projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: fix lockdep deadlock warning due to dev_replace
[linux-2.6-block.git] / fs / btrfs / scrub.c
CommitLineData
a2de733c 1/*
b6bfebc1 2 * Copyright (C) 2011, 2012 STRATO. All rights reserved.
a2de733c
AJ		 3 *
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.
7 *
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.
12 *
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.
17 */
18
a2de733c 19#include <linux/blkdev.h>
558540c1 20#include <linux/ratelimit.h>
a2de733c
AJ		 21#include "ctree.h"
22#include "volumes.h"
23#include "disk-io.h"
24#include "ordered-data.h"
0ef8e451 25#include "transaction.h"
558540c1 26#include "backref.h"
5da6fcbc 27#include "extent_io.h"
ff023aac 28#include "dev-replace.h"
21adbd5c 29#include "check-integrity.h"
606686ee 30#include "rcu-string.h"
53b381b3 31#include "raid56.h"
a2de733c
AJ		 32
33/*
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
37 * any can be found.
38 *
39 * Future enhancements:
a2de733c
AJ		 40 * - In case an unrepairable extent is encountered, track which files are
41 * affected and report them
a2de733c 42 * - track and record media errors, throw out bad devices
a2de733c 43 * - add a mode to also read unallocated space
a2de733c
AJ		 44 */
45
b5d67f64 46struct scrub_block;
d9d181c1 47struct scrub_ctx;
a2de733c 48
ff023aac
SB		 49/*
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.
54 */
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 */
7a9e9987
SB		 58
59/*
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.
63 */
b5d67f64 64#define SCRUB_MAX_PAGES_PER_BLOCK 16 /* 64k per node/leaf/sector */
a2de733c 65
af8e2d1d
MX		 66struct scrub_recover {
67 atomic_t refs;
68 struct btrfs_bio *bbio;
af8e2d1d
MX		 69 u64 map_length;
70};
71
a2de733c 72struct scrub_page {
b5d67f64
SB		 73 struct scrub_block *sblock;
74 struct page *page;
442a4f63 75 struct btrfs_device *dev;
5a6ac9ea 76 struct list_head list;
a2de733c
AJ		 77 u64 flags; /* extent flags */
78 u64 generation;
b5d67f64
SB		 79 u64 logical;
80 u64 physical;
ff023aac 81 u64 physical_for_dev_replace;
57019345 82 atomic_t refs;
b5d67f64
SB		 83 struct {
84 unsigned int mirror_num:8;
85 unsigned int have_csum:1;
86 unsigned int io_error:1;
87 };
a2de733c 88 u8 csum[BTRFS_CSUM_SIZE];
af8e2d1d
MX		 89
90 struct scrub_recover *recover;
a2de733c
AJ		 91};
92
93struct scrub_bio {
94 int index;
d9d181c1 95 struct scrub_ctx *sctx;
a36cf8b8 96 struct btrfs_device *dev;
a2de733c
AJ		 97 struct bio *bio;
98 int err;
99 u64 logical;
100 u64 physical;
ff023aac
SB		 101#if SCRUB_PAGES_PER_WR_BIO >= SCRUB_PAGES_PER_RD_BIO
102 struct scrub_page *pagev[SCRUB_PAGES_PER_WR_BIO];
103#else
104 struct scrub_page *pagev[SCRUB_PAGES_PER_RD_BIO];
105#endif
b5d67f64 106 int page_count;
a2de733c
AJ		 107 int next_free;
108 struct btrfs_work work;
109};
110
b5d67f64 111struct scrub_block {
7a9e9987 112 struct scrub_page *pagev[SCRUB_MAX_PAGES_PER_BLOCK];
b5d67f64
SB		 113 int page_count;
114 atomic_t outstanding_pages;
57019345 115 atomic_t refs; /* free mem on transition to zero */
d9d181c1 116 struct scrub_ctx *sctx;
5a6ac9ea 117 struct scrub_parity *sparity;
b5d67f64
SB		 118 struct {
119 unsigned int header_error:1;
120 unsigned int checksum_error:1;
121 unsigned int no_io_error_seen:1;
442a4f63 122 unsigned int generation_error:1; /* also sets header_error */
5a6ac9ea
MX		 123
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;
b5d67f64 127 };
73ff61db 128 struct btrfs_work work;
b5d67f64
SB		 129};
130
5a6ac9ea
MX		 131/* Used for the chunks with parity stripe such RAID5/6 */
132struct scrub_parity {
133 struct scrub_ctx *sctx;
134
135 struct btrfs_device *scrub_dev;
136
137 u64 logic_start;
138
139 u64 logic_end;
140
141 int nsectors;
142
143 int stripe_len;
144
57019345 145 atomic_t refs;
5a6ac9ea
MX		 146
147 struct list_head spages;
148
149 /* Work of parity check and repair */
150 struct btrfs_work work;
151
152 /* Mark the parity blocks which have data */
153 unsigned long *dbitmap;
154
155 /*
156 * Mark the parity blocks which have data, but errors happen when
157 * read data or check data
158 */
159 unsigned long *ebitmap;
160
161 unsigned long bitmap[0];
162};
163
ff023aac
SB		 164struct 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;
170};
171
d9d181c1 172struct scrub_ctx {
ff023aac 173 struct scrub_bio *bios[SCRUB_BIOS_PER_SCTX];
a36cf8b8 174 struct btrfs_root *dev_root;
a2de733c
AJ		 175 int first_free;
176 int curr;
b6bfebc1
SB		 177 atomic_t bios_in_flight;
178 atomic_t workers_pending;
a2de733c
AJ		 179 spinlock_t list_lock;
180 wait_queue_head_t list_wait;
181 u16 csum_size;
182 struct list_head csum_list;
183 atomic_t cancel_req;
8628764e 184 int readonly;
ff023aac 185 int pages_per_rd_bio;
b5d67f64
SB		 186 u32 sectorsize;
187 u32 nodesize;
63a212ab
SB		 188
189 int is_dev_replace;
ff023aac 190 struct scrub_wr_ctx wr_ctx;
63a212ab 191
a2de733c
AJ		 192 /*
193 * statistics
194 */
195 struct btrfs_scrub_progress stat;
196 spinlock_t stat_lock;
f55985f4
FM		 197
198 /*
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.
204 */
205 atomic_t refs;
a2de733c
AJ		 206};
207
0ef8e451 208struct scrub_fixup_nodatasum {
d9d181c1 209 struct scrub_ctx *sctx;
a36cf8b8 210 struct btrfs_device *dev;
0ef8e451
JS		 211 u64 logical;
212 struct btrfs_root *root;
213 struct btrfs_work work;
214 int mirror_num;
215};
216
652f25a2
JB		 217struct scrub_nocow_inode {
218 u64 inum;
219 u64 offset;
220 u64 root;
221 struct list_head list;
222};
223
ff023aac
SB		 224struct scrub_copy_nocow_ctx {
225 struct scrub_ctx *sctx;
226 u64 logical;
227 u64 len;
228 int mirror_num;
229 u64 physical_for_dev_replace;
652f25a2 230 struct list_head inodes;
ff023aac
SB		 231 struct btrfs_work work;
232};
233
558540c1
JS		 234struct scrub_warning {
235 struct btrfs_path *path;
236 u64 extent_item_size;
558540c1
JS		 237 const char *errstr;
238 sector_t sector;
239 u64 logical;
240 struct btrfs_device *dev;
558540c1
JS		 241};
242
b6bfebc1
SB		 243static void scrub_pending_bio_inc(struct scrub_ctx *sctx);
244static void scrub_pending_bio_dec(struct scrub_ctx *sctx);
245static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx);
246static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx);
b5d67f64 247static int scrub_handle_errored_block(struct scrub_block *sblock_to_check);
be50a8dd 248static int scrub_setup_recheck_block(struct scrub_block *original_sblock,
ff023aac 249 struct scrub_block *sblocks_for_recheck);
34f5c8e9 250static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
affe4a5a
ZL		 251 struct scrub_block *sblock,
252 int retry_failed_mirror);
ba7cf988 253static void scrub_recheck_block_checksum(struct scrub_block *sblock);
b5d67f64 254static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
114ab50d 255 struct scrub_block *sblock_good);
b5d67f64
SB		 256static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
257 struct scrub_block *sblock_good,
258 int page_num, int force_write);
ff023aac
SB		 259static void scrub_write_block_to_dev_replace(struct scrub_block *sblock);
260static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
261 int page_num);
b5d67f64
SB		 262static int scrub_checksum_data(struct scrub_block *sblock);
263static int scrub_checksum_tree_block(struct scrub_block *sblock);
264static int scrub_checksum_super(struct scrub_block *sblock);
265static void scrub_block_get(struct scrub_block *sblock);
266static void scrub_block_put(struct scrub_block *sblock);
7a9e9987
SB		 267static void scrub_page_get(struct scrub_page *spage);
268static void scrub_page_put(struct scrub_page *spage);
5a6ac9ea
MX		 269static void scrub_parity_get(struct scrub_parity *sparity);
270static void scrub_parity_put(struct scrub_parity *sparity);
ff023aac
SB		 271static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
272 struct scrub_page *spage);
d9d181c1 273static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
a36cf8b8 274 u64 physical, struct btrfs_device *dev, u64 flags,
ff023aac
SB		 275 u64 gen, int mirror_num, u8 *csum, int force,
276 u64 physical_for_dev_replace);
4246a0b6 277static void scrub_bio_end_io(struct bio *bio);
b5d67f64
SB		 278static void scrub_bio_end_io_worker(struct btrfs_work *work);
279static void scrub_block_complete(struct scrub_block *sblock);
ff023aac
SB		 280static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
281 u64 extent_logical, u64 extent_len,
282 u64 *extent_physical,
283 struct btrfs_device **extent_dev,
284 int *extent_mirror_num);
285static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
286 struct scrub_wr_ctx *wr_ctx,
287 struct btrfs_fs_info *fs_info,
288 struct btrfs_device *dev,
289 int is_dev_replace);
290static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx);
291static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
292 struct scrub_page *spage);
293static void scrub_wr_submit(struct scrub_ctx *sctx);
4246a0b6 294static void scrub_wr_bio_end_io(struct bio *bio);
ff023aac
SB		 295static void scrub_wr_bio_end_io_worker(struct btrfs_work *work);
296static int write_page_nocow(struct scrub_ctx *sctx,
297 u64 physical_for_dev_replace, struct page *page);
298static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
652f25a2 299 struct scrub_copy_nocow_ctx *ctx);
ff023aac
SB		 300static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
301 int mirror_num, u64 physical_for_dev_replace);
302static void copy_nocow_pages_worker(struct btrfs_work *work);
cb7ab021 303static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
3cb0929a 304static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
f55985f4 305static void scrub_put_ctx(struct scrub_ctx *sctx);
1623edeb
SB		 306
307
b6bfebc1
SB		 308static void scrub_pending_bio_inc(struct scrub_ctx *sctx)
309{
f55985f4 310 atomic_inc(&sctx->refs);
b6bfebc1
SB		 311 atomic_inc(&sctx->bios_in_flight);
312}
313
314static void scrub_pending_bio_dec(struct scrub_ctx *sctx)
315{
316 atomic_dec(&sctx->bios_in_flight);
317 wake_up(&sctx->list_wait);
f55985f4 318 scrub_put_ctx(sctx);
b6bfebc1
SB		 319}
320
cb7ab021 321static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
3cb0929a
WS		 322{
323 while (atomic_read(&fs_info->scrub_pause_req)) {
324 mutex_unlock(&fs_info->scrub_lock);
325 wait_event(fs_info->scrub_pause_wait,
326 atomic_read(&fs_info->scrub_pause_req) == 0);
327 mutex_lock(&fs_info->scrub_lock);
328 }
329}
330
0e22be89 331static void scrub_pause_on(struct btrfs_fs_info *fs_info)
cb7ab021
WS		 332{
333 atomic_inc(&fs_info->scrubs_paused);
334 wake_up(&fs_info->scrub_pause_wait);
0e22be89 335}
cb7ab021 336
0e22be89
Z		 337static void scrub_pause_off(struct btrfs_fs_info *fs_info)
338{
cb7ab021
WS		 339 mutex_lock(&fs_info->scrub_lock);
340 __scrub_blocked_if_needed(fs_info);
341 atomic_dec(&fs_info->scrubs_paused);
342 mutex_unlock(&fs_info->scrub_lock);
343
344 wake_up(&fs_info->scrub_pause_wait);
345}
346
0e22be89
Z		 347static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
348{
349 scrub_pause_on(fs_info);
350 scrub_pause_off(fs_info);
351}
352
b6bfebc1
SB		 353/*
354 * used for workers that require transaction commits (i.e., for the
355 * NOCOW case)
356 */
357static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx)
358{
359 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
360
f55985f4 361 atomic_inc(&sctx->refs);
b6bfebc1
SB		 362 /*
363 * increment scrubs_running to prevent cancel requests from
364 * completing as long as a worker is running. we must also
365 * increment scrubs_paused to prevent deadlocking on pause
366 * requests used for transactions commits (as the worker uses a
367 * transaction context). it is safe to regard the worker
368 * as paused for all matters practical. effectively, we only
369 * avoid cancellation requests from completing.
370 */
371 mutex_lock(&fs_info->scrub_lock);
372 atomic_inc(&fs_info->scrubs_running);
373 atomic_inc(&fs_info->scrubs_paused);
374 mutex_unlock(&fs_info->scrub_lock);
32a44789
WS		 375
376 /*
377 * check if @scrubs_running=@scrubs_paused condition
378 * inside wait_event() is not an atomic operation.
379 * which means we may inc/dec @scrub_running/paused
380 * at any time. Let's wake up @scrub_pause_wait as
381 * much as we can to let commit transaction blocked less.
382 */
383 wake_up(&fs_info->scrub_pause_wait);
384
b6bfebc1
SB		 385 atomic_inc(&sctx->workers_pending);
386}
387
388/* used for workers that require transaction commits */
389static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx)
390{
391 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
392
393 /*
394 * see scrub_pending_trans_workers_inc() why we're pretending
395 * to be paused in the scrub counters
396 */
397 mutex_lock(&fs_info->scrub_lock);
398 atomic_dec(&fs_info->scrubs_running);
399 atomic_dec(&fs_info->scrubs_paused);
400 mutex_unlock(&fs_info->scrub_lock);
401 atomic_dec(&sctx->workers_pending);
402 wake_up(&fs_info->scrub_pause_wait);
403 wake_up(&sctx->list_wait);
f55985f4 404 scrub_put_ctx(sctx);
b6bfebc1
SB		 405}
406
d9d181c1 407static void scrub_free_csums(struct scrub_ctx *sctx)
a2de733c 408{
d9d181c1 409 while (!list_empty(&sctx->csum_list)) {
a2de733c 410 struct btrfs_ordered_sum *sum;
d9d181c1 411 sum = list_first_entry(&sctx->csum_list,
a2de733c
AJ		 412 struct btrfs_ordered_sum, list);
413 list_del(&sum->list);
414 kfree(sum);
415 }
416}
417
d9d181c1 418static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx)
a2de733c
AJ		 419{
420 int i;
a2de733c 421
d9d181c1 422 if (!sctx)
a2de733c
AJ		 423 return;
424
ff023aac
SB		 425 scrub_free_wr_ctx(&sctx->wr_ctx);
426
b5d67f64 427 /* this can happen when scrub is cancelled */
d9d181c1
SB		 428 if (sctx->curr != -1) {
429 struct scrub_bio *sbio = sctx->bios[sctx->curr];
b5d67f64
SB		 430
431 for (i = 0; i < sbio->page_count; i++) {
ff023aac 432 WARN_ON(!sbio->pagev[i]->page);
b5d67f64
SB		 433 scrub_block_put(sbio->pagev[i]->sblock);
434 }
435 bio_put(sbio->bio);
436 }
437
ff023aac 438 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
d9d181c1 439 struct scrub_bio *sbio = sctx->bios[i];
a2de733c
AJ		 440
441 if (!sbio)
442 break;
a2de733c
AJ		 443 kfree(sbio);
444 }
445
d9d181c1
SB		 446 scrub_free_csums(sctx);
447 kfree(sctx);
a2de733c
AJ		 448}
449
f55985f4
FM		 450static void scrub_put_ctx(struct scrub_ctx *sctx)
451{
452 if (atomic_dec_and_test(&sctx->refs))
453 scrub_free_ctx(sctx);
454}
455
a2de733c 456static noinline_for_stack
63a212ab 457struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace)
a2de733c 458{
d9d181c1 459 struct scrub_ctx *sctx;
a2de733c 460 int i;
a2de733c 461 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
ff023aac 462 int ret;
a2de733c 463
d9d181c1
SB		 464 sctx = kzalloc(sizeof(*sctx), GFP_NOFS);
465 if (!sctx)
a2de733c 466 goto nomem;
f55985f4 467 atomic_set(&sctx->refs, 1);
63a212ab 468 sctx->is_dev_replace = is_dev_replace;
b54ffb73 469 sctx->pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO;
d9d181c1 470 sctx->curr = -1;
a36cf8b8 471 sctx->dev_root = dev->dev_root;
ff023aac 472 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
a2de733c
AJ		 473 struct scrub_bio *sbio;
474
475 sbio = kzalloc(sizeof(*sbio), GFP_NOFS);
476 if (!sbio)
477 goto nomem;
d9d181c1 478 sctx->bios[i] = sbio;
a2de733c 479
a2de733c 480 sbio->index = i;
d9d181c1 481 sbio->sctx = sctx;
b5d67f64 482 sbio->page_count = 0;
9e0af237
LB		 483 btrfs_init_work(&sbio->work, btrfs_scrub_helper,
484 scrub_bio_end_io_worker, NULL, NULL);
a2de733c 485
ff023aac 486 if (i != SCRUB_BIOS_PER_SCTX - 1)
d9d181c1 487 sctx->bios[i]->next_free = i + 1;
0ef8e451 488 else
d9d181c1
SB		 489 sctx->bios[i]->next_free = -1;
490 }
491 sctx->first_free = 0;
492 sctx->nodesize = dev->dev_root->nodesize;
d9d181c1 493 sctx->sectorsize = dev->dev_root->sectorsize;
b6bfebc1
SB		 494 atomic_set(&sctx->bios_in_flight, 0);
495 atomic_set(&sctx->workers_pending, 0);
d9d181c1
SB		 496 atomic_set(&sctx->cancel_req, 0);
497 sctx->csum_size = btrfs_super_csum_size(fs_info->super_copy);
498 INIT_LIST_HEAD(&sctx->csum_list);
499
500 spin_lock_init(&sctx->list_lock);
501 spin_lock_init(&sctx->stat_lock);
502 init_waitqueue_head(&sctx->list_wait);
ff023aac
SB		 503
504 ret = scrub_setup_wr_ctx(sctx, &sctx->wr_ctx, fs_info,
505 fs_info->dev_replace.tgtdev, is_dev_replace);
506 if (ret) {
507 scrub_free_ctx(sctx);
508 return ERR_PTR(ret);
509 }
d9d181c1 510 return sctx;
a2de733c
AJ		 511
512nomem:
d9d181c1 513 scrub_free_ctx(sctx);
a2de733c
AJ		 514 return ERR_PTR(-ENOMEM);
515}
516
ff023aac
SB		 517static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root,
518 void *warn_ctx)
558540c1
JS		 519{
520 u64 isize;
521 u32 nlink;
522 int ret;
523 int i;
524 struct extent_buffer *eb;
525 struct btrfs_inode_item *inode_item;
ff023aac 526 struct scrub_warning *swarn = warn_ctx;
558540c1
JS		 527 struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info;
528 struct inode_fs_paths *ipath = NULL;
529 struct btrfs_root *local_root;
530 struct btrfs_key root_key;
1d4c08e0 531 struct btrfs_key key;
558540c1
JS		 532
533 root_key.objectid = root;
534 root_key.type = BTRFS_ROOT_ITEM_KEY;
535 root_key.offset = (u64)-1;
536 local_root = btrfs_read_fs_root_no_name(fs_info, &root_key);
537 if (IS_ERR(local_root)) {
538 ret = PTR_ERR(local_root);
539 goto err;
540 }
541
14692cc1
DS		 542 /*
543 * this makes the path point to (inum INODE_ITEM ioff)
544 */
1d4c08e0
DS		 545 key.objectid = inum;
546 key.type = BTRFS_INODE_ITEM_KEY;
547 key.offset = 0;
548
549 ret = btrfs_search_slot(NULL, local_root, &key, swarn->path, 0, 0);
558540c1
JS		 550 if (ret) {
551 btrfs_release_path(swarn->path);
552 goto err;
553 }
554
555 eb = swarn->path->nodes[0];
556 inode_item = btrfs_item_ptr(eb, swarn->path->slots[0],
557 struct btrfs_inode_item);
558 isize = btrfs_inode_size(eb, inode_item);
559 nlink = btrfs_inode_nlink(eb, inode_item);
560 btrfs_release_path(swarn->path);
561
562 ipath = init_ipath(4096, local_root, swarn->path);
26bdef54
DC		 563 if (IS_ERR(ipath)) {
564 ret = PTR_ERR(ipath);
565 ipath = NULL;
566 goto err;
567 }
558540c1
JS		 568 ret = paths_from_inode(inum, ipath);
569
570 if (ret < 0)
571 goto err;
572
573 /*
574 * we deliberately ignore the bit ipath might have been too small to
575 * hold all of the paths here
576 */
577 for (i = 0; i < ipath->fspath->elem_cnt; ++i)
ecaeb14b 578 btrfs_warn_in_rcu(fs_info, "%s at logical %llu on dev "
558540c1 579 "%s, sector %llu, root %llu, inode %llu, offset %llu, "
ecaeb14b 580 "length %llu, links %u (path: %s)", swarn->errstr,
606686ee 581 swarn->logical, rcu_str_deref(swarn->dev->name),
558540c1
JS		 582 (unsigned long long)swarn->sector, root, inum, offset,
583 min(isize - offset, (u64)PAGE_SIZE), nlink,
745c4d8e 584 (char *)(unsigned long)ipath->fspath->val[i]);
558540c1
JS		 585
586 free_ipath(ipath);
587 return 0;
588
589err:
ecaeb14b 590 btrfs_warn_in_rcu(fs_info, "%s at logical %llu on dev "
558540c1 591 "%s, sector %llu, root %llu, inode %llu, offset %llu: path "
ecaeb14b 592 "resolving failed with ret=%d", swarn->errstr,
606686ee 593 swarn->logical, rcu_str_deref(swarn->dev->name),
558540c1
JS		 594 (unsigned long long)swarn->sector, root, inum, offset, ret);
595
596 free_ipath(ipath);
597 return 0;
598}
599
b5d67f64 600static void scrub_print_warning(const char *errstr, struct scrub_block *sblock)
558540c1 601{
a36cf8b8
SB		 602 struct btrfs_device *dev;
603 struct btrfs_fs_info *fs_info;
558540c1
JS		 604 struct btrfs_path *path;
605 struct btrfs_key found_key;
606 struct extent_buffer *eb;
607 struct btrfs_extent_item *ei;
608 struct scrub_warning swarn;
69917e43
LB		 609 unsigned long ptr = 0;
610 u64 extent_item_pos;
611 u64 flags = 0;
558540c1 612 u64 ref_root;
69917e43 613 u32 item_size;
558540c1 614 u8 ref_level;
69917e43 615 int ret;
558540c1 616
a36cf8b8 617 WARN_ON(sblock->page_count < 1);
7a9e9987 618 dev = sblock->pagev[0]->dev;
a36cf8b8
SB		 619 fs_info = sblock->sctx->dev_root->fs_info;
620
558540c1 621 path = btrfs_alloc_path();
8b9456da
DS		 622 if (!path)
623 return;
558540c1 624
7a9e9987
SB		 625 swarn.sector = (sblock->pagev[0]->physical) >> 9;
626 swarn.logical = sblock->pagev[0]->logical;
558540c1 627 swarn.errstr = errstr;
a36cf8b8 628 swarn.dev = NULL;
558540c1 629
69917e43
LB		 630 ret = extent_from_logical(fs_info, swarn.logical, path, &found_key,
631 &flags);
558540c1
JS		 632 if (ret < 0)
633 goto out;
634
4692cf58 635 extent_item_pos = swarn.logical - found_key.objectid;
558540c1
JS		 636 swarn.extent_item_size = found_key.offset;
637
638 eb = path->nodes[0];
639 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
640 item_size = btrfs_item_size_nr(eb, path->slots[0]);
641
69917e43 642 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
558540c1 643 do {
6eda71d0
LB		 644 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
645 item_size, &ref_root,
646 &ref_level);
ecaeb14b
DS		 647 btrfs_warn_in_rcu(fs_info,
648 "%s at logical %llu on dev %s, "
558540c1 649 "sector %llu: metadata %s (level %d) in tree "
ecaeb14b 650 "%llu", errstr, swarn.logical,
606686ee 651 rcu_str_deref(dev->name),
558540c1
JS		 652 (unsigned long long)swarn.sector,
653 ref_level ? "node" : "leaf",
654 ret < 0 ? -1 : ref_level,
655 ret < 0 ? -1 : ref_root);
656 } while (ret != 1);
d8fe29e9 657 btrfs_release_path(path);
558540c1 658 } else {
d8fe29e9 659 btrfs_release_path(path);
558540c1 660 swarn.path = path;
a36cf8b8 661 swarn.dev = dev;
7a3ae2f8
JS		 662 iterate_extent_inodes(fs_info, found_key.objectid,
663 extent_item_pos, 1,
558540c1
JS		 664 scrub_print_warning_inode, &swarn);
665 }
666
667out:
668 btrfs_free_path(path);
558540c1
JS		 669}
670
ff023aac 671static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx)
0ef8e451 672{
5da6fcbc 673 struct page *page = NULL;
0ef8e451 674 unsigned long index;
ff023aac 675 struct scrub_fixup_nodatasum *fixup = fixup_ctx;
0ef8e451 676 int ret;
5da6fcbc 677 int corrected = 0;
0ef8e451 678 struct btrfs_key key;
5da6fcbc 679 struct inode *inode = NULL;
6f1c3605 680 struct btrfs_fs_info *fs_info;
0ef8e451
JS		 681 u64 end = offset + PAGE_SIZE - 1;
682 struct btrfs_root *local_root;
6f1c3605 683 int srcu_index;
0ef8e451
JS		 684
685 key.objectid = root;
686 key.type = BTRFS_ROOT_ITEM_KEY;
687 key.offset = (u64)-1;
6f1c3605
LB		 688
689 fs_info = fixup->root->fs_info;
690 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
691
692 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
693 if (IS_ERR(local_root)) {
694 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
0ef8e451 695 return PTR_ERR(local_root);
6f1c3605 696 }
0ef8e451
JS		 697
698 key.type = BTRFS_INODE_ITEM_KEY;
699 key.objectid = inum;
700 key.offset = 0;
6f1c3605
LB		 701 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
702 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
0ef8e451
JS		 703 if (IS_ERR(inode))
704 return PTR_ERR(inode);
705
0ef8e451
JS		 706 index = offset >> PAGE_CACHE_SHIFT;
707
708 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
5da6fcbc
JS		 709 if (!page) {
710 ret = -ENOMEM;
711 goto out;
712 }
713
714 if (PageUptodate(page)) {
5da6fcbc
JS		 715 if (PageDirty(page)) {
716 /*
717 * we need to write the data to the defect sector. the
718 * data that was in that sector is not in memory,
719 * because the page was modified. we must not write the
720 * modified page to that sector.
721 *
722 * TODO: what could be done here: wait for the delalloc
723 * runner to write out that page (might involve
724 * COW) and see whether the sector is still
725 * referenced afterwards.
726 *
727 * For the meantime, we'll treat this error
728 * incorrectable, although there is a chance that a
729 * later scrub will find the bad sector again and that
730 * there's no dirty page in memory, then.
731 */
732 ret = -EIO;
733 goto out;
734 }
1203b681 735 ret = repair_io_failure(inode, offset, PAGE_SIZE,
5da6fcbc 736 fixup->logical, page,
ffdd2018 737 offset - page_offset(page),
5da6fcbc
JS		 738 fixup->mirror_num);
739 unlock_page(page);
740 corrected = !ret;
741 } else {
742 /*
743 * we need to get good data first. the general readpage path
744 * will call repair_io_failure for us, we just have to make
745 * sure we read the bad mirror.
746 */
747 ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
748 EXTENT_DAMAGED, GFP_NOFS);
749 if (ret) {
750 /* set_extent_bits should give proper error */
751 WARN_ON(ret > 0);
752 if (ret > 0)
753 ret = -EFAULT;
754 goto out;
755 }
756
757 ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page,
758 btrfs_get_extent,
759 fixup->mirror_num);
760 wait_on_page_locked(page);
761
762 corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset,
763 end, EXTENT_DAMAGED, 0, NULL);
764 if (!corrected)
765 clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
766 EXTENT_DAMAGED, GFP_NOFS);
767 }
768
769out:
770 if (page)
771 put_page(page);
7fb18a06
TK		 772
773 iput(inode);
0ef8e451
JS		 774
775 if (ret < 0)
776 return ret;
777
778 if (ret == 0 && corrected) {
779 /*
780 * we only need to call readpage for one of the inodes belonging
781 * to this extent. so make iterate_extent_inodes stop
782 */
783 return 1;
784 }
785
786 return -EIO;
787}
788
789static void scrub_fixup_nodatasum(struct btrfs_work *work)
790{
791 int ret;
792 struct scrub_fixup_nodatasum *fixup;
d9d181c1 793 struct scrub_ctx *sctx;
0ef8e451 794 struct btrfs_trans_handle *trans = NULL;
0ef8e451
JS		 795 struct btrfs_path *path;
796 int uncorrectable = 0;
797
798 fixup = container_of(work, struct scrub_fixup_nodatasum, work);
d9d181c1 799 sctx = fixup->sctx;
0ef8e451
JS		 800
801 path = btrfs_alloc_path();
802 if (!path) {
d9d181c1
SB		 803 spin_lock(&sctx->stat_lock);
804 ++sctx->stat.malloc_errors;
805 spin_unlock(&sctx->stat_lock);
0ef8e451
JS		 806 uncorrectable = 1;
807 goto out;
808 }
809
810 trans = btrfs_join_transaction(fixup->root);
811 if (IS_ERR(trans)) {
812 uncorrectable = 1;
813 goto out;
814 }
815
816 /*
817 * the idea is to trigger a regular read through the standard path. we
818 * read a page from the (failed) logical address by specifying the
819 * corresponding copynum of the failed sector. thus, that readpage is
820 * expected to fail.
821 * that is the point where on-the-fly error correction will kick in
822 * (once it's finished) and rewrite the failed sector if a good copy
823 * can be found.
824 */
825 ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info,
826 path, scrub_fixup_readpage,
827 fixup);
828 if (ret < 0) {
829 uncorrectable = 1;
830 goto out;
831 }
832 WARN_ON(ret != 1);
833
d9d181c1
SB		 834 spin_lock(&sctx->stat_lock);
835 ++sctx->stat.corrected_errors;
836 spin_unlock(&sctx->stat_lock);
0ef8e451
JS		 837
838out:
839 if (trans && !IS_ERR(trans))
840 btrfs_end_transaction(trans, fixup->root);
841 if (uncorrectable) {
d9d181c1
SB		 842 spin_lock(&sctx->stat_lock);
843 ++sctx->stat.uncorrectable_errors;
844 spin_unlock(&sctx->stat_lock);
ff023aac
SB		 845 btrfs_dev_replace_stats_inc(
846 &sctx->dev_root->fs_info->dev_replace.
847 num_uncorrectable_read_errors);
b14af3b4
DS		 848 btrfs_err_rl_in_rcu(sctx->dev_root->fs_info,
849 "unable to fixup (nodatasum) error at logical %llu on dev %s",
c1c9ff7c 850 fixup->logical, rcu_str_deref(fixup->dev->name));
0ef8e451
JS		 851 }
852
853 btrfs_free_path(path);
854 kfree(fixup);
855
b6bfebc1 856 scrub_pending_trans_workers_dec(sctx);
0ef8e451
JS		 857}
858
af8e2d1d
MX		 859static inline void scrub_get_recover(struct scrub_recover *recover)
860{
861 atomic_inc(&recover->refs);
862}
863
864static inline void scrub_put_recover(struct scrub_recover *recover)
865{
866 if (atomic_dec_and_test(&recover->refs)) {
6e9606d2 867 btrfs_put_bbio(recover->bbio);
af8e2d1d
MX		 868 kfree(recover);
869 }
870}
871
a2de733c 872/*
b5d67f64
SB		 873 * scrub_handle_errored_block gets called when either verification of the
874 * pages failed or the bio failed to read, e.g. with EIO. In the latter
875 * case, this function handles all pages in the bio, even though only one
876 * may be bad.
877 * The goal of this function is to repair the errored block by using the
878 * contents of one of the mirrors.
a2de733c 879 */
b5d67f64 880static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
a2de733c 881{
d9d181c1 882 struct scrub_ctx *sctx = sblock_to_check->sctx;
a36cf8b8 883 struct btrfs_device *dev;
b5d67f64
SB		 884 struct btrfs_fs_info *fs_info;
885 u64 length;
886 u64 logical;
b5d67f64
SB		 887 unsigned int failed_mirror_index;
888 unsigned int is_metadata;
889 unsigned int have_csum;
b5d67f64
SB		 890 struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */
891 struct scrub_block *sblock_bad;
892 int ret;
893 int mirror_index;
894 int page_num;
895 int success;
558540c1 896 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
b5d67f64
SB		 897 DEFAULT_RATELIMIT_BURST);
898
899 BUG_ON(sblock_to_check->page_count < 1);
a36cf8b8 900 fs_info = sctx->dev_root->fs_info;
4ded4f63
SB		 901 if (sblock_to_check->pagev[0]->flags & BTRFS_EXTENT_FLAG_SUPER) {
902 /*
903 * if we find an error in a super block, we just report it.
904 * They will get written with the next transaction commit
905 * anyway
906 */
907 spin_lock(&sctx->stat_lock);
908 ++sctx->stat.super_errors;
909 spin_unlock(&sctx->stat_lock);
910 return 0;
911 }
b5d67f64 912 length = sblock_to_check->page_count * PAGE_SIZE;
7a9e9987 913 logical = sblock_to_check->pagev[0]->logical;
7a9e9987
SB		 914 BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1);
915 failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1;
916 is_metadata = !(sblock_to_check->pagev[0]->flags &
b5d67f64 917 BTRFS_EXTENT_FLAG_DATA);
7a9e9987 918 have_csum = sblock_to_check->pagev[0]->have_csum;
7a9e9987 919 dev = sblock_to_check->pagev[0]->dev;
13db62b7 920
ff023aac
SB		 921 if (sctx->is_dev_replace && !is_metadata && !have_csum) {
922 sblocks_for_recheck = NULL;
923 goto nodatasum_case;
924 }
925
b5d67f64
SB		 926 /*
927 * read all mirrors one after the other. This includes to
928 * re-read the extent or metadata block that failed (that was
929 * the cause that this fixup code is called) another time,
930 * page by page this time in order to know which pages
931 * caused I/O errors and which ones are good (for all mirrors).
932 * It is the goal to handle the situation when more than one
933 * mirror contains I/O errors, but the errors do not
934 * overlap, i.e. the data can be repaired by selecting the
935 * pages from those mirrors without I/O error on the
936 * particular pages. One example (with blocks >= 2 * PAGE_SIZE)
937 * would be that mirror #1 has an I/O error on the first page,
938 * the second page is good, and mirror #2 has an I/O error on
939 * the second page, but the first page is good.
940 * Then the first page of the first mirror can be repaired by
941 * taking the first page of the second mirror, and the
942 * second page of the second mirror can be repaired by
943 * copying the contents of the 2nd page of the 1st mirror.
944 * One more note: if the pages of one mirror contain I/O
945 * errors, the checksum cannot be verified. In order to get
946 * the best data for repairing, the first attempt is to find
947 * a mirror without I/O errors and with a validated checksum.
948 * Only if this is not possible, the pages are picked from
949 * mirrors with I/O errors without considering the checksum.
950 * If the latter is the case, at the end, the checksum of the
951 * repaired area is verified in order to correctly maintain
952 * the statistics.
953 */
954
31e818fe
DS		 955 sblocks_for_recheck = kcalloc(BTRFS_MAX_MIRRORS,
956 sizeof(*sblocks_for_recheck), GFP_NOFS);
b5d67f64 957 if (!sblocks_for_recheck) {
d9d181c1
SB		 958 spin_lock(&sctx->stat_lock);
959 sctx->stat.malloc_errors++;
960 sctx->stat.read_errors++;
961 sctx->stat.uncorrectable_errors++;
962 spin_unlock(&sctx->stat_lock);
a36cf8b8 963 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
b5d67f64 964 goto out;
a2de733c
AJ		 965 }
966
b5d67f64 967 /* setup the context, map the logical blocks and alloc the pages */
be50a8dd 968 ret = scrub_setup_recheck_block(sblock_to_check, sblocks_for_recheck);
b5d67f64 969 if (ret) {
d9d181c1
SB		 970 spin_lock(&sctx->stat_lock);
971 sctx->stat.read_errors++;
972 sctx->stat.uncorrectable_errors++;
973 spin_unlock(&sctx->stat_lock);
a36cf8b8 974 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
b5d67f64
SB		 975 goto out;
976 }
977 BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS);
978 sblock_bad = sblocks_for_recheck + failed_mirror_index;
13db62b7 979
b5d67f64 980 /* build and submit the bios for the failed mirror, check checksums */
affe4a5a 981 scrub_recheck_block(fs_info, sblock_bad, 1);
a2de733c 982
b5d67f64
SB		 983 if (!sblock_bad->header_error && !sblock_bad->checksum_error &&
984 sblock_bad->no_io_error_seen) {
985 /*
986 * the error disappeared after reading page by page, or
987 * the area was part of a huge bio and other parts of the
988 * bio caused I/O errors, or the block layer merged several
989 * read requests into one and the error is caused by a
990 * different bio (usually one of the two latter cases is
991 * the cause)
992 */
d9d181c1
SB		 993 spin_lock(&sctx->stat_lock);
994 sctx->stat.unverified_errors++;
5a6ac9ea 995 sblock_to_check->data_corrected = 1;
d9d181c1 996 spin_unlock(&sctx->stat_lock);
a2de733c 997
ff023aac
SB		 998 if (sctx->is_dev_replace)
999 scrub_write_block_to_dev_replace(sblock_bad);
b5d67f64 1000 goto out;
a2de733c 1001 }
a2de733c 1002
b5d67f64 1003 if (!sblock_bad->no_io_error_seen) {
d9d181c1
SB		 1004 spin_lock(&sctx->stat_lock);
1005 sctx->stat.read_errors++;
1006 spin_unlock(&sctx->stat_lock);
b5d67f64
SB		 1007 if (__ratelimit(&_rs))
1008 scrub_print_warning("i/o error", sblock_to_check);
a36cf8b8 1009 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
b5d67f64 1010 } else if (sblock_bad->checksum_error) {
d9d181c1
SB		 1011 spin_lock(&sctx->stat_lock);
1012 sctx->stat.csum_errors++;
1013 spin_unlock(&sctx->stat_lock);
b5d67f64
SB		 1014 if (__ratelimit(&_rs))
1015 scrub_print_warning("checksum error", sblock_to_check);
a36cf8b8 1016 btrfs_dev_stat_inc_and_print(dev,
442a4f63 1017 BTRFS_DEV_STAT_CORRUPTION_ERRS);
b5d67f64 1018 } else if (sblock_bad->header_error) {
d9d181c1
SB		 1019 spin_lock(&sctx->stat_lock);
1020 sctx->stat.verify_errors++;
1021 spin_unlock(&sctx->stat_lock);
b5d67f64
SB		 1022 if (__ratelimit(&_rs))
1023 scrub_print_warning("checksum/header error",
1024 sblock_to_check);
442a4f63 1025 if (sblock_bad->generation_error)
a36cf8b8 1026 btrfs_dev_stat_inc_and_print(dev,
442a4f63
SB		 1027 BTRFS_DEV_STAT_GENERATION_ERRS);
1028 else
a36cf8b8 1029 btrfs_dev_stat_inc_and_print(dev,
442a4f63 1030 BTRFS_DEV_STAT_CORRUPTION_ERRS);
b5d67f64 1031 }
a2de733c 1032
33ef30ad
ID		 1033 if (sctx->readonly) {
1034 ASSERT(!sctx->is_dev_replace);
1035 goto out;
1036 }
a2de733c 1037
b5d67f64
SB		 1038 if (!is_metadata && !have_csum) {
1039 struct scrub_fixup_nodatasum *fixup_nodatasum;
a2de733c 1040
ff023aac
SB		 1041 WARN_ON(sctx->is_dev_replace);
1042
b25c94c5
ZL		 1043nodatasum_case:
1044
b5d67f64
SB		 1045 /*
1046 * !is_metadata and !have_csum, this means that the data
1047 * might not be COW'ed, that it might be modified
1048 * concurrently. The general strategy to work on the
1049 * commit root does not help in the case when COW is not
1050 * used.
1051 */
1052 fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS);
1053 if (!fixup_nodatasum)
1054 goto did_not_correct_error;
d9d181c1 1055 fixup_nodatasum->sctx = sctx;
a36cf8b8 1056 fixup_nodatasum->dev = dev;
b5d67f64
SB		 1057 fixup_nodatasum->logical = logical;
1058 fixup_nodatasum->root = fs_info->extent_root;
1059 fixup_nodatasum->mirror_num = failed_mirror_index + 1;
b6bfebc1 1060 scrub_pending_trans_workers_inc(sctx);
9e0af237
LB		 1061 btrfs_init_work(&fixup_nodatasum->work, btrfs_scrub_helper,
1062 scrub_fixup_nodatasum, NULL, NULL);
0339ef2f
QW		 1063 btrfs_queue_work(fs_info->scrub_workers,
1064 &fixup_nodatasum->work);
b5d67f64 1065 goto out;
a2de733c
AJ		 1066 }
1067
b5d67f64
SB		 1068 /*
1069 * now build and submit the bios for the other mirrors, check
cb2ced73
SB		 1070 * checksums.
1071 * First try to pick the mirror which is completely without I/O
b5d67f64
SB		 1072 * errors and also does not have a checksum error.
1073 * If one is found, and if a checksum is present, the full block
1074 * that is known to contain an error is rewritten. Afterwards
1075 * the block is known to be corrected.
1076 * If a mirror is found which is completely correct, and no
1077 * checksum is present, only those pages are rewritten that had
1078 * an I/O error in the block to be repaired, since it cannot be
1079 * determined, which copy of the other pages is better (and it
1080 * could happen otherwise that a correct page would be
1081 * overwritten by a bad one).
1082 */
1083 for (mirror_index = 0;
1084 mirror_index < BTRFS_MAX_MIRRORS &&
1085 sblocks_for_recheck[mirror_index].page_count > 0;
1086 mirror_index++) {
cb2ced73 1087 struct scrub_block *sblock_other;
b5d67f64 1088
cb2ced73
SB		 1089 if (mirror_index == failed_mirror_index)
1090 continue;
1091 sblock_other = sblocks_for_recheck + mirror_index;
1092
1093 /* build and submit the bios, check checksums */
affe4a5a 1094 scrub_recheck_block(fs_info, sblock_other, 0);
34f5c8e9
SB		 1095
1096 if (!sblock_other->header_error &&
b5d67f64
SB		 1097 !sblock_other->checksum_error &&
1098 sblock_other->no_io_error_seen) {
ff023aac
SB		 1099 if (sctx->is_dev_replace) {
1100 scrub_write_block_to_dev_replace(sblock_other);
114ab50d 1101 goto corrected_error;
ff023aac 1102 } else {
ff023aac 1103 ret = scrub_repair_block_from_good_copy(
114ab50d
ZL		 1104 sblock_bad, sblock_other);
1105 if (!ret)
1106 goto corrected_error;
ff023aac 1107 }
b5d67f64
SB		 1108 }
1109 }
a2de733c 1110
b968fed1
ZL		 1111 if (sblock_bad->no_io_error_seen && !sctx->is_dev_replace)
1112 goto did_not_correct_error;
ff023aac
SB		 1113
1114 /*
ff023aac 1115 * In case of I/O errors in the area that is supposed to be
b5d67f64
SB		 1116 * repaired, continue by picking good copies of those pages.
1117 * Select the good pages from mirrors to rewrite bad pages from
1118 * the area to fix. Afterwards verify the checksum of the block
1119 * that is supposed to be repaired. This verification step is
1120 * only done for the purpose of statistic counting and for the
1121 * final scrub report, whether errors remain.
1122 * A perfect algorithm could make use of the checksum and try
1123 * all possible combinations of pages from the different mirrors
1124 * until the checksum verification succeeds. For example, when
1125 * the 2nd page of mirror #1 faces I/O errors, and the 2nd page
1126 * of mirror #2 is readable but the final checksum test fails,
1127 * then the 2nd page of mirror #3 could be tried, whether now
1128 * the final checksum succeedes. But this would be a rare
1129 * exception and is therefore not implemented. At least it is
1130 * avoided that the good copy is overwritten.
1131 * A more useful improvement would be to pick the sectors
1132 * without I/O error based on sector sizes (512 bytes on legacy
1133 * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one
1134 * mirror could be repaired by taking 512 byte of a different
1135 * mirror, even if other 512 byte sectors in the same PAGE_SIZE
1136 * area are unreadable.
a2de733c 1137 */
b5d67f64 1138 success = 1;
b968fed1
ZL		 1139 for (page_num = 0; page_num < sblock_bad->page_count;
1140 page_num++) {
7a9e9987 1141 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
b968fed1 1142 struct scrub_block *sblock_other = NULL;
b5d67f64 1143
b968fed1
ZL		 1144 /* skip no-io-error page in scrub */
1145 if (!page_bad->io_error && !sctx->is_dev_replace)
a2de733c 1146 continue;
b5d67f64 1147
b968fed1
ZL		 1148 /* try to find no-io-error page in mirrors */
1149 if (page_bad->io_error) {
1150 for (mirror_index = 0;
1151 mirror_index < BTRFS_MAX_MIRRORS &&
1152 sblocks_for_recheck[mirror_index].page_count > 0;
1153 mirror_index++) {
1154 if (!sblocks_for_recheck[mirror_index].
1155 pagev[page_num]->io_error) {
1156 sblock_other = sblocks_for_recheck +
1157 mirror_index;
1158 break;
b5d67f64
SB		 1159 }
1160 }
b968fed1
ZL		 1161 if (!sblock_other)
1162 success = 0;
96e36920 1163 }
a2de733c 1164
b968fed1
ZL		 1165 if (sctx->is_dev_replace) {
1166 /*
1167 * did not find a mirror to fetch the page
1168 * from. scrub_write_page_to_dev_replace()
1169 * handles this case (page->io_error), by
1170 * filling the block with zeros before
1171 * submitting the write request
1172 */
1173 if (!sblock_other)
1174 sblock_other = sblock_bad;
1175
1176 if (scrub_write_page_to_dev_replace(sblock_other,
1177 page_num) != 0) {
1178 btrfs_dev_replace_stats_inc(
1179 &sctx->dev_root->
1180 fs_info->dev_replace.
1181 num_write_errors);
1182 success = 0;
1183 }
1184 } else if (sblock_other) {
1185 ret = scrub_repair_page_from_good_copy(sblock_bad,
1186 sblock_other,
1187 page_num, 0);
1188 if (0 == ret)
1189 page_bad->io_error = 0;
1190 else
1191 success = 0;
b5d67f64 1192 }
a2de733c 1193 }
a2de733c 1194
b968fed1 1195 if (success && !sctx->is_dev_replace) {
b5d67f64
SB		 1196 if (is_metadata || have_csum) {
1197 /*
1198 * need to verify the checksum now that all
1199 * sectors on disk are repaired (the write
1200 * request for data to be repaired is on its way).
1201 * Just be lazy and use scrub_recheck_block()
1202 * which re-reads the data before the checksum
1203 * is verified, but most likely the data comes out
1204 * of the page cache.
1205 */
affe4a5a 1206 scrub_recheck_block(fs_info, sblock_bad, 1);
34f5c8e9 1207 if (!sblock_bad->header_error &&
b5d67f64
SB		 1208 !sblock_bad->checksum_error &&
1209 sblock_bad->no_io_error_seen)
1210 goto corrected_error;
1211 else
1212 goto did_not_correct_error;
1213 } else {
1214corrected_error:
d9d181c1
SB		 1215 spin_lock(&sctx->stat_lock);
1216 sctx->stat.corrected_errors++;
5a6ac9ea 1217 sblock_to_check->data_corrected = 1;
d9d181c1 1218 spin_unlock(&sctx->stat_lock);
b14af3b4
DS		 1219 btrfs_err_rl_in_rcu(fs_info,
1220 "fixed up error at logical %llu on dev %s",
c1c9ff7c 1221 logical, rcu_str_deref(dev->name));
8628764e 1222 }
b5d67f64
SB		 1223 } else {
1224did_not_correct_error:
d9d181c1
SB		 1225 spin_lock(&sctx->stat_lock);
1226 sctx->stat.uncorrectable_errors++;
1227 spin_unlock(&sctx->stat_lock);
b14af3b4
DS		 1228 btrfs_err_rl_in_rcu(fs_info,
1229 "unable to fixup (regular) error at logical %llu on dev %s",
c1c9ff7c 1230 logical, rcu_str_deref(dev->name));
96e36920 1231 }
a2de733c 1232
b5d67f64
SB		 1233out:
1234 if (sblocks_for_recheck) {
1235 for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS;
1236 mirror_index++) {
1237 struct scrub_block *sblock = sblocks_for_recheck +
1238 mirror_index;
af8e2d1d 1239 struct scrub_recover *recover;
b5d67f64
SB		 1240 int page_index;
1241
7a9e9987
SB		 1242 for (page_index = 0; page_index < sblock->page_count;
1243 page_index++) {
1244 sblock->pagev[page_index]->sblock = NULL;
af8e2d1d
MX		 1245 recover = sblock->pagev[page_index]->recover;
1246 if (recover) {
1247 scrub_put_recover(recover);
1248 sblock->pagev[page_index]->recover =
1249 NULL;
1250 }
7a9e9987
SB		 1251 scrub_page_put(sblock->pagev[page_index]);
1252 }
b5d67f64
SB		 1253 }
1254 kfree(sblocks_for_recheck);
1255 }
a2de733c 1256
b5d67f64
SB		 1257 return 0;
1258}
a2de733c 1259
8e5cfb55 1260static inline int scrub_nr_raid_mirrors(struct btrfs_bio *bbio)
af8e2d1d 1261{
10f11900
ZL		 1262 if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID5)
1263 return 2;
1264 else if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID6)
1265 return 3;
1266 else
af8e2d1d 1267 return (int)bbio->num_stripes;
af8e2d1d
MX		 1268}
1269
10f11900
ZL		 1270static inline void scrub_stripe_index_and_offset(u64 logical, u64 map_type,
1271 u64 *raid_map,
af8e2d1d
MX		 1272 u64 mapped_length,
1273 int nstripes, int mirror,
1274 int *stripe_index,
1275 u64 *stripe_offset)
1276{
1277 int i;
1278
ffe2d203 1279 if (map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
af8e2d1d
MX		 1280 /* RAID5/6 */
1281 for (i = 0; i < nstripes; i++) {
1282 if (raid_map[i] == RAID6_Q_STRIPE ||
1283 raid_map[i] == RAID5_P_STRIPE)
1284 continue;
1285
1286 if (logical >= raid_map[i] &&
1287 logical < raid_map[i] + mapped_length)
1288 break;
1289 }
1290
1291 *stripe_index = i;
1292 *stripe_offset = logical - raid_map[i];
1293 } else {
1294 /* The other RAID type */
1295 *stripe_index = mirror;
1296 *stripe_offset = 0;
1297 }
1298}
1299
be50a8dd 1300static int scrub_setup_recheck_block(struct scrub_block *original_sblock,
b5d67f64
SB		 1301 struct scrub_block *sblocks_for_recheck)
1302{
be50a8dd
ZL		 1303 struct scrub_ctx *sctx = original_sblock->sctx;
1304 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
1305 u64 length = original_sblock->page_count * PAGE_SIZE;
1306 u64 logical = original_sblock->pagev[0]->logical;
4734b7ed
ZL		 1307 u64 generation = original_sblock->pagev[0]->generation;
1308 u64 flags = original_sblock->pagev[0]->flags;
1309 u64 have_csum = original_sblock->pagev[0]->have_csum;
af8e2d1d
MX		 1310 struct scrub_recover *recover;
1311 struct btrfs_bio *bbio;
af8e2d1d
MX		 1312 u64 sublen;
1313 u64 mapped_length;
1314 u64 stripe_offset;
1315 int stripe_index;
be50a8dd 1316 int page_index = 0;
b5d67f64 1317 int mirror_index;
af8e2d1d 1318 int nmirrors;
b5d67f64
SB		 1319 int ret;
1320
1321 /*
57019345 1322 * note: the two members refs and outstanding_pages
b5d67f64
SB		 1323 * are not used (and not set) in the blocks that are used for
1324 * the recheck procedure
1325 */
1326
b5d67f64 1327 while (length > 0) {
af8e2d1d
MX		 1328 sublen = min_t(u64, length, PAGE_SIZE);
1329 mapped_length = sublen;
1330 bbio = NULL;
a2de733c 1331
b5d67f64
SB		 1332 /*
1333 * with a length of PAGE_SIZE, each returned stripe
1334 * represents one mirror
1335 */
af8e2d1d 1336 ret = btrfs_map_sblock(fs_info, REQ_GET_READ_MIRRORS, logical,
8e5cfb55 1337 &mapped_length, &bbio, 0, 1);
b5d67f64 1338 if (ret || !bbio || mapped_length < sublen) {
6e9606d2 1339 btrfs_put_bbio(bbio);
b5d67f64
SB		 1340 return -EIO;
1341 }
a2de733c 1342
af8e2d1d
MX		 1343 recover = kzalloc(sizeof(struct scrub_recover), GFP_NOFS);
1344 if (!recover) {
6e9606d2 1345 btrfs_put_bbio(bbio);
af8e2d1d
MX		 1346 return -ENOMEM;
1347 }
1348
1349 atomic_set(&recover->refs, 1);
1350 recover->bbio = bbio;
af8e2d1d
MX		 1351 recover->map_length = mapped_length;
1352
ff023aac 1353 BUG_ON(page_index >= SCRUB_PAGES_PER_RD_BIO);
af8e2d1d 1354
be50a8dd 1355 nmirrors = min(scrub_nr_raid_mirrors(bbio), BTRFS_MAX_MIRRORS);
10f11900 1356
af8e2d1d 1357 for (mirror_index = 0; mirror_index < nmirrors;
b5d67f64
SB		 1358 mirror_index++) {
1359 struct scrub_block *sblock;
1360 struct scrub_page *page;
1361
b5d67f64 1362 sblock = sblocks_for_recheck + mirror_index;
7a9e9987 1363 sblock->sctx = sctx;
4734b7ed 1364
7a9e9987
SB		 1365 page = kzalloc(sizeof(*page), GFP_NOFS);
1366 if (!page) {
1367leave_nomem:
d9d181c1
SB		 1368 spin_lock(&sctx->stat_lock);
1369 sctx->stat.malloc_errors++;
1370 spin_unlock(&sctx->stat_lock);
af8e2d1d 1371 scrub_put_recover(recover);
b5d67f64
SB		 1372 return -ENOMEM;
1373 }
7a9e9987
SB		 1374 scrub_page_get(page);
1375 sblock->pagev[page_index] = page;
4734b7ed
ZL		 1376 page->sblock = sblock;
1377 page->flags = flags;
1378 page->generation = generation;
7a9e9987 1379 page->logical = logical;
4734b7ed
ZL		 1380 page->have_csum = have_csum;
1381 if (have_csum)
1382 memcpy(page->csum,
1383 original_sblock->pagev[0]->csum,
1384 sctx->csum_size);
af8e2d1d 1385
10f11900
ZL		 1386 scrub_stripe_index_and_offset(logical,
1387 bbio->map_type,
1388 bbio->raid_map,
af8e2d1d 1389 mapped_length,
e34c330d
ZL		 1390 bbio->num_stripes -
1391 bbio->num_tgtdevs,
af8e2d1d
MX		 1392 mirror_index,
1393 &stripe_index,
1394 &stripe_offset);
1395 page->physical = bbio->stripes[stripe_index].physical +
1396 stripe_offset;
1397 page->dev = bbio->stripes[stripe_index].dev;
1398
ff023aac
SB		 1399 BUG_ON(page_index >= original_sblock->page_count);
1400 page->physical_for_dev_replace =
1401 original_sblock->pagev[page_index]->
1402 physical_for_dev_replace;
7a9e9987 1403 /* for missing devices, dev->bdev is NULL */
7a9e9987 1404 page->mirror_num = mirror_index + 1;
b5d67f64 1405 sblock->page_count++;
7a9e9987
SB		 1406 page->page = alloc_page(GFP_NOFS);
1407 if (!page->page)
1408 goto leave_nomem;
af8e2d1d
MX		 1409
1410 scrub_get_recover(recover);
1411 page->recover = recover;
b5d67f64 1412 }
af8e2d1d 1413 scrub_put_recover(recover);
b5d67f64
SB		 1414 length -= sublen;
1415 logical += sublen;
1416 page_index++;
1417 }
1418
1419 return 0;
96e36920
ID		 1420}
1421
af8e2d1d
MX		 1422struct scrub_bio_ret {
1423 struct completion event;
1424 int error;
1425};
1426
4246a0b6 1427static void scrub_bio_wait_endio(struct bio *bio)
af8e2d1d
MX		 1428{
1429 struct scrub_bio_ret *ret = bio->bi_private;
1430
4246a0b6 1431 ret->error = bio->bi_error;
af8e2d1d
MX		 1432 complete(&ret->event);
1433}
1434
1435static inline int scrub_is_page_on_raid56(struct scrub_page *page)
1436{
10f11900 1437 return page->recover &&
ffe2d203 1438 (page->recover->bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK);
af8e2d1d
MX		 1439}
1440
1441static int scrub_submit_raid56_bio_wait(struct btrfs_fs_info *fs_info,
1442 struct bio *bio,
1443 struct scrub_page *page)
1444{
1445 struct scrub_bio_ret done;
1446 int ret;
1447
1448 init_completion(&done.event);
1449 done.error = 0;
1450 bio->bi_iter.bi_sector = page->logical >> 9;
1451 bio->bi_private = &done;
1452 bio->bi_end_io = scrub_bio_wait_endio;
1453
1454 ret = raid56_parity_recover(fs_info->fs_root, bio, page->recover->bbio,
af8e2d1d 1455 page->recover->map_length,
4245215d 1456 page->mirror_num, 0);
af8e2d1d
MX		 1457 if (ret)
1458 return ret;
1459
1460 wait_for_completion(&done.event);
1461 if (done.error)
1462 return -EIO;
1463
1464 return 0;
1465}
1466
b5d67f64
SB		 1467/*
1468 * this function will check the on disk data for checksum errors, header
1469 * errors and read I/O errors. If any I/O errors happen, the exact pages
1470 * which are errored are marked as being bad. The goal is to enable scrub
1471 * to take those pages that are not errored from all the mirrors so that
1472 * the pages that are errored in the just handled mirror can be repaired.
1473 */
34f5c8e9 1474static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
affe4a5a
ZL		 1475 struct scrub_block *sblock,
1476 int retry_failed_mirror)
96e36920 1477{
b5d67f64 1478 int page_num;
96e36920 1479
b5d67f64 1480 sblock->no_io_error_seen = 1;
96e36920 1481
b5d67f64
SB		 1482 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1483 struct bio *bio;
7a9e9987 1484 struct scrub_page *page = sblock->pagev[page_num];
b5d67f64 1485
442a4f63 1486 if (page->dev->bdev == NULL) {
ea9947b4
SB		 1487 page->io_error = 1;
1488 sblock->no_io_error_seen = 0;
1489 continue;
1490 }
1491
7a9e9987 1492 WARN_ON(!page->page);
9be3395b 1493 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
34f5c8e9
SB		 1494 if (!bio) {
1495 page->io_error = 1;
1496 sblock->no_io_error_seen = 0;
1497 continue;
1498 }
442a4f63 1499 bio->bi_bdev = page->dev->bdev;
b5d67f64 1500
34f5c8e9 1501 bio_add_page(bio, page->page, PAGE_SIZE, 0);
af8e2d1d
MX		 1502 if (!retry_failed_mirror && scrub_is_page_on_raid56(page)) {
1503 if (scrub_submit_raid56_bio_wait(fs_info, bio, page))
1504 sblock->no_io_error_seen = 0;
1505 } else {
1506 bio->bi_iter.bi_sector = page->physical >> 9;
1507
1508 if (btrfsic_submit_bio_wait(READ, bio))
1509 sblock->no_io_error_seen = 0;
1510 }
33879d45 1511
b5d67f64
SB		 1512 bio_put(bio);
1513 }
96e36920 1514
b5d67f64 1515 if (sblock->no_io_error_seen)
ba7cf988 1516 scrub_recheck_block_checksum(sblock);
a2de733c
AJ		 1517}
1518
17a9be2f
MX		 1519static inline int scrub_check_fsid(u8 fsid[],
1520 struct scrub_page *spage)
1521{
1522 struct btrfs_fs_devices *fs_devices = spage->dev->fs_devices;
1523 int ret;
1524
1525 ret = memcmp(fsid, fs_devices->fsid, BTRFS_UUID_SIZE);
1526 return !ret;
1527}
1528
ba7cf988 1529static void scrub_recheck_block_checksum(struct scrub_block *sblock)
a2de733c 1530{
ba7cf988
ZL		 1531 sblock->header_error = 0;
1532 sblock->checksum_error = 0;
1533 sblock->generation_error = 0;
b5d67f64 1534
ba7cf988
ZL		 1535 if (sblock->pagev[0]->flags & BTRFS_EXTENT_FLAG_DATA)
1536 scrub_checksum_data(sblock);
1537 else
1538 scrub_checksum_tree_block(sblock);
a2de733c
AJ		 1539}
1540
b5d67f64 1541static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
114ab50d 1542 struct scrub_block *sblock_good)
b5d67f64
SB		 1543{
1544 int page_num;
1545 int ret = 0;
96e36920 1546
b5d67f64
SB		 1547 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
1548 int ret_sub;
96e36920 1549
b5d67f64
SB		 1550 ret_sub = scrub_repair_page_from_good_copy(sblock_bad,
1551 sblock_good,
114ab50d 1552 page_num, 1);
b5d67f64
SB		 1553 if (ret_sub)
1554 ret = ret_sub;
a2de733c 1555 }
b5d67f64
SB		 1556
1557 return ret;
1558}
1559
1560static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
1561 struct scrub_block *sblock_good,
1562 int page_num, int force_write)
1563{
7a9e9987
SB		 1564 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1565 struct scrub_page *page_good = sblock_good->pagev[page_num];
b5d67f64 1566
7a9e9987
SB		 1567 BUG_ON(page_bad->page == NULL);
1568 BUG_ON(page_good->page == NULL);
b5d67f64
SB		 1569 if (force_write || sblock_bad->header_error ||
1570 sblock_bad->checksum_error || page_bad->io_error) {
1571 struct bio *bio;
1572 int ret;
b5d67f64 1573
ff023aac 1574 if (!page_bad->dev->bdev) {
94647322 1575 btrfs_warn_rl(sblock_bad->sctx->dev_root->fs_info,
efe120a0 1576 "scrub_repair_page_from_good_copy(bdev == NULL) "
94647322 1577 "is unexpected");
ff023aac
SB		 1578 return -EIO;
1579 }
1580
9be3395b 1581 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
e627ee7b
TI		 1582 if (!bio)
1583 return -EIO;
442a4f63 1584 bio->bi_bdev = page_bad->dev->bdev;
4f024f37 1585 bio->bi_iter.bi_sector = page_bad->physical >> 9;
b5d67f64
SB		 1586
1587 ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0);
1588 if (PAGE_SIZE != ret) {
1589 bio_put(bio);
1590 return -EIO;
13db62b7 1591 }
b5d67f64 1592
33879d45 1593 if (btrfsic_submit_bio_wait(WRITE, bio)) {
442a4f63
SB		 1594 btrfs_dev_stat_inc_and_print(page_bad->dev,
1595 BTRFS_DEV_STAT_WRITE_ERRS);
ff023aac
SB		 1596 btrfs_dev_replace_stats_inc(
1597 &sblock_bad->sctx->dev_root->fs_info->
1598 dev_replace.num_write_errors);
442a4f63
SB		 1599 bio_put(bio);
1600 return -EIO;
1601 }
b5d67f64 1602 bio_put(bio);
a2de733c
AJ		 1603 }
1604
b5d67f64
SB		 1605 return 0;
1606}
1607
ff023aac
SB		 1608static void scrub_write_block_to_dev_replace(struct scrub_block *sblock)
1609{
1610 int page_num;
1611
5a6ac9ea
MX		 1612 /*
1613 * This block is used for the check of the parity on the source device,
1614 * so the data needn't be written into the destination device.
1615 */
1616 if (sblock->sparity)
1617 return;
1618
ff023aac
SB		 1619 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1620 int ret;
1621
1622 ret = scrub_write_page_to_dev_replace(sblock, page_num);
1623 if (ret)
1624 btrfs_dev_replace_stats_inc(
1625 &sblock->sctx->dev_root->fs_info->dev_replace.
1626 num_write_errors);
1627 }
1628}
1629
1630static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
1631 int page_num)
1632{
1633 struct scrub_page *spage = sblock->pagev[page_num];
1634
1635 BUG_ON(spage->page == NULL);
1636 if (spage->io_error) {
1637 void *mapped_buffer = kmap_atomic(spage->page);
1638
1639 memset(mapped_buffer, 0, PAGE_CACHE_SIZE);
1640 flush_dcache_page(spage->page);
1641 kunmap_atomic(mapped_buffer);
1642 }
1643 return scrub_add_page_to_wr_bio(sblock->sctx, spage);
1644}
1645
1646static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
1647 struct scrub_page *spage)
1648{
1649 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1650 struct scrub_bio *sbio;
1651 int ret;
1652
1653 mutex_lock(&wr_ctx->wr_lock);
1654again:
1655 if (!wr_ctx->wr_curr_bio) {
1656 wr_ctx->wr_curr_bio = kzalloc(sizeof(*wr_ctx->wr_curr_bio),
1657 GFP_NOFS);
1658 if (!wr_ctx->wr_curr_bio) {
1659 mutex_unlock(&wr_ctx->wr_lock);
1660 return -ENOMEM;
1661 }
1662 wr_ctx->wr_curr_bio->sctx = sctx;
1663 wr_ctx->wr_curr_bio->page_count = 0;
1664 }
1665 sbio = wr_ctx->wr_curr_bio;
1666 if (sbio->page_count == 0) {
1667 struct bio *bio;
1668
1669 sbio->physical = spage->physical_for_dev_replace;
1670 sbio->logical = spage->logical;
1671 sbio->dev = wr_ctx->tgtdev;
1672 bio = sbio->bio;
1673 if (!bio) {
9be3395b 1674 bio = btrfs_io_bio_alloc(GFP_NOFS, wr_ctx->pages_per_wr_bio);
ff023aac
SB		 1675 if (!bio) {
1676 mutex_unlock(&wr_ctx->wr_lock);
1677 return -ENOMEM;
1678 }
1679 sbio->bio = bio;
1680 }
1681
1682 bio->bi_private = sbio;
1683 bio->bi_end_io = scrub_wr_bio_end_io;
1684 bio->bi_bdev = sbio->dev->bdev;
4f024f37 1685 bio->bi_iter.bi_sector = sbio->physical >> 9;
ff023aac
SB		 1686 sbio->err = 0;
1687 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
1688 spage->physical_for_dev_replace ||
1689 sbio->logical + sbio->page_count * PAGE_SIZE !=
1690 spage->logical) {
1691 scrub_wr_submit(sctx);
1692 goto again;
1693 }
1694
1695 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
1696 if (ret != PAGE_SIZE) {
1697 if (sbio->page_count < 1) {
1698 bio_put(sbio->bio);
1699 sbio->bio = NULL;
1700 mutex_unlock(&wr_ctx->wr_lock);
1701 return -EIO;
1702 }
1703 scrub_wr_submit(sctx);
1704 goto again;
1705 }
1706
1707 sbio->pagev[sbio->page_count] = spage;
1708 scrub_page_get(spage);
1709 sbio->page_count++;
1710 if (sbio->page_count == wr_ctx->pages_per_wr_bio)
1711 scrub_wr_submit(sctx);
1712 mutex_unlock(&wr_ctx->wr_lock);
1713
1714 return 0;
1715}
1716
1717static void scrub_wr_submit(struct scrub_ctx *sctx)
1718{
1719 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1720 struct scrub_bio *sbio;
1721
1722 if (!wr_ctx->wr_curr_bio)
1723 return;
1724
1725 sbio = wr_ctx->wr_curr_bio;
1726 wr_ctx->wr_curr_bio = NULL;
1727 WARN_ON(!sbio->bio->bi_bdev);
1728 scrub_pending_bio_inc(sctx);
1729 /* process all writes in a single worker thread. Then the block layer
1730 * orders the requests before sending them to the driver which
1731 * doubled the write performance on spinning disks when measured
1732 * with Linux 3.5 */
1733 btrfsic_submit_bio(WRITE, sbio->bio);
1734}
1735
4246a0b6 1736static void scrub_wr_bio_end_io(struct bio *bio)
ff023aac
SB		 1737{
1738 struct scrub_bio *sbio = bio->bi_private;
1739 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
1740
4246a0b6 1741 sbio->err = bio->bi_error;
ff023aac
SB		 1742 sbio->bio = bio;
1743
9e0af237
LB		 1744 btrfs_init_work(&sbio->work, btrfs_scrubwrc_helper,
1745 scrub_wr_bio_end_io_worker, NULL, NULL);
0339ef2f 1746 btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work);
ff023aac
SB		 1747}
1748
1749static void scrub_wr_bio_end_io_worker(struct btrfs_work *work)
1750{
1751 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
1752 struct scrub_ctx *sctx = sbio->sctx;
1753 int i;
1754
1755 WARN_ON(sbio->page_count > SCRUB_PAGES_PER_WR_BIO);
1756 if (sbio->err) {
1757 struct btrfs_dev_replace *dev_replace =
1758 &sbio->sctx->dev_root->fs_info->dev_replace;
1759
1760 for (i = 0; i < sbio->page_count; i++) {
1761 struct scrub_page *spage = sbio->pagev[i];
1762
1763 spage->io_error = 1;
1764 btrfs_dev_replace_stats_inc(&dev_replace->
1765 num_write_errors);
1766 }
1767 }
1768
1769 for (i = 0; i < sbio->page_count; i++)
1770 scrub_page_put(sbio->pagev[i]);
1771
1772 bio_put(sbio->bio);
1773 kfree(sbio);
1774 scrub_pending_bio_dec(sctx);
1775}
1776
1777static int scrub_checksum(struct scrub_block *sblock)
b5d67f64
SB		 1778{
1779 u64 flags;
1780 int ret;
1781
ba7cf988
ZL		 1782 /*
1783 * No need to initialize these stats currently,
1784 * because this function only use return value
1785 * instead of these stats value.
1786 *
1787 * Todo:
1788 * always use stats
1789 */
1790 sblock->header_error = 0;
1791 sblock->generation_error = 0;
1792 sblock->checksum_error = 0;
1793
7a9e9987
SB		 1794 WARN_ON(sblock->page_count < 1);
1795 flags = sblock->pagev[0]->flags;
b5d67f64
SB		 1796 ret = 0;
1797 if (flags & BTRFS_EXTENT_FLAG_DATA)
1798 ret = scrub_checksum_data(sblock);
1799 else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1800 ret = scrub_checksum_tree_block(sblock);
1801 else if (flags & BTRFS_EXTENT_FLAG_SUPER)
1802 (void)scrub_checksum_super(sblock);
1803 else
1804 WARN_ON(1);
1805 if (ret)
1806 scrub_handle_errored_block(sblock);
ff023aac
SB		 1807
1808 return ret;
a2de733c
AJ		 1809}
1810
b5d67f64 1811static int scrub_checksum_data(struct scrub_block *sblock)
a2de733c 1812{
d9d181c1 1813 struct scrub_ctx *sctx = sblock->sctx;
a2de733c 1814 u8 csum[BTRFS_CSUM_SIZE];
b5d67f64
SB		 1815 u8 *on_disk_csum;
1816 struct page *page;
1817 void *buffer;
a2de733c 1818 u32 crc = ~(u32)0;
b5d67f64
SB		 1819 u64 len;
1820 int index;
a2de733c 1821
b5d67f64 1822 BUG_ON(sblock->page_count < 1);
7a9e9987 1823 if (!sblock->pagev[0]->have_csum)
a2de733c
AJ		 1824 return 0;
1825
7a9e9987
SB		 1826 on_disk_csum = sblock->pagev[0]->csum;
1827 page = sblock->pagev[0]->page;
9613bebb 1828 buffer = kmap_atomic(page);
b5d67f64 1829
d9d181c1 1830 len = sctx->sectorsize;
b5d67f64
SB		 1831 index = 0;
1832 for (;;) {
1833 u64 l = min_t(u64, len, PAGE_SIZE);
1834
b0496686 1835 crc = btrfs_csum_data(buffer, crc, l);
9613bebb 1836 kunmap_atomic(buffer);
b5d67f64
SB		 1837 len -= l;
1838 if (len == 0)
1839 break;
1840 index++;
1841 BUG_ON(index >= sblock->page_count);
7a9e9987
SB		 1842 BUG_ON(!sblock->pagev[index]->page);
1843 page = sblock->pagev[index]->page;
9613bebb 1844 buffer = kmap_atomic(page);
b5d67f64
SB		 1845 }
1846
a2de733c 1847 btrfs_csum_final(crc, csum);
d9d181c1 1848 if (memcmp(csum, on_disk_csum, sctx->csum_size))
ba7cf988 1849 sblock->checksum_error = 1;
a2de733c 1850
ba7cf988 1851 return sblock->checksum_error;
a2de733c
AJ		 1852}
1853
b5d67f64 1854static int scrub_checksum_tree_block(struct scrub_block *sblock)
a2de733c 1855{
d9d181c1 1856 struct scrub_ctx *sctx = sblock->sctx;
a2de733c 1857 struct btrfs_header *h;
a36cf8b8 1858 struct btrfs_root *root = sctx->dev_root;
a2de733c 1859 struct btrfs_fs_info *fs_info = root->fs_info;
b5d67f64
SB		 1860 u8 calculated_csum[BTRFS_CSUM_SIZE];
1861 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1862 struct page *page;
1863 void *mapped_buffer;
1864 u64 mapped_size;
1865 void *p;
a2de733c 1866 u32 crc = ~(u32)0;
b5d67f64
SB		 1867 u64 len;
1868 int index;
1869
1870 BUG_ON(sblock->page_count < 1);
7a9e9987 1871 page = sblock->pagev[0]->page;
9613bebb 1872 mapped_buffer = kmap_atomic(page);
b5d67f64 1873 h = (struct btrfs_header *)mapped_buffer;
d9d181c1 1874 memcpy(on_disk_csum, h->csum, sctx->csum_size);
a2de733c
AJ		 1875
1876 /*
1877 * we don't use the getter functions here, as we
1878 * a) don't have an extent buffer and
1879 * b) the page is already kmapped
1880 */
3cae210f 1881 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h))
ba7cf988 1882 sblock->header_error = 1;
a2de733c 1883
ba7cf988
ZL		 1884 if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h)) {
1885 sblock->header_error = 1;
1886 sblock->generation_error = 1;
1887 }
a2de733c 1888
17a9be2f 1889 if (!scrub_check_fsid(h->fsid, sblock->pagev[0]))
ba7cf988 1890 sblock->header_error = 1;
a2de733c
AJ		 1891
1892 if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
1893 BTRFS_UUID_SIZE))
ba7cf988 1894 sblock->header_error = 1;
a2de733c 1895
d9d181c1 1896 len = sctx->nodesize - BTRFS_CSUM_SIZE;
b5d67f64
SB		 1897 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
1898 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
1899 index = 0;
1900 for (;;) {
1901 u64 l = min_t(u64, len, mapped_size);
1902
b0496686 1903 crc = btrfs_csum_data(p, crc, l);
9613bebb 1904 kunmap_atomic(mapped_buffer);
b5d67f64
SB		 1905 len -= l;
1906 if (len == 0)
1907 break;
1908 index++;
1909 BUG_ON(index >= sblock->page_count);
7a9e9987
SB		 1910 BUG_ON(!sblock->pagev[index]->page);
1911 page = sblock->pagev[index]->page;
9613bebb 1912 mapped_buffer = kmap_atomic(page);
b5d67f64
SB		 1913 mapped_size = PAGE_SIZE;
1914 p = mapped_buffer;
1915 }
1916
1917 btrfs_csum_final(crc, calculated_csum);
d9d181c1 1918 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
ba7cf988 1919 sblock->checksum_error = 1;
a2de733c 1920
ba7cf988 1921 return sblock->header_error || sblock->checksum_error;
a2de733c
AJ		 1922}
1923
b5d67f64 1924static int scrub_checksum_super(struct scrub_block *sblock)
a2de733c
AJ		 1925{
1926 struct btrfs_super_block *s;
d9d181c1 1927 struct scrub_ctx *sctx = sblock->sctx;
b5d67f64
SB		 1928 u8 calculated_csum[BTRFS_CSUM_SIZE];
1929 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1930 struct page *page;
1931 void *mapped_buffer;
1932 u64 mapped_size;
1933 void *p;
a2de733c 1934 u32 crc = ~(u32)0;
442a4f63
SB		 1935 int fail_gen = 0;
1936 int fail_cor = 0;
b5d67f64
SB		 1937 u64 len;
1938 int index;
a2de733c 1939
b5d67f64 1940 BUG_ON(sblock->page_count < 1);
7a9e9987 1941 page = sblock->pagev[0]->page;
9613bebb 1942 mapped_buffer = kmap_atomic(page);
b5d67f64 1943 s = (struct btrfs_super_block *)mapped_buffer;
d9d181c1 1944 memcpy(on_disk_csum, s->csum, sctx->csum_size);
a2de733c 1945
3cae210f 1946 if (sblock->pagev[0]->logical != btrfs_super_bytenr(s))
442a4f63 1947 ++fail_cor;
a2de733c 1948
3cae210f 1949 if (sblock->pagev[0]->generation != btrfs_super_generation(s))
442a4f63 1950 ++fail_gen;
a2de733c 1951
17a9be2f 1952 if (!scrub_check_fsid(s->fsid, sblock->pagev[0]))
442a4f63 1953 ++fail_cor;
a2de733c 1954
b5d67f64
SB		 1955 len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE;
1956 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
1957 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
1958 index = 0;
1959 for (;;) {
1960 u64 l = min_t(u64, len, mapped_size);
1961
b0496686 1962 crc = btrfs_csum_data(p, crc, l);
9613bebb 1963 kunmap_atomic(mapped_buffer);
b5d67f64
SB		 1964 len -= l;
1965 if (len == 0)
1966 break;
1967 index++;
1968 BUG_ON(index >= sblock->page_count);
7a9e9987
SB		 1969 BUG_ON(!sblock->pagev[index]->page);
1970 page = sblock->pagev[index]->page;
9613bebb 1971 mapped_buffer = kmap_atomic(page);
b5d67f64
SB		 1972 mapped_size = PAGE_SIZE;
1973 p = mapped_buffer;
1974 }
1975
1976 btrfs_csum_final(crc, calculated_csum);
d9d181c1 1977 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
442a4f63 1978 ++fail_cor;
a2de733c 1979
442a4f63 1980 if (fail_cor + fail_gen) {
a2de733c
AJ		 1981 /*
1982 * if we find an error in a super block, we just report it.
1983 * They will get written with the next transaction commit
1984 * anyway
1985 */
d9d181c1
SB		 1986 spin_lock(&sctx->stat_lock);
1987 ++sctx->stat.super_errors;
1988 spin_unlock(&sctx->stat_lock);
442a4f63 1989 if (fail_cor)
7a9e9987 1990 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
442a4f63
SB		 1991 BTRFS_DEV_STAT_CORRUPTION_ERRS);
1992 else
7a9e9987 1993 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
442a4f63 1994 BTRFS_DEV_STAT_GENERATION_ERRS);
a2de733c
AJ		 1995 }
1996
442a4f63 1997 return fail_cor + fail_gen;
a2de733c
AJ		 1998}
1999
b5d67f64
SB		 2000static void scrub_block_get(struct scrub_block *sblock)
2001{
57019345 2002 atomic_inc(&sblock->refs);
b5d67f64
SB		 2003}
2004
2005static void scrub_block_put(struct scrub_block *sblock)
2006{
57019345 2007 if (atomic_dec_and_test(&sblock->refs)) {
b5d67f64
SB		 2008 int i;
2009
5a6ac9ea
MX		 2010 if (sblock->sparity)
2011 scrub_parity_put(sblock->sparity);
2012
b5d67f64 2013 for (i = 0; i < sblock->page_count; i++)
7a9e9987 2014 scrub_page_put(sblock->pagev[i]);
b5d67f64
SB		 2015 kfree(sblock);
2016 }
2017}
2018
7a9e9987
SB		 2019static void scrub_page_get(struct scrub_page *spage)
2020{
57019345 2021 atomic_inc(&spage->refs);
7a9e9987
SB		 2022}
2023
2024static void scrub_page_put(struct scrub_page *spage)
2025{
57019345 2026 if (atomic_dec_and_test(&spage->refs)) {
7a9e9987
SB		 2027 if (spage->page)
2028 __free_page(spage->page);
2029 kfree(spage);
2030 }
2031}
2032
d9d181c1 2033static void scrub_submit(struct scrub_ctx *sctx)
a2de733c
AJ		 2034{
2035 struct scrub_bio *sbio;
2036
d9d181c1 2037 if (sctx->curr == -1)
1623edeb 2038 return;
a2de733c 2039
d9d181c1
SB		 2040 sbio = sctx->bios[sctx->curr];
2041 sctx->curr = -1;
b6bfebc1 2042 scrub_pending_bio_inc(sctx);
03679ade 2043 btrfsic_submit_bio(READ, sbio->bio);
a2de733c
AJ		 2044}
2045
ff023aac
SB		 2046static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
2047 struct scrub_page *spage)
a2de733c 2048{
b5d67f64 2049 struct scrub_block *sblock = spage->sblock;
a2de733c 2050 struct scrub_bio *sbio;
69f4cb52 2051 int ret;
a2de733c
AJ		 2052
2053again:
2054 /*
2055 * grab a fresh bio or wait for one to become available
2056 */
d9d181c1
SB		 2057 while (sctx->curr == -1) {
2058 spin_lock(&sctx->list_lock);
2059 sctx->curr = sctx->first_free;
2060 if (sctx->curr != -1) {
2061 sctx->first_free = sctx->bios[sctx->curr]->next_free;
2062 sctx->bios[sctx->curr]->next_free = -1;
2063 sctx->bios[sctx->curr]->page_count = 0;
2064 spin_unlock(&sctx->list_lock);
a2de733c 2065 } else {
d9d181c1
SB		 2066 spin_unlock(&sctx->list_lock);
2067 wait_event(sctx->list_wait, sctx->first_free != -1);
a2de733c
AJ		 2068 }
2069 }
d9d181c1 2070 sbio = sctx->bios[sctx->curr];
b5d67f64 2071 if (sbio->page_count == 0) {
69f4cb52
AJ		 2072 struct bio *bio;
2073
b5d67f64
SB		 2074 sbio->physical = spage->physical;
2075 sbio->logical = spage->logical;
a36cf8b8 2076 sbio->dev = spage->dev;
b5d67f64
SB		 2077 bio = sbio->bio;
2078 if (!bio) {
9be3395b 2079 bio = btrfs_io_bio_alloc(GFP_NOFS, sctx->pages_per_rd_bio);
b5d67f64
SB		 2080 if (!bio)
2081 return -ENOMEM;
2082 sbio->bio = bio;
2083 }
69f4cb52
AJ		 2084
2085 bio->bi_private = sbio;
2086 bio->bi_end_io = scrub_bio_end_io;
a36cf8b8 2087 bio->bi_bdev = sbio->dev->bdev;
4f024f37 2088 bio->bi_iter.bi_sector = sbio->physical >> 9;
69f4cb52 2089 sbio->err = 0;
b5d67f64
SB		 2090 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
2091 spage->physical ||
2092 sbio->logical + sbio->page_count * PAGE_SIZE !=
a36cf8b8
SB		 2093 spage->logical ||
2094 sbio->dev != spage->dev) {
d9d181c1 2095 scrub_submit(sctx);
a2de733c
AJ		 2096 goto again;
2097 }
69f4cb52 2098
b5d67f64
SB		 2099 sbio->pagev[sbio->page_count] = spage;
2100 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
2101 if (ret != PAGE_SIZE) {
2102 if (sbio->page_count < 1) {
2103 bio_put(sbio->bio);
2104 sbio->bio = NULL;
2105 return -EIO;
2106 }
d9d181c1 2107 scrub_submit(sctx);
69f4cb52
AJ		 2108 goto again;
2109 }
2110
ff023aac 2111 scrub_block_get(sblock); /* one for the page added to the bio */
b5d67f64
SB		 2112 atomic_inc(&sblock->outstanding_pages);
2113 sbio->page_count++;
ff023aac 2114 if (sbio->page_count == sctx->pages_per_rd_bio)
d9d181c1 2115 scrub_submit(sctx);
b5d67f64
SB		 2116
2117 return 0;
2118}
2119
22365979 2120static void scrub_missing_raid56_end_io(struct bio *bio)
73ff61db
OS		 2121{
2122 struct scrub_block *sblock = bio->bi_private;
2123 struct btrfs_fs_info *fs_info = sblock->sctx->dev_root->fs_info;
2124
22365979 2125 if (bio->bi_error)
73ff61db
OS		 2126 sblock->no_io_error_seen = 0;
2127
2128 btrfs_queue_work(fs_info->scrub_workers, &sblock->work);
2129}
2130
2131static void scrub_missing_raid56_worker(struct btrfs_work *work)
2132{
2133 struct scrub_block *sblock = container_of(work, struct scrub_block, work);
2134 struct scrub_ctx *sctx = sblock->sctx;
73ff61db
OS		 2135 u64 logical;
2136 struct btrfs_device *dev;
2137
73ff61db
OS		 2138 logical = sblock->pagev[0]->logical;
2139 dev = sblock->pagev[0]->dev;
2140
affe4a5a 2141 if (sblock->no_io_error_seen)
ba7cf988 2142 scrub_recheck_block_checksum(sblock);
73ff61db
OS		 2143
2144 if (!sblock->no_io_error_seen) {
2145 spin_lock(&sctx->stat_lock);
2146 sctx->stat.read_errors++;
2147 spin_unlock(&sctx->stat_lock);
ba7cf988 2148 btrfs_err_rl_in_rcu(sctx->dev_root->fs_info,
b14af3b4 2149 "IO error rebuilding logical %llu for dev %s",
73ff61db
OS		 2150 logical, rcu_str_deref(dev->name));
2151 } else if (sblock->header_error || sblock->checksum_error) {
2152 spin_lock(&sctx->stat_lock);
2153 sctx->stat.uncorrectable_errors++;
2154 spin_unlock(&sctx->stat_lock);
ba7cf988 2155 btrfs_err_rl_in_rcu(sctx->dev_root->fs_info,
b14af3b4 2156 "failed to rebuild valid logical %llu for dev %s",
73ff61db
OS		 2157 logical, rcu_str_deref(dev->name));
2158 } else {
2159 scrub_write_block_to_dev_replace(sblock);
2160 }
2161
2162 scrub_block_put(sblock);
2163
2164 if (sctx->is_dev_replace &&
2165 atomic_read(&sctx->wr_ctx.flush_all_writes)) {
2166 mutex_lock(&sctx->wr_ctx.wr_lock);
2167 scrub_wr_submit(sctx);
2168 mutex_unlock(&sctx->wr_ctx.wr_lock);
2169 }
2170
2171 scrub_pending_bio_dec(sctx);
2172}
2173
2174static void scrub_missing_raid56_pages(struct scrub_block *sblock)
2175{
2176 struct scrub_ctx *sctx = sblock->sctx;
2177 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
2178 u64 length = sblock->page_count * PAGE_SIZE;
2179 u64 logical = sblock->pagev[0]->logical;
2180 struct btrfs_bio *bbio;
2181 struct bio *bio;
2182 struct btrfs_raid_bio *rbio;
2183 int ret;
2184 int i;
2185
2186 ret = btrfs_map_sblock(fs_info, REQ_GET_READ_MIRRORS, logical, &length,
2187 &bbio, 0, 1);
2188 if (ret || !bbio || !bbio->raid_map)
2189 goto bbio_out;
2190
2191 if (WARN_ON(!sctx->is_dev_replace ||
2192 !(bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK))) {
2193 /*
2194 * We shouldn't be scrubbing a missing device. Even for dev
2195 * replace, we should only get here for RAID 5/6. We either
2196 * managed to mount something with no mirrors remaining or
2197 * there's a bug in scrub_remap_extent()/btrfs_map_block().
2198 */
2199 goto bbio_out;
2200 }
2201
2202 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
2203 if (!bio)
2204 goto bbio_out;
2205
2206 bio->bi_iter.bi_sector = logical >> 9;
2207 bio->bi_private = sblock;
2208 bio->bi_end_io = scrub_missing_raid56_end_io;
2209
2210 rbio = raid56_alloc_missing_rbio(sctx->dev_root, bio, bbio, length);
2211 if (!rbio)
2212 goto rbio_out;
2213
2214 for (i = 0; i < sblock->page_count; i++) {
2215 struct scrub_page *spage = sblock->pagev[i];
2216
2217 raid56_add_scrub_pages(rbio, spage->page, spage->logical);
2218 }
2219
2220 btrfs_init_work(&sblock->work, btrfs_scrub_helper,
2221 scrub_missing_raid56_worker, NULL, NULL);
2222 scrub_block_get(sblock);
2223 scrub_pending_bio_inc(sctx);
2224 raid56_submit_missing_rbio(rbio);
2225 return;
2226
2227rbio_out:
2228 bio_put(bio);
2229bbio_out:
2230 btrfs_put_bbio(bbio);
2231 spin_lock(&sctx->stat_lock);
2232 sctx->stat.malloc_errors++;
2233 spin_unlock(&sctx->stat_lock);
2234}
2235
d9d181c1 2236static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
a36cf8b8 2237 u64 physical, struct btrfs_device *dev, u64 flags,
ff023aac
SB		 2238 u64 gen, int mirror_num, u8 *csum, int force,
2239 u64 physical_for_dev_replace)
b5d67f64
SB		 2240{
2241 struct scrub_block *sblock;
2242 int index;
2243
2244 sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
2245 if (!sblock) {
d9d181c1
SB		 2246 spin_lock(&sctx->stat_lock);
2247 sctx->stat.malloc_errors++;
2248 spin_unlock(&sctx->stat_lock);
b5d67f64 2249 return -ENOMEM;
a2de733c 2250 }
b5d67f64 2251
7a9e9987
SB		 2252 /* one ref inside this function, plus one for each page added to
2253 * a bio later on */
57019345 2254 atomic_set(&sblock->refs, 1);
d9d181c1 2255 sblock->sctx = sctx;
b5d67f64
SB		 2256 sblock->no_io_error_seen = 1;
2257
2258 for (index = 0; len > 0; index++) {
7a9e9987 2259 struct scrub_page *spage;
b5d67f64
SB		 2260 u64 l = min_t(u64, len, PAGE_SIZE);
2261
7a9e9987
SB		 2262 spage = kzalloc(sizeof(*spage), GFP_NOFS);
2263 if (!spage) {
2264leave_nomem:
d9d181c1
SB		 2265 spin_lock(&sctx->stat_lock);
2266 sctx->stat.malloc_errors++;
2267 spin_unlock(&sctx->stat_lock);
7a9e9987 2268 scrub_block_put(sblock);
b5d67f64
SB		 2269 return -ENOMEM;
2270 }
7a9e9987
SB		 2271 BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
2272 scrub_page_get(spage);
2273 sblock->pagev[index] = spage;
b5d67f64 2274 spage->sblock = sblock;
a36cf8b8 2275 spage->dev = dev;
b5d67f64
SB		 2276 spage->flags = flags;
2277 spage->generation = gen;
2278 spage->logical = logical;
2279 spage->physical = physical;
ff023aac 2280 spage->physical_for_dev_replace = physical_for_dev_replace;
b5d67f64
SB		 2281 spage->mirror_num = mirror_num;
2282 if (csum) {
2283 spage->have_csum = 1;
d9d181c1 2284 memcpy(spage->csum, csum, sctx->csum_size);
b5d67f64
SB		 2285 } else {
2286 spage->have_csum = 0;
2287 }
2288 sblock->page_count++;
7a9e9987
SB		 2289 spage->page = alloc_page(GFP_NOFS);
2290 if (!spage->page)
2291 goto leave_nomem;
b5d67f64
SB		 2292 len -= l;
2293 logical += l;
2294 physical += l;
ff023aac 2295 physical_for_dev_replace += l;
b5d67f64
SB		 2296 }
2297
7a9e9987 2298 WARN_ON(sblock->page_count == 0);
73ff61db
OS		 2299 if (dev->missing) {
2300 /*
2301 * This case should only be hit for RAID 5/6 device replace. See
2302 * the comment in scrub_missing_raid56_pages() for details.
2303 */
2304 scrub_missing_raid56_pages(sblock);
2305 } else {
2306 for (index = 0; index < sblock->page_count; index++) {
2307 struct scrub_page *spage = sblock->pagev[index];
2308 int ret;
1bc87793 2309
73ff61db
OS		 2310 ret = scrub_add_page_to_rd_bio(sctx, spage);
2311 if (ret) {
2312 scrub_block_put(sblock);
2313 return ret;
2314 }
b5d67f64 2315 }
a2de733c 2316
73ff61db
OS		 2317 if (force)
2318 scrub_submit(sctx);
2319 }
a2de733c 2320
b5d67f64
SB		 2321 /* last one frees, either here or in bio completion for last page */
2322 scrub_block_put(sblock);
a2de733c
AJ		 2323 return 0;
2324}
2325
4246a0b6 2326static void scrub_bio_end_io(struct bio *bio)
b5d67f64
SB		 2327{
2328 struct scrub_bio *sbio = bio->bi_private;
a36cf8b8 2329 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
b5d67f64 2330
4246a0b6 2331 sbio->err = bio->bi_error;
b5d67f64
SB		 2332 sbio->bio = bio;
2333
0339ef2f 2334 btrfs_queue_work(fs_info->scrub_workers, &sbio->work);
b5d67f64
SB		 2335}
2336
2337static void scrub_bio_end_io_worker(struct btrfs_work *work)
2338{
2339 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
d9d181c1 2340 struct scrub_ctx *sctx = sbio->sctx;
b5d67f64
SB		 2341 int i;
2342
ff023aac 2343 BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO);
b5d67f64
SB		 2344 if (sbio->err) {
2345 for (i = 0; i < sbio->page_count; i++) {
2346 struct scrub_page *spage = sbio->pagev[i];
2347
2348 spage->io_error = 1;
2349 spage->sblock->no_io_error_seen = 0;
2350 }
2351 }
2352
2353 /* now complete the scrub_block items that have all pages completed */
2354 for (i = 0; i < sbio->page_count; i++) {
2355 struct scrub_page *spage = sbio->pagev[i];
2356 struct scrub_block *sblock = spage->sblock;
2357
2358 if (atomic_dec_and_test(&sblock->outstanding_pages))
2359 scrub_block_complete(sblock);
2360 scrub_block_put(sblock);
2361 }
2362
b5d67f64
SB		 2363 bio_put(sbio->bio);
2364 sbio->bio = NULL;
d9d181c1
SB		 2365 spin_lock(&sctx->list_lock);
2366 sbio->next_free = sctx->first_free;
2367 sctx->first_free = sbio->index;
2368 spin_unlock(&sctx->list_lock);
ff023aac
SB		 2369
2370 if (sctx->is_dev_replace &&
2371 atomic_read(&sctx->wr_ctx.flush_all_writes)) {
2372 mutex_lock(&sctx->wr_ctx.wr_lock);
2373 scrub_wr_submit(sctx);
2374 mutex_unlock(&sctx->wr_ctx.wr_lock);
2375 }
2376
b6bfebc1 2377 scrub_pending_bio_dec(sctx);
b5d67f64
SB		 2378}
2379
5a6ac9ea
MX		 2380static inline void __scrub_mark_bitmap(struct scrub_parity *sparity,
2381 unsigned long *bitmap,
2382 u64 start, u64 len)
2383{
9d644a62 2384 u32 offset;
5a6ac9ea
MX		 2385 int nsectors;
2386 int sectorsize = sparity->sctx->dev_root->sectorsize;
2387
2388 if (len >= sparity->stripe_len) {
2389 bitmap_set(bitmap, 0, sparity->nsectors);
2390 return;
2391 }
2392
2393 start -= sparity->logic_start;
47c5713f 2394 start = div_u64_rem(start, sparity->stripe_len, &offset);
5a6ac9ea
MX		 2395 offset /= sectorsize;
2396 nsectors = (int)len / sectorsize;
2397
2398 if (offset + nsectors <= sparity->nsectors) {
2399 bitmap_set(bitmap, offset, nsectors);
2400 return;
2401 }
2402
2403 bitmap_set(bitmap, offset, sparity->nsectors - offset);
2404 bitmap_set(bitmap, 0, nsectors - (sparity->nsectors - offset));
2405}
2406
2407static inline void scrub_parity_mark_sectors_error(struct scrub_parity *sparity,
2408 u64 start, u64 len)
2409{
2410 __scrub_mark_bitmap(sparity, sparity->ebitmap, start, len);
2411}
2412
2413static inline void scrub_parity_mark_sectors_data(struct scrub_parity *sparity,
2414 u64 start, u64 len)
2415{
2416 __scrub_mark_bitmap(sparity, sparity->dbitmap, start, len);
2417}
2418
b5d67f64
SB		 2419static void scrub_block_complete(struct scrub_block *sblock)
2420{
5a6ac9ea
MX		 2421 int corrupted = 0;
2422
ff023aac 2423 if (!sblock->no_io_error_seen) {
5a6ac9ea 2424 corrupted = 1;
b5d67f64 2425 scrub_handle_errored_block(sblock);
ff023aac
SB		 2426 } else {
2427 /*
2428 * if has checksum error, write via repair mechanism in
2429 * dev replace case, otherwise write here in dev replace
2430 * case.
2431 */
5a6ac9ea
MX		 2432 corrupted = scrub_checksum(sblock);
2433 if (!corrupted && sblock->sctx->is_dev_replace)
ff023aac
SB		 2434 scrub_write_block_to_dev_replace(sblock);
2435 }
5a6ac9ea
MX		 2436
2437 if (sblock->sparity && corrupted && !sblock->data_corrected) {
2438 u64 start = sblock->pagev[0]->logical;
2439 u64 end = sblock->pagev[sblock->page_count - 1]->logical +
2440 PAGE_SIZE;
2441
2442 scrub_parity_mark_sectors_error(sblock->sparity,
2443 start, end - start);
2444 }
b5d67f64
SB		 2445}
2446
3b5753ec 2447static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u8 *csum)
a2de733c
AJ		 2448{
2449 struct btrfs_ordered_sum *sum = NULL;
f51a4a18 2450 unsigned long index;
a2de733c 2451 unsigned long num_sectors;
a2de733c 2452
d9d181c1
SB		 2453 while (!list_empty(&sctx->csum_list)) {
2454 sum = list_first_entry(&sctx->csum_list,
a2de733c
AJ		 2455 struct btrfs_ordered_sum, list);
2456 if (sum->bytenr > logical)
2457 return 0;
2458 if (sum->bytenr + sum->len > logical)
2459 break;
2460
d9d181c1 2461 ++sctx->stat.csum_discards;
a2de733c
AJ		 2462 list_del(&sum->list);
2463 kfree(sum);
2464 sum = NULL;
2465 }
2466 if (!sum)
2467 return 0;
2468
f51a4a18 2469 index = ((u32)(logical - sum->bytenr)) / sctx->sectorsize;
d9d181c1 2470 num_sectors = sum->len / sctx->sectorsize;
f51a4a18
MX		 2471 memcpy(csum, sum->sums + index, sctx->csum_size);
2472 if (index == num_sectors - 1) {
a2de733c
AJ		 2473 list_del(&sum->list);
2474 kfree(sum);
2475 }
f51a4a18 2476 return 1;
a2de733c
AJ		 2477}
2478
2479/* scrub extent tries to collect up to 64 kB for each bio */
d9d181c1 2480static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len,
a36cf8b8 2481 u64 physical, struct btrfs_device *dev, u64 flags,
ff023aac 2482 u64 gen, int mirror_num, u64 physical_for_dev_replace)
a2de733c
AJ		 2483{
2484 int ret;
2485 u8 csum[BTRFS_CSUM_SIZE];
b5d67f64
SB		 2486 u32 blocksize;
2487
2488 if (flags & BTRFS_EXTENT_FLAG_DATA) {
d9d181c1
SB		 2489 blocksize = sctx->sectorsize;
2490 spin_lock(&sctx->stat_lock);
2491 sctx->stat.data_extents_scrubbed++;
2492 sctx->stat.data_bytes_scrubbed += len;
2493 spin_unlock(&sctx->stat_lock);
b5d67f64 2494 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
d9d181c1
SB		 2495 blocksize = sctx->nodesize;
2496 spin_lock(&sctx->stat_lock);
2497 sctx->stat.tree_extents_scrubbed++;
2498 sctx->stat.tree_bytes_scrubbed += len;
2499 spin_unlock(&sctx->stat_lock);
b5d67f64 2500 } else {
d9d181c1 2501 blocksize = sctx->sectorsize;
ff023aac 2502 WARN_ON(1);
b5d67f64 2503 }
a2de733c
AJ		 2504
2505 while (len) {
b5d67f64 2506 u64 l = min_t(u64, len, blocksize);
a2de733c
AJ		 2507 int have_csum = 0;
2508
2509 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2510 /* push csums to sbio */
3b5753ec 2511 have_csum = scrub_find_csum(sctx, logical, csum);
a2de733c 2512 if (have_csum == 0)
d9d181c1 2513 ++sctx->stat.no_csum;
ff023aac
SB		 2514 if (sctx->is_dev_replace && !have_csum) {
2515 ret = copy_nocow_pages(sctx, logical, l,
2516 mirror_num,
2517 physical_for_dev_replace);
2518 goto behind_scrub_pages;
2519 }
a2de733c 2520 }
a36cf8b8 2521 ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen,
ff023aac
SB		 2522 mirror_num, have_csum ? csum : NULL, 0,
2523 physical_for_dev_replace);
2524behind_scrub_pages:
a2de733c
AJ		 2525 if (ret)
2526 return ret;
2527 len -= l;
2528 logical += l;
2529 physical += l;
ff023aac 2530 physical_for_dev_replace += l;
a2de733c
AJ		 2531 }
2532 return 0;
2533}
2534
5a6ac9ea
MX		 2535static int scrub_pages_for_parity(struct scrub_parity *sparity,
2536 u64 logical, u64 len,
2537 u64 physical, struct btrfs_device *dev,
2538 u64 flags, u64 gen, int mirror_num, u8 *csum)
2539{
2540 struct scrub_ctx *sctx = sparity->sctx;
2541 struct scrub_block *sblock;
2542 int index;
2543
2544 sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
2545 if (!sblock) {
2546 spin_lock(&sctx->stat_lock);
2547 sctx->stat.malloc_errors++;
2548 spin_unlock(&sctx->stat_lock);
2549 return -ENOMEM;
2550 }
2551
2552 /* one ref inside this function, plus one for each page added to
2553 * a bio later on */
57019345 2554 atomic_set(&sblock->refs, 1);
5a6ac9ea
MX		 2555 sblock->sctx = sctx;
2556 sblock->no_io_error_seen = 1;
2557 sblock->sparity = sparity;
2558 scrub_parity_get(sparity);
2559
2560 for (index = 0; len > 0; index++) {
2561 struct scrub_page *spage;
2562 u64 l = min_t(u64, len, PAGE_SIZE);
2563
2564 spage = kzalloc(sizeof(*spage), GFP_NOFS);
2565 if (!spage) {
2566leave_nomem:
2567 spin_lock(&sctx->stat_lock);
2568 sctx->stat.malloc_errors++;
2569 spin_unlock(&sctx->stat_lock);
2570 scrub_block_put(sblock);
2571 return -ENOMEM;
2572 }
2573 BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
2574 /* For scrub block */
2575 scrub_page_get(spage);
2576 sblock->pagev[index] = spage;
2577 /* For scrub parity */
2578 scrub_page_get(spage);
2579 list_add_tail(&spage->list, &sparity->spages);
2580 spage->sblock = sblock;
2581 spage->dev = dev;
2582 spage->flags = flags;
2583 spage->generation = gen;
2584 spage->logical = logical;
2585 spage->physical = physical;
2586 spage->mirror_num = mirror_num;
2587 if (csum) {
2588 spage->have_csum = 1;
2589 memcpy(spage->csum, csum, sctx->csum_size);
2590 } else {
2591 spage->have_csum = 0;
2592 }
2593 sblock->page_count++;
2594 spage->page = alloc_page(GFP_NOFS);
2595 if (!spage->page)
2596 goto leave_nomem;
2597 len -= l;
2598 logical += l;
2599 physical += l;
2600 }
2601
2602 WARN_ON(sblock->page_count == 0);
2603 for (index = 0; index < sblock->page_count; index++) {
2604 struct scrub_page *spage = sblock->pagev[index];
2605 int ret;
2606
2607 ret = scrub_add_page_to_rd_bio(sctx, spage);
2608 if (ret) {
2609 scrub_block_put(sblock);
2610 return ret;
2611 }
2612 }
2613
2614 /* last one frees, either here or in bio completion for last page */
2615 scrub_block_put(sblock);
2616 return 0;
2617}
2618
2619static int scrub_extent_for_parity(struct scrub_parity *sparity,
2620 u64 logical, u64 len,
2621 u64 physical, struct btrfs_device *dev,
2622 u64 flags, u64 gen, int mirror_num)
2623{
2624 struct scrub_ctx *sctx = sparity->sctx;
2625 int ret;
2626 u8 csum[BTRFS_CSUM_SIZE];
2627 u32 blocksize;
2628
4a770891
OS		 2629 if (dev->missing) {
2630 scrub_parity_mark_sectors_error(sparity, logical, len);
2631 return 0;
2632 }
2633
5a6ac9ea
MX		 2634 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2635 blocksize = sctx->sectorsize;
2636 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2637 blocksize = sctx->nodesize;
2638 } else {
2639 blocksize = sctx->sectorsize;
2640 WARN_ON(1);
2641 }
2642
2643 while (len) {
2644 u64 l = min_t(u64, len, blocksize);
2645 int have_csum = 0;
2646
2647 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2648 /* push csums to sbio */
3b5753ec 2649 have_csum = scrub_find_csum(sctx, logical, csum);
5a6ac9ea
MX		 2650 if (have_csum == 0)
2651 goto skip;
2652 }
2653 ret = scrub_pages_for_parity(sparity, logical, l, physical, dev,
2654 flags, gen, mirror_num,
2655 have_csum ? csum : NULL);
5a6ac9ea
MX		 2656 if (ret)
2657 return ret;
6b6d24b3 2658skip:
5a6ac9ea
MX		 2659 len -= l;
2660 logical += l;
2661 physical += l;
2662 }
2663 return 0;
2664}
2665
3b080b25
WS		 2666/*
2667 * Given a physical address, this will calculate it's
2668 * logical offset. if this is a parity stripe, it will return
2669 * the most left data stripe's logical offset.
2670 *
2671 * return 0 if it is a data stripe, 1 means parity stripe.
2672 */
2673static int get_raid56_logic_offset(u64 physical, int num,
5a6ac9ea
MX		 2674 struct map_lookup *map, u64 *offset,
2675 u64 *stripe_start)
3b080b25
WS		 2676{
2677 int i;
2678 int j = 0;
2679 u64 stripe_nr;
2680 u64 last_offset;
9d644a62
DS		 2681 u32 stripe_index;
2682 u32 rot;
3b080b25
WS		 2683
2684 last_offset = (physical - map->stripes[num].physical) *
2685 nr_data_stripes(map);
5a6ac9ea
MX		 2686 if (stripe_start)
2687 *stripe_start = last_offset;
2688
3b080b25
WS		 2689 *offset = last_offset;
2690 for (i = 0; i < nr_data_stripes(map); i++) {
2691 *offset = last_offset + i * map->stripe_len;
2692
b8b93add
DS		 2693 stripe_nr = div_u64(*offset, map->stripe_len);
2694 stripe_nr = div_u64(stripe_nr, nr_data_stripes(map));
3b080b25
WS		 2695
2696 /* Work out the disk rotation on this stripe-set */
47c5713f 2697 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, &rot);
3b080b25
WS		 2698 /* calculate which stripe this data locates */
2699 rot += i;
e4fbaee2 2700 stripe_index = rot % map->num_stripes;
3b080b25
WS		 2701 if (stripe_index == num)
2702 return 0;
2703 if (stripe_index < num)
2704 j++;
2705 }
2706 *offset = last_offset + j * map->stripe_len;
2707 return 1;
2708}
2709
5a6ac9ea
MX		 2710static void scrub_free_parity(struct scrub_parity *sparity)
2711{
2712 struct scrub_ctx *sctx = sparity->sctx;
2713 struct scrub_page *curr, *next;
2714 int nbits;
2715
2716 nbits = bitmap_weight(sparity->ebitmap, sparity->nsectors);
2717 if (nbits) {
2718 spin_lock(&sctx->stat_lock);
2719 sctx->stat.read_errors += nbits;
2720 sctx->stat.uncorrectable_errors += nbits;
2721 spin_unlock(&sctx->stat_lock);
2722 }
2723
2724 list_for_each_entry_safe(curr, next, &sparity->spages, list) {
2725 list_del_init(&curr->list);
2726 scrub_page_put(curr);
2727 }
2728
2729 kfree(sparity);
2730}
2731
20b2e302
ZL		 2732static void scrub_parity_bio_endio_worker(struct btrfs_work *work)
2733{
2734 struct scrub_parity *sparity = container_of(work, struct scrub_parity,
2735 work);
2736 struct scrub_ctx *sctx = sparity->sctx;
2737
2738 scrub_free_parity(sparity);
2739 scrub_pending_bio_dec(sctx);
2740}
2741
4246a0b6 2742static void scrub_parity_bio_endio(struct bio *bio)
5a6ac9ea
MX		 2743{
2744 struct scrub_parity *sparity = (struct scrub_parity *)bio->bi_private;
5a6ac9ea 2745
4246a0b6 2746 if (bio->bi_error)
5a6ac9ea
MX		 2747 bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap,
2748 sparity->nsectors);
2749
5a6ac9ea 2750 bio_put(bio);
20b2e302
ZL		 2751
2752 btrfs_init_work(&sparity->work, btrfs_scrubparity_helper,
2753 scrub_parity_bio_endio_worker, NULL, NULL);
2754 btrfs_queue_work(sparity->sctx->dev_root->fs_info->scrub_parity_workers,
2755 &sparity->work);
5a6ac9ea
MX		 2756}
2757
2758static void scrub_parity_check_and_repair(struct scrub_parity *sparity)
2759{
2760 struct scrub_ctx *sctx = sparity->sctx;
2761 struct bio *bio;
2762 struct btrfs_raid_bio *rbio;
2763 struct scrub_page *spage;
2764 struct btrfs_bio *bbio = NULL;
5a6ac9ea
MX		 2765 u64 length;
2766 int ret;
2767
2768 if (!bitmap_andnot(sparity->dbitmap, sparity->dbitmap, sparity->ebitmap,
2769 sparity->nsectors))
2770 goto out;
2771
a0dd59de 2772 length = sparity->logic_end - sparity->logic_start;
76035976 2773 ret = btrfs_map_sblock(sctx->dev_root->fs_info, WRITE,
5a6ac9ea 2774 sparity->logic_start,
8e5cfb55
ZL		 2775 &length, &bbio, 0, 1);
2776 if (ret || !bbio || !bbio->raid_map)
5a6ac9ea
MX		 2777 goto bbio_out;
2778
2779 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
2780 if (!bio)
2781 goto bbio_out;
2782
2783 bio->bi_iter.bi_sector = sparity->logic_start >> 9;
2784 bio->bi_private = sparity;
2785 bio->bi_end_io = scrub_parity_bio_endio;
2786
2787 rbio = raid56_parity_alloc_scrub_rbio(sctx->dev_root, bio, bbio,
8e5cfb55 2788 length, sparity->scrub_dev,
5a6ac9ea
MX		 2789 sparity->dbitmap,
2790 sparity->nsectors);
2791 if (!rbio)
2792 goto rbio_out;
2793
2794 list_for_each_entry(spage, &sparity->spages, list)
b4ee1782 2795 raid56_add_scrub_pages(rbio, spage->page, spage->logical);
5a6ac9ea
MX		 2796
2797 scrub_pending_bio_inc(sctx);
2798 raid56_parity_submit_scrub_rbio(rbio);
2799 return;
2800
2801rbio_out:
2802 bio_put(bio);
2803bbio_out:
6e9606d2 2804 btrfs_put_bbio(bbio);
5a6ac9ea
MX		 2805 bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap,
2806 sparity->nsectors);
2807 spin_lock(&sctx->stat_lock);
2808 sctx->stat.malloc_errors++;
2809 spin_unlock(&sctx->stat_lock);
2810out:
2811 scrub_free_parity(sparity);
2812}
2813
2814static inline int scrub_calc_parity_bitmap_len(int nsectors)
2815{
bfca9a6d 2816 return DIV_ROUND_UP(nsectors, BITS_PER_LONG) * sizeof(long);
5a6ac9ea
MX		 2817}
2818
2819static void scrub_parity_get(struct scrub_parity *sparity)
2820{
57019345 2821 atomic_inc(&sparity->refs);
5a6ac9ea
MX		 2822}
2823
2824static void scrub_parity_put(struct scrub_parity *sparity)
2825{
57019345 2826 if (!atomic_dec_and_test(&sparity->refs))
5a6ac9ea
MX		 2827 return;
2828
2829 scrub_parity_check_and_repair(sparity);
2830}
2831
2832static noinline_for_stack int scrub_raid56_parity(struct scrub_ctx *sctx,
2833 struct map_lookup *map,
2834 struct btrfs_device *sdev,
2835 struct btrfs_path *path,
2836 u64 logic_start,
2837 u64 logic_end)
2838{
2839 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
2840 struct btrfs_root *root = fs_info->extent_root;
2841 struct btrfs_root *csum_root = fs_info->csum_root;
2842 struct btrfs_extent_item *extent;
4a770891 2843 struct btrfs_bio *bbio = NULL;
5a6ac9ea
MX		 2844 u64 flags;
2845 int ret;
2846 int slot;
2847 struct extent_buffer *l;
2848 struct btrfs_key key;
2849 u64 generation;
2850 u64 extent_logical;
2851 u64 extent_physical;
2852 u64 extent_len;
4a770891 2853 u64 mapped_length;
5a6ac9ea
MX		 2854 struct btrfs_device *extent_dev;
2855 struct scrub_parity *sparity;
2856 int nsectors;
2857 int bitmap_len;
2858 int extent_mirror_num;
2859 int stop_loop = 0;
2860
2861 nsectors = map->stripe_len / root->sectorsize;
2862 bitmap_len = scrub_calc_parity_bitmap_len(nsectors);
2863 sparity = kzalloc(sizeof(struct scrub_parity) + 2 * bitmap_len,
2864 GFP_NOFS);
2865 if (!sparity) {
2866 spin_lock(&sctx->stat_lock);
2867 sctx->stat.malloc_errors++;
2868 spin_unlock(&sctx->stat_lock);
2869 return -ENOMEM;
2870 }
2871
2872 sparity->stripe_len = map->stripe_len;
2873 sparity->nsectors = nsectors;
2874 sparity->sctx = sctx;
2875 sparity->scrub_dev = sdev;
2876 sparity->logic_start = logic_start;
2877 sparity->logic_end = logic_end;
57019345 2878 atomic_set(&sparity->refs, 1);
5a6ac9ea
MX		 2879 INIT_LIST_HEAD(&sparity->spages);
2880 sparity->dbitmap = sparity->bitmap;
2881 sparity->ebitmap = (void *)sparity->bitmap + bitmap_len;
2882
2883 ret = 0;
2884 while (logic_start < logic_end) {
2885 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2886 key.type = BTRFS_METADATA_ITEM_KEY;
2887 else
2888 key.type = BTRFS_EXTENT_ITEM_KEY;
2889 key.objectid = logic_start;
2890 key.offset = (u64)-1;
2891
2892 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2893 if (ret < 0)
2894 goto out;
2895
2896 if (ret > 0) {
2897 ret = btrfs_previous_extent_item(root, path, 0);
2898 if (ret < 0)
2899 goto out;
2900 if (ret > 0) {
2901 btrfs_release_path(path);
2902 ret = btrfs_search_slot(NULL, root, &key,
2903 path, 0, 0);
2904 if (ret < 0)
2905 goto out;
2906 }
2907 }
2908
2909 stop_loop = 0;
2910 while (1) {
2911 u64 bytes;
2912
2913 l = path->nodes[0];
2914 slot = path->slots[0];
2915 if (slot >= btrfs_header_nritems(l)) {
2916 ret = btrfs_next_leaf(root, path);
2917 if (ret == 0)
2918 continue;
2919 if (ret < 0)
2920 goto out;
2921
2922 stop_loop = 1;
2923 break;
2924 }
2925 btrfs_item_key_to_cpu(l, &key, slot);
2926
d7cad238
ZL		 2927 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
2928 key.type != BTRFS_METADATA_ITEM_KEY)
2929 goto next;
2930
5a6ac9ea
MX		 2931 if (key.type == BTRFS_METADATA_ITEM_KEY)
2932 bytes = root->nodesize;
2933 else
2934 bytes = key.offset;
2935
2936 if (key.objectid + bytes <= logic_start)
2937 goto next;
2938
a0dd59de 2939 if (key.objectid >= logic_end) {
5a6ac9ea
MX		 2940 stop_loop = 1;
2941 break;
2942 }
2943
2944 while (key.objectid >= logic_start + map->stripe_len)
2945 logic_start += map->stripe_len;
2946
2947 extent = btrfs_item_ptr(l, slot,
2948 struct btrfs_extent_item);
2949 flags = btrfs_extent_flags(l, extent);
2950 generation = btrfs_extent_generation(l, extent);
2951
a323e813
ZL		 2952 if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) &&
2953 (key.objectid < logic_start ||
2954 key.objectid + bytes >
2955 logic_start + map->stripe_len)) {
2956 btrfs_err(fs_info, "scrub: tree block %llu spanning stripes, ignored. logical=%llu",
2957 key.objectid, logic_start);
9799d2c3
ZL		 2958 spin_lock(&sctx->stat_lock);
2959 sctx->stat.uncorrectable_errors++;
2960 spin_unlock(&sctx->stat_lock);
5a6ac9ea
MX		 2961 goto next;
2962 }
2963again:
2964 extent_logical = key.objectid;
2965 extent_len = bytes;
2966
2967 if (extent_logical < logic_start) {
2968 extent_len -= logic_start - extent_logical;
2969 extent_logical = logic_start;
2970 }
2971
2972 if (extent_logical + extent_len >
2973 logic_start + map->stripe_len)
2974 extent_len = logic_start + map->stripe_len -
2975 extent_logical;
2976
2977 scrub_parity_mark_sectors_data(sparity, extent_logical,
2978 extent_len);
2979
4a770891
OS		 2980 mapped_length = extent_len;
2981 ret = btrfs_map_block(fs_info, READ, extent_logical,
2982 &mapped_length, &bbio, 0);
2983 if (!ret) {
2984 if (!bbio || mapped_length < extent_len)
2985 ret = -EIO;
2986 }
2987 if (ret) {
2988 btrfs_put_bbio(bbio);
2989 goto out;
2990 }
2991 extent_physical = bbio->stripes[0].physical;
2992 extent_mirror_num = bbio->mirror_num;
2993 extent_dev = bbio->stripes[0].dev;
2994 btrfs_put_bbio(bbio);
5a6ac9ea
MX		 2995
2996 ret = btrfs_lookup_csums_range(csum_root,
2997 extent_logical,
2998 extent_logical + extent_len - 1,
2999 &sctx->csum_list, 1);
3000 if (ret)
3001 goto out;
3002
3003 ret = scrub_extent_for_parity(sparity, extent_logical,
3004 extent_len,
3005 extent_physical,
3006 extent_dev, flags,
3007 generation,
3008 extent_mirror_num);
6fa96d72
ZL		 3009
3010 scrub_free_csums(sctx);
3011
5a6ac9ea
MX		 3012 if (ret)
3013 goto out;
3014
5a6ac9ea
MX		 3015 if (extent_logical + extent_len <
3016 key.objectid + bytes) {
3017 logic_start += map->stripe_len;
3018
3019 if (logic_start >= logic_end) {
3020 stop_loop = 1;
3021 break;
3022 }
3023
3024 if (logic_start < key.objectid + bytes) {
3025 cond_resched();
3026 goto again;
3027 }
3028 }
3029next:
3030 path->slots[0]++;
3031 }
3032
3033 btrfs_release_path(path);
3034
3035 if (stop_loop)
3036 break;
3037
3038 logic_start += map->stripe_len;
3039 }
3040out:
3041 if (ret < 0)
3042 scrub_parity_mark_sectors_error(sparity, logic_start,
a0dd59de 3043 logic_end - logic_start);
5a6ac9ea
MX		 3044 scrub_parity_put(sparity);
3045 scrub_submit(sctx);
3046 mutex_lock(&sctx->wr_ctx.wr_lock);
3047 scrub_wr_submit(sctx);
3048 mutex_unlock(&sctx->wr_ctx.wr_lock);
3049
3050 btrfs_release_path(path);
3051 return ret < 0 ? ret : 0;
3052}
3053
d9d181c1 3054static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
a36cf8b8
SB		 3055 struct map_lookup *map,
3056 struct btrfs_device *scrub_dev,
ff023aac
SB		 3057 int num, u64 base, u64 length,
3058 int is_dev_replace)
a2de733c 3059{
5a6ac9ea 3060 struct btrfs_path *path, *ppath;
a36cf8b8 3061 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
a2de733c
AJ		 3062 struct btrfs_root *root = fs_info->extent_root;
3063 struct btrfs_root *csum_root = fs_info->csum_root;
3064 struct btrfs_extent_item *extent;
e7786c3a 3065 struct blk_plug plug;
a2de733c
AJ		 3066 u64 flags;
3067 int ret;
3068 int slot;
a2de733c 3069 u64 nstripes;
a2de733c
AJ		 3070 struct extent_buffer *l;
3071 struct btrfs_key key;
3072 u64 physical;
3073 u64 logical;
625f1c8d 3074 u64 logic_end;
3b080b25 3075 u64 physical_end;
a2de733c 3076 u64 generation;
e12fa9cd 3077 int mirror_num;
7a26285e
AJ		 3078 struct reada_control *reada1;
3079 struct reada_control *reada2;
3080 struct btrfs_key key_start;
3081 struct btrfs_key key_end;
a2de733c
AJ		 3082 u64 increment = map->stripe_len;
3083 u64 offset;
ff023aac
SB		 3084 u64 extent_logical;
3085 u64 extent_physical;
3086 u64 extent_len;
5a6ac9ea
MX		 3087 u64 stripe_logical;
3088 u64 stripe_end;
ff023aac
SB		 3089 struct btrfs_device *extent_dev;
3090 int extent_mirror_num;
3b080b25 3091 int stop_loop = 0;
53b381b3 3092
3b080b25 3093 physical = map->stripes[num].physical;
a2de733c 3094 offset = 0;
b8b93add 3095 nstripes = div_u64(length, map->stripe_len);
a2de733c
AJ		 3096 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3097 offset = map->stripe_len * num;
3098 increment = map->stripe_len * map->num_stripes;
193ea74b 3099 mirror_num = 1;
a2de733c
AJ		 3100 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3101 int factor = map->num_stripes / map->sub_stripes;
3102 offset = map->stripe_len * (num / map->sub_stripes);
3103 increment = map->stripe_len * factor;
193ea74b 3104 mirror_num = num % map->sub_stripes + 1;
a2de733c
AJ		 3105 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3106 increment = map->stripe_len;
193ea74b 3107 mirror_num = num % map->num_stripes + 1;
a2de733c
AJ		 3108 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3109 increment = map->stripe_len;
193ea74b 3110 mirror_num = num % map->num_stripes + 1;
ffe2d203 3111 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5a6ac9ea 3112 get_raid56_logic_offset(physical, num, map, &offset, NULL);
3b080b25
WS		 3113 increment = map->stripe_len * nr_data_stripes(map);
3114 mirror_num = 1;
a2de733c
AJ		 3115 } else {
3116 increment = map->stripe_len;
193ea74b 3117 mirror_num = 1;
a2de733c
AJ		 3118 }
3119
3120 path = btrfs_alloc_path();
3121 if (!path)
3122 return -ENOMEM;
3123
5a6ac9ea
MX		 3124 ppath = btrfs_alloc_path();
3125 if (!ppath) {
379d6854 3126 btrfs_free_path(path);
5a6ac9ea
MX		 3127 return -ENOMEM;
3128 }
3129
b5d67f64
SB		 3130 /*
3131 * work on commit root. The related disk blocks are static as
3132 * long as COW is applied. This means, it is save to rewrite
3133 * them to repair disk errors without any race conditions
3134 */
a2de733c
AJ		 3135 path->search_commit_root = 1;
3136 path->skip_locking = 1;
3137
063c54dc
GH		 3138 ppath->search_commit_root = 1;
3139 ppath->skip_locking = 1;
a2de733c 3140 /*
7a26285e
AJ		 3141 * trigger the readahead for extent tree csum tree and wait for
3142 * completion. During readahead, the scrub is officially paused
3143 * to not hold off transaction commits
a2de733c
AJ		 3144 */
3145 logical = base + offset;
3b080b25 3146 physical_end = physical + nstripes * map->stripe_len;
ffe2d203 3147 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3b080b25 3148 get_raid56_logic_offset(physical_end, num,
5a6ac9ea 3149 map, &logic_end, NULL);
3b080b25
WS		 3150 logic_end += base;
3151 } else {
3152 logic_end = logical + increment * nstripes;
3153 }
d9d181c1 3154 wait_event(sctx->list_wait,
b6bfebc1 3155 atomic_read(&sctx->bios_in_flight) == 0);
cb7ab021 3156 scrub_blocked_if_needed(fs_info);
7a26285e
AJ		 3157
3158 /* FIXME it might be better to start readahead at commit root */
3159 key_start.objectid = logical;
3160 key_start.type = BTRFS_EXTENT_ITEM_KEY;
3161 key_start.offset = (u64)0;
3b080b25 3162 key_end.objectid = logic_end;
3173a18f
JB		 3163 key_end.type = BTRFS_METADATA_ITEM_KEY;
3164 key_end.offset = (u64)-1;
7a26285e
AJ		 3165 reada1 = btrfs_reada_add(root, &key_start, &key_end);
3166
3167 key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
3168 key_start.type = BTRFS_EXTENT_CSUM_KEY;
3169 key_start.offset = logical;
3170 key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
3171 key_end.type = BTRFS_EXTENT_CSUM_KEY;
3b080b25 3172 key_end.offset = logic_end;
7a26285e
AJ		 3173 reada2 = btrfs_reada_add(csum_root, &key_start, &key_end);
3174
3175 if (!IS_ERR(reada1))
3176 btrfs_reada_wait(reada1);
3177 if (!IS_ERR(reada2))
3178 btrfs_reada_wait(reada2);
3179
a2de733c
AJ		 3180
3181 /*
3182 * collect all data csums for the stripe to avoid seeking during
3183 * the scrub. This might currently (crc32) end up to be about 1MB
3184 */
e7786c3a 3185 blk_start_plug(&plug);
a2de733c 3186
a2de733c
AJ		 3187 /*
3188 * now find all extents for each stripe and scrub them
3189 */
a2de733c 3190 ret = 0;
3b080b25 3191 while (physical < physical_end) {
a2de733c
AJ		 3192 /*
3193 * canceled?
3194 */
3195 if (atomic_read(&fs_info->scrub_cancel_req) ||
d9d181c1 3196 atomic_read(&sctx->cancel_req)) {
a2de733c
AJ		 3197 ret = -ECANCELED;
3198 goto out;
3199 }
3200 /*
3201 * check to see if we have to pause
3202 */
3203 if (atomic_read(&fs_info->scrub_pause_req)) {
3204 /* push queued extents */
ff023aac 3205 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
d9d181c1 3206 scrub_submit(sctx);
ff023aac
SB		 3207 mutex_lock(&sctx->wr_ctx.wr_lock);
3208 scrub_wr_submit(sctx);
3209 mutex_unlock(&sctx->wr_ctx.wr_lock);
d9d181c1 3210 wait_event(sctx->list_wait,
b6bfebc1 3211 atomic_read(&sctx->bios_in_flight) == 0);
ff023aac 3212 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
3cb0929a 3213 scrub_blocked_if_needed(fs_info);
a2de733c
AJ		 3214 }
3215
f2f66a2f
ZL		 3216 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3217 ret = get_raid56_logic_offset(physical, num, map,
3218 &logical,
3219 &stripe_logical);
3220 logical += base;
3221 if (ret) {
7955323b 3222 /* it is parity strip */
f2f66a2f 3223 stripe_logical += base;
a0dd59de 3224 stripe_end = stripe_logical + increment;
f2f66a2f
ZL		 3225 ret = scrub_raid56_parity(sctx, map, scrub_dev,
3226 ppath, stripe_logical,
3227 stripe_end);
3228 if (ret)
3229 goto out;
3230 goto skip;
3231 }
3232 }
3233
7c76edb7
WS		 3234 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
3235 key.type = BTRFS_METADATA_ITEM_KEY;
3236 else
3237 key.type = BTRFS_EXTENT_ITEM_KEY;
a2de733c 3238 key.objectid = logical;
625f1c8d 3239 key.offset = (u64)-1;
a2de733c
AJ		 3240
3241 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3242 if (ret < 0)
3243 goto out;
3173a18f 3244
8c51032f 3245 if (ret > 0) {
ade2e0b3 3246 ret = btrfs_previous_extent_item(root, path, 0);
a2de733c
AJ		 3247 if (ret < 0)
3248 goto out;
8c51032f
AJ		 3249 if (ret > 0) {
3250 /* there's no smaller item, so stick with the
3251 * larger one */
3252 btrfs_release_path(path);
3253 ret = btrfs_search_slot(NULL, root, &key,
3254 path, 0, 0);
3255 if (ret < 0)
3256 goto out;
3257 }
a2de733c
AJ		 3258 }
3259
625f1c8d 3260 stop_loop = 0;
a2de733c 3261 while (1) {
3173a18f
JB		 3262 u64 bytes;
3263
a2de733c
AJ		 3264 l = path->nodes[0];
3265 slot = path->slots[0];
3266 if (slot >= btrfs_header_nritems(l)) {
3267 ret = btrfs_next_leaf(root, path);
3268 if (ret == 0)
3269 continue;
3270 if (ret < 0)
3271 goto out;
3272
625f1c8d 3273 stop_loop = 1;
a2de733c
AJ		 3274 break;
3275 }
3276 btrfs_item_key_to_cpu(l, &key, slot);
3277
d7cad238
ZL		 3278 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3279 key.type != BTRFS_METADATA_ITEM_KEY)
3280 goto next;
3281
3173a18f 3282 if (key.type == BTRFS_METADATA_ITEM_KEY)
707e8a07 3283 bytes = root->nodesize;
3173a18f
JB		 3284 else
3285 bytes = key.offset;
3286
3287 if (key.objectid + bytes <= logical)
a2de733c
AJ		 3288 goto next;
3289
625f1c8d
LB		 3290 if (key.objectid >= logical + map->stripe_len) {
3291 /* out of this device extent */
3292 if (key.objectid >= logic_end)
3293 stop_loop = 1;
3294 break;
3295 }
a2de733c
AJ		 3296
3297 extent = btrfs_item_ptr(l, slot,
3298 struct btrfs_extent_item);
3299 flags = btrfs_extent_flags(l, extent);
3300 generation = btrfs_extent_generation(l, extent);
3301
a323e813
ZL		 3302 if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) &&
3303 (key.objectid < logical ||
3304 key.objectid + bytes >
3305 logical + map->stripe_len)) {
efe120a0
FH		 3306 btrfs_err(fs_info,
3307 "scrub: tree block %llu spanning "
3308 "stripes, ignored. logical=%llu",
c1c9ff7c 3309 key.objectid, logical);
9799d2c3
ZL		 3310 spin_lock(&sctx->stat_lock);
3311 sctx->stat.uncorrectable_errors++;
3312 spin_unlock(&sctx->stat_lock);
a2de733c
AJ		 3313 goto next;
3314 }
3315
625f1c8d
LB		 3316again:
3317 extent_logical = key.objectid;
3318 extent_len = bytes;
3319
a2de733c
AJ		 3320 /*
3321 * trim extent to this stripe
3322 */
625f1c8d
LB		 3323 if (extent_logical < logical) {
3324 extent_len -= logical - extent_logical;
3325 extent_logical = logical;
a2de733c 3326 }
625f1c8d 3327 if (extent_logical + extent_len >
a2de733c 3328 logical + map->stripe_len) {
625f1c8d
LB		 3329 extent_len = logical + map->stripe_len -
3330 extent_logical;
a2de733c
AJ		 3331 }
3332
625f1c8d 3333 extent_physical = extent_logical - logical + physical;
ff023aac
SB		 3334 extent_dev = scrub_dev;
3335 extent_mirror_num = mirror_num;
3336 if (is_dev_replace)
3337 scrub_remap_extent(fs_info, extent_logical,
3338 extent_len, &extent_physical,
3339 &extent_dev,
3340 &extent_mirror_num);
625f1c8d 3341
fe8cf654
ZL		 3342 ret = btrfs_lookup_csums_range(csum_root,
3343 extent_logical,
3344 extent_logical +
3345 extent_len - 1,
3346 &sctx->csum_list, 1);
625f1c8d
LB		 3347 if (ret)
3348 goto out;
3349
ff023aac
SB		 3350 ret = scrub_extent(sctx, extent_logical, extent_len,
3351 extent_physical, extent_dev, flags,
3352 generation, extent_mirror_num,
115930cb 3353 extent_logical - logical + physical);
6fa96d72
ZL		 3354
3355 scrub_free_csums(sctx);
3356
a2de733c
AJ		 3357 if (ret)
3358 goto out;
3359
625f1c8d
LB		 3360 if (extent_logical + extent_len <
3361 key.objectid + bytes) {
ffe2d203 3362 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3b080b25
WS		 3363 /*
3364 * loop until we find next data stripe
3365 * or we have finished all stripes.
3366 */
5a6ac9ea
MX		 3367loop:
3368 physical += map->stripe_len;
3369 ret = get_raid56_logic_offset(physical,
3370 num, map, &logical,
3371 &stripe_logical);
3372 logical += base;
3373
3374 if (ret && physical < physical_end) {
3375 stripe_logical += base;
3376 stripe_end = stripe_logical +
a0dd59de 3377 increment;
5a6ac9ea
MX		 3378 ret = scrub_raid56_parity(sctx,
3379 map, scrub_dev, ppath,
3380 stripe_logical,
3381 stripe_end);
3382 if (ret)
3383 goto out;
3384 goto loop;
3385 }
3b080b25
WS		 3386 } else {
3387 physical += map->stripe_len;
3388 logical += increment;
3389 }
625f1c8d
LB		 3390 if (logical < key.objectid + bytes) {
3391 cond_resched();
3392 goto again;
3393 }
3394
3b080b25 3395 if (physical >= physical_end) {
625f1c8d
LB		 3396 stop_loop = 1;
3397 break;
3398 }
3399 }
a2de733c
AJ		 3400next:
3401 path->slots[0]++;
3402 }
71267333 3403 btrfs_release_path(path);
3b080b25 3404skip:
a2de733c
AJ		 3405 logical += increment;
3406 physical += map->stripe_len;
d9d181c1 3407 spin_lock(&sctx->stat_lock);
625f1c8d
LB		 3408 if (stop_loop)
3409 sctx->stat.last_physical = map->stripes[num].physical +
3410 length;
3411 else
3412 sctx->stat.last_physical = physical;
d9d181c1 3413 spin_unlock(&sctx->stat_lock);
625f1c8d
LB		 3414 if (stop_loop)
3415 break;
a2de733c 3416 }
ff023aac 3417out:
a2de733c 3418 /* push queued extents */
d9d181c1 3419 scrub_submit(sctx);
ff023aac
SB		 3420 mutex_lock(&sctx->wr_ctx.wr_lock);
3421 scrub_wr_submit(sctx);
3422 mutex_unlock(&sctx->wr_ctx.wr_lock);
a2de733c 3423
e7786c3a 3424 blk_finish_plug(&plug);
a2de733c 3425 btrfs_free_path(path);
5a6ac9ea 3426 btrfs_free_path(ppath);
a2de733c
AJ		 3427 return ret < 0 ? ret : 0;
3428}
3429
d9d181c1 3430static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx,
a36cf8b8 3431 struct btrfs_device *scrub_dev,
a36cf8b8 3432 u64 chunk_offset, u64 length,
020d5b73
FM		 3433 u64 dev_offset,
3434 struct btrfs_block_group_cache *cache,
3435 int is_dev_replace)
a2de733c
AJ		 3436{
3437 struct btrfs_mapping_tree *map_tree =
a36cf8b8 3438 &sctx->dev_root->fs_info->mapping_tree;
a2de733c
AJ		 3439 struct map_lookup *map;
3440 struct extent_map *em;
3441 int i;
ff023aac 3442 int ret = 0;
a2de733c
AJ		 3443
3444 read_lock(&map_tree->map_tree.lock);
3445 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3446 read_unlock(&map_tree->map_tree.lock);
3447
020d5b73
FM		 3448 if (!em) {
3449 /*
3450 * Might have been an unused block group deleted by the cleaner
3451 * kthread or relocation.
3452 */
3453 spin_lock(&cache->lock);
3454 if (!cache->removed)
3455 ret = -EINVAL;
3456 spin_unlock(&cache->lock);
3457
3458 return ret;
3459 }
a2de733c 3460
95617d69 3461 map = em->map_lookup;
a2de733c
AJ		 3462 if (em->start != chunk_offset)
3463 goto out;
3464
3465 if (em->len < length)
3466 goto out;
3467
3468 for (i = 0; i < map->num_stripes; ++i) {
a36cf8b8 3469 if (map->stripes[i].dev->bdev == scrub_dev->bdev &&
859acaf1 3470 map->stripes[i].physical == dev_offset) {
a36cf8b8 3471 ret = scrub_stripe(sctx, map, scrub_dev, i,
ff023aac
SB		 3472 chunk_offset, length,
3473 is_dev_replace);
a2de733c
AJ		 3474 if (ret)
3475 goto out;
3476 }
3477 }
3478out:
3479 free_extent_map(em);
3480
3481 return ret;
3482}
3483
3484static noinline_for_stack
a36cf8b8 3485int scrub_enumerate_chunks(struct scrub_ctx *sctx,
ff023aac
SB		 3486 struct btrfs_device *scrub_dev, u64 start, u64 end,
3487 int is_dev_replace)
a2de733c
AJ		 3488{
3489 struct btrfs_dev_extent *dev_extent = NULL;
3490 struct btrfs_path *path;
a36cf8b8 3491 struct btrfs_root *root = sctx->dev_root;
a2de733c
AJ		 3492 struct btrfs_fs_info *fs_info = root->fs_info;
3493 u64 length;
a2de733c 3494 u64 chunk_offset;
55e3a601 3495 int ret = 0;
76a8efa1 3496 int ro_set;
a2de733c
AJ		 3497 int slot;
3498 struct extent_buffer *l;
3499 struct btrfs_key key;
3500 struct btrfs_key found_key;
3501 struct btrfs_block_group_cache *cache;
ff023aac 3502 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
a2de733c
AJ		 3503
3504 path = btrfs_alloc_path();
3505 if (!path)
3506 return -ENOMEM;
3507
e4058b54 3508 path->reada = READA_FORWARD;
a2de733c
AJ		 3509 path->search_commit_root = 1;
3510 path->skip_locking = 1;
3511
a36cf8b8 3512 key.objectid = scrub_dev->devid;
a2de733c
AJ		 3513 key.offset = 0ull;
3514 key.type = BTRFS_DEV_EXTENT_KEY;
3515
a2de733c
AJ		 3516 while (1) {
3517 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3518 if (ret < 0)
8c51032f
AJ		 3519 break;
3520 if (ret > 0) {
3521 if (path->slots[0] >=
3522 btrfs_header_nritems(path->nodes[0])) {
3523 ret = btrfs_next_leaf(root, path);
55e3a601
Z		 3524 if (ret < 0)
3525 break;
3526 if (ret > 0) {
3527 ret = 0;
8c51032f 3528 break;
55e3a601
Z		 3529 }
3530 } else {
3531 ret = 0;
8c51032f
AJ		 3532 }
3533 }
a2de733c
AJ		 3534
3535 l = path->nodes[0];
3536 slot = path->slots[0];
3537
3538 btrfs_item_key_to_cpu(l, &found_key, slot);
3539
a36cf8b8 3540 if (found_key.objectid != scrub_dev->devid)
a2de733c
AJ		 3541 break;
3542
962a298f 3543 if (found_key.type != BTRFS_DEV_EXTENT_KEY)
a2de733c
AJ		 3544 break;
3545
3546 if (found_key.offset >= end)
3547 break;
3548
3549 if (found_key.offset < key.offset)
3550 break;
3551
3552 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3553 length = btrfs_dev_extent_length(l, dev_extent);
3554
ced96edc
QW		 3555 if (found_key.offset + length <= start)
3556 goto skip;
a2de733c 3557
a2de733c
AJ		 3558 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3559
3560 /*
3561 * get a reference on the corresponding block group to prevent
3562 * the chunk from going away while we scrub it
3563 */
3564 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
ced96edc
QW		 3565
3566 /* some chunks are removed but not committed to disk yet,
3567 * continue scrubbing */
3568 if (!cache)
3569 goto skip;
3570
55e3a601
Z		 3571 /*
3572 * we need call btrfs_inc_block_group_ro() with scrubs_paused,
3573 * to avoid deadlock caused by:
3574 * btrfs_inc_block_group_ro()
3575 * -> btrfs_wait_for_commit()
3576 * -> btrfs_commit_transaction()
3577 * -> btrfs_scrub_pause()
3578 */
3579 scrub_pause_on(fs_info);
3580 ret = btrfs_inc_block_group_ro(root, cache);
3581 scrub_pause_off(fs_info);
76a8efa1
Z		 3582
3583 if (ret == 0) {
3584 ro_set = 1;
3585 } else if (ret == -ENOSPC) {
3586 /*
3587 * btrfs_inc_block_group_ro return -ENOSPC when it
3588 * failed in creating new chunk for metadata.
3589 * It is not a problem for scrub/replace, because
3590 * metadata are always cowed, and our scrub paused
3591 * commit_transactions.
3592 */
3593 ro_set = 0;
3594 } else {
3595 btrfs_warn(fs_info, "failed setting block group ro, ret=%d\n",
3596 ret);
55e3a601
Z		 3597 btrfs_put_block_group(cache);
3598 break;
3599 }
3600
ff023aac
SB		 3601 dev_replace->cursor_right = found_key.offset + length;
3602 dev_replace->cursor_left = found_key.offset;
3603 dev_replace->item_needs_writeback = 1;
8c204c96 3604 ret = scrub_chunk(sctx, scrub_dev, chunk_offset, length,
020d5b73 3605 found_key.offset, cache, is_dev_replace);
ff023aac
SB		 3606
3607 /*
3608 * flush, submit all pending read and write bios, afterwards
3609 * wait for them.
3610 * Note that in the dev replace case, a read request causes
3611 * write requests that are submitted in the read completion
3612 * worker. Therefore in the current situation, it is required
3613 * that all write requests are flushed, so that all read and
3614 * write requests are really completed when bios_in_flight
3615 * changes to 0.
3616 */
3617 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
3618 scrub_submit(sctx);
3619 mutex_lock(&sctx->wr_ctx.wr_lock);
3620 scrub_wr_submit(sctx);
3621 mutex_unlock(&sctx->wr_ctx.wr_lock);
3622
3623 wait_event(sctx->list_wait,
3624 atomic_read(&sctx->bios_in_flight) == 0);
b708ce96
Z		 3625
3626 scrub_pause_on(fs_info);
12cf9372
WS		 3627
3628 /*
3629 * must be called before we decrease @scrub_paused.
3630 * make sure we don't block transaction commit while
3631 * we are waiting pending workers finished.
3632 */
ff023aac
SB		 3633 wait_event(sctx->list_wait,
3634 atomic_read(&sctx->workers_pending) == 0);
12cf9372
WS		 3635 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
3636
b708ce96 3637 scrub_pause_off(fs_info);
ff023aac 3638
76a8efa1
Z		 3639 if (ro_set)
3640 btrfs_dec_block_group_ro(root, cache);
ff023aac 3641
758f2dfc
FM		 3642 /*
3643 * We might have prevented the cleaner kthread from deleting
3644 * this block group if it was already unused because we raced
3645 * and set it to RO mode first. So add it back to the unused
3646 * list, otherwise it might not ever be deleted unless a manual
3647 * balance is triggered or it becomes used and unused again.
3648 */
3649 spin_lock(&cache->lock);
3650 if (!cache->removed && !cache->ro && cache->reserved == 0 &&
3651 btrfs_block_group_used(&cache->item) == 0) {
3652 spin_unlock(&cache->lock);
3653 spin_lock(&fs_info->unused_bgs_lock);
3654 if (list_empty(&cache->bg_list)) {
3655 btrfs_get_block_group(cache);
3656 list_add_tail(&cache->bg_list,
3657 &fs_info->unused_bgs);
3658 }
3659 spin_unlock(&fs_info->unused_bgs_lock);
3660 } else {
3661 spin_unlock(&cache->lock);
3662 }
3663
a2de733c
AJ		 3664 btrfs_put_block_group(cache);
3665 if (ret)
3666 break;
af1be4f8
SB		 3667 if (is_dev_replace &&
3668 atomic64_read(&dev_replace->num_write_errors) > 0) {
ff023aac
SB		 3669 ret = -EIO;
3670 break;
3671 }
3672 if (sctx->stat.malloc_errors > 0) {
3673 ret = -ENOMEM;
3674 break;
3675 }
a2de733c 3676
539f358a
ID		 3677 dev_replace->cursor_left = dev_replace->cursor_right;
3678 dev_replace->item_needs_writeback = 1;
ced96edc 3679skip:
a2de733c 3680 key.offset = found_key.offset + length;
71267333 3681 btrfs_release_path(path);
a2de733c
AJ		 3682 }
3683
a2de733c 3684 btrfs_free_path(path);
8c51032f 3685
55e3a601 3686 return ret;
a2de733c
AJ		 3687}
3688
a36cf8b8
SB		 3689static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx,
3690 struct btrfs_device *scrub_dev)
a2de733c
AJ		 3691{
3692 int i;
3693 u64 bytenr;
3694 u64 gen;
3695 int ret;
a36cf8b8 3696 struct btrfs_root *root = sctx->dev_root;
a2de733c 3697
87533c47 3698 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
79787eaa
JM		 3699 return -EIO;
3700
5f546063
MX		 3701 /* Seed devices of a new filesystem has their own generation. */
3702 if (scrub_dev->fs_devices != root->fs_info->fs_devices)
3703 gen = scrub_dev->generation;
3704 else
3705 gen = root->fs_info->last_trans_committed;
a2de733c
AJ		 3706
3707 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
3708 bytenr = btrfs_sb_offset(i);
935e5cc9
MX		 3709 if (bytenr + BTRFS_SUPER_INFO_SIZE >
3710 scrub_dev->commit_total_bytes)
a2de733c
AJ		 3711 break;
3712
d9d181c1 3713 ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr,
a36cf8b8 3714 scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i,
ff023aac 3715 NULL, 1, bytenr);
a2de733c
AJ		 3716 if (ret)
3717 return ret;
3718 }
b6bfebc1 3719 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
a2de733c
AJ		 3720
3721 return 0;
3722}
3723
3724/*
3725 * get a reference count on fs_info->scrub_workers. start worker if necessary
3726 */
ff023aac
SB		 3727static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info,
3728 int is_dev_replace)
a2de733c 3729{
6f011058 3730 unsigned int flags = WQ_FREEZABLE | WQ_UNBOUND;
0339ef2f 3731 int max_active = fs_info->thread_pool_size;
a2de733c 3732
632dd772 3733 if (fs_info->scrub_workers_refcnt == 0) {
ff023aac 3734 if (is_dev_replace)
0339ef2f 3735 fs_info->scrub_workers =
0de270fa 3736 btrfs_alloc_workqueue("scrub", flags,
0339ef2f 3737 1, 4);
ff023aac 3738 else
0339ef2f 3739 fs_info->scrub_workers =
0de270fa 3740 btrfs_alloc_workqueue("scrub", flags,
0339ef2f 3741 max_active, 4);
e82afc52
ZL		 3742 if (!fs_info->scrub_workers)
3743 goto fail_scrub_workers;
3744
0339ef2f 3745 fs_info->scrub_wr_completion_workers =
0de270fa 3746 btrfs_alloc_workqueue("scrubwrc", flags,
0339ef2f 3747 max_active, 2);
e82afc52
ZL		 3748 if (!fs_info->scrub_wr_completion_workers)
3749 goto fail_scrub_wr_completion_workers;
3750
0339ef2f 3751 fs_info->scrub_nocow_workers =
0de270fa 3752 btrfs_alloc_workqueue("scrubnc", flags, 1, 0);
e82afc52
ZL		 3753 if (!fs_info->scrub_nocow_workers)
3754 goto fail_scrub_nocow_workers;
20b2e302 3755 fs_info->scrub_parity_workers =
0de270fa 3756 btrfs_alloc_workqueue("scrubparity", flags,
20b2e302 3757 max_active, 2);
e82afc52
ZL		 3758 if (!fs_info->scrub_parity_workers)
3759 goto fail_scrub_parity_workers;
632dd772 3760 }
a2de733c 3761 ++fs_info->scrub_workers_refcnt;
e82afc52
ZL		 3762 return 0;
3763
3764fail_scrub_parity_workers:
3765 btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
3766fail_scrub_nocow_workers:
3767 btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
3768fail_scrub_wr_completion_workers:
3769 btrfs_destroy_workqueue(fs_info->scrub_workers);
3770fail_scrub_workers:
3771 return -ENOMEM;
a2de733c
AJ		 3772}
3773
aa1b8cd4 3774static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info)
a2de733c 3775{
ff023aac 3776 if (--fs_info->scrub_workers_refcnt == 0) {
0339ef2f
QW		 3777 btrfs_destroy_workqueue(fs_info->scrub_workers);
3778 btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
3779 btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
20b2e302 3780 btrfs_destroy_workqueue(fs_info->scrub_parity_workers);
ff023aac 3781 }
a2de733c 3782 WARN_ON(fs_info->scrub_workers_refcnt < 0);
a2de733c
AJ		 3783}
3784
aa1b8cd4
SB		 3785int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start,
3786 u64 end, struct btrfs_scrub_progress *progress,
63a212ab 3787 int readonly, int is_dev_replace)
a2de733c 3788{
d9d181c1 3789 struct scrub_ctx *sctx;
a2de733c
AJ		 3790 int ret;
3791 struct btrfs_device *dev;
5d68da3b 3792 struct rcu_string *name;
a2de733c 3793
aa1b8cd4 3794 if (btrfs_fs_closing(fs_info))
a2de733c
AJ		 3795 return -EINVAL;
3796
aa1b8cd4 3797 if (fs_info->chunk_root->nodesize > BTRFS_STRIPE_LEN) {
b5d67f64
SB		 3798 /*
3799 * in this case scrub is unable to calculate the checksum
3800 * the way scrub is implemented. Do not handle this
3801 * situation at all because it won't ever happen.
3802 */
efe120a0
FH		 3803 btrfs_err(fs_info,
3804 "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails",
aa1b8cd4 3805 fs_info->chunk_root->nodesize, BTRFS_STRIPE_LEN);
b5d67f64
SB		 3806 return -EINVAL;
3807 }
3808
aa1b8cd4 3809 if (fs_info->chunk_root->sectorsize != PAGE_SIZE) {
b5d67f64 3810 /* not supported for data w/o checksums */
efe120a0
FH		 3811 btrfs_err(fs_info,
3812 "scrub: size assumption sectorsize != PAGE_SIZE "
3813 "(%d != %lu) fails",
27f9f023 3814 fs_info->chunk_root->sectorsize, PAGE_SIZE);
a2de733c
AJ		 3815 return -EINVAL;
3816 }
3817
7a9e9987
SB		 3818 if (fs_info->chunk_root->nodesize >
3819 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK ||
3820 fs_info->chunk_root->sectorsize >
3821 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) {
3822 /*
3823 * would exhaust the array bounds of pagev member in
3824 * struct scrub_block
3825 */
efe120a0
FH		 3826 btrfs_err(fs_info, "scrub: size assumption nodesize and sectorsize "
3827 "<= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails",
7a9e9987
SB		 3828 fs_info->chunk_root->nodesize,
3829 SCRUB_MAX_PAGES_PER_BLOCK,
3830 fs_info->chunk_root->sectorsize,
3831 SCRUB_MAX_PAGES_PER_BLOCK);
3832 return -EINVAL;
3833 }
3834
a2de733c 3835
aa1b8cd4
SB		 3836 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3837 dev = btrfs_find_device(fs_info, devid, NULL, NULL);
63a212ab 3838 if (!dev || (dev->missing && !is_dev_replace)) {
aa1b8cd4 3839 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
a2de733c
AJ		 3840 return -ENODEV;
3841 }
a2de733c 3842
5d68da3b
MX		 3843 if (!is_dev_replace && !readonly && !dev->writeable) {
3844 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3845 rcu_read_lock();
3846 name = rcu_dereference(dev->name);
3847 btrfs_err(fs_info, "scrub: device %s is not writable",
3848 name->str);
3849 rcu_read_unlock();
3850 return -EROFS;
3851 }
3852
3b7a016f 3853 mutex_lock(&fs_info->scrub_lock);
63a212ab 3854 if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) {
a2de733c 3855 mutex_unlock(&fs_info->scrub_lock);
aa1b8cd4 3856 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
aa1b8cd4 3857 return -EIO;
a2de733c
AJ		 3858 }
3859
73beece9 3860 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
8dabb742
SB		 3861 if (dev->scrub_device ||
3862 (!is_dev_replace &&
3863 btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) {
73beece9 3864 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
a2de733c 3865 mutex_unlock(&fs_info->scrub_lock);
aa1b8cd4 3866 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
a2de733c
AJ		 3867 return -EINPROGRESS;
3868 }
73beece9 3869 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3b7a016f
WS		 3870
3871 ret = scrub_workers_get(fs_info, is_dev_replace);
3872 if (ret) {
3873 mutex_unlock(&fs_info->scrub_lock);
3874 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3875 return ret;
3876 }
3877
63a212ab 3878 sctx = scrub_setup_ctx(dev, is_dev_replace);
d9d181c1 3879 if (IS_ERR(sctx)) {
a2de733c 3880 mutex_unlock(&fs_info->scrub_lock);
aa1b8cd4
SB		 3881 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3882 scrub_workers_put(fs_info);
d9d181c1 3883 return PTR_ERR(sctx);
a2de733c 3884 }
d9d181c1
SB		 3885 sctx->readonly = readonly;
3886 dev->scrub_device = sctx;
3cb0929a 3887 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
a2de733c 3888
3cb0929a
WS		 3889 /*
3890 * checking @scrub_pause_req here, we can avoid
3891 * race between committing transaction and scrubbing.
3892 */
cb7ab021 3893 __scrub_blocked_if_needed(fs_info);
a2de733c
AJ		 3894 atomic_inc(&fs_info->scrubs_running);
3895 mutex_unlock(&fs_info->scrub_lock);
a2de733c 3896
ff023aac 3897 if (!is_dev_replace) {
9b011adf
WS		 3898 /*
3899 * by holding device list mutex, we can
3900 * kick off writing super in log tree sync.
3901 */
3cb0929a 3902 mutex_lock(&fs_info->fs_devices->device_list_mutex);
ff023aac 3903 ret = scrub_supers(sctx, dev);
3cb0929a 3904 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
ff023aac 3905 }
a2de733c
AJ		 3906
3907 if (!ret)
ff023aac
SB		 3908 ret = scrub_enumerate_chunks(sctx, dev, start, end,
3909 is_dev_replace);
a2de733c 3910
b6bfebc1 3911 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
a2de733c
AJ		 3912 atomic_dec(&fs_info->scrubs_running);
3913 wake_up(&fs_info->scrub_pause_wait);
3914
b6bfebc1 3915 wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0);
0ef8e451 3916
a2de733c 3917 if (progress)
d9d181c1 3918 memcpy(progress, &sctx->stat, sizeof(*progress));
a2de733c
AJ		 3919
3920 mutex_lock(&fs_info->scrub_lock);
3921 dev->scrub_device = NULL;
3b7a016f 3922 scrub_workers_put(fs_info);
a2de733c
AJ		 3923 mutex_unlock(&fs_info->scrub_lock);
3924
f55985f4 3925 scrub_put_ctx(sctx);
a2de733c
AJ		 3926
3927 return ret;
3928}
3929
143bede5 3930void btrfs_scrub_pause(struct btrfs_root *root)
a2de733c
AJ		 3931{
3932 struct btrfs_fs_info *fs_info = root->fs_info;
3933
3934 mutex_lock(&fs_info->scrub_lock);
3935 atomic_inc(&fs_info->scrub_pause_req);
3936 while (atomic_read(&fs_info->scrubs_paused) !=
3937 atomic_read(&fs_info->scrubs_running)) {
3938 mutex_unlock(&fs_info->scrub_lock);
3939 wait_event(fs_info->scrub_pause_wait,
3940 atomic_read(&fs_info->scrubs_paused) ==
3941 atomic_read(&fs_info->scrubs_running));
3942 mutex_lock(&fs_info->scrub_lock);
3943 }
3944 mutex_unlock(&fs_info->scrub_lock);
a2de733c
AJ		 3945}
3946
143bede5 3947void btrfs_scrub_continue(struct btrfs_root *root)
a2de733c
AJ		 3948{
3949 struct btrfs_fs_info *fs_info = root->fs_info;
3950
3951 atomic_dec(&fs_info->scrub_pause_req);
3952 wake_up(&fs_info->scrub_pause_wait);
a2de733c
AJ		 3953}
3954
aa1b8cd4 3955int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info)
a2de733c 3956{
a2de733c
AJ		 3957 mutex_lock(&fs_info->scrub_lock);
3958 if (!atomic_read(&fs_info->scrubs_running)) {
3959 mutex_unlock(&fs_info->scrub_lock);
3960 return -ENOTCONN;
3961 }
3962
3963 atomic_inc(&fs_info->scrub_cancel_req);
3964 while (atomic_read(&fs_info->scrubs_running)) {
3965 mutex_unlock(&fs_info->scrub_lock);
3966 wait_event(fs_info->scrub_pause_wait,
3967 atomic_read(&fs_info->scrubs_running) == 0);
3968 mutex_lock(&fs_info->scrub_lock);
3969 }
3970 atomic_dec(&fs_info->scrub_cancel_req);
3971 mutex_unlock(&fs_info->scrub_lock);
3972
3973 return 0;
3974}
3975
aa1b8cd4
SB		 3976int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info,
3977 struct btrfs_device *dev)
49b25e05 3978{
d9d181c1 3979 struct scrub_ctx *sctx;
a2de733c
AJ		 3980
3981 mutex_lock(&fs_info->scrub_lock);
d9d181c1
SB		 3982 sctx = dev->scrub_device;
3983 if (!sctx) {
a2de733c
AJ		 3984 mutex_unlock(&fs_info->scrub_lock);
3985 return -ENOTCONN;
3986 }
d9d181c1 3987 atomic_inc(&sctx->cancel_req);
a2de733c
AJ		 3988 while (dev->scrub_device) {
3989 mutex_unlock(&fs_info->scrub_lock);
3990 wait_event(fs_info->scrub_pause_wait,
3991 dev->scrub_device == NULL);
3992 mutex_lock(&fs_info->scrub_lock);
3993 }
3994 mutex_unlock(&fs_info->scrub_lock);
3995
3996 return 0;
3997}
1623edeb 3998
a2de733c
AJ		 3999int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
4000 struct btrfs_scrub_progress *progress)
4001{
4002 struct btrfs_device *dev;
d9d181c1 4003 struct scrub_ctx *sctx = NULL;
a2de733c
AJ		 4004
4005 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
aa1b8cd4 4006 dev = btrfs_find_device(root->fs_info, devid, NULL, NULL);
a2de733c 4007 if (dev)
d9d181c1
SB		 4008 sctx = dev->scrub_device;
4009 if (sctx)
4010 memcpy(progress, &sctx->stat, sizeof(*progress));
a2de733c
AJ		 4011 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
4012
d9d181c1 4013 return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV;
a2de733c 4014}
ff023aac
SB		 4015
4016static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
4017 u64 extent_logical, u64 extent_len,
4018 u64 *extent_physical,
4019 struct btrfs_device **extent_dev,
4020 int *extent_mirror_num)
4021{
4022 u64 mapped_length;
4023 struct btrfs_bio *bbio = NULL;
4024 int ret;
4025
4026 mapped_length = extent_len;
4027 ret = btrfs_map_block(fs_info, READ, extent_logical,
4028 &mapped_length, &bbio, 0);
4029 if (ret || !bbio || mapped_length < extent_len ||
4030 !bbio->stripes[0].dev->bdev) {
6e9606d2 4031 btrfs_put_bbio(bbio);
ff023aac
SB		 4032 return;
4033 }
4034
4035 *extent_physical = bbio->stripes[0].physical;
4036 *extent_mirror_num = bbio->mirror_num;
4037 *extent_dev = bbio->stripes[0].dev;
6e9606d2 4038 btrfs_put_bbio(bbio);
ff023aac
SB		 4039}
4040
4041static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
4042 struct scrub_wr_ctx *wr_ctx,
4043 struct btrfs_fs_info *fs_info,
4044 struct btrfs_device *dev,
4045 int is_dev_replace)
4046{
4047 WARN_ON(wr_ctx->wr_curr_bio != NULL);
4048
4049 mutex_init(&wr_ctx->wr_lock);
4050 wr_ctx->wr_curr_bio = NULL;
4051 if (!is_dev_replace)
4052 return 0;
4053
4054 WARN_ON(!dev->bdev);
b54ffb73 4055 wr_ctx->pages_per_wr_bio = SCRUB_PAGES_PER_WR_BIO;
ff023aac
SB		 4056 wr_ctx->tgtdev = dev;
4057 atomic_set(&wr_ctx->flush_all_writes, 0);
4058 return 0;
4059}
4060
4061static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx)
4062{
4063 mutex_lock(&wr_ctx->wr_lock);
4064 kfree(wr_ctx->wr_curr_bio);
4065 wr_ctx->wr_curr_bio = NULL;
4066 mutex_unlock(&wr_ctx->wr_lock);
4067}
4068
4069static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
4070 int mirror_num, u64 physical_for_dev_replace)
4071{
4072 struct scrub_copy_nocow_ctx *nocow_ctx;
4073 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
4074
4075 nocow_ctx = kzalloc(sizeof(*nocow_ctx), GFP_NOFS);
4076 if (!nocow_ctx) {
4077 spin_lock(&sctx->stat_lock);
4078 sctx->stat.malloc_errors++;
4079 spin_unlock(&sctx->stat_lock);
4080 return -ENOMEM;
4081 }
4082
4083 scrub_pending_trans_workers_inc(sctx);
4084
4085 nocow_ctx->sctx = sctx;
4086 nocow_ctx->logical = logical;
4087 nocow_ctx->len = len;
4088 nocow_ctx->mirror_num = mirror_num;
4089 nocow_ctx->physical_for_dev_replace = physical_for_dev_replace;
9e0af237
LB		 4090 btrfs_init_work(&nocow_ctx->work, btrfs_scrubnc_helper,
4091 copy_nocow_pages_worker, NULL, NULL);
652f25a2 4092 INIT_LIST_HEAD(&nocow_ctx->inodes);
0339ef2f
QW		 4093 btrfs_queue_work(fs_info->scrub_nocow_workers,
4094 &nocow_ctx->work);
ff023aac
SB		 4095
4096 return 0;
4097}
4098
652f25a2
JB		 4099static int record_inode_for_nocow(u64 inum, u64 offset, u64 root, void *ctx)
4100{
4101 struct scrub_copy_nocow_ctx *nocow_ctx = ctx;
4102 struct scrub_nocow_inode *nocow_inode;
4103
4104 nocow_inode = kzalloc(sizeof(*nocow_inode), GFP_NOFS);
4105 if (!nocow_inode)
4106 return -ENOMEM;
4107 nocow_inode->inum = inum;
4108 nocow_inode->offset = offset;
4109 nocow_inode->root = root;
4110 list_add_tail(&nocow_inode->list, &nocow_ctx->inodes);
4111 return 0;
4112}
4113
4114#define COPY_COMPLETE 1
4115
ff023aac
SB		 4116static void copy_nocow_pages_worker(struct btrfs_work *work)
4117{
4118 struct scrub_copy_nocow_ctx *nocow_ctx =
4119 container_of(work, struct scrub_copy_nocow_ctx, work);
4120 struct scrub_ctx *sctx = nocow_ctx->sctx;
4121 u64 logical = nocow_ctx->logical;
4122 u64 len = nocow_ctx->len;
4123 int mirror_num = nocow_ctx->mirror_num;
4124 u64 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
4125 int ret;
4126 struct btrfs_trans_handle *trans = NULL;
4127 struct btrfs_fs_info *fs_info;
4128 struct btrfs_path *path;
4129 struct btrfs_root *root;
4130 int not_written = 0;
4131
4132 fs_info = sctx->dev_root->fs_info;
4133 root = fs_info->extent_root;
4134
4135 path = btrfs_alloc_path();
4136 if (!path) {
4137 spin_lock(&sctx->stat_lock);
4138 sctx->stat.malloc_errors++;
4139 spin_unlock(&sctx->stat_lock);
4140 not_written = 1;
4141 goto out;
4142 }
4143
4144 trans = btrfs_join_transaction(root);
4145 if (IS_ERR(trans)) {
4146 not_written = 1;
4147 goto out;
4148 }
4149
4150 ret = iterate_inodes_from_logical(logical, fs_info, path,
652f25a2 4151 record_inode_for_nocow, nocow_ctx);
ff023aac 4152 if (ret != 0 && ret != -ENOENT) {
efe120a0
FH		 4153 btrfs_warn(fs_info, "iterate_inodes_from_logical() failed: log %llu, "
4154 "phys %llu, len %llu, mir %u, ret %d",
118a0a25
GU		 4155 logical, physical_for_dev_replace, len, mirror_num,
4156 ret);
ff023aac
SB		 4157 not_written = 1;
4158 goto out;
4159 }
4160
652f25a2
JB		 4161 btrfs_end_transaction(trans, root);
4162 trans = NULL;
4163 while (!list_empty(&nocow_ctx->inodes)) {
4164 struct scrub_nocow_inode *entry;
4165 entry = list_first_entry(&nocow_ctx->inodes,
4166 struct scrub_nocow_inode,
4167 list);
4168 list_del_init(&entry->list);
4169 ret = copy_nocow_pages_for_inode(entry->inum, entry->offset,
4170 entry->root, nocow_ctx);
4171 kfree(entry);
4172 if (ret == COPY_COMPLETE) {
4173 ret = 0;
4174 break;
4175 } else if (ret) {
4176 break;
4177 }
4178 }
ff023aac 4179out:
652f25a2
JB		 4180 while (!list_empty(&nocow_ctx->inodes)) {
4181 struct scrub_nocow_inode *entry;
4182 entry = list_first_entry(&nocow_ctx->inodes,
4183 struct scrub_nocow_inode,
4184 list);
4185 list_del_init(&entry->list);
4186 kfree(entry);
4187 }
ff023aac
SB		 4188 if (trans && !IS_ERR(trans))
4189 btrfs_end_transaction(trans, root);
4190 if (not_written)
4191 btrfs_dev_replace_stats_inc(&fs_info->dev_replace.
4192 num_uncorrectable_read_errors);
4193
4194 btrfs_free_path(path);
4195 kfree(nocow_ctx);
4196
4197 scrub_pending_trans_workers_dec(sctx);
4198}
4199
32159242
GH		 4200static int check_extent_to_block(struct inode *inode, u64 start, u64 len,
4201 u64 logical)
4202{
4203 struct extent_state *cached_state = NULL;
4204 struct btrfs_ordered_extent *ordered;
4205 struct extent_io_tree *io_tree;
4206 struct extent_map *em;
4207 u64 lockstart = start, lockend = start + len - 1;
4208 int ret = 0;
4209
4210 io_tree = &BTRFS_I(inode)->io_tree;
4211
ff13db41 4212 lock_extent_bits(io_tree, lockstart, lockend, &cached_state);
32159242
GH		 4213 ordered = btrfs_lookup_ordered_range(inode, lockstart, len);
4214 if (ordered) {
4215 btrfs_put_ordered_extent(ordered);
4216 ret = 1;
4217 goto out_unlock;
4218 }
4219
4220 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
4221 if (IS_ERR(em)) {
4222 ret = PTR_ERR(em);
4223 goto out_unlock;
4224 }
4225
4226 /*
4227 * This extent does not actually cover the logical extent anymore,
4228 * move on to the next inode.
4229 */
4230 if (em->block_start > logical ||
4231 em->block_start + em->block_len < logical + len) {
4232 free_extent_map(em);
4233 ret = 1;
4234 goto out_unlock;
4235 }
4236 free_extent_map(em);
4237
4238out_unlock:
4239 unlock_extent_cached(io_tree, lockstart, lockend, &cached_state,
4240 GFP_NOFS);
4241 return ret;
4242}
4243
652f25a2
JB		 4244static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
4245 struct scrub_copy_nocow_ctx *nocow_ctx)
ff023aac 4246{
826aa0a8 4247 struct btrfs_fs_info *fs_info = nocow_ctx->sctx->dev_root->fs_info;
ff023aac 4248 struct btrfs_key key;
826aa0a8
MX		 4249 struct inode *inode;
4250 struct page *page;
ff023aac 4251 struct btrfs_root *local_root;
652f25a2 4252 struct extent_io_tree *io_tree;
ff023aac 4253 u64 physical_for_dev_replace;
32159242 4254 u64 nocow_ctx_logical;
652f25a2 4255 u64 len = nocow_ctx->len;
826aa0a8 4256 unsigned long index;
6f1c3605 4257 int srcu_index;
652f25a2
JB		 4258 int ret = 0;
4259 int err = 0;
ff023aac
SB		 4260
4261 key.objectid = root;
4262 key.type = BTRFS_ROOT_ITEM_KEY;
4263 key.offset = (u64)-1;
6f1c3605
LB		 4264
4265 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
4266
ff023aac 4267 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
6f1c3605
LB		 4268 if (IS_ERR(local_root)) {
4269 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
ff023aac 4270 return PTR_ERR(local_root);
6f1c3605 4271 }
ff023aac
SB		 4272
4273 key.type = BTRFS_INODE_ITEM_KEY;
4274 key.objectid = inum;
4275 key.offset = 0;
4276 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
6f1c3605 4277 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
ff023aac
SB		 4278 if (IS_ERR(inode))
4279 return PTR_ERR(inode);
4280
edd1400b 4281 /* Avoid truncate/dio/punch hole.. */
5955102c 4282 inode_lock(inode);
edd1400b
MX		 4283 inode_dio_wait(inode);
4284
ff023aac 4285 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
652f25a2 4286 io_tree = &BTRFS_I(inode)->io_tree;
32159242 4287 nocow_ctx_logical = nocow_ctx->logical;
652f25a2 4288
32159242
GH		 4289 ret = check_extent_to_block(inode, offset, len, nocow_ctx_logical);
4290 if (ret) {
4291 ret = ret > 0 ? 0 : ret;
4292 goto out;
652f25a2 4293 }
652f25a2 4294
ff023aac 4295 while (len >= PAGE_CACHE_SIZE) {
ff023aac 4296 index = offset >> PAGE_CACHE_SHIFT;
edd1400b 4297again:
ff023aac
SB		 4298 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
4299 if (!page) {
efe120a0 4300 btrfs_err(fs_info, "find_or_create_page() failed");
ff023aac 4301 ret = -ENOMEM;
826aa0a8 4302 goto out;
ff023aac
SB		 4303 }
4304
4305 if (PageUptodate(page)) {
4306 if (PageDirty(page))
4307 goto next_page;
4308 } else {
4309 ClearPageError(page);
32159242 4310 err = extent_read_full_page(io_tree, page,
652f25a2
JB		 4311 btrfs_get_extent,
4312 nocow_ctx->mirror_num);
826aa0a8
MX		 4313 if (err) {
4314 ret = err;
ff023aac
SB		 4315 goto next_page;
4316 }
edd1400b 4317
26b25891 4318 lock_page(page);
edd1400b
MX		 4319 /*
4320 * If the page has been remove from the page cache,
4321 * the data on it is meaningless, because it may be
4322 * old one, the new data may be written into the new
4323 * page in the page cache.
4324 */
4325 if (page->mapping != inode->i_mapping) {
652f25a2 4326 unlock_page(page);
edd1400b
MX		 4327 page_cache_release(page);
4328 goto again;
4329 }
ff023aac
SB		 4330 if (!PageUptodate(page)) {
4331 ret = -EIO;
4332 goto next_page;
4333 }
4334 }
32159242
GH		 4335
4336 ret = check_extent_to_block(inode, offset, len,
4337 nocow_ctx_logical);
4338 if (ret) {
4339 ret = ret > 0 ? 0 : ret;
4340 goto next_page;
4341 }
4342
826aa0a8
MX		 4343 err = write_page_nocow(nocow_ctx->sctx,
4344 physical_for_dev_replace, page);
4345 if (err)
4346 ret = err;
ff023aac 4347next_page:
826aa0a8
MX		 4348 unlock_page(page);
4349 page_cache_release(page);
4350
4351 if (ret)
4352 break;
4353
ff023aac
SB		 4354 offset += PAGE_CACHE_SIZE;
4355 physical_for_dev_replace += PAGE_CACHE_SIZE;
32159242 4356 nocow_ctx_logical += PAGE_CACHE_SIZE;
ff023aac
SB		 4357 len -= PAGE_CACHE_SIZE;
4358 }
652f25a2 4359 ret = COPY_COMPLETE;
826aa0a8 4360out:
5955102c 4361 inode_unlock(inode);
826aa0a8 4362 iput(inode);
ff023aac
SB		 4363 return ret;
4364}
4365
4366static int write_page_nocow(struct scrub_ctx *sctx,
4367 u64 physical_for_dev_replace, struct page *page)
4368{
4369 struct bio *bio;
4370 struct btrfs_device *dev;
4371 int ret;
ff023aac
SB		 4372
4373 dev = sctx->wr_ctx.tgtdev;
4374 if (!dev)
4375 return -EIO;
4376 if (!dev->bdev) {
94647322
DS		 4377 btrfs_warn_rl(dev->dev_root->fs_info,
4378 "scrub write_page_nocow(bdev == NULL) is unexpected");
ff023aac
SB		 4379 return -EIO;
4380 }
9be3395b 4381 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
ff023aac
SB		 4382 if (!bio) {
4383 spin_lock(&sctx->stat_lock);
4384 sctx->stat.malloc_errors++;
4385 spin_unlock(&sctx->stat_lock);
4386 return -ENOMEM;
4387 }
4f024f37
KO		 4388 bio->bi_iter.bi_size = 0;
4389 bio->bi_iter.bi_sector = physical_for_dev_replace >> 9;
ff023aac
SB		 4390 bio->bi_bdev = dev->bdev;
4391 ret = bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
4392 if (ret != PAGE_CACHE_SIZE) {
4393leave_with_eio:
4394 bio_put(bio);
4395 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
4396 return -EIO;
4397 }
ff023aac 4398
33879d45 4399 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio))
ff023aac
SB		 4400 goto leave_with_eio;
4401
4402 bio_put(bio);
4403 return 0;
4404}
Linux 6.x block layer and io_uring tree(s)