projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: fix filemap_flush call in btrfs_file_release
[deliverable/linux.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
AJ		 65
66struct scrub_page {
b5d67f64
SB		 67 struct scrub_block *sblock;
68 struct page *page;
442a4f63 69 struct btrfs_device *dev;
a2de733c
AJ		 70 u64 flags; /* extent flags */
71 u64 generation;
b5d67f64
SB		 72 u64 logical;
73 u64 physical;
ff023aac 74 u64 physical_for_dev_replace;
7a9e9987 75 atomic_t ref_count;
b5d67f64
SB		 76 struct {
77 unsigned int mirror_num:8;
78 unsigned int have_csum:1;
79 unsigned int io_error:1;
80 };
a2de733c
AJ		 81 u8 csum[BTRFS_CSUM_SIZE];
82};
83
84struct scrub_bio {
85 int index;
d9d181c1 86 struct scrub_ctx *sctx;
a36cf8b8 87 struct btrfs_device *dev;
a2de733c
AJ		 88 struct bio *bio;
89 int err;
90 u64 logical;
91 u64 physical;
ff023aac
SB		 92#if SCRUB_PAGES_PER_WR_BIO >= SCRUB_PAGES_PER_RD_BIO
93 struct scrub_page *pagev[SCRUB_PAGES_PER_WR_BIO];
94#else
95 struct scrub_page *pagev[SCRUB_PAGES_PER_RD_BIO];
96#endif
b5d67f64 97 int page_count;
a2de733c
AJ		 98 int next_free;
99 struct btrfs_work work;
100};
101
b5d67f64 102struct scrub_block {
7a9e9987 103 struct scrub_page *pagev[SCRUB_MAX_PAGES_PER_BLOCK];
b5d67f64
SB		 104 int page_count;
105 atomic_t outstanding_pages;
106 atomic_t ref_count; /* free mem on transition to zero */
d9d181c1 107 struct scrub_ctx *sctx;
b5d67f64
SB		 108 struct {
109 unsigned int header_error:1;
110 unsigned int checksum_error:1;
111 unsigned int no_io_error_seen:1;
442a4f63 112 unsigned int generation_error:1; /* also sets header_error */
b5d67f64
SB		 113 };
114};
115
ff023aac
SB		 116struct scrub_wr_ctx {
117 struct scrub_bio *wr_curr_bio;
118 struct btrfs_device *tgtdev;
119 int pages_per_wr_bio; /* <= SCRUB_PAGES_PER_WR_BIO */
120 atomic_t flush_all_writes;
121 struct mutex wr_lock;
122};
123
d9d181c1 124struct scrub_ctx {
ff023aac 125 struct scrub_bio *bios[SCRUB_BIOS_PER_SCTX];
a36cf8b8 126 struct btrfs_root *dev_root;
a2de733c
AJ		 127 int first_free;
128 int curr;
b6bfebc1
SB		 129 atomic_t bios_in_flight;
130 atomic_t workers_pending;
a2de733c
AJ		 131 spinlock_t list_lock;
132 wait_queue_head_t list_wait;
133 u16 csum_size;
134 struct list_head csum_list;
135 atomic_t cancel_req;
8628764e 136 int readonly;
ff023aac 137 int pages_per_rd_bio;
b5d67f64
SB		 138 u32 sectorsize;
139 u32 nodesize;
140 u32 leafsize;
63a212ab
SB		 141
142 int is_dev_replace;
ff023aac 143 struct scrub_wr_ctx wr_ctx;
63a212ab 144
a2de733c
AJ		 145 /*
146 * statistics
147 */
148 struct btrfs_scrub_progress stat;
149 spinlock_t stat_lock;
150};
151
0ef8e451 152struct scrub_fixup_nodatasum {
d9d181c1 153 struct scrub_ctx *sctx;
a36cf8b8 154 struct btrfs_device *dev;
0ef8e451
JS		 155 u64 logical;
156 struct btrfs_root *root;
157 struct btrfs_work work;
158 int mirror_num;
159};
160
652f25a2
JB		 161struct scrub_nocow_inode {
162 u64 inum;
163 u64 offset;
164 u64 root;
165 struct list_head list;
166};
167
ff023aac
SB		 168struct scrub_copy_nocow_ctx {
169 struct scrub_ctx *sctx;
170 u64 logical;
171 u64 len;
172 int mirror_num;
173 u64 physical_for_dev_replace;
652f25a2 174 struct list_head inodes;
ff023aac
SB		 175 struct btrfs_work work;
176};
177
558540c1
JS		 178struct scrub_warning {
179 struct btrfs_path *path;
180 u64 extent_item_size;
181 char *scratch_buf;
182 char *msg_buf;
183 const char *errstr;
184 sector_t sector;
185 u64 logical;
186 struct btrfs_device *dev;
187 int msg_bufsize;
188 int scratch_bufsize;
189};
190
b5d67f64 191
b6bfebc1
SB		 192static void scrub_pending_bio_inc(struct scrub_ctx *sctx);
193static void scrub_pending_bio_dec(struct scrub_ctx *sctx);
194static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx);
195static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx);
b5d67f64 196static int scrub_handle_errored_block(struct scrub_block *sblock_to_check);
d9d181c1 197static int scrub_setup_recheck_block(struct scrub_ctx *sctx,
3ec706c8 198 struct btrfs_fs_info *fs_info,
ff023aac 199 struct scrub_block *original_sblock,
b5d67f64 200 u64 length, u64 logical,
ff023aac 201 struct scrub_block *sblocks_for_recheck);
34f5c8e9
SB		 202static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
203 struct scrub_block *sblock, int is_metadata,
204 int have_csum, u8 *csum, u64 generation,
205 u16 csum_size);
b5d67f64
SB		 206static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
207 struct scrub_block *sblock,
208 int is_metadata, int have_csum,
209 const u8 *csum, u64 generation,
210 u16 csum_size);
b5d67f64
SB		 211static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
212 struct scrub_block *sblock_good,
213 int force_write);
214static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
215 struct scrub_block *sblock_good,
216 int page_num, int force_write);
ff023aac
SB		 217static void scrub_write_block_to_dev_replace(struct scrub_block *sblock);
218static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
219 int page_num);
b5d67f64
SB		 220static int scrub_checksum_data(struct scrub_block *sblock);
221static int scrub_checksum_tree_block(struct scrub_block *sblock);
222static int scrub_checksum_super(struct scrub_block *sblock);
223static void scrub_block_get(struct scrub_block *sblock);
224static void scrub_block_put(struct scrub_block *sblock);
7a9e9987
SB		 225static void scrub_page_get(struct scrub_page *spage);
226static void scrub_page_put(struct scrub_page *spage);
ff023aac
SB		 227static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
228 struct scrub_page *spage);
d9d181c1 229static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
a36cf8b8 230 u64 physical, struct btrfs_device *dev, u64 flags,
ff023aac
SB		 231 u64 gen, int mirror_num, u8 *csum, int force,
232 u64 physical_for_dev_replace);
1623edeb 233static void scrub_bio_end_io(struct bio *bio, int err);
b5d67f64
SB		 234static void scrub_bio_end_io_worker(struct btrfs_work *work);
235static void scrub_block_complete(struct scrub_block *sblock);
ff023aac
SB		 236static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
237 u64 extent_logical, u64 extent_len,
238 u64 *extent_physical,
239 struct btrfs_device **extent_dev,
240 int *extent_mirror_num);
241static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
242 struct scrub_wr_ctx *wr_ctx,
243 struct btrfs_fs_info *fs_info,
244 struct btrfs_device *dev,
245 int is_dev_replace);
246static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx);
247static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
248 struct scrub_page *spage);
249static void scrub_wr_submit(struct scrub_ctx *sctx);
250static void scrub_wr_bio_end_io(struct bio *bio, int err);
251static void scrub_wr_bio_end_io_worker(struct btrfs_work *work);
252static int write_page_nocow(struct scrub_ctx *sctx,
253 u64 physical_for_dev_replace, struct page *page);
254static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
652f25a2 255 struct scrub_copy_nocow_ctx *ctx);
ff023aac
SB		 256static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
257 int mirror_num, u64 physical_for_dev_replace);
258static void copy_nocow_pages_worker(struct btrfs_work *work);
cb7ab021 259static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
3cb0929a 260static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
1623edeb
SB		 261
262
b6bfebc1
SB		 263static void scrub_pending_bio_inc(struct scrub_ctx *sctx)
264{
265 atomic_inc(&sctx->bios_in_flight);
266}
267
268static void scrub_pending_bio_dec(struct scrub_ctx *sctx)
269{
270 atomic_dec(&sctx->bios_in_flight);
271 wake_up(&sctx->list_wait);
272}
273
cb7ab021 274static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
3cb0929a
WS		 275{
276 while (atomic_read(&fs_info->scrub_pause_req)) {
277 mutex_unlock(&fs_info->scrub_lock);
278 wait_event(fs_info->scrub_pause_wait,
279 atomic_read(&fs_info->scrub_pause_req) == 0);
280 mutex_lock(&fs_info->scrub_lock);
281 }
282}
283
cb7ab021
WS		 284static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
285{
286 atomic_inc(&fs_info->scrubs_paused);
287 wake_up(&fs_info->scrub_pause_wait);
288
289 mutex_lock(&fs_info->scrub_lock);
290 __scrub_blocked_if_needed(fs_info);
291 atomic_dec(&fs_info->scrubs_paused);
292 mutex_unlock(&fs_info->scrub_lock);
293
294 wake_up(&fs_info->scrub_pause_wait);
295}
296
b6bfebc1
SB		 297/*
298 * used for workers that require transaction commits (i.e., for the
299 * NOCOW case)
300 */
301static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx)
302{
303 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
304
305 /*
306 * increment scrubs_running to prevent cancel requests from
307 * completing as long as a worker is running. we must also
308 * increment scrubs_paused to prevent deadlocking on pause
309 * requests used for transactions commits (as the worker uses a
310 * transaction context). it is safe to regard the worker
311 * as paused for all matters practical. effectively, we only
312 * avoid cancellation requests from completing.
313 */
314 mutex_lock(&fs_info->scrub_lock);
315 atomic_inc(&fs_info->scrubs_running);
316 atomic_inc(&fs_info->scrubs_paused);
317 mutex_unlock(&fs_info->scrub_lock);
32a44789
WS		 318
319 /*
320 * check if @scrubs_running=@scrubs_paused condition
321 * inside wait_event() is not an atomic operation.
322 * which means we may inc/dec @scrub_running/paused
323 * at any time. Let's wake up @scrub_pause_wait as
324 * much as we can to let commit transaction blocked less.
325 */
326 wake_up(&fs_info->scrub_pause_wait);
327
b6bfebc1
SB		 328 atomic_inc(&sctx->workers_pending);
329}
330
331/* used for workers that require transaction commits */
332static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx)
333{
334 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
335
336 /*
337 * see scrub_pending_trans_workers_inc() why we're pretending
338 * to be paused in the scrub counters
339 */
340 mutex_lock(&fs_info->scrub_lock);
341 atomic_dec(&fs_info->scrubs_running);
342 atomic_dec(&fs_info->scrubs_paused);
343 mutex_unlock(&fs_info->scrub_lock);
344 atomic_dec(&sctx->workers_pending);
345 wake_up(&fs_info->scrub_pause_wait);
346 wake_up(&sctx->list_wait);
347}
348
d9d181c1 349static void scrub_free_csums(struct scrub_ctx *sctx)
a2de733c 350{
d9d181c1 351 while (!list_empty(&sctx->csum_list)) {
a2de733c 352 struct btrfs_ordered_sum *sum;
d9d181c1 353 sum = list_first_entry(&sctx->csum_list,
a2de733c
AJ		 354 struct btrfs_ordered_sum, list);
355 list_del(&sum->list);
356 kfree(sum);
357 }
358}
359
d9d181c1 360static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx)
a2de733c
AJ		 361{
362 int i;
a2de733c 363
d9d181c1 364 if (!sctx)
a2de733c
AJ		 365 return;
366
ff023aac
SB		 367 scrub_free_wr_ctx(&sctx->wr_ctx);
368
b5d67f64 369 /* this can happen when scrub is cancelled */
d9d181c1
SB		 370 if (sctx->curr != -1) {
371 struct scrub_bio *sbio = sctx->bios[sctx->curr];
b5d67f64
SB		 372
373 for (i = 0; i < sbio->page_count; i++) {
ff023aac 374 WARN_ON(!sbio->pagev[i]->page);
b5d67f64
SB		 375 scrub_block_put(sbio->pagev[i]->sblock);
376 }
377 bio_put(sbio->bio);
378 }
379
ff023aac 380 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
d9d181c1 381 struct scrub_bio *sbio = sctx->bios[i];
a2de733c
AJ		 382
383 if (!sbio)
384 break;
a2de733c
AJ		 385 kfree(sbio);
386 }
387
d9d181c1
SB		 388 scrub_free_csums(sctx);
389 kfree(sctx);
a2de733c
AJ		 390}
391
392static noinline_for_stack
63a212ab 393struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace)
a2de733c 394{
d9d181c1 395 struct scrub_ctx *sctx;
a2de733c 396 int i;
a2de733c 397 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
ff023aac
SB		 398 int pages_per_rd_bio;
399 int ret;
a2de733c 400
ff023aac
SB		 401 /*
402 * the setting of pages_per_rd_bio is correct for scrub but might
403 * be wrong for the dev_replace code where we might read from
404 * different devices in the initial huge bios. However, that
405 * code is able to correctly handle the case when adding a page
406 * to a bio fails.
407 */
408 if (dev->bdev)
409 pages_per_rd_bio = min_t(int, SCRUB_PAGES_PER_RD_BIO,
410 bio_get_nr_vecs(dev->bdev));
411 else
412 pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO;
d9d181c1
SB		 413 sctx = kzalloc(sizeof(*sctx), GFP_NOFS);
414 if (!sctx)
a2de733c 415 goto nomem;
63a212ab 416 sctx->is_dev_replace = is_dev_replace;
ff023aac 417 sctx->pages_per_rd_bio = pages_per_rd_bio;
d9d181c1 418 sctx->curr = -1;
a36cf8b8 419 sctx->dev_root = dev->dev_root;
ff023aac 420 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
a2de733c
AJ		 421 struct scrub_bio *sbio;
422
423 sbio = kzalloc(sizeof(*sbio), GFP_NOFS);
424 if (!sbio)
425 goto nomem;
d9d181c1 426 sctx->bios[i] = sbio;
a2de733c 427
a2de733c 428 sbio->index = i;
d9d181c1 429 sbio->sctx = sctx;
b5d67f64 430 sbio->page_count = 0;
0339ef2f
QW		 431 btrfs_init_work(&sbio->work, scrub_bio_end_io_worker,
432 NULL, NULL);
a2de733c 433
ff023aac 434 if (i != SCRUB_BIOS_PER_SCTX - 1)
d9d181c1 435 sctx->bios[i]->next_free = i + 1;
0ef8e451 436 else
d9d181c1
SB		 437 sctx->bios[i]->next_free = -1;
438 }
439 sctx->first_free = 0;
440 sctx->nodesize = dev->dev_root->nodesize;
441 sctx->leafsize = dev->dev_root->leafsize;
442 sctx->sectorsize = dev->dev_root->sectorsize;
b6bfebc1
SB		 443 atomic_set(&sctx->bios_in_flight, 0);
444 atomic_set(&sctx->workers_pending, 0);
d9d181c1
SB		 445 atomic_set(&sctx->cancel_req, 0);
446 sctx->csum_size = btrfs_super_csum_size(fs_info->super_copy);
447 INIT_LIST_HEAD(&sctx->csum_list);
448
449 spin_lock_init(&sctx->list_lock);
450 spin_lock_init(&sctx->stat_lock);
451 init_waitqueue_head(&sctx->list_wait);
ff023aac
SB		 452
453 ret = scrub_setup_wr_ctx(sctx, &sctx->wr_ctx, fs_info,
454 fs_info->dev_replace.tgtdev, is_dev_replace);
455 if (ret) {
456 scrub_free_ctx(sctx);
457 return ERR_PTR(ret);
458 }
d9d181c1 459 return sctx;
a2de733c
AJ		 460
461nomem:
d9d181c1 462 scrub_free_ctx(sctx);
a2de733c
AJ		 463 return ERR_PTR(-ENOMEM);
464}
465
ff023aac
SB		 466static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root,
467 void *warn_ctx)
558540c1
JS		 468{
469 u64 isize;
470 u32 nlink;
471 int ret;
472 int i;
473 struct extent_buffer *eb;
474 struct btrfs_inode_item *inode_item;
ff023aac 475 struct scrub_warning *swarn = warn_ctx;
558540c1
JS		 476 struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info;
477 struct inode_fs_paths *ipath = NULL;
478 struct btrfs_root *local_root;
479 struct btrfs_key root_key;
480
481 root_key.objectid = root;
482 root_key.type = BTRFS_ROOT_ITEM_KEY;
483 root_key.offset = (u64)-1;
484 local_root = btrfs_read_fs_root_no_name(fs_info, &root_key);
485 if (IS_ERR(local_root)) {
486 ret = PTR_ERR(local_root);
487 goto err;
488 }
489
490 ret = inode_item_info(inum, 0, local_root, swarn->path);
491 if (ret) {
492 btrfs_release_path(swarn->path);
493 goto err;
494 }
495
496 eb = swarn->path->nodes[0];
497 inode_item = btrfs_item_ptr(eb, swarn->path->slots[0],
498 struct btrfs_inode_item);
499 isize = btrfs_inode_size(eb, inode_item);
500 nlink = btrfs_inode_nlink(eb, inode_item);
501 btrfs_release_path(swarn->path);
502
503 ipath = init_ipath(4096, local_root, swarn->path);
26bdef54
DC		 504 if (IS_ERR(ipath)) {
505 ret = PTR_ERR(ipath);
506 ipath = NULL;
507 goto err;
508 }
558540c1
JS		 509 ret = paths_from_inode(inum, ipath);
510
511 if (ret < 0)
512 goto err;
513
514 /*
515 * we deliberately ignore the bit ipath might have been too small to
516 * hold all of the paths here
517 */
518 for (i = 0; i < ipath->fspath->elem_cnt; ++i)
efe120a0 519 printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
558540c1
JS		 520 "%s, sector %llu, root %llu, inode %llu, offset %llu, "
521 "length %llu, links %u (path: %s)\n", swarn->errstr,
606686ee 522 swarn->logical, rcu_str_deref(swarn->dev->name),
558540c1
JS		 523 (unsigned long long)swarn->sector, root, inum, offset,
524 min(isize - offset, (u64)PAGE_SIZE), nlink,
745c4d8e 525 (char *)(unsigned long)ipath->fspath->val[i]);
558540c1
JS		 526
527 free_ipath(ipath);
528 return 0;
529
530err:
efe120a0 531 printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
558540c1
JS		 532 "%s, sector %llu, root %llu, inode %llu, offset %llu: path "
533 "resolving failed with ret=%d\n", swarn->errstr,
606686ee 534 swarn->logical, rcu_str_deref(swarn->dev->name),
558540c1
JS		 535 (unsigned long long)swarn->sector, root, inum, offset, ret);
536
537 free_ipath(ipath);
538 return 0;
539}
540
b5d67f64 541static void scrub_print_warning(const char *errstr, struct scrub_block *sblock)
558540c1 542{
a36cf8b8
SB		 543 struct btrfs_device *dev;
544 struct btrfs_fs_info *fs_info;
558540c1
JS		 545 struct btrfs_path *path;
546 struct btrfs_key found_key;
547 struct extent_buffer *eb;
548 struct btrfs_extent_item *ei;
549 struct scrub_warning swarn;
69917e43
LB		 550 unsigned long ptr = 0;
551 u64 extent_item_pos;
552 u64 flags = 0;
558540c1 553 u64 ref_root;
69917e43 554 u32 item_size;
558540c1 555 u8 ref_level;
558540c1 556 const int bufsize = 4096;
69917e43 557 int ret;
558540c1 558
a36cf8b8 559 WARN_ON(sblock->page_count < 1);
7a9e9987 560 dev = sblock->pagev[0]->dev;
a36cf8b8
SB		 561 fs_info = sblock->sctx->dev_root->fs_info;
562
558540c1
JS		 563 path = btrfs_alloc_path();
564
565 swarn.scratch_buf = kmalloc(bufsize, GFP_NOFS);
566 swarn.msg_buf = kmalloc(bufsize, GFP_NOFS);
7a9e9987
SB		 567 swarn.sector = (sblock->pagev[0]->physical) >> 9;
568 swarn.logical = sblock->pagev[0]->logical;
558540c1 569 swarn.errstr = errstr;
a36cf8b8 570 swarn.dev = NULL;
558540c1
JS		 571 swarn.msg_bufsize = bufsize;
572 swarn.scratch_bufsize = bufsize;
573
574 if (!path || !swarn.scratch_buf || !swarn.msg_buf)
575 goto out;
576
69917e43
LB		 577 ret = extent_from_logical(fs_info, swarn.logical, path, &found_key,
578 &flags);
558540c1
JS		 579 if (ret < 0)
580 goto out;
581
4692cf58 582 extent_item_pos = swarn.logical - found_key.objectid;
558540c1
JS		 583 swarn.extent_item_size = found_key.offset;
584
585 eb = path->nodes[0];
586 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
587 item_size = btrfs_item_size_nr(eb, path->slots[0]);
588
69917e43 589 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
558540c1 590 do {
6eda71d0
LB		 591 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
592 item_size, &ref_root,
593 &ref_level);
606686ee 594 printk_in_rcu(KERN_WARNING
efe120a0 595 "BTRFS: %s at logical %llu on dev %s, "
558540c1 596 "sector %llu: metadata %s (level %d) in tree "
606686ee
JB		 597 "%llu\n", errstr, swarn.logical,
598 rcu_str_deref(dev->name),
558540c1
JS		 599 (unsigned long long)swarn.sector,
600 ref_level ? "node" : "leaf",
601 ret < 0 ? -1 : ref_level,
602 ret < 0 ? -1 : ref_root);
603 } while (ret != 1);
d8fe29e9 604 btrfs_release_path(path);
558540c1 605 } else {
d8fe29e9 606 btrfs_release_path(path);
558540c1 607 swarn.path = path;
a36cf8b8 608 swarn.dev = dev;
7a3ae2f8
JS		 609 iterate_extent_inodes(fs_info, found_key.objectid,
610 extent_item_pos, 1,
558540c1
JS		 611 scrub_print_warning_inode, &swarn);
612 }
613
614out:
615 btrfs_free_path(path);
616 kfree(swarn.scratch_buf);
617 kfree(swarn.msg_buf);
618}
619
ff023aac 620static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx)
0ef8e451 621{
5da6fcbc 622 struct page *page = NULL;
0ef8e451 623 unsigned long index;
ff023aac 624 struct scrub_fixup_nodatasum *fixup = fixup_ctx;
0ef8e451 625 int ret;
5da6fcbc 626 int corrected = 0;
0ef8e451 627 struct btrfs_key key;
5da6fcbc 628 struct inode *inode = NULL;
6f1c3605 629 struct btrfs_fs_info *fs_info;
0ef8e451
JS		 630 u64 end = offset + PAGE_SIZE - 1;
631 struct btrfs_root *local_root;
6f1c3605 632 int srcu_index;
0ef8e451
JS		 633
634 key.objectid = root;
635 key.type = BTRFS_ROOT_ITEM_KEY;
636 key.offset = (u64)-1;
6f1c3605
LB		 637
638 fs_info = fixup->root->fs_info;
639 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
640
641 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
642 if (IS_ERR(local_root)) {
643 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
0ef8e451 644 return PTR_ERR(local_root);
6f1c3605 645 }
0ef8e451
JS		 646
647 key.type = BTRFS_INODE_ITEM_KEY;
648 key.objectid = inum;
649 key.offset = 0;
6f1c3605
LB		 650 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
651 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
0ef8e451
JS		 652 if (IS_ERR(inode))
653 return PTR_ERR(inode);
654
0ef8e451
JS		 655 index = offset >> PAGE_CACHE_SHIFT;
656
657 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
5da6fcbc
JS		 658 if (!page) {
659 ret = -ENOMEM;
660 goto out;
661 }
662
663 if (PageUptodate(page)) {
5da6fcbc
JS		 664 if (PageDirty(page)) {
665 /*
666 * we need to write the data to the defect sector. the
667 * data that was in that sector is not in memory,
668 * because the page was modified. we must not write the
669 * modified page to that sector.
670 *
671 * TODO: what could be done here: wait for the delalloc
672 * runner to write out that page (might involve
673 * COW) and see whether the sector is still
674 * referenced afterwards.
675 *
676 * For the meantime, we'll treat this error
677 * incorrectable, although there is a chance that a
678 * later scrub will find the bad sector again and that
679 * there's no dirty page in memory, then.
680 */
681 ret = -EIO;
682 goto out;
683 }
3ec706c8
SB		 684 fs_info = BTRFS_I(inode)->root->fs_info;
685 ret = repair_io_failure(fs_info, offset, PAGE_SIZE,
5da6fcbc
JS		 686 fixup->logical, page,
687 fixup->mirror_num);
688 unlock_page(page);
689 corrected = !ret;
690 } else {
691 /*
692 * we need to get good data first. the general readpage path
693 * will call repair_io_failure for us, we just have to make
694 * sure we read the bad mirror.
695 */
696 ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
697 EXTENT_DAMAGED, GFP_NOFS);
698 if (ret) {
699 /* set_extent_bits should give proper error */
700 WARN_ON(ret > 0);
701 if (ret > 0)
702 ret = -EFAULT;
703 goto out;
704 }
705
706 ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page,
707 btrfs_get_extent,
708 fixup->mirror_num);
709 wait_on_page_locked(page);
710
711 corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset,
712 end, EXTENT_DAMAGED, 0, NULL);
713 if (!corrected)
714 clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
715 EXTENT_DAMAGED, GFP_NOFS);
716 }
717
718out:
719 if (page)
720 put_page(page);
7fb18a06
TK		 721
722 iput(inode);
0ef8e451
JS		 723
724 if (ret < 0)
725 return ret;
726
727 if (ret == 0 && corrected) {
728 /*
729 * we only need to call readpage for one of the inodes belonging
730 * to this extent. so make iterate_extent_inodes stop
731 */
732 return 1;
733 }
734
735 return -EIO;
736}
737
738static void scrub_fixup_nodatasum(struct btrfs_work *work)
739{
740 int ret;
741 struct scrub_fixup_nodatasum *fixup;
d9d181c1 742 struct scrub_ctx *sctx;
0ef8e451 743 struct btrfs_trans_handle *trans = NULL;
0ef8e451
JS		 744 struct btrfs_path *path;
745 int uncorrectable = 0;
746
747 fixup = container_of(work, struct scrub_fixup_nodatasum, work);
d9d181c1 748 sctx = fixup->sctx;
0ef8e451
JS		 749
750 path = btrfs_alloc_path();
751 if (!path) {
d9d181c1
SB		 752 spin_lock(&sctx->stat_lock);
753 ++sctx->stat.malloc_errors;
754 spin_unlock(&sctx->stat_lock);
0ef8e451
JS		 755 uncorrectable = 1;
756 goto out;
757 }
758
759 trans = btrfs_join_transaction(fixup->root);
760 if (IS_ERR(trans)) {
761 uncorrectable = 1;
762 goto out;
763 }
764
765 /*
766 * the idea is to trigger a regular read through the standard path. we
767 * read a page from the (failed) logical address by specifying the
768 * corresponding copynum of the failed sector. thus, that readpage is
769 * expected to fail.
770 * that is the point where on-the-fly error correction will kick in
771 * (once it's finished) and rewrite the failed sector if a good copy
772 * can be found.
773 */
774 ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info,
775 path, scrub_fixup_readpage,
776 fixup);
777 if (ret < 0) {
778 uncorrectable = 1;
779 goto out;
780 }
781 WARN_ON(ret != 1);
782
d9d181c1
SB		 783 spin_lock(&sctx->stat_lock);
784 ++sctx->stat.corrected_errors;
785 spin_unlock(&sctx->stat_lock);
0ef8e451
JS		 786
787out:
788 if (trans && !IS_ERR(trans))
789 btrfs_end_transaction(trans, fixup->root);
790 if (uncorrectable) {
d9d181c1
SB		 791 spin_lock(&sctx->stat_lock);
792 ++sctx->stat.uncorrectable_errors;
793 spin_unlock(&sctx->stat_lock);
ff023aac
SB		 794 btrfs_dev_replace_stats_inc(
795 &sctx->dev_root->fs_info->dev_replace.
796 num_uncorrectable_read_errors);
efe120a0
FH		 797 printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
798 "unable to fixup (nodatasum) error at logical %llu on dev %s\n",
c1c9ff7c 799 fixup->logical, rcu_str_deref(fixup->dev->name));
0ef8e451
JS		 800 }
801
802 btrfs_free_path(path);
803 kfree(fixup);
804
b6bfebc1 805 scrub_pending_trans_workers_dec(sctx);
0ef8e451
JS		 806}
807
a2de733c 808/*
b5d67f64
SB		 809 * scrub_handle_errored_block gets called when either verification of the
810 * pages failed or the bio failed to read, e.g. with EIO. In the latter
811 * case, this function handles all pages in the bio, even though only one
812 * may be bad.
813 * The goal of this function is to repair the errored block by using the
814 * contents of one of the mirrors.
a2de733c 815 */
b5d67f64 816static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
a2de733c 817{
d9d181c1 818 struct scrub_ctx *sctx = sblock_to_check->sctx;
a36cf8b8 819 struct btrfs_device *dev;
b5d67f64
SB		 820 struct btrfs_fs_info *fs_info;
821 u64 length;
822 u64 logical;
823 u64 generation;
824 unsigned int failed_mirror_index;
825 unsigned int is_metadata;
826 unsigned int have_csum;
827 u8 *csum;
828 struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */
829 struct scrub_block *sblock_bad;
830 int ret;
831 int mirror_index;
832 int page_num;
833 int success;
558540c1 834 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
b5d67f64
SB		 835 DEFAULT_RATELIMIT_BURST);
836
837 BUG_ON(sblock_to_check->page_count < 1);
a36cf8b8 838 fs_info = sctx->dev_root->fs_info;
4ded4f63
SB		 839 if (sblock_to_check->pagev[0]->flags & BTRFS_EXTENT_FLAG_SUPER) {
840 /*
841 * if we find an error in a super block, we just report it.
842 * They will get written with the next transaction commit
843 * anyway
844 */
845 spin_lock(&sctx->stat_lock);
846 ++sctx->stat.super_errors;
847 spin_unlock(&sctx->stat_lock);
848 return 0;
849 }
b5d67f64 850 length = sblock_to_check->page_count * PAGE_SIZE;
7a9e9987
SB		 851 logical = sblock_to_check->pagev[0]->logical;
852 generation = sblock_to_check->pagev[0]->generation;
853 BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1);
854 failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1;
855 is_metadata = !(sblock_to_check->pagev[0]->flags &
b5d67f64 856 BTRFS_EXTENT_FLAG_DATA);
7a9e9987
SB		 857 have_csum = sblock_to_check->pagev[0]->have_csum;
858 csum = sblock_to_check->pagev[0]->csum;
859 dev = sblock_to_check->pagev[0]->dev;
13db62b7 860
ff023aac
SB		 861 if (sctx->is_dev_replace && !is_metadata && !have_csum) {
862 sblocks_for_recheck = NULL;
863 goto nodatasum_case;
864 }
865
b5d67f64
SB		 866 /*
867 * read all mirrors one after the other. This includes to
868 * re-read the extent or metadata block that failed (that was
869 * the cause that this fixup code is called) another time,
870 * page by page this time in order to know which pages
871 * caused I/O errors and which ones are good (for all mirrors).
872 * It is the goal to handle the situation when more than one
873 * mirror contains I/O errors, but the errors do not
874 * overlap, i.e. the data can be repaired by selecting the
875 * pages from those mirrors without I/O error on the
876 * particular pages. One example (with blocks >= 2 * PAGE_SIZE)
877 * would be that mirror #1 has an I/O error on the first page,
878 * the second page is good, and mirror #2 has an I/O error on
879 * the second page, but the first page is good.
880 * Then the first page of the first mirror can be repaired by
881 * taking the first page of the second mirror, and the
882 * second page of the second mirror can be repaired by
883 * copying the contents of the 2nd page of the 1st mirror.
884 * One more note: if the pages of one mirror contain I/O
885 * errors, the checksum cannot be verified. In order to get
886 * the best data for repairing, the first attempt is to find
887 * a mirror without I/O errors and with a validated checksum.
888 * Only if this is not possible, the pages are picked from
889 * mirrors with I/O errors without considering the checksum.
890 * If the latter is the case, at the end, the checksum of the
891 * repaired area is verified in order to correctly maintain
892 * the statistics.
893 */
894
895 sblocks_for_recheck = kzalloc(BTRFS_MAX_MIRRORS *
896 sizeof(*sblocks_for_recheck),
897 GFP_NOFS);
898 if (!sblocks_for_recheck) {
d9d181c1
SB		 899 spin_lock(&sctx->stat_lock);
900 sctx->stat.malloc_errors++;
901 sctx->stat.read_errors++;
902 sctx->stat.uncorrectable_errors++;
903 spin_unlock(&sctx->stat_lock);
a36cf8b8 904 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
b5d67f64 905 goto out;
a2de733c
AJ		 906 }
907
b5d67f64 908 /* setup the context, map the logical blocks and alloc the pages */
ff023aac 909 ret = scrub_setup_recheck_block(sctx, fs_info, sblock_to_check, length,
b5d67f64
SB		 910 logical, sblocks_for_recheck);
911 if (ret) {
d9d181c1
SB		 912 spin_lock(&sctx->stat_lock);
913 sctx->stat.read_errors++;
914 sctx->stat.uncorrectable_errors++;
915 spin_unlock(&sctx->stat_lock);
a36cf8b8 916 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
b5d67f64
SB		 917 goto out;
918 }
919 BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS);
920 sblock_bad = sblocks_for_recheck + failed_mirror_index;
13db62b7 921
b5d67f64 922 /* build and submit the bios for the failed mirror, check checksums */
34f5c8e9
SB		 923 scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum,
924 csum, generation, sctx->csum_size);
a2de733c 925
b5d67f64
SB		 926 if (!sblock_bad->header_error && !sblock_bad->checksum_error &&
927 sblock_bad->no_io_error_seen) {
928 /*
929 * the error disappeared after reading page by page, or
930 * the area was part of a huge bio and other parts of the
931 * bio caused I/O errors, or the block layer merged several
932 * read requests into one and the error is caused by a
933 * different bio (usually one of the two latter cases is
934 * the cause)
935 */
d9d181c1
SB		 936 spin_lock(&sctx->stat_lock);
937 sctx->stat.unverified_errors++;
938 spin_unlock(&sctx->stat_lock);
a2de733c 939
ff023aac
SB		 940 if (sctx->is_dev_replace)
941 scrub_write_block_to_dev_replace(sblock_bad);
b5d67f64 942 goto out;
a2de733c 943 }
a2de733c 944
b5d67f64 945 if (!sblock_bad->no_io_error_seen) {
d9d181c1
SB		 946 spin_lock(&sctx->stat_lock);
947 sctx->stat.read_errors++;
948 spin_unlock(&sctx->stat_lock);
b5d67f64
SB		 949 if (__ratelimit(&_rs))
950 scrub_print_warning("i/o error", sblock_to_check);
a36cf8b8 951 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
b5d67f64 952 } else if (sblock_bad->checksum_error) {
d9d181c1
SB		 953 spin_lock(&sctx->stat_lock);
954 sctx->stat.csum_errors++;
955 spin_unlock(&sctx->stat_lock);
b5d67f64
SB		 956 if (__ratelimit(&_rs))
957 scrub_print_warning("checksum error", sblock_to_check);
a36cf8b8 958 btrfs_dev_stat_inc_and_print(dev,
442a4f63 959 BTRFS_DEV_STAT_CORRUPTION_ERRS);
b5d67f64 960 } else if (sblock_bad->header_error) {
d9d181c1
SB		 961 spin_lock(&sctx->stat_lock);
962 sctx->stat.verify_errors++;
963 spin_unlock(&sctx->stat_lock);
b5d67f64
SB		 964 if (__ratelimit(&_rs))
965 scrub_print_warning("checksum/header error",
966 sblock_to_check);
442a4f63 967 if (sblock_bad->generation_error)
a36cf8b8 968 btrfs_dev_stat_inc_and_print(dev,
442a4f63
SB		 969 BTRFS_DEV_STAT_GENERATION_ERRS);
970 else
a36cf8b8 971 btrfs_dev_stat_inc_and_print(dev,
442a4f63 972 BTRFS_DEV_STAT_CORRUPTION_ERRS);
b5d67f64 973 }
a2de733c 974
33ef30ad
ID		 975 if (sctx->readonly) {
976 ASSERT(!sctx->is_dev_replace);
977 goto out;
978 }
a2de733c 979
b5d67f64
SB		 980 if (!is_metadata && !have_csum) {
981 struct scrub_fixup_nodatasum *fixup_nodatasum;
a2de733c 982
ff023aac
SB		 983nodatasum_case:
984 WARN_ON(sctx->is_dev_replace);
985
b5d67f64
SB		 986 /*
987 * !is_metadata and !have_csum, this means that the data
988 * might not be COW'ed, that it might be modified
989 * concurrently. The general strategy to work on the
990 * commit root does not help in the case when COW is not
991 * used.
992 */
993 fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS);
994 if (!fixup_nodatasum)
995 goto did_not_correct_error;
d9d181c1 996 fixup_nodatasum->sctx = sctx;
a36cf8b8 997 fixup_nodatasum->dev = dev;
b5d67f64
SB		 998 fixup_nodatasum->logical = logical;
999 fixup_nodatasum->root = fs_info->extent_root;
1000 fixup_nodatasum->mirror_num = failed_mirror_index + 1;
b6bfebc1 1001 scrub_pending_trans_workers_inc(sctx);
0339ef2f
QW		 1002 btrfs_init_work(&fixup_nodatasum->work, scrub_fixup_nodatasum,
1003 NULL, NULL);
1004 btrfs_queue_work(fs_info->scrub_workers,
1005 &fixup_nodatasum->work);
b5d67f64 1006 goto out;
a2de733c
AJ		 1007 }
1008
b5d67f64
SB		 1009 /*
1010 * now build and submit the bios for the other mirrors, check
cb2ced73
SB		 1011 * checksums.
1012 * First try to pick the mirror which is completely without I/O
b5d67f64
SB		 1013 * errors and also does not have a checksum error.
1014 * If one is found, and if a checksum is present, the full block
1015 * that is known to contain an error is rewritten. Afterwards
1016 * the block is known to be corrected.
1017 * If a mirror is found which is completely correct, and no
1018 * checksum is present, only those pages are rewritten that had
1019 * an I/O error in the block to be repaired, since it cannot be
1020 * determined, which copy of the other pages is better (and it
1021 * could happen otherwise that a correct page would be
1022 * overwritten by a bad one).
1023 */
1024 for (mirror_index = 0;
1025 mirror_index < BTRFS_MAX_MIRRORS &&
1026 sblocks_for_recheck[mirror_index].page_count > 0;
1027 mirror_index++) {
cb2ced73 1028 struct scrub_block *sblock_other;
b5d67f64 1029
cb2ced73
SB		 1030 if (mirror_index == failed_mirror_index)
1031 continue;
1032 sblock_other = sblocks_for_recheck + mirror_index;
1033
1034 /* build and submit the bios, check checksums */
34f5c8e9
SB		 1035 scrub_recheck_block(fs_info, sblock_other, is_metadata,
1036 have_csum, csum, generation,
1037 sctx->csum_size);
1038
1039 if (!sblock_other->header_error &&
b5d67f64
SB		 1040 !sblock_other->checksum_error &&
1041 sblock_other->no_io_error_seen) {
ff023aac
SB		 1042 if (sctx->is_dev_replace) {
1043 scrub_write_block_to_dev_replace(sblock_other);
1044 } else {
1045 int force_write = is_metadata || have_csum;
1046
1047 ret = scrub_repair_block_from_good_copy(
1048 sblock_bad, sblock_other,
1049 force_write);
1050 }
b5d67f64
SB		 1051 if (0 == ret)
1052 goto corrected_error;
1053 }
1054 }
a2de733c
AJ		 1055
1056 /*
ff023aac
SB		 1057 * for dev_replace, pick good pages and write to the target device.
1058 */
1059 if (sctx->is_dev_replace) {
1060 success = 1;
1061 for (page_num = 0; page_num < sblock_bad->page_count;
1062 page_num++) {
1063 int sub_success;
1064
1065 sub_success = 0;
1066 for (mirror_index = 0;
1067 mirror_index < BTRFS_MAX_MIRRORS &&
1068 sblocks_for_recheck[mirror_index].page_count > 0;
1069 mirror_index++) {
1070 struct scrub_block *sblock_other =
1071 sblocks_for_recheck + mirror_index;
1072 struct scrub_page *page_other =
1073 sblock_other->pagev[page_num];
1074
1075 if (!page_other->io_error) {
1076 ret = scrub_write_page_to_dev_replace(
1077 sblock_other, page_num);
1078 if (ret == 0) {
1079 /* succeeded for this page */
1080 sub_success = 1;
1081 break;
1082 } else {
1083 btrfs_dev_replace_stats_inc(
1084 &sctx->dev_root->
1085 fs_info->dev_replace.
1086 num_write_errors);
1087 }
1088 }
1089 }
1090
1091 if (!sub_success) {
1092 /*
1093 * did not find a mirror to fetch the page
1094 * from. scrub_write_page_to_dev_replace()
1095 * handles this case (page->io_error), by
1096 * filling the block with zeros before
1097 * submitting the write request
1098 */
1099 success = 0;
1100 ret = scrub_write_page_to_dev_replace(
1101 sblock_bad, page_num);
1102 if (ret)
1103 btrfs_dev_replace_stats_inc(
1104 &sctx->dev_root->fs_info->
1105 dev_replace.num_write_errors);
1106 }
1107 }
1108
1109 goto out;
1110 }
1111
1112 /*
1113 * for regular scrub, repair those pages that are errored.
1114 * In case of I/O errors in the area that is supposed to be
b5d67f64
SB		 1115 * repaired, continue by picking good copies of those pages.
1116 * Select the good pages from mirrors to rewrite bad pages from
1117 * the area to fix. Afterwards verify the checksum of the block
1118 * that is supposed to be repaired. This verification step is
1119 * only done for the purpose of statistic counting and for the
1120 * final scrub report, whether errors remain.
1121 * A perfect algorithm could make use of the checksum and try
1122 * all possible combinations of pages from the different mirrors
1123 * until the checksum verification succeeds. For example, when
1124 * the 2nd page of mirror #1 faces I/O errors, and the 2nd page
1125 * of mirror #2 is readable but the final checksum test fails,
1126 * then the 2nd page of mirror #3 could be tried, whether now
1127 * the final checksum succeedes. But this would be a rare
1128 * exception and is therefore not implemented. At least it is
1129 * avoided that the good copy is overwritten.
1130 * A more useful improvement would be to pick the sectors
1131 * without I/O error based on sector sizes (512 bytes on legacy
1132 * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one
1133 * mirror could be repaired by taking 512 byte of a different
1134 * mirror, even if other 512 byte sectors in the same PAGE_SIZE
1135 * area are unreadable.
a2de733c 1136 */
a2de733c 1137
b5d67f64
SB		 1138 /* can only fix I/O errors from here on */
1139 if (sblock_bad->no_io_error_seen)
1140 goto did_not_correct_error;
1141
1142 success = 1;
1143 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
7a9e9987 1144 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
b5d67f64
SB		 1145
1146 if (!page_bad->io_error)
a2de733c 1147 continue;
b5d67f64
SB		 1148
1149 for (mirror_index = 0;
1150 mirror_index < BTRFS_MAX_MIRRORS &&
1151 sblocks_for_recheck[mirror_index].page_count > 0;
1152 mirror_index++) {
1153 struct scrub_block *sblock_other = sblocks_for_recheck +
1154 mirror_index;
7a9e9987
SB		 1155 struct scrub_page *page_other = sblock_other->pagev[
1156 page_num];
b5d67f64
SB		 1157
1158 if (!page_other->io_error) {
1159 ret = scrub_repair_page_from_good_copy(
1160 sblock_bad, sblock_other, page_num, 0);
1161 if (0 == ret) {
1162 page_bad->io_error = 0;
1163 break; /* succeeded for this page */
1164 }
1165 }
96e36920 1166 }
a2de733c 1167
b5d67f64
SB		 1168 if (page_bad->io_error) {
1169 /* did not find a mirror to copy the page from */
1170 success = 0;
1171 }
a2de733c 1172 }
a2de733c 1173
b5d67f64
SB		 1174 if (success) {
1175 if (is_metadata || have_csum) {
1176 /*
1177 * need to verify the checksum now that all
1178 * sectors on disk are repaired (the write
1179 * request for data to be repaired is on its way).
1180 * Just be lazy and use scrub_recheck_block()
1181 * which re-reads the data before the checksum
1182 * is verified, but most likely the data comes out
1183 * of the page cache.
1184 */
34f5c8e9
SB		 1185 scrub_recheck_block(fs_info, sblock_bad,
1186 is_metadata, have_csum, csum,
1187 generation, sctx->csum_size);
1188 if (!sblock_bad->header_error &&
b5d67f64
SB		 1189 !sblock_bad->checksum_error &&
1190 sblock_bad->no_io_error_seen)
1191 goto corrected_error;
1192 else
1193 goto did_not_correct_error;
1194 } else {
1195corrected_error:
d9d181c1
SB		 1196 spin_lock(&sctx->stat_lock);
1197 sctx->stat.corrected_errors++;
1198 spin_unlock(&sctx->stat_lock);
606686ee 1199 printk_ratelimited_in_rcu(KERN_ERR
efe120a0 1200 "BTRFS: fixed up error at logical %llu on dev %s\n",
c1c9ff7c 1201 logical, rcu_str_deref(dev->name));
8628764e 1202 }
b5d67f64
SB		 1203 } else {
1204did_not_correct_error:
d9d181c1
SB		 1205 spin_lock(&sctx->stat_lock);
1206 sctx->stat.uncorrectable_errors++;
1207 spin_unlock(&sctx->stat_lock);
606686ee 1208 printk_ratelimited_in_rcu(KERN_ERR
efe120a0 1209 "BTRFS: unable to fixup (regular) error at logical %llu on dev %s\n",
c1c9ff7c 1210 logical, rcu_str_deref(dev->name));
96e36920 1211 }
a2de733c 1212
b5d67f64
SB		 1213out:
1214 if (sblocks_for_recheck) {
1215 for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS;
1216 mirror_index++) {
1217 struct scrub_block *sblock = sblocks_for_recheck +
1218 mirror_index;
1219 int page_index;
1220
7a9e9987
SB		 1221 for (page_index = 0; page_index < sblock->page_count;
1222 page_index++) {
1223 sblock->pagev[page_index]->sblock = NULL;
1224 scrub_page_put(sblock->pagev[page_index]);
1225 }
b5d67f64
SB		 1226 }
1227 kfree(sblocks_for_recheck);
1228 }
a2de733c 1229
b5d67f64
SB		 1230 return 0;
1231}
a2de733c 1232
d9d181c1 1233static int scrub_setup_recheck_block(struct scrub_ctx *sctx,
3ec706c8 1234 struct btrfs_fs_info *fs_info,
ff023aac 1235 struct scrub_block *original_sblock,
b5d67f64
SB		 1236 u64 length, u64 logical,
1237 struct scrub_block *sblocks_for_recheck)
1238{
1239 int page_index;
1240 int mirror_index;
1241 int ret;
1242
1243 /*
7a9e9987 1244 * note: the two members ref_count and outstanding_pages
b5d67f64
SB		 1245 * are not used (and not set) in the blocks that are used for
1246 * the recheck procedure
1247 */
1248
1249 page_index = 0;
1250 while (length > 0) {
1251 u64 sublen = min_t(u64, length, PAGE_SIZE);
1252 u64 mapped_length = sublen;
1253 struct btrfs_bio *bbio = NULL;
a2de733c 1254
b5d67f64
SB		 1255 /*
1256 * with a length of PAGE_SIZE, each returned stripe
1257 * represents one mirror
1258 */
29a8d9a0
SB		 1259 ret = btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS, logical,
1260 &mapped_length, &bbio, 0);
b5d67f64
SB		 1261 if (ret || !bbio || mapped_length < sublen) {
1262 kfree(bbio);
1263 return -EIO;
1264 }
a2de733c 1265
ff023aac 1266 BUG_ON(page_index >= SCRUB_PAGES_PER_RD_BIO);
b5d67f64
SB		 1267 for (mirror_index = 0; mirror_index < (int)bbio->num_stripes;
1268 mirror_index++) {
1269 struct scrub_block *sblock;
1270 struct scrub_page *page;
1271
1272 if (mirror_index >= BTRFS_MAX_MIRRORS)
1273 continue;
1274
1275 sblock = sblocks_for_recheck + mirror_index;
7a9e9987
SB		 1276 sblock->sctx = sctx;
1277 page = kzalloc(sizeof(*page), GFP_NOFS);
1278 if (!page) {
1279leave_nomem:
d9d181c1
SB		 1280 spin_lock(&sctx->stat_lock);
1281 sctx->stat.malloc_errors++;
1282 spin_unlock(&sctx->stat_lock);
cf93dcce 1283 kfree(bbio);
b5d67f64
SB		 1284 return -ENOMEM;
1285 }
7a9e9987
SB		 1286 scrub_page_get(page);
1287 sblock->pagev[page_index] = page;
1288 page->logical = logical;
1289 page->physical = bbio->stripes[mirror_index].physical;
ff023aac
SB		 1290 BUG_ON(page_index >= original_sblock->page_count);
1291 page->physical_for_dev_replace =
1292 original_sblock->pagev[page_index]->
1293 physical_for_dev_replace;
7a9e9987
SB		 1294 /* for missing devices, dev->bdev is NULL */
1295 page->dev = bbio->stripes[mirror_index].dev;
1296 page->mirror_num = mirror_index + 1;
b5d67f64 1297 sblock->page_count++;
7a9e9987
SB		 1298 page->page = alloc_page(GFP_NOFS);
1299 if (!page->page)
1300 goto leave_nomem;
b5d67f64
SB		 1301 }
1302 kfree(bbio);
1303 length -= sublen;
1304 logical += sublen;
1305 page_index++;
1306 }
1307
1308 return 0;
96e36920
ID		 1309}
1310
b5d67f64
SB		 1311/*
1312 * this function will check the on disk data for checksum errors, header
1313 * errors and read I/O errors. If any I/O errors happen, the exact pages
1314 * which are errored are marked as being bad. The goal is to enable scrub
1315 * to take those pages that are not errored from all the mirrors so that
1316 * the pages that are errored in the just handled mirror can be repaired.
1317 */
34f5c8e9
SB		 1318static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
1319 struct scrub_block *sblock, int is_metadata,
1320 int have_csum, u8 *csum, u64 generation,
1321 u16 csum_size)
96e36920 1322{
b5d67f64 1323 int page_num;
96e36920 1324
b5d67f64
SB		 1325 sblock->no_io_error_seen = 1;
1326 sblock->header_error = 0;
1327 sblock->checksum_error = 0;
96e36920 1328
b5d67f64
SB		 1329 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1330 struct bio *bio;
7a9e9987 1331 struct scrub_page *page = sblock->pagev[page_num];
b5d67f64 1332
442a4f63 1333 if (page->dev->bdev == NULL) {
ea9947b4
SB		 1334 page->io_error = 1;
1335 sblock->no_io_error_seen = 0;
1336 continue;
1337 }
1338
7a9e9987 1339 WARN_ON(!page->page);
9be3395b 1340 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
34f5c8e9
SB		 1341 if (!bio) {
1342 page->io_error = 1;
1343 sblock->no_io_error_seen = 0;
1344 continue;
1345 }
442a4f63 1346 bio->bi_bdev = page->dev->bdev;
4f024f37 1347 bio->bi_iter.bi_sector = page->physical >> 9;
b5d67f64 1348
34f5c8e9 1349 bio_add_page(bio, page->page, PAGE_SIZE, 0);
33879d45 1350 if (btrfsic_submit_bio_wait(READ, bio))
b5d67f64 1351 sblock->no_io_error_seen = 0;
33879d45 1352
b5d67f64
SB		 1353 bio_put(bio);
1354 }
96e36920 1355
b5d67f64
SB		 1356 if (sblock->no_io_error_seen)
1357 scrub_recheck_block_checksum(fs_info, sblock, is_metadata,
1358 have_csum, csum, generation,
1359 csum_size);
1360
34f5c8e9 1361 return;
a2de733c
AJ		 1362}
1363
b5d67f64
SB		 1364static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
1365 struct scrub_block *sblock,
1366 int is_metadata, int have_csum,
1367 const u8 *csum, u64 generation,
1368 u16 csum_size)
a2de733c 1369{
b5d67f64
SB		 1370 int page_num;
1371 u8 calculated_csum[BTRFS_CSUM_SIZE];
1372 u32 crc = ~(u32)0;
b5d67f64
SB		 1373 void *mapped_buffer;
1374
7a9e9987 1375 WARN_ON(!sblock->pagev[0]->page);
b5d67f64
SB		 1376 if (is_metadata) {
1377 struct btrfs_header *h;
1378
7a9e9987 1379 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
b5d67f64
SB		 1380 h = (struct btrfs_header *)mapped_buffer;
1381
3cae210f 1382 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h) ||
b5d67f64
SB		 1383 memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE) ||
1384 memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
442a4f63 1385 BTRFS_UUID_SIZE)) {
b5d67f64 1386 sblock->header_error = 1;
3cae210f 1387 } else if (generation != btrfs_stack_header_generation(h)) {
442a4f63
SB		 1388 sblock->header_error = 1;
1389 sblock->generation_error = 1;
1390 }
b5d67f64
SB		 1391 csum = h->csum;
1392 } else {
1393 if (!have_csum)
1394 return;
a2de733c 1395
7a9e9987 1396 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
b5d67f64 1397 }
a2de733c 1398
b5d67f64
SB		 1399 for (page_num = 0;;) {
1400 if (page_num == 0 && is_metadata)
b0496686 1401 crc = btrfs_csum_data(
b5d67f64
SB		 1402 ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE,
1403 crc, PAGE_SIZE - BTRFS_CSUM_SIZE);
1404 else
b0496686 1405 crc = btrfs_csum_data(mapped_buffer, crc, PAGE_SIZE);
b5d67f64 1406
9613bebb 1407 kunmap_atomic(mapped_buffer);
b5d67f64
SB		 1408 page_num++;
1409 if (page_num >= sblock->page_count)
1410 break;
7a9e9987 1411 WARN_ON(!sblock->pagev[page_num]->page);
b5d67f64 1412
7a9e9987 1413 mapped_buffer = kmap_atomic(sblock->pagev[page_num]->page);
b5d67f64
SB		 1414 }
1415
1416 btrfs_csum_final(crc, calculated_csum);
1417 if (memcmp(calculated_csum, csum, csum_size))
1418 sblock->checksum_error = 1;
a2de733c
AJ		 1419}
1420
b5d67f64
SB		 1421static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
1422 struct scrub_block *sblock_good,
1423 int force_write)
1424{
1425 int page_num;
1426 int ret = 0;
96e36920 1427
b5d67f64
SB		 1428 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
1429 int ret_sub;
96e36920 1430
b5d67f64
SB		 1431 ret_sub = scrub_repair_page_from_good_copy(sblock_bad,
1432 sblock_good,
1433 page_num,
1434 force_write);
1435 if (ret_sub)
1436 ret = ret_sub;
a2de733c 1437 }
b5d67f64
SB		 1438
1439 return ret;
1440}
1441
1442static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
1443 struct scrub_block *sblock_good,
1444 int page_num, int force_write)
1445{
7a9e9987
SB		 1446 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1447 struct scrub_page *page_good = sblock_good->pagev[page_num];
b5d67f64 1448
7a9e9987
SB		 1449 BUG_ON(page_bad->page == NULL);
1450 BUG_ON(page_good->page == NULL);
b5d67f64
SB		 1451 if (force_write || sblock_bad->header_error ||
1452 sblock_bad->checksum_error || page_bad->io_error) {
1453 struct bio *bio;
1454 int ret;
b5d67f64 1455
ff023aac 1456 if (!page_bad->dev->bdev) {
efe120a0
FH		 1457 printk_ratelimited(KERN_WARNING "BTRFS: "
1458 "scrub_repair_page_from_good_copy(bdev == NULL) "
1459 "is unexpected!\n");
ff023aac
SB		 1460 return -EIO;
1461 }
1462
9be3395b 1463 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
e627ee7b
TI		 1464 if (!bio)
1465 return -EIO;
442a4f63 1466 bio->bi_bdev = page_bad->dev->bdev;
4f024f37 1467 bio->bi_iter.bi_sector = page_bad->physical >> 9;
b5d67f64
SB		 1468
1469 ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0);
1470 if (PAGE_SIZE != ret) {
1471 bio_put(bio);
1472 return -EIO;
13db62b7 1473 }
b5d67f64 1474
33879d45 1475 if (btrfsic_submit_bio_wait(WRITE, bio)) {
442a4f63
SB		 1476 btrfs_dev_stat_inc_and_print(page_bad->dev,
1477 BTRFS_DEV_STAT_WRITE_ERRS);
ff023aac
SB		 1478 btrfs_dev_replace_stats_inc(
1479 &sblock_bad->sctx->dev_root->fs_info->
1480 dev_replace.num_write_errors);
442a4f63
SB		 1481 bio_put(bio);
1482 return -EIO;
1483 }
b5d67f64 1484 bio_put(bio);
a2de733c
AJ		 1485 }
1486
b5d67f64
SB		 1487 return 0;
1488}
1489
ff023aac
SB		 1490static void scrub_write_block_to_dev_replace(struct scrub_block *sblock)
1491{
1492 int page_num;
1493
1494 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1495 int ret;
1496
1497 ret = scrub_write_page_to_dev_replace(sblock, page_num);
1498 if (ret)
1499 btrfs_dev_replace_stats_inc(
1500 &sblock->sctx->dev_root->fs_info->dev_replace.
1501 num_write_errors);
1502 }
1503}
1504
1505static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
1506 int page_num)
1507{
1508 struct scrub_page *spage = sblock->pagev[page_num];
1509
1510 BUG_ON(spage->page == NULL);
1511 if (spage->io_error) {
1512 void *mapped_buffer = kmap_atomic(spage->page);
1513
1514 memset(mapped_buffer, 0, PAGE_CACHE_SIZE);
1515 flush_dcache_page(spage->page);
1516 kunmap_atomic(mapped_buffer);
1517 }
1518 return scrub_add_page_to_wr_bio(sblock->sctx, spage);
1519}
1520
1521static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
1522 struct scrub_page *spage)
1523{
1524 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1525 struct scrub_bio *sbio;
1526 int ret;
1527
1528 mutex_lock(&wr_ctx->wr_lock);
1529again:
1530 if (!wr_ctx->wr_curr_bio) {
1531 wr_ctx->wr_curr_bio = kzalloc(sizeof(*wr_ctx->wr_curr_bio),
1532 GFP_NOFS);
1533 if (!wr_ctx->wr_curr_bio) {
1534 mutex_unlock(&wr_ctx->wr_lock);
1535 return -ENOMEM;
1536 }
1537 wr_ctx->wr_curr_bio->sctx = sctx;
1538 wr_ctx->wr_curr_bio->page_count = 0;
1539 }
1540 sbio = wr_ctx->wr_curr_bio;
1541 if (sbio->page_count == 0) {
1542 struct bio *bio;
1543
1544 sbio->physical = spage->physical_for_dev_replace;
1545 sbio->logical = spage->logical;
1546 sbio->dev = wr_ctx->tgtdev;
1547 bio = sbio->bio;
1548 if (!bio) {
9be3395b 1549 bio = btrfs_io_bio_alloc(GFP_NOFS, wr_ctx->pages_per_wr_bio);
ff023aac
SB		 1550 if (!bio) {
1551 mutex_unlock(&wr_ctx->wr_lock);
1552 return -ENOMEM;
1553 }
1554 sbio->bio = bio;
1555 }
1556
1557 bio->bi_private = sbio;
1558 bio->bi_end_io = scrub_wr_bio_end_io;
1559 bio->bi_bdev = sbio->dev->bdev;
4f024f37 1560 bio->bi_iter.bi_sector = sbio->physical >> 9;
ff023aac
SB		 1561 sbio->err = 0;
1562 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
1563 spage->physical_for_dev_replace ||
1564 sbio->logical + sbio->page_count * PAGE_SIZE !=
1565 spage->logical) {
1566 scrub_wr_submit(sctx);
1567 goto again;
1568 }
1569
1570 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
1571 if (ret != PAGE_SIZE) {
1572 if (sbio->page_count < 1) {
1573 bio_put(sbio->bio);
1574 sbio->bio = NULL;
1575 mutex_unlock(&wr_ctx->wr_lock);
1576 return -EIO;
1577 }
1578 scrub_wr_submit(sctx);
1579 goto again;
1580 }
1581
1582 sbio->pagev[sbio->page_count] = spage;
1583 scrub_page_get(spage);
1584 sbio->page_count++;
1585 if (sbio->page_count == wr_ctx->pages_per_wr_bio)
1586 scrub_wr_submit(sctx);
1587 mutex_unlock(&wr_ctx->wr_lock);
1588
1589 return 0;
1590}
1591
1592static void scrub_wr_submit(struct scrub_ctx *sctx)
1593{
1594 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1595 struct scrub_bio *sbio;
1596
1597 if (!wr_ctx->wr_curr_bio)
1598 return;
1599
1600 sbio = wr_ctx->wr_curr_bio;
1601 wr_ctx->wr_curr_bio = NULL;
1602 WARN_ON(!sbio->bio->bi_bdev);
1603 scrub_pending_bio_inc(sctx);
1604 /* process all writes in a single worker thread. Then the block layer
1605 * orders the requests before sending them to the driver which
1606 * doubled the write performance on spinning disks when measured
1607 * with Linux 3.5 */
1608 btrfsic_submit_bio(WRITE, sbio->bio);
1609}
1610
1611static void scrub_wr_bio_end_io(struct bio *bio, int err)
1612{
1613 struct scrub_bio *sbio = bio->bi_private;
1614 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
1615
1616 sbio->err = err;
1617 sbio->bio = bio;
1618
0339ef2f
QW		 1619 btrfs_init_work(&sbio->work, scrub_wr_bio_end_io_worker, NULL, NULL);
1620 btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work);
ff023aac
SB		 1621}
1622
1623static void scrub_wr_bio_end_io_worker(struct btrfs_work *work)
1624{
1625 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
1626 struct scrub_ctx *sctx = sbio->sctx;
1627 int i;
1628
1629 WARN_ON(sbio->page_count > SCRUB_PAGES_PER_WR_BIO);
1630 if (sbio->err) {
1631 struct btrfs_dev_replace *dev_replace =
1632 &sbio->sctx->dev_root->fs_info->dev_replace;
1633
1634 for (i = 0; i < sbio->page_count; i++) {
1635 struct scrub_page *spage = sbio->pagev[i];
1636
1637 spage->io_error = 1;
1638 btrfs_dev_replace_stats_inc(&dev_replace->
1639 num_write_errors);
1640 }
1641 }
1642
1643 for (i = 0; i < sbio->page_count; i++)
1644 scrub_page_put(sbio->pagev[i]);
1645
1646 bio_put(sbio->bio);
1647 kfree(sbio);
1648 scrub_pending_bio_dec(sctx);
1649}
1650
1651static int scrub_checksum(struct scrub_block *sblock)
b5d67f64
SB		 1652{
1653 u64 flags;
1654 int ret;
1655
7a9e9987
SB		 1656 WARN_ON(sblock->page_count < 1);
1657 flags = sblock->pagev[0]->flags;
b5d67f64
SB		 1658 ret = 0;
1659 if (flags & BTRFS_EXTENT_FLAG_DATA)
1660 ret = scrub_checksum_data(sblock);
1661 else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1662 ret = scrub_checksum_tree_block(sblock);
1663 else if (flags & BTRFS_EXTENT_FLAG_SUPER)
1664 (void)scrub_checksum_super(sblock);
1665 else
1666 WARN_ON(1);
1667 if (ret)
1668 scrub_handle_errored_block(sblock);
ff023aac
SB		 1669
1670 return ret;
a2de733c
AJ		 1671}
1672
b5d67f64 1673static int scrub_checksum_data(struct scrub_block *sblock)
a2de733c 1674{
d9d181c1 1675 struct scrub_ctx *sctx = sblock->sctx;
a2de733c 1676 u8 csum[BTRFS_CSUM_SIZE];
b5d67f64
SB		 1677 u8 *on_disk_csum;
1678 struct page *page;
1679 void *buffer;
a2de733c
AJ		 1680 u32 crc = ~(u32)0;
1681 int fail = 0;
b5d67f64
SB		 1682 u64 len;
1683 int index;
a2de733c 1684
b5d67f64 1685 BUG_ON(sblock->page_count < 1);
7a9e9987 1686 if (!sblock->pagev[0]->have_csum)
a2de733c
AJ		 1687 return 0;
1688
7a9e9987
SB		 1689 on_disk_csum = sblock->pagev[0]->csum;
1690 page = sblock->pagev[0]->page;
9613bebb 1691 buffer = kmap_atomic(page);
b5d67f64 1692
d9d181c1 1693 len = sctx->sectorsize;
b5d67f64
SB		 1694 index = 0;
1695 for (;;) {
1696 u64 l = min_t(u64, len, PAGE_SIZE);
1697
b0496686 1698 crc = btrfs_csum_data(buffer, crc, l);
9613bebb 1699 kunmap_atomic(buffer);
b5d67f64
SB		 1700 len -= l;
1701 if (len == 0)
1702 break;
1703 index++;
1704 BUG_ON(index >= sblock->page_count);
7a9e9987
SB		 1705 BUG_ON(!sblock->pagev[index]->page);
1706 page = sblock->pagev[index]->page;
9613bebb 1707 buffer = kmap_atomic(page);
b5d67f64
SB		 1708 }
1709
a2de733c 1710 btrfs_csum_final(crc, csum);
d9d181c1 1711 if (memcmp(csum, on_disk_csum, sctx->csum_size))
a2de733c
AJ		 1712 fail = 1;
1713
a2de733c
AJ		 1714 return fail;
1715}
1716
b5d67f64 1717static int scrub_checksum_tree_block(struct scrub_block *sblock)
a2de733c 1718{
d9d181c1 1719 struct scrub_ctx *sctx = sblock->sctx;
a2de733c 1720 struct btrfs_header *h;
a36cf8b8 1721 struct btrfs_root *root = sctx->dev_root;
a2de733c 1722 struct btrfs_fs_info *fs_info = root->fs_info;
b5d67f64
SB		 1723 u8 calculated_csum[BTRFS_CSUM_SIZE];
1724 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1725 struct page *page;
1726 void *mapped_buffer;
1727 u64 mapped_size;
1728 void *p;
a2de733c
AJ		 1729 u32 crc = ~(u32)0;
1730 int fail = 0;
1731 int crc_fail = 0;
b5d67f64
SB		 1732 u64 len;
1733 int index;
1734
1735 BUG_ON(sblock->page_count < 1);
7a9e9987 1736 page = sblock->pagev[0]->page;
9613bebb 1737 mapped_buffer = kmap_atomic(page);
b5d67f64 1738 h = (struct btrfs_header *)mapped_buffer;
d9d181c1 1739 memcpy(on_disk_csum, h->csum, sctx->csum_size);
a2de733c
AJ		 1740
1741 /*
1742 * we don't use the getter functions here, as we
1743 * a) don't have an extent buffer and
1744 * b) the page is already kmapped
1745 */
a2de733c 1746
3cae210f 1747 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h))
a2de733c
AJ		 1748 ++fail;
1749
3cae210f 1750 if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h))
a2de733c
AJ		 1751 ++fail;
1752
1753 if (memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
1754 ++fail;
1755
1756 if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
1757 BTRFS_UUID_SIZE))
1758 ++fail;
1759
ff023aac 1760 WARN_ON(sctx->nodesize != sctx->leafsize);
d9d181c1 1761 len = sctx->nodesize - BTRFS_CSUM_SIZE;
b5d67f64
SB		 1762 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
1763 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
1764 index = 0;
1765 for (;;) {
1766 u64 l = min_t(u64, len, mapped_size);
1767
b0496686 1768 crc = btrfs_csum_data(p, crc, l);
9613bebb 1769 kunmap_atomic(mapped_buffer);
b5d67f64
SB		 1770 len -= l;
1771 if (len == 0)
1772 break;
1773 index++;
1774 BUG_ON(index >= sblock->page_count);
7a9e9987
SB		 1775 BUG_ON(!sblock->pagev[index]->page);
1776 page = sblock->pagev[index]->page;
9613bebb 1777 mapped_buffer = kmap_atomic(page);
b5d67f64
SB		 1778 mapped_size = PAGE_SIZE;
1779 p = mapped_buffer;
1780 }
1781
1782 btrfs_csum_final(crc, calculated_csum);
d9d181c1 1783 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
a2de733c
AJ		 1784 ++crc_fail;
1785
a2de733c
AJ		 1786 return fail || crc_fail;
1787}
1788
b5d67f64 1789static int scrub_checksum_super(struct scrub_block *sblock)
a2de733c
AJ		 1790{
1791 struct btrfs_super_block *s;
d9d181c1 1792 struct scrub_ctx *sctx = sblock->sctx;
a36cf8b8 1793 struct btrfs_root *root = sctx->dev_root;
a2de733c 1794 struct btrfs_fs_info *fs_info = root->fs_info;
b5d67f64
SB		 1795 u8 calculated_csum[BTRFS_CSUM_SIZE];
1796 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1797 struct page *page;
1798 void *mapped_buffer;
1799 u64 mapped_size;
1800 void *p;
a2de733c 1801 u32 crc = ~(u32)0;
442a4f63
SB		 1802 int fail_gen = 0;
1803 int fail_cor = 0;
b5d67f64
SB		 1804 u64 len;
1805 int index;
a2de733c 1806
b5d67f64 1807 BUG_ON(sblock->page_count < 1);
7a9e9987 1808 page = sblock->pagev[0]->page;
9613bebb 1809 mapped_buffer = kmap_atomic(page);
b5d67f64 1810 s = (struct btrfs_super_block *)mapped_buffer;
d9d181c1 1811 memcpy(on_disk_csum, s->csum, sctx->csum_size);
a2de733c 1812
3cae210f 1813 if (sblock->pagev[0]->logical != btrfs_super_bytenr(s))
442a4f63 1814 ++fail_cor;
a2de733c 1815
3cae210f 1816 if (sblock->pagev[0]->generation != btrfs_super_generation(s))
442a4f63 1817 ++fail_gen;
a2de733c
AJ		 1818
1819 if (memcmp(s->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
442a4f63 1820 ++fail_cor;
a2de733c 1821
b5d67f64
SB		 1822 len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE;
1823 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
1824 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
1825 index = 0;
1826 for (;;) {
1827 u64 l = min_t(u64, len, mapped_size);
1828
b0496686 1829 crc = btrfs_csum_data(p, crc, l);
9613bebb 1830 kunmap_atomic(mapped_buffer);
b5d67f64
SB		 1831 len -= l;
1832 if (len == 0)
1833 break;
1834 index++;
1835 BUG_ON(index >= sblock->page_count);
7a9e9987
SB		 1836 BUG_ON(!sblock->pagev[index]->page);
1837 page = sblock->pagev[index]->page;
9613bebb 1838 mapped_buffer = kmap_atomic(page);
b5d67f64
SB		 1839 mapped_size = PAGE_SIZE;
1840 p = mapped_buffer;
1841 }
1842
1843 btrfs_csum_final(crc, calculated_csum);
d9d181c1 1844 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
442a4f63 1845 ++fail_cor;
a2de733c 1846
442a4f63 1847 if (fail_cor + fail_gen) {
a2de733c
AJ		 1848 /*
1849 * if we find an error in a super block, we just report it.
1850 * They will get written with the next transaction commit
1851 * anyway
1852 */
d9d181c1
SB		 1853 spin_lock(&sctx->stat_lock);
1854 ++sctx->stat.super_errors;
1855 spin_unlock(&sctx->stat_lock);
442a4f63 1856 if (fail_cor)
7a9e9987 1857 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
442a4f63
SB		 1858 BTRFS_DEV_STAT_CORRUPTION_ERRS);
1859 else
7a9e9987 1860 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
442a4f63 1861 BTRFS_DEV_STAT_GENERATION_ERRS);
a2de733c
AJ		 1862 }
1863
442a4f63 1864 return fail_cor + fail_gen;
a2de733c
AJ		 1865}
1866
b5d67f64
SB		 1867static void scrub_block_get(struct scrub_block *sblock)
1868{
1869 atomic_inc(&sblock->ref_count);
1870}
1871
1872static void scrub_block_put(struct scrub_block *sblock)
1873{
1874 if (atomic_dec_and_test(&sblock->ref_count)) {
1875 int i;
1876
1877 for (i = 0; i < sblock->page_count; i++)
7a9e9987 1878 scrub_page_put(sblock->pagev[i]);
b5d67f64
SB		 1879 kfree(sblock);
1880 }
1881}
1882
7a9e9987
SB		 1883static void scrub_page_get(struct scrub_page *spage)
1884{
1885 atomic_inc(&spage->ref_count);
1886}
1887
1888static void scrub_page_put(struct scrub_page *spage)
1889{
1890 if (atomic_dec_and_test(&spage->ref_count)) {
1891 if (spage->page)
1892 __free_page(spage->page);
1893 kfree(spage);
1894 }
1895}
1896
d9d181c1 1897static void scrub_submit(struct scrub_ctx *sctx)
a2de733c
AJ		 1898{
1899 struct scrub_bio *sbio;
1900
d9d181c1 1901 if (sctx->curr == -1)
1623edeb 1902 return;
a2de733c 1903
d9d181c1
SB		 1904 sbio = sctx->bios[sctx->curr];
1905 sctx->curr = -1;
b6bfebc1 1906 scrub_pending_bio_inc(sctx);
a2de733c 1907
ff023aac
SB		 1908 if (!sbio->bio->bi_bdev) {
1909 /*
1910 * this case should not happen. If btrfs_map_block() is
1911 * wrong, it could happen for dev-replace operations on
1912 * missing devices when no mirrors are available, but in
1913 * this case it should already fail the mount.
1914 * This case is handled correctly (but _very_ slowly).
1915 */
1916 printk_ratelimited(KERN_WARNING
efe120a0 1917 "BTRFS: scrub_submit(bio bdev == NULL) is unexpected!\n");
ff023aac
SB		 1918 bio_endio(sbio->bio, -EIO);
1919 } else {
1920 btrfsic_submit_bio(READ, sbio->bio);
1921 }
a2de733c
AJ		 1922}
1923
ff023aac
SB		 1924static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
1925 struct scrub_page *spage)
a2de733c 1926{
b5d67f64 1927 struct scrub_block *sblock = spage->sblock;
a2de733c 1928 struct scrub_bio *sbio;
69f4cb52 1929 int ret;
a2de733c
AJ		 1930
1931again:
1932 /*
1933 * grab a fresh bio or wait for one to become available
1934 */
d9d181c1
SB		 1935 while (sctx->curr == -1) {
1936 spin_lock(&sctx->list_lock);
1937 sctx->curr = sctx->first_free;
1938 if (sctx->curr != -1) {
1939 sctx->first_free = sctx->bios[sctx->curr]->next_free;
1940 sctx->bios[sctx->curr]->next_free = -1;
1941 sctx->bios[sctx->curr]->page_count = 0;
1942 spin_unlock(&sctx->list_lock);
a2de733c 1943 } else {
d9d181c1
SB		 1944 spin_unlock(&sctx->list_lock);
1945 wait_event(sctx->list_wait, sctx->first_free != -1);
a2de733c
AJ		 1946 }
1947 }
d9d181c1 1948 sbio = sctx->bios[sctx->curr];
b5d67f64 1949 if (sbio->page_count == 0) {
69f4cb52
AJ		 1950 struct bio *bio;
1951
b5d67f64
SB		 1952 sbio->physical = spage->physical;
1953 sbio->logical = spage->logical;
a36cf8b8 1954 sbio->dev = spage->dev;
b5d67f64
SB		 1955 bio = sbio->bio;
1956 if (!bio) {
9be3395b 1957 bio = btrfs_io_bio_alloc(GFP_NOFS, sctx->pages_per_rd_bio);
b5d67f64
SB		 1958 if (!bio)
1959 return -ENOMEM;
1960 sbio->bio = bio;
1961 }
69f4cb52
AJ		 1962
1963 bio->bi_private = sbio;
1964 bio->bi_end_io = scrub_bio_end_io;
a36cf8b8 1965 bio->bi_bdev = sbio->dev->bdev;
4f024f37 1966 bio->bi_iter.bi_sector = sbio->physical >> 9;
69f4cb52 1967 sbio->err = 0;
b5d67f64
SB		 1968 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
1969 spage->physical ||
1970 sbio->logical + sbio->page_count * PAGE_SIZE !=
a36cf8b8
SB		 1971 spage->logical ||
1972 sbio->dev != spage->dev) {
d9d181c1 1973 scrub_submit(sctx);
a2de733c
AJ		 1974 goto again;
1975 }
69f4cb52 1976
b5d67f64
SB		 1977 sbio->pagev[sbio->page_count] = spage;
1978 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
1979 if (ret != PAGE_SIZE) {
1980 if (sbio->page_count < 1) {
1981 bio_put(sbio->bio);
1982 sbio->bio = NULL;
1983 return -EIO;
1984 }
d9d181c1 1985 scrub_submit(sctx);
69f4cb52
AJ		 1986 goto again;
1987 }
1988
ff023aac 1989 scrub_block_get(sblock); /* one for the page added to the bio */
b5d67f64
SB		 1990 atomic_inc(&sblock->outstanding_pages);
1991 sbio->page_count++;
ff023aac 1992 if (sbio->page_count == sctx->pages_per_rd_bio)
d9d181c1 1993 scrub_submit(sctx);
b5d67f64
SB		 1994
1995 return 0;
1996}
1997
d9d181c1 1998static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
a36cf8b8 1999 u64 physical, struct btrfs_device *dev, u64 flags,
ff023aac
SB		 2000 u64 gen, int mirror_num, u8 *csum, int force,
2001 u64 physical_for_dev_replace)
b5d67f64
SB		 2002{
2003 struct scrub_block *sblock;
2004 int index;
2005
2006 sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
2007 if (!sblock) {
d9d181c1
SB		 2008 spin_lock(&sctx->stat_lock);
2009 sctx->stat.malloc_errors++;
2010 spin_unlock(&sctx->stat_lock);
b5d67f64 2011 return -ENOMEM;
a2de733c 2012 }
b5d67f64 2013
7a9e9987
SB		 2014 /* one ref inside this function, plus one for each page added to
2015 * a bio later on */
b5d67f64 2016 atomic_set(&sblock->ref_count, 1);
d9d181c1 2017 sblock->sctx = sctx;
b5d67f64
SB		 2018 sblock->no_io_error_seen = 1;
2019
2020 for (index = 0; len > 0; index++) {
7a9e9987 2021 struct scrub_page *spage;
b5d67f64
SB		 2022 u64 l = min_t(u64, len, PAGE_SIZE);
2023
7a9e9987
SB		 2024 spage = kzalloc(sizeof(*spage), GFP_NOFS);
2025 if (!spage) {
2026leave_nomem:
d9d181c1
SB		 2027 spin_lock(&sctx->stat_lock);
2028 sctx->stat.malloc_errors++;
2029 spin_unlock(&sctx->stat_lock);
7a9e9987 2030 scrub_block_put(sblock);
b5d67f64
SB		 2031 return -ENOMEM;
2032 }
7a9e9987
SB		 2033 BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
2034 scrub_page_get(spage);
2035 sblock->pagev[index] = spage;
b5d67f64 2036 spage->sblock = sblock;
a36cf8b8 2037 spage->dev = dev;
b5d67f64
SB		 2038 spage->flags = flags;
2039 spage->generation = gen;
2040 spage->logical = logical;
2041 spage->physical = physical;
ff023aac 2042 spage->physical_for_dev_replace = physical_for_dev_replace;
b5d67f64
SB		 2043 spage->mirror_num = mirror_num;
2044 if (csum) {
2045 spage->have_csum = 1;
d9d181c1 2046 memcpy(spage->csum, csum, sctx->csum_size);
b5d67f64
SB		 2047 } else {
2048 spage->have_csum = 0;
2049 }
2050 sblock->page_count++;
7a9e9987
SB		 2051 spage->page = alloc_page(GFP_NOFS);
2052 if (!spage->page)
2053 goto leave_nomem;
b5d67f64
SB		 2054 len -= l;
2055 logical += l;
2056 physical += l;
ff023aac 2057 physical_for_dev_replace += l;
b5d67f64
SB		 2058 }
2059
7a9e9987 2060 WARN_ON(sblock->page_count == 0);
b5d67f64 2061 for (index = 0; index < sblock->page_count; index++) {
7a9e9987 2062 struct scrub_page *spage = sblock->pagev[index];
1bc87793
AJ		 2063 int ret;
2064
ff023aac 2065 ret = scrub_add_page_to_rd_bio(sctx, spage);
b5d67f64
SB		 2066 if (ret) {
2067 scrub_block_put(sblock);
1bc87793 2068 return ret;
b5d67f64 2069 }
1bc87793 2070 }
a2de733c 2071
b5d67f64 2072 if (force)
d9d181c1 2073 scrub_submit(sctx);
a2de733c 2074
b5d67f64
SB		 2075 /* last one frees, either here or in bio completion for last page */
2076 scrub_block_put(sblock);
a2de733c
AJ		 2077 return 0;
2078}
2079
b5d67f64
SB		 2080static void scrub_bio_end_io(struct bio *bio, int err)
2081{
2082 struct scrub_bio *sbio = bio->bi_private;
a36cf8b8 2083 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
b5d67f64
SB		 2084
2085 sbio->err = err;
2086 sbio->bio = bio;
2087
0339ef2f 2088 btrfs_queue_work(fs_info->scrub_workers, &sbio->work);
b5d67f64
SB		 2089}
2090
2091static void scrub_bio_end_io_worker(struct btrfs_work *work)
2092{
2093 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
d9d181c1 2094 struct scrub_ctx *sctx = sbio->sctx;
b5d67f64
SB		 2095 int i;
2096
ff023aac 2097 BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO);
b5d67f64
SB		 2098 if (sbio->err) {
2099 for (i = 0; i < sbio->page_count; i++) {
2100 struct scrub_page *spage = sbio->pagev[i];
2101
2102 spage->io_error = 1;
2103 spage->sblock->no_io_error_seen = 0;
2104 }
2105 }
2106
2107 /* now complete the scrub_block items that have all pages completed */
2108 for (i = 0; i < sbio->page_count; i++) {
2109 struct scrub_page *spage = sbio->pagev[i];
2110 struct scrub_block *sblock = spage->sblock;
2111
2112 if (atomic_dec_and_test(&sblock->outstanding_pages))
2113 scrub_block_complete(sblock);
2114 scrub_block_put(sblock);
2115 }
2116
b5d67f64
SB		 2117 bio_put(sbio->bio);
2118 sbio->bio = NULL;
d9d181c1
SB		 2119 spin_lock(&sctx->list_lock);
2120 sbio->next_free = sctx->first_free;
2121 sctx->first_free = sbio->index;
2122 spin_unlock(&sctx->list_lock);
ff023aac
SB		 2123
2124 if (sctx->is_dev_replace &&
2125 atomic_read(&sctx->wr_ctx.flush_all_writes)) {
2126 mutex_lock(&sctx->wr_ctx.wr_lock);
2127 scrub_wr_submit(sctx);
2128 mutex_unlock(&sctx->wr_ctx.wr_lock);
2129 }
2130
b6bfebc1 2131 scrub_pending_bio_dec(sctx);
b5d67f64
SB		 2132}
2133
2134static void scrub_block_complete(struct scrub_block *sblock)
2135{
ff023aac 2136 if (!sblock->no_io_error_seen) {
b5d67f64 2137 scrub_handle_errored_block(sblock);
ff023aac
SB		 2138 } else {
2139 /*
2140 * if has checksum error, write via repair mechanism in
2141 * dev replace case, otherwise write here in dev replace
2142 * case.
2143 */
2144 if (!scrub_checksum(sblock) && sblock->sctx->is_dev_replace)
2145 scrub_write_block_to_dev_replace(sblock);
2146 }
b5d67f64
SB		 2147}
2148
d9d181c1 2149static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len,
a2de733c
AJ		 2150 u8 *csum)
2151{
2152 struct btrfs_ordered_sum *sum = NULL;
f51a4a18 2153 unsigned long index;
a2de733c 2154 unsigned long num_sectors;
a2de733c 2155
d9d181c1
SB		 2156 while (!list_empty(&sctx->csum_list)) {
2157 sum = list_first_entry(&sctx->csum_list,
a2de733c
AJ		 2158 struct btrfs_ordered_sum, list);
2159 if (sum->bytenr > logical)
2160 return 0;
2161 if (sum->bytenr + sum->len > logical)
2162 break;
2163
d9d181c1 2164 ++sctx->stat.csum_discards;
a2de733c
AJ		 2165 list_del(&sum->list);
2166 kfree(sum);
2167 sum = NULL;
2168 }
2169 if (!sum)
2170 return 0;
2171
f51a4a18 2172 index = ((u32)(logical - sum->bytenr)) / sctx->sectorsize;
d9d181c1 2173 num_sectors = sum->len / sctx->sectorsize;
f51a4a18
MX		 2174 memcpy(csum, sum->sums + index, sctx->csum_size);
2175 if (index == num_sectors - 1) {
a2de733c
AJ		 2176 list_del(&sum->list);
2177 kfree(sum);
2178 }
f51a4a18 2179 return 1;
a2de733c
AJ		 2180}
2181
2182/* scrub extent tries to collect up to 64 kB for each bio */
d9d181c1 2183static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len,
a36cf8b8 2184 u64 physical, struct btrfs_device *dev, u64 flags,
ff023aac 2185 u64 gen, int mirror_num, u64 physical_for_dev_replace)
a2de733c
AJ		 2186{
2187 int ret;
2188 u8 csum[BTRFS_CSUM_SIZE];
b5d67f64
SB		 2189 u32 blocksize;
2190
2191 if (flags & BTRFS_EXTENT_FLAG_DATA) {
d9d181c1
SB		 2192 blocksize = sctx->sectorsize;
2193 spin_lock(&sctx->stat_lock);
2194 sctx->stat.data_extents_scrubbed++;
2195 sctx->stat.data_bytes_scrubbed += len;
2196 spin_unlock(&sctx->stat_lock);
b5d67f64 2197 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
ff023aac 2198 WARN_ON(sctx->nodesize != sctx->leafsize);
d9d181c1
SB		 2199 blocksize = sctx->nodesize;
2200 spin_lock(&sctx->stat_lock);
2201 sctx->stat.tree_extents_scrubbed++;
2202 sctx->stat.tree_bytes_scrubbed += len;
2203 spin_unlock(&sctx->stat_lock);
b5d67f64 2204 } else {
d9d181c1 2205 blocksize = sctx->sectorsize;
ff023aac 2206 WARN_ON(1);
b5d67f64 2207 }
a2de733c
AJ		 2208
2209 while (len) {
b5d67f64 2210 u64 l = min_t(u64, len, blocksize);
a2de733c
AJ		 2211 int have_csum = 0;
2212
2213 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2214 /* push csums to sbio */
d9d181c1 2215 have_csum = scrub_find_csum(sctx, logical, l, csum);
a2de733c 2216 if (have_csum == 0)
d9d181c1 2217 ++sctx->stat.no_csum;
ff023aac
SB		 2218 if (sctx->is_dev_replace && !have_csum) {
2219 ret = copy_nocow_pages(sctx, logical, l,
2220 mirror_num,
2221 physical_for_dev_replace);
2222 goto behind_scrub_pages;
2223 }
a2de733c 2224 }
a36cf8b8 2225 ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen,
ff023aac
SB		 2226 mirror_num, have_csum ? csum : NULL, 0,
2227 physical_for_dev_replace);
2228behind_scrub_pages:
a2de733c
AJ		 2229 if (ret)
2230 return ret;
2231 len -= l;
2232 logical += l;
2233 physical += l;
ff023aac 2234 physical_for_dev_replace += l;
a2de733c
AJ		 2235 }
2236 return 0;
2237}
2238
3b080b25
WS		 2239/*
2240 * Given a physical address, this will calculate it's
2241 * logical offset. if this is a parity stripe, it will return
2242 * the most left data stripe's logical offset.
2243 *
2244 * return 0 if it is a data stripe, 1 means parity stripe.
2245 */
2246static int get_raid56_logic_offset(u64 physical, int num,
2247 struct map_lookup *map, u64 *offset)
2248{
2249 int i;
2250 int j = 0;
2251 u64 stripe_nr;
2252 u64 last_offset;
2253 int stripe_index;
2254 int rot;
2255
2256 last_offset = (physical - map->stripes[num].physical) *
2257 nr_data_stripes(map);
2258 *offset = last_offset;
2259 for (i = 0; i < nr_data_stripes(map); i++) {
2260 *offset = last_offset + i * map->stripe_len;
2261
2262 stripe_nr = *offset;
2263 do_div(stripe_nr, map->stripe_len);
2264 do_div(stripe_nr, nr_data_stripes(map));
2265
2266 /* Work out the disk rotation on this stripe-set */
2267 rot = do_div(stripe_nr, map->num_stripes);
2268 /* calculate which stripe this data locates */
2269 rot += i;
e4fbaee2 2270 stripe_index = rot % map->num_stripes;
3b080b25
WS		 2271 if (stripe_index == num)
2272 return 0;
2273 if (stripe_index < num)
2274 j++;
2275 }
2276 *offset = last_offset + j * map->stripe_len;
2277 return 1;
2278}
2279
d9d181c1 2280static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
a36cf8b8
SB		 2281 struct map_lookup *map,
2282 struct btrfs_device *scrub_dev,
ff023aac
SB		 2283 int num, u64 base, u64 length,
2284 int is_dev_replace)
a2de733c
AJ		 2285{
2286 struct btrfs_path *path;
a36cf8b8 2287 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
a2de733c
AJ		 2288 struct btrfs_root *root = fs_info->extent_root;
2289 struct btrfs_root *csum_root = fs_info->csum_root;
2290 struct btrfs_extent_item *extent;
e7786c3a 2291 struct blk_plug plug;
a2de733c
AJ		 2292 u64 flags;
2293 int ret;
2294 int slot;
a2de733c 2295 u64 nstripes;
a2de733c
AJ		 2296 struct extent_buffer *l;
2297 struct btrfs_key key;
2298 u64 physical;
2299 u64 logical;
625f1c8d 2300 u64 logic_end;
3b080b25 2301 u64 physical_end;
a2de733c 2302 u64 generation;
e12fa9cd 2303 int mirror_num;
7a26285e
AJ		 2304 struct reada_control *reada1;
2305 struct reada_control *reada2;
2306 struct btrfs_key key_start;
2307 struct btrfs_key key_end;
a2de733c
AJ		 2308 u64 increment = map->stripe_len;
2309 u64 offset;
ff023aac
SB		 2310 u64 extent_logical;
2311 u64 extent_physical;
2312 u64 extent_len;
2313 struct btrfs_device *extent_dev;
2314 int extent_mirror_num;
3b080b25 2315 int stop_loop = 0;
53b381b3 2316
a2de733c 2317 nstripes = length;
3b080b25 2318 physical = map->stripes[num].physical;
a2de733c
AJ		 2319 offset = 0;
2320 do_div(nstripes, map->stripe_len);
2321 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2322 offset = map->stripe_len * num;
2323 increment = map->stripe_len * map->num_stripes;
193ea74b 2324 mirror_num = 1;
a2de733c
AJ		 2325 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2326 int factor = map->num_stripes / map->sub_stripes;
2327 offset = map->stripe_len * (num / map->sub_stripes);
2328 increment = map->stripe_len * factor;
193ea74b 2329 mirror_num = num % map->sub_stripes + 1;
a2de733c
AJ		 2330 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2331 increment = map->stripe_len;
193ea74b 2332 mirror_num = num % map->num_stripes + 1;
a2de733c
AJ		 2333 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2334 increment = map->stripe_len;
193ea74b 2335 mirror_num = num % map->num_stripes + 1;
3b080b25
WS		 2336 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
2337 BTRFS_BLOCK_GROUP_RAID6)) {
2338 get_raid56_logic_offset(physical, num, map, &offset);
2339 increment = map->stripe_len * nr_data_stripes(map);
2340 mirror_num = 1;
a2de733c
AJ		 2341 } else {
2342 increment = map->stripe_len;
193ea74b 2343 mirror_num = 1;
a2de733c
AJ		 2344 }
2345
2346 path = btrfs_alloc_path();
2347 if (!path)
2348 return -ENOMEM;
2349
b5d67f64
SB		 2350 /*
2351 * work on commit root. The related disk blocks are static as
2352 * long as COW is applied. This means, it is save to rewrite
2353 * them to repair disk errors without any race conditions
2354 */
a2de733c
AJ		 2355 path->search_commit_root = 1;
2356 path->skip_locking = 1;
2357
2358 /*
7a26285e
AJ		 2359 * trigger the readahead for extent tree csum tree and wait for
2360 * completion. During readahead, the scrub is officially paused
2361 * to not hold off transaction commits
a2de733c
AJ		 2362 */
2363 logical = base + offset;
3b080b25
WS		 2364 physical_end = physical + nstripes * map->stripe_len;
2365 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
2366 BTRFS_BLOCK_GROUP_RAID6)) {
2367 get_raid56_logic_offset(physical_end, num,
2368 map, &logic_end);
2369 logic_end += base;
2370 } else {
2371 logic_end = logical + increment * nstripes;
2372 }
d9d181c1 2373 wait_event(sctx->list_wait,
b6bfebc1 2374 atomic_read(&sctx->bios_in_flight) == 0);
cb7ab021 2375 scrub_blocked_if_needed(fs_info);
7a26285e
AJ		 2376
2377 /* FIXME it might be better to start readahead at commit root */
2378 key_start.objectid = logical;
2379 key_start.type = BTRFS_EXTENT_ITEM_KEY;
2380 key_start.offset = (u64)0;
3b080b25 2381 key_end.objectid = logic_end;
3173a18f
JB		 2382 key_end.type = BTRFS_METADATA_ITEM_KEY;
2383 key_end.offset = (u64)-1;
7a26285e
AJ		 2384 reada1 = btrfs_reada_add(root, &key_start, &key_end);
2385
2386 key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
2387 key_start.type = BTRFS_EXTENT_CSUM_KEY;
2388 key_start.offset = logical;
2389 key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
2390 key_end.type = BTRFS_EXTENT_CSUM_KEY;
3b080b25 2391 key_end.offset = logic_end;
7a26285e
AJ		 2392 reada2 = btrfs_reada_add(csum_root, &key_start, &key_end);
2393
2394 if (!IS_ERR(reada1))
2395 btrfs_reada_wait(reada1);
2396 if (!IS_ERR(reada2))
2397 btrfs_reada_wait(reada2);
2398
a2de733c
AJ		 2399
2400 /*
2401 * collect all data csums for the stripe to avoid seeking during
2402 * the scrub. This might currently (crc32) end up to be about 1MB
2403 */
e7786c3a 2404 blk_start_plug(&plug);
a2de733c 2405
a2de733c
AJ		 2406 /*
2407 * now find all extents for each stripe and scrub them
2408 */
a2de733c 2409 ret = 0;
3b080b25
WS		 2410 while (physical < physical_end) {
2411 /* for raid56, we skip parity stripe */
2412 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
2413 BTRFS_BLOCK_GROUP_RAID6)) {
2414 ret = get_raid56_logic_offset(physical, num,
2415 map, &logical);
2416 logical += base;
2417 if (ret)
2418 goto skip;
2419 }
a2de733c
AJ		 2420 /*
2421 * canceled?
2422 */
2423 if (atomic_read(&fs_info->scrub_cancel_req) ||
d9d181c1 2424 atomic_read(&sctx->cancel_req)) {
a2de733c
AJ		 2425 ret = -ECANCELED;
2426 goto out;
2427 }
2428 /*
2429 * check to see if we have to pause
2430 */
2431 if (atomic_read(&fs_info->scrub_pause_req)) {
2432 /* push queued extents */
ff023aac 2433 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
d9d181c1 2434 scrub_submit(sctx);
ff023aac
SB		 2435 mutex_lock(&sctx->wr_ctx.wr_lock);
2436 scrub_wr_submit(sctx);
2437 mutex_unlock(&sctx->wr_ctx.wr_lock);
d9d181c1 2438 wait_event(sctx->list_wait,
b6bfebc1 2439 atomic_read(&sctx->bios_in_flight) == 0);
ff023aac 2440 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
3cb0929a 2441 scrub_blocked_if_needed(fs_info);
a2de733c
AJ		 2442 }
2443
7c76edb7
WS		 2444 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2445 key.type = BTRFS_METADATA_ITEM_KEY;
2446 else
2447 key.type = BTRFS_EXTENT_ITEM_KEY;
a2de733c 2448 key.objectid = logical;
625f1c8d 2449 key.offset = (u64)-1;
a2de733c
AJ		 2450
2451 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2452 if (ret < 0)
2453 goto out;
3173a18f 2454
8c51032f 2455 if (ret > 0) {
ade2e0b3 2456 ret = btrfs_previous_extent_item(root, path, 0);
a2de733c
AJ		 2457 if (ret < 0)
2458 goto out;
8c51032f
AJ		 2459 if (ret > 0) {
2460 /* there's no smaller item, so stick with the
2461 * larger one */
2462 btrfs_release_path(path);
2463 ret = btrfs_search_slot(NULL, root, &key,
2464 path, 0, 0);
2465 if (ret < 0)
2466 goto out;
2467 }
a2de733c
AJ		 2468 }
2469
625f1c8d 2470 stop_loop = 0;
a2de733c 2471 while (1) {
3173a18f
JB		 2472 u64 bytes;
2473
a2de733c
AJ		 2474 l = path->nodes[0];
2475 slot = path->slots[0];
2476 if (slot >= btrfs_header_nritems(l)) {
2477 ret = btrfs_next_leaf(root, path);
2478 if (ret == 0)
2479 continue;
2480 if (ret < 0)
2481 goto out;
2482
625f1c8d 2483 stop_loop = 1;
a2de733c
AJ		 2484 break;
2485 }
2486 btrfs_item_key_to_cpu(l, &key, slot);
2487
3173a18f
JB		 2488 if (key.type == BTRFS_METADATA_ITEM_KEY)
2489 bytes = root->leafsize;
2490 else
2491 bytes = key.offset;
2492
2493 if (key.objectid + bytes <= logical)
a2de733c
AJ		 2494 goto next;
2495
625f1c8d
LB		 2496 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
2497 key.type != BTRFS_METADATA_ITEM_KEY)
2498 goto next;
a2de733c 2499
625f1c8d
LB		 2500 if (key.objectid >= logical + map->stripe_len) {
2501 /* out of this device extent */
2502 if (key.objectid >= logic_end)
2503 stop_loop = 1;
2504 break;
2505 }
a2de733c
AJ		 2506
2507 extent = btrfs_item_ptr(l, slot,
2508 struct btrfs_extent_item);
2509 flags = btrfs_extent_flags(l, extent);
2510 generation = btrfs_extent_generation(l, extent);
2511
2512 if (key.objectid < logical &&
2513 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
efe120a0
FH		 2514 btrfs_err(fs_info,
2515 "scrub: tree block %llu spanning "
2516 "stripes, ignored. logical=%llu",
c1c9ff7c 2517 key.objectid, logical);
a2de733c
AJ		 2518 goto next;
2519 }
2520
625f1c8d
LB		 2521again:
2522 extent_logical = key.objectid;
2523 extent_len = bytes;
2524
a2de733c
AJ		 2525 /*
2526 * trim extent to this stripe
2527 */
625f1c8d
LB		 2528 if (extent_logical < logical) {
2529 extent_len -= logical - extent_logical;
2530 extent_logical = logical;
a2de733c 2531 }
625f1c8d 2532 if (extent_logical + extent_len >
a2de733c 2533 logical + map->stripe_len) {
625f1c8d
LB		 2534 extent_len = logical + map->stripe_len -
2535 extent_logical;
a2de733c
AJ		 2536 }
2537
625f1c8d 2538 extent_physical = extent_logical - logical + physical;
ff023aac
SB		 2539 extent_dev = scrub_dev;
2540 extent_mirror_num = mirror_num;
2541 if (is_dev_replace)
2542 scrub_remap_extent(fs_info, extent_logical,
2543 extent_len, &extent_physical,
2544 &extent_dev,
2545 &extent_mirror_num);
625f1c8d
LB		 2546
2547 ret = btrfs_lookup_csums_range(csum_root, logical,
2548 logical + map->stripe_len - 1,
2549 &sctx->csum_list, 1);
2550 if (ret)
2551 goto out;
2552
ff023aac
SB		 2553 ret = scrub_extent(sctx, extent_logical, extent_len,
2554 extent_physical, extent_dev, flags,
2555 generation, extent_mirror_num,
115930cb 2556 extent_logical - logical + physical);
a2de733c
AJ		 2557 if (ret)
2558 goto out;
2559
d88d46c6 2560 scrub_free_csums(sctx);
625f1c8d
LB		 2561 if (extent_logical + extent_len <
2562 key.objectid + bytes) {
3b080b25
WS		 2563 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
2564 BTRFS_BLOCK_GROUP_RAID6)) {
2565 /*
2566 * loop until we find next data stripe
2567 * or we have finished all stripes.
2568 */
2569 do {
2570 physical += map->stripe_len;
2571 ret = get_raid56_logic_offset(
2572 physical, num,
2573 map, &logical);
2574 logical += base;
2575 } while (physical < physical_end && ret);
2576 } else {
2577 physical += map->stripe_len;
2578 logical += increment;
2579 }
625f1c8d
LB		 2580 if (logical < key.objectid + bytes) {
2581 cond_resched();
2582 goto again;
2583 }
2584
3b080b25 2585 if (physical >= physical_end) {
625f1c8d
LB		 2586 stop_loop = 1;
2587 break;
2588 }
2589 }
a2de733c
AJ		 2590next:
2591 path->slots[0]++;
2592 }
71267333 2593 btrfs_release_path(path);
3b080b25 2594skip:
a2de733c
AJ		 2595 logical += increment;
2596 physical += map->stripe_len;
d9d181c1 2597 spin_lock(&sctx->stat_lock);
625f1c8d
LB		 2598 if (stop_loop)
2599 sctx->stat.last_physical = map->stripes[num].physical +
2600 length;
2601 else
2602 sctx->stat.last_physical = physical;
d9d181c1 2603 spin_unlock(&sctx->stat_lock);
625f1c8d
LB		 2604 if (stop_loop)
2605 break;
a2de733c 2606 }
ff023aac 2607out:
a2de733c 2608 /* push queued extents */
d9d181c1 2609 scrub_submit(sctx);
ff023aac
SB		 2610 mutex_lock(&sctx->wr_ctx.wr_lock);
2611 scrub_wr_submit(sctx);
2612 mutex_unlock(&sctx->wr_ctx.wr_lock);
a2de733c 2613
e7786c3a 2614 blk_finish_plug(&plug);
a2de733c
AJ		 2615 btrfs_free_path(path);
2616 return ret < 0 ? ret : 0;
2617}
2618
d9d181c1 2619static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx,
a36cf8b8
SB		 2620 struct btrfs_device *scrub_dev,
2621 u64 chunk_tree, u64 chunk_objectid,
2622 u64 chunk_offset, u64 length,
ff023aac 2623 u64 dev_offset, int is_dev_replace)
a2de733c
AJ		 2624{
2625 struct btrfs_mapping_tree *map_tree =
a36cf8b8 2626 &sctx->dev_root->fs_info->mapping_tree;
a2de733c
AJ		 2627 struct map_lookup *map;
2628 struct extent_map *em;
2629 int i;
ff023aac 2630 int ret = 0;
a2de733c
AJ		 2631
2632 read_lock(&map_tree->map_tree.lock);
2633 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2634 read_unlock(&map_tree->map_tree.lock);
2635
2636 if (!em)
2637 return -EINVAL;
2638
2639 map = (struct map_lookup *)em->bdev;
2640 if (em->start != chunk_offset)
2641 goto out;
2642
2643 if (em->len < length)
2644 goto out;
2645
2646 for (i = 0; i < map->num_stripes; ++i) {
a36cf8b8 2647 if (map->stripes[i].dev->bdev == scrub_dev->bdev &&
859acaf1 2648 map->stripes[i].physical == dev_offset) {
a36cf8b8 2649 ret = scrub_stripe(sctx, map, scrub_dev, i,
ff023aac
SB		 2650 chunk_offset, length,
2651 is_dev_replace);
a2de733c
AJ		 2652 if (ret)
2653 goto out;
2654 }
2655 }
2656out:
2657 free_extent_map(em);
2658
2659 return ret;
2660}
2661
2662static noinline_for_stack
a36cf8b8 2663int scrub_enumerate_chunks(struct scrub_ctx *sctx,
ff023aac
SB		 2664 struct btrfs_device *scrub_dev, u64 start, u64 end,
2665 int is_dev_replace)
a2de733c
AJ		 2666{
2667 struct btrfs_dev_extent *dev_extent = NULL;
2668 struct btrfs_path *path;
a36cf8b8 2669 struct btrfs_root *root = sctx->dev_root;
a2de733c
AJ		 2670 struct btrfs_fs_info *fs_info = root->fs_info;
2671 u64 length;
2672 u64 chunk_tree;
2673 u64 chunk_objectid;
2674 u64 chunk_offset;
2675 int ret;
2676 int slot;
2677 struct extent_buffer *l;
2678 struct btrfs_key key;
2679 struct btrfs_key found_key;
2680 struct btrfs_block_group_cache *cache;
ff023aac 2681 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
a2de733c
AJ		 2682
2683 path = btrfs_alloc_path();
2684 if (!path)
2685 return -ENOMEM;
2686
2687 path->reada = 2;
2688 path->search_commit_root = 1;
2689 path->skip_locking = 1;
2690
a36cf8b8 2691 key.objectid = scrub_dev->devid;
a2de733c
AJ		 2692 key.offset = 0ull;
2693 key.type = BTRFS_DEV_EXTENT_KEY;
2694
a2de733c
AJ		 2695 while (1) {
2696 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2697 if (ret < 0)
8c51032f
AJ		 2698 break;
2699 if (ret > 0) {
2700 if (path->slots[0] >=
2701 btrfs_header_nritems(path->nodes[0])) {
2702 ret = btrfs_next_leaf(root, path);
2703 if (ret)
2704 break;
2705 }
2706 }
a2de733c
AJ		 2707
2708 l = path->nodes[0];
2709 slot = path->slots[0];
2710
2711 btrfs_item_key_to_cpu(l, &found_key, slot);
2712
a36cf8b8 2713 if (found_key.objectid != scrub_dev->devid)
a2de733c
AJ		 2714 break;
2715
8c51032f 2716 if (btrfs_key_type(&found_key) != BTRFS_DEV_EXTENT_KEY)
a2de733c
AJ		 2717 break;
2718
2719 if (found_key.offset >= end)
2720 break;
2721
2722 if (found_key.offset < key.offset)
2723 break;
2724
2725 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2726 length = btrfs_dev_extent_length(l, dev_extent);
2727
ced96edc
QW		 2728 if (found_key.offset + length <= start)
2729 goto skip;
a2de733c
AJ		 2730
2731 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2732 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2733 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2734
2735 /*
2736 * get a reference on the corresponding block group to prevent
2737 * the chunk from going away while we scrub it
2738 */
2739 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
ced96edc
QW		 2740
2741 /* some chunks are removed but not committed to disk yet,
2742 * continue scrubbing */
2743 if (!cache)
2744 goto skip;
2745
ff023aac
SB		 2746 dev_replace->cursor_right = found_key.offset + length;
2747 dev_replace->cursor_left = found_key.offset;
2748 dev_replace->item_needs_writeback = 1;
a36cf8b8 2749 ret = scrub_chunk(sctx, scrub_dev, chunk_tree, chunk_objectid,
ff023aac
SB		 2750 chunk_offset, length, found_key.offset,
2751 is_dev_replace);
2752
2753 /*
2754 * flush, submit all pending read and write bios, afterwards
2755 * wait for them.
2756 * Note that in the dev replace case, a read request causes
2757 * write requests that are submitted in the read completion
2758 * worker. Therefore in the current situation, it is required
2759 * that all write requests are flushed, so that all read and
2760 * write requests are really completed when bios_in_flight
2761 * changes to 0.
2762 */
2763 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
2764 scrub_submit(sctx);
2765 mutex_lock(&sctx->wr_ctx.wr_lock);
2766 scrub_wr_submit(sctx);
2767 mutex_unlock(&sctx->wr_ctx.wr_lock);
2768
2769 wait_event(sctx->list_wait,
2770 atomic_read(&sctx->bios_in_flight) == 0);
12cf9372
WS		 2771 atomic_inc(&fs_info->scrubs_paused);
2772 wake_up(&fs_info->scrub_pause_wait);
2773
2774 /*
2775 * must be called before we decrease @scrub_paused.
2776 * make sure we don't block transaction commit while
2777 * we are waiting pending workers finished.
2778 */
ff023aac
SB		 2779 wait_event(sctx->list_wait,
2780 atomic_read(&sctx->workers_pending) == 0);
12cf9372
WS		 2781 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
2782
2783 mutex_lock(&fs_info->scrub_lock);
2784 __scrub_blocked_if_needed(fs_info);
2785 atomic_dec(&fs_info->scrubs_paused);
2786 mutex_unlock(&fs_info->scrub_lock);
2787 wake_up(&fs_info->scrub_pause_wait);
ff023aac 2788
a2de733c
AJ		 2789 btrfs_put_block_group(cache);
2790 if (ret)
2791 break;
af1be4f8
SB		 2792 if (is_dev_replace &&
2793 atomic64_read(&dev_replace->num_write_errors) > 0) {
ff023aac
SB		 2794 ret = -EIO;
2795 break;
2796 }
2797 if (sctx->stat.malloc_errors > 0) {
2798 ret = -ENOMEM;
2799 break;
2800 }
a2de733c 2801
539f358a
ID		 2802 dev_replace->cursor_left = dev_replace->cursor_right;
2803 dev_replace->item_needs_writeback = 1;
ced96edc 2804skip:
a2de733c 2805 key.offset = found_key.offset + length;
71267333 2806 btrfs_release_path(path);
a2de733c
AJ		 2807 }
2808
a2de733c 2809 btrfs_free_path(path);
8c51032f
AJ		 2810
2811 /*
2812 * ret can still be 1 from search_slot or next_leaf,
2813 * that's not an error
2814 */
2815 return ret < 0 ? ret : 0;
a2de733c
AJ		 2816}
2817
a36cf8b8
SB		 2818static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx,
2819 struct btrfs_device *scrub_dev)
a2de733c
AJ		 2820{
2821 int i;
2822 u64 bytenr;
2823 u64 gen;
2824 int ret;
a36cf8b8 2825 struct btrfs_root *root = sctx->dev_root;
a2de733c 2826
87533c47 2827 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
79787eaa
JM		 2828 return -EIO;
2829
a2de733c
AJ		 2830 gen = root->fs_info->last_trans_committed;
2831
2832 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2833 bytenr = btrfs_sb_offset(i);
a36cf8b8 2834 if (bytenr + BTRFS_SUPER_INFO_SIZE > scrub_dev->total_bytes)
a2de733c
AJ		 2835 break;
2836
d9d181c1 2837 ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr,
a36cf8b8 2838 scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i,
ff023aac 2839 NULL, 1, bytenr);
a2de733c
AJ		 2840 if (ret)
2841 return ret;
2842 }
b6bfebc1 2843 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
a2de733c
AJ		 2844
2845 return 0;
2846}
2847
2848/*
2849 * get a reference count on fs_info->scrub_workers. start worker if necessary
2850 */
ff023aac
SB		 2851static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info,
2852 int is_dev_replace)
a2de733c 2853{
0dc3b84a 2854 int ret = 0;
0339ef2f
QW		 2855 int flags = WQ_FREEZABLE | WQ_UNBOUND;
2856 int max_active = fs_info->thread_pool_size;
a2de733c 2857
632dd772 2858 if (fs_info->scrub_workers_refcnt == 0) {
ff023aac 2859 if (is_dev_replace)
0339ef2f
QW		 2860 fs_info->scrub_workers =
2861 btrfs_alloc_workqueue("btrfs-scrub", flags,
2862 1, 4);
ff023aac 2863 else
0339ef2f
QW		 2864 fs_info->scrub_workers =
2865 btrfs_alloc_workqueue("btrfs-scrub", flags,
2866 max_active, 4);
2867 if (!fs_info->scrub_workers) {
2868 ret = -ENOMEM;
0dc3b84a 2869 goto out;
0339ef2f
QW		 2870 }
2871 fs_info->scrub_wr_completion_workers =
2872 btrfs_alloc_workqueue("btrfs-scrubwrc", flags,
2873 max_active, 2);
2874 if (!fs_info->scrub_wr_completion_workers) {
2875 ret = -ENOMEM;
ff023aac 2876 goto out;
0339ef2f
QW		 2877 }
2878 fs_info->scrub_nocow_workers =
2879 btrfs_alloc_workqueue("btrfs-scrubnc", flags, 1, 0);
2880 if (!fs_info->scrub_nocow_workers) {
2881 ret = -ENOMEM;
ff023aac 2882 goto out;
0339ef2f 2883 }
632dd772 2884 }
a2de733c 2885 ++fs_info->scrub_workers_refcnt;
0dc3b84a 2886out:
0dc3b84a 2887 return ret;
a2de733c
AJ		 2888}
2889
aa1b8cd4 2890static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info)
a2de733c 2891{
ff023aac 2892 if (--fs_info->scrub_workers_refcnt == 0) {
0339ef2f
QW		 2893 btrfs_destroy_workqueue(fs_info->scrub_workers);
2894 btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
2895 btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
ff023aac 2896 }
a2de733c 2897 WARN_ON(fs_info->scrub_workers_refcnt < 0);
a2de733c
AJ		 2898}
2899
aa1b8cd4
SB		 2900int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start,
2901 u64 end, struct btrfs_scrub_progress *progress,
63a212ab 2902 int readonly, int is_dev_replace)
a2de733c 2903{
d9d181c1 2904 struct scrub_ctx *sctx;
a2de733c
AJ		 2905 int ret;
2906 struct btrfs_device *dev;
5d68da3b 2907 struct rcu_string *name;
a2de733c 2908
aa1b8cd4 2909 if (btrfs_fs_closing(fs_info))
a2de733c
AJ		 2910 return -EINVAL;
2911
2912 /*
2913 * check some assumptions
2914 */
aa1b8cd4 2915 if (fs_info->chunk_root->nodesize != fs_info->chunk_root->leafsize) {
efe120a0
FH		 2916 btrfs_err(fs_info,
2917 "scrub: size assumption nodesize == leafsize (%d == %d) fails",
aa1b8cd4
SB		 2918 fs_info->chunk_root->nodesize,
2919 fs_info->chunk_root->leafsize);
b5d67f64
SB		 2920 return -EINVAL;
2921 }
2922
aa1b8cd4 2923 if (fs_info->chunk_root->nodesize > BTRFS_STRIPE_LEN) {
b5d67f64
SB		 2924 /*
2925 * in this case scrub is unable to calculate the checksum
2926 * the way scrub is implemented. Do not handle this
2927 * situation at all because it won't ever happen.
2928 */
efe120a0
FH		 2929 btrfs_err(fs_info,
2930 "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails",
aa1b8cd4 2931 fs_info->chunk_root->nodesize, BTRFS_STRIPE_LEN);
b5d67f64
SB		 2932 return -EINVAL;
2933 }
2934
aa1b8cd4 2935 if (fs_info->chunk_root->sectorsize != PAGE_SIZE) {
b5d67f64 2936 /* not supported for data w/o checksums */
efe120a0
FH		 2937 btrfs_err(fs_info,
2938 "scrub: size assumption sectorsize != PAGE_SIZE "
2939 "(%d != %lu) fails",
27f9f023 2940 fs_info->chunk_root->sectorsize, PAGE_SIZE);
a2de733c
AJ		 2941 return -EINVAL;
2942 }
2943
7a9e9987
SB		 2944 if (fs_info->chunk_root->nodesize >
2945 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK ||
2946 fs_info->chunk_root->sectorsize >
2947 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) {
2948 /*
2949 * would exhaust the array bounds of pagev member in
2950 * struct scrub_block
2951 */
efe120a0
FH		 2952 btrfs_err(fs_info, "scrub: size assumption nodesize and sectorsize "
2953 "<= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails",
7a9e9987
SB		 2954 fs_info->chunk_root->nodesize,
2955 SCRUB_MAX_PAGES_PER_BLOCK,
2956 fs_info->chunk_root->sectorsize,
2957 SCRUB_MAX_PAGES_PER_BLOCK);
2958 return -EINVAL;
2959 }
2960
a2de733c 2961
aa1b8cd4
SB		 2962 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2963 dev = btrfs_find_device(fs_info, devid, NULL, NULL);
63a212ab 2964 if (!dev || (dev->missing && !is_dev_replace)) {
aa1b8cd4 2965 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
a2de733c
AJ		 2966 return -ENODEV;
2967 }
a2de733c 2968
5d68da3b
MX		 2969 if (!is_dev_replace && !readonly && !dev->writeable) {
2970 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2971 rcu_read_lock();
2972 name = rcu_dereference(dev->name);
2973 btrfs_err(fs_info, "scrub: device %s is not writable",
2974 name->str);
2975 rcu_read_unlock();
2976 return -EROFS;
2977 }
2978
3b7a016f 2979 mutex_lock(&fs_info->scrub_lock);
63a212ab 2980 if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) {
a2de733c 2981 mutex_unlock(&fs_info->scrub_lock);
aa1b8cd4 2982 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
aa1b8cd4 2983 return -EIO;
a2de733c
AJ		 2984 }
2985
8dabb742
SB		 2986 btrfs_dev_replace_lock(&fs_info->dev_replace);
2987 if (dev->scrub_device ||
2988 (!is_dev_replace &&
2989 btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) {
2990 btrfs_dev_replace_unlock(&fs_info->dev_replace);
a2de733c 2991 mutex_unlock(&fs_info->scrub_lock);
aa1b8cd4 2992 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
a2de733c
AJ		 2993 return -EINPROGRESS;
2994 }
8dabb742 2995 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3b7a016f
WS		 2996
2997 ret = scrub_workers_get(fs_info, is_dev_replace);
2998 if (ret) {
2999 mutex_unlock(&fs_info->scrub_lock);
3000 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3001 return ret;
3002 }
3003
63a212ab 3004 sctx = scrub_setup_ctx(dev, is_dev_replace);
d9d181c1 3005 if (IS_ERR(sctx)) {
a2de733c 3006 mutex_unlock(&fs_info->scrub_lock);
aa1b8cd4
SB		 3007 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3008 scrub_workers_put(fs_info);
d9d181c1 3009 return PTR_ERR(sctx);
a2de733c 3010 }
d9d181c1
SB		 3011 sctx->readonly = readonly;
3012 dev->scrub_device = sctx;
3cb0929a 3013 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
a2de733c 3014
3cb0929a
WS		 3015 /*
3016 * checking @scrub_pause_req here, we can avoid
3017 * race between committing transaction and scrubbing.
3018 */
cb7ab021 3019 __scrub_blocked_if_needed(fs_info);
a2de733c
AJ		 3020 atomic_inc(&fs_info->scrubs_running);
3021 mutex_unlock(&fs_info->scrub_lock);
a2de733c 3022
ff023aac 3023 if (!is_dev_replace) {
9b011adf
WS		 3024 /*
3025 * by holding device list mutex, we can
3026 * kick off writing super in log tree sync.
3027 */
3cb0929a 3028 mutex_lock(&fs_info->fs_devices->device_list_mutex);
ff023aac 3029 ret = scrub_supers(sctx, dev);
3cb0929a 3030 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
ff023aac 3031 }
a2de733c
AJ		 3032
3033 if (!ret)
ff023aac
SB		 3034 ret = scrub_enumerate_chunks(sctx, dev, start, end,
3035 is_dev_replace);
a2de733c 3036
b6bfebc1 3037 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
a2de733c
AJ		 3038 atomic_dec(&fs_info->scrubs_running);
3039 wake_up(&fs_info->scrub_pause_wait);
3040
b6bfebc1 3041 wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0);
0ef8e451 3042
a2de733c 3043 if (progress)
d9d181c1 3044 memcpy(progress, &sctx->stat, sizeof(*progress));
a2de733c
AJ		 3045
3046 mutex_lock(&fs_info->scrub_lock);
3047 dev->scrub_device = NULL;
3b7a016f 3048 scrub_workers_put(fs_info);
a2de733c
AJ		 3049 mutex_unlock(&fs_info->scrub_lock);
3050
d9d181c1 3051 scrub_free_ctx(sctx);
a2de733c
AJ		 3052
3053 return ret;
3054}
3055
143bede5 3056void btrfs_scrub_pause(struct btrfs_root *root)
a2de733c
AJ		 3057{
3058 struct btrfs_fs_info *fs_info = root->fs_info;
3059
3060 mutex_lock(&fs_info->scrub_lock);
3061 atomic_inc(&fs_info->scrub_pause_req);
3062 while (atomic_read(&fs_info->scrubs_paused) !=
3063 atomic_read(&fs_info->scrubs_running)) {
3064 mutex_unlock(&fs_info->scrub_lock);
3065 wait_event(fs_info->scrub_pause_wait,
3066 atomic_read(&fs_info->scrubs_paused) ==
3067 atomic_read(&fs_info->scrubs_running));
3068 mutex_lock(&fs_info->scrub_lock);
3069 }
3070 mutex_unlock(&fs_info->scrub_lock);
a2de733c
AJ		 3071}
3072
143bede5 3073void btrfs_scrub_continue(struct btrfs_root *root)
a2de733c
AJ		 3074{
3075 struct btrfs_fs_info *fs_info = root->fs_info;
3076
3077 atomic_dec(&fs_info->scrub_pause_req);
3078 wake_up(&fs_info->scrub_pause_wait);
a2de733c
AJ		 3079}
3080
aa1b8cd4 3081int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info)
a2de733c 3082{
a2de733c
AJ		 3083 mutex_lock(&fs_info->scrub_lock);
3084 if (!atomic_read(&fs_info->scrubs_running)) {
3085 mutex_unlock(&fs_info->scrub_lock);
3086 return -ENOTCONN;
3087 }
3088
3089 atomic_inc(&fs_info->scrub_cancel_req);
3090 while (atomic_read(&fs_info->scrubs_running)) {
3091 mutex_unlock(&fs_info->scrub_lock);
3092 wait_event(fs_info->scrub_pause_wait,
3093 atomic_read(&fs_info->scrubs_running) == 0);
3094 mutex_lock(&fs_info->scrub_lock);
3095 }
3096 atomic_dec(&fs_info->scrub_cancel_req);
3097 mutex_unlock(&fs_info->scrub_lock);
3098
3099 return 0;
3100}
3101
aa1b8cd4
SB		 3102int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info,
3103 struct btrfs_device *dev)
49b25e05 3104{
d9d181c1 3105 struct scrub_ctx *sctx;
a2de733c
AJ		 3106
3107 mutex_lock(&fs_info->scrub_lock);
d9d181c1
SB		 3108 sctx = dev->scrub_device;
3109 if (!sctx) {
a2de733c
AJ		 3110 mutex_unlock(&fs_info->scrub_lock);
3111 return -ENOTCONN;
3112 }
d9d181c1 3113 atomic_inc(&sctx->cancel_req);
a2de733c
AJ		 3114 while (dev->scrub_device) {
3115 mutex_unlock(&fs_info->scrub_lock);
3116 wait_event(fs_info->scrub_pause_wait,
3117 dev->scrub_device == NULL);
3118 mutex_lock(&fs_info->scrub_lock);
3119 }
3120 mutex_unlock(&fs_info->scrub_lock);
3121
3122 return 0;
3123}
1623edeb 3124
a2de733c
AJ		 3125int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
3126 struct btrfs_scrub_progress *progress)
3127{
3128 struct btrfs_device *dev;
d9d181c1 3129 struct scrub_ctx *sctx = NULL;
a2de733c
AJ		 3130
3131 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
aa1b8cd4 3132 dev = btrfs_find_device(root->fs_info, devid, NULL, NULL);
a2de733c 3133 if (dev)
d9d181c1
SB		 3134 sctx = dev->scrub_device;
3135 if (sctx)
3136 memcpy(progress, &sctx->stat, sizeof(*progress));
a2de733c
AJ		 3137 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3138
d9d181c1 3139 return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV;
a2de733c 3140}
ff023aac
SB		 3141
3142static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
3143 u64 extent_logical, u64 extent_len,
3144 u64 *extent_physical,
3145 struct btrfs_device **extent_dev,
3146 int *extent_mirror_num)
3147{
3148 u64 mapped_length;
3149 struct btrfs_bio *bbio = NULL;
3150 int ret;
3151
3152 mapped_length = extent_len;
3153 ret = btrfs_map_block(fs_info, READ, extent_logical,
3154 &mapped_length, &bbio, 0);
3155 if (ret || !bbio || mapped_length < extent_len ||
3156 !bbio->stripes[0].dev->bdev) {
3157 kfree(bbio);
3158 return;
3159 }
3160
3161 *extent_physical = bbio->stripes[0].physical;
3162 *extent_mirror_num = bbio->mirror_num;
3163 *extent_dev = bbio->stripes[0].dev;
3164 kfree(bbio);
3165}
3166
3167static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
3168 struct scrub_wr_ctx *wr_ctx,
3169 struct btrfs_fs_info *fs_info,
3170 struct btrfs_device *dev,
3171 int is_dev_replace)
3172{
3173 WARN_ON(wr_ctx->wr_curr_bio != NULL);
3174
3175 mutex_init(&wr_ctx->wr_lock);
3176 wr_ctx->wr_curr_bio = NULL;
3177 if (!is_dev_replace)
3178 return 0;
3179
3180 WARN_ON(!dev->bdev);
3181 wr_ctx->pages_per_wr_bio = min_t(int, SCRUB_PAGES_PER_WR_BIO,
3182 bio_get_nr_vecs(dev->bdev));
3183 wr_ctx->tgtdev = dev;
3184 atomic_set(&wr_ctx->flush_all_writes, 0);
3185 return 0;
3186}
3187
3188static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx)
3189{
3190 mutex_lock(&wr_ctx->wr_lock);
3191 kfree(wr_ctx->wr_curr_bio);
3192 wr_ctx->wr_curr_bio = NULL;
3193 mutex_unlock(&wr_ctx->wr_lock);
3194}
3195
3196static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
3197 int mirror_num, u64 physical_for_dev_replace)
3198{
3199 struct scrub_copy_nocow_ctx *nocow_ctx;
3200 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
3201
3202 nocow_ctx = kzalloc(sizeof(*nocow_ctx), GFP_NOFS);
3203 if (!nocow_ctx) {
3204 spin_lock(&sctx->stat_lock);
3205 sctx->stat.malloc_errors++;
3206 spin_unlock(&sctx->stat_lock);
3207 return -ENOMEM;
3208 }
3209
3210 scrub_pending_trans_workers_inc(sctx);
3211
3212 nocow_ctx->sctx = sctx;
3213 nocow_ctx->logical = logical;
3214 nocow_ctx->len = len;
3215 nocow_ctx->mirror_num = mirror_num;
3216 nocow_ctx->physical_for_dev_replace = physical_for_dev_replace;
0339ef2f 3217 btrfs_init_work(&nocow_ctx->work, copy_nocow_pages_worker, NULL, NULL);
652f25a2 3218 INIT_LIST_HEAD(&nocow_ctx->inodes);
0339ef2f
QW		 3219 btrfs_queue_work(fs_info->scrub_nocow_workers,
3220 &nocow_ctx->work);
ff023aac
SB		 3221
3222 return 0;
3223}
3224
652f25a2
JB		 3225static int record_inode_for_nocow(u64 inum, u64 offset, u64 root, void *ctx)
3226{
3227 struct scrub_copy_nocow_ctx *nocow_ctx = ctx;
3228 struct scrub_nocow_inode *nocow_inode;
3229
3230 nocow_inode = kzalloc(sizeof(*nocow_inode), GFP_NOFS);
3231 if (!nocow_inode)
3232 return -ENOMEM;
3233 nocow_inode->inum = inum;
3234 nocow_inode->offset = offset;
3235 nocow_inode->root = root;
3236 list_add_tail(&nocow_inode->list, &nocow_ctx->inodes);
3237 return 0;
3238}
3239
3240#define COPY_COMPLETE 1
3241
ff023aac
SB		 3242static void copy_nocow_pages_worker(struct btrfs_work *work)
3243{
3244 struct scrub_copy_nocow_ctx *nocow_ctx =
3245 container_of(work, struct scrub_copy_nocow_ctx, work);
3246 struct scrub_ctx *sctx = nocow_ctx->sctx;
3247 u64 logical = nocow_ctx->logical;
3248 u64 len = nocow_ctx->len;
3249 int mirror_num = nocow_ctx->mirror_num;
3250 u64 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
3251 int ret;
3252 struct btrfs_trans_handle *trans = NULL;
3253 struct btrfs_fs_info *fs_info;
3254 struct btrfs_path *path;
3255 struct btrfs_root *root;
3256 int not_written = 0;
3257
3258 fs_info = sctx->dev_root->fs_info;
3259 root = fs_info->extent_root;
3260
3261 path = btrfs_alloc_path();
3262 if (!path) {
3263 spin_lock(&sctx->stat_lock);
3264 sctx->stat.malloc_errors++;
3265 spin_unlock(&sctx->stat_lock);
3266 not_written = 1;
3267 goto out;
3268 }
3269
3270 trans = btrfs_join_transaction(root);
3271 if (IS_ERR(trans)) {
3272 not_written = 1;
3273 goto out;
3274 }
3275
3276 ret = iterate_inodes_from_logical(logical, fs_info, path,
652f25a2 3277 record_inode_for_nocow, nocow_ctx);
ff023aac 3278 if (ret != 0 && ret != -ENOENT) {
efe120a0
FH		 3279 btrfs_warn(fs_info, "iterate_inodes_from_logical() failed: log %llu, "
3280 "phys %llu, len %llu, mir %u, ret %d",
118a0a25
GU		 3281 logical, physical_for_dev_replace, len, mirror_num,
3282 ret);
ff023aac
SB		 3283 not_written = 1;
3284 goto out;
3285 }
3286
652f25a2
JB		 3287 btrfs_end_transaction(trans, root);
3288 trans = NULL;
3289 while (!list_empty(&nocow_ctx->inodes)) {
3290 struct scrub_nocow_inode *entry;
3291 entry = list_first_entry(&nocow_ctx->inodes,
3292 struct scrub_nocow_inode,
3293 list);
3294 list_del_init(&entry->list);
3295 ret = copy_nocow_pages_for_inode(entry->inum, entry->offset,
3296 entry->root, nocow_ctx);
3297 kfree(entry);
3298 if (ret == COPY_COMPLETE) {
3299 ret = 0;
3300 break;
3301 } else if (ret) {
3302 break;
3303 }
3304 }
ff023aac 3305out:
652f25a2
JB		 3306 while (!list_empty(&nocow_ctx->inodes)) {
3307 struct scrub_nocow_inode *entry;
3308 entry = list_first_entry(&nocow_ctx->inodes,
3309 struct scrub_nocow_inode,
3310 list);
3311 list_del_init(&entry->list);
3312 kfree(entry);
3313 }
ff023aac
SB		 3314 if (trans && !IS_ERR(trans))
3315 btrfs_end_transaction(trans, root);
3316 if (not_written)
3317 btrfs_dev_replace_stats_inc(&fs_info->dev_replace.
3318 num_uncorrectable_read_errors);
3319
3320 btrfs_free_path(path);
3321 kfree(nocow_ctx);
3322
3323 scrub_pending_trans_workers_dec(sctx);
3324}
3325
652f25a2
JB		 3326static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
3327 struct scrub_copy_nocow_ctx *nocow_ctx)
ff023aac 3328{
826aa0a8 3329 struct btrfs_fs_info *fs_info = nocow_ctx->sctx->dev_root->fs_info;
ff023aac 3330 struct btrfs_key key;
826aa0a8
MX		 3331 struct inode *inode;
3332 struct page *page;
ff023aac 3333 struct btrfs_root *local_root;
652f25a2
JB		 3334 struct btrfs_ordered_extent *ordered;
3335 struct extent_map *em;
3336 struct extent_state *cached_state = NULL;
3337 struct extent_io_tree *io_tree;
ff023aac 3338 u64 physical_for_dev_replace;
652f25a2
JB		 3339 u64 len = nocow_ctx->len;
3340 u64 lockstart = offset, lockend = offset + len - 1;
826aa0a8 3341 unsigned long index;
6f1c3605 3342 int srcu_index;
652f25a2
JB		 3343 int ret = 0;
3344 int err = 0;
ff023aac
SB		 3345
3346 key.objectid = root;
3347 key.type = BTRFS_ROOT_ITEM_KEY;
3348 key.offset = (u64)-1;
6f1c3605
LB		 3349
3350 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
3351
ff023aac 3352 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
6f1c3605
LB		 3353 if (IS_ERR(local_root)) {
3354 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
ff023aac 3355 return PTR_ERR(local_root);
6f1c3605 3356 }
ff023aac
SB		 3357
3358 key.type = BTRFS_INODE_ITEM_KEY;
3359 key.objectid = inum;
3360 key.offset = 0;
3361 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
6f1c3605 3362 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
ff023aac
SB		 3363 if (IS_ERR(inode))
3364 return PTR_ERR(inode);
3365
edd1400b
MX		 3366 /* Avoid truncate/dio/punch hole.. */
3367 mutex_lock(&inode->i_mutex);
3368 inode_dio_wait(inode);
3369
ff023aac 3370 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
652f25a2
JB		 3371 io_tree = &BTRFS_I(inode)->io_tree;
3372
3373 lock_extent_bits(io_tree, lockstart, lockend, 0, &cached_state);
3374 ordered = btrfs_lookup_ordered_range(inode, lockstart, len);
3375 if (ordered) {
3376 btrfs_put_ordered_extent(ordered);
3377 goto out_unlock;
3378 }
3379
3380 em = btrfs_get_extent(inode, NULL, 0, lockstart, len, 0);
3381 if (IS_ERR(em)) {
3382 ret = PTR_ERR(em);
3383 goto out_unlock;
3384 }
3385
3386 /*
3387 * This extent does not actually cover the logical extent anymore,
3388 * move on to the next inode.
3389 */
3390 if (em->block_start > nocow_ctx->logical ||
3391 em->block_start + em->block_len < nocow_ctx->logical + len) {
3392 free_extent_map(em);
3393 goto out_unlock;
3394 }
3395 free_extent_map(em);
3396
ff023aac 3397 while (len >= PAGE_CACHE_SIZE) {
ff023aac 3398 index = offset >> PAGE_CACHE_SHIFT;
edd1400b 3399again:
ff023aac
SB		 3400 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
3401 if (!page) {
efe120a0 3402 btrfs_err(fs_info, "find_or_create_page() failed");
ff023aac 3403 ret = -ENOMEM;
826aa0a8 3404 goto out;
ff023aac
SB		 3405 }
3406
3407 if (PageUptodate(page)) {
3408 if (PageDirty(page))
3409 goto next_page;
3410 } else {
3411 ClearPageError(page);
652f25a2
JB		 3412 err = extent_read_full_page_nolock(io_tree, page,
3413 btrfs_get_extent,
3414 nocow_ctx->mirror_num);
826aa0a8
MX		 3415 if (err) {
3416 ret = err;
ff023aac
SB		 3417 goto next_page;
3418 }
edd1400b 3419
26b25891 3420 lock_page(page);
edd1400b
MX		 3421 /*
3422 * If the page has been remove from the page cache,
3423 * the data on it is meaningless, because it may be
3424 * old one, the new data may be written into the new
3425 * page in the page cache.
3426 */
3427 if (page->mapping != inode->i_mapping) {
652f25a2 3428 unlock_page(page);
edd1400b
MX		 3429 page_cache_release(page);
3430 goto again;
3431 }
ff023aac
SB		 3432 if (!PageUptodate(page)) {
3433 ret = -EIO;
3434 goto next_page;
3435 }
3436 }
826aa0a8
MX		 3437 err = write_page_nocow(nocow_ctx->sctx,
3438 physical_for_dev_replace, page);
3439 if (err)
3440 ret = err;
ff023aac 3441next_page:
826aa0a8
MX		 3442 unlock_page(page);
3443 page_cache_release(page);
3444
3445 if (ret)
3446 break;
3447
ff023aac
SB		 3448 offset += PAGE_CACHE_SIZE;
3449 physical_for_dev_replace += PAGE_CACHE_SIZE;
3450 len -= PAGE_CACHE_SIZE;
3451 }
652f25a2
JB		 3452 ret = COPY_COMPLETE;
3453out_unlock:
3454 unlock_extent_cached(io_tree, lockstart, lockend, &cached_state,
3455 GFP_NOFS);
826aa0a8 3456out:
edd1400b 3457 mutex_unlock(&inode->i_mutex);
826aa0a8 3458 iput(inode);
ff023aac
SB		 3459 return ret;
3460}
3461
3462static int write_page_nocow(struct scrub_ctx *sctx,
3463 u64 physical_for_dev_replace, struct page *page)
3464{
3465 struct bio *bio;
3466 struct btrfs_device *dev;
3467 int ret;
ff023aac
SB		 3468
3469 dev = sctx->wr_ctx.tgtdev;
3470 if (!dev)
3471 return -EIO;
3472 if (!dev->bdev) {
3473 printk_ratelimited(KERN_WARNING
efe120a0 3474 "BTRFS: scrub write_page_nocow(bdev == NULL) is unexpected!\n");
ff023aac
SB		 3475 return -EIO;
3476 }
9be3395b 3477 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
ff023aac
SB		 3478 if (!bio) {
3479 spin_lock(&sctx->stat_lock);
3480 sctx->stat.malloc_errors++;
3481 spin_unlock(&sctx->stat_lock);
3482 return -ENOMEM;
3483 }
4f024f37
KO		 3484 bio->bi_iter.bi_size = 0;
3485 bio->bi_iter.bi_sector = physical_for_dev_replace >> 9;
ff023aac
SB		 3486 bio->bi_bdev = dev->bdev;
3487 ret = bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
3488 if (ret != PAGE_CACHE_SIZE) {
3489leave_with_eio:
3490 bio_put(bio);
3491 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
3492 return -EIO;
3493 }
ff023aac 3494
33879d45 3495 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio))
ff023aac
SB		 3496 goto leave_with_eio;
3497
3498 bio_put(bio);
3499 return 0;
3500}
Linux kernel - Deliverables
This page took 0.36739 seconds and 5 git commands to generate.