scrub.c 90.2 KB
Newer Older
A
Arne Jansen 已提交
1
/*
2
 * Copyright (C) 2011, 2012 STRATO.  All rights reserved.
A
Arne Jansen 已提交
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License v2 as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 */

#include <linux/blkdev.h>
20
#include <linux/ratelimit.h>
A
Arne Jansen 已提交
21 22 23 24
#include "ctree.h"
#include "volumes.h"
#include "disk-io.h"
#include "ordered-data.h"
25
#include "transaction.h"
26
#include "backref.h"
27
#include "extent_io.h"
28
#include "dev-replace.h"
29
#include "check-integrity.h"
30
#include "rcu-string.h"
D
David Woodhouse 已提交
31
#include "raid56.h"
A
Arne Jansen 已提交
32 33 34 35 36 37 38 39 40 41 42 43 44 45

/*
 * This is only the first step towards a full-features scrub. It reads all
 * extent and super block and verifies the checksums. In case a bad checksum
 * is found or the extent cannot be read, good data will be written back if
 * any can be found.
 *
 * Future enhancements:
 *  - In case an unrepairable extent is encountered, track which files are
 *    affected and report them
 *  - track and record media errors, throw out bad devices
 *  - add a mode to also read unallocated space
 */

46
struct scrub_block;
47
struct scrub_ctx;
A
Arne Jansen 已提交
48

49 50 51 52 53 54 55 56 57
/*
 * the following three values only influence the performance.
 * The last one configures the number of parallel and outstanding I/O
 * operations. The first two values configure an upper limit for the number
 * of (dynamically allocated) pages that are added to a bio.
 */
#define SCRUB_PAGES_PER_RD_BIO	32	/* 128k per bio */
#define SCRUB_PAGES_PER_WR_BIO	32	/* 128k per bio */
#define SCRUB_BIOS_PER_SCTX	64	/* 8MB per device in flight */
58 59 60 61 62 63

/*
 * the following value times PAGE_SIZE needs to be large enough to match the
 * largest node/leaf/sector size that shall be supported.
 * Values larger than BTRFS_STRIPE_LEN are not supported.
 */
64
#define SCRUB_MAX_PAGES_PER_BLOCK	16	/* 64k per node/leaf/sector */
A
Arne Jansen 已提交
65 66

struct scrub_page {
67 68
	struct scrub_block	*sblock;
	struct page		*page;
69
	struct btrfs_device	*dev;
A
Arne Jansen 已提交
70 71
	u64			flags;  /* extent flags */
	u64			generation;
72 73
	u64			logical;
	u64			physical;
74
	u64			physical_for_dev_replace;
75
	atomic_t		ref_count;
76 77 78 79 80
	struct {
		unsigned int	mirror_num:8;
		unsigned int	have_csum:1;
		unsigned int	io_error:1;
	};
A
Arne Jansen 已提交
81 82 83 84 85
	u8			csum[BTRFS_CSUM_SIZE];
};

struct scrub_bio {
	int			index;
86
	struct scrub_ctx	*sctx;
87
	struct btrfs_device	*dev;
A
Arne Jansen 已提交
88 89 90 91
	struct bio		*bio;
	int			err;
	u64			logical;
	u64			physical;
92 93 94 95 96
#if SCRUB_PAGES_PER_WR_BIO >= SCRUB_PAGES_PER_RD_BIO
	struct scrub_page	*pagev[SCRUB_PAGES_PER_WR_BIO];
#else
	struct scrub_page	*pagev[SCRUB_PAGES_PER_RD_BIO];
#endif
97
	int			page_count;
A
Arne Jansen 已提交
98 99 100 101
	int			next_free;
	struct btrfs_work	work;
};

102
struct scrub_block {
103
	struct scrub_page	*pagev[SCRUB_MAX_PAGES_PER_BLOCK];
104 105 106
	int			page_count;
	atomic_t		outstanding_pages;
	atomic_t		ref_count; /* free mem on transition to zero */
107
	struct scrub_ctx	*sctx;
108 109 110 111
	struct {
		unsigned int	header_error:1;
		unsigned int	checksum_error:1;
		unsigned int	no_io_error_seen:1;
112
		unsigned int	generation_error:1; /* also sets header_error */
113 114 115
	};
};

116 117 118 119 120 121 122 123
struct scrub_wr_ctx {
	struct scrub_bio *wr_curr_bio;
	struct btrfs_device *tgtdev;
	int pages_per_wr_bio; /* <= SCRUB_PAGES_PER_WR_BIO */
	atomic_t flush_all_writes;
	struct mutex wr_lock;
};

124
struct scrub_ctx {
125
	struct scrub_bio	*bios[SCRUB_BIOS_PER_SCTX];
126
	struct btrfs_root	*dev_root;
A
Arne Jansen 已提交
127 128
	int			first_free;
	int			curr;
129 130
	atomic_t		bios_in_flight;
	atomic_t		workers_pending;
A
Arne Jansen 已提交
131 132 133 134 135
	spinlock_t		list_lock;
	wait_queue_head_t	list_wait;
	u16			csum_size;
	struct list_head	csum_list;
	atomic_t		cancel_req;
A
Arne Jansen 已提交
136
	int			readonly;
137
	int			pages_per_rd_bio;
138 139 140
	u32			sectorsize;
	u32			nodesize;
	u32			leafsize;
141 142

	int			is_dev_replace;
143
	struct scrub_wr_ctx	wr_ctx;
144

A
Arne Jansen 已提交
145 146 147 148 149 150 151
	/*
	 * statistics
	 */
	struct btrfs_scrub_progress stat;
	spinlock_t		stat_lock;
};

152
struct scrub_fixup_nodatasum {
153
	struct scrub_ctx	*sctx;
154
	struct btrfs_device	*dev;
155 156 157 158 159 160
	u64			logical;
	struct btrfs_root	*root;
	struct btrfs_work	work;
	int			mirror_num;
};

161 162 163 164 165 166 167
struct scrub_nocow_inode {
	u64			inum;
	u64			offset;
	u64			root;
	struct list_head	list;
};

168 169 170 171 172 173
struct scrub_copy_nocow_ctx {
	struct scrub_ctx	*sctx;
	u64			logical;
	u64			len;
	int			mirror_num;
	u64			physical_for_dev_replace;
174
	struct list_head	inodes;
175 176 177
	struct btrfs_work	work;
};

178 179 180 181 182 183 184 185 186 187 188 189 190
struct scrub_warning {
	struct btrfs_path	*path;
	u64			extent_item_size;
	char			*scratch_buf;
	char			*msg_buf;
	const char		*errstr;
	sector_t		sector;
	u64			logical;
	struct btrfs_device	*dev;
	int			msg_bufsize;
	int			scratch_bufsize;
};

191

192 193 194 195
static void scrub_pending_bio_inc(struct scrub_ctx *sctx);
static void scrub_pending_bio_dec(struct scrub_ctx *sctx);
static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx);
static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx);
196
static int scrub_handle_errored_block(struct scrub_block *sblock_to_check);
197
static int scrub_setup_recheck_block(struct scrub_ctx *sctx,
198
				     struct btrfs_fs_info *fs_info,
199
				     struct scrub_block *original_sblock,
200
				     u64 length, u64 logical,
201
				     struct scrub_block *sblocks_for_recheck);
202 203 204 205
static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
				struct scrub_block *sblock, int is_metadata,
				int have_csum, u8 *csum, u64 generation,
				u16 csum_size);
206 207 208 209 210 211 212 213 214 215 216
static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
					 struct scrub_block *sblock,
					 int is_metadata, int have_csum,
					 const u8 *csum, u64 generation,
					 u16 csum_size);
static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
					     struct scrub_block *sblock_good,
					     int force_write);
static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
					    struct scrub_block *sblock_good,
					    int page_num, int force_write);
217 218 219
static void scrub_write_block_to_dev_replace(struct scrub_block *sblock);
static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
					   int page_num);
220 221 222 223 224
static int scrub_checksum_data(struct scrub_block *sblock);
static int scrub_checksum_tree_block(struct scrub_block *sblock);
static int scrub_checksum_super(struct scrub_block *sblock);
static void scrub_block_get(struct scrub_block *sblock);
static void scrub_block_put(struct scrub_block *sblock);
225 226
static void scrub_page_get(struct scrub_page *spage);
static void scrub_page_put(struct scrub_page *spage);
227 228
static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
				    struct scrub_page *spage);
229
static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
230
		       u64 physical, struct btrfs_device *dev, u64 flags,
231 232
		       u64 gen, int mirror_num, u8 *csum, int force,
		       u64 physical_for_dev_replace);
S
Stefan Behrens 已提交
233
static void scrub_bio_end_io(struct bio *bio, int err);
234 235
static void scrub_bio_end_io_worker(struct btrfs_work *work);
static void scrub_block_complete(struct scrub_block *sblock);
236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254
static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
			       u64 extent_logical, u64 extent_len,
			       u64 *extent_physical,
			       struct btrfs_device **extent_dev,
			       int *extent_mirror_num);
static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
			      struct scrub_wr_ctx *wr_ctx,
			      struct btrfs_fs_info *fs_info,
			      struct btrfs_device *dev,
			      int is_dev_replace);
static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx);
static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
				    struct scrub_page *spage);
static void scrub_wr_submit(struct scrub_ctx *sctx);
static void scrub_wr_bio_end_io(struct bio *bio, int err);
static void scrub_wr_bio_end_io_worker(struct btrfs_work *work);
static int write_page_nocow(struct scrub_ctx *sctx,
			    u64 physical_for_dev_replace, struct page *page);
static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
255
				      struct scrub_copy_nocow_ctx *ctx);
256 257 258
static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
			    int mirror_num, u64 physical_for_dev_replace);
static void copy_nocow_pages_worker(struct btrfs_work *work);
259
static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
260
static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
S
Stefan Behrens 已提交
261 262


263 264 265 266 267 268 269 270 271 272 273
static void scrub_pending_bio_inc(struct scrub_ctx *sctx)
{
	atomic_inc(&sctx->bios_in_flight);
}

static void scrub_pending_bio_dec(struct scrub_ctx *sctx)
{
	atomic_dec(&sctx->bios_in_flight);
	wake_up(&sctx->list_wait);
}

274
static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
275 276 277 278 279 280 281 282 283
{
	while (atomic_read(&fs_info->scrub_pause_req)) {
		mutex_unlock(&fs_info->scrub_lock);
		wait_event(fs_info->scrub_pause_wait,
		   atomic_read(&fs_info->scrub_pause_req) == 0);
		mutex_lock(&fs_info->scrub_lock);
	}
}

284 285 286 287 288 289 290 291 292 293 294 295 296
static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
{
	atomic_inc(&fs_info->scrubs_paused);
	wake_up(&fs_info->scrub_pause_wait);

	mutex_lock(&fs_info->scrub_lock);
	__scrub_blocked_if_needed(fs_info);
	atomic_dec(&fs_info->scrubs_paused);
	mutex_unlock(&fs_info->scrub_lock);

	wake_up(&fs_info->scrub_pause_wait);
}

297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338
/*
 * used for workers that require transaction commits (i.e., for the
 * NOCOW case)
 */
static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx)
{
	struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;

	/*
	 * increment scrubs_running to prevent cancel requests from
	 * completing as long as a worker is running. we must also
	 * increment scrubs_paused to prevent deadlocking on pause
	 * requests used for transactions commits (as the worker uses a
	 * transaction context). it is safe to regard the worker
	 * as paused for all matters practical. effectively, we only
	 * avoid cancellation requests from completing.
	 */
	mutex_lock(&fs_info->scrub_lock);
	atomic_inc(&fs_info->scrubs_running);
	atomic_inc(&fs_info->scrubs_paused);
	mutex_unlock(&fs_info->scrub_lock);
	atomic_inc(&sctx->workers_pending);
}

/* used for workers that require transaction commits */
static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx)
{
	struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;

	/*
	 * see scrub_pending_trans_workers_inc() why we're pretending
	 * to be paused in the scrub counters
	 */
	mutex_lock(&fs_info->scrub_lock);
	atomic_dec(&fs_info->scrubs_running);
	atomic_dec(&fs_info->scrubs_paused);
	mutex_unlock(&fs_info->scrub_lock);
	atomic_dec(&sctx->workers_pending);
	wake_up(&fs_info->scrub_pause_wait);
	wake_up(&sctx->list_wait);
}

339
static void scrub_free_csums(struct scrub_ctx *sctx)
A
Arne Jansen 已提交
340
{
341
	while (!list_empty(&sctx->csum_list)) {
A
Arne Jansen 已提交
342
		struct btrfs_ordered_sum *sum;
343
		sum = list_first_entry(&sctx->csum_list,
A
Arne Jansen 已提交
344 345 346 347 348 349
				       struct btrfs_ordered_sum, list);
		list_del(&sum->list);
		kfree(sum);
	}
}

350
static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx)
A
Arne Jansen 已提交
351 352 353
{
	int i;

354
	if (!sctx)
A
Arne Jansen 已提交
355 356
		return;

357 358
	scrub_free_wr_ctx(&sctx->wr_ctx);

359
	/* this can happen when scrub is cancelled */
360 361
	if (sctx->curr != -1) {
		struct scrub_bio *sbio = sctx->bios[sctx->curr];
362 363

		for (i = 0; i < sbio->page_count; i++) {
364
			WARN_ON(!sbio->pagev[i]->page);
365 366 367 368 369
			scrub_block_put(sbio->pagev[i]->sblock);
		}
		bio_put(sbio->bio);
	}

370
	for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
371
		struct scrub_bio *sbio = sctx->bios[i];
A
Arne Jansen 已提交
372 373 374 375 376 377

		if (!sbio)
			break;
		kfree(sbio);
	}

378 379
	scrub_free_csums(sctx);
	kfree(sctx);
A
Arne Jansen 已提交
380 381 382
}

static noinline_for_stack
383
struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace)
A
Arne Jansen 已提交
384
{
385
	struct scrub_ctx *sctx;
A
Arne Jansen 已提交
386 387
	int		i;
	struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
388 389
	int pages_per_rd_bio;
	int ret;
A
Arne Jansen 已提交
390

391 392 393 394 395 396 397 398 399 400 401 402
	/*
	 * the setting of pages_per_rd_bio is correct for scrub but might
	 * be wrong for the dev_replace code where we might read from
	 * different devices in the initial huge bios. However, that
	 * code is able to correctly handle the case when adding a page
	 * to a bio fails.
	 */
	if (dev->bdev)
		pages_per_rd_bio = min_t(int, SCRUB_PAGES_PER_RD_BIO,
					 bio_get_nr_vecs(dev->bdev));
	else
		pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO;
403 404
	sctx = kzalloc(sizeof(*sctx), GFP_NOFS);
	if (!sctx)
A
Arne Jansen 已提交
405
		goto nomem;
406
	sctx->is_dev_replace = is_dev_replace;
407
	sctx->pages_per_rd_bio = pages_per_rd_bio;
408
	sctx->curr = -1;
409
	sctx->dev_root = dev->dev_root;
410
	for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
A
Arne Jansen 已提交
411 412 413 414 415
		struct scrub_bio *sbio;

		sbio = kzalloc(sizeof(*sbio), GFP_NOFS);
		if (!sbio)
			goto nomem;
416
		sctx->bios[i] = sbio;
A
Arne Jansen 已提交
417 418

		sbio->index = i;
419
		sbio->sctx = sctx;
420 421
		sbio->page_count = 0;
		sbio->work.func = scrub_bio_end_io_worker;
A
Arne Jansen 已提交
422

423
		if (i != SCRUB_BIOS_PER_SCTX - 1)
424
			sctx->bios[i]->next_free = i + 1;
425
		else
426 427 428 429 430 431
			sctx->bios[i]->next_free = -1;
	}
	sctx->first_free = 0;
	sctx->nodesize = dev->dev_root->nodesize;
	sctx->leafsize = dev->dev_root->leafsize;
	sctx->sectorsize = dev->dev_root->sectorsize;
432 433
	atomic_set(&sctx->bios_in_flight, 0);
	atomic_set(&sctx->workers_pending, 0);
434 435 436 437 438 439 440
	atomic_set(&sctx->cancel_req, 0);
	sctx->csum_size = btrfs_super_csum_size(fs_info->super_copy);
	INIT_LIST_HEAD(&sctx->csum_list);

	spin_lock_init(&sctx->list_lock);
	spin_lock_init(&sctx->stat_lock);
	init_waitqueue_head(&sctx->list_wait);
441 442 443 444 445 446 447

	ret = scrub_setup_wr_ctx(sctx, &sctx->wr_ctx, fs_info,
				 fs_info->dev_replace.tgtdev, is_dev_replace);
	if (ret) {
		scrub_free_ctx(sctx);
		return ERR_PTR(ret);
	}
448
	return sctx;
A
Arne Jansen 已提交
449 450

nomem:
451
	scrub_free_ctx(sctx);
A
Arne Jansen 已提交
452 453 454
	return ERR_PTR(-ENOMEM);
}

455 456
static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root,
				     void *warn_ctx)
457 458 459 460 461 462 463
{
	u64 isize;
	u32 nlink;
	int ret;
	int i;
	struct extent_buffer *eb;
	struct btrfs_inode_item *inode_item;
464
	struct scrub_warning *swarn = warn_ctx;
465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492
	struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info;
	struct inode_fs_paths *ipath = NULL;
	struct btrfs_root *local_root;
	struct btrfs_key root_key;

	root_key.objectid = root;
	root_key.type = BTRFS_ROOT_ITEM_KEY;
	root_key.offset = (u64)-1;
	local_root = btrfs_read_fs_root_no_name(fs_info, &root_key);
	if (IS_ERR(local_root)) {
		ret = PTR_ERR(local_root);
		goto err;
	}

	ret = inode_item_info(inum, 0, local_root, swarn->path);
	if (ret) {
		btrfs_release_path(swarn->path);
		goto err;
	}

	eb = swarn->path->nodes[0];
	inode_item = btrfs_item_ptr(eb, swarn->path->slots[0],
					struct btrfs_inode_item);
	isize = btrfs_inode_size(eb, inode_item);
	nlink = btrfs_inode_nlink(eb, inode_item);
	btrfs_release_path(swarn->path);

	ipath = init_ipath(4096, local_root, swarn->path);
493 494 495 496 497
	if (IS_ERR(ipath)) {
		ret = PTR_ERR(ipath);
		ipath = NULL;
		goto err;
	}
498 499 500 501 502 503 504 505 506 507
	ret = paths_from_inode(inum, ipath);

	if (ret < 0)
		goto err;

	/*
	 * we deliberately ignore the bit ipath might have been too small to
	 * hold all of the paths here
	 */
	for (i = 0; i < ipath->fspath->elem_cnt; ++i)
508
		printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
509 510
			"%s, sector %llu, root %llu, inode %llu, offset %llu, "
			"length %llu, links %u (path: %s)\n", swarn->errstr,
511
			swarn->logical, rcu_str_deref(swarn->dev->name),
512 513
			(unsigned long long)swarn->sector, root, inum, offset,
			min(isize - offset, (u64)PAGE_SIZE), nlink,
514
			(char *)(unsigned long)ipath->fspath->val[i]);
515 516 517 518 519

	free_ipath(ipath);
	return 0;

err:
520
	printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
521 522
		"%s, sector %llu, root %llu, inode %llu, offset %llu: path "
		"resolving failed with ret=%d\n", swarn->errstr,
523
		swarn->logical, rcu_str_deref(swarn->dev->name),
524 525 526 527 528 529
		(unsigned long long)swarn->sector, root, inum, offset, ret);

	free_ipath(ipath);
	return 0;
}

530
static void scrub_print_warning(const char *errstr, struct scrub_block *sblock)
531
{
532 533
	struct btrfs_device *dev;
	struct btrfs_fs_info *fs_info;
534 535 536 537 538
	struct btrfs_path *path;
	struct btrfs_key found_key;
	struct extent_buffer *eb;
	struct btrfs_extent_item *ei;
	struct scrub_warning swarn;
539 540 541
	unsigned long ptr = 0;
	u64 extent_item_pos;
	u64 flags = 0;
542
	u64 ref_root;
543
	u32 item_size;
544 545
	u8 ref_level;
	const int bufsize = 4096;
546
	int ret;
547

548
	WARN_ON(sblock->page_count < 1);
549
	dev = sblock->pagev[0]->dev;
550 551
	fs_info = sblock->sctx->dev_root->fs_info;

552 553 554 555
	path = btrfs_alloc_path();

	swarn.scratch_buf = kmalloc(bufsize, GFP_NOFS);
	swarn.msg_buf = kmalloc(bufsize, GFP_NOFS);
556 557
	swarn.sector = (sblock->pagev[0]->physical) >> 9;
	swarn.logical = sblock->pagev[0]->logical;
558
	swarn.errstr = errstr;
559
	swarn.dev = NULL;
560 561 562 563 564 565
	swarn.msg_bufsize = bufsize;
	swarn.scratch_bufsize = bufsize;

	if (!path || !swarn.scratch_buf || !swarn.msg_buf)
		goto out;

566 567
	ret = extent_from_logical(fs_info, swarn.logical, path, &found_key,
				  &flags);
568 569 570
	if (ret < 0)
		goto out;

J
Jan Schmidt 已提交
571
	extent_item_pos = swarn.logical - found_key.objectid;
572 573 574 575 576 577
	swarn.extent_item_size = found_key.offset;

	eb = path->nodes[0];
	ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
	item_size = btrfs_item_size_nr(eb, path->slots[0]);

578
	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
579 580 581
		do {
			ret = tree_backref_for_extent(&ptr, eb, ei, item_size,
							&ref_root, &ref_level);
582
			printk_in_rcu(KERN_WARNING
583
				"BTRFS: %s at logical %llu on dev %s, "
584
				"sector %llu: metadata %s (level %d) in tree "
585 586
				"%llu\n", errstr, swarn.logical,
				rcu_str_deref(dev->name),
587 588 589 590 591
				(unsigned long long)swarn.sector,
				ref_level ? "node" : "leaf",
				ret < 0 ? -1 : ref_level,
				ret < 0 ? -1 : ref_root);
		} while (ret != 1);
592
		btrfs_release_path(path);
593
	} else {
594
		btrfs_release_path(path);
595
		swarn.path = path;
596
		swarn.dev = dev;
597 598
		iterate_extent_inodes(fs_info, found_key.objectid,
					extent_item_pos, 1,
599 600 601 602 603 604 605 606 607
					scrub_print_warning_inode, &swarn);
	}

out:
	btrfs_free_path(path);
	kfree(swarn.scratch_buf);
	kfree(swarn.msg_buf);
}

608
static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx)
609
{
610
	struct page *page = NULL;
611
	unsigned long index;
612
	struct scrub_fixup_nodatasum *fixup = fixup_ctx;
613
	int ret;
614
	int corrected = 0;
615
	struct btrfs_key key;
616
	struct inode *inode = NULL;
617
	struct btrfs_fs_info *fs_info;
618 619
	u64 end = offset + PAGE_SIZE - 1;
	struct btrfs_root *local_root;
620
	int srcu_index;
621 622 623 624

	key.objectid = root;
	key.type = BTRFS_ROOT_ITEM_KEY;
	key.offset = (u64)-1;
625 626 627 628 629 630 631

	fs_info = fixup->root->fs_info;
	srcu_index = srcu_read_lock(&fs_info->subvol_srcu);

	local_root = btrfs_read_fs_root_no_name(fs_info, &key);
	if (IS_ERR(local_root)) {
		srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
632
		return PTR_ERR(local_root);
633
	}
634 635 636 637

	key.type = BTRFS_INODE_ITEM_KEY;
	key.objectid = inum;
	key.offset = 0;
638 639
	inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
	srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
640 641 642 643 644 645
	if (IS_ERR(inode))
		return PTR_ERR(inode);

	index = offset >> PAGE_CACHE_SHIFT;

	page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671
	if (!page) {
		ret = -ENOMEM;
		goto out;
	}

	if (PageUptodate(page)) {
		if (PageDirty(page)) {
			/*
			 * we need to write the data to the defect sector. the
			 * data that was in that sector is not in memory,
			 * because the page was modified. we must not write the
			 * modified page to that sector.
			 *
			 * TODO: what could be done here: wait for the delalloc
			 *       runner to write out that page (might involve
			 *       COW) and see whether the sector is still
			 *       referenced afterwards.
			 *
			 * For the meantime, we'll treat this error
			 * incorrectable, although there is a chance that a
			 * later scrub will find the bad sector again and that
			 * there's no dirty page in memory, then.
			 */
			ret = -EIO;
			goto out;
		}
672 673
		fs_info = BTRFS_I(inode)->root->fs_info;
		ret = repair_io_failure(fs_info, offset, PAGE_SIZE,
674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710
					fixup->logical, page,
					fixup->mirror_num);
		unlock_page(page);
		corrected = !ret;
	} else {
		/*
		 * we need to get good data first. the general readpage path
		 * will call repair_io_failure for us, we just have to make
		 * sure we read the bad mirror.
		 */
		ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
					EXTENT_DAMAGED, GFP_NOFS);
		if (ret) {
			/* set_extent_bits should give proper error */
			WARN_ON(ret > 0);
			if (ret > 0)
				ret = -EFAULT;
			goto out;
		}

		ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page,
						btrfs_get_extent,
						fixup->mirror_num);
		wait_on_page_locked(page);

		corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset,
						end, EXTENT_DAMAGED, 0, NULL);
		if (!corrected)
			clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
						EXTENT_DAMAGED, GFP_NOFS);
	}

out:
	if (page)
		put_page(page);
	if (inode)
		iput(inode);
711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729

	if (ret < 0)
		return ret;

	if (ret == 0 && corrected) {
		/*
		 * we only need to call readpage for one of the inodes belonging
		 * to this extent. so make iterate_extent_inodes stop
		 */
		return 1;
	}

	return -EIO;
}

static void scrub_fixup_nodatasum(struct btrfs_work *work)
{
	int ret;
	struct scrub_fixup_nodatasum *fixup;
730
	struct scrub_ctx *sctx;
731 732 733 734 735
	struct btrfs_trans_handle *trans = NULL;
	struct btrfs_path *path;
	int uncorrectable = 0;

	fixup = container_of(work, struct scrub_fixup_nodatasum, work);
736
	sctx = fixup->sctx;
737 738 739

	path = btrfs_alloc_path();
	if (!path) {
740 741 742
		spin_lock(&sctx->stat_lock);
		++sctx->stat.malloc_errors;
		spin_unlock(&sctx->stat_lock);
743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770
		uncorrectable = 1;
		goto out;
	}

	trans = btrfs_join_transaction(fixup->root);
	if (IS_ERR(trans)) {
		uncorrectable = 1;
		goto out;
	}

	/*
	 * the idea is to trigger a regular read through the standard path. we
	 * read a page from the (failed) logical address by specifying the
	 * corresponding copynum of the failed sector. thus, that readpage is
	 * expected to fail.
	 * that is the point where on-the-fly error correction will kick in
	 * (once it's finished) and rewrite the failed sector if a good copy
	 * can be found.
	 */
	ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info,
						path, scrub_fixup_readpage,
						fixup);
	if (ret < 0) {
		uncorrectable = 1;
		goto out;
	}
	WARN_ON(ret != 1);

771 772 773
	spin_lock(&sctx->stat_lock);
	++sctx->stat.corrected_errors;
	spin_unlock(&sctx->stat_lock);
774 775 776 777 778

out:
	if (trans && !IS_ERR(trans))
		btrfs_end_transaction(trans, fixup->root);
	if (uncorrectable) {
779 780 781
		spin_lock(&sctx->stat_lock);
		++sctx->stat.uncorrectable_errors;
		spin_unlock(&sctx->stat_lock);
782 783 784
		btrfs_dev_replace_stats_inc(
			&sctx->dev_root->fs_info->dev_replace.
			num_uncorrectable_read_errors);
785 786
		printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
		    "unable to fixup (nodatasum) error at logical %llu on dev %s\n",
787
			fixup->logical, rcu_str_deref(fixup->dev->name));
788 789 790 791 792
	}

	btrfs_free_path(path);
	kfree(fixup);

793
	scrub_pending_trans_workers_dec(sctx);
794 795
}

A
Arne Jansen 已提交
796
/*
797 798 799 800 801 802
 * scrub_handle_errored_block gets called when either verification of the
 * pages failed or the bio failed to read, e.g. with EIO. In the latter
 * case, this function handles all pages in the bio, even though only one
 * may be bad.
 * The goal of this function is to repair the errored block by using the
 * contents of one of the mirrors.
A
Arne Jansen 已提交
803
 */
804
static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
A
Arne Jansen 已提交
805
{
806
	struct scrub_ctx *sctx = sblock_to_check->sctx;
807
	struct btrfs_device *dev;
808 809 810 811 812 813 814 815 816 817 818 819 820 821
	struct btrfs_fs_info *fs_info;
	u64 length;
	u64 logical;
	u64 generation;
	unsigned int failed_mirror_index;
	unsigned int is_metadata;
	unsigned int have_csum;
	u8 *csum;
	struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */
	struct scrub_block *sblock_bad;
	int ret;
	int mirror_index;
	int page_num;
	int success;
822
	static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
823 824 825
				      DEFAULT_RATELIMIT_BURST);

	BUG_ON(sblock_to_check->page_count < 1);
826
	fs_info = sctx->dev_root->fs_info;
827 828 829 830 831 832 833 834 835 836 837
	if (sblock_to_check->pagev[0]->flags & BTRFS_EXTENT_FLAG_SUPER) {
		/*
		 * if we find an error in a super block, we just report it.
		 * They will get written with the next transaction commit
		 * anyway
		 */
		spin_lock(&sctx->stat_lock);
		++sctx->stat.super_errors;
		spin_unlock(&sctx->stat_lock);
		return 0;
	}
838
	length = sblock_to_check->page_count * PAGE_SIZE;
839 840 841 842 843
	logical = sblock_to_check->pagev[0]->logical;
	generation = sblock_to_check->pagev[0]->generation;
	BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1);
	failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1;
	is_metadata = !(sblock_to_check->pagev[0]->flags &
844
			BTRFS_EXTENT_FLAG_DATA);
845 846 847
	have_csum = sblock_to_check->pagev[0]->have_csum;
	csum = sblock_to_check->pagev[0]->csum;
	dev = sblock_to_check->pagev[0]->dev;
848

849 850 851 852 853
	if (sctx->is_dev_replace && !is_metadata && !have_csum) {
		sblocks_for_recheck = NULL;
		goto nodatasum_case;
	}

854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886
	/*
	 * read all mirrors one after the other. This includes to
	 * re-read the extent or metadata block that failed (that was
	 * the cause that this fixup code is called) another time,
	 * page by page this time in order to know which pages
	 * caused I/O errors and which ones are good (for all mirrors).
	 * It is the goal to handle the situation when more than one
	 * mirror contains I/O errors, but the errors do not
	 * overlap, i.e. the data can be repaired by selecting the
	 * pages from those mirrors without I/O error on the
	 * particular pages. One example (with blocks >= 2 * PAGE_SIZE)
	 * would be that mirror #1 has an I/O error on the first page,
	 * the second page is good, and mirror #2 has an I/O error on
	 * the second page, but the first page is good.
	 * Then the first page of the first mirror can be repaired by
	 * taking the first page of the second mirror, and the
	 * second page of the second mirror can be repaired by
	 * copying the contents of the 2nd page of the 1st mirror.
	 * One more note: if the pages of one mirror contain I/O
	 * errors, the checksum cannot be verified. In order to get
	 * the best data for repairing, the first attempt is to find
	 * a mirror without I/O errors and with a validated checksum.
	 * Only if this is not possible, the pages are picked from
	 * mirrors with I/O errors without considering the checksum.
	 * If the latter is the case, at the end, the checksum of the
	 * repaired area is verified in order to correctly maintain
	 * the statistics.
	 */

	sblocks_for_recheck = kzalloc(BTRFS_MAX_MIRRORS *
				     sizeof(*sblocks_for_recheck),
				     GFP_NOFS);
	if (!sblocks_for_recheck) {
887 888 889 890 891
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		sctx->stat.read_errors++;
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
892
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
893
		goto out;
A
Arne Jansen 已提交
894 895
	}

896
	/* setup the context, map the logical blocks and alloc the pages */
897
	ret = scrub_setup_recheck_block(sctx, fs_info, sblock_to_check, length,
898 899
					logical, sblocks_for_recheck);
	if (ret) {
900 901 902 903
		spin_lock(&sctx->stat_lock);
		sctx->stat.read_errors++;
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
904
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
905 906 907 908
		goto out;
	}
	BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS);
	sblock_bad = sblocks_for_recheck + failed_mirror_index;
909

910
	/* build and submit the bios for the failed mirror, check checksums */
911 912
	scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum,
			    csum, generation, sctx->csum_size);
A
Arne Jansen 已提交
913

914 915 916 917 918 919 920 921 922 923
	if (!sblock_bad->header_error && !sblock_bad->checksum_error &&
	    sblock_bad->no_io_error_seen) {
		/*
		 * the error disappeared after reading page by page, or
		 * the area was part of a huge bio and other parts of the
		 * bio caused I/O errors, or the block layer merged several
		 * read requests into one and the error is caused by a
		 * different bio (usually one of the two latter cases is
		 * the cause)
		 */
924 925 926
		spin_lock(&sctx->stat_lock);
		sctx->stat.unverified_errors++;
		spin_unlock(&sctx->stat_lock);
A
Arne Jansen 已提交
927

928 929
		if (sctx->is_dev_replace)
			scrub_write_block_to_dev_replace(sblock_bad);
930
		goto out;
A
Arne Jansen 已提交
931 932
	}

933
	if (!sblock_bad->no_io_error_seen) {
934 935 936
		spin_lock(&sctx->stat_lock);
		sctx->stat.read_errors++;
		spin_unlock(&sctx->stat_lock);
937 938
		if (__ratelimit(&_rs))
			scrub_print_warning("i/o error", sblock_to_check);
939
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
940
	} else if (sblock_bad->checksum_error) {
941 942 943
		spin_lock(&sctx->stat_lock);
		sctx->stat.csum_errors++;
		spin_unlock(&sctx->stat_lock);
944 945
		if (__ratelimit(&_rs))
			scrub_print_warning("checksum error", sblock_to_check);
946
		btrfs_dev_stat_inc_and_print(dev,
947
					     BTRFS_DEV_STAT_CORRUPTION_ERRS);
948
	} else if (sblock_bad->header_error) {
949 950 951
		spin_lock(&sctx->stat_lock);
		sctx->stat.verify_errors++;
		spin_unlock(&sctx->stat_lock);
952 953 954
		if (__ratelimit(&_rs))
			scrub_print_warning("checksum/header error",
					    sblock_to_check);
955
		if (sblock_bad->generation_error)
956
			btrfs_dev_stat_inc_and_print(dev,
957 958
				BTRFS_DEV_STAT_GENERATION_ERRS);
		else
959
			btrfs_dev_stat_inc_and_print(dev,
960
				BTRFS_DEV_STAT_CORRUPTION_ERRS);
961
	}
A
Arne Jansen 已提交
962

963 964 965 966
	if (sctx->readonly) {
		ASSERT(!sctx->is_dev_replace);
		goto out;
	}
A
Arne Jansen 已提交
967

968 969
	if (!is_metadata && !have_csum) {
		struct scrub_fixup_nodatasum *fixup_nodatasum;
A
Arne Jansen 已提交
970

971 972 973
nodatasum_case:
		WARN_ON(sctx->is_dev_replace);

974 975 976 977 978 979 980 981 982 983
		/*
		 * !is_metadata and !have_csum, this means that the data
		 * might not be COW'ed, that it might be modified
		 * concurrently. The general strategy to work on the
		 * commit root does not help in the case when COW is not
		 * used.
		 */
		fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS);
		if (!fixup_nodatasum)
			goto did_not_correct_error;
984
		fixup_nodatasum->sctx = sctx;
985
		fixup_nodatasum->dev = dev;
986 987 988
		fixup_nodatasum->logical = logical;
		fixup_nodatasum->root = fs_info->extent_root;
		fixup_nodatasum->mirror_num = failed_mirror_index + 1;
989
		scrub_pending_trans_workers_inc(sctx);
990 991 992 993
		fixup_nodatasum->work.func = scrub_fixup_nodatasum;
		btrfs_queue_worker(&fs_info->scrub_workers,
				   &fixup_nodatasum->work);
		goto out;
A
Arne Jansen 已提交
994 995
	}

996 997
	/*
	 * now build and submit the bios for the other mirrors, check
998 999
	 * checksums.
	 * First try to pick the mirror which is completely without I/O
1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014
	 * errors and also does not have a checksum error.
	 * If one is found, and if a checksum is present, the full block
	 * that is known to contain an error is rewritten. Afterwards
	 * the block is known to be corrected.
	 * If a mirror is found which is completely correct, and no
	 * checksum is present, only those pages are rewritten that had
	 * an I/O error in the block to be repaired, since it cannot be
	 * determined, which copy of the other pages is better (and it
	 * could happen otherwise that a correct page would be
	 * overwritten by a bad one).
	 */
	for (mirror_index = 0;
	     mirror_index < BTRFS_MAX_MIRRORS &&
	     sblocks_for_recheck[mirror_index].page_count > 0;
	     mirror_index++) {
1015
		struct scrub_block *sblock_other;
1016

1017 1018 1019 1020 1021
		if (mirror_index == failed_mirror_index)
			continue;
		sblock_other = sblocks_for_recheck + mirror_index;

		/* build and submit the bios, check checksums */
1022 1023 1024 1025 1026
		scrub_recheck_block(fs_info, sblock_other, is_metadata,
				    have_csum, csum, generation,
				    sctx->csum_size);

		if (!sblock_other->header_error &&
1027 1028
		    !sblock_other->checksum_error &&
		    sblock_other->no_io_error_seen) {
1029 1030 1031 1032 1033 1034 1035 1036 1037
			if (sctx->is_dev_replace) {
				scrub_write_block_to_dev_replace(sblock_other);
			} else {
				int force_write = is_metadata || have_csum;

				ret = scrub_repair_block_from_good_copy(
						sblock_bad, sblock_other,
						force_write);
			}
1038 1039 1040 1041
			if (0 == ret)
				goto corrected_error;
		}
	}
A
Arne Jansen 已提交
1042 1043

	/*
1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101
	 * for dev_replace, pick good pages and write to the target device.
	 */
	if (sctx->is_dev_replace) {
		success = 1;
		for (page_num = 0; page_num < sblock_bad->page_count;
		     page_num++) {
			int sub_success;

			sub_success = 0;
			for (mirror_index = 0;
			     mirror_index < BTRFS_MAX_MIRRORS &&
			     sblocks_for_recheck[mirror_index].page_count > 0;
			     mirror_index++) {
				struct scrub_block *sblock_other =
					sblocks_for_recheck + mirror_index;
				struct scrub_page *page_other =
					sblock_other->pagev[page_num];

				if (!page_other->io_error) {
					ret = scrub_write_page_to_dev_replace(
							sblock_other, page_num);
					if (ret == 0) {
						/* succeeded for this page */
						sub_success = 1;
						break;
					} else {
						btrfs_dev_replace_stats_inc(
							&sctx->dev_root->
							fs_info->dev_replace.
							num_write_errors);
					}
				}
			}

			if (!sub_success) {
				/*
				 * did not find a mirror to fetch the page
				 * from. scrub_write_page_to_dev_replace()
				 * handles this case (page->io_error), by
				 * filling the block with zeros before
				 * submitting the write request
				 */
				success = 0;
				ret = scrub_write_page_to_dev_replace(
						sblock_bad, page_num);
				if (ret)
					btrfs_dev_replace_stats_inc(
						&sctx->dev_root->fs_info->
						dev_replace.num_write_errors);
			}
		}

		goto out;
	}

	/*
	 * for regular scrub, repair those pages that are errored.
	 * In case of I/O errors in the area that is supposed to be
1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122
	 * repaired, continue by picking good copies of those pages.
	 * Select the good pages from mirrors to rewrite bad pages from
	 * the area to fix. Afterwards verify the checksum of the block
	 * that is supposed to be repaired. This verification step is
	 * only done for the purpose of statistic counting and for the
	 * final scrub report, whether errors remain.
	 * A perfect algorithm could make use of the checksum and try
	 * all possible combinations of pages from the different mirrors
	 * until the checksum verification succeeds. For example, when
	 * the 2nd page of mirror #1 faces I/O errors, and the 2nd page
	 * of mirror #2 is readable but the final checksum test fails,
	 * then the 2nd page of mirror #3 could be tried, whether now
	 * the final checksum succeedes. But this would be a rare
	 * exception and is therefore not implemented. At least it is
	 * avoided that the good copy is overwritten.
	 * A more useful improvement would be to pick the sectors
	 * without I/O error based on sector sizes (512 bytes on legacy
	 * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one
	 * mirror could be repaired by taking 512 byte of a different
	 * mirror, even if other 512 byte sectors in the same PAGE_SIZE
	 * area are unreadable.
A
Arne Jansen 已提交
1123 1124
	 */

1125 1126 1127 1128 1129 1130
	/* can only fix I/O errors from here on */
	if (sblock_bad->no_io_error_seen)
		goto did_not_correct_error;

	success = 1;
	for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
1131
		struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1132 1133

		if (!page_bad->io_error)
A
Arne Jansen 已提交
1134
			continue;
1135 1136 1137 1138 1139 1140 1141

		for (mirror_index = 0;
		     mirror_index < BTRFS_MAX_MIRRORS &&
		     sblocks_for_recheck[mirror_index].page_count > 0;
		     mirror_index++) {
			struct scrub_block *sblock_other = sblocks_for_recheck +
							   mirror_index;
1142 1143
			struct scrub_page *page_other = sblock_other->pagev[
							page_num];
1144 1145 1146 1147 1148 1149 1150 1151 1152

			if (!page_other->io_error) {
				ret = scrub_repair_page_from_good_copy(
					sblock_bad, sblock_other, page_num, 0);
				if (0 == ret) {
					page_bad->io_error = 0;
					break; /* succeeded for this page */
				}
			}
I
Ilya Dryomov 已提交
1153
		}
A
Arne Jansen 已提交
1154

1155 1156 1157 1158
		if (page_bad->io_error) {
			/* did not find a mirror to copy the page from */
			success = 0;
		}
A
Arne Jansen 已提交
1159 1160
	}

1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171
	if (success) {
		if (is_metadata || have_csum) {
			/*
			 * need to verify the checksum now that all
			 * sectors on disk are repaired (the write
			 * request for data to be repaired is on its way).
			 * Just be lazy and use scrub_recheck_block()
			 * which re-reads the data before the checksum
			 * is verified, but most likely the data comes out
			 * of the page cache.
			 */
1172 1173 1174 1175
			scrub_recheck_block(fs_info, sblock_bad,
					    is_metadata, have_csum, csum,
					    generation, sctx->csum_size);
			if (!sblock_bad->header_error &&
1176 1177 1178 1179 1180 1181 1182
			    !sblock_bad->checksum_error &&
			    sblock_bad->no_io_error_seen)
				goto corrected_error;
			else
				goto did_not_correct_error;
		} else {
corrected_error:
1183 1184 1185
			spin_lock(&sctx->stat_lock);
			sctx->stat.corrected_errors++;
			spin_unlock(&sctx->stat_lock);
1186
			printk_ratelimited_in_rcu(KERN_ERR
1187
				"BTRFS: fixed up error at logical %llu on dev %s\n",
1188
				logical, rcu_str_deref(dev->name));
A
Arne Jansen 已提交
1189
		}
1190 1191
	} else {
did_not_correct_error:
1192 1193 1194
		spin_lock(&sctx->stat_lock);
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
1195
		printk_ratelimited_in_rcu(KERN_ERR
1196
			"BTRFS: unable to fixup (regular) error at logical %llu on dev %s\n",
1197
			logical, rcu_str_deref(dev->name));
I
Ilya Dryomov 已提交
1198
	}
A
Arne Jansen 已提交
1199

1200 1201 1202 1203 1204 1205 1206 1207
out:
	if (sblocks_for_recheck) {
		for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS;
		     mirror_index++) {
			struct scrub_block *sblock = sblocks_for_recheck +
						     mirror_index;
			int page_index;

1208 1209 1210 1211 1212
			for (page_index = 0; page_index < sblock->page_count;
			     page_index++) {
				sblock->pagev[page_index]->sblock = NULL;
				scrub_page_put(sblock->pagev[page_index]);
			}
1213 1214 1215
		}
		kfree(sblocks_for_recheck);
	}
A
Arne Jansen 已提交
1216

1217 1218
	return 0;
}
A
Arne Jansen 已提交
1219

1220
static int scrub_setup_recheck_block(struct scrub_ctx *sctx,
1221
				     struct btrfs_fs_info *fs_info,
1222
				     struct scrub_block *original_sblock,
1223 1224 1225 1226 1227 1228 1229 1230
				     u64 length, u64 logical,
				     struct scrub_block *sblocks_for_recheck)
{
	int page_index;
	int mirror_index;
	int ret;

	/*
1231
	 * note: the two members ref_count and outstanding_pages
1232 1233 1234 1235 1236 1237 1238 1239 1240
	 * are not used (and not set) in the blocks that are used for
	 * the recheck procedure
	 */

	page_index = 0;
	while (length > 0) {
		u64 sublen = min_t(u64, length, PAGE_SIZE);
		u64 mapped_length = sublen;
		struct btrfs_bio *bbio = NULL;
A
Arne Jansen 已提交
1241

1242 1243 1244 1245
		/*
		 * with a length of PAGE_SIZE, each returned stripe
		 * represents one mirror
		 */
1246 1247
		ret = btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS, logical,
				      &mapped_length, &bbio, 0);
1248 1249 1250 1251
		if (ret || !bbio || mapped_length < sublen) {
			kfree(bbio);
			return -EIO;
		}
A
Arne Jansen 已提交
1252

1253
		BUG_ON(page_index >= SCRUB_PAGES_PER_RD_BIO);
1254 1255 1256 1257 1258 1259 1260 1261 1262
		for (mirror_index = 0; mirror_index < (int)bbio->num_stripes;
		     mirror_index++) {
			struct scrub_block *sblock;
			struct scrub_page *page;

			if (mirror_index >= BTRFS_MAX_MIRRORS)
				continue;

			sblock = sblocks_for_recheck + mirror_index;
1263 1264 1265 1266
			sblock->sctx = sctx;
			page = kzalloc(sizeof(*page), GFP_NOFS);
			if (!page) {
leave_nomem:
1267 1268 1269
				spin_lock(&sctx->stat_lock);
				sctx->stat.malloc_errors++;
				spin_unlock(&sctx->stat_lock);
1270
				kfree(bbio);
1271 1272
				return -ENOMEM;
			}
1273 1274 1275 1276
			scrub_page_get(page);
			sblock->pagev[page_index] = page;
			page->logical = logical;
			page->physical = bbio->stripes[mirror_index].physical;
1277 1278 1279 1280
			BUG_ON(page_index >= original_sblock->page_count);
			page->physical_for_dev_replace =
				original_sblock->pagev[page_index]->
				physical_for_dev_replace;
1281 1282 1283
			/* for missing devices, dev->bdev is NULL */
			page->dev = bbio->stripes[mirror_index].dev;
			page->mirror_num = mirror_index + 1;
1284
			sblock->page_count++;
1285 1286 1287
			page->page = alloc_page(GFP_NOFS);
			if (!page->page)
				goto leave_nomem;
1288 1289 1290 1291 1292 1293 1294 1295
		}
		kfree(bbio);
		length -= sublen;
		logical += sublen;
		page_index++;
	}

	return 0;
I
Ilya Dryomov 已提交
1296 1297
}

1298 1299 1300 1301 1302 1303 1304
/*
 * this function will check the on disk data for checksum errors, header
 * errors and read I/O errors. If any I/O errors happen, the exact pages
 * which are errored are marked as being bad. The goal is to enable scrub
 * to take those pages that are not errored from all the mirrors so that
 * the pages that are errored in the just handled mirror can be repaired.
 */
1305 1306 1307 1308
static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
				struct scrub_block *sblock, int is_metadata,
				int have_csum, u8 *csum, u64 generation,
				u16 csum_size)
I
Ilya Dryomov 已提交
1309
{
1310
	int page_num;
I
Ilya Dryomov 已提交
1311

1312 1313 1314
	sblock->no_io_error_seen = 1;
	sblock->header_error = 0;
	sblock->checksum_error = 0;
I
Ilya Dryomov 已提交
1315

1316 1317
	for (page_num = 0; page_num < sblock->page_count; page_num++) {
		struct bio *bio;
1318
		struct scrub_page *page = sblock->pagev[page_num];
1319

1320
		if (page->dev->bdev == NULL) {
1321 1322 1323 1324 1325
			page->io_error = 1;
			sblock->no_io_error_seen = 0;
			continue;
		}

1326
		WARN_ON(!page->page);
1327
		bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1328 1329 1330 1331 1332
		if (!bio) {
			page->io_error = 1;
			sblock->no_io_error_seen = 0;
			continue;
		}
1333
		bio->bi_bdev = page->dev->bdev;
1334 1335
		bio->bi_sector = page->physical >> 9;

1336
		bio_add_page(bio, page->page, PAGE_SIZE, 0);
1337
		if (btrfsic_submit_bio_wait(READ, bio))
1338
			sblock->no_io_error_seen = 0;
1339

1340 1341
		bio_put(bio);
	}
I
Ilya Dryomov 已提交
1342

1343 1344 1345 1346 1347
	if (sblock->no_io_error_seen)
		scrub_recheck_block_checksum(fs_info, sblock, is_metadata,
					     have_csum, csum, generation,
					     csum_size);

1348
	return;
A
Arne Jansen 已提交
1349 1350
}

1351 1352 1353 1354 1355
static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
					 struct scrub_block *sblock,
					 int is_metadata, int have_csum,
					 const u8 *csum, u64 generation,
					 u16 csum_size)
A
Arne Jansen 已提交
1356
{
1357 1358 1359 1360 1361
	int page_num;
	u8 calculated_csum[BTRFS_CSUM_SIZE];
	u32 crc = ~(u32)0;
	void *mapped_buffer;

1362
	WARN_ON(!sblock->pagev[0]->page);
1363 1364 1365
	if (is_metadata) {
		struct btrfs_header *h;

1366
		mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
1367 1368
		h = (struct btrfs_header *)mapped_buffer;

1369
		if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h) ||
1370 1371
		    memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE) ||
		    memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
1372
			   BTRFS_UUID_SIZE)) {
1373
			sblock->header_error = 1;
1374
		} else if (generation != btrfs_stack_header_generation(h)) {
1375 1376 1377
			sblock->header_error = 1;
			sblock->generation_error = 1;
		}
1378 1379 1380 1381
		csum = h->csum;
	} else {
		if (!have_csum)
			return;
A
Arne Jansen 已提交
1382

1383
		mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
1384
	}
A
Arne Jansen 已提交
1385

1386 1387
	for (page_num = 0;;) {
		if (page_num == 0 && is_metadata)
1388
			crc = btrfs_csum_data(
1389 1390 1391
				((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE,
				crc, PAGE_SIZE - BTRFS_CSUM_SIZE);
		else
1392
			crc = btrfs_csum_data(mapped_buffer, crc, PAGE_SIZE);
1393

1394
		kunmap_atomic(mapped_buffer);
1395 1396 1397
		page_num++;
		if (page_num >= sblock->page_count)
			break;
1398
		WARN_ON(!sblock->pagev[page_num]->page);
1399

1400
		mapped_buffer = kmap_atomic(sblock->pagev[page_num]->page);
1401 1402 1403 1404 1405
	}

	btrfs_csum_final(crc, calculated_csum);
	if (memcmp(calculated_csum, csum, csum_size))
		sblock->checksum_error = 1;
A
Arne Jansen 已提交
1406 1407
}

1408 1409 1410 1411 1412 1413
static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
					     struct scrub_block *sblock_good,
					     int force_write)
{
	int page_num;
	int ret = 0;
I
Ilya Dryomov 已提交
1414

1415 1416
	for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
		int ret_sub;
I
Ilya Dryomov 已提交
1417

1418 1419 1420 1421 1422 1423
		ret_sub = scrub_repair_page_from_good_copy(sblock_bad,
							   sblock_good,
							   page_num,
							   force_write);
		if (ret_sub)
			ret = ret_sub;
A
Arne Jansen 已提交
1424
	}
1425 1426 1427 1428 1429 1430 1431 1432

	return ret;
}

static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
					    struct scrub_block *sblock_good,
					    int page_num, int force_write)
{
1433 1434
	struct scrub_page *page_bad = sblock_bad->pagev[page_num];
	struct scrub_page *page_good = sblock_good->pagev[page_num];
1435

1436 1437
	BUG_ON(page_bad->page == NULL);
	BUG_ON(page_good->page == NULL);
1438 1439 1440 1441 1442
	if (force_write || sblock_bad->header_error ||
	    sblock_bad->checksum_error || page_bad->io_error) {
		struct bio *bio;
		int ret;

1443
		if (!page_bad->dev->bdev) {
1444 1445 1446
			printk_ratelimited(KERN_WARNING "BTRFS: "
				"scrub_repair_page_from_good_copy(bdev == NULL) "
				"is unexpected!\n");
1447 1448 1449
			return -EIO;
		}

1450
		bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1451 1452
		if (!bio)
			return -EIO;
1453
		bio->bi_bdev = page_bad->dev->bdev;
1454 1455 1456 1457 1458 1459
		bio->bi_sector = page_bad->physical >> 9;

		ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0);
		if (PAGE_SIZE != ret) {
			bio_put(bio);
			return -EIO;
1460
		}
1461

1462
		if (btrfsic_submit_bio_wait(WRITE, bio)) {
1463 1464
			btrfs_dev_stat_inc_and_print(page_bad->dev,
				BTRFS_DEV_STAT_WRITE_ERRS);
1465 1466 1467
			btrfs_dev_replace_stats_inc(
				&sblock_bad->sctx->dev_root->fs_info->
				dev_replace.num_write_errors);
1468 1469 1470
			bio_put(bio);
			return -EIO;
		}
1471
		bio_put(bio);
A
Arne Jansen 已提交
1472 1473
	}

1474 1475 1476
	return 0;
}

1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535
static void scrub_write_block_to_dev_replace(struct scrub_block *sblock)
{
	int page_num;

	for (page_num = 0; page_num < sblock->page_count; page_num++) {
		int ret;

		ret = scrub_write_page_to_dev_replace(sblock, page_num);
		if (ret)
			btrfs_dev_replace_stats_inc(
				&sblock->sctx->dev_root->fs_info->dev_replace.
				num_write_errors);
	}
}

static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
					   int page_num)
{
	struct scrub_page *spage = sblock->pagev[page_num];

	BUG_ON(spage->page == NULL);
	if (spage->io_error) {
		void *mapped_buffer = kmap_atomic(spage->page);

		memset(mapped_buffer, 0, PAGE_CACHE_SIZE);
		flush_dcache_page(spage->page);
		kunmap_atomic(mapped_buffer);
	}
	return scrub_add_page_to_wr_bio(sblock->sctx, spage);
}

static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
				    struct scrub_page *spage)
{
	struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
	struct scrub_bio *sbio;
	int ret;

	mutex_lock(&wr_ctx->wr_lock);
again:
	if (!wr_ctx->wr_curr_bio) {
		wr_ctx->wr_curr_bio = kzalloc(sizeof(*wr_ctx->wr_curr_bio),
					      GFP_NOFS);
		if (!wr_ctx->wr_curr_bio) {
			mutex_unlock(&wr_ctx->wr_lock);
			return -ENOMEM;
		}
		wr_ctx->wr_curr_bio->sctx = sctx;
		wr_ctx->wr_curr_bio->page_count = 0;
	}
	sbio = wr_ctx->wr_curr_bio;
	if (sbio->page_count == 0) {
		struct bio *bio;

		sbio->physical = spage->physical_for_dev_replace;
		sbio->logical = spage->logical;
		sbio->dev = wr_ctx->tgtdev;
		bio = sbio->bio;
		if (!bio) {
1536
			bio = btrfs_io_bio_alloc(GFP_NOFS, wr_ctx->pages_per_wr_bio);
1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638
			if (!bio) {
				mutex_unlock(&wr_ctx->wr_lock);
				return -ENOMEM;
			}
			sbio->bio = bio;
		}

		bio->bi_private = sbio;
		bio->bi_end_io = scrub_wr_bio_end_io;
		bio->bi_bdev = sbio->dev->bdev;
		bio->bi_sector = sbio->physical >> 9;
		sbio->err = 0;
	} else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
		   spage->physical_for_dev_replace ||
		   sbio->logical + sbio->page_count * PAGE_SIZE !=
		   spage->logical) {
		scrub_wr_submit(sctx);
		goto again;
	}

	ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
	if (ret != PAGE_SIZE) {
		if (sbio->page_count < 1) {
			bio_put(sbio->bio);
			sbio->bio = NULL;
			mutex_unlock(&wr_ctx->wr_lock);
			return -EIO;
		}
		scrub_wr_submit(sctx);
		goto again;
	}

	sbio->pagev[sbio->page_count] = spage;
	scrub_page_get(spage);
	sbio->page_count++;
	if (sbio->page_count == wr_ctx->pages_per_wr_bio)
		scrub_wr_submit(sctx);
	mutex_unlock(&wr_ctx->wr_lock);

	return 0;
}

static void scrub_wr_submit(struct scrub_ctx *sctx)
{
	struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
	struct scrub_bio *sbio;

	if (!wr_ctx->wr_curr_bio)
		return;

	sbio = wr_ctx->wr_curr_bio;
	wr_ctx->wr_curr_bio = NULL;
	WARN_ON(!sbio->bio->bi_bdev);
	scrub_pending_bio_inc(sctx);
	/* process all writes in a single worker thread. Then the block layer
	 * orders the requests before sending them to the driver which
	 * doubled the write performance on spinning disks when measured
	 * with Linux 3.5 */
	btrfsic_submit_bio(WRITE, sbio->bio);
}

static void scrub_wr_bio_end_io(struct bio *bio, int err)
{
	struct scrub_bio *sbio = bio->bi_private;
	struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;

	sbio->err = err;
	sbio->bio = bio;

	sbio->work.func = scrub_wr_bio_end_io_worker;
	btrfs_queue_worker(&fs_info->scrub_wr_completion_workers, &sbio->work);
}

static void scrub_wr_bio_end_io_worker(struct btrfs_work *work)
{
	struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
	struct scrub_ctx *sctx = sbio->sctx;
	int i;

	WARN_ON(sbio->page_count > SCRUB_PAGES_PER_WR_BIO);
	if (sbio->err) {
		struct btrfs_dev_replace *dev_replace =
			&sbio->sctx->dev_root->fs_info->dev_replace;

		for (i = 0; i < sbio->page_count; i++) {
			struct scrub_page *spage = sbio->pagev[i];

			spage->io_error = 1;
			btrfs_dev_replace_stats_inc(&dev_replace->
						    num_write_errors);
		}
	}

	for (i = 0; i < sbio->page_count; i++)
		scrub_page_put(sbio->pagev[i]);

	bio_put(sbio->bio);
	kfree(sbio);
	scrub_pending_bio_dec(sctx);
}

static int scrub_checksum(struct scrub_block *sblock)
1639 1640 1641 1642
{
	u64 flags;
	int ret;

1643 1644
	WARN_ON(sblock->page_count < 1);
	flags = sblock->pagev[0]->flags;
1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655
	ret = 0;
	if (flags & BTRFS_EXTENT_FLAG_DATA)
		ret = scrub_checksum_data(sblock);
	else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
		ret = scrub_checksum_tree_block(sblock);
	else if (flags & BTRFS_EXTENT_FLAG_SUPER)
		(void)scrub_checksum_super(sblock);
	else
		WARN_ON(1);
	if (ret)
		scrub_handle_errored_block(sblock);
1656 1657

	return ret;
A
Arne Jansen 已提交
1658 1659
}

1660
static int scrub_checksum_data(struct scrub_block *sblock)
A
Arne Jansen 已提交
1661
{
1662
	struct scrub_ctx *sctx = sblock->sctx;
A
Arne Jansen 已提交
1663
	u8 csum[BTRFS_CSUM_SIZE];
1664 1665 1666
	u8 *on_disk_csum;
	struct page *page;
	void *buffer;
A
Arne Jansen 已提交
1667 1668
	u32 crc = ~(u32)0;
	int fail = 0;
1669 1670
	u64 len;
	int index;
A
Arne Jansen 已提交
1671

1672
	BUG_ON(sblock->page_count < 1);
1673
	if (!sblock->pagev[0]->have_csum)
A
Arne Jansen 已提交
1674 1675
		return 0;

1676 1677
	on_disk_csum = sblock->pagev[0]->csum;
	page = sblock->pagev[0]->page;
1678
	buffer = kmap_atomic(page);
1679

1680
	len = sctx->sectorsize;
1681 1682 1683 1684
	index = 0;
	for (;;) {
		u64 l = min_t(u64, len, PAGE_SIZE);

1685
		crc = btrfs_csum_data(buffer, crc, l);
1686
		kunmap_atomic(buffer);
1687 1688 1689 1690 1691
		len -= l;
		if (len == 0)
			break;
		index++;
		BUG_ON(index >= sblock->page_count);
1692 1693
		BUG_ON(!sblock->pagev[index]->page);
		page = sblock->pagev[index]->page;
1694
		buffer = kmap_atomic(page);
1695 1696
	}

A
Arne Jansen 已提交
1697
	btrfs_csum_final(crc, csum);
1698
	if (memcmp(csum, on_disk_csum, sctx->csum_size))
A
Arne Jansen 已提交
1699 1700 1701 1702 1703
		fail = 1;

	return fail;
}

1704
static int scrub_checksum_tree_block(struct scrub_block *sblock)
A
Arne Jansen 已提交
1705
{
1706
	struct scrub_ctx *sctx = sblock->sctx;
A
Arne Jansen 已提交
1707
	struct btrfs_header *h;
1708
	struct btrfs_root *root = sctx->dev_root;
A
Arne Jansen 已提交
1709
	struct btrfs_fs_info *fs_info = root->fs_info;
1710 1711 1712 1713 1714 1715
	u8 calculated_csum[BTRFS_CSUM_SIZE];
	u8 on_disk_csum[BTRFS_CSUM_SIZE];
	struct page *page;
	void *mapped_buffer;
	u64 mapped_size;
	void *p;
A
Arne Jansen 已提交
1716 1717 1718
	u32 crc = ~(u32)0;
	int fail = 0;
	int crc_fail = 0;
1719 1720 1721 1722
	u64 len;
	int index;

	BUG_ON(sblock->page_count < 1);
1723
	page = sblock->pagev[0]->page;
1724
	mapped_buffer = kmap_atomic(page);
1725
	h = (struct btrfs_header *)mapped_buffer;
1726
	memcpy(on_disk_csum, h->csum, sctx->csum_size);
A
Arne Jansen 已提交
1727 1728 1729 1730 1731 1732 1733

	/*
	 * we don't use the getter functions here, as we
	 * a) don't have an extent buffer and
	 * b) the page is already kmapped
	 */

1734
	if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h))
A
Arne Jansen 已提交
1735 1736
		++fail;

1737
	if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h))
A
Arne Jansen 已提交
1738 1739 1740 1741 1742 1743 1744 1745 1746
		++fail;

	if (memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
		++fail;

	if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
		   BTRFS_UUID_SIZE))
		++fail;

1747
	WARN_ON(sctx->nodesize != sctx->leafsize);
1748
	len = sctx->nodesize - BTRFS_CSUM_SIZE;
1749 1750 1751 1752 1753 1754
	mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
	p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
	index = 0;
	for (;;) {
		u64 l = min_t(u64, len, mapped_size);

1755
		crc = btrfs_csum_data(p, crc, l);
1756
		kunmap_atomic(mapped_buffer);
1757 1758 1759 1760 1761
		len -= l;
		if (len == 0)
			break;
		index++;
		BUG_ON(index >= sblock->page_count);
1762 1763
		BUG_ON(!sblock->pagev[index]->page);
		page = sblock->pagev[index]->page;
1764
		mapped_buffer = kmap_atomic(page);
1765 1766 1767 1768 1769
		mapped_size = PAGE_SIZE;
		p = mapped_buffer;
	}

	btrfs_csum_final(crc, calculated_csum);
1770
	if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
A
Arne Jansen 已提交
1771 1772 1773 1774 1775
		++crc_fail;

	return fail || crc_fail;
}

1776
static int scrub_checksum_super(struct scrub_block *sblock)
A
Arne Jansen 已提交
1777 1778
{
	struct btrfs_super_block *s;
1779
	struct scrub_ctx *sctx = sblock->sctx;
1780
	struct btrfs_root *root = sctx->dev_root;
A
Arne Jansen 已提交
1781
	struct btrfs_fs_info *fs_info = root->fs_info;
1782 1783 1784 1785 1786 1787
	u8 calculated_csum[BTRFS_CSUM_SIZE];
	u8 on_disk_csum[BTRFS_CSUM_SIZE];
	struct page *page;
	void *mapped_buffer;
	u64 mapped_size;
	void *p;
A
Arne Jansen 已提交
1788
	u32 crc = ~(u32)0;
1789 1790
	int fail_gen = 0;
	int fail_cor = 0;
1791 1792
	u64 len;
	int index;
A
Arne Jansen 已提交
1793

1794
	BUG_ON(sblock->page_count < 1);
1795
	page = sblock->pagev[0]->page;
1796
	mapped_buffer = kmap_atomic(page);
1797
	s = (struct btrfs_super_block *)mapped_buffer;
1798
	memcpy(on_disk_csum, s->csum, sctx->csum_size);
A
Arne Jansen 已提交
1799

1800
	if (sblock->pagev[0]->logical != btrfs_super_bytenr(s))
1801
		++fail_cor;
A
Arne Jansen 已提交
1802

1803
	if (sblock->pagev[0]->generation != btrfs_super_generation(s))
1804
		++fail_gen;
A
Arne Jansen 已提交
1805 1806

	if (memcmp(s->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
1807
		++fail_cor;
A
Arne Jansen 已提交
1808

1809 1810 1811 1812 1813 1814 1815
	len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE;
	mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
	p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
	index = 0;
	for (;;) {
		u64 l = min_t(u64, len, mapped_size);

1816
		crc = btrfs_csum_data(p, crc, l);
1817
		kunmap_atomic(mapped_buffer);
1818 1819 1820 1821 1822
		len -= l;
		if (len == 0)
			break;
		index++;
		BUG_ON(index >= sblock->page_count);
1823 1824
		BUG_ON(!sblock->pagev[index]->page);
		page = sblock->pagev[index]->page;
1825
		mapped_buffer = kmap_atomic(page);
1826 1827 1828 1829 1830
		mapped_size = PAGE_SIZE;
		p = mapped_buffer;
	}

	btrfs_csum_final(crc, calculated_csum);
1831
	if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
1832
		++fail_cor;
A
Arne Jansen 已提交
1833

1834
	if (fail_cor + fail_gen) {
A
Arne Jansen 已提交
1835 1836 1837 1838 1839
		/*
		 * if we find an error in a super block, we just report it.
		 * They will get written with the next transaction commit
		 * anyway
		 */
1840 1841 1842
		spin_lock(&sctx->stat_lock);
		++sctx->stat.super_errors;
		spin_unlock(&sctx->stat_lock);
1843
		if (fail_cor)
1844
			btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
1845 1846
				BTRFS_DEV_STAT_CORRUPTION_ERRS);
		else
1847
			btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
1848
				BTRFS_DEV_STAT_GENERATION_ERRS);
A
Arne Jansen 已提交
1849 1850
	}

1851
	return fail_cor + fail_gen;
A
Arne Jansen 已提交
1852 1853
}

1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864
static void scrub_block_get(struct scrub_block *sblock)
{
	atomic_inc(&sblock->ref_count);
}

static void scrub_block_put(struct scrub_block *sblock)
{
	if (atomic_dec_and_test(&sblock->ref_count)) {
		int i;

		for (i = 0; i < sblock->page_count; i++)
1865
			scrub_page_put(sblock->pagev[i]);
1866 1867 1868 1869
		kfree(sblock);
	}
}

1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883
static void scrub_page_get(struct scrub_page *spage)
{
	atomic_inc(&spage->ref_count);
}

static void scrub_page_put(struct scrub_page *spage)
{
	if (atomic_dec_and_test(&spage->ref_count)) {
		if (spage->page)
			__free_page(spage->page);
		kfree(spage);
	}
}

1884
static void scrub_submit(struct scrub_ctx *sctx)
A
Arne Jansen 已提交
1885 1886 1887
{
	struct scrub_bio *sbio;

1888
	if (sctx->curr == -1)
S
Stefan Behrens 已提交
1889
		return;
A
Arne Jansen 已提交
1890

1891 1892
	sbio = sctx->bios[sctx->curr];
	sctx->curr = -1;
1893
	scrub_pending_bio_inc(sctx);
A
Arne Jansen 已提交
1894

1895 1896 1897 1898 1899 1900 1901 1902 1903
	if (!sbio->bio->bi_bdev) {
		/*
		 * this case should not happen. If btrfs_map_block() is
		 * wrong, it could happen for dev-replace operations on
		 * missing devices when no mirrors are available, but in
		 * this case it should already fail the mount.
		 * This case is handled correctly (but _very_ slowly).
		 */
		printk_ratelimited(KERN_WARNING
1904
			"BTRFS: scrub_submit(bio bdev == NULL) is unexpected!\n");
1905 1906 1907 1908
		bio_endio(sbio->bio, -EIO);
	} else {
		btrfsic_submit_bio(READ, sbio->bio);
	}
A
Arne Jansen 已提交
1909 1910
}

1911 1912
static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
				    struct scrub_page *spage)
A
Arne Jansen 已提交
1913
{
1914
	struct scrub_block *sblock = spage->sblock;
A
Arne Jansen 已提交
1915
	struct scrub_bio *sbio;
1916
	int ret;
A
Arne Jansen 已提交
1917 1918 1919 1920 1921

again:
	/*
	 * grab a fresh bio or wait for one to become available
	 */
1922 1923 1924 1925 1926 1927 1928 1929
	while (sctx->curr == -1) {
		spin_lock(&sctx->list_lock);
		sctx->curr = sctx->first_free;
		if (sctx->curr != -1) {
			sctx->first_free = sctx->bios[sctx->curr]->next_free;
			sctx->bios[sctx->curr]->next_free = -1;
			sctx->bios[sctx->curr]->page_count = 0;
			spin_unlock(&sctx->list_lock);
A
Arne Jansen 已提交
1930
		} else {
1931 1932
			spin_unlock(&sctx->list_lock);
			wait_event(sctx->list_wait, sctx->first_free != -1);
A
Arne Jansen 已提交
1933 1934
		}
	}
1935
	sbio = sctx->bios[sctx->curr];
1936
	if (sbio->page_count == 0) {
1937 1938
		struct bio *bio;

1939 1940
		sbio->physical = spage->physical;
		sbio->logical = spage->logical;
1941
		sbio->dev = spage->dev;
1942 1943
		bio = sbio->bio;
		if (!bio) {
1944
			bio = btrfs_io_bio_alloc(GFP_NOFS, sctx->pages_per_rd_bio);
1945 1946 1947 1948
			if (!bio)
				return -ENOMEM;
			sbio->bio = bio;
		}
1949 1950 1951

		bio->bi_private = sbio;
		bio->bi_end_io = scrub_bio_end_io;
1952 1953
		bio->bi_bdev = sbio->dev->bdev;
		bio->bi_sector = sbio->physical >> 9;
1954
		sbio->err = 0;
1955 1956 1957
	} else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
		   spage->physical ||
		   sbio->logical + sbio->page_count * PAGE_SIZE !=
1958 1959
		   spage->logical ||
		   sbio->dev != spage->dev) {
1960
		scrub_submit(sctx);
A
Arne Jansen 已提交
1961 1962
		goto again;
	}
1963

1964 1965 1966 1967 1968 1969 1970 1971
	sbio->pagev[sbio->page_count] = spage;
	ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
	if (ret != PAGE_SIZE) {
		if (sbio->page_count < 1) {
			bio_put(sbio->bio);
			sbio->bio = NULL;
			return -EIO;
		}
1972
		scrub_submit(sctx);
1973 1974 1975
		goto again;
	}

1976
	scrub_block_get(sblock); /* one for the page added to the bio */
1977 1978
	atomic_inc(&sblock->outstanding_pages);
	sbio->page_count++;
1979
	if (sbio->page_count == sctx->pages_per_rd_bio)
1980
		scrub_submit(sctx);
1981 1982 1983 1984

	return 0;
}

1985
static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
1986
		       u64 physical, struct btrfs_device *dev, u64 flags,
1987 1988
		       u64 gen, int mirror_num, u8 *csum, int force,
		       u64 physical_for_dev_replace)
1989 1990 1991 1992 1993 1994
{
	struct scrub_block *sblock;
	int index;

	sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
	if (!sblock) {
1995 1996 1997
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		spin_unlock(&sctx->stat_lock);
1998
		return -ENOMEM;
A
Arne Jansen 已提交
1999
	}
2000

2001 2002
	/* one ref inside this function, plus one for each page added to
	 * a bio later on */
2003
	atomic_set(&sblock->ref_count, 1);
2004
	sblock->sctx = sctx;
2005 2006 2007
	sblock->no_io_error_seen = 1;

	for (index = 0; len > 0; index++) {
2008
		struct scrub_page *spage;
2009 2010
		u64 l = min_t(u64, len, PAGE_SIZE);

2011 2012 2013
		spage = kzalloc(sizeof(*spage), GFP_NOFS);
		if (!spage) {
leave_nomem:
2014 2015 2016
			spin_lock(&sctx->stat_lock);
			sctx->stat.malloc_errors++;
			spin_unlock(&sctx->stat_lock);
2017
			scrub_block_put(sblock);
2018 2019
			return -ENOMEM;
		}
2020 2021 2022
		BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
		scrub_page_get(spage);
		sblock->pagev[index] = spage;
2023
		spage->sblock = sblock;
2024
		spage->dev = dev;
2025 2026 2027 2028
		spage->flags = flags;
		spage->generation = gen;
		spage->logical = logical;
		spage->physical = physical;
2029
		spage->physical_for_dev_replace = physical_for_dev_replace;
2030 2031 2032
		spage->mirror_num = mirror_num;
		if (csum) {
			spage->have_csum = 1;
2033
			memcpy(spage->csum, csum, sctx->csum_size);
2034 2035 2036 2037
		} else {
			spage->have_csum = 0;
		}
		sblock->page_count++;
2038 2039 2040
		spage->page = alloc_page(GFP_NOFS);
		if (!spage->page)
			goto leave_nomem;
2041 2042 2043
		len -= l;
		logical += l;
		physical += l;
2044
		physical_for_dev_replace += l;
2045 2046
	}

2047
	WARN_ON(sblock->page_count == 0);
2048
	for (index = 0; index < sblock->page_count; index++) {
2049
		struct scrub_page *spage = sblock->pagev[index];
2050 2051
		int ret;

2052
		ret = scrub_add_page_to_rd_bio(sctx, spage);
2053 2054
		if (ret) {
			scrub_block_put(sblock);
2055
			return ret;
2056
		}
2057
	}
A
Arne Jansen 已提交
2058

2059
	if (force)
2060
		scrub_submit(sctx);
A
Arne Jansen 已提交
2061

2062 2063
	/* last one frees, either here or in bio completion for last page */
	scrub_block_put(sblock);
A
Arne Jansen 已提交
2064 2065 2066
	return 0;
}

2067 2068 2069
static void scrub_bio_end_io(struct bio *bio, int err)
{
	struct scrub_bio *sbio = bio->bi_private;
2070
	struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
2071 2072 2073 2074 2075 2076 2077 2078 2079 2080

	sbio->err = err;
	sbio->bio = bio;

	btrfs_queue_worker(&fs_info->scrub_workers, &sbio->work);
}

static void scrub_bio_end_io_worker(struct btrfs_work *work)
{
	struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
2081
	struct scrub_ctx *sctx = sbio->sctx;
2082 2083
	int i;

2084
	BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO);
2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105
	if (sbio->err) {
		for (i = 0; i < sbio->page_count; i++) {
			struct scrub_page *spage = sbio->pagev[i];

			spage->io_error = 1;
			spage->sblock->no_io_error_seen = 0;
		}
	}

	/* now complete the scrub_block items that have all pages completed */
	for (i = 0; i < sbio->page_count; i++) {
		struct scrub_page *spage = sbio->pagev[i];
		struct scrub_block *sblock = spage->sblock;

		if (atomic_dec_and_test(&sblock->outstanding_pages))
			scrub_block_complete(sblock);
		scrub_block_put(sblock);
	}

	bio_put(sbio->bio);
	sbio->bio = NULL;
2106 2107 2108 2109
	spin_lock(&sctx->list_lock);
	sbio->next_free = sctx->first_free;
	sctx->first_free = sbio->index;
	spin_unlock(&sctx->list_lock);
2110 2111 2112 2113 2114 2115 2116 2117

	if (sctx->is_dev_replace &&
	    atomic_read(&sctx->wr_ctx.flush_all_writes)) {
		mutex_lock(&sctx->wr_ctx.wr_lock);
		scrub_wr_submit(sctx);
		mutex_unlock(&sctx->wr_ctx.wr_lock);
	}

2118
	scrub_pending_bio_dec(sctx);
2119 2120 2121 2122
}

static void scrub_block_complete(struct scrub_block *sblock)
{
2123
	if (!sblock->no_io_error_seen) {
2124
		scrub_handle_errored_block(sblock);
2125 2126 2127 2128 2129 2130 2131 2132 2133
	} else {
		/*
		 * if has checksum error, write via repair mechanism in
		 * dev replace case, otherwise write here in dev replace
		 * case.
		 */
		if (!scrub_checksum(sblock) && sblock->sctx->is_dev_replace)
			scrub_write_block_to_dev_replace(sblock);
	}
2134 2135
}

2136
static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len,
A
Arne Jansen 已提交
2137 2138 2139
			   u8 *csum)
{
	struct btrfs_ordered_sum *sum = NULL;
2140
	unsigned long index;
A
Arne Jansen 已提交
2141 2142
	unsigned long num_sectors;

2143 2144
	while (!list_empty(&sctx->csum_list)) {
		sum = list_first_entry(&sctx->csum_list,
A
Arne Jansen 已提交
2145 2146 2147 2148 2149 2150
				       struct btrfs_ordered_sum, list);
		if (sum->bytenr > logical)
			return 0;
		if (sum->bytenr + sum->len > logical)
			break;

2151
		++sctx->stat.csum_discards;
A
Arne Jansen 已提交
2152 2153 2154 2155 2156 2157 2158
		list_del(&sum->list);
		kfree(sum);
		sum = NULL;
	}
	if (!sum)
		return 0;

2159
	index = ((u32)(logical - sum->bytenr)) / sctx->sectorsize;
2160
	num_sectors = sum->len / sctx->sectorsize;
2161 2162
	memcpy(csum, sum->sums + index, sctx->csum_size);
	if (index == num_sectors - 1) {
A
Arne Jansen 已提交
2163 2164 2165
		list_del(&sum->list);
		kfree(sum);
	}
2166
	return 1;
A
Arne Jansen 已提交
2167 2168 2169
}

/* scrub extent tries to collect up to 64 kB for each bio */
2170
static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len,
2171
			u64 physical, struct btrfs_device *dev, u64 flags,
2172
			u64 gen, int mirror_num, u64 physical_for_dev_replace)
A
Arne Jansen 已提交
2173 2174 2175
{
	int ret;
	u8 csum[BTRFS_CSUM_SIZE];
2176 2177 2178
	u32 blocksize;

	if (flags & BTRFS_EXTENT_FLAG_DATA) {
2179 2180 2181 2182 2183
		blocksize = sctx->sectorsize;
		spin_lock(&sctx->stat_lock);
		sctx->stat.data_extents_scrubbed++;
		sctx->stat.data_bytes_scrubbed += len;
		spin_unlock(&sctx->stat_lock);
2184
	} else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2185
		WARN_ON(sctx->nodesize != sctx->leafsize);
2186 2187 2188 2189 2190
		blocksize = sctx->nodesize;
		spin_lock(&sctx->stat_lock);
		sctx->stat.tree_extents_scrubbed++;
		sctx->stat.tree_bytes_scrubbed += len;
		spin_unlock(&sctx->stat_lock);
2191
	} else {
2192
		blocksize = sctx->sectorsize;
2193
		WARN_ON(1);
2194
	}
A
Arne Jansen 已提交
2195 2196

	while (len) {
2197
		u64 l = min_t(u64, len, blocksize);
A
Arne Jansen 已提交
2198 2199 2200 2201
		int have_csum = 0;

		if (flags & BTRFS_EXTENT_FLAG_DATA) {
			/* push csums to sbio */
2202
			have_csum = scrub_find_csum(sctx, logical, l, csum);
A
Arne Jansen 已提交
2203
			if (have_csum == 0)
2204
				++sctx->stat.no_csum;
2205 2206 2207 2208 2209 2210
			if (sctx->is_dev_replace && !have_csum) {
				ret = copy_nocow_pages(sctx, logical, l,
						       mirror_num,
						      physical_for_dev_replace);
				goto behind_scrub_pages;
			}
A
Arne Jansen 已提交
2211
		}
2212
		ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen,
2213 2214 2215
				  mirror_num, have_csum ? csum : NULL, 0,
				  physical_for_dev_replace);
behind_scrub_pages:
A
Arne Jansen 已提交
2216 2217 2218 2219 2220
		if (ret)
			return ret;
		len -= l;
		logical += l;
		physical += l;
2221
		physical_for_dev_replace += l;
A
Arne Jansen 已提交
2222 2223 2224 2225
	}
	return 0;
}

2226
static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
2227 2228
					   struct map_lookup *map,
					   struct btrfs_device *scrub_dev,
2229 2230
					   int num, u64 base, u64 length,
					   int is_dev_replace)
A
Arne Jansen 已提交
2231 2232
{
	struct btrfs_path *path;
2233
	struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
A
Arne Jansen 已提交
2234 2235 2236
	struct btrfs_root *root = fs_info->extent_root;
	struct btrfs_root *csum_root = fs_info->csum_root;
	struct btrfs_extent_item *extent;
2237
	struct blk_plug plug;
A
Arne Jansen 已提交
2238 2239 2240 2241 2242 2243 2244 2245
	u64 flags;
	int ret;
	int slot;
	u64 nstripes;
	struct extent_buffer *l;
	struct btrfs_key key;
	u64 physical;
	u64 logical;
L
Liu Bo 已提交
2246
	u64 logic_end;
A
Arne Jansen 已提交
2247
	u64 generation;
2248
	int mirror_num;
A
Arne Jansen 已提交
2249 2250 2251 2252
	struct reada_control *reada1;
	struct reada_control *reada2;
	struct btrfs_key key_start;
	struct btrfs_key key_end;
A
Arne Jansen 已提交
2253 2254
	u64 increment = map->stripe_len;
	u64 offset;
2255 2256 2257 2258 2259
	u64 extent_logical;
	u64 extent_physical;
	u64 extent_len;
	struct btrfs_device *extent_dev;
	int extent_mirror_num;
L
Liu Bo 已提交
2260
	int stop_loop;
A
Arne Jansen 已提交
2261

D
David Woodhouse 已提交
2262 2263 2264 2265 2266 2267 2268
	if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
			 BTRFS_BLOCK_GROUP_RAID6)) {
		if (num >= nr_data_stripes(map)) {
			return 0;
		}
	}

A
Arne Jansen 已提交
2269 2270 2271 2272 2273 2274
	nstripes = length;
	offset = 0;
	do_div(nstripes, map->stripe_len);
	if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
		offset = map->stripe_len * num;
		increment = map->stripe_len * map->num_stripes;
2275
		mirror_num = 1;
A
Arne Jansen 已提交
2276 2277 2278 2279
	} else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
		int factor = map->num_stripes / map->sub_stripes;
		offset = map->stripe_len * (num / map->sub_stripes);
		increment = map->stripe_len * factor;
2280
		mirror_num = num % map->sub_stripes + 1;
A
Arne Jansen 已提交
2281 2282
	} else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
		increment = map->stripe_len;
2283
		mirror_num = num % map->num_stripes + 1;
A
Arne Jansen 已提交
2284 2285
	} else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
		increment = map->stripe_len;
2286
		mirror_num = num % map->num_stripes + 1;
A
Arne Jansen 已提交
2287 2288
	} else {
		increment = map->stripe_len;
2289
		mirror_num = 1;
A
Arne Jansen 已提交
2290 2291 2292 2293 2294 2295
	}

	path = btrfs_alloc_path();
	if (!path)
		return -ENOMEM;

2296 2297 2298 2299 2300
	/*
	 * work on commit root. The related disk blocks are static as
	 * long as COW is applied. This means, it is save to rewrite
	 * them to repair disk errors without any race conditions
	 */
A
Arne Jansen 已提交
2301 2302 2303 2304
	path->search_commit_root = 1;
	path->skip_locking = 1;

	/*
A
Arne Jansen 已提交
2305 2306 2307
	 * trigger the readahead for extent tree csum tree and wait for
	 * completion. During readahead, the scrub is officially paused
	 * to not hold off transaction commits
A
Arne Jansen 已提交
2308 2309 2310
	 */
	logical = base + offset;

2311
	wait_event(sctx->list_wait,
2312
		   atomic_read(&sctx->bios_in_flight) == 0);
2313
	scrub_blocked_if_needed(fs_info);
A
Arne Jansen 已提交
2314 2315 2316 2317 2318 2319

	/* FIXME it might be better to start readahead at commit root */
	key_start.objectid = logical;
	key_start.type = BTRFS_EXTENT_ITEM_KEY;
	key_start.offset = (u64)0;
	key_end.objectid = base + offset + nstripes * increment;
2320 2321
	key_end.type = BTRFS_METADATA_ITEM_KEY;
	key_end.offset = (u64)-1;
A
Arne Jansen 已提交
2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336
	reada1 = btrfs_reada_add(root, &key_start, &key_end);

	key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
	key_start.type = BTRFS_EXTENT_CSUM_KEY;
	key_start.offset = logical;
	key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
	key_end.type = BTRFS_EXTENT_CSUM_KEY;
	key_end.offset = base + offset + nstripes * increment;
	reada2 = btrfs_reada_add(csum_root, &key_start, &key_end);

	if (!IS_ERR(reada1))
		btrfs_reada_wait(reada1);
	if (!IS_ERR(reada2))
		btrfs_reada_wait(reada2);

A
Arne Jansen 已提交
2337 2338 2339 2340 2341

	/*
	 * collect all data csums for the stripe to avoid seeking during
	 * the scrub. This might currently (crc32) end up to be about 1MB
	 */
2342
	blk_start_plug(&plug);
A
Arne Jansen 已提交
2343 2344 2345 2346

	/*
	 * now find all extents for each stripe and scrub them
	 */
A
Arne Jansen 已提交
2347 2348
	logical = base + offset;
	physical = map->stripes[num].physical;
L
Liu Bo 已提交
2349
	logic_end = logical + increment * nstripes;
A
Arne Jansen 已提交
2350
	ret = 0;
L
Liu Bo 已提交
2351
	while (logical < logic_end) {
A
Arne Jansen 已提交
2352 2353 2354 2355
		/*
		 * canceled?
		 */
		if (atomic_read(&fs_info->scrub_cancel_req) ||
2356
		    atomic_read(&sctx->cancel_req)) {
A
Arne Jansen 已提交
2357 2358 2359 2360 2361 2362 2363 2364
			ret = -ECANCELED;
			goto out;
		}
		/*
		 * check to see if we have to pause
		 */
		if (atomic_read(&fs_info->scrub_pause_req)) {
			/* push queued extents */
2365
			atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
2366
			scrub_submit(sctx);
2367 2368 2369
			mutex_lock(&sctx->wr_ctx.wr_lock);
			scrub_wr_submit(sctx);
			mutex_unlock(&sctx->wr_ctx.wr_lock);
2370
			wait_event(sctx->list_wait,
2371
				   atomic_read(&sctx->bios_in_flight) == 0);
2372
			atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
2373
			scrub_blocked_if_needed(fs_info);
A
Arne Jansen 已提交
2374 2375 2376 2377
		}

		key.objectid = logical;
		key.type = BTRFS_EXTENT_ITEM_KEY;
L
Liu Bo 已提交
2378
		key.offset = (u64)-1;
A
Arne Jansen 已提交
2379 2380 2381 2382

		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
		if (ret < 0)
			goto out;
2383

2384
		if (ret > 0) {
A
Arne Jansen 已提交
2385 2386 2387 2388
			ret = btrfs_previous_item(root, path, 0,
						  BTRFS_EXTENT_ITEM_KEY);
			if (ret < 0)
				goto out;
2389 2390 2391 2392 2393 2394 2395 2396 2397
			if (ret > 0) {
				/* there's no smaller item, so stick with the
				 * larger one */
				btrfs_release_path(path);
				ret = btrfs_search_slot(NULL, root, &key,
							path, 0, 0);
				if (ret < 0)
					goto out;
			}
A
Arne Jansen 已提交
2398 2399
		}

L
Liu Bo 已提交
2400
		stop_loop = 0;
A
Arne Jansen 已提交
2401
		while (1) {
2402 2403
			u64 bytes;

A
Arne Jansen 已提交
2404 2405 2406 2407 2408 2409 2410 2411 2412
			l = path->nodes[0];
			slot = path->slots[0];
			if (slot >= btrfs_header_nritems(l)) {
				ret = btrfs_next_leaf(root, path);
				if (ret == 0)
					continue;
				if (ret < 0)
					goto out;

L
Liu Bo 已提交
2413
				stop_loop = 1;
A
Arne Jansen 已提交
2414 2415 2416 2417
				break;
			}
			btrfs_item_key_to_cpu(l, &key, slot);

2418 2419 2420 2421 2422 2423
			if (key.type == BTRFS_METADATA_ITEM_KEY)
				bytes = root->leafsize;
			else
				bytes = key.offset;

			if (key.objectid + bytes <= logical)
A
Arne Jansen 已提交
2424 2425
				goto next;

L
Liu Bo 已提交
2426 2427 2428
			if (key.type != BTRFS_EXTENT_ITEM_KEY &&
			    key.type != BTRFS_METADATA_ITEM_KEY)
				goto next;
A
Arne Jansen 已提交
2429

L
Liu Bo 已提交
2430 2431 2432 2433 2434 2435
			if (key.objectid >= logical + map->stripe_len) {
				/* out of this device extent */
				if (key.objectid >= logic_end)
					stop_loop = 1;
				break;
			}
A
Arne Jansen 已提交
2436 2437 2438 2439 2440 2441 2442 2443

			extent = btrfs_item_ptr(l, slot,
						struct btrfs_extent_item);
			flags = btrfs_extent_flags(l, extent);
			generation = btrfs_extent_generation(l, extent);

			if (key.objectid < logical &&
			    (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
2444 2445 2446
				btrfs_err(fs_info,
					   "scrub: tree block %llu spanning "
					   "stripes, ignored. logical=%llu",
2447
				       key.objectid, logical);
A
Arne Jansen 已提交
2448 2449 2450
				goto next;
			}

L
Liu Bo 已提交
2451 2452 2453 2454
again:
			extent_logical = key.objectid;
			extent_len = bytes;

A
Arne Jansen 已提交
2455 2456 2457
			/*
			 * trim extent to this stripe
			 */
L
Liu Bo 已提交
2458 2459 2460
			if (extent_logical < logical) {
				extent_len -= logical - extent_logical;
				extent_logical = logical;
A
Arne Jansen 已提交
2461
			}
L
Liu Bo 已提交
2462
			if (extent_logical + extent_len >
A
Arne Jansen 已提交
2463
			    logical + map->stripe_len) {
L
Liu Bo 已提交
2464 2465
				extent_len = logical + map->stripe_len -
					     extent_logical;
A
Arne Jansen 已提交
2466 2467
			}

L
Liu Bo 已提交
2468
			extent_physical = extent_logical - logical + physical;
2469 2470 2471 2472 2473 2474 2475
			extent_dev = scrub_dev;
			extent_mirror_num = mirror_num;
			if (is_dev_replace)
				scrub_remap_extent(fs_info, extent_logical,
						   extent_len, &extent_physical,
						   &extent_dev,
						   &extent_mirror_num);
L
Liu Bo 已提交
2476 2477 2478 2479 2480 2481 2482

			ret = btrfs_lookup_csums_range(csum_root, logical,
						logical + map->stripe_len - 1,
						&sctx->csum_list, 1);
			if (ret)
				goto out;

2483 2484 2485
			ret = scrub_extent(sctx, extent_logical, extent_len,
					   extent_physical, extent_dev, flags,
					   generation, extent_mirror_num,
2486
					   extent_logical - logical + physical);
A
Arne Jansen 已提交
2487 2488 2489
			if (ret)
				goto out;

2490
			scrub_free_csums(sctx);
L
Liu Bo 已提交
2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505
			if (extent_logical + extent_len <
			    key.objectid + bytes) {
				logical += increment;
				physical += map->stripe_len;

				if (logical < key.objectid + bytes) {
					cond_resched();
					goto again;
				}

				if (logical >= logic_end) {
					stop_loop = 1;
					break;
				}
			}
A
Arne Jansen 已提交
2506 2507 2508
next:
			path->slots[0]++;
		}
C
Chris Mason 已提交
2509
		btrfs_release_path(path);
A
Arne Jansen 已提交
2510 2511
		logical += increment;
		physical += map->stripe_len;
2512
		spin_lock(&sctx->stat_lock);
L
Liu Bo 已提交
2513 2514 2515 2516 2517
		if (stop_loop)
			sctx->stat.last_physical = map->stripes[num].physical +
						   length;
		else
			sctx->stat.last_physical = physical;
2518
		spin_unlock(&sctx->stat_lock);
L
Liu Bo 已提交
2519 2520
		if (stop_loop)
			break;
A
Arne Jansen 已提交
2521
	}
2522
out:
A
Arne Jansen 已提交
2523
	/* push queued extents */
2524
	scrub_submit(sctx);
2525 2526 2527
	mutex_lock(&sctx->wr_ctx.wr_lock);
	scrub_wr_submit(sctx);
	mutex_unlock(&sctx->wr_ctx.wr_lock);
A
Arne Jansen 已提交
2528

2529
	blk_finish_plug(&plug);
A
Arne Jansen 已提交
2530 2531 2532 2533
	btrfs_free_path(path);
	return ret < 0 ? ret : 0;
}

2534
static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx,
2535 2536 2537
					  struct btrfs_device *scrub_dev,
					  u64 chunk_tree, u64 chunk_objectid,
					  u64 chunk_offset, u64 length,
2538
					  u64 dev_offset, int is_dev_replace)
A
Arne Jansen 已提交
2539 2540
{
	struct btrfs_mapping_tree *map_tree =
2541
		&sctx->dev_root->fs_info->mapping_tree;
A
Arne Jansen 已提交
2542 2543 2544
	struct map_lookup *map;
	struct extent_map *em;
	int i;
2545
	int ret = 0;
A
Arne Jansen 已提交
2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561

	read_lock(&map_tree->map_tree.lock);
	em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
	read_unlock(&map_tree->map_tree.lock);

	if (!em)
		return -EINVAL;

	map = (struct map_lookup *)em->bdev;
	if (em->start != chunk_offset)
		goto out;

	if (em->len < length)
		goto out;

	for (i = 0; i < map->num_stripes; ++i) {
2562
		if (map->stripes[i].dev->bdev == scrub_dev->bdev &&
2563
		    map->stripes[i].physical == dev_offset) {
2564
			ret = scrub_stripe(sctx, map, scrub_dev, i,
2565 2566
					   chunk_offset, length,
					   is_dev_replace);
A
Arne Jansen 已提交
2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577
			if (ret)
				goto out;
		}
	}
out:
	free_extent_map(em);

	return ret;
}

static noinline_for_stack
2578
int scrub_enumerate_chunks(struct scrub_ctx *sctx,
2579 2580
			   struct btrfs_device *scrub_dev, u64 start, u64 end,
			   int is_dev_replace)
A
Arne Jansen 已提交
2581 2582 2583
{
	struct btrfs_dev_extent *dev_extent = NULL;
	struct btrfs_path *path;
2584
	struct btrfs_root *root = sctx->dev_root;
A
Arne Jansen 已提交
2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595
	struct btrfs_fs_info *fs_info = root->fs_info;
	u64 length;
	u64 chunk_tree;
	u64 chunk_objectid;
	u64 chunk_offset;
	int ret;
	int slot;
	struct extent_buffer *l;
	struct btrfs_key key;
	struct btrfs_key found_key;
	struct btrfs_block_group_cache *cache;
2596
	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
A
Arne Jansen 已提交
2597 2598 2599 2600 2601 2602 2603 2604 2605

	path = btrfs_alloc_path();
	if (!path)
		return -ENOMEM;

	path->reada = 2;
	path->search_commit_root = 1;
	path->skip_locking = 1;

2606
	key.objectid = scrub_dev->devid;
A
Arne Jansen 已提交
2607 2608 2609 2610 2611 2612
	key.offset = 0ull;
	key.type = BTRFS_DEV_EXTENT_KEY;

	while (1) {
		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
		if (ret < 0)
2613 2614 2615 2616 2617 2618 2619 2620 2621
			break;
		if (ret > 0) {
			if (path->slots[0] >=
			    btrfs_header_nritems(path->nodes[0])) {
				ret = btrfs_next_leaf(root, path);
				if (ret)
					break;
			}
		}
A
Arne Jansen 已提交
2622 2623 2624 2625 2626 2627

		l = path->nodes[0];
		slot = path->slots[0];

		btrfs_item_key_to_cpu(l, &found_key, slot);

2628
		if (found_key.objectid != scrub_dev->devid)
A
Arne Jansen 已提交
2629 2630
			break;

2631
		if (btrfs_key_type(&found_key) != BTRFS_DEV_EXTENT_KEY)
A
Arne Jansen 已提交
2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644
			break;

		if (found_key.offset >= end)
			break;

		if (found_key.offset < key.offset)
			break;

		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
		length = btrfs_dev_extent_length(l, dev_extent);

		if (found_key.offset + length <= start) {
			key.offset = found_key.offset + length;
C
Chris Mason 已提交
2645
			btrfs_release_path(path);
A
Arne Jansen 已提交
2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659
			continue;
		}

		chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
		chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
		chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);

		/*
		 * get a reference on the corresponding block group to prevent
		 * the chunk from going away while we scrub it
		 */
		cache = btrfs_lookup_block_group(fs_info, chunk_offset);
		if (!cache) {
			ret = -ENOENT;
2660
			break;
A
Arne Jansen 已提交
2661
		}
2662 2663 2664
		dev_replace->cursor_right = found_key.offset + length;
		dev_replace->cursor_left = found_key.offset;
		dev_replace->item_needs_writeback = 1;
2665
		ret = scrub_chunk(sctx, scrub_dev, chunk_tree, chunk_objectid,
2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689
				  chunk_offset, length, found_key.offset,
				  is_dev_replace);

		/*
		 * flush, submit all pending read and write bios, afterwards
		 * wait for them.
		 * Note that in the dev replace case, a read request causes
		 * write requests that are submitted in the read completion
		 * worker. Therefore in the current situation, it is required
		 * that all write requests are flushed, so that all read and
		 * write requests are really completed when bios_in_flight
		 * changes to 0.
		 */
		atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
		scrub_submit(sctx);
		mutex_lock(&sctx->wr_ctx.wr_lock);
		scrub_wr_submit(sctx);
		mutex_unlock(&sctx->wr_ctx.wr_lock);

		wait_event(sctx->list_wait,
			   atomic_read(&sctx->bios_in_flight) == 0);
		atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
		wait_event(sctx->list_wait,
			   atomic_read(&sctx->workers_pending) == 0);
2690
		scrub_blocked_if_needed(fs_info);
2691

A
Arne Jansen 已提交
2692 2693 2694
		btrfs_put_block_group(cache);
		if (ret)
			break;
2695 2696
		if (is_dev_replace &&
		    atomic64_read(&dev_replace->num_write_errors) > 0) {
2697 2698 2699 2700 2701 2702 2703
			ret = -EIO;
			break;
		}
		if (sctx->stat.malloc_errors > 0) {
			ret = -ENOMEM;
			break;
		}
A
Arne Jansen 已提交
2704

2705 2706 2707
		dev_replace->cursor_left = dev_replace->cursor_right;
		dev_replace->item_needs_writeback = 1;

A
Arne Jansen 已提交
2708
		key.offset = found_key.offset + length;
C
Chris Mason 已提交
2709
		btrfs_release_path(path);
A
Arne Jansen 已提交
2710 2711 2712
	}

	btrfs_free_path(path);
2713 2714 2715 2716 2717 2718

	/*
	 * ret can still be 1 from search_slot or next_leaf,
	 * that's not an error
	 */
	return ret < 0 ? ret : 0;
A
Arne Jansen 已提交
2719 2720
}

2721 2722
static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx,
					   struct btrfs_device *scrub_dev)
A
Arne Jansen 已提交
2723 2724 2725 2726 2727
{
	int	i;
	u64	bytenr;
	u64	gen;
	int	ret;
2728
	struct btrfs_root *root = sctx->dev_root;
A
Arne Jansen 已提交
2729

2730
	if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
2731 2732
		return -EIO;

A
Arne Jansen 已提交
2733 2734 2735 2736
	gen = root->fs_info->last_trans_committed;

	for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
		bytenr = btrfs_sb_offset(i);
2737
		if (bytenr + BTRFS_SUPER_INFO_SIZE > scrub_dev->total_bytes)
A
Arne Jansen 已提交
2738 2739
			break;

2740
		ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr,
2741
				  scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i,
2742
				  NULL, 1, bytenr);
A
Arne Jansen 已提交
2743 2744 2745
		if (ret)
			return ret;
	}
2746
	wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
A
Arne Jansen 已提交
2747 2748 2749 2750 2751 2752 2753

	return 0;
}

/*
 * get a reference count on fs_info->scrub_workers. start worker if necessary
 */
2754 2755
static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info,
						int is_dev_replace)
A
Arne Jansen 已提交
2756
{
2757
	int ret = 0;
A
Arne Jansen 已提交
2758

A
Arne Jansen 已提交
2759
	if (fs_info->scrub_workers_refcnt == 0) {
2760 2761 2762 2763 2764 2765 2766
		if (is_dev_replace)
			btrfs_init_workers(&fs_info->scrub_workers, "scrub", 1,
					&fs_info->generic_worker);
		else
			btrfs_init_workers(&fs_info->scrub_workers, "scrub",
					fs_info->thread_pool_size,
					&fs_info->generic_worker);
A
Arne Jansen 已提交
2767
		fs_info->scrub_workers.idle_thresh = 4;
2768 2769 2770
		ret = btrfs_start_workers(&fs_info->scrub_workers);
		if (ret)
			goto out;
2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784
		btrfs_init_workers(&fs_info->scrub_wr_completion_workers,
				   "scrubwrc",
				   fs_info->thread_pool_size,
				   &fs_info->generic_worker);
		fs_info->scrub_wr_completion_workers.idle_thresh = 2;
		ret = btrfs_start_workers(
				&fs_info->scrub_wr_completion_workers);
		if (ret)
			goto out;
		btrfs_init_workers(&fs_info->scrub_nocow_workers, "scrubnc", 1,
				   &fs_info->generic_worker);
		ret = btrfs_start_workers(&fs_info->scrub_nocow_workers);
		if (ret)
			goto out;
A
Arne Jansen 已提交
2785
	}
A
Arne Jansen 已提交
2786
	++fs_info->scrub_workers_refcnt;
2787 2788
out:
	return ret;
A
Arne Jansen 已提交
2789 2790
}

2791
static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info)
A
Arne Jansen 已提交
2792
{
2793
	if (--fs_info->scrub_workers_refcnt == 0) {
A
Arne Jansen 已提交
2794
		btrfs_stop_workers(&fs_info->scrub_workers);
2795 2796 2797
		btrfs_stop_workers(&fs_info->scrub_wr_completion_workers);
		btrfs_stop_workers(&fs_info->scrub_nocow_workers);
	}
A
Arne Jansen 已提交
2798 2799 2800
	WARN_ON(fs_info->scrub_workers_refcnt < 0);
}

2801 2802
int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start,
		    u64 end, struct btrfs_scrub_progress *progress,
2803
		    int readonly, int is_dev_replace)
A
Arne Jansen 已提交
2804
{
2805
	struct scrub_ctx *sctx;
A
Arne Jansen 已提交
2806 2807 2808
	int ret;
	struct btrfs_device *dev;

2809
	if (btrfs_fs_closing(fs_info))
A
Arne Jansen 已提交
2810 2811 2812 2813 2814
		return -EINVAL;

	/*
	 * check some assumptions
	 */
2815
	if (fs_info->chunk_root->nodesize != fs_info->chunk_root->leafsize) {
2816 2817
		btrfs_err(fs_info,
			   "scrub: size assumption nodesize == leafsize (%d == %d) fails",
2818 2819
		       fs_info->chunk_root->nodesize,
		       fs_info->chunk_root->leafsize);
2820 2821 2822
		return -EINVAL;
	}

2823
	if (fs_info->chunk_root->nodesize > BTRFS_STRIPE_LEN) {
2824 2825 2826 2827 2828
		/*
		 * in this case scrub is unable to calculate the checksum
		 * the way scrub is implemented. Do not handle this
		 * situation at all because it won't ever happen.
		 */
2829 2830
		btrfs_err(fs_info,
			   "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails",
2831
		       fs_info->chunk_root->nodesize, BTRFS_STRIPE_LEN);
2832 2833 2834
		return -EINVAL;
	}

2835
	if (fs_info->chunk_root->sectorsize != PAGE_SIZE) {
2836
		/* not supported for data w/o checksums */
2837 2838 2839
		btrfs_err(fs_info,
			   "scrub: size assumption sectorsize != PAGE_SIZE "
			   "(%d != %lu) fails",
2840
		       fs_info->chunk_root->sectorsize, PAGE_SIZE);
A
Arne Jansen 已提交
2841 2842 2843
		return -EINVAL;
	}

2844 2845 2846 2847 2848 2849 2850 2851
	if (fs_info->chunk_root->nodesize >
	    PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK ||
	    fs_info->chunk_root->sectorsize >
	    PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) {
		/*
		 * would exhaust the array bounds of pagev member in
		 * struct scrub_block
		 */
2852 2853
		btrfs_err(fs_info, "scrub: size assumption nodesize and sectorsize "
			   "<= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails",
2854 2855 2856 2857 2858 2859 2860
		       fs_info->chunk_root->nodesize,
		       SCRUB_MAX_PAGES_PER_BLOCK,
		       fs_info->chunk_root->sectorsize,
		       SCRUB_MAX_PAGES_PER_BLOCK);
		return -EINVAL;
	}

A
Arne Jansen 已提交
2861

2862 2863
	mutex_lock(&fs_info->fs_devices->device_list_mutex);
	dev = btrfs_find_device(fs_info, devid, NULL, NULL);
2864
	if (!dev || (dev->missing && !is_dev_replace)) {
2865
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
A
Arne Jansen 已提交
2866 2867 2868
		return -ENODEV;
	}

2869
	mutex_lock(&fs_info->scrub_lock);
2870
	if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) {
A
Arne Jansen 已提交
2871
		mutex_unlock(&fs_info->scrub_lock);
2872 2873
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
		return -EIO;
A
Arne Jansen 已提交
2874 2875
	}

2876 2877 2878 2879 2880
	btrfs_dev_replace_lock(&fs_info->dev_replace);
	if (dev->scrub_device ||
	    (!is_dev_replace &&
	     btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) {
		btrfs_dev_replace_unlock(&fs_info->dev_replace);
A
Arne Jansen 已提交
2881
		mutex_unlock(&fs_info->scrub_lock);
2882
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
A
Arne Jansen 已提交
2883 2884
		return -EINPROGRESS;
	}
2885
	btrfs_dev_replace_unlock(&fs_info->dev_replace);
2886 2887 2888 2889 2890 2891 2892 2893

	ret = scrub_workers_get(fs_info, is_dev_replace);
	if (ret) {
		mutex_unlock(&fs_info->scrub_lock);
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
		return ret;
	}

2894
	sctx = scrub_setup_ctx(dev, is_dev_replace);
2895
	if (IS_ERR(sctx)) {
A
Arne Jansen 已提交
2896
		mutex_unlock(&fs_info->scrub_lock);
2897 2898
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
		scrub_workers_put(fs_info);
2899
		return PTR_ERR(sctx);
A
Arne Jansen 已提交
2900
	}
2901 2902
	sctx->readonly = readonly;
	dev->scrub_device = sctx;
2903
	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
A
Arne Jansen 已提交
2904

2905 2906 2907 2908
	/*
	 * checking @scrub_pause_req here, we can avoid
	 * race between committing transaction and scrubbing.
	 */
2909
	__scrub_blocked_if_needed(fs_info);
A
Arne Jansen 已提交
2910 2911 2912
	atomic_inc(&fs_info->scrubs_running);
	mutex_unlock(&fs_info->scrub_lock);

2913
	if (!is_dev_replace) {
2914 2915 2916 2917
		/*
		 * by holding device list mutex, we can
		 * kick off writing super in log tree sync.
		 */
2918
		mutex_lock(&fs_info->fs_devices->device_list_mutex);
2919
		ret = scrub_supers(sctx, dev);
2920
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2921
	}
A
Arne Jansen 已提交
2922 2923

	if (!ret)
2924 2925
		ret = scrub_enumerate_chunks(sctx, dev, start, end,
					     is_dev_replace);
A
Arne Jansen 已提交
2926

2927
	wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
A
Arne Jansen 已提交
2928 2929 2930
	atomic_dec(&fs_info->scrubs_running);
	wake_up(&fs_info->scrub_pause_wait);

2931
	wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0);
2932

A
Arne Jansen 已提交
2933
	if (progress)
2934
		memcpy(progress, &sctx->stat, sizeof(*progress));
A
Arne Jansen 已提交
2935 2936 2937

	mutex_lock(&fs_info->scrub_lock);
	dev->scrub_device = NULL;
2938
	scrub_workers_put(fs_info);
A
Arne Jansen 已提交
2939 2940
	mutex_unlock(&fs_info->scrub_lock);

2941
	scrub_free_ctx(sctx);
A
Arne Jansen 已提交
2942 2943 2944 2945

	return ret;
}

2946
void btrfs_scrub_pause(struct btrfs_root *root)
A
Arne Jansen 已提交
2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962
{
	struct btrfs_fs_info *fs_info = root->fs_info;

	mutex_lock(&fs_info->scrub_lock);
	atomic_inc(&fs_info->scrub_pause_req);
	while (atomic_read(&fs_info->scrubs_paused) !=
	       atomic_read(&fs_info->scrubs_running)) {
		mutex_unlock(&fs_info->scrub_lock);
		wait_event(fs_info->scrub_pause_wait,
			   atomic_read(&fs_info->scrubs_paused) ==
			   atomic_read(&fs_info->scrubs_running));
		mutex_lock(&fs_info->scrub_lock);
	}
	mutex_unlock(&fs_info->scrub_lock);
}

2963
void btrfs_scrub_continue(struct btrfs_root *root)
A
Arne Jansen 已提交
2964 2965 2966 2967 2968 2969 2970
{
	struct btrfs_fs_info *fs_info = root->fs_info;

	atomic_dec(&fs_info->scrub_pause_req);
	wake_up(&fs_info->scrub_pause_wait);
}

2971
int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info)
A
Arne Jansen 已提交
2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991
{
	mutex_lock(&fs_info->scrub_lock);
	if (!atomic_read(&fs_info->scrubs_running)) {
		mutex_unlock(&fs_info->scrub_lock);
		return -ENOTCONN;
	}

	atomic_inc(&fs_info->scrub_cancel_req);
	while (atomic_read(&fs_info->scrubs_running)) {
		mutex_unlock(&fs_info->scrub_lock);
		wait_event(fs_info->scrub_pause_wait,
			   atomic_read(&fs_info->scrubs_running) == 0);
		mutex_lock(&fs_info->scrub_lock);
	}
	atomic_dec(&fs_info->scrub_cancel_req);
	mutex_unlock(&fs_info->scrub_lock);

	return 0;
}

2992 2993
int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info,
			   struct btrfs_device *dev)
2994
{
2995
	struct scrub_ctx *sctx;
A
Arne Jansen 已提交
2996 2997

	mutex_lock(&fs_info->scrub_lock);
2998 2999
	sctx = dev->scrub_device;
	if (!sctx) {
A
Arne Jansen 已提交
3000 3001 3002
		mutex_unlock(&fs_info->scrub_lock);
		return -ENOTCONN;
	}
3003
	atomic_inc(&sctx->cancel_req);
A
Arne Jansen 已提交
3004 3005 3006 3007 3008 3009 3010 3011 3012 3013
	while (dev->scrub_device) {
		mutex_unlock(&fs_info->scrub_lock);
		wait_event(fs_info->scrub_pause_wait,
			   dev->scrub_device == NULL);
		mutex_lock(&fs_info->scrub_lock);
	}
	mutex_unlock(&fs_info->scrub_lock);

	return 0;
}
S
Stefan Behrens 已提交
3014

A
Arne Jansen 已提交
3015 3016 3017 3018
int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
			 struct btrfs_scrub_progress *progress)
{
	struct btrfs_device *dev;
3019
	struct scrub_ctx *sctx = NULL;
A
Arne Jansen 已提交
3020 3021

	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3022
	dev = btrfs_find_device(root->fs_info, devid, NULL, NULL);
A
Arne Jansen 已提交
3023
	if (dev)
3024 3025 3026
		sctx = dev->scrub_device;
	if (sctx)
		memcpy(progress, &sctx->stat, sizeof(*progress));
A
Arne Jansen 已提交
3027 3028
	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);

3029
	return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV;
A
Arne Jansen 已提交
3030
}
3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107

static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
			       u64 extent_logical, u64 extent_len,
			       u64 *extent_physical,
			       struct btrfs_device **extent_dev,
			       int *extent_mirror_num)
{
	u64 mapped_length;
	struct btrfs_bio *bbio = NULL;
	int ret;

	mapped_length = extent_len;
	ret = btrfs_map_block(fs_info, READ, extent_logical,
			      &mapped_length, &bbio, 0);
	if (ret || !bbio || mapped_length < extent_len ||
	    !bbio->stripes[0].dev->bdev) {
		kfree(bbio);
		return;
	}

	*extent_physical = bbio->stripes[0].physical;
	*extent_mirror_num = bbio->mirror_num;
	*extent_dev = bbio->stripes[0].dev;
	kfree(bbio);
}

static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
			      struct scrub_wr_ctx *wr_ctx,
			      struct btrfs_fs_info *fs_info,
			      struct btrfs_device *dev,
			      int is_dev_replace)
{
	WARN_ON(wr_ctx->wr_curr_bio != NULL);

	mutex_init(&wr_ctx->wr_lock);
	wr_ctx->wr_curr_bio = NULL;
	if (!is_dev_replace)
		return 0;

	WARN_ON(!dev->bdev);
	wr_ctx->pages_per_wr_bio = min_t(int, SCRUB_PAGES_PER_WR_BIO,
					 bio_get_nr_vecs(dev->bdev));
	wr_ctx->tgtdev = dev;
	atomic_set(&wr_ctx->flush_all_writes, 0);
	return 0;
}

static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx)
{
	mutex_lock(&wr_ctx->wr_lock);
	kfree(wr_ctx->wr_curr_bio);
	wr_ctx->wr_curr_bio = NULL;
	mutex_unlock(&wr_ctx->wr_lock);
}

static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
			    int mirror_num, u64 physical_for_dev_replace)
{
	struct scrub_copy_nocow_ctx *nocow_ctx;
	struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;

	nocow_ctx = kzalloc(sizeof(*nocow_ctx), GFP_NOFS);
	if (!nocow_ctx) {
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		spin_unlock(&sctx->stat_lock);
		return -ENOMEM;
	}

	scrub_pending_trans_workers_inc(sctx);

	nocow_ctx->sctx = sctx;
	nocow_ctx->logical = logical;
	nocow_ctx->len = len;
	nocow_ctx->mirror_num = mirror_num;
	nocow_ctx->physical_for_dev_replace = physical_for_dev_replace;
	nocow_ctx->work.func = copy_nocow_pages_worker;
3108
	INIT_LIST_HEAD(&nocow_ctx->inodes);
3109 3110 3111 3112 3113 3114
	btrfs_queue_worker(&fs_info->scrub_nocow_workers,
			   &nocow_ctx->work);

	return 0;
}

3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131
static int record_inode_for_nocow(u64 inum, u64 offset, u64 root, void *ctx)
{
	struct scrub_copy_nocow_ctx *nocow_ctx = ctx;
	struct scrub_nocow_inode *nocow_inode;

	nocow_inode = kzalloc(sizeof(*nocow_inode), GFP_NOFS);
	if (!nocow_inode)
		return -ENOMEM;
	nocow_inode->inum = inum;
	nocow_inode->offset = offset;
	nocow_inode->root = root;
	list_add_tail(&nocow_inode->list, &nocow_ctx->inodes);
	return 0;
}

#define COPY_COMPLETE 1

3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166
static void copy_nocow_pages_worker(struct btrfs_work *work)
{
	struct scrub_copy_nocow_ctx *nocow_ctx =
		container_of(work, struct scrub_copy_nocow_ctx, work);
	struct scrub_ctx *sctx = nocow_ctx->sctx;
	u64 logical = nocow_ctx->logical;
	u64 len = nocow_ctx->len;
	int mirror_num = nocow_ctx->mirror_num;
	u64 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
	int ret;
	struct btrfs_trans_handle *trans = NULL;
	struct btrfs_fs_info *fs_info;
	struct btrfs_path *path;
	struct btrfs_root *root;
	int not_written = 0;

	fs_info = sctx->dev_root->fs_info;
	root = fs_info->extent_root;

	path = btrfs_alloc_path();
	if (!path) {
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		spin_unlock(&sctx->stat_lock);
		not_written = 1;
		goto out;
	}

	trans = btrfs_join_transaction(root);
	if (IS_ERR(trans)) {
		not_written = 1;
		goto out;
	}

	ret = iterate_inodes_from_logical(logical, fs_info, path,
3167
					  record_inode_for_nocow, nocow_ctx);
3168
	if (ret != 0 && ret != -ENOENT) {
3169 3170
		btrfs_warn(fs_info, "iterate_inodes_from_logical() failed: log %llu, "
			"phys %llu, len %llu, mir %u, ret %d",
3171 3172
			logical, physical_for_dev_replace, len, mirror_num,
			ret);
3173 3174 3175 3176
		not_written = 1;
		goto out;
	}

3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194
	btrfs_end_transaction(trans, root);
	trans = NULL;
	while (!list_empty(&nocow_ctx->inodes)) {
		struct scrub_nocow_inode *entry;
		entry = list_first_entry(&nocow_ctx->inodes,
					 struct scrub_nocow_inode,
					 list);
		list_del_init(&entry->list);
		ret = copy_nocow_pages_for_inode(entry->inum, entry->offset,
						 entry->root, nocow_ctx);
		kfree(entry);
		if (ret == COPY_COMPLETE) {
			ret = 0;
			break;
		} else if (ret) {
			break;
		}
	}
3195
out:
3196 3197 3198 3199 3200 3201 3202 3203
	while (!list_empty(&nocow_ctx->inodes)) {
		struct scrub_nocow_inode *entry;
		entry = list_first_entry(&nocow_ctx->inodes,
					 struct scrub_nocow_inode,
					 list);
		list_del_init(&entry->list);
		kfree(entry);
	}
3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215
	if (trans && !IS_ERR(trans))
		btrfs_end_transaction(trans, root);
	if (not_written)
		btrfs_dev_replace_stats_inc(&fs_info->dev_replace.
					    num_uncorrectable_read_errors);

	btrfs_free_path(path);
	kfree(nocow_ctx);

	scrub_pending_trans_workers_dec(sctx);
}

3216 3217
static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
				      struct scrub_copy_nocow_ctx *nocow_ctx)
3218
{
3219
	struct btrfs_fs_info *fs_info = nocow_ctx->sctx->dev_root->fs_info;
3220
	struct btrfs_key key;
3221 3222
	struct inode *inode;
	struct page *page;
3223
	struct btrfs_root *local_root;
3224 3225 3226 3227
	struct btrfs_ordered_extent *ordered;
	struct extent_map *em;
	struct extent_state *cached_state = NULL;
	struct extent_io_tree *io_tree;
3228
	u64 physical_for_dev_replace;
3229 3230
	u64 len = nocow_ctx->len;
	u64 lockstart = offset, lockend = offset + len - 1;
3231
	unsigned long index;
3232
	int srcu_index;
3233 3234
	int ret = 0;
	int err = 0;
3235 3236 3237 3238

	key.objectid = root;
	key.type = BTRFS_ROOT_ITEM_KEY;
	key.offset = (u64)-1;
3239 3240 3241

	srcu_index = srcu_read_lock(&fs_info->subvol_srcu);

3242
	local_root = btrfs_read_fs_root_no_name(fs_info, &key);
3243 3244
	if (IS_ERR(local_root)) {
		srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
3245
		return PTR_ERR(local_root);
3246
	}
3247 3248 3249 3250 3251

	key.type = BTRFS_INODE_ITEM_KEY;
	key.objectid = inum;
	key.offset = 0;
	inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
3252
	srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
3253 3254 3255
	if (IS_ERR(inode))
		return PTR_ERR(inode);

3256 3257 3258 3259
	/* Avoid truncate/dio/punch hole.. */
	mutex_lock(&inode->i_mutex);
	inode_dio_wait(inode);

3260
	physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286
	io_tree = &BTRFS_I(inode)->io_tree;

	lock_extent_bits(io_tree, lockstart, lockend, 0, &cached_state);
	ordered = btrfs_lookup_ordered_range(inode, lockstart, len);
	if (ordered) {
		btrfs_put_ordered_extent(ordered);
		goto out_unlock;
	}

	em = btrfs_get_extent(inode, NULL, 0, lockstart, len, 0);
	if (IS_ERR(em)) {
		ret = PTR_ERR(em);
		goto out_unlock;
	}

	/*
	 * This extent does not actually cover the logical extent anymore,
	 * move on to the next inode.
	 */
	if (em->block_start > nocow_ctx->logical ||
	    em->block_start + em->block_len < nocow_ctx->logical + len) {
		free_extent_map(em);
		goto out_unlock;
	}
	free_extent_map(em);

3287 3288
	while (len >= PAGE_CACHE_SIZE) {
		index = offset >> PAGE_CACHE_SHIFT;
3289
again:
3290 3291
		page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
		if (!page) {
3292
			btrfs_err(fs_info, "find_or_create_page() failed");
3293
			ret = -ENOMEM;
3294
			goto out;
3295 3296 3297 3298 3299 3300 3301
		}

		if (PageUptodate(page)) {
			if (PageDirty(page))
				goto next_page;
		} else {
			ClearPageError(page);
3302 3303 3304
			err = extent_read_full_page_nolock(io_tree, page,
							   btrfs_get_extent,
							   nocow_ctx->mirror_num);
3305 3306
			if (err) {
				ret = err;
3307 3308
				goto next_page;
			}
3309

3310
			lock_page(page);
3311 3312 3313 3314 3315 3316 3317
			/*
			 * If the page has been remove from the page cache,
			 * the data on it is meaningless, because it may be
			 * old one, the new data may be written into the new
			 * page in the page cache.
			 */
			if (page->mapping != inode->i_mapping) {
3318
				unlock_page(page);
3319 3320 3321
				page_cache_release(page);
				goto again;
			}
3322 3323 3324 3325 3326
			if (!PageUptodate(page)) {
				ret = -EIO;
				goto next_page;
			}
		}
3327 3328 3329 3330
		err = write_page_nocow(nocow_ctx->sctx,
				       physical_for_dev_replace, page);
		if (err)
			ret = err;
3331
next_page:
3332 3333 3334 3335 3336 3337
		unlock_page(page);
		page_cache_release(page);

		if (ret)
			break;

3338 3339 3340 3341
		offset += PAGE_CACHE_SIZE;
		physical_for_dev_replace += PAGE_CACHE_SIZE;
		len -= PAGE_CACHE_SIZE;
	}
3342 3343 3344 3345
	ret = COPY_COMPLETE;
out_unlock:
	unlock_extent_cached(io_tree, lockstart, lockend, &cached_state,
			     GFP_NOFS);
3346
out:
3347
	mutex_unlock(&inode->i_mutex);
3348
	iput(inode);
3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363
	return ret;
}

static int write_page_nocow(struct scrub_ctx *sctx,
			    u64 physical_for_dev_replace, struct page *page)
{
	struct bio *bio;
	struct btrfs_device *dev;
	int ret;

	dev = sctx->wr_ctx.tgtdev;
	if (!dev)
		return -EIO;
	if (!dev->bdev) {
		printk_ratelimited(KERN_WARNING
3364
			"BTRFS: scrub write_page_nocow(bdev == NULL) is unexpected!\n");
3365 3366
		return -EIO;
	}
3367
	bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384
	if (!bio) {
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		spin_unlock(&sctx->stat_lock);
		return -ENOMEM;
	}
	bio->bi_size = 0;
	bio->bi_sector = physical_for_dev_replace >> 9;
	bio->bi_bdev = dev->bdev;
	ret = bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
	if (ret != PAGE_CACHE_SIZE) {
leave_with_eio:
		bio_put(bio);
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
		return -EIO;
	}

3385
	if (btrfsic_submit_bio_wait(WRITE_SYNC, bio))
3386 3387 3388 3389 3390
		goto leave_with_eio;

	bio_put(bio);
	return 0;
}