2 * Copyright (C) 2011, 2012 STRATO. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/blkdev.h>
20 #include <linux/ratelimit.h>
24 #include "ordered-data.h"
25 #include "transaction.h"
27 #include "extent_io.h"
28 #include "dev-replace.h"
29 #include "check-integrity.h"
30 #include "rcu-string.h"
34 * This is only the first step towards a full-features scrub. It reads all
35 * extent and super block and verifies the checksums. In case a bad checksum
36 * is found or the extent cannot be read, good data will be written back if
39 * Future enhancements:
40 * - In case an unrepairable extent is encountered, track which files are
41 * affected and report them
42 * - track and record media errors, throw out bad devices
43 * - add a mode to also read unallocated space
50 * the following three values only influence the performance.
51 * The last one configures the number of parallel and outstanding I/O
52 * operations. The first two values configure an upper limit for the number
53 * of (dynamically allocated) pages that are added to a bio.
55 #define SCRUB_PAGES_PER_RD_BIO 32 /* 128k per bio */
56 #define SCRUB_PAGES_PER_WR_BIO 32 /* 128k per bio */
57 #define SCRUB_BIOS_PER_SCTX 64 /* 8MB per device in flight */
60 * the following value times PAGE_SIZE needs to be large enough to match the
61 * largest node/leaf/sector size that shall be supported.
62 * Values larger than BTRFS_STRIPE_LEN are not supported.
64 #define SCRUB_MAX_PAGES_PER_BLOCK 16 /* 64k per node/leaf/sector */
66 struct scrub_recover
{
68 struct btrfs_bio
*bbio
;
73 struct scrub_block
*sblock
;
75 struct btrfs_device
*dev
;
76 struct list_head list
;
77 u64 flags
; /* extent flags */
81 u64 physical_for_dev_replace
;
84 unsigned int mirror_num
:8;
85 unsigned int have_csum
:1;
86 unsigned int io_error
:1;
88 u8 csum
[BTRFS_CSUM_SIZE
];
90 struct scrub_recover
*recover
;
95 struct scrub_ctx
*sctx
;
96 struct btrfs_device
*dev
;
101 #if SCRUB_PAGES_PER_WR_BIO >= SCRUB_PAGES_PER_RD_BIO
102 struct scrub_page
*pagev
[SCRUB_PAGES_PER_WR_BIO
];
104 struct scrub_page
*pagev
[SCRUB_PAGES_PER_RD_BIO
];
108 struct btrfs_work work
;
112 struct scrub_page
*pagev
[SCRUB_MAX_PAGES_PER_BLOCK
];
114 atomic_t outstanding_pages
;
115 atomic_t refs
; /* free mem on transition to zero */
116 struct scrub_ctx
*sctx
;
117 struct scrub_parity
*sparity
;
119 unsigned int header_error
:1;
120 unsigned int checksum_error
:1;
121 unsigned int no_io_error_seen
:1;
122 unsigned int generation_error
:1; /* also sets header_error */
124 /* The following is for the data used to check parity */
125 /* It is for the data with checksum */
126 unsigned int data_corrected
:1;
128 struct btrfs_work work
;
131 /* Used for the chunks with parity stripe such RAID5/6 */
132 struct scrub_parity
{
133 struct scrub_ctx
*sctx
;
135 struct btrfs_device
*scrub_dev
;
147 struct list_head spages
;
149 /* Work of parity check and repair */
150 struct btrfs_work work
;
152 /* Mark the parity blocks which have data */
153 unsigned long *dbitmap
;
156 * Mark the parity blocks which have data, but errors happen when
157 * read data or check data
159 unsigned long *ebitmap
;
161 unsigned long bitmap
[0];
164 struct scrub_wr_ctx
{
165 struct scrub_bio
*wr_curr_bio
;
166 struct btrfs_device
*tgtdev
;
167 int pages_per_wr_bio
; /* <= SCRUB_PAGES_PER_WR_BIO */
168 atomic_t flush_all_writes
;
169 struct mutex wr_lock
;
173 struct scrub_bio
*bios
[SCRUB_BIOS_PER_SCTX
];
174 struct btrfs_root
*dev_root
;
177 atomic_t bios_in_flight
;
178 atomic_t workers_pending
;
179 spinlock_t list_lock
;
180 wait_queue_head_t list_wait
;
182 struct list_head csum_list
;
185 int pages_per_rd_bio
;
190 struct scrub_wr_ctx wr_ctx
;
195 struct btrfs_scrub_progress stat
;
196 spinlock_t stat_lock
;
199 * Use a ref counter to avoid use-after-free issues. Scrub workers
200 * decrement bios_in_flight and workers_pending and then do a wakeup
201 * on the list_wait wait queue. We must ensure the main scrub task
202 * doesn't free the scrub context before or while the workers are
203 * doing the wakeup() call.
208 struct scrub_fixup_nodatasum
{
209 struct scrub_ctx
*sctx
;
210 struct btrfs_device
*dev
;
212 struct btrfs_root
*root
;
213 struct btrfs_work work
;
217 struct scrub_nocow_inode
{
221 struct list_head list
;
224 struct scrub_copy_nocow_ctx
{
225 struct scrub_ctx
*sctx
;
229 u64 physical_for_dev_replace
;
230 struct list_head inodes
;
231 struct btrfs_work work
;
234 struct scrub_warning
{
235 struct btrfs_path
*path
;
236 u64 extent_item_size
;
240 struct btrfs_device
*dev
;
243 static void scrub_pending_bio_inc(struct scrub_ctx
*sctx
);
244 static void scrub_pending_bio_dec(struct scrub_ctx
*sctx
);
245 static void scrub_pending_trans_workers_inc(struct scrub_ctx
*sctx
);
246 static void scrub_pending_trans_workers_dec(struct scrub_ctx
*sctx
);
247 static int scrub_handle_errored_block(struct scrub_block
*sblock_to_check
);
248 static int scrub_setup_recheck_block(struct scrub_block
*original_sblock
,
249 struct scrub_block
*sblocks_for_recheck
);
250 static void scrub_recheck_block(struct btrfs_fs_info
*fs_info
,
251 struct scrub_block
*sblock
, int is_metadata
,
252 int have_csum
, u8
*csum
, u64 generation
,
253 u16 csum_size
, int retry_failed_mirror
);
254 static void scrub_recheck_block_checksum(struct btrfs_fs_info
*fs_info
,
255 struct scrub_block
*sblock
,
256 int is_metadata
, int have_csum
,
257 const u8
*csum
, u64 generation
,
259 static int scrub_repair_block_from_good_copy(struct scrub_block
*sblock_bad
,
260 struct scrub_block
*sblock_good
);
261 static int scrub_repair_page_from_good_copy(struct scrub_block
*sblock_bad
,
262 struct scrub_block
*sblock_good
,
263 int page_num
, int force_write
);
264 static void scrub_write_block_to_dev_replace(struct scrub_block
*sblock
);
265 static int scrub_write_page_to_dev_replace(struct scrub_block
*sblock
,
267 static int scrub_checksum_data(struct scrub_block
*sblock
);
268 static int scrub_checksum_tree_block(struct scrub_block
*sblock
);
269 static int scrub_checksum_super(struct scrub_block
*sblock
);
270 static void scrub_block_get(struct scrub_block
*sblock
);
271 static void scrub_block_put(struct scrub_block
*sblock
);
272 static void scrub_page_get(struct scrub_page
*spage
);
273 static void scrub_page_put(struct scrub_page
*spage
);
274 static void scrub_parity_get(struct scrub_parity
*sparity
);
275 static void scrub_parity_put(struct scrub_parity
*sparity
);
276 static int scrub_add_page_to_rd_bio(struct scrub_ctx
*sctx
,
277 struct scrub_page
*spage
);
278 static int scrub_pages(struct scrub_ctx
*sctx
, u64 logical
, u64 len
,
279 u64 physical
, struct btrfs_device
*dev
, u64 flags
,
280 u64 gen
, int mirror_num
, u8
*csum
, int force
,
281 u64 physical_for_dev_replace
);
282 static void scrub_bio_end_io(struct bio
*bio
);
283 static void scrub_bio_end_io_worker(struct btrfs_work
*work
);
284 static void scrub_block_complete(struct scrub_block
*sblock
);
285 static void scrub_remap_extent(struct btrfs_fs_info
*fs_info
,
286 u64 extent_logical
, u64 extent_len
,
287 u64
*extent_physical
,
288 struct btrfs_device
**extent_dev
,
289 int *extent_mirror_num
);
290 static int scrub_setup_wr_ctx(struct scrub_ctx
*sctx
,
291 struct scrub_wr_ctx
*wr_ctx
,
292 struct btrfs_fs_info
*fs_info
,
293 struct btrfs_device
*dev
,
295 static void scrub_free_wr_ctx(struct scrub_wr_ctx
*wr_ctx
);
296 static int scrub_add_page_to_wr_bio(struct scrub_ctx
*sctx
,
297 struct scrub_page
*spage
);
298 static void scrub_wr_submit(struct scrub_ctx
*sctx
);
299 static void scrub_wr_bio_end_io(struct bio
*bio
);
300 static void scrub_wr_bio_end_io_worker(struct btrfs_work
*work
);
301 static int write_page_nocow(struct scrub_ctx
*sctx
,
302 u64 physical_for_dev_replace
, struct page
*page
);
303 static int copy_nocow_pages_for_inode(u64 inum
, u64 offset
, u64 root
,
304 struct scrub_copy_nocow_ctx
*ctx
);
305 static int copy_nocow_pages(struct scrub_ctx
*sctx
, u64 logical
, u64 len
,
306 int mirror_num
, u64 physical_for_dev_replace
);
307 static void copy_nocow_pages_worker(struct btrfs_work
*work
);
308 static void __scrub_blocked_if_needed(struct btrfs_fs_info
*fs_info
);
309 static void scrub_blocked_if_needed(struct btrfs_fs_info
*fs_info
);
310 static void scrub_put_ctx(struct scrub_ctx
*sctx
);
313 static void scrub_pending_bio_inc(struct scrub_ctx
*sctx
)
315 atomic_inc(&sctx
->refs
);
316 atomic_inc(&sctx
->bios_in_flight
);
319 static void scrub_pending_bio_dec(struct scrub_ctx
*sctx
)
321 atomic_dec(&sctx
->bios_in_flight
);
322 wake_up(&sctx
->list_wait
);
326 static void __scrub_blocked_if_needed(struct btrfs_fs_info
*fs_info
)
328 while (atomic_read(&fs_info
->scrub_pause_req
)) {
329 mutex_unlock(&fs_info
->scrub_lock
);
330 wait_event(fs_info
->scrub_pause_wait
,
331 atomic_read(&fs_info
->scrub_pause_req
) == 0);
332 mutex_lock(&fs_info
->scrub_lock
);
336 static void scrub_pause_on(struct btrfs_fs_info
*fs_info
)
338 atomic_inc(&fs_info
->scrubs_paused
);
339 wake_up(&fs_info
->scrub_pause_wait
);
342 static void scrub_pause_off(struct btrfs_fs_info
*fs_info
)
344 mutex_lock(&fs_info
->scrub_lock
);
345 __scrub_blocked_if_needed(fs_info
);
346 atomic_dec(&fs_info
->scrubs_paused
);
347 mutex_unlock(&fs_info
->scrub_lock
);
349 wake_up(&fs_info
->scrub_pause_wait
);
352 static void scrub_blocked_if_needed(struct btrfs_fs_info
*fs_info
)
354 scrub_pause_on(fs_info
);
355 scrub_pause_off(fs_info
);
359 * used for workers that require transaction commits (i.e., for the
362 static void scrub_pending_trans_workers_inc(struct scrub_ctx
*sctx
)
364 struct btrfs_fs_info
*fs_info
= sctx
->dev_root
->fs_info
;
366 atomic_inc(&sctx
->refs
);
368 * increment scrubs_running to prevent cancel requests from
369 * completing as long as a worker is running. we must also
370 * increment scrubs_paused to prevent deadlocking on pause
371 * requests used for transactions commits (as the worker uses a
372 * transaction context). it is safe to regard the worker
373 * as paused for all matters practical. effectively, we only
374 * avoid cancellation requests from completing.
376 mutex_lock(&fs_info
->scrub_lock
);
377 atomic_inc(&fs_info
->scrubs_running
);
378 atomic_inc(&fs_info
->scrubs_paused
);
379 mutex_unlock(&fs_info
->scrub_lock
);
382 * check if @scrubs_running=@scrubs_paused condition
383 * inside wait_event() is not an atomic operation.
384 * which means we may inc/dec @scrub_running/paused
385 * at any time. Let's wake up @scrub_pause_wait as
386 * much as we can to let commit transaction blocked less.
388 wake_up(&fs_info
->scrub_pause_wait
);
390 atomic_inc(&sctx
->workers_pending
);
393 /* used for workers that require transaction commits */
394 static void scrub_pending_trans_workers_dec(struct scrub_ctx
*sctx
)
396 struct btrfs_fs_info
*fs_info
= sctx
->dev_root
->fs_info
;
399 * see scrub_pending_trans_workers_inc() why we're pretending
400 * to be paused in the scrub counters
402 mutex_lock(&fs_info
->scrub_lock
);
403 atomic_dec(&fs_info
->scrubs_running
);
404 atomic_dec(&fs_info
->scrubs_paused
);
405 mutex_unlock(&fs_info
->scrub_lock
);
406 atomic_dec(&sctx
->workers_pending
);
407 wake_up(&fs_info
->scrub_pause_wait
);
408 wake_up(&sctx
->list_wait
);
412 static void scrub_free_csums(struct scrub_ctx
*sctx
)
414 while (!list_empty(&sctx
->csum_list
)) {
415 struct btrfs_ordered_sum
*sum
;
416 sum
= list_first_entry(&sctx
->csum_list
,
417 struct btrfs_ordered_sum
, list
);
418 list_del(&sum
->list
);
423 static noinline_for_stack
void scrub_free_ctx(struct scrub_ctx
*sctx
)
430 scrub_free_wr_ctx(&sctx
->wr_ctx
);
432 /* this can happen when scrub is cancelled */
433 if (sctx
->curr
!= -1) {
434 struct scrub_bio
*sbio
= sctx
->bios
[sctx
->curr
];
436 for (i
= 0; i
< sbio
->page_count
; i
++) {
437 WARN_ON(!sbio
->pagev
[i
]->page
);
438 scrub_block_put(sbio
->pagev
[i
]->sblock
);
443 for (i
= 0; i
< SCRUB_BIOS_PER_SCTX
; ++i
) {
444 struct scrub_bio
*sbio
= sctx
->bios
[i
];
451 scrub_free_csums(sctx
);
455 static void scrub_put_ctx(struct scrub_ctx
*sctx
)
457 if (atomic_dec_and_test(&sctx
->refs
))
458 scrub_free_ctx(sctx
);
461 static noinline_for_stack
462 struct scrub_ctx
*scrub_setup_ctx(struct btrfs_device
*dev
, int is_dev_replace
)
464 struct scrub_ctx
*sctx
;
466 struct btrfs_fs_info
*fs_info
= dev
->dev_root
->fs_info
;
469 sctx
= kzalloc(sizeof(*sctx
), GFP_NOFS
);
472 atomic_set(&sctx
->refs
, 1);
473 sctx
->is_dev_replace
= is_dev_replace
;
474 sctx
->pages_per_rd_bio
= SCRUB_PAGES_PER_RD_BIO
;
476 sctx
->dev_root
= dev
->dev_root
;
477 for (i
= 0; i
< SCRUB_BIOS_PER_SCTX
; ++i
) {
478 struct scrub_bio
*sbio
;
480 sbio
= kzalloc(sizeof(*sbio
), GFP_NOFS
);
483 sctx
->bios
[i
] = sbio
;
487 sbio
->page_count
= 0;
488 btrfs_init_work(&sbio
->work
, btrfs_scrub_helper
,
489 scrub_bio_end_io_worker
, NULL
, NULL
);
491 if (i
!= SCRUB_BIOS_PER_SCTX
- 1)
492 sctx
->bios
[i
]->next_free
= i
+ 1;
494 sctx
->bios
[i
]->next_free
= -1;
496 sctx
->first_free
= 0;
497 sctx
->nodesize
= dev
->dev_root
->nodesize
;
498 sctx
->sectorsize
= dev
->dev_root
->sectorsize
;
499 atomic_set(&sctx
->bios_in_flight
, 0);
500 atomic_set(&sctx
->workers_pending
, 0);
501 atomic_set(&sctx
->cancel_req
, 0);
502 sctx
->csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
503 INIT_LIST_HEAD(&sctx
->csum_list
);
505 spin_lock_init(&sctx
->list_lock
);
506 spin_lock_init(&sctx
->stat_lock
);
507 init_waitqueue_head(&sctx
->list_wait
);
509 ret
= scrub_setup_wr_ctx(sctx
, &sctx
->wr_ctx
, fs_info
,
510 fs_info
->dev_replace
.tgtdev
, is_dev_replace
);
512 scrub_free_ctx(sctx
);
518 scrub_free_ctx(sctx
);
519 return ERR_PTR(-ENOMEM
);
522 static int scrub_print_warning_inode(u64 inum
, u64 offset
, u64 root
,
529 struct extent_buffer
*eb
;
530 struct btrfs_inode_item
*inode_item
;
531 struct scrub_warning
*swarn
= warn_ctx
;
532 struct btrfs_fs_info
*fs_info
= swarn
->dev
->dev_root
->fs_info
;
533 struct inode_fs_paths
*ipath
= NULL
;
534 struct btrfs_root
*local_root
;
535 struct btrfs_key root_key
;
536 struct btrfs_key key
;
538 root_key
.objectid
= root
;
539 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
540 root_key
.offset
= (u64
)-1;
541 local_root
= btrfs_read_fs_root_no_name(fs_info
, &root_key
);
542 if (IS_ERR(local_root
)) {
543 ret
= PTR_ERR(local_root
);
548 * this makes the path point to (inum INODE_ITEM ioff)
551 key
.type
= BTRFS_INODE_ITEM_KEY
;
554 ret
= btrfs_search_slot(NULL
, local_root
, &key
, swarn
->path
, 0, 0);
556 btrfs_release_path(swarn
->path
);
560 eb
= swarn
->path
->nodes
[0];
561 inode_item
= btrfs_item_ptr(eb
, swarn
->path
->slots
[0],
562 struct btrfs_inode_item
);
563 isize
= btrfs_inode_size(eb
, inode_item
);
564 nlink
= btrfs_inode_nlink(eb
, inode_item
);
565 btrfs_release_path(swarn
->path
);
567 ipath
= init_ipath(4096, local_root
, swarn
->path
);
569 ret
= PTR_ERR(ipath
);
573 ret
= paths_from_inode(inum
, ipath
);
579 * we deliberately ignore the bit ipath might have been too small to
580 * hold all of the paths here
582 for (i
= 0; i
< ipath
->fspath
->elem_cnt
; ++i
)
583 btrfs_warn_in_rcu(fs_info
, "%s at logical %llu on dev "
584 "%s, sector %llu, root %llu, inode %llu, offset %llu, "
585 "length %llu, links %u (path: %s)", swarn
->errstr
,
586 swarn
->logical
, rcu_str_deref(swarn
->dev
->name
),
587 (unsigned long long)swarn
->sector
, root
, inum
, offset
,
588 min(isize
- offset
, (u64
)PAGE_SIZE
), nlink
,
589 (char *)(unsigned long)ipath
->fspath
->val
[i
]);
595 btrfs_warn_in_rcu(fs_info
, "%s at logical %llu on dev "
596 "%s, sector %llu, root %llu, inode %llu, offset %llu: path "
597 "resolving failed with ret=%d", swarn
->errstr
,
598 swarn
->logical
, rcu_str_deref(swarn
->dev
->name
),
599 (unsigned long long)swarn
->sector
, root
, inum
, offset
, ret
);
605 static void scrub_print_warning(const char *errstr
, struct scrub_block
*sblock
)
607 struct btrfs_device
*dev
;
608 struct btrfs_fs_info
*fs_info
;
609 struct btrfs_path
*path
;
610 struct btrfs_key found_key
;
611 struct extent_buffer
*eb
;
612 struct btrfs_extent_item
*ei
;
613 struct scrub_warning swarn
;
614 unsigned long ptr
= 0;
622 WARN_ON(sblock
->page_count
< 1);
623 dev
= sblock
->pagev
[0]->dev
;
624 fs_info
= sblock
->sctx
->dev_root
->fs_info
;
626 path
= btrfs_alloc_path();
630 swarn
.sector
= (sblock
->pagev
[0]->physical
) >> 9;
631 swarn
.logical
= sblock
->pagev
[0]->logical
;
632 swarn
.errstr
= errstr
;
635 ret
= extent_from_logical(fs_info
, swarn
.logical
, path
, &found_key
,
640 extent_item_pos
= swarn
.logical
- found_key
.objectid
;
641 swarn
.extent_item_size
= found_key
.offset
;
644 ei
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_extent_item
);
645 item_size
= btrfs_item_size_nr(eb
, path
->slots
[0]);
647 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
649 ret
= tree_backref_for_extent(&ptr
, eb
, &found_key
, ei
,
650 item_size
, &ref_root
,
652 btrfs_warn_in_rcu(fs_info
,
653 "%s at logical %llu on dev %s, "
654 "sector %llu: metadata %s (level %d) in tree "
655 "%llu", errstr
, swarn
.logical
,
656 rcu_str_deref(dev
->name
),
657 (unsigned long long)swarn
.sector
,
658 ref_level
? "node" : "leaf",
659 ret
< 0 ? -1 : ref_level
,
660 ret
< 0 ? -1 : ref_root
);
662 btrfs_release_path(path
);
664 btrfs_release_path(path
);
667 iterate_extent_inodes(fs_info
, found_key
.objectid
,
669 scrub_print_warning_inode
, &swarn
);
673 btrfs_free_path(path
);
676 static int scrub_fixup_readpage(u64 inum
, u64 offset
, u64 root
, void *fixup_ctx
)
678 struct page
*page
= NULL
;
680 struct scrub_fixup_nodatasum
*fixup
= fixup_ctx
;
683 struct btrfs_key key
;
684 struct inode
*inode
= NULL
;
685 struct btrfs_fs_info
*fs_info
;
686 u64 end
= offset
+ PAGE_SIZE
- 1;
687 struct btrfs_root
*local_root
;
691 key
.type
= BTRFS_ROOT_ITEM_KEY
;
692 key
.offset
= (u64
)-1;
694 fs_info
= fixup
->root
->fs_info
;
695 srcu_index
= srcu_read_lock(&fs_info
->subvol_srcu
);
697 local_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
698 if (IS_ERR(local_root
)) {
699 srcu_read_unlock(&fs_info
->subvol_srcu
, srcu_index
);
700 return PTR_ERR(local_root
);
703 key
.type
= BTRFS_INODE_ITEM_KEY
;
706 inode
= btrfs_iget(fs_info
->sb
, &key
, local_root
, NULL
);
707 srcu_read_unlock(&fs_info
->subvol_srcu
, srcu_index
);
709 return PTR_ERR(inode
);
711 index
= offset
>> PAGE_CACHE_SHIFT
;
713 page
= find_or_create_page(inode
->i_mapping
, index
, GFP_NOFS
);
719 if (PageUptodate(page
)) {
720 if (PageDirty(page
)) {
722 * we need to write the data to the defect sector. the
723 * data that was in that sector is not in memory,
724 * because the page was modified. we must not write the
725 * modified page to that sector.
727 * TODO: what could be done here: wait for the delalloc
728 * runner to write out that page (might involve
729 * COW) and see whether the sector is still
730 * referenced afterwards.
732 * For the meantime, we'll treat this error
733 * incorrectable, although there is a chance that a
734 * later scrub will find the bad sector again and that
735 * there's no dirty page in memory, then.
740 ret
= repair_io_failure(inode
, offset
, PAGE_SIZE
,
741 fixup
->logical
, page
,
742 offset
- page_offset(page
),
748 * we need to get good data first. the general readpage path
749 * will call repair_io_failure for us, we just have to make
750 * sure we read the bad mirror.
752 ret
= set_extent_bits(&BTRFS_I(inode
)->io_tree
, offset
, end
,
753 EXTENT_DAMAGED
, GFP_NOFS
);
755 /* set_extent_bits should give proper error */
762 ret
= extent_read_full_page(&BTRFS_I(inode
)->io_tree
, page
,
765 wait_on_page_locked(page
);
767 corrected
= !test_range_bit(&BTRFS_I(inode
)->io_tree
, offset
,
768 end
, EXTENT_DAMAGED
, 0, NULL
);
770 clear_extent_bits(&BTRFS_I(inode
)->io_tree
, offset
, end
,
771 EXTENT_DAMAGED
, GFP_NOFS
);
783 if (ret
== 0 && corrected
) {
785 * we only need to call readpage for one of the inodes belonging
786 * to this extent. so make iterate_extent_inodes stop
794 static void scrub_fixup_nodatasum(struct btrfs_work
*work
)
797 struct scrub_fixup_nodatasum
*fixup
;
798 struct scrub_ctx
*sctx
;
799 struct btrfs_trans_handle
*trans
= NULL
;
800 struct btrfs_path
*path
;
801 int uncorrectable
= 0;
803 fixup
= container_of(work
, struct scrub_fixup_nodatasum
, work
);
806 path
= btrfs_alloc_path();
808 spin_lock(&sctx
->stat_lock
);
809 ++sctx
->stat
.malloc_errors
;
810 spin_unlock(&sctx
->stat_lock
);
815 trans
= btrfs_join_transaction(fixup
->root
);
822 * the idea is to trigger a regular read through the standard path. we
823 * read a page from the (failed) logical address by specifying the
824 * corresponding copynum of the failed sector. thus, that readpage is
826 * that is the point where on-the-fly error correction will kick in
827 * (once it's finished) and rewrite the failed sector if a good copy
830 ret
= iterate_inodes_from_logical(fixup
->logical
, fixup
->root
->fs_info
,
831 path
, scrub_fixup_readpage
,
839 spin_lock(&sctx
->stat_lock
);
840 ++sctx
->stat
.corrected_errors
;
841 spin_unlock(&sctx
->stat_lock
);
844 if (trans
&& !IS_ERR(trans
))
845 btrfs_end_transaction(trans
, fixup
->root
);
847 spin_lock(&sctx
->stat_lock
);
848 ++sctx
->stat
.uncorrectable_errors
;
849 spin_unlock(&sctx
->stat_lock
);
850 btrfs_dev_replace_stats_inc(
851 &sctx
->dev_root
->fs_info
->dev_replace
.
852 num_uncorrectable_read_errors
);
853 btrfs_err_rl_in_rcu(sctx
->dev_root
->fs_info
,
854 "unable to fixup (nodatasum) error at logical %llu on dev %s",
855 fixup
->logical
, rcu_str_deref(fixup
->dev
->name
));
858 btrfs_free_path(path
);
861 scrub_pending_trans_workers_dec(sctx
);
864 static inline void scrub_get_recover(struct scrub_recover
*recover
)
866 atomic_inc(&recover
->refs
);
869 static inline void scrub_put_recover(struct scrub_recover
*recover
)
871 if (atomic_dec_and_test(&recover
->refs
)) {
872 btrfs_put_bbio(recover
->bbio
);
878 * scrub_handle_errored_block gets called when either verification of the
879 * pages failed or the bio failed to read, e.g. with EIO. In the latter
880 * case, this function handles all pages in the bio, even though only one
882 * The goal of this function is to repair the errored block by using the
883 * contents of one of the mirrors.
885 static int scrub_handle_errored_block(struct scrub_block
*sblock_to_check
)
887 struct scrub_ctx
*sctx
= sblock_to_check
->sctx
;
888 struct btrfs_device
*dev
;
889 struct btrfs_fs_info
*fs_info
;
893 unsigned int failed_mirror_index
;
894 unsigned int is_metadata
;
895 unsigned int have_csum
;
897 struct scrub_block
*sblocks_for_recheck
; /* holds one for each mirror */
898 struct scrub_block
*sblock_bad
;
903 static DEFINE_RATELIMIT_STATE(_rs
, DEFAULT_RATELIMIT_INTERVAL
,
904 DEFAULT_RATELIMIT_BURST
);
906 BUG_ON(sblock_to_check
->page_count
< 1);
907 fs_info
= sctx
->dev_root
->fs_info
;
908 if (sblock_to_check
->pagev
[0]->flags
& BTRFS_EXTENT_FLAG_SUPER
) {
910 * if we find an error in a super block, we just report it.
911 * They will get written with the next transaction commit
914 spin_lock(&sctx
->stat_lock
);
915 ++sctx
->stat
.super_errors
;
916 spin_unlock(&sctx
->stat_lock
);
919 length
= sblock_to_check
->page_count
* PAGE_SIZE
;
920 logical
= sblock_to_check
->pagev
[0]->logical
;
921 generation
= sblock_to_check
->pagev
[0]->generation
;
922 BUG_ON(sblock_to_check
->pagev
[0]->mirror_num
< 1);
923 failed_mirror_index
= sblock_to_check
->pagev
[0]->mirror_num
- 1;
924 is_metadata
= !(sblock_to_check
->pagev
[0]->flags
&
925 BTRFS_EXTENT_FLAG_DATA
);
926 have_csum
= sblock_to_check
->pagev
[0]->have_csum
;
927 csum
= sblock_to_check
->pagev
[0]->csum
;
928 dev
= sblock_to_check
->pagev
[0]->dev
;
930 if (sctx
->is_dev_replace
&& !is_metadata
&& !have_csum
) {
931 sblocks_for_recheck
= NULL
;
936 * read all mirrors one after the other. This includes to
937 * re-read the extent or metadata block that failed (that was
938 * the cause that this fixup code is called) another time,
939 * page by page this time in order to know which pages
940 * caused I/O errors and which ones are good (for all mirrors).
941 * It is the goal to handle the situation when more than one
942 * mirror contains I/O errors, but the errors do not
943 * overlap, i.e. the data can be repaired by selecting the
944 * pages from those mirrors without I/O error on the
945 * particular pages. One example (with blocks >= 2 * PAGE_SIZE)
946 * would be that mirror #1 has an I/O error on the first page,
947 * the second page is good, and mirror #2 has an I/O error on
948 * the second page, but the first page is good.
949 * Then the first page of the first mirror can be repaired by
950 * taking the first page of the second mirror, and the
951 * second page of the second mirror can be repaired by
952 * copying the contents of the 2nd page of the 1st mirror.
953 * One more note: if the pages of one mirror contain I/O
954 * errors, the checksum cannot be verified. In order to get
955 * the best data for repairing, the first attempt is to find
956 * a mirror without I/O errors and with a validated checksum.
957 * Only if this is not possible, the pages are picked from
958 * mirrors with I/O errors without considering the checksum.
959 * If the latter is the case, at the end, the checksum of the
960 * repaired area is verified in order to correctly maintain
964 sblocks_for_recheck
= kcalloc(BTRFS_MAX_MIRRORS
,
965 sizeof(*sblocks_for_recheck
), GFP_NOFS
);
966 if (!sblocks_for_recheck
) {
967 spin_lock(&sctx
->stat_lock
);
968 sctx
->stat
.malloc_errors
++;
969 sctx
->stat
.read_errors
++;
970 sctx
->stat
.uncorrectable_errors
++;
971 spin_unlock(&sctx
->stat_lock
);
972 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_READ_ERRS
);
976 /* setup the context, map the logical blocks and alloc the pages */
977 ret
= scrub_setup_recheck_block(sblock_to_check
, sblocks_for_recheck
);
979 spin_lock(&sctx
->stat_lock
);
980 sctx
->stat
.read_errors
++;
981 sctx
->stat
.uncorrectable_errors
++;
982 spin_unlock(&sctx
->stat_lock
);
983 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_READ_ERRS
);
986 BUG_ON(failed_mirror_index
>= BTRFS_MAX_MIRRORS
);
987 sblock_bad
= sblocks_for_recheck
+ failed_mirror_index
;
989 /* build and submit the bios for the failed mirror, check checksums */
990 scrub_recheck_block(fs_info
, sblock_bad
, is_metadata
, have_csum
,
991 csum
, generation
, sctx
->csum_size
, 1);
993 if (!sblock_bad
->header_error
&& !sblock_bad
->checksum_error
&&
994 sblock_bad
->no_io_error_seen
) {
996 * the error disappeared after reading page by page, or
997 * the area was part of a huge bio and other parts of the
998 * bio caused I/O errors, or the block layer merged several
999 * read requests into one and the error is caused by a
1000 * different bio (usually one of the two latter cases is
1003 spin_lock(&sctx
->stat_lock
);
1004 sctx
->stat
.unverified_errors
++;
1005 sblock_to_check
->data_corrected
= 1;
1006 spin_unlock(&sctx
->stat_lock
);
1008 if (sctx
->is_dev_replace
)
1009 scrub_write_block_to_dev_replace(sblock_bad
);
1013 if (!sblock_bad
->no_io_error_seen
) {
1014 spin_lock(&sctx
->stat_lock
);
1015 sctx
->stat
.read_errors
++;
1016 spin_unlock(&sctx
->stat_lock
);
1017 if (__ratelimit(&_rs
))
1018 scrub_print_warning("i/o error", sblock_to_check
);
1019 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_READ_ERRS
);
1020 } else if (sblock_bad
->checksum_error
) {
1021 spin_lock(&sctx
->stat_lock
);
1022 sctx
->stat
.csum_errors
++;
1023 spin_unlock(&sctx
->stat_lock
);
1024 if (__ratelimit(&_rs
))
1025 scrub_print_warning("checksum error", sblock_to_check
);
1026 btrfs_dev_stat_inc_and_print(dev
,
1027 BTRFS_DEV_STAT_CORRUPTION_ERRS
);
1028 } else if (sblock_bad
->header_error
) {
1029 spin_lock(&sctx
->stat_lock
);
1030 sctx
->stat
.verify_errors
++;
1031 spin_unlock(&sctx
->stat_lock
);
1032 if (__ratelimit(&_rs
))
1033 scrub_print_warning("checksum/header error",
1035 if (sblock_bad
->generation_error
)
1036 btrfs_dev_stat_inc_and_print(dev
,
1037 BTRFS_DEV_STAT_GENERATION_ERRS
);
1039 btrfs_dev_stat_inc_and_print(dev
,
1040 BTRFS_DEV_STAT_CORRUPTION_ERRS
);
1043 if (sctx
->readonly
) {
1044 ASSERT(!sctx
->is_dev_replace
);
1048 if (!is_metadata
&& !have_csum
) {
1049 struct scrub_fixup_nodatasum
*fixup_nodatasum
;
1051 WARN_ON(sctx
->is_dev_replace
);
1056 * !is_metadata and !have_csum, this means that the data
1057 * might not be COW'ed, that it might be modified
1058 * concurrently. The general strategy to work on the
1059 * commit root does not help in the case when COW is not
1062 fixup_nodatasum
= kzalloc(sizeof(*fixup_nodatasum
), GFP_NOFS
);
1063 if (!fixup_nodatasum
)
1064 goto did_not_correct_error
;
1065 fixup_nodatasum
->sctx
= sctx
;
1066 fixup_nodatasum
->dev
= dev
;
1067 fixup_nodatasum
->logical
= logical
;
1068 fixup_nodatasum
->root
= fs_info
->extent_root
;
1069 fixup_nodatasum
->mirror_num
= failed_mirror_index
+ 1;
1070 scrub_pending_trans_workers_inc(sctx
);
1071 btrfs_init_work(&fixup_nodatasum
->work
, btrfs_scrub_helper
,
1072 scrub_fixup_nodatasum
, NULL
, NULL
);
1073 btrfs_queue_work(fs_info
->scrub_workers
,
1074 &fixup_nodatasum
->work
);
1079 * now build and submit the bios for the other mirrors, check
1081 * First try to pick the mirror which is completely without I/O
1082 * errors and also does not have a checksum error.
1083 * If one is found, and if a checksum is present, the full block
1084 * that is known to contain an error is rewritten. Afterwards
1085 * the block is known to be corrected.
1086 * If a mirror is found which is completely correct, and no
1087 * checksum is present, only those pages are rewritten that had
1088 * an I/O error in the block to be repaired, since it cannot be
1089 * determined, which copy of the other pages is better (and it
1090 * could happen otherwise that a correct page would be
1091 * overwritten by a bad one).
1093 for (mirror_index
= 0;
1094 mirror_index
< BTRFS_MAX_MIRRORS
&&
1095 sblocks_for_recheck
[mirror_index
].page_count
> 0;
1097 struct scrub_block
*sblock_other
;
1099 if (mirror_index
== failed_mirror_index
)
1101 sblock_other
= sblocks_for_recheck
+ mirror_index
;
1103 /* build and submit the bios, check checksums */
1104 scrub_recheck_block(fs_info
, sblock_other
, is_metadata
,
1105 have_csum
, csum
, generation
,
1106 sctx
->csum_size
, 0);
1108 if (!sblock_other
->header_error
&&
1109 !sblock_other
->checksum_error
&&
1110 sblock_other
->no_io_error_seen
) {
1111 if (sctx
->is_dev_replace
) {
1112 scrub_write_block_to_dev_replace(sblock_other
);
1113 goto corrected_error
;
1115 ret
= scrub_repair_block_from_good_copy(
1116 sblock_bad
, sblock_other
);
1118 goto corrected_error
;
1123 if (sblock_bad
->no_io_error_seen
&& !sctx
->is_dev_replace
)
1124 goto did_not_correct_error
;
1127 * In case of I/O errors in the area that is supposed to be
1128 * repaired, continue by picking good copies of those pages.
1129 * Select the good pages from mirrors to rewrite bad pages from
1130 * the area to fix. Afterwards verify the checksum of the block
1131 * that is supposed to be repaired. This verification step is
1132 * only done for the purpose of statistic counting and for the
1133 * final scrub report, whether errors remain.
1134 * A perfect algorithm could make use of the checksum and try
1135 * all possible combinations of pages from the different mirrors
1136 * until the checksum verification succeeds. For example, when
1137 * the 2nd page of mirror #1 faces I/O errors, and the 2nd page
1138 * of mirror #2 is readable but the final checksum test fails,
1139 * then the 2nd page of mirror #3 could be tried, whether now
1140 * the final checksum succeedes. But this would be a rare
1141 * exception and is therefore not implemented. At least it is
1142 * avoided that the good copy is overwritten.
1143 * A more useful improvement would be to pick the sectors
1144 * without I/O error based on sector sizes (512 bytes on legacy
1145 * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one
1146 * mirror could be repaired by taking 512 byte of a different
1147 * mirror, even if other 512 byte sectors in the same PAGE_SIZE
1148 * area are unreadable.
1151 for (page_num
= 0; page_num
< sblock_bad
->page_count
;
1153 struct scrub_page
*page_bad
= sblock_bad
->pagev
[page_num
];
1154 struct scrub_block
*sblock_other
= NULL
;
1156 /* skip no-io-error page in scrub */
1157 if (!page_bad
->io_error
&& !sctx
->is_dev_replace
)
1160 /* try to find no-io-error page in mirrors */
1161 if (page_bad
->io_error
) {
1162 for (mirror_index
= 0;
1163 mirror_index
< BTRFS_MAX_MIRRORS
&&
1164 sblocks_for_recheck
[mirror_index
].page_count
> 0;
1166 if (!sblocks_for_recheck
[mirror_index
].
1167 pagev
[page_num
]->io_error
) {
1168 sblock_other
= sblocks_for_recheck
+
1177 if (sctx
->is_dev_replace
) {
1179 * did not find a mirror to fetch the page
1180 * from. scrub_write_page_to_dev_replace()
1181 * handles this case (page->io_error), by
1182 * filling the block with zeros before
1183 * submitting the write request
1186 sblock_other
= sblock_bad
;
1188 if (scrub_write_page_to_dev_replace(sblock_other
,
1190 btrfs_dev_replace_stats_inc(
1192 fs_info
->dev_replace
.
1196 } else if (sblock_other
) {
1197 ret
= scrub_repair_page_from_good_copy(sblock_bad
,
1201 page_bad
->io_error
= 0;
1207 if (success
&& !sctx
->is_dev_replace
) {
1208 if (is_metadata
|| have_csum
) {
1210 * need to verify the checksum now that all
1211 * sectors on disk are repaired (the write
1212 * request for data to be repaired is on its way).
1213 * Just be lazy and use scrub_recheck_block()
1214 * which re-reads the data before the checksum
1215 * is verified, but most likely the data comes out
1216 * of the page cache.
1218 scrub_recheck_block(fs_info
, sblock_bad
,
1219 is_metadata
, have_csum
, csum
,
1220 generation
, sctx
->csum_size
, 1);
1221 if (!sblock_bad
->header_error
&&
1222 !sblock_bad
->checksum_error
&&
1223 sblock_bad
->no_io_error_seen
)
1224 goto corrected_error
;
1226 goto did_not_correct_error
;
1229 spin_lock(&sctx
->stat_lock
);
1230 sctx
->stat
.corrected_errors
++;
1231 sblock_to_check
->data_corrected
= 1;
1232 spin_unlock(&sctx
->stat_lock
);
1233 btrfs_err_rl_in_rcu(fs_info
,
1234 "fixed up error at logical %llu on dev %s",
1235 logical
, rcu_str_deref(dev
->name
));
1238 did_not_correct_error
:
1239 spin_lock(&sctx
->stat_lock
);
1240 sctx
->stat
.uncorrectable_errors
++;
1241 spin_unlock(&sctx
->stat_lock
);
1242 btrfs_err_rl_in_rcu(fs_info
,
1243 "unable to fixup (regular) error at logical %llu on dev %s",
1244 logical
, rcu_str_deref(dev
->name
));
1248 if (sblocks_for_recheck
) {
1249 for (mirror_index
= 0; mirror_index
< BTRFS_MAX_MIRRORS
;
1251 struct scrub_block
*sblock
= sblocks_for_recheck
+
1253 struct scrub_recover
*recover
;
1256 for (page_index
= 0; page_index
< sblock
->page_count
;
1258 sblock
->pagev
[page_index
]->sblock
= NULL
;
1259 recover
= sblock
->pagev
[page_index
]->recover
;
1261 scrub_put_recover(recover
);
1262 sblock
->pagev
[page_index
]->recover
=
1265 scrub_page_put(sblock
->pagev
[page_index
]);
1268 kfree(sblocks_for_recheck
);
1274 static inline int scrub_nr_raid_mirrors(struct btrfs_bio
*bbio
)
1276 if (bbio
->map_type
& BTRFS_BLOCK_GROUP_RAID5
)
1278 else if (bbio
->map_type
& BTRFS_BLOCK_GROUP_RAID6
)
1281 return (int)bbio
->num_stripes
;
1284 static inline void scrub_stripe_index_and_offset(u64 logical
, u64 map_type
,
1287 int nstripes
, int mirror
,
1293 if (map_type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
1295 for (i
= 0; i
< nstripes
; i
++) {
1296 if (raid_map
[i
] == RAID6_Q_STRIPE
||
1297 raid_map
[i
] == RAID5_P_STRIPE
)
1300 if (logical
>= raid_map
[i
] &&
1301 logical
< raid_map
[i
] + mapped_length
)
1306 *stripe_offset
= logical
- raid_map
[i
];
1308 /* The other RAID type */
1309 *stripe_index
= mirror
;
1314 static int scrub_setup_recheck_block(struct scrub_block
*original_sblock
,
1315 struct scrub_block
*sblocks_for_recheck
)
1317 struct scrub_ctx
*sctx
= original_sblock
->sctx
;
1318 struct btrfs_fs_info
*fs_info
= sctx
->dev_root
->fs_info
;
1319 u64 length
= original_sblock
->page_count
* PAGE_SIZE
;
1320 u64 logical
= original_sblock
->pagev
[0]->logical
;
1321 u64 generation
= original_sblock
->pagev
[0]->generation
;
1322 u64 flags
= original_sblock
->pagev
[0]->flags
;
1323 u64 have_csum
= original_sblock
->pagev
[0]->have_csum
;
1324 struct scrub_recover
*recover
;
1325 struct btrfs_bio
*bbio
;
1336 * note: the two members refs and outstanding_pages
1337 * are not used (and not set) in the blocks that are used for
1338 * the recheck procedure
1341 while (length
> 0) {
1342 sublen
= min_t(u64
, length
, PAGE_SIZE
);
1343 mapped_length
= sublen
;
1347 * with a length of PAGE_SIZE, each returned stripe
1348 * represents one mirror
1350 ret
= btrfs_map_sblock(fs_info
, REQ_GET_READ_MIRRORS
, logical
,
1351 &mapped_length
, &bbio
, 0, 1);
1352 if (ret
|| !bbio
|| mapped_length
< sublen
) {
1353 btrfs_put_bbio(bbio
);
1357 recover
= kzalloc(sizeof(struct scrub_recover
), GFP_NOFS
);
1359 btrfs_put_bbio(bbio
);
1363 atomic_set(&recover
->refs
, 1);
1364 recover
->bbio
= bbio
;
1365 recover
->map_length
= mapped_length
;
1367 BUG_ON(page_index
>= SCRUB_PAGES_PER_RD_BIO
);
1369 nmirrors
= min(scrub_nr_raid_mirrors(bbio
), BTRFS_MAX_MIRRORS
);
1371 for (mirror_index
= 0; mirror_index
< nmirrors
;
1373 struct scrub_block
*sblock
;
1374 struct scrub_page
*page
;
1376 sblock
= sblocks_for_recheck
+ mirror_index
;
1377 sblock
->sctx
= sctx
;
1379 page
= kzalloc(sizeof(*page
), GFP_NOFS
);
1382 spin_lock(&sctx
->stat_lock
);
1383 sctx
->stat
.malloc_errors
++;
1384 spin_unlock(&sctx
->stat_lock
);
1385 scrub_put_recover(recover
);
1388 scrub_page_get(page
);
1389 sblock
->pagev
[page_index
] = page
;
1390 page
->sblock
= sblock
;
1391 page
->flags
= flags
;
1392 page
->generation
= generation
;
1393 page
->logical
= logical
;
1394 page
->have_csum
= have_csum
;
1397 original_sblock
->pagev
[0]->csum
,
1400 scrub_stripe_index_and_offset(logical
,
1409 page
->physical
= bbio
->stripes
[stripe_index
].physical
+
1411 page
->dev
= bbio
->stripes
[stripe_index
].dev
;
1413 BUG_ON(page_index
>= original_sblock
->page_count
);
1414 page
->physical_for_dev_replace
=
1415 original_sblock
->pagev
[page_index
]->
1416 physical_for_dev_replace
;
1417 /* for missing devices, dev->bdev is NULL */
1418 page
->mirror_num
= mirror_index
+ 1;
1419 sblock
->page_count
++;
1420 page
->page
= alloc_page(GFP_NOFS
);
1424 scrub_get_recover(recover
);
1425 page
->recover
= recover
;
1427 scrub_put_recover(recover
);
1436 struct scrub_bio_ret
{
1437 struct completion event
;
1441 static void scrub_bio_wait_endio(struct bio
*bio
)
1443 struct scrub_bio_ret
*ret
= bio
->bi_private
;
1445 ret
->error
= bio
->bi_error
;
1446 complete(&ret
->event
);
1449 static inline int scrub_is_page_on_raid56(struct scrub_page
*page
)
1451 return page
->recover
&&
1452 (page
->recover
->bbio
->map_type
& BTRFS_BLOCK_GROUP_RAID56_MASK
);
1455 static int scrub_submit_raid56_bio_wait(struct btrfs_fs_info
*fs_info
,
1457 struct scrub_page
*page
)
1459 struct scrub_bio_ret done
;
1462 init_completion(&done
.event
);
1464 bio
->bi_iter
.bi_sector
= page
->logical
>> 9;
1465 bio
->bi_private
= &done
;
1466 bio
->bi_end_io
= scrub_bio_wait_endio
;
1468 ret
= raid56_parity_recover(fs_info
->fs_root
, bio
, page
->recover
->bbio
,
1469 page
->recover
->map_length
,
1470 page
->mirror_num
, 0);
1474 wait_for_completion(&done
.event
);
1482 * this function will check the on disk data for checksum errors, header
1483 * errors and read I/O errors. If any I/O errors happen, the exact pages
1484 * which are errored are marked as being bad. The goal is to enable scrub
1485 * to take those pages that are not errored from all the mirrors so that
1486 * the pages that are errored in the just handled mirror can be repaired.
1488 static void scrub_recheck_block(struct btrfs_fs_info
*fs_info
,
1489 struct scrub_block
*sblock
, int is_metadata
,
1490 int have_csum
, u8
*csum
, u64 generation
,
1491 u16 csum_size
, int retry_failed_mirror
)
1495 sblock
->no_io_error_seen
= 1;
1496 sblock
->header_error
= 0;
1497 sblock
->checksum_error
= 0;
1498 sblock
->generation_error
= 0;
1500 for (page_num
= 0; page_num
< sblock
->page_count
; page_num
++) {
1502 struct scrub_page
*page
= sblock
->pagev
[page_num
];
1504 if (page
->dev
->bdev
== NULL
) {
1506 sblock
->no_io_error_seen
= 0;
1510 WARN_ON(!page
->page
);
1511 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
1514 sblock
->no_io_error_seen
= 0;
1517 bio
->bi_bdev
= page
->dev
->bdev
;
1519 bio_add_page(bio
, page
->page
, PAGE_SIZE
, 0);
1520 if (!retry_failed_mirror
&& scrub_is_page_on_raid56(page
)) {
1521 if (scrub_submit_raid56_bio_wait(fs_info
, bio
, page
))
1522 sblock
->no_io_error_seen
= 0;
1524 bio
->bi_iter
.bi_sector
= page
->physical
>> 9;
1526 if (btrfsic_submit_bio_wait(READ
, bio
))
1527 sblock
->no_io_error_seen
= 0;
1533 if (sblock
->no_io_error_seen
)
1534 scrub_recheck_block_checksum(fs_info
, sblock
, is_metadata
,
1535 have_csum
, csum
, generation
,
1541 static inline int scrub_check_fsid(u8 fsid
[],
1542 struct scrub_page
*spage
)
1544 struct btrfs_fs_devices
*fs_devices
= spage
->dev
->fs_devices
;
1547 ret
= memcmp(fsid
, fs_devices
->fsid
, BTRFS_UUID_SIZE
);
1551 static void scrub_recheck_block_checksum(struct btrfs_fs_info
*fs_info
,
1552 struct scrub_block
*sblock
,
1553 int is_metadata
, int have_csum
,
1554 const u8
*csum
, u64 generation
,
1558 u8 calculated_csum
[BTRFS_CSUM_SIZE
];
1560 void *mapped_buffer
;
1562 WARN_ON(!sblock
->pagev
[0]->page
);
1564 struct btrfs_header
*h
;
1566 mapped_buffer
= kmap_atomic(sblock
->pagev
[0]->page
);
1567 h
= (struct btrfs_header
*)mapped_buffer
;
1569 if (sblock
->pagev
[0]->logical
!= btrfs_stack_header_bytenr(h
) ||
1570 !scrub_check_fsid(h
->fsid
, sblock
->pagev
[0]) ||
1571 memcmp(h
->chunk_tree_uuid
, fs_info
->chunk_tree_uuid
,
1573 sblock
->header_error
= 1;
1574 } else if (generation
!= btrfs_stack_header_generation(h
)) {
1575 sblock
->header_error
= 1;
1576 sblock
->generation_error
= 1;
1583 mapped_buffer
= kmap_atomic(sblock
->pagev
[0]->page
);
1586 for (page_num
= 0;;) {
1587 if (page_num
== 0 && is_metadata
)
1588 crc
= btrfs_csum_data(
1589 ((u8
*)mapped_buffer
) + BTRFS_CSUM_SIZE
,
1590 crc
, PAGE_SIZE
- BTRFS_CSUM_SIZE
);
1592 crc
= btrfs_csum_data(mapped_buffer
, crc
, PAGE_SIZE
);
1594 kunmap_atomic(mapped_buffer
);
1596 if (page_num
>= sblock
->page_count
)
1598 WARN_ON(!sblock
->pagev
[page_num
]->page
);
1600 mapped_buffer
= kmap_atomic(sblock
->pagev
[page_num
]->page
);
1603 btrfs_csum_final(crc
, calculated_csum
);
1604 if (memcmp(calculated_csum
, csum
, csum_size
))
1605 sblock
->checksum_error
= 1;
1608 static int scrub_repair_block_from_good_copy(struct scrub_block
*sblock_bad
,
1609 struct scrub_block
*sblock_good
)
1614 for (page_num
= 0; page_num
< sblock_bad
->page_count
; page_num
++) {
1617 ret_sub
= scrub_repair_page_from_good_copy(sblock_bad
,
1627 static int scrub_repair_page_from_good_copy(struct scrub_block
*sblock_bad
,
1628 struct scrub_block
*sblock_good
,
1629 int page_num
, int force_write
)
1631 struct scrub_page
*page_bad
= sblock_bad
->pagev
[page_num
];
1632 struct scrub_page
*page_good
= sblock_good
->pagev
[page_num
];
1634 BUG_ON(page_bad
->page
== NULL
);
1635 BUG_ON(page_good
->page
== NULL
);
1636 if (force_write
|| sblock_bad
->header_error
||
1637 sblock_bad
->checksum_error
|| page_bad
->io_error
) {
1641 if (!page_bad
->dev
->bdev
) {
1642 btrfs_warn_rl(sblock_bad
->sctx
->dev_root
->fs_info
,
1643 "scrub_repair_page_from_good_copy(bdev == NULL) "
1648 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
1651 bio
->bi_bdev
= page_bad
->dev
->bdev
;
1652 bio
->bi_iter
.bi_sector
= page_bad
->physical
>> 9;
1654 ret
= bio_add_page(bio
, page_good
->page
, PAGE_SIZE
, 0);
1655 if (PAGE_SIZE
!= ret
) {
1660 if (btrfsic_submit_bio_wait(WRITE
, bio
)) {
1661 btrfs_dev_stat_inc_and_print(page_bad
->dev
,
1662 BTRFS_DEV_STAT_WRITE_ERRS
);
1663 btrfs_dev_replace_stats_inc(
1664 &sblock_bad
->sctx
->dev_root
->fs_info
->
1665 dev_replace
.num_write_errors
);
1675 static void scrub_write_block_to_dev_replace(struct scrub_block
*sblock
)
1680 * This block is used for the check of the parity on the source device,
1681 * so the data needn't be written into the destination device.
1683 if (sblock
->sparity
)
1686 for (page_num
= 0; page_num
< sblock
->page_count
; page_num
++) {
1689 ret
= scrub_write_page_to_dev_replace(sblock
, page_num
);
1691 btrfs_dev_replace_stats_inc(
1692 &sblock
->sctx
->dev_root
->fs_info
->dev_replace
.
1697 static int scrub_write_page_to_dev_replace(struct scrub_block
*sblock
,
1700 struct scrub_page
*spage
= sblock
->pagev
[page_num
];
1702 BUG_ON(spage
->page
== NULL
);
1703 if (spage
->io_error
) {
1704 void *mapped_buffer
= kmap_atomic(spage
->page
);
1706 memset(mapped_buffer
, 0, PAGE_CACHE_SIZE
);
1707 flush_dcache_page(spage
->page
);
1708 kunmap_atomic(mapped_buffer
);
1710 return scrub_add_page_to_wr_bio(sblock
->sctx
, spage
);
1713 static int scrub_add_page_to_wr_bio(struct scrub_ctx
*sctx
,
1714 struct scrub_page
*spage
)
1716 struct scrub_wr_ctx
*wr_ctx
= &sctx
->wr_ctx
;
1717 struct scrub_bio
*sbio
;
1720 mutex_lock(&wr_ctx
->wr_lock
);
1722 if (!wr_ctx
->wr_curr_bio
) {
1723 wr_ctx
->wr_curr_bio
= kzalloc(sizeof(*wr_ctx
->wr_curr_bio
),
1725 if (!wr_ctx
->wr_curr_bio
) {
1726 mutex_unlock(&wr_ctx
->wr_lock
);
1729 wr_ctx
->wr_curr_bio
->sctx
= sctx
;
1730 wr_ctx
->wr_curr_bio
->page_count
= 0;
1732 sbio
= wr_ctx
->wr_curr_bio
;
1733 if (sbio
->page_count
== 0) {
1736 sbio
->physical
= spage
->physical_for_dev_replace
;
1737 sbio
->logical
= spage
->logical
;
1738 sbio
->dev
= wr_ctx
->tgtdev
;
1741 bio
= btrfs_io_bio_alloc(GFP_NOFS
, wr_ctx
->pages_per_wr_bio
);
1743 mutex_unlock(&wr_ctx
->wr_lock
);
1749 bio
->bi_private
= sbio
;
1750 bio
->bi_end_io
= scrub_wr_bio_end_io
;
1751 bio
->bi_bdev
= sbio
->dev
->bdev
;
1752 bio
->bi_iter
.bi_sector
= sbio
->physical
>> 9;
1754 } else if (sbio
->physical
+ sbio
->page_count
* PAGE_SIZE
!=
1755 spage
->physical_for_dev_replace
||
1756 sbio
->logical
+ sbio
->page_count
* PAGE_SIZE
!=
1758 scrub_wr_submit(sctx
);
1762 ret
= bio_add_page(sbio
->bio
, spage
->page
, PAGE_SIZE
, 0);
1763 if (ret
!= PAGE_SIZE
) {
1764 if (sbio
->page_count
< 1) {
1767 mutex_unlock(&wr_ctx
->wr_lock
);
1770 scrub_wr_submit(sctx
);
1774 sbio
->pagev
[sbio
->page_count
] = spage
;
1775 scrub_page_get(spage
);
1777 if (sbio
->page_count
== wr_ctx
->pages_per_wr_bio
)
1778 scrub_wr_submit(sctx
);
1779 mutex_unlock(&wr_ctx
->wr_lock
);
1784 static void scrub_wr_submit(struct scrub_ctx
*sctx
)
1786 struct scrub_wr_ctx
*wr_ctx
= &sctx
->wr_ctx
;
1787 struct scrub_bio
*sbio
;
1789 if (!wr_ctx
->wr_curr_bio
)
1792 sbio
= wr_ctx
->wr_curr_bio
;
1793 wr_ctx
->wr_curr_bio
= NULL
;
1794 WARN_ON(!sbio
->bio
->bi_bdev
);
1795 scrub_pending_bio_inc(sctx
);
1796 /* process all writes in a single worker thread. Then the block layer
1797 * orders the requests before sending them to the driver which
1798 * doubled the write performance on spinning disks when measured
1800 btrfsic_submit_bio(WRITE
, sbio
->bio
);
1803 static void scrub_wr_bio_end_io(struct bio
*bio
)
1805 struct scrub_bio
*sbio
= bio
->bi_private
;
1806 struct btrfs_fs_info
*fs_info
= sbio
->dev
->dev_root
->fs_info
;
1808 sbio
->err
= bio
->bi_error
;
1811 btrfs_init_work(&sbio
->work
, btrfs_scrubwrc_helper
,
1812 scrub_wr_bio_end_io_worker
, NULL
, NULL
);
1813 btrfs_queue_work(fs_info
->scrub_wr_completion_workers
, &sbio
->work
);
1816 static void scrub_wr_bio_end_io_worker(struct btrfs_work
*work
)
1818 struct scrub_bio
*sbio
= container_of(work
, struct scrub_bio
, work
);
1819 struct scrub_ctx
*sctx
= sbio
->sctx
;
1822 WARN_ON(sbio
->page_count
> SCRUB_PAGES_PER_WR_BIO
);
1824 struct btrfs_dev_replace
*dev_replace
=
1825 &sbio
->sctx
->dev_root
->fs_info
->dev_replace
;
1827 for (i
= 0; i
< sbio
->page_count
; i
++) {
1828 struct scrub_page
*spage
= sbio
->pagev
[i
];
1830 spage
->io_error
= 1;
1831 btrfs_dev_replace_stats_inc(&dev_replace
->
1836 for (i
= 0; i
< sbio
->page_count
; i
++)
1837 scrub_page_put(sbio
->pagev
[i
]);
1841 scrub_pending_bio_dec(sctx
);
1844 static int scrub_checksum(struct scrub_block
*sblock
)
1849 WARN_ON(sblock
->page_count
< 1);
1850 flags
= sblock
->pagev
[0]->flags
;
1852 if (flags
& BTRFS_EXTENT_FLAG_DATA
)
1853 ret
= scrub_checksum_data(sblock
);
1854 else if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
)
1855 ret
= scrub_checksum_tree_block(sblock
);
1856 else if (flags
& BTRFS_EXTENT_FLAG_SUPER
)
1857 (void)scrub_checksum_super(sblock
);
1861 scrub_handle_errored_block(sblock
);
1866 static int scrub_checksum_data(struct scrub_block
*sblock
)
1868 struct scrub_ctx
*sctx
= sblock
->sctx
;
1869 u8 csum
[BTRFS_CSUM_SIZE
];
1878 BUG_ON(sblock
->page_count
< 1);
1879 if (!sblock
->pagev
[0]->have_csum
)
1882 on_disk_csum
= sblock
->pagev
[0]->csum
;
1883 page
= sblock
->pagev
[0]->page
;
1884 buffer
= kmap_atomic(page
);
1886 len
= sctx
->sectorsize
;
1889 u64 l
= min_t(u64
, len
, PAGE_SIZE
);
1891 crc
= btrfs_csum_data(buffer
, crc
, l
);
1892 kunmap_atomic(buffer
);
1897 BUG_ON(index
>= sblock
->page_count
);
1898 BUG_ON(!sblock
->pagev
[index
]->page
);
1899 page
= sblock
->pagev
[index
]->page
;
1900 buffer
= kmap_atomic(page
);
1903 btrfs_csum_final(crc
, csum
);
1904 if (memcmp(csum
, on_disk_csum
, sctx
->csum_size
))
1910 static int scrub_checksum_tree_block(struct scrub_block
*sblock
)
1912 struct scrub_ctx
*sctx
= sblock
->sctx
;
1913 struct btrfs_header
*h
;
1914 struct btrfs_root
*root
= sctx
->dev_root
;
1915 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1916 u8 calculated_csum
[BTRFS_CSUM_SIZE
];
1917 u8 on_disk_csum
[BTRFS_CSUM_SIZE
];
1919 void *mapped_buffer
;
1928 BUG_ON(sblock
->page_count
< 1);
1929 page
= sblock
->pagev
[0]->page
;
1930 mapped_buffer
= kmap_atomic(page
);
1931 h
= (struct btrfs_header
*)mapped_buffer
;
1932 memcpy(on_disk_csum
, h
->csum
, sctx
->csum_size
);
1935 * we don't use the getter functions here, as we
1936 * a) don't have an extent buffer and
1937 * b) the page is already kmapped
1940 if (sblock
->pagev
[0]->logical
!= btrfs_stack_header_bytenr(h
))
1943 if (sblock
->pagev
[0]->generation
!= btrfs_stack_header_generation(h
))
1946 if (!scrub_check_fsid(h
->fsid
, sblock
->pagev
[0]))
1949 if (memcmp(h
->chunk_tree_uuid
, fs_info
->chunk_tree_uuid
,
1953 len
= sctx
->nodesize
- BTRFS_CSUM_SIZE
;
1954 mapped_size
= PAGE_SIZE
- BTRFS_CSUM_SIZE
;
1955 p
= ((u8
*)mapped_buffer
) + BTRFS_CSUM_SIZE
;
1958 u64 l
= min_t(u64
, len
, mapped_size
);
1960 crc
= btrfs_csum_data(p
, crc
, l
);
1961 kunmap_atomic(mapped_buffer
);
1966 BUG_ON(index
>= sblock
->page_count
);
1967 BUG_ON(!sblock
->pagev
[index
]->page
);
1968 page
= sblock
->pagev
[index
]->page
;
1969 mapped_buffer
= kmap_atomic(page
);
1970 mapped_size
= PAGE_SIZE
;
1974 btrfs_csum_final(crc
, calculated_csum
);
1975 if (memcmp(calculated_csum
, on_disk_csum
, sctx
->csum_size
))
1978 return fail
|| crc_fail
;
1981 static int scrub_checksum_super(struct scrub_block
*sblock
)
1983 struct btrfs_super_block
*s
;
1984 struct scrub_ctx
*sctx
= sblock
->sctx
;
1985 u8 calculated_csum
[BTRFS_CSUM_SIZE
];
1986 u8 on_disk_csum
[BTRFS_CSUM_SIZE
];
1988 void *mapped_buffer
;
1997 BUG_ON(sblock
->page_count
< 1);
1998 page
= sblock
->pagev
[0]->page
;
1999 mapped_buffer
= kmap_atomic(page
);
2000 s
= (struct btrfs_super_block
*)mapped_buffer
;
2001 memcpy(on_disk_csum
, s
->csum
, sctx
->csum_size
);
2003 if (sblock
->pagev
[0]->logical
!= btrfs_super_bytenr(s
))
2006 if (sblock
->pagev
[0]->generation
!= btrfs_super_generation(s
))
2009 if (!scrub_check_fsid(s
->fsid
, sblock
->pagev
[0]))
2012 len
= BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
;
2013 mapped_size
= PAGE_SIZE
- BTRFS_CSUM_SIZE
;
2014 p
= ((u8
*)mapped_buffer
) + BTRFS_CSUM_SIZE
;
2017 u64 l
= min_t(u64
, len
, mapped_size
);
2019 crc
= btrfs_csum_data(p
, crc
, l
);
2020 kunmap_atomic(mapped_buffer
);
2025 BUG_ON(index
>= sblock
->page_count
);
2026 BUG_ON(!sblock
->pagev
[index
]->page
);
2027 page
= sblock
->pagev
[index
]->page
;
2028 mapped_buffer
= kmap_atomic(page
);
2029 mapped_size
= PAGE_SIZE
;
2033 btrfs_csum_final(crc
, calculated_csum
);
2034 if (memcmp(calculated_csum
, on_disk_csum
, sctx
->csum_size
))
2037 if (fail_cor
+ fail_gen
) {
2039 * if we find an error in a super block, we just report it.
2040 * They will get written with the next transaction commit
2043 spin_lock(&sctx
->stat_lock
);
2044 ++sctx
->stat
.super_errors
;
2045 spin_unlock(&sctx
->stat_lock
);
2047 btrfs_dev_stat_inc_and_print(sblock
->pagev
[0]->dev
,
2048 BTRFS_DEV_STAT_CORRUPTION_ERRS
);
2050 btrfs_dev_stat_inc_and_print(sblock
->pagev
[0]->dev
,
2051 BTRFS_DEV_STAT_GENERATION_ERRS
);
2054 return fail_cor
+ fail_gen
;
2057 static void scrub_block_get(struct scrub_block
*sblock
)
2059 atomic_inc(&sblock
->refs
);
2062 static void scrub_block_put(struct scrub_block
*sblock
)
2064 if (atomic_dec_and_test(&sblock
->refs
)) {
2067 if (sblock
->sparity
)
2068 scrub_parity_put(sblock
->sparity
);
2070 for (i
= 0; i
< sblock
->page_count
; i
++)
2071 scrub_page_put(sblock
->pagev
[i
]);
2076 static void scrub_page_get(struct scrub_page
*spage
)
2078 atomic_inc(&spage
->refs
);
2081 static void scrub_page_put(struct scrub_page
*spage
)
2083 if (atomic_dec_and_test(&spage
->refs
)) {
2085 __free_page(spage
->page
);
2090 static void scrub_submit(struct scrub_ctx
*sctx
)
2092 struct scrub_bio
*sbio
;
2094 if (sctx
->curr
== -1)
2097 sbio
= sctx
->bios
[sctx
->curr
];
2099 scrub_pending_bio_inc(sctx
);
2100 btrfsic_submit_bio(READ
, sbio
->bio
);
2103 static int scrub_add_page_to_rd_bio(struct scrub_ctx
*sctx
,
2104 struct scrub_page
*spage
)
2106 struct scrub_block
*sblock
= spage
->sblock
;
2107 struct scrub_bio
*sbio
;
2112 * grab a fresh bio or wait for one to become available
2114 while (sctx
->curr
== -1) {
2115 spin_lock(&sctx
->list_lock
);
2116 sctx
->curr
= sctx
->first_free
;
2117 if (sctx
->curr
!= -1) {
2118 sctx
->first_free
= sctx
->bios
[sctx
->curr
]->next_free
;
2119 sctx
->bios
[sctx
->curr
]->next_free
= -1;
2120 sctx
->bios
[sctx
->curr
]->page_count
= 0;
2121 spin_unlock(&sctx
->list_lock
);
2123 spin_unlock(&sctx
->list_lock
);
2124 wait_event(sctx
->list_wait
, sctx
->first_free
!= -1);
2127 sbio
= sctx
->bios
[sctx
->curr
];
2128 if (sbio
->page_count
== 0) {
2131 sbio
->physical
= spage
->physical
;
2132 sbio
->logical
= spage
->logical
;
2133 sbio
->dev
= spage
->dev
;
2136 bio
= btrfs_io_bio_alloc(GFP_NOFS
, sctx
->pages_per_rd_bio
);
2142 bio
->bi_private
= sbio
;
2143 bio
->bi_end_io
= scrub_bio_end_io
;
2144 bio
->bi_bdev
= sbio
->dev
->bdev
;
2145 bio
->bi_iter
.bi_sector
= sbio
->physical
>> 9;
2147 } else if (sbio
->physical
+ sbio
->page_count
* PAGE_SIZE
!=
2149 sbio
->logical
+ sbio
->page_count
* PAGE_SIZE
!=
2151 sbio
->dev
!= spage
->dev
) {
2156 sbio
->pagev
[sbio
->page_count
] = spage
;
2157 ret
= bio_add_page(sbio
->bio
, spage
->page
, PAGE_SIZE
, 0);
2158 if (ret
!= PAGE_SIZE
) {
2159 if (sbio
->page_count
< 1) {
2168 scrub_block_get(sblock
); /* one for the page added to the bio */
2169 atomic_inc(&sblock
->outstanding_pages
);
2171 if (sbio
->page_count
== sctx
->pages_per_rd_bio
)
2177 static void scrub_missing_raid56_end_io(struct bio
*bio
)
2179 struct scrub_block
*sblock
= bio
->bi_private
;
2180 struct btrfs_fs_info
*fs_info
= sblock
->sctx
->dev_root
->fs_info
;
2183 sblock
->no_io_error_seen
= 0;
2185 btrfs_queue_work(fs_info
->scrub_workers
, &sblock
->work
);
2188 static void scrub_missing_raid56_worker(struct btrfs_work
*work
)
2190 struct scrub_block
*sblock
= container_of(work
, struct scrub_block
, work
);
2191 struct scrub_ctx
*sctx
= sblock
->sctx
;
2192 struct btrfs_fs_info
*fs_info
= sctx
->dev_root
->fs_info
;
2193 unsigned int is_metadata
;
2194 unsigned int have_csum
;
2198 struct btrfs_device
*dev
;
2200 is_metadata
= !(sblock
->pagev
[0]->flags
& BTRFS_EXTENT_FLAG_DATA
);
2201 have_csum
= sblock
->pagev
[0]->have_csum
;
2202 csum
= sblock
->pagev
[0]->csum
;
2203 generation
= sblock
->pagev
[0]->generation
;
2204 logical
= sblock
->pagev
[0]->logical
;
2205 dev
= sblock
->pagev
[0]->dev
;
2207 sblock
->header_error
= 0;
2208 sblock
->checksum_error
= 0;
2209 sblock
->generation_error
= 0;
2210 if (sblock
->no_io_error_seen
) {
2211 scrub_recheck_block_checksum(fs_info
, sblock
, is_metadata
,
2212 have_csum
, csum
, generation
,
2216 if (!sblock
->no_io_error_seen
) {
2217 spin_lock(&sctx
->stat_lock
);
2218 sctx
->stat
.read_errors
++;
2219 spin_unlock(&sctx
->stat_lock
);
2220 btrfs_err_rl_in_rcu(fs_info
,
2221 "IO error rebuilding logical %llu for dev %s",
2222 logical
, rcu_str_deref(dev
->name
));
2223 } else if (sblock
->header_error
|| sblock
->checksum_error
) {
2224 spin_lock(&sctx
->stat_lock
);
2225 sctx
->stat
.uncorrectable_errors
++;
2226 spin_unlock(&sctx
->stat_lock
);
2227 btrfs_err_rl_in_rcu(fs_info
,
2228 "failed to rebuild valid logical %llu for dev %s",
2229 logical
, rcu_str_deref(dev
->name
));
2231 scrub_write_block_to_dev_replace(sblock
);
2234 scrub_block_put(sblock
);
2236 if (sctx
->is_dev_replace
&&
2237 atomic_read(&sctx
->wr_ctx
.flush_all_writes
)) {
2238 mutex_lock(&sctx
->wr_ctx
.wr_lock
);
2239 scrub_wr_submit(sctx
);
2240 mutex_unlock(&sctx
->wr_ctx
.wr_lock
);
2243 scrub_pending_bio_dec(sctx
);
2246 static void scrub_missing_raid56_pages(struct scrub_block
*sblock
)
2248 struct scrub_ctx
*sctx
= sblock
->sctx
;
2249 struct btrfs_fs_info
*fs_info
= sctx
->dev_root
->fs_info
;
2250 u64 length
= sblock
->page_count
* PAGE_SIZE
;
2251 u64 logical
= sblock
->pagev
[0]->logical
;
2252 struct btrfs_bio
*bbio
;
2254 struct btrfs_raid_bio
*rbio
;
2258 ret
= btrfs_map_sblock(fs_info
, REQ_GET_READ_MIRRORS
, logical
, &length
,
2260 if (ret
|| !bbio
|| !bbio
->raid_map
)
2263 if (WARN_ON(!sctx
->is_dev_replace
||
2264 !(bbio
->map_type
& BTRFS_BLOCK_GROUP_RAID56_MASK
))) {
2266 * We shouldn't be scrubbing a missing device. Even for dev
2267 * replace, we should only get here for RAID 5/6. We either
2268 * managed to mount something with no mirrors remaining or
2269 * there's a bug in scrub_remap_extent()/btrfs_map_block().
2274 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 0);
2278 bio
->bi_iter
.bi_sector
= logical
>> 9;
2279 bio
->bi_private
= sblock
;
2280 bio
->bi_end_io
= scrub_missing_raid56_end_io
;
2282 rbio
= raid56_alloc_missing_rbio(sctx
->dev_root
, bio
, bbio
, length
);
2286 for (i
= 0; i
< sblock
->page_count
; i
++) {
2287 struct scrub_page
*spage
= sblock
->pagev
[i
];
2289 raid56_add_scrub_pages(rbio
, spage
->page
, spage
->logical
);
2292 btrfs_init_work(&sblock
->work
, btrfs_scrub_helper
,
2293 scrub_missing_raid56_worker
, NULL
, NULL
);
2294 scrub_block_get(sblock
);
2295 scrub_pending_bio_inc(sctx
);
2296 raid56_submit_missing_rbio(rbio
);
2302 btrfs_put_bbio(bbio
);
2303 spin_lock(&sctx
->stat_lock
);
2304 sctx
->stat
.malloc_errors
++;
2305 spin_unlock(&sctx
->stat_lock
);
2308 static int scrub_pages(struct scrub_ctx
*sctx
, u64 logical
, u64 len
,
2309 u64 physical
, struct btrfs_device
*dev
, u64 flags
,
2310 u64 gen
, int mirror_num
, u8
*csum
, int force
,
2311 u64 physical_for_dev_replace
)
2313 struct scrub_block
*sblock
;
2316 sblock
= kzalloc(sizeof(*sblock
), GFP_NOFS
);
2318 spin_lock(&sctx
->stat_lock
);
2319 sctx
->stat
.malloc_errors
++;
2320 spin_unlock(&sctx
->stat_lock
);
2324 /* one ref inside this function, plus one for each page added to
2326 atomic_set(&sblock
->refs
, 1);
2327 sblock
->sctx
= sctx
;
2328 sblock
->no_io_error_seen
= 1;
2330 for (index
= 0; len
> 0; index
++) {
2331 struct scrub_page
*spage
;
2332 u64 l
= min_t(u64
, len
, PAGE_SIZE
);
2334 spage
= kzalloc(sizeof(*spage
), GFP_NOFS
);
2337 spin_lock(&sctx
->stat_lock
);
2338 sctx
->stat
.malloc_errors
++;
2339 spin_unlock(&sctx
->stat_lock
);
2340 scrub_block_put(sblock
);
2343 BUG_ON(index
>= SCRUB_MAX_PAGES_PER_BLOCK
);
2344 scrub_page_get(spage
);
2345 sblock
->pagev
[index
] = spage
;
2346 spage
->sblock
= sblock
;
2348 spage
->flags
= flags
;
2349 spage
->generation
= gen
;
2350 spage
->logical
= logical
;
2351 spage
->physical
= physical
;
2352 spage
->physical_for_dev_replace
= physical_for_dev_replace
;
2353 spage
->mirror_num
= mirror_num
;
2355 spage
->have_csum
= 1;
2356 memcpy(spage
->csum
, csum
, sctx
->csum_size
);
2358 spage
->have_csum
= 0;
2360 sblock
->page_count
++;
2361 spage
->page
= alloc_page(GFP_NOFS
);
2367 physical_for_dev_replace
+= l
;
2370 WARN_ON(sblock
->page_count
== 0);
2373 * This case should only be hit for RAID 5/6 device replace. See
2374 * the comment in scrub_missing_raid56_pages() for details.
2376 scrub_missing_raid56_pages(sblock
);
2378 for (index
= 0; index
< sblock
->page_count
; index
++) {
2379 struct scrub_page
*spage
= sblock
->pagev
[index
];
2382 ret
= scrub_add_page_to_rd_bio(sctx
, spage
);
2384 scrub_block_put(sblock
);
2393 /* last one frees, either here or in bio completion for last page */
2394 scrub_block_put(sblock
);
2398 static void scrub_bio_end_io(struct bio
*bio
)
2400 struct scrub_bio
*sbio
= bio
->bi_private
;
2401 struct btrfs_fs_info
*fs_info
= sbio
->dev
->dev_root
->fs_info
;
2403 sbio
->err
= bio
->bi_error
;
2406 btrfs_queue_work(fs_info
->scrub_workers
, &sbio
->work
);
2409 static void scrub_bio_end_io_worker(struct btrfs_work
*work
)
2411 struct scrub_bio
*sbio
= container_of(work
, struct scrub_bio
, work
);
2412 struct scrub_ctx
*sctx
= sbio
->sctx
;
2415 BUG_ON(sbio
->page_count
> SCRUB_PAGES_PER_RD_BIO
);
2417 for (i
= 0; i
< sbio
->page_count
; i
++) {
2418 struct scrub_page
*spage
= sbio
->pagev
[i
];
2420 spage
->io_error
= 1;
2421 spage
->sblock
->no_io_error_seen
= 0;
2425 /* now complete the scrub_block items that have all pages completed */
2426 for (i
= 0; i
< sbio
->page_count
; i
++) {
2427 struct scrub_page
*spage
= sbio
->pagev
[i
];
2428 struct scrub_block
*sblock
= spage
->sblock
;
2430 if (atomic_dec_and_test(&sblock
->outstanding_pages
))
2431 scrub_block_complete(sblock
);
2432 scrub_block_put(sblock
);
2437 spin_lock(&sctx
->list_lock
);
2438 sbio
->next_free
= sctx
->first_free
;
2439 sctx
->first_free
= sbio
->index
;
2440 spin_unlock(&sctx
->list_lock
);
2442 if (sctx
->is_dev_replace
&&
2443 atomic_read(&sctx
->wr_ctx
.flush_all_writes
)) {
2444 mutex_lock(&sctx
->wr_ctx
.wr_lock
);
2445 scrub_wr_submit(sctx
);
2446 mutex_unlock(&sctx
->wr_ctx
.wr_lock
);
2449 scrub_pending_bio_dec(sctx
);
2452 static inline void __scrub_mark_bitmap(struct scrub_parity
*sparity
,
2453 unsigned long *bitmap
,
2458 int sectorsize
= sparity
->sctx
->dev_root
->sectorsize
;
2460 if (len
>= sparity
->stripe_len
) {
2461 bitmap_set(bitmap
, 0, sparity
->nsectors
);
2465 start
-= sparity
->logic_start
;
2466 start
= div_u64_rem(start
, sparity
->stripe_len
, &offset
);
2467 offset
/= sectorsize
;
2468 nsectors
= (int)len
/ sectorsize
;
2470 if (offset
+ nsectors
<= sparity
->nsectors
) {
2471 bitmap_set(bitmap
, offset
, nsectors
);
2475 bitmap_set(bitmap
, offset
, sparity
->nsectors
- offset
);
2476 bitmap_set(bitmap
, 0, nsectors
- (sparity
->nsectors
- offset
));
2479 static inline void scrub_parity_mark_sectors_error(struct scrub_parity
*sparity
,
2482 __scrub_mark_bitmap(sparity
, sparity
->ebitmap
, start
, len
);
2485 static inline void scrub_parity_mark_sectors_data(struct scrub_parity
*sparity
,
2488 __scrub_mark_bitmap(sparity
, sparity
->dbitmap
, start
, len
);
2491 static void scrub_block_complete(struct scrub_block
*sblock
)
2495 if (!sblock
->no_io_error_seen
) {
2497 scrub_handle_errored_block(sblock
);
2500 * if has checksum error, write via repair mechanism in
2501 * dev replace case, otherwise write here in dev replace
2504 corrupted
= scrub_checksum(sblock
);
2505 if (!corrupted
&& sblock
->sctx
->is_dev_replace
)
2506 scrub_write_block_to_dev_replace(sblock
);
2509 if (sblock
->sparity
&& corrupted
&& !sblock
->data_corrected
) {
2510 u64 start
= sblock
->pagev
[0]->logical
;
2511 u64 end
= sblock
->pagev
[sblock
->page_count
- 1]->logical
+
2514 scrub_parity_mark_sectors_error(sblock
->sparity
,
2515 start
, end
- start
);
2519 static int scrub_find_csum(struct scrub_ctx
*sctx
, u64 logical
, u64 len
,
2522 struct btrfs_ordered_sum
*sum
= NULL
;
2523 unsigned long index
;
2524 unsigned long num_sectors
;
2526 while (!list_empty(&sctx
->csum_list
)) {
2527 sum
= list_first_entry(&sctx
->csum_list
,
2528 struct btrfs_ordered_sum
, list
);
2529 if (sum
->bytenr
> logical
)
2531 if (sum
->bytenr
+ sum
->len
> logical
)
2534 ++sctx
->stat
.csum_discards
;
2535 list_del(&sum
->list
);
2542 index
= ((u32
)(logical
- sum
->bytenr
)) / sctx
->sectorsize
;
2543 num_sectors
= sum
->len
/ sctx
->sectorsize
;
2544 memcpy(csum
, sum
->sums
+ index
, sctx
->csum_size
);
2545 if (index
== num_sectors
- 1) {
2546 list_del(&sum
->list
);
2552 /* scrub extent tries to collect up to 64 kB for each bio */
2553 static int scrub_extent(struct scrub_ctx
*sctx
, u64 logical
, u64 len
,
2554 u64 physical
, struct btrfs_device
*dev
, u64 flags
,
2555 u64 gen
, int mirror_num
, u64 physical_for_dev_replace
)
2558 u8 csum
[BTRFS_CSUM_SIZE
];
2561 if (flags
& BTRFS_EXTENT_FLAG_DATA
) {
2562 blocksize
= sctx
->sectorsize
;
2563 spin_lock(&sctx
->stat_lock
);
2564 sctx
->stat
.data_extents_scrubbed
++;
2565 sctx
->stat
.data_bytes_scrubbed
+= len
;
2566 spin_unlock(&sctx
->stat_lock
);
2567 } else if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
2568 blocksize
= sctx
->nodesize
;
2569 spin_lock(&sctx
->stat_lock
);
2570 sctx
->stat
.tree_extents_scrubbed
++;
2571 sctx
->stat
.tree_bytes_scrubbed
+= len
;
2572 spin_unlock(&sctx
->stat_lock
);
2574 blocksize
= sctx
->sectorsize
;
2579 u64 l
= min_t(u64
, len
, blocksize
);
2582 if (flags
& BTRFS_EXTENT_FLAG_DATA
) {
2583 /* push csums to sbio */
2584 have_csum
= scrub_find_csum(sctx
, logical
, l
, csum
);
2586 ++sctx
->stat
.no_csum
;
2587 if (sctx
->is_dev_replace
&& !have_csum
) {
2588 ret
= copy_nocow_pages(sctx
, logical
, l
,
2590 physical_for_dev_replace
);
2591 goto behind_scrub_pages
;
2594 ret
= scrub_pages(sctx
, logical
, l
, physical
, dev
, flags
, gen
,
2595 mirror_num
, have_csum
? csum
: NULL
, 0,
2596 physical_for_dev_replace
);
2603 physical_for_dev_replace
+= l
;
2608 static int scrub_pages_for_parity(struct scrub_parity
*sparity
,
2609 u64 logical
, u64 len
,
2610 u64 physical
, struct btrfs_device
*dev
,
2611 u64 flags
, u64 gen
, int mirror_num
, u8
*csum
)
2613 struct scrub_ctx
*sctx
= sparity
->sctx
;
2614 struct scrub_block
*sblock
;
2617 sblock
= kzalloc(sizeof(*sblock
), GFP_NOFS
);
2619 spin_lock(&sctx
->stat_lock
);
2620 sctx
->stat
.malloc_errors
++;
2621 spin_unlock(&sctx
->stat_lock
);
2625 /* one ref inside this function, plus one for each page added to
2627 atomic_set(&sblock
->refs
, 1);
2628 sblock
->sctx
= sctx
;
2629 sblock
->no_io_error_seen
= 1;
2630 sblock
->sparity
= sparity
;
2631 scrub_parity_get(sparity
);
2633 for (index
= 0; len
> 0; index
++) {
2634 struct scrub_page
*spage
;
2635 u64 l
= min_t(u64
, len
, PAGE_SIZE
);
2637 spage
= kzalloc(sizeof(*spage
), GFP_NOFS
);
2640 spin_lock(&sctx
->stat_lock
);
2641 sctx
->stat
.malloc_errors
++;
2642 spin_unlock(&sctx
->stat_lock
);
2643 scrub_block_put(sblock
);
2646 BUG_ON(index
>= SCRUB_MAX_PAGES_PER_BLOCK
);
2647 /* For scrub block */
2648 scrub_page_get(spage
);
2649 sblock
->pagev
[index
] = spage
;
2650 /* For scrub parity */
2651 scrub_page_get(spage
);
2652 list_add_tail(&spage
->list
, &sparity
->spages
);
2653 spage
->sblock
= sblock
;
2655 spage
->flags
= flags
;
2656 spage
->generation
= gen
;
2657 spage
->logical
= logical
;
2658 spage
->physical
= physical
;
2659 spage
->mirror_num
= mirror_num
;
2661 spage
->have_csum
= 1;
2662 memcpy(spage
->csum
, csum
, sctx
->csum_size
);
2664 spage
->have_csum
= 0;
2666 sblock
->page_count
++;
2667 spage
->page
= alloc_page(GFP_NOFS
);
2675 WARN_ON(sblock
->page_count
== 0);
2676 for (index
= 0; index
< sblock
->page_count
; index
++) {
2677 struct scrub_page
*spage
= sblock
->pagev
[index
];
2680 ret
= scrub_add_page_to_rd_bio(sctx
, spage
);
2682 scrub_block_put(sblock
);
2687 /* last one frees, either here or in bio completion for last page */
2688 scrub_block_put(sblock
);
2692 static int scrub_extent_for_parity(struct scrub_parity
*sparity
,
2693 u64 logical
, u64 len
,
2694 u64 physical
, struct btrfs_device
*dev
,
2695 u64 flags
, u64 gen
, int mirror_num
)
2697 struct scrub_ctx
*sctx
= sparity
->sctx
;
2699 u8 csum
[BTRFS_CSUM_SIZE
];
2703 scrub_parity_mark_sectors_error(sparity
, logical
, len
);
2707 if (flags
& BTRFS_EXTENT_FLAG_DATA
) {
2708 blocksize
= sctx
->sectorsize
;
2709 } else if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
2710 blocksize
= sctx
->nodesize
;
2712 blocksize
= sctx
->sectorsize
;
2717 u64 l
= min_t(u64
, len
, blocksize
);
2720 if (flags
& BTRFS_EXTENT_FLAG_DATA
) {
2721 /* push csums to sbio */
2722 have_csum
= scrub_find_csum(sctx
, logical
, l
, csum
);
2726 ret
= scrub_pages_for_parity(sparity
, logical
, l
, physical
, dev
,
2727 flags
, gen
, mirror_num
,
2728 have_csum
? csum
: NULL
);
2740 * Given a physical address, this will calculate it's
2741 * logical offset. if this is a parity stripe, it will return
2742 * the most left data stripe's logical offset.
2744 * return 0 if it is a data stripe, 1 means parity stripe.
2746 static int get_raid56_logic_offset(u64 physical
, int num
,
2747 struct map_lookup
*map
, u64
*offset
,
2757 last_offset
= (physical
- map
->stripes
[num
].physical
) *
2758 nr_data_stripes(map
);
2760 *stripe_start
= last_offset
;
2762 *offset
= last_offset
;
2763 for (i
= 0; i
< nr_data_stripes(map
); i
++) {
2764 *offset
= last_offset
+ i
* map
->stripe_len
;
2766 stripe_nr
= div_u64(*offset
, map
->stripe_len
);
2767 stripe_nr
= div_u64(stripe_nr
, nr_data_stripes(map
));
2769 /* Work out the disk rotation on this stripe-set */
2770 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
, &rot
);
2771 /* calculate which stripe this data locates */
2773 stripe_index
= rot
% map
->num_stripes
;
2774 if (stripe_index
== num
)
2776 if (stripe_index
< num
)
2779 *offset
= last_offset
+ j
* map
->stripe_len
;
2783 static void scrub_free_parity(struct scrub_parity
*sparity
)
2785 struct scrub_ctx
*sctx
= sparity
->sctx
;
2786 struct scrub_page
*curr
, *next
;
2789 nbits
= bitmap_weight(sparity
->ebitmap
, sparity
->nsectors
);
2791 spin_lock(&sctx
->stat_lock
);
2792 sctx
->stat
.read_errors
+= nbits
;
2793 sctx
->stat
.uncorrectable_errors
+= nbits
;
2794 spin_unlock(&sctx
->stat_lock
);
2797 list_for_each_entry_safe(curr
, next
, &sparity
->spages
, list
) {
2798 list_del_init(&curr
->list
);
2799 scrub_page_put(curr
);
2805 static void scrub_parity_bio_endio_worker(struct btrfs_work
*work
)
2807 struct scrub_parity
*sparity
= container_of(work
, struct scrub_parity
,
2809 struct scrub_ctx
*sctx
= sparity
->sctx
;
2811 scrub_free_parity(sparity
);
2812 scrub_pending_bio_dec(sctx
);
2815 static void scrub_parity_bio_endio(struct bio
*bio
)
2817 struct scrub_parity
*sparity
= (struct scrub_parity
*)bio
->bi_private
;
2820 bitmap_or(sparity
->ebitmap
, sparity
->ebitmap
, sparity
->dbitmap
,
2825 btrfs_init_work(&sparity
->work
, btrfs_scrubparity_helper
,
2826 scrub_parity_bio_endio_worker
, NULL
, NULL
);
2827 btrfs_queue_work(sparity
->sctx
->dev_root
->fs_info
->scrub_parity_workers
,
2831 static void scrub_parity_check_and_repair(struct scrub_parity
*sparity
)
2833 struct scrub_ctx
*sctx
= sparity
->sctx
;
2835 struct btrfs_raid_bio
*rbio
;
2836 struct scrub_page
*spage
;
2837 struct btrfs_bio
*bbio
= NULL
;
2841 if (!bitmap_andnot(sparity
->dbitmap
, sparity
->dbitmap
, sparity
->ebitmap
,
2845 length
= sparity
->logic_end
- sparity
->logic_start
;
2846 ret
= btrfs_map_sblock(sctx
->dev_root
->fs_info
, WRITE
,
2847 sparity
->logic_start
,
2848 &length
, &bbio
, 0, 1);
2849 if (ret
|| !bbio
|| !bbio
->raid_map
)
2852 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 0);
2856 bio
->bi_iter
.bi_sector
= sparity
->logic_start
>> 9;
2857 bio
->bi_private
= sparity
;
2858 bio
->bi_end_io
= scrub_parity_bio_endio
;
2860 rbio
= raid56_parity_alloc_scrub_rbio(sctx
->dev_root
, bio
, bbio
,
2861 length
, sparity
->scrub_dev
,
2867 list_for_each_entry(spage
, &sparity
->spages
, list
)
2868 raid56_add_scrub_pages(rbio
, spage
->page
, spage
->logical
);
2870 scrub_pending_bio_inc(sctx
);
2871 raid56_parity_submit_scrub_rbio(rbio
);
2877 btrfs_put_bbio(bbio
);
2878 bitmap_or(sparity
->ebitmap
, sparity
->ebitmap
, sparity
->dbitmap
,
2880 spin_lock(&sctx
->stat_lock
);
2881 sctx
->stat
.malloc_errors
++;
2882 spin_unlock(&sctx
->stat_lock
);
2884 scrub_free_parity(sparity
);
2887 static inline int scrub_calc_parity_bitmap_len(int nsectors
)
2889 return DIV_ROUND_UP(nsectors
, BITS_PER_LONG
) * (BITS_PER_LONG
/ 8);
2892 static void scrub_parity_get(struct scrub_parity
*sparity
)
2894 atomic_inc(&sparity
->refs
);
2897 static void scrub_parity_put(struct scrub_parity
*sparity
)
2899 if (!atomic_dec_and_test(&sparity
->refs
))
2902 scrub_parity_check_and_repair(sparity
);
2905 static noinline_for_stack
int scrub_raid56_parity(struct scrub_ctx
*sctx
,
2906 struct map_lookup
*map
,
2907 struct btrfs_device
*sdev
,
2908 struct btrfs_path
*path
,
2912 struct btrfs_fs_info
*fs_info
= sctx
->dev_root
->fs_info
;
2913 struct btrfs_root
*root
= fs_info
->extent_root
;
2914 struct btrfs_root
*csum_root
= fs_info
->csum_root
;
2915 struct btrfs_extent_item
*extent
;
2916 struct btrfs_bio
*bbio
= NULL
;
2920 struct extent_buffer
*l
;
2921 struct btrfs_key key
;
2924 u64 extent_physical
;
2927 struct btrfs_device
*extent_dev
;
2928 struct scrub_parity
*sparity
;
2931 int extent_mirror_num
;
2934 nsectors
= map
->stripe_len
/ root
->sectorsize
;
2935 bitmap_len
= scrub_calc_parity_bitmap_len(nsectors
);
2936 sparity
= kzalloc(sizeof(struct scrub_parity
) + 2 * bitmap_len
,
2939 spin_lock(&sctx
->stat_lock
);
2940 sctx
->stat
.malloc_errors
++;
2941 spin_unlock(&sctx
->stat_lock
);
2945 sparity
->stripe_len
= map
->stripe_len
;
2946 sparity
->nsectors
= nsectors
;
2947 sparity
->sctx
= sctx
;
2948 sparity
->scrub_dev
= sdev
;
2949 sparity
->logic_start
= logic_start
;
2950 sparity
->logic_end
= logic_end
;
2951 atomic_set(&sparity
->refs
, 1);
2952 INIT_LIST_HEAD(&sparity
->spages
);
2953 sparity
->dbitmap
= sparity
->bitmap
;
2954 sparity
->ebitmap
= (void *)sparity
->bitmap
+ bitmap_len
;
2957 while (logic_start
< logic_end
) {
2958 if (btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
2959 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2961 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2962 key
.objectid
= logic_start
;
2963 key
.offset
= (u64
)-1;
2965 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2970 ret
= btrfs_previous_extent_item(root
, path
, 0);
2974 btrfs_release_path(path
);
2975 ret
= btrfs_search_slot(NULL
, root
, &key
,
2987 slot
= path
->slots
[0];
2988 if (slot
>= btrfs_header_nritems(l
)) {
2989 ret
= btrfs_next_leaf(root
, path
);
2998 btrfs_item_key_to_cpu(l
, &key
, slot
);
3000 if (key
.type
!= BTRFS_EXTENT_ITEM_KEY
&&
3001 key
.type
!= BTRFS_METADATA_ITEM_KEY
)
3004 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
3005 bytes
= root
->nodesize
;
3009 if (key
.objectid
+ bytes
<= logic_start
)
3012 if (key
.objectid
>= logic_end
) {
3017 while (key
.objectid
>= logic_start
+ map
->stripe_len
)
3018 logic_start
+= map
->stripe_len
;
3020 extent
= btrfs_item_ptr(l
, slot
,
3021 struct btrfs_extent_item
);
3022 flags
= btrfs_extent_flags(l
, extent
);
3023 generation
= btrfs_extent_generation(l
, extent
);
3025 if ((flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) &&
3026 (key
.objectid
< logic_start
||
3027 key
.objectid
+ bytes
>
3028 logic_start
+ map
->stripe_len
)) {
3029 btrfs_err(fs_info
, "scrub: tree block %llu spanning stripes, ignored. logical=%llu",
3030 key
.objectid
, logic_start
);
3031 spin_lock(&sctx
->stat_lock
);
3032 sctx
->stat
.uncorrectable_errors
++;
3033 spin_unlock(&sctx
->stat_lock
);
3037 extent_logical
= key
.objectid
;
3040 if (extent_logical
< logic_start
) {
3041 extent_len
-= logic_start
- extent_logical
;
3042 extent_logical
= logic_start
;
3045 if (extent_logical
+ extent_len
>
3046 logic_start
+ map
->stripe_len
)
3047 extent_len
= logic_start
+ map
->stripe_len
-
3050 scrub_parity_mark_sectors_data(sparity
, extent_logical
,
3053 mapped_length
= extent_len
;
3054 ret
= btrfs_map_block(fs_info
, READ
, extent_logical
,
3055 &mapped_length
, &bbio
, 0);
3057 if (!bbio
|| mapped_length
< extent_len
)
3061 btrfs_put_bbio(bbio
);
3064 extent_physical
= bbio
->stripes
[0].physical
;
3065 extent_mirror_num
= bbio
->mirror_num
;
3066 extent_dev
= bbio
->stripes
[0].dev
;
3067 btrfs_put_bbio(bbio
);
3069 ret
= btrfs_lookup_csums_range(csum_root
,
3071 extent_logical
+ extent_len
- 1,
3072 &sctx
->csum_list
, 1);
3076 ret
= scrub_extent_for_parity(sparity
, extent_logical
,
3083 scrub_free_csums(sctx
);
3088 if (extent_logical
+ extent_len
<
3089 key
.objectid
+ bytes
) {
3090 logic_start
+= map
->stripe_len
;
3092 if (logic_start
>= logic_end
) {
3097 if (logic_start
< key
.objectid
+ bytes
) {
3106 btrfs_release_path(path
);
3111 logic_start
+= map
->stripe_len
;
3115 scrub_parity_mark_sectors_error(sparity
, logic_start
,
3116 logic_end
- logic_start
);
3117 scrub_parity_put(sparity
);
3119 mutex_lock(&sctx
->wr_ctx
.wr_lock
);
3120 scrub_wr_submit(sctx
);
3121 mutex_unlock(&sctx
->wr_ctx
.wr_lock
);
3123 btrfs_release_path(path
);
3124 return ret
< 0 ? ret
: 0;
3127 static noinline_for_stack
int scrub_stripe(struct scrub_ctx
*sctx
,
3128 struct map_lookup
*map
,
3129 struct btrfs_device
*scrub_dev
,
3130 int num
, u64 base
, u64 length
,
3133 struct btrfs_path
*path
, *ppath
;
3134 struct btrfs_fs_info
*fs_info
= sctx
->dev_root
->fs_info
;
3135 struct btrfs_root
*root
= fs_info
->extent_root
;
3136 struct btrfs_root
*csum_root
= fs_info
->csum_root
;
3137 struct btrfs_extent_item
*extent
;
3138 struct blk_plug plug
;
3143 struct extent_buffer
*l
;
3144 struct btrfs_key key
;
3151 struct reada_control
*reada1
;
3152 struct reada_control
*reada2
;
3153 struct btrfs_key key_start
;
3154 struct btrfs_key key_end
;
3155 u64 increment
= map
->stripe_len
;
3158 u64 extent_physical
;
3162 struct btrfs_device
*extent_dev
;
3163 int extent_mirror_num
;
3166 physical
= map
->stripes
[num
].physical
;
3168 nstripes
= div_u64(length
, map
->stripe_len
);
3169 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
3170 offset
= map
->stripe_len
* num
;
3171 increment
= map
->stripe_len
* map
->num_stripes
;
3173 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
3174 int factor
= map
->num_stripes
/ map
->sub_stripes
;
3175 offset
= map
->stripe_len
* (num
/ map
->sub_stripes
);
3176 increment
= map
->stripe_len
* factor
;
3177 mirror_num
= num
% map
->sub_stripes
+ 1;
3178 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
3179 increment
= map
->stripe_len
;
3180 mirror_num
= num
% map
->num_stripes
+ 1;
3181 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
3182 increment
= map
->stripe_len
;
3183 mirror_num
= num
% map
->num_stripes
+ 1;
3184 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
3185 get_raid56_logic_offset(physical
, num
, map
, &offset
, NULL
);
3186 increment
= map
->stripe_len
* nr_data_stripes(map
);
3189 increment
= map
->stripe_len
;
3193 path
= btrfs_alloc_path();
3197 ppath
= btrfs_alloc_path();
3199 btrfs_free_path(path
);
3204 * work on commit root. The related disk blocks are static as
3205 * long as COW is applied. This means, it is save to rewrite
3206 * them to repair disk errors without any race conditions
3208 path
->search_commit_root
= 1;
3209 path
->skip_locking
= 1;
3211 ppath
->search_commit_root
= 1;
3212 ppath
->skip_locking
= 1;
3214 * trigger the readahead for extent tree csum tree and wait for
3215 * completion. During readahead, the scrub is officially paused
3216 * to not hold off transaction commits
3218 logical
= base
+ offset
;
3219 physical_end
= physical
+ nstripes
* map
->stripe_len
;
3220 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
3221 get_raid56_logic_offset(physical_end
, num
,
3222 map
, &logic_end
, NULL
);
3225 logic_end
= logical
+ increment
* nstripes
;
3227 wait_event(sctx
->list_wait
,
3228 atomic_read(&sctx
->bios_in_flight
) == 0);
3229 scrub_blocked_if_needed(fs_info
);
3231 /* FIXME it might be better to start readahead at commit root */
3232 key_start
.objectid
= logical
;
3233 key_start
.type
= BTRFS_EXTENT_ITEM_KEY
;
3234 key_start
.offset
= (u64
)0;
3235 key_end
.objectid
= logic_end
;
3236 key_end
.type
= BTRFS_METADATA_ITEM_KEY
;
3237 key_end
.offset
= (u64
)-1;
3238 reada1
= btrfs_reada_add(root
, &key_start
, &key_end
);
3240 key_start
.objectid
= BTRFS_EXTENT_CSUM_OBJECTID
;
3241 key_start
.type
= BTRFS_EXTENT_CSUM_KEY
;
3242 key_start
.offset
= logical
;
3243 key_end
.objectid
= BTRFS_EXTENT_CSUM_OBJECTID
;
3244 key_end
.type
= BTRFS_EXTENT_CSUM_KEY
;
3245 key_end
.offset
= logic_end
;
3246 reada2
= btrfs_reada_add(csum_root
, &key_start
, &key_end
);
3248 if (!IS_ERR(reada1
))
3249 btrfs_reada_wait(reada1
);
3250 if (!IS_ERR(reada2
))
3251 btrfs_reada_wait(reada2
);
3255 * collect all data csums for the stripe to avoid seeking during
3256 * the scrub. This might currently (crc32) end up to be about 1MB
3258 blk_start_plug(&plug
);
3261 * now find all extents for each stripe and scrub them
3264 while (physical
< physical_end
) {
3268 if (atomic_read(&fs_info
->scrub_cancel_req
) ||
3269 atomic_read(&sctx
->cancel_req
)) {
3274 * check to see if we have to pause
3276 if (atomic_read(&fs_info
->scrub_pause_req
)) {
3277 /* push queued extents */
3278 atomic_set(&sctx
->wr_ctx
.flush_all_writes
, 1);
3280 mutex_lock(&sctx
->wr_ctx
.wr_lock
);
3281 scrub_wr_submit(sctx
);
3282 mutex_unlock(&sctx
->wr_ctx
.wr_lock
);
3283 wait_event(sctx
->list_wait
,
3284 atomic_read(&sctx
->bios_in_flight
) == 0);
3285 atomic_set(&sctx
->wr_ctx
.flush_all_writes
, 0);
3286 scrub_blocked_if_needed(fs_info
);
3289 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
3290 ret
= get_raid56_logic_offset(physical
, num
, map
,
3295 /* it is parity strip */
3296 stripe_logical
+= base
;
3297 stripe_end
= stripe_logical
+ increment
;
3298 ret
= scrub_raid56_parity(sctx
, map
, scrub_dev
,
3299 ppath
, stripe_logical
,
3307 if (btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
3308 key
.type
= BTRFS_METADATA_ITEM_KEY
;
3310 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3311 key
.objectid
= logical
;
3312 key
.offset
= (u64
)-1;
3314 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3319 ret
= btrfs_previous_extent_item(root
, path
, 0);
3323 /* there's no smaller item, so stick with the
3325 btrfs_release_path(path
);
3326 ret
= btrfs_search_slot(NULL
, root
, &key
,
3338 slot
= path
->slots
[0];
3339 if (slot
>= btrfs_header_nritems(l
)) {
3340 ret
= btrfs_next_leaf(root
, path
);
3349 btrfs_item_key_to_cpu(l
, &key
, slot
);
3351 if (key
.type
!= BTRFS_EXTENT_ITEM_KEY
&&
3352 key
.type
!= BTRFS_METADATA_ITEM_KEY
)
3355 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
3356 bytes
= root
->nodesize
;
3360 if (key
.objectid
+ bytes
<= logical
)
3363 if (key
.objectid
>= logical
+ map
->stripe_len
) {
3364 /* out of this device extent */
3365 if (key
.objectid
>= logic_end
)
3370 extent
= btrfs_item_ptr(l
, slot
,
3371 struct btrfs_extent_item
);
3372 flags
= btrfs_extent_flags(l
, extent
);
3373 generation
= btrfs_extent_generation(l
, extent
);
3375 if ((flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) &&
3376 (key
.objectid
< logical
||
3377 key
.objectid
+ bytes
>
3378 logical
+ map
->stripe_len
)) {
3380 "scrub: tree block %llu spanning "
3381 "stripes, ignored. logical=%llu",
3382 key
.objectid
, logical
);
3383 spin_lock(&sctx
->stat_lock
);
3384 sctx
->stat
.uncorrectable_errors
++;
3385 spin_unlock(&sctx
->stat_lock
);
3390 extent_logical
= key
.objectid
;
3394 * trim extent to this stripe
3396 if (extent_logical
< logical
) {
3397 extent_len
-= logical
- extent_logical
;
3398 extent_logical
= logical
;
3400 if (extent_logical
+ extent_len
>
3401 logical
+ map
->stripe_len
) {
3402 extent_len
= logical
+ map
->stripe_len
-
3406 extent_physical
= extent_logical
- logical
+ physical
;
3407 extent_dev
= scrub_dev
;
3408 extent_mirror_num
= mirror_num
;
3410 scrub_remap_extent(fs_info
, extent_logical
,
3411 extent_len
, &extent_physical
,
3413 &extent_mirror_num
);
3415 ret
= btrfs_lookup_csums_range(csum_root
,
3419 &sctx
->csum_list
, 1);
3423 ret
= scrub_extent(sctx
, extent_logical
, extent_len
,
3424 extent_physical
, extent_dev
, flags
,
3425 generation
, extent_mirror_num
,
3426 extent_logical
- logical
+ physical
);
3428 scrub_free_csums(sctx
);
3433 if (extent_logical
+ extent_len
<
3434 key
.objectid
+ bytes
) {
3435 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
3437 * loop until we find next data stripe
3438 * or we have finished all stripes.
3441 physical
+= map
->stripe_len
;
3442 ret
= get_raid56_logic_offset(physical
,
3447 if (ret
&& physical
< physical_end
) {
3448 stripe_logical
+= base
;
3449 stripe_end
= stripe_logical
+
3451 ret
= scrub_raid56_parity(sctx
,
3452 map
, scrub_dev
, ppath
,
3460 physical
+= map
->stripe_len
;
3461 logical
+= increment
;
3463 if (logical
< key
.objectid
+ bytes
) {
3468 if (physical
>= physical_end
) {
3476 btrfs_release_path(path
);
3478 logical
+= increment
;
3479 physical
+= map
->stripe_len
;
3480 spin_lock(&sctx
->stat_lock
);
3482 sctx
->stat
.last_physical
= map
->stripes
[num
].physical
+
3485 sctx
->stat
.last_physical
= physical
;
3486 spin_unlock(&sctx
->stat_lock
);
3491 /* push queued extents */
3493 mutex_lock(&sctx
->wr_ctx
.wr_lock
);
3494 scrub_wr_submit(sctx
);
3495 mutex_unlock(&sctx
->wr_ctx
.wr_lock
);
3497 blk_finish_plug(&plug
);
3498 btrfs_free_path(path
);
3499 btrfs_free_path(ppath
);
3500 return ret
< 0 ? ret
: 0;
3503 static noinline_for_stack
int scrub_chunk(struct scrub_ctx
*sctx
,
3504 struct btrfs_device
*scrub_dev
,
3505 u64 chunk_offset
, u64 length
,
3506 u64 dev_offset
, int is_dev_replace
)
3508 struct btrfs_mapping_tree
*map_tree
=
3509 &sctx
->dev_root
->fs_info
->mapping_tree
;
3510 struct map_lookup
*map
;
3511 struct extent_map
*em
;
3515 read_lock(&map_tree
->map_tree
.lock
);
3516 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
3517 read_unlock(&map_tree
->map_tree
.lock
);
3522 map
= (struct map_lookup
*)em
->bdev
;
3523 if (em
->start
!= chunk_offset
)
3526 if (em
->len
< length
)
3529 for (i
= 0; i
< map
->num_stripes
; ++i
) {
3530 if (map
->stripes
[i
].dev
->bdev
== scrub_dev
->bdev
&&
3531 map
->stripes
[i
].physical
== dev_offset
) {
3532 ret
= scrub_stripe(sctx
, map
, scrub_dev
, i
,
3533 chunk_offset
, length
,
3540 free_extent_map(em
);
3545 static noinline_for_stack
3546 int scrub_enumerate_chunks(struct scrub_ctx
*sctx
,
3547 struct btrfs_device
*scrub_dev
, u64 start
, u64 end
,
3550 struct btrfs_dev_extent
*dev_extent
= NULL
;
3551 struct btrfs_path
*path
;
3552 struct btrfs_root
*root
= sctx
->dev_root
;
3553 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3558 struct extent_buffer
*l
;
3559 struct btrfs_key key
;
3560 struct btrfs_key found_key
;
3561 struct btrfs_block_group_cache
*cache
;
3562 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
3564 path
= btrfs_alloc_path();
3569 path
->search_commit_root
= 1;
3570 path
->skip_locking
= 1;
3572 key
.objectid
= scrub_dev
->devid
;
3574 key
.type
= BTRFS_DEV_EXTENT_KEY
;
3577 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3581 if (path
->slots
[0] >=
3582 btrfs_header_nritems(path
->nodes
[0])) {
3583 ret
= btrfs_next_leaf(root
, path
);
3596 slot
= path
->slots
[0];
3598 btrfs_item_key_to_cpu(l
, &found_key
, slot
);
3600 if (found_key
.objectid
!= scrub_dev
->devid
)
3603 if (found_key
.type
!= BTRFS_DEV_EXTENT_KEY
)
3606 if (found_key
.offset
>= end
)
3609 if (found_key
.offset
< key
.offset
)
3612 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
3613 length
= btrfs_dev_extent_length(l
, dev_extent
);
3615 if (found_key
.offset
+ length
<= start
)
3618 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
3621 * get a reference on the corresponding block group to prevent
3622 * the chunk from going away while we scrub it
3624 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3626 /* some chunks are removed but not committed to disk yet,
3627 * continue scrubbing */
3632 * we need call btrfs_inc_block_group_ro() with scrubs_paused,
3633 * to avoid deadlock caused by:
3634 * btrfs_inc_block_group_ro()
3635 * -> btrfs_wait_for_commit()
3636 * -> btrfs_commit_transaction()
3637 * -> btrfs_scrub_pause()
3639 scrub_pause_on(fs_info
);
3640 ret
= btrfs_inc_block_group_ro(root
, cache
);
3641 scrub_pause_off(fs_info
);
3643 btrfs_put_block_group(cache
);
3647 dev_replace
->cursor_right
= found_key
.offset
+ length
;
3648 dev_replace
->cursor_left
= found_key
.offset
;
3649 dev_replace
->item_needs_writeback
= 1;
3650 ret
= scrub_chunk(sctx
, scrub_dev
, chunk_offset
, length
,
3651 found_key
.offset
, is_dev_replace
);
3654 * flush, submit all pending read and write bios, afterwards
3656 * Note that in the dev replace case, a read request causes
3657 * write requests that are submitted in the read completion
3658 * worker. Therefore in the current situation, it is required
3659 * that all write requests are flushed, so that all read and
3660 * write requests are really completed when bios_in_flight
3663 atomic_set(&sctx
->wr_ctx
.flush_all_writes
, 1);
3665 mutex_lock(&sctx
->wr_ctx
.wr_lock
);
3666 scrub_wr_submit(sctx
);
3667 mutex_unlock(&sctx
->wr_ctx
.wr_lock
);
3669 wait_event(sctx
->list_wait
,
3670 atomic_read(&sctx
->bios_in_flight
) == 0);
3672 scrub_pause_on(fs_info
);
3675 * must be called before we decrease @scrub_paused.
3676 * make sure we don't block transaction commit while
3677 * we are waiting pending workers finished.
3679 wait_event(sctx
->list_wait
,
3680 atomic_read(&sctx
->workers_pending
) == 0);
3681 atomic_set(&sctx
->wr_ctx
.flush_all_writes
, 0);
3683 scrub_pause_off(fs_info
);
3685 btrfs_dec_block_group_ro(root
, cache
);
3687 btrfs_put_block_group(cache
);
3690 if (is_dev_replace
&&
3691 atomic64_read(&dev_replace
->num_write_errors
) > 0) {
3695 if (sctx
->stat
.malloc_errors
> 0) {
3700 dev_replace
->cursor_left
= dev_replace
->cursor_right
;
3701 dev_replace
->item_needs_writeback
= 1;
3703 key
.offset
= found_key
.offset
+ length
;
3704 btrfs_release_path(path
);
3707 btrfs_free_path(path
);
3712 static noinline_for_stack
int scrub_supers(struct scrub_ctx
*sctx
,
3713 struct btrfs_device
*scrub_dev
)
3719 struct btrfs_root
*root
= sctx
->dev_root
;
3721 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
3724 /* Seed devices of a new filesystem has their own generation. */
3725 if (scrub_dev
->fs_devices
!= root
->fs_info
->fs_devices
)
3726 gen
= scrub_dev
->generation
;
3728 gen
= root
->fs_info
->last_trans_committed
;
3730 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
3731 bytenr
= btrfs_sb_offset(i
);
3732 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>
3733 scrub_dev
->commit_total_bytes
)
3736 ret
= scrub_pages(sctx
, bytenr
, BTRFS_SUPER_INFO_SIZE
, bytenr
,
3737 scrub_dev
, BTRFS_EXTENT_FLAG_SUPER
, gen
, i
,
3742 wait_event(sctx
->list_wait
, atomic_read(&sctx
->bios_in_flight
) == 0);
3748 * get a reference count on fs_info->scrub_workers. start worker if necessary
3750 static noinline_for_stack
int scrub_workers_get(struct btrfs_fs_info
*fs_info
,
3753 unsigned int flags
= WQ_FREEZABLE
| WQ_UNBOUND
;
3754 int max_active
= fs_info
->thread_pool_size
;
3756 if (fs_info
->scrub_workers_refcnt
== 0) {
3758 fs_info
->scrub_workers
=
3759 btrfs_alloc_workqueue("btrfs-scrub", flags
,
3762 fs_info
->scrub_workers
=
3763 btrfs_alloc_workqueue("btrfs-scrub", flags
,
3765 if (!fs_info
->scrub_workers
)
3766 goto fail_scrub_workers
;
3768 fs_info
->scrub_wr_completion_workers
=
3769 btrfs_alloc_workqueue("btrfs-scrubwrc", flags
,
3771 if (!fs_info
->scrub_wr_completion_workers
)
3772 goto fail_scrub_wr_completion_workers
;
3774 fs_info
->scrub_nocow_workers
=
3775 btrfs_alloc_workqueue("btrfs-scrubnc", flags
, 1, 0);
3776 if (!fs_info
->scrub_nocow_workers
)
3777 goto fail_scrub_nocow_workers
;
3778 fs_info
->scrub_parity_workers
=
3779 btrfs_alloc_workqueue("btrfs-scrubparity", flags
,
3781 if (!fs_info
->scrub_parity_workers
)
3782 goto fail_scrub_parity_workers
;
3784 ++fs_info
->scrub_workers_refcnt
;
3787 fail_scrub_parity_workers
:
3788 btrfs_destroy_workqueue(fs_info
->scrub_nocow_workers
);
3789 fail_scrub_nocow_workers
:
3790 btrfs_destroy_workqueue(fs_info
->scrub_wr_completion_workers
);
3791 fail_scrub_wr_completion_workers
:
3792 btrfs_destroy_workqueue(fs_info
->scrub_workers
);
3797 static noinline_for_stack
void scrub_workers_put(struct btrfs_fs_info
*fs_info
)
3799 if (--fs_info
->scrub_workers_refcnt
== 0) {
3800 btrfs_destroy_workqueue(fs_info
->scrub_workers
);
3801 btrfs_destroy_workqueue(fs_info
->scrub_wr_completion_workers
);
3802 btrfs_destroy_workqueue(fs_info
->scrub_nocow_workers
);
3803 btrfs_destroy_workqueue(fs_info
->scrub_parity_workers
);
3805 WARN_ON(fs_info
->scrub_workers_refcnt
< 0);
3808 int btrfs_scrub_dev(struct btrfs_fs_info
*fs_info
, u64 devid
, u64 start
,
3809 u64 end
, struct btrfs_scrub_progress
*progress
,
3810 int readonly
, int is_dev_replace
)
3812 struct scrub_ctx
*sctx
;
3814 struct btrfs_device
*dev
;
3815 struct rcu_string
*name
;
3817 if (btrfs_fs_closing(fs_info
))
3820 if (fs_info
->chunk_root
->nodesize
> BTRFS_STRIPE_LEN
) {
3822 * in this case scrub is unable to calculate the checksum
3823 * the way scrub is implemented. Do not handle this
3824 * situation at all because it won't ever happen.
3827 "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails",
3828 fs_info
->chunk_root
->nodesize
, BTRFS_STRIPE_LEN
);
3832 if (fs_info
->chunk_root
->sectorsize
!= PAGE_SIZE
) {
3833 /* not supported for data w/o checksums */
3835 "scrub: size assumption sectorsize != PAGE_SIZE "
3836 "(%d != %lu) fails",
3837 fs_info
->chunk_root
->sectorsize
, PAGE_SIZE
);
3841 if (fs_info
->chunk_root
->nodesize
>
3842 PAGE_SIZE
* SCRUB_MAX_PAGES_PER_BLOCK
||
3843 fs_info
->chunk_root
->sectorsize
>
3844 PAGE_SIZE
* SCRUB_MAX_PAGES_PER_BLOCK
) {
3846 * would exhaust the array bounds of pagev member in
3847 * struct scrub_block
3849 btrfs_err(fs_info
, "scrub: size assumption nodesize and sectorsize "
3850 "<= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails",
3851 fs_info
->chunk_root
->nodesize
,
3852 SCRUB_MAX_PAGES_PER_BLOCK
,
3853 fs_info
->chunk_root
->sectorsize
,
3854 SCRUB_MAX_PAGES_PER_BLOCK
);
3859 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
3860 dev
= btrfs_find_device(fs_info
, devid
, NULL
, NULL
);
3861 if (!dev
|| (dev
->missing
&& !is_dev_replace
)) {
3862 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3866 if (!is_dev_replace
&& !readonly
&& !dev
->writeable
) {
3867 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3869 name
= rcu_dereference(dev
->name
);
3870 btrfs_err(fs_info
, "scrub: device %s is not writable",
3876 mutex_lock(&fs_info
->scrub_lock
);
3877 if (!dev
->in_fs_metadata
|| dev
->is_tgtdev_for_dev_replace
) {
3878 mutex_unlock(&fs_info
->scrub_lock
);
3879 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3883 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
3884 if (dev
->scrub_device
||
3886 btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
))) {
3887 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
3888 mutex_unlock(&fs_info
->scrub_lock
);
3889 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3890 return -EINPROGRESS
;
3892 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
3894 ret
= scrub_workers_get(fs_info
, is_dev_replace
);
3896 mutex_unlock(&fs_info
->scrub_lock
);
3897 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3901 sctx
= scrub_setup_ctx(dev
, is_dev_replace
);
3903 mutex_unlock(&fs_info
->scrub_lock
);
3904 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3905 scrub_workers_put(fs_info
);
3906 return PTR_ERR(sctx
);
3908 sctx
->readonly
= readonly
;
3909 dev
->scrub_device
= sctx
;
3910 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3913 * checking @scrub_pause_req here, we can avoid
3914 * race between committing transaction and scrubbing.
3916 __scrub_blocked_if_needed(fs_info
);
3917 atomic_inc(&fs_info
->scrubs_running
);
3918 mutex_unlock(&fs_info
->scrub_lock
);
3920 if (!is_dev_replace
) {
3922 * by holding device list mutex, we can
3923 * kick off writing super in log tree sync.
3925 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
3926 ret
= scrub_supers(sctx
, dev
);
3927 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
3931 ret
= scrub_enumerate_chunks(sctx
, dev
, start
, end
,
3934 wait_event(sctx
->list_wait
, atomic_read(&sctx
->bios_in_flight
) == 0);
3935 atomic_dec(&fs_info
->scrubs_running
);
3936 wake_up(&fs_info
->scrub_pause_wait
);
3938 wait_event(sctx
->list_wait
, atomic_read(&sctx
->workers_pending
) == 0);
3941 memcpy(progress
, &sctx
->stat
, sizeof(*progress
));
3943 mutex_lock(&fs_info
->scrub_lock
);
3944 dev
->scrub_device
= NULL
;
3945 scrub_workers_put(fs_info
);
3946 mutex_unlock(&fs_info
->scrub_lock
);
3948 scrub_put_ctx(sctx
);
3953 void btrfs_scrub_pause(struct btrfs_root
*root
)
3955 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3957 mutex_lock(&fs_info
->scrub_lock
);
3958 atomic_inc(&fs_info
->scrub_pause_req
);
3959 while (atomic_read(&fs_info
->scrubs_paused
) !=
3960 atomic_read(&fs_info
->scrubs_running
)) {
3961 mutex_unlock(&fs_info
->scrub_lock
);
3962 wait_event(fs_info
->scrub_pause_wait
,
3963 atomic_read(&fs_info
->scrubs_paused
) ==
3964 atomic_read(&fs_info
->scrubs_running
));
3965 mutex_lock(&fs_info
->scrub_lock
);
3967 mutex_unlock(&fs_info
->scrub_lock
);
3970 void btrfs_scrub_continue(struct btrfs_root
*root
)
3972 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3974 atomic_dec(&fs_info
->scrub_pause_req
);
3975 wake_up(&fs_info
->scrub_pause_wait
);
3978 int btrfs_scrub_cancel(struct btrfs_fs_info
*fs_info
)
3980 mutex_lock(&fs_info
->scrub_lock
);
3981 if (!atomic_read(&fs_info
->scrubs_running
)) {
3982 mutex_unlock(&fs_info
->scrub_lock
);
3986 atomic_inc(&fs_info
->scrub_cancel_req
);
3987 while (atomic_read(&fs_info
->scrubs_running
)) {
3988 mutex_unlock(&fs_info
->scrub_lock
);
3989 wait_event(fs_info
->scrub_pause_wait
,
3990 atomic_read(&fs_info
->scrubs_running
) == 0);
3991 mutex_lock(&fs_info
->scrub_lock
);
3993 atomic_dec(&fs_info
->scrub_cancel_req
);
3994 mutex_unlock(&fs_info
->scrub_lock
);
3999 int btrfs_scrub_cancel_dev(struct btrfs_fs_info
*fs_info
,
4000 struct btrfs_device
*dev
)
4002 struct scrub_ctx
*sctx
;
4004 mutex_lock(&fs_info
->scrub_lock
);
4005 sctx
= dev
->scrub_device
;
4007 mutex_unlock(&fs_info
->scrub_lock
);
4010 atomic_inc(&sctx
->cancel_req
);
4011 while (dev
->scrub_device
) {
4012 mutex_unlock(&fs_info
->scrub_lock
);
4013 wait_event(fs_info
->scrub_pause_wait
,
4014 dev
->scrub_device
== NULL
);
4015 mutex_lock(&fs_info
->scrub_lock
);
4017 mutex_unlock(&fs_info
->scrub_lock
);
4022 int btrfs_scrub_progress(struct btrfs_root
*root
, u64 devid
,
4023 struct btrfs_scrub_progress
*progress
)
4025 struct btrfs_device
*dev
;
4026 struct scrub_ctx
*sctx
= NULL
;
4028 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
4029 dev
= btrfs_find_device(root
->fs_info
, devid
, NULL
, NULL
);
4031 sctx
= dev
->scrub_device
;
4033 memcpy(progress
, &sctx
->stat
, sizeof(*progress
));
4034 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
4036 return dev
? (sctx
? 0 : -ENOTCONN
) : -ENODEV
;
4039 static void scrub_remap_extent(struct btrfs_fs_info
*fs_info
,
4040 u64 extent_logical
, u64 extent_len
,
4041 u64
*extent_physical
,
4042 struct btrfs_device
**extent_dev
,
4043 int *extent_mirror_num
)
4046 struct btrfs_bio
*bbio
= NULL
;
4049 mapped_length
= extent_len
;
4050 ret
= btrfs_map_block(fs_info
, READ
, extent_logical
,
4051 &mapped_length
, &bbio
, 0);
4052 if (ret
|| !bbio
|| mapped_length
< extent_len
||
4053 !bbio
->stripes
[0].dev
->bdev
) {
4054 btrfs_put_bbio(bbio
);
4058 *extent_physical
= bbio
->stripes
[0].physical
;
4059 *extent_mirror_num
= bbio
->mirror_num
;
4060 *extent_dev
= bbio
->stripes
[0].dev
;
4061 btrfs_put_bbio(bbio
);
4064 static int scrub_setup_wr_ctx(struct scrub_ctx
*sctx
,
4065 struct scrub_wr_ctx
*wr_ctx
,
4066 struct btrfs_fs_info
*fs_info
,
4067 struct btrfs_device
*dev
,
4070 WARN_ON(wr_ctx
->wr_curr_bio
!= NULL
);
4072 mutex_init(&wr_ctx
->wr_lock
);
4073 wr_ctx
->wr_curr_bio
= NULL
;
4074 if (!is_dev_replace
)
4077 WARN_ON(!dev
->bdev
);
4078 wr_ctx
->pages_per_wr_bio
= SCRUB_PAGES_PER_WR_BIO
;
4079 wr_ctx
->tgtdev
= dev
;
4080 atomic_set(&wr_ctx
->flush_all_writes
, 0);
4084 static void scrub_free_wr_ctx(struct scrub_wr_ctx
*wr_ctx
)
4086 mutex_lock(&wr_ctx
->wr_lock
);
4087 kfree(wr_ctx
->wr_curr_bio
);
4088 wr_ctx
->wr_curr_bio
= NULL
;
4089 mutex_unlock(&wr_ctx
->wr_lock
);
4092 static int copy_nocow_pages(struct scrub_ctx
*sctx
, u64 logical
, u64 len
,
4093 int mirror_num
, u64 physical_for_dev_replace
)
4095 struct scrub_copy_nocow_ctx
*nocow_ctx
;
4096 struct btrfs_fs_info
*fs_info
= sctx
->dev_root
->fs_info
;
4098 nocow_ctx
= kzalloc(sizeof(*nocow_ctx
), GFP_NOFS
);
4100 spin_lock(&sctx
->stat_lock
);
4101 sctx
->stat
.malloc_errors
++;
4102 spin_unlock(&sctx
->stat_lock
);
4106 scrub_pending_trans_workers_inc(sctx
);
4108 nocow_ctx
->sctx
= sctx
;
4109 nocow_ctx
->logical
= logical
;
4110 nocow_ctx
->len
= len
;
4111 nocow_ctx
->mirror_num
= mirror_num
;
4112 nocow_ctx
->physical_for_dev_replace
= physical_for_dev_replace
;
4113 btrfs_init_work(&nocow_ctx
->work
, btrfs_scrubnc_helper
,
4114 copy_nocow_pages_worker
, NULL
, NULL
);
4115 INIT_LIST_HEAD(&nocow_ctx
->inodes
);
4116 btrfs_queue_work(fs_info
->scrub_nocow_workers
,
4122 static int record_inode_for_nocow(u64 inum
, u64 offset
, u64 root
, void *ctx
)
4124 struct scrub_copy_nocow_ctx
*nocow_ctx
= ctx
;
4125 struct scrub_nocow_inode
*nocow_inode
;
4127 nocow_inode
= kzalloc(sizeof(*nocow_inode
), GFP_NOFS
);
4130 nocow_inode
->inum
= inum
;
4131 nocow_inode
->offset
= offset
;
4132 nocow_inode
->root
= root
;
4133 list_add_tail(&nocow_inode
->list
, &nocow_ctx
->inodes
);
4137 #define COPY_COMPLETE 1
4139 static void copy_nocow_pages_worker(struct btrfs_work
*work
)
4141 struct scrub_copy_nocow_ctx
*nocow_ctx
=
4142 container_of(work
, struct scrub_copy_nocow_ctx
, work
);
4143 struct scrub_ctx
*sctx
= nocow_ctx
->sctx
;
4144 u64 logical
= nocow_ctx
->logical
;
4145 u64 len
= nocow_ctx
->len
;
4146 int mirror_num
= nocow_ctx
->mirror_num
;
4147 u64 physical_for_dev_replace
= nocow_ctx
->physical_for_dev_replace
;
4149 struct btrfs_trans_handle
*trans
= NULL
;
4150 struct btrfs_fs_info
*fs_info
;
4151 struct btrfs_path
*path
;
4152 struct btrfs_root
*root
;
4153 int not_written
= 0;
4155 fs_info
= sctx
->dev_root
->fs_info
;
4156 root
= fs_info
->extent_root
;
4158 path
= btrfs_alloc_path();
4160 spin_lock(&sctx
->stat_lock
);
4161 sctx
->stat
.malloc_errors
++;
4162 spin_unlock(&sctx
->stat_lock
);
4167 trans
= btrfs_join_transaction(root
);
4168 if (IS_ERR(trans
)) {
4173 ret
= iterate_inodes_from_logical(logical
, fs_info
, path
,
4174 record_inode_for_nocow
, nocow_ctx
);
4175 if (ret
!= 0 && ret
!= -ENOENT
) {
4176 btrfs_warn(fs_info
, "iterate_inodes_from_logical() failed: log %llu, "
4177 "phys %llu, len %llu, mir %u, ret %d",
4178 logical
, physical_for_dev_replace
, len
, mirror_num
,
4184 btrfs_end_transaction(trans
, root
);
4186 while (!list_empty(&nocow_ctx
->inodes
)) {
4187 struct scrub_nocow_inode
*entry
;
4188 entry
= list_first_entry(&nocow_ctx
->inodes
,
4189 struct scrub_nocow_inode
,
4191 list_del_init(&entry
->list
);
4192 ret
= copy_nocow_pages_for_inode(entry
->inum
, entry
->offset
,
4193 entry
->root
, nocow_ctx
);
4195 if (ret
== COPY_COMPLETE
) {
4203 while (!list_empty(&nocow_ctx
->inodes
)) {
4204 struct scrub_nocow_inode
*entry
;
4205 entry
= list_first_entry(&nocow_ctx
->inodes
,
4206 struct scrub_nocow_inode
,
4208 list_del_init(&entry
->list
);
4211 if (trans
&& !IS_ERR(trans
))
4212 btrfs_end_transaction(trans
, root
);
4214 btrfs_dev_replace_stats_inc(&fs_info
->dev_replace
.
4215 num_uncorrectable_read_errors
);
4217 btrfs_free_path(path
);
4220 scrub_pending_trans_workers_dec(sctx
);
4223 static int check_extent_to_block(struct inode
*inode
, u64 start
, u64 len
,
4226 struct extent_state
*cached_state
= NULL
;
4227 struct btrfs_ordered_extent
*ordered
;
4228 struct extent_io_tree
*io_tree
;
4229 struct extent_map
*em
;
4230 u64 lockstart
= start
, lockend
= start
+ len
- 1;
4233 io_tree
= &BTRFS_I(inode
)->io_tree
;
4235 lock_extent_bits(io_tree
, lockstart
, lockend
, 0, &cached_state
);
4236 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
, len
);
4238 btrfs_put_ordered_extent(ordered
);
4243 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
4250 * This extent does not actually cover the logical extent anymore,
4251 * move on to the next inode.
4253 if (em
->block_start
> logical
||
4254 em
->block_start
+ em
->block_len
< logical
+ len
) {
4255 free_extent_map(em
);
4259 free_extent_map(em
);
4262 unlock_extent_cached(io_tree
, lockstart
, lockend
, &cached_state
,
4267 static int copy_nocow_pages_for_inode(u64 inum
, u64 offset
, u64 root
,
4268 struct scrub_copy_nocow_ctx
*nocow_ctx
)
4270 struct btrfs_fs_info
*fs_info
= nocow_ctx
->sctx
->dev_root
->fs_info
;
4271 struct btrfs_key key
;
4272 struct inode
*inode
;
4274 struct btrfs_root
*local_root
;
4275 struct extent_io_tree
*io_tree
;
4276 u64 physical_for_dev_replace
;
4277 u64 nocow_ctx_logical
;
4278 u64 len
= nocow_ctx
->len
;
4279 unsigned long index
;
4284 key
.objectid
= root
;
4285 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4286 key
.offset
= (u64
)-1;
4288 srcu_index
= srcu_read_lock(&fs_info
->subvol_srcu
);
4290 local_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
4291 if (IS_ERR(local_root
)) {
4292 srcu_read_unlock(&fs_info
->subvol_srcu
, srcu_index
);
4293 return PTR_ERR(local_root
);
4296 key
.type
= BTRFS_INODE_ITEM_KEY
;
4297 key
.objectid
= inum
;
4299 inode
= btrfs_iget(fs_info
->sb
, &key
, local_root
, NULL
);
4300 srcu_read_unlock(&fs_info
->subvol_srcu
, srcu_index
);
4302 return PTR_ERR(inode
);
4304 /* Avoid truncate/dio/punch hole.. */
4305 mutex_lock(&inode
->i_mutex
);
4306 inode_dio_wait(inode
);
4308 physical_for_dev_replace
= nocow_ctx
->physical_for_dev_replace
;
4309 io_tree
= &BTRFS_I(inode
)->io_tree
;
4310 nocow_ctx_logical
= nocow_ctx
->logical
;
4312 ret
= check_extent_to_block(inode
, offset
, len
, nocow_ctx_logical
);
4314 ret
= ret
> 0 ? 0 : ret
;
4318 while (len
>= PAGE_CACHE_SIZE
) {
4319 index
= offset
>> PAGE_CACHE_SHIFT
;
4321 page
= find_or_create_page(inode
->i_mapping
, index
, GFP_NOFS
);
4323 btrfs_err(fs_info
, "find_or_create_page() failed");
4328 if (PageUptodate(page
)) {
4329 if (PageDirty(page
))
4332 ClearPageError(page
);
4333 err
= extent_read_full_page(io_tree
, page
,
4335 nocow_ctx
->mirror_num
);
4343 * If the page has been remove from the page cache,
4344 * the data on it is meaningless, because it may be
4345 * old one, the new data may be written into the new
4346 * page in the page cache.
4348 if (page
->mapping
!= inode
->i_mapping
) {
4350 page_cache_release(page
);
4353 if (!PageUptodate(page
)) {
4359 ret
= check_extent_to_block(inode
, offset
, len
,
4362 ret
= ret
> 0 ? 0 : ret
;
4366 err
= write_page_nocow(nocow_ctx
->sctx
,
4367 physical_for_dev_replace
, page
);
4372 page_cache_release(page
);
4377 offset
+= PAGE_CACHE_SIZE
;
4378 physical_for_dev_replace
+= PAGE_CACHE_SIZE
;
4379 nocow_ctx_logical
+= PAGE_CACHE_SIZE
;
4380 len
-= PAGE_CACHE_SIZE
;
4382 ret
= COPY_COMPLETE
;
4384 mutex_unlock(&inode
->i_mutex
);
4389 static int write_page_nocow(struct scrub_ctx
*sctx
,
4390 u64 physical_for_dev_replace
, struct page
*page
)
4393 struct btrfs_device
*dev
;
4396 dev
= sctx
->wr_ctx
.tgtdev
;
4400 btrfs_warn_rl(dev
->dev_root
->fs_info
,
4401 "scrub write_page_nocow(bdev == NULL) is unexpected");
4404 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
4406 spin_lock(&sctx
->stat_lock
);
4407 sctx
->stat
.malloc_errors
++;
4408 spin_unlock(&sctx
->stat_lock
);
4411 bio
->bi_iter
.bi_size
= 0;
4412 bio
->bi_iter
.bi_sector
= physical_for_dev_replace
>> 9;
4413 bio
->bi_bdev
= dev
->bdev
;
4414 ret
= bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0);
4415 if (ret
!= PAGE_CACHE_SIZE
) {
4418 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
4422 if (btrfsic_submit_bio_wait(WRITE_SYNC
, bio
))
4423 goto leave_with_eio
;