scrub.c 111.1 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;
127 128 129
	};
};

130 131 132 133 134 135 136 137 138 139 140 141 142 143
/* Used for the chunks with parity stripe such RAID5/6 */
struct scrub_parity {
	struct scrub_ctx	*sctx;

	struct btrfs_device	*scrub_dev;

	u64			logic_start;

	u64			logic_end;

	int			nsectors;

	int			stripe_len;

144
	atomic_t		refs;
145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162

	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];
};

163 164 165 166 167 168 169 170
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;
};

171
struct scrub_ctx {
172
	struct scrub_bio	*bios[SCRUB_BIOS_PER_SCTX];
173
	struct btrfs_root	*dev_root;
A
Arne Jansen 已提交
174 175
	int			first_free;
	int			curr;
176 177
	atomic_t		bios_in_flight;
	atomic_t		workers_pending;
A
Arne Jansen 已提交
178 179 180 181 182
	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
183
	int			readonly;
184
	int			pages_per_rd_bio;
185 186
	u32			sectorsize;
	u32			nodesize;
187 188

	int			is_dev_replace;
189
	struct scrub_wr_ctx	wr_ctx;
190

A
Arne Jansen 已提交
191 192 193 194 195
	/*
	 * statistics
	 */
	struct btrfs_scrub_progress stat;
	spinlock_t		stat_lock;
196 197 198 199 200 201 202 203 204

	/*
	 * 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
205 206
};

207
struct scrub_fixup_nodatasum {
208
	struct scrub_ctx	*sctx;
209
	struct btrfs_device	*dev;
210 211 212 213 214 215
	u64			logical;
	struct btrfs_root	*root;
	struct btrfs_work	work;
	int			mirror_num;
};

216 217 218 219 220 221 222
struct scrub_nocow_inode {
	u64			inum;
	u64			offset;
	u64			root;
	struct list_head	list;
};

223 224 225 226 227 228
struct scrub_copy_nocow_ctx {
	struct scrub_ctx	*sctx;
	u64			logical;
	u64			len;
	int			mirror_num;
	u64			physical_for_dev_replace;
229
	struct list_head	inodes;
230 231 232
	struct btrfs_work	work;
};

233 234 235 236 237 238 239 240 241
struct scrub_warning {
	struct btrfs_path	*path;
	u64			extent_item_size;
	const char		*errstr;
	sector_t		sector;
	u64			logical;
	struct btrfs_device	*dev;
};

242 243 244 245
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);
246
static int scrub_handle_errored_block(struct scrub_block *sblock_to_check);
247
static int scrub_setup_recheck_block(struct scrub_block *original_sblock,
248
				     struct scrub_block *sblocks_for_recheck);
249 250 251
static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
				struct scrub_block *sblock, int is_metadata,
				int have_csum, u8 *csum, u64 generation,
252
				u16 csum_size, int retry_failed_mirror);
253 254 255 256 257 258
static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
					 struct scrub_block *sblock,
					 int is_metadata, int have_csum,
					 const u8 *csum, u64 generation,
					 u16 csum_size);
static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
259
					     struct scrub_block *sblock_good);
260 261 262
static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
					    struct scrub_block *sblock_good,
					    int page_num, int force_write);
263 264 265
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);
266 267 268 269 270
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);
271 272
static void scrub_page_get(struct scrub_page *spage);
static void scrub_page_put(struct scrub_page *spage);
273 274
static void scrub_parity_get(struct scrub_parity *sparity);
static void scrub_parity_put(struct scrub_parity *sparity);
275 276
static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
				    struct scrub_page *spage);
277
static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
278
		       u64 physical, struct btrfs_device *dev, u64 flags,
279 280
		       u64 gen, int mirror_num, u8 *csum, int force,
		       u64 physical_for_dev_replace);
S
Stefan Behrens 已提交
281
static void scrub_bio_end_io(struct bio *bio, int err);
282 283
static void scrub_bio_end_io_worker(struct btrfs_work *work);
static void scrub_block_complete(struct scrub_block *sblock);
284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302
static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
			       u64 extent_logical, u64 extent_len,
			       u64 *extent_physical,
			       struct btrfs_device **extent_dev,
			       int *extent_mirror_num);
static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
			      struct scrub_wr_ctx *wr_ctx,
			      struct btrfs_fs_info *fs_info,
			      struct btrfs_device *dev,
			      int is_dev_replace);
static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx);
static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
				    struct scrub_page *spage);
static void scrub_wr_submit(struct scrub_ctx *sctx);
static void scrub_wr_bio_end_io(struct bio *bio, int err);
static void scrub_wr_bio_end_io_worker(struct btrfs_work *work);
static int write_page_nocow(struct scrub_ctx *sctx,
			    u64 physical_for_dev_replace, struct page *page);
static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
303
				      struct scrub_copy_nocow_ctx *ctx);
304 305 306
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);
307
static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
308
static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
309
static void scrub_put_ctx(struct scrub_ctx *sctx);
S
Stefan Behrens 已提交
310 311


312 313
static void scrub_pending_bio_inc(struct scrub_ctx *sctx)
{
314
	atomic_inc(&sctx->refs);
315 316 317 318 319 320 321
	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);
322
	scrub_put_ctx(sctx);
323 324
}

325
static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
326 327 328 329 330 331 332 333 334
{
	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);
	}
}

335 336 337 338 339 340 341 342 343 344 345 346 347
static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
{
	atomic_inc(&fs_info->scrubs_paused);
	wake_up(&fs_info->scrub_pause_wait);

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

	wake_up(&fs_info->scrub_pause_wait);
}

348 349 350 351 352 353 354 355
/*
 * 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;

356
	atomic_inc(&sctx->refs);
357 358 359 360 361 362 363 364 365 366 367 368 369
	/*
	 * 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);
370 371 372 373 374 375 376 377 378 379

	/*
	 * 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);

380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398
	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);
399
	scrub_put_ctx(sctx);
400 401
}

402
static void scrub_free_csums(struct scrub_ctx *sctx)
A
Arne Jansen 已提交
403
{
404
	while (!list_empty(&sctx->csum_list)) {
A
Arne Jansen 已提交
405
		struct btrfs_ordered_sum *sum;
406
		sum = list_first_entry(&sctx->csum_list,
A
Arne Jansen 已提交
407 408 409 410 411 412
				       struct btrfs_ordered_sum, list);
		list_del(&sum->list);
		kfree(sum);
	}
}

413
static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx)
A
Arne Jansen 已提交
414 415 416
{
	int i;

417
	if (!sctx)
A
Arne Jansen 已提交
418 419
		return;

420 421
	scrub_free_wr_ctx(&sctx->wr_ctx);

422
	/* this can happen when scrub is cancelled */
423 424
	if (sctx->curr != -1) {
		struct scrub_bio *sbio = sctx->bios[sctx->curr];
425 426

		for (i = 0; i < sbio->page_count; i++) {
427
			WARN_ON(!sbio->pagev[i]->page);
428 429 430 431 432
			scrub_block_put(sbio->pagev[i]->sblock);
		}
		bio_put(sbio->bio);
	}

433
	for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
434
		struct scrub_bio *sbio = sctx->bios[i];
A
Arne Jansen 已提交
435 436 437 438 439 440

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

441 442
	scrub_free_csums(sctx);
	kfree(sctx);
A
Arne Jansen 已提交
443 444
}

445 446 447 448 449 450
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
451
static noinline_for_stack
452
struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace)
A
Arne Jansen 已提交
453
{
454
	struct scrub_ctx *sctx;
A
Arne Jansen 已提交
455 456
	int		i;
	struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
457 458
	int pages_per_rd_bio;
	int ret;
A
Arne Jansen 已提交
459

460 461 462 463 464 465 466 467 468 469 470 471
	/*
	 * the setting of pages_per_rd_bio is correct for scrub but might
	 * be wrong for the dev_replace code where we might read from
	 * different devices in the initial huge bios. However, that
	 * code is able to correctly handle the case when adding a page
	 * to a bio fails.
	 */
	if (dev->bdev)
		pages_per_rd_bio = min_t(int, SCRUB_PAGES_PER_RD_BIO,
					 bio_get_nr_vecs(dev->bdev));
	else
		pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO;
472 473
	sctx = kzalloc(sizeof(*sctx), GFP_NOFS);
	if (!sctx)
A
Arne Jansen 已提交
474
		goto nomem;
475
	atomic_set(&sctx->refs, 1);
476
	sctx->is_dev_replace = is_dev_replace;
477
	sctx->pages_per_rd_bio = pages_per_rd_bio;
478
	sctx->curr = -1;
479
	sctx->dev_root = dev->dev_root;
480
	for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
A
Arne Jansen 已提交
481 482 483 484 485
		struct scrub_bio *sbio;

		sbio = kzalloc(sizeof(*sbio), GFP_NOFS);
		if (!sbio)
			goto nomem;
486
		sctx->bios[i] = sbio;
A
Arne Jansen 已提交
487 488

		sbio->index = i;
489
		sbio->sctx = sctx;
490
		sbio->page_count = 0;
491 492
		btrfs_init_work(&sbio->work, btrfs_scrub_helper,
				scrub_bio_end_io_worker, NULL, NULL);
A
Arne Jansen 已提交
493

494
		if (i != SCRUB_BIOS_PER_SCTX - 1)
495
			sctx->bios[i]->next_free = i + 1;
496
		else
497 498 499 500 501
			sctx->bios[i]->next_free = -1;
	}
	sctx->first_free = 0;
	sctx->nodesize = dev->dev_root->nodesize;
	sctx->sectorsize = dev->dev_root->sectorsize;
502 503
	atomic_set(&sctx->bios_in_flight, 0);
	atomic_set(&sctx->workers_pending, 0);
504 505 506 507 508 509 510
	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);
511 512 513 514 515 516 517

	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);
	}
518
	return sctx;
A
Arne Jansen 已提交
519 520

nomem:
521
	scrub_free_ctx(sctx);
A
Arne Jansen 已提交
522 523 524
	return ERR_PTR(-ENOMEM);
}

525 526
static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root,
				     void *warn_ctx)
527 528 529 530 531 532 533
{
	u64 isize;
	u32 nlink;
	int ret;
	int i;
	struct extent_buffer *eb;
	struct btrfs_inode_item *inode_item;
534
	struct scrub_warning *swarn = warn_ctx;
535 536 537 538
	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;
539
	struct btrfs_key key;
540 541 542 543 544 545 546 547 548 549

	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;
	}

550 551 552
	/*
	 * this makes the path point to (inum INODE_ITEM ioff)
	 */
553 554 555 556 557
	key.objectid = inum;
	key.type = BTRFS_INODE_ITEM_KEY;
	key.offset = 0;

	ret = btrfs_search_slot(NULL, local_root, &key, swarn->path, 0, 0);
558 559 560 561 562 563 564 565 566 567 568 569 570
	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);
571 572 573 574 575
	if (IS_ERR(ipath)) {
		ret = PTR_ERR(ipath);
		ipath = NULL;
		goto err;
	}
576 577 578 579 580 581 582 583 584 585
	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)
586
		printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
587 588
			"%s, sector %llu, root %llu, inode %llu, offset %llu, "
			"length %llu, links %u (path: %s)\n", swarn->errstr,
589
			swarn->logical, rcu_str_deref(swarn->dev->name),
590 591
			(unsigned long long)swarn->sector, root, inum, offset,
			min(isize - offset, (u64)PAGE_SIZE), nlink,
592
			(char *)(unsigned long)ipath->fspath->val[i]);
593 594 595 596 597

	free_ipath(ipath);
	return 0;

err:
598
	printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
599 600
		"%s, sector %llu, root %llu, inode %llu, offset %llu: path "
		"resolving failed with ret=%d\n", swarn->errstr,
601
		swarn->logical, rcu_str_deref(swarn->dev->name),
602 603 604 605 606 607
		(unsigned long long)swarn->sector, root, inum, offset, ret);

	free_ipath(ipath);
	return 0;
}

608
static void scrub_print_warning(const char *errstr, struct scrub_block *sblock)
609
{
610 611
	struct btrfs_device *dev;
	struct btrfs_fs_info *fs_info;
612 613 614 615 616
	struct btrfs_path *path;
	struct btrfs_key found_key;
	struct extent_buffer *eb;
	struct btrfs_extent_item *ei;
	struct scrub_warning swarn;
617 618 619
	unsigned long ptr = 0;
	u64 extent_item_pos;
	u64 flags = 0;
620
	u64 ref_root;
621
	u32 item_size;
622
	u8 ref_level;
623
	int ret;
624

625
	WARN_ON(sblock->page_count < 1);
626
	dev = sblock->pagev[0]->dev;
627 628
	fs_info = sblock->sctx->dev_root->fs_info;

629
	path = btrfs_alloc_path();
630 631
	if (!path)
		return;
632

633 634
	swarn.sector = (sblock->pagev[0]->physical) >> 9;
	swarn.logical = sblock->pagev[0]->logical;
635
	swarn.errstr = errstr;
636
	swarn.dev = NULL;
637

638 639
	ret = extent_from_logical(fs_info, swarn.logical, path, &found_key,
				  &flags);
640 641 642
	if (ret < 0)
		goto out;

J
Jan Schmidt 已提交
643
	extent_item_pos = swarn.logical - found_key.objectid;
644 645 646 647 648 649
	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]);

650
	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
651
		do {
652 653 654
			ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
						      item_size, &ref_root,
						      &ref_level);
655
			printk_in_rcu(KERN_WARNING
656
				"BTRFS: %s at logical %llu on dev %s, "
657
				"sector %llu: metadata %s (level %d) in tree "
658 659
				"%llu\n", errstr, swarn.logical,
				rcu_str_deref(dev->name),
660 661 662 663 664
				(unsigned long long)swarn.sector,
				ref_level ? "node" : "leaf",
				ret < 0 ? -1 : ref_level,
				ret < 0 ? -1 : ref_root);
		} while (ret != 1);
665
		btrfs_release_path(path);
666
	} else {
667
		btrfs_release_path(path);
668
		swarn.path = path;
669
		swarn.dev = dev;
670 671
		iterate_extent_inodes(fs_info, found_key.objectid,
					extent_item_pos, 1,
672 673 674 675 676 677 678
					scrub_print_warning_inode, &swarn);
	}

out:
	btrfs_free_path(path);
}

679
static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx)
680
{
681
	struct page *page = NULL;
682
	unsigned long index;
683
	struct scrub_fixup_nodatasum *fixup = fixup_ctx;
684
	int ret;
685
	int corrected = 0;
686
	struct btrfs_key key;
687
	struct inode *inode = NULL;
688
	struct btrfs_fs_info *fs_info;
689 690
	u64 end = offset + PAGE_SIZE - 1;
	struct btrfs_root *local_root;
691
	int srcu_index;
692 693 694 695

	key.objectid = root;
	key.type = BTRFS_ROOT_ITEM_KEY;
	key.offset = (u64)-1;
696 697 698 699 700 701 702

	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);
703
		return PTR_ERR(local_root);
704
	}
705 706 707 708

	key.type = BTRFS_INODE_ITEM_KEY;
	key.objectid = inum;
	key.offset = 0;
709 710
	inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
	srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
711 712 713 714 715 716
	if (IS_ERR(inode))
		return PTR_ERR(inode);

	index = offset >> PAGE_CACHE_SHIFT;

	page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742
	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;
		}
743
		ret = repair_io_failure(inode, offset, PAGE_SIZE,
744
					fixup->logical, page,
745
					offset - page_offset(page),
746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779
					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);
780 781

	iput(inode);
782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800

	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;
801
	struct scrub_ctx *sctx;
802 803 804 805 806
	struct btrfs_trans_handle *trans = NULL;
	struct btrfs_path *path;
	int uncorrectable = 0;

	fixup = container_of(work, struct scrub_fixup_nodatasum, work);
807
	sctx = fixup->sctx;
808 809 810

	path = btrfs_alloc_path();
	if (!path) {
811 812 813
		spin_lock(&sctx->stat_lock);
		++sctx->stat.malloc_errors;
		spin_unlock(&sctx->stat_lock);
814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841
		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);

842 843 844
	spin_lock(&sctx->stat_lock);
	++sctx->stat.corrected_errors;
	spin_unlock(&sctx->stat_lock);
845 846 847 848 849

out:
	if (trans && !IS_ERR(trans))
		btrfs_end_transaction(trans, fixup->root);
	if (uncorrectable) {
850 851 852
		spin_lock(&sctx->stat_lock);
		++sctx->stat.uncorrectable_errors;
		spin_unlock(&sctx->stat_lock);
853 854 855
		btrfs_dev_replace_stats_inc(
			&sctx->dev_root->fs_info->dev_replace.
			num_uncorrectable_read_errors);
856 857
		printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
		    "unable to fixup (nodatasum) error at logical %llu on dev %s\n",
858
			fixup->logical, rcu_str_deref(fixup->dev->name));
859 860 861 862 863
	}

	btrfs_free_path(path);
	kfree(fixup);

864
	scrub_pending_trans_workers_dec(sctx);
865 866
}

867 868 869 870 871 872 873 874
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)) {
875
		btrfs_put_bbio(recover->bbio);
876 877 878 879
		kfree(recover);
	}
}

A
Arne Jansen 已提交
880
/*
881 882 883 884 885 886
 * 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
887
 */
888
static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
A
Arne Jansen 已提交
889
{
890
	struct scrub_ctx *sctx = sblock_to_check->sctx;
891
	struct btrfs_device *dev;
892 893 894 895 896 897 898 899 900 901 902 903 904 905
	struct btrfs_fs_info *fs_info;
	u64 length;
	u64 logical;
	u64 generation;
	unsigned int failed_mirror_index;
	unsigned int is_metadata;
	unsigned int have_csum;
	u8 *csum;
	struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */
	struct scrub_block *sblock_bad;
	int ret;
	int mirror_index;
	int page_num;
	int success;
906
	static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
907 908 909
				      DEFAULT_RATELIMIT_BURST);

	BUG_ON(sblock_to_check->page_count < 1);
910
	fs_info = sctx->dev_root->fs_info;
911 912 913 914 915 916 917 918 919 920 921
	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;
	}
922
	length = sblock_to_check->page_count * PAGE_SIZE;
923 924 925 926 927
	logical = sblock_to_check->pagev[0]->logical;
	generation = sblock_to_check->pagev[0]->generation;
	BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1);
	failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1;
	is_metadata = !(sblock_to_check->pagev[0]->flags &
928
			BTRFS_EXTENT_FLAG_DATA);
929 930 931
	have_csum = sblock_to_check->pagev[0]->have_csum;
	csum = sblock_to_check->pagev[0]->csum;
	dev = sblock_to_check->pagev[0]->dev;
932

933 934 935 936 937
	if (sctx->is_dev_replace && !is_metadata && !have_csum) {
		sblocks_for_recheck = NULL;
		goto nodatasum_case;
	}

938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966
	/*
	 * 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.
	 */

967 968
	sblocks_for_recheck = kcalloc(BTRFS_MAX_MIRRORS,
				      sizeof(*sblocks_for_recheck), GFP_NOFS);
969
	if (!sblocks_for_recheck) {
970 971 972 973 974
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		sctx->stat.read_errors++;
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
975
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
976
		goto out;
A
Arne Jansen 已提交
977 978
	}

979
	/* setup the context, map the logical blocks and alloc the pages */
980
	ret = scrub_setup_recheck_block(sblock_to_check, sblocks_for_recheck);
981
	if (ret) {
982 983 984 985
		spin_lock(&sctx->stat_lock);
		sctx->stat.read_errors++;
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
986
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
987 988 989 990
		goto out;
	}
	BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS);
	sblock_bad = sblocks_for_recheck + failed_mirror_index;
991

992
	/* build and submit the bios for the failed mirror, check checksums */
993
	scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum,
994
			    csum, generation, sctx->csum_size, 1);
A
Arne Jansen 已提交
995

996 997 998 999 1000 1001 1002 1003 1004 1005
	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)
		 */
1006 1007
		spin_lock(&sctx->stat_lock);
		sctx->stat.unverified_errors++;
1008
		sblock_to_check->data_corrected = 1;
1009
		spin_unlock(&sctx->stat_lock);
A
Arne Jansen 已提交
1010

1011 1012
		if (sctx->is_dev_replace)
			scrub_write_block_to_dev_replace(sblock_bad);
1013
		goto out;
A
Arne Jansen 已提交
1014 1015
	}

1016
	if (!sblock_bad->no_io_error_seen) {
1017 1018 1019
		spin_lock(&sctx->stat_lock);
		sctx->stat.read_errors++;
		spin_unlock(&sctx->stat_lock);
1020 1021
		if (__ratelimit(&_rs))
			scrub_print_warning("i/o error", sblock_to_check);
1022
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
1023
	} else if (sblock_bad->checksum_error) {
1024 1025 1026
		spin_lock(&sctx->stat_lock);
		sctx->stat.csum_errors++;
		spin_unlock(&sctx->stat_lock);
1027 1028
		if (__ratelimit(&_rs))
			scrub_print_warning("checksum error", sblock_to_check);
1029
		btrfs_dev_stat_inc_and_print(dev,
1030
					     BTRFS_DEV_STAT_CORRUPTION_ERRS);
1031
	} else if (sblock_bad->header_error) {
1032 1033 1034
		spin_lock(&sctx->stat_lock);
		sctx->stat.verify_errors++;
		spin_unlock(&sctx->stat_lock);
1035 1036 1037
		if (__ratelimit(&_rs))
			scrub_print_warning("checksum/header error",
					    sblock_to_check);
1038
		if (sblock_bad->generation_error)
1039
			btrfs_dev_stat_inc_and_print(dev,
1040 1041
				BTRFS_DEV_STAT_GENERATION_ERRS);
		else
1042
			btrfs_dev_stat_inc_and_print(dev,
1043
				BTRFS_DEV_STAT_CORRUPTION_ERRS);
1044
	}
A
Arne Jansen 已提交
1045

1046 1047 1048 1049
	if (sctx->readonly) {
		ASSERT(!sctx->is_dev_replace);
		goto out;
	}
A
Arne Jansen 已提交
1050

1051 1052
	if (!is_metadata && !have_csum) {
		struct scrub_fixup_nodatasum *fixup_nodatasum;
A
Arne Jansen 已提交
1053

1054 1055
		WARN_ON(sctx->is_dev_replace);

1056 1057
nodatasum_case:

1058 1059 1060 1061 1062 1063 1064 1065 1066 1067
		/*
		 * !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;
1068
		fixup_nodatasum->sctx = sctx;
1069
		fixup_nodatasum->dev = dev;
1070 1071 1072
		fixup_nodatasum->logical = logical;
		fixup_nodatasum->root = fs_info->extent_root;
		fixup_nodatasum->mirror_num = failed_mirror_index + 1;
1073
		scrub_pending_trans_workers_inc(sctx);
1074 1075
		btrfs_init_work(&fixup_nodatasum->work, btrfs_scrub_helper,
				scrub_fixup_nodatasum, NULL, NULL);
1076 1077
		btrfs_queue_work(fs_info->scrub_workers,
				 &fixup_nodatasum->work);
1078
		goto out;
A
Arne Jansen 已提交
1079 1080
	}

1081 1082
	/*
	 * now build and submit the bios for the other mirrors, check
1083 1084
	 * checksums.
	 * First try to pick the mirror which is completely without I/O
1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099
	 * 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++) {
1100
		struct scrub_block *sblock_other;
1101

1102 1103 1104 1105 1106
		if (mirror_index == failed_mirror_index)
			continue;
		sblock_other = sblocks_for_recheck + mirror_index;

		/* build and submit the bios, check checksums */
1107 1108
		scrub_recheck_block(fs_info, sblock_other, is_metadata,
				    have_csum, csum, generation,
1109
				    sctx->csum_size, 0);
1110 1111

		if (!sblock_other->header_error &&
1112 1113
		    !sblock_other->checksum_error &&
		    sblock_other->no_io_error_seen) {
1114 1115
			if (sctx->is_dev_replace) {
				scrub_write_block_to_dev_replace(sblock_other);
1116
				goto corrected_error;
1117 1118
			} else {
				ret = scrub_repair_block_from_good_copy(
1119 1120 1121
						sblock_bad, sblock_other);
				if (!ret)
					goto corrected_error;
1122
			}
1123 1124
		}
	}
A
Arne Jansen 已提交
1125

1126 1127
	if (sblock_bad->no_io_error_seen && !sctx->is_dev_replace)
		goto did_not_correct_error;
1128 1129 1130

	/*
	 * In case of I/O errors in the area that is supposed to be
1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151
	 * 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
1152
	 */
1153
	success = 1;
1154 1155
	for (page_num = 0; page_num < sblock_bad->page_count;
	     page_num++) {
1156
		struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1157
		struct scrub_block *sblock_other = NULL;
1158

1159 1160
		/* skip no-io-error page in scrub */
		if (!page_bad->io_error && !sctx->is_dev_replace)
A
Arne Jansen 已提交
1161
			continue;
1162

1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173
		/* 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;
1174 1175
				}
			}
1176 1177
			if (!sblock_other)
				success = 0;
I
Ilya Dryomov 已提交
1178
		}
A
Arne Jansen 已提交
1179

1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206
		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;
1207
		}
A
Arne Jansen 已提交
1208 1209
	}

1210
	if (success && !sctx->is_dev_replace) {
1211 1212 1213 1214 1215 1216 1217 1218 1219 1220
		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.
			 */
1221 1222
			scrub_recheck_block(fs_info, sblock_bad,
					    is_metadata, have_csum, csum,
1223
					    generation, sctx->csum_size, 1);
1224
			if (!sblock_bad->header_error &&
1225 1226 1227 1228 1229 1230 1231
			    !sblock_bad->checksum_error &&
			    sblock_bad->no_io_error_seen)
				goto corrected_error;
			else
				goto did_not_correct_error;
		} else {
corrected_error:
1232 1233
			spin_lock(&sctx->stat_lock);
			sctx->stat.corrected_errors++;
1234
			sblock_to_check->data_corrected = 1;
1235
			spin_unlock(&sctx->stat_lock);
1236
			printk_ratelimited_in_rcu(KERN_ERR
1237
				"BTRFS: fixed up error at logical %llu on dev %s\n",
1238
				logical, rcu_str_deref(dev->name));
A
Arne Jansen 已提交
1239
		}
1240 1241
	} else {
did_not_correct_error:
1242 1243 1244
		spin_lock(&sctx->stat_lock);
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
1245
		printk_ratelimited_in_rcu(KERN_ERR
1246
			"BTRFS: unable to fixup (regular) error at logical %llu on dev %s\n",
1247
			logical, rcu_str_deref(dev->name));
I
Ilya Dryomov 已提交
1248
	}
A
Arne Jansen 已提交
1249

1250 1251 1252 1253 1254 1255
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;
1256
			struct scrub_recover *recover;
1257 1258
			int page_index;

1259 1260 1261
			for (page_index = 0; page_index < sblock->page_count;
			     page_index++) {
				sblock->pagev[page_index]->sblock = NULL;
1262 1263 1264 1265 1266 1267
				recover = sblock->pagev[page_index]->recover;
				if (recover) {
					scrub_put_recover(recover);
					sblock->pagev[page_index]->recover =
									NULL;
				}
1268 1269
				scrub_page_put(sblock->pagev[page_index]);
			}
1270 1271 1272
		}
		kfree(sblocks_for_recheck);
	}
A
Arne Jansen 已提交
1273

1274 1275
	return 0;
}
A
Arne Jansen 已提交
1276

1277
static inline int scrub_nr_raid_mirrors(struct btrfs_bio *bbio)
1278
{
Z
Zhao Lei 已提交
1279 1280 1281 1282 1283
	if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID5)
		return 2;
	else if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID6)
		return 3;
	else
1284 1285 1286
		return (int)bbio->num_stripes;
}

Z
Zhao Lei 已提交
1287 1288
static inline void scrub_stripe_index_and_offset(u64 logical, u64 map_type,
						 u64 *raid_map,
1289 1290 1291 1292 1293 1294 1295
						 u64 mapped_length,
						 int nstripes, int mirror,
						 int *stripe_index,
						 u64 *stripe_offset)
{
	int i;

1296
	if (map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316
		/* 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;
	}
}

1317
static int scrub_setup_recheck_block(struct scrub_block *original_sblock,
1318 1319
				     struct scrub_block *sblocks_for_recheck)
{
1320 1321 1322 1323
	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;
1324 1325 1326 1327 1328 1329
	struct scrub_recover *recover;
	struct btrfs_bio *bbio;
	u64 sublen;
	u64 mapped_length;
	u64 stripe_offset;
	int stripe_index;
1330
	int page_index = 0;
1331
	int mirror_index;
1332
	int nmirrors;
1333 1334 1335
	int ret;

	/*
1336
	 * note: the two members refs and outstanding_pages
1337 1338 1339 1340 1341
	 * are not used (and not set) in the blocks that are used for
	 * the recheck procedure
	 */

	while (length > 0) {
1342 1343 1344
		sublen = min_t(u64, length, PAGE_SIZE);
		mapped_length = sublen;
		bbio = NULL;
A
Arne Jansen 已提交
1345

1346 1347 1348 1349
		/*
		 * with a length of PAGE_SIZE, each returned stripe
		 * represents one mirror
		 */
1350
		ret = btrfs_map_sblock(fs_info, REQ_GET_READ_MIRRORS, logical,
1351
				       &mapped_length, &bbio, 0, 1);
1352
		if (ret || !bbio || mapped_length < sublen) {
1353
			btrfs_put_bbio(bbio);
1354 1355
			return -EIO;
		}
A
Arne Jansen 已提交
1356

1357 1358
		recover = kzalloc(sizeof(struct scrub_recover), GFP_NOFS);
		if (!recover) {
1359
			btrfs_put_bbio(bbio);
1360 1361 1362 1363 1364 1365 1366
			return -ENOMEM;
		}

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

1367
		BUG_ON(page_index >= SCRUB_PAGES_PER_RD_BIO);
1368

1369
		nmirrors = min(scrub_nr_raid_mirrors(bbio), BTRFS_MAX_MIRRORS);
Z
Zhao Lei 已提交
1370

1371
		for (mirror_index = 0; mirror_index < nmirrors;
1372 1373 1374 1375 1376
		     mirror_index++) {
			struct scrub_block *sblock;
			struct scrub_page *page;

			sblock = sblocks_for_recheck + mirror_index;
1377 1378 1379 1380
			sblock->sctx = sctx;
			page = kzalloc(sizeof(*page), GFP_NOFS);
			if (!page) {
leave_nomem:
1381 1382 1383
				spin_lock(&sctx->stat_lock);
				sctx->stat.malloc_errors++;
				spin_unlock(&sctx->stat_lock);
1384
				scrub_put_recover(recover);
1385 1386
				return -ENOMEM;
			}
1387 1388 1389
			scrub_page_get(page);
			sblock->pagev[page_index] = page;
			page->logical = logical;
1390

Z
Zhao Lei 已提交
1391 1392 1393
			scrub_stripe_index_and_offset(logical,
						      bbio->map_type,
						      bbio->raid_map,
1394
						      mapped_length,
1395 1396
						      bbio->num_stripes -
						      bbio->num_tgtdevs,
1397 1398 1399 1400 1401 1402 1403
						      mirror_index,
						      &stripe_index,
						      &stripe_offset);
			page->physical = bbio->stripes[stripe_index].physical +
					 stripe_offset;
			page->dev = bbio->stripes[stripe_index].dev;

1404 1405 1406 1407
			BUG_ON(page_index >= original_sblock->page_count);
			page->physical_for_dev_replace =
				original_sblock->pagev[page_index]->
				physical_for_dev_replace;
1408 1409
			/* for missing devices, dev->bdev is NULL */
			page->mirror_num = mirror_index + 1;
1410
			sblock->page_count++;
1411 1412 1413
			page->page = alloc_page(GFP_NOFS);
			if (!page->page)
				goto leave_nomem;
1414 1415 1416

			scrub_get_recover(recover);
			page->recover = recover;
1417
		}
1418
		scrub_put_recover(recover);
1419 1420 1421 1422 1423 1424
		length -= sublen;
		logical += sublen;
		page_index++;
	}

	return 0;
I
Ilya Dryomov 已提交
1425 1426
}

1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441
struct scrub_bio_ret {
	struct completion event;
	int error;
};

static void scrub_bio_wait_endio(struct bio *bio, int error)
{
	struct scrub_bio_ret *ret = bio->bi_private;

	ret->error = error;
	complete(&ret->event);
}

static inline int scrub_is_page_on_raid56(struct scrub_page *page)
{
Z
Zhao Lei 已提交
1442
	return page->recover &&
1443
	       (page->recover->bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK);
1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460
}

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,
1461
				    page->mirror_num, 0);
1462 1463 1464 1465 1466 1467 1468 1469 1470 1471
	if (ret)
		return ret;

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

	return 0;
}

1472 1473 1474 1475 1476 1477 1478
/*
 * 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.
 */
1479 1480 1481
static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
				struct scrub_block *sblock, int is_metadata,
				int have_csum, u8 *csum, u64 generation,
1482
				u16 csum_size, int retry_failed_mirror)
I
Ilya Dryomov 已提交
1483
{
1484
	int page_num;
I
Ilya Dryomov 已提交
1485

1486 1487 1488
	sblock->no_io_error_seen = 1;
	sblock->header_error = 0;
	sblock->checksum_error = 0;
I
Ilya Dryomov 已提交
1489

1490 1491
	for (page_num = 0; page_num < sblock->page_count; page_num++) {
		struct bio *bio;
1492
		struct scrub_page *page = sblock->pagev[page_num];
1493

1494
		if (page->dev->bdev == NULL) {
1495 1496 1497 1498 1499
			page->io_error = 1;
			sblock->no_io_error_seen = 0;
			continue;
		}

1500
		WARN_ON(!page->page);
1501
		bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1502 1503 1504 1505 1506
		if (!bio) {
			page->io_error = 1;
			sblock->no_io_error_seen = 0;
			continue;
		}
1507
		bio->bi_bdev = page->dev->bdev;
1508

1509
		bio_add_page(bio, page->page, PAGE_SIZE, 0);
1510 1511 1512 1513 1514 1515 1516 1517 1518
		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;
		}
1519

1520 1521
		bio_put(bio);
	}
I
Ilya Dryomov 已提交
1522

1523 1524 1525 1526 1527
	if (sblock->no_io_error_seen)
		scrub_recheck_block_checksum(fs_info, sblock, is_metadata,
					     have_csum, csum, generation,
					     csum_size);

1528
	return;
A
Arne Jansen 已提交
1529 1530
}

M
Miao Xie 已提交
1531 1532 1533 1534 1535 1536 1537 1538 1539 1540
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;
}

1541 1542 1543 1544 1545
static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
					 struct scrub_block *sblock,
					 int is_metadata, int have_csum,
					 const u8 *csum, u64 generation,
					 u16 csum_size)
A
Arne Jansen 已提交
1546
{
1547 1548 1549 1550 1551
	int page_num;
	u8 calculated_csum[BTRFS_CSUM_SIZE];
	u32 crc = ~(u32)0;
	void *mapped_buffer;

1552
	WARN_ON(!sblock->pagev[0]->page);
1553 1554 1555
	if (is_metadata) {
		struct btrfs_header *h;

1556
		mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
1557 1558
		h = (struct btrfs_header *)mapped_buffer;

1559
		if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h) ||
M
Miao Xie 已提交
1560
		    !scrub_check_fsid(h->fsid, sblock->pagev[0]) ||
1561
		    memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
1562
			   BTRFS_UUID_SIZE)) {
1563
			sblock->header_error = 1;
1564
		} else if (generation != btrfs_stack_header_generation(h)) {
1565 1566 1567
			sblock->header_error = 1;
			sblock->generation_error = 1;
		}
1568 1569 1570 1571
		csum = h->csum;
	} else {
		if (!have_csum)
			return;
A
Arne Jansen 已提交
1572

1573
		mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
1574
	}
A
Arne Jansen 已提交
1575

1576 1577
	for (page_num = 0;;) {
		if (page_num == 0 && is_metadata)
1578
			crc = btrfs_csum_data(
1579 1580 1581
				((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE,
				crc, PAGE_SIZE - BTRFS_CSUM_SIZE);
		else
1582
			crc = btrfs_csum_data(mapped_buffer, crc, PAGE_SIZE);
1583

1584
		kunmap_atomic(mapped_buffer);
1585 1586 1587
		page_num++;
		if (page_num >= sblock->page_count)
			break;
1588
		WARN_ON(!sblock->pagev[page_num]->page);
1589

1590
		mapped_buffer = kmap_atomic(sblock->pagev[page_num]->page);
1591 1592 1593 1594 1595
	}

	btrfs_csum_final(crc, calculated_csum);
	if (memcmp(calculated_csum, csum, csum_size))
		sblock->checksum_error = 1;
A
Arne Jansen 已提交
1596 1597
}

1598
static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
1599
					     struct scrub_block *sblock_good)
1600 1601 1602
{
	int page_num;
	int ret = 0;
I
Ilya Dryomov 已提交
1603

1604 1605
	for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
		int ret_sub;
I
Ilya Dryomov 已提交
1606

1607 1608
		ret_sub = scrub_repair_page_from_good_copy(sblock_bad,
							   sblock_good,
1609
							   page_num, 1);
1610 1611
		if (ret_sub)
			ret = ret_sub;
A
Arne Jansen 已提交
1612
	}
1613 1614 1615 1616 1617 1618 1619 1620

	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)
{
1621 1622
	struct scrub_page *page_bad = sblock_bad->pagev[page_num];
	struct scrub_page *page_good = sblock_good->pagev[page_num];
1623

1624 1625
	BUG_ON(page_bad->page == NULL);
	BUG_ON(page_good->page == NULL);
1626 1627 1628 1629 1630
	if (force_write || sblock_bad->header_error ||
	    sblock_bad->checksum_error || page_bad->io_error) {
		struct bio *bio;
		int ret;

1631
		if (!page_bad->dev->bdev) {
1632 1633 1634
			printk_ratelimited(KERN_WARNING "BTRFS: "
				"scrub_repair_page_from_good_copy(bdev == NULL) "
				"is unexpected!\n");
1635 1636 1637
			return -EIO;
		}

1638
		bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1639 1640
		if (!bio)
			return -EIO;
1641
		bio->bi_bdev = page_bad->dev->bdev;
1642
		bio->bi_iter.bi_sector = page_bad->physical >> 9;
1643 1644 1645 1646 1647

		ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0);
		if (PAGE_SIZE != ret) {
			bio_put(bio);
			return -EIO;
1648
		}
1649

1650
		if (btrfsic_submit_bio_wait(WRITE, bio)) {
1651 1652
			btrfs_dev_stat_inc_and_print(page_bad->dev,
				BTRFS_DEV_STAT_WRITE_ERRS);
1653 1654 1655
			btrfs_dev_replace_stats_inc(
				&sblock_bad->sctx->dev_root->fs_info->
				dev_replace.num_write_errors);
1656 1657 1658
			bio_put(bio);
			return -EIO;
		}
1659
		bio_put(bio);
A
Arne Jansen 已提交
1660 1661
	}

1662 1663 1664
	return 0;
}

1665 1666 1667 1668
static void scrub_write_block_to_dev_replace(struct scrub_block *sblock)
{
	int page_num;

1669 1670 1671 1672 1673 1674 1675
	/*
	 * 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;

1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 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
	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) {
1731
			bio = btrfs_io_bio_alloc(GFP_NOFS, wr_ctx->pages_per_wr_bio);
1732 1733 1734 1735 1736 1737 1738 1739 1740 1741
			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;
1742
		bio->bi_iter.bi_sector = sbio->physical >> 9;
1743 1744 1745 1746 1747 1748 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 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800
		sbio->err = 0;
	} else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
		   spage->physical_for_dev_replace ||
		   sbio->logical + sbio->page_count * PAGE_SIZE !=
		   spage->logical) {
		scrub_wr_submit(sctx);
		goto again;
	}

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

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

	return 0;
}

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

	if (!wr_ctx->wr_curr_bio)
		return;

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

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

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

1801 1802
	btrfs_init_work(&sbio->work, btrfs_scrubwrc_helper,
			 scrub_wr_bio_end_io_worker, NULL, NULL);
1803
	btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work);
1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834
}

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)
1835 1836 1837 1838
{
	u64 flags;
	int ret;

1839 1840
	WARN_ON(sblock->page_count < 1);
	flags = sblock->pagev[0]->flags;
1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851
	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);
1852 1853

	return ret;
A
Arne Jansen 已提交
1854 1855
}

1856
static int scrub_checksum_data(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
	u8 csum[BTRFS_CSUM_SIZE];
1860 1861 1862
	u8 *on_disk_csum;
	struct page *page;
	void *buffer;
A
Arne Jansen 已提交
1863 1864
	u32 crc = ~(u32)0;
	int fail = 0;
1865 1866
	u64 len;
	int index;
A
Arne Jansen 已提交
1867

1868
	BUG_ON(sblock->page_count < 1);
1869
	if (!sblock->pagev[0]->have_csum)
A
Arne Jansen 已提交
1870 1871
		return 0;

1872 1873
	on_disk_csum = sblock->pagev[0]->csum;
	page = sblock->pagev[0]->page;
1874
	buffer = kmap_atomic(page);
1875

1876
	len = sctx->sectorsize;
1877 1878 1879 1880
	index = 0;
	for (;;) {
		u64 l = min_t(u64, len, PAGE_SIZE);

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

A
Arne Jansen 已提交
1893
	btrfs_csum_final(crc, csum);
1894
	if (memcmp(csum, on_disk_csum, sctx->csum_size))
A
Arne Jansen 已提交
1895 1896 1897 1898 1899
		fail = 1;

	return fail;
}

1900
static int scrub_checksum_tree_block(struct scrub_block *sblock)
A
Arne Jansen 已提交
1901
{
1902
	struct scrub_ctx *sctx = sblock->sctx;
A
Arne Jansen 已提交
1903
	struct btrfs_header *h;
1904
	struct btrfs_root *root = sctx->dev_root;
A
Arne Jansen 已提交
1905
	struct btrfs_fs_info *fs_info = root->fs_info;
1906 1907 1908 1909 1910 1911
	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
1912 1913 1914
	u32 crc = ~(u32)0;
	int fail = 0;
	int crc_fail = 0;
1915 1916 1917 1918
	u64 len;
	int index;

	BUG_ON(sblock->page_count < 1);
1919
	page = sblock->pagev[0]->page;
1920
	mapped_buffer = kmap_atomic(page);
1921
	h = (struct btrfs_header *)mapped_buffer;
1922
	memcpy(on_disk_csum, h->csum, sctx->csum_size);
A
Arne Jansen 已提交
1923 1924 1925 1926 1927 1928 1929

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

1930
	if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h))
A
Arne Jansen 已提交
1931 1932
		++fail;

1933
	if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h))
A
Arne Jansen 已提交
1934 1935
		++fail;

M
Miao Xie 已提交
1936
	if (!scrub_check_fsid(h->fsid, sblock->pagev[0]))
A
Arne Jansen 已提交
1937 1938 1939 1940 1941 1942
		++fail;

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

1943
	len = sctx->nodesize - BTRFS_CSUM_SIZE;
1944 1945 1946 1947 1948 1949
	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);

1950
		crc = btrfs_csum_data(p, crc, l);
1951
		kunmap_atomic(mapped_buffer);
1952 1953 1954 1955 1956
		len -= l;
		if (len == 0)
			break;
		index++;
		BUG_ON(index >= sblock->page_count);
1957 1958
		BUG_ON(!sblock->pagev[index]->page);
		page = sblock->pagev[index]->page;
1959
		mapped_buffer = kmap_atomic(page);
1960 1961 1962 1963 1964
		mapped_size = PAGE_SIZE;
		p = mapped_buffer;
	}

	btrfs_csum_final(crc, calculated_csum);
1965
	if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
A
Arne Jansen 已提交
1966 1967 1968 1969 1970
		++crc_fail;

	return fail || crc_fail;
}

1971
static int scrub_checksum_super(struct scrub_block *sblock)
A
Arne Jansen 已提交
1972 1973
{
	struct btrfs_super_block *s;
1974
	struct scrub_ctx *sctx = sblock->sctx;
1975 1976 1977 1978 1979 1980
	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
1981
	u32 crc = ~(u32)0;
1982 1983
	int fail_gen = 0;
	int fail_cor = 0;
1984 1985
	u64 len;
	int index;
A
Arne Jansen 已提交
1986

1987
	BUG_ON(sblock->page_count < 1);
1988
	page = sblock->pagev[0]->page;
1989
	mapped_buffer = kmap_atomic(page);
1990
	s = (struct btrfs_super_block *)mapped_buffer;
1991
	memcpy(on_disk_csum, s->csum, sctx->csum_size);
A
Arne Jansen 已提交
1992

1993
	if (sblock->pagev[0]->logical != btrfs_super_bytenr(s))
1994
		++fail_cor;
A
Arne Jansen 已提交
1995

1996
	if (sblock->pagev[0]->generation != btrfs_super_generation(s))
1997
		++fail_gen;
A
Arne Jansen 已提交
1998

M
Miao Xie 已提交
1999
	if (!scrub_check_fsid(s->fsid, sblock->pagev[0]))
2000
		++fail_cor;
A
Arne Jansen 已提交
2001

2002 2003 2004 2005 2006 2007 2008
	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);

2009
		crc = btrfs_csum_data(p, crc, l);
2010
		kunmap_atomic(mapped_buffer);
2011 2012 2013 2014 2015
		len -= l;
		if (len == 0)
			break;
		index++;
		BUG_ON(index >= sblock->page_count);
2016 2017
		BUG_ON(!sblock->pagev[index]->page);
		page = sblock->pagev[index]->page;
2018
		mapped_buffer = kmap_atomic(page);
2019 2020 2021 2022 2023
		mapped_size = PAGE_SIZE;
		p = mapped_buffer;
	}

	btrfs_csum_final(crc, calculated_csum);
2024
	if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
2025
		++fail_cor;
A
Arne Jansen 已提交
2026

2027
	if (fail_cor + fail_gen) {
A
Arne Jansen 已提交
2028 2029 2030 2031 2032
		/*
		 * if we find an error in a super block, we just report it.
		 * They will get written with the next transaction commit
		 * anyway
		 */
2033 2034 2035
		spin_lock(&sctx->stat_lock);
		++sctx->stat.super_errors;
		spin_unlock(&sctx->stat_lock);
2036
		if (fail_cor)
2037
			btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
2038 2039
				BTRFS_DEV_STAT_CORRUPTION_ERRS);
		else
2040
			btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
2041
				BTRFS_DEV_STAT_GENERATION_ERRS);
A
Arne Jansen 已提交
2042 2043
	}

2044
	return fail_cor + fail_gen;
A
Arne Jansen 已提交
2045 2046
}

2047 2048
static void scrub_block_get(struct scrub_block *sblock)
{
2049
	atomic_inc(&sblock->refs);
2050 2051 2052 2053
}

static void scrub_block_put(struct scrub_block *sblock)
{
2054
	if (atomic_dec_and_test(&sblock->refs)) {
2055 2056
		int i;

2057 2058 2059
		if (sblock->sparity)
			scrub_parity_put(sblock->sparity);

2060
		for (i = 0; i < sblock->page_count; i++)
2061
			scrub_page_put(sblock->pagev[i]);
2062 2063 2064 2065
		kfree(sblock);
	}
}

2066 2067
static void scrub_page_get(struct scrub_page *spage)
{
2068
	atomic_inc(&spage->refs);
2069 2070 2071 2072
}

static void scrub_page_put(struct scrub_page *spage)
{
2073
	if (atomic_dec_and_test(&spage->refs)) {
2074 2075 2076 2077 2078 2079
		if (spage->page)
			__free_page(spage->page);
		kfree(spage);
	}
}

2080
static void scrub_submit(struct scrub_ctx *sctx)
A
Arne Jansen 已提交
2081 2082 2083
{
	struct scrub_bio *sbio;

2084
	if (sctx->curr == -1)
S
Stefan Behrens 已提交
2085
		return;
A
Arne Jansen 已提交
2086

2087 2088
	sbio = sctx->bios[sctx->curr];
	sctx->curr = -1;
2089
	scrub_pending_bio_inc(sctx);
A
Arne Jansen 已提交
2090

2091 2092 2093 2094 2095 2096 2097 2098 2099
	if (!sbio->bio->bi_bdev) {
		/*
		 * this case should not happen. If btrfs_map_block() is
		 * wrong, it could happen for dev-replace operations on
		 * missing devices when no mirrors are available, but in
		 * this case it should already fail the mount.
		 * This case is handled correctly (but _very_ slowly).
		 */
		printk_ratelimited(KERN_WARNING
2100
			"BTRFS: scrub_submit(bio bdev == NULL) is unexpected!\n");
2101 2102 2103 2104
		bio_endio(sbio->bio, -EIO);
	} else {
		btrfsic_submit_bio(READ, sbio->bio);
	}
A
Arne Jansen 已提交
2105 2106
}

2107 2108
static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
				    struct scrub_page *spage)
A
Arne Jansen 已提交
2109
{
2110
	struct scrub_block *sblock = spage->sblock;
A
Arne Jansen 已提交
2111
	struct scrub_bio *sbio;
2112
	int ret;
A
Arne Jansen 已提交
2113 2114 2115 2116 2117

again:
	/*
	 * grab a fresh bio or wait for one to become available
	 */
2118 2119 2120 2121 2122 2123 2124 2125
	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
2126
		} else {
2127 2128
			spin_unlock(&sctx->list_lock);
			wait_event(sctx->list_wait, sctx->first_free != -1);
A
Arne Jansen 已提交
2129 2130
		}
	}
2131
	sbio = sctx->bios[sctx->curr];
2132
	if (sbio->page_count == 0) {
2133 2134
		struct bio *bio;

2135 2136
		sbio->physical = spage->physical;
		sbio->logical = spage->logical;
2137
		sbio->dev = spage->dev;
2138 2139
		bio = sbio->bio;
		if (!bio) {
2140
			bio = btrfs_io_bio_alloc(GFP_NOFS, sctx->pages_per_rd_bio);
2141 2142 2143 2144
			if (!bio)
				return -ENOMEM;
			sbio->bio = bio;
		}
2145 2146 2147

		bio->bi_private = sbio;
		bio->bi_end_io = scrub_bio_end_io;
2148
		bio->bi_bdev = sbio->dev->bdev;
2149
		bio->bi_iter.bi_sector = sbio->physical >> 9;
2150
		sbio->err = 0;
2151 2152 2153
	} else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
		   spage->physical ||
		   sbio->logical + sbio->page_count * PAGE_SIZE !=
2154 2155
		   spage->logical ||
		   sbio->dev != spage->dev) {
2156
		scrub_submit(sctx);
A
Arne Jansen 已提交
2157 2158
		goto again;
	}
2159

2160 2161 2162 2163 2164 2165 2166 2167
	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;
		}
2168
		scrub_submit(sctx);
2169 2170 2171
		goto again;
	}

2172
	scrub_block_get(sblock); /* one for the page added to the bio */
2173 2174
	atomic_inc(&sblock->outstanding_pages);
	sbio->page_count++;
2175
	if (sbio->page_count == sctx->pages_per_rd_bio)
2176
		scrub_submit(sctx);
2177 2178 2179 2180

	return 0;
}

2181
static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
2182
		       u64 physical, struct btrfs_device *dev, u64 flags,
2183 2184
		       u64 gen, int mirror_num, u8 *csum, int force,
		       u64 physical_for_dev_replace)
2185 2186 2187 2188 2189 2190
{
	struct scrub_block *sblock;
	int index;

	sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
	if (!sblock) {
2191 2192 2193
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		spin_unlock(&sctx->stat_lock);
2194
		return -ENOMEM;
A
Arne Jansen 已提交
2195
	}
2196

2197 2198
	/* one ref inside this function, plus one for each page added to
	 * a bio later on */
2199
	atomic_set(&sblock->refs, 1);
2200
	sblock->sctx = sctx;
2201 2202 2203
	sblock->no_io_error_seen = 1;

	for (index = 0; len > 0; index++) {
2204
		struct scrub_page *spage;
2205 2206
		u64 l = min_t(u64, len, PAGE_SIZE);

2207 2208 2209
		spage = kzalloc(sizeof(*spage), GFP_NOFS);
		if (!spage) {
leave_nomem:
2210 2211 2212
			spin_lock(&sctx->stat_lock);
			sctx->stat.malloc_errors++;
			spin_unlock(&sctx->stat_lock);
2213
			scrub_block_put(sblock);
2214 2215
			return -ENOMEM;
		}
2216 2217 2218
		BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
		scrub_page_get(spage);
		sblock->pagev[index] = spage;
2219
		spage->sblock = sblock;
2220
		spage->dev = dev;
2221 2222 2223 2224
		spage->flags = flags;
		spage->generation = gen;
		spage->logical = logical;
		spage->physical = physical;
2225
		spage->physical_for_dev_replace = physical_for_dev_replace;
2226 2227 2228
		spage->mirror_num = mirror_num;
		if (csum) {
			spage->have_csum = 1;
2229
			memcpy(spage->csum, csum, sctx->csum_size);
2230 2231 2232 2233
		} else {
			spage->have_csum = 0;
		}
		sblock->page_count++;
2234 2235 2236
		spage->page = alloc_page(GFP_NOFS);
		if (!spage->page)
			goto leave_nomem;
2237 2238 2239
		len -= l;
		logical += l;
		physical += l;
2240
		physical_for_dev_replace += l;
2241 2242
	}

2243
	WARN_ON(sblock->page_count == 0);
2244
	for (index = 0; index < sblock->page_count; index++) {
2245
		struct scrub_page *spage = sblock->pagev[index];
2246 2247
		int ret;

2248
		ret = scrub_add_page_to_rd_bio(sctx, spage);
2249 2250
		if (ret) {
			scrub_block_put(sblock);
2251
			return ret;
2252
		}
2253
	}
A
Arne Jansen 已提交
2254

2255
	if (force)
2256
		scrub_submit(sctx);
A
Arne Jansen 已提交
2257

2258 2259
	/* last one frees, either here or in bio completion for last page */
	scrub_block_put(sblock);
A
Arne Jansen 已提交
2260 2261 2262
	return 0;
}

2263 2264 2265
static void scrub_bio_end_io(struct bio *bio, int err)
{
	struct scrub_bio *sbio = bio->bi_private;
2266
	struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
2267 2268 2269 2270

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

2271
	btrfs_queue_work(fs_info->scrub_workers, &sbio->work);
2272 2273 2274 2275 2276
}

static void scrub_bio_end_io_worker(struct btrfs_work *work)
{
	struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
2277
	struct scrub_ctx *sctx = sbio->sctx;
2278 2279
	int i;

2280
	BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO);
2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301
	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;
2302 2303 2304 2305
	spin_lock(&sctx->list_lock);
	sbio->next_free = sctx->first_free;
	sctx->first_free = sbio->index;
	spin_unlock(&sctx->list_lock);
2306 2307 2308 2309 2310 2311 2312 2313

	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);
	}

2314
	scrub_pending_bio_dec(sctx);
2315 2316
}

2317 2318 2319 2320
static inline void __scrub_mark_bitmap(struct scrub_parity *sparity,
				       unsigned long *bitmap,
				       u64 start, u64 len)
{
2321
	u32 offset;
2322 2323 2324 2325 2326 2327 2328 2329 2330
	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;
2331
	start = div_u64_rem(start, sparity->stripe_len, &offset);
2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355
	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);
}

2356 2357
static void scrub_block_complete(struct scrub_block *sblock)
{
2358 2359
	int corrupted = 0;

2360
	if (!sblock->no_io_error_seen) {
2361
		corrupted = 1;
2362
		scrub_handle_errored_block(sblock);
2363 2364 2365 2366 2367 2368
	} else {
		/*
		 * if has checksum error, write via repair mechanism in
		 * dev replace case, otherwise write here in dev replace
		 * case.
		 */
2369 2370
		corrupted = scrub_checksum(sblock);
		if (!corrupted && sblock->sctx->is_dev_replace)
2371 2372
			scrub_write_block_to_dev_replace(sblock);
	}
2373 2374 2375 2376 2377 2378 2379 2380 2381

	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);
	}
2382 2383
}

2384
static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len,
A
Arne Jansen 已提交
2385 2386 2387
			   u8 *csum)
{
	struct btrfs_ordered_sum *sum = NULL;
2388
	unsigned long index;
A
Arne Jansen 已提交
2389 2390
	unsigned long num_sectors;

2391 2392
	while (!list_empty(&sctx->csum_list)) {
		sum = list_first_entry(&sctx->csum_list,
A
Arne Jansen 已提交
2393 2394 2395 2396 2397 2398
				       struct btrfs_ordered_sum, list);
		if (sum->bytenr > logical)
			return 0;
		if (sum->bytenr + sum->len > logical)
			break;

2399
		++sctx->stat.csum_discards;
A
Arne Jansen 已提交
2400 2401 2402 2403 2404 2405 2406
		list_del(&sum->list);
		kfree(sum);
		sum = NULL;
	}
	if (!sum)
		return 0;

2407
	index = ((u32)(logical - sum->bytenr)) / sctx->sectorsize;
2408
	num_sectors = sum->len / sctx->sectorsize;
2409 2410
	memcpy(csum, sum->sums + index, sctx->csum_size);
	if (index == num_sectors - 1) {
A
Arne Jansen 已提交
2411 2412 2413
		list_del(&sum->list);
		kfree(sum);
	}
2414
	return 1;
A
Arne Jansen 已提交
2415 2416 2417
}

/* scrub extent tries to collect up to 64 kB for each bio */
2418
static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len,
2419
			u64 physical, struct btrfs_device *dev, u64 flags,
2420
			u64 gen, int mirror_num, u64 physical_for_dev_replace)
A
Arne Jansen 已提交
2421 2422 2423
{
	int ret;
	u8 csum[BTRFS_CSUM_SIZE];
2424 2425 2426
	u32 blocksize;

	if (flags & BTRFS_EXTENT_FLAG_DATA) {
2427 2428 2429 2430 2431
		blocksize = sctx->sectorsize;
		spin_lock(&sctx->stat_lock);
		sctx->stat.data_extents_scrubbed++;
		sctx->stat.data_bytes_scrubbed += len;
		spin_unlock(&sctx->stat_lock);
2432
	} else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2433 2434 2435 2436 2437
		blocksize = sctx->nodesize;
		spin_lock(&sctx->stat_lock);
		sctx->stat.tree_extents_scrubbed++;
		sctx->stat.tree_bytes_scrubbed += len;
		spin_unlock(&sctx->stat_lock);
2438
	} else {
2439
		blocksize = sctx->sectorsize;
2440
		WARN_ON(1);
2441
	}
A
Arne Jansen 已提交
2442 2443

	while (len) {
2444
		u64 l = min_t(u64, len, blocksize);
A
Arne Jansen 已提交
2445 2446 2447 2448
		int have_csum = 0;

		if (flags & BTRFS_EXTENT_FLAG_DATA) {
			/* push csums to sbio */
2449
			have_csum = scrub_find_csum(sctx, logical, l, csum);
A
Arne Jansen 已提交
2450
			if (have_csum == 0)
2451
				++sctx->stat.no_csum;
2452 2453 2454 2455 2456 2457
			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
2458
		}
2459
		ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen,
2460 2461 2462
				  mirror_num, have_csum ? csum : NULL, 0,
				  physical_for_dev_replace);
behind_scrub_pages:
A
Arne Jansen 已提交
2463 2464 2465 2466 2467
		if (ret)
			return ret;
		len -= l;
		logical += l;
		physical += l;
2468
		physical_for_dev_replace += l;
A
Arne Jansen 已提交
2469 2470 2471 2472
	}
	return 0;
}

2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491
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 */
2492
	atomic_set(&sblock->refs, 1);
2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590
	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;

	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;
2591
skip:
2592 2593 2594 2595 2596 2597 2598
		len -= l;
		logical += l;
		physical += l;
	}
	return 0;
}

2599 2600 2601 2602 2603 2604 2605 2606
/*
 * 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,
2607 2608
				   struct map_lookup *map, u64 *offset,
				   u64 *stripe_start)
2609 2610 2611 2612 2613
{
	int i;
	int j = 0;
	u64 stripe_nr;
	u64 last_offset;
2614 2615
	u32 stripe_index;
	u32 rot;
2616 2617 2618

	last_offset = (physical - map->stripes[num].physical) *
		      nr_data_stripes(map);
2619 2620 2621
	if (stripe_start)
		*stripe_start = last_offset;

2622 2623 2624 2625
	*offset = last_offset;
	for (i = 0; i < nr_data_stripes(map); i++) {
		*offset = last_offset + i * map->stripe_len;

2626 2627
		stripe_nr = div_u64(*offset, map->stripe_len);
		stripe_nr = div_u64(stripe_nr, nr_data_stripes(map));
2628 2629

		/* Work out the disk rotation on this stripe-set */
2630
		stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, &rot);
2631 2632
		/* calculate which stripe this data locates */
		rot += i;
2633
		stripe_index = rot % map->num_stripes;
2634 2635 2636 2637 2638 2639 2640 2641 2642
		if (stripe_index == num)
			return 0;
		if (stripe_index < num)
			j++;
	}
	*offset = last_offset + j * map->stripe_len;
	return 1;
}

2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664
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);
}

2665 2666 2667 2668 2669 2670 2671 2672 2673 2674
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);
}

2675 2676 2677 2678 2679 2680 2681 2682 2683
static void scrub_parity_bio_endio(struct bio *bio, int error)
{
	struct scrub_parity *sparity = (struct scrub_parity *)bio->bi_private;

	if (error)
		bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap,
			  sparity->nsectors);

	bio_put(bio);
2684 2685 2686 2687 2688

	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);
2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705
}

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;

	length = sparity->logic_end - sparity->logic_start + 1;
2706
	ret = btrfs_map_sblock(sctx->dev_root->fs_info, WRITE,
2707
			       sparity->logic_start,
2708 2709
			       &length, &bbio, 0, 1);
	if (ret || !bbio || !bbio->raid_map)
2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720
		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,
2721
					      length, sparity->scrub_dev,
2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737
					      sparity->dbitmap,
					      sparity->nsectors);
	if (!rbio)
		goto rbio_out;

	list_for_each_entry(spage, &sparity->spages, list)
		raid56_parity_add_scrub_pages(rbio, spage->page,
					      spage->logical);

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

rbio_out:
	bio_put(bio);
bbio_out:
2738
	btrfs_put_bbio(bbio);
2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754
	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)
{
2755
	atomic_inc(&sparity->refs);
2756 2757 2758 2759
}

static void scrub_parity_put(struct scrub_parity *sparity)
{
2760
	if (!atomic_dec_and_test(&sparity->refs))
2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809
		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;
	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;
	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;
2810
	atomic_set(&sparity->refs, 1);
2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883
	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);

			if (key.type == BTRFS_METADATA_ITEM_KEY)
				bytes = root->nodesize;
			else
				bytes = key.offset;

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

			if (key.type != BTRFS_EXTENT_ITEM_KEY &&
			    key.type != BTRFS_METADATA_ITEM_KEY)
				goto next;

			if (key.objectid > logic_end) {
				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);

2884 2885 2886 2887 2888 2889
			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);
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
				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);

			scrub_remap_extent(fs_info, extent_logical,
					   extent_len, &extent_physical,
					   &extent_dev,
					   &extent_mirror_num);

			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);
2927 2928 2929

			scrub_free_csums(sctx);

2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971
			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,
						logic_end - logic_start + 1);
	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;
}

2972
static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
2973 2974
					   struct map_lookup *map,
					   struct btrfs_device *scrub_dev,
2975 2976
					   int num, u64 base, u64 length,
					   int is_dev_replace)
A
Arne Jansen 已提交
2977
{
2978
	struct btrfs_path *path, *ppath;
2979
	struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
A
Arne Jansen 已提交
2980 2981 2982
	struct btrfs_root *root = fs_info->extent_root;
	struct btrfs_root *csum_root = fs_info->csum_root;
	struct btrfs_extent_item *extent;
2983
	struct blk_plug plug;
A
Arne Jansen 已提交
2984 2985 2986 2987 2988 2989 2990 2991
	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
2992
	u64 logic_end;
2993
	u64 physical_end;
A
Arne Jansen 已提交
2994
	u64 generation;
2995
	int mirror_num;
A
Arne Jansen 已提交
2996 2997 2998 2999
	struct reada_control *reada1;
	struct reada_control *reada2;
	struct btrfs_key key_start;
	struct btrfs_key key_end;
A
Arne Jansen 已提交
3000 3001
	u64 increment = map->stripe_len;
	u64 offset;
3002 3003 3004
	u64 extent_logical;
	u64 extent_physical;
	u64 extent_len;
3005 3006
	u64 stripe_logical;
	u64 stripe_end;
3007 3008
	struct btrfs_device *extent_dev;
	int extent_mirror_num;
3009
	int stop_loop = 0;
D
David Woodhouse 已提交
3010

3011
	physical = map->stripes[num].physical;
A
Arne Jansen 已提交
3012
	offset = 0;
3013
	nstripes = div_u64(length, map->stripe_len);
A
Arne Jansen 已提交
3014 3015 3016
	if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
		offset = map->stripe_len * num;
		increment = map->stripe_len * map->num_stripes;
3017
		mirror_num = 1;
A
Arne Jansen 已提交
3018 3019 3020 3021
	} 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;
3022
		mirror_num = num % map->sub_stripes + 1;
A
Arne Jansen 已提交
3023 3024
	} else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
		increment = map->stripe_len;
3025
		mirror_num = num % map->num_stripes + 1;
A
Arne Jansen 已提交
3026 3027
	} else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
		increment = map->stripe_len;
3028
		mirror_num = num % map->num_stripes + 1;
3029
	} else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3030
		get_raid56_logic_offset(physical, num, map, &offset, NULL);
3031 3032
		increment = map->stripe_len * nr_data_stripes(map);
		mirror_num = 1;
A
Arne Jansen 已提交
3033 3034
	} else {
		increment = map->stripe_len;
3035
		mirror_num = 1;
A
Arne Jansen 已提交
3036 3037 3038 3039 3040 3041
	}

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

3042 3043
	ppath = btrfs_alloc_path();
	if (!ppath) {
3044
		btrfs_free_path(path);
3045 3046 3047
		return -ENOMEM;
	}

3048 3049 3050 3051 3052
	/*
	 * 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
3053 3054 3055
	path->search_commit_root = 1;
	path->skip_locking = 1;

3056 3057
	ppath->search_commit_root = 1;
	ppath->skip_locking = 1;
A
Arne Jansen 已提交
3058
	/*
A
Arne Jansen 已提交
3059 3060 3061
	 * 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
3062 3063
	 */
	logical = base + offset;
3064
	physical_end = physical + nstripes * map->stripe_len;
3065
	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3066
		get_raid56_logic_offset(physical_end, num,
3067
					map, &logic_end, NULL);
3068 3069 3070 3071
		logic_end += base;
	} else {
		logic_end = logical + increment * nstripes;
	}
3072
	wait_event(sctx->list_wait,
3073
		   atomic_read(&sctx->bios_in_flight) == 0);
3074
	scrub_blocked_if_needed(fs_info);
A
Arne Jansen 已提交
3075 3076 3077 3078 3079

	/* 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;
3080
	key_end.objectid = logic_end;
3081 3082
	key_end.type = BTRFS_METADATA_ITEM_KEY;
	key_end.offset = (u64)-1;
A
Arne Jansen 已提交
3083 3084 3085 3086 3087 3088 3089
	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;
3090
	key_end.offset = logic_end;
A
Arne Jansen 已提交
3091 3092 3093 3094 3095 3096 3097
	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
3098 3099 3100 3101 3102

	/*
	 * collect all data csums for the stripe to avoid seeking during
	 * the scrub. This might currently (crc32) end up to be about 1MB
	 */
3103
	blk_start_plug(&plug);
A
Arne Jansen 已提交
3104 3105 3106 3107 3108

	/*
	 * now find all extents for each stripe and scrub them
	 */
	ret = 0;
3109
	while (physical < physical_end) {
A
Arne Jansen 已提交
3110 3111 3112 3113
		/*
		 * canceled?
		 */
		if (atomic_read(&fs_info->scrub_cancel_req) ||
3114
		    atomic_read(&sctx->cancel_req)) {
A
Arne Jansen 已提交
3115 3116 3117 3118 3119 3120 3121 3122
			ret = -ECANCELED;
			goto out;
		}
		/*
		 * check to see if we have to pause
		 */
		if (atomic_read(&fs_info->scrub_pause_req)) {
			/* push queued extents */
3123
			atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
3124
			scrub_submit(sctx);
3125 3126 3127
			mutex_lock(&sctx->wr_ctx.wr_lock);
			scrub_wr_submit(sctx);
			mutex_unlock(&sctx->wr_ctx.wr_lock);
3128
			wait_event(sctx->list_wait,
3129
				   atomic_read(&sctx->bios_in_flight) == 0);
3130
			atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
3131
			scrub_blocked_if_needed(fs_info);
A
Arne Jansen 已提交
3132 3133
		}

3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151
		/* for raid56, we skip parity stripe */
		if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
			ret = get_raid56_logic_offset(physical, num, map,
						      &logical,
						      &stripe_logical);
			logical += base;
			if (ret) {
				stripe_logical += base;
				stripe_end = stripe_logical + increment - 1;
				ret = scrub_raid56_parity(sctx, map, scrub_dev,
							  ppath, stripe_logical,
							  stripe_end);
				if (ret)
					goto out;
				goto skip;
			}
		}

3152 3153 3154 3155
		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
3156
		key.objectid = logical;
L
Liu Bo 已提交
3157
		key.offset = (u64)-1;
A
Arne Jansen 已提交
3158 3159 3160 3161

		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
		if (ret < 0)
			goto out;
3162

3163
		if (ret > 0) {
3164
			ret = btrfs_previous_extent_item(root, path, 0);
A
Arne Jansen 已提交
3165 3166
			if (ret < 0)
				goto out;
3167 3168 3169 3170 3171 3172 3173 3174 3175
			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
3176 3177
		}

L
Liu Bo 已提交
3178
		stop_loop = 0;
A
Arne Jansen 已提交
3179
		while (1) {
3180 3181
			u64 bytes;

A
Arne Jansen 已提交
3182 3183 3184 3185 3186 3187 3188 3189 3190
			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
3191
				stop_loop = 1;
A
Arne Jansen 已提交
3192 3193 3194 3195
				break;
			}
			btrfs_item_key_to_cpu(l, &key, slot);

3196
			if (key.type == BTRFS_METADATA_ITEM_KEY)
3197
				bytes = root->nodesize;
3198 3199 3200 3201
			else
				bytes = key.offset;

			if (key.objectid + bytes <= logical)
A
Arne Jansen 已提交
3202 3203
				goto next;

L
Liu Bo 已提交
3204 3205 3206
			if (key.type != BTRFS_EXTENT_ITEM_KEY &&
			    key.type != BTRFS_METADATA_ITEM_KEY)
				goto next;
A
Arne Jansen 已提交
3207

L
Liu Bo 已提交
3208 3209 3210 3211 3212 3213
			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
3214 3215 3216 3217 3218 3219

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

3220 3221 3222 3223
			if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) &&
			    (key.objectid < logical ||
			     key.objectid + bytes >
			     logical + map->stripe_len)) {
3224 3225 3226
				btrfs_err(fs_info,
					   "scrub: tree block %llu spanning "
					   "stripes, ignored. logical=%llu",
3227
				       key.objectid, logical);
A
Arne Jansen 已提交
3228 3229 3230
				goto next;
			}

L
Liu Bo 已提交
3231 3232 3233 3234
again:
			extent_logical = key.objectid;
			extent_len = bytes;

A
Arne Jansen 已提交
3235 3236 3237
			/*
			 * trim extent to this stripe
			 */
L
Liu Bo 已提交
3238 3239 3240
			if (extent_logical < logical) {
				extent_len -= logical - extent_logical;
				extent_logical = logical;
A
Arne Jansen 已提交
3241
			}
L
Liu Bo 已提交
3242
			if (extent_logical + extent_len >
A
Arne Jansen 已提交
3243
			    logical + map->stripe_len) {
L
Liu Bo 已提交
3244 3245
				extent_len = logical + map->stripe_len -
					     extent_logical;
A
Arne Jansen 已提交
3246 3247
			}

L
Liu Bo 已提交
3248
			extent_physical = extent_logical - logical + physical;
3249 3250 3251 3252 3253 3254 3255
			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
3256 3257 3258 3259 3260 3261 3262

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

3263 3264 3265
			ret = scrub_extent(sctx, extent_logical, extent_len,
					   extent_physical, extent_dev, flags,
					   generation, extent_mirror_num,
3266
					   extent_logical - logical + physical);
3267 3268 3269

			scrub_free_csums(sctx);

A
Arne Jansen 已提交
3270 3271 3272
			if (ret)
				goto out;

L
Liu Bo 已提交
3273 3274
			if (extent_logical + extent_len <
			    key.objectid + bytes) {
3275
				if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3276 3277 3278 3279
					/*
					 * loop until we find next data stripe
					 * or we have finished all stripes.
					 */
3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298
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 +
								increment - 1;
						ret = scrub_raid56_parity(sctx,
							map, scrub_dev, ppath,
							stripe_logical,
							stripe_end);
						if (ret)
							goto out;
						goto loop;
					}
3299 3300 3301 3302
				} else {
					physical += map->stripe_len;
					logical += increment;
				}
L
Liu Bo 已提交
3303 3304 3305 3306 3307
				if (logical < key.objectid + bytes) {
					cond_resched();
					goto again;
				}

3308
				if (physical >= physical_end) {
L
Liu Bo 已提交
3309 3310 3311 3312
					stop_loop = 1;
					break;
				}
			}
A
Arne Jansen 已提交
3313 3314 3315
next:
			path->slots[0]++;
		}
C
Chris Mason 已提交
3316
		btrfs_release_path(path);
3317
skip:
A
Arne Jansen 已提交
3318 3319
		logical += increment;
		physical += map->stripe_len;
3320
		spin_lock(&sctx->stat_lock);
L
Liu Bo 已提交
3321 3322 3323 3324 3325
		if (stop_loop)
			sctx->stat.last_physical = map->stripes[num].physical +
						   length;
		else
			sctx->stat.last_physical = physical;
3326
		spin_unlock(&sctx->stat_lock);
L
Liu Bo 已提交
3327 3328
		if (stop_loop)
			break;
A
Arne Jansen 已提交
3329
	}
3330
out:
A
Arne Jansen 已提交
3331
	/* push queued extents */
3332
	scrub_submit(sctx);
3333 3334 3335
	mutex_lock(&sctx->wr_ctx.wr_lock);
	scrub_wr_submit(sctx);
	mutex_unlock(&sctx->wr_ctx.wr_lock);
A
Arne Jansen 已提交
3336

3337
	blk_finish_plug(&plug);
A
Arne Jansen 已提交
3338
	btrfs_free_path(path);
3339
	btrfs_free_path(ppath);
A
Arne Jansen 已提交
3340 3341 3342
	return ret < 0 ? ret : 0;
}

3343
static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx,
3344 3345 3346
					  struct btrfs_device *scrub_dev,
					  u64 chunk_tree, u64 chunk_objectid,
					  u64 chunk_offset, u64 length,
3347
					  u64 dev_offset, int is_dev_replace)
A
Arne Jansen 已提交
3348 3349
{
	struct btrfs_mapping_tree *map_tree =
3350
		&sctx->dev_root->fs_info->mapping_tree;
A
Arne Jansen 已提交
3351 3352 3353
	struct map_lookup *map;
	struct extent_map *em;
	int i;
3354
	int ret = 0;
A
Arne Jansen 已提交
3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370

	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) {
3371
		if (map->stripes[i].dev->bdev == scrub_dev->bdev &&
3372
		    map->stripes[i].physical == dev_offset) {
3373
			ret = scrub_stripe(sctx, map, scrub_dev, i,
3374 3375
					   chunk_offset, length,
					   is_dev_replace);
A
Arne Jansen 已提交
3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386
			if (ret)
				goto out;
		}
	}
out:
	free_extent_map(em);

	return ret;
}

static noinline_for_stack
3387
int scrub_enumerate_chunks(struct scrub_ctx *sctx,
3388 3389
			   struct btrfs_device *scrub_dev, u64 start, u64 end,
			   int is_dev_replace)
A
Arne Jansen 已提交
3390 3391 3392
{
	struct btrfs_dev_extent *dev_extent = NULL;
	struct btrfs_path *path;
3393
	struct btrfs_root *root = sctx->dev_root;
A
Arne Jansen 已提交
3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404
	struct btrfs_fs_info *fs_info = root->fs_info;
	u64 length;
	u64 chunk_tree;
	u64 chunk_objectid;
	u64 chunk_offset;
	int ret;
	int slot;
	struct extent_buffer *l;
	struct btrfs_key key;
	struct btrfs_key found_key;
	struct btrfs_block_group_cache *cache;
3405
	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
A
Arne Jansen 已提交
3406 3407 3408 3409 3410 3411 3412 3413 3414

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

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

3415
	key.objectid = scrub_dev->devid;
A
Arne Jansen 已提交
3416 3417 3418 3419 3420 3421
	key.offset = 0ull;
	key.type = BTRFS_DEV_EXTENT_KEY;

	while (1) {
		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
		if (ret < 0)
3422 3423 3424 3425 3426 3427 3428 3429 3430
			break;
		if (ret > 0) {
			if (path->slots[0] >=
			    btrfs_header_nritems(path->nodes[0])) {
				ret = btrfs_next_leaf(root, path);
				if (ret)
					break;
			}
		}
A
Arne Jansen 已提交
3431 3432 3433 3434 3435 3436

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

		btrfs_item_key_to_cpu(l, &found_key, slot);

3437
		if (found_key.objectid != scrub_dev->devid)
A
Arne Jansen 已提交
3438 3439
			break;

3440
		if (found_key.type != BTRFS_DEV_EXTENT_KEY)
A
Arne Jansen 已提交
3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451
			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);

3452 3453
		if (found_key.offset + length <= start)
			goto skip;
A
Arne Jansen 已提交
3454 3455 3456 3457 3458 3459 3460 3461 3462 3463

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

		/*
		 * get a reference on the corresponding block group to prevent
		 * the chunk from going away while we scrub it
		 */
		cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3464 3465 3466 3467 3468 3469

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

3470 3471 3472
		dev_replace->cursor_right = found_key.offset + length;
		dev_replace->cursor_left = found_key.offset;
		dev_replace->item_needs_writeback = 1;
3473
		ret = scrub_chunk(sctx, scrub_dev, chunk_tree, chunk_objectid,
3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494
				  chunk_offset, length, found_key.offset,
				  is_dev_replace);

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

		wait_event(sctx->list_wait,
			   atomic_read(&sctx->bios_in_flight) == 0);
3495 3496 3497 3498 3499 3500 3501 3502
		atomic_inc(&fs_info->scrubs_paused);
		wake_up(&fs_info->scrub_pause_wait);

		/*
		 * must be called before we decrease @scrub_paused.
		 * make sure we don't block transaction commit while
		 * we are waiting pending workers finished.
		 */
3503 3504
		wait_event(sctx->list_wait,
			   atomic_read(&sctx->workers_pending) == 0);
3505 3506 3507 3508 3509 3510 3511
		atomic_set(&sctx->wr_ctx.flush_all_writes, 0);

		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);
3512

A
Arne Jansen 已提交
3513 3514 3515
		btrfs_put_block_group(cache);
		if (ret)
			break;
3516 3517
		if (is_dev_replace &&
		    atomic64_read(&dev_replace->num_write_errors) > 0) {
3518 3519 3520 3521 3522 3523 3524
			ret = -EIO;
			break;
		}
		if (sctx->stat.malloc_errors > 0) {
			ret = -ENOMEM;
			break;
		}
A
Arne Jansen 已提交
3525

3526 3527
		dev_replace->cursor_left = dev_replace->cursor_right;
		dev_replace->item_needs_writeback = 1;
3528
skip:
A
Arne Jansen 已提交
3529
		key.offset = found_key.offset + length;
C
Chris Mason 已提交
3530
		btrfs_release_path(path);
A
Arne Jansen 已提交
3531 3532 3533
	}

	btrfs_free_path(path);
3534 3535 3536 3537 3538 3539

	/*
	 * ret can still be 1 from search_slot or next_leaf,
	 * that's not an error
	 */
	return ret < 0 ? ret : 0;
A
Arne Jansen 已提交
3540 3541
}

3542 3543
static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx,
					   struct btrfs_device *scrub_dev)
A
Arne Jansen 已提交
3544 3545 3546 3547 3548
{
	int	i;
	u64	bytenr;
	u64	gen;
	int	ret;
3549
	struct btrfs_root *root = sctx->dev_root;
A
Arne Jansen 已提交
3550

3551
	if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
3552 3553
		return -EIO;

3554 3555 3556 3557 3558
	/* 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
3559 3560 3561

	for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
		bytenr = btrfs_sb_offset(i);
3562 3563
		if (bytenr + BTRFS_SUPER_INFO_SIZE >
		    scrub_dev->commit_total_bytes)
A
Arne Jansen 已提交
3564 3565
			break;

3566
		ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr,
3567
				  scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i,
3568
				  NULL, 1, bytenr);
A
Arne Jansen 已提交
3569 3570 3571
		if (ret)
			return ret;
	}
3572
	wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
A
Arne Jansen 已提交
3573 3574 3575 3576 3577 3578 3579

	return 0;
}

/*
 * get a reference count on fs_info->scrub_workers. start worker if necessary
 */
3580 3581
static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info,
						int is_dev_replace)
A
Arne Jansen 已提交
3582
{
3583
	unsigned int flags = WQ_FREEZABLE | WQ_UNBOUND;
3584
	int max_active = fs_info->thread_pool_size;
A
Arne Jansen 已提交
3585

A
Arne Jansen 已提交
3586
	if (fs_info->scrub_workers_refcnt == 0) {
3587
		if (is_dev_replace)
3588 3589 3590
			fs_info->scrub_workers =
				btrfs_alloc_workqueue("btrfs-scrub", flags,
						      1, 4);
3591
		else
3592 3593 3594
			fs_info->scrub_workers =
				btrfs_alloc_workqueue("btrfs-scrub", flags,
						      max_active, 4);
3595 3596 3597
		if (!fs_info->scrub_workers)
			goto fail_scrub_workers;

3598 3599 3600
		fs_info->scrub_wr_completion_workers =
			btrfs_alloc_workqueue("btrfs-scrubwrc", flags,
					      max_active, 2);
3601 3602 3603
		if (!fs_info->scrub_wr_completion_workers)
			goto fail_scrub_wr_completion_workers;

3604 3605
		fs_info->scrub_nocow_workers =
			btrfs_alloc_workqueue("btrfs-scrubnc", flags, 1, 0);
3606 3607
		if (!fs_info->scrub_nocow_workers)
			goto fail_scrub_nocow_workers;
3608 3609 3610
		fs_info->scrub_parity_workers =
			btrfs_alloc_workqueue("btrfs-scrubparity", flags,
					      max_active, 2);
3611 3612
		if (!fs_info->scrub_parity_workers)
			goto fail_scrub_parity_workers;
A
Arne Jansen 已提交
3613
	}
A
Arne Jansen 已提交
3614
	++fs_info->scrub_workers_refcnt;
3615 3616 3617 3618 3619 3620 3621 3622 3623 3624
	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
3625 3626
}

3627
static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info)
A
Arne Jansen 已提交
3628
{
3629
	if (--fs_info->scrub_workers_refcnt == 0) {
3630 3631 3632
		btrfs_destroy_workqueue(fs_info->scrub_workers);
		btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
		btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
3633
		btrfs_destroy_workqueue(fs_info->scrub_parity_workers);
3634
	}
A
Arne Jansen 已提交
3635 3636 3637
	WARN_ON(fs_info->scrub_workers_refcnt < 0);
}

3638 3639
int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start,
		    u64 end, struct btrfs_scrub_progress *progress,
3640
		    int readonly, int is_dev_replace)
A
Arne Jansen 已提交
3641
{
3642
	struct scrub_ctx *sctx;
A
Arne Jansen 已提交
3643 3644
	int ret;
	struct btrfs_device *dev;
3645
	struct rcu_string *name;
A
Arne Jansen 已提交
3646

3647
	if (btrfs_fs_closing(fs_info))
A
Arne Jansen 已提交
3648 3649
		return -EINVAL;

3650
	if (fs_info->chunk_root->nodesize > BTRFS_STRIPE_LEN) {
3651 3652 3653 3654 3655
		/*
		 * 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.
		 */
3656 3657
		btrfs_err(fs_info,
			   "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails",
3658
		       fs_info->chunk_root->nodesize, BTRFS_STRIPE_LEN);
3659 3660 3661
		return -EINVAL;
	}

3662
	if (fs_info->chunk_root->sectorsize != PAGE_SIZE) {
3663
		/* not supported for data w/o checksums */
3664 3665 3666
		btrfs_err(fs_info,
			   "scrub: size assumption sectorsize != PAGE_SIZE "
			   "(%d != %lu) fails",
3667
		       fs_info->chunk_root->sectorsize, PAGE_SIZE);
A
Arne Jansen 已提交
3668 3669 3670
		return -EINVAL;
	}

3671 3672 3673 3674 3675 3676 3677 3678
	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
		 */
3679 3680
		btrfs_err(fs_info, "scrub: size assumption nodesize and sectorsize "
			   "<= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails",
3681 3682 3683 3684 3685 3686 3687
		       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
3688

3689 3690
	mutex_lock(&fs_info->fs_devices->device_list_mutex);
	dev = btrfs_find_device(fs_info, devid, NULL, NULL);
3691
	if (!dev || (dev->missing && !is_dev_replace)) {
3692
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
A
Arne Jansen 已提交
3693 3694 3695
		return -ENODEV;
	}

3696 3697 3698 3699 3700 3701 3702 3703 3704 3705
	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;
	}

3706
	mutex_lock(&fs_info->scrub_lock);
3707
	if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) {
A
Arne Jansen 已提交
3708
		mutex_unlock(&fs_info->scrub_lock);
3709 3710
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
		return -EIO;
A
Arne Jansen 已提交
3711 3712
	}

3713 3714 3715 3716 3717
	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
3718
		mutex_unlock(&fs_info->scrub_lock);
3719
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
A
Arne Jansen 已提交
3720 3721
		return -EINPROGRESS;
	}
3722
	btrfs_dev_replace_unlock(&fs_info->dev_replace);
3723 3724 3725 3726 3727 3728 3729 3730

	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;
	}

3731
	sctx = scrub_setup_ctx(dev, is_dev_replace);
3732
	if (IS_ERR(sctx)) {
A
Arne Jansen 已提交
3733
		mutex_unlock(&fs_info->scrub_lock);
3734 3735
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
		scrub_workers_put(fs_info);
3736
		return PTR_ERR(sctx);
A
Arne Jansen 已提交
3737
	}
3738 3739
	sctx->readonly = readonly;
	dev->scrub_device = sctx;
3740
	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
A
Arne Jansen 已提交
3741

3742 3743 3744 3745
	/*
	 * checking @scrub_pause_req here, we can avoid
	 * race between committing transaction and scrubbing.
	 */
3746
	__scrub_blocked_if_needed(fs_info);
A
Arne Jansen 已提交
3747 3748 3749
	atomic_inc(&fs_info->scrubs_running);
	mutex_unlock(&fs_info->scrub_lock);

3750
	if (!is_dev_replace) {
3751 3752 3753 3754
		/*
		 * by holding device list mutex, we can
		 * kick off writing super in log tree sync.
		 */
3755
		mutex_lock(&fs_info->fs_devices->device_list_mutex);
3756
		ret = scrub_supers(sctx, dev);
3757
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3758
	}
A
Arne Jansen 已提交
3759 3760

	if (!ret)
3761 3762
		ret = scrub_enumerate_chunks(sctx, dev, start, end,
					     is_dev_replace);
A
Arne Jansen 已提交
3763

3764
	wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
A
Arne Jansen 已提交
3765 3766 3767
	atomic_dec(&fs_info->scrubs_running);
	wake_up(&fs_info->scrub_pause_wait);

3768
	wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0);
3769

A
Arne Jansen 已提交
3770
	if (progress)
3771
		memcpy(progress, &sctx->stat, sizeof(*progress));
A
Arne Jansen 已提交
3772 3773 3774

	mutex_lock(&fs_info->scrub_lock);
	dev->scrub_device = NULL;
3775
	scrub_workers_put(fs_info);
A
Arne Jansen 已提交
3776 3777
	mutex_unlock(&fs_info->scrub_lock);

3778
	scrub_put_ctx(sctx);
A
Arne Jansen 已提交
3779 3780 3781 3782

	return ret;
}

3783
void btrfs_scrub_pause(struct btrfs_root *root)
A
Arne Jansen 已提交
3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799
{
	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);
}

3800
void btrfs_scrub_continue(struct btrfs_root *root)
A
Arne Jansen 已提交
3801 3802 3803 3804 3805 3806 3807
{
	struct btrfs_fs_info *fs_info = root->fs_info;

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

3808
int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info)
A
Arne Jansen 已提交
3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828
{
	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;
}

3829 3830
int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info,
			   struct btrfs_device *dev)
3831
{
3832
	struct scrub_ctx *sctx;
A
Arne Jansen 已提交
3833 3834

	mutex_lock(&fs_info->scrub_lock);
3835 3836
	sctx = dev->scrub_device;
	if (!sctx) {
A
Arne Jansen 已提交
3837 3838 3839
		mutex_unlock(&fs_info->scrub_lock);
		return -ENOTCONN;
	}
3840
	atomic_inc(&sctx->cancel_req);
A
Arne Jansen 已提交
3841 3842 3843 3844 3845 3846 3847 3848 3849 3850
	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
3851

A
Arne Jansen 已提交
3852 3853 3854 3855
int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
			 struct btrfs_scrub_progress *progress)
{
	struct btrfs_device *dev;
3856
	struct scrub_ctx *sctx = NULL;
A
Arne Jansen 已提交
3857 3858

	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3859
	dev = btrfs_find_device(root->fs_info, devid, NULL, NULL);
A
Arne Jansen 已提交
3860
	if (dev)
3861 3862 3863
		sctx = dev->scrub_device;
	if (sctx)
		memcpy(progress, &sctx->stat, sizeof(*progress));
A
Arne Jansen 已提交
3864 3865
	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);

3866
	return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV;
A
Arne Jansen 已提交
3867
}
3868 3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883

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) {
3884
		btrfs_put_bbio(bbio);
3885 3886 3887 3888 3889 3890
		return;
	}

	*extent_physical = bbio->stripes[0].physical;
	*extent_mirror_num = bbio->mirror_num;
	*extent_dev = bbio->stripes[0].dev;
3891
	btrfs_put_bbio(bbio);
3892 3893 3894 3895 3896 3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943
}

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

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

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

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

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

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

	scrub_pending_trans_workers_inc(sctx);

	nocow_ctx->sctx = sctx;
	nocow_ctx->logical = logical;
	nocow_ctx->len = len;
	nocow_ctx->mirror_num = mirror_num;
	nocow_ctx->physical_for_dev_replace = physical_for_dev_replace;
3944 3945
	btrfs_init_work(&nocow_ctx->work, btrfs_scrubnc_helper,
			copy_nocow_pages_worker, NULL, NULL);
3946
	INIT_LIST_HEAD(&nocow_ctx->inodes);
3947 3948
	btrfs_queue_work(fs_info->scrub_nocow_workers,
			 &nocow_ctx->work);
3949 3950 3951 3952

	return 0;
}

3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969
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

3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004
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,
4005
					  record_inode_for_nocow, nocow_ctx);
4006
	if (ret != 0 && ret != -ENOENT) {
4007 4008
		btrfs_warn(fs_info, "iterate_inodes_from_logical() failed: log %llu, "
			"phys %llu, len %llu, mir %u, ret %d",
4009 4010
			logical, physical_for_dev_replace, len, mirror_num,
			ret);
4011 4012 4013 4014
		not_written = 1;
		goto out;
	}

4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032
	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;
		}
	}
4033
out:
4034 4035 4036 4037 4038 4039 4040 4041
	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);
	}
4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053
	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);
}

4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 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
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;
}

4098 4099
static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
				      struct scrub_copy_nocow_ctx *nocow_ctx)
4100
{
4101
	struct btrfs_fs_info *fs_info = nocow_ctx->sctx->dev_root->fs_info;
4102
	struct btrfs_key key;
4103 4104
	struct inode *inode;
	struct page *page;
4105
	struct btrfs_root *local_root;
4106
	struct extent_io_tree *io_tree;
4107
	u64 physical_for_dev_replace;
4108
	u64 nocow_ctx_logical;
4109
	u64 len = nocow_ctx->len;
4110
	unsigned long index;
4111
	int srcu_index;
4112 4113
	int ret = 0;
	int err = 0;
4114 4115 4116 4117

	key.objectid = root;
	key.type = BTRFS_ROOT_ITEM_KEY;
	key.offset = (u64)-1;
4118 4119 4120

	srcu_index = srcu_read_lock(&fs_info->subvol_srcu);

4121
	local_root = btrfs_read_fs_root_no_name(fs_info, &key);
4122 4123
	if (IS_ERR(local_root)) {
		srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
4124
		return PTR_ERR(local_root);
4125
	}
4126 4127 4128 4129 4130

	key.type = BTRFS_INODE_ITEM_KEY;
	key.objectid = inum;
	key.offset = 0;
	inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
4131
	srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
4132 4133 4134
	if (IS_ERR(inode))
		return PTR_ERR(inode);

4135 4136 4137 4138
	/* Avoid truncate/dio/punch hole.. */
	mutex_lock(&inode->i_mutex);
	inode_dio_wait(inode);

4139
	physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
4140
	io_tree = &BTRFS_I(inode)->io_tree;
4141
	nocow_ctx_logical = nocow_ctx->logical;
4142

4143 4144 4145 4146
	ret = check_extent_to_block(inode, offset, len, nocow_ctx_logical);
	if (ret) {
		ret = ret > 0 ? 0 : ret;
		goto out;
4147 4148
	}

4149 4150
	while (len >= PAGE_CACHE_SIZE) {
		index = offset >> PAGE_CACHE_SHIFT;
4151
again:
4152 4153
		page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
		if (!page) {
4154
			btrfs_err(fs_info, "find_or_create_page() failed");
4155
			ret = -ENOMEM;
4156
			goto out;
4157 4158 4159 4160 4161 4162 4163
		}

		if (PageUptodate(page)) {
			if (PageDirty(page))
				goto next_page;
		} else {
			ClearPageError(page);
4164
			err = extent_read_full_page(io_tree, page,
4165 4166
							   btrfs_get_extent,
							   nocow_ctx->mirror_num);
4167 4168
			if (err) {
				ret = err;
4169 4170
				goto next_page;
			}
4171

4172
			lock_page(page);
4173 4174 4175 4176 4177 4178 4179
			/*
			 * 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) {
4180
				unlock_page(page);
4181 4182 4183
				page_cache_release(page);
				goto again;
			}
4184 4185 4186 4187 4188
			if (!PageUptodate(page)) {
				ret = -EIO;
				goto next_page;
			}
		}
4189 4190 4191 4192 4193 4194 4195 4196

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

4197 4198 4199 4200
		err = write_page_nocow(nocow_ctx->sctx,
				       physical_for_dev_replace, page);
		if (err)
			ret = err;
4201
next_page:
4202 4203 4204 4205 4206 4207
		unlock_page(page);
		page_cache_release(page);

		if (ret)
			break;

4208 4209
		offset += PAGE_CACHE_SIZE;
		physical_for_dev_replace += PAGE_CACHE_SIZE;
4210
		nocow_ctx_logical += PAGE_CACHE_SIZE;
4211 4212
		len -= PAGE_CACHE_SIZE;
	}
4213
	ret = COPY_COMPLETE;
4214
out:
4215
	mutex_unlock(&inode->i_mutex);
4216
	iput(inode);
4217 4218 4219 4220 4221 4222 4223 4224 4225 4226 4227 4228 4229 4230 4231
	return ret;
}

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

	dev = sctx->wr_ctx.tgtdev;
	if (!dev)
		return -EIO;
	if (!dev->bdev) {
		printk_ratelimited(KERN_WARNING
4232
			"BTRFS: scrub write_page_nocow(bdev == NULL) is unexpected!\n");
4233 4234
		return -EIO;
	}
4235
	bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
4236 4237 4238 4239 4240 4241
	if (!bio) {
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		spin_unlock(&sctx->stat_lock);
		return -ENOMEM;
	}
4242 4243
	bio->bi_iter.bi_size = 0;
	bio->bi_iter.bi_sector = physical_for_dev_replace >> 9;
4244 4245 4246 4247 4248 4249 4250 4251 4252
	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;
	}

4253
	if (btrfsic_submit_bio_wait(WRITE_SYNC, bio))
4254 4255 4256 4257 4258
		goto leave_with_eio;

	bio_put(bio);
	return 0;
}