scrub.c 106.2 KB
Newer Older
1
// SPDX-License-Identifier: GPL-2.0
A
Arne Jansen 已提交
2
/*
3
 * Copyright (C) 2011, 2012 STRATO.  All rights reserved.
A
Arne Jansen 已提交
4 5 6
 */

#include <linux/blkdev.h>
7
#include <linux/ratelimit.h>
8
#include <linux/sched/mm.h>
A
Arne Jansen 已提交
9 10 11 12
#include "ctree.h"
#include "volumes.h"
#include "disk-io.h"
#include "ordered-data.h"
13
#include "transaction.h"
14
#include "backref.h"
15
#include "extent_io.h"
16
#include "dev-replace.h"
17
#include "check-integrity.h"
18
#include "rcu-string.h"
D
David Woodhouse 已提交
19
#include "raid56.h"
A
Arne Jansen 已提交
20 21 22 23 24 25 26 27 28 29 30 31 32 33

/*
 * 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
 */

34
struct scrub_block;
35
struct scrub_ctx;
A
Arne Jansen 已提交
36

37 38 39 40 41 42 43 44 45
/*
 * 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 */
46 47 48 49 50 51

/*
 * 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.
 */
52
#define SCRUB_MAX_PAGES_PER_BLOCK	16	/* 64k per node/leaf/sector */
A
Arne Jansen 已提交
53

54
struct scrub_recover {
55
	refcount_t		refs;
56 57 58 59
	struct btrfs_bio	*bbio;
	u64			map_length;
};

A
Arne Jansen 已提交
60
struct scrub_page {
61 62
	struct scrub_block	*sblock;
	struct page		*page;
63
	struct btrfs_device	*dev;
64
	struct list_head	list;
A
Arne Jansen 已提交
65 66
	u64			flags;  /* extent flags */
	u64			generation;
67 68
	u64			logical;
	u64			physical;
69
	u64			physical_for_dev_replace;
70
	atomic_t		refs;
71 72 73 74 75
	struct {
		unsigned int	mirror_num:8;
		unsigned int	have_csum:1;
		unsigned int	io_error:1;
	};
A
Arne Jansen 已提交
76
	u8			csum[BTRFS_CSUM_SIZE];
77 78

	struct scrub_recover	*recover;
A
Arne Jansen 已提交
79 80 81 82
};

struct scrub_bio {
	int			index;
83
	struct scrub_ctx	*sctx;
84
	struct btrfs_device	*dev;
A
Arne Jansen 已提交
85
	struct bio		*bio;
86
	blk_status_t		status;
A
Arne Jansen 已提交
87 88
	u64			logical;
	u64			physical;
89 90 91 92 93
#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
94
	int			page_count;
A
Arne Jansen 已提交
95 96 97 98
	int			next_free;
	struct btrfs_work	work;
};

99
struct scrub_block {
100
	struct scrub_page	*pagev[SCRUB_MAX_PAGES_PER_BLOCK];
101 102
	int			page_count;
	atomic_t		outstanding_pages;
103
	refcount_t		refs; /* free mem on transition to zero */
104
	struct scrub_ctx	*sctx;
105
	struct scrub_parity	*sparity;
106 107 108 109
	struct {
		unsigned int	header_error:1;
		unsigned int	checksum_error:1;
		unsigned int	no_io_error_seen:1;
110
		unsigned int	generation_error:1; /* also sets header_error */
111 112 113 114

		/* The following is for the data used to check parity */
		/* It is for the data with checksum */
		unsigned int	data_corrected:1;
115
	};
116
	struct btrfs_work	work;
117 118
};

119 120 121 122 123 124 125 126 127 128 129 130
/* 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;

131
	u64			stripe_len;
132

133
	refcount_t		refs;
134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151

	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];
};

152
struct scrub_ctx {
153
	struct scrub_bio	*bios[SCRUB_BIOS_PER_SCTX];
154
	struct btrfs_fs_info	*fs_info;
A
Arne Jansen 已提交
155 156
	int			first_free;
	int			curr;
157 158
	atomic_t		bios_in_flight;
	atomic_t		workers_pending;
A
Arne Jansen 已提交
159 160 161 162 163
	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
164
	int			readonly;
165
	int			pages_per_rd_bio;
166 167

	int			is_dev_replace;
168 169 170 171 172

	struct scrub_bio        *wr_curr_bio;
	struct mutex            wr_lock;
	int                     pages_per_wr_bio; /* <= SCRUB_PAGES_PER_WR_BIO */
	struct btrfs_device     *wr_tgtdev;
173
	bool                    flush_all_writes;
174

A
Arne Jansen 已提交
175 176 177 178 179
	/*
	 * statistics
	 */
	struct btrfs_scrub_progress stat;
	spinlock_t		stat_lock;
180 181 182 183 184 185 186 187

	/*
	 * 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.
	 */
188
	refcount_t              refs;
A
Arne Jansen 已提交
189 190
};

191 192 193 194
struct scrub_warning {
	struct btrfs_path	*path;
	u64			extent_item_size;
	const char		*errstr;
D
David Sterba 已提交
195
	u64			physical;
196 197 198 199
	u64			logical;
	struct btrfs_device	*dev;
};

200 201 202 203 204 205 206
struct full_stripe_lock {
	struct rb_node node;
	u64 logical;
	u64 refs;
	struct mutex mutex;
};

207 208
static void scrub_pending_bio_inc(struct scrub_ctx *sctx);
static void scrub_pending_bio_dec(struct scrub_ctx *sctx);
209
static int scrub_handle_errored_block(struct scrub_block *sblock_to_check);
210
static int scrub_setup_recheck_block(struct scrub_block *original_sblock,
211
				     struct scrub_block *sblocks_for_recheck);
212
static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
213 214
				struct scrub_block *sblock,
				int retry_failed_mirror);
215
static void scrub_recheck_block_checksum(struct scrub_block *sblock);
216
static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
217
					     struct scrub_block *sblock_good);
218 219 220
static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
					    struct scrub_block *sblock_good,
					    int page_num, int force_write);
221 222 223
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);
224 225 226 227 228
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);
229 230
static void scrub_page_get(struct scrub_page *spage);
static void scrub_page_put(struct scrub_page *spage);
231 232
static void scrub_parity_get(struct scrub_parity *sparity);
static void scrub_parity_put(struct scrub_parity *sparity);
233 234
static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
				    struct scrub_page *spage);
235
static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
236
		       u64 physical, struct btrfs_device *dev, u64 flags,
237 238
		       u64 gen, int mirror_num, u8 *csum, int force,
		       u64 physical_for_dev_replace);
239
static void scrub_bio_end_io(struct bio *bio);
240 241
static void scrub_bio_end_io_worker(struct btrfs_work *work);
static void scrub_block_complete(struct scrub_block *sblock);
242 243 244 245 246 247 248 249
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);
250
static void scrub_wr_bio_end_io(struct bio *bio);
251
static void scrub_wr_bio_end_io_worker(struct btrfs_work *work);
252
static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
253
static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
254
static void scrub_put_ctx(struct scrub_ctx *sctx);
S
Stefan Behrens 已提交
255

256 257 258 259 260
static inline int scrub_is_page_on_raid56(struct scrub_page *page)
{
	return page->recover &&
	       (page->recover->bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK);
}
S
Stefan Behrens 已提交
261

262 263
static void scrub_pending_bio_inc(struct scrub_ctx *sctx)
{
264
	refcount_inc(&sctx->refs);
265 266 267 268 269 270 271
	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);
272
	scrub_put_ctx(sctx);
273 274
}

275
static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
276 277 278 279 280 281 282 283 284
{
	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);
	}
}

285
static void scrub_pause_on(struct btrfs_fs_info *fs_info)
286 287 288
{
	atomic_inc(&fs_info->scrubs_paused);
	wake_up(&fs_info->scrub_pause_wait);
289
}
290

291 292
static void scrub_pause_off(struct btrfs_fs_info *fs_info)
{
293 294 295 296 297 298 299 300
	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);
}

301 302 303 304 305 306
static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
{
	scrub_pause_on(fs_info);
	scrub_pause_off(fs_info);
}

307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325
/*
 * 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;

326
	lockdep_assert_held(&locks_root->lock);
327 328 329 330 331 332 333 334 335 336 337 338 339 340 341

	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;
		}
	}

342 343 344
	/*
	 * Insert new lock.
	 */
345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369
	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;

370
	lockdep_assert_held(&locks_root->lock);
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

	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;
}

525
static void scrub_free_csums(struct scrub_ctx *sctx)
A
Arne Jansen 已提交
526
{
527
	while (!list_empty(&sctx->csum_list)) {
A
Arne Jansen 已提交
528
		struct btrfs_ordered_sum *sum;
529
		sum = list_first_entry(&sctx->csum_list,
A
Arne Jansen 已提交
530 531 532 533 534 535
				       struct btrfs_ordered_sum, list);
		list_del(&sum->list);
		kfree(sum);
	}
}

536
static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx)
A
Arne Jansen 已提交
537 538 539
{
	int i;

540
	if (!sctx)
A
Arne Jansen 已提交
541 542
		return;

543
	/* this can happen when scrub is cancelled */
544 545
	if (sctx->curr != -1) {
		struct scrub_bio *sbio = sctx->bios[sctx->curr];
546 547

		for (i = 0; i < sbio->page_count; i++) {
548
			WARN_ON(!sbio->pagev[i]->page);
549 550 551 552 553
			scrub_block_put(sbio->pagev[i]->sblock);
		}
		bio_put(sbio->bio);
	}

554
	for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
555
		struct scrub_bio *sbio = sctx->bios[i];
A
Arne Jansen 已提交
556 557 558 559 560 561

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

562
	kfree(sctx->wr_curr_bio);
563 564
	scrub_free_csums(sctx);
	kfree(sctx);
A
Arne Jansen 已提交
565 566
}

567 568
static void scrub_put_ctx(struct scrub_ctx *sctx)
{
569
	if (refcount_dec_and_test(&sctx->refs))
570 571 572
		scrub_free_ctx(sctx);
}

573 574
static noinline_for_stack struct scrub_ctx *scrub_setup_ctx(
		struct btrfs_fs_info *fs_info, int is_dev_replace)
A
Arne Jansen 已提交
575
{
576
	struct scrub_ctx *sctx;
A
Arne Jansen 已提交
577 578
	int		i;

579
	sctx = kzalloc(sizeof(*sctx), GFP_KERNEL);
580
	if (!sctx)
A
Arne Jansen 已提交
581
		goto nomem;
582
	refcount_set(&sctx->refs, 1);
583
	sctx->is_dev_replace = is_dev_replace;
584
	sctx->pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO;
585
	sctx->curr = -1;
586
	sctx->fs_info = fs_info;
587
	for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
A
Arne Jansen 已提交
588 589
		struct scrub_bio *sbio;

590
		sbio = kzalloc(sizeof(*sbio), GFP_KERNEL);
A
Arne Jansen 已提交
591 592
		if (!sbio)
			goto nomem;
593
		sctx->bios[i] = sbio;
A
Arne Jansen 已提交
594 595

		sbio->index = i;
596
		sbio->sctx = sctx;
597
		sbio->page_count = 0;
598 599
		btrfs_init_work(&sbio->work, btrfs_scrub_helper,
				scrub_bio_end_io_worker, NULL, NULL);
A
Arne Jansen 已提交
600

601
		if (i != SCRUB_BIOS_PER_SCTX - 1)
602
			sctx->bios[i]->next_free = i + 1;
603
		else
604 605 606
			sctx->bios[i]->next_free = -1;
	}
	sctx->first_free = 0;
607 608
	atomic_set(&sctx->bios_in_flight, 0);
	atomic_set(&sctx->workers_pending, 0);
609 610 611 612 613 614 615
	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);
616

617 618 619
	WARN_ON(sctx->wr_curr_bio != NULL);
	mutex_init(&sctx->wr_lock);
	sctx->wr_curr_bio = NULL;
620
	if (is_dev_replace) {
621
		WARN_ON(!fs_info->dev_replace.tgtdev);
622
		sctx->pages_per_wr_bio = SCRUB_PAGES_PER_WR_BIO;
623
		sctx->wr_tgtdev = fs_info->dev_replace.tgtdev;
624
		sctx->flush_all_writes = false;
625
	}
626

627
	return sctx;
A
Arne Jansen 已提交
628 629

nomem:
630
	scrub_free_ctx(sctx);
A
Arne Jansen 已提交
631 632 633
	return ERR_PTR(-ENOMEM);
}

634 635
static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root,
				     void *warn_ctx)
636 637 638 639 640
{
	u64 isize;
	u32 nlink;
	int ret;
	int i;
641
	unsigned nofs_flag;
642 643
	struct extent_buffer *eb;
	struct btrfs_inode_item *inode_item;
644
	struct scrub_warning *swarn = warn_ctx;
645
	struct btrfs_fs_info *fs_info = swarn->dev->fs_info;
646 647 648
	struct inode_fs_paths *ipath = NULL;
	struct btrfs_root *local_root;
	struct btrfs_key root_key;
649
	struct btrfs_key key;
650 651 652 653 654 655 656 657 658 659

	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;
	}

660 661 662
	/*
	 * this makes the path point to (inum INODE_ITEM ioff)
	 */
663 664 665 666 667
	key.objectid = inum;
	key.type = BTRFS_INODE_ITEM_KEY;
	key.offset = 0;

	ret = btrfs_search_slot(NULL, local_root, &key, swarn->path, 0, 0);
668 669 670 671 672 673 674 675 676 677 678 679
	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);

680 681 682 683 684 685
	/*
	 * 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();
686
	ipath = init_ipath(4096, local_root, swarn->path);
687
	memalloc_nofs_restore(nofs_flag);
688 689 690 691 692
	if (IS_ERR(ipath)) {
		ret = PTR_ERR(ipath);
		ipath = NULL;
		goto err;
	}
693 694 695 696 697 698 699 700 701 702
	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
703
		btrfs_warn_in_rcu(fs_info,
D
David Sterba 已提交
704
"%s at logical %llu on dev %s, physical %llu, root %llu, inode %llu, offset %llu, length %llu, links %u (path: %s)",
J
Jeff Mahoney 已提交
705 706
				  swarn->errstr, swarn->logical,
				  rcu_str_deref(swarn->dev->name),
D
David Sterba 已提交
707
				  swarn->physical,
J
Jeff Mahoney 已提交
708 709 710
				  root, inum, offset,
				  min(isize - offset, (u64)PAGE_SIZE), nlink,
				  (char *)(unsigned long)ipath->fspath->val[i]);
711 712 713 714 715

	free_ipath(ipath);
	return 0;

err:
J
Jeff Mahoney 已提交
716
	btrfs_warn_in_rcu(fs_info,
D
David Sterba 已提交
717
			  "%s at logical %llu on dev %s, physical %llu, root %llu, inode %llu, offset %llu: path resolving failed with ret=%d",
J
Jeff Mahoney 已提交
718 719
			  swarn->errstr, swarn->logical,
			  rcu_str_deref(swarn->dev->name),
D
David Sterba 已提交
720
			  swarn->physical,
J
Jeff Mahoney 已提交
721
			  root, inum, offset, ret);
722 723 724 725 726

	free_ipath(ipath);
	return 0;
}

727
static void scrub_print_warning(const char *errstr, struct scrub_block *sblock)
728
{
729 730
	struct btrfs_device *dev;
	struct btrfs_fs_info *fs_info;
731 732 733 734 735
	struct btrfs_path *path;
	struct btrfs_key found_key;
	struct extent_buffer *eb;
	struct btrfs_extent_item *ei;
	struct scrub_warning swarn;
736 737 738
	unsigned long ptr = 0;
	u64 extent_item_pos;
	u64 flags = 0;
739
	u64 ref_root;
740
	u32 item_size;
741
	u8 ref_level = 0;
742
	int ret;
743

744
	WARN_ON(sblock->page_count < 1);
745
	dev = sblock->pagev[0]->dev;
746
	fs_info = sblock->sctx->fs_info;
747

748
	path = btrfs_alloc_path();
749 750
	if (!path)
		return;
751

D
David Sterba 已提交
752
	swarn.physical = sblock->pagev[0]->physical;
753
	swarn.logical = sblock->pagev[0]->logical;
754
	swarn.errstr = errstr;
755
	swarn.dev = NULL;
756

757 758
	ret = extent_from_logical(fs_info, swarn.logical, path, &found_key,
				  &flags);
759 760 761
	if (ret < 0)
		goto out;

J
Jan Schmidt 已提交
762
	extent_item_pos = swarn.logical - found_key.objectid;
763 764 765 766 767 768
	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]);

769
	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
770
		do {
771 772 773
			ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
						      item_size, &ref_root,
						      &ref_level);
774
			btrfs_warn_in_rcu(fs_info,
D
David Sterba 已提交
775
"%s at logical %llu on dev %s, physical %llu: metadata %s (level %d) in tree %llu",
J
Jeff Mahoney 已提交
776
				errstr, swarn.logical,
777
				rcu_str_deref(dev->name),
D
David Sterba 已提交
778
				swarn.physical,
779 780 781 782
				ref_level ? "node" : "leaf",
				ret < 0 ? -1 : ref_level,
				ret < 0 ? -1 : ref_root);
		} while (ret != 1);
783
		btrfs_release_path(path);
784
	} else {
785
		btrfs_release_path(path);
786
		swarn.path = path;
787
		swarn.dev = dev;
788 789
		iterate_extent_inodes(fs_info, found_key.objectid,
					extent_item_pos, 1,
790
					scrub_print_warning_inode, &swarn, false);
791 792 793 794 795 796
	}

out:
	btrfs_free_path(path);
}

797 798
static inline void scrub_get_recover(struct scrub_recover *recover)
{
799
	refcount_inc(&recover->refs);
800 801
}

802 803
static inline void scrub_put_recover(struct btrfs_fs_info *fs_info,
				     struct scrub_recover *recover)
804
{
805
	if (refcount_dec_and_test(&recover->refs)) {
806
		btrfs_bio_counter_dec(fs_info);
807
		btrfs_put_bbio(recover->bbio);
808 809 810 811
		kfree(recover);
	}
}

A
Arne Jansen 已提交
812
/*
813 814 815 816 817 818
 * 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
819
 */
820
static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
A
Arne Jansen 已提交
821
{
822
	struct scrub_ctx *sctx = sblock_to_check->sctx;
823
	struct btrfs_device *dev;
824 825 826 827 828 829 830 831 832 833 834
	struct btrfs_fs_info *fs_info;
	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;
835
	bool full_stripe_locked;
836
	unsigned int nofs_flag;
837
	static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
838 839 840
				      DEFAULT_RATELIMIT_BURST);

	BUG_ON(sblock_to_check->page_count < 1);
841
	fs_info = sctx->fs_info;
842 843 844 845 846 847 848 849 850 851 852
	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;
	}
853 854 855 856
	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 &
857
			BTRFS_EXTENT_FLAG_DATA);
858 859
	have_csum = sblock_to_check->pagev[0]->have_csum;
	dev = sblock_to_check->pagev[0]->dev;
860

861 862 863 864 865 866 867 868 869 870
	/*
	 * We must use GFP_NOFS because the scrub task might be waiting for a
	 * worker task executing this function and in turn a transaction commit
	 * might be waiting the scrub task to pause (which needs to wait for all
	 * the worker tasks to complete before pausing).
	 * We do allocations in the workers through insert_full_stripe_lock()
	 * and scrub_add_page_to_wr_bio(), which happens down the call chain of
	 * this function.
	 */
	nofs_flag = memalloc_nofs_save();
871 872 873 874 875 876 877 878 879
	/*
	 * 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) {
880
		memalloc_nofs_restore(nofs_flag);
881 882 883 884 885 886 887 888 889
		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;
	}

890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918
	/*
	 * 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.
	 */

919
	sblocks_for_recheck = kcalloc(BTRFS_MAX_MIRRORS,
920
				      sizeof(*sblocks_for_recheck), GFP_KERNEL);
921
	if (!sblocks_for_recheck) {
922 923 924 925 926
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		sctx->stat.read_errors++;
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
927
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
928
		goto out;
A
Arne Jansen 已提交
929 930
	}

931
	/* setup the context, map the logical blocks and alloc the pages */
932
	ret = scrub_setup_recheck_block(sblock_to_check, sblocks_for_recheck);
933
	if (ret) {
934 935 936 937
		spin_lock(&sctx->stat_lock);
		sctx->stat.read_errors++;
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
938
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
939 940 941 942
		goto out;
	}
	BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS);
	sblock_bad = sblocks_for_recheck + failed_mirror_index;
943

944
	/* build and submit the bios for the failed mirror, check checksums */
945
	scrub_recheck_block(fs_info, sblock_bad, 1);
A
Arne Jansen 已提交
946

947 948 949 950 951 952 953 954 955 956
	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)
		 */
957 958
		spin_lock(&sctx->stat_lock);
		sctx->stat.unverified_errors++;
959
		sblock_to_check->data_corrected = 1;
960
		spin_unlock(&sctx->stat_lock);
A
Arne Jansen 已提交
961

962 963
		if (sctx->is_dev_replace)
			scrub_write_block_to_dev_replace(sblock_bad);
964
		goto out;
A
Arne Jansen 已提交
965 966
	}

967
	if (!sblock_bad->no_io_error_seen) {
968 969 970
		spin_lock(&sctx->stat_lock);
		sctx->stat.read_errors++;
		spin_unlock(&sctx->stat_lock);
971 972
		if (__ratelimit(&_rs))
			scrub_print_warning("i/o error", sblock_to_check);
973
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
974
	} else if (sblock_bad->checksum_error) {
975 976 977
		spin_lock(&sctx->stat_lock);
		sctx->stat.csum_errors++;
		spin_unlock(&sctx->stat_lock);
978 979
		if (__ratelimit(&_rs))
			scrub_print_warning("checksum error", sblock_to_check);
980
		btrfs_dev_stat_inc_and_print(dev,
981
					     BTRFS_DEV_STAT_CORRUPTION_ERRS);
982
	} else if (sblock_bad->header_error) {
983 984 985
		spin_lock(&sctx->stat_lock);
		sctx->stat.verify_errors++;
		spin_unlock(&sctx->stat_lock);
986 987 988
		if (__ratelimit(&_rs))
			scrub_print_warning("checksum/header error",
					    sblock_to_check);
989
		if (sblock_bad->generation_error)
990
			btrfs_dev_stat_inc_and_print(dev,
991 992
				BTRFS_DEV_STAT_GENERATION_ERRS);
		else
993
			btrfs_dev_stat_inc_and_print(dev,
994
				BTRFS_DEV_STAT_CORRUPTION_ERRS);
995
	}
A
Arne Jansen 已提交
996

997 998 999 1000
	if (sctx->readonly) {
		ASSERT(!sctx->is_dev_replace);
		goto out;
	}
A
Arne Jansen 已提交
1001

1002 1003
	/*
	 * now build and submit the bios for the other mirrors, check
1004 1005
	 * checksums.
	 * First try to pick the mirror which is completely without I/O
1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016
	 * 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).
	 */
1017
	for (mirror_index = 0; ;mirror_index++) {
1018
		struct scrub_block *sblock_other;
1019

1020 1021
		if (mirror_index == failed_mirror_index)
			continue;
1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044

		/* raid56's mirror can be more than BTRFS_MAX_MIRRORS */
		if (!scrub_is_page_on_raid56(sblock_bad->pagev[0])) {
			if (mirror_index >= BTRFS_MAX_MIRRORS)
				break;
			if (!sblocks_for_recheck[mirror_index].page_count)
				break;

			sblock_other = sblocks_for_recheck + mirror_index;
		} else {
			struct scrub_recover *r = sblock_bad->pagev[0]->recover;
			int max_allowed = r->bbio->num_stripes -
						r->bbio->num_tgtdevs;

			if (mirror_index >= max_allowed)
				break;
			if (!sblocks_for_recheck[1].page_count)
				break;

			ASSERT(failed_mirror_index == 0);
			sblock_other = sblocks_for_recheck + 1;
			sblock_other->pagev[0]->mirror_num = 1 + mirror_index;
		}
1045 1046

		/* build and submit the bios, check checksums */
1047
		scrub_recheck_block(fs_info, sblock_other, 0);
1048 1049

		if (!sblock_other->header_error &&
1050 1051
		    !sblock_other->checksum_error &&
		    sblock_other->no_io_error_seen) {
1052 1053
			if (sctx->is_dev_replace) {
				scrub_write_block_to_dev_replace(sblock_other);
1054
				goto corrected_error;
1055 1056
			} else {
				ret = scrub_repair_block_from_good_copy(
1057 1058 1059
						sblock_bad, sblock_other);
				if (!ret)
					goto corrected_error;
1060
			}
1061 1062
		}
	}
A
Arne Jansen 已提交
1063

1064 1065
	if (sblock_bad->no_io_error_seen && !sctx->is_dev_replace)
		goto did_not_correct_error;
1066 1067 1068

	/*
	 * In case of I/O errors in the area that is supposed to be
1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080
	 * 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
1081
	 * the final checksum succeeds. But this would be a rare
1082 1083 1084 1085 1086 1087 1088 1089
	 * 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
1090
	 */
1091
	success = 1;
1092 1093
	for (page_num = 0; page_num < sblock_bad->page_count;
	     page_num++) {
1094
		struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1095
		struct scrub_block *sblock_other = NULL;
1096

1097 1098
		/* skip no-io-error page in scrub */
		if (!page_bad->io_error && !sctx->is_dev_replace)
A
Arne Jansen 已提交
1099
			continue;
1100

1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111
		if (scrub_is_page_on_raid56(sblock_bad->pagev[0])) {
			/*
			 * In case of dev replace, if raid56 rebuild process
			 * didn't work out correct data, then copy the content
			 * in sblock_bad to make sure target device is identical
			 * to source device, instead of writing garbage data in
			 * sblock_for_recheck array to target device.
			 */
			sblock_other = NULL;
		} else if (page_bad->io_error) {
			/* try to find no-io-error page in mirrors */
1112 1113 1114 1115 1116 1117 1118 1119 1120
			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;
1121 1122
				}
			}
1123 1124
			if (!sblock_other)
				success = 0;
I
Ilya Dryomov 已提交
1125
		}
A
Arne Jansen 已提交
1126

1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139
		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) {
1140
				atomic64_inc(
1141
					&fs_info->dev_replace.num_write_errors);
1142 1143 1144 1145 1146 1147 1148 1149 1150 1151
				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;
1152
		}
A
Arne Jansen 已提交
1153 1154
	}

1155
	if (success && !sctx->is_dev_replace) {
1156 1157 1158 1159 1160 1161 1162 1163 1164 1165
		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.
			 */
1166
			scrub_recheck_block(fs_info, sblock_bad, 1);
1167
			if (!sblock_bad->header_error &&
1168 1169 1170 1171 1172 1173 1174
			    !sblock_bad->checksum_error &&
			    sblock_bad->no_io_error_seen)
				goto corrected_error;
			else
				goto did_not_correct_error;
		} else {
corrected_error:
1175 1176
			spin_lock(&sctx->stat_lock);
			sctx->stat.corrected_errors++;
1177
			sblock_to_check->data_corrected = 1;
1178
			spin_unlock(&sctx->stat_lock);
1179 1180
			btrfs_err_rl_in_rcu(fs_info,
				"fixed up error at logical %llu on dev %s",
1181
				logical, rcu_str_deref(dev->name));
A
Arne Jansen 已提交
1182
		}
1183 1184
	} else {
did_not_correct_error:
1185 1186 1187
		spin_lock(&sctx->stat_lock);
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
1188 1189
		btrfs_err_rl_in_rcu(fs_info,
			"unable to fixup (regular) error at logical %llu on dev %s",
1190
			logical, rcu_str_deref(dev->name));
I
Ilya Dryomov 已提交
1191
	}
A
Arne Jansen 已提交
1192

1193 1194 1195 1196 1197 1198
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;
1199
			struct scrub_recover *recover;
1200 1201
			int page_index;

1202 1203 1204
			for (page_index = 0; page_index < sblock->page_count;
			     page_index++) {
				sblock->pagev[page_index]->sblock = NULL;
1205 1206
				recover = sblock->pagev[page_index]->recover;
				if (recover) {
1207
					scrub_put_recover(fs_info, recover);
1208 1209 1210
					sblock->pagev[page_index]->recover =
									NULL;
				}
1211 1212
				scrub_page_put(sblock->pagev[page_index]);
			}
1213 1214 1215
		}
		kfree(sblocks_for_recheck);
	}
A
Arne Jansen 已提交
1216

1217
	ret = unlock_full_stripe(fs_info, logical, full_stripe_locked);
1218
	memalloc_nofs_restore(nofs_flag);
1219 1220
	if (ret < 0)
		return ret;
1221 1222
	return 0;
}
A
Arne Jansen 已提交
1223

1224
static inline int scrub_nr_raid_mirrors(struct btrfs_bio *bbio)
1225
{
Z
Zhao Lei 已提交
1226 1227 1228 1229 1230
	if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID5)
		return 2;
	else if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID6)
		return 3;
	else
1231 1232 1233
		return (int)bbio->num_stripes;
}

Z
Zhao Lei 已提交
1234 1235
static inline void scrub_stripe_index_and_offset(u64 logical, u64 map_type,
						 u64 *raid_map,
1236 1237 1238 1239 1240 1241 1242
						 u64 mapped_length,
						 int nstripes, int mirror,
						 int *stripe_index,
						 u64 *stripe_offset)
{
	int i;

1243
	if (map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263
		/* 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;
	}
}

1264
static int scrub_setup_recheck_block(struct scrub_block *original_sblock,
1265 1266
				     struct scrub_block *sblocks_for_recheck)
{
1267
	struct scrub_ctx *sctx = original_sblock->sctx;
1268
	struct btrfs_fs_info *fs_info = sctx->fs_info;
1269 1270
	u64 length = original_sblock->page_count * PAGE_SIZE;
	u64 logical = original_sblock->pagev[0]->logical;
1271 1272 1273
	u64 generation = original_sblock->pagev[0]->generation;
	u64 flags = original_sblock->pagev[0]->flags;
	u64 have_csum = original_sblock->pagev[0]->have_csum;
1274 1275 1276 1277 1278 1279
	struct scrub_recover *recover;
	struct btrfs_bio *bbio;
	u64 sublen;
	u64 mapped_length;
	u64 stripe_offset;
	int stripe_index;
1280
	int page_index = 0;
1281
	int mirror_index;
1282
	int nmirrors;
1283 1284 1285
	int ret;

	/*
1286
	 * note: the two members refs and outstanding_pages
1287 1288 1289 1290 1291
	 * are not used (and not set) in the blocks that are used for
	 * the recheck procedure
	 */

	while (length > 0) {
1292 1293 1294
		sublen = min_t(u64, length, PAGE_SIZE);
		mapped_length = sublen;
		bbio = NULL;
A
Arne Jansen 已提交
1295

1296 1297 1298 1299
		/*
		 * with a length of PAGE_SIZE, each returned stripe
		 * represents one mirror
		 */
1300
		btrfs_bio_counter_inc_blocked(fs_info);
1301
		ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
1302
				logical, &mapped_length, &bbio);
1303
		if (ret || !bbio || mapped_length < sublen) {
1304
			btrfs_put_bbio(bbio);
1305
			btrfs_bio_counter_dec(fs_info);
1306 1307
			return -EIO;
		}
A
Arne Jansen 已提交
1308

1309 1310
		recover = kzalloc(sizeof(struct scrub_recover), GFP_NOFS);
		if (!recover) {
1311
			btrfs_put_bbio(bbio);
1312
			btrfs_bio_counter_dec(fs_info);
1313 1314 1315
			return -ENOMEM;
		}

1316
		refcount_set(&recover->refs, 1);
1317 1318 1319
		recover->bbio = bbio;
		recover->map_length = mapped_length;

1320
		BUG_ON(page_index >= SCRUB_MAX_PAGES_PER_BLOCK);
1321

1322
		nmirrors = min(scrub_nr_raid_mirrors(bbio), BTRFS_MAX_MIRRORS);
Z
Zhao Lei 已提交
1323

1324
		for (mirror_index = 0; mirror_index < nmirrors;
1325 1326 1327 1328 1329
		     mirror_index++) {
			struct scrub_block *sblock;
			struct scrub_page *page;

			sblock = sblocks_for_recheck + mirror_index;
1330
			sblock->sctx = sctx;
1331

1332 1333 1334
			page = kzalloc(sizeof(*page), GFP_NOFS);
			if (!page) {
leave_nomem:
1335 1336 1337
				spin_lock(&sctx->stat_lock);
				sctx->stat.malloc_errors++;
				spin_unlock(&sctx->stat_lock);
1338
				scrub_put_recover(fs_info, recover);
1339 1340
				return -ENOMEM;
			}
1341 1342
			scrub_page_get(page);
			sblock->pagev[page_index] = page;
1343 1344 1345
			page->sblock = sblock;
			page->flags = flags;
			page->generation = generation;
1346
			page->logical = logical;
1347 1348 1349 1350 1351
			page->have_csum = have_csum;
			if (have_csum)
				memcpy(page->csum,
				       original_sblock->pagev[0]->csum,
				       sctx->csum_size);
1352

Z
Zhao Lei 已提交
1353 1354 1355
			scrub_stripe_index_and_offset(logical,
						      bbio->map_type,
						      bbio->raid_map,
1356
						      mapped_length,
1357 1358
						      bbio->num_stripes -
						      bbio->num_tgtdevs,
1359 1360 1361 1362 1363 1364 1365
						      mirror_index,
						      &stripe_index,
						      &stripe_offset);
			page->physical = bbio->stripes[stripe_index].physical +
					 stripe_offset;
			page->dev = bbio->stripes[stripe_index].dev;

1366 1367 1368 1369
			BUG_ON(page_index >= original_sblock->page_count);
			page->physical_for_dev_replace =
				original_sblock->pagev[page_index]->
				physical_for_dev_replace;
1370 1371
			/* for missing devices, dev->bdev is NULL */
			page->mirror_num = mirror_index + 1;
1372
			sblock->page_count++;
1373 1374 1375
			page->page = alloc_page(GFP_NOFS);
			if (!page->page)
				goto leave_nomem;
1376 1377 1378

			scrub_get_recover(recover);
			page->recover = recover;
1379
		}
1380
		scrub_put_recover(fs_info, recover);
1381 1382 1383 1384 1385 1386
		length -= sublen;
		logical += sublen;
		page_index++;
	}

	return 0;
I
Ilya Dryomov 已提交
1387 1388
}

1389
static void scrub_bio_wait_endio(struct bio *bio)
1390
{
1391
	complete(bio->bi_private);
1392 1393 1394 1395 1396 1397
}

static int scrub_submit_raid56_bio_wait(struct btrfs_fs_info *fs_info,
					struct bio *bio,
					struct scrub_page *page)
{
1398
	DECLARE_COMPLETION_ONSTACK(done);
1399
	int ret;
1400
	int mirror_num;
1401 1402 1403 1404 1405

	bio->bi_iter.bi_sector = page->logical >> 9;
	bio->bi_private = &done;
	bio->bi_end_io = scrub_bio_wait_endio;

1406
	mirror_num = page->sblock->pagev[0]->mirror_num;
1407
	ret = raid56_parity_recover(fs_info, bio, page->recover->bbio,
1408
				    page->recover->map_length,
1409
				    mirror_num, 0);
1410 1411 1412
	if (ret)
		return ret;

1413 1414
	wait_for_completion_io(&done);
	return blk_status_to_errno(bio->bi_status);
1415 1416
}

L
Liu Bo 已提交
1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455
static void scrub_recheck_block_on_raid56(struct btrfs_fs_info *fs_info,
					  struct scrub_block *sblock)
{
	struct scrub_page *first_page = sblock->pagev[0];
	struct bio *bio;
	int page_num;

	/* All pages in sblock belong to the same stripe on the same device. */
	ASSERT(first_page->dev);
	if (!first_page->dev->bdev)
		goto out;

	bio = btrfs_io_bio_alloc(BIO_MAX_PAGES);
	bio_set_dev(bio, first_page->dev->bdev);

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

		WARN_ON(!page->page);
		bio_add_page(bio, page->page, PAGE_SIZE, 0);
	}

	if (scrub_submit_raid56_bio_wait(fs_info, bio, first_page)) {
		bio_put(bio);
		goto out;
	}

	bio_put(bio);

	scrub_recheck_block_checksum(sblock);

	return;
out:
	for (page_num = 0; page_num < sblock->page_count; page_num++)
		sblock->pagev[page_num]->io_error = 1;

	sblock->no_io_error_seen = 0;
}

1456 1457 1458 1459 1460 1461 1462
/*
 * 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.
 */
1463
static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
1464 1465
				struct scrub_block *sblock,
				int retry_failed_mirror)
I
Ilya Dryomov 已提交
1466
{
1467
	int page_num;
I
Ilya Dryomov 已提交
1468

1469
	sblock->no_io_error_seen = 1;
I
Ilya Dryomov 已提交
1470

L
Liu Bo 已提交
1471 1472 1473 1474
	/* short cut for raid56 */
	if (!retry_failed_mirror && scrub_is_page_on_raid56(sblock->pagev[0]))
		return scrub_recheck_block_on_raid56(fs_info, sblock);

1475 1476
	for (page_num = 0; page_num < sblock->page_count; page_num++) {
		struct bio *bio;
1477
		struct scrub_page *page = sblock->pagev[page_num];
1478

1479
		if (page->dev->bdev == NULL) {
1480 1481 1482 1483 1484
			page->io_error = 1;
			sblock->no_io_error_seen = 0;
			continue;
		}

1485
		WARN_ON(!page->page);
1486
		bio = btrfs_io_bio_alloc(1);
1487
		bio_set_dev(bio, page->dev->bdev);
1488

1489
		bio_add_page(bio, page->page, PAGE_SIZE, 0);
L
Liu Bo 已提交
1490 1491
		bio->bi_iter.bi_sector = page->physical >> 9;
		bio->bi_opf = REQ_OP_READ;
1492

L
Liu Bo 已提交
1493 1494 1495
		if (btrfsic_submit_bio_wait(bio)) {
			page->io_error = 1;
			sblock->no_io_error_seen = 0;
1496
		}
1497

1498 1499
		bio_put(bio);
	}
I
Ilya Dryomov 已提交
1500

1501
	if (sblock->no_io_error_seen)
1502
		scrub_recheck_block_checksum(sblock);
A
Arne Jansen 已提交
1503 1504
}

M
Miao Xie 已提交
1505 1506 1507 1508 1509 1510
static inline int scrub_check_fsid(u8 fsid[],
				   struct scrub_page *spage)
{
	struct btrfs_fs_devices *fs_devices = spage->dev->fs_devices;
	int ret;

1511
	ret = memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
M
Miao Xie 已提交
1512 1513 1514
	return !ret;
}

1515
static void scrub_recheck_block_checksum(struct scrub_block *sblock)
A
Arne Jansen 已提交
1516
{
1517 1518 1519
	sblock->header_error = 0;
	sblock->checksum_error = 0;
	sblock->generation_error = 0;
1520

1521 1522 1523 1524
	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
1525 1526
}

1527
static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
1528
					     struct scrub_block *sblock_good)
1529 1530 1531
{
	int page_num;
	int ret = 0;
I
Ilya Dryomov 已提交
1532

1533 1534
	for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
		int ret_sub;
I
Ilya Dryomov 已提交
1535

1536 1537
		ret_sub = scrub_repair_page_from_good_copy(sblock_bad,
							   sblock_good,
1538
							   page_num, 1);
1539 1540
		if (ret_sub)
			ret = ret_sub;
A
Arne Jansen 已提交
1541
	}
1542 1543 1544 1545 1546 1547 1548 1549

	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)
{
1550 1551
	struct scrub_page *page_bad = sblock_bad->pagev[page_num];
	struct scrub_page *page_good = sblock_good->pagev[page_num];
1552
	struct btrfs_fs_info *fs_info = sblock_bad->sctx->fs_info;
1553

1554 1555
	BUG_ON(page_bad->page == NULL);
	BUG_ON(page_good->page == NULL);
1556 1557 1558 1559 1560
	if (force_write || sblock_bad->header_error ||
	    sblock_bad->checksum_error || page_bad->io_error) {
		struct bio *bio;
		int ret;

1561
		if (!page_bad->dev->bdev) {
1562
			btrfs_warn_rl(fs_info,
J
Jeff Mahoney 已提交
1563
				"scrub_repair_page_from_good_copy(bdev == NULL) is unexpected");
1564 1565 1566
			return -EIO;
		}

1567
		bio = btrfs_io_bio_alloc(1);
1568
		bio_set_dev(bio, page_bad->dev->bdev);
1569
		bio->bi_iter.bi_sector = page_bad->physical >> 9;
D
David Sterba 已提交
1570
		bio->bi_opf = REQ_OP_WRITE;
1571 1572 1573 1574 1575

		ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0);
		if (PAGE_SIZE != ret) {
			bio_put(bio);
			return -EIO;
1576
		}
1577

1578
		if (btrfsic_submit_bio_wait(bio)) {
1579 1580
			btrfs_dev_stat_inc_and_print(page_bad->dev,
				BTRFS_DEV_STAT_WRITE_ERRS);
1581
			atomic64_inc(&fs_info->dev_replace.num_write_errors);
1582 1583 1584
			bio_put(bio);
			return -EIO;
		}
1585
		bio_put(bio);
A
Arne Jansen 已提交
1586 1587
	}

1588 1589 1590
	return 0;
}

1591 1592
static void scrub_write_block_to_dev_replace(struct scrub_block *sblock)
{
1593
	struct btrfs_fs_info *fs_info = sblock->sctx->fs_info;
1594 1595
	int page_num;

1596 1597 1598 1599 1600 1601 1602
	/*
	 * 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;

1603 1604 1605 1606 1607
	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)
1608
			atomic64_inc(&fs_info->dev_replace.num_write_errors);
1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620
	}
}

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);

1621
		clear_page(mapped_buffer);
1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633
		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;

1634
	mutex_lock(&sctx->wr_lock);
1635
again:
1636 1637
	if (!sctx->wr_curr_bio) {
		sctx->wr_curr_bio = kzalloc(sizeof(*sctx->wr_curr_bio),
1638
					      GFP_KERNEL);
1639 1640
		if (!sctx->wr_curr_bio) {
			mutex_unlock(&sctx->wr_lock);
1641 1642
			return -ENOMEM;
		}
1643 1644
		sctx->wr_curr_bio->sctx = sctx;
		sctx->wr_curr_bio->page_count = 0;
1645
	}
1646
	sbio = sctx->wr_curr_bio;
1647 1648 1649 1650 1651
	if (sbio->page_count == 0) {
		struct bio *bio;

		sbio->physical = spage->physical_for_dev_replace;
		sbio->logical = spage->logical;
1652
		sbio->dev = sctx->wr_tgtdev;
1653 1654
		bio = sbio->bio;
		if (!bio) {
1655
			bio = btrfs_io_bio_alloc(sctx->pages_per_wr_bio);
1656 1657 1658 1659 1660
			sbio->bio = bio;
		}

		bio->bi_private = sbio;
		bio->bi_end_io = scrub_wr_bio_end_io;
1661
		bio_set_dev(bio, sbio->dev->bdev);
1662
		bio->bi_iter.bi_sector = sbio->physical >> 9;
D
David Sterba 已提交
1663
		bio->bi_opf = REQ_OP_WRITE;
1664
		sbio->status = 0;
1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677
	} 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;
1678
			mutex_unlock(&sctx->wr_lock);
1679 1680 1681 1682 1683 1684 1685 1686 1687
			return -EIO;
		}
		scrub_wr_submit(sctx);
		goto again;
	}

	sbio->pagev[sbio->page_count] = spage;
	scrub_page_get(spage);
	sbio->page_count++;
1688
	if (sbio->page_count == sctx->pages_per_wr_bio)
1689
		scrub_wr_submit(sctx);
1690
	mutex_unlock(&sctx->wr_lock);
1691 1692 1693 1694 1695 1696 1697 1698

	return 0;
}

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

1699
	if (!sctx->wr_curr_bio)
1700 1701
		return;

1702 1703
	sbio = sctx->wr_curr_bio;
	sctx->wr_curr_bio = NULL;
1704
	WARN_ON(!sbio->bio->bi_disk);
1705 1706 1707 1708 1709
	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 */
1710
	btrfsic_submit_bio(sbio->bio);
1711 1712
}

1713
static void scrub_wr_bio_end_io(struct bio *bio)
1714 1715
{
	struct scrub_bio *sbio = bio->bi_private;
1716
	struct btrfs_fs_info *fs_info = sbio->dev->fs_info;
1717

1718
	sbio->status = bio->bi_status;
1719 1720
	sbio->bio = bio;

1721 1722
	btrfs_init_work(&sbio->work, btrfs_scrubwrc_helper,
			 scrub_wr_bio_end_io_worker, NULL, NULL);
1723
	btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work);
1724 1725 1726 1727 1728 1729 1730 1731 1732
}

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);
1733
	if (sbio->status) {
1734
		struct btrfs_dev_replace *dev_replace =
1735
			&sbio->sctx->fs_info->dev_replace;
1736 1737 1738 1739 1740

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

			spage->io_error = 1;
1741
			atomic64_inc(&dev_replace->num_write_errors);
1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753
		}
	}

	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)
1754 1755 1756 1757
{
	u64 flags;
	int ret;

1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769
	/*
	 * 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;

1770 1771
	WARN_ON(sblock->page_count < 1);
	flags = sblock->pagev[0]->flags;
1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782
	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);
1783 1784

	return ret;
A
Arne Jansen 已提交
1785 1786
}

1787
static int scrub_checksum_data(struct scrub_block *sblock)
A
Arne Jansen 已提交
1788
{
1789
	struct scrub_ctx *sctx = sblock->sctx;
A
Arne Jansen 已提交
1790
	u8 csum[BTRFS_CSUM_SIZE];
1791 1792 1793
	u8 *on_disk_csum;
	struct page *page;
	void *buffer;
A
Arne Jansen 已提交
1794
	u32 crc = ~(u32)0;
1795 1796
	u64 len;
	int index;
A
Arne Jansen 已提交
1797

1798
	BUG_ON(sblock->page_count < 1);
1799
	if (!sblock->pagev[0]->have_csum)
A
Arne Jansen 已提交
1800 1801
		return 0;

1802 1803
	on_disk_csum = sblock->pagev[0]->csum;
	page = sblock->pagev[0]->page;
1804
	buffer = kmap_atomic(page);
1805

1806
	len = sctx->fs_info->sectorsize;
1807 1808 1809 1810
	index = 0;
	for (;;) {
		u64 l = min_t(u64, len, PAGE_SIZE);

1811
		crc = btrfs_csum_data(buffer, crc, l);
1812
		kunmap_atomic(buffer);
1813 1814 1815 1816 1817
		len -= l;
		if (len == 0)
			break;
		index++;
		BUG_ON(index >= sblock->page_count);
1818 1819
		BUG_ON(!sblock->pagev[index]->page);
		page = sblock->pagev[index]->page;
1820
		buffer = kmap_atomic(page);
1821 1822
	}

A
Arne Jansen 已提交
1823
	btrfs_csum_final(crc, csum);
1824
	if (memcmp(csum, on_disk_csum, sctx->csum_size))
1825
		sblock->checksum_error = 1;
A
Arne Jansen 已提交
1826

1827
	return sblock->checksum_error;
A
Arne Jansen 已提交
1828 1829
}

1830
static int scrub_checksum_tree_block(struct scrub_block *sblock)
A
Arne Jansen 已提交
1831
{
1832
	struct scrub_ctx *sctx = sblock->sctx;
A
Arne Jansen 已提交
1833
	struct btrfs_header *h;
1834
	struct btrfs_fs_info *fs_info = sctx->fs_info;
1835 1836 1837 1838 1839 1840
	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
1841
	u32 crc = ~(u32)0;
1842 1843 1844 1845
	u64 len;
	int index;

	BUG_ON(sblock->page_count < 1);
1846
	page = sblock->pagev[0]->page;
1847
	mapped_buffer = kmap_atomic(page);
1848
	h = (struct btrfs_header *)mapped_buffer;
1849
	memcpy(on_disk_csum, h->csum, sctx->csum_size);
A
Arne Jansen 已提交
1850 1851 1852 1853 1854 1855

	/*
	 * we don't use the getter functions here, as we
	 * a) don't have an extent buffer and
	 * b) the page is already kmapped
	 */
1856
	if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h))
1857
		sblock->header_error = 1;
A
Arne Jansen 已提交
1858

1859 1860 1861 1862
	if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h)) {
		sblock->header_error = 1;
		sblock->generation_error = 1;
	}
A
Arne Jansen 已提交
1863

M
Miao Xie 已提交
1864
	if (!scrub_check_fsid(h->fsid, sblock->pagev[0]))
1865
		sblock->header_error = 1;
A
Arne Jansen 已提交
1866 1867 1868

	if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
		   BTRFS_UUID_SIZE))
1869
		sblock->header_error = 1;
A
Arne Jansen 已提交
1870

1871
	len = sctx->fs_info->nodesize - BTRFS_CSUM_SIZE;
1872 1873 1874 1875 1876 1877
	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);

1878
		crc = btrfs_csum_data(p, crc, l);
1879
		kunmap_atomic(mapped_buffer);
1880 1881 1882 1883 1884
		len -= l;
		if (len == 0)
			break;
		index++;
		BUG_ON(index >= sblock->page_count);
1885 1886
		BUG_ON(!sblock->pagev[index]->page);
		page = sblock->pagev[index]->page;
1887
		mapped_buffer = kmap_atomic(page);
1888 1889 1890 1891 1892
		mapped_size = PAGE_SIZE;
		p = mapped_buffer;
	}

	btrfs_csum_final(crc, calculated_csum);
1893
	if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
1894
		sblock->checksum_error = 1;
A
Arne Jansen 已提交
1895

1896
	return sblock->header_error || sblock->checksum_error;
A
Arne Jansen 已提交
1897 1898
}

1899
static int scrub_checksum_super(struct scrub_block *sblock)
A
Arne Jansen 已提交
1900 1901
{
	struct btrfs_super_block *s;
1902
	struct scrub_ctx *sctx = sblock->sctx;
1903 1904 1905 1906 1907 1908
	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
1909
	u32 crc = ~(u32)0;
1910 1911
	int fail_gen = 0;
	int fail_cor = 0;
1912 1913
	u64 len;
	int index;
A
Arne Jansen 已提交
1914

1915
	BUG_ON(sblock->page_count < 1);
1916
	page = sblock->pagev[0]->page;
1917
	mapped_buffer = kmap_atomic(page);
1918
	s = (struct btrfs_super_block *)mapped_buffer;
1919
	memcpy(on_disk_csum, s->csum, sctx->csum_size);
A
Arne Jansen 已提交
1920

1921
	if (sblock->pagev[0]->logical != btrfs_super_bytenr(s))
1922
		++fail_cor;
A
Arne Jansen 已提交
1923

1924
	if (sblock->pagev[0]->generation != btrfs_super_generation(s))
1925
		++fail_gen;
A
Arne Jansen 已提交
1926

M
Miao Xie 已提交
1927
	if (!scrub_check_fsid(s->fsid, sblock->pagev[0]))
1928
		++fail_cor;
A
Arne Jansen 已提交
1929

1930 1931 1932 1933 1934 1935 1936
	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);

1937
		crc = btrfs_csum_data(p, crc, l);
1938
		kunmap_atomic(mapped_buffer);
1939 1940 1941 1942 1943
		len -= l;
		if (len == 0)
			break;
		index++;
		BUG_ON(index >= sblock->page_count);
1944 1945
		BUG_ON(!sblock->pagev[index]->page);
		page = sblock->pagev[index]->page;
1946
		mapped_buffer = kmap_atomic(page);
1947 1948 1949 1950 1951
		mapped_size = PAGE_SIZE;
		p = mapped_buffer;
	}

	btrfs_csum_final(crc, calculated_csum);
1952
	if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
1953
		++fail_cor;
A
Arne Jansen 已提交
1954

1955
	if (fail_cor + fail_gen) {
A
Arne Jansen 已提交
1956 1957 1958 1959 1960
		/*
		 * if we find an error in a super block, we just report it.
		 * They will get written with the next transaction commit
		 * anyway
		 */
1961 1962 1963
		spin_lock(&sctx->stat_lock);
		++sctx->stat.super_errors;
		spin_unlock(&sctx->stat_lock);
1964
		if (fail_cor)
1965
			btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
1966 1967
				BTRFS_DEV_STAT_CORRUPTION_ERRS);
		else
1968
			btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
1969
				BTRFS_DEV_STAT_GENERATION_ERRS);
A
Arne Jansen 已提交
1970 1971
	}

1972
	return fail_cor + fail_gen;
A
Arne Jansen 已提交
1973 1974
}

1975 1976
static void scrub_block_get(struct scrub_block *sblock)
{
1977
	refcount_inc(&sblock->refs);
1978 1979 1980 1981
}

static void scrub_block_put(struct scrub_block *sblock)
{
1982
	if (refcount_dec_and_test(&sblock->refs)) {
1983 1984
		int i;

1985 1986 1987
		if (sblock->sparity)
			scrub_parity_put(sblock->sparity);

1988
		for (i = 0; i < sblock->page_count; i++)
1989
			scrub_page_put(sblock->pagev[i]);
1990 1991 1992 1993
		kfree(sblock);
	}
}

1994 1995
static void scrub_page_get(struct scrub_page *spage)
{
1996
	atomic_inc(&spage->refs);
1997 1998 1999 2000
}

static void scrub_page_put(struct scrub_page *spage)
{
2001
	if (atomic_dec_and_test(&spage->refs)) {
2002 2003 2004 2005 2006 2007
		if (spage->page)
			__free_page(spage->page);
		kfree(spage);
	}
}

2008
static void scrub_submit(struct scrub_ctx *sctx)
A
Arne Jansen 已提交
2009 2010 2011
{
	struct scrub_bio *sbio;

2012
	if (sctx->curr == -1)
S
Stefan Behrens 已提交
2013
		return;
A
Arne Jansen 已提交
2014

2015 2016
	sbio = sctx->bios[sctx->curr];
	sctx->curr = -1;
2017
	scrub_pending_bio_inc(sctx);
2018
	btrfsic_submit_bio(sbio->bio);
A
Arne Jansen 已提交
2019 2020
}

2021 2022
static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
				    struct scrub_page *spage)
A
Arne Jansen 已提交
2023
{
2024
	struct scrub_block *sblock = spage->sblock;
A
Arne Jansen 已提交
2025
	struct scrub_bio *sbio;
2026
	int ret;
A
Arne Jansen 已提交
2027 2028 2029 2030 2031

again:
	/*
	 * grab a fresh bio or wait for one to become available
	 */
2032 2033 2034 2035 2036 2037 2038 2039
	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
2040
		} else {
2041 2042
			spin_unlock(&sctx->list_lock);
			wait_event(sctx->list_wait, sctx->first_free != -1);
A
Arne Jansen 已提交
2043 2044
		}
	}
2045
	sbio = sctx->bios[sctx->curr];
2046
	if (sbio->page_count == 0) {
2047 2048
		struct bio *bio;

2049 2050
		sbio->physical = spage->physical;
		sbio->logical = spage->logical;
2051
		sbio->dev = spage->dev;
2052 2053
		bio = sbio->bio;
		if (!bio) {
2054
			bio = btrfs_io_bio_alloc(sctx->pages_per_rd_bio);
2055 2056
			sbio->bio = bio;
		}
2057 2058 2059

		bio->bi_private = sbio;
		bio->bi_end_io = scrub_bio_end_io;
2060
		bio_set_dev(bio, sbio->dev->bdev);
2061
		bio->bi_iter.bi_sector = sbio->physical >> 9;
D
David Sterba 已提交
2062
		bio->bi_opf = REQ_OP_READ;
2063
		sbio->status = 0;
2064 2065 2066
	} else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
		   spage->physical ||
		   sbio->logical + sbio->page_count * PAGE_SIZE !=
2067 2068
		   spage->logical ||
		   sbio->dev != spage->dev) {
2069
		scrub_submit(sctx);
A
Arne Jansen 已提交
2070 2071
		goto again;
	}
2072

2073 2074 2075 2076 2077 2078 2079 2080
	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;
		}
2081
		scrub_submit(sctx);
2082 2083 2084
		goto again;
	}

2085
	scrub_block_get(sblock); /* one for the page added to the bio */
2086 2087
	atomic_inc(&sblock->outstanding_pages);
	sbio->page_count++;
2088
	if (sbio->page_count == sctx->pages_per_rd_bio)
2089
		scrub_submit(sctx);
2090 2091 2092 2093

	return 0;
}

2094
static void scrub_missing_raid56_end_io(struct bio *bio)
2095 2096
{
	struct scrub_block *sblock = bio->bi_private;
2097
	struct btrfs_fs_info *fs_info = sblock->sctx->fs_info;
2098

2099
	if (bio->bi_status)
2100 2101
		sblock->no_io_error_seen = 0;

2102 2103
	bio_put(bio);

2104 2105 2106 2107 2108 2109 2110
	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;
2111
	struct btrfs_fs_info *fs_info = sctx->fs_info;
2112 2113 2114 2115 2116 2117
	u64 logical;
	struct btrfs_device *dev;

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

2118
	if (sblock->no_io_error_seen)
2119
		scrub_recheck_block_checksum(sblock);
2120 2121 2122 2123 2124

	if (!sblock->no_io_error_seen) {
		spin_lock(&sctx->stat_lock);
		sctx->stat.read_errors++;
		spin_unlock(&sctx->stat_lock);
2125
		btrfs_err_rl_in_rcu(fs_info,
2126
			"IO error rebuilding logical %llu for dev %s",
2127 2128 2129 2130 2131
			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);
2132
		btrfs_err_rl_in_rcu(fs_info,
2133
			"failed to rebuild valid logical %llu for dev %s",
2134 2135 2136 2137 2138 2139 2140
			logical, rcu_str_deref(dev->name));
	} else {
		scrub_write_block_to_dev_replace(sblock);
	}

	scrub_block_put(sblock);

2141
	if (sctx->is_dev_replace && sctx->flush_all_writes) {
2142
		mutex_lock(&sctx->wr_lock);
2143
		scrub_wr_submit(sctx);
2144
		mutex_unlock(&sctx->wr_lock);
2145 2146 2147 2148 2149 2150 2151 2152
	}

	scrub_pending_bio_dec(sctx);
}

static void scrub_missing_raid56_pages(struct scrub_block *sblock)
{
	struct scrub_ctx *sctx = sblock->sctx;
2153
	struct btrfs_fs_info *fs_info = sctx->fs_info;
2154 2155
	u64 length = sblock->page_count * PAGE_SIZE;
	u64 logical = sblock->pagev[0]->logical;
2156
	struct btrfs_bio *bbio = NULL;
2157 2158 2159 2160 2161
	struct bio *bio;
	struct btrfs_raid_bio *rbio;
	int ret;
	int i;

2162
	btrfs_bio_counter_inc_blocked(fs_info);
2163
	ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical,
2164
			&length, &bbio);
2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178
	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;
	}

2179
	bio = btrfs_io_bio_alloc(0);
2180 2181 2182 2183
	bio->bi_iter.bi_sector = logical >> 9;
	bio->bi_private = sblock;
	bio->bi_end_io = scrub_missing_raid56_end_io;

2184
	rbio = raid56_alloc_missing_rbio(fs_info, bio, bbio, length);
2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203
	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:
2204
	btrfs_bio_counter_dec(fs_info);
2205 2206 2207 2208 2209 2210
	btrfs_put_bbio(bbio);
	spin_lock(&sctx->stat_lock);
	sctx->stat.malloc_errors++;
	spin_unlock(&sctx->stat_lock);
}

2211
static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
2212
		       u64 physical, struct btrfs_device *dev, u64 flags,
2213 2214
		       u64 gen, int mirror_num, u8 *csum, int force,
		       u64 physical_for_dev_replace)
2215 2216 2217 2218
{
	struct scrub_block *sblock;
	int index;

2219
	sblock = kzalloc(sizeof(*sblock), GFP_KERNEL);
2220
	if (!sblock) {
2221 2222 2223
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		spin_unlock(&sctx->stat_lock);
2224
		return -ENOMEM;
A
Arne Jansen 已提交
2225
	}
2226

2227 2228
	/* one ref inside this function, plus one for each page added to
	 * a bio later on */
2229
	refcount_set(&sblock->refs, 1);
2230
	sblock->sctx = sctx;
2231 2232 2233
	sblock->no_io_error_seen = 1;

	for (index = 0; len > 0; index++) {
2234
		struct scrub_page *spage;
2235 2236
		u64 l = min_t(u64, len, PAGE_SIZE);

2237
		spage = kzalloc(sizeof(*spage), GFP_KERNEL);
2238 2239
		if (!spage) {
leave_nomem:
2240 2241 2242
			spin_lock(&sctx->stat_lock);
			sctx->stat.malloc_errors++;
			spin_unlock(&sctx->stat_lock);
2243
			scrub_block_put(sblock);
2244 2245
			return -ENOMEM;
		}
2246 2247 2248
		BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
		scrub_page_get(spage);
		sblock->pagev[index] = spage;
2249
		spage->sblock = sblock;
2250
		spage->dev = dev;
2251 2252 2253 2254
		spage->flags = flags;
		spage->generation = gen;
		spage->logical = logical;
		spage->physical = physical;
2255
		spage->physical_for_dev_replace = physical_for_dev_replace;
2256 2257 2258
		spage->mirror_num = mirror_num;
		if (csum) {
			spage->have_csum = 1;
2259
			memcpy(spage->csum, csum, sctx->csum_size);
2260 2261 2262 2263
		} else {
			spage->have_csum = 0;
		}
		sblock->page_count++;
2264
		spage->page = alloc_page(GFP_KERNEL);
2265 2266
		if (!spage->page)
			goto leave_nomem;
2267 2268 2269
		len -= l;
		logical += l;
		physical += l;
2270
		physical_for_dev_replace += l;
2271 2272
	}

2273
	WARN_ON(sblock->page_count == 0);
2274
	if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) {
2275 2276 2277 2278 2279 2280 2281 2282 2283
		/*
		 * 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;
2284

2285 2286 2287 2288 2289
			ret = scrub_add_page_to_rd_bio(sctx, spage);
			if (ret) {
				scrub_block_put(sblock);
				return ret;
			}
2290
		}
A
Arne Jansen 已提交
2291

2292 2293 2294
		if (force)
			scrub_submit(sctx);
	}
A
Arne Jansen 已提交
2295

2296 2297
	/* last one frees, either here or in bio completion for last page */
	scrub_block_put(sblock);
A
Arne Jansen 已提交
2298 2299 2300
	return 0;
}

2301
static void scrub_bio_end_io(struct bio *bio)
2302 2303
{
	struct scrub_bio *sbio = bio->bi_private;
2304
	struct btrfs_fs_info *fs_info = sbio->dev->fs_info;
2305

2306
	sbio->status = bio->bi_status;
2307 2308
	sbio->bio = bio;

2309
	btrfs_queue_work(fs_info->scrub_workers, &sbio->work);
2310 2311 2312 2313 2314
}

static void scrub_bio_end_io_worker(struct btrfs_work *work)
{
	struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
2315
	struct scrub_ctx *sctx = sbio->sctx;
2316 2317
	int i;

2318
	BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO);
2319
	if (sbio->status) {
2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339
		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;
2340 2341 2342 2343
	spin_lock(&sctx->list_lock);
	sbio->next_free = sctx->first_free;
	sctx->first_free = sbio->index;
	spin_unlock(&sctx->list_lock);
2344

2345
	if (sctx->is_dev_replace && sctx->flush_all_writes) {
2346
		mutex_lock(&sctx->wr_lock);
2347
		scrub_wr_submit(sctx);
2348
		mutex_unlock(&sctx->wr_lock);
2349 2350
	}

2351
	scrub_pending_bio_dec(sctx);
2352 2353
}

2354 2355 2356 2357
static inline void __scrub_mark_bitmap(struct scrub_parity *sparity,
				       unsigned long *bitmap,
				       u64 start, u64 len)
{
2358
	u64 offset;
2359 2360
	u64 nsectors64;
	u32 nsectors;
2361
	int sectorsize = sparity->sctx->fs_info->sectorsize;
2362 2363 2364 2365 2366 2367 2368

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

	start -= sparity->logic_start;
2369 2370
	start = div64_u64_rem(start, sparity->stripe_len, &offset);
	offset = div_u64(offset, sectorsize);
2371 2372 2373 2374
	nsectors64 = div_u64(len, sectorsize);

	ASSERT(nsectors64 < UINT_MAX);
	nsectors = (u32)nsectors64;
2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396

	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);
}

2397 2398
static void scrub_block_complete(struct scrub_block *sblock)
{
2399 2400
	int corrupted = 0;

2401
	if (!sblock->no_io_error_seen) {
2402
		corrupted = 1;
2403
		scrub_handle_errored_block(sblock);
2404 2405 2406 2407 2408 2409
	} else {
		/*
		 * if has checksum error, write via repair mechanism in
		 * dev replace case, otherwise write here in dev replace
		 * case.
		 */
2410 2411
		corrupted = scrub_checksum(sblock);
		if (!corrupted && sblock->sctx->is_dev_replace)
2412 2413
			scrub_write_block_to_dev_replace(sblock);
	}
2414 2415 2416 2417 2418 2419 2420 2421 2422

	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);
	}
2423 2424
}

2425
static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u8 *csum)
A
Arne Jansen 已提交
2426 2427
{
	struct btrfs_ordered_sum *sum = NULL;
2428
	unsigned long index;
A
Arne Jansen 已提交
2429 2430
	unsigned long num_sectors;

2431 2432
	while (!list_empty(&sctx->csum_list)) {
		sum = list_first_entry(&sctx->csum_list,
A
Arne Jansen 已提交
2433 2434 2435 2436 2437 2438
				       struct btrfs_ordered_sum, list);
		if (sum->bytenr > logical)
			return 0;
		if (sum->bytenr + sum->len > logical)
			break;

2439
		++sctx->stat.csum_discards;
A
Arne Jansen 已提交
2440 2441 2442 2443 2444 2445 2446
		list_del(&sum->list);
		kfree(sum);
		sum = NULL;
	}
	if (!sum)
		return 0;

2447 2448 2449
	index = div_u64(logical - sum->bytenr, sctx->fs_info->sectorsize);
	ASSERT(index < UINT_MAX);

2450
	num_sectors = sum->len / sctx->fs_info->sectorsize;
2451 2452
	memcpy(csum, sum->sums + index, sctx->csum_size);
	if (index == num_sectors - 1) {
A
Arne Jansen 已提交
2453 2454 2455
		list_del(&sum->list);
		kfree(sum);
	}
2456
	return 1;
A
Arne Jansen 已提交
2457 2458 2459
}

/* scrub extent tries to collect up to 64 kB for each bio */
L
Liu Bo 已提交
2460 2461
static int scrub_extent(struct scrub_ctx *sctx, struct map_lookup *map,
			u64 logical, u64 len,
2462
			u64 physical, struct btrfs_device *dev, u64 flags,
2463
			u64 gen, int mirror_num, u64 physical_for_dev_replace)
A
Arne Jansen 已提交
2464 2465 2466
{
	int ret;
	u8 csum[BTRFS_CSUM_SIZE];
2467 2468 2469
	u32 blocksize;

	if (flags & BTRFS_EXTENT_FLAG_DATA) {
L
Liu Bo 已提交
2470 2471 2472 2473
		if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
			blocksize = map->stripe_len;
		else
			blocksize = sctx->fs_info->sectorsize;
2474 2475 2476 2477
		spin_lock(&sctx->stat_lock);
		sctx->stat.data_extents_scrubbed++;
		sctx->stat.data_bytes_scrubbed += len;
		spin_unlock(&sctx->stat_lock);
2478
	} else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
L
Liu Bo 已提交
2479 2480 2481 2482
		if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
			blocksize = map->stripe_len;
		else
			blocksize = sctx->fs_info->nodesize;
2483 2484 2485 2486
		spin_lock(&sctx->stat_lock);
		sctx->stat.tree_extents_scrubbed++;
		sctx->stat.tree_bytes_scrubbed += len;
		spin_unlock(&sctx->stat_lock);
2487
	} else {
2488
		blocksize = sctx->fs_info->sectorsize;
2489
		WARN_ON(1);
2490
	}
A
Arne Jansen 已提交
2491 2492

	while (len) {
2493
		u64 l = min_t(u64, len, blocksize);
A
Arne Jansen 已提交
2494 2495 2496 2497
		int have_csum = 0;

		if (flags & BTRFS_EXTENT_FLAG_DATA) {
			/* push csums to sbio */
2498
			have_csum = scrub_find_csum(sctx, logical, csum);
A
Arne Jansen 已提交
2499
			if (have_csum == 0)
2500
				++sctx->stat.no_csum;
A
Arne Jansen 已提交
2501
		}
2502
		ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen,
2503 2504
				  mirror_num, have_csum ? csum : NULL, 0,
				  physical_for_dev_replace);
A
Arne Jansen 已提交
2505 2506 2507 2508 2509
		if (ret)
			return ret;
		len -= l;
		logical += l;
		physical += l;
2510
		physical_for_dev_replace += l;
A
Arne Jansen 已提交
2511 2512 2513 2514
	}
	return 0;
}

2515 2516 2517 2518 2519 2520 2521 2522 2523
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;

2524
	sblock = kzalloc(sizeof(*sblock), GFP_KERNEL);
2525 2526 2527 2528 2529 2530 2531 2532 2533
	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 */
2534
	refcount_set(&sblock->refs, 1);
2535 2536 2537 2538 2539 2540 2541 2542 2543
	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);

2544
		spage = kzalloc(sizeof(*spage), GFP_KERNEL);
2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573
		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++;
2574
		spage->page = alloc_page(GFP_KERNEL);
2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608
		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;

2609
	if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) {
2610 2611 2612 2613
		scrub_parity_mark_sectors_error(sparity, logical, len);
		return 0;
	}

2614
	if (flags & BTRFS_EXTENT_FLAG_DATA) {
L
Liu Bo 已提交
2615
		blocksize = sparity->stripe_len;
2616
	} else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
L
Liu Bo 已提交
2617
		blocksize = sparity->stripe_len;
2618
	} else {
2619
		blocksize = sctx->fs_info->sectorsize;
2620 2621 2622 2623 2624 2625 2626 2627 2628
		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 */
2629
			have_csum = scrub_find_csum(sctx, logical, csum);
2630 2631 2632 2633 2634 2635 2636 2637
			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;
2638
skip:
2639 2640 2641 2642 2643 2644 2645
		len -= l;
		logical += l;
		physical += l;
	}
	return 0;
}

2646 2647 2648 2649 2650 2651 2652 2653
/*
 * 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,
2654 2655
				   struct map_lookup *map, u64 *offset,
				   u64 *stripe_start)
2656 2657 2658 2659 2660
{
	int i;
	int j = 0;
	u64 stripe_nr;
	u64 last_offset;
2661 2662
	u32 stripe_index;
	u32 rot;
2663 2664 2665

	last_offset = (physical - map->stripes[num].physical) *
		      nr_data_stripes(map);
2666 2667 2668
	if (stripe_start)
		*stripe_start = last_offset;

2669 2670 2671 2672
	*offset = last_offset;
	for (i = 0; i < nr_data_stripes(map); i++) {
		*offset = last_offset + i * map->stripe_len;

2673
		stripe_nr = div64_u64(*offset, map->stripe_len);
2674
		stripe_nr = div_u64(stripe_nr, nr_data_stripes(map));
2675 2676

		/* Work out the disk rotation on this stripe-set */
2677
		stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, &rot);
2678 2679
		/* calculate which stripe this data locates */
		rot += i;
2680
		stripe_index = rot % map->num_stripes;
2681 2682 2683 2684 2685 2686 2687 2688 2689
		if (stripe_index == num)
			return 0;
		if (stripe_index < num)
			j++;
	}
	*offset = last_offset + j * map->stripe_len;
	return 1;
}

2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711
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);
}

2712 2713 2714 2715 2716 2717 2718 2719 2720 2721
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);
}

2722
static void scrub_parity_bio_endio(struct bio *bio)
2723 2724
{
	struct scrub_parity *sparity = (struct scrub_parity *)bio->bi_private;
2725
	struct btrfs_fs_info *fs_info = sparity->sctx->fs_info;
2726

2727
	if (bio->bi_status)
2728 2729 2730 2731
		bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap,
			  sparity->nsectors);

	bio_put(bio);
2732 2733 2734

	btrfs_init_work(&sparity->work, btrfs_scrubparity_helper,
			scrub_parity_bio_endio_worker, NULL, NULL);
2735
	btrfs_queue_work(fs_info->scrub_parity_workers, &sparity->work);
2736 2737 2738 2739 2740
}

static void scrub_parity_check_and_repair(struct scrub_parity *sparity)
{
	struct scrub_ctx *sctx = sparity->sctx;
2741
	struct btrfs_fs_info *fs_info = sctx->fs_info;
2742 2743 2744 2745 2746 2747 2748 2749 2750 2751
	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;

2752
	length = sparity->logic_end - sparity->logic_start;
2753 2754

	btrfs_bio_counter_inc_blocked(fs_info);
2755
	ret = btrfs_map_sblock(fs_info, BTRFS_MAP_WRITE, sparity->logic_start,
2756
			       &length, &bbio);
2757
	if (ret || !bbio || !bbio->raid_map)
2758 2759
		goto bbio_out;

2760
	bio = btrfs_io_bio_alloc(0);
2761 2762 2763 2764
	bio->bi_iter.bi_sector = sparity->logic_start >> 9;
	bio->bi_private = sparity;
	bio->bi_end_io = scrub_parity_bio_endio;

2765
	rbio = raid56_parity_alloc_scrub_rbio(fs_info, bio, bbio,
2766
					      length, sparity->scrub_dev,
2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778
					      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:
2779
	btrfs_bio_counter_dec(fs_info);
2780
	btrfs_put_bbio(bbio);
2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791
	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)
{
2792
	return DIV_ROUND_UP(nsectors, BITS_PER_LONG) * sizeof(long);
2793 2794 2795 2796
}

static void scrub_parity_get(struct scrub_parity *sparity)
{
2797
	refcount_inc(&sparity->refs);
2798 2799 2800 2801
}

static void scrub_parity_put(struct scrub_parity *sparity)
{
2802
	if (!refcount_dec_and_test(&sparity->refs))
2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814
		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)
{
2815
	struct btrfs_fs_info *fs_info = sctx->fs_info;
2816 2817 2818
	struct btrfs_root *root = fs_info->extent_root;
	struct btrfs_root *csum_root = fs_info->csum_root;
	struct btrfs_extent_item *extent;
2819
	struct btrfs_bio *bbio = NULL;
2820 2821 2822 2823 2824 2825 2826 2827 2828
	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;
2829
	u64 mapped_length;
2830 2831 2832 2833 2834 2835 2836
	struct btrfs_device *extent_dev;
	struct scrub_parity *sparity;
	int nsectors;
	int bitmap_len;
	int extent_mirror_num;
	int stop_loop = 0;

2837
	nsectors = div_u64(map->stripe_len, fs_info->sectorsize);
2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853
	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;
2854
	refcount_set(&sparity->refs, 1);
2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902
	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);

2903 2904 2905 2906
			if (key.type != BTRFS_EXTENT_ITEM_KEY &&
			    key.type != BTRFS_METADATA_ITEM_KEY)
				goto next;

2907
			if (key.type == BTRFS_METADATA_ITEM_KEY)
2908
				bytes = fs_info->nodesize;
2909 2910 2911 2912 2913 2914
			else
				bytes = key.offset;

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

2915
			if (key.objectid >= logic_end) {
2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927
				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);

2928 2929 2930 2931
			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
2932 2933
				btrfs_err(fs_info,
					  "scrub: tree block %llu spanning stripes, ignored. logical=%llu",
2934
					  key.objectid, logic_start);
2935 2936 2937
				spin_lock(&sctx->stat_lock);
				sctx->stat.uncorrectable_errors++;
				spin_unlock(&sctx->stat_lock);
2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956
				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);

2957
			mapped_length = extent_len;
2958
			bbio = NULL;
2959 2960 2961
			ret = btrfs_map_block(fs_info, BTRFS_MAP_READ,
					extent_logical, &mapped_length, &bbio,
					0);
2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973
			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);
2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987

			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);
2988 2989 2990

			scrub_free_csums(sctx);

2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021
			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,
3022
						logic_end - logic_start);
3023 3024
	scrub_parity_put(sparity);
	scrub_submit(sctx);
3025
	mutex_lock(&sctx->wr_lock);
3026
	scrub_wr_submit(sctx);
3027
	mutex_unlock(&sctx->wr_lock);
3028 3029 3030 3031 3032

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

3033
static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
3034 3035
					   struct map_lookup *map,
					   struct btrfs_device *scrub_dev,
3036
					   int num, u64 base, u64 length)
A
Arne Jansen 已提交
3037
{
3038
	struct btrfs_path *path, *ppath;
3039
	struct btrfs_fs_info *fs_info = sctx->fs_info;
A
Arne Jansen 已提交
3040 3041 3042
	struct btrfs_root *root = fs_info->extent_root;
	struct btrfs_root *csum_root = fs_info->csum_root;
	struct btrfs_extent_item *extent;
3043
	struct blk_plug plug;
A
Arne Jansen 已提交
3044 3045 3046 3047 3048 3049 3050
	u64 flags;
	int ret;
	int slot;
	u64 nstripes;
	struct extent_buffer *l;
	u64 physical;
	u64 logical;
L
Liu Bo 已提交
3051
	u64 logic_end;
3052
	u64 physical_end;
A
Arne Jansen 已提交
3053
	u64 generation;
3054
	int mirror_num;
A
Arne Jansen 已提交
3055 3056
	struct reada_control *reada1;
	struct reada_control *reada2;
3057
	struct btrfs_key key;
A
Arne Jansen 已提交
3058
	struct btrfs_key key_end;
A
Arne Jansen 已提交
3059 3060
	u64 increment = map->stripe_len;
	u64 offset;
3061 3062 3063
	u64 extent_logical;
	u64 extent_physical;
	u64 extent_len;
3064 3065
	u64 stripe_logical;
	u64 stripe_end;
3066 3067
	struct btrfs_device *extent_dev;
	int extent_mirror_num;
3068
	int stop_loop = 0;
D
David Woodhouse 已提交
3069

3070
	physical = map->stripes[num].physical;
A
Arne Jansen 已提交
3071
	offset = 0;
3072
	nstripes = div64_u64(length, map->stripe_len);
A
Arne Jansen 已提交
3073 3074 3075
	if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
		offset = map->stripe_len * num;
		increment = map->stripe_len * map->num_stripes;
3076
		mirror_num = 1;
A
Arne Jansen 已提交
3077 3078 3079 3080
	} 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;
3081
		mirror_num = num % map->sub_stripes + 1;
A
Arne Jansen 已提交
3082 3083
	} else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
		increment = map->stripe_len;
3084
		mirror_num = num % map->num_stripes + 1;
A
Arne Jansen 已提交
3085 3086
	} else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
		increment = map->stripe_len;
3087
		mirror_num = num % map->num_stripes + 1;
3088
	} else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3089
		get_raid56_logic_offset(physical, num, map, &offset, NULL);
3090 3091
		increment = map->stripe_len * nr_data_stripes(map);
		mirror_num = 1;
A
Arne Jansen 已提交
3092 3093
	} else {
		increment = map->stripe_len;
3094
		mirror_num = 1;
A
Arne Jansen 已提交
3095 3096 3097 3098 3099 3100
	}

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

3101 3102
	ppath = btrfs_alloc_path();
	if (!ppath) {
3103
		btrfs_free_path(path);
3104 3105 3106
		return -ENOMEM;
	}

3107 3108 3109 3110 3111
	/*
	 * 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
3112 3113 3114
	path->search_commit_root = 1;
	path->skip_locking = 1;

3115 3116
	ppath->search_commit_root = 1;
	ppath->skip_locking = 1;
A
Arne Jansen 已提交
3117
	/*
A
Arne Jansen 已提交
3118 3119 3120
	 * 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
3121 3122
	 */
	logical = base + offset;
3123
	physical_end = physical + nstripes * map->stripe_len;
3124
	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3125
		get_raid56_logic_offset(physical_end, num,
3126
					map, &logic_end, NULL);
3127 3128 3129 3130
		logic_end += base;
	} else {
		logic_end = logical + increment * nstripes;
	}
3131
	wait_event(sctx->list_wait,
3132
		   atomic_read(&sctx->bios_in_flight) == 0);
3133
	scrub_blocked_if_needed(fs_info);
A
Arne Jansen 已提交
3134 3135

	/* FIXME it might be better to start readahead at commit root */
3136 3137 3138
	key.objectid = logical;
	key.type = BTRFS_EXTENT_ITEM_KEY;
	key.offset = (u64)0;
3139
	key_end.objectid = logic_end;
3140 3141
	key_end.type = BTRFS_METADATA_ITEM_KEY;
	key_end.offset = (u64)-1;
3142
	reada1 = btrfs_reada_add(root, &key, &key_end);
A
Arne Jansen 已提交
3143

3144 3145 3146
	key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
	key.type = BTRFS_EXTENT_CSUM_KEY;
	key.offset = logical;
A
Arne Jansen 已提交
3147 3148
	key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
	key_end.type = BTRFS_EXTENT_CSUM_KEY;
3149
	key_end.offset = logic_end;
3150
	reada2 = btrfs_reada_add(csum_root, &key, &key_end);
A
Arne Jansen 已提交
3151 3152 3153 3154 3155 3156

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

A
Arne Jansen 已提交
3157 3158 3159 3160 3161

	/*
	 * collect all data csums for the stripe to avoid seeking during
	 * the scrub. This might currently (crc32) end up to be about 1MB
	 */
3162
	blk_start_plug(&plug);
A
Arne Jansen 已提交
3163 3164 3165 3166 3167

	/*
	 * now find all extents for each stripe and scrub them
	 */
	ret = 0;
3168
	while (physical < physical_end) {
A
Arne Jansen 已提交
3169 3170 3171 3172
		/*
		 * canceled?
		 */
		if (atomic_read(&fs_info->scrub_cancel_req) ||
3173
		    atomic_read(&sctx->cancel_req)) {
A
Arne Jansen 已提交
3174 3175 3176 3177 3178 3179 3180 3181
			ret = -ECANCELED;
			goto out;
		}
		/*
		 * check to see if we have to pause
		 */
		if (atomic_read(&fs_info->scrub_pause_req)) {
			/* push queued extents */
3182
			sctx->flush_all_writes = true;
3183
			scrub_submit(sctx);
3184
			mutex_lock(&sctx->wr_lock);
3185
			scrub_wr_submit(sctx);
3186
			mutex_unlock(&sctx->wr_lock);
3187
			wait_event(sctx->list_wait,
3188
				   atomic_read(&sctx->bios_in_flight) == 0);
3189
			sctx->flush_all_writes = false;
3190
			scrub_blocked_if_needed(fs_info);
A
Arne Jansen 已提交
3191 3192
		}

3193 3194 3195 3196 3197 3198
		if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
			ret = get_raid56_logic_offset(physical, num, map,
						      &logical,
						      &stripe_logical);
			logical += base;
			if (ret) {
3199
				/* it is parity strip */
3200
				stripe_logical += base;
3201
				stripe_end = stripe_logical + increment;
3202 3203 3204 3205 3206 3207 3208 3209 3210
				ret = scrub_raid56_parity(sctx, map, scrub_dev,
							  ppath, stripe_logical,
							  stripe_end);
				if (ret)
					goto out;
				goto skip;
			}
		}

3211 3212 3213 3214
		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
3215
		key.objectid = logical;
L
Liu Bo 已提交
3216
		key.offset = (u64)-1;
A
Arne Jansen 已提交
3217 3218 3219 3220

		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
		if (ret < 0)
			goto out;
3221

3222
		if (ret > 0) {
3223
			ret = btrfs_previous_extent_item(root, path, 0);
A
Arne Jansen 已提交
3224 3225
			if (ret < 0)
				goto out;
3226 3227 3228 3229 3230 3231 3232 3233 3234
			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
3235 3236
		}

L
Liu Bo 已提交
3237
		stop_loop = 0;
A
Arne Jansen 已提交
3238
		while (1) {
3239 3240
			u64 bytes;

A
Arne Jansen 已提交
3241 3242 3243 3244 3245 3246 3247 3248 3249
			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
3250
				stop_loop = 1;
A
Arne Jansen 已提交
3251 3252 3253 3254
				break;
			}
			btrfs_item_key_to_cpu(l, &key, slot);

3255 3256 3257 3258
			if (key.type != BTRFS_EXTENT_ITEM_KEY &&
			    key.type != BTRFS_METADATA_ITEM_KEY)
				goto next;

3259
			if (key.type == BTRFS_METADATA_ITEM_KEY)
3260
				bytes = fs_info->nodesize;
3261 3262 3263 3264
			else
				bytes = key.offset;

			if (key.objectid + bytes <= logical)
A
Arne Jansen 已提交
3265 3266
				goto next;

L
Liu Bo 已提交
3267 3268 3269 3270 3271 3272
			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
3273 3274 3275 3276 3277 3278

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

3279 3280 3281 3282
			if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) &&
			    (key.objectid < logical ||
			     key.objectid + bytes >
			     logical + map->stripe_len)) {
3283
				btrfs_err(fs_info,
J
Jeff Mahoney 已提交
3284
					   "scrub: tree block %llu spanning stripes, ignored. logical=%llu",
3285
				       key.objectid, logical);
3286 3287 3288
				spin_lock(&sctx->stat_lock);
				sctx->stat.uncorrectable_errors++;
				spin_unlock(&sctx->stat_lock);
A
Arne Jansen 已提交
3289 3290 3291
				goto next;
			}

L
Liu Bo 已提交
3292 3293 3294 3295
again:
			extent_logical = key.objectid;
			extent_len = bytes;

A
Arne Jansen 已提交
3296 3297 3298
			/*
			 * trim extent to this stripe
			 */
L
Liu Bo 已提交
3299 3300 3301
			if (extent_logical < logical) {
				extent_len -= logical - extent_logical;
				extent_logical = logical;
A
Arne Jansen 已提交
3302
			}
L
Liu Bo 已提交
3303
			if (extent_logical + extent_len >
A
Arne Jansen 已提交
3304
			    logical + map->stripe_len) {
L
Liu Bo 已提交
3305 3306
				extent_len = logical + map->stripe_len -
					     extent_logical;
A
Arne Jansen 已提交
3307 3308
			}

L
Liu Bo 已提交
3309
			extent_physical = extent_logical - logical + physical;
3310 3311
			extent_dev = scrub_dev;
			extent_mirror_num = mirror_num;
3312
			if (sctx->is_dev_replace)
3313 3314 3315 3316
				scrub_remap_extent(fs_info, extent_logical,
						   extent_len, &extent_physical,
						   &extent_dev,
						   &extent_mirror_num);
L
Liu Bo 已提交
3317

3318 3319 3320 3321 3322
			ret = btrfs_lookup_csums_range(csum_root,
						       extent_logical,
						       extent_logical +
						       extent_len - 1,
						       &sctx->csum_list, 1);
L
Liu Bo 已提交
3323 3324 3325
			if (ret)
				goto out;

L
Liu Bo 已提交
3326
			ret = scrub_extent(sctx, map, extent_logical, extent_len,
3327 3328
					   extent_physical, extent_dev, flags,
					   generation, extent_mirror_num,
3329
					   extent_logical - logical + physical);
3330 3331 3332

			scrub_free_csums(sctx);

A
Arne Jansen 已提交
3333 3334 3335
			if (ret)
				goto out;

L
Liu Bo 已提交
3336 3337
			if (extent_logical + extent_len <
			    key.objectid + bytes) {
3338
				if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3339 3340 3341 3342
					/*
					 * loop until we find next data stripe
					 * or we have finished all stripes.
					 */
3343 3344 3345 3346 3347 3348 3349 3350 3351 3352
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 +
3353
								increment;
3354 3355 3356 3357 3358 3359 3360 3361
						ret = scrub_raid56_parity(sctx,
							map, scrub_dev, ppath,
							stripe_logical,
							stripe_end);
						if (ret)
							goto out;
						goto loop;
					}
3362 3363 3364 3365
				} else {
					physical += map->stripe_len;
					logical += increment;
				}
L
Liu Bo 已提交
3366 3367 3368 3369 3370
				if (logical < key.objectid + bytes) {
					cond_resched();
					goto again;
				}

3371
				if (physical >= physical_end) {
L
Liu Bo 已提交
3372 3373 3374 3375
					stop_loop = 1;
					break;
				}
			}
A
Arne Jansen 已提交
3376 3377 3378
next:
			path->slots[0]++;
		}
C
Chris Mason 已提交
3379
		btrfs_release_path(path);
3380
skip:
A
Arne Jansen 已提交
3381 3382
		logical += increment;
		physical += map->stripe_len;
3383
		spin_lock(&sctx->stat_lock);
L
Liu Bo 已提交
3384 3385 3386 3387 3388
		if (stop_loop)
			sctx->stat.last_physical = map->stripes[num].physical +
						   length;
		else
			sctx->stat.last_physical = physical;
3389
		spin_unlock(&sctx->stat_lock);
L
Liu Bo 已提交
3390 3391
		if (stop_loop)
			break;
A
Arne Jansen 已提交
3392
	}
3393
out:
A
Arne Jansen 已提交
3394
	/* push queued extents */
3395
	scrub_submit(sctx);
3396
	mutex_lock(&sctx->wr_lock);
3397
	scrub_wr_submit(sctx);
3398
	mutex_unlock(&sctx->wr_lock);
A
Arne Jansen 已提交
3399

3400
	blk_finish_plug(&plug);
A
Arne Jansen 已提交
3401
	btrfs_free_path(path);
3402
	btrfs_free_path(ppath);
A
Arne Jansen 已提交
3403 3404 3405
	return ret < 0 ? ret : 0;
}

3406
static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx,
3407 3408
					  struct btrfs_device *scrub_dev,
					  u64 chunk_offset, u64 length,
3409
					  u64 dev_offset,
3410
					  struct btrfs_block_group_cache *cache)
A
Arne Jansen 已提交
3411
{
3412 3413
	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
3414 3415 3416
	struct map_lookup *map;
	struct extent_map *em;
	int i;
3417
	int ret = 0;
A
Arne Jansen 已提交
3418 3419 3420 3421 3422

	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);

3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434
	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
3435

3436
	map = em->map_lookup;
A
Arne Jansen 已提交
3437 3438 3439 3440 3441 3442 3443
	if (em->start != chunk_offset)
		goto out;

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

	for (i = 0; i < map->num_stripes; ++i) {
3444
		if (map->stripes[i].dev->bdev == scrub_dev->bdev &&
3445
		    map->stripes[i].physical == dev_offset) {
3446
			ret = scrub_stripe(sctx, map, scrub_dev, i,
3447
					   chunk_offset, length);
A
Arne Jansen 已提交
3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458
			if (ret)
				goto out;
		}
	}
out:
	free_extent_map(em);

	return ret;
}

static noinline_for_stack
3459
int scrub_enumerate_chunks(struct scrub_ctx *sctx,
3460
			   struct btrfs_device *scrub_dev, u64 start, u64 end)
A
Arne Jansen 已提交
3461 3462 3463
{
	struct btrfs_dev_extent *dev_extent = NULL;
	struct btrfs_path *path;
3464 3465
	struct btrfs_fs_info *fs_info = sctx->fs_info;
	struct btrfs_root *root = fs_info->dev_root;
A
Arne Jansen 已提交
3466 3467
	u64 length;
	u64 chunk_offset;
3468
	int ret = 0;
3469
	int ro_set;
A
Arne Jansen 已提交
3470 3471 3472 3473 3474
	int slot;
	struct extent_buffer *l;
	struct btrfs_key key;
	struct btrfs_key found_key;
	struct btrfs_block_group_cache *cache;
3475
	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
A
Arne Jansen 已提交
3476 3477 3478 3479 3480

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

3481
	path->reada = READA_FORWARD;
A
Arne Jansen 已提交
3482 3483 3484
	path->search_commit_root = 1;
	path->skip_locking = 1;

3485
	key.objectid = scrub_dev->devid;
A
Arne Jansen 已提交
3486 3487 3488 3489 3490 3491
	key.offset = 0ull;
	key.type = BTRFS_DEV_EXTENT_KEY;

	while (1) {
		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
		if (ret < 0)
3492 3493 3494 3495 3496
			break;
		if (ret > 0) {
			if (path->slots[0] >=
			    btrfs_header_nritems(path->nodes[0])) {
				ret = btrfs_next_leaf(root, path);
3497 3498 3499 3500
				if (ret < 0)
					break;
				if (ret > 0) {
					ret = 0;
3501
					break;
3502 3503 3504
				}
			} else {
				ret = 0;
3505 3506
			}
		}
A
Arne Jansen 已提交
3507 3508 3509 3510 3511 3512

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

		btrfs_item_key_to_cpu(l, &found_key, slot);

3513
		if (found_key.objectid != scrub_dev->devid)
A
Arne Jansen 已提交
3514 3515
			break;

3516
		if (found_key.type != BTRFS_DEV_EXTENT_KEY)
A
Arne Jansen 已提交
3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527
			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);

3528 3529
		if (found_key.offset + length <= start)
			goto skip;
A
Arne Jansen 已提交
3530 3531 3532 3533 3534 3535 3536 3537

		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);
3538 3539 3540 3541 3542 3543

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

3544 3545 3546 3547 3548 3549 3550 3551 3552
		/*
		 * 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);
3553
		ret = btrfs_inc_block_group_ro(cache);
3554
		if (!ret && sctx->is_dev_replace) {
3555 3556
			/*
			 * If we are doing a device replace wait for any tasks
3557
			 * that started delalloc right before we set the block
3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574
			 * 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);
3575
			ret = btrfs_wait_ordered_roots(fs_info, U64_MAX,
3576 3577 3578 3579 3580 3581 3582 3583 3584
						       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
3585
					ret = btrfs_commit_transaction(trans);
3586 3587 3588 3589 3590 3591 3592
				if (ret) {
					scrub_pause_off(fs_info);
					btrfs_put_block_group(cache);
					break;
				}
			}
		}
3593
		scrub_pause_off(fs_info);
3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606

		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
3607
			btrfs_warn(fs_info,
3608
				   "failed setting block group ro: %d", ret);
3609 3610 3611 3612
			btrfs_put_block_group(cache);
			break;
		}

3613
		down_write(&fs_info->dev_replace.rwsem);
3614 3615 3616
		dev_replace->cursor_right = found_key.offset + length;
		dev_replace->cursor_left = found_key.offset;
		dev_replace->item_needs_writeback = 1;
3617 3618
		up_write(&dev_replace->rwsem);

3619
		ret = scrub_chunk(sctx, scrub_dev, chunk_offset, length,
3620
				  found_key.offset, cache);
3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631

		/*
		 * 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.
		 */
3632
		sctx->flush_all_writes = true;
3633
		scrub_submit(sctx);
3634
		mutex_lock(&sctx->wr_lock);
3635
		scrub_wr_submit(sctx);
3636
		mutex_unlock(&sctx->wr_lock);
3637 3638 3639

		wait_event(sctx->list_wait,
			   atomic_read(&sctx->bios_in_flight) == 0);
3640 3641

		scrub_pause_on(fs_info);
3642 3643 3644 3645 3646 3647

		/*
		 * must be called before we decrease @scrub_paused.
		 * make sure we don't block transaction commit while
		 * we are waiting pending workers finished.
		 */
3648 3649
		wait_event(sctx->list_wait,
			   atomic_read(&sctx->workers_pending) == 0);
3650
		sctx->flush_all_writes = false;
3651

3652
		scrub_pause_off(fs_info);
3653

3654
		down_write(&fs_info->dev_replace.rwsem);
3655 3656
		dev_replace->cursor_left = dev_replace->cursor_right;
		dev_replace->item_needs_writeback = 1;
3657
		up_write(&fs_info->dev_replace.rwsem);
3658

3659
		if (ro_set)
3660
			btrfs_dec_block_group_ro(cache);
3661

3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672
		/*
		 * 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);
3673
			btrfs_mark_bg_unused(cache);
3674 3675 3676 3677
		} else {
			spin_unlock(&cache->lock);
		}

A
Arne Jansen 已提交
3678 3679 3680
		btrfs_put_block_group(cache);
		if (ret)
			break;
3681
		if (sctx->is_dev_replace &&
3682
		    atomic64_read(&dev_replace->num_write_errors) > 0) {
3683 3684 3685 3686 3687 3688 3689
			ret = -EIO;
			break;
		}
		if (sctx->stat.malloc_errors > 0) {
			ret = -ENOMEM;
			break;
		}
3690
skip:
A
Arne Jansen 已提交
3691
		key.offset = found_key.offset + length;
C
Chris Mason 已提交
3692
		btrfs_release_path(path);
A
Arne Jansen 已提交
3693 3694 3695
	}

	btrfs_free_path(path);
3696

3697
	return ret;
A
Arne Jansen 已提交
3698 3699
}

3700 3701
static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx,
					   struct btrfs_device *scrub_dev)
A
Arne Jansen 已提交
3702 3703 3704 3705 3706
{
	int	i;
	u64	bytenr;
	u64	gen;
	int	ret;
3707
	struct btrfs_fs_info *fs_info = sctx->fs_info;
A
Arne Jansen 已提交
3708

3709
	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3710 3711
		return -EIO;

3712
	/* Seed devices of a new filesystem has their own generation. */
3713
	if (scrub_dev->fs_devices != fs_info->fs_devices)
3714 3715
		gen = scrub_dev->generation;
	else
3716
		gen = fs_info->last_trans_committed;
A
Arne Jansen 已提交
3717 3718 3719

	for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
		bytenr = btrfs_sb_offset(i);
3720 3721
		if (bytenr + BTRFS_SUPER_INFO_SIZE >
		    scrub_dev->commit_total_bytes)
A
Arne Jansen 已提交
3722 3723
			break;

3724
		ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr,
3725
				  scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i,
3726
				  NULL, 1, bytenr);
A
Arne Jansen 已提交
3727 3728 3729
		if (ret)
			return ret;
	}
3730
	wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
A
Arne Jansen 已提交
3731 3732 3733 3734 3735 3736 3737

	return 0;
}

/*
 * get a reference count on fs_info->scrub_workers. start worker if necessary
 */
3738 3739
static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info,
						int is_dev_replace)
A
Arne Jansen 已提交
3740
{
3741
	unsigned int flags = WQ_FREEZABLE | WQ_UNBOUND;
3742
	int max_active = fs_info->thread_pool_size;
A
Arne Jansen 已提交
3743

3744 3745
	lockdep_assert_held(&fs_info->scrub_lock);

A
Arne Jansen 已提交
3746
	if (fs_info->scrub_workers_refcnt == 0) {
3747 3748
		fs_info->scrub_workers = btrfs_alloc_workqueue(fs_info, "scrub",
				flags, is_dev_replace ? 1 : max_active, 4);
3749 3750 3751
		if (!fs_info->scrub_workers)
			goto fail_scrub_workers;

3752
		fs_info->scrub_wr_completion_workers =
3753
			btrfs_alloc_workqueue(fs_info, "scrubwrc", flags,
3754
					      max_active, 2);
3755 3756 3757
		if (!fs_info->scrub_wr_completion_workers)
			goto fail_scrub_wr_completion_workers;

3758
		fs_info->scrub_parity_workers =
3759
			btrfs_alloc_workqueue(fs_info, "scrubparity", flags,
3760
					      max_active, 2);
3761 3762
		if (!fs_info->scrub_parity_workers)
			goto fail_scrub_parity_workers;
A
Arne Jansen 已提交
3763
	}
A
Arne Jansen 已提交
3764
	++fs_info->scrub_workers_refcnt;
3765 3766 3767 3768 3769 3770 3771 3772
	return 0;

fail_scrub_parity_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
3773 3774
}

3775 3776
int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start,
		    u64 end, struct btrfs_scrub_progress *progress,
3777
		    int readonly, int is_dev_replace)
A
Arne Jansen 已提交
3778
{
3779
	struct scrub_ctx *sctx;
A
Arne Jansen 已提交
3780 3781
	int ret;
	struct btrfs_device *dev;
3782
	unsigned int nofs_flag;
3783 3784 3785
	struct btrfs_workqueue *scrub_workers = NULL;
	struct btrfs_workqueue *scrub_wr_comp = NULL;
	struct btrfs_workqueue *scrub_parity = NULL;
A
Arne Jansen 已提交
3786

3787
	if (btrfs_fs_closing(fs_info))
A
Arne Jansen 已提交
3788 3789
		return -EINVAL;

3790
	if (fs_info->nodesize > BTRFS_STRIPE_LEN) {
3791 3792 3793 3794 3795
		/*
		 * 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.
		 */
3796 3797
		btrfs_err(fs_info,
			   "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails",
3798 3799
		       fs_info->nodesize,
		       BTRFS_STRIPE_LEN);
3800 3801 3802
		return -EINVAL;
	}

3803
	if (fs_info->sectorsize != PAGE_SIZE) {
3804
		/* not supported for data w/o checksums */
3805
		btrfs_err_rl(fs_info,
J
Jeff Mahoney 已提交
3806
			   "scrub: size assumption sectorsize != PAGE_SIZE (%d != %lu) fails",
3807
		       fs_info->sectorsize, PAGE_SIZE);
A
Arne Jansen 已提交
3808 3809 3810
		return -EINVAL;
	}

3811
	if (fs_info->nodesize >
3812
	    PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK ||
3813
	    fs_info->sectorsize > PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) {
3814 3815 3816 3817
		/*
		 * would exhaust the array bounds of pagev member in
		 * struct scrub_block
		 */
J
Jeff Mahoney 已提交
3818 3819
		btrfs_err(fs_info,
			  "scrub: size assumption nodesize and sectorsize <= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails",
3820
		       fs_info->nodesize,
3821
		       SCRUB_MAX_PAGES_PER_BLOCK,
3822
		       fs_info->sectorsize,
3823 3824 3825 3826
		       SCRUB_MAX_PAGES_PER_BLOCK);
		return -EINVAL;
	}

3827 3828 3829 3830
	/* Allocate outside of device_list_mutex */
	sctx = scrub_setup_ctx(fs_info, is_dev_replace);
	if (IS_ERR(sctx))
		return PTR_ERR(sctx);
A
Arne Jansen 已提交
3831

3832
	mutex_lock(&fs_info->fs_devices->device_list_mutex);
3833
	dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
3834 3835
	if (!dev || (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) &&
		     !is_dev_replace)) {
3836
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3837 3838
		ret = -ENODEV;
		goto out_free_ctx;
A
Arne Jansen 已提交
3839 3840
	}

3841 3842
	if (!is_dev_replace && !readonly &&
	    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
3843
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3844 3845
		btrfs_err_in_rcu(fs_info, "scrub: device %s is not writable",
				rcu_str_deref(dev->name));
3846 3847
		ret = -EROFS;
		goto out_free_ctx;
3848 3849
	}

3850
	mutex_lock(&fs_info->scrub_lock);
3851
	if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3852
	    test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &dev->dev_state)) {
A
Arne Jansen 已提交
3853
		mutex_unlock(&fs_info->scrub_lock);
3854
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3855 3856
		ret = -EIO;
		goto out_free_ctx;
A
Arne Jansen 已提交
3857 3858
	}

3859
	down_read(&fs_info->dev_replace.rwsem);
3860
	if (dev->scrub_ctx ||
3861 3862
	    (!is_dev_replace &&
	     btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) {
3863
		up_read(&fs_info->dev_replace.rwsem);
A
Arne Jansen 已提交
3864
		mutex_unlock(&fs_info->scrub_lock);
3865
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3866 3867
		ret = -EINPROGRESS;
		goto out_free_ctx;
A
Arne Jansen 已提交
3868
	}
3869
	up_read(&fs_info->dev_replace.rwsem);
3870 3871 3872 3873 3874

	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);
3875
		goto out_free_ctx;
3876 3877
	}

3878
	sctx->readonly = readonly;
3879
	dev->scrub_ctx = sctx;
3880
	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
A
Arne Jansen 已提交
3881

3882 3883 3884 3885
	/*
	 * checking @scrub_pause_req here, we can avoid
	 * race between committing transaction and scrubbing.
	 */
3886
	__scrub_blocked_if_needed(fs_info);
A
Arne Jansen 已提交
3887 3888 3889
	atomic_inc(&fs_info->scrubs_running);
	mutex_unlock(&fs_info->scrub_lock);

3890 3891 3892 3893 3894 3895 3896 3897 3898 3899
	/*
	 * In order to avoid deadlock with reclaim when there is a transaction
	 * trying to pause scrub, make sure we use GFP_NOFS for all the
	 * allocations done at btrfs_scrub_pages() and scrub_pages_for_parity()
	 * invoked by our callees. The pausing request is done when the
	 * transaction commit starts, and it blocks the transaction until scrub
	 * is paused (done at specific points at scrub_stripe() or right above
	 * before incrementing fs_info->scrubs_running).
	 */
	nofs_flag = memalloc_nofs_save();
3900
	if (!is_dev_replace) {
3901
		btrfs_info(fs_info, "scrub: started on devid %llu", devid);
3902 3903 3904 3905
		/*
		 * by holding device list mutex, we can
		 * kick off writing super in log tree sync.
		 */
3906
		mutex_lock(&fs_info->fs_devices->device_list_mutex);
3907
		ret = scrub_supers(sctx, dev);
3908
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3909
	}
A
Arne Jansen 已提交
3910 3911

	if (!ret)
3912
		ret = scrub_enumerate_chunks(sctx, dev, start, end);
3913
	memalloc_nofs_restore(nofs_flag);
A
Arne Jansen 已提交
3914

3915
	wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
A
Arne Jansen 已提交
3916 3917 3918
	atomic_dec(&fs_info->scrubs_running);
	wake_up(&fs_info->scrub_pause_wait);

3919
	wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0);
3920

A
Arne Jansen 已提交
3921
	if (progress)
3922
		memcpy(progress, &sctx->stat, sizeof(*progress));
A
Arne Jansen 已提交
3923

3924 3925 3926 3927
	if (!is_dev_replace)
		btrfs_info(fs_info, "scrub: %s on devid %llu with status: %d",
			ret ? "not finished" : "finished", devid, ret);

A
Arne Jansen 已提交
3928
	mutex_lock(&fs_info->scrub_lock);
3929
	dev->scrub_ctx = NULL;
3930 3931 3932 3933 3934
	if (--fs_info->scrub_workers_refcnt == 0) {
		scrub_workers = fs_info->scrub_workers;
		scrub_wr_comp = fs_info->scrub_wr_completion_workers;
		scrub_parity = fs_info->scrub_parity_workers;
	}
A
Arne Jansen 已提交
3935 3936
	mutex_unlock(&fs_info->scrub_lock);

3937 3938 3939
	btrfs_destroy_workqueue(scrub_workers);
	btrfs_destroy_workqueue(scrub_wr_comp);
	btrfs_destroy_workqueue(scrub_parity);
3940
	scrub_put_ctx(sctx);
A
Arne Jansen 已提交
3941

3942 3943 3944 3945 3946
	return ret;

out_free_ctx:
	scrub_free_ctx(sctx);

A
Arne Jansen 已提交
3947 3948 3949
	return ret;
}

3950
void btrfs_scrub_pause(struct btrfs_fs_info *fs_info)
A
Arne Jansen 已提交
3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964
{
	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);
}

3965
void btrfs_scrub_continue(struct btrfs_fs_info *fs_info)
A
Arne Jansen 已提交
3966 3967 3968 3969 3970
{
	atomic_dec(&fs_info->scrub_pause_req);
	wake_up(&fs_info->scrub_pause_wait);
}

3971
int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info)
A
Arne Jansen 已提交
3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991
{
	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;
}

3992 3993
int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info,
			   struct btrfs_device *dev)
3994
{
3995
	struct scrub_ctx *sctx;
A
Arne Jansen 已提交
3996 3997

	mutex_lock(&fs_info->scrub_lock);
3998
	sctx = dev->scrub_ctx;
3999
	if (!sctx) {
A
Arne Jansen 已提交
4000 4001 4002
		mutex_unlock(&fs_info->scrub_lock);
		return -ENOTCONN;
	}
4003
	atomic_inc(&sctx->cancel_req);
4004
	while (dev->scrub_ctx) {
A
Arne Jansen 已提交
4005 4006
		mutex_unlock(&fs_info->scrub_lock);
		wait_event(fs_info->scrub_pause_wait,
4007
			   dev->scrub_ctx == NULL);
A
Arne Jansen 已提交
4008 4009 4010 4011 4012 4013
		mutex_lock(&fs_info->scrub_lock);
	}
	mutex_unlock(&fs_info->scrub_lock);

	return 0;
}
S
Stefan Behrens 已提交
4014

4015
int btrfs_scrub_progress(struct btrfs_fs_info *fs_info, u64 devid,
A
Arne Jansen 已提交
4016 4017 4018
			 struct btrfs_scrub_progress *progress)
{
	struct btrfs_device *dev;
4019
	struct scrub_ctx *sctx = NULL;
A
Arne Jansen 已提交
4020

4021
	mutex_lock(&fs_info->fs_devices->device_list_mutex);
4022
	dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
A
Arne Jansen 已提交
4023
	if (dev)
4024
		sctx = dev->scrub_ctx;
4025 4026
	if (sctx)
		memcpy(progress, &sctx->stat, sizeof(*progress));
4027
	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
A
Arne Jansen 已提交
4028

4029
	return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV;
A
Arne Jansen 已提交
4030
}
4031 4032 4033 4034 4035 4036 4037 4038 4039 4040 4041 4042

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;
4043
	ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, extent_logical,
4044 4045 4046
			      &mapped_length, &bbio, 0);
	if (ret || !bbio || mapped_length < extent_len ||
	    !bbio->stripes[0].dev->bdev) {
4047
		btrfs_put_bbio(bbio);
4048 4049 4050 4051 4052 4053
		return;
	}

	*extent_physical = bbio->stripes[0].physical;
	*extent_mirror_num = bbio->mirror_num;
	*extent_dev = bbio->stripes[0].dev;
4054
	btrfs_put_bbio(bbio);
4055
}