]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/btrfs/scrub.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: Update out-of-date "skip parity stripe" comment
[mirror_ubuntu-artful-kernel.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
SB		 250static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
251 struct scrub_block *sblock, int is_metadata,
252 int have_csum, u8 *csum, u64 generation,
af8e2d1d 253 u16 csum_size, int retry_failed_mirror);
b5d67f64
SB		 254static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
255 struct scrub_block *sblock,
256 int is_metadata, int have_csum,
257 const u8 *csum, u64 generation,
258 u16 csum_size);
b5d67f64 259static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
114ab50d 260 struct scrub_block *sblock_good);
b5d67f64
SB		 261static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
262 struct scrub_block *sblock_good,
263 int page_num, int force_write);
ff023aac
SB		 264static void scrub_write_block_to_dev_replace(struct scrub_block *sblock);
265static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
266 int page_num);
b5d67f64
SB		 267static int scrub_checksum_data(struct scrub_block *sblock);
268static int scrub_checksum_tree_block(struct scrub_block *sblock);
269static int scrub_checksum_super(struct scrub_block *sblock);
270static void scrub_block_get(struct scrub_block *sblock);
271static void scrub_block_put(struct scrub_block *sblock);
7a9e9987
SB		 272static void scrub_page_get(struct scrub_page *spage);
273static void scrub_page_put(struct scrub_page *spage);
5a6ac9ea
MX		 274static void scrub_parity_get(struct scrub_parity *sparity);
275static void scrub_parity_put(struct scrub_parity *sparity);
ff023aac
SB		 276static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
277 struct scrub_page *spage);
d9d181c1 278static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
a36cf8b8 279 u64 physical, struct btrfs_device *dev, u64 flags,
ff023aac
SB		 280 u64 gen, int mirror_num, u8 *csum, int force,
281 u64 physical_for_dev_replace);
1623edeb 282static void scrub_bio_end_io(struct bio *bio, int err);
b5d67f64
SB		 283static void scrub_bio_end_io_worker(struct btrfs_work *work);
284static void scrub_block_complete(struct scrub_block *sblock);
ff023aac
SB		 285static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
286 u64 extent_logical, u64 extent_len,
287 u64 *extent_physical,
288 struct btrfs_device **extent_dev,
289 int *extent_mirror_num);
290static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
291 struct scrub_wr_ctx *wr_ctx,
292 struct btrfs_fs_info *fs_info,
293 struct btrfs_device *dev,
294 int is_dev_replace);
295static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx);
296static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
297 struct scrub_page *spage);
298static void scrub_wr_submit(struct scrub_ctx *sctx);
299static void scrub_wr_bio_end_io(struct bio *bio, int err);
300static void scrub_wr_bio_end_io_worker(struct btrfs_work *work);
301static int write_page_nocow(struct scrub_ctx *sctx,
302 u64 physical_for_dev_replace, struct page *page);
303static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
652f25a2 304 struct scrub_copy_nocow_ctx *ctx);
ff023aac
SB		 305static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
306 int mirror_num, u64 physical_for_dev_replace);
307static void copy_nocow_pages_worker(struct btrfs_work *work);
cb7ab021 308static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
3cb0929a 309static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
f55985f4 310static void scrub_put_ctx(struct scrub_ctx *sctx);
1623edeb
SB		 311
312
b6bfebc1
SB		 313static void scrub_pending_bio_inc(struct scrub_ctx *sctx)
314{
f55985f4 315 atomic_inc(&sctx->refs);
b6bfebc1
SB		 316 atomic_inc(&sctx->bios_in_flight);
317}
318
319static void scrub_pending_bio_dec(struct scrub_ctx *sctx)
320{
321 atomic_dec(&sctx->bios_in_flight);
322 wake_up(&sctx->list_wait);
f55985f4 323 scrub_put_ctx(sctx);
b6bfebc1
SB		 324}
325
cb7ab021 326static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
3cb0929a
WS		 327{
328 while (atomic_read(&fs_info->scrub_pause_req)) {
329 mutex_unlock(&fs_info->scrub_lock);
330 wait_event(fs_info->scrub_pause_wait,
331 atomic_read(&fs_info->scrub_pause_req) == 0);
332 mutex_lock(&fs_info->scrub_lock);
333 }
334}
335
0e22be89 336static void scrub_pause_on(struct btrfs_fs_info *fs_info)
cb7ab021
WS		 337{
338 atomic_inc(&fs_info->scrubs_paused);
339 wake_up(&fs_info->scrub_pause_wait);
0e22be89 340}
cb7ab021 341
0e22be89
Z		 342static void scrub_pause_off(struct btrfs_fs_info *fs_info)
343{
cb7ab021
WS		 344 mutex_lock(&fs_info->scrub_lock);
345 __scrub_blocked_if_needed(fs_info);
346 atomic_dec(&fs_info->scrubs_paused);
347 mutex_unlock(&fs_info->scrub_lock);
348
349 wake_up(&fs_info->scrub_pause_wait);
350}
351
0e22be89
Z		 352static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
353{
354 scrub_pause_on(fs_info);
355 scrub_pause_off(fs_info);
356}
357
b6bfebc1
SB		 358/*
359 * used for workers that require transaction commits (i.e., for the
360 * NOCOW case)
361 */
362static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx)
363{
364 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
365
f55985f4 366 atomic_inc(&sctx->refs);
b6bfebc1
SB		 367 /*
368 * increment scrubs_running to prevent cancel requests from
369 * completing as long as a worker is running. we must also
370 * increment scrubs_paused to prevent deadlocking on pause
371 * requests used for transactions commits (as the worker uses a
372 * transaction context). it is safe to regard the worker
373 * as paused for all matters practical. effectively, we only
374 * avoid cancellation requests from completing.
375 */
376 mutex_lock(&fs_info->scrub_lock);
377 atomic_inc(&fs_info->scrubs_running);
378 atomic_inc(&fs_info->scrubs_paused);
379 mutex_unlock(&fs_info->scrub_lock);
32a44789
WS		 380
381 /*
382 * check if @scrubs_running=@scrubs_paused condition
383 * inside wait_event() is not an atomic operation.
384 * which means we may inc/dec @scrub_running/paused
385 * at any time. Let's wake up @scrub_pause_wait as
386 * much as we can to let commit transaction blocked less.
387 */
388 wake_up(&fs_info->scrub_pause_wait);
389
b6bfebc1
SB		 390 atomic_inc(&sctx->workers_pending);
391}
392
393/* used for workers that require transaction commits */
394static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx)
395{
396 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
397
398 /*
399 * see scrub_pending_trans_workers_inc() why we're pretending
400 * to be paused in the scrub counters
401 */
402 mutex_lock(&fs_info->scrub_lock);
403 atomic_dec(&fs_info->scrubs_running);
404 atomic_dec(&fs_info->scrubs_paused);
405 mutex_unlock(&fs_info->scrub_lock);
406 atomic_dec(&sctx->workers_pending);
407 wake_up(&fs_info->scrub_pause_wait);
408 wake_up(&sctx->list_wait);
f55985f4 409 scrub_put_ctx(sctx);
b6bfebc1
SB		 410}
411
d9d181c1 412static void scrub_free_csums(struct scrub_ctx *sctx)
a2de733c 413{
d9d181c1 414 while (!list_empty(&sctx->csum_list)) {
a2de733c 415 struct btrfs_ordered_sum *sum;
d9d181c1 416 sum = list_first_entry(&sctx->csum_list,
a2de733c
AJ		 417 struct btrfs_ordered_sum, list);
418 list_del(&sum->list);
419 kfree(sum);
420 }
421}
422
d9d181c1 423static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx)
a2de733c
AJ		 424{
425 int i;
a2de733c 426
d9d181c1 427 if (!sctx)
a2de733c
AJ		 428 return;
429
ff023aac
SB		 430 scrub_free_wr_ctx(&sctx->wr_ctx);
431
b5d67f64 432 /* this can happen when scrub is cancelled */
d9d181c1
SB		 433 if (sctx->curr != -1) {
434 struct scrub_bio *sbio = sctx->bios[sctx->curr];
b5d67f64
SB		 435
436 for (i = 0; i < sbio->page_count; i++) {
ff023aac 437 WARN_ON(!sbio->pagev[i]->page);
b5d67f64
SB		 438 scrub_block_put(sbio->pagev[i]->sblock);
439 }
440 bio_put(sbio->bio);
441 }
442
ff023aac 443 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
d9d181c1 444 struct scrub_bio *sbio = sctx->bios[i];
a2de733c
AJ		 445
446 if (!sbio)
447 break;
a2de733c
AJ		 448 kfree(sbio);
449 }
450
d9d181c1
SB		 451 scrub_free_csums(sctx);
452 kfree(sctx);
a2de733c
AJ		 453}
454
f55985f4
FM		 455static void scrub_put_ctx(struct scrub_ctx *sctx)
456{
457 if (atomic_dec_and_test(&sctx->refs))
458 scrub_free_ctx(sctx);
459}
460
a2de733c 461static noinline_for_stack
63a212ab 462struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace)
a2de733c 463{
d9d181c1 464 struct scrub_ctx *sctx;
a2de733c 465 int i;
a2de733c 466 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
ff023aac
SB		 467 int pages_per_rd_bio;
468 int ret;
a2de733c 469
ff023aac
SB		 470 /*
471 * the setting of pages_per_rd_bio is correct for scrub but might
472 * be wrong for the dev_replace code where we might read from
473 * different devices in the initial huge bios. However, that
474 * code is able to correctly handle the case when adding a page
475 * to a bio fails.
476 */
477 if (dev->bdev)
478 pages_per_rd_bio = min_t(int, SCRUB_PAGES_PER_RD_BIO,
479 bio_get_nr_vecs(dev->bdev));
480 else
481 pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO;
d9d181c1
SB		 482 sctx = kzalloc(sizeof(*sctx), GFP_NOFS);
483 if (!sctx)
a2de733c 484 goto nomem;
f55985f4 485 atomic_set(&sctx->refs, 1);
63a212ab 486 sctx->is_dev_replace = is_dev_replace;
ff023aac 487 sctx->pages_per_rd_bio = pages_per_rd_bio;
d9d181c1 488 sctx->curr = -1;
a36cf8b8 489 sctx->dev_root = dev->dev_root;
ff023aac 490 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
a2de733c
AJ		 491 struct scrub_bio *sbio;
492
493 sbio = kzalloc(sizeof(*sbio), GFP_NOFS);
494 if (!sbio)
495 goto nomem;
d9d181c1 496 sctx->bios[i] = sbio;
a2de733c 497
a2de733c 498 sbio->index = i;
d9d181c1 499 sbio->sctx = sctx;
b5d67f64 500 sbio->page_count = 0;
9e0af237
LB		 501 btrfs_init_work(&sbio->work, btrfs_scrub_helper,
502 scrub_bio_end_io_worker, NULL, NULL);
a2de733c 503
ff023aac 504 if (i != SCRUB_BIOS_PER_SCTX - 1)
d9d181c1 505 sctx->bios[i]->next_free = i + 1;
0ef8e451 506 else
d9d181c1
SB		 507 sctx->bios[i]->next_free = -1;
508 }
509 sctx->first_free = 0;
510 sctx->nodesize = dev->dev_root->nodesize;
d9d181c1 511 sctx->sectorsize = dev->dev_root->sectorsize;
b6bfebc1
SB		 512 atomic_set(&sctx->bios_in_flight, 0);
513 atomic_set(&sctx->workers_pending, 0);
d9d181c1
SB		 514 atomic_set(&sctx->cancel_req, 0);
515 sctx->csum_size = btrfs_super_csum_size(fs_info->super_copy);
516 INIT_LIST_HEAD(&sctx->csum_list);
517
518 spin_lock_init(&sctx->list_lock);
519 spin_lock_init(&sctx->stat_lock);
520 init_waitqueue_head(&sctx->list_wait);
ff023aac
SB		 521
522 ret = scrub_setup_wr_ctx(sctx, &sctx->wr_ctx, fs_info,
523 fs_info->dev_replace.tgtdev, is_dev_replace);
524 if (ret) {
525 scrub_free_ctx(sctx);
526 return ERR_PTR(ret);
527 }
d9d181c1 528 return sctx;
a2de733c
AJ		 529
530nomem:
d9d181c1 531 scrub_free_ctx(sctx);
a2de733c
AJ		 532 return ERR_PTR(-ENOMEM);
533}
534
ff023aac
SB		 535static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root,
536 void *warn_ctx)
558540c1
JS		 537{
538 u64 isize;
539 u32 nlink;
540 int ret;
541 int i;
542 struct extent_buffer *eb;
543 struct btrfs_inode_item *inode_item;
ff023aac 544 struct scrub_warning *swarn = warn_ctx;
558540c1
JS		 545 struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info;
546 struct inode_fs_paths *ipath = NULL;
547 struct btrfs_root *local_root;
548 struct btrfs_key root_key;
1d4c08e0 549 struct btrfs_key key;
558540c1
JS		 550
551 root_key.objectid = root;
552 root_key.type = BTRFS_ROOT_ITEM_KEY;
553 root_key.offset = (u64)-1;
554 local_root = btrfs_read_fs_root_no_name(fs_info, &root_key);
555 if (IS_ERR(local_root)) {
556 ret = PTR_ERR(local_root);
557 goto err;
558 }
559
14692cc1
DS		 560 /*
561 * this makes the path point to (inum INODE_ITEM ioff)
562 */
1d4c08e0
DS		 563 key.objectid = inum;
564 key.type = BTRFS_INODE_ITEM_KEY;
565 key.offset = 0;
566
567 ret = btrfs_search_slot(NULL, local_root, &key, swarn->path, 0, 0);
558540c1
JS		 568 if (ret) {
569 btrfs_release_path(swarn->path);
570 goto err;
571 }
572
573 eb = swarn->path->nodes[0];
574 inode_item = btrfs_item_ptr(eb, swarn->path->slots[0],
575 struct btrfs_inode_item);
576 isize = btrfs_inode_size(eb, inode_item);
577 nlink = btrfs_inode_nlink(eb, inode_item);
578 btrfs_release_path(swarn->path);
579
580 ipath = init_ipath(4096, local_root, swarn->path);
26bdef54
DC		 581 if (IS_ERR(ipath)) {
582 ret = PTR_ERR(ipath);
583 ipath = NULL;
584 goto err;
585 }
558540c1
JS		 586 ret = paths_from_inode(inum, ipath);
587
588 if (ret < 0)
589 goto err;
590
591 /*
592 * we deliberately ignore the bit ipath might have been too small to
593 * hold all of the paths here
594 */
595 for (i = 0; i < ipath->fspath->elem_cnt; ++i)
efe120a0 596 printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
558540c1
JS		 597 "%s, sector %llu, root %llu, inode %llu, offset %llu, "
598 "length %llu, links %u (path: %s)\n", swarn->errstr,
606686ee 599 swarn->logical, rcu_str_deref(swarn->dev->name),
558540c1
JS		 600 (unsigned long long)swarn->sector, root, inum, offset,
601 min(isize - offset, (u64)PAGE_SIZE), nlink,
745c4d8e 602 (char *)(unsigned long)ipath->fspath->val[i]);
558540c1
JS		 603
604 free_ipath(ipath);
605 return 0;
606
607err:
efe120a0 608 printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
558540c1
JS		 609 "%s, sector %llu, root %llu, inode %llu, offset %llu: path "
610 "resolving failed with ret=%d\n", swarn->errstr,
606686ee 611 swarn->logical, rcu_str_deref(swarn->dev->name),
558540c1
JS		 612 (unsigned long long)swarn->sector, root, inum, offset, ret);
613
614 free_ipath(ipath);
615 return 0;
616}
617
b5d67f64 618static void scrub_print_warning(const char *errstr, struct scrub_block *sblock)
558540c1 619{
a36cf8b8
SB		 620 struct btrfs_device *dev;
621 struct btrfs_fs_info *fs_info;
558540c1
JS		 622 struct btrfs_path *path;
623 struct btrfs_key found_key;
624 struct extent_buffer *eb;
625 struct btrfs_extent_item *ei;
626 struct scrub_warning swarn;
69917e43
LB		 627 unsigned long ptr = 0;
628 u64 extent_item_pos;
629 u64 flags = 0;
558540c1 630 u64 ref_root;
69917e43 631 u32 item_size;
558540c1 632 u8 ref_level;
69917e43 633 int ret;
558540c1 634
a36cf8b8 635 WARN_ON(sblock->page_count < 1);
7a9e9987 636 dev = sblock->pagev[0]->dev;
a36cf8b8
SB		 637 fs_info = sblock->sctx->dev_root->fs_info;
638
558540c1 639 path = btrfs_alloc_path();
8b9456da
DS		 640 if (!path)
641 return;
558540c1 642
7a9e9987
SB		 643 swarn.sector = (sblock->pagev[0]->physical) >> 9;
644 swarn.logical = sblock->pagev[0]->logical;
558540c1 645 swarn.errstr = errstr;
a36cf8b8 646 swarn.dev = NULL;
558540c1 647
69917e43
LB		 648 ret = extent_from_logical(fs_info, swarn.logical, path, &found_key,
649 &flags);
558540c1
JS		 650 if (ret < 0)
651 goto out;
652
4692cf58 653 extent_item_pos = swarn.logical - found_key.objectid;
558540c1
JS		 654 swarn.extent_item_size = found_key.offset;
655
656 eb = path->nodes[0];
657 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
658 item_size = btrfs_item_size_nr(eb, path->slots[0]);
659
69917e43 660 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
558540c1 661 do {
6eda71d0
LB		 662 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
663 item_size, &ref_root,
664 &ref_level);
606686ee 665 printk_in_rcu(KERN_WARNING
efe120a0 666 "BTRFS: %s at logical %llu on dev %s, "
558540c1 667 "sector %llu: metadata %s (level %d) in tree "
606686ee
JB		 668 "%llu\n", errstr, swarn.logical,
669 rcu_str_deref(dev->name),
558540c1
JS		 670 (unsigned long long)swarn.sector,
671 ref_level ? "node" : "leaf",
672 ret < 0 ? -1 : ref_level,
673 ret < 0 ? -1 : ref_root);
674 } while (ret != 1);
d8fe29e9 675 btrfs_release_path(path);
558540c1 676 } else {
d8fe29e9 677 btrfs_release_path(path);
558540c1 678 swarn.path = path;
a36cf8b8 679 swarn.dev = dev;
7a3ae2f8
JS		 680 iterate_extent_inodes(fs_info, found_key.objectid,
681 extent_item_pos, 1,
558540c1
JS		 682 scrub_print_warning_inode, &swarn);
683 }
684
685out:
686 btrfs_free_path(path);
558540c1
JS		 687}
688
ff023aac 689static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx)
0ef8e451 690{
5da6fcbc 691 struct page *page = NULL;
0ef8e451 692 unsigned long index;
ff023aac 693 struct scrub_fixup_nodatasum *fixup = fixup_ctx;
0ef8e451 694 int ret;
5da6fcbc 695 int corrected = 0;
0ef8e451 696 struct btrfs_key key;
5da6fcbc 697 struct inode *inode = NULL;
6f1c3605 698 struct btrfs_fs_info *fs_info;
0ef8e451
JS		 699 u64 end = offset + PAGE_SIZE - 1;
700 struct btrfs_root *local_root;
6f1c3605 701 int srcu_index;
0ef8e451
JS		 702
703 key.objectid = root;
704 key.type = BTRFS_ROOT_ITEM_KEY;
705 key.offset = (u64)-1;
6f1c3605
LB		 706
707 fs_info = fixup->root->fs_info;
708 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
709
710 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
711 if (IS_ERR(local_root)) {
712 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
0ef8e451 713 return PTR_ERR(local_root);
6f1c3605 714 }
0ef8e451
JS		 715
716 key.type = BTRFS_INODE_ITEM_KEY;
717 key.objectid = inum;
718 key.offset = 0;
6f1c3605
LB		 719 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
720 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
0ef8e451
JS		 721 if (IS_ERR(inode))
722 return PTR_ERR(inode);
723
0ef8e451
JS		 724 index = offset >> PAGE_CACHE_SHIFT;
725
726 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
5da6fcbc
JS		 727 if (!page) {
728 ret = -ENOMEM;
729 goto out;
730 }
731
732 if (PageUptodate(page)) {
5da6fcbc
JS		 733 if (PageDirty(page)) {
734 /*
735 * we need to write the data to the defect sector. the
736 * data that was in that sector is not in memory,
737 * because the page was modified. we must not write the
738 * modified page to that sector.
739 *
740 * TODO: what could be done here: wait for the delalloc
741 * runner to write out that page (might involve
742 * COW) and see whether the sector is still
743 * referenced afterwards.
744 *
745 * For the meantime, we'll treat this error
746 * incorrectable, although there is a chance that a
747 * later scrub will find the bad sector again and that
748 * there's no dirty page in memory, then.
749 */
750 ret = -EIO;
751 goto out;
752 }
1203b681 753 ret = repair_io_failure(inode, offset, PAGE_SIZE,
5da6fcbc 754 fixup->logical, page,
ffdd2018 755 offset - page_offset(page),
5da6fcbc
JS		 756 fixup->mirror_num);
757 unlock_page(page);
758 corrected = !ret;
759 } else {
760 /*
761 * we need to get good data first. the general readpage path
762 * will call repair_io_failure for us, we just have to make
763 * sure we read the bad mirror.
764 */
765 ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
766 EXTENT_DAMAGED, GFP_NOFS);
767 if (ret) {
768 /* set_extent_bits should give proper error */
769 WARN_ON(ret > 0);
770 if (ret > 0)
771 ret = -EFAULT;
772 goto out;
773 }
774
775 ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page,
776 btrfs_get_extent,
777 fixup->mirror_num);
778 wait_on_page_locked(page);
779
780 corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset,
781 end, EXTENT_DAMAGED, 0, NULL);
782 if (!corrected)
783 clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
784 EXTENT_DAMAGED, GFP_NOFS);
785 }
786
787out:
788 if (page)
789 put_page(page);
7fb18a06
TK		 790
791 iput(inode);
0ef8e451
JS		 792
793 if (ret < 0)
794 return ret;
795
796 if (ret == 0 && corrected) {
797 /*
798 * we only need to call readpage for one of the inodes belonging
799 * to this extent. so make iterate_extent_inodes stop
800 */
801 return 1;
802 }
803
804 return -EIO;
805}
806
807static void scrub_fixup_nodatasum(struct btrfs_work *work)
808{
809 int ret;
810 struct scrub_fixup_nodatasum *fixup;
d9d181c1 811 struct scrub_ctx *sctx;
0ef8e451 812 struct btrfs_trans_handle *trans = NULL;
0ef8e451
JS		 813 struct btrfs_path *path;
814 int uncorrectable = 0;
815
816 fixup = container_of(work, struct scrub_fixup_nodatasum, work);
d9d181c1 817 sctx = fixup->sctx;
0ef8e451
JS		 818
819 path = btrfs_alloc_path();
820 if (!path) {
d9d181c1
SB		 821 spin_lock(&sctx->stat_lock);
822 ++sctx->stat.malloc_errors;
823 spin_unlock(&sctx->stat_lock);
0ef8e451
JS		 824 uncorrectable = 1;
825 goto out;
826 }
827
828 trans = btrfs_join_transaction(fixup->root);
829 if (IS_ERR(trans)) {
830 uncorrectable = 1;
831 goto out;
832 }
833
834 /*
835 * the idea is to trigger a regular read through the standard path. we
836 * read a page from the (failed) logical address by specifying the
837 * corresponding copynum of the failed sector. thus, that readpage is
838 * expected to fail.
839 * that is the point where on-the-fly error correction will kick in
840 * (once it's finished) and rewrite the failed sector if a good copy
841 * can be found.
842 */
843 ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info,
844 path, scrub_fixup_readpage,
845 fixup);
846 if (ret < 0) {
847 uncorrectable = 1;
848 goto out;
849 }
850 WARN_ON(ret != 1);
851
d9d181c1
SB		 852 spin_lock(&sctx->stat_lock);
853 ++sctx->stat.corrected_errors;
854 spin_unlock(&sctx->stat_lock);
0ef8e451
JS		 855
856out:
857 if (trans && !IS_ERR(trans))
858 btrfs_end_transaction(trans, fixup->root);
859 if (uncorrectable) {
d9d181c1
SB		 860 spin_lock(&sctx->stat_lock);
861 ++sctx->stat.uncorrectable_errors;
862 spin_unlock(&sctx->stat_lock);
ff023aac
SB		 863 btrfs_dev_replace_stats_inc(
864 &sctx->dev_root->fs_info->dev_replace.
865 num_uncorrectable_read_errors);
efe120a0
FH		 866 printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
867 "unable to fixup (nodatasum) error at logical %llu on dev %s\n",
c1c9ff7c 868 fixup->logical, rcu_str_deref(fixup->dev->name));
0ef8e451
JS		 869 }
870
871 btrfs_free_path(path);
872 kfree(fixup);
873
b6bfebc1 874 scrub_pending_trans_workers_dec(sctx);
0ef8e451
JS		 875}
876
af8e2d1d
MX		 877static inline void scrub_get_recover(struct scrub_recover *recover)
878{
879 atomic_inc(&recover->refs);
880}
881
882static inline void scrub_put_recover(struct scrub_recover *recover)
883{
884 if (atomic_dec_and_test(&recover->refs)) {
6e9606d2 885 btrfs_put_bbio(recover->bbio);
af8e2d1d
MX		 886 kfree(recover);
887 }
888}
889
a2de733c 890/*
b5d67f64
SB		 891 * scrub_handle_errored_block gets called when either verification of the
892 * pages failed or the bio failed to read, e.g. with EIO. In the latter
893 * case, this function handles all pages in the bio, even though only one
894 * may be bad.
895 * The goal of this function is to repair the errored block by using the
896 * contents of one of the mirrors.
a2de733c 897 */
b5d67f64 898static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
a2de733c 899{
d9d181c1 900 struct scrub_ctx *sctx = sblock_to_check->sctx;
a36cf8b8 901 struct btrfs_device *dev;
b5d67f64
SB		 902 struct btrfs_fs_info *fs_info;
903 u64 length;
904 u64 logical;
905 u64 generation;
906 unsigned int failed_mirror_index;
907 unsigned int is_metadata;
908 unsigned int have_csum;
909 u8 *csum;
910 struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */
911 struct scrub_block *sblock_bad;
912 int ret;
913 int mirror_index;
914 int page_num;
915 int success;
558540c1 916 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
b5d67f64
SB		 917 DEFAULT_RATELIMIT_BURST);
918
919 BUG_ON(sblock_to_check->page_count < 1);
a36cf8b8 920 fs_info = sctx->dev_root->fs_info;
4ded4f63
SB		 921 if (sblock_to_check->pagev[0]->flags & BTRFS_EXTENT_FLAG_SUPER) {
922 /*
923 * if we find an error in a super block, we just report it.
924 * They will get written with the next transaction commit
925 * anyway
926 */
927 spin_lock(&sctx->stat_lock);
928 ++sctx->stat.super_errors;
929 spin_unlock(&sctx->stat_lock);
930 return 0;
931 }
b5d67f64 932 length = sblock_to_check->page_count * PAGE_SIZE;
7a9e9987
SB		 933 logical = sblock_to_check->pagev[0]->logical;
934 generation = sblock_to_check->pagev[0]->generation;
935 BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1);
936 failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1;
937 is_metadata = !(sblock_to_check->pagev[0]->flags &
b5d67f64 938 BTRFS_EXTENT_FLAG_DATA);
7a9e9987
SB		 939 have_csum = sblock_to_check->pagev[0]->have_csum;
940 csum = sblock_to_check->pagev[0]->csum;
941 dev = sblock_to_check->pagev[0]->dev;
13db62b7 942
ff023aac
SB		 943 if (sctx->is_dev_replace && !is_metadata && !have_csum) {
944 sblocks_for_recheck = NULL;
945 goto nodatasum_case;
946 }
947
b5d67f64
SB		 948 /*
949 * read all mirrors one after the other. This includes to
950 * re-read the extent or metadata block that failed (that was
951 * the cause that this fixup code is called) another time,
952 * page by page this time in order to know which pages
953 * caused I/O errors and which ones are good (for all mirrors).
954 * It is the goal to handle the situation when more than one
955 * mirror contains I/O errors, but the errors do not
956 * overlap, i.e. the data can be repaired by selecting the
957 * pages from those mirrors without I/O error on the
958 * particular pages. One example (with blocks >= 2 * PAGE_SIZE)
959 * would be that mirror #1 has an I/O error on the first page,
960 * the second page is good, and mirror #2 has an I/O error on
961 * the second page, but the first page is good.
962 * Then the first page of the first mirror can be repaired by
963 * taking the first page of the second mirror, and the
964 * second page of the second mirror can be repaired by
965 * copying the contents of the 2nd page of the 1st mirror.
966 * One more note: if the pages of one mirror contain I/O
967 * errors, the checksum cannot be verified. In order to get
968 * the best data for repairing, the first attempt is to find
969 * a mirror without I/O errors and with a validated checksum.
970 * Only if this is not possible, the pages are picked from
971 * mirrors with I/O errors without considering the checksum.
972 * If the latter is the case, at the end, the checksum of the
973 * repaired area is verified in order to correctly maintain
974 * the statistics.
975 */
976
31e818fe
DS		 977 sblocks_for_recheck = kcalloc(BTRFS_MAX_MIRRORS,
978 sizeof(*sblocks_for_recheck), GFP_NOFS);
b5d67f64 979 if (!sblocks_for_recheck) {
d9d181c1
SB		 980 spin_lock(&sctx->stat_lock);
981 sctx->stat.malloc_errors++;
982 sctx->stat.read_errors++;
983 sctx->stat.uncorrectable_errors++;
984 spin_unlock(&sctx->stat_lock);
a36cf8b8 985 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
b5d67f64 986 goto out;
a2de733c
AJ		 987 }
988
b5d67f64 989 /* setup the context, map the logical blocks and alloc the pages */
be50a8dd 990 ret = scrub_setup_recheck_block(sblock_to_check, sblocks_for_recheck);
b5d67f64 991 if (ret) {
d9d181c1
SB		 992 spin_lock(&sctx->stat_lock);
993 sctx->stat.read_errors++;
994 sctx->stat.uncorrectable_errors++;
995 spin_unlock(&sctx->stat_lock);
a36cf8b8 996 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
b5d67f64
SB		 997 goto out;
998 }
999 BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS);
1000 sblock_bad = sblocks_for_recheck + failed_mirror_index;
13db62b7 1001
b5d67f64 1002 /* build and submit the bios for the failed mirror, check checksums */
34f5c8e9 1003 scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum,
af8e2d1d 1004 csum, generation, sctx->csum_size, 1);
a2de733c 1005
b5d67f64
SB		 1006 if (!sblock_bad->header_error && !sblock_bad->checksum_error &&
1007 sblock_bad->no_io_error_seen) {
1008 /*
1009 * the error disappeared after reading page by page, or
1010 * the area was part of a huge bio and other parts of the
1011 * bio caused I/O errors, or the block layer merged several
1012 * read requests into one and the error is caused by a
1013 * different bio (usually one of the two latter cases is
1014 * the cause)
1015 */
d9d181c1
SB		 1016 spin_lock(&sctx->stat_lock);
1017 sctx->stat.unverified_errors++;
5a6ac9ea 1018 sblock_to_check->data_corrected = 1;
d9d181c1 1019 spin_unlock(&sctx->stat_lock);
a2de733c 1020
ff023aac
SB		 1021 if (sctx->is_dev_replace)
1022 scrub_write_block_to_dev_replace(sblock_bad);
b5d67f64 1023 goto out;
a2de733c 1024 }
a2de733c 1025
b5d67f64 1026 if (!sblock_bad->no_io_error_seen) {
d9d181c1
SB		 1027 spin_lock(&sctx->stat_lock);
1028 sctx->stat.read_errors++;
1029 spin_unlock(&sctx->stat_lock);
b5d67f64
SB		 1030 if (__ratelimit(&_rs))
1031 scrub_print_warning("i/o error", sblock_to_check);
a36cf8b8 1032 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
b5d67f64 1033 } else if (sblock_bad->checksum_error) {
d9d181c1
SB		 1034 spin_lock(&sctx->stat_lock);
1035 sctx->stat.csum_errors++;
1036 spin_unlock(&sctx->stat_lock);
b5d67f64
SB		 1037 if (__ratelimit(&_rs))
1038 scrub_print_warning("checksum error", sblock_to_check);
a36cf8b8 1039 btrfs_dev_stat_inc_and_print(dev,
442a4f63 1040 BTRFS_DEV_STAT_CORRUPTION_ERRS);
b5d67f64 1041 } else if (sblock_bad->header_error) {
d9d181c1
SB		 1042 spin_lock(&sctx->stat_lock);
1043 sctx->stat.verify_errors++;
1044 spin_unlock(&sctx->stat_lock);
b5d67f64
SB		 1045 if (__ratelimit(&_rs))
1046 scrub_print_warning("checksum/header error",
1047 sblock_to_check);
442a4f63 1048 if (sblock_bad->generation_error)
a36cf8b8 1049 btrfs_dev_stat_inc_and_print(dev,
442a4f63
SB		 1050 BTRFS_DEV_STAT_GENERATION_ERRS);
1051 else
a36cf8b8 1052 btrfs_dev_stat_inc_and_print(dev,
442a4f63 1053 BTRFS_DEV_STAT_CORRUPTION_ERRS);
b5d67f64 1054 }
a2de733c 1055
33ef30ad
ID		 1056 if (sctx->readonly) {
1057 ASSERT(!sctx->is_dev_replace);
1058 goto out;
1059 }
a2de733c 1060
b5d67f64
SB		 1061 if (!is_metadata && !have_csum) {
1062 struct scrub_fixup_nodatasum *fixup_nodatasum;
a2de733c 1063
ff023aac
SB		 1064 WARN_ON(sctx->is_dev_replace);
1065
b25c94c5
ZL		 1066nodatasum_case:
1067
b5d67f64
SB		 1068 /*
1069 * !is_metadata and !have_csum, this means that the data
1070 * might not be COW'ed, that it might be modified
1071 * concurrently. The general strategy to work on the
1072 * commit root does not help in the case when COW is not
1073 * used.
1074 */
1075 fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS);
1076 if (!fixup_nodatasum)
1077 goto did_not_correct_error;
d9d181c1 1078 fixup_nodatasum->sctx = sctx;
a36cf8b8 1079 fixup_nodatasum->dev = dev;
b5d67f64
SB		 1080 fixup_nodatasum->logical = logical;
1081 fixup_nodatasum->root = fs_info->extent_root;
1082 fixup_nodatasum->mirror_num = failed_mirror_index + 1;
b6bfebc1 1083 scrub_pending_trans_workers_inc(sctx);
9e0af237
LB		 1084 btrfs_init_work(&fixup_nodatasum->work, btrfs_scrub_helper,
1085 scrub_fixup_nodatasum, NULL, NULL);
0339ef2f
QW		 1086 btrfs_queue_work(fs_info->scrub_workers,
1087 &fixup_nodatasum->work);
b5d67f64 1088 goto out;
a2de733c
AJ		 1089 }
1090
b5d67f64
SB		 1091 /*
1092 * now build and submit the bios for the other mirrors, check
cb2ced73
SB		 1093 * checksums.
1094 * First try to pick the mirror which is completely without I/O
b5d67f64
SB		 1095 * errors and also does not have a checksum error.
1096 * If one is found, and if a checksum is present, the full block
1097 * that is known to contain an error is rewritten. Afterwards
1098 * the block is known to be corrected.
1099 * If a mirror is found which is completely correct, and no
1100 * checksum is present, only those pages are rewritten that had
1101 * an I/O error in the block to be repaired, since it cannot be
1102 * determined, which copy of the other pages is better (and it
1103 * could happen otherwise that a correct page would be
1104 * overwritten by a bad one).
1105 */
1106 for (mirror_index = 0;
1107 mirror_index < BTRFS_MAX_MIRRORS &&
1108 sblocks_for_recheck[mirror_index].page_count > 0;
1109 mirror_index++) {
cb2ced73 1110 struct scrub_block *sblock_other;
b5d67f64 1111
cb2ced73
SB		 1112 if (mirror_index == failed_mirror_index)
1113 continue;
1114 sblock_other = sblocks_for_recheck + mirror_index;
1115
1116 /* build and submit the bios, check checksums */
34f5c8e9
SB		 1117 scrub_recheck_block(fs_info, sblock_other, is_metadata,
1118 have_csum, csum, generation,
af8e2d1d 1119 sctx->csum_size, 0);
34f5c8e9
SB		 1120
1121 if (!sblock_other->header_error &&
b5d67f64
SB		 1122 !sblock_other->checksum_error &&
1123 sblock_other->no_io_error_seen) {
ff023aac
SB		 1124 if (sctx->is_dev_replace) {
1125 scrub_write_block_to_dev_replace(sblock_other);
114ab50d 1126 goto corrected_error;
ff023aac 1127 } else {
ff023aac 1128 ret = scrub_repair_block_from_good_copy(
114ab50d
ZL		 1129 sblock_bad, sblock_other);
1130 if (!ret)
1131 goto corrected_error;
ff023aac 1132 }
b5d67f64
SB		 1133 }
1134 }
a2de733c 1135
b968fed1
ZL		 1136 if (sblock_bad->no_io_error_seen && !sctx->is_dev_replace)
1137 goto did_not_correct_error;
ff023aac
SB		 1138
1139 /*
ff023aac 1140 * In case of I/O errors in the area that is supposed to be
b5d67f64
SB		 1141 * repaired, continue by picking good copies of those pages.
1142 * Select the good pages from mirrors to rewrite bad pages from
1143 * the area to fix. Afterwards verify the checksum of the block
1144 * that is supposed to be repaired. This verification step is
1145 * only done for the purpose of statistic counting and for the
1146 * final scrub report, whether errors remain.
1147 * A perfect algorithm could make use of the checksum and try
1148 * all possible combinations of pages from the different mirrors
1149 * until the checksum verification succeeds. For example, when
1150 * the 2nd page of mirror #1 faces I/O errors, and the 2nd page
1151 * of mirror #2 is readable but the final checksum test fails,
1152 * then the 2nd page of mirror #3 could be tried, whether now
1153 * the final checksum succeedes. But this would be a rare
1154 * exception and is therefore not implemented. At least it is
1155 * avoided that the good copy is overwritten.
1156 * A more useful improvement would be to pick the sectors
1157 * without I/O error based on sector sizes (512 bytes on legacy
1158 * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one
1159 * mirror could be repaired by taking 512 byte of a different
1160 * mirror, even if other 512 byte sectors in the same PAGE_SIZE
1161 * area are unreadable.
a2de733c 1162 */
b5d67f64 1163 success = 1;
b968fed1
ZL		 1164 for (page_num = 0; page_num < sblock_bad->page_count;
1165 page_num++) {
7a9e9987 1166 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
b968fed1 1167 struct scrub_block *sblock_other = NULL;
b5d67f64 1168
b968fed1
ZL		 1169 /* skip no-io-error page in scrub */
1170 if (!page_bad->io_error && !sctx->is_dev_replace)
a2de733c 1171 continue;
b5d67f64 1172
b968fed1
ZL		 1173 /* try to find no-io-error page in mirrors */
1174 if (page_bad->io_error) {
1175 for (mirror_index = 0;
1176 mirror_index < BTRFS_MAX_MIRRORS &&
1177 sblocks_for_recheck[mirror_index].page_count > 0;
1178 mirror_index++) {
1179 if (!sblocks_for_recheck[mirror_index].
1180 pagev[page_num]->io_error) {
1181 sblock_other = sblocks_for_recheck +
1182 mirror_index;
1183 break;
b5d67f64
SB		 1184 }
1185 }
b968fed1
ZL		 1186 if (!sblock_other)
1187 success = 0;
96e36920 1188 }
a2de733c 1189
b968fed1
ZL		 1190 if (sctx->is_dev_replace) {
1191 /*
1192 * did not find a mirror to fetch the page
1193 * from. scrub_write_page_to_dev_replace()
1194 * handles this case (page->io_error), by
1195 * filling the block with zeros before
1196 * submitting the write request
1197 */
1198 if (!sblock_other)
1199 sblock_other = sblock_bad;
1200
1201 if (scrub_write_page_to_dev_replace(sblock_other,
1202 page_num) != 0) {
1203 btrfs_dev_replace_stats_inc(
1204 &sctx->dev_root->
1205 fs_info->dev_replace.
1206 num_write_errors);
1207 success = 0;
1208 }
1209 } else if (sblock_other) {
1210 ret = scrub_repair_page_from_good_copy(sblock_bad,
1211 sblock_other,
1212 page_num, 0);
1213 if (0 == ret)
1214 page_bad->io_error = 0;
1215 else
1216 success = 0;
b5d67f64 1217 }
a2de733c 1218 }
a2de733c 1219
b968fed1 1220 if (success && !sctx->is_dev_replace) {
b5d67f64
SB		 1221 if (is_metadata || have_csum) {
1222 /*
1223 * need to verify the checksum now that all
1224 * sectors on disk are repaired (the write
1225 * request for data to be repaired is on its way).
1226 * Just be lazy and use scrub_recheck_block()
1227 * which re-reads the data before the checksum
1228 * is verified, but most likely the data comes out
1229 * of the page cache.
1230 */
34f5c8e9
SB		 1231 scrub_recheck_block(fs_info, sblock_bad,
1232 is_metadata, have_csum, csum,
af8e2d1d 1233 generation, sctx->csum_size, 1);
34f5c8e9 1234 if (!sblock_bad->header_error &&
b5d67f64
SB		 1235 !sblock_bad->checksum_error &&
1236 sblock_bad->no_io_error_seen)
1237 goto corrected_error;
1238 else
1239 goto did_not_correct_error;
1240 } else {
1241corrected_error:
d9d181c1
SB		 1242 spin_lock(&sctx->stat_lock);
1243 sctx->stat.corrected_errors++;
5a6ac9ea 1244 sblock_to_check->data_corrected = 1;
d9d181c1 1245 spin_unlock(&sctx->stat_lock);
606686ee 1246 printk_ratelimited_in_rcu(KERN_ERR
efe120a0 1247 "BTRFS: fixed up error at logical %llu on dev %s\n",
c1c9ff7c 1248 logical, rcu_str_deref(dev->name));
8628764e 1249 }
b5d67f64
SB		 1250 } else {
1251did_not_correct_error:
d9d181c1
SB		 1252 spin_lock(&sctx->stat_lock);
1253 sctx->stat.uncorrectable_errors++;
1254 spin_unlock(&sctx->stat_lock);
606686ee 1255 printk_ratelimited_in_rcu(KERN_ERR
efe120a0 1256 "BTRFS: unable to fixup (regular) error at logical %llu on dev %s\n",
c1c9ff7c 1257 logical, rcu_str_deref(dev->name));
96e36920 1258 }
a2de733c 1259
b5d67f64
SB		 1260out:
1261 if (sblocks_for_recheck) {
1262 for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS;
1263 mirror_index++) {
1264 struct scrub_block *sblock = sblocks_for_recheck +
1265 mirror_index;
af8e2d1d 1266 struct scrub_recover *recover;
b5d67f64
SB		 1267 int page_index;
1268
7a9e9987
SB		 1269 for (page_index = 0; page_index < sblock->page_count;
1270 page_index++) {
1271 sblock->pagev[page_index]->sblock = NULL;
af8e2d1d
MX		 1272 recover = sblock->pagev[page_index]->recover;
1273 if (recover) {
1274 scrub_put_recover(recover);
1275 sblock->pagev[page_index]->recover =
1276 NULL;
1277 }
7a9e9987
SB		 1278 scrub_page_put(sblock->pagev[page_index]);
1279 }
b5d67f64
SB		 1280 }
1281 kfree(sblocks_for_recheck);
1282 }
a2de733c 1283
b5d67f64
SB		 1284 return 0;
1285}
a2de733c 1286
8e5cfb55 1287static inline int scrub_nr_raid_mirrors(struct btrfs_bio *bbio)
af8e2d1d 1288{
10f11900
ZL		 1289 if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID5)
1290 return 2;
1291 else if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID6)
1292 return 3;
1293 else
af8e2d1d 1294 return (int)bbio->num_stripes;
af8e2d1d
MX		 1295}
1296
10f11900
ZL		 1297static inline void scrub_stripe_index_and_offset(u64 logical, u64 map_type,
1298 u64 *raid_map,
af8e2d1d
MX		 1299 u64 mapped_length,
1300 int nstripes, int mirror,
1301 int *stripe_index,
1302 u64 *stripe_offset)
1303{
1304 int i;
1305
ffe2d203 1306 if (map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
af8e2d1d
MX		 1307 /* RAID5/6 */
1308 for (i = 0; i < nstripes; i++) {
1309 if (raid_map[i] == RAID6_Q_STRIPE ||
1310 raid_map[i] == RAID5_P_STRIPE)
1311 continue;
1312
1313 if (logical >= raid_map[i] &&
1314 logical < raid_map[i] + mapped_length)
1315 break;
1316 }
1317
1318 *stripe_index = i;
1319 *stripe_offset = logical - raid_map[i];
1320 } else {
1321 /* The other RAID type */
1322 *stripe_index = mirror;
1323 *stripe_offset = 0;
1324 }
1325}
1326
be50a8dd 1327static int scrub_setup_recheck_block(struct scrub_block *original_sblock,
b5d67f64
SB		 1328 struct scrub_block *sblocks_for_recheck)
1329{
be50a8dd
ZL		 1330 struct scrub_ctx *sctx = original_sblock->sctx;
1331 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
1332 u64 length = original_sblock->page_count * PAGE_SIZE;
1333 u64 logical = original_sblock->pagev[0]->logical;
af8e2d1d
MX		 1334 struct scrub_recover *recover;
1335 struct btrfs_bio *bbio;
af8e2d1d
MX		 1336 u64 sublen;
1337 u64 mapped_length;
1338 u64 stripe_offset;
1339 int stripe_index;
be50a8dd 1340 int page_index = 0;
b5d67f64 1341 int mirror_index;
af8e2d1d 1342 int nmirrors;
b5d67f64
SB		 1343 int ret;
1344
1345 /*
57019345 1346 * note: the two members refs and outstanding_pages
b5d67f64
SB		 1347 * are not used (and not set) in the blocks that are used for
1348 * the recheck procedure
1349 */
1350
b5d67f64 1351 while (length > 0) {
af8e2d1d
MX		 1352 sublen = min_t(u64, length, PAGE_SIZE);
1353 mapped_length = sublen;
1354 bbio = NULL;
a2de733c 1355
b5d67f64
SB		 1356 /*
1357 * with a length of PAGE_SIZE, each returned stripe
1358 * represents one mirror
1359 */
af8e2d1d 1360 ret = btrfs_map_sblock(fs_info, REQ_GET_READ_MIRRORS, logical,
8e5cfb55 1361 &mapped_length, &bbio, 0, 1);
b5d67f64 1362 if (ret || !bbio || mapped_length < sublen) {
6e9606d2 1363 btrfs_put_bbio(bbio);
b5d67f64
SB		 1364 return -EIO;
1365 }
a2de733c 1366
af8e2d1d
MX		 1367 recover = kzalloc(sizeof(struct scrub_recover), GFP_NOFS);
1368 if (!recover) {
6e9606d2 1369 btrfs_put_bbio(bbio);
af8e2d1d
MX		 1370 return -ENOMEM;
1371 }
1372
1373 atomic_set(&recover->refs, 1);
1374 recover->bbio = bbio;
af8e2d1d
MX		 1375 recover->map_length = mapped_length;
1376
ff023aac 1377 BUG_ON(page_index >= SCRUB_PAGES_PER_RD_BIO);
af8e2d1d 1378
be50a8dd 1379 nmirrors = min(scrub_nr_raid_mirrors(bbio), BTRFS_MAX_MIRRORS);
10f11900 1380
af8e2d1d 1381 for (mirror_index = 0; mirror_index < nmirrors;
b5d67f64
SB		 1382 mirror_index++) {
1383 struct scrub_block *sblock;
1384 struct scrub_page *page;
1385
b5d67f64 1386 sblock = sblocks_for_recheck + mirror_index;
7a9e9987
SB		 1387 sblock->sctx = sctx;
1388 page = kzalloc(sizeof(*page), GFP_NOFS);
1389 if (!page) {
1390leave_nomem:
d9d181c1
SB		 1391 spin_lock(&sctx->stat_lock);
1392 sctx->stat.malloc_errors++;
1393 spin_unlock(&sctx->stat_lock);
af8e2d1d 1394 scrub_put_recover(recover);
b5d67f64
SB		 1395 return -ENOMEM;
1396 }
7a9e9987
SB		 1397 scrub_page_get(page);
1398 sblock->pagev[page_index] = page;
1399 page->logical = logical;
af8e2d1d 1400
10f11900
ZL		 1401 scrub_stripe_index_and_offset(logical,
1402 bbio->map_type,
1403 bbio->raid_map,
af8e2d1d 1404 mapped_length,
e34c330d
ZL		 1405 bbio->num_stripes -
1406 bbio->num_tgtdevs,
af8e2d1d
MX		 1407 mirror_index,
1408 &stripe_index,
1409 &stripe_offset);
1410 page->physical = bbio->stripes[stripe_index].physical +
1411 stripe_offset;
1412 page->dev = bbio->stripes[stripe_index].dev;
1413
ff023aac
SB		 1414 BUG_ON(page_index >= original_sblock->page_count);
1415 page->physical_for_dev_replace =
1416 original_sblock->pagev[page_index]->
1417 physical_for_dev_replace;
7a9e9987 1418 /* for missing devices, dev->bdev is NULL */
7a9e9987 1419 page->mirror_num = mirror_index + 1;
b5d67f64 1420 sblock->page_count++;
7a9e9987
SB		 1421 page->page = alloc_page(GFP_NOFS);
1422 if (!page->page)
1423 goto leave_nomem;
af8e2d1d
MX		 1424
1425 scrub_get_recover(recover);
1426 page->recover = recover;
b5d67f64 1427 }
af8e2d1d 1428 scrub_put_recover(recover);
b5d67f64
SB		 1429 length -= sublen;
1430 logical += sublen;
1431 page_index++;
1432 }
1433
1434 return 0;
96e36920
ID		 1435}
1436
af8e2d1d
MX		 1437struct scrub_bio_ret {
1438 struct completion event;
1439 int error;
1440};
1441
1442static void scrub_bio_wait_endio(struct bio *bio, int error)
1443{
1444 struct scrub_bio_ret *ret = bio->bi_private;
1445
1446 ret->error = error;
1447 complete(&ret->event);
1448}
1449
1450static inline int scrub_is_page_on_raid56(struct scrub_page *page)
1451{
10f11900 1452 return page->recover &&
ffe2d203 1453 (page->recover->bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK);
af8e2d1d
MX		 1454}
1455
1456static int scrub_submit_raid56_bio_wait(struct btrfs_fs_info *fs_info,
1457 struct bio *bio,
1458 struct scrub_page *page)
1459{
1460 struct scrub_bio_ret done;
1461 int ret;
1462
1463 init_completion(&done.event);
1464 done.error = 0;
1465 bio->bi_iter.bi_sector = page->logical >> 9;
1466 bio->bi_private = &done;
1467 bio->bi_end_io = scrub_bio_wait_endio;
1468
1469 ret = raid56_parity_recover(fs_info->fs_root, bio, page->recover->bbio,
af8e2d1d 1470 page->recover->map_length,
4245215d 1471 page->mirror_num, 0);
af8e2d1d
MX		 1472 if (ret)
1473 return ret;
1474
1475 wait_for_completion(&done.event);
1476 if (done.error)
1477 return -EIO;
1478
1479 return 0;
1480}
1481
b5d67f64
SB		 1482/*
1483 * this function will check the on disk data for checksum errors, header
1484 * errors and read I/O errors. If any I/O errors happen, the exact pages
1485 * which are errored are marked as being bad. The goal is to enable scrub
1486 * to take those pages that are not errored from all the mirrors so that
1487 * the pages that are errored in the just handled mirror can be repaired.
1488 */
34f5c8e9
SB		 1489static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
1490 struct scrub_block *sblock, int is_metadata,
1491 int have_csum, u8 *csum, u64 generation,
af8e2d1d 1492 u16 csum_size, int retry_failed_mirror)
96e36920 1493{
b5d67f64 1494 int page_num;
96e36920 1495
b5d67f64
SB		 1496 sblock->no_io_error_seen = 1;
1497 sblock->header_error = 0;
1498 sblock->checksum_error = 0;
96e36920 1499
b5d67f64
SB		 1500 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1501 struct bio *bio;
7a9e9987 1502 struct scrub_page *page = sblock->pagev[page_num];
b5d67f64 1503
442a4f63 1504 if (page->dev->bdev == NULL) {
ea9947b4
SB		 1505 page->io_error = 1;
1506 sblock->no_io_error_seen = 0;
1507 continue;
1508 }
1509
7a9e9987 1510 WARN_ON(!page->page);
9be3395b 1511 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
34f5c8e9
SB		 1512 if (!bio) {
1513 page->io_error = 1;
1514 sblock->no_io_error_seen = 0;
1515 continue;
1516 }
442a4f63 1517 bio->bi_bdev = page->dev->bdev;
b5d67f64 1518
34f5c8e9 1519 bio_add_page(bio, page->page, PAGE_SIZE, 0);
af8e2d1d
MX		 1520 if (!retry_failed_mirror && scrub_is_page_on_raid56(page)) {
1521 if (scrub_submit_raid56_bio_wait(fs_info, bio, page))
1522 sblock->no_io_error_seen = 0;
1523 } else {
1524 bio->bi_iter.bi_sector = page->physical >> 9;
1525
1526 if (btrfsic_submit_bio_wait(READ, bio))
1527 sblock->no_io_error_seen = 0;
1528 }
33879d45 1529
b5d67f64
SB		 1530 bio_put(bio);
1531 }
96e36920 1532
b5d67f64
SB		 1533 if (sblock->no_io_error_seen)
1534 scrub_recheck_block_checksum(fs_info, sblock, is_metadata,
1535 have_csum, csum, generation,
1536 csum_size);
1537
34f5c8e9 1538 return;
a2de733c
AJ		 1539}
1540
17a9be2f
MX		 1541static inline int scrub_check_fsid(u8 fsid[],
1542 struct scrub_page *spage)
1543{
1544 struct btrfs_fs_devices *fs_devices = spage->dev->fs_devices;
1545 int ret;
1546
1547 ret = memcmp(fsid, fs_devices->fsid, BTRFS_UUID_SIZE);
1548 return !ret;
1549}
1550
b5d67f64
SB		 1551static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
1552 struct scrub_block *sblock,
1553 int is_metadata, int have_csum,
1554 const u8 *csum, u64 generation,
1555 u16 csum_size)
a2de733c 1556{
b5d67f64
SB		 1557 int page_num;
1558 u8 calculated_csum[BTRFS_CSUM_SIZE];
1559 u32 crc = ~(u32)0;
b5d67f64
SB		 1560 void *mapped_buffer;
1561
7a9e9987 1562 WARN_ON(!sblock->pagev[0]->page);
b5d67f64
SB		 1563 if (is_metadata) {
1564 struct btrfs_header *h;
1565
7a9e9987 1566 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
b5d67f64
SB		 1567 h = (struct btrfs_header *)mapped_buffer;
1568
3cae210f 1569 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h) ||
17a9be2f 1570 !scrub_check_fsid(h->fsid, sblock->pagev[0]) ||
b5d67f64 1571 memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
442a4f63 1572 BTRFS_UUID_SIZE)) {
b5d67f64 1573 sblock->header_error = 1;
3cae210f 1574 } else if (generation != btrfs_stack_header_generation(h)) {
442a4f63
SB		 1575 sblock->header_error = 1;
1576 sblock->generation_error = 1;
1577 }
b5d67f64
SB		 1578 csum = h->csum;
1579 } else {
1580 if (!have_csum)
1581 return;
a2de733c 1582
7a9e9987 1583 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
b5d67f64 1584 }
a2de733c 1585
b5d67f64
SB		 1586 for (page_num = 0;;) {
1587 if (page_num == 0 && is_metadata)
b0496686 1588 crc = btrfs_csum_data(
b5d67f64
SB		 1589 ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE,
1590 crc, PAGE_SIZE - BTRFS_CSUM_SIZE);
1591 else
b0496686 1592 crc = btrfs_csum_data(mapped_buffer, crc, PAGE_SIZE);
b5d67f64 1593
9613bebb 1594 kunmap_atomic(mapped_buffer);
b5d67f64
SB		 1595 page_num++;
1596 if (page_num >= sblock->page_count)
1597 break;
7a9e9987 1598 WARN_ON(!sblock->pagev[page_num]->page);
b5d67f64 1599
7a9e9987 1600 mapped_buffer = kmap_atomic(sblock->pagev[page_num]->page);
b5d67f64
SB		 1601 }
1602
1603 btrfs_csum_final(crc, calculated_csum);
1604 if (memcmp(calculated_csum, csum, csum_size))
1605 sblock->checksum_error = 1;
a2de733c
AJ		 1606}
1607
b5d67f64 1608static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
114ab50d 1609 struct scrub_block *sblock_good)
b5d67f64
SB		 1610{
1611 int page_num;
1612 int ret = 0;
96e36920 1613
b5d67f64
SB		 1614 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
1615 int ret_sub;
96e36920 1616
b5d67f64
SB		 1617 ret_sub = scrub_repair_page_from_good_copy(sblock_bad,
1618 sblock_good,
114ab50d 1619 page_num, 1);
b5d67f64
SB		 1620 if (ret_sub)
1621 ret = ret_sub;
a2de733c 1622 }
b5d67f64
SB		 1623
1624 return ret;
1625}
1626
1627static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
1628 struct scrub_block *sblock_good,
1629 int page_num, int force_write)
1630{
7a9e9987
SB		 1631 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1632 struct scrub_page *page_good = sblock_good->pagev[page_num];
b5d67f64 1633
7a9e9987
SB		 1634 BUG_ON(page_bad->page == NULL);
1635 BUG_ON(page_good->page == NULL);
b5d67f64
SB		 1636 if (force_write || sblock_bad->header_error ||
1637 sblock_bad->checksum_error || page_bad->io_error) {
1638 struct bio *bio;
1639 int ret;
b5d67f64 1640
ff023aac 1641 if (!page_bad->dev->bdev) {
efe120a0
FH		 1642 printk_ratelimited(KERN_WARNING "BTRFS: "
1643 "scrub_repair_page_from_good_copy(bdev == NULL) "
1644 "is unexpected!\n");
ff023aac
SB		 1645 return -EIO;
1646 }
1647
9be3395b 1648 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
e627ee7b
TI		 1649 if (!bio)
1650 return -EIO;
442a4f63 1651 bio->bi_bdev = page_bad->dev->bdev;
4f024f37 1652 bio->bi_iter.bi_sector = page_bad->physical >> 9;
b5d67f64
SB		 1653
1654 ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0);
1655 if (PAGE_SIZE != ret) {
1656 bio_put(bio);
1657 return -EIO;
13db62b7 1658 }
b5d67f64 1659
33879d45 1660 if (btrfsic_submit_bio_wait(WRITE, bio)) {
442a4f63
SB		 1661 btrfs_dev_stat_inc_and_print(page_bad->dev,
1662 BTRFS_DEV_STAT_WRITE_ERRS);
ff023aac
SB		 1663 btrfs_dev_replace_stats_inc(
1664 &sblock_bad->sctx->dev_root->fs_info->
1665 dev_replace.num_write_errors);
442a4f63
SB		 1666 bio_put(bio);
1667 return -EIO;
1668 }
b5d67f64 1669 bio_put(bio);
a2de733c
AJ		 1670 }
1671
b5d67f64
SB		 1672 return 0;
1673}
1674
ff023aac
SB		 1675static void scrub_write_block_to_dev_replace(struct scrub_block *sblock)
1676{
1677 int page_num;
1678
5a6ac9ea
MX		 1679 /*
1680 * This block is used for the check of the parity on the source device,
1681 * so the data needn't be written into the destination device.
1682 */
1683 if (sblock->sparity)
1684 return;
1685
ff023aac
SB		 1686 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1687 int ret;
1688
1689 ret = scrub_write_page_to_dev_replace(sblock, page_num);
1690 if (ret)
1691 btrfs_dev_replace_stats_inc(
1692 &sblock->sctx->dev_root->fs_info->dev_replace.
1693 num_write_errors);
1694 }
1695}
1696
1697static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
1698 int page_num)
1699{
1700 struct scrub_page *spage = sblock->pagev[page_num];
1701
1702 BUG_ON(spage->page == NULL);
1703 if (spage->io_error) {
1704 void *mapped_buffer = kmap_atomic(spage->page);
1705
1706 memset(mapped_buffer, 0, PAGE_CACHE_SIZE);
1707 flush_dcache_page(spage->page);
1708 kunmap_atomic(mapped_buffer);
1709 }
1710 return scrub_add_page_to_wr_bio(sblock->sctx, spage);
1711}
1712
1713static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
1714 struct scrub_page *spage)
1715{
1716 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1717 struct scrub_bio *sbio;
1718 int ret;
1719
1720 mutex_lock(&wr_ctx->wr_lock);
1721again:
1722 if (!wr_ctx->wr_curr_bio) {
1723 wr_ctx->wr_curr_bio = kzalloc(sizeof(*wr_ctx->wr_curr_bio),
1724 GFP_NOFS);
1725 if (!wr_ctx->wr_curr_bio) {
1726 mutex_unlock(&wr_ctx->wr_lock);
1727 return -ENOMEM;
1728 }
1729 wr_ctx->wr_curr_bio->sctx = sctx;
1730 wr_ctx->wr_curr_bio->page_count = 0;
1731 }
1732 sbio = wr_ctx->wr_curr_bio;
1733 if (sbio->page_count == 0) {
1734 struct bio *bio;
1735
1736 sbio->physical = spage->physical_for_dev_replace;
1737 sbio->logical = spage->logical;
1738 sbio->dev = wr_ctx->tgtdev;
1739 bio = sbio->bio;
1740 if (!bio) {
9be3395b 1741 bio = btrfs_io_bio_alloc(GFP_NOFS, wr_ctx->pages_per_wr_bio);
ff023aac
SB		 1742 if (!bio) {
1743 mutex_unlock(&wr_ctx->wr_lock);
1744 return -ENOMEM;
1745 }
1746 sbio->bio = bio;
1747 }
1748
1749 bio->bi_private = sbio;
1750 bio->bi_end_io = scrub_wr_bio_end_io;
1751 bio->bi_bdev = sbio->dev->bdev;
4f024f37 1752 bio->bi_iter.bi_sector = sbio->physical >> 9;
ff023aac
SB		 1753 sbio->err = 0;
1754 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
1755 spage->physical_for_dev_replace ||
1756 sbio->logical + sbio->page_count * PAGE_SIZE !=
1757 spage->logical) {
1758 scrub_wr_submit(sctx);
1759 goto again;
1760 }
1761
1762 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
1763 if (ret != PAGE_SIZE) {
1764 if (sbio->page_count < 1) {
1765 bio_put(sbio->bio);
1766 sbio->bio = NULL;
1767 mutex_unlock(&wr_ctx->wr_lock);
1768 return -EIO;
1769 }
1770 scrub_wr_submit(sctx);
1771 goto again;
1772 }
1773
1774 sbio->pagev[sbio->page_count] = spage;
1775 scrub_page_get(spage);
1776 sbio->page_count++;
1777 if (sbio->page_count == wr_ctx->pages_per_wr_bio)
1778 scrub_wr_submit(sctx);
1779 mutex_unlock(&wr_ctx->wr_lock);
1780
1781 return 0;
1782}
1783
1784static void scrub_wr_submit(struct scrub_ctx *sctx)
1785{
1786 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1787 struct scrub_bio *sbio;
1788
1789 if (!wr_ctx->wr_curr_bio)
1790 return;
1791
1792 sbio = wr_ctx->wr_curr_bio;
1793 wr_ctx->wr_curr_bio = NULL;
1794 WARN_ON(!sbio->bio->bi_bdev);
1795 scrub_pending_bio_inc(sctx);
1796 /* process all writes in a single worker thread. Then the block layer
1797 * orders the requests before sending them to the driver which
1798 * doubled the write performance on spinning disks when measured
1799 * with Linux 3.5 */
1800 btrfsic_submit_bio(WRITE, sbio->bio);
1801}
1802
1803static void scrub_wr_bio_end_io(struct bio *bio, int err)
1804{
1805 struct scrub_bio *sbio = bio->bi_private;
1806 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
1807
1808 sbio->err = err;
1809 sbio->bio = bio;
1810
9e0af237
LB		 1811 btrfs_init_work(&sbio->work, btrfs_scrubwrc_helper,
1812 scrub_wr_bio_end_io_worker, NULL, NULL);
0339ef2f 1813 btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work);
ff023aac
SB		 1814}
1815
1816static void scrub_wr_bio_end_io_worker(struct btrfs_work *work)
1817{
1818 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
1819 struct scrub_ctx *sctx = sbio->sctx;
1820 int i;
1821
1822 WARN_ON(sbio->page_count > SCRUB_PAGES_PER_WR_BIO);
1823 if (sbio->err) {
1824 struct btrfs_dev_replace *dev_replace =
1825 &sbio->sctx->dev_root->fs_info->dev_replace;
1826
1827 for (i = 0; i < sbio->page_count; i++) {
1828 struct scrub_page *spage = sbio->pagev[i];
1829
1830 spage->io_error = 1;
1831 btrfs_dev_replace_stats_inc(&dev_replace->
1832 num_write_errors);
1833 }
1834 }
1835
1836 for (i = 0; i < sbio->page_count; i++)
1837 scrub_page_put(sbio->pagev[i]);
1838
1839 bio_put(sbio->bio);
1840 kfree(sbio);
1841 scrub_pending_bio_dec(sctx);
1842}
1843
1844static int scrub_checksum(struct scrub_block *sblock)
b5d67f64
SB		 1845{
1846 u64 flags;
1847 int ret;
1848
7a9e9987
SB		 1849 WARN_ON(sblock->page_count < 1);
1850 flags = sblock->pagev[0]->flags;
b5d67f64
SB		 1851 ret = 0;
1852 if (flags & BTRFS_EXTENT_FLAG_DATA)
1853 ret = scrub_checksum_data(sblock);
1854 else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1855 ret = scrub_checksum_tree_block(sblock);
1856 else if (flags & BTRFS_EXTENT_FLAG_SUPER)
1857 (void)scrub_checksum_super(sblock);
1858 else
1859 WARN_ON(1);
1860 if (ret)
1861 scrub_handle_errored_block(sblock);
ff023aac
SB		 1862
1863 return ret;
a2de733c
AJ		 1864}
1865
b5d67f64 1866static int scrub_checksum_data(struct scrub_block *sblock)
a2de733c 1867{
d9d181c1 1868 struct scrub_ctx *sctx = sblock->sctx;
a2de733c 1869 u8 csum[BTRFS_CSUM_SIZE];
b5d67f64
SB		 1870 u8 *on_disk_csum;
1871 struct page *page;
1872 void *buffer;
a2de733c
AJ		 1873 u32 crc = ~(u32)0;
1874 int fail = 0;
b5d67f64
SB		 1875 u64 len;
1876 int index;
a2de733c 1877
b5d67f64 1878 BUG_ON(sblock->page_count < 1);
7a9e9987 1879 if (!sblock->pagev[0]->have_csum)
a2de733c
AJ		 1880 return 0;
1881
7a9e9987
SB		 1882 on_disk_csum = sblock->pagev[0]->csum;
1883 page = sblock->pagev[0]->page;
9613bebb 1884 buffer = kmap_atomic(page);
b5d67f64 1885
d9d181c1 1886 len = sctx->sectorsize;
b5d67f64
SB		 1887 index = 0;
1888 for (;;) {
1889 u64 l = min_t(u64, len, PAGE_SIZE);
1890
b0496686 1891 crc = btrfs_csum_data(buffer, crc, l);
9613bebb 1892 kunmap_atomic(buffer);
b5d67f64
SB		 1893 len -= l;
1894 if (len == 0)
1895 break;
1896 index++;
1897 BUG_ON(index >= sblock->page_count);
7a9e9987
SB		 1898 BUG_ON(!sblock->pagev[index]->page);
1899 page = sblock->pagev[index]->page;
9613bebb 1900 buffer = kmap_atomic(page);
b5d67f64
SB		 1901 }
1902
a2de733c 1903 btrfs_csum_final(crc, csum);
d9d181c1 1904 if (memcmp(csum, on_disk_csum, sctx->csum_size))
a2de733c
AJ		 1905 fail = 1;
1906
a2de733c
AJ		 1907 return fail;
1908}
1909
b5d67f64 1910static int scrub_checksum_tree_block(struct scrub_block *sblock)
a2de733c 1911{
d9d181c1 1912 struct scrub_ctx *sctx = sblock->sctx;
a2de733c 1913 struct btrfs_header *h;
a36cf8b8 1914 struct btrfs_root *root = sctx->dev_root;
a2de733c 1915 struct btrfs_fs_info *fs_info = root->fs_info;
b5d67f64
SB		 1916 u8 calculated_csum[BTRFS_CSUM_SIZE];
1917 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1918 struct page *page;
1919 void *mapped_buffer;
1920 u64 mapped_size;
1921 void *p;
a2de733c
AJ		 1922 u32 crc = ~(u32)0;
1923 int fail = 0;
1924 int crc_fail = 0;
b5d67f64
SB		 1925 u64 len;
1926 int index;
1927
1928 BUG_ON(sblock->page_count < 1);
7a9e9987 1929 page = sblock->pagev[0]->page;
9613bebb 1930 mapped_buffer = kmap_atomic(page);
b5d67f64 1931 h = (struct btrfs_header *)mapped_buffer;
d9d181c1 1932 memcpy(on_disk_csum, h->csum, sctx->csum_size);
a2de733c
AJ		 1933
1934 /*
1935 * we don't use the getter functions here, as we
1936 * a) don't have an extent buffer and
1937 * b) the page is already kmapped
1938 */
a2de733c 1939
3cae210f 1940 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h))
a2de733c
AJ		 1941 ++fail;
1942
3cae210f 1943 if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h))
a2de733c
AJ		 1944 ++fail;
1945
17a9be2f 1946 if (!scrub_check_fsid(h->fsid, sblock->pagev[0]))
a2de733c
AJ		 1947 ++fail;
1948
1949 if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
1950 BTRFS_UUID_SIZE))
1951 ++fail;
1952
d9d181c1 1953 len = sctx->nodesize - BTRFS_CSUM_SIZE;
b5d67f64
SB		 1954 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
1955 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
1956 index = 0;
1957 for (;;) {
1958 u64 l = min_t(u64, len, mapped_size);
1959
b0496686 1960 crc = btrfs_csum_data(p, crc, l);
9613bebb 1961 kunmap_atomic(mapped_buffer);
b5d67f64
SB		 1962 len -= l;
1963 if (len == 0)
1964 break;
1965 index++;
1966 BUG_ON(index >= sblock->page_count);
7a9e9987
SB		 1967 BUG_ON(!sblock->pagev[index]->page);
1968 page = sblock->pagev[index]->page;
9613bebb 1969 mapped_buffer = kmap_atomic(page);
b5d67f64
SB		 1970 mapped_size = PAGE_SIZE;
1971 p = mapped_buffer;
1972 }
1973
1974 btrfs_csum_final(crc, calculated_csum);
d9d181c1 1975 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
a2de733c
AJ		 1976 ++crc_fail;
1977
a2de733c
AJ		 1978 return fail || crc_fail;
1979}
1980
b5d67f64 1981static int scrub_checksum_super(struct scrub_block *sblock)
a2de733c
AJ		 1982{
1983 struct btrfs_super_block *s;
d9d181c1 1984 struct scrub_ctx *sctx = sblock->sctx;
b5d67f64
SB		 1985 u8 calculated_csum[BTRFS_CSUM_SIZE];
1986 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1987 struct page *page;
1988 void *mapped_buffer;
1989 u64 mapped_size;
1990 void *p;
a2de733c 1991 u32 crc = ~(u32)0;
442a4f63
SB		 1992 int fail_gen = 0;
1993 int fail_cor = 0;
b5d67f64
SB		 1994 u64 len;
1995 int index;
a2de733c 1996
b5d67f64 1997 BUG_ON(sblock->page_count < 1);
7a9e9987 1998 page = sblock->pagev[0]->page;
9613bebb 1999 mapped_buffer = kmap_atomic(page);
b5d67f64 2000 s = (struct btrfs_super_block *)mapped_buffer;
d9d181c1 2001 memcpy(on_disk_csum, s->csum, sctx->csum_size);
a2de733c 2002
3cae210f 2003 if (sblock->pagev[0]->logical != btrfs_super_bytenr(s))
442a4f63 2004 ++fail_cor;
a2de733c 2005
3cae210f 2006 if (sblock->pagev[0]->generation != btrfs_super_generation(s))
442a4f63 2007 ++fail_gen;
a2de733c 2008
17a9be2f 2009 if (!scrub_check_fsid(s->fsid, sblock->pagev[0]))
442a4f63 2010 ++fail_cor;
a2de733c 2011
b5d67f64
SB		 2012 len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE;
2013 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
2014 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
2015 index = 0;
2016 for (;;) {
2017 u64 l = min_t(u64, len, mapped_size);
2018
b0496686 2019 crc = btrfs_csum_data(p, crc, l);
9613bebb 2020 kunmap_atomic(mapped_buffer);
b5d67f64
SB		 2021 len -= l;
2022 if (len == 0)
2023 break;
2024 index++;
2025 BUG_ON(index >= sblock->page_count);
7a9e9987
SB		 2026 BUG_ON(!sblock->pagev[index]->page);
2027 page = sblock->pagev[index]->page;
9613bebb 2028 mapped_buffer = kmap_atomic(page);
b5d67f64
SB		 2029 mapped_size = PAGE_SIZE;
2030 p = mapped_buffer;
2031 }
2032
2033 btrfs_csum_final(crc, calculated_csum);
d9d181c1 2034 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
442a4f63 2035 ++fail_cor;
a2de733c 2036
442a4f63 2037 if (fail_cor + fail_gen) {
a2de733c
AJ		 2038 /*
2039 * if we find an error in a super block, we just report it.
2040 * They will get written with the next transaction commit
2041 * anyway
2042 */
d9d181c1
SB		 2043 spin_lock(&sctx->stat_lock);
2044 ++sctx->stat.super_errors;
2045 spin_unlock(&sctx->stat_lock);
442a4f63 2046 if (fail_cor)
7a9e9987 2047 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
442a4f63
SB		 2048 BTRFS_DEV_STAT_CORRUPTION_ERRS);
2049 else
7a9e9987 2050 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
442a4f63 2051 BTRFS_DEV_STAT_GENERATION_ERRS);
a2de733c
AJ		 2052 }
2053
442a4f63 2054 return fail_cor + fail_gen;
a2de733c
AJ		 2055}
2056
b5d67f64
SB		 2057static void scrub_block_get(struct scrub_block *sblock)
2058{
57019345 2059 atomic_inc(&sblock->refs);
b5d67f64
SB		 2060}
2061
2062static void scrub_block_put(struct scrub_block *sblock)
2063{
57019345 2064 if (atomic_dec_and_test(&sblock->refs)) {
b5d67f64
SB		 2065 int i;
2066
5a6ac9ea
MX		 2067 if (sblock->sparity)
2068 scrub_parity_put(sblock->sparity);
2069
b5d67f64 2070 for (i = 0; i < sblock->page_count; i++)
7a9e9987 2071 scrub_page_put(sblock->pagev[i]);
b5d67f64
SB		 2072 kfree(sblock);
2073 }
2074}
2075
7a9e9987
SB		 2076static void scrub_page_get(struct scrub_page *spage)
2077{
57019345 2078 atomic_inc(&spage->refs);
7a9e9987
SB		 2079}
2080
2081static void scrub_page_put(struct scrub_page *spage)
2082{
57019345 2083 if (atomic_dec_and_test(&spage->refs)) {
7a9e9987
SB		 2084 if (spage->page)
2085 __free_page(spage->page);
2086 kfree(spage);
2087 }
2088}
2089
d9d181c1 2090static void scrub_submit(struct scrub_ctx *sctx)
a2de733c
AJ		 2091{
2092 struct scrub_bio *sbio;
2093
d9d181c1 2094 if (sctx->curr == -1)
1623edeb 2095 return;
a2de733c 2096
d9d181c1
SB		 2097 sbio = sctx->bios[sctx->curr];
2098 sctx->curr = -1;
b6bfebc1 2099 scrub_pending_bio_inc(sctx);
03679ade 2100 btrfsic_submit_bio(READ, sbio->bio);
a2de733c
AJ		 2101}
2102
ff023aac
SB		 2103static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
2104 struct scrub_page *spage)
a2de733c 2105{
b5d67f64 2106 struct scrub_block *sblock = spage->sblock;
a2de733c 2107 struct scrub_bio *sbio;
69f4cb52 2108 int ret;
a2de733c
AJ		 2109
2110again:
2111 /*
2112 * grab a fresh bio or wait for one to become available
2113 */
d9d181c1
SB		 2114 while (sctx->curr == -1) {
2115 spin_lock(&sctx->list_lock);
2116 sctx->curr = sctx->first_free;
2117 if (sctx->curr != -1) {
2118 sctx->first_free = sctx->bios[sctx->curr]->next_free;
2119 sctx->bios[sctx->curr]->next_free = -1;
2120 sctx->bios[sctx->curr]->page_count = 0;
2121 spin_unlock(&sctx->list_lock);
a2de733c 2122 } else {
d9d181c1
SB		 2123 spin_unlock(&sctx->list_lock);
2124 wait_event(sctx->list_wait, sctx->first_free != -1);
a2de733c
AJ		 2125 }
2126 }
d9d181c1 2127 sbio = sctx->bios[sctx->curr];
b5d67f64 2128 if (sbio->page_count == 0) {
69f4cb52
AJ		 2129 struct bio *bio;
2130
b5d67f64
SB		 2131 sbio->physical = spage->physical;
2132 sbio->logical = spage->logical;
a36cf8b8 2133 sbio->dev = spage->dev;
b5d67f64
SB		 2134 bio = sbio->bio;
2135 if (!bio) {
9be3395b 2136 bio = btrfs_io_bio_alloc(GFP_NOFS, sctx->pages_per_rd_bio);
b5d67f64
SB		 2137 if (!bio)
2138 return -ENOMEM;
2139 sbio->bio = bio;
2140 }
69f4cb52
AJ		 2141
2142 bio->bi_private = sbio;
2143 bio->bi_end_io = scrub_bio_end_io;
a36cf8b8 2144 bio->bi_bdev = sbio->dev->bdev;
4f024f37 2145 bio->bi_iter.bi_sector = sbio->physical >> 9;
69f4cb52 2146 sbio->err = 0;
b5d67f64
SB		 2147 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
2148 spage->physical ||
2149 sbio->logical + sbio->page_count * PAGE_SIZE !=
a36cf8b8
SB		 2150 spage->logical ||
2151 sbio->dev != spage->dev) {
d9d181c1 2152 scrub_submit(sctx);
a2de733c
AJ		 2153 goto again;
2154 }
69f4cb52 2155
b5d67f64
SB		 2156 sbio->pagev[sbio->page_count] = spage;
2157 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
2158 if (ret != PAGE_SIZE) {
2159 if (sbio->page_count < 1) {
2160 bio_put(sbio->bio);
2161 sbio->bio = NULL;
2162 return -EIO;
2163 }
d9d181c1 2164 scrub_submit(sctx);
69f4cb52
AJ		 2165 goto again;
2166 }
2167
ff023aac 2168 scrub_block_get(sblock); /* one for the page added to the bio */
b5d67f64
SB		 2169 atomic_inc(&sblock->outstanding_pages);
2170 sbio->page_count++;
ff023aac 2171 if (sbio->page_count == sctx->pages_per_rd_bio)
d9d181c1 2172 scrub_submit(sctx);
b5d67f64
SB		 2173
2174 return 0;
2175}
2176
73ff61db
OS		 2177static void scrub_missing_raid56_end_io(struct bio *bio, int error)
2178{
2179 struct scrub_block *sblock = bio->bi_private;
2180 struct btrfs_fs_info *fs_info = sblock->sctx->dev_root->fs_info;
2181
2182 if (error)
2183 sblock->no_io_error_seen = 0;
2184
2185 btrfs_queue_work(fs_info->scrub_workers, &sblock->work);
2186}
2187
2188static void scrub_missing_raid56_worker(struct btrfs_work *work)
2189{
2190 struct scrub_block *sblock = container_of(work, struct scrub_block, work);
2191 struct scrub_ctx *sctx = sblock->sctx;
2192 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
2193 unsigned int is_metadata;
2194 unsigned int have_csum;
2195 u8 *csum;
2196 u64 generation;
2197 u64 logical;
2198 struct btrfs_device *dev;
2199
2200 is_metadata = !(sblock->pagev[0]->flags & BTRFS_EXTENT_FLAG_DATA);
2201 have_csum = sblock->pagev[0]->have_csum;
2202 csum = sblock->pagev[0]->csum;
2203 generation = sblock->pagev[0]->generation;
2204 logical = sblock->pagev[0]->logical;
2205 dev = sblock->pagev[0]->dev;
2206
2207 if (sblock->no_io_error_seen) {
2208 scrub_recheck_block_checksum(fs_info, sblock, is_metadata,
2209 have_csum, csum, generation,
2210 sctx->csum_size);
2211 }
2212
2213 if (!sblock->no_io_error_seen) {
2214 spin_lock(&sctx->stat_lock);
2215 sctx->stat.read_errors++;
2216 spin_unlock(&sctx->stat_lock);
2217 printk_ratelimited_in_rcu(KERN_ERR
2218 "BTRFS: I/O error rebulding logical %llu for dev %s\n",
2219 logical, rcu_str_deref(dev->name));
2220 } else if (sblock->header_error || sblock->checksum_error) {
2221 spin_lock(&sctx->stat_lock);
2222 sctx->stat.uncorrectable_errors++;
2223 spin_unlock(&sctx->stat_lock);
2224 printk_ratelimited_in_rcu(KERN_ERR
2225 "BTRFS: failed to rebuild valid logical %llu for dev %s\n",
2226 logical, rcu_str_deref(dev->name));
2227 } else {
2228 scrub_write_block_to_dev_replace(sblock);
2229 }
2230
2231 scrub_block_put(sblock);
2232
2233 if (sctx->is_dev_replace &&
2234 atomic_read(&sctx->wr_ctx.flush_all_writes)) {
2235 mutex_lock(&sctx->wr_ctx.wr_lock);
2236 scrub_wr_submit(sctx);
2237 mutex_unlock(&sctx->wr_ctx.wr_lock);
2238 }
2239
2240 scrub_pending_bio_dec(sctx);
2241}
2242
2243static void scrub_missing_raid56_pages(struct scrub_block *sblock)
2244{
2245 struct scrub_ctx *sctx = sblock->sctx;
2246 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
2247 u64 length = sblock->page_count * PAGE_SIZE;
2248 u64 logical = sblock->pagev[0]->logical;
2249 struct btrfs_bio *bbio;
2250 struct bio *bio;
2251 struct btrfs_raid_bio *rbio;
2252 int ret;
2253 int i;
2254
2255 ret = btrfs_map_sblock(fs_info, REQ_GET_READ_MIRRORS, logical, &length,
2256 &bbio, 0, 1);
2257 if (ret || !bbio || !bbio->raid_map)
2258 goto bbio_out;
2259
2260 if (WARN_ON(!sctx->is_dev_replace ||
2261 !(bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK))) {
2262 /*
2263 * We shouldn't be scrubbing a missing device. Even for dev
2264 * replace, we should only get here for RAID 5/6. We either
2265 * managed to mount something with no mirrors remaining or
2266 * there's a bug in scrub_remap_extent()/btrfs_map_block().
2267 */
2268 goto bbio_out;
2269 }
2270
2271 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
2272 if (!bio)
2273 goto bbio_out;
2274
2275 bio->bi_iter.bi_sector = logical >> 9;
2276 bio->bi_private = sblock;
2277 bio->bi_end_io = scrub_missing_raid56_end_io;
2278
2279 rbio = raid56_alloc_missing_rbio(sctx->dev_root, bio, bbio, length);
2280 if (!rbio)
2281 goto rbio_out;
2282
2283 for (i = 0; i < sblock->page_count; i++) {
2284 struct scrub_page *spage = sblock->pagev[i];
2285
2286 raid56_add_scrub_pages(rbio, spage->page, spage->logical);
2287 }
2288
2289 btrfs_init_work(&sblock->work, btrfs_scrub_helper,
2290 scrub_missing_raid56_worker, NULL, NULL);
2291 scrub_block_get(sblock);
2292 scrub_pending_bio_inc(sctx);
2293 raid56_submit_missing_rbio(rbio);
2294 return;
2295
2296rbio_out:
2297 bio_put(bio);
2298bbio_out:
2299 btrfs_put_bbio(bbio);
2300 spin_lock(&sctx->stat_lock);
2301 sctx->stat.malloc_errors++;
2302 spin_unlock(&sctx->stat_lock);
2303}
2304
d9d181c1 2305static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
a36cf8b8 2306 u64 physical, struct btrfs_device *dev, u64 flags,
ff023aac
SB		 2307 u64 gen, int mirror_num, u8 *csum, int force,
2308 u64 physical_for_dev_replace)
b5d67f64
SB		 2309{
2310 struct scrub_block *sblock;
2311 int index;
2312
2313 sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
2314 if (!sblock) {
d9d181c1
SB		 2315 spin_lock(&sctx->stat_lock);
2316 sctx->stat.malloc_errors++;
2317 spin_unlock(&sctx->stat_lock);
b5d67f64 2318 return -ENOMEM;
a2de733c 2319 }
b5d67f64 2320
7a9e9987
SB		 2321 /* one ref inside this function, plus one for each page added to
2322 * a bio later on */
57019345 2323 atomic_set(&sblock->refs, 1);
d9d181c1 2324 sblock->sctx = sctx;
b5d67f64
SB		 2325 sblock->no_io_error_seen = 1;
2326
2327 for (index = 0; len > 0; index++) {
7a9e9987 2328 struct scrub_page *spage;
b5d67f64
SB		 2329 u64 l = min_t(u64, len, PAGE_SIZE);
2330
7a9e9987
SB		 2331 spage = kzalloc(sizeof(*spage), GFP_NOFS);
2332 if (!spage) {
2333leave_nomem:
d9d181c1
SB		 2334 spin_lock(&sctx->stat_lock);
2335 sctx->stat.malloc_errors++;
2336 spin_unlock(&sctx->stat_lock);
7a9e9987 2337 scrub_block_put(sblock);
b5d67f64
SB		 2338 return -ENOMEM;
2339 }
7a9e9987
SB		 2340 BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
2341 scrub_page_get(spage);
2342 sblock->pagev[index] = spage;
b5d67f64 2343 spage->sblock = sblock;
a36cf8b8 2344 spage->dev = dev;
b5d67f64
SB		 2345 spage->flags = flags;
2346 spage->generation = gen;
2347 spage->logical = logical;
2348 spage->physical = physical;
ff023aac 2349 spage->physical_for_dev_replace = physical_for_dev_replace;
b5d67f64
SB		 2350 spage->mirror_num = mirror_num;
2351 if (csum) {
2352 spage->have_csum = 1;
d9d181c1 2353 memcpy(spage->csum, csum, sctx->csum_size);
b5d67f64
SB		 2354 } else {
2355 spage->have_csum = 0;
2356 }
2357 sblock->page_count++;
7a9e9987
SB		 2358 spage->page = alloc_page(GFP_NOFS);
2359 if (!spage->page)
2360 goto leave_nomem;
b5d67f64
SB		 2361 len -= l;
2362 logical += l;
2363 physical += l;
ff023aac 2364 physical_for_dev_replace += l;
b5d67f64
SB		 2365 }
2366
7a9e9987 2367 WARN_ON(sblock->page_count == 0);
73ff61db
OS		 2368 if (dev->missing) {
2369 /*
2370 * This case should only be hit for RAID 5/6 device replace. See
2371 * the comment in scrub_missing_raid56_pages() for details.
2372 */
2373 scrub_missing_raid56_pages(sblock);
2374 } else {
2375 for (index = 0; index < sblock->page_count; index++) {
2376 struct scrub_page *spage = sblock->pagev[index];
2377 int ret;
2378
2379 ret = scrub_add_page_to_rd_bio(sctx, spage);
2380 if (ret) {
2381 scrub_block_put(sblock);
2382 return ret;
2383 }
b5d67f64 2384 }
a2de733c 2385
73ff61db
OS		 2386 if (force)
2387 scrub_submit(sctx);
2388 }
a2de733c 2389
b5d67f64
SB		 2390 /* last one frees, either here or in bio completion for last page */
2391 scrub_block_put(sblock);
a2de733c
AJ		 2392 return 0;
2393}
2394
b5d67f64
SB		 2395static void scrub_bio_end_io(struct bio *bio, int err)
2396{
2397 struct scrub_bio *sbio = bio->bi_private;
a36cf8b8 2398 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
b5d67f64
SB		 2399
2400 sbio->err = err;
2401 sbio->bio = bio;
2402
0339ef2f 2403 btrfs_queue_work(fs_info->scrub_workers, &sbio->work);
b5d67f64
SB		 2404}
2405
2406static void scrub_bio_end_io_worker(struct btrfs_work *work)
2407{
2408 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
d9d181c1 2409 struct scrub_ctx *sctx = sbio->sctx;
b5d67f64
SB		 2410 int i;
2411
ff023aac 2412 BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO);
b5d67f64
SB		 2413 if (sbio->err) {
2414 for (i = 0; i < sbio->page_count; i++) {
2415 struct scrub_page *spage = sbio->pagev[i];
2416
2417 spage->io_error = 1;
2418 spage->sblock->no_io_error_seen = 0;
2419 }
2420 }
2421
2422 /* now complete the scrub_block items that have all pages completed */
2423 for (i = 0; i < sbio->page_count; i++) {
2424 struct scrub_page *spage = sbio->pagev[i];
2425 struct scrub_block *sblock = spage->sblock;
2426
2427 if (atomic_dec_and_test(&sblock->outstanding_pages))
2428 scrub_block_complete(sblock);
2429 scrub_block_put(sblock);
2430 }
2431
b5d67f64
SB		 2432 bio_put(sbio->bio);
2433 sbio->bio = NULL;
d9d181c1
SB		 2434 spin_lock(&sctx->list_lock);
2435 sbio->next_free = sctx->first_free;
2436 sctx->first_free = sbio->index;
2437 spin_unlock(&sctx->list_lock);
ff023aac
SB		 2438
2439 if (sctx->is_dev_replace &&
2440 atomic_read(&sctx->wr_ctx.flush_all_writes)) {
2441 mutex_lock(&sctx->wr_ctx.wr_lock);
2442 scrub_wr_submit(sctx);
2443 mutex_unlock(&sctx->wr_ctx.wr_lock);
2444 }
2445
b6bfebc1 2446 scrub_pending_bio_dec(sctx);
b5d67f64
SB		 2447}
2448
5a6ac9ea
MX		 2449static inline void __scrub_mark_bitmap(struct scrub_parity *sparity,
2450 unsigned long *bitmap,
2451 u64 start, u64 len)
2452{
9d644a62 2453 u32 offset;
5a6ac9ea
MX		 2454 int nsectors;
2455 int sectorsize = sparity->sctx->dev_root->sectorsize;
2456
2457 if (len >= sparity->stripe_len) {
2458 bitmap_set(bitmap, 0, sparity->nsectors);
2459 return;
2460 }
2461
2462 start -= sparity->logic_start;
47c5713f 2463 start = div_u64_rem(start, sparity->stripe_len, &offset);
5a6ac9ea
MX		 2464 offset /= sectorsize;
2465 nsectors = (int)len / sectorsize;
2466
2467 if (offset + nsectors <= sparity->nsectors) {
2468 bitmap_set(bitmap, offset, nsectors);
2469 return;
2470 }
2471
2472 bitmap_set(bitmap, offset, sparity->nsectors - offset);
2473 bitmap_set(bitmap, 0, nsectors - (sparity->nsectors - offset));
2474}
2475
2476static inline void scrub_parity_mark_sectors_error(struct scrub_parity *sparity,
2477 u64 start, u64 len)
2478{
2479 __scrub_mark_bitmap(sparity, sparity->ebitmap, start, len);
2480}
2481
2482static inline void scrub_parity_mark_sectors_data(struct scrub_parity *sparity,
2483 u64 start, u64 len)
2484{
2485 __scrub_mark_bitmap(sparity, sparity->dbitmap, start, len);
2486}
2487
b5d67f64
SB		 2488static void scrub_block_complete(struct scrub_block *sblock)
2489{
5a6ac9ea
MX		 2490 int corrupted = 0;
2491
ff023aac 2492 if (!sblock->no_io_error_seen) {
5a6ac9ea 2493 corrupted = 1;
b5d67f64 2494 scrub_handle_errored_block(sblock);
ff023aac
SB		 2495 } else {
2496 /*
2497 * if has checksum error, write via repair mechanism in
2498 * dev replace case, otherwise write here in dev replace
2499 * case.
2500 */
5a6ac9ea
MX		 2501 corrupted = scrub_checksum(sblock);
2502 if (!corrupted && sblock->sctx->is_dev_replace)
ff023aac
SB		 2503 scrub_write_block_to_dev_replace(sblock);
2504 }
5a6ac9ea
MX		 2505
2506 if (sblock->sparity && corrupted && !sblock->data_corrected) {
2507 u64 start = sblock->pagev[0]->logical;
2508 u64 end = sblock->pagev[sblock->page_count - 1]->logical +
2509 PAGE_SIZE;
2510
2511 scrub_parity_mark_sectors_error(sblock->sparity,
2512 start, end - start);
2513 }
b5d67f64
SB		 2514}
2515
d9d181c1 2516static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len,
a2de733c
AJ		 2517 u8 *csum)
2518{
2519 struct btrfs_ordered_sum *sum = NULL;
f51a4a18 2520 unsigned long index;
a2de733c 2521 unsigned long num_sectors;
a2de733c 2522
d9d181c1
SB		 2523 while (!list_empty(&sctx->csum_list)) {
2524 sum = list_first_entry(&sctx->csum_list,
a2de733c
AJ		 2525 struct btrfs_ordered_sum, list);
2526 if (sum->bytenr > logical)
2527 return 0;
2528 if (sum->bytenr + sum->len > logical)
2529 break;
2530
d9d181c1 2531 ++sctx->stat.csum_discards;
a2de733c
AJ		 2532 list_del(&sum->list);
2533 kfree(sum);
2534 sum = NULL;
2535 }
2536 if (!sum)
2537 return 0;
2538
f51a4a18 2539 index = ((u32)(logical - sum->bytenr)) / sctx->sectorsize;
d9d181c1 2540 num_sectors = sum->len / sctx->sectorsize;
f51a4a18
MX		 2541 memcpy(csum, sum->sums + index, sctx->csum_size);
2542 if (index == num_sectors - 1) {
a2de733c
AJ		 2543 list_del(&sum->list);
2544 kfree(sum);
2545 }
f51a4a18 2546 return 1;
a2de733c
AJ		 2547}
2548
2549/* scrub extent tries to collect up to 64 kB for each bio */
d9d181c1 2550static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len,
a36cf8b8 2551 u64 physical, struct btrfs_device *dev, u64 flags,
ff023aac 2552 u64 gen, int mirror_num, u64 physical_for_dev_replace)
a2de733c
AJ		 2553{
2554 int ret;
2555 u8 csum[BTRFS_CSUM_SIZE];
b5d67f64
SB		 2556 u32 blocksize;
2557
2558 if (flags & BTRFS_EXTENT_FLAG_DATA) {
d9d181c1
SB		 2559 blocksize = sctx->sectorsize;
2560 spin_lock(&sctx->stat_lock);
2561 sctx->stat.data_extents_scrubbed++;
2562 sctx->stat.data_bytes_scrubbed += len;
2563 spin_unlock(&sctx->stat_lock);
b5d67f64 2564 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
d9d181c1
SB		 2565 blocksize = sctx->nodesize;
2566 spin_lock(&sctx->stat_lock);
2567 sctx->stat.tree_extents_scrubbed++;
2568 sctx->stat.tree_bytes_scrubbed += len;
2569 spin_unlock(&sctx->stat_lock);
b5d67f64 2570 } else {
d9d181c1 2571 blocksize = sctx->sectorsize;
ff023aac 2572 WARN_ON(1);
b5d67f64 2573 }
a2de733c
AJ		 2574
2575 while (len) {
b5d67f64 2576 u64 l = min_t(u64, len, blocksize);
a2de733c
AJ		 2577 int have_csum = 0;
2578
2579 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2580 /* push csums to sbio */
d9d181c1 2581 have_csum = scrub_find_csum(sctx, logical, l, csum);
a2de733c 2582 if (have_csum == 0)
d9d181c1 2583 ++sctx->stat.no_csum;
ff023aac
SB		 2584 if (sctx->is_dev_replace && !have_csum) {
2585 ret = copy_nocow_pages(sctx, logical, l,
2586 mirror_num,
2587 physical_for_dev_replace);
2588 goto behind_scrub_pages;
2589 }
a2de733c 2590 }
a36cf8b8 2591 ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen,
ff023aac
SB		 2592 mirror_num, have_csum ? csum : NULL, 0,
2593 physical_for_dev_replace);
2594behind_scrub_pages:
a2de733c
AJ		 2595 if (ret)
2596 return ret;
2597 len -= l;
2598 logical += l;
2599 physical += l;
ff023aac 2600 physical_for_dev_replace += l;
a2de733c
AJ		 2601 }
2602 return 0;
2603}
2604
5a6ac9ea
MX		 2605static int scrub_pages_for_parity(struct scrub_parity *sparity,
2606 u64 logical, u64 len,
2607 u64 physical, struct btrfs_device *dev,
2608 u64 flags, u64 gen, int mirror_num, u8 *csum)
2609{
2610 struct scrub_ctx *sctx = sparity->sctx;
2611 struct scrub_block *sblock;
2612 int index;
2613
2614 sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
2615 if (!sblock) {
2616 spin_lock(&sctx->stat_lock);
2617 sctx->stat.malloc_errors++;
2618 spin_unlock(&sctx->stat_lock);
2619 return -ENOMEM;
2620 }
2621
2622 /* one ref inside this function, plus one for each page added to
2623 * a bio later on */
57019345 2624 atomic_set(&sblock->refs, 1);
5a6ac9ea
MX		 2625 sblock->sctx = sctx;
2626 sblock->no_io_error_seen = 1;
2627 sblock->sparity = sparity;
2628 scrub_parity_get(sparity);
2629
2630 for (index = 0; len > 0; index++) {
2631 struct scrub_page *spage;
2632 u64 l = min_t(u64, len, PAGE_SIZE);
2633
2634 spage = kzalloc(sizeof(*spage), GFP_NOFS);
2635 if (!spage) {
2636leave_nomem:
2637 spin_lock(&sctx->stat_lock);
2638 sctx->stat.malloc_errors++;
2639 spin_unlock(&sctx->stat_lock);
2640 scrub_block_put(sblock);
2641 return -ENOMEM;
2642 }
2643 BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
2644 /* For scrub block */
2645 scrub_page_get(spage);
2646 sblock->pagev[index] = spage;
2647 /* For scrub parity */
2648 scrub_page_get(spage);
2649 list_add_tail(&spage->list, &sparity->spages);
2650 spage->sblock = sblock;
2651 spage->dev = dev;
2652 spage->flags = flags;
2653 spage->generation = gen;
2654 spage->logical = logical;
2655 spage->physical = physical;
2656 spage->mirror_num = mirror_num;
2657 if (csum) {
2658 spage->have_csum = 1;
2659 memcpy(spage->csum, csum, sctx->csum_size);
2660 } else {
2661 spage->have_csum = 0;
2662 }
2663 sblock->page_count++;
2664 spage->page = alloc_page(GFP_NOFS);
2665 if (!spage->page)
2666 goto leave_nomem;
2667 len -= l;
2668 logical += l;
2669 physical += l;
2670 }
2671
2672 WARN_ON(sblock->page_count == 0);
2673 for (index = 0; index < sblock->page_count; index++) {
2674 struct scrub_page *spage = sblock->pagev[index];
2675 int ret;
2676
2677 ret = scrub_add_page_to_rd_bio(sctx, spage);
2678 if (ret) {
2679 scrub_block_put(sblock);
2680 return ret;
2681 }
2682 }
2683
2684 /* last one frees, either here or in bio completion for last page */
2685 scrub_block_put(sblock);
2686 return 0;
2687}
2688
2689static int scrub_extent_for_parity(struct scrub_parity *sparity,
2690 u64 logical, u64 len,
2691 u64 physical, struct btrfs_device *dev,
2692 u64 flags, u64 gen, int mirror_num)
2693{
2694 struct scrub_ctx *sctx = sparity->sctx;
2695 int ret;
2696 u8 csum[BTRFS_CSUM_SIZE];
2697 u32 blocksize;
2698
4a770891
OS		 2699 if (dev->missing) {
2700 scrub_parity_mark_sectors_error(sparity, logical, len);
2701 return 0;
2702 }
2703
5a6ac9ea
MX		 2704 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2705 blocksize = sctx->sectorsize;
2706 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2707 blocksize = sctx->nodesize;
2708 } else {
2709 blocksize = sctx->sectorsize;
2710 WARN_ON(1);
2711 }
2712
2713 while (len) {
2714 u64 l = min_t(u64, len, blocksize);
2715 int have_csum = 0;
2716
2717 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2718 /* push csums to sbio */
2719 have_csum = scrub_find_csum(sctx, logical, l, csum);
2720 if (have_csum == 0)
2721 goto skip;
2722 }
2723 ret = scrub_pages_for_parity(sparity, logical, l, physical, dev,
2724 flags, gen, mirror_num,
2725 have_csum ? csum : NULL);
5a6ac9ea
MX		 2726 if (ret)
2727 return ret;
6b6d24b3 2728skip:
5a6ac9ea
MX		 2729 len -= l;
2730 logical += l;
2731 physical += l;
2732 }
2733 return 0;
2734}
2735
3b080b25
WS		 2736/*
2737 * Given a physical address, this will calculate it's
2738 * logical offset. if this is a parity stripe, it will return
2739 * the most left data stripe's logical offset.
2740 *
2741 * return 0 if it is a data stripe, 1 means parity stripe.
2742 */
2743static int get_raid56_logic_offset(u64 physical, int num,
5a6ac9ea
MX		 2744 struct map_lookup *map, u64 *offset,
2745 u64 *stripe_start)
3b080b25
WS		 2746{
2747 int i;
2748 int j = 0;
2749 u64 stripe_nr;
2750 u64 last_offset;
9d644a62
DS		 2751 u32 stripe_index;
2752 u32 rot;
3b080b25
WS		 2753
2754 last_offset = (physical - map->stripes[num].physical) *
2755 nr_data_stripes(map);
5a6ac9ea
MX		 2756 if (stripe_start)
2757 *stripe_start = last_offset;
2758
3b080b25
WS		 2759 *offset = last_offset;
2760 for (i = 0; i < nr_data_stripes(map); i++) {
2761 *offset = last_offset + i * map->stripe_len;
2762
b8b93add
DS		 2763 stripe_nr = div_u64(*offset, map->stripe_len);
2764 stripe_nr = div_u64(stripe_nr, nr_data_stripes(map));
3b080b25
WS		 2765
2766 /* Work out the disk rotation on this stripe-set */
47c5713f 2767 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, &rot);
3b080b25
WS		 2768 /* calculate which stripe this data locates */
2769 rot += i;
e4fbaee2 2770 stripe_index = rot % map->num_stripes;
3b080b25
WS		 2771 if (stripe_index == num)
2772 return 0;
2773 if (stripe_index < num)
2774 j++;
2775 }
2776 *offset = last_offset + j * map->stripe_len;
2777 return 1;
2778}
2779
5a6ac9ea
MX		 2780static void scrub_free_parity(struct scrub_parity *sparity)
2781{
2782 struct scrub_ctx *sctx = sparity->sctx;
2783 struct scrub_page *curr, *next;
2784 int nbits;
2785
2786 nbits = bitmap_weight(sparity->ebitmap, sparity->nsectors);
2787 if (nbits) {
2788 spin_lock(&sctx->stat_lock);
2789 sctx->stat.read_errors += nbits;
2790 sctx->stat.uncorrectable_errors += nbits;
2791 spin_unlock(&sctx->stat_lock);
2792 }
2793
2794 list_for_each_entry_safe(curr, next, &sparity->spages, list) {
2795 list_del_init(&curr->list);
2796 scrub_page_put(curr);
2797 }
2798
2799 kfree(sparity);
2800}
2801
20b2e302
ZL		 2802static void scrub_parity_bio_endio_worker(struct btrfs_work *work)
2803{
2804 struct scrub_parity *sparity = container_of(work, struct scrub_parity,
2805 work);
2806 struct scrub_ctx *sctx = sparity->sctx;
2807
2808 scrub_free_parity(sparity);
2809 scrub_pending_bio_dec(sctx);
2810}
2811
5a6ac9ea
MX		 2812static void scrub_parity_bio_endio(struct bio *bio, int error)
2813{
2814 struct scrub_parity *sparity = (struct scrub_parity *)bio->bi_private;
5a6ac9ea
MX		 2815
2816 if (error)
2817 bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap,
2818 sparity->nsectors);
2819
5a6ac9ea 2820 bio_put(bio);
20b2e302
ZL		 2821
2822 btrfs_init_work(&sparity->work, btrfs_scrubparity_helper,
2823 scrub_parity_bio_endio_worker, NULL, NULL);
2824 btrfs_queue_work(sparity->sctx->dev_root->fs_info->scrub_parity_workers,
2825 &sparity->work);
5a6ac9ea
MX		 2826}
2827
2828static void scrub_parity_check_and_repair(struct scrub_parity *sparity)
2829{
2830 struct scrub_ctx *sctx = sparity->sctx;
2831 struct bio *bio;
2832 struct btrfs_raid_bio *rbio;
2833 struct scrub_page *spage;
2834 struct btrfs_bio *bbio = NULL;
5a6ac9ea
MX		 2835 u64 length;
2836 int ret;
2837
2838 if (!bitmap_andnot(sparity->dbitmap, sparity->dbitmap, sparity->ebitmap,
2839 sparity->nsectors))
2840 goto out;
2841
a0dd59de 2842 length = sparity->logic_end - sparity->logic_start;
76035976 2843 ret = btrfs_map_sblock(sctx->dev_root->fs_info, WRITE,
5a6ac9ea 2844 sparity->logic_start,
8e5cfb55
ZL		 2845 &length, &bbio, 0, 1);
2846 if (ret || !bbio || !bbio->raid_map)
5a6ac9ea
MX		 2847 goto bbio_out;
2848
2849 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
2850 if (!bio)
2851 goto bbio_out;
2852
2853 bio->bi_iter.bi_sector = sparity->logic_start >> 9;
2854 bio->bi_private = sparity;
2855 bio->bi_end_io = scrub_parity_bio_endio;
2856
2857 rbio = raid56_parity_alloc_scrub_rbio(sctx->dev_root, bio, bbio,
8e5cfb55 2858 length, sparity->scrub_dev,
5a6ac9ea
MX		 2859 sparity->dbitmap,
2860 sparity->nsectors);
2861 if (!rbio)
2862 goto rbio_out;
2863
2864 list_for_each_entry(spage, &sparity->spages, list)
b4ee1782 2865 raid56_add_scrub_pages(rbio, spage->page, spage->logical);
5a6ac9ea
MX		 2866
2867 scrub_pending_bio_inc(sctx);
2868 raid56_parity_submit_scrub_rbio(rbio);
2869 return;
2870
2871rbio_out:
2872 bio_put(bio);
2873bbio_out:
6e9606d2 2874 btrfs_put_bbio(bbio);
5a6ac9ea
MX		 2875 bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap,
2876 sparity->nsectors);
2877 spin_lock(&sctx->stat_lock);
2878 sctx->stat.malloc_errors++;
2879 spin_unlock(&sctx->stat_lock);
2880out:
2881 scrub_free_parity(sparity);
2882}
2883
2884static inline int scrub_calc_parity_bitmap_len(int nsectors)
2885{
2886 return DIV_ROUND_UP(nsectors, BITS_PER_LONG) * (BITS_PER_LONG / 8);
2887}
2888
2889static void scrub_parity_get(struct scrub_parity *sparity)
2890{
57019345 2891 atomic_inc(&sparity->refs);
5a6ac9ea
MX		 2892}
2893
2894static void scrub_parity_put(struct scrub_parity *sparity)
2895{
57019345 2896 if (!atomic_dec_and_test(&sparity->refs))
5a6ac9ea
MX		 2897 return;
2898
2899 scrub_parity_check_and_repair(sparity);
2900}
2901
2902static noinline_for_stack int scrub_raid56_parity(struct scrub_ctx *sctx,
2903 struct map_lookup *map,
2904 struct btrfs_device *sdev,
2905 struct btrfs_path *path,
2906 u64 logic_start,
2907 u64 logic_end)
2908{
2909 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
2910 struct btrfs_root *root = fs_info->extent_root;
2911 struct btrfs_root *csum_root = fs_info->csum_root;
2912 struct btrfs_extent_item *extent;
4a770891 2913 struct btrfs_bio *bbio = NULL;
5a6ac9ea
MX		 2914 u64 flags;
2915 int ret;
2916 int slot;
2917 struct extent_buffer *l;
2918 struct btrfs_key key;
2919 u64 generation;
2920 u64 extent_logical;
2921 u64 extent_physical;
2922 u64 extent_len;
4a770891 2923 u64 mapped_length;
5a6ac9ea
MX		 2924 struct btrfs_device *extent_dev;
2925 struct scrub_parity *sparity;
2926 int nsectors;
2927 int bitmap_len;
2928 int extent_mirror_num;
2929 int stop_loop = 0;
2930
2931 nsectors = map->stripe_len / root->sectorsize;
2932 bitmap_len = scrub_calc_parity_bitmap_len(nsectors);
2933 sparity = kzalloc(sizeof(struct scrub_parity) + 2 * bitmap_len,
2934 GFP_NOFS);
2935 if (!sparity) {
2936 spin_lock(&sctx->stat_lock);
2937 sctx->stat.malloc_errors++;
2938 spin_unlock(&sctx->stat_lock);
2939 return -ENOMEM;
2940 }
2941
2942 sparity->stripe_len = map->stripe_len;
2943 sparity->nsectors = nsectors;
2944 sparity->sctx = sctx;
2945 sparity->scrub_dev = sdev;
2946 sparity->logic_start = logic_start;
2947 sparity->logic_end = logic_end;
57019345 2948 atomic_set(&sparity->refs, 1);
5a6ac9ea
MX		 2949 INIT_LIST_HEAD(&sparity->spages);
2950 sparity->dbitmap = sparity->bitmap;
2951 sparity->ebitmap = (void *)sparity->bitmap + bitmap_len;
2952
2953 ret = 0;
2954 while (logic_start < logic_end) {
2955 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2956 key.type = BTRFS_METADATA_ITEM_KEY;
2957 else
2958 key.type = BTRFS_EXTENT_ITEM_KEY;
2959 key.objectid = logic_start;
2960 key.offset = (u64)-1;
2961
2962 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2963 if (ret < 0)
2964 goto out;
2965
2966 if (ret > 0) {
2967 ret = btrfs_previous_extent_item(root, path, 0);
2968 if (ret < 0)
2969 goto out;
2970 if (ret > 0) {
2971 btrfs_release_path(path);
2972 ret = btrfs_search_slot(NULL, root, &key,
2973 path, 0, 0);
2974 if (ret < 0)
2975 goto out;
2976 }
2977 }
2978
2979 stop_loop = 0;
2980 while (1) {
2981 u64 bytes;
2982
2983 l = path->nodes[0];
2984 slot = path->slots[0];
2985 if (slot >= btrfs_header_nritems(l)) {
2986 ret = btrfs_next_leaf(root, path);
2987 if (ret == 0)
2988 continue;
2989 if (ret < 0)
2990 goto out;
2991
2992 stop_loop = 1;
2993 break;
2994 }
2995 btrfs_item_key_to_cpu(l, &key, slot);
2996
d7cad238
ZL		 2997 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
2998 key.type != BTRFS_METADATA_ITEM_KEY)
2999 goto next;
3000
5a6ac9ea
MX		 3001 if (key.type == BTRFS_METADATA_ITEM_KEY)
3002 bytes = root->nodesize;
3003 else
3004 bytes = key.offset;
3005
3006 if (key.objectid + bytes <= logic_start)
3007 goto next;
3008
a0dd59de 3009 if (key.objectid >= logic_end) {
5a6ac9ea
MX		 3010 stop_loop = 1;
3011 break;
3012 }
3013
3014 while (key.objectid >= logic_start + map->stripe_len)
3015 logic_start += map->stripe_len;
3016
3017 extent = btrfs_item_ptr(l, slot,
3018 struct btrfs_extent_item);
3019 flags = btrfs_extent_flags(l, extent);
3020 generation = btrfs_extent_generation(l, extent);
3021
a323e813
ZL		 3022 if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) &&
3023 (key.objectid < logic_start ||
3024 key.objectid + bytes >
3025 logic_start + map->stripe_len)) {
3026 btrfs_err(fs_info, "scrub: tree block %llu spanning stripes, ignored. logical=%llu",
3027 key.objectid, logic_start);
5a6ac9ea
MX		 3028 goto next;
3029 }
3030again:
3031 extent_logical = key.objectid;
3032 extent_len = bytes;
3033
3034 if (extent_logical < logic_start) {
3035 extent_len -= logic_start - extent_logical;
3036 extent_logical = logic_start;
3037 }
3038
3039 if (extent_logical + extent_len >
3040 logic_start + map->stripe_len)
3041 extent_len = logic_start + map->stripe_len -
3042 extent_logical;
3043
3044 scrub_parity_mark_sectors_data(sparity, extent_logical,
3045 extent_len);
3046
4a770891
OS		 3047 mapped_length = extent_len;
3048 ret = btrfs_map_block(fs_info, READ, extent_logical,
3049 &mapped_length, &bbio, 0);
3050 if (!ret) {
3051 if (!bbio || mapped_length < extent_len)
3052 ret = -EIO;
3053 }
3054 if (ret) {
3055 btrfs_put_bbio(bbio);
3056 goto out;
3057 }
3058 extent_physical = bbio->stripes[0].physical;
3059 extent_mirror_num = bbio->mirror_num;
3060 extent_dev = bbio->stripes[0].dev;
3061 btrfs_put_bbio(bbio);
5a6ac9ea
MX		 3062
3063 ret = btrfs_lookup_csums_range(csum_root,
3064 extent_logical,
3065 extent_logical + extent_len - 1,
3066 &sctx->csum_list, 1);
3067 if (ret)
3068 goto out;
3069
3070 ret = scrub_extent_for_parity(sparity, extent_logical,
3071 extent_len,
3072 extent_physical,
3073 extent_dev, flags,
3074 generation,
3075 extent_mirror_num);
6fa96d72
ZL		 3076
3077 scrub_free_csums(sctx);
3078
5a6ac9ea
MX		 3079 if (ret)
3080 goto out;
3081
5a6ac9ea
MX		 3082 if (extent_logical + extent_len <
3083 key.objectid + bytes) {
3084 logic_start += map->stripe_len;
3085
3086 if (logic_start >= logic_end) {
3087 stop_loop = 1;
3088 break;
3089 }
3090
3091 if (logic_start < key.objectid + bytes) {
3092 cond_resched();
3093 goto again;
3094 }
3095 }
3096next:
3097 path->slots[0]++;
3098 }
3099
3100 btrfs_release_path(path);
3101
3102 if (stop_loop)
3103 break;
3104
3105 logic_start += map->stripe_len;
3106 }
3107out:
3108 if (ret < 0)
3109 scrub_parity_mark_sectors_error(sparity, logic_start,
a0dd59de 3110 logic_end - logic_start);
5a6ac9ea
MX		 3111 scrub_parity_put(sparity);
3112 scrub_submit(sctx);
3113 mutex_lock(&sctx->wr_ctx.wr_lock);
3114 scrub_wr_submit(sctx);
3115 mutex_unlock(&sctx->wr_ctx.wr_lock);
3116
3117 btrfs_release_path(path);
3118 return ret < 0 ? ret : 0;
3119}
3120
d9d181c1 3121static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
a36cf8b8
SB		 3122 struct map_lookup *map,
3123 struct btrfs_device *scrub_dev,
ff023aac
SB		 3124 int num, u64 base, u64 length,
3125 int is_dev_replace)
a2de733c 3126{
5a6ac9ea 3127 struct btrfs_path *path, *ppath;
a36cf8b8 3128 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
a2de733c
AJ		 3129 struct btrfs_root *root = fs_info->extent_root;
3130 struct btrfs_root *csum_root = fs_info->csum_root;
3131 struct btrfs_extent_item *extent;
e7786c3a 3132 struct blk_plug plug;
a2de733c
AJ		 3133 u64 flags;
3134 int ret;
3135 int slot;
a2de733c 3136 u64 nstripes;
a2de733c
AJ		 3137 struct extent_buffer *l;
3138 struct btrfs_key key;
3139 u64 physical;
3140 u64 logical;
625f1c8d 3141 u64 logic_end;
3b080b25 3142 u64 physical_end;
a2de733c 3143 u64 generation;
e12fa9cd 3144 int mirror_num;
7a26285e
AJ		 3145 struct reada_control *reada1;
3146 struct reada_control *reada2;
3147 struct btrfs_key key_start;
3148 struct btrfs_key key_end;
a2de733c
AJ		 3149 u64 increment = map->stripe_len;
3150 u64 offset;
ff023aac
SB		 3151 u64 extent_logical;
3152 u64 extent_physical;
3153 u64 extent_len;
5a6ac9ea
MX		 3154 u64 stripe_logical;
3155 u64 stripe_end;
ff023aac
SB		 3156 struct btrfs_device *extent_dev;
3157 int extent_mirror_num;
3b080b25 3158 int stop_loop = 0;
53b381b3 3159
3b080b25 3160 physical = map->stripes[num].physical;
a2de733c 3161 offset = 0;
b8b93add 3162 nstripes = div_u64(length, map->stripe_len);
a2de733c
AJ		 3163 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3164 offset = map->stripe_len * num;
3165 increment = map->stripe_len * map->num_stripes;
193ea74b 3166 mirror_num = 1;
a2de733c
AJ		 3167 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3168 int factor = map->num_stripes / map->sub_stripes;
3169 offset = map->stripe_len * (num / map->sub_stripes);
3170 increment = map->stripe_len * factor;
193ea74b 3171 mirror_num = num % map->sub_stripes + 1;
a2de733c
AJ		 3172 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3173 increment = map->stripe_len;
193ea74b 3174 mirror_num = num % map->num_stripes + 1;
a2de733c
AJ		 3175 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3176 increment = map->stripe_len;
193ea74b 3177 mirror_num = num % map->num_stripes + 1;
ffe2d203 3178 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5a6ac9ea 3179 get_raid56_logic_offset(physical, num, map, &offset, NULL);
3b080b25
WS		 3180 increment = map->stripe_len * nr_data_stripes(map);
3181 mirror_num = 1;
a2de733c
AJ		 3182 } else {
3183 increment = map->stripe_len;
193ea74b 3184 mirror_num = 1;
a2de733c
AJ		 3185 }
3186
3187 path = btrfs_alloc_path();
3188 if (!path)
3189 return -ENOMEM;
3190
5a6ac9ea
MX		 3191 ppath = btrfs_alloc_path();
3192 if (!ppath) {
379d6854 3193 btrfs_free_path(path);
5a6ac9ea
MX		 3194 return -ENOMEM;
3195 }
3196
b5d67f64
SB		 3197 /*
3198 * work on commit root. The related disk blocks are static as
3199 * long as COW is applied. This means, it is save to rewrite
3200 * them to repair disk errors without any race conditions
3201 */
a2de733c
AJ		 3202 path->search_commit_root = 1;
3203 path->skip_locking = 1;
3204
063c54dc
GH		 3205 ppath->search_commit_root = 1;
3206 ppath->skip_locking = 1;
a2de733c 3207 /*
7a26285e
AJ		 3208 * trigger the readahead for extent tree csum tree and wait for
3209 * completion. During readahead, the scrub is officially paused
3210 * to not hold off transaction commits
a2de733c
AJ		 3211 */
3212 logical = base + offset;
3b080b25 3213 physical_end = physical + nstripes * map->stripe_len;
ffe2d203 3214 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3b080b25 3215 get_raid56_logic_offset(physical_end, num,
5a6ac9ea 3216 map, &logic_end, NULL);
3b080b25
WS		 3217 logic_end += base;
3218 } else {
3219 logic_end = logical + increment * nstripes;
3220 }
d9d181c1 3221 wait_event(sctx->list_wait,
b6bfebc1 3222 atomic_read(&sctx->bios_in_flight) == 0);
cb7ab021 3223 scrub_blocked_if_needed(fs_info);
7a26285e
AJ		 3224
3225 /* FIXME it might be better to start readahead at commit root */
3226 key_start.objectid = logical;
3227 key_start.type = BTRFS_EXTENT_ITEM_KEY;
3228 key_start.offset = (u64)0;
3b080b25 3229 key_end.objectid = logic_end;
3173a18f
JB		 3230 key_end.type = BTRFS_METADATA_ITEM_KEY;
3231 key_end.offset = (u64)-1;
7a26285e
AJ		 3232 reada1 = btrfs_reada_add(root, &key_start, &key_end);
3233
3234 key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
3235 key_start.type = BTRFS_EXTENT_CSUM_KEY;
3236 key_start.offset = logical;
3237 key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
3238 key_end.type = BTRFS_EXTENT_CSUM_KEY;
3b080b25 3239 key_end.offset = logic_end;
7a26285e
AJ		 3240 reada2 = btrfs_reada_add(csum_root, &key_start, &key_end);
3241
3242 if (!IS_ERR(reada1))
3243 btrfs_reada_wait(reada1);
3244 if (!IS_ERR(reada2))
3245 btrfs_reada_wait(reada2);
3246
a2de733c
AJ		 3247
3248 /*
3249 * collect all data csums for the stripe to avoid seeking during
3250 * the scrub. This might currently (crc32) end up to be about 1MB
3251 */
e7786c3a 3252 blk_start_plug(&plug);
a2de733c 3253
a2de733c
AJ		 3254 /*
3255 * now find all extents for each stripe and scrub them
3256 */
a2de733c 3257 ret = 0;
3b080b25 3258 while (physical < physical_end) {
a2de733c
AJ		 3259 /*
3260 * canceled?
3261 */
3262 if (atomic_read(&fs_info->scrub_cancel_req) ||
d9d181c1 3263 atomic_read(&sctx->cancel_req)) {
a2de733c
AJ		 3264 ret = -ECANCELED;
3265 goto out;
3266 }
3267 /*
3268 * check to see if we have to pause
3269 */
3270 if (atomic_read(&fs_info->scrub_pause_req)) {
3271 /* push queued extents */
ff023aac 3272 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
d9d181c1 3273 scrub_submit(sctx);
ff023aac
SB		 3274 mutex_lock(&sctx->wr_ctx.wr_lock);
3275 scrub_wr_submit(sctx);
3276 mutex_unlock(&sctx->wr_ctx.wr_lock);
d9d181c1 3277 wait_event(sctx->list_wait,
b6bfebc1 3278 atomic_read(&sctx->bios_in_flight) == 0);
ff023aac 3279 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
3cb0929a 3280 scrub_blocked_if_needed(fs_info);
a2de733c
AJ		 3281 }
3282
f2f66a2f
ZL		 3283 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3284 ret = get_raid56_logic_offset(physical, num, map,
3285 &logical,
3286 &stripe_logical);
3287 logical += base;
3288 if (ret) {
7955323b 3289 /* it is parity strip */
f2f66a2f 3290 stripe_logical += base;
a0dd59de 3291 stripe_end = stripe_logical + increment;
f2f66a2f
ZL		 3292 ret = scrub_raid56_parity(sctx, map, scrub_dev,
3293 ppath, stripe_logical,
3294 stripe_end);
3295 if (ret)
3296 goto out;
3297 goto skip;
3298 }
3299 }
3300
7c76edb7
WS		 3301 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
3302 key.type = BTRFS_METADATA_ITEM_KEY;
3303 else
3304 key.type = BTRFS_EXTENT_ITEM_KEY;
a2de733c 3305 key.objectid = logical;
625f1c8d 3306 key.offset = (u64)-1;
a2de733c
AJ		 3307
3308 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3309 if (ret < 0)
3310 goto out;
3173a18f 3311
8c51032f 3312 if (ret > 0) {
ade2e0b3 3313 ret = btrfs_previous_extent_item(root, path, 0);
a2de733c
AJ		 3314 if (ret < 0)
3315 goto out;
8c51032f
AJ		 3316 if (ret > 0) {
3317 /* there's no smaller item, so stick with the
3318 * larger one */
3319 btrfs_release_path(path);
3320 ret = btrfs_search_slot(NULL, root, &key,
3321 path, 0, 0);
3322 if (ret < 0)
3323 goto out;
3324 }
a2de733c
AJ		 3325 }
3326
625f1c8d 3327 stop_loop = 0;
a2de733c 3328 while (1) {
3173a18f
JB		 3329 u64 bytes;
3330
a2de733c
AJ		 3331 l = path->nodes[0];
3332 slot = path->slots[0];
3333 if (slot >= btrfs_header_nritems(l)) {
3334 ret = btrfs_next_leaf(root, path);
3335 if (ret == 0)
3336 continue;
3337 if (ret < 0)
3338 goto out;
3339
625f1c8d 3340 stop_loop = 1;
a2de733c
AJ		 3341 break;
3342 }
3343 btrfs_item_key_to_cpu(l, &key, slot);
3344
d7cad238
ZL		 3345 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3346 key.type != BTRFS_METADATA_ITEM_KEY)
3347 goto next;
3348
3173a18f 3349 if (key.type == BTRFS_METADATA_ITEM_KEY)
707e8a07 3350 bytes = root->nodesize;
3173a18f
JB		 3351 else
3352 bytes = key.offset;
3353
3354 if (key.objectid + bytes <= logical)
a2de733c
AJ		 3355 goto next;
3356
625f1c8d
LB		 3357 if (key.objectid >= logical + map->stripe_len) {
3358 /* out of this device extent */
3359 if (key.objectid >= logic_end)
3360 stop_loop = 1;
3361 break;
3362 }
a2de733c
AJ		 3363
3364 extent = btrfs_item_ptr(l, slot,
3365 struct btrfs_extent_item);
3366 flags = btrfs_extent_flags(l, extent);
3367 generation = btrfs_extent_generation(l, extent);
3368
a323e813
ZL		 3369 if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) &&
3370 (key.objectid < logical ||
3371 key.objectid + bytes >
3372 logical + map->stripe_len)) {
efe120a0
FH		 3373 btrfs_err(fs_info,
3374 "scrub: tree block %llu spanning "
3375 "stripes, ignored. logical=%llu",
c1c9ff7c 3376 key.objectid, logical);
a2de733c
AJ		 3377 goto next;
3378 }
3379
625f1c8d
LB		 3380again:
3381 extent_logical = key.objectid;
3382 extent_len = bytes;
3383
a2de733c
AJ		 3384 /*
3385 * trim extent to this stripe
3386 */
625f1c8d
LB		 3387 if (extent_logical < logical) {
3388 extent_len -= logical - extent_logical;
3389 extent_logical = logical;
a2de733c 3390 }
625f1c8d 3391 if (extent_logical + extent_len >
a2de733c 3392 logical + map->stripe_len) {
625f1c8d
LB		 3393 extent_len = logical + map->stripe_len -
3394 extent_logical;
a2de733c
AJ		 3395 }
3396
625f1c8d 3397 extent_physical = extent_logical - logical + physical;
ff023aac
SB		 3398 extent_dev = scrub_dev;
3399 extent_mirror_num = mirror_num;
3400 if (is_dev_replace)
3401 scrub_remap_extent(fs_info, extent_logical,
3402 extent_len, &extent_physical,
3403 &extent_dev,
3404 &extent_mirror_num);
625f1c8d 3405
fe8cf654
ZL		 3406 ret = btrfs_lookup_csums_range(csum_root,
3407 extent_logical,
3408 extent_logical +
3409 extent_len - 1,
3410 &sctx->csum_list, 1);
625f1c8d
LB		 3411 if (ret)
3412 goto out;
3413
ff023aac
SB		 3414 ret = scrub_extent(sctx, extent_logical, extent_len,
3415 extent_physical, extent_dev, flags,
3416 generation, extent_mirror_num,
115930cb 3417 extent_logical - logical + physical);
6fa96d72
ZL		 3418
3419 scrub_free_csums(sctx);
3420
a2de733c
AJ		 3421 if (ret)
3422 goto out;
3423
625f1c8d
LB		 3424 if (extent_logical + extent_len <
3425 key.objectid + bytes) {
ffe2d203 3426 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3b080b25
WS		 3427 /*
3428 * loop until we find next data stripe
3429 * or we have finished all stripes.
3430 */
5a6ac9ea
MX		 3431loop:
3432 physical += map->stripe_len;
3433 ret = get_raid56_logic_offset(physical,
3434 num, map, &logical,
3435 &stripe_logical);
3436 logical += base;
3437
3438 if (ret && physical < physical_end) {
3439 stripe_logical += base;
3440 stripe_end = stripe_logical +
a0dd59de 3441 increment;
5a6ac9ea
MX		 3442 ret = scrub_raid56_parity(sctx,
3443 map, scrub_dev, ppath,
3444 stripe_logical,
3445 stripe_end);
3446 if (ret)
3447 goto out;
3448 goto loop;
3449 }
3b080b25
WS		 3450 } else {
3451 physical += map->stripe_len;
3452 logical += increment;
3453 }
625f1c8d
LB		 3454 if (logical < key.objectid + bytes) {
3455 cond_resched();
3456 goto again;
3457 }
3458
3b080b25 3459 if (physical >= physical_end) {
625f1c8d
LB		 3460 stop_loop = 1;
3461 break;
3462 }
3463 }
a2de733c
AJ		 3464next:
3465 path->slots[0]++;
3466 }
71267333 3467 btrfs_release_path(path);
3b080b25 3468skip:
a2de733c
AJ		 3469 logical += increment;
3470 physical += map->stripe_len;
d9d181c1 3471 spin_lock(&sctx->stat_lock);
625f1c8d
LB		 3472 if (stop_loop)
3473 sctx->stat.last_physical = map->stripes[num].physical +
3474 length;
3475 else
3476 sctx->stat.last_physical = physical;
d9d181c1 3477 spin_unlock(&sctx->stat_lock);
625f1c8d
LB		 3478 if (stop_loop)
3479 break;
a2de733c 3480 }
ff023aac 3481out:
a2de733c 3482 /* push queued extents */
d9d181c1 3483 scrub_submit(sctx);
ff023aac
SB		 3484 mutex_lock(&sctx->wr_ctx.wr_lock);
3485 scrub_wr_submit(sctx);
3486 mutex_unlock(&sctx->wr_ctx.wr_lock);
a2de733c 3487
e7786c3a 3488 blk_finish_plug(&plug);
a2de733c 3489 btrfs_free_path(path);
5a6ac9ea 3490 btrfs_free_path(ppath);
a2de733c
AJ		 3491 return ret < 0 ? ret : 0;
3492}
3493
d9d181c1 3494static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx,
a36cf8b8
SB		 3495 struct btrfs_device *scrub_dev,
3496 u64 chunk_tree, u64 chunk_objectid,
3497 u64 chunk_offset, u64 length,
ff023aac 3498 u64 dev_offset, int is_dev_replace)
a2de733c
AJ		 3499{
3500 struct btrfs_mapping_tree *map_tree =
a36cf8b8 3501 &sctx->dev_root->fs_info->mapping_tree;
a2de733c
AJ		 3502 struct map_lookup *map;
3503 struct extent_map *em;
3504 int i;
ff023aac 3505 int ret = 0;
a2de733c
AJ		 3506
3507 read_lock(&map_tree->map_tree.lock);
3508 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3509 read_unlock(&map_tree->map_tree.lock);
3510
3511 if (!em)
3512 return -EINVAL;
3513
3514 map = (struct map_lookup *)em->bdev;
3515 if (em->start != chunk_offset)
3516 goto out;
3517
3518 if (em->len < length)
3519 goto out;
3520
3521 for (i = 0; i < map->num_stripes; ++i) {
a36cf8b8 3522 if (map->stripes[i].dev->bdev == scrub_dev->bdev &&
859acaf1 3523 map->stripes[i].physical == dev_offset) {
a36cf8b8 3524 ret = scrub_stripe(sctx, map, scrub_dev, i,
ff023aac
SB		 3525 chunk_offset, length,
3526 is_dev_replace);
a2de733c
AJ		 3527 if (ret)
3528 goto out;
3529 }
3530 }
3531out:
3532 free_extent_map(em);
3533
3534 return ret;
3535}
3536
3537static noinline_for_stack
a36cf8b8 3538int scrub_enumerate_chunks(struct scrub_ctx *sctx,
ff023aac
SB		 3539 struct btrfs_device *scrub_dev, u64 start, u64 end,
3540 int is_dev_replace)
a2de733c
AJ		 3541{
3542 struct btrfs_dev_extent *dev_extent = NULL;
3543 struct btrfs_path *path;
a36cf8b8 3544 struct btrfs_root *root = sctx->dev_root;
a2de733c
AJ		 3545 struct btrfs_fs_info *fs_info = root->fs_info;
3546 u64 length;
3547 u64 chunk_tree;
3548 u64 chunk_objectid;
3549 u64 chunk_offset;
55e3a601 3550 int ret = 0;
a2de733c
AJ		 3551 int slot;
3552 struct extent_buffer *l;
3553 struct btrfs_key key;
3554 struct btrfs_key found_key;
3555 struct btrfs_block_group_cache *cache;
ff023aac 3556 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
a2de733c
AJ		 3557
3558 path = btrfs_alloc_path();
3559 if (!path)
3560 return -ENOMEM;
3561
3562 path->reada = 2;
3563 path->search_commit_root = 1;
3564 path->skip_locking = 1;
3565
a36cf8b8 3566 key.objectid = scrub_dev->devid;
a2de733c
AJ		 3567 key.offset = 0ull;
3568 key.type = BTRFS_DEV_EXTENT_KEY;
3569
a2de733c
AJ		 3570 while (1) {
3571 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3572 if (ret < 0)
8c51032f
AJ		 3573 break;
3574 if (ret > 0) {
3575 if (path->slots[0] >=
3576 btrfs_header_nritems(path->nodes[0])) {
3577 ret = btrfs_next_leaf(root, path);
55e3a601
Z		 3578 if (ret < 0)
3579 break;
3580 if (ret > 0) {
3581 ret = 0;
8c51032f 3582 break;
55e3a601
Z		 3583 }
3584 } else {
3585 ret = 0;
8c51032f
AJ		 3586 }
3587 }
a2de733c
AJ		 3588
3589 l = path->nodes[0];
3590 slot = path->slots[0];
3591
3592 btrfs_item_key_to_cpu(l, &found_key, slot);
3593
a36cf8b8 3594 if (found_key.objectid != scrub_dev->devid)
a2de733c
AJ		 3595 break;
3596
962a298f 3597 if (found_key.type != BTRFS_DEV_EXTENT_KEY)
a2de733c
AJ		 3598 break;
3599
3600 if (found_key.offset >= end)
3601 break;
3602
3603 if (found_key.offset < key.offset)
3604 break;
3605
3606 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3607 length = btrfs_dev_extent_length(l, dev_extent);
3608
ced96edc
QW		 3609 if (found_key.offset + length <= start)
3610 goto skip;
a2de733c
AJ		 3611
3612 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3613 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3614 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3615
3616 /*
3617 * get a reference on the corresponding block group to prevent
3618 * the chunk from going away while we scrub it
3619 */
3620 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
ced96edc
QW		 3621
3622 /* some chunks are removed but not committed to disk yet,
3623 * continue scrubbing */
3624 if (!cache)
3625 goto skip;
3626
55e3a601
Z		 3627 /*
3628 * we need call btrfs_inc_block_group_ro() with scrubs_paused,
3629 * to avoid deadlock caused by:
3630 * btrfs_inc_block_group_ro()
3631 * -> btrfs_wait_for_commit()
3632 * -> btrfs_commit_transaction()
3633 * -> btrfs_scrub_pause()
3634 */
3635 scrub_pause_on(fs_info);
3636 ret = btrfs_inc_block_group_ro(root, cache);
3637 scrub_pause_off(fs_info);
3638 if (ret) {
3639 btrfs_put_block_group(cache);
3640 break;
3641 }
3642
ff023aac
SB		 3643 dev_replace->cursor_right = found_key.offset + length;
3644 dev_replace->cursor_left = found_key.offset;
3645 dev_replace->item_needs_writeback = 1;
a36cf8b8 3646 ret = scrub_chunk(sctx, scrub_dev, chunk_tree, chunk_objectid,
ff023aac
SB		 3647 chunk_offset, length, found_key.offset,
3648 is_dev_replace);
3649
3650 /*
3651 * flush, submit all pending read and write bios, afterwards
3652 * wait for them.
3653 * Note that in the dev replace case, a read request causes
3654 * write requests that are submitted in the read completion
3655 * worker. Therefore in the current situation, it is required
3656 * that all write requests are flushed, so that all read and
3657 * write requests are really completed when bios_in_flight
3658 * changes to 0.
3659 */
3660 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
3661 scrub_submit(sctx);
3662 mutex_lock(&sctx->wr_ctx.wr_lock);
3663 scrub_wr_submit(sctx);
3664 mutex_unlock(&sctx->wr_ctx.wr_lock);
3665
3666 wait_event(sctx->list_wait,
3667 atomic_read(&sctx->bios_in_flight) == 0);
b708ce96
Z		 3668
3669 scrub_pause_on(fs_info);
12cf9372
WS		 3670
3671 /*
3672 * must be called before we decrease @scrub_paused.
3673 * make sure we don't block transaction commit while
3674 * we are waiting pending workers finished.
3675 */
ff023aac
SB		 3676 wait_event(sctx->list_wait,
3677 atomic_read(&sctx->workers_pending) == 0);
12cf9372
WS		 3678 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
3679
b708ce96 3680 scrub_pause_off(fs_info);
ff023aac 3681
55e3a601
Z		 3682 btrfs_dec_block_group_ro(root, cache);
3683
a2de733c
AJ		 3684 btrfs_put_block_group(cache);
3685 if (ret)
3686 break;
af1be4f8
SB		 3687 if (is_dev_replace &&
3688 atomic64_read(&dev_replace->num_write_errors) > 0) {
ff023aac
SB		 3689 ret = -EIO;
3690 break;
3691 }
3692 if (sctx->stat.malloc_errors > 0) {
3693 ret = -ENOMEM;
3694 break;
3695 }
a2de733c 3696
539f358a
ID		 3697 dev_replace->cursor_left = dev_replace->cursor_right;
3698 dev_replace->item_needs_writeback = 1;
ced96edc 3699skip:
a2de733c 3700 key.offset = found_key.offset + length;
71267333 3701 btrfs_release_path(path);
a2de733c
AJ		 3702 }
3703
a2de733c 3704 btrfs_free_path(path);
8c51032f 3705
55e3a601 3706 return ret;
a2de733c
AJ		 3707}
3708
a36cf8b8
SB		 3709static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx,
3710 struct btrfs_device *scrub_dev)
a2de733c
AJ		 3711{
3712 int i;
3713 u64 bytenr;
3714 u64 gen;
3715 int ret;
a36cf8b8 3716 struct btrfs_root *root = sctx->dev_root;
a2de733c 3717
87533c47 3718 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
79787eaa
JM		 3719 return -EIO;
3720
5f546063
MX		 3721 /* Seed devices of a new filesystem has their own generation. */
3722 if (scrub_dev->fs_devices != root->fs_info->fs_devices)
3723 gen = scrub_dev->generation;
3724 else
3725 gen = root->fs_info->last_trans_committed;
a2de733c
AJ		 3726
3727 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
3728 bytenr = btrfs_sb_offset(i);
935e5cc9
MX		 3729 if (bytenr + BTRFS_SUPER_INFO_SIZE >
3730 scrub_dev->commit_total_bytes)
a2de733c
AJ		 3731 break;
3732
d9d181c1 3733 ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr,
a36cf8b8 3734 scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i,
ff023aac 3735 NULL, 1, bytenr);
a2de733c
AJ		 3736 if (ret)
3737 return ret;
3738 }
b6bfebc1 3739 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
a2de733c
AJ		 3740
3741 return 0;
3742}
3743
3744/*
3745 * get a reference count on fs_info->scrub_workers. start worker if necessary
3746 */
ff023aac
SB		 3747static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info,
3748 int is_dev_replace)
a2de733c 3749{
6f011058 3750 unsigned int flags = WQ_FREEZABLE | WQ_UNBOUND;
0339ef2f 3751 int max_active = fs_info->thread_pool_size;
a2de733c 3752
632dd772 3753 if (fs_info->scrub_workers_refcnt == 0) {
ff023aac 3754 if (is_dev_replace)
0339ef2f
QW		 3755 fs_info->scrub_workers =
3756 btrfs_alloc_workqueue("btrfs-scrub", flags,
3757 1, 4);
ff023aac 3758 else
0339ef2f
QW		 3759 fs_info->scrub_workers =
3760 btrfs_alloc_workqueue("btrfs-scrub", flags,
3761 max_active, 4);
e82afc52
ZL		 3762 if (!fs_info->scrub_workers)
3763 goto fail_scrub_workers;
3764
0339ef2f
QW		 3765 fs_info->scrub_wr_completion_workers =
3766 btrfs_alloc_workqueue("btrfs-scrubwrc", flags,
3767 max_active, 2);
e82afc52
ZL		 3768 if (!fs_info->scrub_wr_completion_workers)
3769 goto fail_scrub_wr_completion_workers;
3770
0339ef2f
QW		 3771 fs_info->scrub_nocow_workers =
3772 btrfs_alloc_workqueue("btrfs-scrubnc", flags, 1, 0);
e82afc52
ZL		 3773 if (!fs_info->scrub_nocow_workers)
3774 goto fail_scrub_nocow_workers;
20b2e302
ZL		 3775 fs_info->scrub_parity_workers =
3776 btrfs_alloc_workqueue("btrfs-scrubparity", flags,
3777 max_active, 2);
e82afc52
ZL		 3778 if (!fs_info->scrub_parity_workers)
3779 goto fail_scrub_parity_workers;
632dd772 3780 }
a2de733c 3781 ++fs_info->scrub_workers_refcnt;
e82afc52
ZL		 3782 return 0;
3783
3784fail_scrub_parity_workers:
3785 btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
3786fail_scrub_nocow_workers:
3787 btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
3788fail_scrub_wr_completion_workers:
3789 btrfs_destroy_workqueue(fs_info->scrub_workers);
3790fail_scrub_workers:
3791 return -ENOMEM;
a2de733c
AJ		 3792}
3793
aa1b8cd4 3794static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info)
a2de733c 3795{
ff023aac 3796 if (--fs_info->scrub_workers_refcnt == 0) {
0339ef2f
QW		 3797 btrfs_destroy_workqueue(fs_info->scrub_workers);
3798 btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
3799 btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
20b2e302 3800 btrfs_destroy_workqueue(fs_info->scrub_parity_workers);
ff023aac 3801 }
a2de733c 3802 WARN_ON(fs_info->scrub_workers_refcnt < 0);
a2de733c
AJ		 3803}
3804
aa1b8cd4
SB		 3805int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start,
3806 u64 end, struct btrfs_scrub_progress *progress,
63a212ab 3807 int readonly, int is_dev_replace)
a2de733c 3808{
d9d181c1 3809 struct scrub_ctx *sctx;
a2de733c
AJ		 3810 int ret;
3811 struct btrfs_device *dev;
5d68da3b 3812 struct rcu_string *name;
a2de733c 3813
aa1b8cd4 3814 if (btrfs_fs_closing(fs_info))
a2de733c
AJ		 3815 return -EINVAL;
3816
aa1b8cd4 3817 if (fs_info->chunk_root->nodesize > BTRFS_STRIPE_LEN) {
b5d67f64
SB		 3818 /*
3819 * in this case scrub is unable to calculate the checksum
3820 * the way scrub is implemented. Do not handle this
3821 * situation at all because it won't ever happen.
3822 */
efe120a0
FH		 3823 btrfs_err(fs_info,
3824 "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails",
aa1b8cd4 3825 fs_info->chunk_root->nodesize, BTRFS_STRIPE_LEN);
b5d67f64
SB		 3826 return -EINVAL;
3827 }
3828
aa1b8cd4 3829 if (fs_info->chunk_root->sectorsize != PAGE_SIZE) {
b5d67f64 3830 /* not supported for data w/o checksums */
efe120a0
FH		 3831 btrfs_err(fs_info,
3832 "scrub: size assumption sectorsize != PAGE_SIZE "
3833 "(%d != %lu) fails",
27f9f023 3834 fs_info->chunk_root->sectorsize, PAGE_SIZE);
a2de733c
AJ		 3835 return -EINVAL;
3836 }
3837
7a9e9987
SB		 3838 if (fs_info->chunk_root->nodesize >
3839 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK ||
3840 fs_info->chunk_root->sectorsize >
3841 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) {
3842 /*
3843 * would exhaust the array bounds of pagev member in
3844 * struct scrub_block
3845 */
efe120a0
FH		 3846 btrfs_err(fs_info, "scrub: size assumption nodesize and sectorsize "
3847 "<= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails",
7a9e9987
SB		 3848 fs_info->chunk_root->nodesize,
3849 SCRUB_MAX_PAGES_PER_BLOCK,
3850 fs_info->chunk_root->sectorsize,
3851 SCRUB_MAX_PAGES_PER_BLOCK);
3852 return -EINVAL;
3853 }
3854
a2de733c 3855
aa1b8cd4
SB		 3856 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3857 dev = btrfs_find_device(fs_info, devid, NULL, NULL);
63a212ab 3858 if (!dev || (dev->missing && !is_dev_replace)) {
aa1b8cd4 3859 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
a2de733c
AJ		 3860 return -ENODEV;
3861 }
a2de733c 3862
5d68da3b
MX		 3863 if (!is_dev_replace && !readonly && !dev->writeable) {
3864 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3865 rcu_read_lock();
3866 name = rcu_dereference(dev->name);
3867 btrfs_err(fs_info, "scrub: device %s is not writable",
3868 name->str);
3869 rcu_read_unlock();
3870 return -EROFS;
3871 }
3872
3b7a016f 3873 mutex_lock(&fs_info->scrub_lock);
63a212ab 3874 if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) {
a2de733c 3875 mutex_unlock(&fs_info->scrub_lock);
aa1b8cd4 3876 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
aa1b8cd4 3877 return -EIO;
a2de733c
AJ		 3878 }
3879
8dabb742
SB		 3880 btrfs_dev_replace_lock(&fs_info->dev_replace);
3881 if (dev->scrub_device ||
3882 (!is_dev_replace &&
3883 btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) {
3884 btrfs_dev_replace_unlock(&fs_info->dev_replace);
a2de733c 3885 mutex_unlock(&fs_info->scrub_lock);
aa1b8cd4 3886 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
a2de733c
AJ		 3887 return -EINPROGRESS;
3888 }
8dabb742 3889 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3b7a016f
WS		 3890
3891 ret = scrub_workers_get(fs_info, is_dev_replace);
3892 if (ret) {
3893 mutex_unlock(&fs_info->scrub_lock);
3894 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3895 return ret;
3896 }
3897
63a212ab 3898 sctx = scrub_setup_ctx(dev, is_dev_replace);
d9d181c1 3899 if (IS_ERR(sctx)) {
a2de733c 3900 mutex_unlock(&fs_info->scrub_lock);
aa1b8cd4
SB		 3901 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3902 scrub_workers_put(fs_info);
d9d181c1 3903 return PTR_ERR(sctx);
a2de733c 3904 }
d9d181c1
SB		 3905 sctx->readonly = readonly;
3906 dev->scrub_device = sctx;
3cb0929a 3907 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
a2de733c 3908
3cb0929a
WS		 3909 /*
3910 * checking @scrub_pause_req here, we can avoid
3911 * race between committing transaction and scrubbing.
3912 */
cb7ab021 3913 __scrub_blocked_if_needed(fs_info);
a2de733c
AJ		 3914 atomic_inc(&fs_info->scrubs_running);
3915 mutex_unlock(&fs_info->scrub_lock);
a2de733c 3916
ff023aac 3917 if (!is_dev_replace) {
9b011adf
WS		 3918 /*
3919 * by holding device list mutex, we can
3920 * kick off writing super in log tree sync.
3921 */
3cb0929a 3922 mutex_lock(&fs_info->fs_devices->device_list_mutex);
ff023aac 3923 ret = scrub_supers(sctx, dev);
3cb0929a 3924 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
ff023aac 3925 }
a2de733c
AJ		 3926
3927 if (!ret)
ff023aac
SB		 3928 ret = scrub_enumerate_chunks(sctx, dev, start, end,
3929 is_dev_replace);
a2de733c 3930
b6bfebc1 3931 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
a2de733c
AJ		 3932 atomic_dec(&fs_info->scrubs_running);
3933 wake_up(&fs_info->scrub_pause_wait);
3934
b6bfebc1 3935 wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0);
0ef8e451 3936
a2de733c 3937 if (progress)
d9d181c1 3938 memcpy(progress, &sctx->stat, sizeof(*progress));
a2de733c
AJ		 3939
3940 mutex_lock(&fs_info->scrub_lock);
3941 dev->scrub_device = NULL;
3b7a016f 3942 scrub_workers_put(fs_info);
a2de733c
AJ		 3943 mutex_unlock(&fs_info->scrub_lock);
3944
f55985f4 3945 scrub_put_ctx(sctx);
a2de733c
AJ		 3946
3947 return ret;
3948}
3949
143bede5 3950void btrfs_scrub_pause(struct btrfs_root *root)
a2de733c
AJ		 3951{
3952 struct btrfs_fs_info *fs_info = root->fs_info;
3953
3954 mutex_lock(&fs_info->scrub_lock);
3955 atomic_inc(&fs_info->scrub_pause_req);
3956 while (atomic_read(&fs_info->scrubs_paused) !=
3957 atomic_read(&fs_info->scrubs_running)) {
3958 mutex_unlock(&fs_info->scrub_lock);
3959 wait_event(fs_info->scrub_pause_wait,
3960 atomic_read(&fs_info->scrubs_paused) ==
3961 atomic_read(&fs_info->scrubs_running));
3962 mutex_lock(&fs_info->scrub_lock);
3963 }
3964 mutex_unlock(&fs_info->scrub_lock);
a2de733c
AJ		 3965}
3966
143bede5 3967void btrfs_scrub_continue(struct btrfs_root *root)
a2de733c
AJ		 3968{
3969 struct btrfs_fs_info *fs_info = root->fs_info;
3970
3971 atomic_dec(&fs_info->scrub_pause_req);
3972 wake_up(&fs_info->scrub_pause_wait);
a2de733c
AJ		 3973}
3974
aa1b8cd4 3975int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info)
a2de733c 3976{
a2de733c
AJ		 3977 mutex_lock(&fs_info->scrub_lock);
3978 if (!atomic_read(&fs_info->scrubs_running)) {
3979 mutex_unlock(&fs_info->scrub_lock);
3980 return -ENOTCONN;
3981 }
3982
3983 atomic_inc(&fs_info->scrub_cancel_req);
3984 while (atomic_read(&fs_info->scrubs_running)) {
3985 mutex_unlock(&fs_info->scrub_lock);
3986 wait_event(fs_info->scrub_pause_wait,
3987 atomic_read(&fs_info->scrubs_running) == 0);
3988 mutex_lock(&fs_info->scrub_lock);
3989 }
3990 atomic_dec(&fs_info->scrub_cancel_req);
3991 mutex_unlock(&fs_info->scrub_lock);
3992
3993 return 0;
3994}
3995
aa1b8cd4
SB		 3996int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info,
3997 struct btrfs_device *dev)
49b25e05 3998{
d9d181c1 3999 struct scrub_ctx *sctx;
a2de733c
AJ		 4000
4001 mutex_lock(&fs_info->scrub_lock);
d9d181c1
SB		 4002 sctx = dev->scrub_device;
4003 if (!sctx) {
a2de733c
AJ		 4004 mutex_unlock(&fs_info->scrub_lock);
4005 return -ENOTCONN;
4006 }
d9d181c1 4007 atomic_inc(&sctx->cancel_req);
a2de733c
AJ		 4008 while (dev->scrub_device) {
4009 mutex_unlock(&fs_info->scrub_lock);
4010 wait_event(fs_info->scrub_pause_wait,
4011 dev->scrub_device == NULL);
4012 mutex_lock(&fs_info->scrub_lock);
4013 }
4014 mutex_unlock(&fs_info->scrub_lock);
4015
4016 return 0;
4017}
1623edeb 4018
a2de733c
AJ		 4019int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
4020 struct btrfs_scrub_progress *progress)
4021{
4022 struct btrfs_device *dev;
d9d181c1 4023 struct scrub_ctx *sctx = NULL;
a2de733c
AJ		 4024
4025 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
aa1b8cd4 4026 dev = btrfs_find_device(root->fs_info, devid, NULL, NULL);
a2de733c 4027 if (dev)
d9d181c1
SB		 4028 sctx = dev->scrub_device;
4029 if (sctx)
4030 memcpy(progress, &sctx->stat, sizeof(*progress));
a2de733c
AJ		 4031 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
4032
d9d181c1 4033 return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV;
a2de733c 4034}
ff023aac
SB		 4035
4036static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
4037 u64 extent_logical, u64 extent_len,
4038 u64 *extent_physical,
4039 struct btrfs_device **extent_dev,
4040 int *extent_mirror_num)
4041{
4042 u64 mapped_length;
4043 struct btrfs_bio *bbio = NULL;
4044 int ret;
4045
4046 mapped_length = extent_len;
4047 ret = btrfs_map_block(fs_info, READ, extent_logical,
4048 &mapped_length, &bbio, 0);
4049 if (ret || !bbio || mapped_length < extent_len ||
4050 !bbio->stripes[0].dev->bdev) {
6e9606d2 4051 btrfs_put_bbio(bbio);
ff023aac
SB		 4052 return;
4053 }
4054
4055 *extent_physical = bbio->stripes[0].physical;
4056 *extent_mirror_num = bbio->mirror_num;
4057 *extent_dev = bbio->stripes[0].dev;
6e9606d2 4058 btrfs_put_bbio(bbio);
ff023aac
SB		 4059}
4060
4061static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
4062 struct scrub_wr_ctx *wr_ctx,
4063 struct btrfs_fs_info *fs_info,
4064 struct btrfs_device *dev,
4065 int is_dev_replace)
4066{
4067 WARN_ON(wr_ctx->wr_curr_bio != NULL);
4068
4069 mutex_init(&wr_ctx->wr_lock);
4070 wr_ctx->wr_curr_bio = NULL;
4071 if (!is_dev_replace)
4072 return 0;
4073
4074 WARN_ON(!dev->bdev);
4075 wr_ctx->pages_per_wr_bio = min_t(int, SCRUB_PAGES_PER_WR_BIO,
4076 bio_get_nr_vecs(dev->bdev));
4077 wr_ctx->tgtdev = dev;
4078 atomic_set(&wr_ctx->flush_all_writes, 0);
4079 return 0;
4080}
4081
4082static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx)
4083{
4084 mutex_lock(&wr_ctx->wr_lock);
4085 kfree(wr_ctx->wr_curr_bio);
4086 wr_ctx->wr_curr_bio = NULL;
4087 mutex_unlock(&wr_ctx->wr_lock);
4088}
4089
4090static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
4091 int mirror_num, u64 physical_for_dev_replace)
4092{
4093 struct scrub_copy_nocow_ctx *nocow_ctx;
4094 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
4095
4096 nocow_ctx = kzalloc(sizeof(*nocow_ctx), GFP_NOFS);
4097 if (!nocow_ctx) {
4098 spin_lock(&sctx->stat_lock);
4099 sctx->stat.malloc_errors++;
4100 spin_unlock(&sctx->stat_lock);
4101 return -ENOMEM;
4102 }
4103
4104 scrub_pending_trans_workers_inc(sctx);
4105
4106 nocow_ctx->sctx = sctx;
4107 nocow_ctx->logical = logical;
4108 nocow_ctx->len = len;
4109 nocow_ctx->mirror_num = mirror_num;
4110 nocow_ctx->physical_for_dev_replace = physical_for_dev_replace;
9e0af237
LB		 4111 btrfs_init_work(&nocow_ctx->work, btrfs_scrubnc_helper,
4112 copy_nocow_pages_worker, NULL, NULL);
652f25a2 4113 INIT_LIST_HEAD(&nocow_ctx->inodes);
0339ef2f
QW		 4114 btrfs_queue_work(fs_info->scrub_nocow_workers,
4115 &nocow_ctx->work);
ff023aac
SB		 4116
4117 return 0;
4118}
4119
652f25a2
JB		 4120static int record_inode_for_nocow(u64 inum, u64 offset, u64 root, void *ctx)
4121{
4122 struct scrub_copy_nocow_ctx *nocow_ctx = ctx;
4123 struct scrub_nocow_inode *nocow_inode;
4124
4125 nocow_inode = kzalloc(sizeof(*nocow_inode), GFP_NOFS);
4126 if (!nocow_inode)
4127 return -ENOMEM;
4128 nocow_inode->inum = inum;
4129 nocow_inode->offset = offset;
4130 nocow_inode->root = root;
4131 list_add_tail(&nocow_inode->list, &nocow_ctx->inodes);
4132 return 0;
4133}
4134
4135#define COPY_COMPLETE 1
4136
ff023aac
SB		 4137static void copy_nocow_pages_worker(struct btrfs_work *work)
4138{
4139 struct scrub_copy_nocow_ctx *nocow_ctx =
4140 container_of(work, struct scrub_copy_nocow_ctx, work);
4141 struct scrub_ctx *sctx = nocow_ctx->sctx;
4142 u64 logical = nocow_ctx->logical;
4143 u64 len = nocow_ctx->len;
4144 int mirror_num = nocow_ctx->mirror_num;
4145 u64 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
4146 int ret;
4147 struct btrfs_trans_handle *trans = NULL;
4148 struct btrfs_fs_info *fs_info;
4149 struct btrfs_path *path;
4150 struct btrfs_root *root;
4151 int not_written = 0;
4152
4153 fs_info = sctx->dev_root->fs_info;
4154 root = fs_info->extent_root;
4155
4156 path = btrfs_alloc_path();
4157 if (!path) {
4158 spin_lock(&sctx->stat_lock);
4159 sctx->stat.malloc_errors++;
4160 spin_unlock(&sctx->stat_lock);
4161 not_written = 1;
4162 goto out;
4163 }
4164
4165 trans = btrfs_join_transaction(root);
4166 if (IS_ERR(trans)) {
4167 not_written = 1;
4168 goto out;
4169 }
4170
4171 ret = iterate_inodes_from_logical(logical, fs_info, path,
652f25a2 4172 record_inode_for_nocow, nocow_ctx);
ff023aac 4173 if (ret != 0 && ret != -ENOENT) {
efe120a0
FH		 4174 btrfs_warn(fs_info, "iterate_inodes_from_logical() failed: log %llu, "
4175 "phys %llu, len %llu, mir %u, ret %d",
118a0a25
GU		 4176 logical, physical_for_dev_replace, len, mirror_num,
4177 ret);
ff023aac
SB		 4178 not_written = 1;
4179 goto out;
4180 }
4181
652f25a2
JB		 4182 btrfs_end_transaction(trans, root);
4183 trans = NULL;
4184 while (!list_empty(&nocow_ctx->inodes)) {
4185 struct scrub_nocow_inode *entry;
4186 entry = list_first_entry(&nocow_ctx->inodes,
4187 struct scrub_nocow_inode,
4188 list);
4189 list_del_init(&entry->list);
4190 ret = copy_nocow_pages_for_inode(entry->inum, entry->offset,
4191 entry->root, nocow_ctx);
4192 kfree(entry);
4193 if (ret == COPY_COMPLETE) {
4194 ret = 0;
4195 break;
4196 } else if (ret) {
4197 break;
4198 }
4199 }
ff023aac 4200out:
652f25a2
JB		 4201 while (!list_empty(&nocow_ctx->inodes)) {
4202 struct scrub_nocow_inode *entry;
4203 entry = list_first_entry(&nocow_ctx->inodes,
4204 struct scrub_nocow_inode,
4205 list);
4206 list_del_init(&entry->list);
4207 kfree(entry);
4208 }
ff023aac
SB		 4209 if (trans && !IS_ERR(trans))
4210 btrfs_end_transaction(trans, root);
4211 if (not_written)
4212 btrfs_dev_replace_stats_inc(&fs_info->dev_replace.
4213 num_uncorrectable_read_errors);
4214
4215 btrfs_free_path(path);
4216 kfree(nocow_ctx);
4217
4218 scrub_pending_trans_workers_dec(sctx);
4219}
4220
32159242
GH		 4221static int check_extent_to_block(struct inode *inode, u64 start, u64 len,
4222 u64 logical)
4223{
4224 struct extent_state *cached_state = NULL;
4225 struct btrfs_ordered_extent *ordered;
4226 struct extent_io_tree *io_tree;
4227 struct extent_map *em;
4228 u64 lockstart = start, lockend = start + len - 1;
4229 int ret = 0;
4230
4231 io_tree = &BTRFS_I(inode)->io_tree;
4232
4233 lock_extent_bits(io_tree, lockstart, lockend, 0, &cached_state);
4234 ordered = btrfs_lookup_ordered_range(inode, lockstart, len);
4235 if (ordered) {
4236 btrfs_put_ordered_extent(ordered);
4237 ret = 1;
4238 goto out_unlock;
4239 }
4240
4241 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
4242 if (IS_ERR(em)) {
4243 ret = PTR_ERR(em);
4244 goto out_unlock;
4245 }
4246
4247 /*
4248 * This extent does not actually cover the logical extent anymore,
4249 * move on to the next inode.
4250 */
4251 if (em->block_start > logical ||
4252 em->block_start + em->block_len < logical + len) {
4253 free_extent_map(em);
4254 ret = 1;
4255 goto out_unlock;
4256 }
4257 free_extent_map(em);
4258
4259out_unlock:
4260 unlock_extent_cached(io_tree, lockstart, lockend, &cached_state,
4261 GFP_NOFS);
4262 return ret;
4263}
4264
652f25a2
JB		 4265static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
4266 struct scrub_copy_nocow_ctx *nocow_ctx)
ff023aac 4267{
826aa0a8 4268 struct btrfs_fs_info *fs_info = nocow_ctx->sctx->dev_root->fs_info;
ff023aac 4269 struct btrfs_key key;
826aa0a8
MX		 4270 struct inode *inode;
4271 struct page *page;
ff023aac 4272 struct btrfs_root *local_root;
652f25a2 4273 struct extent_io_tree *io_tree;
ff023aac 4274 u64 physical_for_dev_replace;
32159242 4275 u64 nocow_ctx_logical;
652f25a2 4276 u64 len = nocow_ctx->len;
826aa0a8 4277 unsigned long index;
6f1c3605 4278 int srcu_index;
652f25a2
JB		 4279 int ret = 0;
4280 int err = 0;
ff023aac
SB		 4281
4282 key.objectid = root;
4283 key.type = BTRFS_ROOT_ITEM_KEY;
4284 key.offset = (u64)-1;
6f1c3605
LB		 4285
4286 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
4287
ff023aac 4288 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
6f1c3605
LB		 4289 if (IS_ERR(local_root)) {
4290 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
ff023aac 4291 return PTR_ERR(local_root);
6f1c3605 4292 }
ff023aac
SB		 4293
4294 key.type = BTRFS_INODE_ITEM_KEY;
4295 key.objectid = inum;
4296 key.offset = 0;
4297 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
6f1c3605 4298 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
ff023aac
SB		 4299 if (IS_ERR(inode))
4300 return PTR_ERR(inode);
4301
edd1400b
MX		 4302 /* Avoid truncate/dio/punch hole.. */
4303 mutex_lock(&inode->i_mutex);
4304 inode_dio_wait(inode);
4305
ff023aac 4306 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
652f25a2 4307 io_tree = &BTRFS_I(inode)->io_tree;
32159242 4308 nocow_ctx_logical = nocow_ctx->logical;
652f25a2 4309
32159242
GH		 4310 ret = check_extent_to_block(inode, offset, len, nocow_ctx_logical);
4311 if (ret) {
4312 ret = ret > 0 ? 0 : ret;
4313 goto out;
652f25a2 4314 }
652f25a2 4315
ff023aac 4316 while (len >= PAGE_CACHE_SIZE) {
ff023aac 4317 index = offset >> PAGE_CACHE_SHIFT;
edd1400b 4318again:
ff023aac
SB		 4319 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
4320 if (!page) {
efe120a0 4321 btrfs_err(fs_info, "find_or_create_page() failed");
ff023aac 4322 ret = -ENOMEM;
826aa0a8 4323 goto out;
ff023aac
SB		 4324 }
4325
4326 if (PageUptodate(page)) {
4327 if (PageDirty(page))
4328 goto next_page;
4329 } else {
4330 ClearPageError(page);
32159242 4331 err = extent_read_full_page(io_tree, page,
652f25a2
JB		 4332 btrfs_get_extent,
4333 nocow_ctx->mirror_num);
826aa0a8
MX		 4334 if (err) {
4335 ret = err;
ff023aac
SB		 4336 goto next_page;
4337 }
edd1400b 4338
26b25891 4339 lock_page(page);
edd1400b
MX		 4340 /*
4341 * If the page has been remove from the page cache,
4342 * the data on it is meaningless, because it may be
4343 * old one, the new data may be written into the new
4344 * page in the page cache.
4345 */
4346 if (page->mapping != inode->i_mapping) {
652f25a2 4347 unlock_page(page);
edd1400b
MX		 4348 page_cache_release(page);
4349 goto again;
4350 }
ff023aac
SB		 4351 if (!PageUptodate(page)) {
4352 ret = -EIO;
4353 goto next_page;
4354 }
4355 }
32159242
GH		 4356
4357 ret = check_extent_to_block(inode, offset, len,
4358 nocow_ctx_logical);
4359 if (ret) {
4360 ret = ret > 0 ? 0 : ret;
4361 goto next_page;
4362 }
4363
826aa0a8
MX		 4364 err = write_page_nocow(nocow_ctx->sctx,
4365 physical_for_dev_replace, page);
4366 if (err)
4367 ret = err;
ff023aac 4368next_page:
826aa0a8
MX		 4369 unlock_page(page);
4370 page_cache_release(page);
4371
4372 if (ret)
4373 break;
4374
ff023aac
SB		 4375 offset += PAGE_CACHE_SIZE;
4376 physical_for_dev_replace += PAGE_CACHE_SIZE;
32159242 4377 nocow_ctx_logical += PAGE_CACHE_SIZE;
ff023aac
SB		 4378 len -= PAGE_CACHE_SIZE;
4379 }
652f25a2 4380 ret = COPY_COMPLETE;
826aa0a8 4381out:
edd1400b 4382 mutex_unlock(&inode->i_mutex);
826aa0a8 4383 iput(inode);
ff023aac
SB		 4384 return ret;
4385}
4386
4387static int write_page_nocow(struct scrub_ctx *sctx,
4388 u64 physical_for_dev_replace, struct page *page)
4389{
4390 struct bio *bio;
4391 struct btrfs_device *dev;
4392 int ret;
ff023aac
SB		 4393
4394 dev = sctx->wr_ctx.tgtdev;
4395 if (!dev)
4396 return -EIO;
4397 if (!dev->bdev) {
4398 printk_ratelimited(KERN_WARNING
efe120a0 4399 "BTRFS: scrub write_page_nocow(bdev == NULL) is unexpected!\n");
ff023aac
SB		 4400 return -EIO;
4401 }
9be3395b 4402 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
ff023aac
SB		 4403 if (!bio) {
4404 spin_lock(&sctx->stat_lock);
4405 sctx->stat.malloc_errors++;
4406 spin_unlock(&sctx->stat_lock);
4407 return -ENOMEM;
4408 }
4f024f37
KO		 4409 bio->bi_iter.bi_size = 0;
4410 bio->bi_iter.bi_sector = physical_for_dev_replace >> 9;
ff023aac
SB		 4411 bio->bi_bdev = dev->bdev;
4412 ret = bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
4413 if (ret != PAGE_CACHE_SIZE) {
4414leave_with_eio:
4415 bio_put(bio);
4416 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
4417 return -EIO;
4418 }
ff023aac 4419
33879d45 4420 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio))
ff023aac
SB		 4421 goto leave_with_eio;
4422
4423 bio_put(bio);
4424 return 0;
4425}
Ubuntu Artful kernel mirror