projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: make static code static & remove dead code
[deliverable/linux.git] / fs / btrfs / scrub.c
CommitLineData
a2de733c 1/*
b6bfebc1 2 * Copyright (C) 2011, 2012 STRATO. All rights reserved.
a2de733c
AJ		 3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
a2de733c 19#include <linux/blkdev.h>
558540c1 20#include <linux/ratelimit.h>
a2de733c
AJ		 21#include "ctree.h"
22#include "volumes.h"
23#include "disk-io.h"
24#include "ordered-data.h"
0ef8e451 25#include "transaction.h"
558540c1 26#include "backref.h"
5da6fcbc 27#include "extent_io.h"
ff023aac 28#include "dev-replace.h"
21adbd5c 29#include "check-integrity.h"
606686ee 30#include "rcu-string.h"
53b381b3 31#include "raid56.h"
a2de733c
AJ		 32
33/*
34 * This is only the first step towards a full-features scrub. It reads all
35 * extent and super block and verifies the checksums. In case a bad checksum
36 * is found or the extent cannot be read, good data will be written back if
37 * any can be found.
38 *
39 * Future enhancements:
a2de733c
AJ		 40 * - In case an unrepairable extent is encountered, track which files are
41 * affected and report them
a2de733c 42 * - track and record media errors, throw out bad devices
a2de733c 43 * - add a mode to also read unallocated space
a2de733c
AJ		 44 */
45
b5d67f64 46struct scrub_block;
d9d181c1 47struct scrub_ctx;
a2de733c 48
ff023aac
SB		 49/*
50 * the following three values only influence the performance.
51 * The last one configures the number of parallel and outstanding I/O
52 * operations. The first two values configure an upper limit for the number
53 * of (dynamically allocated) pages that are added to a bio.
54 */
55#define SCRUB_PAGES_PER_RD_BIO 32 /* 128k per bio */
56#define SCRUB_PAGES_PER_WR_BIO 32 /* 128k per bio */
57#define SCRUB_BIOS_PER_SCTX 64 /* 8MB per device in flight */
7a9e9987
SB		 58
59/*
60 * the following value times PAGE_SIZE needs to be large enough to match the
61 * largest node/leaf/sector size that shall be supported.
62 * Values larger than BTRFS_STRIPE_LEN are not supported.
63 */
b5d67f64 64#define SCRUB_MAX_PAGES_PER_BLOCK 16 /* 64k per node/leaf/sector */
a2de733c
AJ		 65
66struct scrub_page {
b5d67f64
SB		 67 struct scrub_block *sblock;
68 struct page *page;
442a4f63 69 struct btrfs_device *dev;
a2de733c
AJ		 70 u64 flags; /* extent flags */
71 u64 generation;
b5d67f64
SB		 72 u64 logical;
73 u64 physical;
ff023aac 74 u64 physical_for_dev_replace;
7a9e9987 75 atomic_t ref_count;
b5d67f64
SB		 76 struct {
77 unsigned int mirror_num:8;
78 unsigned int have_csum:1;
79 unsigned int io_error:1;
80 };
a2de733c
AJ		 81 u8 csum[BTRFS_CSUM_SIZE];
82};
83
84struct scrub_bio {
85 int index;
d9d181c1 86 struct scrub_ctx *sctx;
a36cf8b8 87 struct btrfs_device *dev;
a2de733c
AJ		 88 struct bio *bio;
89 int err;
90 u64 logical;
91 u64 physical;
ff023aac
SB		 92#if SCRUB_PAGES_PER_WR_BIO >= SCRUB_PAGES_PER_RD_BIO
93 struct scrub_page *pagev[SCRUB_PAGES_PER_WR_BIO];
94#else
95 struct scrub_page *pagev[SCRUB_PAGES_PER_RD_BIO];
96#endif
b5d67f64 97 int page_count;
a2de733c
AJ		 98 int next_free;
99 struct btrfs_work work;
100};
101
b5d67f64 102struct scrub_block {
7a9e9987 103 struct scrub_page *pagev[SCRUB_MAX_PAGES_PER_BLOCK];
b5d67f64
SB		 104 int page_count;
105 atomic_t outstanding_pages;
106 atomic_t ref_count; /* free mem on transition to zero */
d9d181c1 107 struct scrub_ctx *sctx;
b5d67f64
SB		 108 struct {
109 unsigned int header_error:1;
110 unsigned int checksum_error:1;
111 unsigned int no_io_error_seen:1;
442a4f63 112 unsigned int generation_error:1; /* also sets header_error */
b5d67f64
SB		 113 };
114};
115
ff023aac
SB		 116struct scrub_wr_ctx {
117 struct scrub_bio *wr_curr_bio;
118 struct btrfs_device *tgtdev;
119 int pages_per_wr_bio; /* <= SCRUB_PAGES_PER_WR_BIO */
120 atomic_t flush_all_writes;
121 struct mutex wr_lock;
122};
123
d9d181c1 124struct scrub_ctx {
ff023aac 125 struct scrub_bio *bios[SCRUB_BIOS_PER_SCTX];
a36cf8b8 126 struct btrfs_root *dev_root;
a2de733c
AJ		 127 int first_free;
128 int curr;
b6bfebc1
SB		 129 atomic_t bios_in_flight;
130 atomic_t workers_pending;
a2de733c
AJ		 131 spinlock_t list_lock;
132 wait_queue_head_t list_wait;
133 u16 csum_size;
134 struct list_head csum_list;
135 atomic_t cancel_req;
8628764e 136 int readonly;
ff023aac 137 int pages_per_rd_bio;
b5d67f64
SB		 138 u32 sectorsize;
139 u32 nodesize;
140 u32 leafsize;
63a212ab
SB		 141
142 int is_dev_replace;
ff023aac 143 struct scrub_wr_ctx wr_ctx;
63a212ab 144
a2de733c
AJ		 145 /*
146 * statistics
147 */
148 struct btrfs_scrub_progress stat;
149 spinlock_t stat_lock;
150};
151
0ef8e451 152struct scrub_fixup_nodatasum {
d9d181c1 153 struct scrub_ctx *sctx;
a36cf8b8 154 struct btrfs_device *dev;
0ef8e451
JS		 155 u64 logical;
156 struct btrfs_root *root;
157 struct btrfs_work work;
158 int mirror_num;
159};
160
ff023aac
SB		 161struct scrub_copy_nocow_ctx {
162 struct scrub_ctx *sctx;
163 u64 logical;
164 u64 len;
165 int mirror_num;
166 u64 physical_for_dev_replace;
167 struct btrfs_work work;
168};
169
558540c1
JS		 170struct scrub_warning {
171 struct btrfs_path *path;
172 u64 extent_item_size;
173 char *scratch_buf;
174 char *msg_buf;
175 const char *errstr;
176 sector_t sector;
177 u64 logical;
178 struct btrfs_device *dev;
179 int msg_bufsize;
180 int scratch_bufsize;
181};
182
b5d67f64 183
b6bfebc1
SB		 184static void scrub_pending_bio_inc(struct scrub_ctx *sctx);
185static void scrub_pending_bio_dec(struct scrub_ctx *sctx);
186static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx);
187static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx);
b5d67f64 188static int scrub_handle_errored_block(struct scrub_block *sblock_to_check);
d9d181c1 189static int scrub_setup_recheck_block(struct scrub_ctx *sctx,
3ec706c8 190 struct btrfs_fs_info *fs_info,
ff023aac 191 struct scrub_block *original_sblock,
b5d67f64 192 u64 length, u64 logical,
ff023aac 193 struct scrub_block *sblocks_for_recheck);
34f5c8e9
SB		 194static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
195 struct scrub_block *sblock, int is_metadata,
196 int have_csum, u8 *csum, u64 generation,
197 u16 csum_size);
b5d67f64
SB		 198static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
199 struct scrub_block *sblock,
200 int is_metadata, int have_csum,
201 const u8 *csum, u64 generation,
202 u16 csum_size);
203static void scrub_complete_bio_end_io(struct bio *bio, int err);
204static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
205 struct scrub_block *sblock_good,
206 int force_write);
207static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
208 struct scrub_block *sblock_good,
209 int page_num, int force_write);
ff023aac
SB		 210static void scrub_write_block_to_dev_replace(struct scrub_block *sblock);
211static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
212 int page_num);
b5d67f64
SB		 213static int scrub_checksum_data(struct scrub_block *sblock);
214static int scrub_checksum_tree_block(struct scrub_block *sblock);
215static int scrub_checksum_super(struct scrub_block *sblock);
216static void scrub_block_get(struct scrub_block *sblock);
217static void scrub_block_put(struct scrub_block *sblock);
7a9e9987
SB		 218static void scrub_page_get(struct scrub_page *spage);
219static void scrub_page_put(struct scrub_page *spage);
ff023aac
SB		 220static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
221 struct scrub_page *spage);
d9d181c1 222static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
a36cf8b8 223 u64 physical, struct btrfs_device *dev, u64 flags,
ff023aac
SB		 224 u64 gen, int mirror_num, u8 *csum, int force,
225 u64 physical_for_dev_replace);
1623edeb 226static void scrub_bio_end_io(struct bio *bio, int err);
b5d67f64
SB		 227static void scrub_bio_end_io_worker(struct btrfs_work *work);
228static void scrub_block_complete(struct scrub_block *sblock);
ff023aac
SB		 229static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
230 u64 extent_logical, u64 extent_len,
231 u64 *extent_physical,
232 struct btrfs_device **extent_dev,
233 int *extent_mirror_num);
234static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
235 struct scrub_wr_ctx *wr_ctx,
236 struct btrfs_fs_info *fs_info,
237 struct btrfs_device *dev,
238 int is_dev_replace);
239static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx);
240static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
241 struct scrub_page *spage);
242static void scrub_wr_submit(struct scrub_ctx *sctx);
243static void scrub_wr_bio_end_io(struct bio *bio, int err);
244static void scrub_wr_bio_end_io_worker(struct btrfs_work *work);
245static int write_page_nocow(struct scrub_ctx *sctx,
246 u64 physical_for_dev_replace, struct page *page);
247static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
248 void *ctx);
249static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
250 int mirror_num, u64 physical_for_dev_replace);
251static void copy_nocow_pages_worker(struct btrfs_work *work);
1623edeb
SB		 252
253
b6bfebc1
SB		 254static void scrub_pending_bio_inc(struct scrub_ctx *sctx)
255{
256 atomic_inc(&sctx->bios_in_flight);
257}
258
259static void scrub_pending_bio_dec(struct scrub_ctx *sctx)
260{
261 atomic_dec(&sctx->bios_in_flight);
262 wake_up(&sctx->list_wait);
263}
264
265/*
266 * used for workers that require transaction commits (i.e., for the
267 * NOCOW case)
268 */
269static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx)
270{
271 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
272
273 /*
274 * increment scrubs_running to prevent cancel requests from
275 * completing as long as a worker is running. we must also
276 * increment scrubs_paused to prevent deadlocking on pause
277 * requests used for transactions commits (as the worker uses a
278 * transaction context). it is safe to regard the worker
279 * as paused for all matters practical. effectively, we only
280 * avoid cancellation requests from completing.
281 */
282 mutex_lock(&fs_info->scrub_lock);
283 atomic_inc(&fs_info->scrubs_running);
284 atomic_inc(&fs_info->scrubs_paused);
285 mutex_unlock(&fs_info->scrub_lock);
286 atomic_inc(&sctx->workers_pending);
287}
288
289/* used for workers that require transaction commits */
290static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx)
291{
292 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
293
294 /*
295 * see scrub_pending_trans_workers_inc() why we're pretending
296 * to be paused in the scrub counters
297 */
298 mutex_lock(&fs_info->scrub_lock);
299 atomic_dec(&fs_info->scrubs_running);
300 atomic_dec(&fs_info->scrubs_paused);
301 mutex_unlock(&fs_info->scrub_lock);
302 atomic_dec(&sctx->workers_pending);
303 wake_up(&fs_info->scrub_pause_wait);
304 wake_up(&sctx->list_wait);
305}
306
d9d181c1 307static void scrub_free_csums(struct scrub_ctx *sctx)
a2de733c 308{
d9d181c1 309 while (!list_empty(&sctx->csum_list)) {
a2de733c 310 struct btrfs_ordered_sum *sum;
d9d181c1 311 sum = list_first_entry(&sctx->csum_list,
a2de733c
AJ		 312 struct btrfs_ordered_sum, list);
313 list_del(&sum->list);
314 kfree(sum);
315 }
316}
317
d9d181c1 318static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx)
a2de733c
AJ		 319{
320 int i;
a2de733c 321
d9d181c1 322 if (!sctx)
a2de733c
AJ		 323 return;
324
ff023aac
SB		 325 scrub_free_wr_ctx(&sctx->wr_ctx);
326
b5d67f64 327 /* this can happen when scrub is cancelled */
d9d181c1
SB		 328 if (sctx->curr != -1) {
329 struct scrub_bio *sbio = sctx->bios[sctx->curr];
b5d67f64
SB		 330
331 for (i = 0; i < sbio->page_count; i++) {
ff023aac 332 WARN_ON(!sbio->pagev[i]->page);
b5d67f64
SB		 333 scrub_block_put(sbio->pagev[i]->sblock);
334 }
335 bio_put(sbio->bio);
336 }
337
ff023aac 338 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
d9d181c1 339 struct scrub_bio *sbio = sctx->bios[i];
a2de733c
AJ		 340
341 if (!sbio)
342 break;
a2de733c
AJ		 343 kfree(sbio);
344 }
345
d9d181c1
SB		 346 scrub_free_csums(sctx);
347 kfree(sctx);
a2de733c
AJ		 348}
349
350static noinline_for_stack
63a212ab 351struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace)
a2de733c 352{
d9d181c1 353 struct scrub_ctx *sctx;
a2de733c 354 int i;
a2de733c 355 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
ff023aac
SB		 356 int pages_per_rd_bio;
357 int ret;
a2de733c 358
ff023aac
SB		 359 /*
360 * the setting of pages_per_rd_bio is correct for scrub but might
361 * be wrong for the dev_replace code where we might read from
362 * different devices in the initial huge bios. However, that
363 * code is able to correctly handle the case when adding a page
364 * to a bio fails.
365 */
366 if (dev->bdev)
367 pages_per_rd_bio = min_t(int, SCRUB_PAGES_PER_RD_BIO,
368 bio_get_nr_vecs(dev->bdev));
369 else
370 pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO;
d9d181c1
SB		 371 sctx = kzalloc(sizeof(*sctx), GFP_NOFS);
372 if (!sctx)
a2de733c 373 goto nomem;
63a212ab 374 sctx->is_dev_replace = is_dev_replace;
ff023aac 375 sctx->pages_per_rd_bio = pages_per_rd_bio;
d9d181c1 376 sctx->curr = -1;
a36cf8b8 377 sctx->dev_root = dev->dev_root;
ff023aac 378 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
a2de733c
AJ		 379 struct scrub_bio *sbio;
380
381 sbio = kzalloc(sizeof(*sbio), GFP_NOFS);
382 if (!sbio)
383 goto nomem;
d9d181c1 384 sctx->bios[i] = sbio;
a2de733c 385
a2de733c 386 sbio->index = i;
d9d181c1 387 sbio->sctx = sctx;
b5d67f64
SB		 388 sbio->page_count = 0;
389 sbio->work.func = scrub_bio_end_io_worker;
a2de733c 390
ff023aac 391 if (i != SCRUB_BIOS_PER_SCTX - 1)
d9d181c1 392 sctx->bios[i]->next_free = i + 1;
0ef8e451 393 else
d9d181c1
SB		 394 sctx->bios[i]->next_free = -1;
395 }
396 sctx->first_free = 0;
397 sctx->nodesize = dev->dev_root->nodesize;
398 sctx->leafsize = dev->dev_root->leafsize;
399 sctx->sectorsize = dev->dev_root->sectorsize;
b6bfebc1
SB		 400 atomic_set(&sctx->bios_in_flight, 0);
401 atomic_set(&sctx->workers_pending, 0);
d9d181c1
SB		 402 atomic_set(&sctx->cancel_req, 0);
403 sctx->csum_size = btrfs_super_csum_size(fs_info->super_copy);
404 INIT_LIST_HEAD(&sctx->csum_list);
405
406 spin_lock_init(&sctx->list_lock);
407 spin_lock_init(&sctx->stat_lock);
408 init_waitqueue_head(&sctx->list_wait);
ff023aac
SB		 409
410 ret = scrub_setup_wr_ctx(sctx, &sctx->wr_ctx, fs_info,
411 fs_info->dev_replace.tgtdev, is_dev_replace);
412 if (ret) {
413 scrub_free_ctx(sctx);
414 return ERR_PTR(ret);
415 }
d9d181c1 416 return sctx;
a2de733c
AJ		 417
418nomem:
d9d181c1 419 scrub_free_ctx(sctx);
a2de733c
AJ		 420 return ERR_PTR(-ENOMEM);
421}
422
ff023aac
SB		 423static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root,
424 void *warn_ctx)
558540c1
JS		 425{
426 u64 isize;
427 u32 nlink;
428 int ret;
429 int i;
430 struct extent_buffer *eb;
431 struct btrfs_inode_item *inode_item;
ff023aac 432 struct scrub_warning *swarn = warn_ctx;
558540c1
JS		 433 struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info;
434 struct inode_fs_paths *ipath = NULL;
435 struct btrfs_root *local_root;
436 struct btrfs_key root_key;
437
438 root_key.objectid = root;
439 root_key.type = BTRFS_ROOT_ITEM_KEY;
440 root_key.offset = (u64)-1;
441 local_root = btrfs_read_fs_root_no_name(fs_info, &root_key);
442 if (IS_ERR(local_root)) {
443 ret = PTR_ERR(local_root);
444 goto err;
445 }
446
447 ret = inode_item_info(inum, 0, local_root, swarn->path);
448 if (ret) {
449 btrfs_release_path(swarn->path);
450 goto err;
451 }
452
453 eb = swarn->path->nodes[0];
454 inode_item = btrfs_item_ptr(eb, swarn->path->slots[0],
455 struct btrfs_inode_item);
456 isize = btrfs_inode_size(eb, inode_item);
457 nlink = btrfs_inode_nlink(eb, inode_item);
458 btrfs_release_path(swarn->path);
459
460 ipath = init_ipath(4096, local_root, swarn->path);
26bdef54
DC		 461 if (IS_ERR(ipath)) {
462 ret = PTR_ERR(ipath);
463 ipath = NULL;
464 goto err;
465 }
558540c1
JS		 466 ret = paths_from_inode(inum, ipath);
467
468 if (ret < 0)
469 goto err;
470
471 /*
472 * we deliberately ignore the bit ipath might have been too small to
473 * hold all of the paths here
474 */
475 for (i = 0; i < ipath->fspath->elem_cnt; ++i)
606686ee 476 printk_in_rcu(KERN_WARNING "btrfs: %s at logical %llu on dev "
558540c1
JS		 477 "%s, sector %llu, root %llu, inode %llu, offset %llu, "
478 "length %llu, links %u (path: %s)\n", swarn->errstr,
606686ee 479 swarn->logical, rcu_str_deref(swarn->dev->name),
558540c1
JS		 480 (unsigned long long)swarn->sector, root, inum, offset,
481 min(isize - offset, (u64)PAGE_SIZE), nlink,
745c4d8e 482 (char *)(unsigned long)ipath->fspath->val[i]);
558540c1
JS		 483
484 free_ipath(ipath);
485 return 0;
486
487err:
606686ee 488 printk_in_rcu(KERN_WARNING "btrfs: %s at logical %llu on dev "
558540c1
JS		 489 "%s, sector %llu, root %llu, inode %llu, offset %llu: path "
490 "resolving failed with ret=%d\n", swarn->errstr,
606686ee 491 swarn->logical, rcu_str_deref(swarn->dev->name),
558540c1
JS		 492 (unsigned long long)swarn->sector, root, inum, offset, ret);
493
494 free_ipath(ipath);
495 return 0;
496}
497
b5d67f64 498static void scrub_print_warning(const char *errstr, struct scrub_block *sblock)
558540c1 499{
a36cf8b8
SB		 500 struct btrfs_device *dev;
501 struct btrfs_fs_info *fs_info;
558540c1
JS		 502 struct btrfs_path *path;
503 struct btrfs_key found_key;
504 struct extent_buffer *eb;
505 struct btrfs_extent_item *ei;
506 struct scrub_warning swarn;
69917e43
LB		 507 unsigned long ptr = 0;
508 u64 extent_item_pos;
509 u64 flags = 0;
558540c1 510 u64 ref_root;
69917e43 511 u32 item_size;
558540c1 512 u8 ref_level;
558540c1 513 const int bufsize = 4096;
69917e43 514 int ret;
558540c1 515
a36cf8b8 516 WARN_ON(sblock->page_count < 1);
7a9e9987 517 dev = sblock->pagev[0]->dev;
a36cf8b8
SB		 518 fs_info = sblock->sctx->dev_root->fs_info;
519
558540c1
JS		 520 path = btrfs_alloc_path();
521
522 swarn.scratch_buf = kmalloc(bufsize, GFP_NOFS);
523 swarn.msg_buf = kmalloc(bufsize, GFP_NOFS);
7a9e9987
SB		 524 swarn.sector = (sblock->pagev[0]->physical) >> 9;
525 swarn.logical = sblock->pagev[0]->logical;
558540c1 526 swarn.errstr = errstr;
a36cf8b8 527 swarn.dev = NULL;
558540c1
JS		 528 swarn.msg_bufsize = bufsize;
529 swarn.scratch_bufsize = bufsize;
530
531 if (!path || !swarn.scratch_buf || !swarn.msg_buf)
532 goto out;
533
69917e43
LB		 534 ret = extent_from_logical(fs_info, swarn.logical, path, &found_key,
535 &flags);
558540c1
JS		 536 if (ret < 0)
537 goto out;
538
4692cf58 539 extent_item_pos = swarn.logical - found_key.objectid;
558540c1
JS		 540 swarn.extent_item_size = found_key.offset;
541
542 eb = path->nodes[0];
543 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
544 item_size = btrfs_item_size_nr(eb, path->slots[0]);
545
69917e43 546 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
558540c1
JS		 547 do {
548 ret = tree_backref_for_extent(&ptr, eb, ei, item_size,
549 &ref_root, &ref_level);
606686ee 550 printk_in_rcu(KERN_WARNING
1623edeb 551 "btrfs: %s at logical %llu on dev %s, "
558540c1 552 "sector %llu: metadata %s (level %d) in tree "
606686ee
JB		 553 "%llu\n", errstr, swarn.logical,
554 rcu_str_deref(dev->name),
558540c1
JS		 555 (unsigned long long)swarn.sector,
556 ref_level ? "node" : "leaf",
557 ret < 0 ? -1 : ref_level,
558 ret < 0 ? -1 : ref_root);
559 } while (ret != 1);
d8fe29e9 560 btrfs_release_path(path);
558540c1 561 } else {
d8fe29e9 562 btrfs_release_path(path);
558540c1 563 swarn.path = path;
a36cf8b8 564 swarn.dev = dev;
7a3ae2f8
JS		 565 iterate_extent_inodes(fs_info, found_key.objectid,
566 extent_item_pos, 1,
558540c1
JS		 567 scrub_print_warning_inode, &swarn);
568 }
569
570out:
571 btrfs_free_path(path);
572 kfree(swarn.scratch_buf);
573 kfree(swarn.msg_buf);
574}
575
ff023aac 576static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx)
0ef8e451 577{
5da6fcbc 578 struct page *page = NULL;
0ef8e451 579 unsigned long index;
ff023aac 580 struct scrub_fixup_nodatasum *fixup = fixup_ctx;
0ef8e451 581 int ret;
5da6fcbc 582 int corrected = 0;
0ef8e451 583 struct btrfs_key key;
5da6fcbc 584 struct inode *inode = NULL;
6f1c3605 585 struct btrfs_fs_info *fs_info;
0ef8e451
JS		 586 u64 end = offset + PAGE_SIZE - 1;
587 struct btrfs_root *local_root;
6f1c3605 588 int srcu_index;
0ef8e451
JS		 589
590 key.objectid = root;
591 key.type = BTRFS_ROOT_ITEM_KEY;
592 key.offset = (u64)-1;
6f1c3605
LB		 593
594 fs_info = fixup->root->fs_info;
595 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
596
597 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
598 if (IS_ERR(local_root)) {
599 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
0ef8e451 600 return PTR_ERR(local_root);
6f1c3605 601 }
0ef8e451
JS		 602
603 key.type = BTRFS_INODE_ITEM_KEY;
604 key.objectid = inum;
605 key.offset = 0;
6f1c3605
LB		 606 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
607 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
0ef8e451
JS		 608 if (IS_ERR(inode))
609 return PTR_ERR(inode);
610
0ef8e451
JS		 611 index = offset >> PAGE_CACHE_SHIFT;
612
613 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
5da6fcbc
JS		 614 if (!page) {
615 ret = -ENOMEM;
616 goto out;
617 }
618
619 if (PageUptodate(page)) {
5da6fcbc
JS		 620 if (PageDirty(page)) {
621 /*
622 * we need to write the data to the defect sector. the
623 * data that was in that sector is not in memory,
624 * because the page was modified. we must not write the
625 * modified page to that sector.
626 *
627 * TODO: what could be done here: wait for the delalloc
628 * runner to write out that page (might involve
629 * COW) and see whether the sector is still
630 * referenced afterwards.
631 *
632 * For the meantime, we'll treat this error
633 * incorrectable, although there is a chance that a
634 * later scrub will find the bad sector again and that
635 * there's no dirty page in memory, then.
636 */
637 ret = -EIO;
638 goto out;
639 }
3ec706c8
SB		 640 fs_info = BTRFS_I(inode)->root->fs_info;
641 ret = repair_io_failure(fs_info, offset, PAGE_SIZE,
5da6fcbc
JS		 642 fixup->logical, page,
643 fixup->mirror_num);
644 unlock_page(page);
645 corrected = !ret;
646 } else {
647 /*
648 * we need to get good data first. the general readpage path
649 * will call repair_io_failure for us, we just have to make
650 * sure we read the bad mirror.
651 */
652 ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
653 EXTENT_DAMAGED, GFP_NOFS);
654 if (ret) {
655 /* set_extent_bits should give proper error */
656 WARN_ON(ret > 0);
657 if (ret > 0)
658 ret = -EFAULT;
659 goto out;
660 }
661
662 ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page,
663 btrfs_get_extent,
664 fixup->mirror_num);
665 wait_on_page_locked(page);
666
667 corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset,
668 end, EXTENT_DAMAGED, 0, NULL);
669 if (!corrected)
670 clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
671 EXTENT_DAMAGED, GFP_NOFS);
672 }
673
674out:
675 if (page)
676 put_page(page);
677 if (inode)
678 iput(inode);
0ef8e451
JS		 679
680 if (ret < 0)
681 return ret;
682
683 if (ret == 0 && corrected) {
684 /*
685 * we only need to call readpage for one of the inodes belonging
686 * to this extent. so make iterate_extent_inodes stop
687 */
688 return 1;
689 }
690
691 return -EIO;
692}
693
694static void scrub_fixup_nodatasum(struct btrfs_work *work)
695{
696 int ret;
697 struct scrub_fixup_nodatasum *fixup;
d9d181c1 698 struct scrub_ctx *sctx;
0ef8e451
JS		 699 struct btrfs_trans_handle *trans = NULL;
700 struct btrfs_fs_info *fs_info;
701 struct btrfs_path *path;
702 int uncorrectable = 0;
703
704 fixup = container_of(work, struct scrub_fixup_nodatasum, work);
d9d181c1 705 sctx = fixup->sctx;
0ef8e451
JS		 706 fs_info = fixup->root->fs_info;
707
708 path = btrfs_alloc_path();
709 if (!path) {
d9d181c1
SB		 710 spin_lock(&sctx->stat_lock);
711 ++sctx->stat.malloc_errors;
712 spin_unlock(&sctx->stat_lock);
0ef8e451
JS		 713 uncorrectable = 1;
714 goto out;
715 }
716
717 trans = btrfs_join_transaction(fixup->root);
718 if (IS_ERR(trans)) {
719 uncorrectable = 1;
720 goto out;
721 }
722
723 /*
724 * the idea is to trigger a regular read through the standard path. we
725 * read a page from the (failed) logical address by specifying the
726 * corresponding copynum of the failed sector. thus, that readpage is
727 * expected to fail.
728 * that is the point where on-the-fly error correction will kick in
729 * (once it's finished) and rewrite the failed sector if a good copy
730 * can be found.
731 */
732 ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info,
733 path, scrub_fixup_readpage,
734 fixup);
735 if (ret < 0) {
736 uncorrectable = 1;
737 goto out;
738 }
739 WARN_ON(ret != 1);
740
d9d181c1
SB		 741 spin_lock(&sctx->stat_lock);
742 ++sctx->stat.corrected_errors;
743 spin_unlock(&sctx->stat_lock);
0ef8e451
JS		 744
745out:
746 if (trans && !IS_ERR(trans))
747 btrfs_end_transaction(trans, fixup->root);
748 if (uncorrectable) {
d9d181c1
SB		 749 spin_lock(&sctx->stat_lock);
750 ++sctx->stat.uncorrectable_errors;
751 spin_unlock(&sctx->stat_lock);
ff023aac
SB		 752 btrfs_dev_replace_stats_inc(
753 &sctx->dev_root->fs_info->dev_replace.
754 num_uncorrectable_read_errors);
606686ee 755 printk_ratelimited_in_rcu(KERN_ERR
b5d67f64 756 "btrfs: unable to fixup (nodatasum) error at logical %llu on dev %s\n",
606686ee 757 (unsigned long long)fixup->logical,
a36cf8b8 758 rcu_str_deref(fixup->dev->name));
0ef8e451
JS		 759 }
760
761 btrfs_free_path(path);
762 kfree(fixup);
763
b6bfebc1 764 scrub_pending_trans_workers_dec(sctx);
0ef8e451
JS		 765}
766
a2de733c 767/*
b5d67f64
SB		 768 * scrub_handle_errored_block gets called when either verification of the
769 * pages failed or the bio failed to read, e.g. with EIO. In the latter
770 * case, this function handles all pages in the bio, even though only one
771 * may be bad.
772 * The goal of this function is to repair the errored block by using the
773 * contents of one of the mirrors.
a2de733c 774 */
b5d67f64 775static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
a2de733c 776{
d9d181c1 777 struct scrub_ctx *sctx = sblock_to_check->sctx;
a36cf8b8 778 struct btrfs_device *dev;
b5d67f64
SB		 779 struct btrfs_fs_info *fs_info;
780 u64 length;
781 u64 logical;
782 u64 generation;
783 unsigned int failed_mirror_index;
784 unsigned int is_metadata;
785 unsigned int have_csum;
786 u8 *csum;
787 struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */
788 struct scrub_block *sblock_bad;
789 int ret;
790 int mirror_index;
791 int page_num;
792 int success;
558540c1 793 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
b5d67f64
SB		 794 DEFAULT_RATELIMIT_BURST);
795
796 BUG_ON(sblock_to_check->page_count < 1);
a36cf8b8 797 fs_info = sctx->dev_root->fs_info;
4ded4f63
SB		 798 if (sblock_to_check->pagev[0]->flags & BTRFS_EXTENT_FLAG_SUPER) {
799 /*
800 * if we find an error in a super block, we just report it.
801 * They will get written with the next transaction commit
802 * anyway
803 */
804 spin_lock(&sctx->stat_lock);
805 ++sctx->stat.super_errors;
806 spin_unlock(&sctx->stat_lock);
807 return 0;
808 }
b5d67f64 809 length = sblock_to_check->page_count * PAGE_SIZE;
7a9e9987
SB		 810 logical = sblock_to_check->pagev[0]->logical;
811 generation = sblock_to_check->pagev[0]->generation;
812 BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1);
813 failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1;
814 is_metadata = !(sblock_to_check->pagev[0]->flags &
b5d67f64 815 BTRFS_EXTENT_FLAG_DATA);
7a9e9987
SB		 816 have_csum = sblock_to_check->pagev[0]->have_csum;
817 csum = sblock_to_check->pagev[0]->csum;
818 dev = sblock_to_check->pagev[0]->dev;
13db62b7 819
ff023aac
SB		 820 if (sctx->is_dev_replace && !is_metadata && !have_csum) {
821 sblocks_for_recheck = NULL;
822 goto nodatasum_case;
823 }
824
b5d67f64
SB		 825 /*
826 * read all mirrors one after the other. This includes to
827 * re-read the extent or metadata block that failed (that was
828 * the cause that this fixup code is called) another time,
829 * page by page this time in order to know which pages
830 * caused I/O errors and which ones are good (for all mirrors).
831 * It is the goal to handle the situation when more than one
832 * mirror contains I/O errors, but the errors do not
833 * overlap, i.e. the data can be repaired by selecting the
834 * pages from those mirrors without I/O error on the
835 * particular pages. One example (with blocks >= 2 * PAGE_SIZE)
836 * would be that mirror #1 has an I/O error on the first page,
837 * the second page is good, and mirror #2 has an I/O error on
838 * the second page, but the first page is good.
839 * Then the first page of the first mirror can be repaired by
840 * taking the first page of the second mirror, and the
841 * second page of the second mirror can be repaired by
842 * copying the contents of the 2nd page of the 1st mirror.
843 * One more note: if the pages of one mirror contain I/O
844 * errors, the checksum cannot be verified. In order to get
845 * the best data for repairing, the first attempt is to find
846 * a mirror without I/O errors and with a validated checksum.
847 * Only if this is not possible, the pages are picked from
848 * mirrors with I/O errors without considering the checksum.
849 * If the latter is the case, at the end, the checksum of the
850 * repaired area is verified in order to correctly maintain
851 * the statistics.
852 */
853
854 sblocks_for_recheck = kzalloc(BTRFS_MAX_MIRRORS *
855 sizeof(*sblocks_for_recheck),
856 GFP_NOFS);
857 if (!sblocks_for_recheck) {
d9d181c1
SB		 858 spin_lock(&sctx->stat_lock);
859 sctx->stat.malloc_errors++;
860 sctx->stat.read_errors++;
861 sctx->stat.uncorrectable_errors++;
862 spin_unlock(&sctx->stat_lock);
a36cf8b8 863 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
b5d67f64 864 goto out;
a2de733c
AJ		 865 }
866
b5d67f64 867 /* setup the context, map the logical blocks and alloc the pages */
ff023aac 868 ret = scrub_setup_recheck_block(sctx, fs_info, sblock_to_check, length,
b5d67f64
SB		 869 logical, sblocks_for_recheck);
870 if (ret) {
d9d181c1
SB		 871 spin_lock(&sctx->stat_lock);
872 sctx->stat.read_errors++;
873 sctx->stat.uncorrectable_errors++;
874 spin_unlock(&sctx->stat_lock);
a36cf8b8 875 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
b5d67f64
SB		 876 goto out;
877 }
878 BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS);
879 sblock_bad = sblocks_for_recheck + failed_mirror_index;
13db62b7 880
b5d67f64 881 /* build and submit the bios for the failed mirror, check checksums */
34f5c8e9
SB		 882 scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum,
883 csum, generation, sctx->csum_size);
a2de733c 884
b5d67f64
SB		 885 if (!sblock_bad->header_error && !sblock_bad->checksum_error &&
886 sblock_bad->no_io_error_seen) {
887 /*
888 * the error disappeared after reading page by page, or
889 * the area was part of a huge bio and other parts of the
890 * bio caused I/O errors, or the block layer merged several
891 * read requests into one and the error is caused by a
892 * different bio (usually one of the two latter cases is
893 * the cause)
894 */
d9d181c1
SB		 895 spin_lock(&sctx->stat_lock);
896 sctx->stat.unverified_errors++;
897 spin_unlock(&sctx->stat_lock);
a2de733c 898
ff023aac
SB		 899 if (sctx->is_dev_replace)
900 scrub_write_block_to_dev_replace(sblock_bad);
b5d67f64 901 goto out;
a2de733c 902 }
a2de733c 903
b5d67f64 904 if (!sblock_bad->no_io_error_seen) {
d9d181c1
SB		 905 spin_lock(&sctx->stat_lock);
906 sctx->stat.read_errors++;
907 spin_unlock(&sctx->stat_lock);
b5d67f64
SB		 908 if (__ratelimit(&_rs))
909 scrub_print_warning("i/o error", sblock_to_check);
a36cf8b8 910 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
b5d67f64 911 } else if (sblock_bad->checksum_error) {
d9d181c1
SB		 912 spin_lock(&sctx->stat_lock);
913 sctx->stat.csum_errors++;
914 spin_unlock(&sctx->stat_lock);
b5d67f64
SB		 915 if (__ratelimit(&_rs))
916 scrub_print_warning("checksum error", sblock_to_check);
a36cf8b8 917 btrfs_dev_stat_inc_and_print(dev,
442a4f63 918 BTRFS_DEV_STAT_CORRUPTION_ERRS);
b5d67f64 919 } else if (sblock_bad->header_error) {
d9d181c1
SB		 920 spin_lock(&sctx->stat_lock);
921 sctx->stat.verify_errors++;
922 spin_unlock(&sctx->stat_lock);
b5d67f64
SB		 923 if (__ratelimit(&_rs))
924 scrub_print_warning("checksum/header error",
925 sblock_to_check);
442a4f63 926 if (sblock_bad->generation_error)
a36cf8b8 927 btrfs_dev_stat_inc_and_print(dev,
442a4f63
SB		 928 BTRFS_DEV_STAT_GENERATION_ERRS);
929 else
a36cf8b8 930 btrfs_dev_stat_inc_and_print(dev,
442a4f63 931 BTRFS_DEV_STAT_CORRUPTION_ERRS);
b5d67f64 932 }
a2de733c 933
ff023aac 934 if (sctx->readonly && !sctx->is_dev_replace)
b5d67f64 935 goto did_not_correct_error;
a2de733c 936
b5d67f64
SB		 937 if (!is_metadata && !have_csum) {
938 struct scrub_fixup_nodatasum *fixup_nodatasum;
a2de733c 939
ff023aac
SB		 940nodatasum_case:
941 WARN_ON(sctx->is_dev_replace);
942
b5d67f64
SB		 943 /*
944 * !is_metadata and !have_csum, this means that the data
945 * might not be COW'ed, that it might be modified
946 * concurrently. The general strategy to work on the
947 * commit root does not help in the case when COW is not
948 * used.
949 */
950 fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS);
951 if (!fixup_nodatasum)
952 goto did_not_correct_error;
d9d181c1 953 fixup_nodatasum->sctx = sctx;
a36cf8b8 954 fixup_nodatasum->dev = dev;
b5d67f64
SB		 955 fixup_nodatasum->logical = logical;
956 fixup_nodatasum->root = fs_info->extent_root;
957 fixup_nodatasum->mirror_num = failed_mirror_index + 1;
b6bfebc1 958 scrub_pending_trans_workers_inc(sctx);
b5d67f64
SB		 959 fixup_nodatasum->work.func = scrub_fixup_nodatasum;
960 btrfs_queue_worker(&fs_info->scrub_workers,
961 &fixup_nodatasum->work);
962 goto out;
a2de733c
AJ		 963 }
964
b5d67f64
SB		 965 /*
966 * now build and submit the bios for the other mirrors, check
cb2ced73
SB		 967 * checksums.
968 * First try to pick the mirror which is completely without I/O
b5d67f64
SB		 969 * errors and also does not have a checksum error.
970 * If one is found, and if a checksum is present, the full block
971 * that is known to contain an error is rewritten. Afterwards
972 * the block is known to be corrected.
973 * If a mirror is found which is completely correct, and no
974 * checksum is present, only those pages are rewritten that had
975 * an I/O error in the block to be repaired, since it cannot be
976 * determined, which copy of the other pages is better (and it
977 * could happen otherwise that a correct page would be
978 * overwritten by a bad one).
979 */
980 for (mirror_index = 0;
981 mirror_index < BTRFS_MAX_MIRRORS &&
982 sblocks_for_recheck[mirror_index].page_count > 0;
983 mirror_index++) {
cb2ced73 984 struct scrub_block *sblock_other;
b5d67f64 985
cb2ced73
SB		 986 if (mirror_index == failed_mirror_index)
987 continue;
988 sblock_other = sblocks_for_recheck + mirror_index;
989
990 /* build and submit the bios, check checksums */
34f5c8e9
SB		 991 scrub_recheck_block(fs_info, sblock_other, is_metadata,
992 have_csum, csum, generation,
993 sctx->csum_size);
994
995 if (!sblock_other->header_error &&
b5d67f64
SB		 996 !sblock_other->checksum_error &&
997 sblock_other->no_io_error_seen) {
ff023aac
SB		 998 if (sctx->is_dev_replace) {
999 scrub_write_block_to_dev_replace(sblock_other);
1000 } else {
1001 int force_write = is_metadata || have_csum;
1002
1003 ret = scrub_repair_block_from_good_copy(
1004 sblock_bad, sblock_other,
1005 force_write);
1006 }
b5d67f64
SB		 1007 if (0 == ret)
1008 goto corrected_error;
1009 }
1010 }
a2de733c
AJ		 1011
1012 /*
ff023aac
SB		 1013 * for dev_replace, pick good pages and write to the target device.
1014 */
1015 if (sctx->is_dev_replace) {
1016 success = 1;
1017 for (page_num = 0; page_num < sblock_bad->page_count;
1018 page_num++) {
1019 int sub_success;
1020
1021 sub_success = 0;
1022 for (mirror_index = 0;
1023 mirror_index < BTRFS_MAX_MIRRORS &&
1024 sblocks_for_recheck[mirror_index].page_count > 0;
1025 mirror_index++) {
1026 struct scrub_block *sblock_other =
1027 sblocks_for_recheck + mirror_index;
1028 struct scrub_page *page_other =
1029 sblock_other->pagev[page_num];
1030
1031 if (!page_other->io_error) {
1032 ret = scrub_write_page_to_dev_replace(
1033 sblock_other, page_num);
1034 if (ret == 0) {
1035 /* succeeded for this page */
1036 sub_success = 1;
1037 break;
1038 } else {
1039 btrfs_dev_replace_stats_inc(
1040 &sctx->dev_root->
1041 fs_info->dev_replace.
1042 num_write_errors);
1043 }
1044 }
1045 }
1046
1047 if (!sub_success) {
1048 /*
1049 * did not find a mirror to fetch the page
1050 * from. scrub_write_page_to_dev_replace()
1051 * handles this case (page->io_error), by
1052 * filling the block with zeros before
1053 * submitting the write request
1054 */
1055 success = 0;
1056 ret = scrub_write_page_to_dev_replace(
1057 sblock_bad, page_num);
1058 if (ret)
1059 btrfs_dev_replace_stats_inc(
1060 &sctx->dev_root->fs_info->
1061 dev_replace.num_write_errors);
1062 }
1063 }
1064
1065 goto out;
1066 }
1067
1068 /*
1069 * for regular scrub, repair those pages that are errored.
1070 * In case of I/O errors in the area that is supposed to be
b5d67f64
SB		 1071 * repaired, continue by picking good copies of those pages.
1072 * Select the good pages from mirrors to rewrite bad pages from
1073 * the area to fix. Afterwards verify the checksum of the block
1074 * that is supposed to be repaired. This verification step is
1075 * only done for the purpose of statistic counting and for the
1076 * final scrub report, whether errors remain.
1077 * A perfect algorithm could make use of the checksum and try
1078 * all possible combinations of pages from the different mirrors
1079 * until the checksum verification succeeds. For example, when
1080 * the 2nd page of mirror #1 faces I/O errors, and the 2nd page
1081 * of mirror #2 is readable but the final checksum test fails,
1082 * then the 2nd page of mirror #3 could be tried, whether now
1083 * the final checksum succeedes. But this would be a rare
1084 * exception and is therefore not implemented. At least it is
1085 * avoided that the good copy is overwritten.
1086 * A more useful improvement would be to pick the sectors
1087 * without I/O error based on sector sizes (512 bytes on legacy
1088 * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one
1089 * mirror could be repaired by taking 512 byte of a different
1090 * mirror, even if other 512 byte sectors in the same PAGE_SIZE
1091 * area are unreadable.
a2de733c 1092 */
a2de733c 1093
b5d67f64
SB		 1094 /* can only fix I/O errors from here on */
1095 if (sblock_bad->no_io_error_seen)
1096 goto did_not_correct_error;
1097
1098 success = 1;
1099 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
7a9e9987 1100 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
b5d67f64
SB		 1101
1102 if (!page_bad->io_error)
a2de733c 1103 continue;
b5d67f64
SB		 1104
1105 for (mirror_index = 0;
1106 mirror_index < BTRFS_MAX_MIRRORS &&
1107 sblocks_for_recheck[mirror_index].page_count > 0;
1108 mirror_index++) {
1109 struct scrub_block *sblock_other = sblocks_for_recheck +
1110 mirror_index;
7a9e9987
SB		 1111 struct scrub_page *page_other = sblock_other->pagev[
1112 page_num];
b5d67f64
SB		 1113
1114 if (!page_other->io_error) {
1115 ret = scrub_repair_page_from_good_copy(
1116 sblock_bad, sblock_other, page_num, 0);
1117 if (0 == ret) {
1118 page_bad->io_error = 0;
1119 break; /* succeeded for this page */
1120 }
1121 }
96e36920 1122 }
a2de733c 1123
b5d67f64
SB		 1124 if (page_bad->io_error) {
1125 /* did not find a mirror to copy the page from */
1126 success = 0;
1127 }
a2de733c 1128 }
a2de733c 1129
b5d67f64
SB		 1130 if (success) {
1131 if (is_metadata || have_csum) {
1132 /*
1133 * need to verify the checksum now that all
1134 * sectors on disk are repaired (the write
1135 * request for data to be repaired is on its way).
1136 * Just be lazy and use scrub_recheck_block()
1137 * which re-reads the data before the checksum
1138 * is verified, but most likely the data comes out
1139 * of the page cache.
1140 */
34f5c8e9
SB		 1141 scrub_recheck_block(fs_info, sblock_bad,
1142 is_metadata, have_csum, csum,
1143 generation, sctx->csum_size);
1144 if (!sblock_bad->header_error &&
b5d67f64
SB		 1145 !sblock_bad->checksum_error &&
1146 sblock_bad->no_io_error_seen)
1147 goto corrected_error;
1148 else
1149 goto did_not_correct_error;
1150 } else {
1151corrected_error:
d9d181c1
SB		 1152 spin_lock(&sctx->stat_lock);
1153 sctx->stat.corrected_errors++;
1154 spin_unlock(&sctx->stat_lock);
606686ee 1155 printk_ratelimited_in_rcu(KERN_ERR
b5d67f64 1156 "btrfs: fixed up error at logical %llu on dev %s\n",
606686ee 1157 (unsigned long long)logical,
a36cf8b8 1158 rcu_str_deref(dev->name));
8628764e 1159 }
b5d67f64
SB		 1160 } else {
1161did_not_correct_error:
d9d181c1
SB		 1162 spin_lock(&sctx->stat_lock);
1163 sctx->stat.uncorrectable_errors++;
1164 spin_unlock(&sctx->stat_lock);
606686ee 1165 printk_ratelimited_in_rcu(KERN_ERR
b5d67f64 1166 "btrfs: unable to fixup (regular) error at logical %llu on dev %s\n",
606686ee 1167 (unsigned long long)logical,
a36cf8b8 1168 rcu_str_deref(dev->name));
96e36920 1169 }
a2de733c 1170
b5d67f64
SB		 1171out:
1172 if (sblocks_for_recheck) {
1173 for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS;
1174 mirror_index++) {
1175 struct scrub_block *sblock = sblocks_for_recheck +
1176 mirror_index;
1177 int page_index;
1178
7a9e9987
SB		 1179 for (page_index = 0; page_index < sblock->page_count;
1180 page_index++) {
1181 sblock->pagev[page_index]->sblock = NULL;
1182 scrub_page_put(sblock->pagev[page_index]);
1183 }
b5d67f64
SB		 1184 }
1185 kfree(sblocks_for_recheck);
1186 }
a2de733c 1187
b5d67f64
SB		 1188 return 0;
1189}
a2de733c 1190
d9d181c1 1191static int scrub_setup_recheck_block(struct scrub_ctx *sctx,
3ec706c8 1192 struct btrfs_fs_info *fs_info,
ff023aac 1193 struct scrub_block *original_sblock,
b5d67f64
SB		 1194 u64 length, u64 logical,
1195 struct scrub_block *sblocks_for_recheck)
1196{
1197 int page_index;
1198 int mirror_index;
1199 int ret;
1200
1201 /*
7a9e9987 1202 * note: the two members ref_count and outstanding_pages
b5d67f64
SB		 1203 * are not used (and not set) in the blocks that are used for
1204 * the recheck procedure
1205 */
1206
1207 page_index = 0;
1208 while (length > 0) {
1209 u64 sublen = min_t(u64, length, PAGE_SIZE);
1210 u64 mapped_length = sublen;
1211 struct btrfs_bio *bbio = NULL;
a2de733c 1212
b5d67f64
SB		 1213 /*
1214 * with a length of PAGE_SIZE, each returned stripe
1215 * represents one mirror
1216 */
29a8d9a0
SB		 1217 ret = btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS, logical,
1218 &mapped_length, &bbio, 0);
b5d67f64
SB		 1219 if (ret || !bbio || mapped_length < sublen) {
1220 kfree(bbio);
1221 return -EIO;
1222 }
a2de733c 1223
ff023aac 1224 BUG_ON(page_index >= SCRUB_PAGES_PER_RD_BIO);
b5d67f64
SB		 1225 for (mirror_index = 0; mirror_index < (int)bbio->num_stripes;
1226 mirror_index++) {
1227 struct scrub_block *sblock;
1228 struct scrub_page *page;
1229
1230 if (mirror_index >= BTRFS_MAX_MIRRORS)
1231 continue;
1232
1233 sblock = sblocks_for_recheck + mirror_index;
7a9e9987
SB		 1234 sblock->sctx = sctx;
1235 page = kzalloc(sizeof(*page), GFP_NOFS);
1236 if (!page) {
1237leave_nomem:
d9d181c1
SB		 1238 spin_lock(&sctx->stat_lock);
1239 sctx->stat.malloc_errors++;
1240 spin_unlock(&sctx->stat_lock);
cf93dcce 1241 kfree(bbio);
b5d67f64
SB		 1242 return -ENOMEM;
1243 }
7a9e9987
SB		 1244 scrub_page_get(page);
1245 sblock->pagev[page_index] = page;
1246 page->logical = logical;
1247 page->physical = bbio->stripes[mirror_index].physical;
ff023aac
SB		 1248 BUG_ON(page_index >= original_sblock->page_count);
1249 page->physical_for_dev_replace =
1250 original_sblock->pagev[page_index]->
1251 physical_for_dev_replace;
7a9e9987
SB		 1252 /* for missing devices, dev->bdev is NULL */
1253 page->dev = bbio->stripes[mirror_index].dev;
1254 page->mirror_num = mirror_index + 1;
b5d67f64 1255 sblock->page_count++;
7a9e9987
SB		 1256 page->page = alloc_page(GFP_NOFS);
1257 if (!page->page)
1258 goto leave_nomem;
b5d67f64
SB		 1259 }
1260 kfree(bbio);
1261 length -= sublen;
1262 logical += sublen;
1263 page_index++;
1264 }
1265
1266 return 0;
96e36920
ID		 1267}
1268
b5d67f64
SB		 1269/*
1270 * this function will check the on disk data for checksum errors, header
1271 * errors and read I/O errors. If any I/O errors happen, the exact pages
1272 * which are errored are marked as being bad. The goal is to enable scrub
1273 * to take those pages that are not errored from all the mirrors so that
1274 * the pages that are errored in the just handled mirror can be repaired.
1275 */
34f5c8e9
SB		 1276static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
1277 struct scrub_block *sblock, int is_metadata,
1278 int have_csum, u8 *csum, u64 generation,
1279 u16 csum_size)
96e36920 1280{
b5d67f64 1281 int page_num;
96e36920 1282
b5d67f64
SB		 1283 sblock->no_io_error_seen = 1;
1284 sblock->header_error = 0;
1285 sblock->checksum_error = 0;
96e36920 1286
b5d67f64
SB		 1287 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1288 struct bio *bio;
7a9e9987 1289 struct scrub_page *page = sblock->pagev[page_num];
b5d67f64
SB		 1290 DECLARE_COMPLETION_ONSTACK(complete);
1291
442a4f63 1292 if (page->dev->bdev == NULL) {
ea9947b4
SB		 1293 page->io_error = 1;
1294 sblock->no_io_error_seen = 0;
1295 continue;
1296 }
1297
7a9e9987 1298 WARN_ON(!page->page);
b5d67f64 1299 bio = bio_alloc(GFP_NOFS, 1);
34f5c8e9
SB		 1300 if (!bio) {
1301 page->io_error = 1;
1302 sblock->no_io_error_seen = 0;
1303 continue;
1304 }
442a4f63 1305 bio->bi_bdev = page->dev->bdev;
b5d67f64
SB		 1306 bio->bi_sector = page->physical >> 9;
1307 bio->bi_end_io = scrub_complete_bio_end_io;
1308 bio->bi_private = &complete;
1309
34f5c8e9 1310 bio_add_page(bio, page->page, PAGE_SIZE, 0);
b5d67f64 1311 btrfsic_submit_bio(READ, bio);
96e36920 1312
b5d67f64
SB		 1313 /* this will also unplug the queue */
1314 wait_for_completion(&complete);
96e36920 1315
b5d67f64
SB		 1316 page->io_error = !test_bit(BIO_UPTODATE, &bio->bi_flags);
1317 if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1318 sblock->no_io_error_seen = 0;
1319 bio_put(bio);
1320 }
96e36920 1321
b5d67f64
SB		 1322 if (sblock->no_io_error_seen)
1323 scrub_recheck_block_checksum(fs_info, sblock, is_metadata,
1324 have_csum, csum, generation,
1325 csum_size);
1326
34f5c8e9 1327 return;
a2de733c
AJ		 1328}
1329
b5d67f64
SB		 1330static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
1331 struct scrub_block *sblock,
1332 int is_metadata, int have_csum,
1333 const u8 *csum, u64 generation,
1334 u16 csum_size)
a2de733c 1335{
b5d67f64
SB		 1336 int page_num;
1337 u8 calculated_csum[BTRFS_CSUM_SIZE];
1338 u32 crc = ~(u32)0;
b5d67f64
SB		 1339 void *mapped_buffer;
1340
7a9e9987 1341 WARN_ON(!sblock->pagev[0]->page);
b5d67f64
SB		 1342 if (is_metadata) {
1343 struct btrfs_header *h;
1344
7a9e9987 1345 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
b5d67f64
SB		 1346 h = (struct btrfs_header *)mapped_buffer;
1347
7a9e9987 1348 if (sblock->pagev[0]->logical != le64_to_cpu(h->bytenr) ||
b5d67f64
SB		 1349 memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE) ||
1350 memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
442a4f63 1351 BTRFS_UUID_SIZE)) {
b5d67f64 1352 sblock->header_error = 1;
442a4f63
SB		 1353 } else if (generation != le64_to_cpu(h->generation)) {
1354 sblock->header_error = 1;
1355 sblock->generation_error = 1;
1356 }
b5d67f64
SB		 1357 csum = h->csum;
1358 } else {
1359 if (!have_csum)
1360 return;
a2de733c 1361
7a9e9987 1362 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
b5d67f64 1363 }
a2de733c 1364
b5d67f64
SB		 1365 for (page_num = 0;;) {
1366 if (page_num == 0 && is_metadata)
b0496686 1367 crc = btrfs_csum_data(
b5d67f64
SB		 1368 ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE,
1369 crc, PAGE_SIZE - BTRFS_CSUM_SIZE);
1370 else
b0496686 1371 crc = btrfs_csum_data(mapped_buffer, crc, PAGE_SIZE);
b5d67f64 1372
9613bebb 1373 kunmap_atomic(mapped_buffer);
b5d67f64
SB		 1374 page_num++;
1375 if (page_num >= sblock->page_count)
1376 break;
7a9e9987 1377 WARN_ON(!sblock->pagev[page_num]->page);
b5d67f64 1378
7a9e9987 1379 mapped_buffer = kmap_atomic(sblock->pagev[page_num]->page);
b5d67f64
SB		 1380 }
1381
1382 btrfs_csum_final(crc, calculated_csum);
1383 if (memcmp(calculated_csum, csum, csum_size))
1384 sblock->checksum_error = 1;
a2de733c
AJ		 1385}
1386
b5d67f64 1387static void scrub_complete_bio_end_io(struct bio *bio, int err)
a2de733c 1388{
b5d67f64
SB		 1389 complete((struct completion *)bio->bi_private);
1390}
a2de733c 1391
b5d67f64
SB		 1392static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
1393 struct scrub_block *sblock_good,
1394 int force_write)
1395{
1396 int page_num;
1397 int ret = 0;
96e36920 1398
b5d67f64
SB		 1399 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
1400 int ret_sub;
96e36920 1401
b5d67f64
SB		 1402 ret_sub = scrub_repair_page_from_good_copy(sblock_bad,
1403 sblock_good,
1404 page_num,
1405 force_write);
1406 if (ret_sub)
1407 ret = ret_sub;
a2de733c 1408 }
b5d67f64
SB		 1409
1410 return ret;
1411}
1412
1413static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
1414 struct scrub_block *sblock_good,
1415 int page_num, int force_write)
1416{
7a9e9987
SB		 1417 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1418 struct scrub_page *page_good = sblock_good->pagev[page_num];
b5d67f64 1419
7a9e9987
SB		 1420 BUG_ON(page_bad->page == NULL);
1421 BUG_ON(page_good->page == NULL);
b5d67f64
SB		 1422 if (force_write || sblock_bad->header_error ||
1423 sblock_bad->checksum_error || page_bad->io_error) {
1424 struct bio *bio;
1425 int ret;
1426 DECLARE_COMPLETION_ONSTACK(complete);
1427
ff023aac
SB		 1428 if (!page_bad->dev->bdev) {
1429 printk_ratelimited(KERN_WARNING
1430 "btrfs: scrub_repair_page_from_good_copy(bdev == NULL) is unexpected!\n");
1431 return -EIO;
1432 }
1433
b5d67f64 1434 bio = bio_alloc(GFP_NOFS, 1);
e627ee7b
TI		 1435 if (!bio)
1436 return -EIO;
442a4f63 1437 bio->bi_bdev = page_bad->dev->bdev;
b5d67f64
SB		 1438 bio->bi_sector = page_bad->physical >> 9;
1439 bio->bi_end_io = scrub_complete_bio_end_io;
1440 bio->bi_private = &complete;
1441
1442 ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0);
1443 if (PAGE_SIZE != ret) {
1444 bio_put(bio);
1445 return -EIO;
13db62b7 1446 }
b5d67f64
SB		 1447 btrfsic_submit_bio(WRITE, bio);
1448
1449 /* this will also unplug the queue */
1450 wait_for_completion(&complete);
442a4f63
SB		 1451 if (!bio_flagged(bio, BIO_UPTODATE)) {
1452 btrfs_dev_stat_inc_and_print(page_bad->dev,
1453 BTRFS_DEV_STAT_WRITE_ERRS);
ff023aac
SB		 1454 btrfs_dev_replace_stats_inc(
1455 &sblock_bad->sctx->dev_root->fs_info->
1456 dev_replace.num_write_errors);
442a4f63
SB		 1457 bio_put(bio);
1458 return -EIO;
1459 }
b5d67f64 1460 bio_put(bio);
a2de733c
AJ		 1461 }
1462
b5d67f64
SB		 1463 return 0;
1464}
1465
ff023aac
SB		 1466static void scrub_write_block_to_dev_replace(struct scrub_block *sblock)
1467{
1468 int page_num;
1469
1470 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1471 int ret;
1472
1473 ret = scrub_write_page_to_dev_replace(sblock, page_num);
1474 if (ret)
1475 btrfs_dev_replace_stats_inc(
1476 &sblock->sctx->dev_root->fs_info->dev_replace.
1477 num_write_errors);
1478 }
1479}
1480
1481static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
1482 int page_num)
1483{
1484 struct scrub_page *spage = sblock->pagev[page_num];
1485
1486 BUG_ON(spage->page == NULL);
1487 if (spage->io_error) {
1488 void *mapped_buffer = kmap_atomic(spage->page);
1489
1490 memset(mapped_buffer, 0, PAGE_CACHE_SIZE);
1491 flush_dcache_page(spage->page);
1492 kunmap_atomic(mapped_buffer);
1493 }
1494 return scrub_add_page_to_wr_bio(sblock->sctx, spage);
1495}
1496
1497static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
1498 struct scrub_page *spage)
1499{
1500 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1501 struct scrub_bio *sbio;
1502 int ret;
1503
1504 mutex_lock(&wr_ctx->wr_lock);
1505again:
1506 if (!wr_ctx->wr_curr_bio) {
1507 wr_ctx->wr_curr_bio = kzalloc(sizeof(*wr_ctx->wr_curr_bio),
1508 GFP_NOFS);
1509 if (!wr_ctx->wr_curr_bio) {
1510 mutex_unlock(&wr_ctx->wr_lock);
1511 return -ENOMEM;
1512 }
1513 wr_ctx->wr_curr_bio->sctx = sctx;
1514 wr_ctx->wr_curr_bio->page_count = 0;
1515 }
1516 sbio = wr_ctx->wr_curr_bio;
1517 if (sbio->page_count == 0) {
1518 struct bio *bio;
1519
1520 sbio->physical = spage->physical_for_dev_replace;
1521 sbio->logical = spage->logical;
1522 sbio->dev = wr_ctx->tgtdev;
1523 bio = sbio->bio;
1524 if (!bio) {
1525 bio = bio_alloc(GFP_NOFS, wr_ctx->pages_per_wr_bio);
1526 if (!bio) {
1527 mutex_unlock(&wr_ctx->wr_lock);
1528 return -ENOMEM;
1529 }
1530 sbio->bio = bio;
1531 }
1532
1533 bio->bi_private = sbio;
1534 bio->bi_end_io = scrub_wr_bio_end_io;
1535 bio->bi_bdev = sbio->dev->bdev;
1536 bio->bi_sector = sbio->physical >> 9;
1537 sbio->err = 0;
1538 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
1539 spage->physical_for_dev_replace ||
1540 sbio->logical + sbio->page_count * PAGE_SIZE !=
1541 spage->logical) {
1542 scrub_wr_submit(sctx);
1543 goto again;
1544 }
1545
1546 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
1547 if (ret != PAGE_SIZE) {
1548 if (sbio->page_count < 1) {
1549 bio_put(sbio->bio);
1550 sbio->bio = NULL;
1551 mutex_unlock(&wr_ctx->wr_lock);
1552 return -EIO;
1553 }
1554 scrub_wr_submit(sctx);
1555 goto again;
1556 }
1557
1558 sbio->pagev[sbio->page_count] = spage;
1559 scrub_page_get(spage);
1560 sbio->page_count++;
1561 if (sbio->page_count == wr_ctx->pages_per_wr_bio)
1562 scrub_wr_submit(sctx);
1563 mutex_unlock(&wr_ctx->wr_lock);
1564
1565 return 0;
1566}
1567
1568static void scrub_wr_submit(struct scrub_ctx *sctx)
1569{
1570 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1571 struct scrub_bio *sbio;
1572
1573 if (!wr_ctx->wr_curr_bio)
1574 return;
1575
1576 sbio = wr_ctx->wr_curr_bio;
1577 wr_ctx->wr_curr_bio = NULL;
1578 WARN_ON(!sbio->bio->bi_bdev);
1579 scrub_pending_bio_inc(sctx);
1580 /* process all writes in a single worker thread. Then the block layer
1581 * orders the requests before sending them to the driver which
1582 * doubled the write performance on spinning disks when measured
1583 * with Linux 3.5 */
1584 btrfsic_submit_bio(WRITE, sbio->bio);
1585}
1586
1587static void scrub_wr_bio_end_io(struct bio *bio, int err)
1588{
1589 struct scrub_bio *sbio = bio->bi_private;
1590 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
1591
1592 sbio->err = err;
1593 sbio->bio = bio;
1594
1595 sbio->work.func = scrub_wr_bio_end_io_worker;
1596 btrfs_queue_worker(&fs_info->scrub_wr_completion_workers, &sbio->work);
1597}
1598
1599static void scrub_wr_bio_end_io_worker(struct btrfs_work *work)
1600{
1601 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
1602 struct scrub_ctx *sctx = sbio->sctx;
1603 int i;
1604
1605 WARN_ON(sbio->page_count > SCRUB_PAGES_PER_WR_BIO);
1606 if (sbio->err) {
1607 struct btrfs_dev_replace *dev_replace =
1608 &sbio->sctx->dev_root->fs_info->dev_replace;
1609
1610 for (i = 0; i < sbio->page_count; i++) {
1611 struct scrub_page *spage = sbio->pagev[i];
1612
1613 spage->io_error = 1;
1614 btrfs_dev_replace_stats_inc(&dev_replace->
1615 num_write_errors);
1616 }
1617 }
1618
1619 for (i = 0; i < sbio->page_count; i++)
1620 scrub_page_put(sbio->pagev[i]);
1621
1622 bio_put(sbio->bio);
1623 kfree(sbio);
1624 scrub_pending_bio_dec(sctx);
1625}
1626
1627static int scrub_checksum(struct scrub_block *sblock)
b5d67f64
SB		 1628{
1629 u64 flags;
1630 int ret;
1631
7a9e9987
SB		 1632 WARN_ON(sblock->page_count < 1);
1633 flags = sblock->pagev[0]->flags;
b5d67f64
SB		 1634 ret = 0;
1635 if (flags & BTRFS_EXTENT_FLAG_DATA)
1636 ret = scrub_checksum_data(sblock);
1637 else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1638 ret = scrub_checksum_tree_block(sblock);
1639 else if (flags & BTRFS_EXTENT_FLAG_SUPER)
1640 (void)scrub_checksum_super(sblock);
1641 else
1642 WARN_ON(1);
1643 if (ret)
1644 scrub_handle_errored_block(sblock);
ff023aac
SB		 1645
1646 return ret;
a2de733c
AJ		 1647}
1648
b5d67f64 1649static int scrub_checksum_data(struct scrub_block *sblock)
a2de733c 1650{
d9d181c1 1651 struct scrub_ctx *sctx = sblock->sctx;
a2de733c 1652 u8 csum[BTRFS_CSUM_SIZE];
b5d67f64
SB		 1653 u8 *on_disk_csum;
1654 struct page *page;
1655 void *buffer;
a2de733c
AJ		 1656 u32 crc = ~(u32)0;
1657 int fail = 0;
b5d67f64
SB		 1658 u64 len;
1659 int index;
a2de733c 1660
b5d67f64 1661 BUG_ON(sblock->page_count < 1);
7a9e9987 1662 if (!sblock->pagev[0]->have_csum)
a2de733c
AJ		 1663 return 0;
1664
7a9e9987
SB		 1665 on_disk_csum = sblock->pagev[0]->csum;
1666 page = sblock->pagev[0]->page;
9613bebb 1667 buffer = kmap_atomic(page);
b5d67f64 1668
d9d181c1 1669 len = sctx->sectorsize;
b5d67f64
SB		 1670 index = 0;
1671 for (;;) {
1672 u64 l = min_t(u64, len, PAGE_SIZE);
1673
b0496686 1674 crc = btrfs_csum_data(buffer, crc, l);
9613bebb 1675 kunmap_atomic(buffer);
b5d67f64
SB		 1676 len -= l;
1677 if (len == 0)
1678 break;
1679 index++;
1680 BUG_ON(index >= sblock->page_count);
7a9e9987
SB		 1681 BUG_ON(!sblock->pagev[index]->page);
1682 page = sblock->pagev[index]->page;
9613bebb 1683 buffer = kmap_atomic(page);
b5d67f64
SB		 1684 }
1685
a2de733c 1686 btrfs_csum_final(crc, csum);
d9d181c1 1687 if (memcmp(csum, on_disk_csum, sctx->csum_size))
a2de733c
AJ		 1688 fail = 1;
1689
a2de733c
AJ		 1690 return fail;
1691}
1692
b5d67f64 1693static int scrub_checksum_tree_block(struct scrub_block *sblock)
a2de733c 1694{
d9d181c1 1695 struct scrub_ctx *sctx = sblock->sctx;
a2de733c 1696 struct btrfs_header *h;
a36cf8b8 1697 struct btrfs_root *root = sctx->dev_root;
a2de733c 1698 struct btrfs_fs_info *fs_info = root->fs_info;
b5d67f64
SB		 1699 u8 calculated_csum[BTRFS_CSUM_SIZE];
1700 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1701 struct page *page;
1702 void *mapped_buffer;
1703 u64 mapped_size;
1704 void *p;
a2de733c
AJ		 1705 u32 crc = ~(u32)0;
1706 int fail = 0;
1707 int crc_fail = 0;
b5d67f64
SB		 1708 u64 len;
1709 int index;
1710
1711 BUG_ON(sblock->page_count < 1);
7a9e9987 1712 page = sblock->pagev[0]->page;
9613bebb 1713 mapped_buffer = kmap_atomic(page);
b5d67f64 1714 h = (struct btrfs_header *)mapped_buffer;
d9d181c1 1715 memcpy(on_disk_csum, h->csum, sctx->csum_size);
a2de733c
AJ		 1716
1717 /*
1718 * we don't use the getter functions here, as we
1719 * a) don't have an extent buffer and
1720 * b) the page is already kmapped
1721 */
a2de733c 1722
7a9e9987 1723 if (sblock->pagev[0]->logical != le64_to_cpu(h->bytenr))
a2de733c
AJ		 1724 ++fail;
1725
7a9e9987 1726 if (sblock->pagev[0]->generation != le64_to_cpu(h->generation))
a2de733c
AJ		 1727 ++fail;
1728
1729 if (memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
1730 ++fail;
1731
1732 if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
1733 BTRFS_UUID_SIZE))
1734 ++fail;
1735
ff023aac 1736 WARN_ON(sctx->nodesize != sctx->leafsize);
d9d181c1 1737 len = sctx->nodesize - BTRFS_CSUM_SIZE;
b5d67f64
SB		 1738 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
1739 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
1740 index = 0;
1741 for (;;) {
1742 u64 l = min_t(u64, len, mapped_size);
1743
b0496686 1744 crc = btrfs_csum_data(p, crc, l);
9613bebb 1745 kunmap_atomic(mapped_buffer);
b5d67f64
SB		 1746 len -= l;
1747 if (len == 0)
1748 break;
1749 index++;
1750 BUG_ON(index >= sblock->page_count);
7a9e9987
SB		 1751 BUG_ON(!sblock->pagev[index]->page);
1752 page = sblock->pagev[index]->page;
9613bebb 1753 mapped_buffer = kmap_atomic(page);
b5d67f64
SB		 1754 mapped_size = PAGE_SIZE;
1755 p = mapped_buffer;
1756 }
1757
1758 btrfs_csum_final(crc, calculated_csum);
d9d181c1 1759 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
a2de733c
AJ		 1760 ++crc_fail;
1761
a2de733c
AJ		 1762 return fail || crc_fail;
1763}
1764
b5d67f64 1765static int scrub_checksum_super(struct scrub_block *sblock)
a2de733c
AJ		 1766{
1767 struct btrfs_super_block *s;
d9d181c1 1768 struct scrub_ctx *sctx = sblock->sctx;
a36cf8b8 1769 struct btrfs_root *root = sctx->dev_root;
a2de733c 1770 struct btrfs_fs_info *fs_info = root->fs_info;
b5d67f64
SB		 1771 u8 calculated_csum[BTRFS_CSUM_SIZE];
1772 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1773 struct page *page;
1774 void *mapped_buffer;
1775 u64 mapped_size;
1776 void *p;
a2de733c 1777 u32 crc = ~(u32)0;
442a4f63
SB		 1778 int fail_gen = 0;
1779 int fail_cor = 0;
b5d67f64
SB		 1780 u64 len;
1781 int index;
a2de733c 1782
b5d67f64 1783 BUG_ON(sblock->page_count < 1);
7a9e9987 1784 page = sblock->pagev[0]->page;
9613bebb 1785 mapped_buffer = kmap_atomic(page);
b5d67f64 1786 s = (struct btrfs_super_block *)mapped_buffer;
d9d181c1 1787 memcpy(on_disk_csum, s->csum, sctx->csum_size);
a2de733c 1788
7a9e9987 1789 if (sblock->pagev[0]->logical != le64_to_cpu(s->bytenr))
442a4f63 1790 ++fail_cor;
a2de733c 1791
7a9e9987 1792 if (sblock->pagev[0]->generation != le64_to_cpu(s->generation))
442a4f63 1793 ++fail_gen;
a2de733c
AJ		 1794
1795 if (memcmp(s->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
442a4f63 1796 ++fail_cor;
a2de733c 1797
b5d67f64
SB		 1798 len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE;
1799 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
1800 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
1801 index = 0;
1802 for (;;) {
1803 u64 l = min_t(u64, len, mapped_size);
1804
b0496686 1805 crc = btrfs_csum_data(p, crc, l);
9613bebb 1806 kunmap_atomic(mapped_buffer);
b5d67f64
SB		 1807 len -= l;
1808 if (len == 0)
1809 break;
1810 index++;
1811 BUG_ON(index >= sblock->page_count);
7a9e9987
SB		 1812 BUG_ON(!sblock->pagev[index]->page);
1813 page = sblock->pagev[index]->page;
9613bebb 1814 mapped_buffer = kmap_atomic(page);
b5d67f64
SB		 1815 mapped_size = PAGE_SIZE;
1816 p = mapped_buffer;
1817 }
1818
1819 btrfs_csum_final(crc, calculated_csum);
d9d181c1 1820 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
442a4f63 1821 ++fail_cor;
a2de733c 1822
442a4f63 1823 if (fail_cor + fail_gen) {
a2de733c
AJ		 1824 /*
1825 * if we find an error in a super block, we just report it.
1826 * They will get written with the next transaction commit
1827 * anyway
1828 */
d9d181c1
SB		 1829 spin_lock(&sctx->stat_lock);
1830 ++sctx->stat.super_errors;
1831 spin_unlock(&sctx->stat_lock);
442a4f63 1832 if (fail_cor)
7a9e9987 1833 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
442a4f63
SB		 1834 BTRFS_DEV_STAT_CORRUPTION_ERRS);
1835 else
7a9e9987 1836 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
442a4f63 1837 BTRFS_DEV_STAT_GENERATION_ERRS);
a2de733c
AJ		 1838 }
1839
442a4f63 1840 return fail_cor + fail_gen;
a2de733c
AJ		 1841}
1842
b5d67f64
SB		 1843static void scrub_block_get(struct scrub_block *sblock)
1844{
1845 atomic_inc(&sblock->ref_count);
1846}
1847
1848static void scrub_block_put(struct scrub_block *sblock)
1849{
1850 if (atomic_dec_and_test(&sblock->ref_count)) {
1851 int i;
1852
1853 for (i = 0; i < sblock->page_count; i++)
7a9e9987 1854 scrub_page_put(sblock->pagev[i]);
b5d67f64
SB		 1855 kfree(sblock);
1856 }
1857}
1858
7a9e9987
SB		 1859static void scrub_page_get(struct scrub_page *spage)
1860{
1861 atomic_inc(&spage->ref_count);
1862}
1863
1864static void scrub_page_put(struct scrub_page *spage)
1865{
1866 if (atomic_dec_and_test(&spage->ref_count)) {
1867 if (spage->page)
1868 __free_page(spage->page);
1869 kfree(spage);
1870 }
1871}
1872
d9d181c1 1873static void scrub_submit(struct scrub_ctx *sctx)
a2de733c
AJ		 1874{
1875 struct scrub_bio *sbio;
1876
d9d181c1 1877 if (sctx->curr == -1)
1623edeb 1878 return;
a2de733c 1879
d9d181c1
SB		 1880 sbio = sctx->bios[sctx->curr];
1881 sctx->curr = -1;
b6bfebc1 1882 scrub_pending_bio_inc(sctx);
a2de733c 1883
ff023aac
SB		 1884 if (!sbio->bio->bi_bdev) {
1885 /*
1886 * this case should not happen. If btrfs_map_block() is
1887 * wrong, it could happen for dev-replace operations on
1888 * missing devices when no mirrors are available, but in
1889 * this case it should already fail the mount.
1890 * This case is handled correctly (but _very_ slowly).
1891 */
1892 printk_ratelimited(KERN_WARNING
1893 "btrfs: scrub_submit(bio bdev == NULL) is unexpected!\n");
1894 bio_endio(sbio->bio, -EIO);
1895 } else {
1896 btrfsic_submit_bio(READ, sbio->bio);
1897 }
a2de733c
AJ		 1898}
1899
ff023aac
SB		 1900static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
1901 struct scrub_page *spage)
a2de733c 1902{
b5d67f64 1903 struct scrub_block *sblock = spage->sblock;
a2de733c 1904 struct scrub_bio *sbio;
69f4cb52 1905 int ret;
a2de733c
AJ		 1906
1907again:
1908 /*
1909 * grab a fresh bio or wait for one to become available
1910 */
d9d181c1
SB		 1911 while (sctx->curr == -1) {
1912 spin_lock(&sctx->list_lock);
1913 sctx->curr = sctx->first_free;
1914 if (sctx->curr != -1) {
1915 sctx->first_free = sctx->bios[sctx->curr]->next_free;
1916 sctx->bios[sctx->curr]->next_free = -1;
1917 sctx->bios[sctx->curr]->page_count = 0;
1918 spin_unlock(&sctx->list_lock);
a2de733c 1919 } else {
d9d181c1
SB		 1920 spin_unlock(&sctx->list_lock);
1921 wait_event(sctx->list_wait, sctx->first_free != -1);
a2de733c
AJ		 1922 }
1923 }
d9d181c1 1924 sbio = sctx->bios[sctx->curr];
b5d67f64 1925 if (sbio->page_count == 0) {
69f4cb52
AJ		 1926 struct bio *bio;
1927
b5d67f64
SB		 1928 sbio->physical = spage->physical;
1929 sbio->logical = spage->logical;
a36cf8b8 1930 sbio->dev = spage->dev;
b5d67f64
SB		 1931 bio = sbio->bio;
1932 if (!bio) {
ff023aac 1933 bio = bio_alloc(GFP_NOFS, sctx->pages_per_rd_bio);
b5d67f64
SB		 1934 if (!bio)
1935 return -ENOMEM;
1936 sbio->bio = bio;
1937 }
69f4cb52
AJ		 1938
1939 bio->bi_private = sbio;
1940 bio->bi_end_io = scrub_bio_end_io;
a36cf8b8
SB		 1941 bio->bi_bdev = sbio->dev->bdev;
1942 bio->bi_sector = sbio->physical >> 9;
69f4cb52 1943 sbio->err = 0;
b5d67f64
SB		 1944 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
1945 spage->physical ||
1946 sbio->logical + sbio->page_count * PAGE_SIZE !=
a36cf8b8
SB		 1947 spage->logical ||
1948 sbio->dev != spage->dev) {
d9d181c1 1949 scrub_submit(sctx);
a2de733c
AJ		 1950 goto again;
1951 }
69f4cb52 1952
b5d67f64
SB		 1953 sbio->pagev[sbio->page_count] = spage;
1954 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
1955 if (ret != PAGE_SIZE) {
1956 if (sbio->page_count < 1) {
1957 bio_put(sbio->bio);
1958 sbio->bio = NULL;
1959 return -EIO;
1960 }
d9d181c1 1961 scrub_submit(sctx);
69f4cb52
AJ		 1962 goto again;
1963 }
1964
ff023aac 1965 scrub_block_get(sblock); /* one for the page added to the bio */
b5d67f64
SB		 1966 atomic_inc(&sblock->outstanding_pages);
1967 sbio->page_count++;
ff023aac 1968 if (sbio->page_count == sctx->pages_per_rd_bio)
d9d181c1 1969 scrub_submit(sctx);
b5d67f64
SB		 1970
1971 return 0;
1972}
1973
d9d181c1 1974static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
a36cf8b8 1975 u64 physical, struct btrfs_device *dev, u64 flags,
ff023aac
SB		 1976 u64 gen, int mirror_num, u8 *csum, int force,
1977 u64 physical_for_dev_replace)
b5d67f64
SB		 1978{
1979 struct scrub_block *sblock;
1980 int index;
1981
1982 sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
1983 if (!sblock) {
d9d181c1
SB		 1984 spin_lock(&sctx->stat_lock);
1985 sctx->stat.malloc_errors++;
1986 spin_unlock(&sctx->stat_lock);
b5d67f64 1987 return -ENOMEM;
a2de733c 1988 }
b5d67f64 1989
7a9e9987
SB		 1990 /* one ref inside this function, plus one for each page added to
1991 * a bio later on */
b5d67f64 1992 atomic_set(&sblock->ref_count, 1);
d9d181c1 1993 sblock->sctx = sctx;
b5d67f64
SB		 1994 sblock->no_io_error_seen = 1;
1995
1996 for (index = 0; len > 0; index++) {
7a9e9987 1997 struct scrub_page *spage;
b5d67f64
SB		 1998 u64 l = min_t(u64, len, PAGE_SIZE);
1999
7a9e9987
SB		 2000 spage = kzalloc(sizeof(*spage), GFP_NOFS);
2001 if (!spage) {
2002leave_nomem:
d9d181c1
SB		 2003 spin_lock(&sctx->stat_lock);
2004 sctx->stat.malloc_errors++;
2005 spin_unlock(&sctx->stat_lock);
7a9e9987 2006 scrub_block_put(sblock);
b5d67f64
SB		 2007 return -ENOMEM;
2008 }
7a9e9987
SB		 2009 BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
2010 scrub_page_get(spage);
2011 sblock->pagev[index] = spage;
b5d67f64 2012 spage->sblock = sblock;
a36cf8b8 2013 spage->dev = dev;
b5d67f64
SB		 2014 spage->flags = flags;
2015 spage->generation = gen;
2016 spage->logical = logical;
2017 spage->physical = physical;
ff023aac 2018 spage->physical_for_dev_replace = physical_for_dev_replace;
b5d67f64
SB		 2019 spage->mirror_num = mirror_num;
2020 if (csum) {
2021 spage->have_csum = 1;
d9d181c1 2022 memcpy(spage->csum, csum, sctx->csum_size);
b5d67f64
SB		 2023 } else {
2024 spage->have_csum = 0;
2025 }
2026 sblock->page_count++;
7a9e9987
SB		 2027 spage->page = alloc_page(GFP_NOFS);
2028 if (!spage->page)
2029 goto leave_nomem;
b5d67f64
SB		 2030 len -= l;
2031 logical += l;
2032 physical += l;
ff023aac 2033 physical_for_dev_replace += l;
b5d67f64
SB		 2034 }
2035
7a9e9987 2036 WARN_ON(sblock->page_count == 0);
b5d67f64 2037 for (index = 0; index < sblock->page_count; index++) {
7a9e9987 2038 struct scrub_page *spage = sblock->pagev[index];
1bc87793
AJ		 2039 int ret;
2040
ff023aac 2041 ret = scrub_add_page_to_rd_bio(sctx, spage);
b5d67f64
SB		 2042 if (ret) {
2043 scrub_block_put(sblock);
1bc87793 2044 return ret;
b5d67f64 2045 }
1bc87793 2046 }
a2de733c 2047
b5d67f64 2048 if (force)
d9d181c1 2049 scrub_submit(sctx);
a2de733c 2050
b5d67f64
SB		 2051 /* last one frees, either here or in bio completion for last page */
2052 scrub_block_put(sblock);
a2de733c
AJ		 2053 return 0;
2054}
2055
b5d67f64
SB		 2056static void scrub_bio_end_io(struct bio *bio, int err)
2057{
2058 struct scrub_bio *sbio = bio->bi_private;
a36cf8b8 2059 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
b5d67f64
SB		 2060
2061 sbio->err = err;
2062 sbio->bio = bio;
2063
2064 btrfs_queue_worker(&fs_info->scrub_workers, &sbio->work);
2065}
2066
2067static void scrub_bio_end_io_worker(struct btrfs_work *work)
2068{
2069 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
d9d181c1 2070 struct scrub_ctx *sctx = sbio->sctx;
b5d67f64
SB		 2071 int i;
2072
ff023aac 2073 BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO);
b5d67f64
SB		 2074 if (sbio->err) {
2075 for (i = 0; i < sbio->page_count; i++) {
2076 struct scrub_page *spage = sbio->pagev[i];
2077
2078 spage->io_error = 1;
2079 spage->sblock->no_io_error_seen = 0;
2080 }
2081 }
2082
2083 /* now complete the scrub_block items that have all pages completed */
2084 for (i = 0; i < sbio->page_count; i++) {
2085 struct scrub_page *spage = sbio->pagev[i];
2086 struct scrub_block *sblock = spage->sblock;
2087
2088 if (atomic_dec_and_test(&sblock->outstanding_pages))
2089 scrub_block_complete(sblock);
2090 scrub_block_put(sblock);
2091 }
2092
b5d67f64
SB		 2093 bio_put(sbio->bio);
2094 sbio->bio = NULL;
d9d181c1
SB		 2095 spin_lock(&sctx->list_lock);
2096 sbio->next_free = sctx->first_free;
2097 sctx->first_free = sbio->index;
2098 spin_unlock(&sctx->list_lock);
ff023aac
SB		 2099
2100 if (sctx->is_dev_replace &&
2101 atomic_read(&sctx->wr_ctx.flush_all_writes)) {
2102 mutex_lock(&sctx->wr_ctx.wr_lock);
2103 scrub_wr_submit(sctx);
2104 mutex_unlock(&sctx->wr_ctx.wr_lock);
2105 }
2106
b6bfebc1 2107 scrub_pending_bio_dec(sctx);
b5d67f64
SB		 2108}
2109
2110static void scrub_block_complete(struct scrub_block *sblock)
2111{
ff023aac 2112 if (!sblock->no_io_error_seen) {
b5d67f64 2113 scrub_handle_errored_block(sblock);
ff023aac
SB		 2114 } else {
2115 /*
2116 * if has checksum error, write via repair mechanism in
2117 * dev replace case, otherwise write here in dev replace
2118 * case.
2119 */
2120 if (!scrub_checksum(sblock) && sblock->sctx->is_dev_replace)
2121 scrub_write_block_to_dev_replace(sblock);
2122 }
b5d67f64
SB		 2123}
2124
d9d181c1 2125static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len,
a2de733c
AJ		 2126 u8 *csum)
2127{
2128 struct btrfs_ordered_sum *sum = NULL;
2129 int ret = 0;
2130 unsigned long i;
2131 unsigned long num_sectors;
a2de733c 2132
d9d181c1
SB		 2133 while (!list_empty(&sctx->csum_list)) {
2134 sum = list_first_entry(&sctx->csum_list,
a2de733c
AJ		 2135 struct btrfs_ordered_sum, list);
2136 if (sum->bytenr > logical)
2137 return 0;
2138 if (sum->bytenr + sum->len > logical)
2139 break;
2140
d9d181c1 2141 ++sctx->stat.csum_discards;
a2de733c
AJ		 2142 list_del(&sum->list);
2143 kfree(sum);
2144 sum = NULL;
2145 }
2146 if (!sum)
2147 return 0;
2148
d9d181c1 2149 num_sectors = sum->len / sctx->sectorsize;
a2de733c
AJ		 2150 for (i = 0; i < num_sectors; ++i) {
2151 if (sum->sums[i].bytenr == logical) {
d9d181c1 2152 memcpy(csum, &sum->sums[i].sum, sctx->csum_size);
a2de733c
AJ		 2153 ret = 1;
2154 break;
2155 }
2156 }
2157 if (ret && i == num_sectors - 1) {
2158 list_del(&sum->list);
2159 kfree(sum);
2160 }
2161 return ret;
2162}
2163
2164/* scrub extent tries to collect up to 64 kB for each bio */
d9d181c1 2165static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len,
a36cf8b8 2166 u64 physical, struct btrfs_device *dev, u64 flags,
ff023aac 2167 u64 gen, int mirror_num, u64 physical_for_dev_replace)
a2de733c
AJ		 2168{
2169 int ret;
2170 u8 csum[BTRFS_CSUM_SIZE];
b5d67f64
SB		 2171 u32 blocksize;
2172
2173 if (flags & BTRFS_EXTENT_FLAG_DATA) {
d9d181c1
SB		 2174 blocksize = sctx->sectorsize;
2175 spin_lock(&sctx->stat_lock);
2176 sctx->stat.data_extents_scrubbed++;
2177 sctx->stat.data_bytes_scrubbed += len;
2178 spin_unlock(&sctx->stat_lock);
b5d67f64 2179 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
ff023aac 2180 WARN_ON(sctx->nodesize != sctx->leafsize);
d9d181c1
SB		 2181 blocksize = sctx->nodesize;
2182 spin_lock(&sctx->stat_lock);
2183 sctx->stat.tree_extents_scrubbed++;
2184 sctx->stat.tree_bytes_scrubbed += len;
2185 spin_unlock(&sctx->stat_lock);
b5d67f64 2186 } else {
d9d181c1 2187 blocksize = sctx->sectorsize;
ff023aac 2188 WARN_ON(1);
b5d67f64 2189 }
a2de733c
AJ		 2190
2191 while (len) {
b5d67f64 2192 u64 l = min_t(u64, len, blocksize);
a2de733c
AJ		 2193 int have_csum = 0;
2194
2195 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2196 /* push csums to sbio */
d9d181c1 2197 have_csum = scrub_find_csum(sctx, logical, l, csum);
a2de733c 2198 if (have_csum == 0)
d9d181c1 2199 ++sctx->stat.no_csum;
ff023aac
SB		 2200 if (sctx->is_dev_replace && !have_csum) {
2201 ret = copy_nocow_pages(sctx, logical, l,
2202 mirror_num,
2203 physical_for_dev_replace);
2204 goto behind_scrub_pages;
2205 }
a2de733c 2206 }
a36cf8b8 2207 ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen,
ff023aac
SB		 2208 mirror_num, have_csum ? csum : NULL, 0,
2209 physical_for_dev_replace);
2210behind_scrub_pages:
a2de733c
AJ		 2211 if (ret)
2212 return ret;
2213 len -= l;
2214 logical += l;
2215 physical += l;
ff023aac 2216 physical_for_dev_replace += l;
a2de733c
AJ		 2217 }
2218 return 0;
2219}
2220
d9d181c1 2221static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
a36cf8b8
SB		 2222 struct map_lookup *map,
2223 struct btrfs_device *scrub_dev,
ff023aac
SB		 2224 int num, u64 base, u64 length,
2225 int is_dev_replace)
a2de733c
AJ		 2226{
2227 struct btrfs_path *path;
a36cf8b8 2228 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
a2de733c
AJ		 2229 struct btrfs_root *root = fs_info->extent_root;
2230 struct btrfs_root *csum_root = fs_info->csum_root;
2231 struct btrfs_extent_item *extent;
e7786c3a 2232 struct blk_plug plug;
a2de733c
AJ		 2233 u64 flags;
2234 int ret;
2235 int slot;
2236 int i;
2237 u64 nstripes;
a2de733c
AJ		 2238 struct extent_buffer *l;
2239 struct btrfs_key key;
2240 u64 physical;
2241 u64 logical;
2242 u64 generation;
e12fa9cd 2243 int mirror_num;
7a26285e
AJ		 2244 struct reada_control *reada1;
2245 struct reada_control *reada2;
2246 struct btrfs_key key_start;
2247 struct btrfs_key key_end;
a2de733c
AJ		 2248 u64 increment = map->stripe_len;
2249 u64 offset;
ff023aac
SB		 2250 u64 extent_logical;
2251 u64 extent_physical;
2252 u64 extent_len;
2253 struct btrfs_device *extent_dev;
2254 int extent_mirror_num;
a2de733c 2255
53b381b3
DW		 2256 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
2257 BTRFS_BLOCK_GROUP_RAID6)) {
2258 if (num >= nr_data_stripes(map)) {
2259 return 0;
2260 }
2261 }
2262
a2de733c
AJ		 2263 nstripes = length;
2264 offset = 0;
2265 do_div(nstripes, map->stripe_len);
2266 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2267 offset = map->stripe_len * num;
2268 increment = map->stripe_len * map->num_stripes;
193ea74b 2269 mirror_num = 1;
a2de733c
AJ		 2270 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2271 int factor = map->num_stripes / map->sub_stripes;
2272 offset = map->stripe_len * (num / map->sub_stripes);
2273 increment = map->stripe_len * factor;
193ea74b 2274 mirror_num = num % map->sub_stripes + 1;
a2de733c
AJ		 2275 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2276 increment = map->stripe_len;
193ea74b 2277 mirror_num = num % map->num_stripes + 1;
a2de733c
AJ		 2278 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2279 increment = map->stripe_len;
193ea74b 2280 mirror_num = num % map->num_stripes + 1;
a2de733c
AJ		 2281 } else {
2282 increment = map->stripe_len;
193ea74b 2283 mirror_num = 1;
a2de733c
AJ		 2284 }
2285
2286 path = btrfs_alloc_path();
2287 if (!path)
2288 return -ENOMEM;
2289
b5d67f64
SB		 2290 /*
2291 * work on commit root. The related disk blocks are static as
2292 * long as COW is applied. This means, it is save to rewrite
2293 * them to repair disk errors without any race conditions
2294 */
a2de733c
AJ		 2295 path->search_commit_root = 1;
2296 path->skip_locking = 1;
2297
2298 /*
7a26285e
AJ		 2299 * trigger the readahead for extent tree csum tree and wait for
2300 * completion. During readahead, the scrub is officially paused
2301 * to not hold off transaction commits
a2de733c
AJ		 2302 */
2303 logical = base + offset;
a2de733c 2304
d9d181c1 2305 wait_event(sctx->list_wait,
b6bfebc1 2306 atomic_read(&sctx->bios_in_flight) == 0);
7a26285e
AJ		 2307 atomic_inc(&fs_info->scrubs_paused);
2308 wake_up(&fs_info->scrub_pause_wait);
2309
2310 /* FIXME it might be better to start readahead at commit root */
2311 key_start.objectid = logical;
2312 key_start.type = BTRFS_EXTENT_ITEM_KEY;
2313 key_start.offset = (u64)0;
2314 key_end.objectid = base + offset + nstripes * increment;
3173a18f
JB		 2315 key_end.type = BTRFS_METADATA_ITEM_KEY;
2316 key_end.offset = (u64)-1;
7a26285e
AJ		 2317 reada1 = btrfs_reada_add(root, &key_start, &key_end);
2318
2319 key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
2320 key_start.type = BTRFS_EXTENT_CSUM_KEY;
2321 key_start.offset = logical;
2322 key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
2323 key_end.type = BTRFS_EXTENT_CSUM_KEY;
2324 key_end.offset = base + offset + nstripes * increment;
2325 reada2 = btrfs_reada_add(csum_root, &key_start, &key_end);
2326
2327 if (!IS_ERR(reada1))
2328 btrfs_reada_wait(reada1);
2329 if (!IS_ERR(reada2))
2330 btrfs_reada_wait(reada2);
2331
2332 mutex_lock(&fs_info->scrub_lock);
2333 while (atomic_read(&fs_info->scrub_pause_req)) {
2334 mutex_unlock(&fs_info->scrub_lock);
2335 wait_event(fs_info->scrub_pause_wait,
2336 atomic_read(&fs_info->scrub_pause_req) == 0);
2337 mutex_lock(&fs_info->scrub_lock);
a2de733c 2338 }
7a26285e
AJ		 2339 atomic_dec(&fs_info->scrubs_paused);
2340 mutex_unlock(&fs_info->scrub_lock);
2341 wake_up(&fs_info->scrub_pause_wait);
a2de733c
AJ		 2342
2343 /*
2344 * collect all data csums for the stripe to avoid seeking during
2345 * the scrub. This might currently (crc32) end up to be about 1MB
2346 */
e7786c3a 2347 blk_start_plug(&plug);
a2de733c 2348
a2de733c
AJ		 2349 /*
2350 * now find all extents for each stripe and scrub them
2351 */
7a26285e
AJ		 2352 logical = base + offset;
2353 physical = map->stripes[num].physical;
a2de733c 2354 ret = 0;
7a26285e 2355 for (i = 0; i < nstripes; ++i) {
a2de733c
AJ		 2356 /*
2357 * canceled?
2358 */
2359 if (atomic_read(&fs_info->scrub_cancel_req) ||
d9d181c1 2360 atomic_read(&sctx->cancel_req)) {
a2de733c
AJ		 2361 ret = -ECANCELED;
2362 goto out;
2363 }
2364 /*
2365 * check to see if we have to pause
2366 */
2367 if (atomic_read(&fs_info->scrub_pause_req)) {
2368 /* push queued extents */
ff023aac 2369 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
d9d181c1 2370 scrub_submit(sctx);
ff023aac
SB		 2371 mutex_lock(&sctx->wr_ctx.wr_lock);
2372 scrub_wr_submit(sctx);
2373 mutex_unlock(&sctx->wr_ctx.wr_lock);
d9d181c1 2374 wait_event(sctx->list_wait,
b6bfebc1 2375 atomic_read(&sctx->bios_in_flight) == 0);
ff023aac 2376 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
a2de733c
AJ		 2377 atomic_inc(&fs_info->scrubs_paused);
2378 wake_up(&fs_info->scrub_pause_wait);
2379 mutex_lock(&fs_info->scrub_lock);
2380 while (atomic_read(&fs_info->scrub_pause_req)) {
2381 mutex_unlock(&fs_info->scrub_lock);
2382 wait_event(fs_info->scrub_pause_wait,
2383 atomic_read(&fs_info->scrub_pause_req) == 0);
2384 mutex_lock(&fs_info->scrub_lock);
2385 }
2386 atomic_dec(&fs_info->scrubs_paused);
2387 mutex_unlock(&fs_info->scrub_lock);
2388 wake_up(&fs_info->scrub_pause_wait);
a2de733c
AJ		 2389 }
2390
7a26285e
AJ		 2391 ret = btrfs_lookup_csums_range(csum_root, logical,
2392 logical + map->stripe_len - 1,
d9d181c1 2393 &sctx->csum_list, 1);
7a26285e
AJ		 2394 if (ret)
2395 goto out;
2396
a2de733c
AJ		 2397 key.objectid = logical;
2398 key.type = BTRFS_EXTENT_ITEM_KEY;
2399 key.offset = (u64)0;
2400
2401 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2402 if (ret < 0)
2403 goto out;
3173a18f 2404
8c51032f 2405 if (ret > 0) {
a2de733c
AJ		 2406 ret = btrfs_previous_item(root, path, 0,
2407 BTRFS_EXTENT_ITEM_KEY);
2408 if (ret < 0)
2409 goto out;
8c51032f
AJ		 2410 if (ret > 0) {
2411 /* there's no smaller item, so stick with the
2412 * larger one */
2413 btrfs_release_path(path);
2414 ret = btrfs_search_slot(NULL, root, &key,
2415 path, 0, 0);
2416 if (ret < 0)
2417 goto out;
2418 }
a2de733c
AJ		 2419 }
2420
2421 while (1) {
3173a18f
JB		 2422 u64 bytes;
2423
a2de733c
AJ		 2424 l = path->nodes[0];
2425 slot = path->slots[0];
2426 if (slot >= btrfs_header_nritems(l)) {
2427 ret = btrfs_next_leaf(root, path);
2428 if (ret == 0)
2429 continue;
2430 if (ret < 0)
2431 goto out;
2432
2433 break;
2434 }
2435 btrfs_item_key_to_cpu(l, &key, slot);
2436
3173a18f
JB		 2437 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
2438 key.type != BTRFS_METADATA_ITEM_KEY)
2439 goto next;
2440
2441 if (key.type == BTRFS_METADATA_ITEM_KEY)
2442 bytes = root->leafsize;
2443 else
2444 bytes = key.offset;
2445
2446 if (key.objectid + bytes <= logical)
a2de733c
AJ		 2447 goto next;
2448
2449 if (key.objectid >= logical + map->stripe_len)
2450 break;
2451
a2de733c
AJ		 2452
2453 extent = btrfs_item_ptr(l, slot,
2454 struct btrfs_extent_item);
2455 flags = btrfs_extent_flags(l, extent);
2456 generation = btrfs_extent_generation(l, extent);
2457
2458 if (key.objectid < logical &&
2459 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
2460 printk(KERN_ERR
2461 "btrfs scrub: tree block %llu spanning "
2462 "stripes, ignored. logical=%llu\n",
2463 (unsigned long long)key.objectid,
2464 (unsigned long long)logical);
2465 goto next;
2466 }
2467
2468 /*
2469 * trim extent to this stripe
2470 */
2471 if (key.objectid < logical) {
3173a18f 2472 bytes -= logical - key.objectid;
a2de733c
AJ		 2473 key.objectid = logical;
2474 }
3173a18f 2475 if (key.objectid + bytes >
a2de733c 2476 logical + map->stripe_len) {
3173a18f
JB		 2477 bytes = logical + map->stripe_len -
2478 key.objectid;
a2de733c
AJ		 2479 }
2480
ff023aac
SB		 2481 extent_logical = key.objectid;
2482 extent_physical = key.objectid - logical + physical;
3173a18f 2483 extent_len = bytes;
ff023aac
SB		 2484 extent_dev = scrub_dev;
2485 extent_mirror_num = mirror_num;
2486 if (is_dev_replace)
2487 scrub_remap_extent(fs_info, extent_logical,
2488 extent_len, &extent_physical,
2489 &extent_dev,
2490 &extent_mirror_num);
2491 ret = scrub_extent(sctx, extent_logical, extent_len,
2492 extent_physical, extent_dev, flags,
2493 generation, extent_mirror_num,
2494 key.objectid - logical + physical);
a2de733c
AJ		 2495 if (ret)
2496 goto out;
2497
2498next:
2499 path->slots[0]++;
2500 }
71267333 2501 btrfs_release_path(path);
a2de733c
AJ		 2502 logical += increment;
2503 physical += map->stripe_len;
d9d181c1
SB		 2504 spin_lock(&sctx->stat_lock);
2505 sctx->stat.last_physical = physical;
2506 spin_unlock(&sctx->stat_lock);
a2de733c 2507 }
ff023aac 2508out:
a2de733c 2509 /* push queued extents */
d9d181c1 2510 scrub_submit(sctx);
ff023aac
SB		 2511 mutex_lock(&sctx->wr_ctx.wr_lock);
2512 scrub_wr_submit(sctx);
2513 mutex_unlock(&sctx->wr_ctx.wr_lock);
a2de733c 2514
e7786c3a 2515 blk_finish_plug(&plug);
a2de733c
AJ		 2516 btrfs_free_path(path);
2517 return ret < 0 ? ret : 0;
2518}
2519
d9d181c1 2520static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx,
a36cf8b8
SB		 2521 struct btrfs_device *scrub_dev,
2522 u64 chunk_tree, u64 chunk_objectid,
2523 u64 chunk_offset, u64 length,
ff023aac 2524 u64 dev_offset, int is_dev_replace)
a2de733c
AJ		 2525{
2526 struct btrfs_mapping_tree *map_tree =
a36cf8b8 2527 &sctx->dev_root->fs_info->mapping_tree;
a2de733c
AJ		 2528 struct map_lookup *map;
2529 struct extent_map *em;
2530 int i;
ff023aac 2531 int ret = 0;
a2de733c
AJ		 2532
2533 read_lock(&map_tree->map_tree.lock);
2534 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2535 read_unlock(&map_tree->map_tree.lock);
2536
2537 if (!em)
2538 return -EINVAL;
2539
2540 map = (struct map_lookup *)em->bdev;
2541 if (em->start != chunk_offset)
2542 goto out;
2543
2544 if (em->len < length)
2545 goto out;
2546
2547 for (i = 0; i < map->num_stripes; ++i) {
a36cf8b8 2548 if (map->stripes[i].dev->bdev == scrub_dev->bdev &&
859acaf1 2549 map->stripes[i].physical == dev_offset) {
a36cf8b8 2550 ret = scrub_stripe(sctx, map, scrub_dev, i,
ff023aac
SB		 2551 chunk_offset, length,
2552 is_dev_replace);
a2de733c
AJ		 2553 if (ret)
2554 goto out;
2555 }
2556 }
2557out:
2558 free_extent_map(em);
2559
2560 return ret;
2561}
2562
2563static noinline_for_stack
a36cf8b8 2564int scrub_enumerate_chunks(struct scrub_ctx *sctx,
ff023aac
SB		 2565 struct btrfs_device *scrub_dev, u64 start, u64 end,
2566 int is_dev_replace)
a2de733c
AJ		 2567{
2568 struct btrfs_dev_extent *dev_extent = NULL;
2569 struct btrfs_path *path;
a36cf8b8 2570 struct btrfs_root *root = sctx->dev_root;
a2de733c
AJ		 2571 struct btrfs_fs_info *fs_info = root->fs_info;
2572 u64 length;
2573 u64 chunk_tree;
2574 u64 chunk_objectid;
2575 u64 chunk_offset;
2576 int ret;
2577 int slot;
2578 struct extent_buffer *l;
2579 struct btrfs_key key;
2580 struct btrfs_key found_key;
2581 struct btrfs_block_group_cache *cache;
ff023aac 2582 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
a2de733c
AJ		 2583
2584 path = btrfs_alloc_path();
2585 if (!path)
2586 return -ENOMEM;
2587
2588 path->reada = 2;
2589 path->search_commit_root = 1;
2590 path->skip_locking = 1;
2591
a36cf8b8 2592 key.objectid = scrub_dev->devid;
a2de733c
AJ		 2593 key.offset = 0ull;
2594 key.type = BTRFS_DEV_EXTENT_KEY;
2595
a2de733c
AJ		 2596 while (1) {
2597 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2598 if (ret < 0)
8c51032f
AJ		 2599 break;
2600 if (ret > 0) {
2601 if (path->slots[0] >=
2602 btrfs_header_nritems(path->nodes[0])) {
2603 ret = btrfs_next_leaf(root, path);
2604 if (ret)
2605 break;
2606 }
2607 }
a2de733c
AJ		 2608
2609 l = path->nodes[0];
2610 slot = path->slots[0];
2611
2612 btrfs_item_key_to_cpu(l, &found_key, slot);
2613
a36cf8b8 2614 if (found_key.objectid != scrub_dev->devid)
a2de733c
AJ		 2615 break;
2616
8c51032f 2617 if (btrfs_key_type(&found_key) != BTRFS_DEV_EXTENT_KEY)
a2de733c
AJ		 2618 break;
2619
2620 if (found_key.offset >= end)
2621 break;
2622
2623 if (found_key.offset < key.offset)
2624 break;
2625
2626 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2627 length = btrfs_dev_extent_length(l, dev_extent);
2628
2629 if (found_key.offset + length <= start) {
2630 key.offset = found_key.offset + length;
71267333 2631 btrfs_release_path(path);
a2de733c
AJ		 2632 continue;
2633 }
2634
2635 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2636 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2637 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2638
2639 /*
2640 * get a reference on the corresponding block group to prevent
2641 * the chunk from going away while we scrub it
2642 */
2643 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2644 if (!cache) {
2645 ret = -ENOENT;
8c51032f 2646 break;
a2de733c 2647 }
ff023aac
SB		 2648 dev_replace->cursor_right = found_key.offset + length;
2649 dev_replace->cursor_left = found_key.offset;
2650 dev_replace->item_needs_writeback = 1;
a36cf8b8 2651 ret = scrub_chunk(sctx, scrub_dev, chunk_tree, chunk_objectid,
ff023aac
SB		 2652 chunk_offset, length, found_key.offset,
2653 is_dev_replace);
2654
2655 /*
2656 * flush, submit all pending read and write bios, afterwards
2657 * wait for them.
2658 * Note that in the dev replace case, a read request causes
2659 * write requests that are submitted in the read completion
2660 * worker. Therefore in the current situation, it is required
2661 * that all write requests are flushed, so that all read and
2662 * write requests are really completed when bios_in_flight
2663 * changes to 0.
2664 */
2665 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
2666 scrub_submit(sctx);
2667 mutex_lock(&sctx->wr_ctx.wr_lock);
2668 scrub_wr_submit(sctx);
2669 mutex_unlock(&sctx->wr_ctx.wr_lock);
2670
2671 wait_event(sctx->list_wait,
2672 atomic_read(&sctx->bios_in_flight) == 0);
2673 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
2674 atomic_inc(&fs_info->scrubs_paused);
2675 wake_up(&fs_info->scrub_pause_wait);
2676 wait_event(sctx->list_wait,
2677 atomic_read(&sctx->workers_pending) == 0);
2678
2679 mutex_lock(&fs_info->scrub_lock);
2680 while (atomic_read(&fs_info->scrub_pause_req)) {
2681 mutex_unlock(&fs_info->scrub_lock);
2682 wait_event(fs_info->scrub_pause_wait,
2683 atomic_read(&fs_info->scrub_pause_req) == 0);
2684 mutex_lock(&fs_info->scrub_lock);
2685 }
2686 atomic_dec(&fs_info->scrubs_paused);
2687 mutex_unlock(&fs_info->scrub_lock);
2688 wake_up(&fs_info->scrub_pause_wait);
2689
2690 dev_replace->cursor_left = dev_replace->cursor_right;
2691 dev_replace->item_needs_writeback = 1;
a2de733c
AJ		 2692 btrfs_put_block_group(cache);
2693 if (ret)
2694 break;
af1be4f8
SB		 2695 if (is_dev_replace &&
2696 atomic64_read(&dev_replace->num_write_errors) > 0) {
ff023aac
SB		 2697 ret = -EIO;
2698 break;
2699 }
2700 if (sctx->stat.malloc_errors > 0) {
2701 ret = -ENOMEM;
2702 break;
2703 }
a2de733c
AJ		 2704
2705 key.offset = found_key.offset + length;
71267333 2706 btrfs_release_path(path);
a2de733c
AJ		 2707 }
2708
a2de733c 2709 btrfs_free_path(path);
8c51032f
AJ		 2710
2711 /*
2712 * ret can still be 1 from search_slot or next_leaf,
2713 * that's not an error
2714 */
2715 return ret < 0 ? ret : 0;
a2de733c
AJ		 2716}
2717
a36cf8b8
SB		 2718static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx,
2719 struct btrfs_device *scrub_dev)
a2de733c
AJ		 2720{
2721 int i;
2722 u64 bytenr;
2723 u64 gen;
2724 int ret;
a36cf8b8 2725 struct btrfs_root *root = sctx->dev_root;
a2de733c 2726
87533c47 2727 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
79787eaa
JM		 2728 return -EIO;
2729
a2de733c
AJ		 2730 gen = root->fs_info->last_trans_committed;
2731
2732 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2733 bytenr = btrfs_sb_offset(i);
a36cf8b8 2734 if (bytenr + BTRFS_SUPER_INFO_SIZE > scrub_dev->total_bytes)
a2de733c
AJ		 2735 break;
2736
d9d181c1 2737 ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr,
a36cf8b8 2738 scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i,
ff023aac 2739 NULL, 1, bytenr);
a2de733c
AJ		 2740 if (ret)
2741 return ret;
2742 }
b6bfebc1 2743 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
a2de733c
AJ		 2744
2745 return 0;
2746}
2747
2748/*
2749 * get a reference count on fs_info->scrub_workers. start worker if necessary
2750 */
ff023aac
SB		 2751static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info,
2752 int is_dev_replace)
a2de733c 2753{
0dc3b84a 2754 int ret = 0;
a2de733c
AJ		 2755
2756 mutex_lock(&fs_info->scrub_lock);
632dd772 2757 if (fs_info->scrub_workers_refcnt == 0) {
ff023aac
SB		 2758 if (is_dev_replace)
2759 btrfs_init_workers(&fs_info->scrub_workers, "scrub", 1,
2760 &fs_info->generic_worker);
2761 else
2762 btrfs_init_workers(&fs_info->scrub_workers, "scrub",
2763 fs_info->thread_pool_size,
2764 &fs_info->generic_worker);
632dd772 2765 fs_info->scrub_workers.idle_thresh = 4;
0dc3b84a
JB		 2766 ret = btrfs_start_workers(&fs_info->scrub_workers);
2767 if (ret)
2768 goto out;
ff023aac
SB		 2769 btrfs_init_workers(&fs_info->scrub_wr_completion_workers,
2770 "scrubwrc",
2771 fs_info->thread_pool_size,
2772 &fs_info->generic_worker);
2773 fs_info->scrub_wr_completion_workers.idle_thresh = 2;
2774 ret = btrfs_start_workers(
2775 &fs_info->scrub_wr_completion_workers);
2776 if (ret)
2777 goto out;
2778 btrfs_init_workers(&fs_info->scrub_nocow_workers, "scrubnc", 1,
2779 &fs_info->generic_worker);
2780 ret = btrfs_start_workers(&fs_info->scrub_nocow_workers);
2781 if (ret)
2782 goto out;
632dd772 2783 }
a2de733c 2784 ++fs_info->scrub_workers_refcnt;
0dc3b84a 2785out:
a2de733c
AJ		 2786 mutex_unlock(&fs_info->scrub_lock);
2787
0dc3b84a 2788 return ret;
a2de733c
AJ		 2789}
2790
aa1b8cd4 2791static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info)
a2de733c 2792{
a2de733c 2793 mutex_lock(&fs_info->scrub_lock);
ff023aac 2794 if (--fs_info->scrub_workers_refcnt == 0) {
a2de733c 2795 btrfs_stop_workers(&fs_info->scrub_workers);
ff023aac
SB		 2796 btrfs_stop_workers(&fs_info->scrub_wr_completion_workers);
2797 btrfs_stop_workers(&fs_info->scrub_nocow_workers);
2798 }
a2de733c
AJ		 2799 WARN_ON(fs_info->scrub_workers_refcnt < 0);
2800 mutex_unlock(&fs_info->scrub_lock);
2801}
2802
aa1b8cd4
SB		 2803int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start,
2804 u64 end, struct btrfs_scrub_progress *progress,
63a212ab 2805 int readonly, int is_dev_replace)
a2de733c 2806{
d9d181c1 2807 struct scrub_ctx *sctx;
a2de733c
AJ		 2808 int ret;
2809 struct btrfs_device *dev;
2810
aa1b8cd4 2811 if (btrfs_fs_closing(fs_info))
a2de733c
AJ		 2812 return -EINVAL;
2813
2814 /*
2815 * check some assumptions
2816 */
aa1b8cd4 2817 if (fs_info->chunk_root->nodesize != fs_info->chunk_root->leafsize) {
b5d67f64
SB		 2818 printk(KERN_ERR
2819 "btrfs_scrub: size assumption nodesize == leafsize (%d == %d) fails\n",
aa1b8cd4
SB		 2820 fs_info->chunk_root->nodesize,
2821 fs_info->chunk_root->leafsize);
b5d67f64
SB		 2822 return -EINVAL;
2823 }
2824
aa1b8cd4 2825 if (fs_info->chunk_root->nodesize > BTRFS_STRIPE_LEN) {
b5d67f64
SB		 2826 /*
2827 * in this case scrub is unable to calculate the checksum
2828 * the way scrub is implemented. Do not handle this
2829 * situation at all because it won't ever happen.
2830 */
2831 printk(KERN_ERR
2832 "btrfs_scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails\n",
aa1b8cd4 2833 fs_info->chunk_root->nodesize, BTRFS_STRIPE_LEN);
b5d67f64
SB		 2834 return -EINVAL;
2835 }
2836
aa1b8cd4 2837 if (fs_info->chunk_root->sectorsize != PAGE_SIZE) {
b5d67f64
SB		 2838 /* not supported for data w/o checksums */
2839 printk(KERN_ERR
2840 "btrfs_scrub: size assumption sectorsize != PAGE_SIZE (%d != %lld) fails\n",
aa1b8cd4
SB		 2841 fs_info->chunk_root->sectorsize,
2842 (unsigned long long)PAGE_SIZE);
a2de733c
AJ		 2843 return -EINVAL;
2844 }
2845
7a9e9987
SB		 2846 if (fs_info->chunk_root->nodesize >
2847 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK ||
2848 fs_info->chunk_root->sectorsize >
2849 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) {
2850 /*
2851 * would exhaust the array bounds of pagev member in
2852 * struct scrub_block
2853 */
2854 pr_err("btrfs_scrub: size assumption nodesize and sectorsize <= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails\n",
2855 fs_info->chunk_root->nodesize,
2856 SCRUB_MAX_PAGES_PER_BLOCK,
2857 fs_info->chunk_root->sectorsize,
2858 SCRUB_MAX_PAGES_PER_BLOCK);
2859 return -EINVAL;
2860 }
2861
ff023aac 2862 ret = scrub_workers_get(fs_info, is_dev_replace);
a2de733c
AJ		 2863 if (ret)
2864 return ret;
2865
aa1b8cd4
SB		 2866 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2867 dev = btrfs_find_device(fs_info, devid, NULL, NULL);
63a212ab 2868 if (!dev || (dev->missing && !is_dev_replace)) {
aa1b8cd4
SB		 2869 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2870 scrub_workers_put(fs_info);
a2de733c
AJ		 2871 return -ENODEV;
2872 }
2873 mutex_lock(&fs_info->scrub_lock);
2874
63a212ab 2875 if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) {
a2de733c 2876 mutex_unlock(&fs_info->scrub_lock);
aa1b8cd4
SB		 2877 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2878 scrub_workers_put(fs_info);
2879 return -EIO;
a2de733c
AJ		 2880 }
2881
8dabb742
SB		 2882 btrfs_dev_replace_lock(&fs_info->dev_replace);
2883 if (dev->scrub_device ||
2884 (!is_dev_replace &&
2885 btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) {
2886 btrfs_dev_replace_unlock(&fs_info->dev_replace);
a2de733c 2887 mutex_unlock(&fs_info->scrub_lock);
aa1b8cd4
SB		 2888 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2889 scrub_workers_put(fs_info);
a2de733c
AJ		 2890 return -EINPROGRESS;
2891 }
8dabb742 2892 btrfs_dev_replace_unlock(&fs_info->dev_replace);
63a212ab 2893 sctx = scrub_setup_ctx(dev, is_dev_replace);
d9d181c1 2894 if (IS_ERR(sctx)) {
a2de733c 2895 mutex_unlock(&fs_info->scrub_lock);
aa1b8cd4
SB		 2896 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2897 scrub_workers_put(fs_info);
d9d181c1 2898 return PTR_ERR(sctx);
a2de733c 2899 }
d9d181c1
SB		 2900 sctx->readonly = readonly;
2901 dev->scrub_device = sctx;
a2de733c
AJ		 2902
2903 atomic_inc(&fs_info->scrubs_running);
2904 mutex_unlock(&fs_info->scrub_lock);
aa1b8cd4 2905 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
a2de733c 2906
ff023aac
SB		 2907 if (!is_dev_replace) {
2908 down_read(&fs_info->scrub_super_lock);
2909 ret = scrub_supers(sctx, dev);
2910 up_read(&fs_info->scrub_super_lock);
2911 }
a2de733c
AJ		 2912
2913 if (!ret)
ff023aac
SB		 2914 ret = scrub_enumerate_chunks(sctx, dev, start, end,
2915 is_dev_replace);
a2de733c 2916
b6bfebc1 2917 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
a2de733c
AJ		 2918 atomic_dec(&fs_info->scrubs_running);
2919 wake_up(&fs_info->scrub_pause_wait);
2920
b6bfebc1 2921 wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0);
0ef8e451 2922
a2de733c 2923 if (progress)
d9d181c1 2924 memcpy(progress, &sctx->stat, sizeof(*progress));
a2de733c
AJ		 2925
2926 mutex_lock(&fs_info->scrub_lock);
2927 dev->scrub_device = NULL;
2928 mutex_unlock(&fs_info->scrub_lock);
2929
d9d181c1 2930 scrub_free_ctx(sctx);
aa1b8cd4 2931 scrub_workers_put(fs_info);
a2de733c
AJ		 2932
2933 return ret;
2934}
2935
143bede5 2936void btrfs_scrub_pause(struct btrfs_root *root)
a2de733c
AJ		 2937{
2938 struct btrfs_fs_info *fs_info = root->fs_info;
2939
2940 mutex_lock(&fs_info->scrub_lock);
2941 atomic_inc(&fs_info->scrub_pause_req);
2942 while (atomic_read(&fs_info->scrubs_paused) !=
2943 atomic_read(&fs_info->scrubs_running)) {
2944 mutex_unlock(&fs_info->scrub_lock);
2945 wait_event(fs_info->scrub_pause_wait,
2946 atomic_read(&fs_info->scrubs_paused) ==
2947 atomic_read(&fs_info->scrubs_running));
2948 mutex_lock(&fs_info->scrub_lock);
2949 }
2950 mutex_unlock(&fs_info->scrub_lock);
a2de733c
AJ		 2951}
2952
143bede5 2953void btrfs_scrub_continue(struct btrfs_root *root)
a2de733c
AJ		 2954{
2955 struct btrfs_fs_info *fs_info = root->fs_info;
2956
2957 atomic_dec(&fs_info->scrub_pause_req);
2958 wake_up(&fs_info->scrub_pause_wait);
a2de733c
AJ		 2959}
2960
143bede5 2961void btrfs_scrub_pause_super(struct btrfs_root *root)
a2de733c
AJ		 2962{
2963 down_write(&root->fs_info->scrub_super_lock);
a2de733c
AJ		 2964}
2965
143bede5 2966void btrfs_scrub_continue_super(struct btrfs_root *root)
a2de733c
AJ		 2967{
2968 up_write(&root->fs_info->scrub_super_lock);
a2de733c
AJ		 2969}
2970
aa1b8cd4 2971int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info)
a2de733c 2972{
a2de733c
AJ		 2973 mutex_lock(&fs_info->scrub_lock);
2974 if (!atomic_read(&fs_info->scrubs_running)) {
2975 mutex_unlock(&fs_info->scrub_lock);
2976 return -ENOTCONN;
2977 }
2978
2979 atomic_inc(&fs_info->scrub_cancel_req);
2980 while (atomic_read(&fs_info->scrubs_running)) {
2981 mutex_unlock(&fs_info->scrub_lock);
2982 wait_event(fs_info->scrub_pause_wait,
2983 atomic_read(&fs_info->scrubs_running) == 0);
2984 mutex_lock(&fs_info->scrub_lock);
2985 }
2986 atomic_dec(&fs_info->scrub_cancel_req);
2987 mutex_unlock(&fs_info->scrub_lock);
2988
2989 return 0;
2990}
2991
aa1b8cd4
SB		 2992int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info,
2993 struct btrfs_device *dev)
49b25e05 2994{
d9d181c1 2995 struct scrub_ctx *sctx;
a2de733c
AJ		 2996
2997 mutex_lock(&fs_info->scrub_lock);
d9d181c1
SB		 2998 sctx = dev->scrub_device;
2999 if (!sctx) {
a2de733c
AJ		 3000 mutex_unlock(&fs_info->scrub_lock);
3001 return -ENOTCONN;
3002 }
d9d181c1 3003 atomic_inc(&sctx->cancel_req);
a2de733c
AJ		 3004 while (dev->scrub_device) {
3005 mutex_unlock(&fs_info->scrub_lock);
3006 wait_event(fs_info->scrub_pause_wait,
3007 dev->scrub_device == NULL);
3008 mutex_lock(&fs_info->scrub_lock);
3009 }
3010 mutex_unlock(&fs_info->scrub_lock);
3011
3012 return 0;
3013}
1623edeb 3014
a2de733c
AJ		 3015int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
3016 struct btrfs_scrub_progress *progress)
3017{
3018 struct btrfs_device *dev;
d9d181c1 3019 struct scrub_ctx *sctx = NULL;
a2de733c
AJ		 3020
3021 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
aa1b8cd4 3022 dev = btrfs_find_device(root->fs_info, devid, NULL, NULL);
a2de733c 3023 if (dev)
d9d181c1
SB		 3024 sctx = dev->scrub_device;
3025 if (sctx)
3026 memcpy(progress, &sctx->stat, sizeof(*progress));
a2de733c
AJ		 3027 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3028
d9d181c1 3029 return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV;
a2de733c 3030}
ff023aac
SB		 3031
3032static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
3033 u64 extent_logical, u64 extent_len,
3034 u64 *extent_physical,
3035 struct btrfs_device **extent_dev,
3036 int *extent_mirror_num)
3037{
3038 u64 mapped_length;
3039 struct btrfs_bio *bbio = NULL;
3040 int ret;
3041
3042 mapped_length = extent_len;
3043 ret = btrfs_map_block(fs_info, READ, extent_logical,
3044 &mapped_length, &bbio, 0);
3045 if (ret || !bbio || mapped_length < extent_len ||
3046 !bbio->stripes[0].dev->bdev) {
3047 kfree(bbio);
3048 return;
3049 }
3050
3051 *extent_physical = bbio->stripes[0].physical;
3052 *extent_mirror_num = bbio->mirror_num;
3053 *extent_dev = bbio->stripes[0].dev;
3054 kfree(bbio);
3055}
3056
3057static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
3058 struct scrub_wr_ctx *wr_ctx,
3059 struct btrfs_fs_info *fs_info,
3060 struct btrfs_device *dev,
3061 int is_dev_replace)
3062{
3063 WARN_ON(wr_ctx->wr_curr_bio != NULL);
3064
3065 mutex_init(&wr_ctx->wr_lock);
3066 wr_ctx->wr_curr_bio = NULL;
3067 if (!is_dev_replace)
3068 return 0;
3069
3070 WARN_ON(!dev->bdev);
3071 wr_ctx->pages_per_wr_bio = min_t(int, SCRUB_PAGES_PER_WR_BIO,
3072 bio_get_nr_vecs(dev->bdev));
3073 wr_ctx->tgtdev = dev;
3074 atomic_set(&wr_ctx->flush_all_writes, 0);
3075 return 0;
3076}
3077
3078static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx)
3079{
3080 mutex_lock(&wr_ctx->wr_lock);
3081 kfree(wr_ctx->wr_curr_bio);
3082 wr_ctx->wr_curr_bio = NULL;
3083 mutex_unlock(&wr_ctx->wr_lock);
3084}
3085
3086static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
3087 int mirror_num, u64 physical_for_dev_replace)
3088{
3089 struct scrub_copy_nocow_ctx *nocow_ctx;
3090 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
3091
3092 nocow_ctx = kzalloc(sizeof(*nocow_ctx), GFP_NOFS);
3093 if (!nocow_ctx) {
3094 spin_lock(&sctx->stat_lock);
3095 sctx->stat.malloc_errors++;
3096 spin_unlock(&sctx->stat_lock);
3097 return -ENOMEM;
3098 }
3099
3100 scrub_pending_trans_workers_inc(sctx);
3101
3102 nocow_ctx->sctx = sctx;
3103 nocow_ctx->logical = logical;
3104 nocow_ctx->len = len;
3105 nocow_ctx->mirror_num = mirror_num;
3106 nocow_ctx->physical_for_dev_replace = physical_for_dev_replace;
3107 nocow_ctx->work.func = copy_nocow_pages_worker;
3108 btrfs_queue_worker(&fs_info->scrub_nocow_workers,
3109 &nocow_ctx->work);
3110
3111 return 0;
3112}
3113
3114static void copy_nocow_pages_worker(struct btrfs_work *work)
3115{
3116 struct scrub_copy_nocow_ctx *nocow_ctx =
3117 container_of(work, struct scrub_copy_nocow_ctx, work);
3118 struct scrub_ctx *sctx = nocow_ctx->sctx;
3119 u64 logical = nocow_ctx->logical;
3120 u64 len = nocow_ctx->len;
3121 int mirror_num = nocow_ctx->mirror_num;
3122 u64 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
3123 int ret;
3124 struct btrfs_trans_handle *trans = NULL;
3125 struct btrfs_fs_info *fs_info;
3126 struct btrfs_path *path;
3127 struct btrfs_root *root;
3128 int not_written = 0;
3129
3130 fs_info = sctx->dev_root->fs_info;
3131 root = fs_info->extent_root;
3132
3133 path = btrfs_alloc_path();
3134 if (!path) {
3135 spin_lock(&sctx->stat_lock);
3136 sctx->stat.malloc_errors++;
3137 spin_unlock(&sctx->stat_lock);
3138 not_written = 1;
3139 goto out;
3140 }
3141
3142 trans = btrfs_join_transaction(root);
3143 if (IS_ERR(trans)) {
3144 not_written = 1;
3145 goto out;
3146 }
3147
3148 ret = iterate_inodes_from_logical(logical, fs_info, path,
3149 copy_nocow_pages_for_inode,
3150 nocow_ctx);
3151 if (ret != 0 && ret != -ENOENT) {
3152 pr_warn("iterate_inodes_from_logical() failed: log %llu, phys %llu, len %llu, mir %llu, ret %d\n",
3153 (unsigned long long)logical,
3154 (unsigned long long)physical_for_dev_replace,
3155 (unsigned long long)len,
3156 (unsigned long long)mirror_num, ret);
3157 not_written = 1;
3158 goto out;
3159 }
3160
3161out:
3162 if (trans && !IS_ERR(trans))
3163 btrfs_end_transaction(trans, root);
3164 if (not_written)
3165 btrfs_dev_replace_stats_inc(&fs_info->dev_replace.
3166 num_uncorrectable_read_errors);
3167
3168 btrfs_free_path(path);
3169 kfree(nocow_ctx);
3170
3171 scrub_pending_trans_workers_dec(sctx);
3172}
3173
3174static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root, void *ctx)
3175{
3176 unsigned long index;
3177 struct scrub_copy_nocow_ctx *nocow_ctx = ctx;
3178 int ret = 0;
3179 struct btrfs_key key;
3180 struct inode *inode = NULL;
3181 struct btrfs_root *local_root;
3182 u64 physical_for_dev_replace;
3183 u64 len;
3184 struct btrfs_fs_info *fs_info = nocow_ctx->sctx->dev_root->fs_info;
6f1c3605 3185 int srcu_index;
ff023aac
SB		 3186
3187 key.objectid = root;
3188 key.type = BTRFS_ROOT_ITEM_KEY;
3189 key.offset = (u64)-1;
6f1c3605
LB		 3190
3191 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
3192
ff023aac 3193 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
6f1c3605
LB		 3194 if (IS_ERR(local_root)) {
3195 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
ff023aac 3196 return PTR_ERR(local_root);
6f1c3605 3197 }
ff023aac
SB		 3198
3199 key.type = BTRFS_INODE_ITEM_KEY;
3200 key.objectid = inum;
3201 key.offset = 0;
3202 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
6f1c3605 3203 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
ff023aac
SB		 3204 if (IS_ERR(inode))
3205 return PTR_ERR(inode);
3206
3207 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
3208 len = nocow_ctx->len;
3209 while (len >= PAGE_CACHE_SIZE) {
3210 struct page *page = NULL;
3211 int ret_sub;
3212
3213 index = offset >> PAGE_CACHE_SHIFT;
3214
3215 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
3216 if (!page) {
3217 pr_err("find_or_create_page() failed\n");
3218 ret = -ENOMEM;
3219 goto next_page;
3220 }
3221
3222 if (PageUptodate(page)) {
3223 if (PageDirty(page))
3224 goto next_page;
3225 } else {
3226 ClearPageError(page);
3227 ret_sub = extent_read_full_page(&BTRFS_I(inode)->
3228 io_tree,
3229 page, btrfs_get_extent,
3230 nocow_ctx->mirror_num);
3231 if (ret_sub) {
3232 ret = ret_sub;
3233 goto next_page;
3234 }
3235 wait_on_page_locked(page);
3236 if (!PageUptodate(page)) {
3237 ret = -EIO;
3238 goto next_page;
3239 }
3240 }
3241 ret_sub = write_page_nocow(nocow_ctx->sctx,
3242 physical_for_dev_replace, page);
3243 if (ret_sub) {
3244 ret = ret_sub;
3245 goto next_page;
3246 }
3247
3248next_page:
3249 if (page) {
3250 unlock_page(page);
3251 put_page(page);
3252 }
3253 offset += PAGE_CACHE_SIZE;
3254 physical_for_dev_replace += PAGE_CACHE_SIZE;
3255 len -= PAGE_CACHE_SIZE;
3256 }
3257
3258 if (inode)
3259 iput(inode);
3260 return ret;
3261}
3262
3263static int write_page_nocow(struct scrub_ctx *sctx,
3264 u64 physical_for_dev_replace, struct page *page)
3265{
3266 struct bio *bio;
3267 struct btrfs_device *dev;
3268 int ret;
3269 DECLARE_COMPLETION_ONSTACK(compl);
3270
3271 dev = sctx->wr_ctx.tgtdev;
3272 if (!dev)
3273 return -EIO;
3274 if (!dev->bdev) {
3275 printk_ratelimited(KERN_WARNING
3276 "btrfs: scrub write_page_nocow(bdev == NULL) is unexpected!\n");
3277 return -EIO;
3278 }
3279 bio = bio_alloc(GFP_NOFS, 1);
3280 if (!bio) {
3281 spin_lock(&sctx->stat_lock);
3282 sctx->stat.malloc_errors++;
3283 spin_unlock(&sctx->stat_lock);
3284 return -ENOMEM;
3285 }
3286 bio->bi_private = &compl;
3287 bio->bi_end_io = scrub_complete_bio_end_io;
3288 bio->bi_size = 0;
3289 bio->bi_sector = physical_for_dev_replace >> 9;
3290 bio->bi_bdev = dev->bdev;
3291 ret = bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
3292 if (ret != PAGE_CACHE_SIZE) {
3293leave_with_eio:
3294 bio_put(bio);
3295 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
3296 return -EIO;
3297 }
3298 btrfsic_submit_bio(WRITE_SYNC, bio);
3299 wait_for_completion(&compl);
3300
3301 if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
3302 goto leave_with_eio;
3303
3304 bio_put(bio);
3305 return 0;
3306}
Linux kernel - Deliverables
This page took 0.268701 seconds and 5 git commands to generate.