projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: Replace fs_info->scrub_* workqueue with btrfs_workqueue.
[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 98 int next_free;
0339ef2f
QW		 99 struct btrfs_work_struct
100 work;
a2de733c
AJ		 101};
102
b5d67f64 103struct scrub_block {
7a9e9987 104 struct scrub_page *pagev[SCRUB_MAX_PAGES_PER_BLOCK];
b5d67f64
SB		 105 int page_count;
106 atomic_t outstanding_pages;
107 atomic_t ref_count; /* free mem on transition to zero */
d9d181c1 108 struct scrub_ctx *sctx;
b5d67f64
SB		 109 struct {
110 unsigned int header_error:1;
111 unsigned int checksum_error:1;
112 unsigned int no_io_error_seen:1;
442a4f63 113 unsigned int generation_error:1; /* also sets header_error */
b5d67f64
SB		 114 };
115};
116
ff023aac
SB		 117struct scrub_wr_ctx {
118 struct scrub_bio *wr_curr_bio;
119 struct btrfs_device *tgtdev;
120 int pages_per_wr_bio; /* <= SCRUB_PAGES_PER_WR_BIO */
121 atomic_t flush_all_writes;
122 struct mutex wr_lock;
123};
124
d9d181c1 125struct scrub_ctx {
ff023aac 126 struct scrub_bio *bios[SCRUB_BIOS_PER_SCTX];
a36cf8b8 127 struct btrfs_root *dev_root;
a2de733c
AJ		 128 int first_free;
129 int curr;
b6bfebc1
SB		 130 atomic_t bios_in_flight;
131 atomic_t workers_pending;
a2de733c
AJ		 132 spinlock_t list_lock;
133 wait_queue_head_t list_wait;
134 u16 csum_size;
135 struct list_head csum_list;
136 atomic_t cancel_req;
8628764e 137 int readonly;
ff023aac 138 int pages_per_rd_bio;
b5d67f64
SB		 139 u32 sectorsize;
140 u32 nodesize;
141 u32 leafsize;
63a212ab
SB		 142
143 int is_dev_replace;
ff023aac 144 struct scrub_wr_ctx wr_ctx;
63a212ab 145
a2de733c
AJ		 146 /*
147 * statistics
148 */
149 struct btrfs_scrub_progress stat;
150 spinlock_t stat_lock;
151};
152
0ef8e451 153struct scrub_fixup_nodatasum {
d9d181c1 154 struct scrub_ctx *sctx;
a36cf8b8 155 struct btrfs_device *dev;
0ef8e451
JS		 156 u64 logical;
157 struct btrfs_root *root;
0339ef2f
QW		 158 struct btrfs_work_struct
159 work;
0ef8e451
JS		 160 int mirror_num;
161};
162
652f25a2
JB		 163struct scrub_nocow_inode {
164 u64 inum;
165 u64 offset;
166 u64 root;
167 struct list_head list;
168};
169
ff023aac
SB		 170struct scrub_copy_nocow_ctx {
171 struct scrub_ctx *sctx;
172 u64 logical;
173 u64 len;
174 int mirror_num;
175 u64 physical_for_dev_replace;
652f25a2 176 struct list_head inodes;
0339ef2f
QW		 177 struct btrfs_work_struct
178 work;
ff023aac
SB		 179};
180
558540c1
JS		 181struct scrub_warning {
182 struct btrfs_path *path;
183 u64 extent_item_size;
184 char *scratch_buf;
185 char *msg_buf;
186 const char *errstr;
187 sector_t sector;
188 u64 logical;
189 struct btrfs_device *dev;
190 int msg_bufsize;
191 int scratch_bufsize;
192};
193
b5d67f64 194
b6bfebc1
SB		 195static void scrub_pending_bio_inc(struct scrub_ctx *sctx);
196static void scrub_pending_bio_dec(struct scrub_ctx *sctx);
197static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx);
198static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx);
b5d67f64 199static int scrub_handle_errored_block(struct scrub_block *sblock_to_check);
d9d181c1 200static int scrub_setup_recheck_block(struct scrub_ctx *sctx,
3ec706c8 201 struct btrfs_fs_info *fs_info,
ff023aac 202 struct scrub_block *original_sblock,
b5d67f64 203 u64 length, u64 logical,
ff023aac 204 struct scrub_block *sblocks_for_recheck);
34f5c8e9
SB		 205static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
206 struct scrub_block *sblock, int is_metadata,
207 int have_csum, u8 *csum, u64 generation,
208 u16 csum_size);
b5d67f64
SB		 209static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
210 struct scrub_block *sblock,
211 int is_metadata, int have_csum,
212 const u8 *csum, u64 generation,
213 u16 csum_size);
b5d67f64
SB		 214static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
215 struct scrub_block *sblock_good,
216 int force_write);
217static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
218 struct scrub_block *sblock_good,
219 int page_num, int force_write);
ff023aac
SB		 220static void scrub_write_block_to_dev_replace(struct scrub_block *sblock);
221static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
222 int page_num);
b5d67f64
SB		 223static int scrub_checksum_data(struct scrub_block *sblock);
224static int scrub_checksum_tree_block(struct scrub_block *sblock);
225static int scrub_checksum_super(struct scrub_block *sblock);
226static void scrub_block_get(struct scrub_block *sblock);
227static void scrub_block_put(struct scrub_block *sblock);
7a9e9987
SB		 228static void scrub_page_get(struct scrub_page *spage);
229static void scrub_page_put(struct scrub_page *spage);
ff023aac
SB		 230static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
231 struct scrub_page *spage);
d9d181c1 232static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
a36cf8b8 233 u64 physical, struct btrfs_device *dev, u64 flags,
ff023aac
SB		 234 u64 gen, int mirror_num, u8 *csum, int force,
235 u64 physical_for_dev_replace);
1623edeb 236static void scrub_bio_end_io(struct bio *bio, int err);
0339ef2f 237static void scrub_bio_end_io_worker(struct btrfs_work_struct *work);
b5d67f64 238static void scrub_block_complete(struct scrub_block *sblock);
ff023aac
SB		 239static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
240 u64 extent_logical, u64 extent_len,
241 u64 *extent_physical,
242 struct btrfs_device **extent_dev,
243 int *extent_mirror_num);
244static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
245 struct scrub_wr_ctx *wr_ctx,
246 struct btrfs_fs_info *fs_info,
247 struct btrfs_device *dev,
248 int is_dev_replace);
249static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx);
250static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
251 struct scrub_page *spage);
252static void scrub_wr_submit(struct scrub_ctx *sctx);
253static void scrub_wr_bio_end_io(struct bio *bio, int err);
0339ef2f 254static void scrub_wr_bio_end_io_worker(struct btrfs_work_struct *work);
ff023aac
SB		 255static int write_page_nocow(struct scrub_ctx *sctx,
256 u64 physical_for_dev_replace, struct page *page);
257static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
652f25a2 258 struct scrub_copy_nocow_ctx *ctx);
ff023aac
SB		 259static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
260 int mirror_num, u64 physical_for_dev_replace);
0339ef2f 261static void copy_nocow_pages_worker(struct btrfs_work_struct *work);
cb7ab021 262static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
3cb0929a 263static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
1623edeb
SB		 264
265
b6bfebc1
SB		 266static void scrub_pending_bio_inc(struct scrub_ctx *sctx)
267{
268 atomic_inc(&sctx->bios_in_flight);
269}
270
271static void scrub_pending_bio_dec(struct scrub_ctx *sctx)
272{
273 atomic_dec(&sctx->bios_in_flight);
274 wake_up(&sctx->list_wait);
275}
276
cb7ab021 277static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
3cb0929a
WS		 278{
279 while (atomic_read(&fs_info->scrub_pause_req)) {
280 mutex_unlock(&fs_info->scrub_lock);
281 wait_event(fs_info->scrub_pause_wait,
282 atomic_read(&fs_info->scrub_pause_req) == 0);
283 mutex_lock(&fs_info->scrub_lock);
284 }
285}
286
cb7ab021
WS		 287static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
288{
289 atomic_inc(&fs_info->scrubs_paused);
290 wake_up(&fs_info->scrub_pause_wait);
291
292 mutex_lock(&fs_info->scrub_lock);
293 __scrub_blocked_if_needed(fs_info);
294 atomic_dec(&fs_info->scrubs_paused);
295 mutex_unlock(&fs_info->scrub_lock);
296
297 wake_up(&fs_info->scrub_pause_wait);
298}
299
b6bfebc1
SB		 300/*
301 * used for workers that require transaction commits (i.e., for the
302 * NOCOW case)
303 */
304static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx)
305{
306 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
307
308 /*
309 * increment scrubs_running to prevent cancel requests from
310 * completing as long as a worker is running. we must also
311 * increment scrubs_paused to prevent deadlocking on pause
312 * requests used for transactions commits (as the worker uses a
313 * transaction context). it is safe to regard the worker
314 * as paused for all matters practical. effectively, we only
315 * avoid cancellation requests from completing.
316 */
317 mutex_lock(&fs_info->scrub_lock);
318 atomic_inc(&fs_info->scrubs_running);
319 atomic_inc(&fs_info->scrubs_paused);
320 mutex_unlock(&fs_info->scrub_lock);
32a44789
WS		 321
322 /*
323 * check if @scrubs_running=@scrubs_paused condition
324 * inside wait_event() is not an atomic operation.
325 * which means we may inc/dec @scrub_running/paused
326 * at any time. Let's wake up @scrub_pause_wait as
327 * much as we can to let commit transaction blocked less.
328 */
329 wake_up(&fs_info->scrub_pause_wait);
330
b6bfebc1
SB		 331 atomic_inc(&sctx->workers_pending);
332}
333
334/* used for workers that require transaction commits */
335static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx)
336{
337 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
338
339 /*
340 * see scrub_pending_trans_workers_inc() why we're pretending
341 * to be paused in the scrub counters
342 */
343 mutex_lock(&fs_info->scrub_lock);
344 atomic_dec(&fs_info->scrubs_running);
345 atomic_dec(&fs_info->scrubs_paused);
346 mutex_unlock(&fs_info->scrub_lock);
347 atomic_dec(&sctx->workers_pending);
348 wake_up(&fs_info->scrub_pause_wait);
349 wake_up(&sctx->list_wait);
350}
351
d9d181c1 352static void scrub_free_csums(struct scrub_ctx *sctx)
a2de733c 353{
d9d181c1 354 while (!list_empty(&sctx->csum_list)) {
a2de733c 355 struct btrfs_ordered_sum *sum;
d9d181c1 356 sum = list_first_entry(&sctx->csum_list,
a2de733c
AJ		 357 struct btrfs_ordered_sum, list);
358 list_del(&sum->list);
359 kfree(sum);
360 }
361}
362
d9d181c1 363static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx)
a2de733c
AJ		 364{
365 int i;
a2de733c 366
d9d181c1 367 if (!sctx)
a2de733c
AJ		 368 return;
369
ff023aac
SB		 370 scrub_free_wr_ctx(&sctx->wr_ctx);
371
b5d67f64 372 /* this can happen when scrub is cancelled */
d9d181c1
SB		 373 if (sctx->curr != -1) {
374 struct scrub_bio *sbio = sctx->bios[sctx->curr];
b5d67f64
SB		 375
376 for (i = 0; i < sbio->page_count; i++) {
ff023aac 377 WARN_ON(!sbio->pagev[i]->page);
b5d67f64
SB		 378 scrub_block_put(sbio->pagev[i]->sblock);
379 }
380 bio_put(sbio->bio);
381 }
382
ff023aac 383 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
d9d181c1 384 struct scrub_bio *sbio = sctx->bios[i];
a2de733c
AJ		 385
386 if (!sbio)
387 break;
a2de733c
AJ		 388 kfree(sbio);
389 }
390
d9d181c1
SB		 391 scrub_free_csums(sctx);
392 kfree(sctx);
a2de733c
AJ		 393}
394
395static noinline_for_stack
63a212ab 396struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace)
a2de733c 397{
d9d181c1 398 struct scrub_ctx *sctx;
a2de733c 399 int i;
a2de733c 400 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
ff023aac
SB		 401 int pages_per_rd_bio;
402 int ret;
a2de733c 403
ff023aac
SB		 404 /*
405 * the setting of pages_per_rd_bio is correct for scrub but might
406 * be wrong for the dev_replace code where we might read from
407 * different devices in the initial huge bios. However, that
408 * code is able to correctly handle the case when adding a page
409 * to a bio fails.
410 */
411 if (dev->bdev)
412 pages_per_rd_bio = min_t(int, SCRUB_PAGES_PER_RD_BIO,
413 bio_get_nr_vecs(dev->bdev));
414 else
415 pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO;
d9d181c1
SB		 416 sctx = kzalloc(sizeof(*sctx), GFP_NOFS);
417 if (!sctx)
a2de733c 418 goto nomem;
63a212ab 419 sctx->is_dev_replace = is_dev_replace;
ff023aac 420 sctx->pages_per_rd_bio = pages_per_rd_bio;
d9d181c1 421 sctx->curr = -1;
a36cf8b8 422 sctx->dev_root = dev->dev_root;
ff023aac 423 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
a2de733c
AJ		 424 struct scrub_bio *sbio;
425
426 sbio = kzalloc(sizeof(*sbio), GFP_NOFS);
427 if (!sbio)
428 goto nomem;
d9d181c1 429 sctx->bios[i] = sbio;
a2de733c 430
a2de733c 431 sbio->index = i;
d9d181c1 432 sbio->sctx = sctx;
b5d67f64 433 sbio->page_count = 0;
0339ef2f
QW		 434 btrfs_init_work(&sbio->work, scrub_bio_end_io_worker,
435 NULL, NULL);
a2de733c 436
ff023aac 437 if (i != SCRUB_BIOS_PER_SCTX - 1)
d9d181c1 438 sctx->bios[i]->next_free = i + 1;
0ef8e451 439 else
d9d181c1
SB		 440 sctx->bios[i]->next_free = -1;
441 }
442 sctx->first_free = 0;
443 sctx->nodesize = dev->dev_root->nodesize;
444 sctx->leafsize = dev->dev_root->leafsize;
445 sctx->sectorsize = dev->dev_root->sectorsize;
b6bfebc1
SB		 446 atomic_set(&sctx->bios_in_flight, 0);
447 atomic_set(&sctx->workers_pending, 0);
d9d181c1
SB		 448 atomic_set(&sctx->cancel_req, 0);
449 sctx->csum_size = btrfs_super_csum_size(fs_info->super_copy);
450 INIT_LIST_HEAD(&sctx->csum_list);
451
452 spin_lock_init(&sctx->list_lock);
453 spin_lock_init(&sctx->stat_lock);
454 init_waitqueue_head(&sctx->list_wait);
ff023aac
SB		 455
456 ret = scrub_setup_wr_ctx(sctx, &sctx->wr_ctx, fs_info,
457 fs_info->dev_replace.tgtdev, is_dev_replace);
458 if (ret) {
459 scrub_free_ctx(sctx);
460 return ERR_PTR(ret);
461 }
d9d181c1 462 return sctx;
a2de733c
AJ		 463
464nomem:
d9d181c1 465 scrub_free_ctx(sctx);
a2de733c
AJ		 466 return ERR_PTR(-ENOMEM);
467}
468
ff023aac
SB		 469static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root,
470 void *warn_ctx)
558540c1
JS		 471{
472 u64 isize;
473 u32 nlink;
474 int ret;
475 int i;
476 struct extent_buffer *eb;
477 struct btrfs_inode_item *inode_item;
ff023aac 478 struct scrub_warning *swarn = warn_ctx;
558540c1
JS		 479 struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info;
480 struct inode_fs_paths *ipath = NULL;
481 struct btrfs_root *local_root;
482 struct btrfs_key root_key;
483
484 root_key.objectid = root;
485 root_key.type = BTRFS_ROOT_ITEM_KEY;
486 root_key.offset = (u64)-1;
487 local_root = btrfs_read_fs_root_no_name(fs_info, &root_key);
488 if (IS_ERR(local_root)) {
489 ret = PTR_ERR(local_root);
490 goto err;
491 }
492
493 ret = inode_item_info(inum, 0, local_root, swarn->path);
494 if (ret) {
495 btrfs_release_path(swarn->path);
496 goto err;
497 }
498
499 eb = swarn->path->nodes[0];
500 inode_item = btrfs_item_ptr(eb, swarn->path->slots[0],
501 struct btrfs_inode_item);
502 isize = btrfs_inode_size(eb, inode_item);
503 nlink = btrfs_inode_nlink(eb, inode_item);
504 btrfs_release_path(swarn->path);
505
506 ipath = init_ipath(4096, local_root, swarn->path);
26bdef54
DC		 507 if (IS_ERR(ipath)) {
508 ret = PTR_ERR(ipath);
509 ipath = NULL;
510 goto err;
511 }
558540c1
JS		 512 ret = paths_from_inode(inum, ipath);
513
514 if (ret < 0)
515 goto err;
516
517 /*
518 * we deliberately ignore the bit ipath might have been too small to
519 * hold all of the paths here
520 */
521 for (i = 0; i < ipath->fspath->elem_cnt; ++i)
efe120a0 522 printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
558540c1
JS		 523 "%s, sector %llu, root %llu, inode %llu, offset %llu, "
524 "length %llu, links %u (path: %s)\n", swarn->errstr,
606686ee 525 swarn->logical, rcu_str_deref(swarn->dev->name),
558540c1
JS		 526 (unsigned long long)swarn->sector, root, inum, offset,
527 min(isize - offset, (u64)PAGE_SIZE), nlink,
745c4d8e 528 (char *)(unsigned long)ipath->fspath->val[i]);
558540c1
JS		 529
530 free_ipath(ipath);
531 return 0;
532
533err:
efe120a0 534 printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
558540c1
JS		 535 "%s, sector %llu, root %llu, inode %llu, offset %llu: path "
536 "resolving failed with ret=%d\n", swarn->errstr,
606686ee 537 swarn->logical, rcu_str_deref(swarn->dev->name),
558540c1
JS		 538 (unsigned long long)swarn->sector, root, inum, offset, ret);
539
540 free_ipath(ipath);
541 return 0;
542}
543
b5d67f64 544static void scrub_print_warning(const char *errstr, struct scrub_block *sblock)
558540c1 545{
a36cf8b8
SB		 546 struct btrfs_device *dev;
547 struct btrfs_fs_info *fs_info;
558540c1
JS		 548 struct btrfs_path *path;
549 struct btrfs_key found_key;
550 struct extent_buffer *eb;
551 struct btrfs_extent_item *ei;
552 struct scrub_warning swarn;
69917e43
LB		 553 unsigned long ptr = 0;
554 u64 extent_item_pos;
555 u64 flags = 0;
558540c1 556 u64 ref_root;
69917e43 557 u32 item_size;
558540c1 558 u8 ref_level;
558540c1 559 const int bufsize = 4096;
69917e43 560 int ret;
558540c1 561
a36cf8b8 562 WARN_ON(sblock->page_count < 1);
7a9e9987 563 dev = sblock->pagev[0]->dev;
a36cf8b8
SB		 564 fs_info = sblock->sctx->dev_root->fs_info;
565
558540c1
JS		 566 path = btrfs_alloc_path();
567
568 swarn.scratch_buf = kmalloc(bufsize, GFP_NOFS);
569 swarn.msg_buf = kmalloc(bufsize, GFP_NOFS);
7a9e9987
SB		 570 swarn.sector = (sblock->pagev[0]->physical) >> 9;
571 swarn.logical = sblock->pagev[0]->logical;
558540c1 572 swarn.errstr = errstr;
a36cf8b8 573 swarn.dev = NULL;
558540c1
JS		 574 swarn.msg_bufsize = bufsize;
575 swarn.scratch_bufsize = bufsize;
576
577 if (!path || !swarn.scratch_buf || !swarn.msg_buf)
578 goto out;
579
69917e43
LB		 580 ret = extent_from_logical(fs_info, swarn.logical, path, &found_key,
581 &flags);
558540c1
JS		 582 if (ret < 0)
583 goto out;
584
4692cf58 585 extent_item_pos = swarn.logical - found_key.objectid;
558540c1
JS		 586 swarn.extent_item_size = found_key.offset;
587
588 eb = path->nodes[0];
589 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
590 item_size = btrfs_item_size_nr(eb, path->slots[0]);
591
69917e43 592 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
558540c1
JS		 593 do {
594 ret = tree_backref_for_extent(&ptr, eb, ei, item_size,
595 &ref_root, &ref_level);
606686ee 596 printk_in_rcu(KERN_WARNING
efe120a0 597 "BTRFS: %s at logical %llu on dev %s, "
558540c1 598 "sector %llu: metadata %s (level %d) in tree "
606686ee
JB		 599 "%llu\n", errstr, swarn.logical,
600 rcu_str_deref(dev->name),
558540c1
JS		 601 (unsigned long long)swarn.sector,
602 ref_level ? "node" : "leaf",
603 ret < 0 ? -1 : ref_level,
604 ret < 0 ? -1 : ref_root);
605 } while (ret != 1);
d8fe29e9 606 btrfs_release_path(path);
558540c1 607 } else {
d8fe29e9 608 btrfs_release_path(path);
558540c1 609 swarn.path = path;
a36cf8b8 610 swarn.dev = dev;
7a3ae2f8
JS		 611 iterate_extent_inodes(fs_info, found_key.objectid,
612 extent_item_pos, 1,
558540c1
JS		 613 scrub_print_warning_inode, &swarn);
614 }
615
616out:
617 btrfs_free_path(path);
618 kfree(swarn.scratch_buf);
619 kfree(swarn.msg_buf);
620}
621
ff023aac 622static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx)
0ef8e451 623{
5da6fcbc 624 struct page *page = NULL;
0ef8e451 625 unsigned long index;
ff023aac 626 struct scrub_fixup_nodatasum *fixup = fixup_ctx;
0ef8e451 627 int ret;
5da6fcbc 628 int corrected = 0;
0ef8e451 629 struct btrfs_key key;
5da6fcbc 630 struct inode *inode = NULL;
6f1c3605 631 struct btrfs_fs_info *fs_info;
0ef8e451
JS		 632 u64 end = offset + PAGE_SIZE - 1;
633 struct btrfs_root *local_root;
6f1c3605 634 int srcu_index;
0ef8e451
JS		 635
636 key.objectid = root;
637 key.type = BTRFS_ROOT_ITEM_KEY;
638 key.offset = (u64)-1;
6f1c3605
LB		 639
640 fs_info = fixup->root->fs_info;
641 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
642
643 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
644 if (IS_ERR(local_root)) {
645 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
0ef8e451 646 return PTR_ERR(local_root);
6f1c3605 647 }
0ef8e451
JS		 648
649 key.type = BTRFS_INODE_ITEM_KEY;
650 key.objectid = inum;
651 key.offset = 0;
6f1c3605
LB		 652 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
653 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
0ef8e451
JS		 654 if (IS_ERR(inode))
655 return PTR_ERR(inode);
656
0ef8e451
JS		 657 index = offset >> PAGE_CACHE_SHIFT;
658
659 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
5da6fcbc
JS		 660 if (!page) {
661 ret = -ENOMEM;
662 goto out;
663 }
664
665 if (PageUptodate(page)) {
5da6fcbc
JS		 666 if (PageDirty(page)) {
667 /*
668 * we need to write the data to the defect sector. the
669 * data that was in that sector is not in memory,
670 * because the page was modified. we must not write the
671 * modified page to that sector.
672 *
673 * TODO: what could be done here: wait for the delalloc
674 * runner to write out that page (might involve
675 * COW) and see whether the sector is still
676 * referenced afterwards.
677 *
678 * For the meantime, we'll treat this error
679 * incorrectable, although there is a chance that a
680 * later scrub will find the bad sector again and that
681 * there's no dirty page in memory, then.
682 */
683 ret = -EIO;
684 goto out;
685 }
3ec706c8
SB		 686 fs_info = BTRFS_I(inode)->root->fs_info;
687 ret = repair_io_failure(fs_info, offset, PAGE_SIZE,
5da6fcbc
JS		 688 fixup->logical, page,
689 fixup->mirror_num);
690 unlock_page(page);
691 corrected = !ret;
692 } else {
693 /*
694 * we need to get good data first. the general readpage path
695 * will call repair_io_failure for us, we just have to make
696 * sure we read the bad mirror.
697 */
698 ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
699 EXTENT_DAMAGED, GFP_NOFS);
700 if (ret) {
701 /* set_extent_bits should give proper error */
702 WARN_ON(ret > 0);
703 if (ret > 0)
704 ret = -EFAULT;
705 goto out;
706 }
707
708 ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page,
709 btrfs_get_extent,
710 fixup->mirror_num);
711 wait_on_page_locked(page);
712
713 corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset,
714 end, EXTENT_DAMAGED, 0, NULL);
715 if (!corrected)
716 clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
717 EXTENT_DAMAGED, GFP_NOFS);
718 }
719
720out:
721 if (page)
722 put_page(page);
723 if (inode)
724 iput(inode);
0ef8e451
JS		 725
726 if (ret < 0)
727 return ret;
728
729 if (ret == 0 && corrected) {
730 /*
731 * we only need to call readpage for one of the inodes belonging
732 * to this extent. so make iterate_extent_inodes stop
733 */
734 return 1;
735 }
736
737 return -EIO;
738}
739
0339ef2f 740static void scrub_fixup_nodatasum(struct btrfs_work_struct *work)
0ef8e451
JS		 741{
742 int ret;
743 struct scrub_fixup_nodatasum *fixup;
d9d181c1 744 struct scrub_ctx *sctx;
0ef8e451 745 struct btrfs_trans_handle *trans = NULL;
0ef8e451
JS		 746 struct btrfs_path *path;
747 int uncorrectable = 0;
748
749 fixup = container_of(work, struct scrub_fixup_nodatasum, work);
d9d181c1 750 sctx = fixup->sctx;
0ef8e451
JS		 751
752 path = btrfs_alloc_path();
753 if (!path) {
d9d181c1
SB		 754 spin_lock(&sctx->stat_lock);
755 ++sctx->stat.malloc_errors;
756 spin_unlock(&sctx->stat_lock);
0ef8e451
JS		 757 uncorrectable = 1;
758 goto out;
759 }
760
761 trans = btrfs_join_transaction(fixup->root);
762 if (IS_ERR(trans)) {
763 uncorrectable = 1;
764 goto out;
765 }
766
767 /*
768 * the idea is to trigger a regular read through the standard path. we
769 * read a page from the (failed) logical address by specifying the
770 * corresponding copynum of the failed sector. thus, that readpage is
771 * expected to fail.
772 * that is the point where on-the-fly error correction will kick in
773 * (once it's finished) and rewrite the failed sector if a good copy
774 * can be found.
775 */
776 ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info,
777 path, scrub_fixup_readpage,
778 fixup);
779 if (ret < 0) {
780 uncorrectable = 1;
781 goto out;
782 }
783 WARN_ON(ret != 1);
784
d9d181c1
SB		 785 spin_lock(&sctx->stat_lock);
786 ++sctx->stat.corrected_errors;
787 spin_unlock(&sctx->stat_lock);
0ef8e451
JS		 788
789out:
790 if (trans && !IS_ERR(trans))
791 btrfs_end_transaction(trans, fixup->root);
792 if (uncorrectable) {
d9d181c1
SB		 793 spin_lock(&sctx->stat_lock);
794 ++sctx->stat.uncorrectable_errors;
795 spin_unlock(&sctx->stat_lock);
ff023aac
SB		 796 btrfs_dev_replace_stats_inc(
797 &sctx->dev_root->fs_info->dev_replace.
798 num_uncorrectable_read_errors);
efe120a0
FH		 799 printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
800 "unable to fixup (nodatasum) error at logical %llu on dev %s\n",
c1c9ff7c 801 fixup->logical, rcu_str_deref(fixup->dev->name));
0ef8e451
JS		 802 }
803
804 btrfs_free_path(path);
805 kfree(fixup);
806
b6bfebc1 807 scrub_pending_trans_workers_dec(sctx);
0ef8e451
JS		 808}
809
a2de733c 810/*
b5d67f64
SB		 811 * scrub_handle_errored_block gets called when either verification of the
812 * pages failed or the bio failed to read, e.g. with EIO. In the latter
813 * case, this function handles all pages in the bio, even though only one
814 * may be bad.
815 * The goal of this function is to repair the errored block by using the
816 * contents of one of the mirrors.
a2de733c 817 */
b5d67f64 818static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
a2de733c 819{
d9d181c1 820 struct scrub_ctx *sctx = sblock_to_check->sctx;
a36cf8b8 821 struct btrfs_device *dev;
b5d67f64
SB		 822 struct btrfs_fs_info *fs_info;
823 u64 length;
824 u64 logical;
825 u64 generation;
826 unsigned int failed_mirror_index;
827 unsigned int is_metadata;
828 unsigned int have_csum;
829 u8 *csum;
830 struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */
831 struct scrub_block *sblock_bad;
832 int ret;
833 int mirror_index;
834 int page_num;
835 int success;
558540c1 836 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
b5d67f64
SB		 837 DEFAULT_RATELIMIT_BURST);
838
839 BUG_ON(sblock_to_check->page_count < 1);
a36cf8b8 840 fs_info = sctx->dev_root->fs_info;
4ded4f63
SB		 841 if (sblock_to_check->pagev[0]->flags & BTRFS_EXTENT_FLAG_SUPER) {
842 /*
843 * if we find an error in a super block, we just report it.
844 * They will get written with the next transaction commit
845 * anyway
846 */
847 spin_lock(&sctx->stat_lock);
848 ++sctx->stat.super_errors;
849 spin_unlock(&sctx->stat_lock);
850 return 0;
851 }
b5d67f64 852 length = sblock_to_check->page_count * PAGE_SIZE;
7a9e9987
SB		 853 logical = sblock_to_check->pagev[0]->logical;
854 generation = sblock_to_check->pagev[0]->generation;
855 BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1);
856 failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1;
857 is_metadata = !(sblock_to_check->pagev[0]->flags &
b5d67f64 858 BTRFS_EXTENT_FLAG_DATA);
7a9e9987
SB		 859 have_csum = sblock_to_check->pagev[0]->have_csum;
860 csum = sblock_to_check->pagev[0]->csum;
861 dev = sblock_to_check->pagev[0]->dev;
13db62b7 862
ff023aac
SB		 863 if (sctx->is_dev_replace && !is_metadata && !have_csum) {
864 sblocks_for_recheck = NULL;
865 goto nodatasum_case;
866 }
867
b5d67f64
SB		 868 /*
869 * read all mirrors one after the other. This includes to
870 * re-read the extent or metadata block that failed (that was
871 * the cause that this fixup code is called) another time,
872 * page by page this time in order to know which pages
873 * caused I/O errors and which ones are good (for all mirrors).
874 * It is the goal to handle the situation when more than one
875 * mirror contains I/O errors, but the errors do not
876 * overlap, i.e. the data can be repaired by selecting the
877 * pages from those mirrors without I/O error on the
878 * particular pages. One example (with blocks >= 2 * PAGE_SIZE)
879 * would be that mirror #1 has an I/O error on the first page,
880 * the second page is good, and mirror #2 has an I/O error on
881 * the second page, but the first page is good.
882 * Then the first page of the first mirror can be repaired by
883 * taking the first page of the second mirror, and the
884 * second page of the second mirror can be repaired by
885 * copying the contents of the 2nd page of the 1st mirror.
886 * One more note: if the pages of one mirror contain I/O
887 * errors, the checksum cannot be verified. In order to get
888 * the best data for repairing, the first attempt is to find
889 * a mirror without I/O errors and with a validated checksum.
890 * Only if this is not possible, the pages are picked from
891 * mirrors with I/O errors without considering the checksum.
892 * If the latter is the case, at the end, the checksum of the
893 * repaired area is verified in order to correctly maintain
894 * the statistics.
895 */
896
897 sblocks_for_recheck = kzalloc(BTRFS_MAX_MIRRORS *
898 sizeof(*sblocks_for_recheck),
899 GFP_NOFS);
900 if (!sblocks_for_recheck) {
d9d181c1
SB		 901 spin_lock(&sctx->stat_lock);
902 sctx->stat.malloc_errors++;
903 sctx->stat.read_errors++;
904 sctx->stat.uncorrectable_errors++;
905 spin_unlock(&sctx->stat_lock);
a36cf8b8 906 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
b5d67f64 907 goto out;
a2de733c
AJ		 908 }
909
b5d67f64 910 /* setup the context, map the logical blocks and alloc the pages */
ff023aac 911 ret = scrub_setup_recheck_block(sctx, fs_info, sblock_to_check, length,
b5d67f64
SB		 912 logical, sblocks_for_recheck);
913 if (ret) {
d9d181c1
SB		 914 spin_lock(&sctx->stat_lock);
915 sctx->stat.read_errors++;
916 sctx->stat.uncorrectable_errors++;
917 spin_unlock(&sctx->stat_lock);
a36cf8b8 918 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
b5d67f64
SB		 919 goto out;
920 }
921 BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS);
922 sblock_bad = sblocks_for_recheck + failed_mirror_index;
13db62b7 923
b5d67f64 924 /* build and submit the bios for the failed mirror, check checksums */
34f5c8e9
SB		 925 scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum,
926 csum, generation, sctx->csum_size);
a2de733c 927
b5d67f64
SB		 928 if (!sblock_bad->header_error && !sblock_bad->checksum_error &&
929 sblock_bad->no_io_error_seen) {
930 /*
931 * the error disappeared after reading page by page, or
932 * the area was part of a huge bio and other parts of the
933 * bio caused I/O errors, or the block layer merged several
934 * read requests into one and the error is caused by a
935 * different bio (usually one of the two latter cases is
936 * the cause)
937 */
d9d181c1
SB		 938 spin_lock(&sctx->stat_lock);
939 sctx->stat.unverified_errors++;
940 spin_unlock(&sctx->stat_lock);
a2de733c 941
ff023aac
SB		 942 if (sctx->is_dev_replace)
943 scrub_write_block_to_dev_replace(sblock_bad);
b5d67f64 944 goto out;
a2de733c 945 }
a2de733c 946
b5d67f64 947 if (!sblock_bad->no_io_error_seen) {
d9d181c1
SB		 948 spin_lock(&sctx->stat_lock);
949 sctx->stat.read_errors++;
950 spin_unlock(&sctx->stat_lock);
b5d67f64
SB		 951 if (__ratelimit(&_rs))
952 scrub_print_warning("i/o error", sblock_to_check);
a36cf8b8 953 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
b5d67f64 954 } else if (sblock_bad->checksum_error) {
d9d181c1
SB		 955 spin_lock(&sctx->stat_lock);
956 sctx->stat.csum_errors++;
957 spin_unlock(&sctx->stat_lock);
b5d67f64
SB		 958 if (__ratelimit(&_rs))
959 scrub_print_warning("checksum error", sblock_to_check);
a36cf8b8 960 btrfs_dev_stat_inc_and_print(dev,
442a4f63 961 BTRFS_DEV_STAT_CORRUPTION_ERRS);
b5d67f64 962 } else if (sblock_bad->header_error) {
d9d181c1
SB		 963 spin_lock(&sctx->stat_lock);
964 sctx->stat.verify_errors++;
965 spin_unlock(&sctx->stat_lock);
b5d67f64
SB		 966 if (__ratelimit(&_rs))
967 scrub_print_warning("checksum/header error",
968 sblock_to_check);
442a4f63 969 if (sblock_bad->generation_error)
a36cf8b8 970 btrfs_dev_stat_inc_and_print(dev,
442a4f63
SB		 971 BTRFS_DEV_STAT_GENERATION_ERRS);
972 else
a36cf8b8 973 btrfs_dev_stat_inc_and_print(dev,
442a4f63 974 BTRFS_DEV_STAT_CORRUPTION_ERRS);
b5d67f64 975 }
a2de733c 976
33ef30ad
ID		 977 if (sctx->readonly) {
978 ASSERT(!sctx->is_dev_replace);
979 goto out;
980 }
a2de733c 981
b5d67f64
SB		 982 if (!is_metadata && !have_csum) {
983 struct scrub_fixup_nodatasum *fixup_nodatasum;
a2de733c 984
ff023aac
SB		 985nodatasum_case:
986 WARN_ON(sctx->is_dev_replace);
987
b5d67f64
SB		 988 /*
989 * !is_metadata and !have_csum, this means that the data
990 * might not be COW'ed, that it might be modified
991 * concurrently. The general strategy to work on the
992 * commit root does not help in the case when COW is not
993 * used.
994 */
995 fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS);
996 if (!fixup_nodatasum)
997 goto did_not_correct_error;
d9d181c1 998 fixup_nodatasum->sctx = sctx;
a36cf8b8 999 fixup_nodatasum->dev = dev;
b5d67f64
SB		 1000 fixup_nodatasum->logical = logical;
1001 fixup_nodatasum->root = fs_info->extent_root;
1002 fixup_nodatasum->mirror_num = failed_mirror_index + 1;
b6bfebc1 1003 scrub_pending_trans_workers_inc(sctx);
0339ef2f
QW		 1004 btrfs_init_work(&fixup_nodatasum->work, scrub_fixup_nodatasum,
1005 NULL, NULL);
1006 btrfs_queue_work(fs_info->scrub_workers,
1007 &fixup_nodatasum->work);
b5d67f64 1008 goto out;
a2de733c
AJ		 1009 }
1010
b5d67f64
SB		 1011 /*
1012 * now build and submit the bios for the other mirrors, check
cb2ced73
SB		 1013 * checksums.
1014 * First try to pick the mirror which is completely without I/O
b5d67f64
SB		 1015 * errors and also does not have a checksum error.
1016 * If one is found, and if a checksum is present, the full block
1017 * that is known to contain an error is rewritten. Afterwards
1018 * the block is known to be corrected.
1019 * If a mirror is found which is completely correct, and no
1020 * checksum is present, only those pages are rewritten that had
1021 * an I/O error in the block to be repaired, since it cannot be
1022 * determined, which copy of the other pages is better (and it
1023 * could happen otherwise that a correct page would be
1024 * overwritten by a bad one).
1025 */
1026 for (mirror_index = 0;
1027 mirror_index < BTRFS_MAX_MIRRORS &&
1028 sblocks_for_recheck[mirror_index].page_count > 0;
1029 mirror_index++) {
cb2ced73 1030 struct scrub_block *sblock_other;
b5d67f64 1031
cb2ced73
SB		 1032 if (mirror_index == failed_mirror_index)
1033 continue;
1034 sblock_other = sblocks_for_recheck + mirror_index;
1035
1036 /* build and submit the bios, check checksums */
34f5c8e9
SB		 1037 scrub_recheck_block(fs_info, sblock_other, is_metadata,
1038 have_csum, csum, generation,
1039 sctx->csum_size);
1040
1041 if (!sblock_other->header_error &&
b5d67f64
SB		 1042 !sblock_other->checksum_error &&
1043 sblock_other->no_io_error_seen) {
ff023aac
SB		 1044 if (sctx->is_dev_replace) {
1045 scrub_write_block_to_dev_replace(sblock_other);
1046 } else {
1047 int force_write = is_metadata || have_csum;
1048
1049 ret = scrub_repair_block_from_good_copy(
1050 sblock_bad, sblock_other,
1051 force_write);
1052 }
b5d67f64
SB		 1053 if (0 == ret)
1054 goto corrected_error;
1055 }
1056 }
a2de733c
AJ		 1057
1058 /*
ff023aac
SB		 1059 * for dev_replace, pick good pages and write to the target device.
1060 */
1061 if (sctx->is_dev_replace) {
1062 success = 1;
1063 for (page_num = 0; page_num < sblock_bad->page_count;
1064 page_num++) {
1065 int sub_success;
1066
1067 sub_success = 0;
1068 for (mirror_index = 0;
1069 mirror_index < BTRFS_MAX_MIRRORS &&
1070 sblocks_for_recheck[mirror_index].page_count > 0;
1071 mirror_index++) {
1072 struct scrub_block *sblock_other =
1073 sblocks_for_recheck + mirror_index;
1074 struct scrub_page *page_other =
1075 sblock_other->pagev[page_num];
1076
1077 if (!page_other->io_error) {
1078 ret = scrub_write_page_to_dev_replace(
1079 sblock_other, page_num);
1080 if (ret == 0) {
1081 /* succeeded for this page */
1082 sub_success = 1;
1083 break;
1084 } else {
1085 btrfs_dev_replace_stats_inc(
1086 &sctx->dev_root->
1087 fs_info->dev_replace.
1088 num_write_errors);
1089 }
1090 }
1091 }
1092
1093 if (!sub_success) {
1094 /*
1095 * did not find a mirror to fetch the page
1096 * from. scrub_write_page_to_dev_replace()
1097 * handles this case (page->io_error), by
1098 * filling the block with zeros before
1099 * submitting the write request
1100 */
1101 success = 0;
1102 ret = scrub_write_page_to_dev_replace(
1103 sblock_bad, page_num);
1104 if (ret)
1105 btrfs_dev_replace_stats_inc(
1106 &sctx->dev_root->fs_info->
1107 dev_replace.num_write_errors);
1108 }
1109 }
1110
1111 goto out;
1112 }
1113
1114 /*
1115 * for regular scrub, repair those pages that are errored.
1116 * In case of I/O errors in the area that is supposed to be
b5d67f64
SB		 1117 * repaired, continue by picking good copies of those pages.
1118 * Select the good pages from mirrors to rewrite bad pages from
1119 * the area to fix. Afterwards verify the checksum of the block
1120 * that is supposed to be repaired. This verification step is
1121 * only done for the purpose of statistic counting and for the
1122 * final scrub report, whether errors remain.
1123 * A perfect algorithm could make use of the checksum and try
1124 * all possible combinations of pages from the different mirrors
1125 * until the checksum verification succeeds. For example, when
1126 * the 2nd page of mirror #1 faces I/O errors, and the 2nd page
1127 * of mirror #2 is readable but the final checksum test fails,
1128 * then the 2nd page of mirror #3 could be tried, whether now
1129 * the final checksum succeedes. But this would be a rare
1130 * exception and is therefore not implemented. At least it is
1131 * avoided that the good copy is overwritten.
1132 * A more useful improvement would be to pick the sectors
1133 * without I/O error based on sector sizes (512 bytes on legacy
1134 * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one
1135 * mirror could be repaired by taking 512 byte of a different
1136 * mirror, even if other 512 byte sectors in the same PAGE_SIZE
1137 * area are unreadable.
a2de733c 1138 */
a2de733c 1139
b5d67f64
SB		 1140 /* can only fix I/O errors from here on */
1141 if (sblock_bad->no_io_error_seen)
1142 goto did_not_correct_error;
1143
1144 success = 1;
1145 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
7a9e9987 1146 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
b5d67f64
SB		 1147
1148 if (!page_bad->io_error)
a2de733c 1149 continue;
b5d67f64
SB		 1150
1151 for (mirror_index = 0;
1152 mirror_index < BTRFS_MAX_MIRRORS &&
1153 sblocks_for_recheck[mirror_index].page_count > 0;
1154 mirror_index++) {
1155 struct scrub_block *sblock_other = sblocks_for_recheck +
1156 mirror_index;
7a9e9987
SB		 1157 struct scrub_page *page_other = sblock_other->pagev[
1158 page_num];
b5d67f64
SB		 1159
1160 if (!page_other->io_error) {
1161 ret = scrub_repair_page_from_good_copy(
1162 sblock_bad, sblock_other, page_num, 0);
1163 if (0 == ret) {
1164 page_bad->io_error = 0;
1165 break; /* succeeded for this page */
1166 }
1167 }
96e36920 1168 }
a2de733c 1169
b5d67f64
SB		 1170 if (page_bad->io_error) {
1171 /* did not find a mirror to copy the page from */
1172 success = 0;
1173 }
a2de733c 1174 }
a2de733c 1175
b5d67f64
SB		 1176 if (success) {
1177 if (is_metadata || have_csum) {
1178 /*
1179 * need to verify the checksum now that all
1180 * sectors on disk are repaired (the write
1181 * request for data to be repaired is on its way).
1182 * Just be lazy and use scrub_recheck_block()
1183 * which re-reads the data before the checksum
1184 * is verified, but most likely the data comes out
1185 * of the page cache.
1186 */
34f5c8e9
SB		 1187 scrub_recheck_block(fs_info, sblock_bad,
1188 is_metadata, have_csum, csum,
1189 generation, sctx->csum_size);
1190 if (!sblock_bad->header_error &&
b5d67f64
SB		 1191 !sblock_bad->checksum_error &&
1192 sblock_bad->no_io_error_seen)
1193 goto corrected_error;
1194 else
1195 goto did_not_correct_error;
1196 } else {
1197corrected_error:
d9d181c1
SB		 1198 spin_lock(&sctx->stat_lock);
1199 sctx->stat.corrected_errors++;
1200 spin_unlock(&sctx->stat_lock);
606686ee 1201 printk_ratelimited_in_rcu(KERN_ERR
efe120a0 1202 "BTRFS: fixed up error at logical %llu on dev %s\n",
c1c9ff7c 1203 logical, rcu_str_deref(dev->name));
8628764e 1204 }
b5d67f64
SB		 1205 } else {
1206did_not_correct_error:
d9d181c1
SB		 1207 spin_lock(&sctx->stat_lock);
1208 sctx->stat.uncorrectable_errors++;
1209 spin_unlock(&sctx->stat_lock);
606686ee 1210 printk_ratelimited_in_rcu(KERN_ERR
efe120a0 1211 "BTRFS: unable to fixup (regular) error at logical %llu on dev %s\n",
c1c9ff7c 1212 logical, rcu_str_deref(dev->name));
96e36920 1213 }
a2de733c 1214
b5d67f64
SB		 1215out:
1216 if (sblocks_for_recheck) {
1217 for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS;
1218 mirror_index++) {
1219 struct scrub_block *sblock = sblocks_for_recheck +
1220 mirror_index;
1221 int page_index;
1222
7a9e9987
SB		 1223 for (page_index = 0; page_index < sblock->page_count;
1224 page_index++) {
1225 sblock->pagev[page_index]->sblock = NULL;
1226 scrub_page_put(sblock->pagev[page_index]);
1227 }
b5d67f64
SB		 1228 }
1229 kfree(sblocks_for_recheck);
1230 }
a2de733c 1231
b5d67f64
SB		 1232 return 0;
1233}
a2de733c 1234
d9d181c1 1235static int scrub_setup_recheck_block(struct scrub_ctx *sctx,
3ec706c8 1236 struct btrfs_fs_info *fs_info,
ff023aac 1237 struct scrub_block *original_sblock,
b5d67f64
SB		 1238 u64 length, u64 logical,
1239 struct scrub_block *sblocks_for_recheck)
1240{
1241 int page_index;
1242 int mirror_index;
1243 int ret;
1244
1245 /*
7a9e9987 1246 * note: the two members ref_count and outstanding_pages
b5d67f64
SB		 1247 * are not used (and not set) in the blocks that are used for
1248 * the recheck procedure
1249 */
1250
1251 page_index = 0;
1252 while (length > 0) {
1253 u64 sublen = min_t(u64, length, PAGE_SIZE);
1254 u64 mapped_length = sublen;
1255 struct btrfs_bio *bbio = NULL;
a2de733c 1256
b5d67f64
SB		 1257 /*
1258 * with a length of PAGE_SIZE, each returned stripe
1259 * represents one mirror
1260 */
29a8d9a0
SB		 1261 ret = btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS, logical,
1262 &mapped_length, &bbio, 0);
b5d67f64
SB		 1263 if (ret || !bbio || mapped_length < sublen) {
1264 kfree(bbio);
1265 return -EIO;
1266 }
a2de733c 1267
ff023aac 1268 BUG_ON(page_index >= SCRUB_PAGES_PER_RD_BIO);
b5d67f64
SB		 1269 for (mirror_index = 0; mirror_index < (int)bbio->num_stripes;
1270 mirror_index++) {
1271 struct scrub_block *sblock;
1272 struct scrub_page *page;
1273
1274 if (mirror_index >= BTRFS_MAX_MIRRORS)
1275 continue;
1276
1277 sblock = sblocks_for_recheck + mirror_index;
7a9e9987
SB		 1278 sblock->sctx = sctx;
1279 page = kzalloc(sizeof(*page), GFP_NOFS);
1280 if (!page) {
1281leave_nomem:
d9d181c1
SB		 1282 spin_lock(&sctx->stat_lock);
1283 sctx->stat.malloc_errors++;
1284 spin_unlock(&sctx->stat_lock);
cf93dcce 1285 kfree(bbio);
b5d67f64
SB		 1286 return -ENOMEM;
1287 }
7a9e9987
SB		 1288 scrub_page_get(page);
1289 sblock->pagev[page_index] = page;
1290 page->logical = logical;
1291 page->physical = bbio->stripes[mirror_index].physical;
ff023aac
SB		 1292 BUG_ON(page_index >= original_sblock->page_count);
1293 page->physical_for_dev_replace =
1294 original_sblock->pagev[page_index]->
1295 physical_for_dev_replace;
7a9e9987
SB		 1296 /* for missing devices, dev->bdev is NULL */
1297 page->dev = bbio->stripes[mirror_index].dev;
1298 page->mirror_num = mirror_index + 1;
b5d67f64 1299 sblock->page_count++;
7a9e9987
SB		 1300 page->page = alloc_page(GFP_NOFS);
1301 if (!page->page)
1302 goto leave_nomem;
b5d67f64
SB		 1303 }
1304 kfree(bbio);
1305 length -= sublen;
1306 logical += sublen;
1307 page_index++;
1308 }
1309
1310 return 0;
96e36920
ID		 1311}
1312
b5d67f64
SB		 1313/*
1314 * this function will check the on disk data for checksum errors, header
1315 * errors and read I/O errors. If any I/O errors happen, the exact pages
1316 * which are errored are marked as being bad. The goal is to enable scrub
1317 * to take those pages that are not errored from all the mirrors so that
1318 * the pages that are errored in the just handled mirror can be repaired.
1319 */
34f5c8e9
SB		 1320static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
1321 struct scrub_block *sblock, int is_metadata,
1322 int have_csum, u8 *csum, u64 generation,
1323 u16 csum_size)
96e36920 1324{
b5d67f64 1325 int page_num;
96e36920 1326
b5d67f64
SB		 1327 sblock->no_io_error_seen = 1;
1328 sblock->header_error = 0;
1329 sblock->checksum_error = 0;
96e36920 1330
b5d67f64
SB		 1331 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1332 struct bio *bio;
7a9e9987 1333 struct scrub_page *page = sblock->pagev[page_num];
b5d67f64 1334
442a4f63 1335 if (page->dev->bdev == NULL) {
ea9947b4
SB		 1336 page->io_error = 1;
1337 sblock->no_io_error_seen = 0;
1338 continue;
1339 }
1340
7a9e9987 1341 WARN_ON(!page->page);
9be3395b 1342 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
34f5c8e9
SB		 1343 if (!bio) {
1344 page->io_error = 1;
1345 sblock->no_io_error_seen = 0;
1346 continue;
1347 }
442a4f63 1348 bio->bi_bdev = page->dev->bdev;
b5d67f64 1349 bio->bi_sector = page->physical >> 9;
b5d67f64 1350
34f5c8e9 1351 bio_add_page(bio, page->page, PAGE_SIZE, 0);
c170bbb4 1352 if (btrfsic_submit_bio_wait(READ, bio))
b5d67f64 1353 sblock->no_io_error_seen = 0;
c170bbb4 1354
b5d67f64
SB		 1355 bio_put(bio);
1356 }
96e36920 1357
b5d67f64
SB		 1358 if (sblock->no_io_error_seen)
1359 scrub_recheck_block_checksum(fs_info, sblock, is_metadata,
1360 have_csum, csum, generation,
1361 csum_size);
1362
34f5c8e9 1363 return;
a2de733c
AJ		 1364}
1365
b5d67f64
SB		 1366static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
1367 struct scrub_block *sblock,
1368 int is_metadata, int have_csum,
1369 const u8 *csum, u64 generation,
1370 u16 csum_size)
a2de733c 1371{
b5d67f64
SB		 1372 int page_num;
1373 u8 calculated_csum[BTRFS_CSUM_SIZE];
1374 u32 crc = ~(u32)0;
b5d67f64
SB		 1375 void *mapped_buffer;
1376
7a9e9987 1377 WARN_ON(!sblock->pagev[0]->page);
b5d67f64
SB		 1378 if (is_metadata) {
1379 struct btrfs_header *h;
1380
7a9e9987 1381 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
b5d67f64
SB		 1382 h = (struct btrfs_header *)mapped_buffer;
1383
3cae210f 1384 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h) ||
b5d67f64
SB		 1385 memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE) ||
1386 memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
442a4f63 1387 BTRFS_UUID_SIZE)) {
b5d67f64 1388 sblock->header_error = 1;
3cae210f 1389 } else if (generation != btrfs_stack_header_generation(h)) {
442a4f63
SB		 1390 sblock->header_error = 1;
1391 sblock->generation_error = 1;
1392 }
b5d67f64
SB		 1393 csum = h->csum;
1394 } else {
1395 if (!have_csum)
1396 return;
a2de733c 1397
7a9e9987 1398 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
b5d67f64 1399 }
a2de733c 1400
b5d67f64
SB		 1401 for (page_num = 0;;) {
1402 if (page_num == 0 && is_metadata)
b0496686 1403 crc = btrfs_csum_data(
b5d67f64
SB		 1404 ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE,
1405 crc, PAGE_SIZE - BTRFS_CSUM_SIZE);
1406 else
b0496686 1407 crc = btrfs_csum_data(mapped_buffer, crc, PAGE_SIZE);
b5d67f64 1408
9613bebb 1409 kunmap_atomic(mapped_buffer);
b5d67f64
SB		 1410 page_num++;
1411 if (page_num >= sblock->page_count)
1412 break;
7a9e9987 1413 WARN_ON(!sblock->pagev[page_num]->page);
b5d67f64 1414
7a9e9987 1415 mapped_buffer = kmap_atomic(sblock->pagev[page_num]->page);
b5d67f64
SB		 1416 }
1417
1418 btrfs_csum_final(crc, calculated_csum);
1419 if (memcmp(calculated_csum, csum, csum_size))
1420 sblock->checksum_error = 1;
a2de733c
AJ		 1421}
1422
b5d67f64
SB		 1423static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
1424 struct scrub_block *sblock_good,
1425 int force_write)
1426{
1427 int page_num;
1428 int ret = 0;
96e36920 1429
b5d67f64
SB		 1430 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
1431 int ret_sub;
96e36920 1432
b5d67f64
SB		 1433 ret_sub = scrub_repair_page_from_good_copy(sblock_bad,
1434 sblock_good,
1435 page_num,
1436 force_write);
1437 if (ret_sub)
1438 ret = ret_sub;
a2de733c 1439 }
b5d67f64
SB		 1440
1441 return ret;
1442}
1443
1444static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
1445 struct scrub_block *sblock_good,
1446 int page_num, int force_write)
1447{
7a9e9987
SB		 1448 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1449 struct scrub_page *page_good = sblock_good->pagev[page_num];
b5d67f64 1450
7a9e9987
SB		 1451 BUG_ON(page_bad->page == NULL);
1452 BUG_ON(page_good->page == NULL);
b5d67f64
SB		 1453 if (force_write || sblock_bad->header_error ||
1454 sblock_bad->checksum_error || page_bad->io_error) {
1455 struct bio *bio;
1456 int ret;
b5d67f64 1457
ff023aac 1458 if (!page_bad->dev->bdev) {
efe120a0
FH		 1459 printk_ratelimited(KERN_WARNING "BTRFS: "
1460 "scrub_repair_page_from_good_copy(bdev == NULL) "
1461 "is unexpected!\n");
ff023aac
SB		 1462 return -EIO;
1463 }
1464
9be3395b 1465 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
e627ee7b
TI		 1466 if (!bio)
1467 return -EIO;
442a4f63 1468 bio->bi_bdev = page_bad->dev->bdev;
b5d67f64 1469 bio->bi_sector = page_bad->physical >> 9;
b5d67f64
SB		 1470
1471 ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0);
1472 if (PAGE_SIZE != ret) {
1473 bio_put(bio);
1474 return -EIO;
13db62b7 1475 }
b5d67f64 1476
c170bbb4 1477 if (btrfsic_submit_bio_wait(WRITE, bio)) {
442a4f63
SB		 1478 btrfs_dev_stat_inc_and_print(page_bad->dev,
1479 BTRFS_DEV_STAT_WRITE_ERRS);
ff023aac
SB		 1480 btrfs_dev_replace_stats_inc(
1481 &sblock_bad->sctx->dev_root->fs_info->
1482 dev_replace.num_write_errors);
442a4f63
SB		 1483 bio_put(bio);
1484 return -EIO;
1485 }
b5d67f64 1486 bio_put(bio);
a2de733c
AJ		 1487 }
1488
b5d67f64
SB		 1489 return 0;
1490}
1491
ff023aac
SB		 1492static void scrub_write_block_to_dev_replace(struct scrub_block *sblock)
1493{
1494 int page_num;
1495
1496 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1497 int ret;
1498
1499 ret = scrub_write_page_to_dev_replace(sblock, page_num);
1500 if (ret)
1501 btrfs_dev_replace_stats_inc(
1502 &sblock->sctx->dev_root->fs_info->dev_replace.
1503 num_write_errors);
1504 }
1505}
1506
1507static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
1508 int page_num)
1509{
1510 struct scrub_page *spage = sblock->pagev[page_num];
1511
1512 BUG_ON(spage->page == NULL);
1513 if (spage->io_error) {
1514 void *mapped_buffer = kmap_atomic(spage->page);
1515
1516 memset(mapped_buffer, 0, PAGE_CACHE_SIZE);
1517 flush_dcache_page(spage->page);
1518 kunmap_atomic(mapped_buffer);
1519 }
1520 return scrub_add_page_to_wr_bio(sblock->sctx, spage);
1521}
1522
1523static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
1524 struct scrub_page *spage)
1525{
1526 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1527 struct scrub_bio *sbio;
1528 int ret;
1529
1530 mutex_lock(&wr_ctx->wr_lock);
1531again:
1532 if (!wr_ctx->wr_curr_bio) {
1533 wr_ctx->wr_curr_bio = kzalloc(sizeof(*wr_ctx->wr_curr_bio),
1534 GFP_NOFS);
1535 if (!wr_ctx->wr_curr_bio) {
1536 mutex_unlock(&wr_ctx->wr_lock);
1537 return -ENOMEM;
1538 }
1539 wr_ctx->wr_curr_bio->sctx = sctx;
1540 wr_ctx->wr_curr_bio->page_count = 0;
1541 }
1542 sbio = wr_ctx->wr_curr_bio;
1543 if (sbio->page_count == 0) {
1544 struct bio *bio;
1545
1546 sbio->physical = spage->physical_for_dev_replace;
1547 sbio->logical = spage->logical;
1548 sbio->dev = wr_ctx->tgtdev;
1549 bio = sbio->bio;
1550 if (!bio) {
9be3395b 1551 bio = btrfs_io_bio_alloc(GFP_NOFS, wr_ctx->pages_per_wr_bio);
ff023aac
SB		 1552 if (!bio) {
1553 mutex_unlock(&wr_ctx->wr_lock);
1554 return -ENOMEM;
1555 }
1556 sbio->bio = bio;
1557 }
1558
1559 bio->bi_private = sbio;
1560 bio->bi_end_io = scrub_wr_bio_end_io;
1561 bio->bi_bdev = sbio->dev->bdev;
1562 bio->bi_sector = sbio->physical >> 9;
1563 sbio->err = 0;
1564 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
1565 spage->physical_for_dev_replace ||
1566 sbio->logical + sbio->page_count * PAGE_SIZE !=
1567 spage->logical) {
1568 scrub_wr_submit(sctx);
1569 goto again;
1570 }
1571
1572 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
1573 if (ret != PAGE_SIZE) {
1574 if (sbio->page_count < 1) {
1575 bio_put(sbio->bio);
1576 sbio->bio = NULL;
1577 mutex_unlock(&wr_ctx->wr_lock);
1578 return -EIO;
1579 }
1580 scrub_wr_submit(sctx);
1581 goto again;
1582 }
1583
1584 sbio->pagev[sbio->page_count] = spage;
1585 scrub_page_get(spage);
1586 sbio->page_count++;
1587 if (sbio->page_count == wr_ctx->pages_per_wr_bio)
1588 scrub_wr_submit(sctx);
1589 mutex_unlock(&wr_ctx->wr_lock);
1590
1591 return 0;
1592}
1593
1594static void scrub_wr_submit(struct scrub_ctx *sctx)
1595{
1596 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1597 struct scrub_bio *sbio;
1598
1599 if (!wr_ctx->wr_curr_bio)
1600 return;
1601
1602 sbio = wr_ctx->wr_curr_bio;
1603 wr_ctx->wr_curr_bio = NULL;
1604 WARN_ON(!sbio->bio->bi_bdev);
1605 scrub_pending_bio_inc(sctx);
1606 /* process all writes in a single worker thread. Then the block layer
1607 * orders the requests before sending them to the driver which
1608 * doubled the write performance on spinning disks when measured
1609 * with Linux 3.5 */
1610 btrfsic_submit_bio(WRITE, sbio->bio);
1611}
1612
1613static void scrub_wr_bio_end_io(struct bio *bio, int err)
1614{
1615 struct scrub_bio *sbio = bio->bi_private;
1616 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
1617
1618 sbio->err = err;
1619 sbio->bio = bio;
1620
0339ef2f
QW		 1621 btrfs_init_work(&sbio->work, scrub_wr_bio_end_io_worker, NULL, NULL);
1622 btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work);
ff023aac
SB		 1623}
1624
0339ef2f 1625static void scrub_wr_bio_end_io_worker(struct btrfs_work_struct *work)
ff023aac
SB		 1626{
1627 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
1628 struct scrub_ctx *sctx = sbio->sctx;
1629 int i;
1630
1631 WARN_ON(sbio->page_count > SCRUB_PAGES_PER_WR_BIO);
1632 if (sbio->err) {
1633 struct btrfs_dev_replace *dev_replace =
1634 &sbio->sctx->dev_root->fs_info->dev_replace;
1635
1636 for (i = 0; i < sbio->page_count; i++) {
1637 struct scrub_page *spage = sbio->pagev[i];
1638
1639 spage->io_error = 1;
1640 btrfs_dev_replace_stats_inc(&dev_replace->
1641 num_write_errors);
1642 }
1643 }
1644
1645 for (i = 0; i < sbio->page_count; i++)
1646 scrub_page_put(sbio->pagev[i]);
1647
1648 bio_put(sbio->bio);
1649 kfree(sbio);
1650 scrub_pending_bio_dec(sctx);
1651}
1652
1653static int scrub_checksum(struct scrub_block *sblock)
b5d67f64
SB		 1654{
1655 u64 flags;
1656 int ret;
1657
7a9e9987
SB		 1658 WARN_ON(sblock->page_count < 1);
1659 flags = sblock->pagev[0]->flags;
b5d67f64
SB		 1660 ret = 0;
1661 if (flags & BTRFS_EXTENT_FLAG_DATA)
1662 ret = scrub_checksum_data(sblock);
1663 else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1664 ret = scrub_checksum_tree_block(sblock);
1665 else if (flags & BTRFS_EXTENT_FLAG_SUPER)
1666 (void)scrub_checksum_super(sblock);
1667 else
1668 WARN_ON(1);
1669 if (ret)
1670 scrub_handle_errored_block(sblock);
ff023aac
SB		 1671
1672 return ret;
a2de733c
AJ		 1673}
1674
b5d67f64 1675static int scrub_checksum_data(struct scrub_block *sblock)
a2de733c 1676{
d9d181c1 1677 struct scrub_ctx *sctx = sblock->sctx;
a2de733c 1678 u8 csum[BTRFS_CSUM_SIZE];
b5d67f64
SB		 1679 u8 *on_disk_csum;
1680 struct page *page;
1681 void *buffer;
a2de733c
AJ		 1682 u32 crc = ~(u32)0;
1683 int fail = 0;
b5d67f64
SB		 1684 u64 len;
1685 int index;
a2de733c 1686
b5d67f64 1687 BUG_ON(sblock->page_count < 1);
7a9e9987 1688 if (!sblock->pagev[0]->have_csum)
a2de733c
AJ		 1689 return 0;
1690
7a9e9987
SB		 1691 on_disk_csum = sblock->pagev[0]->csum;
1692 page = sblock->pagev[0]->page;
9613bebb 1693 buffer = kmap_atomic(page);
b5d67f64 1694
d9d181c1 1695 len = sctx->sectorsize;
b5d67f64
SB		 1696 index = 0;
1697 for (;;) {
1698 u64 l = min_t(u64, len, PAGE_SIZE);
1699
b0496686 1700 crc = btrfs_csum_data(buffer, crc, l);
9613bebb 1701 kunmap_atomic(buffer);
b5d67f64
SB		 1702 len -= l;
1703 if (len == 0)
1704 break;
1705 index++;
1706 BUG_ON(index >= sblock->page_count);
7a9e9987
SB		 1707 BUG_ON(!sblock->pagev[index]->page);
1708 page = sblock->pagev[index]->page;
9613bebb 1709 buffer = kmap_atomic(page);
b5d67f64
SB		 1710 }
1711
a2de733c 1712 btrfs_csum_final(crc, csum);
d9d181c1 1713 if (memcmp(csum, on_disk_csum, sctx->csum_size))
a2de733c
AJ		 1714 fail = 1;
1715
a2de733c
AJ		 1716 return fail;
1717}
1718
b5d67f64 1719static int scrub_checksum_tree_block(struct scrub_block *sblock)
a2de733c 1720{
d9d181c1 1721 struct scrub_ctx *sctx = sblock->sctx;
a2de733c 1722 struct btrfs_header *h;
a36cf8b8 1723 struct btrfs_root *root = sctx->dev_root;
a2de733c 1724 struct btrfs_fs_info *fs_info = root->fs_info;
b5d67f64
SB		 1725 u8 calculated_csum[BTRFS_CSUM_SIZE];
1726 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1727 struct page *page;
1728 void *mapped_buffer;
1729 u64 mapped_size;
1730 void *p;
a2de733c
AJ		 1731 u32 crc = ~(u32)0;
1732 int fail = 0;
1733 int crc_fail = 0;
b5d67f64
SB		 1734 u64 len;
1735 int index;
1736
1737 BUG_ON(sblock->page_count < 1);
7a9e9987 1738 page = sblock->pagev[0]->page;
9613bebb 1739 mapped_buffer = kmap_atomic(page);
b5d67f64 1740 h = (struct btrfs_header *)mapped_buffer;
d9d181c1 1741 memcpy(on_disk_csum, h->csum, sctx->csum_size);
a2de733c
AJ		 1742
1743 /*
1744 * we don't use the getter functions here, as we
1745 * a) don't have an extent buffer and
1746 * b) the page is already kmapped
1747 */
a2de733c 1748
3cae210f 1749 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h))
a2de733c
AJ		 1750 ++fail;
1751
3cae210f 1752 if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h))
a2de733c
AJ		 1753 ++fail;
1754
1755 if (memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
1756 ++fail;
1757
1758 if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
1759 BTRFS_UUID_SIZE))
1760 ++fail;
1761
ff023aac 1762 WARN_ON(sctx->nodesize != sctx->leafsize);
d9d181c1 1763 len = sctx->nodesize - BTRFS_CSUM_SIZE;
b5d67f64
SB		 1764 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
1765 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
1766 index = 0;
1767 for (;;) {
1768 u64 l = min_t(u64, len, mapped_size);
1769
b0496686 1770 crc = btrfs_csum_data(p, crc, l);
9613bebb 1771 kunmap_atomic(mapped_buffer);
b5d67f64
SB		 1772 len -= l;
1773 if (len == 0)
1774 break;
1775 index++;
1776 BUG_ON(index >= sblock->page_count);
7a9e9987
SB		 1777 BUG_ON(!sblock->pagev[index]->page);
1778 page = sblock->pagev[index]->page;
9613bebb 1779 mapped_buffer = kmap_atomic(page);
b5d67f64
SB		 1780 mapped_size = PAGE_SIZE;
1781 p = mapped_buffer;
1782 }
1783
1784 btrfs_csum_final(crc, calculated_csum);
d9d181c1 1785 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
a2de733c
AJ		 1786 ++crc_fail;
1787
a2de733c
AJ		 1788 return fail || crc_fail;
1789}
1790
b5d67f64 1791static int scrub_checksum_super(struct scrub_block *sblock)
a2de733c
AJ		 1792{
1793 struct btrfs_super_block *s;
d9d181c1 1794 struct scrub_ctx *sctx = sblock->sctx;
a36cf8b8 1795 struct btrfs_root *root = sctx->dev_root;
a2de733c 1796 struct btrfs_fs_info *fs_info = root->fs_info;
b5d67f64
SB		 1797 u8 calculated_csum[BTRFS_CSUM_SIZE];
1798 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1799 struct page *page;
1800 void *mapped_buffer;
1801 u64 mapped_size;
1802 void *p;
a2de733c 1803 u32 crc = ~(u32)0;
442a4f63
SB		 1804 int fail_gen = 0;
1805 int fail_cor = 0;
b5d67f64
SB		 1806 u64 len;
1807 int index;
a2de733c 1808
b5d67f64 1809 BUG_ON(sblock->page_count < 1);
7a9e9987 1810 page = sblock->pagev[0]->page;
9613bebb 1811 mapped_buffer = kmap_atomic(page);
b5d67f64 1812 s = (struct btrfs_super_block *)mapped_buffer;
d9d181c1 1813 memcpy(on_disk_csum, s->csum, sctx->csum_size);
a2de733c 1814
3cae210f 1815 if (sblock->pagev[0]->logical != btrfs_super_bytenr(s))
442a4f63 1816 ++fail_cor;
a2de733c 1817
3cae210f 1818 if (sblock->pagev[0]->generation != btrfs_super_generation(s))
442a4f63 1819 ++fail_gen;
a2de733c
AJ		 1820
1821 if (memcmp(s->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
442a4f63 1822 ++fail_cor;
a2de733c 1823
b5d67f64
SB		 1824 len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE;
1825 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
1826 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
1827 index = 0;
1828 for (;;) {
1829 u64 l = min_t(u64, len, mapped_size);
1830
b0496686 1831 crc = btrfs_csum_data(p, crc, l);
9613bebb 1832 kunmap_atomic(mapped_buffer);
b5d67f64
SB		 1833 len -= l;
1834 if (len == 0)
1835 break;
1836 index++;
1837 BUG_ON(index >= sblock->page_count);
7a9e9987
SB		 1838 BUG_ON(!sblock->pagev[index]->page);
1839 page = sblock->pagev[index]->page;
9613bebb 1840 mapped_buffer = kmap_atomic(page);
b5d67f64
SB		 1841 mapped_size = PAGE_SIZE;
1842 p = mapped_buffer;
1843 }
1844
1845 btrfs_csum_final(crc, calculated_csum);
d9d181c1 1846 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
442a4f63 1847 ++fail_cor;
a2de733c 1848
442a4f63 1849 if (fail_cor + fail_gen) {
a2de733c
AJ		 1850 /*
1851 * if we find an error in a super block, we just report it.
1852 * They will get written with the next transaction commit
1853 * anyway
1854 */
d9d181c1
SB		 1855 spin_lock(&sctx->stat_lock);
1856 ++sctx->stat.super_errors;
1857 spin_unlock(&sctx->stat_lock);
442a4f63 1858 if (fail_cor)
7a9e9987 1859 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
442a4f63
SB		 1860 BTRFS_DEV_STAT_CORRUPTION_ERRS);
1861 else
7a9e9987 1862 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
442a4f63 1863 BTRFS_DEV_STAT_GENERATION_ERRS);
a2de733c
AJ		 1864 }
1865
442a4f63 1866 return fail_cor + fail_gen;
a2de733c
AJ		 1867}
1868
b5d67f64
SB		 1869static void scrub_block_get(struct scrub_block *sblock)
1870{
1871 atomic_inc(&sblock->ref_count);
1872}
1873
1874static void scrub_block_put(struct scrub_block *sblock)
1875{
1876 if (atomic_dec_and_test(&sblock->ref_count)) {
1877 int i;
1878
1879 for (i = 0; i < sblock->page_count; i++)
7a9e9987 1880 scrub_page_put(sblock->pagev[i]);
b5d67f64
SB		 1881 kfree(sblock);
1882 }
1883}
1884
7a9e9987
SB		 1885static void scrub_page_get(struct scrub_page *spage)
1886{
1887 atomic_inc(&spage->ref_count);
1888}
1889
1890static void scrub_page_put(struct scrub_page *spage)
1891{
1892 if (atomic_dec_and_test(&spage->ref_count)) {
1893 if (spage->page)
1894 __free_page(spage->page);
1895 kfree(spage);
1896 }
1897}
1898
d9d181c1 1899static void scrub_submit(struct scrub_ctx *sctx)
a2de733c
AJ		 1900{
1901 struct scrub_bio *sbio;
1902
d9d181c1 1903 if (sctx->curr == -1)
1623edeb 1904 return;
a2de733c 1905
d9d181c1
SB		 1906 sbio = sctx->bios[sctx->curr];
1907 sctx->curr = -1;
b6bfebc1 1908 scrub_pending_bio_inc(sctx);
a2de733c 1909
ff023aac
SB		 1910 if (!sbio->bio->bi_bdev) {
1911 /*
1912 * this case should not happen. If btrfs_map_block() is
1913 * wrong, it could happen for dev-replace operations on
1914 * missing devices when no mirrors are available, but in
1915 * this case it should already fail the mount.
1916 * This case is handled correctly (but _very_ slowly).
1917 */
1918 printk_ratelimited(KERN_WARNING
efe120a0 1919 "BTRFS: scrub_submit(bio bdev == NULL) is unexpected!\n");
ff023aac
SB		 1920 bio_endio(sbio->bio, -EIO);
1921 } else {
1922 btrfsic_submit_bio(READ, sbio->bio);
1923 }
a2de733c
AJ		 1924}
1925
ff023aac
SB		 1926static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
1927 struct scrub_page *spage)
a2de733c 1928{
b5d67f64 1929 struct scrub_block *sblock = spage->sblock;
a2de733c 1930 struct scrub_bio *sbio;
69f4cb52 1931 int ret;
a2de733c
AJ		 1932
1933again:
1934 /*
1935 * grab a fresh bio or wait for one to become available
1936 */
d9d181c1
SB		 1937 while (sctx->curr == -1) {
1938 spin_lock(&sctx->list_lock);
1939 sctx->curr = sctx->first_free;
1940 if (sctx->curr != -1) {
1941 sctx->first_free = sctx->bios[sctx->curr]->next_free;
1942 sctx->bios[sctx->curr]->next_free = -1;
1943 sctx->bios[sctx->curr]->page_count = 0;
1944 spin_unlock(&sctx->list_lock);
a2de733c 1945 } else {
d9d181c1
SB		 1946 spin_unlock(&sctx->list_lock);
1947 wait_event(sctx->list_wait, sctx->first_free != -1);
a2de733c
AJ		 1948 }
1949 }
d9d181c1 1950 sbio = sctx->bios[sctx->curr];
b5d67f64 1951 if (sbio->page_count == 0) {
69f4cb52
AJ		 1952 struct bio *bio;
1953
b5d67f64
SB		 1954 sbio->physical = spage->physical;
1955 sbio->logical = spage->logical;
a36cf8b8 1956 sbio->dev = spage->dev;
b5d67f64
SB		 1957 bio = sbio->bio;
1958 if (!bio) {
9be3395b 1959 bio = btrfs_io_bio_alloc(GFP_NOFS, sctx->pages_per_rd_bio);
b5d67f64
SB		 1960 if (!bio)
1961 return -ENOMEM;
1962 sbio->bio = bio;
1963 }
69f4cb52
AJ		 1964
1965 bio->bi_private = sbio;
1966 bio->bi_end_io = scrub_bio_end_io;
a36cf8b8
SB		 1967 bio->bi_bdev = sbio->dev->bdev;
1968 bio->bi_sector = sbio->physical >> 9;
69f4cb52 1969 sbio->err = 0;
b5d67f64
SB		 1970 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
1971 spage->physical ||
1972 sbio->logical + sbio->page_count * PAGE_SIZE !=
a36cf8b8
SB		 1973 spage->logical ||
1974 sbio->dev != spage->dev) {
d9d181c1 1975 scrub_submit(sctx);
a2de733c
AJ		 1976 goto again;
1977 }
69f4cb52 1978
b5d67f64
SB		 1979 sbio->pagev[sbio->page_count] = spage;
1980 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
1981 if (ret != PAGE_SIZE) {
1982 if (sbio->page_count < 1) {
1983 bio_put(sbio->bio);
1984 sbio->bio = NULL;
1985 return -EIO;
1986 }
d9d181c1 1987 scrub_submit(sctx);
69f4cb52
AJ		 1988 goto again;
1989 }
1990
ff023aac 1991 scrub_block_get(sblock); /* one for the page added to the bio */
b5d67f64
SB		 1992 atomic_inc(&sblock->outstanding_pages);
1993 sbio->page_count++;
ff023aac 1994 if (sbio->page_count == sctx->pages_per_rd_bio)
d9d181c1 1995 scrub_submit(sctx);
b5d67f64
SB		 1996
1997 return 0;
1998}
1999
d9d181c1 2000static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
a36cf8b8 2001 u64 physical, struct btrfs_device *dev, u64 flags,
ff023aac
SB		 2002 u64 gen, int mirror_num, u8 *csum, int force,
2003 u64 physical_for_dev_replace)
b5d67f64
SB		 2004{
2005 struct scrub_block *sblock;
2006 int index;
2007
2008 sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
2009 if (!sblock) {
d9d181c1
SB		 2010 spin_lock(&sctx->stat_lock);
2011 sctx->stat.malloc_errors++;
2012 spin_unlock(&sctx->stat_lock);
b5d67f64 2013 return -ENOMEM;
a2de733c 2014 }
b5d67f64 2015
7a9e9987
SB		 2016 /* one ref inside this function, plus one for each page added to
2017 * a bio later on */
b5d67f64 2018 atomic_set(&sblock->ref_count, 1);
d9d181c1 2019 sblock->sctx = sctx;
b5d67f64
SB		 2020 sblock->no_io_error_seen = 1;
2021
2022 for (index = 0; len > 0; index++) {
7a9e9987 2023 struct scrub_page *spage;
b5d67f64
SB		 2024 u64 l = min_t(u64, len, PAGE_SIZE);
2025
7a9e9987
SB		 2026 spage = kzalloc(sizeof(*spage), GFP_NOFS);
2027 if (!spage) {
2028leave_nomem:
d9d181c1
SB		 2029 spin_lock(&sctx->stat_lock);
2030 sctx->stat.malloc_errors++;
2031 spin_unlock(&sctx->stat_lock);
7a9e9987 2032 scrub_block_put(sblock);
b5d67f64
SB		 2033 return -ENOMEM;
2034 }
7a9e9987
SB		 2035 BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
2036 scrub_page_get(spage);
2037 sblock->pagev[index] = spage;
b5d67f64 2038 spage->sblock = sblock;
a36cf8b8 2039 spage->dev = dev;
b5d67f64
SB		 2040 spage->flags = flags;
2041 spage->generation = gen;
2042 spage->logical = logical;
2043 spage->physical = physical;
ff023aac 2044 spage->physical_for_dev_replace = physical_for_dev_replace;
b5d67f64
SB		 2045 spage->mirror_num = mirror_num;
2046 if (csum) {
2047 spage->have_csum = 1;
d9d181c1 2048 memcpy(spage->csum, csum, sctx->csum_size);
b5d67f64
SB		 2049 } else {
2050 spage->have_csum = 0;
2051 }
2052 sblock->page_count++;
7a9e9987
SB		 2053 spage->page = alloc_page(GFP_NOFS);
2054 if (!spage->page)
2055 goto leave_nomem;
b5d67f64
SB		 2056 len -= l;
2057 logical += l;
2058 physical += l;
ff023aac 2059 physical_for_dev_replace += l;
b5d67f64
SB		 2060 }
2061
7a9e9987 2062 WARN_ON(sblock->page_count == 0);
b5d67f64 2063 for (index = 0; index < sblock->page_count; index++) {
7a9e9987 2064 struct scrub_page *spage = sblock->pagev[index];
1bc87793
AJ		 2065 int ret;
2066
ff023aac 2067 ret = scrub_add_page_to_rd_bio(sctx, spage);
b5d67f64
SB		 2068 if (ret) {
2069 scrub_block_put(sblock);
1bc87793 2070 return ret;
b5d67f64 2071 }
1bc87793 2072 }
a2de733c 2073
b5d67f64 2074 if (force)
d9d181c1 2075 scrub_submit(sctx);
a2de733c 2076
b5d67f64
SB		 2077 /* last one frees, either here or in bio completion for last page */
2078 scrub_block_put(sblock);
a2de733c
AJ		 2079 return 0;
2080}
2081
b5d67f64
SB		 2082static void scrub_bio_end_io(struct bio *bio, int err)
2083{
2084 struct scrub_bio *sbio = bio->bi_private;
a36cf8b8 2085 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
b5d67f64
SB		 2086
2087 sbio->err = err;
2088 sbio->bio = bio;
2089
0339ef2f 2090 btrfs_queue_work(fs_info->scrub_workers, &sbio->work);
b5d67f64
SB		 2091}
2092
0339ef2f 2093static void scrub_bio_end_io_worker(struct btrfs_work_struct *work)
b5d67f64
SB		 2094{
2095 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
d9d181c1 2096 struct scrub_ctx *sctx = sbio->sctx;
b5d67f64
SB		 2097 int i;
2098
ff023aac 2099 BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO);
b5d67f64
SB		 2100 if (sbio->err) {
2101 for (i = 0; i < sbio->page_count; i++) {
2102 struct scrub_page *spage = sbio->pagev[i];
2103
2104 spage->io_error = 1;
2105 spage->sblock->no_io_error_seen = 0;
2106 }
2107 }
2108
2109 /* now complete the scrub_block items that have all pages completed */
2110 for (i = 0; i < sbio->page_count; i++) {
2111 struct scrub_page *spage = sbio->pagev[i];
2112 struct scrub_block *sblock = spage->sblock;
2113
2114 if (atomic_dec_and_test(&sblock->outstanding_pages))
2115 scrub_block_complete(sblock);
2116 scrub_block_put(sblock);
2117 }
2118
b5d67f64
SB		 2119 bio_put(sbio->bio);
2120 sbio->bio = NULL;
d9d181c1
SB		 2121 spin_lock(&sctx->list_lock);
2122 sbio->next_free = sctx->first_free;
2123 sctx->first_free = sbio->index;
2124 spin_unlock(&sctx->list_lock);
ff023aac
SB		 2125
2126 if (sctx->is_dev_replace &&
2127 atomic_read(&sctx->wr_ctx.flush_all_writes)) {
2128 mutex_lock(&sctx->wr_ctx.wr_lock);
2129 scrub_wr_submit(sctx);
2130 mutex_unlock(&sctx->wr_ctx.wr_lock);
2131 }
2132
b6bfebc1 2133 scrub_pending_bio_dec(sctx);
b5d67f64
SB		 2134}
2135
2136static void scrub_block_complete(struct scrub_block *sblock)
2137{
ff023aac 2138 if (!sblock->no_io_error_seen) {
b5d67f64 2139 scrub_handle_errored_block(sblock);
ff023aac
SB		 2140 } else {
2141 /*
2142 * if has checksum error, write via repair mechanism in
2143 * dev replace case, otherwise write here in dev replace
2144 * case.
2145 */
2146 if (!scrub_checksum(sblock) && sblock->sctx->is_dev_replace)
2147 scrub_write_block_to_dev_replace(sblock);
2148 }
b5d67f64
SB		 2149}
2150
d9d181c1 2151static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len,
a2de733c
AJ		 2152 u8 *csum)
2153{
2154 struct btrfs_ordered_sum *sum = NULL;
f51a4a18 2155 unsigned long index;
a2de733c 2156 unsigned long num_sectors;
a2de733c 2157
d9d181c1
SB		 2158 while (!list_empty(&sctx->csum_list)) {
2159 sum = list_first_entry(&sctx->csum_list,
a2de733c
AJ		 2160 struct btrfs_ordered_sum, list);
2161 if (sum->bytenr > logical)
2162 return 0;
2163 if (sum->bytenr + sum->len > logical)
2164 break;
2165
d9d181c1 2166 ++sctx->stat.csum_discards;
a2de733c
AJ		 2167 list_del(&sum->list);
2168 kfree(sum);
2169 sum = NULL;
2170 }
2171 if (!sum)
2172 return 0;
2173
f51a4a18 2174 index = ((u32)(logical - sum->bytenr)) / sctx->sectorsize;
d9d181c1 2175 num_sectors = sum->len / sctx->sectorsize;
f51a4a18
MX		 2176 memcpy(csum, sum->sums + index, sctx->csum_size);
2177 if (index == num_sectors - 1) {
a2de733c
AJ		 2178 list_del(&sum->list);
2179 kfree(sum);
2180 }
f51a4a18 2181 return 1;
a2de733c
AJ		 2182}
2183
2184/* scrub extent tries to collect up to 64 kB for each bio */
d9d181c1 2185static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len,
a36cf8b8 2186 u64 physical, struct btrfs_device *dev, u64 flags,
ff023aac 2187 u64 gen, int mirror_num, u64 physical_for_dev_replace)
a2de733c
AJ		 2188{
2189 int ret;
2190 u8 csum[BTRFS_CSUM_SIZE];
b5d67f64
SB		 2191 u32 blocksize;
2192
2193 if (flags & BTRFS_EXTENT_FLAG_DATA) {
d9d181c1
SB		 2194 blocksize = sctx->sectorsize;
2195 spin_lock(&sctx->stat_lock);
2196 sctx->stat.data_extents_scrubbed++;
2197 sctx->stat.data_bytes_scrubbed += len;
2198 spin_unlock(&sctx->stat_lock);
b5d67f64 2199 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
ff023aac 2200 WARN_ON(sctx->nodesize != sctx->leafsize);
d9d181c1
SB		 2201 blocksize = sctx->nodesize;
2202 spin_lock(&sctx->stat_lock);
2203 sctx->stat.tree_extents_scrubbed++;
2204 sctx->stat.tree_bytes_scrubbed += len;
2205 spin_unlock(&sctx->stat_lock);
b5d67f64 2206 } else {
d9d181c1 2207 blocksize = sctx->sectorsize;
ff023aac 2208 WARN_ON(1);
b5d67f64 2209 }
a2de733c
AJ		 2210
2211 while (len) {
b5d67f64 2212 u64 l = min_t(u64, len, blocksize);
a2de733c
AJ		 2213 int have_csum = 0;
2214
2215 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2216 /* push csums to sbio */
d9d181c1 2217 have_csum = scrub_find_csum(sctx, logical, l, csum);
a2de733c 2218 if (have_csum == 0)
d9d181c1 2219 ++sctx->stat.no_csum;
ff023aac
SB		 2220 if (sctx->is_dev_replace && !have_csum) {
2221 ret = copy_nocow_pages(sctx, logical, l,
2222 mirror_num,
2223 physical_for_dev_replace);
2224 goto behind_scrub_pages;
2225 }
a2de733c 2226 }
a36cf8b8 2227 ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen,
ff023aac
SB		 2228 mirror_num, have_csum ? csum : NULL, 0,
2229 physical_for_dev_replace);
2230behind_scrub_pages:
a2de733c
AJ		 2231 if (ret)
2232 return ret;
2233 len -= l;
2234 logical += l;
2235 physical += l;
ff023aac 2236 physical_for_dev_replace += l;
a2de733c
AJ		 2237 }
2238 return 0;
2239}
2240
d9d181c1 2241static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
a36cf8b8
SB		 2242 struct map_lookup *map,
2243 struct btrfs_device *scrub_dev,
ff023aac
SB		 2244 int num, u64 base, u64 length,
2245 int is_dev_replace)
a2de733c
AJ		 2246{
2247 struct btrfs_path *path;
a36cf8b8 2248 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
a2de733c
AJ		 2249 struct btrfs_root *root = fs_info->extent_root;
2250 struct btrfs_root *csum_root = fs_info->csum_root;
2251 struct btrfs_extent_item *extent;
e7786c3a 2252 struct blk_plug plug;
a2de733c
AJ		 2253 u64 flags;
2254 int ret;
2255 int slot;
a2de733c 2256 u64 nstripes;
a2de733c
AJ		 2257 struct extent_buffer *l;
2258 struct btrfs_key key;
2259 u64 physical;
2260 u64 logical;
625f1c8d 2261 u64 logic_end;
a2de733c 2262 u64 generation;
e12fa9cd 2263 int mirror_num;
7a26285e
AJ		 2264 struct reada_control *reada1;
2265 struct reada_control *reada2;
2266 struct btrfs_key key_start;
2267 struct btrfs_key key_end;
a2de733c
AJ		 2268 u64 increment = map->stripe_len;
2269 u64 offset;
ff023aac
SB		 2270 u64 extent_logical;
2271 u64 extent_physical;
2272 u64 extent_len;
2273 struct btrfs_device *extent_dev;
2274 int extent_mirror_num;
625f1c8d 2275 int stop_loop;
a2de733c 2276
53b381b3
DW		 2277 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
2278 BTRFS_BLOCK_GROUP_RAID6)) {
2279 if (num >= nr_data_stripes(map)) {
2280 return 0;
2281 }
2282 }
2283
a2de733c
AJ		 2284 nstripes = length;
2285 offset = 0;
2286 do_div(nstripes, map->stripe_len);
2287 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2288 offset = map->stripe_len * num;
2289 increment = map->stripe_len * map->num_stripes;
193ea74b 2290 mirror_num = 1;
a2de733c
AJ		 2291 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2292 int factor = map->num_stripes / map->sub_stripes;
2293 offset = map->stripe_len * (num / map->sub_stripes);
2294 increment = map->stripe_len * factor;
193ea74b 2295 mirror_num = num % map->sub_stripes + 1;
a2de733c
AJ		 2296 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2297 increment = map->stripe_len;
193ea74b 2298 mirror_num = num % map->num_stripes + 1;
a2de733c
AJ		 2299 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2300 increment = map->stripe_len;
193ea74b 2301 mirror_num = num % map->num_stripes + 1;
a2de733c
AJ		 2302 } else {
2303 increment = map->stripe_len;
193ea74b 2304 mirror_num = 1;
a2de733c
AJ		 2305 }
2306
2307 path = btrfs_alloc_path();
2308 if (!path)
2309 return -ENOMEM;
2310
b5d67f64
SB		 2311 /*
2312 * work on commit root. The related disk blocks are static as
2313 * long as COW is applied. This means, it is save to rewrite
2314 * them to repair disk errors without any race conditions
2315 */
a2de733c
AJ		 2316 path->search_commit_root = 1;
2317 path->skip_locking = 1;
2318
2319 /*
7a26285e
AJ		 2320 * trigger the readahead for extent tree csum tree and wait for
2321 * completion. During readahead, the scrub is officially paused
2322 * to not hold off transaction commits
a2de733c
AJ		 2323 */
2324 logical = base + offset;
a2de733c 2325
d9d181c1 2326 wait_event(sctx->list_wait,
b6bfebc1 2327 atomic_read(&sctx->bios_in_flight) == 0);
cb7ab021 2328 scrub_blocked_if_needed(fs_info);
7a26285e
AJ		 2329
2330 /* FIXME it might be better to start readahead at commit root */
2331 key_start.objectid = logical;
2332 key_start.type = BTRFS_EXTENT_ITEM_KEY;
2333 key_start.offset = (u64)0;
2334 key_end.objectid = base + offset + nstripes * increment;
3173a18f
JB		 2335 key_end.type = BTRFS_METADATA_ITEM_KEY;
2336 key_end.offset = (u64)-1;
7a26285e
AJ		 2337 reada1 = btrfs_reada_add(root, &key_start, &key_end);
2338
2339 key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
2340 key_start.type = BTRFS_EXTENT_CSUM_KEY;
2341 key_start.offset = logical;
2342 key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
2343 key_end.type = BTRFS_EXTENT_CSUM_KEY;
2344 key_end.offset = base + offset + nstripes * increment;
2345 reada2 = btrfs_reada_add(csum_root, &key_start, &key_end);
2346
2347 if (!IS_ERR(reada1))
2348 btrfs_reada_wait(reada1);
2349 if (!IS_ERR(reada2))
2350 btrfs_reada_wait(reada2);
2351
a2de733c
AJ		 2352
2353 /*
2354 * collect all data csums for the stripe to avoid seeking during
2355 * the scrub. This might currently (crc32) end up to be about 1MB
2356 */
e7786c3a 2357 blk_start_plug(&plug);
a2de733c 2358
a2de733c
AJ		 2359 /*
2360 * now find all extents for each stripe and scrub them
2361 */
7a26285e
AJ		 2362 logical = base + offset;
2363 physical = map->stripes[num].physical;
625f1c8d 2364 logic_end = logical + increment * nstripes;
a2de733c 2365 ret = 0;
625f1c8d 2366 while (logical < logic_end) {
a2de733c
AJ		 2367 /*
2368 * canceled?
2369 */
2370 if (atomic_read(&fs_info->scrub_cancel_req) ||
d9d181c1 2371 atomic_read(&sctx->cancel_req)) {
a2de733c
AJ		 2372 ret = -ECANCELED;
2373 goto out;
2374 }
2375 /*
2376 * check to see if we have to pause
2377 */
2378 if (atomic_read(&fs_info->scrub_pause_req)) {
2379 /* push queued extents */
ff023aac 2380 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
d9d181c1 2381 scrub_submit(sctx);
ff023aac
SB		 2382 mutex_lock(&sctx->wr_ctx.wr_lock);
2383 scrub_wr_submit(sctx);
2384 mutex_unlock(&sctx->wr_ctx.wr_lock);
d9d181c1 2385 wait_event(sctx->list_wait,
b6bfebc1 2386 atomic_read(&sctx->bios_in_flight) == 0);
ff023aac 2387 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
3cb0929a 2388 scrub_blocked_if_needed(fs_info);
a2de733c
AJ		 2389 }
2390
7c76edb7
WS		 2391 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2392 key.type = BTRFS_METADATA_ITEM_KEY;
2393 else
2394 key.type = BTRFS_EXTENT_ITEM_KEY;
a2de733c 2395 key.objectid = logical;
625f1c8d 2396 key.offset = (u64)-1;
a2de733c
AJ		 2397
2398 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2399 if (ret < 0)
2400 goto out;
3173a18f 2401
8c51032f 2402 if (ret > 0) {
ade2e0b3 2403 ret = btrfs_previous_extent_item(root, path, 0);
a2de733c
AJ		 2404 if (ret < 0)
2405 goto out;
8c51032f
AJ		 2406 if (ret > 0) {
2407 /* there's no smaller item, so stick with the
2408 * larger one */
2409 btrfs_release_path(path);
2410 ret = btrfs_search_slot(NULL, root, &key,
2411 path, 0, 0);
2412 if (ret < 0)
2413 goto out;
2414 }
a2de733c
AJ		 2415 }
2416
625f1c8d 2417 stop_loop = 0;
a2de733c 2418 while (1) {
3173a18f
JB		 2419 u64 bytes;
2420
a2de733c
AJ		 2421 l = path->nodes[0];
2422 slot = path->slots[0];
2423 if (slot >= btrfs_header_nritems(l)) {
2424 ret = btrfs_next_leaf(root, path);
2425 if (ret == 0)
2426 continue;
2427 if (ret < 0)
2428 goto out;
2429
625f1c8d 2430 stop_loop = 1;
a2de733c
AJ		 2431 break;
2432 }
2433 btrfs_item_key_to_cpu(l, &key, slot);
2434
3173a18f
JB		 2435 if (key.type == BTRFS_METADATA_ITEM_KEY)
2436 bytes = root->leafsize;
2437 else
2438 bytes = key.offset;
2439
2440 if (key.objectid + bytes <= logical)
a2de733c
AJ		 2441 goto next;
2442
625f1c8d
LB		 2443 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
2444 key.type != BTRFS_METADATA_ITEM_KEY)
2445 goto next;
a2de733c 2446
625f1c8d
LB		 2447 if (key.objectid >= logical + map->stripe_len) {
2448 /* out of this device extent */
2449 if (key.objectid >= logic_end)
2450 stop_loop = 1;
2451 break;
2452 }
a2de733c
AJ		 2453
2454 extent = btrfs_item_ptr(l, slot,
2455 struct btrfs_extent_item);
2456 flags = btrfs_extent_flags(l, extent);
2457 generation = btrfs_extent_generation(l, extent);
2458
2459 if (key.objectid < logical &&
2460 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
efe120a0
FH		 2461 btrfs_err(fs_info,
2462 "scrub: tree block %llu spanning "
2463 "stripes, ignored. logical=%llu",
c1c9ff7c 2464 key.objectid, logical);
a2de733c
AJ		 2465 goto next;
2466 }
2467
625f1c8d
LB		 2468again:
2469 extent_logical = key.objectid;
2470 extent_len = bytes;
2471
a2de733c
AJ		 2472 /*
2473 * trim extent to this stripe
2474 */
625f1c8d
LB		 2475 if (extent_logical < logical) {
2476 extent_len -= logical - extent_logical;
2477 extent_logical = logical;
a2de733c 2478 }
625f1c8d 2479 if (extent_logical + extent_len >
a2de733c 2480 logical + map->stripe_len) {
625f1c8d
LB		 2481 extent_len = logical + map->stripe_len -
2482 extent_logical;
a2de733c
AJ		 2483 }
2484
625f1c8d 2485 extent_physical = extent_logical - logical + physical;
ff023aac
SB		 2486 extent_dev = scrub_dev;
2487 extent_mirror_num = mirror_num;
2488 if (is_dev_replace)
2489 scrub_remap_extent(fs_info, extent_logical,
2490 extent_len, &extent_physical,
2491 &extent_dev,
2492 &extent_mirror_num);
625f1c8d
LB		 2493
2494 ret = btrfs_lookup_csums_range(csum_root, logical,
2495 logical + map->stripe_len - 1,
2496 &sctx->csum_list, 1);
2497 if (ret)
2498 goto out;
2499
ff023aac
SB		 2500 ret = scrub_extent(sctx, extent_logical, extent_len,
2501 extent_physical, extent_dev, flags,
2502 generation, extent_mirror_num,
115930cb 2503 extent_logical - logical + physical);
a2de733c
AJ		 2504 if (ret)
2505 goto out;
2506
d88d46c6 2507 scrub_free_csums(sctx);
625f1c8d
LB		 2508 if (extent_logical + extent_len <
2509 key.objectid + bytes) {
2510 logical += increment;
2511 physical += map->stripe_len;
2512
2513 if (logical < key.objectid + bytes) {
2514 cond_resched();
2515 goto again;
2516 }
2517
2518 if (logical >= logic_end) {
2519 stop_loop = 1;
2520 break;
2521 }
2522 }
a2de733c
AJ		 2523next:
2524 path->slots[0]++;
2525 }
71267333 2526 btrfs_release_path(path);
a2de733c
AJ		 2527 logical += increment;
2528 physical += map->stripe_len;
d9d181c1 2529 spin_lock(&sctx->stat_lock);
625f1c8d
LB		 2530 if (stop_loop)
2531 sctx->stat.last_physical = map->stripes[num].physical +
2532 length;
2533 else
2534 sctx->stat.last_physical = physical;
d9d181c1 2535 spin_unlock(&sctx->stat_lock);
625f1c8d
LB		 2536 if (stop_loop)
2537 break;
a2de733c 2538 }
ff023aac 2539out:
a2de733c 2540 /* push queued extents */
d9d181c1 2541 scrub_submit(sctx);
ff023aac
SB		 2542 mutex_lock(&sctx->wr_ctx.wr_lock);
2543 scrub_wr_submit(sctx);
2544 mutex_unlock(&sctx->wr_ctx.wr_lock);
a2de733c 2545
e7786c3a 2546 blk_finish_plug(&plug);
a2de733c
AJ		 2547 btrfs_free_path(path);
2548 return ret < 0 ? ret : 0;
2549}
2550
d9d181c1 2551static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx,
a36cf8b8
SB		 2552 struct btrfs_device *scrub_dev,
2553 u64 chunk_tree, u64 chunk_objectid,
2554 u64 chunk_offset, u64 length,
ff023aac 2555 u64 dev_offset, int is_dev_replace)
a2de733c
AJ		 2556{
2557 struct btrfs_mapping_tree *map_tree =
a36cf8b8 2558 &sctx->dev_root->fs_info->mapping_tree;
a2de733c
AJ		 2559 struct map_lookup *map;
2560 struct extent_map *em;
2561 int i;
ff023aac 2562 int ret = 0;
a2de733c
AJ		 2563
2564 read_lock(&map_tree->map_tree.lock);
2565 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2566 read_unlock(&map_tree->map_tree.lock);
2567
2568 if (!em)
2569 return -EINVAL;
2570
2571 map = (struct map_lookup *)em->bdev;
2572 if (em->start != chunk_offset)
2573 goto out;
2574
2575 if (em->len < length)
2576 goto out;
2577
2578 for (i = 0; i < map->num_stripes; ++i) {
a36cf8b8 2579 if (map->stripes[i].dev->bdev == scrub_dev->bdev &&
859acaf1 2580 map->stripes[i].physical == dev_offset) {
a36cf8b8 2581 ret = scrub_stripe(sctx, map, scrub_dev, i,
ff023aac
SB		 2582 chunk_offset, length,
2583 is_dev_replace);
a2de733c
AJ		 2584 if (ret)
2585 goto out;
2586 }
2587 }
2588out:
2589 free_extent_map(em);
2590
2591 return ret;
2592}
2593
2594static noinline_for_stack
a36cf8b8 2595int scrub_enumerate_chunks(struct scrub_ctx *sctx,
ff023aac
SB		 2596 struct btrfs_device *scrub_dev, u64 start, u64 end,
2597 int is_dev_replace)
a2de733c
AJ		 2598{
2599 struct btrfs_dev_extent *dev_extent = NULL;
2600 struct btrfs_path *path;
a36cf8b8 2601 struct btrfs_root *root = sctx->dev_root;
a2de733c
AJ		 2602 struct btrfs_fs_info *fs_info = root->fs_info;
2603 u64 length;
2604 u64 chunk_tree;
2605 u64 chunk_objectid;
2606 u64 chunk_offset;
2607 int ret;
2608 int slot;
2609 struct extent_buffer *l;
2610 struct btrfs_key key;
2611 struct btrfs_key found_key;
2612 struct btrfs_block_group_cache *cache;
ff023aac 2613 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
a2de733c
AJ		 2614
2615 path = btrfs_alloc_path();
2616 if (!path)
2617 return -ENOMEM;
2618
2619 path->reada = 2;
2620 path->search_commit_root = 1;
2621 path->skip_locking = 1;
2622
a36cf8b8 2623 key.objectid = scrub_dev->devid;
a2de733c
AJ		 2624 key.offset = 0ull;
2625 key.type = BTRFS_DEV_EXTENT_KEY;
2626
a2de733c
AJ		 2627 while (1) {
2628 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2629 if (ret < 0)
8c51032f
AJ		 2630 break;
2631 if (ret > 0) {
2632 if (path->slots[0] >=
2633 btrfs_header_nritems(path->nodes[0])) {
2634 ret = btrfs_next_leaf(root, path);
2635 if (ret)
2636 break;
2637 }
2638 }
a2de733c
AJ		 2639
2640 l = path->nodes[0];
2641 slot = path->slots[0];
2642
2643 btrfs_item_key_to_cpu(l, &found_key, slot);
2644
a36cf8b8 2645 if (found_key.objectid != scrub_dev->devid)
a2de733c
AJ		 2646 break;
2647
8c51032f 2648 if (btrfs_key_type(&found_key) != BTRFS_DEV_EXTENT_KEY)
a2de733c
AJ		 2649 break;
2650
2651 if (found_key.offset >= end)
2652 break;
2653
2654 if (found_key.offset < key.offset)
2655 break;
2656
2657 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2658 length = btrfs_dev_extent_length(l, dev_extent);
2659
2660 if (found_key.offset + length <= start) {
2661 key.offset = found_key.offset + length;
71267333 2662 btrfs_release_path(path);
a2de733c
AJ		 2663 continue;
2664 }
2665
2666 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2667 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2668 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2669
2670 /*
2671 * get a reference on the corresponding block group to prevent
2672 * the chunk from going away while we scrub it
2673 */
2674 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2675 if (!cache) {
2676 ret = -ENOENT;
8c51032f 2677 break;
a2de733c 2678 }
ff023aac
SB		 2679 dev_replace->cursor_right = found_key.offset + length;
2680 dev_replace->cursor_left = found_key.offset;
2681 dev_replace->item_needs_writeback = 1;
a36cf8b8 2682 ret = scrub_chunk(sctx, scrub_dev, chunk_tree, chunk_objectid,
ff023aac
SB		 2683 chunk_offset, length, found_key.offset,
2684 is_dev_replace);
2685
2686 /*
2687 * flush, submit all pending read and write bios, afterwards
2688 * wait for them.
2689 * Note that in the dev replace case, a read request causes
2690 * write requests that are submitted in the read completion
2691 * worker. Therefore in the current situation, it is required
2692 * that all write requests are flushed, so that all read and
2693 * write requests are really completed when bios_in_flight
2694 * changes to 0.
2695 */
2696 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
2697 scrub_submit(sctx);
2698 mutex_lock(&sctx->wr_ctx.wr_lock);
2699 scrub_wr_submit(sctx);
2700 mutex_unlock(&sctx->wr_ctx.wr_lock);
2701
2702 wait_event(sctx->list_wait,
2703 atomic_read(&sctx->bios_in_flight) == 0);
12cf9372
WS		 2704 atomic_inc(&fs_info->scrubs_paused);
2705 wake_up(&fs_info->scrub_pause_wait);
2706
2707 /*
2708 * must be called before we decrease @scrub_paused.
2709 * make sure we don't block transaction commit while
2710 * we are waiting pending workers finished.
2711 */
ff023aac
SB		 2712 wait_event(sctx->list_wait,
2713 atomic_read(&sctx->workers_pending) == 0);
12cf9372
WS		 2714 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
2715
2716 mutex_lock(&fs_info->scrub_lock);
2717 __scrub_blocked_if_needed(fs_info);
2718 atomic_dec(&fs_info->scrubs_paused);
2719 mutex_unlock(&fs_info->scrub_lock);
2720 wake_up(&fs_info->scrub_pause_wait);
ff023aac 2721
a2de733c
AJ		 2722 btrfs_put_block_group(cache);
2723 if (ret)
2724 break;
af1be4f8
SB		 2725 if (is_dev_replace &&
2726 atomic64_read(&dev_replace->num_write_errors) > 0) {
ff023aac
SB		 2727 ret = -EIO;
2728 break;
2729 }
2730 if (sctx->stat.malloc_errors > 0) {
2731 ret = -ENOMEM;
2732 break;
2733 }
a2de733c 2734
539f358a
ID		 2735 dev_replace->cursor_left = dev_replace->cursor_right;
2736 dev_replace->item_needs_writeback = 1;
2737
a2de733c 2738 key.offset = found_key.offset + length;
71267333 2739 btrfs_release_path(path);
a2de733c
AJ		 2740 }
2741
a2de733c 2742 btrfs_free_path(path);
8c51032f
AJ		 2743
2744 /*
2745 * ret can still be 1 from search_slot or next_leaf,
2746 * that's not an error
2747 */
2748 return ret < 0 ? ret : 0;
a2de733c
AJ		 2749}
2750
a36cf8b8
SB		 2751static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx,
2752 struct btrfs_device *scrub_dev)
a2de733c
AJ		 2753{
2754 int i;
2755 u64 bytenr;
2756 u64 gen;
2757 int ret;
a36cf8b8 2758 struct btrfs_root *root = sctx->dev_root;
a2de733c 2759
87533c47 2760 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
79787eaa
JM		 2761 return -EIO;
2762
a2de733c
AJ		 2763 gen = root->fs_info->last_trans_committed;
2764
2765 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2766 bytenr = btrfs_sb_offset(i);
a36cf8b8 2767 if (bytenr + BTRFS_SUPER_INFO_SIZE > scrub_dev->total_bytes)
a2de733c
AJ		 2768 break;
2769
d9d181c1 2770 ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr,
a36cf8b8 2771 scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i,
ff023aac 2772 NULL, 1, bytenr);
a2de733c
AJ		 2773 if (ret)
2774 return ret;
2775 }
b6bfebc1 2776 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
a2de733c
AJ		 2777
2778 return 0;
2779}
2780
2781/*
2782 * get a reference count on fs_info->scrub_workers. start worker if necessary
2783 */
ff023aac
SB		 2784static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info,
2785 int is_dev_replace)
a2de733c 2786{
0dc3b84a 2787 int ret = 0;
0339ef2f
QW		 2788 int flags = WQ_FREEZABLE | WQ_UNBOUND;
2789 int max_active = fs_info->thread_pool_size;
a2de733c 2790
632dd772 2791 if (fs_info->scrub_workers_refcnt == 0) {
ff023aac 2792 if (is_dev_replace)
0339ef2f
QW		 2793 fs_info->scrub_workers =
2794 btrfs_alloc_workqueue("btrfs-scrub", flags,
2795 1, 4);
ff023aac 2796 else
0339ef2f
QW		 2797 fs_info->scrub_workers =
2798 btrfs_alloc_workqueue("btrfs-scrub", flags,
2799 max_active, 4);
2800 if (!fs_info->scrub_workers) {
2801 ret = -ENOMEM;
0dc3b84a 2802 goto out;
0339ef2f
QW		 2803 }
2804 fs_info->scrub_wr_completion_workers =
2805 btrfs_alloc_workqueue("btrfs-scrubwrc", flags,
2806 max_active, 2);
2807 if (!fs_info->scrub_wr_completion_workers) {
2808 ret = -ENOMEM;
ff023aac 2809 goto out;
0339ef2f
QW		 2810 }
2811 fs_info->scrub_nocow_workers =
2812 btrfs_alloc_workqueue("btrfs-scrubnc", flags, 1, 0);
2813 if (!fs_info->scrub_nocow_workers) {
2814 ret = -ENOMEM;
ff023aac 2815 goto out;
0339ef2f 2816 }
632dd772 2817 }
a2de733c 2818 ++fs_info->scrub_workers_refcnt;
0dc3b84a 2819out:
0dc3b84a 2820 return ret;
a2de733c
AJ		 2821}
2822
aa1b8cd4 2823static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info)
a2de733c 2824{
ff023aac 2825 if (--fs_info->scrub_workers_refcnt == 0) {
0339ef2f
QW		 2826 btrfs_destroy_workqueue(fs_info->scrub_workers);
2827 btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
2828 btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
ff023aac 2829 }
a2de733c 2830 WARN_ON(fs_info->scrub_workers_refcnt < 0);
a2de733c
AJ		 2831}
2832
aa1b8cd4
SB		 2833int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start,
2834 u64 end, struct btrfs_scrub_progress *progress,
63a212ab 2835 int readonly, int is_dev_replace)
a2de733c 2836{
d9d181c1 2837 struct scrub_ctx *sctx;
a2de733c
AJ		 2838 int ret;
2839 struct btrfs_device *dev;
2840
aa1b8cd4 2841 if (btrfs_fs_closing(fs_info))
a2de733c
AJ		 2842 return -EINVAL;
2843
2844 /*
2845 * check some assumptions
2846 */
aa1b8cd4 2847 if (fs_info->chunk_root->nodesize != fs_info->chunk_root->leafsize) {
efe120a0
FH		 2848 btrfs_err(fs_info,
2849 "scrub: size assumption nodesize == leafsize (%d == %d) fails",
aa1b8cd4
SB		 2850 fs_info->chunk_root->nodesize,
2851 fs_info->chunk_root->leafsize);
b5d67f64
SB		 2852 return -EINVAL;
2853 }
2854
aa1b8cd4 2855 if (fs_info->chunk_root->nodesize > BTRFS_STRIPE_LEN) {
b5d67f64
SB		 2856 /*
2857 * in this case scrub is unable to calculate the checksum
2858 * the way scrub is implemented. Do not handle this
2859 * situation at all because it won't ever happen.
2860 */
efe120a0
FH		 2861 btrfs_err(fs_info,
2862 "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails",
aa1b8cd4 2863 fs_info->chunk_root->nodesize, BTRFS_STRIPE_LEN);
b5d67f64
SB		 2864 return -EINVAL;
2865 }
2866
aa1b8cd4 2867 if (fs_info->chunk_root->sectorsize != PAGE_SIZE) {
b5d67f64 2868 /* not supported for data w/o checksums */
efe120a0
FH		 2869 btrfs_err(fs_info,
2870 "scrub: size assumption sectorsize != PAGE_SIZE "
2871 "(%d != %lu) fails",
27f9f023 2872 fs_info->chunk_root->sectorsize, PAGE_SIZE);
a2de733c
AJ		 2873 return -EINVAL;
2874 }
2875
7a9e9987
SB		 2876 if (fs_info->chunk_root->nodesize >
2877 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK ||
2878 fs_info->chunk_root->sectorsize >
2879 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) {
2880 /*
2881 * would exhaust the array bounds of pagev member in
2882 * struct scrub_block
2883 */
efe120a0
FH		 2884 btrfs_err(fs_info, "scrub: size assumption nodesize and sectorsize "
2885 "<= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails",
7a9e9987
SB		 2886 fs_info->chunk_root->nodesize,
2887 SCRUB_MAX_PAGES_PER_BLOCK,
2888 fs_info->chunk_root->sectorsize,
2889 SCRUB_MAX_PAGES_PER_BLOCK);
2890 return -EINVAL;
2891 }
2892
a2de733c 2893
aa1b8cd4
SB		 2894 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2895 dev = btrfs_find_device(fs_info, devid, NULL, NULL);
63a212ab 2896 if (!dev || (dev->missing && !is_dev_replace)) {
aa1b8cd4 2897 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
a2de733c
AJ		 2898 return -ENODEV;
2899 }
a2de733c 2900
3b7a016f 2901 mutex_lock(&fs_info->scrub_lock);
63a212ab 2902 if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) {
a2de733c 2903 mutex_unlock(&fs_info->scrub_lock);
aa1b8cd4 2904 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
aa1b8cd4 2905 return -EIO;
a2de733c
AJ		 2906 }
2907
8dabb742
SB		 2908 btrfs_dev_replace_lock(&fs_info->dev_replace);
2909 if (dev->scrub_device ||
2910 (!is_dev_replace &&
2911 btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) {
2912 btrfs_dev_replace_unlock(&fs_info->dev_replace);
a2de733c 2913 mutex_unlock(&fs_info->scrub_lock);
aa1b8cd4 2914 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
a2de733c
AJ		 2915 return -EINPROGRESS;
2916 }
8dabb742 2917 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3b7a016f
WS		 2918
2919 ret = scrub_workers_get(fs_info, is_dev_replace);
2920 if (ret) {
2921 mutex_unlock(&fs_info->scrub_lock);
2922 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2923 return ret;
2924 }
2925
63a212ab 2926 sctx = scrub_setup_ctx(dev, is_dev_replace);
d9d181c1 2927 if (IS_ERR(sctx)) {
a2de733c 2928 mutex_unlock(&fs_info->scrub_lock);
aa1b8cd4
SB		 2929 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2930 scrub_workers_put(fs_info);
d9d181c1 2931 return PTR_ERR(sctx);
a2de733c 2932 }
d9d181c1
SB		 2933 sctx->readonly = readonly;
2934 dev->scrub_device = sctx;
3cb0929a 2935 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
a2de733c 2936
3cb0929a
WS		 2937 /*
2938 * checking @scrub_pause_req here, we can avoid
2939 * race between committing transaction and scrubbing.
2940 */
cb7ab021 2941 __scrub_blocked_if_needed(fs_info);
a2de733c
AJ		 2942 atomic_inc(&fs_info->scrubs_running);
2943 mutex_unlock(&fs_info->scrub_lock);
a2de733c 2944
ff023aac 2945 if (!is_dev_replace) {
9b011adf
WS		 2946 /*
2947 * by holding device list mutex, we can
2948 * kick off writing super in log tree sync.
2949 */
3cb0929a 2950 mutex_lock(&fs_info->fs_devices->device_list_mutex);
ff023aac 2951 ret = scrub_supers(sctx, dev);
3cb0929a 2952 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
ff023aac 2953 }
a2de733c
AJ		 2954
2955 if (!ret)
ff023aac
SB		 2956 ret = scrub_enumerate_chunks(sctx, dev, start, end,
2957 is_dev_replace);
a2de733c 2958
b6bfebc1 2959 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
a2de733c
AJ		 2960 atomic_dec(&fs_info->scrubs_running);
2961 wake_up(&fs_info->scrub_pause_wait);
2962
b6bfebc1 2963 wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0);
0ef8e451 2964
a2de733c 2965 if (progress)
d9d181c1 2966 memcpy(progress, &sctx->stat, sizeof(*progress));
a2de733c
AJ		 2967
2968 mutex_lock(&fs_info->scrub_lock);
2969 dev->scrub_device = NULL;
3b7a016f 2970 scrub_workers_put(fs_info);
a2de733c
AJ		 2971 mutex_unlock(&fs_info->scrub_lock);
2972
d9d181c1 2973 scrub_free_ctx(sctx);
a2de733c
AJ		 2974
2975 return ret;
2976}
2977
143bede5 2978void btrfs_scrub_pause(struct btrfs_root *root)
a2de733c
AJ		 2979{
2980 struct btrfs_fs_info *fs_info = root->fs_info;
2981
2982 mutex_lock(&fs_info->scrub_lock);
2983 atomic_inc(&fs_info->scrub_pause_req);
2984 while (atomic_read(&fs_info->scrubs_paused) !=
2985 atomic_read(&fs_info->scrubs_running)) {
2986 mutex_unlock(&fs_info->scrub_lock);
2987 wait_event(fs_info->scrub_pause_wait,
2988 atomic_read(&fs_info->scrubs_paused) ==
2989 atomic_read(&fs_info->scrubs_running));
2990 mutex_lock(&fs_info->scrub_lock);
2991 }
2992 mutex_unlock(&fs_info->scrub_lock);
a2de733c
AJ		 2993}
2994
143bede5 2995void btrfs_scrub_continue(struct btrfs_root *root)
a2de733c
AJ		 2996{
2997 struct btrfs_fs_info *fs_info = root->fs_info;
2998
2999 atomic_dec(&fs_info->scrub_pause_req);
3000 wake_up(&fs_info->scrub_pause_wait);
a2de733c
AJ		 3001}
3002
aa1b8cd4 3003int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info)
a2de733c 3004{
a2de733c
AJ		 3005 mutex_lock(&fs_info->scrub_lock);
3006 if (!atomic_read(&fs_info->scrubs_running)) {
3007 mutex_unlock(&fs_info->scrub_lock);
3008 return -ENOTCONN;
3009 }
3010
3011 atomic_inc(&fs_info->scrub_cancel_req);
3012 while (atomic_read(&fs_info->scrubs_running)) {
3013 mutex_unlock(&fs_info->scrub_lock);
3014 wait_event(fs_info->scrub_pause_wait,
3015 atomic_read(&fs_info->scrubs_running) == 0);
3016 mutex_lock(&fs_info->scrub_lock);
3017 }
3018 atomic_dec(&fs_info->scrub_cancel_req);
3019 mutex_unlock(&fs_info->scrub_lock);
3020
3021 return 0;
3022}
3023
aa1b8cd4
SB		 3024int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info,
3025 struct btrfs_device *dev)
49b25e05 3026{
d9d181c1 3027 struct scrub_ctx *sctx;
a2de733c
AJ		 3028
3029 mutex_lock(&fs_info->scrub_lock);
d9d181c1
SB		 3030 sctx = dev->scrub_device;
3031 if (!sctx) {
a2de733c
AJ		 3032 mutex_unlock(&fs_info->scrub_lock);
3033 return -ENOTCONN;
3034 }
d9d181c1 3035 atomic_inc(&sctx->cancel_req);
a2de733c
AJ		 3036 while (dev->scrub_device) {
3037 mutex_unlock(&fs_info->scrub_lock);
3038 wait_event(fs_info->scrub_pause_wait,
3039 dev->scrub_device == NULL);
3040 mutex_lock(&fs_info->scrub_lock);
3041 }
3042 mutex_unlock(&fs_info->scrub_lock);
3043
3044 return 0;
3045}
1623edeb 3046
a2de733c
AJ		 3047int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
3048 struct btrfs_scrub_progress *progress)
3049{
3050 struct btrfs_device *dev;
d9d181c1 3051 struct scrub_ctx *sctx = NULL;
a2de733c
AJ		 3052
3053 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
aa1b8cd4 3054 dev = btrfs_find_device(root->fs_info, devid, NULL, NULL);
a2de733c 3055 if (dev)
d9d181c1
SB		 3056 sctx = dev->scrub_device;
3057 if (sctx)
3058 memcpy(progress, &sctx->stat, sizeof(*progress));
a2de733c
AJ		 3059 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3060
d9d181c1 3061 return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV;
a2de733c 3062}
ff023aac
SB		 3063
3064static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
3065 u64 extent_logical, u64 extent_len,
3066 u64 *extent_physical,
3067 struct btrfs_device **extent_dev,
3068 int *extent_mirror_num)
3069{
3070 u64 mapped_length;
3071 struct btrfs_bio *bbio = NULL;
3072 int ret;
3073
3074 mapped_length = extent_len;
3075 ret = btrfs_map_block(fs_info, READ, extent_logical,
3076 &mapped_length, &bbio, 0);
3077 if (ret || !bbio || mapped_length < extent_len ||
3078 !bbio->stripes[0].dev->bdev) {
3079 kfree(bbio);
3080 return;
3081 }
3082
3083 *extent_physical = bbio->stripes[0].physical;
3084 *extent_mirror_num = bbio->mirror_num;
3085 *extent_dev = bbio->stripes[0].dev;
3086 kfree(bbio);
3087}
3088
3089static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
3090 struct scrub_wr_ctx *wr_ctx,
3091 struct btrfs_fs_info *fs_info,
3092 struct btrfs_device *dev,
3093 int is_dev_replace)
3094{
3095 WARN_ON(wr_ctx->wr_curr_bio != NULL);
3096
3097 mutex_init(&wr_ctx->wr_lock);
3098 wr_ctx->wr_curr_bio = NULL;
3099 if (!is_dev_replace)
3100 return 0;
3101
3102 WARN_ON(!dev->bdev);
3103 wr_ctx->pages_per_wr_bio = min_t(int, SCRUB_PAGES_PER_WR_BIO,
3104 bio_get_nr_vecs(dev->bdev));
3105 wr_ctx->tgtdev = dev;
3106 atomic_set(&wr_ctx->flush_all_writes, 0);
3107 return 0;
3108}
3109
3110static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx)
3111{
3112 mutex_lock(&wr_ctx->wr_lock);
3113 kfree(wr_ctx->wr_curr_bio);
3114 wr_ctx->wr_curr_bio = NULL;
3115 mutex_unlock(&wr_ctx->wr_lock);
3116}
3117
3118static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
3119 int mirror_num, u64 physical_for_dev_replace)
3120{
3121 struct scrub_copy_nocow_ctx *nocow_ctx;
3122 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
3123
3124 nocow_ctx = kzalloc(sizeof(*nocow_ctx), GFP_NOFS);
3125 if (!nocow_ctx) {
3126 spin_lock(&sctx->stat_lock);
3127 sctx->stat.malloc_errors++;
3128 spin_unlock(&sctx->stat_lock);
3129 return -ENOMEM;
3130 }
3131
3132 scrub_pending_trans_workers_inc(sctx);
3133
3134 nocow_ctx->sctx = sctx;
3135 nocow_ctx->logical = logical;
3136 nocow_ctx->len = len;
3137 nocow_ctx->mirror_num = mirror_num;
3138 nocow_ctx->physical_for_dev_replace = physical_for_dev_replace;
0339ef2f 3139 btrfs_init_work(&nocow_ctx->work, copy_nocow_pages_worker, NULL, NULL);
652f25a2 3140 INIT_LIST_HEAD(&nocow_ctx->inodes);
0339ef2f
QW		 3141 btrfs_queue_work(fs_info->scrub_nocow_workers,
3142 &nocow_ctx->work);
ff023aac
SB		 3143
3144 return 0;
3145}
3146
652f25a2
JB		 3147static int record_inode_for_nocow(u64 inum, u64 offset, u64 root, void *ctx)
3148{
3149 struct scrub_copy_nocow_ctx *nocow_ctx = ctx;
3150 struct scrub_nocow_inode *nocow_inode;
3151
3152 nocow_inode = kzalloc(sizeof(*nocow_inode), GFP_NOFS);
3153 if (!nocow_inode)
3154 return -ENOMEM;
3155 nocow_inode->inum = inum;
3156 nocow_inode->offset = offset;
3157 nocow_inode->root = root;
3158 list_add_tail(&nocow_inode->list, &nocow_ctx->inodes);
3159 return 0;
3160}
3161
3162#define COPY_COMPLETE 1
3163
0339ef2f 3164static void copy_nocow_pages_worker(struct btrfs_work_struct *work)
ff023aac
SB		 3165{
3166 struct scrub_copy_nocow_ctx *nocow_ctx =
3167 container_of(work, struct scrub_copy_nocow_ctx, work);
3168 struct scrub_ctx *sctx = nocow_ctx->sctx;
3169 u64 logical = nocow_ctx->logical;
3170 u64 len = nocow_ctx->len;
3171 int mirror_num = nocow_ctx->mirror_num;
3172 u64 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
3173 int ret;
3174 struct btrfs_trans_handle *trans = NULL;
3175 struct btrfs_fs_info *fs_info;
3176 struct btrfs_path *path;
3177 struct btrfs_root *root;
3178 int not_written = 0;
3179
3180 fs_info = sctx->dev_root->fs_info;
3181 root = fs_info->extent_root;
3182
3183 path = btrfs_alloc_path();
3184 if (!path) {
3185 spin_lock(&sctx->stat_lock);
3186 sctx->stat.malloc_errors++;
3187 spin_unlock(&sctx->stat_lock);
3188 not_written = 1;
3189 goto out;
3190 }
3191
3192 trans = btrfs_join_transaction(root);
3193 if (IS_ERR(trans)) {
3194 not_written = 1;
3195 goto out;
3196 }
3197
3198 ret = iterate_inodes_from_logical(logical, fs_info, path,
652f25a2 3199 record_inode_for_nocow, nocow_ctx);
ff023aac 3200 if (ret != 0 && ret != -ENOENT) {
efe120a0
FH		 3201 btrfs_warn(fs_info, "iterate_inodes_from_logical() failed: log %llu, "
3202 "phys %llu, len %llu, mir %u, ret %d",
118a0a25
GU		 3203 logical, physical_for_dev_replace, len, mirror_num,
3204 ret);
ff023aac
SB		 3205 not_written = 1;
3206 goto out;
3207 }
3208
652f25a2
JB		 3209 btrfs_end_transaction(trans, root);
3210 trans = NULL;
3211 while (!list_empty(&nocow_ctx->inodes)) {
3212 struct scrub_nocow_inode *entry;
3213 entry = list_first_entry(&nocow_ctx->inodes,
3214 struct scrub_nocow_inode,
3215 list);
3216 list_del_init(&entry->list);
3217 ret = copy_nocow_pages_for_inode(entry->inum, entry->offset,
3218 entry->root, nocow_ctx);
3219 kfree(entry);
3220 if (ret == COPY_COMPLETE) {
3221 ret = 0;
3222 break;
3223 } else if (ret) {
3224 break;
3225 }
3226 }
ff023aac 3227out:
652f25a2
JB		 3228 while (!list_empty(&nocow_ctx->inodes)) {
3229 struct scrub_nocow_inode *entry;
3230 entry = list_first_entry(&nocow_ctx->inodes,
3231 struct scrub_nocow_inode,
3232 list);
3233 list_del_init(&entry->list);
3234 kfree(entry);
3235 }
ff023aac
SB		 3236 if (trans && !IS_ERR(trans))
3237 btrfs_end_transaction(trans, root);
3238 if (not_written)
3239 btrfs_dev_replace_stats_inc(&fs_info->dev_replace.
3240 num_uncorrectable_read_errors);
3241
3242 btrfs_free_path(path);
3243 kfree(nocow_ctx);
3244
3245 scrub_pending_trans_workers_dec(sctx);
3246}
3247
652f25a2
JB		 3248static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
3249 struct scrub_copy_nocow_ctx *nocow_ctx)
ff023aac 3250{
826aa0a8 3251 struct btrfs_fs_info *fs_info = nocow_ctx->sctx->dev_root->fs_info;
ff023aac 3252 struct btrfs_key key;
826aa0a8
MX		 3253 struct inode *inode;
3254 struct page *page;
ff023aac 3255 struct btrfs_root *local_root;
652f25a2
JB		 3256 struct btrfs_ordered_extent *ordered;
3257 struct extent_map *em;
3258 struct extent_state *cached_state = NULL;
3259 struct extent_io_tree *io_tree;
ff023aac 3260 u64 physical_for_dev_replace;
652f25a2
JB		 3261 u64 len = nocow_ctx->len;
3262 u64 lockstart = offset, lockend = offset + len - 1;
826aa0a8 3263 unsigned long index;
6f1c3605 3264 int srcu_index;
652f25a2
JB		 3265 int ret = 0;
3266 int err = 0;
ff023aac
SB		 3267
3268 key.objectid = root;
3269 key.type = BTRFS_ROOT_ITEM_KEY;
3270 key.offset = (u64)-1;
6f1c3605
LB		 3271
3272 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
3273
ff023aac 3274 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
6f1c3605
LB		 3275 if (IS_ERR(local_root)) {
3276 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
ff023aac 3277 return PTR_ERR(local_root);
6f1c3605 3278 }
ff023aac
SB		 3279
3280 key.type = BTRFS_INODE_ITEM_KEY;
3281 key.objectid = inum;
3282 key.offset = 0;
3283 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
6f1c3605 3284 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
ff023aac
SB		 3285 if (IS_ERR(inode))
3286 return PTR_ERR(inode);
3287
edd1400b
MX		 3288 /* Avoid truncate/dio/punch hole.. */
3289 mutex_lock(&inode->i_mutex);
3290 inode_dio_wait(inode);
3291
ff023aac 3292 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
652f25a2
JB		 3293 io_tree = &BTRFS_I(inode)->io_tree;
3294
3295 lock_extent_bits(io_tree, lockstart, lockend, 0, &cached_state);
3296 ordered = btrfs_lookup_ordered_range(inode, lockstart, len);
3297 if (ordered) {
3298 btrfs_put_ordered_extent(ordered);
3299 goto out_unlock;
3300 }
3301
3302 em = btrfs_get_extent(inode, NULL, 0, lockstart, len, 0);
3303 if (IS_ERR(em)) {
3304 ret = PTR_ERR(em);
3305 goto out_unlock;
3306 }
3307
3308 /*
3309 * This extent does not actually cover the logical extent anymore,
3310 * move on to the next inode.
3311 */
3312 if (em->block_start > nocow_ctx->logical ||
3313 em->block_start + em->block_len < nocow_ctx->logical + len) {
3314 free_extent_map(em);
3315 goto out_unlock;
3316 }
3317 free_extent_map(em);
3318
ff023aac 3319 while (len >= PAGE_CACHE_SIZE) {
ff023aac 3320 index = offset >> PAGE_CACHE_SHIFT;
edd1400b 3321again:
ff023aac
SB		 3322 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
3323 if (!page) {
efe120a0 3324 btrfs_err(fs_info, "find_or_create_page() failed");
ff023aac 3325 ret = -ENOMEM;
826aa0a8 3326 goto out;
ff023aac
SB		 3327 }
3328
3329 if (PageUptodate(page)) {
3330 if (PageDirty(page))
3331 goto next_page;
3332 } else {
3333 ClearPageError(page);
652f25a2
JB		 3334 err = extent_read_full_page_nolock(io_tree, page,
3335 btrfs_get_extent,
3336 nocow_ctx->mirror_num);
826aa0a8
MX		 3337 if (err) {
3338 ret = err;
ff023aac
SB		 3339 goto next_page;
3340 }
edd1400b 3341
26b25891 3342 lock_page(page);
edd1400b
MX		 3343 /*
3344 * If the page has been remove from the page cache,
3345 * the data on it is meaningless, because it may be
3346 * old one, the new data may be written into the new
3347 * page in the page cache.
3348 */
3349 if (page->mapping != inode->i_mapping) {
652f25a2 3350 unlock_page(page);
edd1400b
MX		 3351 page_cache_release(page);
3352 goto again;
3353 }
ff023aac
SB		 3354 if (!PageUptodate(page)) {
3355 ret = -EIO;
3356 goto next_page;
3357 }
3358 }
826aa0a8
MX		 3359 err = write_page_nocow(nocow_ctx->sctx,
3360 physical_for_dev_replace, page);
3361 if (err)
3362 ret = err;
ff023aac 3363next_page:
826aa0a8
MX		 3364 unlock_page(page);
3365 page_cache_release(page);
3366
3367 if (ret)
3368 break;
3369
ff023aac
SB		 3370 offset += PAGE_CACHE_SIZE;
3371 physical_for_dev_replace += PAGE_CACHE_SIZE;
3372 len -= PAGE_CACHE_SIZE;
3373 }
652f25a2
JB		 3374 ret = COPY_COMPLETE;
3375out_unlock:
3376 unlock_extent_cached(io_tree, lockstart, lockend, &cached_state,
3377 GFP_NOFS);
826aa0a8 3378out:
edd1400b 3379 mutex_unlock(&inode->i_mutex);
826aa0a8 3380 iput(inode);
ff023aac
SB		 3381 return ret;
3382}
3383
3384static int write_page_nocow(struct scrub_ctx *sctx,
3385 u64 physical_for_dev_replace, struct page *page)
3386{
3387 struct bio *bio;
3388 struct btrfs_device *dev;
3389 int ret;
ff023aac
SB		 3390
3391 dev = sctx->wr_ctx.tgtdev;
3392 if (!dev)
3393 return -EIO;
3394 if (!dev->bdev) {
3395 printk_ratelimited(KERN_WARNING
efe120a0 3396 "BTRFS: scrub write_page_nocow(bdev == NULL) is unexpected!\n");
ff023aac
SB		 3397 return -EIO;
3398 }
9be3395b 3399 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
ff023aac
SB		 3400 if (!bio) {
3401 spin_lock(&sctx->stat_lock);
3402 sctx->stat.malloc_errors++;
3403 spin_unlock(&sctx->stat_lock);
3404 return -ENOMEM;
3405 }
ff023aac
SB		 3406 bio->bi_size = 0;
3407 bio->bi_sector = physical_for_dev_replace >> 9;
3408 bio->bi_bdev = dev->bdev;
3409 ret = bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
3410 if (ret != PAGE_CACHE_SIZE) {
3411leave_with_eio:
3412 bio_put(bio);
3413 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
3414 return -EIO;
3415 }
ff023aac 3416
c170bbb4 3417 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio))
ff023aac
SB		 3418 goto leave_with_eio;
3419
3420 bio_put(bio);
3421 return 0;
3422}
Linux kernel - Deliverables
This page took 0.324398 seconds and 5 git commands to generate.