scrub.c 115.7 KB
Newer Older
A
Arne Jansen 已提交
1
/*
2
 * Copyright (C) 2011, 2012 STRATO.  All rights reserved.
A
Arne Jansen 已提交
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License v2 as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 */

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

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

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

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

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

66
struct scrub_recover {
67
	refcount_t		refs;
68 69 70 71
	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
	refcount_t		refs; /* free mem on transition to zero */
116
	struct scrub_ctx	*sctx;
117
	struct scrub_parity	*sparity;
118 119 120 121
	struct {
		unsigned int	header_error:1;
		unsigned int	checksum_error:1;
		unsigned int	no_io_error_seen:1;
122
		unsigned int	generation_error:1; /* also sets header_error */
123 124 125 126

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

131 132 133 134 135 136 137 138 139 140 141 142 143 144
/* Used for the chunks with parity stripe such RAID5/6 */
struct scrub_parity {
	struct scrub_ctx	*sctx;

	struct btrfs_device	*scrub_dev;

	u64			logic_start;

	u64			logic_end;

	int			nsectors;

	int			stripe_len;

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

	struct list_head	spages;

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

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

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

	unsigned long		bitmap[0];
};

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

172
struct scrub_ctx {
173
	struct scrub_bio	*bios[SCRUB_BIOS_PER_SCTX];
174
	struct btrfs_fs_info	*fs_info;
A
Arne Jansen 已提交
175 176
	int			first_free;
	int			curr;
177 178
	atomic_t		bios_in_flight;
	atomic_t		workers_pending;
A
Arne Jansen 已提交
179 180 181 182 183
	spinlock_t		list_lock;
	wait_queue_head_t	list_wait;
	u16			csum_size;
	struct list_head	csum_list;
	atomic_t		cancel_req;
A
Arne Jansen 已提交
184
	int			readonly;
185
	int			pages_per_rd_bio;
186 187
	u32			sectorsize;
	u32			nodesize;
188 189

	int			is_dev_replace;
190
	struct scrub_wr_ctx	wr_ctx;
191

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

	/*
	 * Use a ref counter to avoid use-after-free issues. Scrub workers
	 * decrement bios_in_flight and workers_pending and then do a wakeup
	 * on the list_wait wait queue. We must ensure the main scrub task
	 * doesn't free the scrub context before or while the workers are
	 * doing the wakeup() call.
	 */
	atomic_t                refs;
A
Arne Jansen 已提交
206 207
};

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

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

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

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

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


306 307
static void scrub_pending_bio_inc(struct scrub_ctx *sctx)
{
308
	atomic_inc(&sctx->refs);
309 310 311 312 313 314 315
	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);
316
	scrub_put_ctx(sctx);
317 318
}

319
static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
320 321 322 323 324 325 326 327 328
{
	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);
	}
}

329
static void scrub_pause_on(struct btrfs_fs_info *fs_info)
330 331 332
{
	atomic_inc(&fs_info->scrubs_paused);
	wake_up(&fs_info->scrub_pause_wait);
333
}
334

335 336
static void scrub_pause_off(struct btrfs_fs_info *fs_info)
{
337 338 339 340 341 342 343 344
	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);
}

345 346 347 348 349 350
static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
{
	scrub_pause_on(fs_info);
	scrub_pause_off(fs_info);
}

351 352 353 354 355 356
/*
 * used for workers that require transaction commits (i.e., for the
 * NOCOW case)
 */
static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx)
{
357
	struct btrfs_fs_info *fs_info = sctx->fs_info;
358

359
	atomic_inc(&sctx->refs);
360 361 362 363 364 365 366 367 368 369 370 371 372
	/*
	 * 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);
373 374 375 376 377 378 379 380 381 382

	/*
	 * 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);

383 384 385 386 387 388
	atomic_inc(&sctx->workers_pending);
}

/* used for workers that require transaction commits */
static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx)
{
389
	struct btrfs_fs_info *fs_info = sctx->fs_info;
390 391 392 393 394 395 396 397 398 399 400 401

	/*
	 * 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);
402
	scrub_put_ctx(sctx);
403 404
}

405
static void scrub_free_csums(struct scrub_ctx *sctx)
A
Arne Jansen 已提交
406
{
407
	while (!list_empty(&sctx->csum_list)) {
A
Arne Jansen 已提交
408
		struct btrfs_ordered_sum *sum;
409
		sum = list_first_entry(&sctx->csum_list,
A
Arne Jansen 已提交
410 411 412 413 414 415
				       struct btrfs_ordered_sum, list);
		list_del(&sum->list);
		kfree(sum);
	}
}

416
static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx)
A
Arne Jansen 已提交
417 418 419
{
	int i;

420
	if (!sctx)
A
Arne Jansen 已提交
421 422
		return;

423 424
	scrub_free_wr_ctx(&sctx->wr_ctx);

425
	/* this can happen when scrub is cancelled */
426 427
	if (sctx->curr != -1) {
		struct scrub_bio *sbio = sctx->bios[sctx->curr];
428 429

		for (i = 0; i < sbio->page_count; i++) {
430
			WARN_ON(!sbio->pagev[i]->page);
431 432 433 434 435
			scrub_block_put(sbio->pagev[i]->sblock);
		}
		bio_put(sbio->bio);
	}

436
	for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
437
		struct scrub_bio *sbio = sctx->bios[i];
A
Arne Jansen 已提交
438 439 440 441 442 443

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

444 445
	scrub_free_csums(sctx);
	kfree(sctx);
A
Arne Jansen 已提交
446 447
}

448 449 450 451 452 453
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
454
static noinline_for_stack
455
struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace)
A
Arne Jansen 已提交
456
{
457
	struct scrub_ctx *sctx;
A
Arne Jansen 已提交
458
	int		i;
459
	struct btrfs_fs_info *fs_info = dev->fs_info;
460
	int ret;
A
Arne Jansen 已提交
461

462
	sctx = kzalloc(sizeof(*sctx), GFP_KERNEL);
463
	if (!sctx)
A
Arne Jansen 已提交
464
		goto nomem;
465
	atomic_set(&sctx->refs, 1);
466
	sctx->is_dev_replace = is_dev_replace;
467
	sctx->pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO;
468
	sctx->curr = -1;
469
	sctx->fs_info = dev->fs_info;
470
	for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
A
Arne Jansen 已提交
471 472
		struct scrub_bio *sbio;

473
		sbio = kzalloc(sizeof(*sbio), GFP_KERNEL);
A
Arne Jansen 已提交
474 475
		if (!sbio)
			goto nomem;
476
		sctx->bios[i] = sbio;
A
Arne Jansen 已提交
477 478

		sbio->index = i;
479
		sbio->sctx = sctx;
480
		sbio->page_count = 0;
481 482
		btrfs_init_work(&sbio->work, btrfs_scrub_helper,
				scrub_bio_end_io_worker, NULL, NULL);
A
Arne Jansen 已提交
483

484
		if (i != SCRUB_BIOS_PER_SCTX - 1)
485
			sctx->bios[i]->next_free = i + 1;
486
		else
487 488 489
			sctx->bios[i]->next_free = -1;
	}
	sctx->first_free = 0;
490 491
	sctx->nodesize = fs_info->nodesize;
	sctx->sectorsize = fs_info->sectorsize;
492 493
	atomic_set(&sctx->bios_in_flight, 0);
	atomic_set(&sctx->workers_pending, 0);
494 495 496 497 498 499 500
	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);
501

502
	ret = scrub_setup_wr_ctx(&sctx->wr_ctx,
503 504 505 506 507
				 fs_info->dev_replace.tgtdev, is_dev_replace);
	if (ret) {
		scrub_free_ctx(sctx);
		return ERR_PTR(ret);
	}
508
	return sctx;
A
Arne Jansen 已提交
509 510

nomem:
511
	scrub_free_ctx(sctx);
A
Arne Jansen 已提交
512 513 514
	return ERR_PTR(-ENOMEM);
}

515 516
static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root,
				     void *warn_ctx)
517 518 519 520 521 522 523
{
	u64 isize;
	u32 nlink;
	int ret;
	int i;
	struct extent_buffer *eb;
	struct btrfs_inode_item *inode_item;
524
	struct scrub_warning *swarn = warn_ctx;
525
	struct btrfs_fs_info *fs_info = swarn->dev->fs_info;
526 527 528
	struct inode_fs_paths *ipath = NULL;
	struct btrfs_root *local_root;
	struct btrfs_key root_key;
529
	struct btrfs_key key;
530 531 532 533 534 535 536 537 538 539

	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;
	}

540 541 542
	/*
	 * this makes the path point to (inum INODE_ITEM ioff)
	 */
543 544 545 546 547
	key.objectid = inum;
	key.type = BTRFS_INODE_ITEM_KEY;
	key.offset = 0;

	ret = btrfs_search_slot(NULL, local_root, &key, swarn->path, 0, 0);
548 549 550 551 552 553 554 555 556 557 558 559 560
	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);
561 562 563 564 565
	if (IS_ERR(ipath)) {
		ret = PTR_ERR(ipath);
		ipath = NULL;
		goto err;
	}
566 567 568 569 570 571 572 573 574 575
	ret = paths_from_inode(inum, ipath);

	if (ret < 0)
		goto err;

	/*
	 * we deliberately ignore the bit ipath might have been too small to
	 * hold all of the paths here
	 */
	for (i = 0; i < ipath->fspath->elem_cnt; ++i)
J
Jeff Mahoney 已提交
576 577 578 579 580 581 582 583
		btrfs_warn_in_rcu(fs_info,
				  "%s at logical %llu on dev %s, sector %llu, root %llu, inode %llu, offset %llu, length %llu, links %u (path: %s)",
				  swarn->errstr, swarn->logical,
				  rcu_str_deref(swarn->dev->name),
				  (unsigned long long)swarn->sector,
				  root, inum, offset,
				  min(isize - offset, (u64)PAGE_SIZE), nlink,
				  (char *)(unsigned long)ipath->fspath->val[i]);
584 585 586 587 588

	free_ipath(ipath);
	return 0;

err:
J
Jeff Mahoney 已提交
589 590 591 592 593 594
	btrfs_warn_in_rcu(fs_info,
			  "%s at logical %llu on dev %s, sector %llu, root %llu, inode %llu, offset %llu: path resolving failed with ret=%d",
			  swarn->errstr, swarn->logical,
			  rcu_str_deref(swarn->dev->name),
			  (unsigned long long)swarn->sector,
			  root, inum, offset, ret);
595 596 597 598 599

	free_ipath(ipath);
	return 0;
}

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

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

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

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

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

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

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

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

out:
	btrfs_free_path(path);
}

670
static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx)
671
{
672
	struct page *page = NULL;
673
	unsigned long index;
674
	struct scrub_fixup_nodatasum *fixup = fixup_ctx;
675
	int ret;
676
	int corrected = 0;
677
	struct btrfs_key key;
678
	struct inode *inode = NULL;
679
	struct btrfs_fs_info *fs_info;
680 681
	u64 end = offset + PAGE_SIZE - 1;
	struct btrfs_root *local_root;
682
	int srcu_index;
683 684 685 686

	key.objectid = root;
	key.type = BTRFS_ROOT_ITEM_KEY;
	key.offset = (u64)-1;
687 688 689 690 691 692 693

	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);
694
		return PTR_ERR(local_root);
695
	}
696 697 698 699

	key.type = BTRFS_INODE_ITEM_KEY;
	key.objectid = inum;
	key.offset = 0;
700 701
	inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
	srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
702 703 704
	if (IS_ERR(inode))
		return PTR_ERR(inode);

705
	index = offset >> PAGE_SHIFT;
706 707

	page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733
	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;
		}
734
		ret = repair_io_failure(BTRFS_I(inode), offset, PAGE_SIZE,
735
					fixup->logical, page,
736
					offset - page_offset(page),
737 738 739 740 741 742 743 744 745 746
					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,
747
					EXTENT_DAMAGED);
748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764
		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,
765
						EXTENT_DAMAGED);
766 767 768 769 770
	}

out:
	if (page)
		put_page(page);
771 772

	iput(inode);
773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789

	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)
{
790
	struct btrfs_fs_info *fs_info;
791 792
	int ret;
	struct scrub_fixup_nodatasum *fixup;
793
	struct scrub_ctx *sctx;
794 795 796 797 798
	struct btrfs_trans_handle *trans = NULL;
	struct btrfs_path *path;
	int uncorrectable = 0;

	fixup = container_of(work, struct scrub_fixup_nodatasum, work);
799
	sctx = fixup->sctx;
800
	fs_info = fixup->root->fs_info;
801 802 803

	path = btrfs_alloc_path();
	if (!path) {
804 805 806
		spin_lock(&sctx->stat_lock);
		++sctx->stat.malloc_errors;
		spin_unlock(&sctx->stat_lock);
807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825
		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.
	 */
826 827
	ret = iterate_inodes_from_logical(fixup->logical, fs_info, path,
					  scrub_fixup_readpage, fixup);
828 829 830 831 832 833
	if (ret < 0) {
		uncorrectable = 1;
		goto out;
	}
	WARN_ON(ret != 1);

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

out:
	if (trans && !IS_ERR(trans))
840
		btrfs_end_transaction(trans);
841
	if (uncorrectable) {
842 843 844
		spin_lock(&sctx->stat_lock);
		++sctx->stat.uncorrectable_errors;
		spin_unlock(&sctx->stat_lock);
845
		btrfs_dev_replace_stats_inc(
846 847
			&fs_info->dev_replace.num_uncorrectable_read_errors);
		btrfs_err_rl_in_rcu(fs_info,
848
		    "unable to fixup (nodatasum) error at logical %llu on dev %s",
849
			fixup->logical, rcu_str_deref(fixup->dev->name));
850 851 852 853 854
	}

	btrfs_free_path(path);
	kfree(fixup);

855
	scrub_pending_trans_workers_dec(sctx);
856 857
}

858 859
static inline void scrub_get_recover(struct scrub_recover *recover)
{
860
	refcount_inc(&recover->refs);
861 862 863 864
}

static inline void scrub_put_recover(struct scrub_recover *recover)
{
865
	if (refcount_dec_and_test(&recover->refs)) {
866
		btrfs_put_bbio(recover->bbio);
867 868 869 870
		kfree(recover);
	}
}

A
Arne Jansen 已提交
871
/*
872 873 874 875 876 877
 * 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
878
 */
879
static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
A
Arne Jansen 已提交
880
{
881
	struct scrub_ctx *sctx = sblock_to_check->sctx;
882
	struct btrfs_device *dev;
883 884 885 886 887 888 889 890 891 892 893 894
	struct btrfs_fs_info *fs_info;
	u64 length;
	u64 logical;
	unsigned int failed_mirror_index;
	unsigned int is_metadata;
	unsigned int have_csum;
	struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */
	struct scrub_block *sblock_bad;
	int ret;
	int mirror_index;
	int page_num;
	int success;
895
	static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
896 897 898
				      DEFAULT_RATELIMIT_BURST);

	BUG_ON(sblock_to_check->page_count < 1);
899
	fs_info = sctx->fs_info;
900 901 902 903 904 905 906 907 908 909 910
	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;
	}
911
	length = sblock_to_check->page_count * PAGE_SIZE;
912 913 914 915
	logical = sblock_to_check->pagev[0]->logical;
	BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1);
	failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1;
	is_metadata = !(sblock_to_check->pagev[0]->flags &
916
			BTRFS_EXTENT_FLAG_DATA);
917 918
	have_csum = sblock_to_check->pagev[0]->have_csum;
	dev = sblock_to_check->pagev[0]->dev;
919

920 921 922 923 924
	if (sctx->is_dev_replace && !is_metadata && !have_csum) {
		sblocks_for_recheck = NULL;
		goto nodatasum_case;
	}

925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953
	/*
	 * 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.
	 */

954 955
	sblocks_for_recheck = kcalloc(BTRFS_MAX_MIRRORS,
				      sizeof(*sblocks_for_recheck), GFP_NOFS);
956
	if (!sblocks_for_recheck) {
957 958 959 960 961
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		sctx->stat.read_errors++;
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
962
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
963
		goto out;
A
Arne Jansen 已提交
964 965
	}

966
	/* setup the context, map the logical blocks and alloc the pages */
967
	ret = scrub_setup_recheck_block(sblock_to_check, sblocks_for_recheck);
968
	if (ret) {
969 970 971 972
		spin_lock(&sctx->stat_lock);
		sctx->stat.read_errors++;
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
973
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
974 975 976 977
		goto out;
	}
	BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS);
	sblock_bad = sblocks_for_recheck + failed_mirror_index;
978

979
	/* build and submit the bios for the failed mirror, check checksums */
980
	scrub_recheck_block(fs_info, sblock_bad, 1);
A
Arne Jansen 已提交
981

982 983 984 985 986 987 988 989 990 991
	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)
		 */
992 993
		spin_lock(&sctx->stat_lock);
		sctx->stat.unverified_errors++;
994
		sblock_to_check->data_corrected = 1;
995
		spin_unlock(&sctx->stat_lock);
A
Arne Jansen 已提交
996

997 998
		if (sctx->is_dev_replace)
			scrub_write_block_to_dev_replace(sblock_bad);
999
		goto out;
A
Arne Jansen 已提交
1000 1001
	}

1002
	if (!sblock_bad->no_io_error_seen) {
1003 1004 1005
		spin_lock(&sctx->stat_lock);
		sctx->stat.read_errors++;
		spin_unlock(&sctx->stat_lock);
1006 1007
		if (__ratelimit(&_rs))
			scrub_print_warning("i/o error", sblock_to_check);
1008
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
1009
	} else if (sblock_bad->checksum_error) {
1010 1011 1012
		spin_lock(&sctx->stat_lock);
		sctx->stat.csum_errors++;
		spin_unlock(&sctx->stat_lock);
1013 1014
		if (__ratelimit(&_rs))
			scrub_print_warning("checksum error", sblock_to_check);
1015
		btrfs_dev_stat_inc_and_print(dev,
1016
					     BTRFS_DEV_STAT_CORRUPTION_ERRS);
1017
	} else if (sblock_bad->header_error) {
1018 1019 1020
		spin_lock(&sctx->stat_lock);
		sctx->stat.verify_errors++;
		spin_unlock(&sctx->stat_lock);
1021 1022 1023
		if (__ratelimit(&_rs))
			scrub_print_warning("checksum/header error",
					    sblock_to_check);
1024
		if (sblock_bad->generation_error)
1025
			btrfs_dev_stat_inc_and_print(dev,
1026 1027
				BTRFS_DEV_STAT_GENERATION_ERRS);
		else
1028
			btrfs_dev_stat_inc_and_print(dev,
1029
				BTRFS_DEV_STAT_CORRUPTION_ERRS);
1030
	}
A
Arne Jansen 已提交
1031

1032 1033 1034 1035
	if (sctx->readonly) {
		ASSERT(!sctx->is_dev_replace);
		goto out;
	}
A
Arne Jansen 已提交
1036

1037 1038
	if (!is_metadata && !have_csum) {
		struct scrub_fixup_nodatasum *fixup_nodatasum;
A
Arne Jansen 已提交
1039

1040 1041
		WARN_ON(sctx->is_dev_replace);

1042 1043
nodatasum_case:

1044 1045
		/*
		 * !is_metadata and !have_csum, this means that the data
1046
		 * might not be COWed, that it might be modified
1047 1048 1049 1050 1051 1052 1053
		 * 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;
1054
		fixup_nodatasum->sctx = sctx;
1055
		fixup_nodatasum->dev = dev;
1056 1057 1058
		fixup_nodatasum->logical = logical;
		fixup_nodatasum->root = fs_info->extent_root;
		fixup_nodatasum->mirror_num = failed_mirror_index + 1;
1059
		scrub_pending_trans_workers_inc(sctx);
1060 1061
		btrfs_init_work(&fixup_nodatasum->work, btrfs_scrub_helper,
				scrub_fixup_nodatasum, NULL, NULL);
1062 1063
		btrfs_queue_work(fs_info->scrub_workers,
				 &fixup_nodatasum->work);
1064
		goto out;
A
Arne Jansen 已提交
1065 1066
	}

1067 1068
	/*
	 * now build and submit the bios for the other mirrors, check
1069 1070
	 * checksums.
	 * First try to pick the mirror which is completely without I/O
1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085
	 * 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++) {
1086
		struct scrub_block *sblock_other;
1087

1088 1089 1090 1091 1092
		if (mirror_index == failed_mirror_index)
			continue;
		sblock_other = sblocks_for_recheck + mirror_index;

		/* build and submit the bios, check checksums */
1093
		scrub_recheck_block(fs_info, sblock_other, 0);
1094 1095

		if (!sblock_other->header_error &&
1096 1097
		    !sblock_other->checksum_error &&
		    sblock_other->no_io_error_seen) {
1098 1099
			if (sctx->is_dev_replace) {
				scrub_write_block_to_dev_replace(sblock_other);
1100
				goto corrected_error;
1101 1102
			} else {
				ret = scrub_repair_block_from_good_copy(
1103 1104 1105
						sblock_bad, sblock_other);
				if (!ret)
					goto corrected_error;
1106
			}
1107 1108
		}
	}
A
Arne Jansen 已提交
1109

1110 1111
	if (sblock_bad->no_io_error_seen && !sctx->is_dev_replace)
		goto did_not_correct_error;
1112 1113 1114

	/*
	 * In case of I/O errors in the area that is supposed to be
1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126
	 * 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
1127
	 * the final checksum succeeds. But this would be a rare
1128 1129 1130 1131 1132 1133 1134 1135
	 * 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
1136
	 */
1137
	success = 1;
1138 1139
	for (page_num = 0; page_num < sblock_bad->page_count;
	     page_num++) {
1140
		struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1141
		struct scrub_block *sblock_other = NULL;
1142

1143 1144
		/* skip no-io-error page in scrub */
		if (!page_bad->io_error && !sctx->is_dev_replace)
A
Arne Jansen 已提交
1145
			continue;
1146

1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157
		/* 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;
1158 1159
				}
			}
1160 1161
			if (!sblock_other)
				success = 0;
I
Ilya Dryomov 已提交
1162
		}
A
Arne Jansen 已提交
1163

1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177
		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(
1178
					&fs_info->dev_replace.num_write_errors);
1179 1180 1181 1182 1183 1184 1185 1186 1187 1188
				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;
1189
		}
A
Arne Jansen 已提交
1190 1191
	}

1192
	if (success && !sctx->is_dev_replace) {
1193 1194 1195 1196 1197 1198 1199 1200 1201 1202
		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.
			 */
1203
			scrub_recheck_block(fs_info, sblock_bad, 1);
1204
			if (!sblock_bad->header_error &&
1205 1206 1207 1208 1209 1210 1211
			    !sblock_bad->checksum_error &&
			    sblock_bad->no_io_error_seen)
				goto corrected_error;
			else
				goto did_not_correct_error;
		} else {
corrected_error:
1212 1213
			spin_lock(&sctx->stat_lock);
			sctx->stat.corrected_errors++;
1214
			sblock_to_check->data_corrected = 1;
1215
			spin_unlock(&sctx->stat_lock);
1216 1217
			btrfs_err_rl_in_rcu(fs_info,
				"fixed up error at logical %llu on dev %s",
1218
				logical, rcu_str_deref(dev->name));
A
Arne Jansen 已提交
1219
		}
1220 1221
	} else {
did_not_correct_error:
1222 1223 1224
		spin_lock(&sctx->stat_lock);
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
1225 1226
		btrfs_err_rl_in_rcu(fs_info,
			"unable to fixup (regular) error at logical %llu on dev %s",
1227
			logical, rcu_str_deref(dev->name));
I
Ilya Dryomov 已提交
1228
	}
A
Arne Jansen 已提交
1229

1230 1231 1232 1233 1234 1235
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;
1236
			struct scrub_recover *recover;
1237 1238
			int page_index;

1239 1240 1241
			for (page_index = 0; page_index < sblock->page_count;
			     page_index++) {
				sblock->pagev[page_index]->sblock = NULL;
1242 1243 1244 1245 1246 1247
				recover = sblock->pagev[page_index]->recover;
				if (recover) {
					scrub_put_recover(recover);
					sblock->pagev[page_index]->recover =
									NULL;
				}
1248 1249
				scrub_page_put(sblock->pagev[page_index]);
			}
1250 1251 1252
		}
		kfree(sblocks_for_recheck);
	}
A
Arne Jansen 已提交
1253

1254 1255
	return 0;
}
A
Arne Jansen 已提交
1256

1257
static inline int scrub_nr_raid_mirrors(struct btrfs_bio *bbio)
1258
{
Z
Zhao Lei 已提交
1259 1260 1261 1262 1263
	if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID5)
		return 2;
	else if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID6)
		return 3;
	else
1264 1265 1266
		return (int)bbio->num_stripes;
}

Z
Zhao Lei 已提交
1267 1268
static inline void scrub_stripe_index_and_offset(u64 logical, u64 map_type,
						 u64 *raid_map,
1269 1270 1271 1272 1273 1274 1275
						 u64 mapped_length,
						 int nstripes, int mirror,
						 int *stripe_index,
						 u64 *stripe_offset)
{
	int i;

1276
	if (map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296
		/* 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;
	}
}

1297
static int scrub_setup_recheck_block(struct scrub_block *original_sblock,
1298 1299
				     struct scrub_block *sblocks_for_recheck)
{
1300
	struct scrub_ctx *sctx = original_sblock->sctx;
1301
	struct btrfs_fs_info *fs_info = sctx->fs_info;
1302 1303
	u64 length = original_sblock->page_count * PAGE_SIZE;
	u64 logical = original_sblock->pagev[0]->logical;
1304 1305 1306
	u64 generation = original_sblock->pagev[0]->generation;
	u64 flags = original_sblock->pagev[0]->flags;
	u64 have_csum = original_sblock->pagev[0]->have_csum;
1307 1308 1309 1310 1311 1312
	struct scrub_recover *recover;
	struct btrfs_bio *bbio;
	u64 sublen;
	u64 mapped_length;
	u64 stripe_offset;
	int stripe_index;
1313
	int page_index = 0;
1314
	int mirror_index;
1315
	int nmirrors;
1316 1317 1318
	int ret;

	/*
1319
	 * note: the two members refs and outstanding_pages
1320 1321 1322 1323 1324
	 * are not used (and not set) in the blocks that are used for
	 * the recheck procedure
	 */

	while (length > 0) {
1325 1326 1327
		sublen = min_t(u64, length, PAGE_SIZE);
		mapped_length = sublen;
		bbio = NULL;
A
Arne Jansen 已提交
1328

1329 1330 1331 1332
		/*
		 * with a length of PAGE_SIZE, each returned stripe
		 * represents one mirror
		 */
1333 1334
		ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
				logical, &mapped_length, &bbio, 0, 1);
1335
		if (ret || !bbio || mapped_length < sublen) {
1336
			btrfs_put_bbio(bbio);
1337 1338
			return -EIO;
		}
A
Arne Jansen 已提交
1339

1340 1341
		recover = kzalloc(sizeof(struct scrub_recover), GFP_NOFS);
		if (!recover) {
1342
			btrfs_put_bbio(bbio);
1343 1344 1345
			return -ENOMEM;
		}

1346
		refcount_set(&recover->refs, 1);
1347 1348 1349
		recover->bbio = bbio;
		recover->map_length = mapped_length;

1350
		BUG_ON(page_index >= SCRUB_MAX_PAGES_PER_BLOCK);
1351

1352
		nmirrors = min(scrub_nr_raid_mirrors(bbio), BTRFS_MAX_MIRRORS);
Z
Zhao Lei 已提交
1353

1354
		for (mirror_index = 0; mirror_index < nmirrors;
1355 1356 1357 1358 1359
		     mirror_index++) {
			struct scrub_block *sblock;
			struct scrub_page *page;

			sblock = sblocks_for_recheck + mirror_index;
1360
			sblock->sctx = sctx;
1361

1362 1363 1364
			page = kzalloc(sizeof(*page), GFP_NOFS);
			if (!page) {
leave_nomem:
1365 1366 1367
				spin_lock(&sctx->stat_lock);
				sctx->stat.malloc_errors++;
				spin_unlock(&sctx->stat_lock);
1368
				scrub_put_recover(recover);
1369 1370
				return -ENOMEM;
			}
1371 1372
			scrub_page_get(page);
			sblock->pagev[page_index] = page;
1373 1374 1375
			page->sblock = sblock;
			page->flags = flags;
			page->generation = generation;
1376
			page->logical = logical;
1377 1378 1379 1380 1381
			page->have_csum = have_csum;
			if (have_csum)
				memcpy(page->csum,
				       original_sblock->pagev[0]->csum,
				       sctx->csum_size);
1382

Z
Zhao Lei 已提交
1383 1384 1385
			scrub_stripe_index_and_offset(logical,
						      bbio->map_type,
						      bbio->raid_map,
1386
						      mapped_length,
1387 1388
						      bbio->num_stripes -
						      bbio->num_tgtdevs,
1389 1390 1391 1392 1393 1394 1395
						      mirror_index,
						      &stripe_index,
						      &stripe_offset);
			page->physical = bbio->stripes[stripe_index].physical +
					 stripe_offset;
			page->dev = bbio->stripes[stripe_index].dev;

1396 1397 1398 1399
			BUG_ON(page_index >= original_sblock->page_count);
			page->physical_for_dev_replace =
				original_sblock->pagev[page_index]->
				physical_for_dev_replace;
1400 1401
			/* for missing devices, dev->bdev is NULL */
			page->mirror_num = mirror_index + 1;
1402
			sblock->page_count++;
1403 1404 1405
			page->page = alloc_page(GFP_NOFS);
			if (!page->page)
				goto leave_nomem;
1406 1407 1408

			scrub_get_recover(recover);
			page->recover = recover;
1409
		}
1410
		scrub_put_recover(recover);
1411 1412 1413 1414 1415 1416
		length -= sublen;
		logical += sublen;
		page_index++;
	}

	return 0;
I
Ilya Dryomov 已提交
1417 1418
}

1419 1420 1421 1422 1423
struct scrub_bio_ret {
	struct completion event;
	int error;
};

1424
static void scrub_bio_wait_endio(struct bio *bio)
1425 1426 1427
{
	struct scrub_bio_ret *ret = bio->bi_private;

1428
	ret->error = bio->bi_error;
1429 1430 1431 1432 1433
	complete(&ret->event);
}

static inline int scrub_is_page_on_raid56(struct scrub_page *page)
{
Z
Zhao Lei 已提交
1434
	return page->recover &&
1435
	       (page->recover->bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK);
1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450
}

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;

1451
	ret = raid56_parity_recover(fs_info, bio, page->recover->bbio,
1452
				    page->recover->map_length,
1453
				    page->mirror_num, 0);
1454 1455 1456 1457 1458 1459 1460 1461 1462 1463
	if (ret)
		return ret;

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

	return 0;
}

1464 1465 1466 1467 1468 1469 1470
/*
 * 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.
 */
1471
static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
1472 1473
				struct scrub_block *sblock,
				int retry_failed_mirror)
I
Ilya Dryomov 已提交
1474
{
1475
	int page_num;
I
Ilya Dryomov 已提交
1476

1477
	sblock->no_io_error_seen = 1;
I
Ilya Dryomov 已提交
1478

1479 1480
	for (page_num = 0; page_num < sblock->page_count; page_num++) {
		struct bio *bio;
1481
		struct scrub_page *page = sblock->pagev[page_num];
1482

1483
		if (page->dev->bdev == NULL) {
1484 1485 1486 1487 1488
			page->io_error = 1;
			sblock->no_io_error_seen = 0;
			continue;
		}

1489
		WARN_ON(!page->page);
1490
		bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1491 1492 1493 1494 1495
		if (!bio) {
			page->io_error = 1;
			sblock->no_io_error_seen = 0;
			continue;
		}
1496
		bio->bi_bdev = page->dev->bdev;
1497

1498
		bio_add_page(bio, page->page, PAGE_SIZE, 0);
1499 1500 1501 1502 1503
		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;
M
Mike Christie 已提交
1504
			bio_set_op_attrs(bio, REQ_OP_READ, 0);
1505

1506
			if (btrfsic_submit_bio_wait(bio))
1507 1508
				sblock->no_io_error_seen = 0;
		}
1509

1510 1511
		bio_put(bio);
	}
I
Ilya Dryomov 已提交
1512

1513
	if (sblock->no_io_error_seen)
1514
		scrub_recheck_block_checksum(sblock);
A
Arne Jansen 已提交
1515 1516
}

M
Miao Xie 已提交
1517 1518 1519 1520 1521 1522 1523 1524 1525 1526
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;
}

1527
static void scrub_recheck_block_checksum(struct scrub_block *sblock)
A
Arne Jansen 已提交
1528
{
1529 1530 1531
	sblock->header_error = 0;
	sblock->checksum_error = 0;
	sblock->generation_error = 0;
1532

1533 1534 1535 1536
	if (sblock->pagev[0]->flags & BTRFS_EXTENT_FLAG_DATA)
		scrub_checksum_data(sblock);
	else
		scrub_checksum_tree_block(sblock);
A
Arne Jansen 已提交
1537 1538
}

1539
static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
1540
					     struct scrub_block *sblock_good)
1541 1542 1543
{
	int page_num;
	int ret = 0;
I
Ilya Dryomov 已提交
1544

1545 1546
	for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
		int ret_sub;
I
Ilya Dryomov 已提交
1547

1548 1549
		ret_sub = scrub_repair_page_from_good_copy(sblock_bad,
							   sblock_good,
1550
							   page_num, 1);
1551 1552
		if (ret_sub)
			ret = ret_sub;
A
Arne Jansen 已提交
1553
	}
1554 1555 1556 1557 1558 1559 1560 1561

	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)
{
1562 1563
	struct scrub_page *page_bad = sblock_bad->pagev[page_num];
	struct scrub_page *page_good = sblock_good->pagev[page_num];
1564
	struct btrfs_fs_info *fs_info = sblock_bad->sctx->fs_info;
1565

1566 1567
	BUG_ON(page_bad->page == NULL);
	BUG_ON(page_good->page == NULL);
1568 1569 1570 1571 1572
	if (force_write || sblock_bad->header_error ||
	    sblock_bad->checksum_error || page_bad->io_error) {
		struct bio *bio;
		int ret;

1573
		if (!page_bad->dev->bdev) {
1574
			btrfs_warn_rl(fs_info,
J
Jeff Mahoney 已提交
1575
				"scrub_repair_page_from_good_copy(bdev == NULL) is unexpected");
1576 1577 1578
			return -EIO;
		}

1579
		bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1580 1581
		if (!bio)
			return -EIO;
1582
		bio->bi_bdev = page_bad->dev->bdev;
1583
		bio->bi_iter.bi_sector = page_bad->physical >> 9;
M
Mike Christie 已提交
1584
		bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
1585 1586 1587 1588 1589

		ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0);
		if (PAGE_SIZE != ret) {
			bio_put(bio);
			return -EIO;
1590
		}
1591

1592
		if (btrfsic_submit_bio_wait(bio)) {
1593 1594
			btrfs_dev_stat_inc_and_print(page_bad->dev,
				BTRFS_DEV_STAT_WRITE_ERRS);
1595
			btrfs_dev_replace_stats_inc(
1596
				&fs_info->dev_replace.num_write_errors);
1597 1598 1599
			bio_put(bio);
			return -EIO;
		}
1600
		bio_put(bio);
A
Arne Jansen 已提交
1601 1602
	}

1603 1604 1605
	return 0;
}

1606 1607
static void scrub_write_block_to_dev_replace(struct scrub_block *sblock)
{
1608
	struct btrfs_fs_info *fs_info = sblock->sctx->fs_info;
1609 1610
	int page_num;

1611 1612 1613 1614 1615 1616 1617
	/*
	 * 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;

1618 1619 1620 1621 1622 1623
	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(
1624
				&fs_info->dev_replace.num_write_errors);
1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636
	}
}

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);

1637
		memset(mapped_buffer, 0, PAGE_SIZE);
1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654
		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),
1655
					      GFP_KERNEL);
1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671
		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) {
1672 1673
			bio = btrfs_io_bio_alloc(GFP_KERNEL,
					wr_ctx->pages_per_wr_bio);
1674 1675 1676 1677 1678 1679 1680 1681 1682 1683
			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;
1684
		bio->bi_iter.bi_sector = sbio->physical >> 9;
M
Mike Christie 已提交
1685
		bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
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 1731 1732
		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 */
1733
	btrfsic_submit_bio(sbio->bio);
1734 1735
}

1736
static void scrub_wr_bio_end_io(struct bio *bio)
1737 1738
{
	struct scrub_bio *sbio = bio->bi_private;
1739
	struct btrfs_fs_info *fs_info = sbio->dev->fs_info;
1740

1741
	sbio->err = bio->bi_error;
1742 1743
	sbio->bio = bio;

1744 1745
	btrfs_init_work(&sbio->work, btrfs_scrubwrc_helper,
			 scrub_wr_bio_end_io_worker, NULL, NULL);
1746
	btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work);
1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757
}

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 =
1758
			&sbio->sctx->fs_info->dev_replace;
1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777

		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)
1778 1779 1780 1781
{
	u64 flags;
	int ret;

1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793
	/*
	 * No need to initialize these stats currently,
	 * because this function only use return value
	 * instead of these stats value.
	 *
	 * Todo:
	 * always use stats
	 */
	sblock->header_error = 0;
	sblock->generation_error = 0;
	sblock->checksum_error = 0;

1794 1795
	WARN_ON(sblock->page_count < 1);
	flags = sblock->pagev[0]->flags;
1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806
	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);
1807 1808

	return ret;
A
Arne Jansen 已提交
1809 1810
}

1811
static int scrub_checksum_data(struct scrub_block *sblock)
A
Arne Jansen 已提交
1812
{
1813
	struct scrub_ctx *sctx = sblock->sctx;
A
Arne Jansen 已提交
1814
	u8 csum[BTRFS_CSUM_SIZE];
1815 1816 1817
	u8 *on_disk_csum;
	struct page *page;
	void *buffer;
A
Arne Jansen 已提交
1818
	u32 crc = ~(u32)0;
1819 1820
	u64 len;
	int index;
A
Arne Jansen 已提交
1821

1822
	BUG_ON(sblock->page_count < 1);
1823
	if (!sblock->pagev[0]->have_csum)
A
Arne Jansen 已提交
1824 1825
		return 0;

1826 1827
	on_disk_csum = sblock->pagev[0]->csum;
	page = sblock->pagev[0]->page;
1828
	buffer = kmap_atomic(page);
1829

1830
	len = sctx->sectorsize;
1831 1832 1833 1834
	index = 0;
	for (;;) {
		u64 l = min_t(u64, len, PAGE_SIZE);

1835
		crc = btrfs_csum_data(buffer, crc, l);
1836
		kunmap_atomic(buffer);
1837 1838 1839 1840 1841
		len -= l;
		if (len == 0)
			break;
		index++;
		BUG_ON(index >= sblock->page_count);
1842 1843
		BUG_ON(!sblock->pagev[index]->page);
		page = sblock->pagev[index]->page;
1844
		buffer = kmap_atomic(page);
1845 1846
	}

A
Arne Jansen 已提交
1847
	btrfs_csum_final(crc, csum);
1848
	if (memcmp(csum, on_disk_csum, sctx->csum_size))
1849
		sblock->checksum_error = 1;
A
Arne Jansen 已提交
1850

1851
	return sblock->checksum_error;
A
Arne Jansen 已提交
1852 1853
}

1854
static int scrub_checksum_tree_block(struct scrub_block *sblock)
A
Arne Jansen 已提交
1855
{
1856
	struct scrub_ctx *sctx = sblock->sctx;
A
Arne Jansen 已提交
1857
	struct btrfs_header *h;
1858
	struct btrfs_fs_info *fs_info = sctx->fs_info;
1859 1860 1861 1862 1863 1864
	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
1865
	u32 crc = ~(u32)0;
1866 1867 1868 1869
	u64 len;
	int index;

	BUG_ON(sblock->page_count < 1);
1870
	page = sblock->pagev[0]->page;
1871
	mapped_buffer = kmap_atomic(page);
1872
	h = (struct btrfs_header *)mapped_buffer;
1873
	memcpy(on_disk_csum, h->csum, sctx->csum_size);
A
Arne Jansen 已提交
1874 1875 1876 1877 1878 1879

	/*
	 * we don't use the getter functions here, as we
	 * a) don't have an extent buffer and
	 * b) the page is already kmapped
	 */
1880
	if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h))
1881
		sblock->header_error = 1;
A
Arne Jansen 已提交
1882

1883 1884 1885 1886
	if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h)) {
		sblock->header_error = 1;
		sblock->generation_error = 1;
	}
A
Arne Jansen 已提交
1887

M
Miao Xie 已提交
1888
	if (!scrub_check_fsid(h->fsid, sblock->pagev[0]))
1889
		sblock->header_error = 1;
A
Arne Jansen 已提交
1890 1891 1892

	if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
		   BTRFS_UUID_SIZE))
1893
		sblock->header_error = 1;
A
Arne Jansen 已提交
1894

1895
	len = sctx->nodesize - BTRFS_CSUM_SIZE;
1896 1897 1898 1899 1900 1901
	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);

1902
		crc = btrfs_csum_data(p, crc, l);
1903
		kunmap_atomic(mapped_buffer);
1904 1905 1906 1907 1908
		len -= l;
		if (len == 0)
			break;
		index++;
		BUG_ON(index >= sblock->page_count);
1909 1910
		BUG_ON(!sblock->pagev[index]->page);
		page = sblock->pagev[index]->page;
1911
		mapped_buffer = kmap_atomic(page);
1912 1913 1914 1915 1916
		mapped_size = PAGE_SIZE;
		p = mapped_buffer;
	}

	btrfs_csum_final(crc, calculated_csum);
1917
	if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
1918
		sblock->checksum_error = 1;
A
Arne Jansen 已提交
1919

1920
	return sblock->header_error || sblock->checksum_error;
A
Arne Jansen 已提交
1921 1922
}

1923
static int scrub_checksum_super(struct scrub_block *sblock)
A
Arne Jansen 已提交
1924 1925
{
	struct btrfs_super_block *s;
1926
	struct scrub_ctx *sctx = sblock->sctx;
1927 1928 1929 1930 1931 1932
	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
1933
	u32 crc = ~(u32)0;
1934 1935
	int fail_gen = 0;
	int fail_cor = 0;
1936 1937
	u64 len;
	int index;
A
Arne Jansen 已提交
1938

1939
	BUG_ON(sblock->page_count < 1);
1940
	page = sblock->pagev[0]->page;
1941
	mapped_buffer = kmap_atomic(page);
1942
	s = (struct btrfs_super_block *)mapped_buffer;
1943
	memcpy(on_disk_csum, s->csum, sctx->csum_size);
A
Arne Jansen 已提交
1944

1945
	if (sblock->pagev[0]->logical != btrfs_super_bytenr(s))
1946
		++fail_cor;
A
Arne Jansen 已提交
1947

1948
	if (sblock->pagev[0]->generation != btrfs_super_generation(s))
1949
		++fail_gen;
A
Arne Jansen 已提交
1950

M
Miao Xie 已提交
1951
	if (!scrub_check_fsid(s->fsid, sblock->pagev[0]))
1952
		++fail_cor;
A
Arne Jansen 已提交
1953

1954 1955 1956 1957 1958 1959 1960
	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);

1961
		crc = btrfs_csum_data(p, crc, l);
1962
		kunmap_atomic(mapped_buffer);
1963 1964 1965 1966 1967
		len -= l;
		if (len == 0)
			break;
		index++;
		BUG_ON(index >= sblock->page_count);
1968 1969
		BUG_ON(!sblock->pagev[index]->page);
		page = sblock->pagev[index]->page;
1970
		mapped_buffer = kmap_atomic(page);
1971 1972 1973 1974 1975
		mapped_size = PAGE_SIZE;
		p = mapped_buffer;
	}

	btrfs_csum_final(crc, calculated_csum);
1976
	if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
1977
		++fail_cor;
A
Arne Jansen 已提交
1978

1979
	if (fail_cor + fail_gen) {
A
Arne Jansen 已提交
1980 1981 1982 1983 1984
		/*
		 * if we find an error in a super block, we just report it.
		 * They will get written with the next transaction commit
		 * anyway
		 */
1985 1986 1987
		spin_lock(&sctx->stat_lock);
		++sctx->stat.super_errors;
		spin_unlock(&sctx->stat_lock);
1988
		if (fail_cor)
1989
			btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
1990 1991
				BTRFS_DEV_STAT_CORRUPTION_ERRS);
		else
1992
			btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
1993
				BTRFS_DEV_STAT_GENERATION_ERRS);
A
Arne Jansen 已提交
1994 1995
	}

1996
	return fail_cor + fail_gen;
A
Arne Jansen 已提交
1997 1998
}

1999 2000
static void scrub_block_get(struct scrub_block *sblock)
{
2001
	refcount_inc(&sblock->refs);
2002 2003 2004 2005
}

static void scrub_block_put(struct scrub_block *sblock)
{
2006
	if (refcount_dec_and_test(&sblock->refs)) {
2007 2008
		int i;

2009 2010 2011
		if (sblock->sparity)
			scrub_parity_put(sblock->sparity);

2012
		for (i = 0; i < sblock->page_count; i++)
2013
			scrub_page_put(sblock->pagev[i]);
2014 2015 2016 2017
		kfree(sblock);
	}
}

2018 2019
static void scrub_page_get(struct scrub_page *spage)
{
2020
	atomic_inc(&spage->refs);
2021 2022 2023 2024
}

static void scrub_page_put(struct scrub_page *spage)
{
2025
	if (atomic_dec_and_test(&spage->refs)) {
2026 2027 2028 2029 2030 2031
		if (spage->page)
			__free_page(spage->page);
		kfree(spage);
	}
}

2032
static void scrub_submit(struct scrub_ctx *sctx)
A
Arne Jansen 已提交
2033 2034 2035
{
	struct scrub_bio *sbio;

2036
	if (sctx->curr == -1)
S
Stefan Behrens 已提交
2037
		return;
A
Arne Jansen 已提交
2038

2039 2040
	sbio = sctx->bios[sctx->curr];
	sctx->curr = -1;
2041
	scrub_pending_bio_inc(sctx);
2042
	btrfsic_submit_bio(sbio->bio);
A
Arne Jansen 已提交
2043 2044
}

2045 2046
static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
				    struct scrub_page *spage)
A
Arne Jansen 已提交
2047
{
2048
	struct scrub_block *sblock = spage->sblock;
A
Arne Jansen 已提交
2049
	struct scrub_bio *sbio;
2050
	int ret;
A
Arne Jansen 已提交
2051 2052 2053 2054 2055

again:
	/*
	 * grab a fresh bio or wait for one to become available
	 */
2056 2057 2058 2059 2060 2061 2062 2063
	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
2064
		} else {
2065 2066
			spin_unlock(&sctx->list_lock);
			wait_event(sctx->list_wait, sctx->first_free != -1);
A
Arne Jansen 已提交
2067 2068
		}
	}
2069
	sbio = sctx->bios[sctx->curr];
2070
	if (sbio->page_count == 0) {
2071 2072
		struct bio *bio;

2073 2074
		sbio->physical = spage->physical;
		sbio->logical = spage->logical;
2075
		sbio->dev = spage->dev;
2076 2077
		bio = sbio->bio;
		if (!bio) {
2078 2079
			bio = btrfs_io_bio_alloc(GFP_KERNEL,
					sctx->pages_per_rd_bio);
2080 2081 2082 2083
			if (!bio)
				return -ENOMEM;
			sbio->bio = bio;
		}
2084 2085 2086

		bio->bi_private = sbio;
		bio->bi_end_io = scrub_bio_end_io;
2087
		bio->bi_bdev = sbio->dev->bdev;
2088
		bio->bi_iter.bi_sector = sbio->physical >> 9;
M
Mike Christie 已提交
2089
		bio_set_op_attrs(bio, REQ_OP_READ, 0);
2090
		sbio->err = 0;
2091 2092 2093
	} else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
		   spage->physical ||
		   sbio->logical + sbio->page_count * PAGE_SIZE !=
2094 2095
		   spage->logical ||
		   sbio->dev != spage->dev) {
2096
		scrub_submit(sctx);
A
Arne Jansen 已提交
2097 2098
		goto again;
	}
2099

2100 2101 2102 2103 2104 2105 2106 2107
	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;
		}
2108
		scrub_submit(sctx);
2109 2110 2111
		goto again;
	}

2112
	scrub_block_get(sblock); /* one for the page added to the bio */
2113 2114
	atomic_inc(&sblock->outstanding_pages);
	sbio->page_count++;
2115
	if (sbio->page_count == sctx->pages_per_rd_bio)
2116
		scrub_submit(sctx);
2117 2118 2119 2120

	return 0;
}

2121
static void scrub_missing_raid56_end_io(struct bio *bio)
2122 2123
{
	struct scrub_block *sblock = bio->bi_private;
2124
	struct btrfs_fs_info *fs_info = sblock->sctx->fs_info;
2125

2126
	if (bio->bi_error)
2127 2128
		sblock->no_io_error_seen = 0;

2129 2130
	bio_put(bio);

2131 2132 2133 2134 2135 2136 2137
	btrfs_queue_work(fs_info->scrub_workers, &sblock->work);
}

static void scrub_missing_raid56_worker(struct btrfs_work *work)
{
	struct scrub_block *sblock = container_of(work, struct scrub_block, work);
	struct scrub_ctx *sctx = sblock->sctx;
2138
	struct btrfs_fs_info *fs_info = sctx->fs_info;
2139 2140 2141 2142 2143 2144
	u64 logical;
	struct btrfs_device *dev;

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

2145
	if (sblock->no_io_error_seen)
2146
		scrub_recheck_block_checksum(sblock);
2147 2148 2149 2150 2151

	if (!sblock->no_io_error_seen) {
		spin_lock(&sctx->stat_lock);
		sctx->stat.read_errors++;
		spin_unlock(&sctx->stat_lock);
2152
		btrfs_err_rl_in_rcu(fs_info,
2153
			"IO error rebuilding logical %llu for dev %s",
2154 2155 2156 2157 2158
			logical, rcu_str_deref(dev->name));
	} else if (sblock->header_error || sblock->checksum_error) {
		spin_lock(&sctx->stat_lock);
		sctx->stat.uncorrectable_errors++;
		spin_unlock(&sctx->stat_lock);
2159
		btrfs_err_rl_in_rcu(fs_info,
2160
			"failed to rebuild valid logical %llu for dev %s",
2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180
			logical, rcu_str_deref(dev->name));
	} else {
		scrub_write_block_to_dev_replace(sblock);
	}

	scrub_block_put(sblock);

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

	scrub_pending_bio_dec(sctx);
}

static void scrub_missing_raid56_pages(struct scrub_block *sblock)
{
	struct scrub_ctx *sctx = sblock->sctx;
2181
	struct btrfs_fs_info *fs_info = sctx->fs_info;
2182 2183
	u64 length = sblock->page_count * PAGE_SIZE;
	u64 logical = sblock->pagev[0]->logical;
2184
	struct btrfs_bio *bbio = NULL;
2185 2186 2187 2188 2189
	struct bio *bio;
	struct btrfs_raid_bio *rbio;
	int ret;
	int i;

2190 2191
	ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical,
			&length, &bbio, 0, 1);
2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213
	if (ret || !bbio || !bbio->raid_map)
		goto bbio_out;

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

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

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

2214
	rbio = raid56_alloc_missing_rbio(fs_info, bio, bbio, length);
2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239
	if (!rbio)
		goto rbio_out;

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

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

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

rbio_out:
	bio_put(bio);
bbio_out:
	btrfs_put_bbio(bbio);
	spin_lock(&sctx->stat_lock);
	sctx->stat.malloc_errors++;
	spin_unlock(&sctx->stat_lock);
}

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

2248
	sblock = kzalloc(sizeof(*sblock), GFP_KERNEL);
2249
	if (!sblock) {
2250 2251 2252
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		spin_unlock(&sctx->stat_lock);
2253
		return -ENOMEM;
A
Arne Jansen 已提交
2254
	}
2255

2256 2257
	/* one ref inside this function, plus one for each page added to
	 * a bio later on */
2258
	refcount_set(&sblock->refs, 1);
2259
	sblock->sctx = sctx;
2260 2261 2262
	sblock->no_io_error_seen = 1;

	for (index = 0; len > 0; index++) {
2263
		struct scrub_page *spage;
2264 2265
		u64 l = min_t(u64, len, PAGE_SIZE);

2266
		spage = kzalloc(sizeof(*spage), GFP_KERNEL);
2267 2268
		if (!spage) {
leave_nomem:
2269 2270 2271
			spin_lock(&sctx->stat_lock);
			sctx->stat.malloc_errors++;
			spin_unlock(&sctx->stat_lock);
2272
			scrub_block_put(sblock);
2273 2274
			return -ENOMEM;
		}
2275 2276 2277
		BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
		scrub_page_get(spage);
		sblock->pagev[index] = spage;
2278
		spage->sblock = sblock;
2279
		spage->dev = dev;
2280 2281 2282 2283
		spage->flags = flags;
		spage->generation = gen;
		spage->logical = logical;
		spage->physical = physical;
2284
		spage->physical_for_dev_replace = physical_for_dev_replace;
2285 2286 2287
		spage->mirror_num = mirror_num;
		if (csum) {
			spage->have_csum = 1;
2288
			memcpy(spage->csum, csum, sctx->csum_size);
2289 2290 2291 2292
		} else {
			spage->have_csum = 0;
		}
		sblock->page_count++;
2293
		spage->page = alloc_page(GFP_KERNEL);
2294 2295
		if (!spage->page)
			goto leave_nomem;
2296 2297 2298
		len -= l;
		logical += l;
		physical += l;
2299
		physical_for_dev_replace += l;
2300 2301
	}

2302
	WARN_ON(sblock->page_count == 0);
2303 2304 2305 2306 2307 2308 2309 2310 2311 2312
	if (dev->missing) {
		/*
		 * This case should only be hit for RAID 5/6 device replace. See
		 * the comment in scrub_missing_raid56_pages() for details.
		 */
		scrub_missing_raid56_pages(sblock);
	} else {
		for (index = 0; index < sblock->page_count; index++) {
			struct scrub_page *spage = sblock->pagev[index];
			int ret;
2313

2314 2315 2316 2317 2318
			ret = scrub_add_page_to_rd_bio(sctx, spage);
			if (ret) {
				scrub_block_put(sblock);
				return ret;
			}
2319
		}
A
Arne Jansen 已提交
2320

2321 2322 2323
		if (force)
			scrub_submit(sctx);
	}
A
Arne Jansen 已提交
2324

2325 2326
	/* last one frees, either here or in bio completion for last page */
	scrub_block_put(sblock);
A
Arne Jansen 已提交
2327 2328 2329
	return 0;
}

2330
static void scrub_bio_end_io(struct bio *bio)
2331 2332
{
	struct scrub_bio *sbio = bio->bi_private;
2333
	struct btrfs_fs_info *fs_info = sbio->dev->fs_info;
2334

2335
	sbio->err = bio->bi_error;
2336 2337
	sbio->bio = bio;

2338
	btrfs_queue_work(fs_info->scrub_workers, &sbio->work);
2339 2340 2341 2342 2343
}

static void scrub_bio_end_io_worker(struct btrfs_work *work)
{
	struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
2344
	struct scrub_ctx *sctx = sbio->sctx;
2345 2346
	int i;

2347
	BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO);
2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368
	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;
2369 2370 2371 2372
	spin_lock(&sctx->list_lock);
	sbio->next_free = sctx->first_free;
	sctx->first_free = sbio->index;
	spin_unlock(&sctx->list_lock);
2373 2374 2375 2376 2377 2378 2379 2380

	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);
	}

2381
	scrub_pending_bio_dec(sctx);
2382 2383
}

2384 2385 2386 2387
static inline void __scrub_mark_bitmap(struct scrub_parity *sparity,
				       unsigned long *bitmap,
				       u64 start, u64 len)
{
2388
	u32 offset;
2389
	int nsectors;
2390
	int sectorsize = sparity->sctx->fs_info->sectorsize;
2391 2392 2393 2394 2395 2396 2397

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

	start -= sparity->logic_start;
2398
	start = div_u64_rem(start, sparity->stripe_len, &offset);
2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422
	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);
}

2423 2424
static void scrub_block_complete(struct scrub_block *sblock)
{
2425 2426
	int corrupted = 0;

2427
	if (!sblock->no_io_error_seen) {
2428
		corrupted = 1;
2429
		scrub_handle_errored_block(sblock);
2430 2431 2432 2433 2434 2435
	} else {
		/*
		 * if has checksum error, write via repair mechanism in
		 * dev replace case, otherwise write here in dev replace
		 * case.
		 */
2436 2437
		corrupted = scrub_checksum(sblock);
		if (!corrupted && sblock->sctx->is_dev_replace)
2438 2439
			scrub_write_block_to_dev_replace(sblock);
	}
2440 2441 2442 2443 2444 2445 2446 2447 2448

	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);
	}
2449 2450
}

2451
static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u8 *csum)
A
Arne Jansen 已提交
2452 2453
{
	struct btrfs_ordered_sum *sum = NULL;
2454
	unsigned long index;
A
Arne Jansen 已提交
2455 2456
	unsigned long num_sectors;

2457 2458
	while (!list_empty(&sctx->csum_list)) {
		sum = list_first_entry(&sctx->csum_list,
A
Arne Jansen 已提交
2459 2460 2461 2462 2463 2464
				       struct btrfs_ordered_sum, list);
		if (sum->bytenr > logical)
			return 0;
		if (sum->bytenr + sum->len > logical)
			break;

2465
		++sctx->stat.csum_discards;
A
Arne Jansen 已提交
2466 2467 2468 2469 2470 2471 2472
		list_del(&sum->list);
		kfree(sum);
		sum = NULL;
	}
	if (!sum)
		return 0;

2473
	index = ((u32)(logical - sum->bytenr)) / sctx->sectorsize;
2474
	num_sectors = sum->len / sctx->sectorsize;
2475 2476
	memcpy(csum, sum->sums + index, sctx->csum_size);
	if (index == num_sectors - 1) {
A
Arne Jansen 已提交
2477 2478 2479
		list_del(&sum->list);
		kfree(sum);
	}
2480
	return 1;
A
Arne Jansen 已提交
2481 2482 2483
}

/* scrub extent tries to collect up to 64 kB for each bio */
2484
static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len,
2485
			u64 physical, struct btrfs_device *dev, u64 flags,
2486
			u64 gen, int mirror_num, u64 physical_for_dev_replace)
A
Arne Jansen 已提交
2487 2488 2489
{
	int ret;
	u8 csum[BTRFS_CSUM_SIZE];
2490 2491 2492
	u32 blocksize;

	if (flags & BTRFS_EXTENT_FLAG_DATA) {
2493 2494 2495 2496 2497
		blocksize = sctx->sectorsize;
		spin_lock(&sctx->stat_lock);
		sctx->stat.data_extents_scrubbed++;
		sctx->stat.data_bytes_scrubbed += len;
		spin_unlock(&sctx->stat_lock);
2498
	} else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2499 2500 2501 2502 2503
		blocksize = sctx->nodesize;
		spin_lock(&sctx->stat_lock);
		sctx->stat.tree_extents_scrubbed++;
		sctx->stat.tree_bytes_scrubbed += len;
		spin_unlock(&sctx->stat_lock);
2504
	} else {
2505
		blocksize = sctx->sectorsize;
2506
		WARN_ON(1);
2507
	}
A
Arne Jansen 已提交
2508 2509

	while (len) {
2510
		u64 l = min_t(u64, len, blocksize);
A
Arne Jansen 已提交
2511 2512 2513 2514
		int have_csum = 0;

		if (flags & BTRFS_EXTENT_FLAG_DATA) {
			/* push csums to sbio */
2515
			have_csum = scrub_find_csum(sctx, logical, csum);
A
Arne Jansen 已提交
2516
			if (have_csum == 0)
2517
				++sctx->stat.no_csum;
2518 2519 2520 2521 2522 2523
			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
2524
		}
2525
		ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen,
2526 2527 2528
				  mirror_num, have_csum ? csum : NULL, 0,
				  physical_for_dev_replace);
behind_scrub_pages:
A
Arne Jansen 已提交
2529 2530 2531 2532 2533
		if (ret)
			return ret;
		len -= l;
		logical += l;
		physical += l;
2534
		physical_for_dev_replace += l;
A
Arne Jansen 已提交
2535 2536 2537 2538
	}
	return 0;
}

2539 2540 2541 2542 2543 2544 2545 2546 2547
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;

2548
	sblock = kzalloc(sizeof(*sblock), GFP_KERNEL);
2549 2550 2551 2552 2553 2554 2555 2556 2557
	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 */
2558
	refcount_set(&sblock->refs, 1);
2559 2560 2561 2562 2563 2564 2565 2566 2567
	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);

2568
		spage = kzalloc(sizeof(*spage), GFP_KERNEL);
2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597
		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++;
2598
		spage->page = alloc_page(GFP_KERNEL);
2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632
		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;

2633 2634 2635 2636 2637
	if (dev->missing) {
		scrub_parity_mark_sectors_error(sparity, logical, len);
		return 0;
	}

2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652
	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 */
2653
			have_csum = scrub_find_csum(sctx, logical, csum);
2654 2655 2656 2657 2658 2659 2660 2661
			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;
2662
skip:
2663 2664 2665 2666 2667 2668 2669
		len -= l;
		logical += l;
		physical += l;
	}
	return 0;
}

2670 2671 2672 2673 2674 2675 2676 2677
/*
 * 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,
2678 2679
				   struct map_lookup *map, u64 *offset,
				   u64 *stripe_start)
2680 2681 2682 2683 2684
{
	int i;
	int j = 0;
	u64 stripe_nr;
	u64 last_offset;
2685 2686
	u32 stripe_index;
	u32 rot;
2687 2688 2689

	last_offset = (physical - map->stripes[num].physical) *
		      nr_data_stripes(map);
2690 2691 2692
	if (stripe_start)
		*stripe_start = last_offset;

2693 2694 2695 2696
	*offset = last_offset;
	for (i = 0; i < nr_data_stripes(map); i++) {
		*offset = last_offset + i * map->stripe_len;

2697 2698
		stripe_nr = div_u64(*offset, map->stripe_len);
		stripe_nr = div_u64(stripe_nr, nr_data_stripes(map));
2699 2700

		/* Work out the disk rotation on this stripe-set */
2701
		stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, &rot);
2702 2703
		/* calculate which stripe this data locates */
		rot += i;
2704
		stripe_index = rot % map->num_stripes;
2705 2706 2707 2708 2709 2710 2711 2712 2713
		if (stripe_index == num)
			return 0;
		if (stripe_index < num)
			j++;
	}
	*offset = last_offset + j * map->stripe_len;
	return 1;
}

2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735
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);
}

2736 2737 2738 2739 2740 2741 2742 2743 2744 2745
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);
}

2746
static void scrub_parity_bio_endio(struct bio *bio)
2747 2748
{
	struct scrub_parity *sparity = (struct scrub_parity *)bio->bi_private;
2749
	struct btrfs_fs_info *fs_info = sparity->sctx->fs_info;
2750

2751
	if (bio->bi_error)
2752 2753 2754 2755
		bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap,
			  sparity->nsectors);

	bio_put(bio);
2756 2757 2758

	btrfs_init_work(&sparity->work, btrfs_scrubparity_helper,
			scrub_parity_bio_endio_worker, NULL, NULL);
2759
	btrfs_queue_work(fs_info->scrub_parity_workers, &sparity->work);
2760 2761 2762 2763 2764
}

static void scrub_parity_check_and_repair(struct scrub_parity *sparity)
{
	struct scrub_ctx *sctx = sparity->sctx;
2765
	struct btrfs_fs_info *fs_info = sctx->fs_info;
2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776
	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;

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

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

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

2791
	rbio = raid56_parity_alloc_scrub_rbio(fs_info, bio, bbio,
2792
					      length, sparity->scrub_dev,
2793 2794 2795 2796 2797 2798
					      sparity->dbitmap,
					      sparity->nsectors);
	if (!rbio)
		goto rbio_out;

	list_for_each_entry(spage, &sparity->spages, list)
2799
		raid56_add_scrub_pages(rbio, spage->page, spage->logical);
2800 2801 2802 2803 2804 2805 2806 2807

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

rbio_out:
	bio_put(bio);
bbio_out:
2808
	btrfs_put_bbio(bbio);
2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819
	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)
{
2820
	return DIV_ROUND_UP(nsectors, BITS_PER_LONG) * sizeof(long);
2821 2822 2823 2824
}

static void scrub_parity_get(struct scrub_parity *sparity)
{
2825
	atomic_inc(&sparity->refs);
2826 2827 2828 2829
}

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

2865
	nsectors = div_u64(map->stripe_len, fs_info->sectorsize);
2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881
	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;
2882
	atomic_set(&sparity->refs, 1);
2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930
	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);

2931 2932 2933 2934
			if (key.type != BTRFS_EXTENT_ITEM_KEY &&
			    key.type != BTRFS_METADATA_ITEM_KEY)
				goto next;

2935
			if (key.type == BTRFS_METADATA_ITEM_KEY)
2936
				bytes = fs_info->nodesize;
2937 2938 2939 2940 2941 2942
			else
				bytes = key.offset;

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

2943
			if (key.objectid >= logic_end) {
2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955
				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);

2956 2957 2958 2959
			if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) &&
			    (key.objectid < logic_start ||
			     key.objectid + bytes >
			     logic_start + map->stripe_len)) {
J
Jeff Mahoney 已提交
2960 2961
				btrfs_err(fs_info,
					  "scrub: tree block %llu spanning stripes, ignored. logical=%llu",
2962
					  key.objectid, logic_start);
2963 2964 2965
				spin_lock(&sctx->stat_lock);
				sctx->stat.uncorrectable_errors++;
				spin_unlock(&sctx->stat_lock);
2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984
				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);

2985
			mapped_length = extent_len;
2986
			bbio = NULL;
2987 2988 2989
			ret = btrfs_map_block(fs_info, BTRFS_MAP_READ,
					extent_logical, &mapped_length, &bbio,
					0);
2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001
			if (!ret) {
				if (!bbio || mapped_length < extent_len)
					ret = -EIO;
			}
			if (ret) {
				btrfs_put_bbio(bbio);
				goto out;
			}
			extent_physical = bbio->stripes[0].physical;
			extent_mirror_num = bbio->mirror_num;
			extent_dev = bbio->stripes[0].dev;
			btrfs_put_bbio(bbio);
3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015

			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);
3016 3017 3018

			scrub_free_csums(sctx);

3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049
			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,
3050
						logic_end - logic_start);
3051 3052 3053 3054 3055 3056 3057 3058 3059 3060
	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;
}

3061
static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
3062 3063
					   struct map_lookup *map,
					   struct btrfs_device *scrub_dev,
3064 3065
					   int num, u64 base, u64 length,
					   int is_dev_replace)
A
Arne Jansen 已提交
3066
{
3067
	struct btrfs_path *path, *ppath;
3068
	struct btrfs_fs_info *fs_info = sctx->fs_info;
A
Arne Jansen 已提交
3069 3070 3071
	struct btrfs_root *root = fs_info->extent_root;
	struct btrfs_root *csum_root = fs_info->csum_root;
	struct btrfs_extent_item *extent;
3072
	struct blk_plug plug;
A
Arne Jansen 已提交
3073 3074 3075 3076 3077 3078 3079
	u64 flags;
	int ret;
	int slot;
	u64 nstripes;
	struct extent_buffer *l;
	u64 physical;
	u64 logical;
L
Liu Bo 已提交
3080
	u64 logic_end;
3081
	u64 physical_end;
A
Arne Jansen 已提交
3082
	u64 generation;
3083
	int mirror_num;
A
Arne Jansen 已提交
3084 3085
	struct reada_control *reada1;
	struct reada_control *reada2;
3086
	struct btrfs_key key;
A
Arne Jansen 已提交
3087
	struct btrfs_key key_end;
A
Arne Jansen 已提交
3088 3089
	u64 increment = map->stripe_len;
	u64 offset;
3090 3091 3092
	u64 extent_logical;
	u64 extent_physical;
	u64 extent_len;
3093 3094
	u64 stripe_logical;
	u64 stripe_end;
3095 3096
	struct btrfs_device *extent_dev;
	int extent_mirror_num;
3097
	int stop_loop = 0;
D
David Woodhouse 已提交
3098

3099
	physical = map->stripes[num].physical;
A
Arne Jansen 已提交
3100
	offset = 0;
3101
	nstripes = div_u64(length, map->stripe_len);
A
Arne Jansen 已提交
3102 3103 3104
	if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
		offset = map->stripe_len * num;
		increment = map->stripe_len * map->num_stripes;
3105
		mirror_num = 1;
A
Arne Jansen 已提交
3106 3107 3108 3109
	} 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;
3110
		mirror_num = num % map->sub_stripes + 1;
A
Arne Jansen 已提交
3111 3112
	} else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
		increment = map->stripe_len;
3113
		mirror_num = num % map->num_stripes + 1;
A
Arne Jansen 已提交
3114 3115
	} else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
		increment = map->stripe_len;
3116
		mirror_num = num % map->num_stripes + 1;
3117
	} else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3118
		get_raid56_logic_offset(physical, num, map, &offset, NULL);
3119 3120
		increment = map->stripe_len * nr_data_stripes(map);
		mirror_num = 1;
A
Arne Jansen 已提交
3121 3122
	} else {
		increment = map->stripe_len;
3123
		mirror_num = 1;
A
Arne Jansen 已提交
3124 3125 3126 3127 3128 3129
	}

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

3130 3131
	ppath = btrfs_alloc_path();
	if (!ppath) {
3132
		btrfs_free_path(path);
3133 3134 3135
		return -ENOMEM;
	}

3136 3137 3138 3139 3140
	/*
	 * 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
3141 3142 3143
	path->search_commit_root = 1;
	path->skip_locking = 1;

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

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

3173 3174 3175
	key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
	key.type = BTRFS_EXTENT_CSUM_KEY;
	key.offset = logical;
A
Arne Jansen 已提交
3176 3177
	key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
	key_end.type = BTRFS_EXTENT_CSUM_KEY;
3178
	key_end.offset = logic_end;
3179
	reada2 = btrfs_reada_add(csum_root, &key, &key_end);
A
Arne Jansen 已提交
3180 3181 3182 3183 3184 3185

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

A
Arne Jansen 已提交
3186 3187 3188 3189 3190

	/*
	 * collect all data csums for the stripe to avoid seeking during
	 * the scrub. This might currently (crc32) end up to be about 1MB
	 */
3191
	blk_start_plug(&plug);
A
Arne Jansen 已提交
3192 3193 3194 3195 3196

	/*
	 * now find all extents for each stripe and scrub them
	 */
	ret = 0;
3197
	while (physical < physical_end) {
A
Arne Jansen 已提交
3198 3199 3200 3201
		/*
		 * canceled?
		 */
		if (atomic_read(&fs_info->scrub_cancel_req) ||
3202
		    atomic_read(&sctx->cancel_req)) {
A
Arne Jansen 已提交
3203 3204 3205 3206 3207 3208 3209 3210
			ret = -ECANCELED;
			goto out;
		}
		/*
		 * check to see if we have to pause
		 */
		if (atomic_read(&fs_info->scrub_pause_req)) {
			/* push queued extents */
3211
			atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
3212
			scrub_submit(sctx);
3213 3214 3215
			mutex_lock(&sctx->wr_ctx.wr_lock);
			scrub_wr_submit(sctx);
			mutex_unlock(&sctx->wr_ctx.wr_lock);
3216
			wait_event(sctx->list_wait,
3217
				   atomic_read(&sctx->bios_in_flight) == 0);
3218
			atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
3219
			scrub_blocked_if_needed(fs_info);
A
Arne Jansen 已提交
3220 3221
		}

3222 3223 3224 3225 3226 3227
		if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
			ret = get_raid56_logic_offset(physical, num, map,
						      &logical,
						      &stripe_logical);
			logical += base;
			if (ret) {
3228
				/* it is parity strip */
3229
				stripe_logical += base;
3230
				stripe_end = stripe_logical + increment;
3231 3232 3233 3234 3235 3236 3237 3238 3239
				ret = scrub_raid56_parity(sctx, map, scrub_dev,
							  ppath, stripe_logical,
							  stripe_end);
				if (ret)
					goto out;
				goto skip;
			}
		}

3240 3241 3242 3243
		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
3244
		key.objectid = logical;
L
Liu Bo 已提交
3245
		key.offset = (u64)-1;
A
Arne Jansen 已提交
3246 3247 3248 3249

		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
		if (ret < 0)
			goto out;
3250

3251
		if (ret > 0) {
3252
			ret = btrfs_previous_extent_item(root, path, 0);
A
Arne Jansen 已提交
3253 3254
			if (ret < 0)
				goto out;
3255 3256 3257 3258 3259 3260 3261 3262 3263
			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
3264 3265
		}

L
Liu Bo 已提交
3266
		stop_loop = 0;
A
Arne Jansen 已提交
3267
		while (1) {
3268 3269
			u64 bytes;

A
Arne Jansen 已提交
3270 3271 3272 3273 3274 3275 3276 3277 3278
			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
3279
				stop_loop = 1;
A
Arne Jansen 已提交
3280 3281 3282 3283
				break;
			}
			btrfs_item_key_to_cpu(l, &key, slot);

3284 3285 3286 3287
			if (key.type != BTRFS_EXTENT_ITEM_KEY &&
			    key.type != BTRFS_METADATA_ITEM_KEY)
				goto next;

3288
			if (key.type == BTRFS_METADATA_ITEM_KEY)
3289
				bytes = fs_info->nodesize;
3290 3291 3292 3293
			else
				bytes = key.offset;

			if (key.objectid + bytes <= logical)
A
Arne Jansen 已提交
3294 3295
				goto next;

L
Liu Bo 已提交
3296 3297 3298 3299 3300 3301
			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
3302 3303 3304 3305 3306 3307

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

3308 3309 3310 3311
			if ((flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) &&
			    (key.objectid < logical ||
			     key.objectid + bytes >
			     logical + map->stripe_len)) {
3312
				btrfs_err(fs_info,
J
Jeff Mahoney 已提交
3313
					   "scrub: tree block %llu spanning stripes, ignored. logical=%llu",
3314
				       key.objectid, logical);
3315 3316 3317
				spin_lock(&sctx->stat_lock);
				sctx->stat.uncorrectable_errors++;
				spin_unlock(&sctx->stat_lock);
A
Arne Jansen 已提交
3318 3319 3320
				goto next;
			}

L
Liu Bo 已提交
3321 3322 3323 3324
again:
			extent_logical = key.objectid;
			extent_len = bytes;

A
Arne Jansen 已提交
3325 3326 3327
			/*
			 * trim extent to this stripe
			 */
L
Liu Bo 已提交
3328 3329 3330
			if (extent_logical < logical) {
				extent_len -= logical - extent_logical;
				extent_logical = logical;
A
Arne Jansen 已提交
3331
			}
L
Liu Bo 已提交
3332
			if (extent_logical + extent_len >
A
Arne Jansen 已提交
3333
			    logical + map->stripe_len) {
L
Liu Bo 已提交
3334 3335
				extent_len = logical + map->stripe_len -
					     extent_logical;
A
Arne Jansen 已提交
3336 3337
			}

L
Liu Bo 已提交
3338
			extent_physical = extent_logical - logical + physical;
3339 3340 3341 3342 3343 3344 3345
			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
3346

3347 3348 3349 3350 3351
			ret = btrfs_lookup_csums_range(csum_root,
						       extent_logical,
						       extent_logical +
						       extent_len - 1,
						       &sctx->csum_list, 1);
L
Liu Bo 已提交
3352 3353 3354
			if (ret)
				goto out;

3355 3356 3357
			ret = scrub_extent(sctx, extent_logical, extent_len,
					   extent_physical, extent_dev, flags,
					   generation, extent_mirror_num,
3358
					   extent_logical - logical + physical);
3359 3360 3361

			scrub_free_csums(sctx);

A
Arne Jansen 已提交
3362 3363 3364
			if (ret)
				goto out;

L
Liu Bo 已提交
3365 3366
			if (extent_logical + extent_len <
			    key.objectid + bytes) {
3367
				if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3368 3369 3370 3371
					/*
					 * loop until we find next data stripe
					 * or we have finished all stripes.
					 */
3372 3373 3374 3375 3376 3377 3378 3379 3380 3381
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 +
3382
								increment;
3383 3384 3385 3386 3387 3388 3389 3390
						ret = scrub_raid56_parity(sctx,
							map, scrub_dev, ppath,
							stripe_logical,
							stripe_end);
						if (ret)
							goto out;
						goto loop;
					}
3391 3392 3393 3394
				} else {
					physical += map->stripe_len;
					logical += increment;
				}
L
Liu Bo 已提交
3395 3396 3397 3398 3399
				if (logical < key.objectid + bytes) {
					cond_resched();
					goto again;
				}

3400
				if (physical >= physical_end) {
L
Liu Bo 已提交
3401 3402 3403 3404
					stop_loop = 1;
					break;
				}
			}
A
Arne Jansen 已提交
3405 3406 3407
next:
			path->slots[0]++;
		}
C
Chris Mason 已提交
3408
		btrfs_release_path(path);
3409
skip:
A
Arne Jansen 已提交
3410 3411
		logical += increment;
		physical += map->stripe_len;
3412
		spin_lock(&sctx->stat_lock);
L
Liu Bo 已提交
3413 3414 3415 3416 3417
		if (stop_loop)
			sctx->stat.last_physical = map->stripes[num].physical +
						   length;
		else
			sctx->stat.last_physical = physical;
3418
		spin_unlock(&sctx->stat_lock);
L
Liu Bo 已提交
3419 3420
		if (stop_loop)
			break;
A
Arne Jansen 已提交
3421
	}
3422
out:
A
Arne Jansen 已提交
3423
	/* push queued extents */
3424
	scrub_submit(sctx);
3425 3426 3427
	mutex_lock(&sctx->wr_ctx.wr_lock);
	scrub_wr_submit(sctx);
	mutex_unlock(&sctx->wr_ctx.wr_lock);
A
Arne Jansen 已提交
3428

3429
	blk_finish_plug(&plug);
A
Arne Jansen 已提交
3430
	btrfs_free_path(path);
3431
	btrfs_free_path(ppath);
A
Arne Jansen 已提交
3432 3433 3434
	return ret < 0 ? ret : 0;
}

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

	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);

3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464
	if (!em) {
		/*
		 * Might have been an unused block group deleted by the cleaner
		 * kthread or relocation.
		 */
		spin_lock(&cache->lock);
		if (!cache->removed)
			ret = -EINVAL;
		spin_unlock(&cache->lock);

		return ret;
	}
A
Arne Jansen 已提交
3465

3466
	map = em->map_lookup;
A
Arne Jansen 已提交
3467 3468 3469 3470 3471 3472 3473
	if (em->start != chunk_offset)
		goto out;

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

	for (i = 0; i < map->num_stripes; ++i) {
3474
		if (map->stripes[i].dev->bdev == scrub_dev->bdev &&
3475
		    map->stripes[i].physical == dev_offset) {
3476
			ret = scrub_stripe(sctx, map, scrub_dev, i,
3477 3478
					   chunk_offset, length,
					   is_dev_replace);
A
Arne Jansen 已提交
3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489
			if (ret)
				goto out;
		}
	}
out:
	free_extent_map(em);

	return ret;
}

static noinline_for_stack
3490
int scrub_enumerate_chunks(struct scrub_ctx *sctx,
3491 3492
			   struct btrfs_device *scrub_dev, u64 start, u64 end,
			   int is_dev_replace)
A
Arne Jansen 已提交
3493 3494 3495
{
	struct btrfs_dev_extent *dev_extent = NULL;
	struct btrfs_path *path;
3496 3497
	struct btrfs_fs_info *fs_info = sctx->fs_info;
	struct btrfs_root *root = fs_info->dev_root;
A
Arne Jansen 已提交
3498 3499
	u64 length;
	u64 chunk_offset;
3500
	int ret = 0;
3501
	int ro_set;
A
Arne Jansen 已提交
3502 3503 3504 3505 3506
	int slot;
	struct extent_buffer *l;
	struct btrfs_key key;
	struct btrfs_key found_key;
	struct btrfs_block_group_cache *cache;
3507
	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
A
Arne Jansen 已提交
3508 3509 3510 3511 3512

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

3513
	path->reada = READA_FORWARD;
A
Arne Jansen 已提交
3514 3515 3516
	path->search_commit_root = 1;
	path->skip_locking = 1;

3517
	key.objectid = scrub_dev->devid;
A
Arne Jansen 已提交
3518 3519 3520 3521 3522 3523
	key.offset = 0ull;
	key.type = BTRFS_DEV_EXTENT_KEY;

	while (1) {
		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
		if (ret < 0)
3524 3525 3526 3527 3528
			break;
		if (ret > 0) {
			if (path->slots[0] >=
			    btrfs_header_nritems(path->nodes[0])) {
				ret = btrfs_next_leaf(root, path);
3529 3530 3531 3532
				if (ret < 0)
					break;
				if (ret > 0) {
					ret = 0;
3533
					break;
3534 3535 3536
				}
			} else {
				ret = 0;
3537 3538
			}
		}
A
Arne Jansen 已提交
3539 3540 3541 3542 3543 3544

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

		btrfs_item_key_to_cpu(l, &found_key, slot);

3545
		if (found_key.objectid != scrub_dev->devid)
A
Arne Jansen 已提交
3546 3547
			break;

3548
		if (found_key.type != BTRFS_DEV_EXTENT_KEY)
A
Arne Jansen 已提交
3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559
			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);

3560 3561
		if (found_key.offset + length <= start)
			goto skip;
A
Arne Jansen 已提交
3562 3563 3564 3565 3566 3567 3568 3569

		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);
3570 3571 3572 3573 3574 3575

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

3576 3577 3578 3579 3580 3581 3582 3583 3584
		/*
		 * we need call btrfs_inc_block_group_ro() with scrubs_paused,
		 * to avoid deadlock caused by:
		 * btrfs_inc_block_group_ro()
		 * -> btrfs_wait_for_commit()
		 * -> btrfs_commit_transaction()
		 * -> btrfs_scrub_pause()
		 */
		scrub_pause_on(fs_info);
3585
		ret = btrfs_inc_block_group_ro(fs_info, cache);
3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616
		if (!ret && is_dev_replace) {
			/*
			 * If we are doing a device replace wait for any tasks
			 * that started dellaloc right before we set the block
			 * group to RO mode, as they might have just allocated
			 * an extent from it or decided they could do a nocow
			 * write. And if any such tasks did that, wait for their
			 * ordered extents to complete and then commit the
			 * current transaction, so that we can later see the new
			 * extent items in the extent tree - the ordered extents
			 * create delayed data references (for cow writes) when
			 * they complete, which will be run and insert the
			 * corresponding extent items into the extent tree when
			 * we commit the transaction they used when running
			 * inode.c:btrfs_finish_ordered_io(). We later use
			 * the commit root of the extent tree to find extents
			 * to copy from the srcdev into the tgtdev, and we don't
			 * want to miss any new extents.
			 */
			btrfs_wait_block_group_reservations(cache);
			btrfs_wait_nocow_writers(cache);
			ret = btrfs_wait_ordered_roots(fs_info, -1,
						       cache->key.objectid,
						       cache->key.offset);
			if (ret > 0) {
				struct btrfs_trans_handle *trans;

				trans = btrfs_join_transaction(root);
				if (IS_ERR(trans))
					ret = PTR_ERR(trans);
				else
3617
					ret = btrfs_commit_transaction(trans);
3618 3619 3620 3621 3622 3623 3624
				if (ret) {
					scrub_pause_off(fs_info);
					btrfs_put_block_group(cache);
					break;
				}
			}
		}
3625
		scrub_pause_off(fs_info);
3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638

		if (ret == 0) {
			ro_set = 1;
		} else if (ret == -ENOSPC) {
			/*
			 * btrfs_inc_block_group_ro return -ENOSPC when it
			 * failed in creating new chunk for metadata.
			 * It is not a problem for scrub/replace, because
			 * metadata are always cowed, and our scrub paused
			 * commit_transactions.
			 */
			ro_set = 0;
		} else {
J
Jeff Mahoney 已提交
3639 3640
			btrfs_warn(fs_info,
				   "failed setting block group ro, ret=%d\n",
3641
				   ret);
3642 3643 3644 3645
			btrfs_put_block_group(cache);
			break;
		}

3646
		btrfs_dev_replace_lock(&fs_info->dev_replace, 1);
3647 3648 3649
		dev_replace->cursor_right = found_key.offset + length;
		dev_replace->cursor_left = found_key.offset;
		dev_replace->item_needs_writeback = 1;
3650
		btrfs_dev_replace_unlock(&fs_info->dev_replace, 1);
3651
		ret = scrub_chunk(sctx, scrub_dev, chunk_offset, length,
3652
				  found_key.offset, cache, is_dev_replace);
3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671

		/*
		 * 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);
3672 3673

		scrub_pause_on(fs_info);
3674 3675 3676 3677 3678 3679

		/*
		 * must be called before we decrease @scrub_paused.
		 * make sure we don't block transaction commit while
		 * we are waiting pending workers finished.
		 */
3680 3681
		wait_event(sctx->list_wait,
			   atomic_read(&sctx->workers_pending) == 0);
3682 3683
		atomic_set(&sctx->wr_ctx.flush_all_writes, 0);

3684
		scrub_pause_off(fs_info);
3685

3686 3687 3688 3689 3690
		btrfs_dev_replace_lock(&fs_info->dev_replace, 1);
		dev_replace->cursor_left = dev_replace->cursor_right;
		dev_replace->item_needs_writeback = 1;
		btrfs_dev_replace_unlock(&fs_info->dev_replace, 1);

3691
		if (ro_set)
3692
			btrfs_dec_block_group_ro(cache);
3693

3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715
		/*
		 * We might have prevented the cleaner kthread from deleting
		 * this block group if it was already unused because we raced
		 * and set it to RO mode first. So add it back to the unused
		 * list, otherwise it might not ever be deleted unless a manual
		 * balance is triggered or it becomes used and unused again.
		 */
		spin_lock(&cache->lock);
		if (!cache->removed && !cache->ro && cache->reserved == 0 &&
		    btrfs_block_group_used(&cache->item) == 0) {
			spin_unlock(&cache->lock);
			spin_lock(&fs_info->unused_bgs_lock);
			if (list_empty(&cache->bg_list)) {
				btrfs_get_block_group(cache);
				list_add_tail(&cache->bg_list,
					      &fs_info->unused_bgs);
			}
			spin_unlock(&fs_info->unused_bgs_lock);
		} else {
			spin_unlock(&cache->lock);
		}

A
Arne Jansen 已提交
3716 3717 3718
		btrfs_put_block_group(cache);
		if (ret)
			break;
3719 3720
		if (is_dev_replace &&
		    atomic64_read(&dev_replace->num_write_errors) > 0) {
3721 3722 3723 3724 3725 3726 3727
			ret = -EIO;
			break;
		}
		if (sctx->stat.malloc_errors > 0) {
			ret = -ENOMEM;
			break;
		}
3728
skip:
A
Arne Jansen 已提交
3729
		key.offset = found_key.offset + length;
C
Chris Mason 已提交
3730
		btrfs_release_path(path);
A
Arne Jansen 已提交
3731 3732 3733
	}

	btrfs_free_path(path);
3734

3735
	return ret;
A
Arne Jansen 已提交
3736 3737
}

3738 3739
static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx,
					   struct btrfs_device *scrub_dev)
A
Arne Jansen 已提交
3740 3741 3742 3743 3744
{
	int	i;
	u64	bytenr;
	u64	gen;
	int	ret;
3745
	struct btrfs_fs_info *fs_info = sctx->fs_info;
A
Arne Jansen 已提交
3746

3747
	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3748 3749
		return -EIO;

3750
	/* Seed devices of a new filesystem has their own generation. */
3751
	if (scrub_dev->fs_devices != fs_info->fs_devices)
3752 3753
		gen = scrub_dev->generation;
	else
3754
		gen = fs_info->last_trans_committed;
A
Arne Jansen 已提交
3755 3756 3757

	for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
		bytenr = btrfs_sb_offset(i);
3758 3759
		if (bytenr + BTRFS_SUPER_INFO_SIZE >
		    scrub_dev->commit_total_bytes)
A
Arne Jansen 已提交
3760 3761
			break;

3762
		ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr,
3763
				  scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i,
3764
				  NULL, 1, bytenr);
A
Arne Jansen 已提交
3765 3766 3767
		if (ret)
			return ret;
	}
3768
	wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
A
Arne Jansen 已提交
3769 3770 3771 3772 3773 3774 3775

	return 0;
}

/*
 * get a reference count on fs_info->scrub_workers. start worker if necessary
 */
3776 3777
static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info,
						int is_dev_replace)
A
Arne Jansen 已提交
3778
{
3779
	unsigned int flags = WQ_FREEZABLE | WQ_UNBOUND;
3780
	int max_active = fs_info->thread_pool_size;
A
Arne Jansen 已提交
3781

A
Arne Jansen 已提交
3782
	if (fs_info->scrub_workers_refcnt == 0) {
3783
		if (is_dev_replace)
3784
			fs_info->scrub_workers =
3785
				btrfs_alloc_workqueue(fs_info, "scrub", flags,
3786
						      1, 4);
3787
		else
3788
			fs_info->scrub_workers =
3789
				btrfs_alloc_workqueue(fs_info, "scrub", flags,
3790
						      max_active, 4);
3791 3792 3793
		if (!fs_info->scrub_workers)
			goto fail_scrub_workers;

3794
		fs_info->scrub_wr_completion_workers =
3795
			btrfs_alloc_workqueue(fs_info, "scrubwrc", flags,
3796
					      max_active, 2);
3797 3798 3799
		if (!fs_info->scrub_wr_completion_workers)
			goto fail_scrub_wr_completion_workers;

3800
		fs_info->scrub_nocow_workers =
3801
			btrfs_alloc_workqueue(fs_info, "scrubnc", flags, 1, 0);
3802 3803
		if (!fs_info->scrub_nocow_workers)
			goto fail_scrub_nocow_workers;
3804
		fs_info->scrub_parity_workers =
3805
			btrfs_alloc_workqueue(fs_info, "scrubparity", flags,
3806
					      max_active, 2);
3807 3808
		if (!fs_info->scrub_parity_workers)
			goto fail_scrub_parity_workers;
A
Arne Jansen 已提交
3809
	}
A
Arne Jansen 已提交
3810
	++fs_info->scrub_workers_refcnt;
3811 3812 3813 3814 3815 3816 3817 3818 3819 3820
	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
3821 3822
}

3823
static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info)
A
Arne Jansen 已提交
3824
{
3825
	if (--fs_info->scrub_workers_refcnt == 0) {
3826 3827 3828
		btrfs_destroy_workqueue(fs_info->scrub_workers);
		btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
		btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
3829
		btrfs_destroy_workqueue(fs_info->scrub_parity_workers);
3830
	}
A
Arne Jansen 已提交
3831 3832 3833
	WARN_ON(fs_info->scrub_workers_refcnt < 0);
}

3834 3835
int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start,
		    u64 end, struct btrfs_scrub_progress *progress,
3836
		    int readonly, int is_dev_replace)
A
Arne Jansen 已提交
3837
{
3838
	struct scrub_ctx *sctx;
A
Arne Jansen 已提交
3839 3840
	int ret;
	struct btrfs_device *dev;
3841
	struct rcu_string *name;
A
Arne Jansen 已提交
3842

3843
	if (btrfs_fs_closing(fs_info))
A
Arne Jansen 已提交
3844 3845
		return -EINVAL;

3846
	if (fs_info->nodesize > BTRFS_STRIPE_LEN) {
3847 3848 3849 3850 3851
		/*
		 * 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.
		 */
3852 3853
		btrfs_err(fs_info,
			   "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails",
3854 3855
		       fs_info->nodesize,
		       BTRFS_STRIPE_LEN);
3856 3857 3858
		return -EINVAL;
	}

3859
	if (fs_info->sectorsize != PAGE_SIZE) {
3860
		/* not supported for data w/o checksums */
3861
		btrfs_err_rl(fs_info,
J
Jeff Mahoney 已提交
3862
			   "scrub: size assumption sectorsize != PAGE_SIZE (%d != %lu) fails",
3863
		       fs_info->sectorsize, PAGE_SIZE);
A
Arne Jansen 已提交
3864 3865 3866
		return -EINVAL;
	}

3867
	if (fs_info->nodesize >
3868
	    PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK ||
3869
	    fs_info->sectorsize > PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) {
3870 3871 3872 3873
		/*
		 * would exhaust the array bounds of pagev member in
		 * struct scrub_block
		 */
J
Jeff Mahoney 已提交
3874 3875
		btrfs_err(fs_info,
			  "scrub: size assumption nodesize and sectorsize <= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails",
3876
		       fs_info->nodesize,
3877
		       SCRUB_MAX_PAGES_PER_BLOCK,
3878
		       fs_info->sectorsize,
3879 3880 3881 3882
		       SCRUB_MAX_PAGES_PER_BLOCK);
		return -EINVAL;
	}

A
Arne Jansen 已提交
3883

3884 3885
	mutex_lock(&fs_info->fs_devices->device_list_mutex);
	dev = btrfs_find_device(fs_info, devid, NULL, NULL);
3886
	if (!dev || (dev->missing && !is_dev_replace)) {
3887
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
A
Arne Jansen 已提交
3888 3889 3890
		return -ENODEV;
	}

3891 3892 3893 3894 3895 3896 3897 3898 3899 3900
	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;
	}

3901
	mutex_lock(&fs_info->scrub_lock);
3902
	if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) {
A
Arne Jansen 已提交
3903
		mutex_unlock(&fs_info->scrub_lock);
3904 3905
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
		return -EIO;
A
Arne Jansen 已提交
3906 3907
	}

3908
	btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3909 3910 3911
	if (dev->scrub_device ||
	    (!is_dev_replace &&
	     btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) {
3912
		btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
A
Arne Jansen 已提交
3913
		mutex_unlock(&fs_info->scrub_lock);
3914
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
A
Arne Jansen 已提交
3915 3916
		return -EINPROGRESS;
	}
3917
	btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3918 3919 3920 3921 3922 3923 3924 3925

	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;
	}

3926
	sctx = scrub_setup_ctx(dev, is_dev_replace);
3927
	if (IS_ERR(sctx)) {
A
Arne Jansen 已提交
3928
		mutex_unlock(&fs_info->scrub_lock);
3929 3930
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
		scrub_workers_put(fs_info);
3931
		return PTR_ERR(sctx);
A
Arne Jansen 已提交
3932
	}
3933 3934
	sctx->readonly = readonly;
	dev->scrub_device = sctx;
3935
	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
A
Arne Jansen 已提交
3936

3937 3938 3939 3940
	/*
	 * checking @scrub_pause_req here, we can avoid
	 * race between committing transaction and scrubbing.
	 */
3941
	__scrub_blocked_if_needed(fs_info);
A
Arne Jansen 已提交
3942 3943 3944
	atomic_inc(&fs_info->scrubs_running);
	mutex_unlock(&fs_info->scrub_lock);

3945
	if (!is_dev_replace) {
3946 3947 3948 3949
		/*
		 * by holding device list mutex, we can
		 * kick off writing super in log tree sync.
		 */
3950
		mutex_lock(&fs_info->fs_devices->device_list_mutex);
3951
		ret = scrub_supers(sctx, dev);
3952
		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3953
	}
A
Arne Jansen 已提交
3954 3955

	if (!ret)
3956 3957
		ret = scrub_enumerate_chunks(sctx, dev, start, end,
					     is_dev_replace);
A
Arne Jansen 已提交
3958

3959
	wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
A
Arne Jansen 已提交
3960 3961 3962
	atomic_dec(&fs_info->scrubs_running);
	wake_up(&fs_info->scrub_pause_wait);

3963
	wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0);
3964

A
Arne Jansen 已提交
3965
	if (progress)
3966
		memcpy(progress, &sctx->stat, sizeof(*progress));
A
Arne Jansen 已提交
3967 3968 3969

	mutex_lock(&fs_info->scrub_lock);
	dev->scrub_device = NULL;
3970
	scrub_workers_put(fs_info);
A
Arne Jansen 已提交
3971 3972
	mutex_unlock(&fs_info->scrub_lock);

3973
	scrub_put_ctx(sctx);
A
Arne Jansen 已提交
3974 3975 3976 3977

	return ret;
}

3978
void btrfs_scrub_pause(struct btrfs_fs_info *fs_info)
A
Arne Jansen 已提交
3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991 3992
{
	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);
}

3993
void btrfs_scrub_continue(struct btrfs_fs_info *fs_info)
A
Arne Jansen 已提交
3994 3995 3996 3997 3998
{
	atomic_dec(&fs_info->scrub_pause_req);
	wake_up(&fs_info->scrub_pause_wait);
}

3999
int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info)
A
Arne Jansen 已提交
4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019
{
	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;
}

4020 4021
int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info,
			   struct btrfs_device *dev)
4022
{
4023
	struct scrub_ctx *sctx;
A
Arne Jansen 已提交
4024 4025

	mutex_lock(&fs_info->scrub_lock);
4026 4027
	sctx = dev->scrub_device;
	if (!sctx) {
A
Arne Jansen 已提交
4028 4029 4030
		mutex_unlock(&fs_info->scrub_lock);
		return -ENOTCONN;
	}
4031
	atomic_inc(&sctx->cancel_req);
A
Arne Jansen 已提交
4032 4033 4034 4035 4036 4037 4038 4039 4040 4041
	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
4042

4043
int btrfs_scrub_progress(struct btrfs_fs_info *fs_info, u64 devid,
A
Arne Jansen 已提交
4044 4045 4046
			 struct btrfs_scrub_progress *progress)
{
	struct btrfs_device *dev;
4047
	struct scrub_ctx *sctx = NULL;
A
Arne Jansen 已提交
4048

4049 4050
	mutex_lock(&fs_info->fs_devices->device_list_mutex);
	dev = btrfs_find_device(fs_info, devid, NULL, NULL);
A
Arne Jansen 已提交
4051
	if (dev)
4052 4053 4054
		sctx = dev->scrub_device;
	if (sctx)
		memcpy(progress, &sctx->stat, sizeof(*progress));
4055
	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
A
Arne Jansen 已提交
4056

4057
	return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV;
A
Arne Jansen 已提交
4058
}
4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070

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;
4071
	ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, extent_logical,
4072 4073 4074
			      &mapped_length, &bbio, 0);
	if (ret || !bbio || mapped_length < extent_len ||
	    !bbio->stripes[0].dev->bdev) {
4075
		btrfs_put_bbio(bbio);
4076 4077 4078 4079 4080 4081
		return;
	}

	*extent_physical = bbio->stripes[0].physical;
	*extent_mirror_num = bbio->mirror_num;
	*extent_dev = bbio->stripes[0].dev;
4082
	btrfs_put_bbio(bbio);
4083 4084
}

4085
static int scrub_setup_wr_ctx(struct scrub_wr_ctx *wr_ctx,
4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096
			      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);
4097
	wr_ctx->pages_per_wr_bio = SCRUB_PAGES_PER_WR_BIO;
4098 4099 4100 4101 4102 4103 4104 4105 4106 4107 4108 4109 4110 4111 4112 4113 4114
	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;
4115
	struct btrfs_fs_info *fs_info = sctx->fs_info;
4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130 4131

	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;
4132 4133
	btrfs_init_work(&nocow_ctx->work, btrfs_scrubnc_helper,
			copy_nocow_pages_worker, NULL, NULL);
4134
	INIT_LIST_HEAD(&nocow_ctx->inodes);
4135 4136
	btrfs_queue_work(fs_info->scrub_nocow_workers,
			 &nocow_ctx->work);
4137 4138 4139 4140

	return 0;
}

4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152 4153 4154 4155 4156 4157
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

4158 4159 4160 4161 4162
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;
4163 4164
	struct btrfs_fs_info *fs_info = sctx->fs_info;
	struct btrfs_root *root = fs_info->extent_root;
4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189
	u64 logical = nocow_ctx->logical;
	u64 len = nocow_ctx->len;
	int mirror_num = nocow_ctx->mirror_num;
	u64 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
	int ret;
	struct btrfs_trans_handle *trans = NULL;
	struct btrfs_path *path;
	int not_written = 0;

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

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

	ret = iterate_inodes_from_logical(logical, fs_info, path,
4190
					  record_inode_for_nocow, nocow_ctx);
4191
	if (ret != 0 && ret != -ENOENT) {
J
Jeff Mahoney 已提交
4192 4193 4194 4195
		btrfs_warn(fs_info,
			   "iterate_inodes_from_logical() failed: log %llu, phys %llu, len %llu, mir %u, ret %d",
			   logical, physical_for_dev_replace, len, mirror_num,
			   ret);
4196 4197 4198 4199
		not_written = 1;
		goto out;
	}

4200
	btrfs_end_transaction(trans);
4201 4202 4203 4204 4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216 4217
	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;
		}
	}
4218
out:
4219 4220 4221 4222 4223 4224 4225 4226
	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);
	}
4227
	if (trans && !IS_ERR(trans))
4228
		btrfs_end_transaction(trans);
4229 4230 4231 4232 4233 4234 4235 4236 4237 4238
	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);
}

4239
static int check_extent_to_block(struct btrfs_inode *inode, u64 start, u64 len,
4240 4241 4242 4243 4244 4245 4246 4247 4248
				 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;

4249
	io_tree = &inode->io_tree;
4250

4251
	lock_extent_bits(io_tree, lockstart, lockend, &cached_state);
4252
	ordered = btrfs_lookup_ordered_range(inode, lockstart, len);
4253 4254 4255 4256 4257 4258 4259 4260 4261 4262 4263 4264 4265 4266 4267 4268 4269 4270 4271 4272 4273 4274 4275 4276 4277 4278 4279 4280 4281 4282
	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;
}

4283 4284
static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
				      struct scrub_copy_nocow_ctx *nocow_ctx)
4285
{
4286
	struct btrfs_fs_info *fs_info = nocow_ctx->sctx->fs_info;
4287
	struct btrfs_key key;
4288 4289
	struct inode *inode;
	struct page *page;
4290
	struct btrfs_root *local_root;
4291
	struct extent_io_tree *io_tree;
4292
	u64 physical_for_dev_replace;
4293
	u64 nocow_ctx_logical;
4294
	u64 len = nocow_ctx->len;
4295
	unsigned long index;
4296
	int srcu_index;
4297 4298
	int ret = 0;
	int err = 0;
4299 4300 4301 4302

	key.objectid = root;
	key.type = BTRFS_ROOT_ITEM_KEY;
	key.offset = (u64)-1;
4303 4304 4305

	srcu_index = srcu_read_lock(&fs_info->subvol_srcu);

4306
	local_root = btrfs_read_fs_root_no_name(fs_info, &key);
4307 4308
	if (IS_ERR(local_root)) {
		srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
4309
		return PTR_ERR(local_root);
4310
	}
4311 4312 4313 4314 4315

	key.type = BTRFS_INODE_ITEM_KEY;
	key.objectid = inum;
	key.offset = 0;
	inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
4316
	srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
4317 4318 4319
	if (IS_ERR(inode))
		return PTR_ERR(inode);

4320
	/* Avoid truncate/dio/punch hole.. */
A
Al Viro 已提交
4321
	inode_lock(inode);
4322 4323
	inode_dio_wait(inode);

4324
	physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
4325
	io_tree = &BTRFS_I(inode)->io_tree;
4326
	nocow_ctx_logical = nocow_ctx->logical;
4327

4328 4329
	ret = check_extent_to_block(BTRFS_I(inode), offset, len,
			nocow_ctx_logical);
4330 4331 4332
	if (ret) {
		ret = ret > 0 ? 0 : ret;
		goto out;
4333 4334
	}

4335 4336
	while (len >= PAGE_SIZE) {
		index = offset >> PAGE_SHIFT;
4337
again:
4338 4339
		page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
		if (!page) {
4340
			btrfs_err(fs_info, "find_or_create_page() failed");
4341
			ret = -ENOMEM;
4342
			goto out;
4343 4344 4345 4346 4347 4348 4349
		}

		if (PageUptodate(page)) {
			if (PageDirty(page))
				goto next_page;
		} else {
			ClearPageError(page);
4350
			err = extent_read_full_page(io_tree, page,
4351 4352
							   btrfs_get_extent,
							   nocow_ctx->mirror_num);
4353 4354
			if (err) {
				ret = err;
4355 4356
				goto next_page;
			}
4357

4358
			lock_page(page);
4359 4360 4361 4362 4363 4364 4365
			/*
			 * 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) {
4366
				unlock_page(page);
4367
				put_page(page);
4368 4369
				goto again;
			}
4370 4371 4372 4373 4374
			if (!PageUptodate(page)) {
				ret = -EIO;
				goto next_page;
			}
		}
4375

4376
		ret = check_extent_to_block(BTRFS_I(inode), offset, len,
4377 4378 4379 4380 4381 4382
					    nocow_ctx_logical);
		if (ret) {
			ret = ret > 0 ? 0 : ret;
			goto next_page;
		}

4383 4384 4385 4386
		err = write_page_nocow(nocow_ctx->sctx,
				       physical_for_dev_replace, page);
		if (err)
			ret = err;
4387
next_page:
4388
		unlock_page(page);
4389
		put_page(page);
4390 4391 4392 4393

		if (ret)
			break;

4394 4395 4396 4397
		offset += PAGE_SIZE;
		physical_for_dev_replace += PAGE_SIZE;
		nocow_ctx_logical += PAGE_SIZE;
		len -= PAGE_SIZE;
4398
	}
4399
	ret = COPY_COMPLETE;
4400
out:
A
Al Viro 已提交
4401
	inode_unlock(inode);
4402
	iput(inode);
4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416
	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) {
4417
		btrfs_warn_rl(dev->fs_info,
4418
			"scrub write_page_nocow(bdev == NULL) is unexpected");
4419 4420
		return -EIO;
	}
4421
	bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
4422 4423 4424 4425 4426 4427
	if (!bio) {
		spin_lock(&sctx->stat_lock);
		sctx->stat.malloc_errors++;
		spin_unlock(&sctx->stat_lock);
		return -ENOMEM;
	}
4428 4429
	bio->bi_iter.bi_size = 0;
	bio->bi_iter.bi_sector = physical_for_dev_replace >> 9;
4430
	bio->bi_bdev = dev->bdev;
4431
	bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
4432 4433
	ret = bio_add_page(bio, page, PAGE_SIZE, 0);
	if (ret != PAGE_SIZE) {
4434 4435 4436 4437 4438 4439
leave_with_eio:
		bio_put(bio);
		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
		return -EIO;
	}

4440
	if (btrfsic_submit_bio_wait(bio))
4441 4442 4443 4444 4445
		goto leave_with_eio;

	bio_put(bio);
	return 0;
}