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 */
67 struct scrub_block
*sblock
;
69 struct btrfs_device
*dev
;
70 u64 flags
; /* extent flags */
74 u64 physical_for_dev_replace
;
77 unsigned int mirror_num
:8;
78 unsigned int have_csum
:1;
79 unsigned int io_error
:1;
81 u8 csum
[BTRFS_CSUM_SIZE
];
86 struct scrub_ctx
*sctx
;
87 struct btrfs_device
*dev
;
92 #if SCRUB_PAGES_PER_WR_BIO >= SCRUB_PAGES_PER_RD_BIO
93 struct scrub_page
*pagev
[SCRUB_PAGES_PER_WR_BIO
];
95 struct scrub_page
*pagev
[SCRUB_PAGES_PER_RD_BIO
];
99 struct btrfs_work work
;
103 struct scrub_page
*pagev
[SCRUB_MAX_PAGES_PER_BLOCK
];
105 atomic_t outstanding_pages
;
106 atomic_t ref_count
; /* free mem on transition to zero */
107 struct scrub_ctx
*sctx
;
109 unsigned int header_error
:1;
110 unsigned int checksum_error
:1;
111 unsigned int no_io_error_seen
:1;
112 unsigned int generation_error
:1; /* also sets header_error */
116 struct scrub_wr_ctx
{
117 struct scrub_bio
*wr_curr_bio
;
118 struct btrfs_device
*tgtdev
;
119 int pages_per_wr_bio
; /* <= SCRUB_PAGES_PER_WR_BIO */
120 atomic_t flush_all_writes
;
121 struct mutex wr_lock
;
125 struct scrub_bio
*bios
[SCRUB_BIOS_PER_SCTX
];
126 struct btrfs_root
*dev_root
;
129 atomic_t bios_in_flight
;
130 atomic_t workers_pending
;
131 spinlock_t list_lock
;
132 wait_queue_head_t list_wait
;
134 struct list_head csum_list
;
137 int pages_per_rd_bio
;
143 struct scrub_wr_ctx wr_ctx
;
148 struct btrfs_scrub_progress stat
;
149 spinlock_t stat_lock
;
152 struct scrub_fixup_nodatasum
{
153 struct scrub_ctx
*sctx
;
154 struct btrfs_device
*dev
;
156 struct btrfs_root
*root
;
157 struct btrfs_work work
;
161 struct scrub_nocow_inode
{
165 struct list_head list
;
168 struct scrub_copy_nocow_ctx
{
169 struct scrub_ctx
*sctx
;
173 u64 physical_for_dev_replace
;
174 struct list_head inodes
;
175 struct btrfs_work work
;
178 struct scrub_warning
{
179 struct btrfs_path
*path
;
180 u64 extent_item_size
;
186 struct btrfs_device
*dev
;
192 static void scrub_pending_bio_inc(struct scrub_ctx
*sctx
);
193 static void scrub_pending_bio_dec(struct scrub_ctx
*sctx
);
194 static void scrub_pending_trans_workers_inc(struct scrub_ctx
*sctx
);
195 static void scrub_pending_trans_workers_dec(struct scrub_ctx
*sctx
);
196 static int scrub_handle_errored_block(struct scrub_block
*sblock_to_check
);
197 static int scrub_setup_recheck_block(struct scrub_ctx
*sctx
,
198 struct btrfs_fs_info
*fs_info
,
199 struct scrub_block
*original_sblock
,
200 u64 length
, u64 logical
,
201 struct scrub_block
*sblocks_for_recheck
);
202 static void scrub_recheck_block(struct btrfs_fs_info
*fs_info
,
203 struct scrub_block
*sblock
, int is_metadata
,
204 int have_csum
, u8
*csum
, u64 generation
,
206 static void scrub_recheck_block_checksum(struct btrfs_fs_info
*fs_info
,
207 struct scrub_block
*sblock
,
208 int is_metadata
, int have_csum
,
209 const u8
*csum
, u64 generation
,
211 static int scrub_repair_block_from_good_copy(struct scrub_block
*sblock_bad
,
212 struct scrub_block
*sblock_good
,
214 static int scrub_repair_page_from_good_copy(struct scrub_block
*sblock_bad
,
215 struct scrub_block
*sblock_good
,
216 int page_num
, int force_write
);
217 static void scrub_write_block_to_dev_replace(struct scrub_block
*sblock
);
218 static int scrub_write_page_to_dev_replace(struct scrub_block
*sblock
,
220 static int scrub_checksum_data(struct scrub_block
*sblock
);
221 static int scrub_checksum_tree_block(struct scrub_block
*sblock
);
222 static int scrub_checksum_super(struct scrub_block
*sblock
);
223 static void scrub_block_get(struct scrub_block
*sblock
);
224 static void scrub_block_put(struct scrub_block
*sblock
);
225 static void scrub_page_get(struct scrub_page
*spage
);
226 static void scrub_page_put(struct scrub_page
*spage
);
227 static int scrub_add_page_to_rd_bio(struct scrub_ctx
*sctx
,
228 struct scrub_page
*spage
);
229 static int scrub_pages(struct scrub_ctx
*sctx
, u64 logical
, u64 len
,
230 u64 physical
, struct btrfs_device
*dev
, u64 flags
,
231 u64 gen
, int mirror_num
, u8
*csum
, int force
,
232 u64 physical_for_dev_replace
);
233 static void scrub_bio_end_io(struct bio
*bio
, int err
);
234 static void scrub_bio_end_io_worker(struct btrfs_work
*work
);
235 static void scrub_block_complete(struct scrub_block
*sblock
);
236 static void scrub_remap_extent(struct btrfs_fs_info
*fs_info
,
237 u64 extent_logical
, u64 extent_len
,
238 u64
*extent_physical
,
239 struct btrfs_device
**extent_dev
,
240 int *extent_mirror_num
);
241 static int scrub_setup_wr_ctx(struct scrub_ctx
*sctx
,
242 struct scrub_wr_ctx
*wr_ctx
,
243 struct btrfs_fs_info
*fs_info
,
244 struct btrfs_device
*dev
,
246 static void scrub_free_wr_ctx(struct scrub_wr_ctx
*wr_ctx
);
247 static int scrub_add_page_to_wr_bio(struct scrub_ctx
*sctx
,
248 struct scrub_page
*spage
);
249 static void scrub_wr_submit(struct scrub_ctx
*sctx
);
250 static void scrub_wr_bio_end_io(struct bio
*bio
, int err
);
251 static void scrub_wr_bio_end_io_worker(struct btrfs_work
*work
);
252 static int write_page_nocow(struct scrub_ctx
*sctx
,
253 u64 physical_for_dev_replace
, struct page
*page
);
254 static int copy_nocow_pages_for_inode(u64 inum
, u64 offset
, u64 root
,
255 struct scrub_copy_nocow_ctx
*ctx
);
256 static int copy_nocow_pages(struct scrub_ctx
*sctx
, u64 logical
, u64 len
,
257 int mirror_num
, u64 physical_for_dev_replace
);
258 static void copy_nocow_pages_worker(struct btrfs_work
*work
);
259 static void __scrub_blocked_if_needed(struct btrfs_fs_info
*fs_info
);
260 static void scrub_blocked_if_needed(struct btrfs_fs_info
*fs_info
);
263 static void scrub_pending_bio_inc(struct scrub_ctx
*sctx
)
265 atomic_inc(&sctx
->bios_in_flight
);
268 static void scrub_pending_bio_dec(struct scrub_ctx
*sctx
)
270 atomic_dec(&sctx
->bios_in_flight
);
271 wake_up(&sctx
->list_wait
);
274 static void __scrub_blocked_if_needed(struct btrfs_fs_info
*fs_info
)
276 while (atomic_read(&fs_info
->scrub_pause_req
)) {
277 mutex_unlock(&fs_info
->scrub_lock
);
278 wait_event(fs_info
->scrub_pause_wait
,
279 atomic_read(&fs_info
->scrub_pause_req
) == 0);
280 mutex_lock(&fs_info
->scrub_lock
);
284 static void scrub_blocked_if_needed(struct btrfs_fs_info
*fs_info
)
286 atomic_inc(&fs_info
->scrubs_paused
);
287 wake_up(&fs_info
->scrub_pause_wait
);
289 mutex_lock(&fs_info
->scrub_lock
);
290 __scrub_blocked_if_needed(fs_info
);
291 atomic_dec(&fs_info
->scrubs_paused
);
292 mutex_unlock(&fs_info
->scrub_lock
);
294 wake_up(&fs_info
->scrub_pause_wait
);
298 * used for workers that require transaction commits (i.e., for the
301 static void scrub_pending_trans_workers_inc(struct scrub_ctx
*sctx
)
303 struct btrfs_fs_info
*fs_info
= sctx
->dev_root
->fs_info
;
306 * increment scrubs_running to prevent cancel requests from
307 * completing as long as a worker is running. we must also
308 * increment scrubs_paused to prevent deadlocking on pause
309 * requests used for transactions commits (as the worker uses a
310 * transaction context). it is safe to regard the worker
311 * as paused for all matters practical. effectively, we only
312 * avoid cancellation requests from completing.
314 mutex_lock(&fs_info
->scrub_lock
);
315 atomic_inc(&fs_info
->scrubs_running
);
316 atomic_inc(&fs_info
->scrubs_paused
);
317 mutex_unlock(&fs_info
->scrub_lock
);
320 * check if @scrubs_running=@scrubs_paused condition
321 * inside wait_event() is not an atomic operation.
322 * which means we may inc/dec @scrub_running/paused
323 * at any time. Let's wake up @scrub_pause_wait as
324 * much as we can to let commit transaction blocked less.
326 wake_up(&fs_info
->scrub_pause_wait
);
328 atomic_inc(&sctx
->workers_pending
);
331 /* used for workers that require transaction commits */
332 static void scrub_pending_trans_workers_dec(struct scrub_ctx
*sctx
)
334 struct btrfs_fs_info
*fs_info
= sctx
->dev_root
->fs_info
;
337 * see scrub_pending_trans_workers_inc() why we're pretending
338 * to be paused in the scrub counters
340 mutex_lock(&fs_info
->scrub_lock
);
341 atomic_dec(&fs_info
->scrubs_running
);
342 atomic_dec(&fs_info
->scrubs_paused
);
343 mutex_unlock(&fs_info
->scrub_lock
);
344 atomic_dec(&sctx
->workers_pending
);
345 wake_up(&fs_info
->scrub_pause_wait
);
346 wake_up(&sctx
->list_wait
);
349 static void scrub_free_csums(struct scrub_ctx
*sctx
)
351 while (!list_empty(&sctx
->csum_list
)) {
352 struct btrfs_ordered_sum
*sum
;
353 sum
= list_first_entry(&sctx
->csum_list
,
354 struct btrfs_ordered_sum
, list
);
355 list_del(&sum
->list
);
360 static noinline_for_stack
void scrub_free_ctx(struct scrub_ctx
*sctx
)
367 scrub_free_wr_ctx(&sctx
->wr_ctx
);
369 /* this can happen when scrub is cancelled */
370 if (sctx
->curr
!= -1) {
371 struct scrub_bio
*sbio
= sctx
->bios
[sctx
->curr
];
373 for (i
= 0; i
< sbio
->page_count
; i
++) {
374 WARN_ON(!sbio
->pagev
[i
]->page
);
375 scrub_block_put(sbio
->pagev
[i
]->sblock
);
380 for (i
= 0; i
< SCRUB_BIOS_PER_SCTX
; ++i
) {
381 struct scrub_bio
*sbio
= sctx
->bios
[i
];
388 scrub_free_csums(sctx
);
392 static noinline_for_stack
393 struct scrub_ctx
*scrub_setup_ctx(struct btrfs_device
*dev
, int is_dev_replace
)
395 struct scrub_ctx
*sctx
;
397 struct btrfs_fs_info
*fs_info
= dev
->dev_root
->fs_info
;
398 int pages_per_rd_bio
;
402 * the setting of pages_per_rd_bio is correct for scrub but might
403 * be wrong for the dev_replace code where we might read from
404 * different devices in the initial huge bios. However, that
405 * code is able to correctly handle the case when adding a page
409 pages_per_rd_bio
= min_t(int, SCRUB_PAGES_PER_RD_BIO
,
410 bio_get_nr_vecs(dev
->bdev
));
412 pages_per_rd_bio
= SCRUB_PAGES_PER_RD_BIO
;
413 sctx
= kzalloc(sizeof(*sctx
), GFP_NOFS
);
416 sctx
->is_dev_replace
= is_dev_replace
;
417 sctx
->pages_per_rd_bio
= pages_per_rd_bio
;
419 sctx
->dev_root
= dev
->dev_root
;
420 for (i
= 0; i
< SCRUB_BIOS_PER_SCTX
; ++i
) {
421 struct scrub_bio
*sbio
;
423 sbio
= kzalloc(sizeof(*sbio
), GFP_NOFS
);
426 sctx
->bios
[i
] = sbio
;
430 sbio
->page_count
= 0;
431 btrfs_init_work(&sbio
->work
, scrub_bio_end_io_worker
,
434 if (i
!= SCRUB_BIOS_PER_SCTX
- 1)
435 sctx
->bios
[i
]->next_free
= i
+ 1;
437 sctx
->bios
[i
]->next_free
= -1;
439 sctx
->first_free
= 0;
440 sctx
->nodesize
= dev
->dev_root
->nodesize
;
441 sctx
->leafsize
= dev
->dev_root
->leafsize
;
442 sctx
->sectorsize
= dev
->dev_root
->sectorsize
;
443 atomic_set(&sctx
->bios_in_flight
, 0);
444 atomic_set(&sctx
->workers_pending
, 0);
445 atomic_set(&sctx
->cancel_req
, 0);
446 sctx
->csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
447 INIT_LIST_HEAD(&sctx
->csum_list
);
449 spin_lock_init(&sctx
->list_lock
);
450 spin_lock_init(&sctx
->stat_lock
);
451 init_waitqueue_head(&sctx
->list_wait
);
453 ret
= scrub_setup_wr_ctx(sctx
, &sctx
->wr_ctx
, fs_info
,
454 fs_info
->dev_replace
.tgtdev
, is_dev_replace
);
456 scrub_free_ctx(sctx
);
462 scrub_free_ctx(sctx
);
463 return ERR_PTR(-ENOMEM
);
466 static int scrub_print_warning_inode(u64 inum
, u64 offset
, u64 root
,
473 struct extent_buffer
*eb
;
474 struct btrfs_inode_item
*inode_item
;
475 struct scrub_warning
*swarn
= warn_ctx
;
476 struct btrfs_fs_info
*fs_info
= swarn
->dev
->dev_root
->fs_info
;
477 struct inode_fs_paths
*ipath
= NULL
;
478 struct btrfs_root
*local_root
;
479 struct btrfs_key root_key
;
481 root_key
.objectid
= root
;
482 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
483 root_key
.offset
= (u64
)-1;
484 local_root
= btrfs_read_fs_root_no_name(fs_info
, &root_key
);
485 if (IS_ERR(local_root
)) {
486 ret
= PTR_ERR(local_root
);
490 ret
= inode_item_info(inum
, 0, local_root
, swarn
->path
);
492 btrfs_release_path(swarn
->path
);
496 eb
= swarn
->path
->nodes
[0];
497 inode_item
= btrfs_item_ptr(eb
, swarn
->path
->slots
[0],
498 struct btrfs_inode_item
);
499 isize
= btrfs_inode_size(eb
, inode_item
);
500 nlink
= btrfs_inode_nlink(eb
, inode_item
);
501 btrfs_release_path(swarn
->path
);
503 ipath
= init_ipath(4096, local_root
, swarn
->path
);
505 ret
= PTR_ERR(ipath
);
509 ret
= paths_from_inode(inum
, ipath
);
515 * we deliberately ignore the bit ipath might have been too small to
516 * hold all of the paths here
518 for (i
= 0; i
< ipath
->fspath
->elem_cnt
; ++i
)
519 printk_in_rcu(KERN_WARNING
"BTRFS: %s at logical %llu on dev "
520 "%s, sector %llu, root %llu, inode %llu, offset %llu, "
521 "length %llu, links %u (path: %s)\n", swarn
->errstr
,
522 swarn
->logical
, rcu_str_deref(swarn
->dev
->name
),
523 (unsigned long long)swarn
->sector
, root
, inum
, offset
,
524 min(isize
- offset
, (u64
)PAGE_SIZE
), nlink
,
525 (char *)(unsigned long)ipath
->fspath
->val
[i
]);
531 printk_in_rcu(KERN_WARNING
"BTRFS: %s at logical %llu on dev "
532 "%s, sector %llu, root %llu, inode %llu, offset %llu: path "
533 "resolving failed with ret=%d\n", swarn
->errstr
,
534 swarn
->logical
, rcu_str_deref(swarn
->dev
->name
),
535 (unsigned long long)swarn
->sector
, root
, inum
, offset
, ret
);
541 static void scrub_print_warning(const char *errstr
, struct scrub_block
*sblock
)
543 struct btrfs_device
*dev
;
544 struct btrfs_fs_info
*fs_info
;
545 struct btrfs_path
*path
;
546 struct btrfs_key found_key
;
547 struct extent_buffer
*eb
;
548 struct btrfs_extent_item
*ei
;
549 struct scrub_warning swarn
;
550 unsigned long ptr
= 0;
556 const int bufsize
= 4096;
559 WARN_ON(sblock
->page_count
< 1);
560 dev
= sblock
->pagev
[0]->dev
;
561 fs_info
= sblock
->sctx
->dev_root
->fs_info
;
563 path
= btrfs_alloc_path();
565 swarn
.scratch_buf
= kmalloc(bufsize
, GFP_NOFS
);
566 swarn
.msg_buf
= kmalloc(bufsize
, GFP_NOFS
);
567 swarn
.sector
= (sblock
->pagev
[0]->physical
) >> 9;
568 swarn
.logical
= sblock
->pagev
[0]->logical
;
569 swarn
.errstr
= errstr
;
571 swarn
.msg_bufsize
= bufsize
;
572 swarn
.scratch_bufsize
= bufsize
;
574 if (!path
|| !swarn
.scratch_buf
|| !swarn
.msg_buf
)
577 ret
= extent_from_logical(fs_info
, swarn
.logical
, path
, &found_key
,
582 extent_item_pos
= swarn
.logical
- found_key
.objectid
;
583 swarn
.extent_item_size
= found_key
.offset
;
586 ei
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_extent_item
);
587 item_size
= btrfs_item_size_nr(eb
, path
->slots
[0]);
589 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
591 ret
= tree_backref_for_extent(&ptr
, eb
, ei
, item_size
,
592 &ref_root
, &ref_level
);
593 printk_in_rcu(KERN_WARNING
594 "BTRFS: %s at logical %llu on dev %s, "
595 "sector %llu: metadata %s (level %d) in tree "
596 "%llu\n", errstr
, swarn
.logical
,
597 rcu_str_deref(dev
->name
),
598 (unsigned long long)swarn
.sector
,
599 ref_level
? "node" : "leaf",
600 ret
< 0 ? -1 : ref_level
,
601 ret
< 0 ? -1 : ref_root
);
603 btrfs_release_path(path
);
605 btrfs_release_path(path
);
608 iterate_extent_inodes(fs_info
, found_key
.objectid
,
610 scrub_print_warning_inode
, &swarn
);
614 btrfs_free_path(path
);
615 kfree(swarn
.scratch_buf
);
616 kfree(swarn
.msg_buf
);
619 static int scrub_fixup_readpage(u64 inum
, u64 offset
, u64 root
, void *fixup_ctx
)
621 struct page
*page
= NULL
;
623 struct scrub_fixup_nodatasum
*fixup
= fixup_ctx
;
626 struct btrfs_key key
;
627 struct inode
*inode
= NULL
;
628 struct btrfs_fs_info
*fs_info
;
629 u64 end
= offset
+ PAGE_SIZE
- 1;
630 struct btrfs_root
*local_root
;
634 key
.type
= BTRFS_ROOT_ITEM_KEY
;
635 key
.offset
= (u64
)-1;
637 fs_info
= fixup
->root
->fs_info
;
638 srcu_index
= srcu_read_lock(&fs_info
->subvol_srcu
);
640 local_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
641 if (IS_ERR(local_root
)) {
642 srcu_read_unlock(&fs_info
->subvol_srcu
, srcu_index
);
643 return PTR_ERR(local_root
);
646 key
.type
= BTRFS_INODE_ITEM_KEY
;
649 inode
= btrfs_iget(fs_info
->sb
, &key
, local_root
, NULL
);
650 srcu_read_unlock(&fs_info
->subvol_srcu
, srcu_index
);
652 return PTR_ERR(inode
);
654 index
= offset
>> PAGE_CACHE_SHIFT
;
656 page
= find_or_create_page(inode
->i_mapping
, index
, GFP_NOFS
);
662 if (PageUptodate(page
)) {
663 if (PageDirty(page
)) {
665 * we need to write the data to the defect sector. the
666 * data that was in that sector is not in memory,
667 * because the page was modified. we must not write the
668 * modified page to that sector.
670 * TODO: what could be done here: wait for the delalloc
671 * runner to write out that page (might involve
672 * COW) and see whether the sector is still
673 * referenced afterwards.
675 * For the meantime, we'll treat this error
676 * incorrectable, although there is a chance that a
677 * later scrub will find the bad sector again and that
678 * there's no dirty page in memory, then.
683 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
684 ret
= repair_io_failure(fs_info
, offset
, PAGE_SIZE
,
685 fixup
->logical
, page
,
691 * we need to get good data first. the general readpage path
692 * will call repair_io_failure for us, we just have to make
693 * sure we read the bad mirror.
695 ret
= set_extent_bits(&BTRFS_I(inode
)->io_tree
, offset
, end
,
696 EXTENT_DAMAGED
, GFP_NOFS
);
698 /* set_extent_bits should give proper error */
705 ret
= extent_read_full_page(&BTRFS_I(inode
)->io_tree
, page
,
708 wait_on_page_locked(page
);
710 corrected
= !test_range_bit(&BTRFS_I(inode
)->io_tree
, offset
,
711 end
, EXTENT_DAMAGED
, 0, NULL
);
713 clear_extent_bits(&BTRFS_I(inode
)->io_tree
, offset
, end
,
714 EXTENT_DAMAGED
, GFP_NOFS
);
726 if (ret
== 0 && corrected
) {
728 * we only need to call readpage for one of the inodes belonging
729 * to this extent. so make iterate_extent_inodes stop
737 static void scrub_fixup_nodatasum(struct btrfs_work
*work
)
740 struct scrub_fixup_nodatasum
*fixup
;
741 struct scrub_ctx
*sctx
;
742 struct btrfs_trans_handle
*trans
= NULL
;
743 struct btrfs_path
*path
;
744 int uncorrectable
= 0;
746 fixup
= container_of(work
, struct scrub_fixup_nodatasum
, work
);
749 path
= btrfs_alloc_path();
751 spin_lock(&sctx
->stat_lock
);
752 ++sctx
->stat
.malloc_errors
;
753 spin_unlock(&sctx
->stat_lock
);
758 trans
= btrfs_join_transaction(fixup
->root
);
765 * the idea is to trigger a regular read through the standard path. we
766 * read a page from the (failed) logical address by specifying the
767 * corresponding copynum of the failed sector. thus, that readpage is
769 * that is the point where on-the-fly error correction will kick in
770 * (once it's finished) and rewrite the failed sector if a good copy
773 ret
= iterate_inodes_from_logical(fixup
->logical
, fixup
->root
->fs_info
,
774 path
, scrub_fixup_readpage
,
782 spin_lock(&sctx
->stat_lock
);
783 ++sctx
->stat
.corrected_errors
;
784 spin_unlock(&sctx
->stat_lock
);
787 if (trans
&& !IS_ERR(trans
))
788 btrfs_end_transaction(trans
, fixup
->root
);
790 spin_lock(&sctx
->stat_lock
);
791 ++sctx
->stat
.uncorrectable_errors
;
792 spin_unlock(&sctx
->stat_lock
);
793 btrfs_dev_replace_stats_inc(
794 &sctx
->dev_root
->fs_info
->dev_replace
.
795 num_uncorrectable_read_errors
);
796 printk_ratelimited_in_rcu(KERN_ERR
"BTRFS: "
797 "unable to fixup (nodatasum) error at logical %llu on dev %s\n",
798 fixup
->logical
, rcu_str_deref(fixup
->dev
->name
));
801 btrfs_free_path(path
);
804 scrub_pending_trans_workers_dec(sctx
);
808 * scrub_handle_errored_block gets called when either verification of the
809 * pages failed or the bio failed to read, e.g. with EIO. In the latter
810 * case, this function handles all pages in the bio, even though only one
812 * The goal of this function is to repair the errored block by using the
813 * contents of one of the mirrors.
815 static int scrub_handle_errored_block(struct scrub_block
*sblock_to_check
)
817 struct scrub_ctx
*sctx
= sblock_to_check
->sctx
;
818 struct btrfs_device
*dev
;
819 struct btrfs_fs_info
*fs_info
;
823 unsigned int failed_mirror_index
;
824 unsigned int is_metadata
;
825 unsigned int have_csum
;
827 struct scrub_block
*sblocks_for_recheck
; /* holds one for each mirror */
828 struct scrub_block
*sblock_bad
;
833 static DEFINE_RATELIMIT_STATE(_rs
, DEFAULT_RATELIMIT_INTERVAL
,
834 DEFAULT_RATELIMIT_BURST
);
836 BUG_ON(sblock_to_check
->page_count
< 1);
837 fs_info
= sctx
->dev_root
->fs_info
;
838 if (sblock_to_check
->pagev
[0]->flags
& BTRFS_EXTENT_FLAG_SUPER
) {
840 * if we find an error in a super block, we just report it.
841 * They will get written with the next transaction commit
844 spin_lock(&sctx
->stat_lock
);
845 ++sctx
->stat
.super_errors
;
846 spin_unlock(&sctx
->stat_lock
);
849 length
= sblock_to_check
->page_count
* PAGE_SIZE
;
850 logical
= sblock_to_check
->pagev
[0]->logical
;
851 generation
= sblock_to_check
->pagev
[0]->generation
;
852 BUG_ON(sblock_to_check
->pagev
[0]->mirror_num
< 1);
853 failed_mirror_index
= sblock_to_check
->pagev
[0]->mirror_num
- 1;
854 is_metadata
= !(sblock_to_check
->pagev
[0]->flags
&
855 BTRFS_EXTENT_FLAG_DATA
);
856 have_csum
= sblock_to_check
->pagev
[0]->have_csum
;
857 csum
= sblock_to_check
->pagev
[0]->csum
;
858 dev
= sblock_to_check
->pagev
[0]->dev
;
860 if (sctx
->is_dev_replace
&& !is_metadata
&& !have_csum
) {
861 sblocks_for_recheck
= NULL
;
866 * read all mirrors one after the other. This includes to
867 * re-read the extent or metadata block that failed (that was
868 * the cause that this fixup code is called) another time,
869 * page by page this time in order to know which pages
870 * caused I/O errors and which ones are good (for all mirrors).
871 * It is the goal to handle the situation when more than one
872 * mirror contains I/O errors, but the errors do not
873 * overlap, i.e. the data can be repaired by selecting the
874 * pages from those mirrors without I/O error on the
875 * particular pages. One example (with blocks >= 2 * PAGE_SIZE)
876 * would be that mirror #1 has an I/O error on the first page,
877 * the second page is good, and mirror #2 has an I/O error on
878 * the second page, but the first page is good.
879 * Then the first page of the first mirror can be repaired by
880 * taking the first page of the second mirror, and the
881 * second page of the second mirror can be repaired by
882 * copying the contents of the 2nd page of the 1st mirror.
883 * One more note: if the pages of one mirror contain I/O
884 * errors, the checksum cannot be verified. In order to get
885 * the best data for repairing, the first attempt is to find
886 * a mirror without I/O errors and with a validated checksum.
887 * Only if this is not possible, the pages are picked from
888 * mirrors with I/O errors without considering the checksum.
889 * If the latter is the case, at the end, the checksum of the
890 * repaired area is verified in order to correctly maintain
894 sblocks_for_recheck
= kzalloc(BTRFS_MAX_MIRRORS
*
895 sizeof(*sblocks_for_recheck
),
897 if (!sblocks_for_recheck
) {
898 spin_lock(&sctx
->stat_lock
);
899 sctx
->stat
.malloc_errors
++;
900 sctx
->stat
.read_errors
++;
901 sctx
->stat
.uncorrectable_errors
++;
902 spin_unlock(&sctx
->stat_lock
);
903 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_READ_ERRS
);
907 /* setup the context, map the logical blocks and alloc the pages */
908 ret
= scrub_setup_recheck_block(sctx
, fs_info
, sblock_to_check
, length
,
909 logical
, sblocks_for_recheck
);
911 spin_lock(&sctx
->stat_lock
);
912 sctx
->stat
.read_errors
++;
913 sctx
->stat
.uncorrectable_errors
++;
914 spin_unlock(&sctx
->stat_lock
);
915 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_READ_ERRS
);
918 BUG_ON(failed_mirror_index
>= BTRFS_MAX_MIRRORS
);
919 sblock_bad
= sblocks_for_recheck
+ failed_mirror_index
;
921 /* build and submit the bios for the failed mirror, check checksums */
922 scrub_recheck_block(fs_info
, sblock_bad
, is_metadata
, have_csum
,
923 csum
, generation
, sctx
->csum_size
);
925 if (!sblock_bad
->header_error
&& !sblock_bad
->checksum_error
&&
926 sblock_bad
->no_io_error_seen
) {
928 * the error disappeared after reading page by page, or
929 * the area was part of a huge bio and other parts of the
930 * bio caused I/O errors, or the block layer merged several
931 * read requests into one and the error is caused by a
932 * different bio (usually one of the two latter cases is
935 spin_lock(&sctx
->stat_lock
);
936 sctx
->stat
.unverified_errors
++;
937 spin_unlock(&sctx
->stat_lock
);
939 if (sctx
->is_dev_replace
)
940 scrub_write_block_to_dev_replace(sblock_bad
);
944 if (!sblock_bad
->no_io_error_seen
) {
945 spin_lock(&sctx
->stat_lock
);
946 sctx
->stat
.read_errors
++;
947 spin_unlock(&sctx
->stat_lock
);
948 if (__ratelimit(&_rs
))
949 scrub_print_warning("i/o error", sblock_to_check
);
950 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_READ_ERRS
);
951 } else if (sblock_bad
->checksum_error
) {
952 spin_lock(&sctx
->stat_lock
);
953 sctx
->stat
.csum_errors
++;
954 spin_unlock(&sctx
->stat_lock
);
955 if (__ratelimit(&_rs
))
956 scrub_print_warning("checksum error", sblock_to_check
);
957 btrfs_dev_stat_inc_and_print(dev
,
958 BTRFS_DEV_STAT_CORRUPTION_ERRS
);
959 } else if (sblock_bad
->header_error
) {
960 spin_lock(&sctx
->stat_lock
);
961 sctx
->stat
.verify_errors
++;
962 spin_unlock(&sctx
->stat_lock
);
963 if (__ratelimit(&_rs
))
964 scrub_print_warning("checksum/header error",
966 if (sblock_bad
->generation_error
)
967 btrfs_dev_stat_inc_and_print(dev
,
968 BTRFS_DEV_STAT_GENERATION_ERRS
);
970 btrfs_dev_stat_inc_and_print(dev
,
971 BTRFS_DEV_STAT_CORRUPTION_ERRS
);
974 if (sctx
->readonly
) {
975 ASSERT(!sctx
->is_dev_replace
);
979 if (!is_metadata
&& !have_csum
) {
980 struct scrub_fixup_nodatasum
*fixup_nodatasum
;
983 WARN_ON(sctx
->is_dev_replace
);
986 * !is_metadata and !have_csum, this means that the data
987 * might not be COW'ed, that it might be modified
988 * concurrently. The general strategy to work on the
989 * commit root does not help in the case when COW is not
992 fixup_nodatasum
= kzalloc(sizeof(*fixup_nodatasum
), GFP_NOFS
);
993 if (!fixup_nodatasum
)
994 goto did_not_correct_error
;
995 fixup_nodatasum
->sctx
= sctx
;
996 fixup_nodatasum
->dev
= dev
;
997 fixup_nodatasum
->logical
= logical
;
998 fixup_nodatasum
->root
= fs_info
->extent_root
;
999 fixup_nodatasum
->mirror_num
= failed_mirror_index
+ 1;
1000 scrub_pending_trans_workers_inc(sctx
);
1001 btrfs_init_work(&fixup_nodatasum
->work
, scrub_fixup_nodatasum
,
1003 btrfs_queue_work(fs_info
->scrub_workers
,
1004 &fixup_nodatasum
->work
);
1009 * now build and submit the bios for the other mirrors, check
1011 * First try to pick the mirror which is completely without I/O
1012 * errors and also does not have a checksum error.
1013 * If one is found, and if a checksum is present, the full block
1014 * that is known to contain an error is rewritten. Afterwards
1015 * the block is known to be corrected.
1016 * If a mirror is found which is completely correct, and no
1017 * checksum is present, only those pages are rewritten that had
1018 * an I/O error in the block to be repaired, since it cannot be
1019 * determined, which copy of the other pages is better (and it
1020 * could happen otherwise that a correct page would be
1021 * overwritten by a bad one).
1023 for (mirror_index
= 0;
1024 mirror_index
< BTRFS_MAX_MIRRORS
&&
1025 sblocks_for_recheck
[mirror_index
].page_count
> 0;
1027 struct scrub_block
*sblock_other
;
1029 if (mirror_index
== failed_mirror_index
)
1031 sblock_other
= sblocks_for_recheck
+ mirror_index
;
1033 /* build and submit the bios, check checksums */
1034 scrub_recheck_block(fs_info
, sblock_other
, is_metadata
,
1035 have_csum
, csum
, generation
,
1038 if (!sblock_other
->header_error
&&
1039 !sblock_other
->checksum_error
&&
1040 sblock_other
->no_io_error_seen
) {
1041 if (sctx
->is_dev_replace
) {
1042 scrub_write_block_to_dev_replace(sblock_other
);
1044 int force_write
= is_metadata
|| have_csum
;
1046 ret
= scrub_repair_block_from_good_copy(
1047 sblock_bad
, sblock_other
,
1051 goto corrected_error
;
1056 * for dev_replace, pick good pages and write to the target device.
1058 if (sctx
->is_dev_replace
) {
1060 for (page_num
= 0; page_num
< sblock_bad
->page_count
;
1065 for (mirror_index
= 0;
1066 mirror_index
< BTRFS_MAX_MIRRORS
&&
1067 sblocks_for_recheck
[mirror_index
].page_count
> 0;
1069 struct scrub_block
*sblock_other
=
1070 sblocks_for_recheck
+ mirror_index
;
1071 struct scrub_page
*page_other
=
1072 sblock_other
->pagev
[page_num
];
1074 if (!page_other
->io_error
) {
1075 ret
= scrub_write_page_to_dev_replace(
1076 sblock_other
, page_num
);
1078 /* succeeded for this page */
1082 btrfs_dev_replace_stats_inc(
1084 fs_info
->dev_replace
.
1092 * did not find a mirror to fetch the page
1093 * from. scrub_write_page_to_dev_replace()
1094 * handles this case (page->io_error), by
1095 * filling the block with zeros before
1096 * submitting the write request
1099 ret
= scrub_write_page_to_dev_replace(
1100 sblock_bad
, page_num
);
1102 btrfs_dev_replace_stats_inc(
1103 &sctx
->dev_root
->fs_info
->
1104 dev_replace
.num_write_errors
);
1112 * for regular scrub, repair those pages that are errored.
1113 * In case of I/O errors in the area that is supposed to be
1114 * repaired, continue by picking good copies of those pages.
1115 * Select the good pages from mirrors to rewrite bad pages from
1116 * the area to fix. Afterwards verify the checksum of the block
1117 * that is supposed to be repaired. This verification step is
1118 * only done for the purpose of statistic counting and for the
1119 * final scrub report, whether errors remain.
1120 * A perfect algorithm could make use of the checksum and try
1121 * all possible combinations of pages from the different mirrors
1122 * until the checksum verification succeeds. For example, when
1123 * the 2nd page of mirror #1 faces I/O errors, and the 2nd page
1124 * of mirror #2 is readable but the final checksum test fails,
1125 * then the 2nd page of mirror #3 could be tried, whether now
1126 * the final checksum succeedes. But this would be a rare
1127 * exception and is therefore not implemented. At least it is
1128 * avoided that the good copy is overwritten.
1129 * A more useful improvement would be to pick the sectors
1130 * without I/O error based on sector sizes (512 bytes on legacy
1131 * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one
1132 * mirror could be repaired by taking 512 byte of a different
1133 * mirror, even if other 512 byte sectors in the same PAGE_SIZE
1134 * area are unreadable.
1137 /* can only fix I/O errors from here on */
1138 if (sblock_bad
->no_io_error_seen
)
1139 goto did_not_correct_error
;
1142 for (page_num
= 0; page_num
< sblock_bad
->page_count
; page_num
++) {
1143 struct scrub_page
*page_bad
= sblock_bad
->pagev
[page_num
];
1145 if (!page_bad
->io_error
)
1148 for (mirror_index
= 0;
1149 mirror_index
< BTRFS_MAX_MIRRORS
&&
1150 sblocks_for_recheck
[mirror_index
].page_count
> 0;
1152 struct scrub_block
*sblock_other
= sblocks_for_recheck
+
1154 struct scrub_page
*page_other
= sblock_other
->pagev
[
1157 if (!page_other
->io_error
) {
1158 ret
= scrub_repair_page_from_good_copy(
1159 sblock_bad
, sblock_other
, page_num
, 0);
1161 page_bad
->io_error
= 0;
1162 break; /* succeeded for this page */
1167 if (page_bad
->io_error
) {
1168 /* did not find a mirror to copy the page from */
1174 if (is_metadata
|| have_csum
) {
1176 * need to verify the checksum now that all
1177 * sectors on disk are repaired (the write
1178 * request for data to be repaired is on its way).
1179 * Just be lazy and use scrub_recheck_block()
1180 * which re-reads the data before the checksum
1181 * is verified, but most likely the data comes out
1182 * of the page cache.
1184 scrub_recheck_block(fs_info
, sblock_bad
,
1185 is_metadata
, have_csum
, csum
,
1186 generation
, sctx
->csum_size
);
1187 if (!sblock_bad
->header_error
&&
1188 !sblock_bad
->checksum_error
&&
1189 sblock_bad
->no_io_error_seen
)
1190 goto corrected_error
;
1192 goto did_not_correct_error
;
1195 spin_lock(&sctx
->stat_lock
);
1196 sctx
->stat
.corrected_errors
++;
1197 spin_unlock(&sctx
->stat_lock
);
1198 printk_ratelimited_in_rcu(KERN_ERR
1199 "BTRFS: fixed up error at logical %llu on dev %s\n",
1200 logical
, rcu_str_deref(dev
->name
));
1203 did_not_correct_error
:
1204 spin_lock(&sctx
->stat_lock
);
1205 sctx
->stat
.uncorrectable_errors
++;
1206 spin_unlock(&sctx
->stat_lock
);
1207 printk_ratelimited_in_rcu(KERN_ERR
1208 "BTRFS: unable to fixup (regular) error at logical %llu on dev %s\n",
1209 logical
, rcu_str_deref(dev
->name
));
1213 if (sblocks_for_recheck
) {
1214 for (mirror_index
= 0; mirror_index
< BTRFS_MAX_MIRRORS
;
1216 struct scrub_block
*sblock
= sblocks_for_recheck
+
1220 for (page_index
= 0; page_index
< sblock
->page_count
;
1222 sblock
->pagev
[page_index
]->sblock
= NULL
;
1223 scrub_page_put(sblock
->pagev
[page_index
]);
1226 kfree(sblocks_for_recheck
);
1232 static int scrub_setup_recheck_block(struct scrub_ctx
*sctx
,
1233 struct btrfs_fs_info
*fs_info
,
1234 struct scrub_block
*original_sblock
,
1235 u64 length
, u64 logical
,
1236 struct scrub_block
*sblocks_for_recheck
)
1243 * note: the two members ref_count and outstanding_pages
1244 * are not used (and not set) in the blocks that are used for
1245 * the recheck procedure
1249 while (length
> 0) {
1250 u64 sublen
= min_t(u64
, length
, PAGE_SIZE
);
1251 u64 mapped_length
= sublen
;
1252 struct btrfs_bio
*bbio
= NULL
;
1255 * with a length of PAGE_SIZE, each returned stripe
1256 * represents one mirror
1258 ret
= btrfs_map_block(fs_info
, REQ_GET_READ_MIRRORS
, logical
,
1259 &mapped_length
, &bbio
, 0);
1260 if (ret
|| !bbio
|| mapped_length
< sublen
) {
1265 BUG_ON(page_index
>= SCRUB_PAGES_PER_RD_BIO
);
1266 for (mirror_index
= 0; mirror_index
< (int)bbio
->num_stripes
;
1268 struct scrub_block
*sblock
;
1269 struct scrub_page
*page
;
1271 if (mirror_index
>= BTRFS_MAX_MIRRORS
)
1274 sblock
= sblocks_for_recheck
+ mirror_index
;
1275 sblock
->sctx
= sctx
;
1276 page
= kzalloc(sizeof(*page
), GFP_NOFS
);
1279 spin_lock(&sctx
->stat_lock
);
1280 sctx
->stat
.malloc_errors
++;
1281 spin_unlock(&sctx
->stat_lock
);
1285 scrub_page_get(page
);
1286 sblock
->pagev
[page_index
] = page
;
1287 page
->logical
= logical
;
1288 page
->physical
= bbio
->stripes
[mirror_index
].physical
;
1289 BUG_ON(page_index
>= original_sblock
->page_count
);
1290 page
->physical_for_dev_replace
=
1291 original_sblock
->pagev
[page_index
]->
1292 physical_for_dev_replace
;
1293 /* for missing devices, dev->bdev is NULL */
1294 page
->dev
= bbio
->stripes
[mirror_index
].dev
;
1295 page
->mirror_num
= mirror_index
+ 1;
1296 sblock
->page_count
++;
1297 page
->page
= alloc_page(GFP_NOFS
);
1311 * this function will check the on disk data for checksum errors, header
1312 * errors and read I/O errors. If any I/O errors happen, the exact pages
1313 * which are errored are marked as being bad. The goal is to enable scrub
1314 * to take those pages that are not errored from all the mirrors so that
1315 * the pages that are errored in the just handled mirror can be repaired.
1317 static void scrub_recheck_block(struct btrfs_fs_info
*fs_info
,
1318 struct scrub_block
*sblock
, int is_metadata
,
1319 int have_csum
, u8
*csum
, u64 generation
,
1324 sblock
->no_io_error_seen
= 1;
1325 sblock
->header_error
= 0;
1326 sblock
->checksum_error
= 0;
1328 for (page_num
= 0; page_num
< sblock
->page_count
; page_num
++) {
1330 struct scrub_page
*page
= sblock
->pagev
[page_num
];
1332 if (page
->dev
->bdev
== NULL
) {
1334 sblock
->no_io_error_seen
= 0;
1338 WARN_ON(!page
->page
);
1339 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
1342 sblock
->no_io_error_seen
= 0;
1345 bio
->bi_bdev
= page
->dev
->bdev
;
1346 bio
->bi_iter
.bi_sector
= page
->physical
>> 9;
1348 bio_add_page(bio
, page
->page
, PAGE_SIZE
, 0);
1349 if (btrfsic_submit_bio_wait(READ
, bio
))
1350 sblock
->no_io_error_seen
= 0;
1355 if (sblock
->no_io_error_seen
)
1356 scrub_recheck_block_checksum(fs_info
, sblock
, is_metadata
,
1357 have_csum
, csum
, generation
,
1363 static void scrub_recheck_block_checksum(struct btrfs_fs_info
*fs_info
,
1364 struct scrub_block
*sblock
,
1365 int is_metadata
, int have_csum
,
1366 const u8
*csum
, u64 generation
,
1370 u8 calculated_csum
[BTRFS_CSUM_SIZE
];
1372 void *mapped_buffer
;
1374 WARN_ON(!sblock
->pagev
[0]->page
);
1376 struct btrfs_header
*h
;
1378 mapped_buffer
= kmap_atomic(sblock
->pagev
[0]->page
);
1379 h
= (struct btrfs_header
*)mapped_buffer
;
1381 if (sblock
->pagev
[0]->logical
!= btrfs_stack_header_bytenr(h
) ||
1382 memcmp(h
->fsid
, fs_info
->fsid
, BTRFS_UUID_SIZE
) ||
1383 memcmp(h
->chunk_tree_uuid
, fs_info
->chunk_tree_uuid
,
1385 sblock
->header_error
= 1;
1386 } else if (generation
!= btrfs_stack_header_generation(h
)) {
1387 sblock
->header_error
= 1;
1388 sblock
->generation_error
= 1;
1395 mapped_buffer
= kmap_atomic(sblock
->pagev
[0]->page
);
1398 for (page_num
= 0;;) {
1399 if (page_num
== 0 && is_metadata
)
1400 crc
= btrfs_csum_data(
1401 ((u8
*)mapped_buffer
) + BTRFS_CSUM_SIZE
,
1402 crc
, PAGE_SIZE
- BTRFS_CSUM_SIZE
);
1404 crc
= btrfs_csum_data(mapped_buffer
, crc
, PAGE_SIZE
);
1406 kunmap_atomic(mapped_buffer
);
1408 if (page_num
>= sblock
->page_count
)
1410 WARN_ON(!sblock
->pagev
[page_num
]->page
);
1412 mapped_buffer
= kmap_atomic(sblock
->pagev
[page_num
]->page
);
1415 btrfs_csum_final(crc
, calculated_csum
);
1416 if (memcmp(calculated_csum
, csum
, csum_size
))
1417 sblock
->checksum_error
= 1;
1420 static int scrub_repair_block_from_good_copy(struct scrub_block
*sblock_bad
,
1421 struct scrub_block
*sblock_good
,
1427 for (page_num
= 0; page_num
< sblock_bad
->page_count
; page_num
++) {
1430 ret_sub
= scrub_repair_page_from_good_copy(sblock_bad
,
1441 static int scrub_repair_page_from_good_copy(struct scrub_block
*sblock_bad
,
1442 struct scrub_block
*sblock_good
,
1443 int page_num
, int force_write
)
1445 struct scrub_page
*page_bad
= sblock_bad
->pagev
[page_num
];
1446 struct scrub_page
*page_good
= sblock_good
->pagev
[page_num
];
1448 BUG_ON(page_bad
->page
== NULL
);
1449 BUG_ON(page_good
->page
== NULL
);
1450 if (force_write
|| sblock_bad
->header_error
||
1451 sblock_bad
->checksum_error
|| page_bad
->io_error
) {
1455 if (!page_bad
->dev
->bdev
) {
1456 printk_ratelimited(KERN_WARNING
"BTRFS: "
1457 "scrub_repair_page_from_good_copy(bdev == NULL) "
1458 "is unexpected!\n");
1462 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
1465 bio
->bi_bdev
= page_bad
->dev
->bdev
;
1466 bio
->bi_iter
.bi_sector
= page_bad
->physical
>> 9;
1468 ret
= bio_add_page(bio
, page_good
->page
, PAGE_SIZE
, 0);
1469 if (PAGE_SIZE
!= ret
) {
1474 if (btrfsic_submit_bio_wait(WRITE
, bio
)) {
1475 btrfs_dev_stat_inc_and_print(page_bad
->dev
,
1476 BTRFS_DEV_STAT_WRITE_ERRS
);
1477 btrfs_dev_replace_stats_inc(
1478 &sblock_bad
->sctx
->dev_root
->fs_info
->
1479 dev_replace
.num_write_errors
);
1489 static void scrub_write_block_to_dev_replace(struct scrub_block
*sblock
)
1493 for (page_num
= 0; page_num
< sblock
->page_count
; page_num
++) {
1496 ret
= scrub_write_page_to_dev_replace(sblock
, page_num
);
1498 btrfs_dev_replace_stats_inc(
1499 &sblock
->sctx
->dev_root
->fs_info
->dev_replace
.
1504 static int scrub_write_page_to_dev_replace(struct scrub_block
*sblock
,
1507 struct scrub_page
*spage
= sblock
->pagev
[page_num
];
1509 BUG_ON(spage
->page
== NULL
);
1510 if (spage
->io_error
) {
1511 void *mapped_buffer
= kmap_atomic(spage
->page
);
1513 memset(mapped_buffer
, 0, PAGE_CACHE_SIZE
);
1514 flush_dcache_page(spage
->page
);
1515 kunmap_atomic(mapped_buffer
);
1517 return scrub_add_page_to_wr_bio(sblock
->sctx
, spage
);
1520 static int scrub_add_page_to_wr_bio(struct scrub_ctx
*sctx
,
1521 struct scrub_page
*spage
)
1523 struct scrub_wr_ctx
*wr_ctx
= &sctx
->wr_ctx
;
1524 struct scrub_bio
*sbio
;
1527 mutex_lock(&wr_ctx
->wr_lock
);
1529 if (!wr_ctx
->wr_curr_bio
) {
1530 wr_ctx
->wr_curr_bio
= kzalloc(sizeof(*wr_ctx
->wr_curr_bio
),
1532 if (!wr_ctx
->wr_curr_bio
) {
1533 mutex_unlock(&wr_ctx
->wr_lock
);
1536 wr_ctx
->wr_curr_bio
->sctx
= sctx
;
1537 wr_ctx
->wr_curr_bio
->page_count
= 0;
1539 sbio
= wr_ctx
->wr_curr_bio
;
1540 if (sbio
->page_count
== 0) {
1543 sbio
->physical
= spage
->physical_for_dev_replace
;
1544 sbio
->logical
= spage
->logical
;
1545 sbio
->dev
= wr_ctx
->tgtdev
;
1548 bio
= btrfs_io_bio_alloc(GFP_NOFS
, wr_ctx
->pages_per_wr_bio
);
1550 mutex_unlock(&wr_ctx
->wr_lock
);
1556 bio
->bi_private
= sbio
;
1557 bio
->bi_end_io
= scrub_wr_bio_end_io
;
1558 bio
->bi_bdev
= sbio
->dev
->bdev
;
1559 bio
->bi_iter
.bi_sector
= sbio
->physical
>> 9;
1561 } else if (sbio
->physical
+ sbio
->page_count
* PAGE_SIZE
!=
1562 spage
->physical_for_dev_replace
||
1563 sbio
->logical
+ sbio
->page_count
* PAGE_SIZE
!=
1565 scrub_wr_submit(sctx
);
1569 ret
= bio_add_page(sbio
->bio
, spage
->page
, PAGE_SIZE
, 0);
1570 if (ret
!= PAGE_SIZE
) {
1571 if (sbio
->page_count
< 1) {
1574 mutex_unlock(&wr_ctx
->wr_lock
);
1577 scrub_wr_submit(sctx
);
1581 sbio
->pagev
[sbio
->page_count
] = spage
;
1582 scrub_page_get(spage
);
1584 if (sbio
->page_count
== wr_ctx
->pages_per_wr_bio
)
1585 scrub_wr_submit(sctx
);
1586 mutex_unlock(&wr_ctx
->wr_lock
);
1591 static void scrub_wr_submit(struct scrub_ctx
*sctx
)
1593 struct scrub_wr_ctx
*wr_ctx
= &sctx
->wr_ctx
;
1594 struct scrub_bio
*sbio
;
1596 if (!wr_ctx
->wr_curr_bio
)
1599 sbio
= wr_ctx
->wr_curr_bio
;
1600 wr_ctx
->wr_curr_bio
= NULL
;
1601 WARN_ON(!sbio
->bio
->bi_bdev
);
1602 scrub_pending_bio_inc(sctx
);
1603 /* process all writes in a single worker thread. Then the block layer
1604 * orders the requests before sending them to the driver which
1605 * doubled the write performance on spinning disks when measured
1607 btrfsic_submit_bio(WRITE
, sbio
->bio
);
1610 static void scrub_wr_bio_end_io(struct bio
*bio
, int err
)
1612 struct scrub_bio
*sbio
= bio
->bi_private
;
1613 struct btrfs_fs_info
*fs_info
= sbio
->dev
->dev_root
->fs_info
;
1618 btrfs_init_work(&sbio
->work
, scrub_wr_bio_end_io_worker
, NULL
, NULL
);
1619 btrfs_queue_work(fs_info
->scrub_wr_completion_workers
, &sbio
->work
);
1622 static void scrub_wr_bio_end_io_worker(struct btrfs_work
*work
)
1624 struct scrub_bio
*sbio
= container_of(work
, struct scrub_bio
, work
);
1625 struct scrub_ctx
*sctx
= sbio
->sctx
;
1628 WARN_ON(sbio
->page_count
> SCRUB_PAGES_PER_WR_BIO
);
1630 struct btrfs_dev_replace
*dev_replace
=
1631 &sbio
->sctx
->dev_root
->fs_info
->dev_replace
;
1633 for (i
= 0; i
< sbio
->page_count
; i
++) {
1634 struct scrub_page
*spage
= sbio
->pagev
[i
];
1636 spage
->io_error
= 1;
1637 btrfs_dev_replace_stats_inc(&dev_replace
->
1642 for (i
= 0; i
< sbio
->page_count
; i
++)
1643 scrub_page_put(sbio
->pagev
[i
]);
1647 scrub_pending_bio_dec(sctx
);
1650 static int scrub_checksum(struct scrub_block
*sblock
)
1655 WARN_ON(sblock
->page_count
< 1);
1656 flags
= sblock
->pagev
[0]->flags
;
1658 if (flags
& BTRFS_EXTENT_FLAG_DATA
)
1659 ret
= scrub_checksum_data(sblock
);
1660 else if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
)
1661 ret
= scrub_checksum_tree_block(sblock
);
1662 else if (flags
& BTRFS_EXTENT_FLAG_SUPER
)
1663 (void)scrub_checksum_super(sblock
);
1667 scrub_handle_errored_block(sblock
);
1672 static int scrub_checksum_data(struct scrub_block
*sblock
)
1674 struct scrub_ctx
*sctx
= sblock
->sctx
;
1675 u8 csum
[BTRFS_CSUM_SIZE
];
1684 BUG_ON(sblock
->page_count
< 1);
1685 if (!sblock
->pagev
[0]->have_csum
)
1688 on_disk_csum
= sblock
->pagev
[0]->csum
;
1689 page
= sblock
->pagev
[0]->page
;
1690 buffer
= kmap_atomic(page
);
1692 len
= sctx
->sectorsize
;
1695 u64 l
= min_t(u64
, len
, PAGE_SIZE
);
1697 crc
= btrfs_csum_data(buffer
, crc
, l
);
1698 kunmap_atomic(buffer
);
1703 BUG_ON(index
>= sblock
->page_count
);
1704 BUG_ON(!sblock
->pagev
[index
]->page
);
1705 page
= sblock
->pagev
[index
]->page
;
1706 buffer
= kmap_atomic(page
);
1709 btrfs_csum_final(crc
, csum
);
1710 if (memcmp(csum
, on_disk_csum
, sctx
->csum_size
))
1716 static int scrub_checksum_tree_block(struct scrub_block
*sblock
)
1718 struct scrub_ctx
*sctx
= sblock
->sctx
;
1719 struct btrfs_header
*h
;
1720 struct btrfs_root
*root
= sctx
->dev_root
;
1721 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1722 u8 calculated_csum
[BTRFS_CSUM_SIZE
];
1723 u8 on_disk_csum
[BTRFS_CSUM_SIZE
];
1725 void *mapped_buffer
;
1734 BUG_ON(sblock
->page_count
< 1);
1735 page
= sblock
->pagev
[0]->page
;
1736 mapped_buffer
= kmap_atomic(page
);
1737 h
= (struct btrfs_header
*)mapped_buffer
;
1738 memcpy(on_disk_csum
, h
->csum
, sctx
->csum_size
);
1741 * we don't use the getter functions here, as we
1742 * a) don't have an extent buffer and
1743 * b) the page is already kmapped
1746 if (sblock
->pagev
[0]->logical
!= btrfs_stack_header_bytenr(h
))
1749 if (sblock
->pagev
[0]->generation
!= btrfs_stack_header_generation(h
))
1752 if (memcmp(h
->fsid
, fs_info
->fsid
, BTRFS_UUID_SIZE
))
1755 if (memcmp(h
->chunk_tree_uuid
, fs_info
->chunk_tree_uuid
,
1759 WARN_ON(sctx
->nodesize
!= sctx
->leafsize
);
1760 len
= sctx
->nodesize
- BTRFS_CSUM_SIZE
;
1761 mapped_size
= PAGE_SIZE
- BTRFS_CSUM_SIZE
;
1762 p
= ((u8
*)mapped_buffer
) + BTRFS_CSUM_SIZE
;
1765 u64 l
= min_t(u64
, len
, mapped_size
);
1767 crc
= btrfs_csum_data(p
, crc
, l
);
1768 kunmap_atomic(mapped_buffer
);
1773 BUG_ON(index
>= sblock
->page_count
);
1774 BUG_ON(!sblock
->pagev
[index
]->page
);
1775 page
= sblock
->pagev
[index
]->page
;
1776 mapped_buffer
= kmap_atomic(page
);
1777 mapped_size
= PAGE_SIZE
;
1781 btrfs_csum_final(crc
, calculated_csum
);
1782 if (memcmp(calculated_csum
, on_disk_csum
, sctx
->csum_size
))
1785 return fail
|| crc_fail
;
1788 static int scrub_checksum_super(struct scrub_block
*sblock
)
1790 struct btrfs_super_block
*s
;
1791 struct scrub_ctx
*sctx
= sblock
->sctx
;
1792 struct btrfs_root
*root
= sctx
->dev_root
;
1793 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1794 u8 calculated_csum
[BTRFS_CSUM_SIZE
];
1795 u8 on_disk_csum
[BTRFS_CSUM_SIZE
];
1797 void *mapped_buffer
;
1806 BUG_ON(sblock
->page_count
< 1);
1807 page
= sblock
->pagev
[0]->page
;
1808 mapped_buffer
= kmap_atomic(page
);
1809 s
= (struct btrfs_super_block
*)mapped_buffer
;
1810 memcpy(on_disk_csum
, s
->csum
, sctx
->csum_size
);
1812 if (sblock
->pagev
[0]->logical
!= btrfs_super_bytenr(s
))
1815 if (sblock
->pagev
[0]->generation
!= btrfs_super_generation(s
))
1818 if (memcmp(s
->fsid
, fs_info
->fsid
, BTRFS_UUID_SIZE
))
1821 len
= BTRFS_SUPER_INFO_SIZE
- BTRFS_CSUM_SIZE
;
1822 mapped_size
= PAGE_SIZE
- BTRFS_CSUM_SIZE
;
1823 p
= ((u8
*)mapped_buffer
) + BTRFS_CSUM_SIZE
;
1826 u64 l
= min_t(u64
, len
, mapped_size
);
1828 crc
= btrfs_csum_data(p
, crc
, l
);
1829 kunmap_atomic(mapped_buffer
);
1834 BUG_ON(index
>= sblock
->page_count
);
1835 BUG_ON(!sblock
->pagev
[index
]->page
);
1836 page
= sblock
->pagev
[index
]->page
;
1837 mapped_buffer
= kmap_atomic(page
);
1838 mapped_size
= PAGE_SIZE
;
1842 btrfs_csum_final(crc
, calculated_csum
);
1843 if (memcmp(calculated_csum
, on_disk_csum
, sctx
->csum_size
))
1846 if (fail_cor
+ fail_gen
) {
1848 * if we find an error in a super block, we just report it.
1849 * They will get written with the next transaction commit
1852 spin_lock(&sctx
->stat_lock
);
1853 ++sctx
->stat
.super_errors
;
1854 spin_unlock(&sctx
->stat_lock
);
1856 btrfs_dev_stat_inc_and_print(sblock
->pagev
[0]->dev
,
1857 BTRFS_DEV_STAT_CORRUPTION_ERRS
);
1859 btrfs_dev_stat_inc_and_print(sblock
->pagev
[0]->dev
,
1860 BTRFS_DEV_STAT_GENERATION_ERRS
);
1863 return fail_cor
+ fail_gen
;
1866 static void scrub_block_get(struct scrub_block
*sblock
)
1868 atomic_inc(&sblock
->ref_count
);
1871 static void scrub_block_put(struct scrub_block
*sblock
)
1873 if (atomic_dec_and_test(&sblock
->ref_count
)) {
1876 for (i
= 0; i
< sblock
->page_count
; i
++)
1877 scrub_page_put(sblock
->pagev
[i
]);
1882 static void scrub_page_get(struct scrub_page
*spage
)
1884 atomic_inc(&spage
->ref_count
);
1887 static void scrub_page_put(struct scrub_page
*spage
)
1889 if (atomic_dec_and_test(&spage
->ref_count
)) {
1891 __free_page(spage
->page
);
1896 static void scrub_submit(struct scrub_ctx
*sctx
)
1898 struct scrub_bio
*sbio
;
1900 if (sctx
->curr
== -1)
1903 sbio
= sctx
->bios
[sctx
->curr
];
1905 scrub_pending_bio_inc(sctx
);
1907 if (!sbio
->bio
->bi_bdev
) {
1909 * this case should not happen. If btrfs_map_block() is
1910 * wrong, it could happen for dev-replace operations on
1911 * missing devices when no mirrors are available, but in
1912 * this case it should already fail the mount.
1913 * This case is handled correctly (but _very_ slowly).
1915 printk_ratelimited(KERN_WARNING
1916 "BTRFS: scrub_submit(bio bdev == NULL) is unexpected!\n");
1917 bio_endio(sbio
->bio
, -EIO
);
1919 btrfsic_submit_bio(READ
, sbio
->bio
);
1923 static int scrub_add_page_to_rd_bio(struct scrub_ctx
*sctx
,
1924 struct scrub_page
*spage
)
1926 struct scrub_block
*sblock
= spage
->sblock
;
1927 struct scrub_bio
*sbio
;
1932 * grab a fresh bio or wait for one to become available
1934 while (sctx
->curr
== -1) {
1935 spin_lock(&sctx
->list_lock
);
1936 sctx
->curr
= sctx
->first_free
;
1937 if (sctx
->curr
!= -1) {
1938 sctx
->first_free
= sctx
->bios
[sctx
->curr
]->next_free
;
1939 sctx
->bios
[sctx
->curr
]->next_free
= -1;
1940 sctx
->bios
[sctx
->curr
]->page_count
= 0;
1941 spin_unlock(&sctx
->list_lock
);
1943 spin_unlock(&sctx
->list_lock
);
1944 wait_event(sctx
->list_wait
, sctx
->first_free
!= -1);
1947 sbio
= sctx
->bios
[sctx
->curr
];
1948 if (sbio
->page_count
== 0) {
1951 sbio
->physical
= spage
->physical
;
1952 sbio
->logical
= spage
->logical
;
1953 sbio
->dev
= spage
->dev
;
1956 bio
= btrfs_io_bio_alloc(GFP_NOFS
, sctx
->pages_per_rd_bio
);
1962 bio
->bi_private
= sbio
;
1963 bio
->bi_end_io
= scrub_bio_end_io
;
1964 bio
->bi_bdev
= sbio
->dev
->bdev
;
1965 bio
->bi_iter
.bi_sector
= sbio
->physical
>> 9;
1967 } else if (sbio
->physical
+ sbio
->page_count
* PAGE_SIZE
!=
1969 sbio
->logical
+ sbio
->page_count
* PAGE_SIZE
!=
1971 sbio
->dev
!= spage
->dev
) {
1976 sbio
->pagev
[sbio
->page_count
] = spage
;
1977 ret
= bio_add_page(sbio
->bio
, spage
->page
, PAGE_SIZE
, 0);
1978 if (ret
!= PAGE_SIZE
) {
1979 if (sbio
->page_count
< 1) {
1988 scrub_block_get(sblock
); /* one for the page added to the bio */
1989 atomic_inc(&sblock
->outstanding_pages
);
1991 if (sbio
->page_count
== sctx
->pages_per_rd_bio
)
1997 static int scrub_pages(struct scrub_ctx
*sctx
, u64 logical
, u64 len
,
1998 u64 physical
, struct btrfs_device
*dev
, u64 flags
,
1999 u64 gen
, int mirror_num
, u8
*csum
, int force
,
2000 u64 physical_for_dev_replace
)
2002 struct scrub_block
*sblock
;
2005 sblock
= kzalloc(sizeof(*sblock
), GFP_NOFS
);
2007 spin_lock(&sctx
->stat_lock
);
2008 sctx
->stat
.malloc_errors
++;
2009 spin_unlock(&sctx
->stat_lock
);
2013 /* one ref inside this function, plus one for each page added to
2015 atomic_set(&sblock
->ref_count
, 1);
2016 sblock
->sctx
= sctx
;
2017 sblock
->no_io_error_seen
= 1;
2019 for (index
= 0; len
> 0; index
++) {
2020 struct scrub_page
*spage
;
2021 u64 l
= min_t(u64
, len
, PAGE_SIZE
);
2023 spage
= kzalloc(sizeof(*spage
), GFP_NOFS
);
2026 spin_lock(&sctx
->stat_lock
);
2027 sctx
->stat
.malloc_errors
++;
2028 spin_unlock(&sctx
->stat_lock
);
2029 scrub_block_put(sblock
);
2032 BUG_ON(index
>= SCRUB_MAX_PAGES_PER_BLOCK
);
2033 scrub_page_get(spage
);
2034 sblock
->pagev
[index
] = spage
;
2035 spage
->sblock
= sblock
;
2037 spage
->flags
= flags
;
2038 spage
->generation
= gen
;
2039 spage
->logical
= logical
;
2040 spage
->physical
= physical
;
2041 spage
->physical_for_dev_replace
= physical_for_dev_replace
;
2042 spage
->mirror_num
= mirror_num
;
2044 spage
->have_csum
= 1;
2045 memcpy(spage
->csum
, csum
, sctx
->csum_size
);
2047 spage
->have_csum
= 0;
2049 sblock
->page_count
++;
2050 spage
->page
= alloc_page(GFP_NOFS
);
2056 physical_for_dev_replace
+= l
;
2059 WARN_ON(sblock
->page_count
== 0);
2060 for (index
= 0; index
< sblock
->page_count
; index
++) {
2061 struct scrub_page
*spage
= sblock
->pagev
[index
];
2064 ret
= scrub_add_page_to_rd_bio(sctx
, spage
);
2066 scrub_block_put(sblock
);
2074 /* last one frees, either here or in bio completion for last page */
2075 scrub_block_put(sblock
);
2079 static void scrub_bio_end_io(struct bio
*bio
, int err
)
2081 struct scrub_bio
*sbio
= bio
->bi_private
;
2082 struct btrfs_fs_info
*fs_info
= sbio
->dev
->dev_root
->fs_info
;
2087 btrfs_queue_work(fs_info
->scrub_workers
, &sbio
->work
);
2090 static void scrub_bio_end_io_worker(struct btrfs_work
*work
)
2092 struct scrub_bio
*sbio
= container_of(work
, struct scrub_bio
, work
);
2093 struct scrub_ctx
*sctx
= sbio
->sctx
;
2096 BUG_ON(sbio
->page_count
> SCRUB_PAGES_PER_RD_BIO
);
2098 for (i
= 0; i
< sbio
->page_count
; i
++) {
2099 struct scrub_page
*spage
= sbio
->pagev
[i
];
2101 spage
->io_error
= 1;
2102 spage
->sblock
->no_io_error_seen
= 0;
2106 /* now complete the scrub_block items that have all pages completed */
2107 for (i
= 0; i
< sbio
->page_count
; i
++) {
2108 struct scrub_page
*spage
= sbio
->pagev
[i
];
2109 struct scrub_block
*sblock
= spage
->sblock
;
2111 if (atomic_dec_and_test(&sblock
->outstanding_pages
))
2112 scrub_block_complete(sblock
);
2113 scrub_block_put(sblock
);
2118 spin_lock(&sctx
->list_lock
);
2119 sbio
->next_free
= sctx
->first_free
;
2120 sctx
->first_free
= sbio
->index
;
2121 spin_unlock(&sctx
->list_lock
);
2123 if (sctx
->is_dev_replace
&&
2124 atomic_read(&sctx
->wr_ctx
.flush_all_writes
)) {
2125 mutex_lock(&sctx
->wr_ctx
.wr_lock
);
2126 scrub_wr_submit(sctx
);
2127 mutex_unlock(&sctx
->wr_ctx
.wr_lock
);
2130 scrub_pending_bio_dec(sctx
);
2133 static void scrub_block_complete(struct scrub_block
*sblock
)
2135 if (!sblock
->no_io_error_seen
) {
2136 scrub_handle_errored_block(sblock
);
2139 * if has checksum error, write via repair mechanism in
2140 * dev replace case, otherwise write here in dev replace
2143 if (!scrub_checksum(sblock
) && sblock
->sctx
->is_dev_replace
)
2144 scrub_write_block_to_dev_replace(sblock
);
2148 static int scrub_find_csum(struct scrub_ctx
*sctx
, u64 logical
, u64 len
,
2151 struct btrfs_ordered_sum
*sum
= NULL
;
2152 unsigned long index
;
2153 unsigned long num_sectors
;
2155 while (!list_empty(&sctx
->csum_list
)) {
2156 sum
= list_first_entry(&sctx
->csum_list
,
2157 struct btrfs_ordered_sum
, list
);
2158 if (sum
->bytenr
> logical
)
2160 if (sum
->bytenr
+ sum
->len
> logical
)
2163 ++sctx
->stat
.csum_discards
;
2164 list_del(&sum
->list
);
2171 index
= ((u32
)(logical
- sum
->bytenr
)) / sctx
->sectorsize
;
2172 num_sectors
= sum
->len
/ sctx
->sectorsize
;
2173 memcpy(csum
, sum
->sums
+ index
, sctx
->csum_size
);
2174 if (index
== num_sectors
- 1) {
2175 list_del(&sum
->list
);
2181 /* scrub extent tries to collect up to 64 kB for each bio */
2182 static int scrub_extent(struct scrub_ctx
*sctx
, u64 logical
, u64 len
,
2183 u64 physical
, struct btrfs_device
*dev
, u64 flags
,
2184 u64 gen
, int mirror_num
, u64 physical_for_dev_replace
)
2187 u8 csum
[BTRFS_CSUM_SIZE
];
2190 if (flags
& BTRFS_EXTENT_FLAG_DATA
) {
2191 blocksize
= sctx
->sectorsize
;
2192 spin_lock(&sctx
->stat_lock
);
2193 sctx
->stat
.data_extents_scrubbed
++;
2194 sctx
->stat
.data_bytes_scrubbed
+= len
;
2195 spin_unlock(&sctx
->stat_lock
);
2196 } else if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
2197 WARN_ON(sctx
->nodesize
!= sctx
->leafsize
);
2198 blocksize
= sctx
->nodesize
;
2199 spin_lock(&sctx
->stat_lock
);
2200 sctx
->stat
.tree_extents_scrubbed
++;
2201 sctx
->stat
.tree_bytes_scrubbed
+= len
;
2202 spin_unlock(&sctx
->stat_lock
);
2204 blocksize
= sctx
->sectorsize
;
2209 u64 l
= min_t(u64
, len
, blocksize
);
2212 if (flags
& BTRFS_EXTENT_FLAG_DATA
) {
2213 /* push csums to sbio */
2214 have_csum
= scrub_find_csum(sctx
, logical
, l
, csum
);
2216 ++sctx
->stat
.no_csum
;
2217 if (sctx
->is_dev_replace
&& !have_csum
) {
2218 ret
= copy_nocow_pages(sctx
, logical
, l
,
2220 physical_for_dev_replace
);
2221 goto behind_scrub_pages
;
2224 ret
= scrub_pages(sctx
, logical
, l
, physical
, dev
, flags
, gen
,
2225 mirror_num
, have_csum
? csum
: NULL
, 0,
2226 physical_for_dev_replace
);
2233 physical_for_dev_replace
+= l
;
2238 static noinline_for_stack
int scrub_stripe(struct scrub_ctx
*sctx
,
2239 struct map_lookup
*map
,
2240 struct btrfs_device
*scrub_dev
,
2241 int num
, u64 base
, u64 length
,
2244 struct btrfs_path
*path
;
2245 struct btrfs_fs_info
*fs_info
= sctx
->dev_root
->fs_info
;
2246 struct btrfs_root
*root
= fs_info
->extent_root
;
2247 struct btrfs_root
*csum_root
= fs_info
->csum_root
;
2248 struct btrfs_extent_item
*extent
;
2249 struct blk_plug plug
;
2254 struct extent_buffer
*l
;
2255 struct btrfs_key key
;
2261 struct reada_control
*reada1
;
2262 struct reada_control
*reada2
;
2263 struct btrfs_key key_start
;
2264 struct btrfs_key key_end
;
2265 u64 increment
= map
->stripe_len
;
2268 u64 extent_physical
;
2270 struct btrfs_device
*extent_dev
;
2271 int extent_mirror_num
;
2274 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
2275 BTRFS_BLOCK_GROUP_RAID6
)) {
2276 if (num
>= nr_data_stripes(map
)) {
2283 do_div(nstripes
, map
->stripe_len
);
2284 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
2285 offset
= map
->stripe_len
* num
;
2286 increment
= map
->stripe_len
* map
->num_stripes
;
2288 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
2289 int factor
= map
->num_stripes
/ map
->sub_stripes
;
2290 offset
= map
->stripe_len
* (num
/ map
->sub_stripes
);
2291 increment
= map
->stripe_len
* factor
;
2292 mirror_num
= num
% map
->sub_stripes
+ 1;
2293 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
2294 increment
= map
->stripe_len
;
2295 mirror_num
= num
% map
->num_stripes
+ 1;
2296 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
2297 increment
= map
->stripe_len
;
2298 mirror_num
= num
% map
->num_stripes
+ 1;
2300 increment
= map
->stripe_len
;
2304 path
= btrfs_alloc_path();
2309 * work on commit root. The related disk blocks are static as
2310 * long as COW is applied. This means, it is save to rewrite
2311 * them to repair disk errors without any race conditions
2313 path
->search_commit_root
= 1;
2314 path
->skip_locking
= 1;
2317 * trigger the readahead for extent tree csum tree and wait for
2318 * completion. During readahead, the scrub is officially paused
2319 * to not hold off transaction commits
2321 logical
= base
+ offset
;
2323 wait_event(sctx
->list_wait
,
2324 atomic_read(&sctx
->bios_in_flight
) == 0);
2325 scrub_blocked_if_needed(fs_info
);
2327 /* FIXME it might be better to start readahead at commit root */
2328 key_start
.objectid
= logical
;
2329 key_start
.type
= BTRFS_EXTENT_ITEM_KEY
;
2330 key_start
.offset
= (u64
)0;
2331 key_end
.objectid
= base
+ offset
+ nstripes
* increment
;
2332 key_end
.type
= BTRFS_METADATA_ITEM_KEY
;
2333 key_end
.offset
= (u64
)-1;
2334 reada1
= btrfs_reada_add(root
, &key_start
, &key_end
);
2336 key_start
.objectid
= BTRFS_EXTENT_CSUM_OBJECTID
;
2337 key_start
.type
= BTRFS_EXTENT_CSUM_KEY
;
2338 key_start
.offset
= logical
;
2339 key_end
.objectid
= BTRFS_EXTENT_CSUM_OBJECTID
;
2340 key_end
.type
= BTRFS_EXTENT_CSUM_KEY
;
2341 key_end
.offset
= base
+ offset
+ nstripes
* increment
;
2342 reada2
= btrfs_reada_add(csum_root
, &key_start
, &key_end
);
2344 if (!IS_ERR(reada1
))
2345 btrfs_reada_wait(reada1
);
2346 if (!IS_ERR(reada2
))
2347 btrfs_reada_wait(reada2
);
2351 * collect all data csums for the stripe to avoid seeking during
2352 * the scrub. This might currently (crc32) end up to be about 1MB
2354 blk_start_plug(&plug
);
2357 * now find all extents for each stripe and scrub them
2359 logical
= base
+ offset
;
2360 physical
= map
->stripes
[num
].physical
;
2361 logic_end
= logical
+ increment
* nstripes
;
2363 while (logical
< logic_end
) {
2367 if (atomic_read(&fs_info
->scrub_cancel_req
) ||
2368 atomic_read(&sctx
->cancel_req
)) {
2373 * check to see if we have to pause
2375 if (atomic_read(&fs_info
->scrub_pause_req
)) {
2376 /* push queued extents */
2377 atomic_set(&sctx
->wr_ctx
.flush_all_writes
, 1);
2379 mutex_lock(&sctx
->wr_ctx
.wr_lock
);
2380 scrub_wr_submit(sctx
);
2381 mutex_unlock(&sctx
->wr_ctx
.wr_lock
);
2382 wait_event(sctx
->list_wait
,
2383 atomic_read(&sctx
->bios_in_flight
) == 0);
2384 atomic_set(&sctx
->wr_ctx
.flush_all_writes
, 0);
2385 scrub_blocked_if_needed(fs_info
);
2388 if (btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
2389 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2391 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2392 key
.objectid
= logical
;
2393 key
.offset
= (u64
)-1;
2395 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2400 ret
= btrfs_previous_extent_item(root
, path
, 0);
2404 /* there's no smaller item, so stick with the
2406 btrfs_release_path(path
);
2407 ret
= btrfs_search_slot(NULL
, root
, &key
,
2419 slot
= path
->slots
[0];
2420 if (slot
>= btrfs_header_nritems(l
)) {
2421 ret
= btrfs_next_leaf(root
, path
);
2430 btrfs_item_key_to_cpu(l
, &key
, slot
);
2432 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
2433 bytes
= root
->leafsize
;
2437 if (key
.objectid
+ bytes
<= logical
)
2440 if (key
.type
!= BTRFS_EXTENT_ITEM_KEY
&&
2441 key
.type
!= BTRFS_METADATA_ITEM_KEY
)
2444 if (key
.objectid
>= logical
+ map
->stripe_len
) {
2445 /* out of this device extent */
2446 if (key
.objectid
>= logic_end
)
2451 extent
= btrfs_item_ptr(l
, slot
,
2452 struct btrfs_extent_item
);
2453 flags
= btrfs_extent_flags(l
, extent
);
2454 generation
= btrfs_extent_generation(l
, extent
);
2456 if (key
.objectid
< logical
&&
2457 (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
)) {
2459 "scrub: tree block %llu spanning "
2460 "stripes, ignored. logical=%llu",
2461 key
.objectid
, logical
);
2466 extent_logical
= key
.objectid
;
2470 * trim extent to this stripe
2472 if (extent_logical
< logical
) {
2473 extent_len
-= logical
- extent_logical
;
2474 extent_logical
= logical
;
2476 if (extent_logical
+ extent_len
>
2477 logical
+ map
->stripe_len
) {
2478 extent_len
= logical
+ map
->stripe_len
-
2482 extent_physical
= extent_logical
- logical
+ physical
;
2483 extent_dev
= scrub_dev
;
2484 extent_mirror_num
= mirror_num
;
2486 scrub_remap_extent(fs_info
, extent_logical
,
2487 extent_len
, &extent_physical
,
2489 &extent_mirror_num
);
2491 ret
= btrfs_lookup_csums_range(csum_root
, logical
,
2492 logical
+ map
->stripe_len
- 1,
2493 &sctx
->csum_list
, 1);
2497 ret
= scrub_extent(sctx
, extent_logical
, extent_len
,
2498 extent_physical
, extent_dev
, flags
,
2499 generation
, extent_mirror_num
,
2500 extent_logical
- logical
+ physical
);
2504 scrub_free_csums(sctx
);
2505 if (extent_logical
+ extent_len
<
2506 key
.objectid
+ bytes
) {
2507 logical
+= increment
;
2508 physical
+= map
->stripe_len
;
2510 if (logical
< key
.objectid
+ bytes
) {
2515 if (logical
>= logic_end
) {
2523 btrfs_release_path(path
);
2524 logical
+= increment
;
2525 physical
+= map
->stripe_len
;
2526 spin_lock(&sctx
->stat_lock
);
2528 sctx
->stat
.last_physical
= map
->stripes
[num
].physical
+
2531 sctx
->stat
.last_physical
= physical
;
2532 spin_unlock(&sctx
->stat_lock
);
2537 /* push queued extents */
2539 mutex_lock(&sctx
->wr_ctx
.wr_lock
);
2540 scrub_wr_submit(sctx
);
2541 mutex_unlock(&sctx
->wr_ctx
.wr_lock
);
2543 blk_finish_plug(&plug
);
2544 btrfs_free_path(path
);
2545 return ret
< 0 ? ret
: 0;
2548 static noinline_for_stack
int scrub_chunk(struct scrub_ctx
*sctx
,
2549 struct btrfs_device
*scrub_dev
,
2550 u64 chunk_tree
, u64 chunk_objectid
,
2551 u64 chunk_offset
, u64 length
,
2552 u64 dev_offset
, int is_dev_replace
)
2554 struct btrfs_mapping_tree
*map_tree
=
2555 &sctx
->dev_root
->fs_info
->mapping_tree
;
2556 struct map_lookup
*map
;
2557 struct extent_map
*em
;
2561 read_lock(&map_tree
->map_tree
.lock
);
2562 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
2563 read_unlock(&map_tree
->map_tree
.lock
);
2568 map
= (struct map_lookup
*)em
->bdev
;
2569 if (em
->start
!= chunk_offset
)
2572 if (em
->len
< length
)
2575 for (i
= 0; i
< map
->num_stripes
; ++i
) {
2576 if (map
->stripes
[i
].dev
->bdev
== scrub_dev
->bdev
&&
2577 map
->stripes
[i
].physical
== dev_offset
) {
2578 ret
= scrub_stripe(sctx
, map
, scrub_dev
, i
,
2579 chunk_offset
, length
,
2586 free_extent_map(em
);
2591 static noinline_for_stack
2592 int scrub_enumerate_chunks(struct scrub_ctx
*sctx
,
2593 struct btrfs_device
*scrub_dev
, u64 start
, u64 end
,
2596 struct btrfs_dev_extent
*dev_extent
= NULL
;
2597 struct btrfs_path
*path
;
2598 struct btrfs_root
*root
= sctx
->dev_root
;
2599 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2606 struct extent_buffer
*l
;
2607 struct btrfs_key key
;
2608 struct btrfs_key found_key
;
2609 struct btrfs_block_group_cache
*cache
;
2610 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
2612 path
= btrfs_alloc_path();
2617 path
->search_commit_root
= 1;
2618 path
->skip_locking
= 1;
2620 key
.objectid
= scrub_dev
->devid
;
2622 key
.type
= BTRFS_DEV_EXTENT_KEY
;
2625 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2629 if (path
->slots
[0] >=
2630 btrfs_header_nritems(path
->nodes
[0])) {
2631 ret
= btrfs_next_leaf(root
, path
);
2638 slot
= path
->slots
[0];
2640 btrfs_item_key_to_cpu(l
, &found_key
, slot
);
2642 if (found_key
.objectid
!= scrub_dev
->devid
)
2645 if (btrfs_key_type(&found_key
) != BTRFS_DEV_EXTENT_KEY
)
2648 if (found_key
.offset
>= end
)
2651 if (found_key
.offset
< key
.offset
)
2654 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
2655 length
= btrfs_dev_extent_length(l
, dev_extent
);
2657 if (found_key
.offset
+ length
<= start
) {
2658 key
.offset
= found_key
.offset
+ length
;
2659 btrfs_release_path(path
);
2663 chunk_tree
= btrfs_dev_extent_chunk_tree(l
, dev_extent
);
2664 chunk_objectid
= btrfs_dev_extent_chunk_objectid(l
, dev_extent
);
2665 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
2668 * get a reference on the corresponding block group to prevent
2669 * the chunk from going away while we scrub it
2671 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
2676 dev_replace
->cursor_right
= found_key
.offset
+ length
;
2677 dev_replace
->cursor_left
= found_key
.offset
;
2678 dev_replace
->item_needs_writeback
= 1;
2679 ret
= scrub_chunk(sctx
, scrub_dev
, chunk_tree
, chunk_objectid
,
2680 chunk_offset
, length
, found_key
.offset
,
2684 * flush, submit all pending read and write bios, afterwards
2686 * Note that in the dev replace case, a read request causes
2687 * write requests that are submitted in the read completion
2688 * worker. Therefore in the current situation, it is required
2689 * that all write requests are flushed, so that all read and
2690 * write requests are really completed when bios_in_flight
2693 atomic_set(&sctx
->wr_ctx
.flush_all_writes
, 1);
2695 mutex_lock(&sctx
->wr_ctx
.wr_lock
);
2696 scrub_wr_submit(sctx
);
2697 mutex_unlock(&sctx
->wr_ctx
.wr_lock
);
2699 wait_event(sctx
->list_wait
,
2700 atomic_read(&sctx
->bios_in_flight
) == 0);
2701 atomic_inc(&fs_info
->scrubs_paused
);
2702 wake_up(&fs_info
->scrub_pause_wait
);
2705 * must be called before we decrease @scrub_paused.
2706 * make sure we don't block transaction commit while
2707 * we are waiting pending workers finished.
2709 wait_event(sctx
->list_wait
,
2710 atomic_read(&sctx
->workers_pending
) == 0);
2711 atomic_set(&sctx
->wr_ctx
.flush_all_writes
, 0);
2713 mutex_lock(&fs_info
->scrub_lock
);
2714 __scrub_blocked_if_needed(fs_info
);
2715 atomic_dec(&fs_info
->scrubs_paused
);
2716 mutex_unlock(&fs_info
->scrub_lock
);
2717 wake_up(&fs_info
->scrub_pause_wait
);
2719 btrfs_put_block_group(cache
);
2722 if (is_dev_replace
&&
2723 atomic64_read(&dev_replace
->num_write_errors
) > 0) {
2727 if (sctx
->stat
.malloc_errors
> 0) {
2732 dev_replace
->cursor_left
= dev_replace
->cursor_right
;
2733 dev_replace
->item_needs_writeback
= 1;
2735 key
.offset
= found_key
.offset
+ length
;
2736 btrfs_release_path(path
);
2739 btrfs_free_path(path
);
2742 * ret can still be 1 from search_slot or next_leaf,
2743 * that's not an error
2745 return ret
< 0 ? ret
: 0;
2748 static noinline_for_stack
int scrub_supers(struct scrub_ctx
*sctx
,
2749 struct btrfs_device
*scrub_dev
)
2755 struct btrfs_root
*root
= sctx
->dev_root
;
2757 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
2760 gen
= root
->fs_info
->last_trans_committed
;
2762 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
2763 bytenr
= btrfs_sb_offset(i
);
2764 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
> scrub_dev
->total_bytes
)
2767 ret
= scrub_pages(sctx
, bytenr
, BTRFS_SUPER_INFO_SIZE
, bytenr
,
2768 scrub_dev
, BTRFS_EXTENT_FLAG_SUPER
, gen
, i
,
2773 wait_event(sctx
->list_wait
, atomic_read(&sctx
->bios_in_flight
) == 0);
2779 * get a reference count on fs_info->scrub_workers. start worker if necessary
2781 static noinline_for_stack
int scrub_workers_get(struct btrfs_fs_info
*fs_info
,
2785 int flags
= WQ_FREEZABLE
| WQ_UNBOUND
;
2786 int max_active
= fs_info
->thread_pool_size
;
2788 if (fs_info
->scrub_workers_refcnt
== 0) {
2790 fs_info
->scrub_workers
=
2791 btrfs_alloc_workqueue("btrfs-scrub", flags
,
2794 fs_info
->scrub_workers
=
2795 btrfs_alloc_workqueue("btrfs-scrub", flags
,
2797 if (!fs_info
->scrub_workers
) {
2801 fs_info
->scrub_wr_completion_workers
=
2802 btrfs_alloc_workqueue("btrfs-scrubwrc", flags
,
2804 if (!fs_info
->scrub_wr_completion_workers
) {
2808 fs_info
->scrub_nocow_workers
=
2809 btrfs_alloc_workqueue("btrfs-scrubnc", flags
, 1, 0);
2810 if (!fs_info
->scrub_nocow_workers
) {
2815 ++fs_info
->scrub_workers_refcnt
;
2820 static noinline_for_stack
void scrub_workers_put(struct btrfs_fs_info
*fs_info
)
2822 if (--fs_info
->scrub_workers_refcnt
== 0) {
2823 btrfs_destroy_workqueue(fs_info
->scrub_workers
);
2824 btrfs_destroy_workqueue(fs_info
->scrub_wr_completion_workers
);
2825 btrfs_destroy_workqueue(fs_info
->scrub_nocow_workers
);
2827 WARN_ON(fs_info
->scrub_workers_refcnt
< 0);
2830 int btrfs_scrub_dev(struct btrfs_fs_info
*fs_info
, u64 devid
, u64 start
,
2831 u64 end
, struct btrfs_scrub_progress
*progress
,
2832 int readonly
, int is_dev_replace
)
2834 struct scrub_ctx
*sctx
;
2836 struct btrfs_device
*dev
;
2838 if (btrfs_fs_closing(fs_info
))
2842 * check some assumptions
2844 if (fs_info
->chunk_root
->nodesize
!= fs_info
->chunk_root
->leafsize
) {
2846 "scrub: size assumption nodesize == leafsize (%d == %d) fails",
2847 fs_info
->chunk_root
->nodesize
,
2848 fs_info
->chunk_root
->leafsize
);
2852 if (fs_info
->chunk_root
->nodesize
> BTRFS_STRIPE_LEN
) {
2854 * in this case scrub is unable to calculate the checksum
2855 * the way scrub is implemented. Do not handle this
2856 * situation at all because it won't ever happen.
2859 "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails",
2860 fs_info
->chunk_root
->nodesize
, BTRFS_STRIPE_LEN
);
2864 if (fs_info
->chunk_root
->sectorsize
!= PAGE_SIZE
) {
2865 /* not supported for data w/o checksums */
2867 "scrub: size assumption sectorsize != PAGE_SIZE "
2868 "(%d != %lu) fails",
2869 fs_info
->chunk_root
->sectorsize
, PAGE_SIZE
);
2873 if (fs_info
->chunk_root
->nodesize
>
2874 PAGE_SIZE
* SCRUB_MAX_PAGES_PER_BLOCK
||
2875 fs_info
->chunk_root
->sectorsize
>
2876 PAGE_SIZE
* SCRUB_MAX_PAGES_PER_BLOCK
) {
2878 * would exhaust the array bounds of pagev member in
2879 * struct scrub_block
2881 btrfs_err(fs_info
, "scrub: size assumption nodesize and sectorsize "
2882 "<= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails",
2883 fs_info
->chunk_root
->nodesize
,
2884 SCRUB_MAX_PAGES_PER_BLOCK
,
2885 fs_info
->chunk_root
->sectorsize
,
2886 SCRUB_MAX_PAGES_PER_BLOCK
);
2891 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2892 dev
= btrfs_find_device(fs_info
, devid
, NULL
, NULL
);
2893 if (!dev
|| (dev
->missing
&& !is_dev_replace
)) {
2894 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2898 mutex_lock(&fs_info
->scrub_lock
);
2899 if (!dev
->in_fs_metadata
|| dev
->is_tgtdev_for_dev_replace
) {
2900 mutex_unlock(&fs_info
->scrub_lock
);
2901 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2905 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
2906 if (dev
->scrub_device
||
2908 btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
))) {
2909 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
2910 mutex_unlock(&fs_info
->scrub_lock
);
2911 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2912 return -EINPROGRESS
;
2914 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
2916 ret
= scrub_workers_get(fs_info
, is_dev_replace
);
2918 mutex_unlock(&fs_info
->scrub_lock
);
2919 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2923 sctx
= scrub_setup_ctx(dev
, is_dev_replace
);
2925 mutex_unlock(&fs_info
->scrub_lock
);
2926 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2927 scrub_workers_put(fs_info
);
2928 return PTR_ERR(sctx
);
2930 sctx
->readonly
= readonly
;
2931 dev
->scrub_device
= sctx
;
2932 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2935 * checking @scrub_pause_req here, we can avoid
2936 * race between committing transaction and scrubbing.
2938 __scrub_blocked_if_needed(fs_info
);
2939 atomic_inc(&fs_info
->scrubs_running
);
2940 mutex_unlock(&fs_info
->scrub_lock
);
2942 if (!is_dev_replace
) {
2944 * by holding device list mutex, we can
2945 * kick off writing super in log tree sync.
2947 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2948 ret
= scrub_supers(sctx
, dev
);
2949 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2953 ret
= scrub_enumerate_chunks(sctx
, dev
, start
, end
,
2956 wait_event(sctx
->list_wait
, atomic_read(&sctx
->bios_in_flight
) == 0);
2957 atomic_dec(&fs_info
->scrubs_running
);
2958 wake_up(&fs_info
->scrub_pause_wait
);
2960 wait_event(sctx
->list_wait
, atomic_read(&sctx
->workers_pending
) == 0);
2963 memcpy(progress
, &sctx
->stat
, sizeof(*progress
));
2965 mutex_lock(&fs_info
->scrub_lock
);
2966 dev
->scrub_device
= NULL
;
2967 scrub_workers_put(fs_info
);
2968 mutex_unlock(&fs_info
->scrub_lock
);
2970 scrub_free_ctx(sctx
);
2975 void btrfs_scrub_pause(struct btrfs_root
*root
)
2977 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2979 mutex_lock(&fs_info
->scrub_lock
);
2980 atomic_inc(&fs_info
->scrub_pause_req
);
2981 while (atomic_read(&fs_info
->scrubs_paused
) !=
2982 atomic_read(&fs_info
->scrubs_running
)) {
2983 mutex_unlock(&fs_info
->scrub_lock
);
2984 wait_event(fs_info
->scrub_pause_wait
,
2985 atomic_read(&fs_info
->scrubs_paused
) ==
2986 atomic_read(&fs_info
->scrubs_running
));
2987 mutex_lock(&fs_info
->scrub_lock
);
2989 mutex_unlock(&fs_info
->scrub_lock
);
2992 void btrfs_scrub_continue(struct btrfs_root
*root
)
2994 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2996 atomic_dec(&fs_info
->scrub_pause_req
);
2997 wake_up(&fs_info
->scrub_pause_wait
);
3000 int btrfs_scrub_cancel(struct btrfs_fs_info
*fs_info
)
3002 mutex_lock(&fs_info
->scrub_lock
);
3003 if (!atomic_read(&fs_info
->scrubs_running
)) {
3004 mutex_unlock(&fs_info
->scrub_lock
);
3008 atomic_inc(&fs_info
->scrub_cancel_req
);
3009 while (atomic_read(&fs_info
->scrubs_running
)) {
3010 mutex_unlock(&fs_info
->scrub_lock
);
3011 wait_event(fs_info
->scrub_pause_wait
,
3012 atomic_read(&fs_info
->scrubs_running
) == 0);
3013 mutex_lock(&fs_info
->scrub_lock
);
3015 atomic_dec(&fs_info
->scrub_cancel_req
);
3016 mutex_unlock(&fs_info
->scrub_lock
);
3021 int btrfs_scrub_cancel_dev(struct btrfs_fs_info
*fs_info
,
3022 struct btrfs_device
*dev
)
3024 struct scrub_ctx
*sctx
;
3026 mutex_lock(&fs_info
->scrub_lock
);
3027 sctx
= dev
->scrub_device
;
3029 mutex_unlock(&fs_info
->scrub_lock
);
3032 atomic_inc(&sctx
->cancel_req
);
3033 while (dev
->scrub_device
) {
3034 mutex_unlock(&fs_info
->scrub_lock
);
3035 wait_event(fs_info
->scrub_pause_wait
,
3036 dev
->scrub_device
== NULL
);
3037 mutex_lock(&fs_info
->scrub_lock
);
3039 mutex_unlock(&fs_info
->scrub_lock
);
3044 int btrfs_scrub_progress(struct btrfs_root
*root
, u64 devid
,
3045 struct btrfs_scrub_progress
*progress
)
3047 struct btrfs_device
*dev
;
3048 struct scrub_ctx
*sctx
= NULL
;
3050 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3051 dev
= btrfs_find_device(root
->fs_info
, devid
, NULL
, NULL
);
3053 sctx
= dev
->scrub_device
;
3055 memcpy(progress
, &sctx
->stat
, sizeof(*progress
));
3056 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
3058 return dev
? (sctx
? 0 : -ENOTCONN
) : -ENODEV
;
3061 static void scrub_remap_extent(struct btrfs_fs_info
*fs_info
,
3062 u64 extent_logical
, u64 extent_len
,
3063 u64
*extent_physical
,
3064 struct btrfs_device
**extent_dev
,
3065 int *extent_mirror_num
)
3068 struct btrfs_bio
*bbio
= NULL
;
3071 mapped_length
= extent_len
;
3072 ret
= btrfs_map_block(fs_info
, READ
, extent_logical
,
3073 &mapped_length
, &bbio
, 0);
3074 if (ret
|| !bbio
|| mapped_length
< extent_len
||
3075 !bbio
->stripes
[0].dev
->bdev
) {
3080 *extent_physical
= bbio
->stripes
[0].physical
;
3081 *extent_mirror_num
= bbio
->mirror_num
;
3082 *extent_dev
= bbio
->stripes
[0].dev
;
3086 static int scrub_setup_wr_ctx(struct scrub_ctx
*sctx
,
3087 struct scrub_wr_ctx
*wr_ctx
,
3088 struct btrfs_fs_info
*fs_info
,
3089 struct btrfs_device
*dev
,
3092 WARN_ON(wr_ctx
->wr_curr_bio
!= NULL
);
3094 mutex_init(&wr_ctx
->wr_lock
);
3095 wr_ctx
->wr_curr_bio
= NULL
;
3096 if (!is_dev_replace
)
3099 WARN_ON(!dev
->bdev
);
3100 wr_ctx
->pages_per_wr_bio
= min_t(int, SCRUB_PAGES_PER_WR_BIO
,
3101 bio_get_nr_vecs(dev
->bdev
));
3102 wr_ctx
->tgtdev
= dev
;
3103 atomic_set(&wr_ctx
->flush_all_writes
, 0);
3107 static void scrub_free_wr_ctx(struct scrub_wr_ctx
*wr_ctx
)
3109 mutex_lock(&wr_ctx
->wr_lock
);
3110 kfree(wr_ctx
->wr_curr_bio
);
3111 wr_ctx
->wr_curr_bio
= NULL
;
3112 mutex_unlock(&wr_ctx
->wr_lock
);
3115 static int copy_nocow_pages(struct scrub_ctx
*sctx
, u64 logical
, u64 len
,
3116 int mirror_num
, u64 physical_for_dev_replace
)
3118 struct scrub_copy_nocow_ctx
*nocow_ctx
;
3119 struct btrfs_fs_info
*fs_info
= sctx
->dev_root
->fs_info
;
3121 nocow_ctx
= kzalloc(sizeof(*nocow_ctx
), GFP_NOFS
);
3123 spin_lock(&sctx
->stat_lock
);
3124 sctx
->stat
.malloc_errors
++;
3125 spin_unlock(&sctx
->stat_lock
);
3129 scrub_pending_trans_workers_inc(sctx
);
3131 nocow_ctx
->sctx
= sctx
;
3132 nocow_ctx
->logical
= logical
;
3133 nocow_ctx
->len
= len
;
3134 nocow_ctx
->mirror_num
= mirror_num
;
3135 nocow_ctx
->physical_for_dev_replace
= physical_for_dev_replace
;
3136 btrfs_init_work(&nocow_ctx
->work
, copy_nocow_pages_worker
, NULL
, NULL
);
3137 INIT_LIST_HEAD(&nocow_ctx
->inodes
);
3138 btrfs_queue_work(fs_info
->scrub_nocow_workers
,
3144 static int record_inode_for_nocow(u64 inum
, u64 offset
, u64 root
, void *ctx
)
3146 struct scrub_copy_nocow_ctx
*nocow_ctx
= ctx
;
3147 struct scrub_nocow_inode
*nocow_inode
;
3149 nocow_inode
= kzalloc(sizeof(*nocow_inode
), GFP_NOFS
);
3152 nocow_inode
->inum
= inum
;
3153 nocow_inode
->offset
= offset
;
3154 nocow_inode
->root
= root
;
3155 list_add_tail(&nocow_inode
->list
, &nocow_ctx
->inodes
);
3159 #define COPY_COMPLETE 1
3161 static void copy_nocow_pages_worker(struct btrfs_work
*work
)
3163 struct scrub_copy_nocow_ctx
*nocow_ctx
=
3164 container_of(work
, struct scrub_copy_nocow_ctx
, work
);
3165 struct scrub_ctx
*sctx
= nocow_ctx
->sctx
;
3166 u64 logical
= nocow_ctx
->logical
;
3167 u64 len
= nocow_ctx
->len
;
3168 int mirror_num
= nocow_ctx
->mirror_num
;
3169 u64 physical_for_dev_replace
= nocow_ctx
->physical_for_dev_replace
;
3171 struct btrfs_trans_handle
*trans
= NULL
;
3172 struct btrfs_fs_info
*fs_info
;
3173 struct btrfs_path
*path
;
3174 struct btrfs_root
*root
;
3175 int not_written
= 0;
3177 fs_info
= sctx
->dev_root
->fs_info
;
3178 root
= fs_info
->extent_root
;
3180 path
= btrfs_alloc_path();
3182 spin_lock(&sctx
->stat_lock
);
3183 sctx
->stat
.malloc_errors
++;
3184 spin_unlock(&sctx
->stat_lock
);
3189 trans
= btrfs_join_transaction(root
);
3190 if (IS_ERR(trans
)) {
3195 ret
= iterate_inodes_from_logical(logical
, fs_info
, path
,
3196 record_inode_for_nocow
, nocow_ctx
);
3197 if (ret
!= 0 && ret
!= -ENOENT
) {
3198 btrfs_warn(fs_info
, "iterate_inodes_from_logical() failed: log %llu, "
3199 "phys %llu, len %llu, mir %u, ret %d",
3200 logical
, physical_for_dev_replace
, len
, mirror_num
,
3206 btrfs_end_transaction(trans
, root
);
3208 while (!list_empty(&nocow_ctx
->inodes
)) {
3209 struct scrub_nocow_inode
*entry
;
3210 entry
= list_first_entry(&nocow_ctx
->inodes
,
3211 struct scrub_nocow_inode
,
3213 list_del_init(&entry
->list
);
3214 ret
= copy_nocow_pages_for_inode(entry
->inum
, entry
->offset
,
3215 entry
->root
, nocow_ctx
);
3217 if (ret
== COPY_COMPLETE
) {
3225 while (!list_empty(&nocow_ctx
->inodes
)) {
3226 struct scrub_nocow_inode
*entry
;
3227 entry
= list_first_entry(&nocow_ctx
->inodes
,
3228 struct scrub_nocow_inode
,
3230 list_del_init(&entry
->list
);
3233 if (trans
&& !IS_ERR(trans
))
3234 btrfs_end_transaction(trans
, root
);
3236 btrfs_dev_replace_stats_inc(&fs_info
->dev_replace
.
3237 num_uncorrectable_read_errors
);
3239 btrfs_free_path(path
);
3242 scrub_pending_trans_workers_dec(sctx
);
3245 static int copy_nocow_pages_for_inode(u64 inum
, u64 offset
, u64 root
,
3246 struct scrub_copy_nocow_ctx
*nocow_ctx
)
3248 struct btrfs_fs_info
*fs_info
= nocow_ctx
->sctx
->dev_root
->fs_info
;
3249 struct btrfs_key key
;
3250 struct inode
*inode
;
3252 struct btrfs_root
*local_root
;
3253 struct btrfs_ordered_extent
*ordered
;
3254 struct extent_map
*em
;
3255 struct extent_state
*cached_state
= NULL
;
3256 struct extent_io_tree
*io_tree
;
3257 u64 physical_for_dev_replace
;
3258 u64 len
= nocow_ctx
->len
;
3259 u64 lockstart
= offset
, lockend
= offset
+ len
- 1;
3260 unsigned long index
;
3265 key
.objectid
= root
;
3266 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3267 key
.offset
= (u64
)-1;
3269 srcu_index
= srcu_read_lock(&fs_info
->subvol_srcu
);
3271 local_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
3272 if (IS_ERR(local_root
)) {
3273 srcu_read_unlock(&fs_info
->subvol_srcu
, srcu_index
);
3274 return PTR_ERR(local_root
);
3277 key
.type
= BTRFS_INODE_ITEM_KEY
;
3278 key
.objectid
= inum
;
3280 inode
= btrfs_iget(fs_info
->sb
, &key
, local_root
, NULL
);
3281 srcu_read_unlock(&fs_info
->subvol_srcu
, srcu_index
);
3283 return PTR_ERR(inode
);
3285 /* Avoid truncate/dio/punch hole.. */
3286 mutex_lock(&inode
->i_mutex
);
3287 inode_dio_wait(inode
);
3289 physical_for_dev_replace
= nocow_ctx
->physical_for_dev_replace
;
3290 io_tree
= &BTRFS_I(inode
)->io_tree
;
3292 lock_extent_bits(io_tree
, lockstart
, lockend
, 0, &cached_state
);
3293 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
, len
);
3295 btrfs_put_ordered_extent(ordered
);
3299 em
= btrfs_get_extent(inode
, NULL
, 0, lockstart
, len
, 0);
3306 * This extent does not actually cover the logical extent anymore,
3307 * move on to the next inode.
3309 if (em
->block_start
> nocow_ctx
->logical
||
3310 em
->block_start
+ em
->block_len
< nocow_ctx
->logical
+ len
) {
3311 free_extent_map(em
);
3314 free_extent_map(em
);
3316 while (len
>= PAGE_CACHE_SIZE
) {
3317 index
= offset
>> PAGE_CACHE_SHIFT
;
3319 page
= find_or_create_page(inode
->i_mapping
, index
, GFP_NOFS
);
3321 btrfs_err(fs_info
, "find_or_create_page() failed");
3326 if (PageUptodate(page
)) {
3327 if (PageDirty(page
))
3330 ClearPageError(page
);
3331 err
= extent_read_full_page_nolock(io_tree
, page
,
3333 nocow_ctx
->mirror_num
);
3341 * If the page has been remove from the page cache,
3342 * the data on it is meaningless, because it may be
3343 * old one, the new data may be written into the new
3344 * page in the page cache.
3346 if (page
->mapping
!= inode
->i_mapping
) {
3348 page_cache_release(page
);
3351 if (!PageUptodate(page
)) {
3356 err
= write_page_nocow(nocow_ctx
->sctx
,
3357 physical_for_dev_replace
, page
);
3362 page_cache_release(page
);
3367 offset
+= PAGE_CACHE_SIZE
;
3368 physical_for_dev_replace
+= PAGE_CACHE_SIZE
;
3369 len
-= PAGE_CACHE_SIZE
;
3371 ret
= COPY_COMPLETE
;
3373 unlock_extent_cached(io_tree
, lockstart
, lockend
, &cached_state
,
3376 mutex_unlock(&inode
->i_mutex
);
3381 static int write_page_nocow(struct scrub_ctx
*sctx
,
3382 u64 physical_for_dev_replace
, struct page
*page
)
3385 struct btrfs_device
*dev
;
3388 dev
= sctx
->wr_ctx
.tgtdev
;
3392 printk_ratelimited(KERN_WARNING
3393 "BTRFS: scrub write_page_nocow(bdev == NULL) is unexpected!\n");
3396 bio
= btrfs_io_bio_alloc(GFP_NOFS
, 1);
3398 spin_lock(&sctx
->stat_lock
);
3399 sctx
->stat
.malloc_errors
++;
3400 spin_unlock(&sctx
->stat_lock
);
3403 bio
->bi_iter
.bi_size
= 0;
3404 bio
->bi_iter
.bi_sector
= physical_for_dev_replace
>> 9;
3405 bio
->bi_bdev
= dev
->bdev
;
3406 ret
= bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0);
3407 if (ret
!= PAGE_CACHE_SIZE
) {
3410 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_WRITE_ERRS
);
3414 if (btrfsic_submit_bio_wait(WRITE_SYNC
, bio
))
3415 goto leave_with_eio
;