scrub.c 121.0 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>
21
#include <linux/sched/mm.h>
A
Arne Jansen 已提交
22 23 24 25
#include "ctree.h"
#include "volumes.h"
#include "disk-io.h"
#include "ordered-data.h"
26
#include "transaction.h"
27
#include "backref.h"
28
#include "extent_io.h"
29
#include "dev-replace.h"
30
#include "check-integrity.h"
31
#include "rcu-string.h"
D
David Woodhouse 已提交
32
#include "raid56.h"
A
Arne Jansen 已提交
33 34 35 36 37 38 39 40 41 42 43 44 45 46

/*
 * 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
 */

47
struct scrub_block;
48
struct scrub_ctx;
A
Arne Jansen 已提交
49

50 51 52 53 54 55 56 57 58
/*
 * 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 */
59 60 61 62 63 64

/*
 * 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.
 */
65
#define SCRUB_MAX_PAGES_PER_BLOCK	16	/* 64k per node/leaf/sector */
A
Arne Jansen 已提交
66

67
struct scrub_recover {
68
	refcount_t		refs;
69 70 71 72
	struct btrfs_bio	*bbio;
	u64			map_length;
};

A
Arne Jansen 已提交
73
struct scrub_page {
74 75
	struct scrub_block	*sblock;
	struct page		*page;
76
	struct btrfs_device	*dev;
77
	struct list_head	list;
A
Arne Jansen 已提交
78 79
	u64			flags;  /* extent flags */
	u64			generation;
80 81
	u64			logical;
	u64			physical;
82
	u64			physical_for_dev_replace;
83
	atomic_t		refs;
84 85 86 87 88
	struct {
		unsigned int	mirror_num:8;
		unsigned int	have_csum:1;
		unsigned int	io_error:1;
	};
A
Arne Jansen 已提交
89
	u8			csum[BTRFS_CSUM_SIZE];
90 91

	struct scrub_recover	*recover;
A
Arne Jansen 已提交
92 93 94 95
};

struct scrub_bio {
	int			index;
96
	struct scrub_ctx	*sctx;
97
	struct btrfs_device	*dev;
A
Arne Jansen 已提交
98 99 100 101
	struct bio		*bio;
	int			err;
	u64			logical;
	u64			physical;
102 103 104 105 106
#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
107
	int			page_count;
A
Arne Jansen 已提交
108 109 110 111
	int			next_free;
	struct btrfs_work	work;
};

112
struct scrub_block {
113
	struct scrub_page	*pagev[SCRUB_MAX_PAGES_PER_BLOCK];
114 115
	int			page_count;
	atomic_t		outstanding_pages;
116
	refcount_t		refs; /* free mem on transition to zero */
117
	struct scrub_ctx	*sctx;
118
	struct scrub_parity	*sparity;
119 120 121 122
	struct {
		unsigned int	header_error:1;
		unsigned int	checksum_error:1;
		unsigned int	no_io_error_seen:1;
123
		unsigned int	generation_error:1; /* also sets header_error */
124 125 126 127

		/* The following is for the data used to check parity */
		/* It is for the data with checksum */
		unsigned int	data_corrected:1;
128
	};
129
	struct btrfs_work	work;
130 131
};

132 133 134 135 136 137 138 139 140 141 142 143
/* Used for the chunks with parity stripe such RAID5/6 */
struct scrub_parity {
	struct scrub_ctx	*sctx;

	struct btrfs_device	*scrub_dev;

	u64			logic_start;

	u64			logic_end;

	int			nsectors;

144
	u64			stripe_len;
145

146
	refcount_t		refs;
147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164

	struct list_head	spages;

	/* Work of parity check and repair */
	struct btrfs_work	work;

	/* Mark the parity blocks which have data */
	unsigned long		*dbitmap;

	/*
	 * Mark the parity blocks which have data, but errors happen when
	 * read data or check data
	 */
	unsigned long		*ebitmap;

	unsigned long		bitmap[0];
};

165
struct scrub_ctx {
166
	struct scrub_bio	*bios[SCRUB_BIOS_PER_SCTX];
167
	struct btrfs_fs_info	*fs_info;
A
Arne Jansen 已提交
168 169
	int			first_free;
	int			curr;
170 171
	atomic_t		bios_in_flight;
	atomic_t		workers_pending;
A
Arne Jansen 已提交
172 173 174 175 176
	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
177
	int			readonly;
178
	int			pages_per_rd_bio;
179 180

	int			is_dev_replace;
181 182 183 184 185 186

	struct scrub_bio        *wr_curr_bio;
	struct mutex            wr_lock;
	int                     pages_per_wr_bio; /* <= SCRUB_PAGES_PER_WR_BIO */
	atomic_t                flush_all_writes;
	struct btrfs_device     *wr_tgtdev;
187

A
Arne Jansen 已提交
188 189 190 191 192
	/*
	 * statistics
	 */
	struct btrfs_scrub_progress stat;
	spinlock_t		stat_lock;
193 194 195 196 197 198 199 200

	/*
	 * Use a ref counter to avoid use-after-free issues. Scrub workers
	 * decrement bios_in_flight and workers_pending and then do a wakeup
	 * on the list_wait wait queue. We must ensure the main scrub task
	 * doesn't free the scrub context before or while the workers are
	 * doing the wakeup() call.
	 */
201
	refcount_t              refs;
A
Arne Jansen 已提交
202 203
};

204
struct scrub_fixup_nodatasum {
205
	struct scrub_ctx	*sctx;
206
	struct btrfs_device	*dev;
207 208 209 210 211 212
	u64			logical;
	struct btrfs_root	*root;
	struct btrfs_work	work;
	int			mirror_num;
};

213 214 215 216 217 218 219
struct scrub_nocow_inode {
	u64			inum;
	u64			offset;
	u64			root;
	struct list_head	list;
};

220 221 222 223 224 225
struct scrub_copy_nocow_ctx {
	struct scrub_ctx	*sctx;
	u64			logical;
	u64			len;
	int			mirror_num;
	u64			physical_for_dev_replace;
226
	struct list_head	inodes;
227 228 229
	struct btrfs_work	work;
};

230 231 232 233 234 235 236 237 238
struct scrub_warning {
	struct btrfs_path	*path;
	u64			extent_item_size;
	const char		*errstr;
	sector_t		sector;
	u64			logical;
	struct btrfs_device	*dev;
};

239 240 241 242 243 244 245
struct full_stripe_lock {
	struct rb_node node;
	u64 logical;
	u64 refs;
	struct mutex mutex;
};

246 247 248 249
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);
250
static int scrub_handle_errored_block(struct scrub_block *sblock_to_check);
251
static int scrub_setup_recheck_block(struct scrub_block *original_sblock,
252
				     struct scrub_block *sblocks_for_recheck);
253
static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
254 255
				struct scrub_block *sblock,
				int retry_failed_mirror);
256
static void scrub_recheck_block_checksum(struct scrub_block *sblock);
257
static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
258
					     struct scrub_block *sblock_good);
259 260 261
static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
					    struct scrub_block *sblock_good,
					    int page_num, int force_write);
262 263 264
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);
265 266 267 268 269
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);
270 271
static void scrub_page_get(struct scrub_page *spage);
static void scrub_page_put(struct scrub_page *spage);
272 273
static void scrub_parity_get(struct scrub_parity *sparity);
static void scrub_parity_put(struct scrub_parity *sparity);
274 275
static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
				    struct scrub_page *spage);
276
static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
277
		       u64 physical, struct btrfs_device *dev, u64 flags,
278 279
		       u64 gen, int mirror_num, u8 *csum, int force,
		       u64 physical_for_dev_replace);
280
static void scrub_bio_end_io(struct bio *bio);
281 282
static void scrub_bio_end_io_worker(struct btrfs_work *work);
static void scrub_block_complete(struct scrub_block *sblock);
283 284 285 286 287 288 289 290
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_add_page_to_wr_bio(struct scrub_ctx *sctx,
				    struct scrub_page *spage);
static void scrub_wr_submit(struct scrub_ctx *sctx);
291
static void scrub_wr_bio_end_io(struct bio *bio);
292 293 294 295
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,
296
				      struct scrub_copy_nocow_ctx *ctx);
297 298 299
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);
300
static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
301
static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
302
static void scrub_put_ctx(struct scrub_ctx *sctx);
S
Stefan Behrens 已提交
303 304


305 306
static void scrub_pending_bio_inc(struct scrub_ctx *sctx)
{
307
	refcount_inc(&sctx->refs);
308 309 310 311 312 313 314
	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);
315
	scrub_put_ctx(sctx);
316 317
}

318
static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
319 320 321 322 323 324 325 326 327
{
	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);
	}
}

328
static void scrub_pause_on(struct btrfs_fs_info *fs_info)
329 330 331
{
	atomic_inc(&fs_info->scrubs_paused);
	wake_up(&fs_info->scrub_pause_wait);
332
}
333

334 335
static void scrub_pause_off(struct btrfs_fs_info *fs_info)
{
336 337 338 339 340 341 342 343
	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);
}

344 345 346 347 348 349
static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
{
	scrub_pause_on(fs_info);
	scrub_pause_off(fs_info);
}

350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565
/*
 * Insert new full stripe lock into full stripe locks tree
 *
 * Return pointer to existing or newly inserted full_stripe_lock structure if
 * everything works well.
 * Return ERR_PTR(-ENOMEM) if we failed to allocate memory
 *
 * NOTE: caller must hold full_stripe_locks_root->lock before calling this
 * function
 */
static struct full_stripe_lock *insert_full_stripe_lock(
		struct btrfs_full_stripe_locks_tree *locks_root,
		u64 fstripe_logical)
{
	struct rb_node **p;
	struct rb_node *parent = NULL;
	struct full_stripe_lock *entry;
	struct full_stripe_lock *ret;

	WARN_ON(!mutex_is_locked(&locks_root->lock));

	p = &locks_root->root.rb_node;
	while (*p) {
		parent = *p;
		entry = rb_entry(parent, struct full_stripe_lock, node);
		if (fstripe_logical < entry->logical) {
			p = &(*p)->rb_left;
		} else if (fstripe_logical > entry->logical) {
			p = &(*p)->rb_right;
		} else {
			entry->refs++;
			return entry;
		}
	}

	/* Insert new lock */
	ret = kmalloc(sizeof(*ret), GFP_KERNEL);
	if (!ret)
		return ERR_PTR(-ENOMEM);
	ret->logical = fstripe_logical;
	ret->refs = 1;
	mutex_init(&ret->mutex);

	rb_link_node(&ret->node, parent, p);
	rb_insert_color(&ret->node, &locks_root->root);
	return ret;
}

/*
 * Search for a full stripe lock of a block group
 *
 * Return pointer to existing full stripe lock if found
 * Return NULL if not found
 */
static struct full_stripe_lock *search_full_stripe_lock(
		struct btrfs_full_stripe_locks_tree *locks_root,
		u64 fstripe_logical)
{
	struct rb_node *node;
	struct full_stripe_lock *entry;

	WARN_ON(!mutex_is_locked(&locks_root->lock));

	node = locks_root->root.rb_node;
	while (node) {
		entry = rb_entry(node, struct full_stripe_lock, node);
		if (fstripe_logical < entry->logical)
			node = node->rb_left;
		else if (fstripe_logical > entry->logical)
			node = node->rb_right;
		else
			return entry;
	}
	return NULL;
}

/*
 * Helper to get full stripe logical from a normal bytenr.
 *
 * Caller must ensure @cache is a RAID56 block group.
 */
static u64 get_full_stripe_logical(struct btrfs_block_group_cache *cache,
				   u64 bytenr)
{
	u64 ret;

	/*
	 * Due to chunk item size limit, full stripe length should not be
	 * larger than U32_MAX. Just a sanity check here.
	 */
	WARN_ON_ONCE(cache->full_stripe_len >= U32_MAX);

	/*
	 * round_down() can only handle power of 2, while RAID56 full
	 * stripe length can be 64KiB * n, so we need to manually round down.
	 */
	ret = div64_u64(bytenr - cache->key.objectid, cache->full_stripe_len) *
		cache->full_stripe_len + cache->key.objectid;
	return ret;
}

/*
 * Lock a full stripe to avoid concurrency of recovery and read
 *
 * It's only used for profiles with parities (RAID5/6), for other profiles it
 * does nothing.
 *
 * Return 0 if we locked full stripe covering @bytenr, with a mutex held.
 * So caller must call unlock_full_stripe() at the same context.
 *
 * Return <0 if encounters error.
 */
static int lock_full_stripe(struct btrfs_fs_info *fs_info, u64 bytenr,
			    bool *locked_ret)
{
	struct btrfs_block_group_cache *bg_cache;
	struct btrfs_full_stripe_locks_tree *locks_root;
	struct full_stripe_lock *existing;
	u64 fstripe_start;
	int ret = 0;

	*locked_ret = false;
	bg_cache = btrfs_lookup_block_group(fs_info, bytenr);
	if (!bg_cache) {
		ASSERT(0);
		return -ENOENT;
	}

	/* Profiles not based on parity don't need full stripe lock */
	if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_RAID56_MASK))
		goto out;
	locks_root = &bg_cache->full_stripe_locks_root;

	fstripe_start = get_full_stripe_logical(bg_cache, bytenr);

	/* Now insert the full stripe lock */
	mutex_lock(&locks_root->lock);
	existing = insert_full_stripe_lock(locks_root, fstripe_start);
	mutex_unlock(&locks_root->lock);
	if (IS_ERR(existing)) {
		ret = PTR_ERR(existing);
		goto out;
	}
	mutex_lock(&existing->mutex);
	*locked_ret = true;
out:
	btrfs_put_block_group(bg_cache);
	return ret;
}

/*
 * Unlock a full stripe.
 *
 * NOTE: Caller must ensure it's the same context calling corresponding
 * lock_full_stripe().
 *
 * Return 0 if we unlock full stripe without problem.
 * Return <0 for error
 */
static int unlock_full_stripe(struct btrfs_fs_info *fs_info, u64 bytenr,
			      bool locked)
{
	struct btrfs_block_group_cache *bg_cache;
	struct btrfs_full_stripe_locks_tree *locks_root;
	struct full_stripe_lock *fstripe_lock;
	u64 fstripe_start;
	bool freeit = false;
	int ret = 0;

	/* If we didn't acquire full stripe lock, no need to continue */
	if (!locked)
		return 0;

	bg_cache = btrfs_lookup_block_group(fs_info, bytenr);
	if (!bg_cache) {
		ASSERT(0);
		return -ENOENT;
	}
	if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_RAID56_MASK))
		goto out;

	locks_root = &bg_cache->full_stripe_locks_root;
	fstripe_start = get_full_stripe_logical(bg_cache, bytenr);

	mutex_lock(&locks_root->lock);
	fstripe_lock = search_full_stripe_lock(locks_root, fstripe_start);
	/* Unpaired unlock_full_stripe() detected */
	if (!fstripe_lock) {
		WARN_ON(1);
		ret = -ENOENT;
		mutex_unlock(&locks_root->lock);
		goto out;
	}

	if (fstripe_lock->refs == 0) {
		WARN_ON(1);
		btrfs_warn(fs_info, "full stripe lock at %llu refcount underflow",
			fstripe_lock->logical);
	} else {
		fstripe_lock->refs--;
	}

	if (fstripe_lock->refs == 0) {
		rb_erase(&fstripe_lock->node, &locks_root->root);
		freeit = true;
	}
	mutex_unlock(&locks_root->lock);

	mutex_unlock(&fstripe_lock->mutex);
	if (freeit)
		kfree(fstripe_lock);
out:
	btrfs_put_block_group(bg_cache);
	return ret;
}

566 567 568 569 570 571
/*
 * used for workers that require transaction commits (i.e., for the
 * NOCOW case)
 */
static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx)
{
572
	struct btrfs_fs_info *fs_info = sctx->fs_info;
573

574
	refcount_inc(&sctx->refs);
575 576 577 578 579 580 581 582 583 584 585 586 587
	/*
	 * 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);
588 589 590 591 592 593 594 595 596 597

	/*
	 * check if @scrubs_running=@scrubs_paused condition
	 * inside wait_event() is not an atomic operation.
	 * which means we may inc/dec @scrub_running/paused
	 * at any time. Let's wake up @scrub_pause_wait as
	 * much as we can to let commit transaction blocked less.
	 */
	wake_up(&fs_info->scrub_pause_wait);

598 599 600 601 602 603
	atomic_inc(&sctx->workers_pending);
}

/* used for workers that require transaction commits */
static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx)
{
604
	struct btrfs_fs_info *fs_info = sctx->fs_info;
605 606 607 608 609 610 611 612 613 614 615 616

	/*
	 * 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);
617
	scrub_put_ctx(sctx);
618 619
}

620
static void scrub_free_csums(struct scrub_ctx *sctx)
A
Arne Jansen 已提交
621
{
622
	while (!list_empty(&sctx->csum_list)) {
A
Arne Jansen 已提交
623
		struct btrfs_ordered_sum *sum;
624
		sum = list_first_entry(&sctx->csum_list,
A
Arne Jansen 已提交
625 626 627 628 629 630
				       struct btrfs_ordered_sum, list);
		list_del(&sum->list);
		kfree(sum);
	}
}

631
static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx)
A
Arne Jansen 已提交
632 633 634
{
	int i;

635
	if (!sctx)
A
Arne Jansen 已提交
636 637
		return;

638
	/* this can happen when scrub is cancelled */
639 640
	if (sctx->curr != -1) {
		struct scrub_bio *sbio = sctx->bios[sctx->curr];
641 642

		for (i = 0; i < sbio->page_count; i++) {
643
			WARN_ON(!sbio->pagev[i]->page);
644 645 646 647 648
			scrub_block_put(sbio->pagev[i]->sblock);
		}
		bio_put(sbio->bio);
	}

649
	for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
650
		struct scrub_bio *sbio = sctx->bios[i];
A
Arne Jansen 已提交
651 652 653 654 655 656

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

657
	kfree(sctx->wr_curr_bio);
658 659
	scrub_free_csums(sctx);
	kfree(sctx);
A
Arne Jansen 已提交
660 661
}

662 663
static void scrub_put_ctx(struct scrub_ctx *sctx)
{
664
	if (refcount_dec_and_test(&sctx->refs))
665 666 667
		scrub_free_ctx(sctx);
}

A
Arne Jansen 已提交
668
static noinline_for_stack
669
struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace)
A
Arne Jansen 已提交
670
{
671
	struct scrub_ctx *sctx;
A
Arne Jansen 已提交
672
	int		i;
673
	struct btrfs_fs_info *fs_info = dev->fs_info;
A
Arne Jansen 已提交
674

675
	sctx = kzalloc(sizeof(*sctx), GFP_KERNEL);
676
	if (!sctx)
A
Arne Jansen 已提交
677
		goto nomem;
678
	refcount_set(&sctx->refs, 1);
679
	sctx->is_dev_replace = is_dev_replace;
680
	sctx->pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO;
681
	sctx->curr = -1;
682
	sctx->fs_info = dev->fs_info;
683
	for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
A
Arne Jansen 已提交
684 685
		struct scrub_bio *sbio;

686
		sbio = kzalloc(sizeof(*sbio), GFP_KERNEL);
A
Arne Jansen 已提交
687 688
		if (!sbio)
			goto nomem;
689
		sctx->bios[i] = sbio;
A
Arne Jansen 已提交
690 691

		sbio->index = i;
692
		sbio->sctx = sctx;
693
		sbio->page_count = 0;
694 695
		btrfs_init_work(&sbio->work, btrfs_scrub_helper,
				scrub_bio_end_io_worker, NULL, NULL);
A
Arne Jansen 已提交
696

697
		if (i != SCRUB_BIOS_PER_SCTX - 1)
698
			sctx->bios[i]->next_free = i + 1;
699
		else
700 701 702
			sctx->bios[i]->next_free = -1;
	}
	sctx->first_free = 0;
703 704
	atomic_set(&sctx->bios_in_flight, 0);
	atomic_set(&sctx->workers_pending, 0);
705 706 707 708 709 710 711
	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);
712

713 714 715
	WARN_ON(sctx->wr_curr_bio != NULL);
	mutex_init(&sctx->wr_lock);
	sctx->wr_curr_bio = NULL;
716 717
	if (is_dev_replace) {
		WARN_ON(!dev->bdev);
718 719 720
		sctx->pages_per_wr_bio = SCRUB_PAGES_PER_WR_BIO;
		sctx->wr_tgtdev = dev;
		atomic_set(&sctx->flush_all_writes, 0);
721
	}
722

723
	return sctx;
A
Arne Jansen 已提交
724 725

nomem:
726
	scrub_free_ctx(sctx);
A
Arne Jansen 已提交
727 728 729
	return ERR_PTR(-ENOMEM);
}

730 731
static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root,
				     void *warn_ctx)
732 733 734 735 736
{
	u64 isize;
	u32 nlink;
	int ret;
	int i;
737
	unsigned nofs_flag;
738 739
	struct extent_buffer *eb;
	struct btrfs_inode_item *inode_item;
740
	struct scrub_warning *swarn = warn_ctx;
741
	struct btrfs_fs_info *fs_info = swarn->dev->fs_info;
742 743 744
	struct inode_fs_paths *ipath = NULL;
	struct btrfs_root *local_root;
	struct btrfs_key root_key;
745
	struct btrfs_key key;
746 747 748 749 750 751 752 753 754 755

	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;
	}

756 757 758
	/*
	 * this makes the path point to (inum INODE_ITEM ioff)
	 */
759 760 761 762 763
	key.objectid = inum;
	key.type = BTRFS_INODE_ITEM_KEY;
	key.offset = 0;

	ret = btrfs_search_slot(NULL, local_root, &key, swarn->path, 0, 0);
764 765 766 767 768 769 770 771 772 773 774 775
	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);

776 777 778 779 780 781
	/*
	 * init_path might indirectly call vmalloc, or use GFP_KERNEL. Scrub
	 * uses GFP_NOFS in this context, so we keep it consistent but it does
	 * not seem to be strictly necessary.
	 */
	nofs_flag = memalloc_nofs_save();
782
	ipath = init_ipath(4096, local_root, swarn->path);
783
	memalloc_nofs_restore(nofs_flag);
784 785 786 787 788
	if (IS_ERR(ipath)) {
		ret = PTR_ERR(ipath);
		ipath = NULL;
		goto err;
	}
789 790 791 792 793 794 795 796 797 798
	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)
J
Jeff Mahoney 已提交
799 800 801 802 803 804 805 806
		btrfs_warn_in_rcu(fs_info,
				  "%s at logical %llu on dev %s, sector %llu, root %llu, inode %llu, offset %llu, length %llu, links %u (path: %s)",
				  swarn->errstr, swarn->logical,
				  rcu_str_deref(swarn->dev->name),
				  (unsigned long long)swarn->sector,
				  root, inum, offset,
				  min(isize - offset, (u64)PAGE_SIZE), nlink,
				  (char *)(unsigned long)ipath->fspath->val[i]);
807 808 809 810 811

	free_ipath(ipath);
	return 0;

err:
J
Jeff Mahoney 已提交
812 813 814 815 816 817
	btrfs_warn_in_rcu(fs_info,
			  "%s at logical %llu on dev %s, sector %llu, root %llu, inode %llu, offset %llu: path resolving failed with ret=%d",
			  swarn->errstr, swarn->logical,
			  rcu_str_deref(swarn->dev->name),
			  (unsigned long long)swarn->sector,
			  root, inum, offset, ret);
818 819 820 821 822

	free_ipath(ipath);
	return 0;
}

823
static void scrub_print_warning(const char *errstr, struct scrub_block *sblock)
824
{
825 826
	struct btrfs_device *dev;
	struct btrfs_fs_info *fs_info;
827 828 829 830 831
	struct btrfs_path *path;
	struct btrfs_key found_key;
	struct extent_buffer *eb;
	struct btrfs_extent_item *ei;
	struct scrub_warning swarn;
832 833 834
	unsigned long ptr = 0;
	u64 extent_item_pos;
	u64 flags = 0;
835
	u64 ref_root;
836
	u32 item_size;
837
	u8 ref_level = 0;
838
	int ret;
839

840
	WARN_ON(sblock->page_count < 1);
841
	dev = sblock->pagev[0]->dev;
842
	fs_info = sblock->sctx->fs_info;
843

844
	path = btrfs_alloc_path();
845 846
	if (!path)
		return;
847

848 849
	swarn.sector = (sblock->pagev[0]->physical) >> 9;
	swarn.logical = sblock->pagev[0]->logical;
850
	swarn.errstr = errstr;
851
	swarn.dev = NULL;
852

853 854
	ret = extent_from_logical(fs_info, swarn.logical, path, &found_key,
				  &flags);
855 856 857
	if (ret < 0)
		goto out;

J
Jan Schmidt 已提交
858
	extent_item_pos = swarn.logical - found_key.objectid;
859 860 861 862 863 864
	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]);

865
	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
866
		do {
867 868 869
			ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
						      item_size, &ref_root,
						      &ref_level);
870
			btrfs_warn_in_rcu(fs_info,
J
Jeff Mahoney 已提交
871 872
				"%s at logical %llu on dev %s, sector %llu: metadata %s (level %d) in tree %llu",
				errstr, swarn.logical,
873
				rcu_str_deref(dev->name),
874 875 876 877 878
				(unsigned long long)swarn.sector,
				ref_level ? "node" : "leaf",
				ret < 0 ? -1 : ref_level,
				ret < 0 ? -1 : ref_root);
		} while (ret != 1);
879
		btrfs_release_path(path);
880
	} else {
881
		btrfs_release_path(path);
882
		swarn.path = path;
883
		swarn.dev = dev;
884 885
		iterate_extent_inodes(fs_info, found_key.objectid,
					extent_item_pos, 1,
886 887 888 889 890 891 892
					scrub_print_warning_inode, &swarn);
	}

out:
	btrfs_free_path(path);
}

893
static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx)
894
{
895
	struct page *page = NULL;
896
	unsigned long index;
897
	struct scrub_fixup_nodatasum *fixup = fixup_ctx;
898
	int ret;
899
	int corrected = 0;
900
	struct btrfs_key key;
901
	struct inode *inode = NULL;
902
	struct btrfs_fs_info *fs_info;
903 904
	u64 end = offset + PAGE_SIZE - 1;
	struct btrfs_root *local_root;
905
	int srcu_index;
906 907 908 909

	key.objectid = root;
	key.type = BTRFS_ROOT_ITEM_KEY;
	key.offset = (u64)-1;
910 911 912 913 914 915 916

	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);
917
		return PTR_ERR(local_root);
918
	}
919 920 921 922

	key.type = BTRFS_INODE_ITEM_KEY;
	key.objectid = inum;
	key.offset = 0;
923 924
	inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
	srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
925 926 927
	if (IS_ERR(inode))
		return PTR_ERR(inode);

928
	index = offset >> PAGE_SHIFT;
929 930

	page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956
	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;
		}
957
		ret = repair_io_failure(fs_info, inum, offset, PAGE_SIZE,
958
					fixup->logical, page,
959
					offset - page_offset(page),
960 961 962 963 964 965 966 967 968 969
					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,
970
					EXTENT_DAMAGED);
971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987
		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,
988
						EXTENT_DAMAGED);
989 990 991 992 993
	}

out:
	if (page)
		put_page(page);
994 995

	iput(inode);
996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012

	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)
{
1013
	struct btrfs_fs_info *fs_info;
1014 1015
	int ret;
	struct scrub_fixup_nodatasum *fixup;
1016
	struct scrub_ctx *sctx;
1017 1018 1019 1020 1021
	struct btrfs_trans_handle *trans = NULL;
	struct btrfs_path *path;
	int uncorrectable = 0;

	fixup = container_of(work, struct scrub_fixup_nodatasum, work);
1022
	sctx = fixup->sctx;
1023
	fs_info = fixup->root->fs_info;
1024 1025 1026

	path = btrfs_alloc_path();
	if (!path) {
1027 1028 1029
		spin_lock(&sctx->stat_lock);
		++sctx->stat.malloc_errors;
		spin_unlock(&sctx->stat_lock);
1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048
		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.
	 */
1049 1050
	ret = iterate_inodes_from_logical(fixup->logical, fs_info, path,
					  scrub_fixup_readpage, fixup);
1051 1052 1053 1054 1055 1056
	if (ret < 0) {
		uncorrectable = 1;
		goto out;
	}
	WARN_ON(ret != 1);

1057 1058 1059
	spin_lock(&sctx->stat_lock);
	++sctx->stat.corrected_errors;
	spin_unlock(&sctx->stat_lock);
1060 1061 1062

out:
	if (trans && !IS_ERR(trans))
1063
		btrfs_end_transaction(trans);
1064
	if (uncorrectable) {
1065 1066 1067
		spin_lock(&sctx->stat_lock);
		++sctx->stat.uncorrectable_errors;
		spin_unlock(&sctx->stat_lock);
1068
		btrfs_dev_replace_stats_inc(
1069 1070
			&fs_info->dev_replace.num_uncorrectable_read_errors);
		btrfs_err_rl_in_rcu(fs_info,
1071
		    "unable to fixup (nodatasum) error at logical %llu on dev %s",
1072
			fixup->logical, rcu_str_deref(fixup->dev->name));
1073 1074 1075 1076 1077
	}

	btrfs_free_path(path);
	kfree(fixup);

1078
	scrub_pending_trans_workers_dec(sctx);
1079 1080
}

1081 1082
static inline void scrub_get_recover(struct scrub_recover *recover)
{
1083
	refcount_inc(&recover->refs);
1084 1085
}

1086 1087
static inline void scrub_put_recover(struct btrfs_fs_info *fs_info,
				     struct scrub_recover *recover)
1088
{
1089
	if (refcount_dec_and_test(&recover->refs)) {
1090
		btrfs_bio_counter_dec(fs_info);
1091
		btrfs_put_bbio(recover->bbio);
1092 1093 1094 1095
		kfree(recover);
	}
}

A
Arne Jansen 已提交
1096
/*
1097 1098 1099 1100 1101 1102
 * 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
1103
 */
1104
static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
A
Arne Jansen 已提交
1105
{
1106
	struct scrub_ctx *sctx = sblock_to_check->sctx;
1107
	struct btrfs_device *dev;
1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119
	struct btrfs_fs_info *fs_info;
	u64 length;
	u64 logical;
	unsigned int failed_mirror_index;
	unsigned int is_metadata;
	unsigned int have_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;
1120
	bool full_stripe_locked;
1121
	static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
1122 1123 1124
				      DEFAULT_RATELIMIT_BURST);

	BUG_ON(sblock_to_check->page_count < 1);
1125
	fs_info = sctx->fs_info;
1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136
	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;
	}
1137
	length = sblock_to_check->page_count * PAGE_SIZE;
1138 1139 1140 1141
	logical = sblock_to_check->pagev[0]->logical;
	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 &
1142
			BTRFS_EXTENT_FLAG_DATA);
1143 1144
	have_csum = sblock_to_check->pagev[0]->have_csum;
	dev = sblock_to_check->pagev[0]->dev;
1145

1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163
	/*
	 * For RAID5/6, race can happen for a different device scrub thread.
	 * For data corruption, Parity and Data threads will both try
	 * to recovery the data.
	 * Race can lead to doubly added csum error, or even unrecoverable
	 * error.
	 */
	ret = lock_full_stripe(fs_info, logical, &full_stripe_locked);
	if (ret < 0) {
		spin_lock(&sctx->stat_lock);
		if (ret == -ENOMEM)
			sctx->stat.malloc_errors++;
		sctx->stat.read_errors++;
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
		return ret;
	}

1164 1165 1166 1167 1168
	if (sctx->is_dev_replace && !is_metadata && !have_csum) {
		sblocks_for_recheck = NULL;
		goto nodatasum_case;
	}

1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197
	/*
	 * 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.
	 */

1198 1199
	sblocks_for_recheck = kcalloc(BTRFS_MAX_MIRRORS,
				      sizeof(*sblocks_for_recheck), GFP_NOFS);
1200
	if (!sblocks_for_recheck) {
1201 1202 1203 1204 1205
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		sctx->stat.read_errors++;
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
1206
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
1207
		goto out;
A
Arne Jansen 已提交
1208 1209
	}

1210
	/* setup the context, map the logical blocks and alloc the pages */
1211
	ret = scrub_setup_recheck_block(sblock_to_check, sblocks_for_recheck);
1212
	if (ret) {
1213 1214 1215 1216
		spin_lock(&sctx->stat_lock);
		sctx->stat.read_errors++;
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
1217
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
1218 1219 1220 1221
		goto out;
	}
	BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS);
	sblock_bad = sblocks_for_recheck + failed_mirror_index;
1222

1223
	/* build and submit the bios for the failed mirror, check checksums */
1224
	scrub_recheck_block(fs_info, sblock_bad, 1);
A
Arne Jansen 已提交
1225

1226 1227 1228 1229 1230 1231 1232 1233 1234 1235
	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)
		 */
1236 1237
		spin_lock(&sctx->stat_lock);
		sctx->stat.unverified_errors++;
1238
		sblock_to_check->data_corrected = 1;
1239
		spin_unlock(&sctx->stat_lock);
A
Arne Jansen 已提交
1240

1241 1242
		if (sctx->is_dev_replace)
			scrub_write_block_to_dev_replace(sblock_bad);
1243
		goto out;
A
Arne Jansen 已提交
1244 1245
	}

1246
	if (!sblock_bad->no_io_error_seen) {
1247 1248 1249
		spin_lock(&sctx->stat_lock);
		sctx->stat.read_errors++;
		spin_unlock(&sctx->stat_lock);
1250 1251
		if (__ratelimit(&_rs))
			scrub_print_warning("i/o error", sblock_to_check);
1252
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
1253
	} else if (sblock_bad->checksum_error) {
1254 1255 1256
		spin_lock(&sctx->stat_lock);
		sctx->stat.csum_errors++;
		spin_unlock(&sctx->stat_lock);
1257 1258
		if (__ratelimit(&_rs))
			scrub_print_warning("checksum error", sblock_to_check);
1259
		btrfs_dev_stat_inc_and_print(dev,
1260
					     BTRFS_DEV_STAT_CORRUPTION_ERRS);
1261
	} else if (sblock_bad->header_error) {
1262 1263 1264
		spin_lock(&sctx->stat_lock);
		sctx->stat.verify_errors++;
		spin_unlock(&sctx->stat_lock);
1265 1266 1267
		if (__ratelimit(&_rs))
			scrub_print_warning("checksum/header error",
					    sblock_to_check);
1268
		if (sblock_bad->generation_error)
1269
			btrfs_dev_stat_inc_and_print(dev,
1270 1271
				BTRFS_DEV_STAT_GENERATION_ERRS);
		else
1272
			btrfs_dev_stat_inc_and_print(dev,
1273
				BTRFS_DEV_STAT_CORRUPTION_ERRS);
1274
	}
A
Arne Jansen 已提交
1275

1276 1277 1278 1279
	if (sctx->readonly) {
		ASSERT(!sctx->is_dev_replace);
		goto out;
	}
A
Arne Jansen 已提交
1280

1281 1282
	if (!is_metadata && !have_csum) {
		struct scrub_fixup_nodatasum *fixup_nodatasum;
A
Arne Jansen 已提交
1283

1284 1285
		WARN_ON(sctx->is_dev_replace);

1286 1287
nodatasum_case:

1288 1289
		/*
		 * !is_metadata and !have_csum, this means that the data
1290
		 * might not be COWed, that it might be modified
1291 1292 1293 1294 1295 1296 1297
		 * 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;
1298
		fixup_nodatasum->sctx = sctx;
1299
		fixup_nodatasum->dev = dev;
1300 1301 1302
		fixup_nodatasum->logical = logical;
		fixup_nodatasum->root = fs_info->extent_root;
		fixup_nodatasum->mirror_num = failed_mirror_index + 1;
1303
		scrub_pending_trans_workers_inc(sctx);
1304 1305
		btrfs_init_work(&fixup_nodatasum->work, btrfs_scrub_helper,
				scrub_fixup_nodatasum, NULL, NULL);
1306 1307
		btrfs_queue_work(fs_info->scrub_workers,
				 &fixup_nodatasum->work);
1308
		goto out;
A
Arne Jansen 已提交
1309 1310
	}

1311 1312
	/*
	 * now build and submit the bios for the other mirrors, check
1313 1314
	 * checksums.
	 * First try to pick the mirror which is completely without I/O
1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329
	 * 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++) {
1330
		struct scrub_block *sblock_other;
1331

1332 1333 1334 1335 1336
		if (mirror_index == failed_mirror_index)
			continue;
		sblock_other = sblocks_for_recheck + mirror_index;

		/* build and submit the bios, check checksums */
1337
		scrub_recheck_block(fs_info, sblock_other, 0);
1338 1339

		if (!sblock_other->header_error &&
1340 1341
		    !sblock_other->checksum_error &&
		    sblock_other->no_io_error_seen) {
1342 1343
			if (sctx->is_dev_replace) {
				scrub_write_block_to_dev_replace(sblock_other);
1344
				goto corrected_error;
1345 1346
			} else {
				ret = scrub_repair_block_from_good_copy(
1347 1348 1349
						sblock_bad, sblock_other);
				if (!ret)
					goto corrected_error;
1350
			}
1351 1352
		}
	}
A
Arne Jansen 已提交
1353

1354 1355
	if (sblock_bad->no_io_error_seen && !sctx->is_dev_replace)
		goto did_not_correct_error;
1356 1357 1358

	/*
	 * In case of I/O errors in the area that is supposed to be
1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370
	 * 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
1371
	 * the final checksum succeeds. But this would be a rare
1372 1373 1374 1375 1376 1377 1378 1379
	 * 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
1380
	 */
1381
	success = 1;
1382 1383
	for (page_num = 0; page_num < sblock_bad->page_count;
	     page_num++) {
1384
		struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1385
		struct scrub_block *sblock_other = NULL;
1386

1387 1388
		/* skip no-io-error page in scrub */
		if (!page_bad->io_error && !sctx->is_dev_replace)
A
Arne Jansen 已提交
1389
			continue;
1390

1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401
		/* try to find no-io-error page in mirrors */
		if (page_bad->io_error) {
			for (mirror_index = 0;
			     mirror_index < BTRFS_MAX_MIRRORS &&
			     sblocks_for_recheck[mirror_index].page_count > 0;
			     mirror_index++) {
				if (!sblocks_for_recheck[mirror_index].
				    pagev[page_num]->io_error) {
					sblock_other = sblocks_for_recheck +
						       mirror_index;
					break;
1402 1403
				}
			}
1404 1405
			if (!sblock_other)
				success = 0;
I
Ilya Dryomov 已提交
1406
		}
A
Arne Jansen 已提交
1407

1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421
		if (sctx->is_dev_replace) {
			/*
			 * 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
			 */
			if (!sblock_other)
				sblock_other = sblock_bad;

			if (scrub_write_page_to_dev_replace(sblock_other,
							    page_num) != 0) {
				btrfs_dev_replace_stats_inc(
1422
					&fs_info->dev_replace.num_write_errors);
1423 1424 1425 1426 1427 1428 1429 1430 1431 1432
				success = 0;
			}
		} else if (sblock_other) {
			ret = scrub_repair_page_from_good_copy(sblock_bad,
							       sblock_other,
							       page_num, 0);
			if (0 == ret)
				page_bad->io_error = 0;
			else
				success = 0;
1433
		}
A
Arne Jansen 已提交
1434 1435
	}

1436
	if (success && !sctx->is_dev_replace) {
1437 1438 1439 1440 1441 1442 1443 1444 1445 1446
		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.
			 */
1447
			scrub_recheck_block(fs_info, sblock_bad, 1);
1448
			if (!sblock_bad->header_error &&
1449 1450 1451 1452 1453 1454 1455
			    !sblock_bad->checksum_error &&
			    sblock_bad->no_io_error_seen)
				goto corrected_error;
			else
				goto did_not_correct_error;
		} else {
corrected_error:
1456 1457
			spin_lock(&sctx->stat_lock);
			sctx->stat.corrected_errors++;
1458
			sblock_to_check->data_corrected = 1;
1459
			spin_unlock(&sctx->stat_lock);
1460 1461
			btrfs_err_rl_in_rcu(fs_info,
				"fixed up error at logical %llu on dev %s",
1462
				logical, rcu_str_deref(dev->name));
A
Arne Jansen 已提交
1463
		}
1464 1465
	} else {
did_not_correct_error:
1466 1467 1468
		spin_lock(&sctx->stat_lock);
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
1469 1470
		btrfs_err_rl_in_rcu(fs_info,
			"unable to fixup (regular) error at logical %llu on dev %s",
1471
			logical, rcu_str_deref(dev->name));
I
Ilya Dryomov 已提交
1472
	}
A
Arne Jansen 已提交
1473

1474 1475 1476 1477 1478 1479
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;
1480
			struct scrub_recover *recover;
1481 1482
			int page_index;

1483 1484 1485
			for (page_index = 0; page_index < sblock->page_count;
			     page_index++) {
				sblock->pagev[page_index]->sblock = NULL;
1486 1487
				recover = sblock->pagev[page_index]->recover;
				if (recover) {
1488
					scrub_put_recover(fs_info, recover);
1489 1490 1491
					sblock->pagev[page_index]->recover =
									NULL;
				}
1492 1493
				scrub_page_put(sblock->pagev[page_index]);
			}
1494 1495 1496
		}
		kfree(sblocks_for_recheck);
	}
A
Arne Jansen 已提交
1497

1498 1499 1500
	ret = unlock_full_stripe(fs_info, logical, full_stripe_locked);
	if (ret < 0)
		return ret;
1501 1502
	return 0;
}
A
Arne Jansen 已提交
1503

1504
static inline int scrub_nr_raid_mirrors(struct btrfs_bio *bbio)
1505
{
Z
Zhao Lei 已提交
1506 1507 1508 1509 1510
	if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID5)
		return 2;
	else if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID6)
		return 3;
	else
1511 1512 1513
		return (int)bbio->num_stripes;
}

Z
Zhao Lei 已提交
1514 1515
static inline void scrub_stripe_index_and_offset(u64 logical, u64 map_type,
						 u64 *raid_map,
1516 1517 1518 1519 1520 1521 1522
						 u64 mapped_length,
						 int nstripes, int mirror,
						 int *stripe_index,
						 u64 *stripe_offset)
{
	int i;

1523
	if (map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543
		/* RAID5/6 */
		for (i = 0; i < nstripes; i++) {
			if (raid_map[i] == RAID6_Q_STRIPE ||
			    raid_map[i] == RAID5_P_STRIPE)
				continue;

			if (logical >= raid_map[i] &&
			    logical < raid_map[i] + mapped_length)
				break;
		}

		*stripe_index = i;
		*stripe_offset = logical - raid_map[i];
	} else {
		/* The other RAID type */
		*stripe_index = mirror;
		*stripe_offset = 0;
	}
}

1544
static int scrub_setup_recheck_block(struct scrub_block *original_sblock,
1545 1546
				     struct scrub_block *sblocks_for_recheck)
{
1547
	struct scrub_ctx *sctx = original_sblock->sctx;
1548
	struct btrfs_fs_info *fs_info = sctx->fs_info;
1549 1550
	u64 length = original_sblock->page_count * PAGE_SIZE;
	u64 logical = original_sblock->pagev[0]->logical;
1551 1552 1553
	u64 generation = original_sblock->pagev[0]->generation;
	u64 flags = original_sblock->pagev[0]->flags;
	u64 have_csum = original_sblock->pagev[0]->have_csum;
1554 1555 1556 1557 1558 1559
	struct scrub_recover *recover;
	struct btrfs_bio *bbio;
	u64 sublen;
	u64 mapped_length;
	u64 stripe_offset;
	int stripe_index;
1560
	int page_index = 0;
1561
	int mirror_index;
1562
	int nmirrors;
1563 1564 1565
	int ret;

	/*
1566
	 * note: the two members refs and outstanding_pages
1567 1568 1569 1570 1571
	 * are not used (and not set) in the blocks that are used for
	 * the recheck procedure
	 */

	while (length > 0) {
1572 1573 1574
		sublen = min_t(u64, length, PAGE_SIZE);
		mapped_length = sublen;
		bbio = NULL;
A
Arne Jansen 已提交
1575

1576 1577 1578 1579
		/*
		 * with a length of PAGE_SIZE, each returned stripe
		 * represents one mirror
		 */
1580
		btrfs_bio_counter_inc_blocked(fs_info);
1581
		ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
1582
				logical, &mapped_length, &bbio);
1583
		if (ret || !bbio || mapped_length < sublen) {
1584
			btrfs_put_bbio(bbio);
1585
			btrfs_bio_counter_dec(fs_info);
1586 1587
			return -EIO;
		}
A
Arne Jansen 已提交
1588

1589 1590
		recover = kzalloc(sizeof(struct scrub_recover), GFP_NOFS);
		if (!recover) {
1591
			btrfs_put_bbio(bbio);
1592
			btrfs_bio_counter_dec(fs_info);
1593 1594 1595
			return -ENOMEM;
		}

1596
		refcount_set(&recover->refs, 1);
1597 1598 1599
		recover->bbio = bbio;
		recover->map_length = mapped_length;

1600
		BUG_ON(page_index >= SCRUB_MAX_PAGES_PER_BLOCK);
1601

1602
		nmirrors = min(scrub_nr_raid_mirrors(bbio), BTRFS_MAX_MIRRORS);
Z
Zhao Lei 已提交
1603

1604
		for (mirror_index = 0; mirror_index < nmirrors;
1605 1606 1607 1608 1609
		     mirror_index++) {
			struct scrub_block *sblock;
			struct scrub_page *page;

			sblock = sblocks_for_recheck + mirror_index;
1610
			sblock->sctx = sctx;
1611

1612 1613 1614
			page = kzalloc(sizeof(*page), GFP_NOFS);
			if (!page) {
leave_nomem:
1615 1616 1617
				spin_lock(&sctx->stat_lock);
				sctx->stat.malloc_errors++;
				spin_unlock(&sctx->stat_lock);
1618
				scrub_put_recover(fs_info, recover);
1619 1620
				return -ENOMEM;
			}
1621 1622
			scrub_page_get(page);
			sblock->pagev[page_index] = page;
1623 1624 1625
			page->sblock = sblock;
			page->flags = flags;
			page->generation = generation;
1626
			page->logical = logical;
1627 1628 1629 1630 1631
			page->have_csum = have_csum;
			if (have_csum)
				memcpy(page->csum,
				       original_sblock->pagev[0]->csum,
				       sctx->csum_size);
1632

Z
Zhao Lei 已提交
1633 1634 1635
			scrub_stripe_index_and_offset(logical,
						      bbio->map_type,
						      bbio->raid_map,
1636
						      mapped_length,
1637 1638
						      bbio->num_stripes -
						      bbio->num_tgtdevs,
1639 1640 1641 1642 1643 1644 1645
						      mirror_index,
						      &stripe_index,
						      &stripe_offset);
			page->physical = bbio->stripes[stripe_index].physical +
					 stripe_offset;
			page->dev = bbio->stripes[stripe_index].dev;

1646 1647 1648 1649
			BUG_ON(page_index >= original_sblock->page_count);
			page->physical_for_dev_replace =
				original_sblock->pagev[page_index]->
				physical_for_dev_replace;
1650 1651
			/* for missing devices, dev->bdev is NULL */
			page->mirror_num = mirror_index + 1;
1652
			sblock->page_count++;
1653 1654 1655
			page->page = alloc_page(GFP_NOFS);
			if (!page->page)
				goto leave_nomem;
1656 1657 1658

			scrub_get_recover(recover);
			page->recover = recover;
1659
		}
1660
		scrub_put_recover(fs_info, recover);
1661 1662 1663 1664 1665 1666
		length -= sublen;
		logical += sublen;
		page_index++;
	}

	return 0;
I
Ilya Dryomov 已提交
1667 1668
}

1669 1670 1671 1672 1673
struct scrub_bio_ret {
	struct completion event;
	int error;
};

1674
static void scrub_bio_wait_endio(struct bio *bio)
1675 1676 1677
{
	struct scrub_bio_ret *ret = bio->bi_private;

1678
	ret->error = bio->bi_error;
1679 1680 1681 1682 1683
	complete(&ret->event);
}

static inline int scrub_is_page_on_raid56(struct scrub_page *page)
{
Z
Zhao Lei 已提交
1684
	return page->recover &&
1685
	       (page->recover->bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK);
1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700
}

static int scrub_submit_raid56_bio_wait(struct btrfs_fs_info *fs_info,
					struct bio *bio,
					struct scrub_page *page)
{
	struct scrub_bio_ret done;
	int ret;

	init_completion(&done.event);
	done.error = 0;
	bio->bi_iter.bi_sector = page->logical >> 9;
	bio->bi_private = &done;
	bio->bi_end_io = scrub_bio_wait_endio;

1701
	ret = raid56_parity_recover(fs_info, bio, page->recover->bbio,
1702
				    page->recover->map_length,
1703
				    page->mirror_num, 0);
1704 1705 1706 1707 1708 1709 1710 1711 1712 1713
	if (ret)
		return ret;

	wait_for_completion(&done.event);
	if (done.error)
		return -EIO;

	return 0;
}

1714 1715 1716 1717 1718 1719 1720
/*
 * 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.
 */
1721
static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
1722 1723
				struct scrub_block *sblock,
				int retry_failed_mirror)
I
Ilya Dryomov 已提交
1724
{
1725
	int page_num;
I
Ilya Dryomov 已提交
1726

1727
	sblock->no_io_error_seen = 1;
I
Ilya Dryomov 已提交
1728

1729 1730
	for (page_num = 0; page_num < sblock->page_count; page_num++) {
		struct bio *bio;
1731
		struct scrub_page *page = sblock->pagev[page_num];
1732

1733
		if (page->dev->bdev == NULL) {
1734 1735 1736 1737 1738
			page->io_error = 1;
			sblock->no_io_error_seen = 0;
			continue;
		}

1739
		WARN_ON(!page->page);
1740
		bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1741
		bio->bi_bdev = page->dev->bdev;
1742

1743
		bio_add_page(bio, page->page, PAGE_SIZE, 0);
1744
		if (!retry_failed_mirror && scrub_is_page_on_raid56(page)) {
1745 1746
			if (scrub_submit_raid56_bio_wait(fs_info, bio, page)) {
				page->io_error = 1;
1747
				sblock->no_io_error_seen = 0;
1748
			}
1749 1750
		} else {
			bio->bi_iter.bi_sector = page->physical >> 9;
M
Mike Christie 已提交
1751
			bio_set_op_attrs(bio, REQ_OP_READ, 0);
1752

1753 1754
			if (btrfsic_submit_bio_wait(bio)) {
				page->io_error = 1;
1755
				sblock->no_io_error_seen = 0;
1756
			}
1757
		}
1758

1759 1760
		bio_put(bio);
	}
I
Ilya Dryomov 已提交
1761

1762
	if (sblock->no_io_error_seen)
1763
		scrub_recheck_block_checksum(sblock);
A
Arne Jansen 已提交
1764 1765
}

M
Miao Xie 已提交
1766 1767 1768 1769 1770 1771 1772 1773 1774 1775
static inline int scrub_check_fsid(u8 fsid[],
				   struct scrub_page *spage)
{
	struct btrfs_fs_devices *fs_devices = spage->dev->fs_devices;
	int ret;

	ret = memcmp(fsid, fs_devices->fsid, BTRFS_UUID_SIZE);
	return !ret;
}

1776
static void scrub_recheck_block_checksum(struct scrub_block *sblock)
A
Arne Jansen 已提交
1777
{
1778 1779 1780
	sblock->header_error = 0;
	sblock->checksum_error = 0;
	sblock->generation_error = 0;
1781

1782 1783 1784 1785
	if (sblock->pagev[0]->flags & BTRFS_EXTENT_FLAG_DATA)
		scrub_checksum_data(sblock);
	else
		scrub_checksum_tree_block(sblock);
A
Arne Jansen 已提交
1786 1787
}

1788
static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
1789
					     struct scrub_block *sblock_good)
1790 1791 1792
{
	int page_num;
	int ret = 0;
I
Ilya Dryomov 已提交
1793

1794 1795
	for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
		int ret_sub;
I
Ilya Dryomov 已提交
1796

1797 1798
		ret_sub = scrub_repair_page_from_good_copy(sblock_bad,
							   sblock_good,
1799
							   page_num, 1);
1800 1801
		if (ret_sub)
			ret = ret_sub;
A
Arne Jansen 已提交
1802
	}
1803 1804 1805 1806 1807 1808 1809 1810

	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)
{
1811 1812
	struct scrub_page *page_bad = sblock_bad->pagev[page_num];
	struct scrub_page *page_good = sblock_good->pagev[page_num];
1813
	struct btrfs_fs_info *fs_info = sblock_bad->sctx->fs_info;
1814

1815 1816
	BUG_ON(page_bad->page == NULL);
	BUG_ON(page_good->page == NULL);
1817 1818 1819 1820 1821
	if (force_write || sblock_bad->header_error ||
	    sblock_bad->checksum_error || page_bad->io_error) {
		struct bio *bio;
		int ret;

1822
		if (!page_bad->dev->bdev) {
1823
			btrfs_warn_rl(fs_info,
J
Jeff Mahoney 已提交
1824
				"scrub_repair_page_from_good_copy(bdev == NULL) is unexpected");
1825 1826 1827
			return -EIO;
		}

1828
		bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1829
		bio->bi_bdev = page_bad->dev->bdev;
1830
		bio->bi_iter.bi_sector = page_bad->physical >> 9;
M
Mike Christie 已提交
1831
		bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
1832 1833 1834 1835 1836

		ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0);
		if (PAGE_SIZE != ret) {
			bio_put(bio);
			return -EIO;
1837
		}
1838

1839
		if (btrfsic_submit_bio_wait(bio)) {
1840 1841
			btrfs_dev_stat_inc_and_print(page_bad->dev,
				BTRFS_DEV_STAT_WRITE_ERRS);
1842
			btrfs_dev_replace_stats_inc(
1843
				&fs_info->dev_replace.num_write_errors);
1844 1845 1846
			bio_put(bio);
			return -EIO;
		}
1847
		bio_put(bio);
A
Arne Jansen 已提交
1848 1849
	}

1850 1851 1852
	return 0;
}

1853 1854
static void scrub_write_block_to_dev_replace(struct scrub_block *sblock)
{
1855
	struct btrfs_fs_info *fs_info = sblock->sctx->fs_info;
1856 1857
	int page_num;

1858 1859 1860 1861 1862 1863 1864
	/*
	 * This block is used for the check of the parity on the source device,
	 * so the data needn't be written into the destination device.
	 */
	if (sblock->sparity)
		return;

1865 1866 1867 1868 1869 1870
	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(
1871
				&fs_info->dev_replace.num_write_errors);
1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883
	}
}

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);

1884
		clear_page(mapped_buffer);
1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896
		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_bio *sbio;
	int ret;

1897
	mutex_lock(&sctx->wr_lock);
1898
again:
1899 1900
	if (!sctx->wr_curr_bio) {
		sctx->wr_curr_bio = kzalloc(sizeof(*sctx->wr_curr_bio),
1901
					      GFP_KERNEL);
1902 1903
		if (!sctx->wr_curr_bio) {
			mutex_unlock(&sctx->wr_lock);
1904 1905
			return -ENOMEM;
		}
1906 1907
		sctx->wr_curr_bio->sctx = sctx;
		sctx->wr_curr_bio->page_count = 0;
1908
	}
1909
	sbio = sctx->wr_curr_bio;
1910 1911 1912 1913 1914
	if (sbio->page_count == 0) {
		struct bio *bio;

		sbio->physical = spage->physical_for_dev_replace;
		sbio->logical = spage->logical;
1915
		sbio->dev = sctx->wr_tgtdev;
1916 1917
		bio = sbio->bio;
		if (!bio) {
1918
			bio = btrfs_io_bio_alloc(GFP_KERNEL,
1919
					sctx->pages_per_wr_bio);
1920 1921 1922 1923 1924 1925
			sbio->bio = bio;
		}

		bio->bi_private = sbio;
		bio->bi_end_io = scrub_wr_bio_end_io;
		bio->bi_bdev = sbio->dev->bdev;
1926
		bio->bi_iter.bi_sector = sbio->physical >> 9;
M
Mike Christie 已提交
1927
		bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941
		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;
1942
			mutex_unlock(&sctx->wr_lock);
1943 1944 1945 1946 1947 1948 1949 1950 1951
			return -EIO;
		}
		scrub_wr_submit(sctx);
		goto again;
	}

	sbio->pagev[sbio->page_count] = spage;
	scrub_page_get(spage);
	sbio->page_count++;
1952
	if (sbio->page_count == sctx->pages_per_wr_bio)
1953
		scrub_wr_submit(sctx);
1954
	mutex_unlock(&sctx->wr_lock);
1955 1956 1957 1958 1959 1960 1961 1962

	return 0;
}

static void scrub_wr_submit(struct scrub_ctx *sctx)
{
	struct scrub_bio *sbio;

1963
	if (!sctx->wr_curr_bio)
1964 1965
		return;

1966 1967
	sbio = sctx->wr_curr_bio;
	sctx->wr_curr_bio = NULL;
1968 1969 1970 1971 1972 1973
	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 */
1974
	btrfsic_submit_bio(sbio->bio);
1975 1976
}

1977
static void scrub_wr_bio_end_io(struct bio *bio)
1978 1979
{
	struct scrub_bio *sbio = bio->bi_private;
1980
	struct btrfs_fs_info *fs_info = sbio->dev->fs_info;
1981

1982
	sbio->err = bio->bi_error;
1983 1984
	sbio->bio = bio;

1985 1986
	btrfs_init_work(&sbio->work, btrfs_scrubwrc_helper,
			 scrub_wr_bio_end_io_worker, NULL, NULL);
1987
	btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work);
1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998
}

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 =
1999
			&sbio->sctx->fs_info->dev_replace;
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

		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)
2019 2020 2021 2022
{
	u64 flags;
	int ret;

2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034
	/*
	 * No need to initialize these stats currently,
	 * because this function only use return value
	 * instead of these stats value.
	 *
	 * Todo:
	 * always use stats
	 */
	sblock->header_error = 0;
	sblock->generation_error = 0;
	sblock->checksum_error = 0;

2035 2036
	WARN_ON(sblock->page_count < 1);
	flags = sblock->pagev[0]->flags;
2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047
	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);
2048 2049

	return ret;
A
Arne Jansen 已提交
2050 2051
}

2052
static int scrub_checksum_data(struct scrub_block *sblock)
A
Arne Jansen 已提交
2053
{
2054
	struct scrub_ctx *sctx = sblock->sctx;
A
Arne Jansen 已提交
2055
	u8 csum[BTRFS_CSUM_SIZE];
2056 2057 2058
	u8 *on_disk_csum;
	struct page *page;
	void *buffer;
A
Arne Jansen 已提交
2059
	u32 crc = ~(u32)0;
2060 2061
	u64 len;
	int index;
A
Arne Jansen 已提交
2062

2063
	BUG_ON(sblock->page_count < 1);
2064
	if (!sblock->pagev[0]->have_csum)
A
Arne Jansen 已提交
2065 2066
		return 0;

2067 2068
	on_disk_csum = sblock->pagev[0]->csum;
	page = sblock->pagev[0]->page;
2069
	buffer = kmap_atomic(page);
2070

2071
	len = sctx->fs_info->sectorsize;
2072 2073 2074 2075
	index = 0;
	for (;;) {
		u64 l = min_t(u64, len, PAGE_SIZE);

2076
		crc = btrfs_csum_data(buffer, crc, l);
2077
		kunmap_atomic(buffer);
2078 2079 2080 2081 2082
		len -= l;
		if (len == 0)
			break;
		index++;
		BUG_ON(index >= sblock->page_count);
2083 2084
		BUG_ON(!sblock->pagev[index]->page);
		page = sblock->pagev[index]->page;
2085
		buffer = kmap_atomic(page);
2086 2087
	}

A
Arne Jansen 已提交
2088
	btrfs_csum_final(crc, csum);
2089
	if (memcmp(csum, on_disk_csum, sctx->csum_size))
2090
		sblock->checksum_error = 1;
A
Arne Jansen 已提交
2091

2092
	return sblock->checksum_error;
A
Arne Jansen 已提交
2093 2094
}

2095
static int scrub_checksum_tree_block(struct scrub_block *sblock)
A
Arne Jansen 已提交
2096
{
2097
	struct scrub_ctx *sctx = sblock->sctx;
A
Arne Jansen 已提交
2098
	struct btrfs_header *h;
2099
	struct btrfs_fs_info *fs_info = sctx->fs_info;
2100 2101 2102 2103 2104 2105
	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
2106
	u32 crc = ~(u32)0;
2107 2108 2109 2110
	u64 len;
	int index;

	BUG_ON(sblock->page_count < 1);
2111
	page = sblock->pagev[0]->page;
2112
	mapped_buffer = kmap_atomic(page);
2113
	h = (struct btrfs_header *)mapped_buffer;
2114
	memcpy(on_disk_csum, h->csum, sctx->csum_size);
A
Arne Jansen 已提交
2115 2116 2117 2118 2119 2120

	/*
	 * we don't use the getter functions here, as we
	 * a) don't have an extent buffer and
	 * b) the page is already kmapped
	 */
2121
	if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h))
2122
		sblock->header_error = 1;
A
Arne Jansen 已提交
2123

2124 2125 2126 2127
	if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h)) {
		sblock->header_error = 1;
		sblock->generation_error = 1;
	}
A
Arne Jansen 已提交
2128

M
Miao Xie 已提交
2129
	if (!scrub_check_fsid(h->fsid, sblock->pagev[0]))
2130
		sblock->header_error = 1;
A
Arne Jansen 已提交
2131 2132 2133

	if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
		   BTRFS_UUID_SIZE))
2134
		sblock->header_error = 1;
A
Arne Jansen 已提交
2135

2136
	len = sctx->fs_info->nodesize - BTRFS_CSUM_SIZE;
2137 2138 2139 2140 2141 2142
	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);

2143
		crc = btrfs_csum_data(p, crc, l);
2144
		kunmap_atomic(mapped_buffer);
2145 2146 2147 2148 2149
		len -= l;
		if (len == 0)
			break;
		index++;
		BUG_ON(index >= sblock->page_count);
2150 2151
		BUG_ON(!sblock->pagev[index]->page);
		page = sblock->pagev[index]->page;
2152
		mapped_buffer = kmap_atomic(page);
2153 2154 2155 2156 2157
		mapped_size = PAGE_SIZE;
		p = mapped_buffer;
	}

	btrfs_csum_final(crc, calculated_csum);
2158
	if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
2159
		sblock->checksum_error = 1;
A
Arne Jansen 已提交
2160

2161
	return sblock->header_error || sblock->checksum_error;
A
Arne Jansen 已提交
2162 2163
}

2164
static int scrub_checksum_super(struct scrub_block *sblock)
A
Arne Jansen 已提交
2165 2166
{
	struct btrfs_super_block *s;
2167
	struct scrub_ctx *sctx = sblock->sctx;
2168 2169 2170 2171 2172 2173
	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
2174
	u32 crc = ~(u32)0;
2175 2176
	int fail_gen = 0;
	int fail_cor = 0;
2177 2178
	u64 len;
	int index;
A
Arne Jansen 已提交
2179

2180
	BUG_ON(sblock->page_count < 1);
2181
	page = sblock->pagev[0]->page;
2182
	mapped_buffer = kmap_atomic(page);
2183
	s = (struct btrfs_super_block *)mapped_buffer;
2184
	memcpy(on_disk_csum, s->csum, sctx->csum_size);
A
Arne Jansen 已提交
2185

2186
	if (sblock->pagev[0]->logical != btrfs_super_bytenr(s))
2187
		++fail_cor;
A
Arne Jansen 已提交
2188

2189
	if (sblock->pagev[0]->generation != btrfs_super_generation(s))
2190
		++fail_gen;
A
Arne Jansen 已提交
2191

M
Miao Xie 已提交
2192
	if (!scrub_check_fsid(s->fsid, sblock->pagev[0]))
2193
		++fail_cor;
A
Arne Jansen 已提交
2194

2195 2196 2197 2198 2199 2200 2201
	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);

2202
		crc = btrfs_csum_data(p, crc, l);
2203
		kunmap_atomic(mapped_buffer);
2204 2205 2206 2207 2208
		len -= l;
		if (len == 0)
			break;
		index++;
		BUG_ON(index >= sblock->page_count);
2209 2210
		BUG_ON(!sblock->pagev[index]->page);
		page = sblock->pagev[index]->page;
2211
		mapped_buffer = kmap_atomic(page);
2212 2213 2214 2215 2216
		mapped_size = PAGE_SIZE;
		p = mapped_buffer;
	}

	btrfs_csum_final(crc, calculated_csum);
2217
	if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
2218
		++fail_cor;
A
Arne Jansen 已提交
2219

2220
	if (fail_cor + fail_gen) {
A
Arne Jansen 已提交
2221 2222 2223 2224 2225
		/*
		 * if we find an error in a super block, we just report it.
		 * They will get written with the next transaction commit
		 * anyway
		 */
2226 2227 2228
		spin_lock(&sctx->stat_lock);
		++sctx->stat.super_errors;
		spin_unlock(&sctx->stat_lock);
2229
		if (fail_cor)
2230
			btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
2231 2232
				BTRFS_DEV_STAT_CORRUPTION_ERRS);
		else
2233
			btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
2234
				BTRFS_DEV_STAT_GENERATION_ERRS);
A
Arne Jansen 已提交
2235 2236
	}

2237
	return fail_cor + fail_gen;
A
Arne Jansen 已提交
2238 2239
}

2240 2241
static void scrub_block_get(struct scrub_block *sblock)
{
2242
	refcount_inc(&sblock->refs);
2243 2244 2245 2246
}

static void scrub_block_put(struct scrub_block *sblock)
{
2247
	if (refcount_dec_and_test(&sblock->refs)) {
2248 2249
		int i;

2250 2251 2252
		if (sblock->sparity)
			scrub_parity_put(sblock->sparity);

2253
		for (i = 0; i < sblock->page_count; i++)
2254
			scrub_page_put(sblock->pagev[i]);
2255 2256 2257 2258
		kfree(sblock);
	}
}

2259 2260
static void scrub_page_get(struct scrub_page *spage)
{
2261
	atomic_inc(&spage->refs);
2262 2263 2264 2265
}

static void scrub_page_put(struct scrub_page *spage)
{
2266
	if (atomic_dec_and_test(&spage->refs)) {
2267 2268 2269 2270 2271 2272
		if (spage->page)
			__free_page(spage->page);
		kfree(spage);
	}
}

2273
static void scrub_submit(struct scrub_ctx *sctx)
A
Arne Jansen 已提交
2274 2275 2276
{
	struct scrub_bio *sbio;

2277
	if (sctx->curr == -1)
S
Stefan Behrens 已提交
2278
		return;
A
Arne Jansen 已提交
2279

2280 2281
	sbio = sctx->bios[sctx->curr];
	sctx->curr = -1;
2282
	scrub_pending_bio_inc(sctx);
2283
	btrfsic_submit_bio(sbio->bio);
A
Arne Jansen 已提交
2284 2285
}

2286 2287
static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
				    struct scrub_page *spage)
A
Arne Jansen 已提交
2288
{
2289
	struct scrub_block *sblock = spage->sblock;
A
Arne Jansen 已提交
2290
	struct scrub_bio *sbio;
2291
	int ret;
A
Arne Jansen 已提交
2292 2293 2294 2295 2296

again:
	/*
	 * grab a fresh bio or wait for one to become available
	 */
2297 2298 2299 2300 2301 2302 2303 2304
	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
2305
		} else {
2306 2307
			spin_unlock(&sctx->list_lock);
			wait_event(sctx->list_wait, sctx->first_free != -1);
A
Arne Jansen 已提交
2308 2309
		}
	}
2310
	sbio = sctx->bios[sctx->curr];
2311
	if (sbio->page_count == 0) {
2312 2313
		struct bio *bio;

2314 2315
		sbio->physical = spage->physical;
		sbio->logical = spage->logical;
2316
		sbio->dev = spage->dev;
2317 2318
		bio = sbio->bio;
		if (!bio) {
2319 2320
			bio = btrfs_io_bio_alloc(GFP_KERNEL,
					sctx->pages_per_rd_bio);
2321 2322
			sbio->bio = bio;
		}
2323 2324 2325

		bio->bi_private = sbio;
		bio->bi_end_io = scrub_bio_end_io;
2326
		bio->bi_bdev = sbio->dev->bdev;
2327
		bio->bi_iter.bi_sector = sbio->physical >> 9;
M
Mike Christie 已提交
2328
		bio_set_op_attrs(bio, REQ_OP_READ, 0);
2329
		sbio->err = 0;
2330 2331 2332
	} else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
		   spage->physical ||
		   sbio->logical + sbio->page_count * PAGE_SIZE !=
2333 2334
		   spage->logical ||
		   sbio->dev != spage->dev) {
2335
		scrub_submit(sctx);
A
Arne Jansen 已提交
2336 2337
		goto again;
	}
2338

2339 2340 2341 2342 2343 2344 2345 2346
	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;
		}
2347
		scrub_submit(sctx);
2348 2349 2350
		goto again;
	}

2351
	scrub_block_get(sblock); /* one for the page added to the bio */
2352 2353
	atomic_inc(&sblock->outstanding_pages);
	sbio->page_count++;
2354
	if (sbio->page_count == sctx->pages_per_rd_bio)
2355
		scrub_submit(sctx);
2356 2357 2358 2359

	return 0;
}

2360
static void scrub_missing_raid56_end_io(struct bio *bio)
2361 2362
{
	struct scrub_block *sblock = bio->bi_private;
2363
	struct btrfs_fs_info *fs_info = sblock->sctx->fs_info;
2364

2365
	if (bio->bi_error)
2366 2367
		sblock->no_io_error_seen = 0;

2368 2369
	bio_put(bio);

2370 2371 2372 2373 2374 2375 2376
	btrfs_queue_work(fs_info->scrub_workers, &sblock->work);
}

static void scrub_missing_raid56_worker(struct btrfs_work *work)
{
	struct scrub_block *sblock = container_of(work, struct scrub_block, work);
	struct scrub_ctx *sctx = sblock->sctx;
2377
	struct btrfs_fs_info *fs_info = sctx->fs_info;
2378 2379 2380 2381 2382 2383
	u64 logical;
	struct btrfs_device *dev;

	logical = sblock->pagev[0]->logical;
	dev = sblock->pagev[0]->dev;

2384
	if (sblock->no_io_error_seen)
2385
		scrub_recheck_block_checksum(sblock);
2386 2387 2388 2389 2390

	if (!sblock->no_io_error_seen) {
		spin_lock(&sctx->stat_lock);
		sctx->stat.read_errors++;
		spin_unlock(&sctx->stat_lock);
2391
		btrfs_err_rl_in_rcu(fs_info,
2392
			"IO error rebuilding logical %llu for dev %s",
2393 2394 2395 2396 2397
			logical, rcu_str_deref(dev->name));
	} else if (sblock->header_error || sblock->checksum_error) {
		spin_lock(&sctx->stat_lock);
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
2398
		btrfs_err_rl_in_rcu(fs_info,
2399
			"failed to rebuild valid logical %llu for dev %s",
2400 2401 2402 2403 2404 2405 2406 2407
			logical, rcu_str_deref(dev->name));
	} else {
		scrub_write_block_to_dev_replace(sblock);
	}

	scrub_block_put(sblock);

	if (sctx->is_dev_replace &&
2408 2409
	    atomic_read(&sctx->flush_all_writes)) {
		mutex_lock(&sctx->wr_lock);
2410
		scrub_wr_submit(sctx);
2411
		mutex_unlock(&sctx->wr_lock);
2412 2413 2414 2415 2416 2417 2418 2419
	}

	scrub_pending_bio_dec(sctx);
}

static void scrub_missing_raid56_pages(struct scrub_block *sblock)
{
	struct scrub_ctx *sctx = sblock->sctx;
2420
	struct btrfs_fs_info *fs_info = sctx->fs_info;
2421 2422
	u64 length = sblock->page_count * PAGE_SIZE;
	u64 logical = sblock->pagev[0]->logical;
2423
	struct btrfs_bio *bbio = NULL;
2424 2425 2426 2427 2428
	struct bio *bio;
	struct btrfs_raid_bio *rbio;
	int ret;
	int i;

2429
	btrfs_bio_counter_inc_blocked(fs_info);
2430
	ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical,
2431
			&length, &bbio);
2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450
	if (ret || !bbio || !bbio->raid_map)
		goto bbio_out;

	if (WARN_ON(!sctx->is_dev_replace ||
		    !(bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK))) {
		/*
		 * We shouldn't be scrubbing a missing device. Even for dev
		 * replace, we should only get here for RAID 5/6. We either
		 * managed to mount something with no mirrors remaining or
		 * there's a bug in scrub_remap_extent()/btrfs_map_block().
		 */
		goto bbio_out;
	}

	bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
	bio->bi_iter.bi_sector = logical >> 9;
	bio->bi_private = sblock;
	bio->bi_end_io = scrub_missing_raid56_end_io;

2451
	rbio = raid56_alloc_missing_rbio(fs_info, bio, bbio, length);
2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470
	if (!rbio)
		goto rbio_out;

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

		raid56_add_scrub_pages(rbio, spage->page, spage->logical);
	}

	btrfs_init_work(&sblock->work, btrfs_scrub_helper,
			scrub_missing_raid56_worker, NULL, NULL);
	scrub_block_get(sblock);
	scrub_pending_bio_inc(sctx);
	raid56_submit_missing_rbio(rbio);
	return;

rbio_out:
	bio_put(bio);
bbio_out:
2471
	btrfs_bio_counter_dec(fs_info);
2472 2473 2474 2475 2476 2477
	btrfs_put_bbio(bbio);
	spin_lock(&sctx->stat_lock);
	sctx->stat.malloc_errors++;
	spin_unlock(&sctx->stat_lock);
}

2478
static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
2479
		       u64 physical, struct btrfs_device *dev, u64 flags,
2480 2481
		       u64 gen, int mirror_num, u8 *csum, int force,
		       u64 physical_for_dev_replace)
2482 2483 2484 2485
{
	struct scrub_block *sblock;
	int index;

2486
	sblock = kzalloc(sizeof(*sblock), GFP_KERNEL);
2487
	if (!sblock) {
2488 2489 2490
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		spin_unlock(&sctx->stat_lock);
2491
		return -ENOMEM;
A
Arne Jansen 已提交
2492
	}
2493

2494 2495
	/* one ref inside this function, plus one for each page added to
	 * a bio later on */
2496
	refcount_set(&sblock->refs, 1);
2497
	sblock->sctx = sctx;
2498 2499 2500
	sblock->no_io_error_seen = 1;

	for (index = 0; len > 0; index++) {
2501
		struct scrub_page *spage;
2502 2503
		u64 l = min_t(u64, len, PAGE_SIZE);

2504
		spage = kzalloc(sizeof(*spage), GFP_KERNEL);
2505 2506
		if (!spage) {
leave_nomem:
2507 2508 2509
			spin_lock(&sctx->stat_lock);
			sctx->stat.malloc_errors++;
			spin_unlock(&sctx->stat_lock);
2510
			scrub_block_put(sblock);
2511 2512
			return -ENOMEM;
		}
2513 2514 2515
		BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
		scrub_page_get(spage);
		sblock->pagev[index] = spage;
2516
		spage->sblock = sblock;
2517
		spage->dev = dev;
2518 2519 2520 2521
		spage->flags = flags;
		spage->generation = gen;
		spage->logical = logical;
		spage->physical = physical;
2522
		spage->physical_for_dev_replace = physical_for_dev_replace;
2523 2524 2525
		spage->mirror_num = mirror_num;
		if (csum) {
			spage->have_csum = 1;
2526
			memcpy(spage->csum, csum, sctx->csum_size);
2527 2528 2529 2530
		} else {
			spage->have_csum = 0;
		}
		sblock->page_count++;
2531
		spage->page = alloc_page(GFP_KERNEL);
2532 2533
		if (!spage->page)
			goto leave_nomem;
2534 2535 2536
		len -= l;
		logical += l;
		physical += l;
2537
		physical_for_dev_replace += l;
2538 2539
	}

2540
	WARN_ON(sblock->page_count == 0);
2541 2542 2543 2544 2545 2546 2547 2548 2549 2550
	if (dev->missing) {
		/*
		 * This case should only be hit for RAID 5/6 device replace. See
		 * the comment in scrub_missing_raid56_pages() for details.
		 */
		scrub_missing_raid56_pages(sblock);
	} else {
		for (index = 0; index < sblock->page_count; index++) {
			struct scrub_page *spage = sblock->pagev[index];
			int ret;
2551

2552 2553 2554 2555 2556
			ret = scrub_add_page_to_rd_bio(sctx, spage);
			if (ret) {
				scrub_block_put(sblock);
				return ret;
			}
2557
		}
A
Arne Jansen 已提交
2558

2559 2560 2561
		if (force)
			scrub_submit(sctx);
	}
A
Arne Jansen 已提交
2562

2563 2564
	/* last one frees, either here or in bio completion for last page */
	scrub_block_put(sblock);
A
Arne Jansen 已提交
2565 2566 2567
	return 0;
}

2568
static void scrub_bio_end_io(struct bio *bio)
2569 2570
{
	struct scrub_bio *sbio = bio->bi_private;
2571
	struct btrfs_fs_info *fs_info = sbio->dev->fs_info;
2572

2573
	sbio->err = bio->bi_error;
2574 2575
	sbio->bio = bio;

2576
	btrfs_queue_work(fs_info->scrub_workers, &sbio->work);
2577 2578 2579 2580 2581
}

static void scrub_bio_end_io_worker(struct btrfs_work *work)
{
	struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
2582
	struct scrub_ctx *sctx = sbio->sctx;
2583 2584
	int i;

2585
	BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO);
2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606
	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;
2607 2608 2609 2610
	spin_lock(&sctx->list_lock);
	sbio->next_free = sctx->first_free;
	sctx->first_free = sbio->index;
	spin_unlock(&sctx->list_lock);
2611 2612

	if (sctx->is_dev_replace &&
2613 2614
	    atomic_read(&sctx->flush_all_writes)) {
		mutex_lock(&sctx->wr_lock);
2615
		scrub_wr_submit(sctx);
2616
		mutex_unlock(&sctx->wr_lock);
2617 2618
	}

2619
	scrub_pending_bio_dec(sctx);
2620 2621
}

2622 2623 2624 2625
static inline void __scrub_mark_bitmap(struct scrub_parity *sparity,
				       unsigned long *bitmap,
				       u64 start, u64 len)
{
2626
	u64 offset;
2627
	int nsectors;
2628
	int sectorsize = sparity->sctx->fs_info->sectorsize;
2629 2630 2631 2632 2633 2634 2635

	if (len >= sparity->stripe_len) {
		bitmap_set(bitmap, 0, sparity->nsectors);
		return;
	}

	start -= sparity->logic_start;
2636 2637
	start = div64_u64_rem(start, sparity->stripe_len, &offset);
	offset = div_u64(offset, sectorsize);
2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660
	nsectors = (int)len / sectorsize;

	if (offset + nsectors <= sparity->nsectors) {
		bitmap_set(bitmap, offset, nsectors);
		return;
	}

	bitmap_set(bitmap, offset, sparity->nsectors - offset);
	bitmap_set(bitmap, 0, nsectors - (sparity->nsectors - offset));
}

static inline void scrub_parity_mark_sectors_error(struct scrub_parity *sparity,
						   u64 start, u64 len)
{
	__scrub_mark_bitmap(sparity, sparity->ebitmap, start, len);
}

static inline void scrub_parity_mark_sectors_data(struct scrub_parity *sparity,
						  u64 start, u64 len)
{
	__scrub_mark_bitmap(sparity, sparity->dbitmap, start, len);
}

2661 2662
static void scrub_block_complete(struct scrub_block *sblock)
{
2663 2664
	int corrupted = 0;

2665
	if (!sblock->no_io_error_seen) {
2666
		corrupted = 1;
2667
		scrub_handle_errored_block(sblock);
2668 2669 2670 2671 2672 2673
	} else {
		/*
		 * if has checksum error, write via repair mechanism in
		 * dev replace case, otherwise write here in dev replace
		 * case.
		 */
2674 2675
		corrupted = scrub_checksum(sblock);
		if (!corrupted && sblock->sctx->is_dev_replace)
2676 2677
			scrub_write_block_to_dev_replace(sblock);
	}
2678 2679 2680 2681 2682 2683 2684 2685 2686

	if (sblock->sparity && corrupted && !sblock->data_corrected) {
		u64 start = sblock->pagev[0]->logical;
		u64 end = sblock->pagev[sblock->page_count - 1]->logical +
			  PAGE_SIZE;

		scrub_parity_mark_sectors_error(sblock->sparity,
						start, end - start);
	}
2687 2688
}

2689
static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u8 *csum)
A
Arne Jansen 已提交
2690 2691
{
	struct btrfs_ordered_sum *sum = NULL;
2692
	unsigned long index;
A
Arne Jansen 已提交
2693 2694
	unsigned long num_sectors;

2695 2696
	while (!list_empty(&sctx->csum_list)) {
		sum = list_first_entry(&sctx->csum_list,
A
Arne Jansen 已提交
2697 2698 2699 2700 2701 2702
				       struct btrfs_ordered_sum, list);
		if (sum->bytenr > logical)
			return 0;
		if (sum->bytenr + sum->len > logical)
			break;

2703
		++sctx->stat.csum_discards;
A
Arne Jansen 已提交
2704 2705 2706 2707 2708 2709 2710
		list_del(&sum->list);
		kfree(sum);
		sum = NULL;
	}
	if (!sum)
		return 0;

2711 2712
	index = ((u32)(logical - sum->bytenr)) / sctx->fs_info->sectorsize;
	num_sectors = sum->len / sctx->fs_info->sectorsize;
2713 2714
	memcpy(csum, sum->sums + index, sctx->csum_size);
	if (index == num_sectors - 1) {
A
Arne Jansen 已提交
2715 2716 2717
		list_del(&sum->list);
		kfree(sum);
	}
2718
	return 1;
A
Arne Jansen 已提交
2719 2720 2721
}

/* scrub extent tries to collect up to 64 kB for each bio */
2722
static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len,
2723
			u64 physical, struct btrfs_device *dev, u64 flags,
2724
			u64 gen, int mirror_num, u64 physical_for_dev_replace)
A
Arne Jansen 已提交
2725 2726 2727
{
	int ret;
	u8 csum[BTRFS_CSUM_SIZE];
2728 2729 2730
	u32 blocksize;

	if (flags & BTRFS_EXTENT_FLAG_DATA) {
2731
		blocksize = sctx->fs_info->sectorsize;
2732 2733 2734 2735
		spin_lock(&sctx->stat_lock);
		sctx->stat.data_extents_scrubbed++;
		sctx->stat.data_bytes_scrubbed += len;
		spin_unlock(&sctx->stat_lock);
2736
	} else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2737
		blocksize = sctx->fs_info->nodesize;
2738 2739 2740 2741
		spin_lock(&sctx->stat_lock);
		sctx->stat.tree_extents_scrubbed++;
		sctx->stat.tree_bytes_scrubbed += len;
		spin_unlock(&sctx->stat_lock);
2742
	} else {
2743
		blocksize = sctx->fs_info->sectorsize;
2744
		WARN_ON(1);
2745
	}
A
Arne Jansen 已提交
2746 2747

	while (len) {
2748
		u64 l = min_t(u64, len, blocksize);
A
Arne Jansen 已提交
2749 2750 2751 2752
		int have_csum = 0;

		if (flags & BTRFS_EXTENT_FLAG_DATA) {
			/* push csums to sbio */
2753
			have_csum = scrub_find_csum(sctx, logical, csum);
A
Arne Jansen 已提交
2754
			if (have_csum == 0)
2755
				++sctx->stat.no_csum;
2756 2757 2758 2759 2760 2761
			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
2762
		}
2763
		ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen,
2764 2765 2766
				  mirror_num, have_csum ? csum : NULL, 0,
				  physical_for_dev_replace);
behind_scrub_pages:
A
Arne Jansen 已提交
2767 2768 2769 2770 2771
		if (ret)
			return ret;
		len -= l;
		logical += l;
		physical += l;
2772
		physical_for_dev_replace += l;
A
Arne Jansen 已提交
2773 2774 2775 2776
	}
	return 0;
}

2777 2778 2779 2780 2781 2782 2783 2784 2785
static int scrub_pages_for_parity(struct scrub_parity *sparity,
				  u64 logical, u64 len,
				  u64 physical, struct btrfs_device *dev,
				  u64 flags, u64 gen, int mirror_num, u8 *csum)
{
	struct scrub_ctx *sctx = sparity->sctx;
	struct scrub_block *sblock;
	int index;

2786
	sblock = kzalloc(sizeof(*sblock), GFP_KERNEL);
2787 2788 2789 2790 2791 2792 2793 2794 2795
	if (!sblock) {
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		spin_unlock(&sctx->stat_lock);
		return -ENOMEM;
	}

	/* one ref inside this function, plus one for each page added to
	 * a bio later on */
2796
	refcount_set(&sblock->refs, 1);
2797 2798 2799 2800 2801 2802 2803 2804 2805
	sblock->sctx = sctx;
	sblock->no_io_error_seen = 1;
	sblock->sparity = sparity;
	scrub_parity_get(sparity);

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

2806
		spage = kzalloc(sizeof(*spage), GFP_KERNEL);
2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835
		if (!spage) {
leave_nomem:
			spin_lock(&sctx->stat_lock);
			sctx->stat.malloc_errors++;
			spin_unlock(&sctx->stat_lock);
			scrub_block_put(sblock);
			return -ENOMEM;
		}
		BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
		/* For scrub block */
		scrub_page_get(spage);
		sblock->pagev[index] = spage;
		/* For scrub parity */
		scrub_page_get(spage);
		list_add_tail(&spage->list, &sparity->spages);
		spage->sblock = sblock;
		spage->dev = dev;
		spage->flags = flags;
		spage->generation = gen;
		spage->logical = logical;
		spage->physical = physical;
		spage->mirror_num = mirror_num;
		if (csum) {
			spage->have_csum = 1;
			memcpy(spage->csum, csum, sctx->csum_size);
		} else {
			spage->have_csum = 0;
		}
		sblock->page_count++;
2836
		spage->page = alloc_page(GFP_KERNEL);
2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870
		if (!spage->page)
			goto leave_nomem;
		len -= l;
		logical += l;
		physical += l;
	}

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

		ret = scrub_add_page_to_rd_bio(sctx, spage);
		if (ret) {
			scrub_block_put(sblock);
			return ret;
		}
	}

	/* last one frees, either here or in bio completion for last page */
	scrub_block_put(sblock);
	return 0;
}

static int scrub_extent_for_parity(struct scrub_parity *sparity,
				   u64 logical, u64 len,
				   u64 physical, struct btrfs_device *dev,
				   u64 flags, u64 gen, int mirror_num)
{
	struct scrub_ctx *sctx = sparity->sctx;
	int ret;
	u8 csum[BTRFS_CSUM_SIZE];
	u32 blocksize;

2871 2872 2873 2874 2875
	if (dev->missing) {
		scrub_parity_mark_sectors_error(sparity, logical, len);
		return 0;
	}

2876
	if (flags & BTRFS_EXTENT_FLAG_DATA) {
2877
		blocksize = sctx->fs_info->sectorsize;
2878
	} else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2879
		blocksize = sctx->fs_info->nodesize;
2880
	} else {
2881
		blocksize = sctx->fs_info->sectorsize;
2882 2883 2884 2885 2886 2887 2888 2889 2890
		WARN_ON(1);
	}

	while (len) {
		u64 l = min_t(u64, len, blocksize);
		int have_csum = 0;

		if (flags & BTRFS_EXTENT_FLAG_DATA) {
			/* push csums to sbio */
2891
			have_csum = scrub_find_csum(sctx, logical, csum);
2892 2893 2894 2895 2896 2897 2898 2899
			if (have_csum == 0)
				goto skip;
		}
		ret = scrub_pages_for_parity(sparity, logical, l, physical, dev,
					     flags, gen, mirror_num,
					     have_csum ? csum : NULL);
		if (ret)
			return ret;
2900
skip:
2901 2902 2903 2904 2905 2906 2907
		len -= l;
		logical += l;
		physical += l;
	}
	return 0;
}

2908 2909 2910 2911 2912 2913 2914 2915
/*
 * Given a physical address, this will calculate it's
 * logical offset. if this is a parity stripe, it will return
 * the most left data stripe's logical offset.
 *
 * return 0 if it is a data stripe, 1 means parity stripe.
 */
static int get_raid56_logic_offset(u64 physical, int num,
2916 2917
				   struct map_lookup *map, u64 *offset,
				   u64 *stripe_start)
2918 2919 2920 2921 2922
{
	int i;
	int j = 0;
	u64 stripe_nr;
	u64 last_offset;
2923 2924
	u32 stripe_index;
	u32 rot;
2925 2926 2927

	last_offset = (physical - map->stripes[num].physical) *
		      nr_data_stripes(map);
2928 2929 2930
	if (stripe_start)
		*stripe_start = last_offset;

2931 2932 2933 2934
	*offset = last_offset;
	for (i = 0; i < nr_data_stripes(map); i++) {
		*offset = last_offset + i * map->stripe_len;

2935
		stripe_nr = div64_u64(*offset, map->stripe_len);
2936
		stripe_nr = div_u64(stripe_nr, nr_data_stripes(map));
2937 2938

		/* Work out the disk rotation on this stripe-set */
2939
		stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, &rot);
2940 2941
		/* calculate which stripe this data locates */
		rot += i;
2942
		stripe_index = rot % map->num_stripes;
2943 2944 2945 2946 2947 2948 2949 2950 2951
		if (stripe_index == num)
			return 0;
		if (stripe_index < num)
			j++;
	}
	*offset = last_offset + j * map->stripe_len;
	return 1;
}

2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973
static void scrub_free_parity(struct scrub_parity *sparity)
{
	struct scrub_ctx *sctx = sparity->sctx;
	struct scrub_page *curr, *next;
	int nbits;

	nbits = bitmap_weight(sparity->ebitmap, sparity->nsectors);
	if (nbits) {
		spin_lock(&sctx->stat_lock);
		sctx->stat.read_errors += nbits;
		sctx->stat.uncorrectable_errors += nbits;
		spin_unlock(&sctx->stat_lock);
	}

	list_for_each_entry_safe(curr, next, &sparity->spages, list) {
		list_del_init(&curr->list);
		scrub_page_put(curr);
	}

	kfree(sparity);
}

2974 2975 2976 2977 2978 2979 2980 2981 2982 2983
static void scrub_parity_bio_endio_worker(struct btrfs_work *work)
{
	struct scrub_parity *sparity = container_of(work, struct scrub_parity,
						    work);
	struct scrub_ctx *sctx = sparity->sctx;

	scrub_free_parity(sparity);
	scrub_pending_bio_dec(sctx);
}

2984
static void scrub_parity_bio_endio(struct bio *bio)
2985 2986
{
	struct scrub_parity *sparity = (struct scrub_parity *)bio->bi_private;
2987
	struct btrfs_fs_info *fs_info = sparity->sctx->fs_info;
2988

2989
	if (bio->bi_error)
2990 2991 2992 2993
		bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap,
			  sparity->nsectors);

	bio_put(bio);
2994 2995 2996

	btrfs_init_work(&sparity->work, btrfs_scrubparity_helper,
			scrub_parity_bio_endio_worker, NULL, NULL);
2997
	btrfs_queue_work(fs_info->scrub_parity_workers, &sparity->work);
2998 2999 3000 3001 3002
}

static void scrub_parity_check_and_repair(struct scrub_parity *sparity)
{
	struct scrub_ctx *sctx = sparity->sctx;
3003
	struct btrfs_fs_info *fs_info = sctx->fs_info;
3004 3005 3006 3007 3008 3009 3010 3011 3012 3013
	struct bio *bio;
	struct btrfs_raid_bio *rbio;
	struct btrfs_bio *bbio = NULL;
	u64 length;
	int ret;

	if (!bitmap_andnot(sparity->dbitmap, sparity->dbitmap, sparity->ebitmap,
			   sparity->nsectors))
		goto out;

3014
	length = sparity->logic_end - sparity->logic_start;
3015 3016

	btrfs_bio_counter_inc_blocked(fs_info);
3017
	ret = btrfs_map_sblock(fs_info, BTRFS_MAP_WRITE, sparity->logic_start,
3018
			       &length, &bbio);
3019
	if (ret || !bbio || !bbio->raid_map)
3020 3021 3022 3023 3024 3025 3026
		goto bbio_out;

	bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
	bio->bi_iter.bi_sector = sparity->logic_start >> 9;
	bio->bi_private = sparity;
	bio->bi_end_io = scrub_parity_bio_endio;

3027
	rbio = raid56_parity_alloc_scrub_rbio(fs_info, bio, bbio,
3028
					      length, sparity->scrub_dev,
3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040
					      sparity->dbitmap,
					      sparity->nsectors);
	if (!rbio)
		goto rbio_out;

	scrub_pending_bio_inc(sctx);
	raid56_parity_submit_scrub_rbio(rbio);
	return;

rbio_out:
	bio_put(bio);
bbio_out:
3041
	btrfs_bio_counter_dec(fs_info);
3042
	btrfs_put_bbio(bbio);
3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053
	bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap,
		  sparity->nsectors);
	spin_lock(&sctx->stat_lock);
	sctx->stat.malloc_errors++;
	spin_unlock(&sctx->stat_lock);
out:
	scrub_free_parity(sparity);
}

static inline int scrub_calc_parity_bitmap_len(int nsectors)
{
3054
	return DIV_ROUND_UP(nsectors, BITS_PER_LONG) * sizeof(long);
3055 3056 3057 3058
}

static void scrub_parity_get(struct scrub_parity *sparity)
{
3059
	refcount_inc(&sparity->refs);
3060 3061 3062 3063
}

static void scrub_parity_put(struct scrub_parity *sparity)
{
3064
	if (!refcount_dec_and_test(&sparity->refs))
3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076
		return;

	scrub_parity_check_and_repair(sparity);
}

static noinline_for_stack int scrub_raid56_parity(struct scrub_ctx *sctx,
						  struct map_lookup *map,
						  struct btrfs_device *sdev,
						  struct btrfs_path *path,
						  u64 logic_start,
						  u64 logic_end)
{
3077
	struct btrfs_fs_info *fs_info = sctx->fs_info;
3078 3079 3080
	struct btrfs_root *root = fs_info->extent_root;
	struct btrfs_root *csum_root = fs_info->csum_root;
	struct btrfs_extent_item *extent;
3081
	struct btrfs_bio *bbio = NULL;
3082 3083 3084 3085 3086 3087 3088 3089 3090
	u64 flags;
	int ret;
	int slot;
	struct extent_buffer *l;
	struct btrfs_key key;
	u64 generation;
	u64 extent_logical;
	u64 extent_physical;
	u64 extent_len;
3091
	u64 mapped_length;
3092 3093 3094 3095 3096 3097 3098
	struct btrfs_device *extent_dev;
	struct scrub_parity *sparity;
	int nsectors;
	int bitmap_len;
	int extent_mirror_num;
	int stop_loop = 0;

3099
	nsectors = div_u64(map->stripe_len, fs_info->sectorsize);
3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115
	bitmap_len = scrub_calc_parity_bitmap_len(nsectors);
	sparity = kzalloc(sizeof(struct scrub_parity) + 2 * bitmap_len,
			  GFP_NOFS);
	if (!sparity) {
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		spin_unlock(&sctx->stat_lock);
		return -ENOMEM;
	}

	sparity->stripe_len = map->stripe_len;
	sparity->nsectors = nsectors;
	sparity->sctx = sctx;
	sparity->scrub_dev = sdev;
	sparity->logic_start = logic_start;
	sparity->logic_end = logic_end;
3116
	refcount_set(&sparity->refs, 1);
3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 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
	INIT_LIST_HEAD(&sparity->spages);
	sparity->dbitmap = sparity->bitmap;
	sparity->ebitmap = (void *)sparity->bitmap + bitmap_len;

	ret = 0;
	while (logic_start < logic_end) {
		if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
			key.type = BTRFS_METADATA_ITEM_KEY;
		else
			key.type = BTRFS_EXTENT_ITEM_KEY;
		key.objectid = logic_start;
		key.offset = (u64)-1;

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

		if (ret > 0) {
			ret = btrfs_previous_extent_item(root, path, 0);
			if (ret < 0)
				goto out;
			if (ret > 0) {
				btrfs_release_path(path);
				ret = btrfs_search_slot(NULL, root, &key,
							path, 0, 0);
				if (ret < 0)
					goto out;
			}
		}

		stop_loop = 0;
		while (1) {
			u64 bytes;

			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;

				stop_loop = 1;
				break;
			}
			btrfs_item_key_to_cpu(l, &key, slot);

3165 3166 3167 3168
			if (key.type != BTRFS_EXTENT_ITEM_KEY &&
			    key.type != BTRFS_METADATA_ITEM_KEY)
				goto next;

3169
			if (key.type == BTRFS_METADATA_ITEM_KEY)
3170
				bytes = fs_info->nodesize;
3171 3172 3173 3174 3175 3176
			else
				bytes = key.offset;

			if (key.objectid + bytes <= logic_start)
				goto next;

3177
			if (key.objectid >= logic_end) {
3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189
				stop_loop = 1;
				break;
			}

			while (key.objectid >= logic_start + map->stripe_len)
				logic_start += map->stripe_len;

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

3190 3191 3192 3193
			if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) &&
			    (key.objectid < logic_start ||
			     key.objectid + bytes >
			     logic_start + map->stripe_len)) {
J
Jeff Mahoney 已提交
3194 3195
				btrfs_err(fs_info,
					  "scrub: tree block %llu spanning stripes, ignored. logical=%llu",
3196
					  key.objectid, logic_start);
3197 3198 3199
				spin_lock(&sctx->stat_lock);
				sctx->stat.uncorrectable_errors++;
				spin_unlock(&sctx->stat_lock);
3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218
				goto next;
			}
again:
			extent_logical = key.objectid;
			extent_len = bytes;

			if (extent_logical < logic_start) {
				extent_len -= logic_start - extent_logical;
				extent_logical = logic_start;
			}

			if (extent_logical + extent_len >
			    logic_start + map->stripe_len)
				extent_len = logic_start + map->stripe_len -
					     extent_logical;

			scrub_parity_mark_sectors_data(sparity, extent_logical,
						       extent_len);

3219
			mapped_length = extent_len;
3220
			bbio = NULL;
3221 3222 3223
			ret = btrfs_map_block(fs_info, BTRFS_MAP_READ,
					extent_logical, &mapped_length, &bbio,
					0);
3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235
			if (!ret) {
				if (!bbio || mapped_length < extent_len)
					ret = -EIO;
			}
			if (ret) {
				btrfs_put_bbio(bbio);
				goto out;
			}
			extent_physical = bbio->stripes[0].physical;
			extent_mirror_num = bbio->mirror_num;
			extent_dev = bbio->stripes[0].dev;
			btrfs_put_bbio(bbio);
3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249

			ret = btrfs_lookup_csums_range(csum_root,
						extent_logical,
						extent_logical + extent_len - 1,
						&sctx->csum_list, 1);
			if (ret)
				goto out;

			ret = scrub_extent_for_parity(sparity, extent_logical,
						      extent_len,
						      extent_physical,
						      extent_dev, flags,
						      generation,
						      extent_mirror_num);
3250 3251 3252

			scrub_free_csums(sctx);

3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283
			if (ret)
				goto out;

			if (extent_logical + extent_len <
			    key.objectid + bytes) {
				logic_start += map->stripe_len;

				if (logic_start >= logic_end) {
					stop_loop = 1;
					break;
				}

				if (logic_start < key.objectid + bytes) {
					cond_resched();
					goto again;
				}
			}
next:
			path->slots[0]++;
		}

		btrfs_release_path(path);

		if (stop_loop)
			break;

		logic_start += map->stripe_len;
	}
out:
	if (ret < 0)
		scrub_parity_mark_sectors_error(sparity, logic_start,
3284
						logic_end - logic_start);
3285 3286
	scrub_parity_put(sparity);
	scrub_submit(sctx);
3287
	mutex_lock(&sctx->wr_lock);
3288
	scrub_wr_submit(sctx);
3289
	mutex_unlock(&sctx->wr_lock);
3290 3291 3292 3293 3294

	btrfs_release_path(path);
	return ret < 0 ? ret : 0;
}

3295
static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
3296 3297
					   struct map_lookup *map,
					   struct btrfs_device *scrub_dev,
3298 3299
					   int num, u64 base, u64 length,
					   int is_dev_replace)
A
Arne Jansen 已提交
3300
{
3301
	struct btrfs_path *path, *ppath;
3302
	struct btrfs_fs_info *fs_info = sctx->fs_info;
A
Arne Jansen 已提交
3303 3304 3305
	struct btrfs_root *root = fs_info->extent_root;
	struct btrfs_root *csum_root = fs_info->csum_root;
	struct btrfs_extent_item *extent;
3306
	struct blk_plug plug;
A
Arne Jansen 已提交
3307 3308 3309 3310 3311 3312 3313
	u64 flags;
	int ret;
	int slot;
	u64 nstripes;
	struct extent_buffer *l;
	u64 physical;
	u64 logical;
L
Liu Bo 已提交
3314
	u64 logic_end;
3315
	u64 physical_end;
A
Arne Jansen 已提交
3316
	u64 generation;
3317
	int mirror_num;
A
Arne Jansen 已提交
3318 3319
	struct reada_control *reada1;
	struct reada_control *reada2;
3320
	struct btrfs_key key;
A
Arne Jansen 已提交
3321
	struct btrfs_key key_end;
A
Arne Jansen 已提交
3322 3323
	u64 increment = map->stripe_len;
	u64 offset;
3324 3325 3326
	u64 extent_logical;
	u64 extent_physical;
	u64 extent_len;
3327 3328
	u64 stripe_logical;
	u64 stripe_end;
3329 3330
	struct btrfs_device *extent_dev;
	int extent_mirror_num;
3331
	int stop_loop = 0;
D
David Woodhouse 已提交
3332

3333
	physical = map->stripes[num].physical;
A
Arne Jansen 已提交
3334
	offset = 0;
3335
	nstripes = div64_u64(length, map->stripe_len);
A
Arne Jansen 已提交
3336 3337 3338
	if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
		offset = map->stripe_len * num;
		increment = map->stripe_len * map->num_stripes;
3339
		mirror_num = 1;
A
Arne Jansen 已提交
3340 3341 3342 3343
	} 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;
3344
		mirror_num = num % map->sub_stripes + 1;
A
Arne Jansen 已提交
3345 3346
	} else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
		increment = map->stripe_len;
3347
		mirror_num = num % map->num_stripes + 1;
A
Arne Jansen 已提交
3348 3349
	} else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
		increment = map->stripe_len;
3350
		mirror_num = num % map->num_stripes + 1;
3351
	} else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3352
		get_raid56_logic_offset(physical, num, map, &offset, NULL);
3353 3354
		increment = map->stripe_len * nr_data_stripes(map);
		mirror_num = 1;
A
Arne Jansen 已提交
3355 3356
	} else {
		increment = map->stripe_len;
3357
		mirror_num = 1;
A
Arne Jansen 已提交
3358 3359 3360 3361 3362 3363
	}

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

3364 3365
	ppath = btrfs_alloc_path();
	if (!ppath) {
3366
		btrfs_free_path(path);
3367 3368 3369
		return -ENOMEM;
	}

3370 3371 3372 3373 3374
	/*
	 * 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
3375 3376 3377
	path->search_commit_root = 1;
	path->skip_locking = 1;

3378 3379
	ppath->search_commit_root = 1;
	ppath->skip_locking = 1;
A
Arne Jansen 已提交
3380
	/*
A
Arne Jansen 已提交
3381 3382 3383
	 * 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
3384 3385
	 */
	logical = base + offset;
3386
	physical_end = physical + nstripes * map->stripe_len;
3387
	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3388
		get_raid56_logic_offset(physical_end, num,
3389
					map, &logic_end, NULL);
3390 3391 3392 3393
		logic_end += base;
	} else {
		logic_end = logical + increment * nstripes;
	}
3394
	wait_event(sctx->list_wait,
3395
		   atomic_read(&sctx->bios_in_flight) == 0);
3396
	scrub_blocked_if_needed(fs_info);
A
Arne Jansen 已提交
3397 3398

	/* FIXME it might be better to start readahead at commit root */
3399 3400 3401
	key.objectid = logical;
	key.type = BTRFS_EXTENT_ITEM_KEY;
	key.offset = (u64)0;
3402
	key_end.objectid = logic_end;
3403 3404
	key_end.type = BTRFS_METADATA_ITEM_KEY;
	key_end.offset = (u64)-1;
3405
	reada1 = btrfs_reada_add(root, &key, &key_end);
A
Arne Jansen 已提交
3406

3407 3408 3409
	key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
	key.type = BTRFS_EXTENT_CSUM_KEY;
	key.offset = logical;
A
Arne Jansen 已提交
3410 3411
	key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
	key_end.type = BTRFS_EXTENT_CSUM_KEY;
3412
	key_end.offset = logic_end;
3413
	reada2 = btrfs_reada_add(csum_root, &key, &key_end);
A
Arne Jansen 已提交
3414 3415 3416 3417 3418 3419

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

A
Arne Jansen 已提交
3420 3421 3422 3423 3424

	/*
	 * collect all data csums for the stripe to avoid seeking during
	 * the scrub. This might currently (crc32) end up to be about 1MB
	 */
3425
	blk_start_plug(&plug);
A
Arne Jansen 已提交
3426 3427 3428 3429 3430

	/*
	 * now find all extents for each stripe and scrub them
	 */
	ret = 0;
3431
	while (physical < physical_end) {
A
Arne Jansen 已提交
3432 3433 3434 3435
		/*
		 * canceled?
		 */
		if (atomic_read(&fs_info->scrub_cancel_req) ||
3436
		    atomic_read(&sctx->cancel_req)) {
A
Arne Jansen 已提交
3437 3438 3439 3440 3441 3442 3443 3444
			ret = -ECANCELED;
			goto out;
		}
		/*
		 * check to see if we have to pause
		 */
		if (atomic_read(&fs_info->scrub_pause_req)) {
			/* push queued extents */
3445
			atomic_set(&sctx->flush_all_writes, 1);
3446
			scrub_submit(sctx);
3447
			mutex_lock(&sctx->wr_lock);
3448
			scrub_wr_submit(sctx);
3449
			mutex_unlock(&sctx->wr_lock);
3450
			wait_event(sctx->list_wait,
3451
				   atomic_read(&sctx->bios_in_flight) == 0);
3452
			atomic_set(&sctx->flush_all_writes, 0);
3453
			scrub_blocked_if_needed(fs_info);
A
Arne Jansen 已提交
3454 3455
		}

3456 3457 3458 3459 3460 3461
		if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
			ret = get_raid56_logic_offset(physical, num, map,
						      &logical,
						      &stripe_logical);
			logical += base;
			if (ret) {
3462
				/* it is parity strip */
3463
				stripe_logical += base;
3464
				stripe_end = stripe_logical + increment;
3465 3466 3467 3468 3469 3470 3471 3472 3473
				ret = scrub_raid56_parity(sctx, map, scrub_dev,
							  ppath, stripe_logical,
							  stripe_end);
				if (ret)
					goto out;
				goto skip;
			}
		}

3474 3475 3476 3477
		if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
			key.type = BTRFS_METADATA_ITEM_KEY;
		else
			key.type = BTRFS_EXTENT_ITEM_KEY;
A
Arne Jansen 已提交
3478
		key.objectid = logical;
L
Liu Bo 已提交
3479
		key.offset = (u64)-1;
A
Arne Jansen 已提交
3480 3481 3482 3483

		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
		if (ret < 0)
			goto out;
3484

3485
		if (ret > 0) {
3486
			ret = btrfs_previous_extent_item(root, path, 0);
A
Arne Jansen 已提交
3487 3488
			if (ret < 0)
				goto out;
3489 3490 3491 3492 3493 3494 3495 3496 3497
			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
3498 3499
		}

L
Liu Bo 已提交
3500
		stop_loop = 0;
A
Arne Jansen 已提交
3501
		while (1) {
3502 3503
			u64 bytes;

A
Arne Jansen 已提交
3504 3505 3506 3507 3508 3509 3510 3511 3512
			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
3513
				stop_loop = 1;
A
Arne Jansen 已提交
3514 3515 3516 3517
				break;
			}
			btrfs_item_key_to_cpu(l, &key, slot);

3518 3519 3520 3521
			if (key.type != BTRFS_EXTENT_ITEM_KEY &&
			    key.type != BTRFS_METADATA_ITEM_KEY)
				goto next;

3522
			if (key.type == BTRFS_METADATA_ITEM_KEY)
3523
				bytes = fs_info->nodesize;
3524 3525 3526 3527
			else
				bytes = key.offset;

			if (key.objectid + bytes <= logical)
A
Arne Jansen 已提交
3528 3529
				goto next;

L
Liu Bo 已提交
3530 3531 3532 3533 3534 3535
			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
3536 3537 3538 3539 3540 3541

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

3542 3543 3544 3545
			if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) &&
			    (key.objectid < logical ||
			     key.objectid + bytes >
			     logical + map->stripe_len)) {
3546
				btrfs_err(fs_info,
J
Jeff Mahoney 已提交
3547
					   "scrub: tree block %llu spanning stripes, ignored. logical=%llu",
3548
				       key.objectid, logical);
3549 3550 3551
				spin_lock(&sctx->stat_lock);
				sctx->stat.uncorrectable_errors++;
				spin_unlock(&sctx->stat_lock);
A
Arne Jansen 已提交
3552 3553 3554
				goto next;
			}

L
Liu Bo 已提交
3555 3556 3557 3558
again:
			extent_logical = key.objectid;
			extent_len = bytes;

A
Arne Jansen 已提交
3559 3560 3561
			/*
			 * trim extent to this stripe
			 */
L
Liu Bo 已提交
3562 3563 3564
			if (extent_logical < logical) {
				extent_len -= logical - extent_logical;
				extent_logical = logical;
A
Arne Jansen 已提交
3565
			}
L
Liu Bo 已提交
3566
			if (extent_logical + extent_len >
A
Arne Jansen 已提交
3567
			    logical + map->stripe_len) {
L
Liu Bo 已提交
3568 3569
				extent_len = logical + map->stripe_len -
					     extent_logical;
A
Arne Jansen 已提交
3570 3571
			}

L
Liu Bo 已提交
3572
			extent_physical = extent_logical - logical + physical;
3573 3574 3575 3576 3577 3578 3579
			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
3580

3581 3582 3583 3584 3585
			ret = btrfs_lookup_csums_range(csum_root,
						       extent_logical,
						       extent_logical +
						       extent_len - 1,
						       &sctx->csum_list, 1);
L
Liu Bo 已提交
3586 3587 3588
			if (ret)
				goto out;

3589 3590 3591
			ret = scrub_extent(sctx, extent_logical, extent_len,
					   extent_physical, extent_dev, flags,
					   generation, extent_mirror_num,
3592
					   extent_logical - logical + physical);
3593 3594 3595

			scrub_free_csums(sctx);

A
Arne Jansen 已提交
3596 3597 3598
			if (ret)
				goto out;

L
Liu Bo 已提交
3599 3600
			if (extent_logical + extent_len <
			    key.objectid + bytes) {
3601
				if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3602 3603 3604 3605
					/*
					 * loop until we find next data stripe
					 * or we have finished all stripes.
					 */
3606 3607 3608 3609 3610 3611 3612 3613 3614 3615
loop:
					physical += map->stripe_len;
					ret = get_raid56_logic_offset(physical,
							num, map, &logical,
							&stripe_logical);
					logical += base;

					if (ret && physical < physical_end) {
						stripe_logical += base;
						stripe_end = stripe_logical +
3616
								increment;
3617 3618 3619 3620 3621 3622 3623 3624
						ret = scrub_raid56_parity(sctx,
							map, scrub_dev, ppath,
							stripe_logical,
							stripe_end);
						if (ret)
							goto out;
						goto loop;
					}
3625 3626 3627 3628
				} else {
					physical += map->stripe_len;
					logical += increment;
				}
L
Liu Bo 已提交
3629 3630 3631 3632 3633
				if (logical < key.objectid + bytes) {
					cond_resched();
					goto again;
				}

3634
				if (physical >= physical_end) {
L
Liu Bo 已提交
3635 3636 3637 3638
					stop_loop = 1;
					break;
				}
			}
A
Arne Jansen 已提交
3639 3640 3641
next:
			path->slots[0]++;
		}
C
Chris Mason 已提交
3642
		btrfs_release_path(path);
3643
skip:
A
Arne Jansen 已提交
3644 3645
		logical += increment;
		physical += map->stripe_len;
3646
		spin_lock(&sctx->stat_lock);
L
Liu Bo 已提交
3647 3648 3649 3650 3651
		if (stop_loop)
			sctx->stat.last_physical = map->stripes[num].physical +
						   length;
		else
			sctx->stat.last_physical = physical;
3652
		spin_unlock(&sctx->stat_lock);
L
Liu Bo 已提交
3653 3654
		if (stop_loop)
			break;
A
Arne Jansen 已提交
3655
	}
3656
out:
A
Arne Jansen 已提交
3657
	/* push queued extents */
3658
	scrub_submit(sctx);
3659
	mutex_lock(&sctx->wr_lock);
3660
	scrub_wr_submit(sctx);
3661
	mutex_unlock(&sctx->wr_lock);
A
Arne Jansen 已提交
3662

3663
	blk_finish_plug(&plug);
A
Arne Jansen 已提交
3664
	btrfs_free_path(path);
3665
	btrfs_free_path(ppath);
A
Arne Jansen 已提交
3666 3667 3668
	return ret < 0 ? ret : 0;
}

3669
static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx,
3670 3671
					  struct btrfs_device *scrub_dev,
					  u64 chunk_offset, u64 length,
3672 3673 3674
					  u64 dev_offset,
					  struct btrfs_block_group_cache *cache,
					  int is_dev_replace)
A
Arne Jansen 已提交
3675
{
3676 3677
	struct btrfs_fs_info *fs_info = sctx->fs_info;
	struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
A
Arne Jansen 已提交
3678 3679 3680
	struct map_lookup *map;
	struct extent_map *em;
	int i;
3681
	int ret = 0;
A
Arne Jansen 已提交
3682 3683 3684 3685 3686

	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);

3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698
	if (!em) {
		/*
		 * Might have been an unused block group deleted by the cleaner
		 * kthread or relocation.
		 */
		spin_lock(&cache->lock);
		if (!cache->removed)
			ret = -EINVAL;
		spin_unlock(&cache->lock);

		return ret;
	}
A
Arne Jansen 已提交
3699

3700
	map = em->map_lookup;
A
Arne Jansen 已提交
3701 3702 3703 3704 3705 3706 3707
	if (em->start != chunk_offset)
		goto out;

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

	for (i = 0; i < map->num_stripes; ++i) {
3708
		if (map->stripes[i].dev->bdev == scrub_dev->bdev &&
3709
		    map->stripes[i].physical == dev_offset) {
3710
			ret = scrub_stripe(sctx, map, scrub_dev, i,
3711 3712
					   chunk_offset, length,
					   is_dev_replace);
A
Arne Jansen 已提交
3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723
			if (ret)
				goto out;
		}
	}
out:
	free_extent_map(em);

	return ret;
}

static noinline_for_stack
3724
int scrub_enumerate_chunks(struct scrub_ctx *sctx,
3725 3726
			   struct btrfs_device *scrub_dev, u64 start, u64 end,
			   int is_dev_replace)
A
Arne Jansen 已提交
3727 3728 3729
{
	struct btrfs_dev_extent *dev_extent = NULL;
	struct btrfs_path *path;
3730 3731
	struct btrfs_fs_info *fs_info = sctx->fs_info;
	struct btrfs_root *root = fs_info->dev_root;
A
Arne Jansen 已提交
3732 3733
	u64 length;
	u64 chunk_offset;
3734
	int ret = 0;
3735
	int ro_set;
A
Arne Jansen 已提交
3736 3737 3738 3739 3740
	int slot;
	struct extent_buffer *l;
	struct btrfs_key key;
	struct btrfs_key found_key;
	struct btrfs_block_group_cache *cache;
3741
	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
A
Arne Jansen 已提交
3742 3743 3744 3745 3746

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

3747
	path->reada = READA_FORWARD;
A
Arne Jansen 已提交
3748 3749 3750
	path->search_commit_root = 1;
	path->skip_locking = 1;

3751
	key.objectid = scrub_dev->devid;
A
Arne Jansen 已提交
3752 3753 3754 3755 3756 3757
	key.offset = 0ull;
	key.type = BTRFS_DEV_EXTENT_KEY;

	while (1) {
		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
		if (ret < 0)
3758 3759 3760 3761 3762
			break;
		if (ret > 0) {
			if (path->slots[0] >=
			    btrfs_header_nritems(path->nodes[0])) {
				ret = btrfs_next_leaf(root, path);
3763 3764 3765 3766
				if (ret < 0)
					break;
				if (ret > 0) {
					ret = 0;
3767
					break;
3768 3769 3770
				}
			} else {
				ret = 0;
3771 3772
			}
		}
A
Arne Jansen 已提交
3773 3774 3775 3776 3777 3778

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

		btrfs_item_key_to_cpu(l, &found_key, slot);

3779
		if (found_key.objectid != scrub_dev->devid)
A
Arne Jansen 已提交
3780 3781
			break;

3782
		if (found_key.type != BTRFS_DEV_EXTENT_KEY)
A
Arne Jansen 已提交
3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793
			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);

3794 3795
		if (found_key.offset + length <= start)
			goto skip;
A
Arne Jansen 已提交
3796 3797 3798 3799 3800 3801 3802 3803

		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);
3804 3805 3806 3807 3808 3809

		/* some chunks are removed but not committed to disk yet,
		 * continue scrubbing */
		if (!cache)
			goto skip;

3810 3811 3812 3813 3814 3815 3816 3817 3818
		/*
		 * we need call btrfs_inc_block_group_ro() with scrubs_paused,
		 * to avoid deadlock caused by:
		 * btrfs_inc_block_group_ro()
		 * -> btrfs_wait_for_commit()
		 * -> btrfs_commit_transaction()
		 * -> btrfs_scrub_pause()
		 */
		scrub_pause_on(fs_info);
3819
		ret = btrfs_inc_block_group_ro(fs_info, cache);
3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850
		if (!ret && is_dev_replace) {
			/*
			 * If we are doing a device replace wait for any tasks
			 * that started dellaloc right before we set the block
			 * group to RO mode, as they might have just allocated
			 * an extent from it or decided they could do a nocow
			 * write. And if any such tasks did that, wait for their
			 * ordered extents to complete and then commit the
			 * current transaction, so that we can later see the new
			 * extent items in the extent tree - the ordered extents
			 * create delayed data references (for cow writes) when
			 * they complete, which will be run and insert the
			 * corresponding extent items into the extent tree when
			 * we commit the transaction they used when running
			 * inode.c:btrfs_finish_ordered_io(). We later use
			 * the commit root of the extent tree to find extents
			 * to copy from the srcdev into the tgtdev, and we don't
			 * want to miss any new extents.
			 */
			btrfs_wait_block_group_reservations(cache);
			btrfs_wait_nocow_writers(cache);
			ret = btrfs_wait_ordered_roots(fs_info, -1,
						       cache->key.objectid,
						       cache->key.offset);
			if (ret > 0) {
				struct btrfs_trans_handle *trans;

				trans = btrfs_join_transaction(root);
				if (IS_ERR(trans))
					ret = PTR_ERR(trans);
				else
3851
					ret = btrfs_commit_transaction(trans);
3852 3853 3854 3855 3856 3857 3858
				if (ret) {
					scrub_pause_off(fs_info);
					btrfs_put_block_group(cache);
					break;
				}
			}
		}
3859
		scrub_pause_off(fs_info);
3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871 3872

		if (ret == 0) {
			ro_set = 1;
		} else if (ret == -ENOSPC) {
			/*
			 * btrfs_inc_block_group_ro return -ENOSPC when it
			 * failed in creating new chunk for metadata.
			 * It is not a problem for scrub/replace, because
			 * metadata are always cowed, and our scrub paused
			 * commit_transactions.
			 */
			ro_set = 0;
		} else {
J
Jeff Mahoney 已提交
3873 3874
			btrfs_warn(fs_info,
				   "failed setting block group ro, ret=%d\n",
3875
				   ret);
3876 3877 3878 3879
			btrfs_put_block_group(cache);
			break;
		}

3880
		btrfs_dev_replace_lock(&fs_info->dev_replace, 1);
3881 3882 3883
		dev_replace->cursor_right = found_key.offset + length;
		dev_replace->cursor_left = found_key.offset;
		dev_replace->item_needs_writeback = 1;
3884
		btrfs_dev_replace_unlock(&fs_info->dev_replace, 1);
3885
		ret = scrub_chunk(sctx, scrub_dev, chunk_offset, length,
3886
				  found_key.offset, cache, is_dev_replace);
3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897

		/*
		 * 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.
		 */
3898
		atomic_set(&sctx->flush_all_writes, 1);
3899
		scrub_submit(sctx);
3900
		mutex_lock(&sctx->wr_lock);
3901
		scrub_wr_submit(sctx);
3902
		mutex_unlock(&sctx->wr_lock);
3903 3904 3905

		wait_event(sctx->list_wait,
			   atomic_read(&sctx->bios_in_flight) == 0);
3906 3907

		scrub_pause_on(fs_info);
3908 3909 3910 3911 3912 3913

		/*
		 * must be called before we decrease @scrub_paused.
		 * make sure we don't block transaction commit while
		 * we are waiting pending workers finished.
		 */
3914 3915
		wait_event(sctx->list_wait,
			   atomic_read(&sctx->workers_pending) == 0);
3916
		atomic_set(&sctx->flush_all_writes, 0);
3917

3918
		scrub_pause_off(fs_info);
3919

3920 3921 3922 3923 3924
		btrfs_dev_replace_lock(&fs_info->dev_replace, 1);
		dev_replace->cursor_left = dev_replace->cursor_right;
		dev_replace->item_needs_writeback = 1;
		btrfs_dev_replace_unlock(&fs_info->dev_replace, 1);

3925
		if (ro_set)
3926
			btrfs_dec_block_group_ro(cache);
3927

3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949
		/*
		 * We might have prevented the cleaner kthread from deleting
		 * this block group if it was already unused because we raced
		 * and set it to RO mode first. So add it back to the unused
		 * list, otherwise it might not ever be deleted unless a manual
		 * balance is triggered or it becomes used and unused again.
		 */
		spin_lock(&cache->lock);
		if (!cache->removed && !cache->ro && cache->reserved == 0 &&
		    btrfs_block_group_used(&cache->item) == 0) {
			spin_unlock(&cache->lock);
			spin_lock(&fs_info->unused_bgs_lock);
			if (list_empty(&cache->bg_list)) {
				btrfs_get_block_group(cache);
				list_add_tail(&cache->bg_list,
					      &fs_info->unused_bgs);
			}
			spin_unlock(&fs_info->unused_bgs_lock);
		} else {
			spin_unlock(&cache->lock);
		}

A
Arne Jansen 已提交
3950 3951 3952
		btrfs_put_block_group(cache);
		if (ret)
			break;
3953 3954
		if (is_dev_replace &&
		    atomic64_read(&dev_replace->num_write_errors) > 0) {
3955 3956 3957 3958 3959 3960 3961
			ret = -EIO;
			break;
		}
		if (sctx->stat.malloc_errors > 0) {
			ret = -ENOMEM;
			break;
		}
3962
skip:
A
Arne Jansen 已提交
3963
		key.offset = found_key.offset + length;
C
Chris Mason 已提交
3964
		btrfs_release_path(path);
A
Arne Jansen 已提交
3965 3966 3967
	}

	btrfs_free_path(path);
3968

3969
	return ret;
A
Arne Jansen 已提交
3970 3971
}

3972 3973
static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx,
					   struct btrfs_device *scrub_dev)
A
Arne Jansen 已提交
3974 3975 3976 3977 3978
{
	int	i;
	u64	bytenr;
	u64	gen;
	int	ret;
3979
	struct btrfs_fs_info *fs_info = sctx->fs_info;
A
Arne Jansen 已提交
3980

3981
	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3982 3983
		return -EIO;

3984
	/* Seed devices of a new filesystem has their own generation. */
3985
	if (scrub_dev->fs_devices != fs_info->fs_devices)
3986 3987
		gen = scrub_dev->generation;
	else
3988
		gen = fs_info->last_trans_committed;
A
Arne Jansen 已提交
3989 3990 3991

	for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
		bytenr = btrfs_sb_offset(i);
3992 3993
		if (bytenr + BTRFS_SUPER_INFO_SIZE >
		    scrub_dev->commit_total_bytes)
A
Arne Jansen 已提交
3994 3995
			break;

3996
		ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr,
3997
				  scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i,
3998
				  NULL, 1, bytenr);
A
Arne Jansen 已提交
3999 4000 4001
		if (ret)
			return ret;
	}
4002
	wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
A
Arne Jansen 已提交
4003 4004 4005 4006 4007 4008 4009

	return 0;
}

/*
 * get a reference count on fs_info->scrub_workers. start worker if necessary
 */
4010 4011
static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info,
						int is_dev_replace)
A
Arne Jansen 已提交
4012
{
4013
	unsigned int flags = WQ_FREEZABLE | WQ_UNBOUND;
4014
	int max_active = fs_info->thread_pool_size;
A
Arne Jansen 已提交
4015

A
Arne Jansen 已提交
4016
	if (fs_info->scrub_workers_refcnt == 0) {
4017
		if (is_dev_replace)
4018
			fs_info->scrub_workers =
4019
				btrfs_alloc_workqueue(fs_info, "scrub", flags,
4020
						      1, 4);
4021
		else
4022
			fs_info->scrub_workers =
4023
				btrfs_alloc_workqueue(fs_info, "scrub", flags,
4024
						      max_active, 4);
4025 4026 4027
		if (!fs_info->scrub_workers)
			goto fail_scrub_workers;

4028
		fs_info->scrub_wr_completion_workers =
4029
			btrfs_alloc_workqueue(fs_info, "scrubwrc", flags,
4030
					      max_active, 2);
4031 4032 4033
		if (!fs_info->scrub_wr_completion_workers)
			goto fail_scrub_wr_completion_workers;

4034
		fs_info->scrub_nocow_workers =
4035
			btrfs_alloc_workqueue(fs_info, "scrubnc", flags, 1, 0);
4036 4037
		if (!fs_info->scrub_nocow_workers)
			goto fail_scrub_nocow_workers;
4038
		fs_info->scrub_parity_workers =
4039
			btrfs_alloc_workqueue(fs_info, "scrubparity", flags,
4040
					      max_active, 2);
4041 4042
		if (!fs_info->scrub_parity_workers)
			goto fail_scrub_parity_workers;
A
Arne Jansen 已提交
4043
	}
A
Arne Jansen 已提交
4044
	++fs_info->scrub_workers_refcnt;
4045 4046 4047 4048 4049 4050 4051 4052 4053 4054
	return 0;

fail_scrub_parity_workers:
	btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
fail_scrub_nocow_workers:
	btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
fail_scrub_wr_completion_workers:
	btrfs_destroy_workqueue(fs_info->scrub_workers);
fail_scrub_workers:
	return -ENOMEM;
A
Arne Jansen 已提交
4055 4056
}

4057
static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info)
A
Arne Jansen 已提交
4058
{
4059
	if (--fs_info->scrub_workers_refcnt == 0) {
4060 4061 4062
		btrfs_destroy_workqueue(fs_info->scrub_workers);
		btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
		btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
4063
		btrfs_destroy_workqueue(fs_info->scrub_parity_workers);
4064
	}
A
Arne Jansen 已提交
4065 4066 4067
	WARN_ON(fs_info->scrub_workers_refcnt < 0);
}

4068 4069
int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start,
		    u64 end, struct btrfs_scrub_progress *progress,
4070
		    int readonly, int is_dev_replace)
A
Arne Jansen 已提交
4071
{
4072
	struct scrub_ctx *sctx;
A
Arne Jansen 已提交
4073 4074
	int ret;
	struct btrfs_device *dev;
4075
	struct rcu_string *name;
A
Arne Jansen 已提交
4076

4077
	if (btrfs_fs_closing(fs_info))
A
Arne Jansen 已提交
4078 4079
		return -EINVAL;

4080
	if (fs_info->nodesize > BTRFS_STRIPE_LEN) {
4081 4082 4083 4084 4085
		/*
		 * 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.
		 */
4086 4087
		btrfs_err(fs_info,
			   "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails",
4088 4089
		       fs_info->nodesize,
		       BTRFS_STRIPE_LEN);
4090 4091 4092
		return -EINVAL;
	}

4093
	if (fs_info->sectorsize != PAGE_SIZE) {
4094
		/* not supported for data w/o checksums */
4095
		btrfs_err_rl(fs_info,
J
Jeff Mahoney 已提交
4096
			   "scrub: size assumption sectorsize != PAGE_SIZE (%d != %lu) fails",
4097
		       fs_info->sectorsize, PAGE_SIZE);
A
Arne Jansen 已提交
4098 4099 4100
		return -EINVAL;
	}

4101
	if (fs_info->nodesize >
4102
	    PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK ||
4103
	    fs_info->sectorsize > PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) {
4104 4105 4106 4107
		/*
		 * would exhaust the array bounds of pagev member in
		 * struct scrub_block
		 */
J
Jeff Mahoney 已提交
4108 4109
		btrfs_err(fs_info,
			  "scrub: size assumption nodesize and sectorsize <= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails",
4110
		       fs_info->nodesize,
4111
		       SCRUB_MAX_PAGES_PER_BLOCK,
4112
		       fs_info->sectorsize,
4113 4114 4115 4116
		       SCRUB_MAX_PAGES_PER_BLOCK);
		return -EINVAL;
	}

A
Arne Jansen 已提交
4117

4118 4119
	mutex_lock(&fs_info->fs_devices->device_list_mutex);
	dev = btrfs_find_device(fs_info, devid, NULL, NULL);
4120
	if (!dev || (dev->missing && !is_dev_replace)) {
4121
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
A
Arne Jansen 已提交
4122 4123 4124
		return -ENODEV;
	}

4125 4126 4127 4128 4129 4130 4131 4132 4133 4134
	if (!is_dev_replace && !readonly && !dev->writeable) {
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
		rcu_read_lock();
		name = rcu_dereference(dev->name);
		btrfs_err(fs_info, "scrub: device %s is not writable",
			  name->str);
		rcu_read_unlock();
		return -EROFS;
	}

4135
	mutex_lock(&fs_info->scrub_lock);
4136
	if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) {
A
Arne Jansen 已提交
4137
		mutex_unlock(&fs_info->scrub_lock);
4138 4139
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
		return -EIO;
A
Arne Jansen 已提交
4140 4141
	}

4142
	btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
4143 4144 4145
	if (dev->scrub_device ||
	    (!is_dev_replace &&
	     btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) {
4146
		btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
A
Arne Jansen 已提交
4147
		mutex_unlock(&fs_info->scrub_lock);
4148
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
A
Arne Jansen 已提交
4149 4150
		return -EINPROGRESS;
	}
4151
	btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
4152 4153 4154 4155 4156 4157 4158 4159

	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;
	}

4160
	sctx = scrub_setup_ctx(dev, is_dev_replace);
4161
	if (IS_ERR(sctx)) {
A
Arne Jansen 已提交
4162
		mutex_unlock(&fs_info->scrub_lock);
4163 4164
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
		scrub_workers_put(fs_info);
4165
		return PTR_ERR(sctx);
A
Arne Jansen 已提交
4166
	}
4167 4168
	sctx->readonly = readonly;
	dev->scrub_device = sctx;
4169
	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
A
Arne Jansen 已提交
4170

4171 4172 4173 4174
	/*
	 * checking @scrub_pause_req here, we can avoid
	 * race between committing transaction and scrubbing.
	 */
4175
	__scrub_blocked_if_needed(fs_info);
A
Arne Jansen 已提交
4176 4177 4178
	atomic_inc(&fs_info->scrubs_running);
	mutex_unlock(&fs_info->scrub_lock);

4179
	if (!is_dev_replace) {
4180 4181 4182 4183
		/*
		 * by holding device list mutex, we can
		 * kick off writing super in log tree sync.
		 */
4184
		mutex_lock(&fs_info->fs_devices->device_list_mutex);
4185
		ret = scrub_supers(sctx, dev);
4186
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4187
	}
A
Arne Jansen 已提交
4188 4189

	if (!ret)
4190 4191
		ret = scrub_enumerate_chunks(sctx, dev, start, end,
					     is_dev_replace);
A
Arne Jansen 已提交
4192

4193
	wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
A
Arne Jansen 已提交
4194 4195 4196
	atomic_dec(&fs_info->scrubs_running);
	wake_up(&fs_info->scrub_pause_wait);

4197
	wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0);
4198

A
Arne Jansen 已提交
4199
	if (progress)
4200
		memcpy(progress, &sctx->stat, sizeof(*progress));
A
Arne Jansen 已提交
4201 4202 4203

	mutex_lock(&fs_info->scrub_lock);
	dev->scrub_device = NULL;
4204
	scrub_workers_put(fs_info);
A
Arne Jansen 已提交
4205 4206
	mutex_unlock(&fs_info->scrub_lock);

4207
	scrub_put_ctx(sctx);
A
Arne Jansen 已提交
4208 4209 4210 4211

	return ret;
}

4212
void btrfs_scrub_pause(struct btrfs_fs_info *fs_info)
A
Arne Jansen 已提交
4213 4214 4215 4216 4217 4218 4219 4220 4221 4222 4223 4224 4225 4226
{
	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);
}

4227
void btrfs_scrub_continue(struct btrfs_fs_info *fs_info)
A
Arne Jansen 已提交
4228 4229 4230 4231 4232
{
	atomic_dec(&fs_info->scrub_pause_req);
	wake_up(&fs_info->scrub_pause_wait);
}

4233
int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info)
A
Arne Jansen 已提交
4234 4235 4236 4237 4238 4239 4240 4241 4242 4243 4244 4245 4246 4247 4248 4249 4250 4251 4252 4253
{
	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;
}

4254 4255
int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info,
			   struct btrfs_device *dev)
4256
{
4257
	struct scrub_ctx *sctx;
A
Arne Jansen 已提交
4258 4259

	mutex_lock(&fs_info->scrub_lock);
4260 4261
	sctx = dev->scrub_device;
	if (!sctx) {
A
Arne Jansen 已提交
4262 4263 4264
		mutex_unlock(&fs_info->scrub_lock);
		return -ENOTCONN;
	}
4265
	atomic_inc(&sctx->cancel_req);
A
Arne Jansen 已提交
4266 4267 4268 4269 4270 4271 4272 4273 4274 4275
	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
4276

4277
int btrfs_scrub_progress(struct btrfs_fs_info *fs_info, u64 devid,
A
Arne Jansen 已提交
4278 4279 4280
			 struct btrfs_scrub_progress *progress)
{
	struct btrfs_device *dev;
4281
	struct scrub_ctx *sctx = NULL;
A
Arne Jansen 已提交
4282

4283 4284
	mutex_lock(&fs_info->fs_devices->device_list_mutex);
	dev = btrfs_find_device(fs_info, devid, NULL, NULL);
A
Arne Jansen 已提交
4285
	if (dev)
4286 4287 4288
		sctx = dev->scrub_device;
	if (sctx)
		memcpy(progress, &sctx->stat, sizeof(*progress));
4289
	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
A
Arne Jansen 已提交
4290

4291
	return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV;
A
Arne Jansen 已提交
4292
}
4293 4294 4295 4296 4297 4298 4299 4300 4301 4302 4303 4304

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;
4305
	ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, extent_logical,
4306 4307 4308
			      &mapped_length, &bbio, 0);
	if (ret || !bbio || mapped_length < extent_len ||
	    !bbio->stripes[0].dev->bdev) {
4309
		btrfs_put_bbio(bbio);
4310 4311 4312 4313 4314 4315
		return;
	}

	*extent_physical = bbio->stripes[0].physical;
	*extent_mirror_num = bbio->mirror_num;
	*extent_dev = bbio->stripes[0].dev;
4316
	btrfs_put_bbio(bbio);
4317 4318 4319 4320 4321 4322
}

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;
4323
	struct btrfs_fs_info *fs_info = sctx->fs_info;
4324 4325 4326 4327 4328 4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339

	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;
4340 4341
	btrfs_init_work(&nocow_ctx->work, btrfs_scrubnc_helper,
			copy_nocow_pages_worker, NULL, NULL);
4342
	INIT_LIST_HEAD(&nocow_ctx->inodes);
4343 4344
	btrfs_queue_work(fs_info->scrub_nocow_workers,
			 &nocow_ctx->work);
4345 4346 4347 4348

	return 0;
}

4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365
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

4366 4367 4368 4369 4370
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;
4371 4372
	struct btrfs_fs_info *fs_info = sctx->fs_info;
	struct btrfs_root *root = fs_info->extent_root;
4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389 4390 4391 4392 4393 4394 4395 4396 4397
	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_path *path;
	int not_written = 0;

	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,
4398
					  record_inode_for_nocow, nocow_ctx);
4399
	if (ret != 0 && ret != -ENOENT) {
J
Jeff Mahoney 已提交
4400 4401 4402 4403
		btrfs_warn(fs_info,
			   "iterate_inodes_from_logical() failed: log %llu, phys %llu, len %llu, mir %u, ret %d",
			   logical, physical_for_dev_replace, len, mirror_num,
			   ret);
4404 4405 4406 4407
		not_written = 1;
		goto out;
	}

4408
	btrfs_end_transaction(trans);
4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425
	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;
		}
	}
4426
out:
4427 4428 4429 4430 4431 4432 4433 4434
	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);
	}
4435
	if (trans && !IS_ERR(trans))
4436
		btrfs_end_transaction(trans);
4437 4438 4439 4440 4441 4442 4443 4444 4445 4446
	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);
}

4447
static int check_extent_to_block(struct btrfs_inode *inode, u64 start, u64 len,
4448 4449 4450 4451 4452 4453 4454 4455 4456
				 u64 logical)
{
	struct extent_state *cached_state = NULL;
	struct btrfs_ordered_extent *ordered;
	struct extent_io_tree *io_tree;
	struct extent_map *em;
	u64 lockstart = start, lockend = start + len - 1;
	int ret = 0;

4457
	io_tree = &inode->io_tree;
4458

4459
	lock_extent_bits(io_tree, lockstart, lockend, &cached_state);
4460
	ordered = btrfs_lookup_ordered_range(inode, lockstart, len);
4461 4462 4463 4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490
	if (ordered) {
		btrfs_put_ordered_extent(ordered);
		ret = 1;
		goto out_unlock;
	}

	em = btrfs_get_extent(inode, NULL, 0, start, 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 > logical ||
	    em->block_start + em->block_len < logical + len) {
		free_extent_map(em);
		ret = 1;
		goto out_unlock;
	}
	free_extent_map(em);

out_unlock:
	unlock_extent_cached(io_tree, lockstart, lockend, &cached_state,
			     GFP_NOFS);
	return ret;
}

4491 4492
static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
				      struct scrub_copy_nocow_ctx *nocow_ctx)
4493
{
4494
	struct btrfs_fs_info *fs_info = nocow_ctx->sctx->fs_info;
4495
	struct btrfs_key key;
4496 4497
	struct inode *inode;
	struct page *page;
4498
	struct btrfs_root *local_root;
4499
	struct extent_io_tree *io_tree;
4500
	u64 physical_for_dev_replace;
4501
	u64 nocow_ctx_logical;
4502
	u64 len = nocow_ctx->len;
4503
	unsigned long index;
4504
	int srcu_index;
4505 4506
	int ret = 0;
	int err = 0;
4507 4508 4509 4510

	key.objectid = root;
	key.type = BTRFS_ROOT_ITEM_KEY;
	key.offset = (u64)-1;
4511 4512 4513

	srcu_index = srcu_read_lock(&fs_info->subvol_srcu);

4514
	local_root = btrfs_read_fs_root_no_name(fs_info, &key);
4515 4516
	if (IS_ERR(local_root)) {
		srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
4517
		return PTR_ERR(local_root);
4518
	}
4519 4520 4521 4522 4523

	key.type = BTRFS_INODE_ITEM_KEY;
	key.objectid = inum;
	key.offset = 0;
	inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
4524
	srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
4525 4526 4527
	if (IS_ERR(inode))
		return PTR_ERR(inode);

4528
	/* Avoid truncate/dio/punch hole.. */
A
Al Viro 已提交
4529
	inode_lock(inode);
4530 4531
	inode_dio_wait(inode);

4532
	physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
4533
	io_tree = &BTRFS_I(inode)->io_tree;
4534
	nocow_ctx_logical = nocow_ctx->logical;
4535

4536 4537
	ret = check_extent_to_block(BTRFS_I(inode), offset, len,
			nocow_ctx_logical);
4538 4539 4540
	if (ret) {
		ret = ret > 0 ? 0 : ret;
		goto out;
4541 4542
	}

4543 4544
	while (len >= PAGE_SIZE) {
		index = offset >> PAGE_SHIFT;
4545
again:
4546 4547
		page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
		if (!page) {
4548
			btrfs_err(fs_info, "find_or_create_page() failed");
4549
			ret = -ENOMEM;
4550
			goto out;
4551 4552 4553 4554 4555 4556 4557
		}

		if (PageUptodate(page)) {
			if (PageDirty(page))
				goto next_page;
		} else {
			ClearPageError(page);
4558
			err = extent_read_full_page(io_tree, page,
4559 4560
							   btrfs_get_extent,
							   nocow_ctx->mirror_num);
4561 4562
			if (err) {
				ret = err;
4563 4564
				goto next_page;
			}
4565

4566
			lock_page(page);
4567 4568 4569 4570 4571 4572 4573
			/*
			 * 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) {
4574
				unlock_page(page);
4575
				put_page(page);
4576 4577
				goto again;
			}
4578 4579 4580 4581 4582
			if (!PageUptodate(page)) {
				ret = -EIO;
				goto next_page;
			}
		}
4583

4584
		ret = check_extent_to_block(BTRFS_I(inode), offset, len,
4585 4586 4587 4588 4589 4590
					    nocow_ctx_logical);
		if (ret) {
			ret = ret > 0 ? 0 : ret;
			goto next_page;
		}

4591 4592 4593 4594
		err = write_page_nocow(nocow_ctx->sctx,
				       physical_for_dev_replace, page);
		if (err)
			ret = err;
4595
next_page:
4596
		unlock_page(page);
4597
		put_page(page);
4598 4599 4600 4601

		if (ret)
			break;

4602 4603 4604 4605
		offset += PAGE_SIZE;
		physical_for_dev_replace += PAGE_SIZE;
		nocow_ctx_logical += PAGE_SIZE;
		len -= PAGE_SIZE;
4606
	}
4607
	ret = COPY_COMPLETE;
4608
out:
A
Al Viro 已提交
4609
	inode_unlock(inode);
4610
	iput(inode);
4611 4612 4613 4614 4615 4616 4617 4618 4619 4620
	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;

4621
	dev = sctx->wr_tgtdev;
4622 4623 4624
	if (!dev)
		return -EIO;
	if (!dev->bdev) {
4625
		btrfs_warn_rl(dev->fs_info,
4626
			"scrub write_page_nocow(bdev == NULL) is unexpected");
4627 4628
		return -EIO;
	}
4629
	bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
4630 4631
	bio->bi_iter.bi_size = 0;
	bio->bi_iter.bi_sector = physical_for_dev_replace >> 9;
4632
	bio->bi_bdev = dev->bdev;
4633
	bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
4634 4635
	ret = bio_add_page(bio, page, PAGE_SIZE, 0);
	if (ret != PAGE_SIZE) {
4636 4637 4638 4639 4640 4641
leave_with_eio:
		bio_put(bio);
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
		return -EIO;
	}

4642
	if (btrfsic_submit_bio_wait(bio))
4643 4644 4645 4646 4647
		goto leave_with_eio;

	bio_put(bio);
	return 0;
}