_dev_replace, struct page *page)
4371 struct btrfs_device *dev;
4374 dev = sctx->wr_ctx.tgtdev;
4378 printk_ratelimited(KERN_WARNING
4379 "BTRFS: scrub write_page_nocow(bdev == NULL) is unexpected!\n");
4382 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
4384 spin_lock(&sctx->stat_lock);
4385 sctx->stat.malloc_errors++;
4386 spin_unlock(&sctx->stat_lock);
4389 bio->bi_iter.bi_size = 0;
4390 bio->bi_iter.bi_sector = physical_for_dev_replace >> 9;
4391 bio->bi_bdev = dev->bdev;
4392 ret = bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
4393 if (ret != PAGE_CACHE_SIZE) {
4396 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
4400 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio))
4401 goto leave_with_eio;