scrub.c 113.0 KB
Newer Older
A
Arne Jansen 已提交
1
/*
2
 * Copyright (C) 2011, 2012 STRATO.  All rights reserved.
A
Arne Jansen 已提交
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License v2 as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 */

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

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

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

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

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

66 67 68 69 70 71
struct scrub_recover {
	atomic_t		refs;
	struct btrfs_bio	*bbio;
	u64			map_length;
};

A
Arne Jansen 已提交
72
struct scrub_page {
73 74
	struct scrub_block	*sblock;
	struct page		*page;
75
	struct btrfs_device	*dev;
76
	struct list_head	list;
A
Arne Jansen 已提交
77 78
	u64			flags;  /* extent flags */
	u64			generation;
79 80
	u64			logical;
	u64			physical;
81
	u64			physical_for_dev_replace;
82
	atomic_t		refs;
83 84 85 86 87
	struct {
		unsigned int	mirror_num:8;
		unsigned int	have_csum:1;
		unsigned int	io_error:1;
	};
A
Arne Jansen 已提交
88
	u8			csum[BTRFS_CSUM_SIZE];
89 90

	struct scrub_recover	*recover;
A
Arne Jansen 已提交
91 92 93 94
};

struct scrub_bio {
	int			index;
95
	struct scrub_ctx	*sctx;
96
	struct btrfs_device	*dev;
A
Arne Jansen 已提交
97 98 99 100
	struct bio		*bio;
	int			err;
	u64			logical;
	u64			physical;
101 102 103 104 105
#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
106
	int			page_count;
A
Arne Jansen 已提交
107 108 109 110
	int			next_free;
	struct btrfs_work	work;
};

111
struct scrub_block {
112
	struct scrub_page	*pagev[SCRUB_MAX_PAGES_PER_BLOCK];
113 114
	int			page_count;
	atomic_t		outstanding_pages;
115
	atomic_t		refs; /* free mem on transition to zero */
116
	struct scrub_ctx	*sctx;
117
	struct scrub_parity	*sparity;
118 119 120 121
	struct {
		unsigned int	header_error:1;
		unsigned int	checksum_error:1;
		unsigned int	no_io_error_seen:1;
122
		unsigned int	generation_error:1; /* also sets header_error */
123 124 125 126

		/* The following is for the data used to check parity */
		/* It is for the data with checksum */
		unsigned int	data_corrected:1;
127
	};
128
	struct btrfs_work	work;
129 130
};

131 132 133 134 135 136 137 138 139 140 141 142 143 144
/* 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;

	int			stripe_len;

145
	atomic_t		refs;
146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163

	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];
};

164 165 166 167 168 169 170 171
struct scrub_wr_ctx {
	struct scrub_bio *wr_curr_bio;
	struct btrfs_device *tgtdev;
	int pages_per_wr_bio; /* <= SCRUB_PAGES_PER_WR_BIO */
	atomic_t flush_all_writes;
	struct mutex wr_lock;
};

172
struct scrub_ctx {
173
	struct scrub_bio	*bios[SCRUB_BIOS_PER_SCTX];
174
	struct btrfs_root	*dev_root;
A
Arne Jansen 已提交
175 176
	int			first_free;
	int			curr;
177 178
	atomic_t		bios_in_flight;
	atomic_t		workers_pending;
A
Arne Jansen 已提交
179 180 181 182 183
	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
184
	int			readonly;
185
	int			pages_per_rd_bio;
186 187
	u32			sectorsize;
	u32			nodesize;
188 189

	int			is_dev_replace;
190
	struct scrub_wr_ctx	wr_ctx;
191

A
Arne Jansen 已提交
192 193 194 195 196
	/*
	 * statistics
	 */
	struct btrfs_scrub_progress stat;
	spinlock_t		stat_lock;
197 198 199 200 201 202 203 204 205

	/*
	 * 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.
	 */
	atomic_t                refs;
A
Arne Jansen 已提交
206 207
};

208
struct scrub_fixup_nodatasum {
209
	struct scrub_ctx	*sctx;
210
	struct btrfs_device	*dev;
211 212 213 214 215 216
	u64			logical;
	struct btrfs_root	*root;
	struct btrfs_work	work;
	int			mirror_num;
};

217 218 219 220 221 222 223
struct scrub_nocow_inode {
	u64			inum;
	u64			offset;
	u64			root;
	struct list_head	list;
};

224 225 226 227 228 229
struct scrub_copy_nocow_ctx {
	struct scrub_ctx	*sctx;
	u64			logical;
	u64			len;
	int			mirror_num;
	u64			physical_for_dev_replace;
230
	struct list_head	inodes;
231 232 233
	struct btrfs_work	work;
};

234 235 236 237 238 239 240 241 242
struct scrub_warning {
	struct btrfs_path	*path;
	u64			extent_item_size;
	const char		*errstr;
	sector_t		sector;
	u64			logical;
	struct btrfs_device	*dev;
};

243 244 245 246
static void scrub_pending_bio_inc(struct scrub_ctx *sctx);
static void scrub_pending_bio_dec(struct scrub_ctx *sctx);
static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx);
static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx);
247
static int scrub_handle_errored_block(struct scrub_block *sblock_to_check);
248
static int scrub_setup_recheck_block(struct scrub_block *original_sblock,
249
				     struct scrub_block *sblocks_for_recheck);
250
static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
251 252
				struct scrub_block *sblock,
				int retry_failed_mirror);
253
static void scrub_recheck_block_checksum(struct scrub_block *sblock);
254
static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
255
					     struct scrub_block *sblock_good);
256 257 258
static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
					    struct scrub_block *sblock_good,
					    int page_num, int force_write);
259 260 261
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);
262 263 264 265 266
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);
267 268
static void scrub_page_get(struct scrub_page *spage);
static void scrub_page_put(struct scrub_page *spage);
269 270
static void scrub_parity_get(struct scrub_parity *sparity);
static void scrub_parity_put(struct scrub_parity *sparity);
271 272
static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
				    struct scrub_page *spage);
273
static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
274
		       u64 physical, struct btrfs_device *dev, u64 flags,
275 276
		       u64 gen, int mirror_num, u8 *csum, int force,
		       u64 physical_for_dev_replace);
277
static void scrub_bio_end_io(struct bio *bio);
278 279
static void scrub_bio_end_io_worker(struct btrfs_work *work);
static void scrub_block_complete(struct scrub_block *sblock);
280 281 282 283 284 285 286 287 288 289 290 291 292 293
static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
			       u64 extent_logical, u64 extent_len,
			       u64 *extent_physical,
			       struct btrfs_device **extent_dev,
			       int *extent_mirror_num);
static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
			      struct scrub_wr_ctx *wr_ctx,
			      struct btrfs_fs_info *fs_info,
			      struct btrfs_device *dev,
			      int is_dev_replace);
static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx);
static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
				    struct scrub_page *spage);
static void scrub_wr_submit(struct scrub_ctx *sctx);
294
static void scrub_wr_bio_end_io(struct bio *bio);
295 296 297 298
static void scrub_wr_bio_end_io_worker(struct btrfs_work *work);
static int write_page_nocow(struct scrub_ctx *sctx,
			    u64 physical_for_dev_replace, struct page *page);
static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
299
				      struct scrub_copy_nocow_ctx *ctx);
300 301 302
static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
			    int mirror_num, u64 physical_for_dev_replace);
static void copy_nocow_pages_worker(struct btrfs_work *work);
303
static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
304
static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
305
static void scrub_put_ctx(struct scrub_ctx *sctx);
S
Stefan Behrens 已提交
306 307


308 309
static void scrub_pending_bio_inc(struct scrub_ctx *sctx)
{
310
	atomic_inc(&sctx->refs);
311 312 313 314 315 316 317
	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);
318
	scrub_put_ctx(sctx);
319 320
}

321
static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
322 323 324 325 326 327 328 329 330
{
	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);
	}
}

331
static void scrub_pause_on(struct btrfs_fs_info *fs_info)
332 333 334
{
	atomic_inc(&fs_info->scrubs_paused);
	wake_up(&fs_info->scrub_pause_wait);
335
}
336

337 338
static void scrub_pause_off(struct btrfs_fs_info *fs_info)
{
339 340 341 342 343 344 345 346
	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);
}

347 348 349 350 351 352
static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
{
	scrub_pause_on(fs_info);
	scrub_pause_off(fs_info);
}

353 354 355 356 357 358 359 360
/*
 * used for workers that require transaction commits (i.e., for the
 * NOCOW case)
 */
static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx)
{
	struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;

361
	atomic_inc(&sctx->refs);
362 363 364 365 366 367 368 369 370 371 372 373 374
	/*
	 * increment scrubs_running to prevent cancel requests from
	 * completing as long as a worker is running. we must also
	 * increment scrubs_paused to prevent deadlocking on pause
	 * requests used for transactions commits (as the worker uses a
	 * transaction context). it is safe to regard the worker
	 * as paused for all matters practical. effectively, we only
	 * avoid cancellation requests from completing.
	 */
	mutex_lock(&fs_info->scrub_lock);
	atomic_inc(&fs_info->scrubs_running);
	atomic_inc(&fs_info->scrubs_paused);
	mutex_unlock(&fs_info->scrub_lock);
375 376 377 378 379 380 381 382 383 384

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

385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403
	atomic_inc(&sctx->workers_pending);
}

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

	/*
	 * see scrub_pending_trans_workers_inc() why we're pretending
	 * to be paused in the scrub counters
	 */
	mutex_lock(&fs_info->scrub_lock);
	atomic_dec(&fs_info->scrubs_running);
	atomic_dec(&fs_info->scrubs_paused);
	mutex_unlock(&fs_info->scrub_lock);
	atomic_dec(&sctx->workers_pending);
	wake_up(&fs_info->scrub_pause_wait);
	wake_up(&sctx->list_wait);
404
	scrub_put_ctx(sctx);
405 406
}

407
static void scrub_free_csums(struct scrub_ctx *sctx)
A
Arne Jansen 已提交
408
{
409
	while (!list_empty(&sctx->csum_list)) {
A
Arne Jansen 已提交
410
		struct btrfs_ordered_sum *sum;
411
		sum = list_first_entry(&sctx->csum_list,
A
Arne Jansen 已提交
412 413 414 415 416 417
				       struct btrfs_ordered_sum, list);
		list_del(&sum->list);
		kfree(sum);
	}
}

418
static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx)
A
Arne Jansen 已提交
419 420 421
{
	int i;

422
	if (!sctx)
A
Arne Jansen 已提交
423 424
		return;

425 426
	scrub_free_wr_ctx(&sctx->wr_ctx);

427
	/* this can happen when scrub is cancelled */
428 429
	if (sctx->curr != -1) {
		struct scrub_bio *sbio = sctx->bios[sctx->curr];
430 431

		for (i = 0; i < sbio->page_count; i++) {
432
			WARN_ON(!sbio->pagev[i]->page);
433 434 435 436 437
			scrub_block_put(sbio->pagev[i]->sblock);
		}
		bio_put(sbio->bio);
	}

438
	for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
439
		struct scrub_bio *sbio = sctx->bios[i];
A
Arne Jansen 已提交
440 441 442 443 444 445

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

446 447
	scrub_free_csums(sctx);
	kfree(sctx);
A
Arne Jansen 已提交
448 449
}

450 451 452 453 454 455
static void scrub_put_ctx(struct scrub_ctx *sctx)
{
	if (atomic_dec_and_test(&sctx->refs))
		scrub_free_ctx(sctx);
}

A
Arne Jansen 已提交
456
static noinline_for_stack
457
struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace)
A
Arne Jansen 已提交
458
{
459
	struct scrub_ctx *sctx;
A
Arne Jansen 已提交
460 461
	int		i;
	struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
462
	int ret;
A
Arne Jansen 已提交
463

464 465
	sctx = kzalloc(sizeof(*sctx), GFP_NOFS);
	if (!sctx)
A
Arne Jansen 已提交
466
		goto nomem;
467
	atomic_set(&sctx->refs, 1);
468
	sctx->is_dev_replace = is_dev_replace;
469
	sctx->pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO;
470
	sctx->curr = -1;
471
	sctx->dev_root = dev->dev_root;
472
	for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
A
Arne Jansen 已提交
473 474 475 476 477
		struct scrub_bio *sbio;

		sbio = kzalloc(sizeof(*sbio), GFP_NOFS);
		if (!sbio)
			goto nomem;
478
		sctx->bios[i] = sbio;
A
Arne Jansen 已提交
479 480

		sbio->index = i;
481
		sbio->sctx = sctx;
482
		sbio->page_count = 0;
483 484
		btrfs_init_work(&sbio->work, btrfs_scrub_helper,
				scrub_bio_end_io_worker, NULL, NULL);
A
Arne Jansen 已提交
485

486
		if (i != SCRUB_BIOS_PER_SCTX - 1)
487
			sctx->bios[i]->next_free = i + 1;
488
		else
489 490 491 492 493
			sctx->bios[i]->next_free = -1;
	}
	sctx->first_free = 0;
	sctx->nodesize = dev->dev_root->nodesize;
	sctx->sectorsize = dev->dev_root->sectorsize;
494 495
	atomic_set(&sctx->bios_in_flight, 0);
	atomic_set(&sctx->workers_pending, 0);
496 497 498 499 500 501 502
	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);
503 504 505 506 507 508 509

	ret = scrub_setup_wr_ctx(sctx, &sctx->wr_ctx, fs_info,
				 fs_info->dev_replace.tgtdev, is_dev_replace);
	if (ret) {
		scrub_free_ctx(sctx);
		return ERR_PTR(ret);
	}
510
	return sctx;
A
Arne Jansen 已提交
511 512

nomem:
513
	scrub_free_ctx(sctx);
A
Arne Jansen 已提交
514 515 516
	return ERR_PTR(-ENOMEM);
}

517 518
static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root,
				     void *warn_ctx)
519 520 521 522 523 524 525
{
	u64 isize;
	u32 nlink;
	int ret;
	int i;
	struct extent_buffer *eb;
	struct btrfs_inode_item *inode_item;
526
	struct scrub_warning *swarn = warn_ctx;
527 528 529 530
	struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info;
	struct inode_fs_paths *ipath = NULL;
	struct btrfs_root *local_root;
	struct btrfs_key root_key;
531
	struct btrfs_key key;
532 533 534 535 536 537 538 539 540 541

	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;
	}

542 543 544
	/*
	 * this makes the path point to (inum INODE_ITEM ioff)
	 */
545 546 547 548 549
	key.objectid = inum;
	key.type = BTRFS_INODE_ITEM_KEY;
	key.offset = 0;

	ret = btrfs_search_slot(NULL, local_root, &key, swarn->path, 0, 0);
550 551 552 553 554 555 556 557 558 559 560 561 562
	if (ret) {
		btrfs_release_path(swarn->path);
		goto err;
	}

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

	ipath = init_ipath(4096, local_root, swarn->path);
563 564 565 566 567
	if (IS_ERR(ipath)) {
		ret = PTR_ERR(ipath);
		ipath = NULL;
		goto err;
	}
568 569 570 571 572 573 574 575 576 577
	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)
578
		btrfs_warn_in_rcu(fs_info, "%s at logical %llu on dev "
579
			"%s, sector %llu, root %llu, inode %llu, offset %llu, "
580
			"length %llu, links %u (path: %s)", swarn->errstr,
581
			swarn->logical, rcu_str_deref(swarn->dev->name),
582 583
			(unsigned long long)swarn->sector, root, inum, offset,
			min(isize - offset, (u64)PAGE_SIZE), nlink,
584
			(char *)(unsigned long)ipath->fspath->val[i]);
585 586 587 588 589

	free_ipath(ipath);
	return 0;

err:
590
	btrfs_warn_in_rcu(fs_info, "%s at logical %llu on dev "
591
		"%s, sector %llu, root %llu, inode %llu, offset %llu: path "
592
		"resolving failed with ret=%d", swarn->errstr,
593
		swarn->logical, rcu_str_deref(swarn->dev->name),
594 595 596 597 598 599
		(unsigned long long)swarn->sector, root, inum, offset, ret);

	free_ipath(ipath);
	return 0;
}

600
static void scrub_print_warning(const char *errstr, struct scrub_block *sblock)
601
{
602 603
	struct btrfs_device *dev;
	struct btrfs_fs_info *fs_info;
604 605 606 607 608
	struct btrfs_path *path;
	struct btrfs_key found_key;
	struct extent_buffer *eb;
	struct btrfs_extent_item *ei;
	struct scrub_warning swarn;
609 610 611
	unsigned long ptr = 0;
	u64 extent_item_pos;
	u64 flags = 0;
612
	u64 ref_root;
613
	u32 item_size;
614
	u8 ref_level;
615
	int ret;
616

617
	WARN_ON(sblock->page_count < 1);
618
	dev = sblock->pagev[0]->dev;
619 620
	fs_info = sblock->sctx->dev_root->fs_info;

621
	path = btrfs_alloc_path();
622 623
	if (!path)
		return;
624

625 626
	swarn.sector = (sblock->pagev[0]->physical) >> 9;
	swarn.logical = sblock->pagev[0]->logical;
627
	swarn.errstr = errstr;
628
	swarn.dev = NULL;
629

630 631
	ret = extent_from_logical(fs_info, swarn.logical, path, &found_key,
				  &flags);
632 633 634
	if (ret < 0)
		goto out;

J
Jan Schmidt 已提交
635
	extent_item_pos = swarn.logical - found_key.objectid;
636 637 638 639 640 641
	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]);

642
	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
643
		do {
644 645 646
			ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
						      item_size, &ref_root,
						      &ref_level);
647 648
			btrfs_warn_in_rcu(fs_info,
				"%s at logical %llu on dev %s, "
649
				"sector %llu: metadata %s (level %d) in tree "
650
				"%llu", errstr, swarn.logical,
651
				rcu_str_deref(dev->name),
652 653 654 655 656
				(unsigned long long)swarn.sector,
				ref_level ? "node" : "leaf",
				ret < 0 ? -1 : ref_level,
				ret < 0 ? -1 : ref_root);
		} while (ret != 1);
657
		btrfs_release_path(path);
658
	} else {
659
		btrfs_release_path(path);
660
		swarn.path = path;
661
		swarn.dev = dev;
662 663
		iterate_extent_inodes(fs_info, found_key.objectid,
					extent_item_pos, 1,
664 665 666 667 668 669 670
					scrub_print_warning_inode, &swarn);
	}

out:
	btrfs_free_path(path);
}

671
static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx)
672
{
673
	struct page *page = NULL;
674
	unsigned long index;
675
	struct scrub_fixup_nodatasum *fixup = fixup_ctx;
676
	int ret;
677
	int corrected = 0;
678
	struct btrfs_key key;
679
	struct inode *inode = NULL;
680
	struct btrfs_fs_info *fs_info;
681 682
	u64 end = offset + PAGE_SIZE - 1;
	struct btrfs_root *local_root;
683
	int srcu_index;
684 685 686 687

	key.objectid = root;
	key.type = BTRFS_ROOT_ITEM_KEY;
	key.offset = (u64)-1;
688 689 690 691 692 693 694

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

	local_root = btrfs_read_fs_root_no_name(fs_info, &key);
	if (IS_ERR(local_root)) {
		srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
695
		return PTR_ERR(local_root);
696
	}
697 698 699 700

	key.type = BTRFS_INODE_ITEM_KEY;
	key.objectid = inum;
	key.offset = 0;
701 702
	inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
	srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
703 704 705 706 707 708
	if (IS_ERR(inode))
		return PTR_ERR(inode);

	index = offset >> PAGE_CACHE_SHIFT;

	page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734
	if (!page) {
		ret = -ENOMEM;
		goto out;
	}

	if (PageUptodate(page)) {
		if (PageDirty(page)) {
			/*
			 * we need to write the data to the defect sector. the
			 * data that was in that sector is not in memory,
			 * because the page was modified. we must not write the
			 * modified page to that sector.
			 *
			 * TODO: what could be done here: wait for the delalloc
			 *       runner to write out that page (might involve
			 *       COW) and see whether the sector is still
			 *       referenced afterwards.
			 *
			 * For the meantime, we'll treat this error
			 * incorrectable, although there is a chance that a
			 * later scrub will find the bad sector again and that
			 * there's no dirty page in memory, then.
			 */
			ret = -EIO;
			goto out;
		}
735
		ret = repair_io_failure(inode, offset, PAGE_SIZE,
736
					fixup->logical, page,
737
					offset - page_offset(page),
738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771
					fixup->mirror_num);
		unlock_page(page);
		corrected = !ret;
	} else {
		/*
		 * we need to get good data first. the general readpage path
		 * will call repair_io_failure for us, we just have to make
		 * sure we read the bad mirror.
		 */
		ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
					EXTENT_DAMAGED, GFP_NOFS);
		if (ret) {
			/* set_extent_bits should give proper error */
			WARN_ON(ret > 0);
			if (ret > 0)
				ret = -EFAULT;
			goto out;
		}

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

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

out:
	if (page)
		put_page(page);
772 773

	iput(inode);
774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792

	if (ret < 0)
		return ret;

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

	return -EIO;
}

static void scrub_fixup_nodatasum(struct btrfs_work *work)
{
	int ret;
	struct scrub_fixup_nodatasum *fixup;
793
	struct scrub_ctx *sctx;
794 795 796 797 798
	struct btrfs_trans_handle *trans = NULL;
	struct btrfs_path *path;
	int uncorrectable = 0;

	fixup = container_of(work, struct scrub_fixup_nodatasum, work);
799
	sctx = fixup->sctx;
800 801 802

	path = btrfs_alloc_path();
	if (!path) {
803 804 805
		spin_lock(&sctx->stat_lock);
		++sctx->stat.malloc_errors;
		spin_unlock(&sctx->stat_lock);
806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833
		uncorrectable = 1;
		goto out;
	}

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

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

834 835 836
	spin_lock(&sctx->stat_lock);
	++sctx->stat.corrected_errors;
	spin_unlock(&sctx->stat_lock);
837 838 839 840 841

out:
	if (trans && !IS_ERR(trans))
		btrfs_end_transaction(trans, fixup->root);
	if (uncorrectable) {
842 843 844
		spin_lock(&sctx->stat_lock);
		++sctx->stat.uncorrectable_errors;
		spin_unlock(&sctx->stat_lock);
845 846 847
		btrfs_dev_replace_stats_inc(
			&sctx->dev_root->fs_info->dev_replace.
			num_uncorrectable_read_errors);
848 849
		btrfs_err_rl_in_rcu(sctx->dev_root->fs_info,
		    "unable to fixup (nodatasum) error at logical %llu on dev %s",
850
			fixup->logical, rcu_str_deref(fixup->dev->name));
851 852 853 854 855
	}

	btrfs_free_path(path);
	kfree(fixup);

856
	scrub_pending_trans_workers_dec(sctx);
857 858
}

859 860 861 862 863 864 865 866
static inline void scrub_get_recover(struct scrub_recover *recover)
{
	atomic_inc(&recover->refs);
}

static inline void scrub_put_recover(struct scrub_recover *recover)
{
	if (atomic_dec_and_test(&recover->refs)) {
867
		btrfs_put_bbio(recover->bbio);
868 869 870 871
		kfree(recover);
	}
}

A
Arne Jansen 已提交
872
/*
873 874 875 876 877 878
 * 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
879
 */
880
static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
A
Arne Jansen 已提交
881
{
882
	struct scrub_ctx *sctx = sblock_to_check->sctx;
883
	struct btrfs_device *dev;
884 885 886 887 888 889 890 891 892 893 894 895
	struct btrfs_fs_info *fs_info;
	u64 length;
	u64 logical;
	unsigned int failed_mirror_index;
	unsigned int is_metadata;
	unsigned int have_csum;
	struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */
	struct scrub_block *sblock_bad;
	int ret;
	int mirror_index;
	int page_num;
	int success;
896
	static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
897 898 899
				      DEFAULT_RATELIMIT_BURST);

	BUG_ON(sblock_to_check->page_count < 1);
900
	fs_info = sctx->dev_root->fs_info;
901 902 903 904 905 906 907 908 909 910 911
	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;
	}
912
	length = sblock_to_check->page_count * PAGE_SIZE;
913 914 915 916
	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 &
917
			BTRFS_EXTENT_FLAG_DATA);
918 919
	have_csum = sblock_to_check->pagev[0]->have_csum;
	dev = sblock_to_check->pagev[0]->dev;
920

921 922 923 924 925
	if (sctx->is_dev_replace && !is_metadata && !have_csum) {
		sblocks_for_recheck = NULL;
		goto nodatasum_case;
	}

926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954
	/*
	 * 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.
	 */

955 956
	sblocks_for_recheck = kcalloc(BTRFS_MAX_MIRRORS,
				      sizeof(*sblocks_for_recheck), GFP_NOFS);
957
	if (!sblocks_for_recheck) {
958 959 960 961 962
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		sctx->stat.read_errors++;
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
963
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
964
		goto out;
A
Arne Jansen 已提交
965 966
	}

967
	/* setup the context, map the logical blocks and alloc the pages */
968
	ret = scrub_setup_recheck_block(sblock_to_check, sblocks_for_recheck);
969
	if (ret) {
970 971 972 973
		spin_lock(&sctx->stat_lock);
		sctx->stat.read_errors++;
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
974
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
975 976 977 978
		goto out;
	}
	BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS);
	sblock_bad = sblocks_for_recheck + failed_mirror_index;
979

980
	/* build and submit the bios for the failed mirror, check checksums */
981
	scrub_recheck_block(fs_info, sblock_bad, 1);
A
Arne Jansen 已提交
982

983 984 985 986 987 988 989 990 991 992
	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)
		 */
993 994
		spin_lock(&sctx->stat_lock);
		sctx->stat.unverified_errors++;
995
		sblock_to_check->data_corrected = 1;
996
		spin_unlock(&sctx->stat_lock);
A
Arne Jansen 已提交
997

998 999
		if (sctx->is_dev_replace)
			scrub_write_block_to_dev_replace(sblock_bad);
1000
		goto out;
A
Arne Jansen 已提交
1001 1002
	}

1003
	if (!sblock_bad->no_io_error_seen) {
1004 1005 1006
		spin_lock(&sctx->stat_lock);
		sctx->stat.read_errors++;
		spin_unlock(&sctx->stat_lock);
1007 1008
		if (__ratelimit(&_rs))
			scrub_print_warning("i/o error", sblock_to_check);
1009
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
1010
	} else if (sblock_bad->checksum_error) {
1011 1012 1013
		spin_lock(&sctx->stat_lock);
		sctx->stat.csum_errors++;
		spin_unlock(&sctx->stat_lock);
1014 1015
		if (__ratelimit(&_rs))
			scrub_print_warning("checksum error", sblock_to_check);
1016
		btrfs_dev_stat_inc_and_print(dev,
1017
					     BTRFS_DEV_STAT_CORRUPTION_ERRS);
1018
	} else if (sblock_bad->header_error) {
1019 1020 1021
		spin_lock(&sctx->stat_lock);
		sctx->stat.verify_errors++;
		spin_unlock(&sctx->stat_lock);
1022 1023 1024
		if (__ratelimit(&_rs))
			scrub_print_warning("checksum/header error",
					    sblock_to_check);
1025
		if (sblock_bad->generation_error)
1026
			btrfs_dev_stat_inc_and_print(dev,
1027 1028
				BTRFS_DEV_STAT_GENERATION_ERRS);
		else
1029
			btrfs_dev_stat_inc_and_print(dev,
1030
				BTRFS_DEV_STAT_CORRUPTION_ERRS);
1031
	}
A
Arne Jansen 已提交
1032

1033 1034 1035 1036
	if (sctx->readonly) {
		ASSERT(!sctx->is_dev_replace);
		goto out;
	}
A
Arne Jansen 已提交
1037

1038 1039
	if (!is_metadata && !have_csum) {
		struct scrub_fixup_nodatasum *fixup_nodatasum;
A
Arne Jansen 已提交
1040

1041 1042
		WARN_ON(sctx->is_dev_replace);

1043 1044
nodatasum_case:

1045 1046 1047 1048 1049 1050 1051 1052 1053 1054
		/*
		 * !is_metadata and !have_csum, this means that the data
		 * might not be COW'ed, that it might be modified
		 * concurrently. The general strategy to work on the
		 * commit root does not help in the case when COW is not
		 * used.
		 */
		fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS);
		if (!fixup_nodatasum)
			goto did_not_correct_error;
1055
		fixup_nodatasum->sctx = sctx;
1056
		fixup_nodatasum->dev = dev;
1057 1058 1059
		fixup_nodatasum->logical = logical;
		fixup_nodatasum->root = fs_info->extent_root;
		fixup_nodatasum->mirror_num = failed_mirror_index + 1;
1060
		scrub_pending_trans_workers_inc(sctx);
1061 1062
		btrfs_init_work(&fixup_nodatasum->work, btrfs_scrub_helper,
				scrub_fixup_nodatasum, NULL, NULL);
1063 1064
		btrfs_queue_work(fs_info->scrub_workers,
				 &fixup_nodatasum->work);
1065
		goto out;
A
Arne Jansen 已提交
1066 1067
	}

1068 1069
	/*
	 * now build and submit the bios for the other mirrors, check
1070 1071
	 * checksums.
	 * First try to pick the mirror which is completely without I/O
1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086
	 * errors and also does not have a checksum error.
	 * If one is found, and if a checksum is present, the full block
	 * that is known to contain an error is rewritten. Afterwards
	 * the block is known to be corrected.
	 * If a mirror is found which is completely correct, and no
	 * checksum is present, only those pages are rewritten that had
	 * an I/O error in the block to be repaired, since it cannot be
	 * determined, which copy of the other pages is better (and it
	 * could happen otherwise that a correct page would be
	 * overwritten by a bad one).
	 */
	for (mirror_index = 0;
	     mirror_index < BTRFS_MAX_MIRRORS &&
	     sblocks_for_recheck[mirror_index].page_count > 0;
	     mirror_index++) {
1087
		struct scrub_block *sblock_other;
1088

1089 1090 1091 1092 1093
		if (mirror_index == failed_mirror_index)
			continue;
		sblock_other = sblocks_for_recheck + mirror_index;

		/* build and submit the bios, check checksums */
1094
		scrub_recheck_block(fs_info, sblock_other, 0);
1095 1096

		if (!sblock_other->header_error &&
1097 1098
		    !sblock_other->checksum_error &&
		    sblock_other->no_io_error_seen) {
1099 1100
			if (sctx->is_dev_replace) {
				scrub_write_block_to_dev_replace(sblock_other);
1101
				goto corrected_error;
1102 1103
			} else {
				ret = scrub_repair_block_from_good_copy(
1104 1105 1106
						sblock_bad, sblock_other);
				if (!ret)
					goto corrected_error;
1107
			}
1108 1109
		}
	}
A
Arne Jansen 已提交
1110

1111 1112
	if (sblock_bad->no_io_error_seen && !sctx->is_dev_replace)
		goto did_not_correct_error;
1113 1114 1115

	/*
	 * In case of I/O errors in the area that is supposed to be
1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136
	 * repaired, continue by picking good copies of those pages.
	 * Select the good pages from mirrors to rewrite bad pages from
	 * the area to fix. Afterwards verify the checksum of the block
	 * that is supposed to be repaired. This verification step is
	 * only done for the purpose of statistic counting and for the
	 * final scrub report, whether errors remain.
	 * A perfect algorithm could make use of the checksum and try
	 * all possible combinations of pages from the different mirrors
	 * until the checksum verification succeeds. For example, when
	 * the 2nd page of mirror #1 faces I/O errors, and the 2nd page
	 * of mirror #2 is readable but the final checksum test fails,
	 * then the 2nd page of mirror #3 could be tried, whether now
	 * the final checksum succeedes. But this would be a rare
	 * exception and is therefore not implemented. At least it is
	 * avoided that the good copy is overwritten.
	 * A more useful improvement would be to pick the sectors
	 * without I/O error based on sector sizes (512 bytes on legacy
	 * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one
	 * mirror could be repaired by taking 512 byte of a different
	 * mirror, even if other 512 byte sectors in the same PAGE_SIZE
	 * area are unreadable.
A
Arne Jansen 已提交
1137
	 */
1138
	success = 1;
1139 1140
	for (page_num = 0; page_num < sblock_bad->page_count;
	     page_num++) {
1141
		struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1142
		struct scrub_block *sblock_other = NULL;
1143

1144 1145
		/* skip no-io-error page in scrub */
		if (!page_bad->io_error && !sctx->is_dev_replace)
A
Arne Jansen 已提交
1146
			continue;
1147

1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158
		/* try to find no-io-error page in mirrors */
		if (page_bad->io_error) {
			for (mirror_index = 0;
			     mirror_index < BTRFS_MAX_MIRRORS &&
			     sblocks_for_recheck[mirror_index].page_count > 0;
			     mirror_index++) {
				if (!sblocks_for_recheck[mirror_index].
				    pagev[page_num]->io_error) {
					sblock_other = sblocks_for_recheck +
						       mirror_index;
					break;
1159 1160
				}
			}
1161 1162
			if (!sblock_other)
				success = 0;
I
Ilya Dryomov 已提交
1163
		}
A
Arne Jansen 已提交
1164

1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191
		if (sctx->is_dev_replace) {
			/*
			 * did not find a mirror to fetch the page
			 * from. scrub_write_page_to_dev_replace()
			 * handles this case (page->io_error), by
			 * filling the block with zeros before
			 * submitting the write request
			 */
			if (!sblock_other)
				sblock_other = sblock_bad;

			if (scrub_write_page_to_dev_replace(sblock_other,
							    page_num) != 0) {
				btrfs_dev_replace_stats_inc(
					&sctx->dev_root->
					fs_info->dev_replace.
					num_write_errors);
				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;
1192
		}
A
Arne Jansen 已提交
1193 1194
	}

1195
	if (success && !sctx->is_dev_replace) {
1196 1197 1198 1199 1200 1201 1202 1203 1204 1205
		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.
			 */
1206
			scrub_recheck_block(fs_info, sblock_bad, 1);
1207
			if (!sblock_bad->header_error &&
1208 1209 1210 1211 1212 1213 1214
			    !sblock_bad->checksum_error &&
			    sblock_bad->no_io_error_seen)
				goto corrected_error;
			else
				goto did_not_correct_error;
		} else {
corrected_error:
1215 1216
			spin_lock(&sctx->stat_lock);
			sctx->stat.corrected_errors++;
1217
			sblock_to_check->data_corrected = 1;
1218
			spin_unlock(&sctx->stat_lock);
1219 1220
			btrfs_err_rl_in_rcu(fs_info,
				"fixed up error at logical %llu on dev %s",
1221
				logical, rcu_str_deref(dev->name));
A
Arne Jansen 已提交
1222
		}
1223 1224
	} else {
did_not_correct_error:
1225 1226 1227
		spin_lock(&sctx->stat_lock);
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
1228 1229
		btrfs_err_rl_in_rcu(fs_info,
			"unable to fixup (regular) error at logical %llu on dev %s",
1230
			logical, rcu_str_deref(dev->name));
I
Ilya Dryomov 已提交
1231
	}
A
Arne Jansen 已提交
1232

1233 1234 1235 1236 1237 1238
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;
1239
			struct scrub_recover *recover;
1240 1241
			int page_index;

1242 1243 1244
			for (page_index = 0; page_index < sblock->page_count;
			     page_index++) {
				sblock->pagev[page_index]->sblock = NULL;
1245 1246 1247 1248 1249 1250
				recover = sblock->pagev[page_index]->recover;
				if (recover) {
					scrub_put_recover(recover);
					sblock->pagev[page_index]->recover =
									NULL;
				}
1251 1252
				scrub_page_put(sblock->pagev[page_index]);
			}
1253 1254 1255
		}
		kfree(sblocks_for_recheck);
	}
A
Arne Jansen 已提交
1256

1257 1258
	return 0;
}
A
Arne Jansen 已提交
1259

1260
static inline int scrub_nr_raid_mirrors(struct btrfs_bio *bbio)
1261
{
Z
Zhao Lei 已提交
1262 1263 1264 1265 1266
	if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID5)
		return 2;
	else if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID6)
		return 3;
	else
1267 1268 1269
		return (int)bbio->num_stripes;
}

Z
Zhao Lei 已提交
1270 1271
static inline void scrub_stripe_index_and_offset(u64 logical, u64 map_type,
						 u64 *raid_map,
1272 1273 1274 1275 1276 1277 1278
						 u64 mapped_length,
						 int nstripes, int mirror,
						 int *stripe_index,
						 u64 *stripe_offset)
{
	int i;

1279
	if (map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299
		/* 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;
	}
}

1300
static int scrub_setup_recheck_block(struct scrub_block *original_sblock,
1301 1302
				     struct scrub_block *sblocks_for_recheck)
{
1303 1304 1305 1306
	struct scrub_ctx *sctx = original_sblock->sctx;
	struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
	u64 length = original_sblock->page_count * PAGE_SIZE;
	u64 logical = original_sblock->pagev[0]->logical;
1307 1308 1309
	u64 generation = original_sblock->pagev[0]->generation;
	u64 flags = original_sblock->pagev[0]->flags;
	u64 have_csum = original_sblock->pagev[0]->have_csum;
1310 1311 1312 1313 1314 1315
	struct scrub_recover *recover;
	struct btrfs_bio *bbio;
	u64 sublen;
	u64 mapped_length;
	u64 stripe_offset;
	int stripe_index;
1316
	int page_index = 0;
1317
	int mirror_index;
1318
	int nmirrors;
1319 1320 1321
	int ret;

	/*
1322
	 * note: the two members refs and outstanding_pages
1323 1324 1325 1326 1327
	 * are not used (and not set) in the blocks that are used for
	 * the recheck procedure
	 */

	while (length > 0) {
1328 1329 1330
		sublen = min_t(u64, length, PAGE_SIZE);
		mapped_length = sublen;
		bbio = NULL;
A
Arne Jansen 已提交
1331

1332 1333 1334 1335
		/*
		 * with a length of PAGE_SIZE, each returned stripe
		 * represents one mirror
		 */
1336
		ret = btrfs_map_sblock(fs_info, REQ_GET_READ_MIRRORS, logical,
1337
				       &mapped_length, &bbio, 0, 1);
1338
		if (ret || !bbio || mapped_length < sublen) {
1339
			btrfs_put_bbio(bbio);
1340 1341
			return -EIO;
		}
A
Arne Jansen 已提交
1342

1343 1344
		recover = kzalloc(sizeof(struct scrub_recover), GFP_NOFS);
		if (!recover) {
1345
			btrfs_put_bbio(bbio);
1346 1347 1348 1349 1350 1351 1352
			return -ENOMEM;
		}

		atomic_set(&recover->refs, 1);
		recover->bbio = bbio;
		recover->map_length = mapped_length;

1353
		BUG_ON(page_index >= SCRUB_PAGES_PER_RD_BIO);
1354

1355
		nmirrors = min(scrub_nr_raid_mirrors(bbio), BTRFS_MAX_MIRRORS);
Z
Zhao Lei 已提交
1356

1357
		for (mirror_index = 0; mirror_index < nmirrors;
1358 1359 1360 1361 1362
		     mirror_index++) {
			struct scrub_block *sblock;
			struct scrub_page *page;

			sblock = sblocks_for_recheck + mirror_index;
1363
			sblock->sctx = sctx;
1364

1365 1366 1367
			page = kzalloc(sizeof(*page), GFP_NOFS);
			if (!page) {
leave_nomem:
1368 1369 1370
				spin_lock(&sctx->stat_lock);
				sctx->stat.malloc_errors++;
				spin_unlock(&sctx->stat_lock);
1371
				scrub_put_recover(recover);
1372 1373
				return -ENOMEM;
			}
1374 1375
			scrub_page_get(page);
			sblock->pagev[page_index] = page;
1376 1377 1378
			page->sblock = sblock;
			page->flags = flags;
			page->generation = generation;
1379
			page->logical = logical;
1380 1381 1382 1383 1384
			page->have_csum = have_csum;
			if (have_csum)
				memcpy(page->csum,
				       original_sblock->pagev[0]->csum,
				       sctx->csum_size);
1385

Z
Zhao Lei 已提交
1386 1387 1388
			scrub_stripe_index_and_offset(logical,
						      bbio->map_type,
						      bbio->raid_map,
1389
						      mapped_length,
1390 1391
						      bbio->num_stripes -
						      bbio->num_tgtdevs,
1392 1393 1394 1395 1396 1397 1398
						      mirror_index,
						      &stripe_index,
						      &stripe_offset);
			page->physical = bbio->stripes[stripe_index].physical +
					 stripe_offset;
			page->dev = bbio->stripes[stripe_index].dev;

1399 1400 1401 1402
			BUG_ON(page_index >= original_sblock->page_count);
			page->physical_for_dev_replace =
				original_sblock->pagev[page_index]->
				physical_for_dev_replace;
1403 1404
			/* for missing devices, dev->bdev is NULL */
			page->mirror_num = mirror_index + 1;
1405
			sblock->page_count++;
1406 1407 1408
			page->page = alloc_page(GFP_NOFS);
			if (!page->page)
				goto leave_nomem;
1409 1410 1411

			scrub_get_recover(recover);
			page->recover = recover;
1412
		}
1413
		scrub_put_recover(recover);
1414 1415 1416 1417 1418 1419
		length -= sublen;
		logical += sublen;
		page_index++;
	}

	return 0;
I
Ilya Dryomov 已提交
1420 1421
}

1422 1423 1424 1425 1426
struct scrub_bio_ret {
	struct completion event;
	int error;
};

1427
static void scrub_bio_wait_endio(struct bio *bio)
1428 1429 1430
{
	struct scrub_bio_ret *ret = bio->bi_private;

1431
	ret->error = bio->bi_error;
1432 1433 1434 1435 1436
	complete(&ret->event);
}

static inline int scrub_is_page_on_raid56(struct scrub_page *page)
{
Z
Zhao Lei 已提交
1437
	return page->recover &&
1438
	       (page->recover->bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK);
1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455
}

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

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

	ret = raid56_parity_recover(fs_info->fs_root, bio, page->recover->bbio,
				    page->recover->map_length,
1456
				    page->mirror_num, 0);
1457 1458 1459 1460 1461 1462 1463 1464 1465 1466
	if (ret)
		return ret;

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

	return 0;
}

1467 1468 1469 1470 1471 1472 1473
/*
 * 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.
 */
1474
static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
1475 1476
				struct scrub_block *sblock,
				int retry_failed_mirror)
I
Ilya Dryomov 已提交
1477
{
1478
	int page_num;
I
Ilya Dryomov 已提交
1479

1480
	sblock->no_io_error_seen = 1;
I
Ilya Dryomov 已提交
1481

1482 1483
	for (page_num = 0; page_num < sblock->page_count; page_num++) {
		struct bio *bio;
1484
		struct scrub_page *page = sblock->pagev[page_num];
1485

1486
		if (page->dev->bdev == NULL) {
1487 1488 1489 1490 1491
			page->io_error = 1;
			sblock->no_io_error_seen = 0;
			continue;
		}

1492
		WARN_ON(!page->page);
1493
		bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1494 1495 1496 1497 1498
		if (!bio) {
			page->io_error = 1;
			sblock->no_io_error_seen = 0;
			continue;
		}
1499
		bio->bi_bdev = page->dev->bdev;
1500

1501
		bio_add_page(bio, page->page, PAGE_SIZE, 0);
1502 1503 1504 1505 1506 1507 1508 1509 1510
		if (!retry_failed_mirror && scrub_is_page_on_raid56(page)) {
			if (scrub_submit_raid56_bio_wait(fs_info, bio, page))
				sblock->no_io_error_seen = 0;
		} else {
			bio->bi_iter.bi_sector = page->physical >> 9;

			if (btrfsic_submit_bio_wait(READ, bio))
				sblock->no_io_error_seen = 0;
		}
1511

1512 1513
		bio_put(bio);
	}
I
Ilya Dryomov 已提交
1514

1515
	if (sblock->no_io_error_seen)
1516
		scrub_recheck_block_checksum(sblock);
1517

1518
	return;
A
Arne Jansen 已提交
1519 1520
}

M
Miao Xie 已提交
1521 1522 1523 1524 1525 1526 1527 1528 1529 1530
static inline int scrub_check_fsid(u8 fsid[],
				   struct scrub_page *spage)
{
	struct btrfs_fs_devices *fs_devices = spage->dev->fs_devices;
	int ret;

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

1531
static void scrub_recheck_block_checksum(struct scrub_block *sblock)
A
Arne Jansen 已提交
1532
{
1533 1534 1535
	sblock->header_error = 0;
	sblock->checksum_error = 0;
	sblock->generation_error = 0;
1536

1537 1538 1539 1540
	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
1541 1542
}

1543
static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
1544
					     struct scrub_block *sblock_good)
1545 1546 1547
{
	int page_num;
	int ret = 0;
I
Ilya Dryomov 已提交
1548

1549 1550
	for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
		int ret_sub;
I
Ilya Dryomov 已提交
1551

1552 1553
		ret_sub = scrub_repair_page_from_good_copy(sblock_bad,
							   sblock_good,
1554
							   page_num, 1);
1555 1556
		if (ret_sub)
			ret = ret_sub;
A
Arne Jansen 已提交
1557
	}
1558 1559 1560 1561 1562 1563 1564 1565

	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)
{
1566 1567
	struct scrub_page *page_bad = sblock_bad->pagev[page_num];
	struct scrub_page *page_good = sblock_good->pagev[page_num];
1568

1569 1570
	BUG_ON(page_bad->page == NULL);
	BUG_ON(page_good->page == NULL);
1571 1572 1573 1574 1575
	if (force_write || sblock_bad->header_error ||
	    sblock_bad->checksum_error || page_bad->io_error) {
		struct bio *bio;
		int ret;

1576
		if (!page_bad->dev->bdev) {
1577
			btrfs_warn_rl(sblock_bad->sctx->dev_root->fs_info,
1578
				"scrub_repair_page_from_good_copy(bdev == NULL) "
1579
				"is unexpected");
1580 1581 1582
			return -EIO;
		}

1583
		bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1584 1585
		if (!bio)
			return -EIO;
1586
		bio->bi_bdev = page_bad->dev->bdev;
1587
		bio->bi_iter.bi_sector = page_bad->physical >> 9;
1588 1589 1590 1591 1592

		ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0);
		if (PAGE_SIZE != ret) {
			bio_put(bio);
			return -EIO;
1593
		}
1594

1595
		if (btrfsic_submit_bio_wait(WRITE, bio)) {
1596 1597
			btrfs_dev_stat_inc_and_print(page_bad->dev,
				BTRFS_DEV_STAT_WRITE_ERRS);
1598 1599 1600
			btrfs_dev_replace_stats_inc(
				&sblock_bad->sctx->dev_root->fs_info->
				dev_replace.num_write_errors);
1601 1602 1603
			bio_put(bio);
			return -EIO;
		}
1604
		bio_put(bio);
A
Arne Jansen 已提交
1605 1606
	}

1607 1608 1609
	return 0;
}

1610 1611 1612 1613
static void scrub_write_block_to_dev_replace(struct scrub_block *sblock)
{
	int page_num;

1614 1615 1616 1617 1618 1619 1620
	/*
	 * 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;

1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675
	for (page_num = 0; page_num < sblock->page_count; page_num++) {
		int ret;

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

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

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

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

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

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

		sbio->physical = spage->physical_for_dev_replace;
		sbio->logical = spage->logical;
		sbio->dev = wr_ctx->tgtdev;
		bio = sbio->bio;
		if (!bio) {
1676
			bio = btrfs_io_bio_alloc(GFP_NOFS, wr_ctx->pages_per_wr_bio);
1677 1678 1679 1680 1681 1682 1683 1684 1685 1686
			if (!bio) {
				mutex_unlock(&wr_ctx->wr_lock);
				return -ENOMEM;
			}
			sbio->bio = bio;
		}

		bio->bi_private = sbio;
		bio->bi_end_io = scrub_wr_bio_end_io;
		bio->bi_bdev = sbio->dev->bdev;
1687
		bio->bi_iter.bi_sector = sbio->physical >> 9;
1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737
		sbio->err = 0;
	} else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
		   spage->physical_for_dev_replace ||
		   sbio->logical + sbio->page_count * PAGE_SIZE !=
		   spage->logical) {
		scrub_wr_submit(sctx);
		goto again;
	}

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

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

	return 0;
}

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

	if (!wr_ctx->wr_curr_bio)
		return;

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

1738
static void scrub_wr_bio_end_io(struct bio *bio)
1739 1740 1741 1742
{
	struct scrub_bio *sbio = bio->bi_private;
	struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;

1743
	sbio->err = bio->bi_error;
1744 1745
	sbio->bio = bio;

1746 1747
	btrfs_init_work(&sbio->work, btrfs_scrubwrc_helper,
			 scrub_wr_bio_end_io_worker, NULL, NULL);
1748
	btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work);
1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779
}

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

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

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

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

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

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

static int scrub_checksum(struct scrub_block *sblock)
1780 1781 1782 1783
{
	u64 flags;
	int ret;

1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795
	/*
	 * 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;

1796 1797
	WARN_ON(sblock->page_count < 1);
	flags = sblock->pagev[0]->flags;
1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808
	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);
1809 1810

	return ret;
A
Arne Jansen 已提交
1811 1812
}

1813
static int scrub_checksum_data(struct scrub_block *sblock)
A
Arne Jansen 已提交
1814
{
1815
	struct scrub_ctx *sctx = sblock->sctx;
A
Arne Jansen 已提交
1816
	u8 csum[BTRFS_CSUM_SIZE];
1817 1818 1819
	u8 *on_disk_csum;
	struct page *page;
	void *buffer;
A
Arne Jansen 已提交
1820
	u32 crc = ~(u32)0;
1821 1822
	u64 len;
	int index;
A
Arne Jansen 已提交
1823

1824
	BUG_ON(sblock->page_count < 1);
1825
	if (!sblock->pagev[0]->have_csum)
A
Arne Jansen 已提交
1826 1827
		return 0;

1828 1829
	on_disk_csum = sblock->pagev[0]->csum;
	page = sblock->pagev[0]->page;
1830
	buffer = kmap_atomic(page);
1831

1832
	len = sctx->sectorsize;
1833 1834 1835 1836
	index = 0;
	for (;;) {
		u64 l = min_t(u64, len, PAGE_SIZE);

1837
		crc = btrfs_csum_data(buffer, crc, l);
1838
		kunmap_atomic(buffer);
1839 1840 1841 1842 1843
		len -= l;
		if (len == 0)
			break;
		index++;
		BUG_ON(index >= sblock->page_count);
1844 1845
		BUG_ON(!sblock->pagev[index]->page);
		page = sblock->pagev[index]->page;
1846
		buffer = kmap_atomic(page);
1847 1848
	}

A
Arne Jansen 已提交
1849
	btrfs_csum_final(crc, csum);
1850
	if (memcmp(csum, on_disk_csum, sctx->csum_size))
1851
		sblock->checksum_error = 1;
A
Arne Jansen 已提交
1852

1853
	return sblock->checksum_error;
A
Arne Jansen 已提交
1854 1855
}

1856
static int scrub_checksum_tree_block(struct scrub_block *sblock)
A
Arne Jansen 已提交
1857
{
1858
	struct scrub_ctx *sctx = sblock->sctx;
A
Arne Jansen 已提交
1859
	struct btrfs_header *h;
1860
	struct btrfs_root *root = sctx->dev_root;
A
Arne Jansen 已提交
1861
	struct btrfs_fs_info *fs_info = root->fs_info;
1862 1863 1864 1865 1866 1867
	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
1868
	u32 crc = ~(u32)0;
1869 1870 1871 1872
	u64 len;
	int index;

	BUG_ON(sblock->page_count < 1);
1873
	page = sblock->pagev[0]->page;
1874
	mapped_buffer = kmap_atomic(page);
1875
	h = (struct btrfs_header *)mapped_buffer;
1876
	memcpy(on_disk_csum, h->csum, sctx->csum_size);
A
Arne Jansen 已提交
1877 1878 1879 1880 1881 1882

	/*
	 * we don't use the getter functions here, as we
	 * a) don't have an extent buffer and
	 * b) the page is already kmapped
	 */
1883
	if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h))
1884
		sblock->header_error = 1;
A
Arne Jansen 已提交
1885

1886 1887 1888 1889
	if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h)) {
		sblock->header_error = 1;
		sblock->generation_error = 1;
	}
A
Arne Jansen 已提交
1890

M
Miao Xie 已提交
1891
	if (!scrub_check_fsid(h->fsid, sblock->pagev[0]))
1892
		sblock->header_error = 1;
A
Arne Jansen 已提交
1893 1894 1895

	if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
		   BTRFS_UUID_SIZE))
1896
		sblock->header_error = 1;
A
Arne Jansen 已提交
1897

1898
	len = sctx->nodesize - BTRFS_CSUM_SIZE;
1899 1900 1901 1902 1903 1904
	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);

1905
		crc = btrfs_csum_data(p, crc, l);
1906
		kunmap_atomic(mapped_buffer);
1907 1908 1909 1910 1911
		len -= l;
		if (len == 0)
			break;
		index++;
		BUG_ON(index >= sblock->page_count);
1912 1913
		BUG_ON(!sblock->pagev[index]->page);
		page = sblock->pagev[index]->page;
1914
		mapped_buffer = kmap_atomic(page);
1915 1916 1917 1918 1919
		mapped_size = PAGE_SIZE;
		p = mapped_buffer;
	}

	btrfs_csum_final(crc, calculated_csum);
1920
	if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
1921
		sblock->checksum_error = 1;
A
Arne Jansen 已提交
1922

1923
	return sblock->header_error || sblock->checksum_error;
A
Arne Jansen 已提交
1924 1925
}

1926
static int scrub_checksum_super(struct scrub_block *sblock)
A
Arne Jansen 已提交
1927 1928
{
	struct btrfs_super_block *s;
1929
	struct scrub_ctx *sctx = sblock->sctx;
1930 1931 1932 1933 1934 1935
	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
1936
	u32 crc = ~(u32)0;
1937 1938
	int fail_gen = 0;
	int fail_cor = 0;
1939 1940
	u64 len;
	int index;
A
Arne Jansen 已提交
1941

1942
	BUG_ON(sblock->page_count < 1);
1943
	page = sblock->pagev[0]->page;
1944
	mapped_buffer = kmap_atomic(page);
1945
	s = (struct btrfs_super_block *)mapped_buffer;
1946
	memcpy(on_disk_csum, s->csum, sctx->csum_size);
A
Arne Jansen 已提交
1947

1948
	if (sblock->pagev[0]->logical != btrfs_super_bytenr(s))
1949
		++fail_cor;
A
Arne Jansen 已提交
1950

1951
	if (sblock->pagev[0]->generation != btrfs_super_generation(s))
1952
		++fail_gen;
A
Arne Jansen 已提交
1953

M
Miao Xie 已提交
1954
	if (!scrub_check_fsid(s->fsid, sblock->pagev[0]))
1955
		++fail_cor;
A
Arne Jansen 已提交
1956

1957 1958 1959 1960 1961 1962 1963
	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);

1964
		crc = btrfs_csum_data(p, crc, l);
1965
		kunmap_atomic(mapped_buffer);
1966 1967 1968 1969 1970
		len -= l;
		if (len == 0)
			break;
		index++;
		BUG_ON(index >= sblock->page_count);
1971 1972
		BUG_ON(!sblock->pagev[index]->page);
		page = sblock->pagev[index]->page;
1973
		mapped_buffer = kmap_atomic(page);
1974 1975 1976 1977 1978
		mapped_size = PAGE_SIZE;
		p = mapped_buffer;
	}

	btrfs_csum_final(crc, calculated_csum);
1979
	if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
1980
		++fail_cor;
A
Arne Jansen 已提交
1981

1982
	if (fail_cor + fail_gen) {
A
Arne Jansen 已提交
1983 1984 1985 1986 1987
		/*
		 * if we find an error in a super block, we just report it.
		 * They will get written with the next transaction commit
		 * anyway
		 */
1988 1989 1990
		spin_lock(&sctx->stat_lock);
		++sctx->stat.super_errors;
		spin_unlock(&sctx->stat_lock);
1991
		if (fail_cor)
1992
			btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
1993 1994
				BTRFS_DEV_STAT_CORRUPTION_ERRS);
		else
1995
			btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
1996
				BTRFS_DEV_STAT_GENERATION_ERRS);
A
Arne Jansen 已提交
1997 1998
	}

1999
	return fail_cor + fail_gen;
A
Arne Jansen 已提交
2000 2001
}

2002 2003
static void scrub_block_get(struct scrub_block *sblock)
{
2004
	atomic_inc(&sblock->refs);
2005 2006 2007 2008
}

static void scrub_block_put(struct scrub_block *sblock)
{
2009
	if (atomic_dec_and_test(&sblock->refs)) {
2010 2011
		int i;

2012 2013 2014
		if (sblock->sparity)
			scrub_parity_put(sblock->sparity);

2015
		for (i = 0; i < sblock->page_count; i++)
2016
			scrub_page_put(sblock->pagev[i]);
2017 2018 2019 2020
		kfree(sblock);
	}
}

2021 2022
static void scrub_page_get(struct scrub_page *spage)
{
2023
	atomic_inc(&spage->refs);
2024 2025 2026 2027
}

static void scrub_page_put(struct scrub_page *spage)
{
2028
	if (atomic_dec_and_test(&spage->refs)) {
2029 2030 2031 2032 2033 2034
		if (spage->page)
			__free_page(spage->page);
		kfree(spage);
	}
}

2035
static void scrub_submit(struct scrub_ctx *sctx)
A
Arne Jansen 已提交
2036 2037 2038
{
	struct scrub_bio *sbio;

2039
	if (sctx->curr == -1)
S
Stefan Behrens 已提交
2040
		return;
A
Arne Jansen 已提交
2041

2042 2043
	sbio = sctx->bios[sctx->curr];
	sctx->curr = -1;
2044
	scrub_pending_bio_inc(sctx);
2045
	btrfsic_submit_bio(READ, sbio->bio);
A
Arne Jansen 已提交
2046 2047
}

2048 2049
static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
				    struct scrub_page *spage)
A
Arne Jansen 已提交
2050
{
2051
	struct scrub_block *sblock = spage->sblock;
A
Arne Jansen 已提交
2052
	struct scrub_bio *sbio;
2053
	int ret;
A
Arne Jansen 已提交
2054 2055 2056 2057 2058

again:
	/*
	 * grab a fresh bio or wait for one to become available
	 */
2059 2060 2061 2062 2063 2064 2065 2066
	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
2067
		} else {
2068 2069
			spin_unlock(&sctx->list_lock);
			wait_event(sctx->list_wait, sctx->first_free != -1);
A
Arne Jansen 已提交
2070 2071
		}
	}
2072
	sbio = sctx->bios[sctx->curr];
2073
	if (sbio->page_count == 0) {
2074 2075
		struct bio *bio;

2076 2077
		sbio->physical = spage->physical;
		sbio->logical = spage->logical;
2078
		sbio->dev = spage->dev;
2079 2080
		bio = sbio->bio;
		if (!bio) {
2081
			bio = btrfs_io_bio_alloc(GFP_NOFS, sctx->pages_per_rd_bio);
2082 2083 2084 2085
			if (!bio)
				return -ENOMEM;
			sbio->bio = bio;
		}
2086 2087 2088

		bio->bi_private = sbio;
		bio->bi_end_io = scrub_bio_end_io;
2089
		bio->bi_bdev = sbio->dev->bdev;
2090
		bio->bi_iter.bi_sector = sbio->physical >> 9;
2091
		sbio->err = 0;
2092 2093 2094
	} else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
		   spage->physical ||
		   sbio->logical + sbio->page_count * PAGE_SIZE !=
2095 2096
		   spage->logical ||
		   sbio->dev != spage->dev) {
2097
		scrub_submit(sctx);
A
Arne Jansen 已提交
2098 2099
		goto again;
	}
2100

2101 2102 2103 2104 2105 2106 2107 2108
	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;
		}
2109
		scrub_submit(sctx);
2110 2111 2112
		goto again;
	}

2113
	scrub_block_get(sblock); /* one for the page added to the bio */
2114 2115
	atomic_inc(&sblock->outstanding_pages);
	sbio->page_count++;
2116
	if (sbio->page_count == sctx->pages_per_rd_bio)
2117
		scrub_submit(sctx);
2118 2119 2120 2121

	return 0;
}

2122
static void scrub_missing_raid56_end_io(struct bio *bio)
2123 2124 2125 2126
{
	struct scrub_block *sblock = bio->bi_private;
	struct btrfs_fs_info *fs_info = sblock->sctx->dev_root->fs_info;

2127
	if (bio->bi_error)
2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142
		sblock->no_io_error_seen = 0;

	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;
	u64 logical;
	struct btrfs_device *dev;

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

2143
	if (sblock->no_io_error_seen)
2144
		scrub_recheck_block_checksum(sblock);
2145 2146 2147 2148 2149

	if (!sblock->no_io_error_seen) {
		spin_lock(&sctx->stat_lock);
		sctx->stat.read_errors++;
		spin_unlock(&sctx->stat_lock);
2150
		btrfs_err_rl_in_rcu(sctx->dev_root->fs_info,
2151
			"IO error rebuilding logical %llu for dev %s",
2152 2153 2154 2155 2156
			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);
2157
		btrfs_err_rl_in_rcu(sctx->dev_root->fs_info,
2158
			"failed to rebuild valid logical %llu for dev %s",
2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237
			logical, rcu_str_deref(dev->name));
	} else {
		scrub_write_block_to_dev_replace(sblock);
	}

	scrub_block_put(sblock);

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

	scrub_pending_bio_dec(sctx);
}

static void scrub_missing_raid56_pages(struct scrub_block *sblock)
{
	struct scrub_ctx *sctx = sblock->sctx;
	struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
	u64 length = sblock->page_count * PAGE_SIZE;
	u64 logical = sblock->pagev[0]->logical;
	struct btrfs_bio *bbio;
	struct bio *bio;
	struct btrfs_raid_bio *rbio;
	int ret;
	int i;

	ret = btrfs_map_sblock(fs_info, REQ_GET_READ_MIRRORS, logical, &length,
			       &bbio, 0, 1);
	if (ret || !bbio || !bbio->raid_map)
		goto bbio_out;

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

	bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
	if (!bio)
		goto bbio_out;

	bio->bi_iter.bi_sector = logical >> 9;
	bio->bi_private = sblock;
	bio->bi_end_io = scrub_missing_raid56_end_io;

	rbio = raid56_alloc_missing_rbio(sctx->dev_root, bio, bbio, length);
	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:
	btrfs_put_bbio(bbio);
	spin_lock(&sctx->stat_lock);
	sctx->stat.malloc_errors++;
	spin_unlock(&sctx->stat_lock);
}

2238
static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
2239
		       u64 physical, struct btrfs_device *dev, u64 flags,
2240 2241
		       u64 gen, int mirror_num, u8 *csum, int force,
		       u64 physical_for_dev_replace)
2242 2243 2244 2245 2246 2247
{
	struct scrub_block *sblock;
	int index;

	sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
	if (!sblock) {
2248 2249 2250
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		spin_unlock(&sctx->stat_lock);
2251
		return -ENOMEM;
A
Arne Jansen 已提交
2252
	}
2253

2254 2255
	/* one ref inside this function, plus one for each page added to
	 * a bio later on */
2256
	atomic_set(&sblock->refs, 1);
2257
	sblock->sctx = sctx;
2258 2259 2260
	sblock->no_io_error_seen = 1;

	for (index = 0; len > 0; index++) {
2261
		struct scrub_page *spage;
2262 2263
		u64 l = min_t(u64, len, PAGE_SIZE);

2264 2265 2266
		spage = kzalloc(sizeof(*spage), GFP_NOFS);
		if (!spage) {
leave_nomem:
2267 2268 2269
			spin_lock(&sctx->stat_lock);
			sctx->stat.malloc_errors++;
			spin_unlock(&sctx->stat_lock);
2270
			scrub_block_put(sblock);
2271 2272
			return -ENOMEM;
		}
2273 2274 2275
		BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
		scrub_page_get(spage);
		sblock->pagev[index] = spage;
2276
		spage->sblock = sblock;
2277
		spage->dev = dev;
2278 2279 2280 2281
		spage->flags = flags;
		spage->generation = gen;
		spage->logical = logical;
		spage->physical = physical;
2282
		spage->physical_for_dev_replace = physical_for_dev_replace;
2283 2284 2285
		spage->mirror_num = mirror_num;
		if (csum) {
			spage->have_csum = 1;
2286
			memcpy(spage->csum, csum, sctx->csum_size);
2287 2288 2289 2290
		} else {
			spage->have_csum = 0;
		}
		sblock->page_count++;
2291 2292 2293
		spage->page = alloc_page(GFP_NOFS);
		if (!spage->page)
			goto leave_nomem;
2294 2295 2296
		len -= l;
		logical += l;
		physical += l;
2297
		physical_for_dev_replace += l;
2298 2299
	}

2300
	WARN_ON(sblock->page_count == 0);
2301 2302 2303 2304 2305 2306 2307 2308 2309 2310
	if (dev->missing) {
		/*
		 * This case should only be hit for RAID 5/6 device replace. See
		 * the comment in scrub_missing_raid56_pages() for details.
		 */
		scrub_missing_raid56_pages(sblock);
	} else {
		for (index = 0; index < sblock->page_count; index++) {
			struct scrub_page *spage = sblock->pagev[index];
			int ret;
2311

2312 2313 2314 2315 2316
			ret = scrub_add_page_to_rd_bio(sctx, spage);
			if (ret) {
				scrub_block_put(sblock);
				return ret;
			}
2317
		}
A
Arne Jansen 已提交
2318

2319 2320 2321
		if (force)
			scrub_submit(sctx);
	}
A
Arne Jansen 已提交
2322

2323 2324
	/* last one frees, either here or in bio completion for last page */
	scrub_block_put(sblock);
A
Arne Jansen 已提交
2325 2326 2327
	return 0;
}

2328
static void scrub_bio_end_io(struct bio *bio)
2329 2330
{
	struct scrub_bio *sbio = bio->bi_private;
2331
	struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
2332

2333
	sbio->err = bio->bi_error;
2334 2335
	sbio->bio = bio;

2336
	btrfs_queue_work(fs_info->scrub_workers, &sbio->work);
2337 2338 2339 2340 2341
}

static void scrub_bio_end_io_worker(struct btrfs_work *work)
{
	struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
2342
	struct scrub_ctx *sctx = sbio->sctx;
2343 2344
	int i;

2345
	BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO);
2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366
	if (sbio->err) {
		for (i = 0; i < sbio->page_count; i++) {
			struct scrub_page *spage = sbio->pagev[i];

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

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

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

	bio_put(sbio->bio);
	sbio->bio = NULL;
2367 2368 2369 2370
	spin_lock(&sctx->list_lock);
	sbio->next_free = sctx->first_free;
	sctx->first_free = sbio->index;
	spin_unlock(&sctx->list_lock);
2371 2372 2373 2374 2375 2376 2377 2378

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

2379
	scrub_pending_bio_dec(sctx);
2380 2381
}

2382 2383 2384 2385
static inline void __scrub_mark_bitmap(struct scrub_parity *sparity,
				       unsigned long *bitmap,
				       u64 start, u64 len)
{
2386
	u32 offset;
2387 2388 2389 2390 2391 2392 2393 2394 2395
	int nsectors;
	int sectorsize = sparity->sctx->dev_root->sectorsize;

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

	start -= sparity->logic_start;
2396
	start = div_u64_rem(start, sparity->stripe_len, &offset);
2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420
	offset /= sectorsize;
	nsectors = (int)len / sectorsize;

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

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

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

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

2421 2422
static void scrub_block_complete(struct scrub_block *sblock)
{
2423 2424
	int corrupted = 0;

2425
	if (!sblock->no_io_error_seen) {
2426
		corrupted = 1;
2427
		scrub_handle_errored_block(sblock);
2428 2429 2430 2431 2432 2433
	} else {
		/*
		 * if has checksum error, write via repair mechanism in
		 * dev replace case, otherwise write here in dev replace
		 * case.
		 */
2434 2435
		corrupted = scrub_checksum(sblock);
		if (!corrupted && sblock->sctx->is_dev_replace)
2436 2437
			scrub_write_block_to_dev_replace(sblock);
	}
2438 2439 2440 2441 2442 2443 2444 2445 2446

	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);
	}
2447 2448
}

2449
static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len,
A
Arne Jansen 已提交
2450 2451 2452
			   u8 *csum)
{
	struct btrfs_ordered_sum *sum = NULL;
2453
	unsigned long index;
A
Arne Jansen 已提交
2454 2455
	unsigned long num_sectors;

2456 2457
	while (!list_empty(&sctx->csum_list)) {
		sum = list_first_entry(&sctx->csum_list,
A
Arne Jansen 已提交
2458 2459 2460 2461 2462 2463
				       struct btrfs_ordered_sum, list);
		if (sum->bytenr > logical)
			return 0;
		if (sum->bytenr + sum->len > logical)
			break;

2464
		++sctx->stat.csum_discards;
A
Arne Jansen 已提交
2465 2466 2467 2468 2469 2470 2471
		list_del(&sum->list);
		kfree(sum);
		sum = NULL;
	}
	if (!sum)
		return 0;

2472
	index = ((u32)(logical - sum->bytenr)) / sctx->sectorsize;
2473
	num_sectors = sum->len / sctx->sectorsize;
2474 2475
	memcpy(csum, sum->sums + index, sctx->csum_size);
	if (index == num_sectors - 1) {
A
Arne Jansen 已提交
2476 2477 2478
		list_del(&sum->list);
		kfree(sum);
	}
2479
	return 1;
A
Arne Jansen 已提交
2480 2481 2482
}

/* scrub extent tries to collect up to 64 kB for each bio */
2483
static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len,
2484
			u64 physical, struct btrfs_device *dev, u64 flags,
2485
			u64 gen, int mirror_num, u64 physical_for_dev_replace)
A
Arne Jansen 已提交
2486 2487 2488
{
	int ret;
	u8 csum[BTRFS_CSUM_SIZE];
2489 2490 2491
	u32 blocksize;

	if (flags & BTRFS_EXTENT_FLAG_DATA) {
2492 2493 2494 2495 2496
		blocksize = sctx->sectorsize;
		spin_lock(&sctx->stat_lock);
		sctx->stat.data_extents_scrubbed++;
		sctx->stat.data_bytes_scrubbed += len;
		spin_unlock(&sctx->stat_lock);
2497
	} else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2498 2499 2500 2501 2502
		blocksize = sctx->nodesize;
		spin_lock(&sctx->stat_lock);
		sctx->stat.tree_extents_scrubbed++;
		sctx->stat.tree_bytes_scrubbed += len;
		spin_unlock(&sctx->stat_lock);
2503
	} else {
2504
		blocksize = sctx->sectorsize;
2505
		WARN_ON(1);
2506
	}
A
Arne Jansen 已提交
2507 2508

	while (len) {
2509
		u64 l = min_t(u64, len, blocksize);
A
Arne Jansen 已提交
2510 2511 2512 2513
		int have_csum = 0;

		if (flags & BTRFS_EXTENT_FLAG_DATA) {
			/* push csums to sbio */
2514
			have_csum = scrub_find_csum(sctx, logical, l, csum);
A
Arne Jansen 已提交
2515
			if (have_csum == 0)
2516
				++sctx->stat.no_csum;
2517 2518 2519 2520 2521 2522
			if (sctx->is_dev_replace && !have_csum) {
				ret = copy_nocow_pages(sctx, logical, l,
						       mirror_num,
						      physical_for_dev_replace);
				goto behind_scrub_pages;
			}
A
Arne Jansen 已提交
2523
		}
2524
		ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen,
2525 2526 2527
				  mirror_num, have_csum ? csum : NULL, 0,
				  physical_for_dev_replace);
behind_scrub_pages:
A
Arne Jansen 已提交
2528 2529 2530 2531 2532
		if (ret)
			return ret;
		len -= l;
		logical += l;
		physical += l;
2533
		physical_for_dev_replace += l;
A
Arne Jansen 已提交
2534 2535 2536 2537
	}
	return 0;
}

2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556
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;

	sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
	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 */
2557
	atomic_set(&sblock->refs, 1);
2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 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 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631
	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);

		spage = kzalloc(sizeof(*spage), GFP_NOFS);
		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++;
		spage->page = alloc_page(GFP_NOFS);
		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;

2632 2633 2634 2635 2636
	if (dev->missing) {
		scrub_parity_mark_sectors_error(sparity, logical, len);
		return 0;
	}

2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660
	if (flags & BTRFS_EXTENT_FLAG_DATA) {
		blocksize = sctx->sectorsize;
	} else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
		blocksize = sctx->nodesize;
	} else {
		blocksize = sctx->sectorsize;
		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 */
			have_csum = scrub_find_csum(sctx, logical, l, csum);
			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;
2661
skip:
2662 2663 2664 2665 2666 2667 2668
		len -= l;
		logical += l;
		physical += l;
	}
	return 0;
}

2669 2670 2671 2672 2673 2674 2675 2676
/*
 * 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,
2677 2678
				   struct map_lookup *map, u64 *offset,
				   u64 *stripe_start)
2679 2680 2681 2682 2683
{
	int i;
	int j = 0;
	u64 stripe_nr;
	u64 last_offset;
2684 2685
	u32 stripe_index;
	u32 rot;
2686 2687 2688

	last_offset = (physical - map->stripes[num].physical) *
		      nr_data_stripes(map);
2689 2690 2691
	if (stripe_start)
		*stripe_start = last_offset;

2692 2693 2694 2695
	*offset = last_offset;
	for (i = 0; i < nr_data_stripes(map); i++) {
		*offset = last_offset + i * map->stripe_len;

2696 2697
		stripe_nr = div_u64(*offset, map->stripe_len);
		stripe_nr = div_u64(stripe_nr, nr_data_stripes(map));
2698 2699

		/* Work out the disk rotation on this stripe-set */
2700
		stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, &rot);
2701 2702
		/* calculate which stripe this data locates */
		rot += i;
2703
		stripe_index = rot % map->num_stripes;
2704 2705 2706 2707 2708 2709 2710 2711 2712
		if (stripe_index == num)
			return 0;
		if (stripe_index < num)
			j++;
	}
	*offset = last_offset + j * map->stripe_len;
	return 1;
}

2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734
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);
}

2735 2736 2737 2738 2739 2740 2741 2742 2743 2744
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);
}

2745
static void scrub_parity_bio_endio(struct bio *bio)
2746 2747 2748
{
	struct scrub_parity *sparity = (struct scrub_parity *)bio->bi_private;

2749
	if (bio->bi_error)
2750 2751 2752 2753
		bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap,
			  sparity->nsectors);

	bio_put(bio);
2754 2755 2756 2757 2758

	btrfs_init_work(&sparity->work, btrfs_scrubparity_helper,
			scrub_parity_bio_endio_worker, NULL, NULL);
	btrfs_queue_work(sparity->sctx->dev_root->fs_info->scrub_parity_workers,
			 &sparity->work);
2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774
}

static void scrub_parity_check_and_repair(struct scrub_parity *sparity)
{
	struct scrub_ctx *sctx = sparity->sctx;
	struct bio *bio;
	struct btrfs_raid_bio *rbio;
	struct scrub_page *spage;
	struct btrfs_bio *bbio = NULL;
	u64 length;
	int ret;

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

2775
	length = sparity->logic_end - sparity->logic_start;
2776
	ret = btrfs_map_sblock(sctx->dev_root->fs_info, WRITE,
2777
			       sparity->logic_start,
2778 2779
			       &length, &bbio, 0, 1);
	if (ret || !bbio || !bbio->raid_map)
2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790
		goto bbio_out;

	bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
	if (!bio)
		goto bbio_out;

	bio->bi_iter.bi_sector = sparity->logic_start >> 9;
	bio->bi_private = sparity;
	bio->bi_end_io = scrub_parity_bio_endio;

	rbio = raid56_parity_alloc_scrub_rbio(sctx->dev_root, bio, bbio,
2791
					      length, sparity->scrub_dev,
2792 2793 2794 2795 2796 2797
					      sparity->dbitmap,
					      sparity->nsectors);
	if (!rbio)
		goto rbio_out;

	list_for_each_entry(spage, &sparity->spages, list)
2798
		raid56_add_scrub_pages(rbio, spage->page, spage->logical);
2799 2800 2801 2802 2803 2804 2805 2806

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

rbio_out:
	bio_put(bio);
bbio_out:
2807
	btrfs_put_bbio(bbio);
2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823
	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)
{
	return DIV_ROUND_UP(nsectors, BITS_PER_LONG) * (BITS_PER_LONG / 8);
}

static void scrub_parity_get(struct scrub_parity *sparity)
{
2824
	atomic_inc(&sparity->refs);
2825 2826 2827 2828
}

static void scrub_parity_put(struct scrub_parity *sparity)
{
2829
	if (!atomic_dec_and_test(&sparity->refs))
2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845
		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)
{
	struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
	struct btrfs_root *root = fs_info->extent_root;
	struct btrfs_root *csum_root = fs_info->csum_root;
	struct btrfs_extent_item *extent;
2846
	struct btrfs_bio *bbio = NULL;
2847 2848 2849 2850 2851 2852 2853 2854 2855
	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;
2856
	u64 mapped_length;
2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880
	struct btrfs_device *extent_dev;
	struct scrub_parity *sparity;
	int nsectors;
	int bitmap_len;
	int extent_mirror_num;
	int stop_loop = 0;

	nsectors = map->stripe_len / root->sectorsize;
	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;
2881
	atomic_set(&sparity->refs, 1);
2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929
	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);

2930 2931 2932 2933
			if (key.type != BTRFS_EXTENT_ITEM_KEY &&
			    key.type != BTRFS_METADATA_ITEM_KEY)
				goto next;

2934 2935 2936 2937 2938 2939 2940 2941
			if (key.type == BTRFS_METADATA_ITEM_KEY)
				bytes = root->nodesize;
			else
				bytes = key.offset;

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

2942
			if (key.objectid >= logic_end) {
2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954
				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);

2955 2956 2957 2958 2959 2960
			if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) &&
			    (key.objectid < logic_start ||
			     key.objectid + bytes >
			     logic_start + map->stripe_len)) {
				btrfs_err(fs_info, "scrub: tree block %llu spanning stripes, ignored. logical=%llu",
					  key.objectid, logic_start);
2961 2962 2963
				spin_lock(&sctx->stat_lock);
				sctx->stat.uncorrectable_errors++;
				spin_unlock(&sctx->stat_lock);
2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982
				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);

2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997
			mapped_length = extent_len;
			ret = btrfs_map_block(fs_info, READ, extent_logical,
					      &mapped_length, &bbio, 0);
			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);
2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011

			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);
3012 3013 3014

			scrub_free_csums(sctx);

3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045
			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,
3046
						logic_end - logic_start);
3047 3048 3049 3050 3051 3052 3053 3054 3055 3056
	scrub_parity_put(sparity);
	scrub_submit(sctx);
	mutex_lock(&sctx->wr_ctx.wr_lock);
	scrub_wr_submit(sctx);
	mutex_unlock(&sctx->wr_ctx.wr_lock);

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

3057
static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
3058 3059
					   struct map_lookup *map,
					   struct btrfs_device *scrub_dev,
3060 3061
					   int num, u64 base, u64 length,
					   int is_dev_replace)
A
Arne Jansen 已提交
3062
{
3063
	struct btrfs_path *path, *ppath;
3064
	struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
A
Arne Jansen 已提交
3065 3066 3067
	struct btrfs_root *root = fs_info->extent_root;
	struct btrfs_root *csum_root = fs_info->csum_root;
	struct btrfs_extent_item *extent;
3068
	struct blk_plug plug;
A
Arne Jansen 已提交
3069 3070 3071 3072 3073 3074 3075 3076
	u64 flags;
	int ret;
	int slot;
	u64 nstripes;
	struct extent_buffer *l;
	struct btrfs_key key;
	u64 physical;
	u64 logical;
L
Liu Bo 已提交
3077
	u64 logic_end;
3078
	u64 physical_end;
A
Arne Jansen 已提交
3079
	u64 generation;
3080
	int mirror_num;
A
Arne Jansen 已提交
3081 3082 3083 3084
	struct reada_control *reada1;
	struct reada_control *reada2;
	struct btrfs_key key_start;
	struct btrfs_key key_end;
A
Arne Jansen 已提交
3085 3086
	u64 increment = map->stripe_len;
	u64 offset;
3087 3088 3089
	u64 extent_logical;
	u64 extent_physical;
	u64 extent_len;
3090 3091
	u64 stripe_logical;
	u64 stripe_end;
3092 3093
	struct btrfs_device *extent_dev;
	int extent_mirror_num;
3094
	int stop_loop = 0;
D
David Woodhouse 已提交
3095

3096
	physical = map->stripes[num].physical;
A
Arne Jansen 已提交
3097
	offset = 0;
3098
	nstripes = div_u64(length, map->stripe_len);
A
Arne Jansen 已提交
3099 3100 3101
	if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
		offset = map->stripe_len * num;
		increment = map->stripe_len * map->num_stripes;
3102
		mirror_num = 1;
A
Arne Jansen 已提交
3103 3104 3105 3106
	} 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;
3107
		mirror_num = num % map->sub_stripes + 1;
A
Arne Jansen 已提交
3108 3109
	} else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
		increment = map->stripe_len;
3110
		mirror_num = num % map->num_stripes + 1;
A
Arne Jansen 已提交
3111 3112
	} else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
		increment = map->stripe_len;
3113
		mirror_num = num % map->num_stripes + 1;
3114
	} else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3115
		get_raid56_logic_offset(physical, num, map, &offset, NULL);
3116 3117
		increment = map->stripe_len * nr_data_stripes(map);
		mirror_num = 1;
A
Arne Jansen 已提交
3118 3119
	} else {
		increment = map->stripe_len;
3120
		mirror_num = 1;
A
Arne Jansen 已提交
3121 3122 3123 3124 3125 3126
	}

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

3127 3128
	ppath = btrfs_alloc_path();
	if (!ppath) {
3129
		btrfs_free_path(path);
3130 3131 3132
		return -ENOMEM;
	}

3133 3134 3135 3136 3137
	/*
	 * 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
3138 3139 3140
	path->search_commit_root = 1;
	path->skip_locking = 1;

3141 3142
	ppath->search_commit_root = 1;
	ppath->skip_locking = 1;
A
Arne Jansen 已提交
3143
	/*
A
Arne Jansen 已提交
3144 3145 3146
	 * 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
3147 3148
	 */
	logical = base + offset;
3149
	physical_end = physical + nstripes * map->stripe_len;
3150
	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3151
		get_raid56_logic_offset(physical_end, num,
3152
					map, &logic_end, NULL);
3153 3154 3155 3156
		logic_end += base;
	} else {
		logic_end = logical + increment * nstripes;
	}
3157
	wait_event(sctx->list_wait,
3158
		   atomic_read(&sctx->bios_in_flight) == 0);
3159
	scrub_blocked_if_needed(fs_info);
A
Arne Jansen 已提交
3160 3161 3162 3163 3164

	/* FIXME it might be better to start readahead at commit root */
	key_start.objectid = logical;
	key_start.type = BTRFS_EXTENT_ITEM_KEY;
	key_start.offset = (u64)0;
3165
	key_end.objectid = logic_end;
3166 3167
	key_end.type = BTRFS_METADATA_ITEM_KEY;
	key_end.offset = (u64)-1;
A
Arne Jansen 已提交
3168 3169 3170 3171 3172 3173 3174
	reada1 = btrfs_reada_add(root, &key_start, &key_end);

	key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
	key_start.type = BTRFS_EXTENT_CSUM_KEY;
	key_start.offset = logical;
	key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
	key_end.type = BTRFS_EXTENT_CSUM_KEY;
3175
	key_end.offset = logic_end;
A
Arne Jansen 已提交
3176 3177 3178 3179 3180 3181 3182
	reada2 = btrfs_reada_add(csum_root, &key_start, &key_end);

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

A
Arne Jansen 已提交
3183 3184 3185 3186 3187

	/*
	 * collect all data csums for the stripe to avoid seeking during
	 * the scrub. This might currently (crc32) end up to be about 1MB
	 */
3188
	blk_start_plug(&plug);
A
Arne Jansen 已提交
3189 3190 3191 3192 3193

	/*
	 * now find all extents for each stripe and scrub them
	 */
	ret = 0;
3194
	while (physical < physical_end) {
A
Arne Jansen 已提交
3195 3196 3197 3198
		/*
		 * canceled?
		 */
		if (atomic_read(&fs_info->scrub_cancel_req) ||
3199
		    atomic_read(&sctx->cancel_req)) {
A
Arne Jansen 已提交
3200 3201 3202 3203 3204 3205 3206 3207
			ret = -ECANCELED;
			goto out;
		}
		/*
		 * check to see if we have to pause
		 */
		if (atomic_read(&fs_info->scrub_pause_req)) {
			/* push queued extents */
3208
			atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
3209
			scrub_submit(sctx);
3210 3211 3212
			mutex_lock(&sctx->wr_ctx.wr_lock);
			scrub_wr_submit(sctx);
			mutex_unlock(&sctx->wr_ctx.wr_lock);
3213
			wait_event(sctx->list_wait,
3214
				   atomic_read(&sctx->bios_in_flight) == 0);
3215
			atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
3216
			scrub_blocked_if_needed(fs_info);
A
Arne Jansen 已提交
3217 3218
		}

3219 3220 3221 3222 3223 3224
		if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
			ret = get_raid56_logic_offset(physical, num, map,
						      &logical,
						      &stripe_logical);
			logical += base;
			if (ret) {
3225
				/* it is parity strip */
3226
				stripe_logical += base;
3227
				stripe_end = stripe_logical + increment;
3228 3229 3230 3231 3232 3233 3234 3235 3236
				ret = scrub_raid56_parity(sctx, map, scrub_dev,
							  ppath, stripe_logical,
							  stripe_end);
				if (ret)
					goto out;
				goto skip;
			}
		}

3237 3238 3239 3240
		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
3241
		key.objectid = logical;
L
Liu Bo 已提交
3242
		key.offset = (u64)-1;
A
Arne Jansen 已提交
3243 3244 3245 3246

		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
		if (ret < 0)
			goto out;
3247

3248
		if (ret > 0) {
3249
			ret = btrfs_previous_extent_item(root, path, 0);
A
Arne Jansen 已提交
3250 3251
			if (ret < 0)
				goto out;
3252 3253 3254 3255 3256 3257 3258 3259 3260
			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
3261 3262
		}

L
Liu Bo 已提交
3263
		stop_loop = 0;
A
Arne Jansen 已提交
3264
		while (1) {
3265 3266
			u64 bytes;

A
Arne Jansen 已提交
3267 3268 3269 3270 3271 3272 3273 3274 3275
			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
3276
				stop_loop = 1;
A
Arne Jansen 已提交
3277 3278 3279 3280
				break;
			}
			btrfs_item_key_to_cpu(l, &key, slot);

3281 3282 3283 3284
			if (key.type != BTRFS_EXTENT_ITEM_KEY &&
			    key.type != BTRFS_METADATA_ITEM_KEY)
				goto next;

3285
			if (key.type == BTRFS_METADATA_ITEM_KEY)
3286
				bytes = root->nodesize;
3287 3288 3289 3290
			else
				bytes = key.offset;

			if (key.objectid + bytes <= logical)
A
Arne Jansen 已提交
3291 3292
				goto next;

L
Liu Bo 已提交
3293 3294 3295 3296 3297 3298
			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
3299 3300 3301 3302 3303 3304

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

3305 3306 3307 3308
			if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) &&
			    (key.objectid < logical ||
			     key.objectid + bytes >
			     logical + map->stripe_len)) {
3309 3310 3311
				btrfs_err(fs_info,
					   "scrub: tree block %llu spanning "
					   "stripes, ignored. logical=%llu",
3312
				       key.objectid, logical);
3313 3314 3315
				spin_lock(&sctx->stat_lock);
				sctx->stat.uncorrectable_errors++;
				spin_unlock(&sctx->stat_lock);
A
Arne Jansen 已提交
3316 3317 3318
				goto next;
			}

L
Liu Bo 已提交
3319 3320 3321 3322
again:
			extent_logical = key.objectid;
			extent_len = bytes;

A
Arne Jansen 已提交
3323 3324 3325
			/*
			 * trim extent to this stripe
			 */
L
Liu Bo 已提交
3326 3327 3328
			if (extent_logical < logical) {
				extent_len -= logical - extent_logical;
				extent_logical = logical;
A
Arne Jansen 已提交
3329
			}
L
Liu Bo 已提交
3330
			if (extent_logical + extent_len >
A
Arne Jansen 已提交
3331
			    logical + map->stripe_len) {
L
Liu Bo 已提交
3332 3333
				extent_len = logical + map->stripe_len -
					     extent_logical;
A
Arne Jansen 已提交
3334 3335
			}

L
Liu Bo 已提交
3336
			extent_physical = extent_logical - logical + physical;
3337 3338 3339 3340 3341 3342 3343
			extent_dev = scrub_dev;
			extent_mirror_num = mirror_num;
			if (is_dev_replace)
				scrub_remap_extent(fs_info, extent_logical,
						   extent_len, &extent_physical,
						   &extent_dev,
						   &extent_mirror_num);
L
Liu Bo 已提交
3344

3345 3346 3347 3348 3349
			ret = btrfs_lookup_csums_range(csum_root,
						       extent_logical,
						       extent_logical +
						       extent_len - 1,
						       &sctx->csum_list, 1);
L
Liu Bo 已提交
3350 3351 3352
			if (ret)
				goto out;

3353 3354 3355
			ret = scrub_extent(sctx, extent_logical, extent_len,
					   extent_physical, extent_dev, flags,
					   generation, extent_mirror_num,
3356
					   extent_logical - logical + physical);
3357 3358 3359

			scrub_free_csums(sctx);

A
Arne Jansen 已提交
3360 3361 3362
			if (ret)
				goto out;

L
Liu Bo 已提交
3363 3364
			if (extent_logical + extent_len <
			    key.objectid + bytes) {
3365
				if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3366 3367 3368 3369
					/*
					 * loop until we find next data stripe
					 * or we have finished all stripes.
					 */
3370 3371 3372 3373 3374 3375 3376 3377 3378 3379
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 +
3380
								increment;
3381 3382 3383 3384 3385 3386 3387 3388
						ret = scrub_raid56_parity(sctx,
							map, scrub_dev, ppath,
							stripe_logical,
							stripe_end);
						if (ret)
							goto out;
						goto loop;
					}
3389 3390 3391 3392
				} else {
					physical += map->stripe_len;
					logical += increment;
				}
L
Liu Bo 已提交
3393 3394 3395 3396 3397
				if (logical < key.objectid + bytes) {
					cond_resched();
					goto again;
				}

3398
				if (physical >= physical_end) {
L
Liu Bo 已提交
3399 3400 3401 3402
					stop_loop = 1;
					break;
				}
			}
A
Arne Jansen 已提交
3403 3404 3405
next:
			path->slots[0]++;
		}
C
Chris Mason 已提交
3406
		btrfs_release_path(path);
3407
skip:
A
Arne Jansen 已提交
3408 3409
		logical += increment;
		physical += map->stripe_len;
3410
		spin_lock(&sctx->stat_lock);
L
Liu Bo 已提交
3411 3412 3413 3414 3415
		if (stop_loop)
			sctx->stat.last_physical = map->stripes[num].physical +
						   length;
		else
			sctx->stat.last_physical = physical;
3416
		spin_unlock(&sctx->stat_lock);
L
Liu Bo 已提交
3417 3418
		if (stop_loop)
			break;
A
Arne Jansen 已提交
3419
	}
3420
out:
A
Arne Jansen 已提交
3421
	/* push queued extents */
3422
	scrub_submit(sctx);
3423 3424 3425
	mutex_lock(&sctx->wr_ctx.wr_lock);
	scrub_wr_submit(sctx);
	mutex_unlock(&sctx->wr_ctx.wr_lock);
A
Arne Jansen 已提交
3426

3427
	blk_finish_plug(&plug);
A
Arne Jansen 已提交
3428
	btrfs_free_path(path);
3429
	btrfs_free_path(ppath);
A
Arne Jansen 已提交
3430 3431 3432
	return ret < 0 ? ret : 0;
}

3433
static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx,
3434 3435
					  struct btrfs_device *scrub_dev,
					  u64 chunk_offset, u64 length,
3436
					  u64 dev_offset, int is_dev_replace)
A
Arne Jansen 已提交
3437 3438
{
	struct btrfs_mapping_tree *map_tree =
3439
		&sctx->dev_root->fs_info->mapping_tree;
A
Arne Jansen 已提交
3440 3441 3442
	struct map_lookup *map;
	struct extent_map *em;
	int i;
3443
	int ret = 0;
A
Arne Jansen 已提交
3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459

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

	if (!em)
		return -EINVAL;

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

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

	for (i = 0; i < map->num_stripes; ++i) {
3460
		if (map->stripes[i].dev->bdev == scrub_dev->bdev &&
3461
		    map->stripes[i].physical == dev_offset) {
3462
			ret = scrub_stripe(sctx, map, scrub_dev, i,
3463 3464
					   chunk_offset, length,
					   is_dev_replace);
A
Arne Jansen 已提交
3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475
			if (ret)
				goto out;
		}
	}
out:
	free_extent_map(em);

	return ret;
}

static noinline_for_stack
3476
int scrub_enumerate_chunks(struct scrub_ctx *sctx,
3477 3478
			   struct btrfs_device *scrub_dev, u64 start, u64 end,
			   int is_dev_replace)
A
Arne Jansen 已提交
3479 3480 3481
{
	struct btrfs_dev_extent *dev_extent = NULL;
	struct btrfs_path *path;
3482
	struct btrfs_root *root = sctx->dev_root;
A
Arne Jansen 已提交
3483 3484 3485
	struct btrfs_fs_info *fs_info = root->fs_info;
	u64 length;
	u64 chunk_offset;
3486
	int ret = 0;
A
Arne Jansen 已提交
3487 3488 3489 3490 3491
	int slot;
	struct extent_buffer *l;
	struct btrfs_key key;
	struct btrfs_key found_key;
	struct btrfs_block_group_cache *cache;
3492
	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
A
Arne Jansen 已提交
3493 3494 3495 3496 3497 3498 3499 3500 3501

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

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

3502
	key.objectid = scrub_dev->devid;
A
Arne Jansen 已提交
3503 3504 3505 3506 3507 3508
	key.offset = 0ull;
	key.type = BTRFS_DEV_EXTENT_KEY;

	while (1) {
		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
		if (ret < 0)
3509 3510 3511 3512 3513
			break;
		if (ret > 0) {
			if (path->slots[0] >=
			    btrfs_header_nritems(path->nodes[0])) {
				ret = btrfs_next_leaf(root, path);
3514 3515 3516 3517
				if (ret < 0)
					break;
				if (ret > 0) {
					ret = 0;
3518
					break;
3519 3520 3521
				}
			} else {
				ret = 0;
3522 3523
			}
		}
A
Arne Jansen 已提交
3524 3525 3526 3527 3528 3529

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

		btrfs_item_key_to_cpu(l, &found_key, slot);

3530
		if (found_key.objectid != scrub_dev->devid)
A
Arne Jansen 已提交
3531 3532
			break;

3533
		if (found_key.type != BTRFS_DEV_EXTENT_KEY)
A
Arne Jansen 已提交
3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544
			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);

3545 3546
		if (found_key.offset + length <= start)
			goto skip;
A
Arne Jansen 已提交
3547 3548 3549 3550 3551 3552 3553 3554

		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);
3555 3556 3557 3558 3559 3560

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

3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576
		/*
		 * 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);
		ret = btrfs_inc_block_group_ro(root, cache);
		scrub_pause_off(fs_info);
		if (ret) {
			btrfs_put_block_group(cache);
			break;
		}

3577 3578 3579
		dev_replace->cursor_right = found_key.offset + length;
		dev_replace->cursor_left = found_key.offset;
		dev_replace->item_needs_writeback = 1;
3580 3581
		ret = scrub_chunk(sctx, scrub_dev, chunk_offset, length,
				  found_key.offset, is_dev_replace);
3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600

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

		wait_event(sctx->list_wait,
			   atomic_read(&sctx->bios_in_flight) == 0);
3601 3602

		scrub_pause_on(fs_info);
3603 3604 3605 3606 3607 3608

		/*
		 * must be called before we decrease @scrub_paused.
		 * make sure we don't block transaction commit while
		 * we are waiting pending workers finished.
		 */
3609 3610
		wait_event(sctx->list_wait,
			   atomic_read(&sctx->workers_pending) == 0);
3611 3612
		atomic_set(&sctx->wr_ctx.flush_all_writes, 0);

3613
		scrub_pause_off(fs_info);
3614

3615
		btrfs_dec_block_group_ro(root, cache);
3616

A
Arne Jansen 已提交
3617 3618 3619
		btrfs_put_block_group(cache);
		if (ret)
			break;
3620 3621
		if (is_dev_replace &&
		    atomic64_read(&dev_replace->num_write_errors) > 0) {
3622 3623 3624 3625 3626 3627 3628
			ret = -EIO;
			break;
		}
		if (sctx->stat.malloc_errors > 0) {
			ret = -ENOMEM;
			break;
		}
A
Arne Jansen 已提交
3629

3630 3631
		dev_replace->cursor_left = dev_replace->cursor_right;
		dev_replace->item_needs_writeback = 1;
3632
skip:
A
Arne Jansen 已提交
3633
		key.offset = found_key.offset + length;
C
Chris Mason 已提交
3634
		btrfs_release_path(path);
A
Arne Jansen 已提交
3635 3636 3637
	}

	btrfs_free_path(path);
3638

3639
	return ret;
A
Arne Jansen 已提交
3640 3641
}

3642 3643
static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx,
					   struct btrfs_device *scrub_dev)
A
Arne Jansen 已提交
3644 3645 3646 3647 3648
{
	int	i;
	u64	bytenr;
	u64	gen;
	int	ret;
3649
	struct btrfs_root *root = sctx->dev_root;
A
Arne Jansen 已提交
3650

3651
	if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
3652 3653
		return -EIO;

3654 3655 3656 3657 3658
	/* Seed devices of a new filesystem has their own generation. */
	if (scrub_dev->fs_devices != root->fs_info->fs_devices)
		gen = scrub_dev->generation;
	else
		gen = root->fs_info->last_trans_committed;
A
Arne Jansen 已提交
3659 3660 3661

	for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
		bytenr = btrfs_sb_offset(i);
3662 3663
		if (bytenr + BTRFS_SUPER_INFO_SIZE >
		    scrub_dev->commit_total_bytes)
A
Arne Jansen 已提交
3664 3665
			break;

3666
		ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr,
3667
				  scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i,
3668
				  NULL, 1, bytenr);
A
Arne Jansen 已提交
3669 3670 3671
		if (ret)
			return ret;
	}
3672
	wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
A
Arne Jansen 已提交
3673 3674 3675 3676 3677 3678 3679

	return 0;
}

/*
 * get a reference count on fs_info->scrub_workers. start worker if necessary
 */
3680 3681
static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info,
						int is_dev_replace)
A
Arne Jansen 已提交
3682
{
3683
	unsigned int flags = WQ_FREEZABLE | WQ_UNBOUND;
3684
	int max_active = fs_info->thread_pool_size;
A
Arne Jansen 已提交
3685

A
Arne Jansen 已提交
3686
	if (fs_info->scrub_workers_refcnt == 0) {
3687
		if (is_dev_replace)
3688 3689 3690
			fs_info->scrub_workers =
				btrfs_alloc_workqueue("btrfs-scrub", flags,
						      1, 4);
3691
		else
3692 3693 3694
			fs_info->scrub_workers =
				btrfs_alloc_workqueue("btrfs-scrub", flags,
						      max_active, 4);
3695 3696 3697
		if (!fs_info->scrub_workers)
			goto fail_scrub_workers;

3698 3699 3700
		fs_info->scrub_wr_completion_workers =
			btrfs_alloc_workqueue("btrfs-scrubwrc", flags,
					      max_active, 2);
3701 3702 3703
		if (!fs_info->scrub_wr_completion_workers)
			goto fail_scrub_wr_completion_workers;

3704 3705
		fs_info->scrub_nocow_workers =
			btrfs_alloc_workqueue("btrfs-scrubnc", flags, 1, 0);
3706 3707
		if (!fs_info->scrub_nocow_workers)
			goto fail_scrub_nocow_workers;
3708 3709 3710
		fs_info->scrub_parity_workers =
			btrfs_alloc_workqueue("btrfs-scrubparity", flags,
					      max_active, 2);
3711 3712
		if (!fs_info->scrub_parity_workers)
			goto fail_scrub_parity_workers;
A
Arne Jansen 已提交
3713
	}
A
Arne Jansen 已提交
3714
	++fs_info->scrub_workers_refcnt;
3715 3716 3717 3718 3719 3720 3721 3722 3723 3724
	return 0;

fail_scrub_parity_workers:
	btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
fail_scrub_nocow_workers:
	btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
fail_scrub_wr_completion_workers:
	btrfs_destroy_workqueue(fs_info->scrub_workers);
fail_scrub_workers:
	return -ENOMEM;
A
Arne Jansen 已提交
3725 3726
}

3727
static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info)
A
Arne Jansen 已提交
3728
{
3729
	if (--fs_info->scrub_workers_refcnt == 0) {
3730 3731 3732
		btrfs_destroy_workqueue(fs_info->scrub_workers);
		btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
		btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
3733
		btrfs_destroy_workqueue(fs_info->scrub_parity_workers);
3734
	}
A
Arne Jansen 已提交
3735 3736 3737
	WARN_ON(fs_info->scrub_workers_refcnt < 0);
}

3738 3739
int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start,
		    u64 end, struct btrfs_scrub_progress *progress,
3740
		    int readonly, int is_dev_replace)
A
Arne Jansen 已提交
3741
{
3742
	struct scrub_ctx *sctx;
A
Arne Jansen 已提交
3743 3744
	int ret;
	struct btrfs_device *dev;
3745
	struct rcu_string *name;
A
Arne Jansen 已提交
3746

3747
	if (btrfs_fs_closing(fs_info))
A
Arne Jansen 已提交
3748 3749
		return -EINVAL;

3750
	if (fs_info->chunk_root->nodesize > BTRFS_STRIPE_LEN) {
3751 3752 3753 3754 3755
		/*
		 * 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.
		 */
3756 3757
		btrfs_err(fs_info,
			   "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails",
3758
		       fs_info->chunk_root->nodesize, BTRFS_STRIPE_LEN);
3759 3760 3761
		return -EINVAL;
	}

3762
	if (fs_info->chunk_root->sectorsize != PAGE_SIZE) {
3763
		/* not supported for data w/o checksums */
3764 3765 3766
		btrfs_err(fs_info,
			   "scrub: size assumption sectorsize != PAGE_SIZE "
			   "(%d != %lu) fails",
3767
		       fs_info->chunk_root->sectorsize, PAGE_SIZE);
A
Arne Jansen 已提交
3768 3769 3770
		return -EINVAL;
	}

3771 3772 3773 3774 3775 3776 3777 3778
	if (fs_info->chunk_root->nodesize >
	    PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK ||
	    fs_info->chunk_root->sectorsize >
	    PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) {
		/*
		 * would exhaust the array bounds of pagev member in
		 * struct scrub_block
		 */
3779 3780
		btrfs_err(fs_info, "scrub: size assumption nodesize and sectorsize "
			   "<= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails",
3781 3782 3783 3784 3785 3786 3787
		       fs_info->chunk_root->nodesize,
		       SCRUB_MAX_PAGES_PER_BLOCK,
		       fs_info->chunk_root->sectorsize,
		       SCRUB_MAX_PAGES_PER_BLOCK);
		return -EINVAL;
	}

A
Arne Jansen 已提交
3788

3789 3790
	mutex_lock(&fs_info->fs_devices->device_list_mutex);
	dev = btrfs_find_device(fs_info, devid, NULL, NULL);
3791
	if (!dev || (dev->missing && !is_dev_replace)) {
3792
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
A
Arne Jansen 已提交
3793 3794 3795
		return -ENODEV;
	}

3796 3797 3798 3799 3800 3801 3802 3803 3804 3805
	if (!is_dev_replace && !readonly && !dev->writeable) {
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
		rcu_read_lock();
		name = rcu_dereference(dev->name);
		btrfs_err(fs_info, "scrub: device %s is not writable",
			  name->str);
		rcu_read_unlock();
		return -EROFS;
	}

3806
	mutex_lock(&fs_info->scrub_lock);
3807
	if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) {
A
Arne Jansen 已提交
3808
		mutex_unlock(&fs_info->scrub_lock);
3809 3810
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
		return -EIO;
A
Arne Jansen 已提交
3811 3812
	}

3813 3814 3815 3816 3817
	btrfs_dev_replace_lock(&fs_info->dev_replace);
	if (dev->scrub_device ||
	    (!is_dev_replace &&
	     btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) {
		btrfs_dev_replace_unlock(&fs_info->dev_replace);
A
Arne Jansen 已提交
3818
		mutex_unlock(&fs_info->scrub_lock);
3819
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
A
Arne Jansen 已提交
3820 3821
		return -EINPROGRESS;
	}
3822
	btrfs_dev_replace_unlock(&fs_info->dev_replace);
3823 3824 3825 3826 3827 3828 3829 3830

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

3831
	sctx = scrub_setup_ctx(dev, is_dev_replace);
3832
	if (IS_ERR(sctx)) {
A
Arne Jansen 已提交
3833
		mutex_unlock(&fs_info->scrub_lock);
3834 3835
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
		scrub_workers_put(fs_info);
3836
		return PTR_ERR(sctx);
A
Arne Jansen 已提交
3837
	}
3838 3839
	sctx->readonly = readonly;
	dev->scrub_device = sctx;
3840
	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
A
Arne Jansen 已提交
3841

3842 3843 3844 3845
	/*
	 * checking @scrub_pause_req here, we can avoid
	 * race between committing transaction and scrubbing.
	 */
3846
	__scrub_blocked_if_needed(fs_info);
A
Arne Jansen 已提交
3847 3848 3849
	atomic_inc(&fs_info->scrubs_running);
	mutex_unlock(&fs_info->scrub_lock);

3850
	if (!is_dev_replace) {
3851 3852 3853 3854
		/*
		 * by holding device list mutex, we can
		 * kick off writing super in log tree sync.
		 */
3855
		mutex_lock(&fs_info->fs_devices->device_list_mutex);
3856
		ret = scrub_supers(sctx, dev);
3857
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3858
	}
A
Arne Jansen 已提交
3859 3860

	if (!ret)
3861 3862
		ret = scrub_enumerate_chunks(sctx, dev, start, end,
					     is_dev_replace);
A
Arne Jansen 已提交
3863

3864
	wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
A
Arne Jansen 已提交
3865 3866 3867
	atomic_dec(&fs_info->scrubs_running);
	wake_up(&fs_info->scrub_pause_wait);

3868
	wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0);
3869

A
Arne Jansen 已提交
3870
	if (progress)
3871
		memcpy(progress, &sctx->stat, sizeof(*progress));
A
Arne Jansen 已提交
3872 3873 3874

	mutex_lock(&fs_info->scrub_lock);
	dev->scrub_device = NULL;
3875
	scrub_workers_put(fs_info);
A
Arne Jansen 已提交
3876 3877
	mutex_unlock(&fs_info->scrub_lock);

3878
	scrub_put_ctx(sctx);
A
Arne Jansen 已提交
3879 3880 3881 3882

	return ret;
}

3883
void btrfs_scrub_pause(struct btrfs_root *root)
A
Arne Jansen 已提交
3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899
{
	struct btrfs_fs_info *fs_info = root->fs_info;

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

3900
void btrfs_scrub_continue(struct btrfs_root *root)
A
Arne Jansen 已提交
3901 3902 3903 3904 3905 3906 3907
{
	struct btrfs_fs_info *fs_info = root->fs_info;

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

3908
int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info)
A
Arne Jansen 已提交
3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928
{
	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;
}

3929 3930
int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info,
			   struct btrfs_device *dev)
3931
{
3932
	struct scrub_ctx *sctx;
A
Arne Jansen 已提交
3933 3934

	mutex_lock(&fs_info->scrub_lock);
3935 3936
	sctx = dev->scrub_device;
	if (!sctx) {
A
Arne Jansen 已提交
3937 3938 3939
		mutex_unlock(&fs_info->scrub_lock);
		return -ENOTCONN;
	}
3940
	atomic_inc(&sctx->cancel_req);
A
Arne Jansen 已提交
3941 3942 3943 3944 3945 3946 3947 3948 3949 3950
	while (dev->scrub_device) {
		mutex_unlock(&fs_info->scrub_lock);
		wait_event(fs_info->scrub_pause_wait,
			   dev->scrub_device == NULL);
		mutex_lock(&fs_info->scrub_lock);
	}
	mutex_unlock(&fs_info->scrub_lock);

	return 0;
}
S
Stefan Behrens 已提交
3951

A
Arne Jansen 已提交
3952 3953 3954 3955
int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
			 struct btrfs_scrub_progress *progress)
{
	struct btrfs_device *dev;
3956
	struct scrub_ctx *sctx = NULL;
A
Arne Jansen 已提交
3957 3958

	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3959
	dev = btrfs_find_device(root->fs_info, devid, NULL, NULL);
A
Arne Jansen 已提交
3960
	if (dev)
3961 3962 3963
		sctx = dev->scrub_device;
	if (sctx)
		memcpy(progress, &sctx->stat, sizeof(*progress));
A
Arne Jansen 已提交
3964 3965
	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);

3966
	return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV;
A
Arne Jansen 已提交
3967
}
3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983

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

	mapped_length = extent_len;
	ret = btrfs_map_block(fs_info, READ, extent_logical,
			      &mapped_length, &bbio, 0);
	if (ret || !bbio || mapped_length < extent_len ||
	    !bbio->stripes[0].dev->bdev) {
3984
		btrfs_put_bbio(bbio);
3985 3986 3987 3988 3989 3990
		return;
	}

	*extent_physical = bbio->stripes[0].physical;
	*extent_mirror_num = bbio->mirror_num;
	*extent_dev = bbio->stripes[0].dev;
3991
	btrfs_put_bbio(bbio);
3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007
}

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

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

	WARN_ON(!dev->bdev);
4008
	wr_ctx->pages_per_wr_bio = SCRUB_PAGES_PER_WR_BIO;
4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038 4039 4040 4041 4042
	wr_ctx->tgtdev = dev;
	atomic_set(&wr_ctx->flush_all_writes, 0);
	return 0;
}

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

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

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

	scrub_pending_trans_workers_inc(sctx);

	nocow_ctx->sctx = sctx;
	nocow_ctx->logical = logical;
	nocow_ctx->len = len;
	nocow_ctx->mirror_num = mirror_num;
	nocow_ctx->physical_for_dev_replace = physical_for_dev_replace;
4043 4044
	btrfs_init_work(&nocow_ctx->work, btrfs_scrubnc_helper,
			copy_nocow_pages_worker, NULL, NULL);
4045
	INIT_LIST_HEAD(&nocow_ctx->inodes);
4046 4047
	btrfs_queue_work(fs_info->scrub_nocow_workers,
			 &nocow_ctx->work);
4048 4049 4050 4051

	return 0;
}

4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068
static int record_inode_for_nocow(u64 inum, u64 offset, u64 root, void *ctx)
{
	struct scrub_copy_nocow_ctx *nocow_ctx = ctx;
	struct scrub_nocow_inode *nocow_inode;

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

#define COPY_COMPLETE 1

4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103
static void copy_nocow_pages_worker(struct btrfs_work *work)
{
	struct scrub_copy_nocow_ctx *nocow_ctx =
		container_of(work, struct scrub_copy_nocow_ctx, work);
	struct scrub_ctx *sctx = nocow_ctx->sctx;
	u64 logical = nocow_ctx->logical;
	u64 len = nocow_ctx->len;
	int mirror_num = nocow_ctx->mirror_num;
	u64 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
	int ret;
	struct btrfs_trans_handle *trans = NULL;
	struct btrfs_fs_info *fs_info;
	struct btrfs_path *path;
	struct btrfs_root *root;
	int not_written = 0;

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

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

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

	ret = iterate_inodes_from_logical(logical, fs_info, path,
4104
					  record_inode_for_nocow, nocow_ctx);
4105
	if (ret != 0 && ret != -ENOENT) {
4106 4107
		btrfs_warn(fs_info, "iterate_inodes_from_logical() failed: log %llu, "
			"phys %llu, len %llu, mir %u, ret %d",
4108 4109
			logical, physical_for_dev_replace, len, mirror_num,
			ret);
4110 4111 4112 4113
		not_written = 1;
		goto out;
	}

4114 4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130 4131
	btrfs_end_transaction(trans, root);
	trans = NULL;
	while (!list_empty(&nocow_ctx->inodes)) {
		struct scrub_nocow_inode *entry;
		entry = list_first_entry(&nocow_ctx->inodes,
					 struct scrub_nocow_inode,
					 list);
		list_del_init(&entry->list);
		ret = copy_nocow_pages_for_inode(entry->inum, entry->offset,
						 entry->root, nocow_ctx);
		kfree(entry);
		if (ret == COPY_COMPLETE) {
			ret = 0;
			break;
		} else if (ret) {
			break;
		}
	}
4132
out:
4133 4134 4135 4136 4137 4138 4139 4140
	while (!list_empty(&nocow_ctx->inodes)) {
		struct scrub_nocow_inode *entry;
		entry = list_first_entry(&nocow_ctx->inodes,
					 struct scrub_nocow_inode,
					 list);
		list_del_init(&entry->list);
		kfree(entry);
	}
4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152
	if (trans && !IS_ERR(trans))
		btrfs_end_transaction(trans, root);
	if (not_written)
		btrfs_dev_replace_stats_inc(&fs_info->dev_replace.
					    num_uncorrectable_read_errors);

	btrfs_free_path(path);
	kfree(nocow_ctx);

	scrub_pending_trans_workers_dec(sctx);
}

4153 4154 4155 4156 4157 4158 4159 4160 4161 4162 4163 4164 4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191 4192 4193 4194 4195 4196
static int check_extent_to_block(struct inode *inode, u64 start, u64 len,
				 u64 logical)
{
	struct extent_state *cached_state = NULL;
	struct btrfs_ordered_extent *ordered;
	struct extent_io_tree *io_tree;
	struct extent_map *em;
	u64 lockstart = start, lockend = start + len - 1;
	int ret = 0;

	io_tree = &BTRFS_I(inode)->io_tree;

	lock_extent_bits(io_tree, lockstart, lockend, 0, &cached_state);
	ordered = btrfs_lookup_ordered_range(inode, lockstart, len);
	if (ordered) {
		btrfs_put_ordered_extent(ordered);
		ret = 1;
		goto out_unlock;
	}

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

	/*
	 * This extent does not actually cover the logical extent anymore,
	 * move on to the next inode.
	 */
	if (em->block_start > logical ||
	    em->block_start + em->block_len < logical + len) {
		free_extent_map(em);
		ret = 1;
		goto out_unlock;
	}
	free_extent_map(em);

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

4197 4198
static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
				      struct scrub_copy_nocow_ctx *nocow_ctx)
4199
{
4200
	struct btrfs_fs_info *fs_info = nocow_ctx->sctx->dev_root->fs_info;
4201
	struct btrfs_key key;
4202 4203
	struct inode *inode;
	struct page *page;
4204
	struct btrfs_root *local_root;
4205
	struct extent_io_tree *io_tree;
4206
	u64 physical_for_dev_replace;
4207
	u64 nocow_ctx_logical;
4208
	u64 len = nocow_ctx->len;
4209
	unsigned long index;
4210
	int srcu_index;
4211 4212
	int ret = 0;
	int err = 0;
4213 4214 4215 4216

	key.objectid = root;
	key.type = BTRFS_ROOT_ITEM_KEY;
	key.offset = (u64)-1;
4217 4218 4219

	srcu_index = srcu_read_lock(&fs_info->subvol_srcu);

4220
	local_root = btrfs_read_fs_root_no_name(fs_info, &key);
4221 4222
	if (IS_ERR(local_root)) {
		srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
4223
		return PTR_ERR(local_root);
4224
	}
4225 4226 4227 4228 4229

	key.type = BTRFS_INODE_ITEM_KEY;
	key.objectid = inum;
	key.offset = 0;
	inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
4230
	srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
4231 4232 4233
	if (IS_ERR(inode))
		return PTR_ERR(inode);

4234 4235 4236 4237
	/* Avoid truncate/dio/punch hole.. */
	mutex_lock(&inode->i_mutex);
	inode_dio_wait(inode);

4238
	physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
4239
	io_tree = &BTRFS_I(inode)->io_tree;
4240
	nocow_ctx_logical = nocow_ctx->logical;
4241

4242 4243 4244 4245
	ret = check_extent_to_block(inode, offset, len, nocow_ctx_logical);
	if (ret) {
		ret = ret > 0 ? 0 : ret;
		goto out;
4246 4247
	}

4248 4249
	while (len >= PAGE_CACHE_SIZE) {
		index = offset >> PAGE_CACHE_SHIFT;
4250
again:
4251 4252
		page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
		if (!page) {
4253
			btrfs_err(fs_info, "find_or_create_page() failed");
4254
			ret = -ENOMEM;
4255
			goto out;
4256 4257 4258 4259 4260 4261 4262
		}

		if (PageUptodate(page)) {
			if (PageDirty(page))
				goto next_page;
		} else {
			ClearPageError(page);
4263
			err = extent_read_full_page(io_tree, page,
4264 4265
							   btrfs_get_extent,
							   nocow_ctx->mirror_num);
4266 4267
			if (err) {
				ret = err;
4268 4269
				goto next_page;
			}
4270

4271
			lock_page(page);
4272 4273 4274 4275 4276 4277 4278
			/*
			 * If the page has been remove from the page cache,
			 * the data on it is meaningless, because it may be
			 * old one, the new data may be written into the new
			 * page in the page cache.
			 */
			if (page->mapping != inode->i_mapping) {
4279
				unlock_page(page);
4280 4281 4282
				page_cache_release(page);
				goto again;
			}
4283 4284 4285 4286 4287
			if (!PageUptodate(page)) {
				ret = -EIO;
				goto next_page;
			}
		}
4288 4289 4290 4291 4292 4293 4294 4295

		ret = check_extent_to_block(inode, offset, len,
					    nocow_ctx_logical);
		if (ret) {
			ret = ret > 0 ? 0 : ret;
			goto next_page;
		}

4296 4297 4298 4299
		err = write_page_nocow(nocow_ctx->sctx,
				       physical_for_dev_replace, page);
		if (err)
			ret = err;
4300
next_page:
4301 4302 4303 4304 4305 4306
		unlock_page(page);
		page_cache_release(page);

		if (ret)
			break;

4307 4308
		offset += PAGE_CACHE_SIZE;
		physical_for_dev_replace += PAGE_CACHE_SIZE;
4309
		nocow_ctx_logical += PAGE_CACHE_SIZE;
4310 4311
		len -= PAGE_CACHE_SIZE;
	}
4312
	ret = COPY_COMPLETE;
4313
out:
4314
	mutex_unlock(&inode->i_mutex);
4315
	iput(inode);
4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326 4327 4328 4329
	return ret;
}

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

	dev = sctx->wr_ctx.tgtdev;
	if (!dev)
		return -EIO;
	if (!dev->bdev) {
4330 4331
		btrfs_warn_rl(dev->dev_root->fs_info,
			"scrub write_page_nocow(bdev == NULL) is unexpected");
4332 4333
		return -EIO;
	}
4334
	bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
4335 4336 4337 4338 4339 4340
	if (!bio) {
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		spin_unlock(&sctx->stat_lock);
		return -ENOMEM;
	}
4341 4342
	bio->bi_iter.bi_size = 0;
	bio->bi_iter.bi_sector = physical_for_dev_replace >> 9;
4343 4344 4345 4346 4347 4348 4349 4350 4351
	bio->bi_bdev = dev->bdev;
	ret = bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
	if (ret != PAGE_CACHE_SIZE) {
leave_with_eio:
		bio_put(bio);
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
		return -EIO;
	}

4352
	if (btrfsic_submit_bio_wait(WRITE_SYNC, bio))
4353 4354 4355 4356 4357
		goto leave_with_eio;

	bio_put(bio);
	return 0;
}