raid1.c 86.9 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7 8 9 10 11
/*
 * raid1.c : Multiple Devices driver for Linux
 *
 * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat
 *
 * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman
 *
 * RAID-1 management functions.
 *
 * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000
 *
12
 * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk>
L
Linus Torvalds 已提交
13 14
 * Various fixes by Neil Brown <neilb@cse.unsw.edu.au>
 *
15 16 17 18 19 20 21 22 23
 * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support
 * bitmapped intelligence in resync:
 *
 *      - bitmap marked during normal i/o
 *      - bitmap used to skip nondirty blocks during sync
 *
 * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology:
 * - persistent bitmap code
 *
L
Linus Torvalds 已提交
24 25 26 27 28 29 30 31 32 33
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2, or (at your option)
 * any later version.
 *
 * You should have received a copy of the GNU General Public License
 * (for example /usr/src/linux/COPYING); if not, write to the Free
 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */

34
#include <linux/slab.h>
35
#include <linux/delay.h>
36
#include <linux/blkdev.h>
37
#include <linux/module.h>
38
#include <linux/seq_file.h>
39
#include <linux/ratelimit.h>
40
#include "md.h"
41 42
#include "raid1.h"
#include "bitmap.h"
43

L
Linus Torvalds 已提交
44 45 46 47 48
/*
 * Number of guaranteed r1bios in case of extreme VM load:
 */
#define	NR_RAID1_BIOS 256

49 50 51 52 53 54 55 56 57 58 59 60 61 62
/* when we get a read error on a read-only array, we redirect to another
 * device without failing the first device, or trying to over-write to
 * correct the read error.  To keep track of bad blocks on a per-bio
 * level, we store IO_BLOCKED in the appropriate 'bios' pointer
 */
#define IO_BLOCKED ((struct bio *)1)
/* When we successfully write to a known bad-block, we need to remove the
 * bad-block marking which must be done from process context.  So we record
 * the success by setting devs[n].bio to IO_MADE_GOOD
 */
#define IO_MADE_GOOD ((struct bio *)2)

#define BIO_SPECIAL(bio) ((unsigned long)bio <= 2)

63 64 65 66 67
/* When there are this many requests queue to be written by
 * the raid1 thread, we become 'congested' to provide back-pressure
 * for writeback.
 */
static int max_queued_requests = 1024;
L
Linus Torvalds 已提交
68

69 70
static void allow_barrier(struct r1conf *conf, sector_t start_next_window,
			  sector_t bi_sector);
71
static void lower_barrier(struct r1conf *conf);
L
Linus Torvalds 已提交
72

A
Al Viro 已提交
73
static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data)
L
Linus Torvalds 已提交
74 75
{
	struct pool_info *pi = data;
76
	int size = offsetof(struct r1bio, bios[pi->raid_disks]);
L
Linus Torvalds 已提交
77 78

	/* allocate a r1bio with room for raid_disks entries in the bios array */
J
Jens Axboe 已提交
79
	return kzalloc(size, gfp_flags);
L
Linus Torvalds 已提交
80 81 82 83 84 85 86 87
}

static void r1bio_pool_free(void *r1_bio, void *data)
{
	kfree(r1_bio);
}

#define RESYNC_BLOCK_SIZE (64*1024)
88
#define RESYNC_DEPTH 32
L
Linus Torvalds 已提交
89 90
#define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
#define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
91 92
#define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH)
#define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9)
93 94
#define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW)
#define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9)
95
#define NEXT_NORMALIO_DISTANCE (3 * RESYNC_WINDOW_SECTORS)
L
Linus Torvalds 已提交
96

A
Al Viro 已提交
97
static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
L
Linus Torvalds 已提交
98 99
{
	struct pool_info *pi = data;
100
	struct r1bio *r1_bio;
L
Linus Torvalds 已提交
101
	struct bio *bio;
102
	int need_pages;
L
Linus Torvalds 已提交
103 104 105
	int i, j;

	r1_bio = r1bio_pool_alloc(gfp_flags, pi);
J
Jens Axboe 已提交
106
	if (!r1_bio)
L
Linus Torvalds 已提交
107 108 109 110 111 112
		return NULL;

	/*
	 * Allocate bios : 1 for reading, n-1 for writing
	 */
	for (j = pi->raid_disks ; j-- ; ) {
113
		bio = bio_kmalloc(gfp_flags, RESYNC_PAGES);
L
Linus Torvalds 已提交
114 115 116 117 118 119
		if (!bio)
			goto out_free_bio;
		r1_bio->bios[j] = bio;
	}
	/*
	 * Allocate RESYNC_PAGES data pages and attach them to
120 121 122
	 * the first bio.
	 * If this is a user-requested check/repair, allocate
	 * RESYNC_PAGES for each bio.
L
Linus Torvalds 已提交
123
	 */
124
	if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery))
125
		need_pages = pi->raid_disks;
126
	else
127 128
		need_pages = 1;
	for (j = 0; j < need_pages; j++) {
129
		bio = r1_bio->bios[j];
K
Kent Overstreet 已提交
130
		bio->bi_vcnt = RESYNC_PAGES;
131

K
Kent Overstreet 已提交
132
		if (bio_alloc_pages(bio, gfp_flags))
133
			goto out_free_pages;
134 135 136 137 138 139 140
	}
	/* If not user-requests, copy the page pointers to all bios */
	if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) {
		for (i=0; i<RESYNC_PAGES ; i++)
			for (j=1; j<pi->raid_disks; j++)
				r1_bio->bios[j]->bi_io_vec[i].bv_page =
					r1_bio->bios[0]->bi_io_vec[i].bv_page;
L
Linus Torvalds 已提交
141 142 143 144 145 146
	}

	r1_bio->master_bio = NULL;

	return r1_bio;

147 148 149 150 151 152 153 154
out_free_pages:
	while (--j >= 0) {
		struct bio_vec *bv;

		bio_for_each_segment_all(bv, r1_bio->bios[j], i)
			__free_page(bv->bv_page);
	}

L
Linus Torvalds 已提交
155
out_free_bio:
156
	while (++j < pi->raid_disks)
L
Linus Torvalds 已提交
157 158 159 160 161 162 163 164
		bio_put(r1_bio->bios[j]);
	r1bio_pool_free(r1_bio, data);
	return NULL;
}

static void r1buf_pool_free(void *__r1_bio, void *data)
{
	struct pool_info *pi = data;
165
	int i,j;
166
	struct r1bio *r1bio = __r1_bio;
L
Linus Torvalds 已提交
167

168 169 170 171 172
	for (i = 0; i < RESYNC_PAGES; i++)
		for (j = pi->raid_disks; j-- ;) {
			if (j == 0 ||
			    r1bio->bios[j]->bi_io_vec[i].bv_page !=
			    r1bio->bios[0]->bi_io_vec[i].bv_page)
173
				safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page);
174
		}
L
Linus Torvalds 已提交
175 176 177 178 179 180
	for (i=0 ; i < pi->raid_disks; i++)
		bio_put(r1bio->bios[i]);

	r1bio_pool_free(r1bio, data);
}

181
static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio)
L
Linus Torvalds 已提交
182 183 184
{
	int i;

185
	for (i = 0; i < conf->raid_disks * 2; i++) {
L
Linus Torvalds 已提交
186
		struct bio **bio = r1_bio->bios + i;
187
		if (!BIO_SPECIAL(*bio))
L
Linus Torvalds 已提交
188 189 190 191 192
			bio_put(*bio);
		*bio = NULL;
	}
}

193
static void free_r1bio(struct r1bio *r1_bio)
L
Linus Torvalds 已提交
194
{
195
	struct r1conf *conf = r1_bio->mddev->private;
L
Linus Torvalds 已提交
196 197 198 199 200

	put_all_bios(conf, r1_bio);
	mempool_free(r1_bio, conf->r1bio_pool);
}

201
static void put_buf(struct r1bio *r1_bio)
L
Linus Torvalds 已提交
202
{
203
	struct r1conf *conf = r1_bio->mddev->private;
204 205
	int i;

206
	for (i = 0; i < conf->raid_disks * 2; i++) {
207 208 209 210
		struct bio *bio = r1_bio->bios[i];
		if (bio->bi_end_io)
			rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
	}
L
Linus Torvalds 已提交
211 212 213

	mempool_free(r1_bio, conf->r1buf_pool);

214
	lower_barrier(conf);
L
Linus Torvalds 已提交
215 216
}

217
static void reschedule_retry(struct r1bio *r1_bio)
L
Linus Torvalds 已提交
218 219
{
	unsigned long flags;
220
	struct mddev *mddev = r1_bio->mddev;
221
	struct r1conf *conf = mddev->private;
L
Linus Torvalds 已提交
222 223 224

	spin_lock_irqsave(&conf->device_lock, flags);
	list_add(&r1_bio->retry_list, &conf->retry_list);
225
	conf->nr_queued ++;
L
Linus Torvalds 已提交
226 227
	spin_unlock_irqrestore(&conf->device_lock, flags);

228
	wake_up(&conf->wait_barrier);
L
Linus Torvalds 已提交
229 230 231 232 233 234 235 236
	md_wakeup_thread(mddev->thread);
}

/*
 * raid_end_bio_io() is called when we have finished servicing a mirrored
 * operation and are ready to return a success/failure code to the buffer
 * cache layer.
 */
237
static void call_bio_endio(struct r1bio *r1_bio)
238 239 240
{
	struct bio *bio = r1_bio->master_bio;
	int done;
241
	struct r1conf *conf = r1_bio->mddev->private;
242
	sector_t start_next_window = r1_bio->start_next_window;
243
	sector_t bi_sector = bio->bi_iter.bi_sector;
244 245 246 247 248 249 250

	if (bio->bi_phys_segments) {
		unsigned long flags;
		spin_lock_irqsave(&conf->device_lock, flags);
		bio->bi_phys_segments--;
		done = (bio->bi_phys_segments == 0);
		spin_unlock_irqrestore(&conf->device_lock, flags);
251 252 253 254 255
		/*
		 * make_request() might be waiting for
		 * bi_phys_segments to decrease
		 */
		wake_up(&conf->wait_barrier);
256 257 258 259
	} else
		done = 1;

	if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
260 261
		bio->bi_error = -EIO;

262
	if (done) {
263
		bio_endio(bio);
264 265 266 267
		/*
		 * Wake up any possible resync thread that waits for the device
		 * to go idle.
		 */
268
		allow_barrier(conf, start_next_window, bi_sector);
269 270 271
	}
}

272
static void raid_end_bio_io(struct r1bio *r1_bio)
L
Linus Torvalds 已提交
273 274 275
{
	struct bio *bio = r1_bio->master_bio;

276 277
	/* if nobody has done the final endio yet, do it now */
	if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
278 279
		pr_debug("raid1: sync end %s on sectors %llu-%llu\n",
			 (bio_data_dir(bio) == WRITE) ? "write" : "read",
280 281
			 (unsigned long long) bio->bi_iter.bi_sector,
			 (unsigned long long) bio_end_sector(bio) - 1);
282

283
		call_bio_endio(r1_bio);
284
	}
L
Linus Torvalds 已提交
285 286 287 288 289 290
	free_r1bio(r1_bio);
}

/*
 * Update disk head position estimator based on IRQ completion info.
 */
291
static inline void update_head_pos(int disk, struct r1bio *r1_bio)
L
Linus Torvalds 已提交
292
{
293
	struct r1conf *conf = r1_bio->mddev->private;
L
Linus Torvalds 已提交
294 295 296 297 298

	conf->mirrors[disk].head_position =
		r1_bio->sector + (r1_bio->sectors);
}

299 300 301
/*
 * Find the disk number which triggered given bio
 */
302
static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio)
303 304
{
	int mirror;
305 306
	struct r1conf *conf = r1_bio->mddev->private;
	int raid_disks = conf->raid_disks;
307

308
	for (mirror = 0; mirror < raid_disks * 2; mirror++)
309 310 311
		if (r1_bio->bios[mirror] == bio)
			break;

312
	BUG_ON(mirror == raid_disks * 2);
313 314 315 316 317
	update_head_pos(mirror, r1_bio);

	return mirror;
}

318
static void raid1_end_read_request(struct bio *bio)
L
Linus Torvalds 已提交
319
{
320
	int uptodate = !bio->bi_error;
321
	struct r1bio *r1_bio = bio->bi_private;
L
Linus Torvalds 已提交
322
	int mirror;
323
	struct r1conf *conf = r1_bio->mddev->private;
L
Linus Torvalds 已提交
324 325 326 327 328

	mirror = r1_bio->read_disk;
	/*
	 * this branch is our 'one mirror IO has finished' event handler:
	 */
329 330
	update_head_pos(mirror, r1_bio);

331 332 333 334 335 336
	if (uptodate)
		set_bit(R1BIO_Uptodate, &r1_bio->state);
	else {
		/* If all other devices have failed, we want to return
		 * the error upwards rather than fail the last device.
		 * Here we redefine "uptodate" to mean "Don't want to retry"
L
Linus Torvalds 已提交
337
		 */
338 339 340 341
		unsigned long flags;
		spin_lock_irqsave(&conf->device_lock, flags);
		if (r1_bio->mddev->degraded == conf->raid_disks ||
		    (r1_bio->mddev->degraded == conf->raid_disks-1 &&
342
		     test_bit(In_sync, &conf->mirrors[mirror].rdev->flags)))
343 344 345
			uptodate = 1;
		spin_unlock_irqrestore(&conf->device_lock, flags);
	}
L
Linus Torvalds 已提交
346

347
	if (uptodate) {
L
Linus Torvalds 已提交
348
		raid_end_bio_io(r1_bio);
349 350
		rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
	} else {
L
Linus Torvalds 已提交
351 352 353 354
		/*
		 * oops, read error:
		 */
		char b[BDEVNAME_SIZE];
355 356 357 358 359 360 361
		printk_ratelimited(
			KERN_ERR "md/raid1:%s: %s: "
			"rescheduling sector %llu\n",
			mdname(conf->mddev),
			bdevname(conf->mirrors[mirror].rdev->bdev,
				 b),
			(unsigned long long)r1_bio->sector);
362
		set_bit(R1BIO_ReadError, &r1_bio->state);
L
Linus Torvalds 已提交
363
		reschedule_retry(r1_bio);
364
		/* don't drop the reference on read_disk yet */
L
Linus Torvalds 已提交
365 366 367
	}
}

368
static void close_write(struct r1bio *r1_bio)
369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386
{
	/* it really is the end of this request */
	if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
		/* free extra copy of the data pages */
		int i = r1_bio->behind_page_count;
		while (i--)
			safe_put_page(r1_bio->behind_bvecs[i].bv_page);
		kfree(r1_bio->behind_bvecs);
		r1_bio->behind_bvecs = NULL;
	}
	/* clear the bitmap if all writes complete successfully */
	bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
			r1_bio->sectors,
			!test_bit(R1BIO_Degraded, &r1_bio->state),
			test_bit(R1BIO_BehindIO, &r1_bio->state));
	md_write_end(r1_bio->mddev);
}

387
static void r1_bio_write_done(struct r1bio *r1_bio)
388
{
389 390 391 392 393 394 395
	if (!atomic_dec_and_test(&r1_bio->remaining))
		return;

	if (test_bit(R1BIO_WriteError, &r1_bio->state))
		reschedule_retry(r1_bio);
	else {
		close_write(r1_bio);
396 397 398 399
		if (test_bit(R1BIO_MadeGood, &r1_bio->state))
			reschedule_retry(r1_bio);
		else
			raid_end_bio_io(r1_bio);
400 401 402
	}
}

403
static void raid1_end_write_request(struct bio *bio)
L
Linus Torvalds 已提交
404
{
405
	struct r1bio *r1_bio = bio->bi_private;
406
	int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
407
	struct r1conf *conf = r1_bio->mddev->private;
408
	struct bio *to_put = NULL;
L
Linus Torvalds 已提交
409

410
	mirror = find_bio_disk(r1_bio, bio);
L
Linus Torvalds 已提交
411

T
Tejun Heo 已提交
412 413 414
	/*
	 * 'one mirror IO has finished' event handler:
	 */
415
	if (bio->bi_error) {
416 417
		set_bit(WriteErrorSeen,
			&conf->mirrors[mirror].rdev->flags);
418 419 420 421 422
		if (!test_and_set_bit(WantReplacement,
				      &conf->mirrors[mirror].rdev->flags))
			set_bit(MD_RECOVERY_NEEDED, &
				conf->mddev->recovery);

423
		set_bit(R1BIO_WriteError, &r1_bio->state);
424
	} else {
L
Linus Torvalds 已提交
425
		/*
T
Tejun Heo 已提交
426 427 428 429 430 431 432 433
		 * Set R1BIO_Uptodate in our master bio, so that we
		 * will return a good error code for to the higher
		 * levels even if IO on some other mirrored buffer
		 * fails.
		 *
		 * The 'master' represents the composite IO operation
		 * to user-side. So if something waits for IO, then it
		 * will wait for the 'master' bio.
L
Linus Torvalds 已提交
434
		 */
435 436 437
		sector_t first_bad;
		int bad_sectors;

438 439
		r1_bio->bios[mirror] = NULL;
		to_put = bio;
440 441 442 443 444 445 446 447 448 449 450
		/*
		 * Do not set R1BIO_Uptodate if the current device is
		 * rebuilding or Faulty. This is because we cannot use
		 * such device for properly reading the data back (we could
		 * potentially use it, if the current write would have felt
		 * before rdev->recovery_offset, but for simplicity we don't
		 * check this here.
		 */
		if (test_bit(In_sync, &conf->mirrors[mirror].rdev->flags) &&
		    !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags))
			set_bit(R1BIO_Uptodate, &r1_bio->state);
T
Tejun Heo 已提交
451

452 453 454 455 456 457 458 459 460
		/* Maybe we can clear some bad blocks. */
		if (is_badblock(conf->mirrors[mirror].rdev,
				r1_bio->sector, r1_bio->sectors,
				&first_bad, &bad_sectors)) {
			r1_bio->bios[mirror] = IO_MADE_GOOD;
			set_bit(R1BIO_MadeGood, &r1_bio->state);
		}
	}

T
Tejun Heo 已提交
461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476
	if (behind) {
		if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags))
			atomic_dec(&r1_bio->behind_remaining);

		/*
		 * In behind mode, we ACK the master bio once the I/O
		 * has safely reached all non-writemostly
		 * disks. Setting the Returned bit ensures that this
		 * gets done only once -- we don't ever want to return
		 * -EIO here, instead we'll wait
		 */
		if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
		    test_bit(R1BIO_Uptodate, &r1_bio->state)) {
			/* Maybe we can return now */
			if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
				struct bio *mbio = r1_bio->master_bio;
477 478
				pr_debug("raid1: behind end write sectors"
					 " %llu-%llu\n",
479 480
					 (unsigned long long) mbio->bi_iter.bi_sector,
					 (unsigned long long) bio_end_sector(mbio) - 1);
481
				call_bio_endio(r1_bio);
482 483 484
			}
		}
	}
485 486 487
	if (r1_bio->bios[mirror] == NULL)
		rdev_dec_pending(conf->mirrors[mirror].rdev,
				 conf->mddev);
T
Tejun Heo 已提交
488

L
Linus Torvalds 已提交
489 490 491 492
	/*
	 * Let's see if all mirrored write operations have finished
	 * already.
	 */
493
	r1_bio_write_done(r1_bio);
494

495 496
	if (to_put)
		bio_put(to_put);
L
Linus Torvalds 已提交
497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512
}

/*
 * This routine returns the disk from which the requested read should
 * be done. There is a per-array 'next expected sequential IO' sector
 * number - if this matches on the next IO then we use the last disk.
 * There is also a per-disk 'last know head position' sector that is
 * maintained from IRQ contexts, both the normal and the resync IO
 * completion handlers update this position correctly. If there is no
 * perfect sequential match then we pick the disk whose head is closest.
 *
 * If there are 2 mirrors in the same 2 devices, performance degrades
 * because position is mirror, not device based.
 *
 * The rdev for the device selected will have nr_pending incremented.
 */
513
static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors)
L
Linus Torvalds 已提交
514
{
515
	const sector_t this_sector = r1_bio->sector;
516 517
	int sectors;
	int best_good_sectors;
518 519
	int best_disk, best_dist_disk, best_pending_disk;
	int has_nonrot_disk;
520
	int disk;
N
NeilBrown 已提交
521
	sector_t best_dist;
522
	unsigned int min_pending;
523
	struct md_rdev *rdev;
524
	int choose_first;
525
	int choose_next_idle;
L
Linus Torvalds 已提交
526 527 528

	rcu_read_lock();
	/*
529
	 * Check if we can balance. We can balance on the whole
L
Linus Torvalds 已提交
530 531 532 533
	 * device if no resync is going on, or below the resync window.
	 * We take the first readable disk when above the resync window.
	 */
 retry:
534
	sectors = r1_bio->sectors;
N
NeilBrown 已提交
535
	best_disk = -1;
536
	best_dist_disk = -1;
N
NeilBrown 已提交
537
	best_dist = MaxSector;
538 539
	best_pending_disk = -1;
	min_pending = UINT_MAX;
540
	best_good_sectors = 0;
541
	has_nonrot_disk = 0;
542
	choose_next_idle = 0;
543

544 545
	if ((conf->mddev->recovery_cp < this_sector + sectors) ||
	    (mddev_is_clustered(conf->mddev) &&
546
	    md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector,
547 548 549 550
		    this_sector + sectors)))
		choose_first = 1;
	else
		choose_first = 0;
L
Linus Torvalds 已提交
551

552
	for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {
N
NeilBrown 已提交
553
		sector_t dist;
554 555
		sector_t first_bad;
		int bad_sectors;
556
		unsigned int pending;
557
		bool nonrot;
558

559 560 561
		rdev = rcu_dereference(conf->mirrors[disk].rdev);
		if (r1_bio->bios[disk] == IO_BLOCKED
		    || rdev == NULL
N
NeilBrown 已提交
562
		    || test_bit(Faulty, &rdev->flags))
563
			continue;
N
NeilBrown 已提交
564 565
		if (!test_bit(In_sync, &rdev->flags) &&
		    rdev->recovery_offset < this_sector + sectors)
L
Linus Torvalds 已提交
566
			continue;
N
NeilBrown 已提交
567 568 569
		if (test_bit(WriteMostly, &rdev->flags)) {
			/* Don't balance among write-mostly, just
			 * use the first as a last resort */
570
			if (best_dist_disk < 0) {
571 572
				if (is_badblock(rdev, this_sector, sectors,
						&first_bad, &bad_sectors)) {
573
					if (first_bad <= this_sector)
574 575 576 577 578
						/* Cannot use this */
						continue;
					best_good_sectors = first_bad - this_sector;
				} else
					best_good_sectors = sectors;
579 580
				best_dist_disk = disk;
				best_pending_disk = disk;
581
			}
N
NeilBrown 已提交
582 583 584 585 586
			continue;
		}
		/* This is a reasonable device to use.  It might
		 * even be best.
		 */
587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615
		if (is_badblock(rdev, this_sector, sectors,
				&first_bad, &bad_sectors)) {
			if (best_dist < MaxSector)
				/* already have a better device */
				continue;
			if (first_bad <= this_sector) {
				/* cannot read here. If this is the 'primary'
				 * device, then we must not read beyond
				 * bad_sectors from another device..
				 */
				bad_sectors -= (this_sector - first_bad);
				if (choose_first && sectors > bad_sectors)
					sectors = bad_sectors;
				if (best_good_sectors > sectors)
					best_good_sectors = sectors;

			} else {
				sector_t good_sectors = first_bad - this_sector;
				if (good_sectors > best_good_sectors) {
					best_good_sectors = good_sectors;
					best_disk = disk;
				}
				if (choose_first)
					break;
			}
			continue;
		} else
			best_good_sectors = sectors;

616 617
		nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev));
		has_nonrot_disk |= nonrot;
618
		pending = atomic_read(&rdev->nr_pending);
N
NeilBrown 已提交
619
		dist = abs(this_sector - conf->mirrors[disk].head_position);
620
		if (choose_first) {
N
NeilBrown 已提交
621
			best_disk = disk;
L
Linus Torvalds 已提交
622 623
			break;
		}
624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661
		/* Don't change to another disk for sequential reads */
		if (conf->mirrors[disk].next_seq_sect == this_sector
		    || dist == 0) {
			int opt_iosize = bdev_io_opt(rdev->bdev) >> 9;
			struct raid1_info *mirror = &conf->mirrors[disk];

			best_disk = disk;
			/*
			 * If buffered sequential IO size exceeds optimal
			 * iosize, check if there is idle disk. If yes, choose
			 * the idle disk. read_balance could already choose an
			 * idle disk before noticing it's a sequential IO in
			 * this disk. This doesn't matter because this disk
			 * will idle, next time it will be utilized after the
			 * first disk has IO size exceeds optimal iosize. In
			 * this way, iosize of the first disk will be optimal
			 * iosize at least. iosize of the second disk might be
			 * small, but not a big deal since when the second disk
			 * starts IO, the first disk is likely still busy.
			 */
			if (nonrot && opt_iosize > 0 &&
			    mirror->seq_start != MaxSector &&
			    mirror->next_seq_sect > opt_iosize &&
			    mirror->next_seq_sect - opt_iosize >=
			    mirror->seq_start) {
				choose_next_idle = 1;
				continue;
			}
			break;
		}
		/* If device is idle, use it */
		if (pending == 0) {
			best_disk = disk;
			break;
		}

		if (choose_next_idle)
			continue;
662 663 664 665 666 667

		if (min_pending > pending) {
			min_pending = pending;
			best_pending_disk = disk;
		}

N
NeilBrown 已提交
668 669
		if (dist < best_dist) {
			best_dist = dist;
670
			best_dist_disk = disk;
L
Linus Torvalds 已提交
671
		}
672
	}
L
Linus Torvalds 已提交
673

674 675 676 677 678 679 680 681 682 683 684 685 686
	/*
	 * If all disks are rotational, choose the closest disk. If any disk is
	 * non-rotational, choose the disk with less pending request even the
	 * disk is rotational, which might/might not be optimal for raids with
	 * mixed ratation/non-rotational disks depending on workload.
	 */
	if (best_disk == -1) {
		if (has_nonrot_disk)
			best_disk = best_pending_disk;
		else
			best_disk = best_dist_disk;
	}

N
NeilBrown 已提交
687 688
	if (best_disk >= 0) {
		rdev = rcu_dereference(conf->mirrors[best_disk].rdev);
689 690 691
		if (!rdev)
			goto retry;
		atomic_inc(&rdev->nr_pending);
N
NeilBrown 已提交
692
		if (test_bit(Faulty, &rdev->flags)) {
L
Linus Torvalds 已提交
693 694 695
			/* cannot risk returning a device that failed
			 * before we inc'ed nr_pending
			 */
696
			rdev_dec_pending(rdev, conf->mddev);
L
Linus Torvalds 已提交
697 698
			goto retry;
		}
699
		sectors = best_good_sectors;
700 701 702 703

		if (conf->mirrors[best_disk].next_seq_sect != this_sector)
			conf->mirrors[best_disk].seq_start = this_sector;

704
		conf->mirrors[best_disk].next_seq_sect = this_sector + sectors;
L
Linus Torvalds 已提交
705 706
	}
	rcu_read_unlock();
707
	*max_sectors = sectors;
L
Linus Torvalds 已提交
708

N
NeilBrown 已提交
709
	return best_disk;
L
Linus Torvalds 已提交
710 711
}

712
static int raid1_congested(struct mddev *mddev, int bits)
713
{
714
	struct r1conf *conf = mddev->private;
715 716
	int i, ret = 0;

717
	if ((bits & (1 << WB_async_congested)) &&
718 719 720
	    conf->pending_count >= max_queued_requests)
		return 1;

721
	rcu_read_lock();
722
	for (i = 0; i < conf->raid_disks * 2; i++) {
723
		struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
724
		if (rdev && !test_bit(Faulty, &rdev->flags)) {
725
			struct request_queue *q = bdev_get_queue(rdev->bdev);
726

727 728
			BUG_ON(!q);

729 730 731
			/* Note the '|| 1' - when read_balance prefers
			 * non-congested targets, it can be removed
			 */
732
			if ((bits & (1 << WB_async_congested)) || 1)
733 734 735 736 737 738 739 740 741
				ret |= bdi_congested(&q->backing_dev_info, bits);
			else
				ret &= bdi_congested(&q->backing_dev_info, bits);
		}
	}
	rcu_read_unlock();
	return ret;
}

742
static void flush_pending_writes(struct r1conf *conf)
743 744 745 746 747 748 749 750 751
{
	/* Any writes that have been queued but are awaiting
	 * bitmap updates get flushed here.
	 */
	spin_lock_irq(&conf->device_lock);

	if (conf->pending_bio_list.head) {
		struct bio *bio;
		bio = bio_list_get(&conf->pending_bio_list);
752
		conf->pending_count = 0;
753 754 755 756
		spin_unlock_irq(&conf->device_lock);
		/* flush any pending bitmap writes to
		 * disk before proceeding w/ I/O */
		bitmap_unplug(conf->mddev->bitmap);
757
		wake_up(&conf->wait_barrier);
758 759 760 761

		while (bio) { /* submit pending writes */
			struct bio *next = bio->bi_next;
			bio->bi_next = NULL;
M
Mike Christie 已提交
762
			if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
S
Shaohua Li 已提交
763 764
			    !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
				/* Just ignore it */
765
				bio_endio(bio);
S
Shaohua Li 已提交
766 767
			else
				generic_make_request(bio);
768 769 770 771
			bio = next;
		}
	} else
		spin_unlock_irq(&conf->device_lock);
J
Jens Axboe 已提交
772 773
}

774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793
/* Barriers....
 * Sometimes we need to suspend IO while we do something else,
 * either some resync/recovery, or reconfigure the array.
 * To do this we raise a 'barrier'.
 * The 'barrier' is a counter that can be raised multiple times
 * to count how many activities are happening which preclude
 * normal IO.
 * We can only raise the barrier if there is no pending IO.
 * i.e. if nr_pending == 0.
 * We choose only to raise the barrier if no-one is waiting for the
 * barrier to go down.  This means that as soon as an IO request
 * is ready, no other operations which require a barrier will start
 * until the IO request has had a chance.
 *
 * So: regular IO calls 'wait_barrier'.  When that returns there
 *    is no backgroup IO happening,  It must arrange to call
 *    allow_barrier when it has finished its IO.
 * backgroup IO calls must call raise_barrier.  Once that returns
 *    there is no normal IO happeing.  It must arrange to call
 *    lower_barrier when the particular background IO completes.
L
Linus Torvalds 已提交
794
 */
795
static void raise_barrier(struct r1conf *conf, sector_t sector_nr)
L
Linus Torvalds 已提交
796 797
{
	spin_lock_irq(&conf->resync_lock);
798 799 800

	/* Wait until no block IO is waiting */
	wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
801
			    conf->resync_lock);
802 803 804

	/* block any new IO from starting */
	conf->barrier++;
805
	conf->next_resync = sector_nr;
806

807 808 809 810 811 812 813
	/* For these conditions we must wait:
	 * A: while the array is in frozen state
	 * B: while barrier >= RESYNC_DEPTH, meaning resync reach
	 *    the max count which allowed.
	 * C: next_resync + RESYNC_SECTORS > start_next_window, meaning
	 *    next resync will reach to the window which normal bios are
	 *    handling.
814
	 * D: while there are any active requests in the current window.
815
	 */
816
	wait_event_lock_irq(conf->wait_barrier,
817
			    !conf->array_frozen &&
818
			    conf->barrier < RESYNC_DEPTH &&
819
			    conf->current_window_requests == 0 &&
820 821
			    (conf->start_next_window >=
			     conf->next_resync + RESYNC_SECTORS),
822
			    conf->resync_lock);
823

824
	conf->nr_pending++;
825 826 827
	spin_unlock_irq(&conf->resync_lock);
}

828
static void lower_barrier(struct r1conf *conf)
829 830
{
	unsigned long flags;
831
	BUG_ON(conf->barrier <= 0);
832 833
	spin_lock_irqsave(&conf->resync_lock, flags);
	conf->barrier--;
834
	conf->nr_pending--;
835 836 837 838
	spin_unlock_irqrestore(&conf->resync_lock, flags);
	wake_up(&conf->wait_barrier);
}

839
static bool need_to_wait_for_sync(struct r1conf *conf, struct bio *bio)
840
{
841 842 843 844 845
	bool wait = false;

	if (conf->array_frozen || !bio)
		wait = true;
	else if (conf->barrier && bio_data_dir(bio) == WRITE) {
846 847 848 849
		if ((conf->mddev->curr_resync_completed
		     >= bio_end_sector(bio)) ||
		    (conf->next_resync + NEXT_NORMALIO_DISTANCE
		     <= bio->bi_iter.bi_sector))
850 851 852 853 854 855 856 857 858 859 860 861
			wait = false;
		else
			wait = true;
	}

	return wait;
}

static sector_t wait_barrier(struct r1conf *conf, struct bio *bio)
{
	sector_t sector = 0;

862
	spin_lock_irq(&conf->resync_lock);
863
	if (need_to_wait_for_sync(conf, bio)) {
864
		conf->nr_waiting++;
865 866 867 868
		/* Wait for the barrier to drop.
		 * However if there are already pending
		 * requests (preventing the barrier from
		 * rising completely), and the
869
		 * per-process bio queue isn't empty,
870
		 * then don't wait, as we need to empty
871 872
		 * that queue to allow conf->start_next_window
		 * to increase.
873 874
		 */
		wait_event_lock_irq(conf->wait_barrier,
875 876
				    !conf->array_frozen &&
				    (!conf->barrier ||
877 878 879 880
				     ((conf->start_next_window <
				       conf->next_resync + RESYNC_SECTORS) &&
				      current->bio_list &&
				      !bio_list_empty(current->bio_list))),
881
				    conf->resync_lock);
882
		conf->nr_waiting--;
L
Linus Torvalds 已提交
883
	}
884 885

	if (bio && bio_data_dir(bio) == WRITE) {
886
		if (bio->bi_iter.bi_sector >= conf->next_resync) {
887 888 889 890 891 892
			if (conf->start_next_window == MaxSector)
				conf->start_next_window =
					conf->next_resync +
					NEXT_NORMALIO_DISTANCE;

			if ((conf->start_next_window + NEXT_NORMALIO_DISTANCE)
893
			    <= bio->bi_iter.bi_sector)
894 895 896 897
				conf->next_window_requests++;
			else
				conf->current_window_requests++;
			sector = conf->start_next_window;
898
		}
899 900
	}

901
	conf->nr_pending++;
L
Linus Torvalds 已提交
902
	spin_unlock_irq(&conf->resync_lock);
903
	return sector;
L
Linus Torvalds 已提交
904 905
}

906 907
static void allow_barrier(struct r1conf *conf, sector_t start_next_window,
			  sector_t bi_sector)
908 909
{
	unsigned long flags;
910

911 912
	spin_lock_irqsave(&conf->resync_lock, flags);
	conf->nr_pending--;
913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933
	if (start_next_window) {
		if (start_next_window == conf->start_next_window) {
			if (conf->start_next_window + NEXT_NORMALIO_DISTANCE
			    <= bi_sector)
				conf->next_window_requests--;
			else
				conf->current_window_requests--;
		} else
			conf->current_window_requests--;

		if (!conf->current_window_requests) {
			if (conf->next_window_requests) {
				conf->current_window_requests =
					conf->next_window_requests;
				conf->next_window_requests = 0;
				conf->start_next_window +=
					NEXT_NORMALIO_DISTANCE;
			} else
				conf->start_next_window = MaxSector;
		}
	}
934 935 936 937
	spin_unlock_irqrestore(&conf->resync_lock, flags);
	wake_up(&conf->wait_barrier);
}

938
static void freeze_array(struct r1conf *conf, int extra)
939 940 941
{
	/* stop syncio and normal IO and wait for everything to
	 * go quite.
942
	 * We wait until nr_pending match nr_queued+extra
943 944 945 946
	 * This is called in the context of one normal IO request
	 * that has failed. Thus any sync request that might be pending
	 * will be blocked by nr_pending, and we need to wait for
	 * pending IO requests to complete or be queued for re-try.
947
	 * Thus the number queued (nr_queued) plus this request (extra)
948 949
	 * must match the number of pending IOs (nr_pending) before
	 * we continue.
950 951
	 */
	spin_lock_irq(&conf->resync_lock);
952
	conf->array_frozen = 1;
953
	wait_event_lock_irq_cmd(conf->wait_barrier,
954
				conf->nr_pending == conf->nr_queued+extra,
955 956
				conf->resync_lock,
				flush_pending_writes(conf));
957 958
	spin_unlock_irq(&conf->resync_lock);
}
959
static void unfreeze_array(struct r1conf *conf)
960 961 962
{
	/* reverse the effect of the freeze */
	spin_lock_irq(&conf->resync_lock);
963
	conf->array_frozen = 0;
964 965 966 967
	wake_up(&conf->wait_barrier);
	spin_unlock_irq(&conf->resync_lock);
}

968
/* duplicate the data pages for behind I/O
969
 */
970
static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio)
971 972 973
{
	int i;
	struct bio_vec *bvec;
974
	struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec),
975
					GFP_NOIO);
976
	if (unlikely(!bvecs))
977
		return;
978

979
	bio_for_each_segment_all(bvec, bio, i) {
980 981 982
		bvecs[i] = *bvec;
		bvecs[i].bv_page = alloc_page(GFP_NOIO);
		if (unlikely(!bvecs[i].bv_page))
983
			goto do_sync_io;
984 985 986
		memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset,
		       kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
		kunmap(bvecs[i].bv_page);
987 988
		kunmap(bvec->bv_page);
	}
989
	r1_bio->behind_bvecs = bvecs;
990 991 992
	r1_bio->behind_page_count = bio->bi_vcnt;
	set_bit(R1BIO_BehindIO, &r1_bio->state);
	return;
993 994

do_sync_io:
995
	for (i = 0; i < bio->bi_vcnt; i++)
996 997 998
		if (bvecs[i].bv_page)
			put_page(bvecs[i].bv_page);
	kfree(bvecs);
999 1000
	pr_debug("%dB behind alloc failed, doing sync I/O\n",
		 bio->bi_iter.bi_size);
1001 1002
}

1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016
struct raid1_plug_cb {
	struct blk_plug_cb	cb;
	struct bio_list		pending;
	int			pending_cnt;
};

static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule)
{
	struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb,
						  cb);
	struct mddev *mddev = plug->cb.data;
	struct r1conf *conf = mddev->private;
	struct bio *bio;

1017
	if (from_schedule || current->bio_list) {
1018 1019 1020 1021
		spin_lock_irq(&conf->device_lock);
		bio_list_merge(&conf->pending_bio_list, &plug->pending);
		conf->pending_count += plug->pending_cnt;
		spin_unlock_irq(&conf->device_lock);
1022
		wake_up(&conf->wait_barrier);
1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035
		md_wakeup_thread(mddev->thread);
		kfree(plug);
		return;
	}

	/* we aren't scheduling, so we can do the write-out directly. */
	bio = bio_list_get(&plug->pending);
	bitmap_unplug(mddev->bitmap);
	wake_up(&conf->wait_barrier);

	while (bio) { /* submit pending writes */
		struct bio *next = bio->bi_next;
		bio->bi_next = NULL;
M
Mike Christie 已提交
1036
		if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
1037 1038
		    !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
			/* Just ignore it */
1039
			bio_endio(bio);
1040 1041
		else
			generic_make_request(bio);
1042 1043 1044 1045 1046
		bio = next;
	}
	kfree(plug);
}

S
Shaohua Li 已提交
1047
static void raid1_make_request(struct mddev *mddev, struct bio * bio)
L
Linus Torvalds 已提交
1048
{
1049
	struct r1conf *conf = mddev->private;
1050
	struct raid1_info *mirror;
1051
	struct r1bio *r1_bio;
L
Linus Torvalds 已提交
1052
	struct bio *read_bio;
1053
	int i, disks;
1054
	struct bitmap *bitmap;
1055
	unsigned long flags;
M
Mike Christie 已提交
1056
	const int op = bio_op(bio);
1057
	const int rw = bio_data_dir(bio);
1058
	const unsigned long do_sync = (bio->bi_rw & REQ_SYNC);
T
Tejun Heo 已提交
1059
	const unsigned long do_flush_fua = (bio->bi_rw & (REQ_FLUSH | REQ_FUA));
M
Mike Christie 已提交
1060
	const unsigned long do_sec = (bio->bi_rw & REQ_SECURE);
1061
	struct md_rdev *blocked_rdev;
1062 1063
	struct blk_plug_cb *cb;
	struct raid1_plug_cb *plug = NULL;
1064 1065 1066
	int first_clone;
	int sectors_handled;
	int max_sectors;
1067
	sector_t start_next_window;
1068

L
Linus Torvalds 已提交
1069 1070 1071 1072 1073
	/*
	 * Register the new request and wait if the reconstruction
	 * thread has put up a bar for new requests.
	 * Continue immediately if no resync is active currently.
	 */
1074

1075 1076
	md_write_start(mddev, bio); /* wait on superblock update early */

1077
	if (bio_data_dir(bio) == WRITE &&
1078 1079 1080
	    ((bio_end_sector(bio) > mddev->suspend_lo &&
	    bio->bi_iter.bi_sector < mddev->suspend_hi) ||
	    (mddev_is_clustered(mddev) &&
1081 1082
	     md_cluster_ops->area_resyncing(mddev, WRITE,
		     bio->bi_iter.bi_sector, bio_end_sector(bio))))) {
1083 1084 1085 1086 1087 1088 1089 1090 1091
		/* As the suspend_* range is controlled by
		 * userspace, we want an interruptible
		 * wait.
		 */
		DEFINE_WAIT(w);
		for (;;) {
			flush_signals(current);
			prepare_to_wait(&conf->wait_barrier,
					&w, TASK_INTERRUPTIBLE);
K
Kent Overstreet 已提交
1092
			if (bio_end_sector(bio) <= mddev->suspend_lo ||
1093 1094
			    bio->bi_iter.bi_sector >= mddev->suspend_hi ||
			    (mddev_is_clustered(mddev) &&
1095
			     !md_cluster_ops->area_resyncing(mddev, WRITE,
1096
				     bio->bi_iter.bi_sector, bio_end_sector(bio))))
1097 1098 1099 1100 1101
				break;
			schedule();
		}
		finish_wait(&conf->wait_barrier, &w);
	}
1102

1103
	start_next_window = wait_barrier(conf, bio);
L
Linus Torvalds 已提交
1104

1105 1106
	bitmap = mddev->bitmap;

L
Linus Torvalds 已提交
1107 1108 1109 1110 1111 1112 1113 1114
	/*
	 * make_request() can abort the operation when READA is being
	 * used and no empty request is available.
	 *
	 */
	r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);

	r1_bio->master_bio = bio;
1115
	r1_bio->sectors = bio_sectors(bio);
1116
	r1_bio->state = 0;
L
Linus Torvalds 已提交
1117
	r1_bio->mddev = mddev;
1118
	r1_bio->sector = bio->bi_iter.bi_sector;
L
Linus Torvalds 已提交
1119

1120 1121 1122 1123 1124 1125 1126 1127
	/* We might need to issue multiple reads to different
	 * devices if there are bad blocks around, so we keep
	 * track of the number of reads in bio->bi_phys_segments.
	 * If this is 0, there is only one r1_bio and no locking
	 * will be needed when requests complete.  If it is
	 * non-zero, then it is the number of not-completed requests.
	 */
	bio->bi_phys_segments = 0;
1128
	bio_clear_flag(bio, BIO_SEG_VALID);
1129

1130
	if (rw == READ) {
L
Linus Torvalds 已提交
1131 1132 1133
		/*
		 * read balancing logic:
		 */
1134 1135 1136 1137
		int rdisk;

read_again:
		rdisk = read_balance(conf, r1_bio, &max_sectors);
L
Linus Torvalds 已提交
1138 1139 1140 1141

		if (rdisk < 0) {
			/* couldn't find anywhere to read from */
			raid_end_bio_io(r1_bio);
1142
			return;
L
Linus Torvalds 已提交
1143 1144 1145
		}
		mirror = conf->mirrors + rdisk;

1146 1147 1148 1149 1150 1151 1152 1153 1154
		if (test_bit(WriteMostly, &mirror->rdev->flags) &&
		    bitmap) {
			/* Reading from a write-mostly device must
			 * take care not to over-take any writes
			 * that are 'behind'
			 */
			wait_event(bitmap->behind_wait,
				   atomic_read(&bitmap->behind_writes) == 0);
		}
L
Linus Torvalds 已提交
1155
		r1_bio->read_disk = rdisk;
1156
		r1_bio->start_next_window = 0;
L
Linus Torvalds 已提交
1157

1158
		read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev);
1159
		bio_trim(read_bio, r1_bio->sector - bio->bi_iter.bi_sector,
1160
			 max_sectors);
L
Linus Torvalds 已提交
1161 1162 1163

		r1_bio->bios[rdisk] = read_bio;

1164 1165
		read_bio->bi_iter.bi_sector = r1_bio->sector +
			mirror->rdev->data_offset;
L
Linus Torvalds 已提交
1166 1167
		read_bio->bi_bdev = mirror->rdev->bdev;
		read_bio->bi_end_io = raid1_end_read_request;
M
Mike Christie 已提交
1168
		bio_set_op_attrs(read_bio, op, do_sync);
L
Linus Torvalds 已提交
1169 1170
		read_bio->bi_private = r1_bio;

1171 1172 1173 1174 1175 1176
		if (max_sectors < r1_bio->sectors) {
			/* could not read all from this device, so we will
			 * need another r1_bio.
			 */

			sectors_handled = (r1_bio->sector + max_sectors
1177
					   - bio->bi_iter.bi_sector);
1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194
			r1_bio->sectors = max_sectors;
			spin_lock_irq(&conf->device_lock);
			if (bio->bi_phys_segments == 0)
				bio->bi_phys_segments = 2;
			else
				bio->bi_phys_segments++;
			spin_unlock_irq(&conf->device_lock);
			/* Cannot call generic_make_request directly
			 * as that will be queued in __make_request
			 * and subsequent mempool_alloc might block waiting
			 * for it.  So hand bio over to raid1d.
			 */
			reschedule_retry(r1_bio);

			r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);

			r1_bio->master_bio = bio;
1195
			r1_bio->sectors = bio_sectors(bio) - sectors_handled;
1196 1197
			r1_bio->state = 0;
			r1_bio->mddev = mddev;
1198 1199
			r1_bio->sector = bio->bi_iter.bi_sector +
				sectors_handled;
1200 1201 1202
			goto read_again;
		} else
			generic_make_request(read_bio);
1203
		return;
L
Linus Torvalds 已提交
1204 1205 1206 1207 1208
	}

	/*
	 * WRITE:
	 */
1209 1210 1211 1212 1213
	if (conf->pending_count >= max_queued_requests) {
		md_wakeup_thread(mddev->thread);
		wait_event(conf->wait_barrier,
			   conf->pending_count < max_queued_requests);
	}
1214
	/* first select target devices under rcu_lock and
L
Linus Torvalds 已提交
1215 1216
	 * inc refcount on their rdev.  Record them by setting
	 * bios[x] to bio
1217 1218 1219 1220 1221 1222
	 * If there are known/acknowledged bad blocks on any device on
	 * which we have seen a write error, we want to avoid writing those
	 * blocks.
	 * This potentially requires several writes to write around
	 * the bad blocks.  Each set of writes gets it's own r1bio
	 * with a set of bios attached.
L
Linus Torvalds 已提交
1223
	 */
N
NeilBrown 已提交
1224

1225
	disks = conf->raid_disks * 2;
1226
 retry_write:
1227
	r1_bio->start_next_window = start_next_window;
1228
	blocked_rdev = NULL;
L
Linus Torvalds 已提交
1229
	rcu_read_lock();
1230
	max_sectors = r1_bio->sectors;
L
Linus Torvalds 已提交
1231
	for (i = 0;  i < disks; i++) {
1232
		struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
1233 1234 1235 1236 1237
		if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
			atomic_inc(&rdev->nr_pending);
			blocked_rdev = rdev;
			break;
		}
1238
		r1_bio->bios[i] = NULL;
1239
		if (!rdev || test_bit(Faulty, &rdev->flags)) {
1240 1241
			if (i < conf->raid_disks)
				set_bit(R1BIO_Degraded, &r1_bio->state);
1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268
			continue;
		}

		atomic_inc(&rdev->nr_pending);
		if (test_bit(WriteErrorSeen, &rdev->flags)) {
			sector_t first_bad;
			int bad_sectors;
			int is_bad;

			is_bad = is_badblock(rdev, r1_bio->sector,
					     max_sectors,
					     &first_bad, &bad_sectors);
			if (is_bad < 0) {
				/* mustn't write here until the bad block is
				 * acknowledged*/
				set_bit(BlockedBadBlocks, &rdev->flags);
				blocked_rdev = rdev;
				break;
			}
			if (is_bad && first_bad <= r1_bio->sector) {
				/* Cannot write here at all */
				bad_sectors -= (r1_bio->sector - first_bad);
				if (bad_sectors < max_sectors)
					/* mustn't write more than bad_sectors
					 * to other devices yet
					 */
					max_sectors = bad_sectors;
1269
				rdev_dec_pending(rdev, mddev);
1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280
				/* We don't set R1BIO_Degraded as that
				 * only applies if the disk is
				 * missing, so it might be re-added,
				 * and we want to know to recover this
				 * chunk.
				 * In this case the device is here,
				 * and the fact that this chunk is not
				 * in-sync is recorded in the bad
				 * block log
				 */
				continue;
1281
			}
1282 1283 1284 1285 1286 1287 1288
			if (is_bad) {
				int good_sectors = first_bad - r1_bio->sector;
				if (good_sectors < max_sectors)
					max_sectors = good_sectors;
			}
		}
		r1_bio->bios[i] = bio;
L
Linus Torvalds 已提交
1289 1290 1291
	}
	rcu_read_unlock();

1292 1293 1294
	if (unlikely(blocked_rdev)) {
		/* Wait for this device to become unblocked */
		int j;
1295
		sector_t old = start_next_window;
1296 1297 1298 1299

		for (j = 0; j < i; j++)
			if (r1_bio->bios[j])
				rdev_dec_pending(conf->mirrors[j].rdev, mddev);
1300
		r1_bio->state = 0;
1301
		allow_barrier(conf, start_next_window, bio->bi_iter.bi_sector);
1302
		md_wait_for_blocked_rdev(blocked_rdev, mddev);
1303 1304 1305 1306 1307 1308 1309 1310 1311 1312
		start_next_window = wait_barrier(conf, bio);
		/*
		 * We must make sure the multi r1bios of bio have
		 * the same value of bi_phys_segments
		 */
		if (bio->bi_phys_segments && old &&
		    old != start_next_window)
			/* Wait for the former r1bio(s) to complete */
			wait_event(conf->wait_barrier,
				   bio->bi_phys_segments == 1);
1313 1314 1315
		goto retry_write;
	}

1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326
	if (max_sectors < r1_bio->sectors) {
		/* We are splitting this write into multiple parts, so
		 * we need to prepare for allocating another r1_bio.
		 */
		r1_bio->sectors = max_sectors;
		spin_lock_irq(&conf->device_lock);
		if (bio->bi_phys_segments == 0)
			bio->bi_phys_segments = 2;
		else
			bio->bi_phys_segments++;
		spin_unlock_irq(&conf->device_lock);
1327
	}
1328
	sectors_handled = r1_bio->sector + max_sectors - bio->bi_iter.bi_sector;
1329

1330
	atomic_set(&r1_bio->remaining, 1);
1331
	atomic_set(&r1_bio->behind_remaining, 0);
1332

1333
	first_clone = 1;
L
Linus Torvalds 已提交
1334 1335 1336 1337 1338
	for (i = 0; i < disks; i++) {
		struct bio *mbio;
		if (!r1_bio->bios[i])
			continue;

1339
		mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
1340
		bio_trim(mbio, r1_bio->sector - bio->bi_iter.bi_sector, max_sectors);
1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358

		if (first_clone) {
			/* do behind I/O ?
			 * Not if there are too many, or cannot
			 * allocate memory, or a reader on WriteMostly
			 * is waiting for behind writes to flush */
			if (bitmap &&
			    (atomic_read(&bitmap->behind_writes)
			     < mddev->bitmap_info.max_write_behind) &&
			    !waitqueue_active(&bitmap->behind_wait))
				alloc_behind_pages(mbio, r1_bio);

			bitmap_startwrite(bitmap, r1_bio->sector,
					  r1_bio->sectors,
					  test_bit(R1BIO_BehindIO,
						   &r1_bio->state));
			first_clone = 0;
		}
1359
		if (r1_bio->behind_bvecs) {
1360 1361 1362
			struct bio_vec *bvec;
			int j;

1363 1364
			/*
			 * We trimmed the bio, so _all is legit
1365
			 */
1366
			bio_for_each_segment_all(bvec, mbio, j)
1367
				bvec->bv_page = r1_bio->behind_bvecs[j].bv_page;
1368 1369 1370 1371
			if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
				atomic_inc(&r1_bio->behind_remaining);
		}

1372 1373
		r1_bio->bios[i] = mbio;

1374
		mbio->bi_iter.bi_sector	= (r1_bio->sector +
1375 1376 1377
				   conf->mirrors[i].rdev->data_offset);
		mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
		mbio->bi_end_io	= raid1_end_write_request;
M
Mike Christie 已提交
1378
		bio_set_op_attrs(mbio, op, do_flush_fua | do_sync | do_sec);
1379 1380
		mbio->bi_private = r1_bio;

L
Linus Torvalds 已提交
1381
		atomic_inc(&r1_bio->remaining);
1382 1383 1384 1385 1386 1387

		cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug));
		if (cb)
			plug = container_of(cb, struct raid1_plug_cb, cb);
		else
			plug = NULL;
1388
		spin_lock_irqsave(&conf->device_lock, flags);
1389 1390 1391 1392 1393 1394 1395
		if (plug) {
			bio_list_add(&plug->pending, mbio);
			plug->pending_cnt++;
		} else {
			bio_list_add(&conf->pending_bio_list, mbio);
			conf->pending_count++;
		}
1396
		spin_unlock_irqrestore(&conf->device_lock, flags);
1397
		if (!plug)
N
NeilBrown 已提交
1398
			md_wakeup_thread(mddev->thread);
L
Linus Torvalds 已提交
1399
	}
1400 1401 1402
	/* Mustn't call r1_bio_write_done before this next test,
	 * as it could result in the bio being freed.
	 */
1403
	if (sectors_handled < bio_sectors(bio)) {
1404
		r1_bio_write_done(r1_bio);
1405 1406 1407 1408 1409
		/* We need another r1_bio.  It has already been counted
		 * in bio->bi_phys_segments
		 */
		r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
		r1_bio->master_bio = bio;
1410
		r1_bio->sectors = bio_sectors(bio) - sectors_handled;
1411 1412
		r1_bio->state = 0;
		r1_bio->mddev = mddev;
1413
		r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled;
1414 1415 1416
		goto retry_write;
	}

1417 1418 1419 1420
	r1_bio_write_done(r1_bio);

	/* In case raid1d snuck in to freeze_array */
	wake_up(&conf->wait_barrier);
L
Linus Torvalds 已提交
1421 1422
}

S
Shaohua Li 已提交
1423
static void raid1_status(struct seq_file *seq, struct mddev *mddev)
L
Linus Torvalds 已提交
1424
{
1425
	struct r1conf *conf = mddev->private;
L
Linus Torvalds 已提交
1426 1427 1428
	int i;

	seq_printf(seq, " [%d/%d] [", conf->raid_disks,
1429
		   conf->raid_disks - mddev->degraded);
1430 1431
	rcu_read_lock();
	for (i = 0; i < conf->raid_disks; i++) {
1432
		struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
L
Linus Torvalds 已提交
1433
		seq_printf(seq, "%s",
1434 1435 1436
			   rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
	}
	rcu_read_unlock();
L
Linus Torvalds 已提交
1437 1438 1439
	seq_printf(seq, "]");
}

S
Shaohua Li 已提交
1440
static void raid1_error(struct mddev *mddev, struct md_rdev *rdev)
L
Linus Torvalds 已提交
1441 1442
{
	char b[BDEVNAME_SIZE];
1443
	struct r1conf *conf = mddev->private;
1444
	unsigned long flags;
L
Linus Torvalds 已提交
1445 1446 1447 1448 1449 1450 1451

	/*
	 * If it is not operational, then we have already marked it as dead
	 * else if it is the last working disks, ignore the error, let the
	 * next level up know.
	 * else mark the drive as failed
	 */
1452
	if (test_bit(In_sync, &rdev->flags)
1453
	    && (conf->raid_disks - mddev->degraded) == 1) {
L
Linus Torvalds 已提交
1454 1455
		/*
		 * Don't fail the drive, act as though we were just a
1456 1457 1458
		 * normal single drive.
		 * However don't try a recovery from this drive as
		 * it is very likely to fail.
L
Linus Torvalds 已提交
1459
		 */
1460
		conf->recovery_disabled = mddev->recovery_disabled;
L
Linus Torvalds 已提交
1461
		return;
1462
	}
1463
	set_bit(Blocked, &rdev->flags);
1464
	spin_lock_irqsave(&conf->device_lock, flags);
1465
	if (test_and_clear_bit(In_sync, &rdev->flags)) {
L
Linus Torvalds 已提交
1466
		mddev->degraded++;
1467 1468 1469
		set_bit(Faulty, &rdev->flags);
	} else
		set_bit(Faulty, &rdev->flags);
1470
	spin_unlock_irqrestore(&conf->device_lock, flags);
1471 1472 1473 1474
	/*
	 * if recovery is running, make sure it aborts.
	 */
	set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1475 1476
	set_mask_bits(&mddev->flags, 0,
		      BIT(MD_CHANGE_DEVS) | BIT(MD_CHANGE_PENDING));
1477 1478 1479
	printk(KERN_ALERT
	       "md/raid1:%s: Disk failure on %s, disabling device.\n"
	       "md/raid1:%s: Operation continuing on %d devices.\n",
N
NeilBrown 已提交
1480 1481
	       mdname(mddev), bdevname(rdev->bdev, b),
	       mdname(mddev), conf->raid_disks - mddev->degraded);
L
Linus Torvalds 已提交
1482 1483
}

1484
static void print_conf(struct r1conf *conf)
L
Linus Torvalds 已提交
1485 1486 1487
{
	int i;

N
NeilBrown 已提交
1488
	printk(KERN_DEBUG "RAID1 conf printout:\n");
L
Linus Torvalds 已提交
1489
	if (!conf) {
N
NeilBrown 已提交
1490
		printk(KERN_DEBUG "(!conf)\n");
L
Linus Torvalds 已提交
1491 1492
		return;
	}
N
NeilBrown 已提交
1493
	printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
L
Linus Torvalds 已提交
1494 1495
		conf->raid_disks);

1496
	rcu_read_lock();
L
Linus Torvalds 已提交
1497 1498
	for (i = 0; i < conf->raid_disks; i++) {
		char b[BDEVNAME_SIZE];
1499
		struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
1500
		if (rdev)
N
NeilBrown 已提交
1501
			printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n",
1502 1503 1504
			       i, !test_bit(In_sync, &rdev->flags),
			       !test_bit(Faulty, &rdev->flags),
			       bdevname(rdev->bdev,b));
L
Linus Torvalds 已提交
1505
	}
1506
	rcu_read_unlock();
L
Linus Torvalds 已提交
1507 1508
}

1509
static void close_sync(struct r1conf *conf)
L
Linus Torvalds 已提交
1510
{
1511 1512
	wait_barrier(conf, NULL);
	allow_barrier(conf, 0, 0);
L
Linus Torvalds 已提交
1513 1514 1515

	mempool_destroy(conf->r1buf_pool);
	conf->r1buf_pool = NULL;
1516

1517
	spin_lock_irq(&conf->resync_lock);
1518
	conf->next_resync = MaxSector - 2 * NEXT_NORMALIO_DISTANCE;
1519
	conf->start_next_window = MaxSector;
1520 1521 1522 1523
	conf->current_window_requests +=
		conf->next_window_requests;
	conf->next_window_requests = 0;
	spin_unlock_irq(&conf->resync_lock);
L
Linus Torvalds 已提交
1524 1525
}

1526
static int raid1_spare_active(struct mddev *mddev)
L
Linus Torvalds 已提交
1527 1528
{
	int i;
1529
	struct r1conf *conf = mddev->private;
1530 1531
	int count = 0;
	unsigned long flags;
L
Linus Torvalds 已提交
1532 1533

	/*
1534
	 * Find all failed disks within the RAID1 configuration
1535 1536
	 * and mark them readable.
	 * Called under mddev lock, so rcu protection not needed.
1537 1538
	 * device_lock used to avoid races with raid1_end_read_request
	 * which expects 'In_sync' flags and ->degraded to be consistent.
L
Linus Torvalds 已提交
1539
	 */
1540
	spin_lock_irqsave(&conf->device_lock, flags);
L
Linus Torvalds 已提交
1541
	for (i = 0; i < conf->raid_disks; i++) {
1542
		struct md_rdev *rdev = conf->mirrors[i].rdev;
1543 1544
		struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev;
		if (repl
1545
		    && !test_bit(Candidate, &repl->flags)
1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562
		    && repl->recovery_offset == MaxSector
		    && !test_bit(Faulty, &repl->flags)
		    && !test_and_set_bit(In_sync, &repl->flags)) {
			/* replacement has just become active */
			if (!rdev ||
			    !test_and_clear_bit(In_sync, &rdev->flags))
				count++;
			if (rdev) {
				/* Replaced device not technically
				 * faulty, but we need to be sure
				 * it gets removed and never re-added
				 */
				set_bit(Faulty, &rdev->flags);
				sysfs_notify_dirent_safe(
					rdev->sysfs_state);
			}
		}
1563
		if (rdev
1564
		    && rdev->recovery_offset == MaxSector
1565
		    && !test_bit(Faulty, &rdev->flags)
1566
		    && !test_and_set_bit(In_sync, &rdev->flags)) {
1567
			count++;
1568
			sysfs_notify_dirent_safe(rdev->sysfs_state);
L
Linus Torvalds 已提交
1569 1570
		}
	}
1571 1572
	mddev->degraded -= count;
	spin_unlock_irqrestore(&conf->device_lock, flags);
L
Linus Torvalds 已提交
1573 1574

	print_conf(conf);
1575
	return count;
L
Linus Torvalds 已提交
1576 1577
}

1578
static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev)
L
Linus Torvalds 已提交
1579
{
1580
	struct r1conf *conf = mddev->private;
1581
	int err = -EEXIST;
1582
	int mirror = 0;
1583
	struct raid1_info *p;
1584
	int first = 0;
1585
	int last = conf->raid_disks - 1;
L
Linus Torvalds 已提交
1586

1587 1588 1589
	if (mddev->recovery_disabled == conf->recovery_disabled)
		return -EBUSY;

1590 1591 1592
	if (md_integrity_add_rdev(rdev, mddev))
		return -ENXIO;

1593 1594 1595
	if (rdev->raid_disk >= 0)
		first = last = rdev->raid_disk;

1596 1597 1598 1599 1600 1601 1602 1603 1604
	/*
	 * find the disk ... but prefer rdev->saved_raid_disk
	 * if possible.
	 */
	if (rdev->saved_raid_disk >= 0 &&
	    rdev->saved_raid_disk >= first &&
	    conf->mirrors[rdev->saved_raid_disk].rdev == NULL)
		first = last = rdev->saved_raid_disk;

1605 1606 1607
	for (mirror = first; mirror <= last; mirror++) {
		p = conf->mirrors+mirror;
		if (!p->rdev) {
L
Linus Torvalds 已提交
1608

1609 1610 1611
			if (mddev->gendisk)
				disk_stack_limits(mddev->gendisk, rdev->bdev,
						  rdev->data_offset << 9);
L
Linus Torvalds 已提交
1612 1613 1614

			p->head_position = 0;
			rdev->raid_disk = mirror;
1615
			err = 0;
1616 1617 1618 1619
			/* As all devices are equivalent, we don't need a full recovery
			 * if this was recently any drive of the array
			 */
			if (rdev->saved_raid_disk < 0)
1620
				conf->fullsync = 1;
1621
			rcu_assign_pointer(p->rdev, rdev);
L
Linus Torvalds 已提交
1622 1623
			break;
		}
1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635
		if (test_bit(WantReplacement, &p->rdev->flags) &&
		    p[conf->raid_disks].rdev == NULL) {
			/* Add this device as a replacement */
			clear_bit(In_sync, &rdev->flags);
			set_bit(Replacement, &rdev->flags);
			rdev->raid_disk = mirror;
			err = 0;
			conf->fullsync = 1;
			rcu_assign_pointer(p[conf->raid_disks].rdev, rdev);
			break;
		}
	}
1636
	if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev)))
S
Shaohua Li 已提交
1637
		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
L
Linus Torvalds 已提交
1638
	print_conf(conf);
1639
	return err;
L
Linus Torvalds 已提交
1640 1641
}

1642
static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
L
Linus Torvalds 已提交
1643
{
1644
	struct r1conf *conf = mddev->private;
L
Linus Torvalds 已提交
1645
	int err = 0;
1646
	int number = rdev->raid_disk;
1647
	struct raid1_info *p = conf->mirrors + number;
L
Linus Torvalds 已提交
1648

1649 1650 1651
	if (rdev != p->rdev)
		p = conf->mirrors + conf->raid_disks + number;

L
Linus Torvalds 已提交
1652
	print_conf(conf);
1653
	if (rdev == p->rdev) {
1654
		if (test_bit(In_sync, &rdev->flags) ||
L
Linus Torvalds 已提交
1655 1656 1657 1658
		    atomic_read(&rdev->nr_pending)) {
			err = -EBUSY;
			goto abort;
		}
N
NeilBrown 已提交
1659
		/* Only remove non-faulty devices if recovery
1660 1661 1662
		 * is not possible.
		 */
		if (!test_bit(Faulty, &rdev->flags) &&
1663
		    mddev->recovery_disabled != conf->recovery_disabled &&
1664 1665 1666 1667
		    mddev->degraded < conf->raid_disks) {
			err = -EBUSY;
			goto abort;
		}
L
Linus Torvalds 已提交
1668
		p->rdev = NULL;
1669
		synchronize_rcu();
L
Linus Torvalds 已提交
1670 1671 1672 1673
		if (atomic_read(&rdev->nr_pending)) {
			/* lost the race, try later */
			err = -EBUSY;
			p->rdev = rdev;
1674
			goto abort;
1675 1676 1677 1678 1679 1680 1681
		} else if (conf->mirrors[conf->raid_disks + number].rdev) {
			/* We just removed a device that is being replaced.
			 * Move down the replacement.  We drain all IO before
			 * doing this to avoid confusion.
			 */
			struct md_rdev *repl =
				conf->mirrors[conf->raid_disks + number].rdev;
1682
			freeze_array(conf, 0);
1683 1684 1685
			clear_bit(Replacement, &repl->flags);
			p->rdev = repl;
			conf->mirrors[conf->raid_disks + number].rdev = NULL;
1686
			unfreeze_array(conf);
1687 1688
			clear_bit(WantReplacement, &rdev->flags);
		} else
1689
			clear_bit(WantReplacement, &rdev->flags);
1690
		err = md_integrity_register(mddev);
L
Linus Torvalds 已提交
1691 1692 1693 1694 1695 1696 1697
	}
abort:

	print_conf(conf);
	return err;
}

1698
static void end_sync_read(struct bio *bio)
L
Linus Torvalds 已提交
1699
{
1700
	struct r1bio *r1_bio = bio->bi_private;
L
Linus Torvalds 已提交
1701

1702
	update_head_pos(r1_bio->read_disk, r1_bio);
1703

L
Linus Torvalds 已提交
1704 1705 1706 1707 1708
	/*
	 * we have read a block, now it needs to be re-written,
	 * or re-read if the read failed.
	 * We don't do much here, just schedule handling by raid1d
	 */
1709
	if (!bio->bi_error)
L
Linus Torvalds 已提交
1710
		set_bit(R1BIO_Uptodate, &r1_bio->state);
1711 1712 1713

	if (atomic_dec_and_test(&r1_bio->remaining))
		reschedule_retry(r1_bio);
L
Linus Torvalds 已提交
1714 1715
}

1716
static void end_sync_write(struct bio *bio)
L
Linus Torvalds 已提交
1717
{
1718
	int uptodate = !bio->bi_error;
1719
	struct r1bio *r1_bio = bio->bi_private;
1720
	struct mddev *mddev = r1_bio->mddev;
1721
	struct r1conf *conf = mddev->private;
L
Linus Torvalds 已提交
1722
	int mirror=0;
1723 1724
	sector_t first_bad;
	int bad_sectors;
L
Linus Torvalds 已提交
1725

1726 1727
	mirror = find_bio_disk(r1_bio, bio);

1728
	if (!uptodate) {
N
NeilBrown 已提交
1729
		sector_t sync_blocks = 0;
1730 1731 1732 1733
		sector_t s = r1_bio->sector;
		long sectors_to_go = r1_bio->sectors;
		/* make sure these bits doesn't get cleared. */
		do {
1734
			bitmap_end_sync(mddev->bitmap, s,
1735 1736 1737 1738
					&sync_blocks, 1);
			s += sync_blocks;
			sectors_to_go -= sync_blocks;
		} while (sectors_to_go > 0);
1739 1740
		set_bit(WriteErrorSeen,
			&conf->mirrors[mirror].rdev->flags);
1741 1742 1743 1744
		if (!test_and_set_bit(WantReplacement,
				      &conf->mirrors[mirror].rdev->flags))
			set_bit(MD_RECOVERY_NEEDED, &
				mddev->recovery);
1745
		set_bit(R1BIO_WriteError, &r1_bio->state);
1746 1747 1748
	} else if (is_badblock(conf->mirrors[mirror].rdev,
			       r1_bio->sector,
			       r1_bio->sectors,
1749 1750 1751 1752 1753 1754
			       &first_bad, &bad_sectors) &&
		   !is_badblock(conf->mirrors[r1_bio->read_disk].rdev,
				r1_bio->sector,
				r1_bio->sectors,
				&first_bad, &bad_sectors)
		)
1755
		set_bit(R1BIO_MadeGood, &r1_bio->state);
1756

L
Linus Torvalds 已提交
1757
	if (atomic_dec_and_test(&r1_bio->remaining)) {
1758
		int s = r1_bio->sectors;
1759 1760
		if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
		    test_bit(R1BIO_WriteError, &r1_bio->state))
1761 1762 1763 1764 1765
			reschedule_retry(r1_bio);
		else {
			put_buf(r1_bio);
			md_done_sync(mddev, s, uptodate);
		}
L
Linus Torvalds 已提交
1766 1767 1768
	}
}

1769
static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector,
1770 1771
			    int sectors, struct page *page, int rw)
{
M
Mike Christie 已提交
1772
	if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false))
1773 1774
		/* success */
		return 1;
1775
	if (rw == WRITE) {
1776
		set_bit(WriteErrorSeen, &rdev->flags);
1777 1778 1779 1780 1781
		if (!test_and_set_bit(WantReplacement,
				      &rdev->flags))
			set_bit(MD_RECOVERY_NEEDED, &
				rdev->mddev->recovery);
	}
1782 1783 1784 1785 1786 1787
	/* need to record an error - either for the block or the device */
	if (!rdev_set_badblocks(rdev, sector, sectors, 0))
		md_error(rdev->mddev, rdev);
	return 0;
}

1788
static int fix_sync_read_error(struct r1bio *r1_bio)
L
Linus Torvalds 已提交
1789
{
1790 1791 1792 1793 1794 1795 1796
	/* Try some synchronous reads of other devices to get
	 * good data, much like with normal read errors.  Only
	 * read into the pages we already have so we don't
	 * need to re-issue the read request.
	 * We don't need to freeze the array, because being in an
	 * active sync request, there is no normal IO, and
	 * no overlapping syncs.
1797 1798 1799
	 * We don't need to check is_badblock() again as we
	 * made sure that anything with a bad block in range
	 * will have bi_end_io clear.
1800
	 */
1801
	struct mddev *mddev = r1_bio->mddev;
1802
	struct r1conf *conf = mddev->private;
1803 1804 1805 1806 1807 1808 1809 1810 1811
	struct bio *bio = r1_bio->bios[r1_bio->read_disk];
	sector_t sect = r1_bio->sector;
	int sectors = r1_bio->sectors;
	int idx = 0;

	while(sectors) {
		int s = sectors;
		int d = r1_bio->read_disk;
		int success = 0;
1812
		struct md_rdev *rdev;
1813
		int start;
1814 1815 1816 1817 1818 1819 1820 1821 1822 1823

		if (s > (PAGE_SIZE>>9))
			s = PAGE_SIZE >> 9;
		do {
			if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
				/* No rcu protection needed here devices
				 * can only be removed when no resync is
				 * active, and resync is currently active
				 */
				rdev = conf->mirrors[d].rdev;
1824
				if (sync_page_io(rdev, sect, s<<9,
1825
						 bio->bi_io_vec[idx].bv_page,
M
Mike Christie 已提交
1826
						 REQ_OP_READ, 0, false)) {
1827 1828 1829 1830 1831
					success = 1;
					break;
				}
			}
			d++;
1832
			if (d == conf->raid_disks * 2)
1833 1834 1835
				d = 0;
		} while (!success && d != r1_bio->read_disk);

1836
		if (!success) {
1837
			char b[BDEVNAME_SIZE];
1838 1839 1840 1841 1842 1843
			int abort = 0;
			/* Cannot read from anywhere, this block is lost.
			 * Record a bad block on each device.  If that doesn't
			 * work just disable and interrupt the recovery.
			 * Don't fail devices as that won't really help.
			 */
1844 1845 1846 1847 1848
			printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error"
			       " for block %llu\n",
			       mdname(mddev),
			       bdevname(bio->bi_bdev, b),
			       (unsigned long long)r1_bio->sector);
1849
			for (d = 0; d < conf->raid_disks * 2; d++) {
1850 1851 1852 1853 1854 1855 1856
				rdev = conf->mirrors[d].rdev;
				if (!rdev || test_bit(Faulty, &rdev->flags))
					continue;
				if (!rdev_set_badblocks(rdev, sect, s, 0))
					abort = 1;
			}
			if (abort) {
1857 1858
				conf->recovery_disabled =
					mddev->recovery_disabled;
1859 1860 1861 1862 1863 1864 1865 1866 1867 1868
				set_bit(MD_RECOVERY_INTR, &mddev->recovery);
				md_done_sync(mddev, r1_bio->sectors, 0);
				put_buf(r1_bio);
				return 0;
			}
			/* Try next page */
			sectors -= s;
			sect += s;
			idx++;
			continue;
1869
		}
1870 1871 1872 1873 1874

		start = d;
		/* write it back and re-read */
		while (d != r1_bio->read_disk) {
			if (d == 0)
1875
				d = conf->raid_disks * 2;
1876 1877 1878 1879
			d--;
			if (r1_bio->bios[d]->bi_end_io != end_sync_read)
				continue;
			rdev = conf->mirrors[d].rdev;
1880 1881 1882
			if (r1_sync_page_io(rdev, sect, s,
					    bio->bi_io_vec[idx].bv_page,
					    WRITE) == 0) {
1883 1884
				r1_bio->bios[d]->bi_end_io = NULL;
				rdev_dec_pending(rdev, mddev);
1885
			}
1886 1887 1888 1889
		}
		d = start;
		while (d != r1_bio->read_disk) {
			if (d == 0)
1890
				d = conf->raid_disks * 2;
1891 1892 1893 1894
			d--;
			if (r1_bio->bios[d]->bi_end_io != end_sync_read)
				continue;
			rdev = conf->mirrors[d].rdev;
1895 1896 1897
			if (r1_sync_page_io(rdev, sect, s,
					    bio->bi_io_vec[idx].bv_page,
					    READ) != 0)
1898
				atomic_add(s, &rdev->corrected_errors);
1899
		}
1900 1901 1902 1903
		sectors -= s;
		sect += s;
		idx ++;
	}
1904
	set_bit(R1BIO_Uptodate, &r1_bio->state);
1905
	bio->bi_error = 0;
1906 1907 1908
	return 1;
}

1909
static void process_checks(struct r1bio *r1_bio)
1910 1911 1912 1913 1914 1915 1916 1917
{
	/* We have read all readable devices.  If we haven't
	 * got the block, then there is no hope left.
	 * If we have, then we want to do a comparison
	 * and skip the write if everything is the same.
	 * If any blocks failed to read, then we need to
	 * attempt an over-write
	 */
1918
	struct mddev *mddev = r1_bio->mddev;
1919
	struct r1conf *conf = mddev->private;
1920 1921
	int primary;
	int i;
1922
	int vcnt;
1923

1924 1925 1926 1927 1928
	/* Fix variable parts of all bios */
	vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9);
	for (i = 0; i < conf->raid_disks * 2; i++) {
		int j;
		int size;
1929
		int error;
1930 1931 1932
		struct bio *b = r1_bio->bios[i];
		if (b->bi_end_io != end_sync_read)
			continue;
1933 1934
		/* fixup the bio for reuse, but preserve errno */
		error = b->bi_error;
1935
		bio_reset(b);
1936
		b->bi_error = error;
1937
		b->bi_vcnt = vcnt;
1938 1939
		b->bi_iter.bi_size = r1_bio->sectors << 9;
		b->bi_iter.bi_sector = r1_bio->sector +
1940 1941 1942 1943 1944
			conf->mirrors[i].rdev->data_offset;
		b->bi_bdev = conf->mirrors[i].rdev->bdev;
		b->bi_end_io = end_sync_read;
		b->bi_private = r1_bio;

1945
		size = b->bi_iter.bi_size;
1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956
		for (j = 0; j < vcnt ; j++) {
			struct bio_vec *bi;
			bi = &b->bi_io_vec[j];
			bi->bv_offset = 0;
			if (size > PAGE_SIZE)
				bi->bv_len = PAGE_SIZE;
			else
				bi->bv_len = size;
			size -= PAGE_SIZE;
		}
	}
1957
	for (primary = 0; primary < conf->raid_disks * 2; primary++)
1958
		if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
1959
		    !r1_bio->bios[primary]->bi_error) {
1960 1961 1962 1963 1964
			r1_bio->bios[primary]->bi_end_io = NULL;
			rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
			break;
		}
	r1_bio->read_disk = primary;
1965
	for (i = 0; i < conf->raid_disks * 2; i++) {
1966 1967 1968
		int j;
		struct bio *pbio = r1_bio->bios[primary];
		struct bio *sbio = r1_bio->bios[i];
1969
		int error = sbio->bi_error;
1970

K
Kent Overstreet 已提交
1971
		if (sbio->bi_end_io != end_sync_read)
1972
			continue;
1973 1974
		/* Now we can 'fixup' the error value */
		sbio->bi_error = 0;
1975

1976
		if (!error) {
1977 1978 1979 1980 1981 1982
			for (j = vcnt; j-- ; ) {
				struct page *p, *s;
				p = pbio->bi_io_vec[j].bv_page;
				s = sbio->bi_io_vec[j].bv_page;
				if (memcmp(page_address(p),
					   page_address(s),
1983
					   sbio->bi_io_vec[j].bv_len))
1984
					break;
1985
			}
1986 1987 1988
		} else
			j = 0;
		if (j >= 0)
1989
			atomic64_add(r1_bio->sectors, &mddev->resync_mismatches);
1990
		if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)
1991
			      && !error)) {
1992 1993 1994 1995 1996
			/* No need to write to this device. */
			sbio->bi_end_io = NULL;
			rdev_dec_pending(conf->mirrors[i].rdev, mddev);
			continue;
		}
K
Kent Overstreet 已提交
1997 1998

		bio_copy_data(sbio, pbio);
1999
	}
2000 2001
}

2002
static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio)
2003
{
2004
	struct r1conf *conf = mddev->private;
2005
	int i;
2006
	int disks = conf->raid_disks * 2;
2007 2008 2009 2010 2011 2012 2013 2014
	struct bio *bio, *wbio;

	bio = r1_bio->bios[r1_bio->read_disk];

	if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
		/* ouch - failed to read all of that. */
		if (!fix_sync_read_error(r1_bio))
			return;
2015 2016

	if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2017 2018
		process_checks(r1_bio);

2019 2020 2021
	/*
	 * schedule writes
	 */
L
Linus Torvalds 已提交
2022 2023 2024
	atomic_set(&r1_bio->remaining, 1);
	for (i = 0; i < disks ; i++) {
		wbio = r1_bio->bios[i];
2025 2026 2027 2028
		if (wbio->bi_end_io == NULL ||
		    (wbio->bi_end_io == end_sync_read &&
		     (i == r1_bio->read_disk ||
		      !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
L
Linus Torvalds 已提交
2029 2030
			continue;

M
Mike Christie 已提交
2031
		bio_set_op_attrs(wbio, REQ_OP_WRITE, 0);
2032
		wbio->bi_end_io = end_sync_write;
L
Linus Torvalds 已提交
2033
		atomic_inc(&r1_bio->remaining);
2034
		md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio));
2035

L
Linus Torvalds 已提交
2036 2037 2038 2039
		generic_make_request(wbio);
	}

	if (atomic_dec_and_test(&r1_bio->remaining)) {
2040
		/* if we're here, all write(s) have completed, so clean up */
2041 2042 2043 2044 2045 2046 2047 2048
		int s = r1_bio->sectors;
		if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
		    test_bit(R1BIO_WriteError, &r1_bio->state))
			reschedule_retry(r1_bio);
		else {
			put_buf(r1_bio);
			md_done_sync(mddev, s, 1);
		}
L
Linus Torvalds 已提交
2049 2050 2051 2052 2053 2054 2055 2056
	}
}

/*
 * This is a kernel thread which:
 *
 *	1.	Retries failed read operations on working mirrors.
 *	2.	Updates the raid superblock when problems encounter.
2057
 *	3.	Performs writes following reads for array synchronising.
L
Linus Torvalds 已提交
2058 2059
 */

2060
static void fix_read_error(struct r1conf *conf, int read_disk,
2061 2062
			   sector_t sect, int sectors)
{
2063
	struct mddev *mddev = conf->mddev;
2064 2065 2066 2067 2068
	while(sectors) {
		int s = sectors;
		int d = read_disk;
		int success = 0;
		int start;
2069
		struct md_rdev *rdev;
2070 2071 2072 2073 2074 2075 2076 2077 2078 2079

		if (s > (PAGE_SIZE>>9))
			s = PAGE_SIZE >> 9;

		do {
			/* Note: no rcu protection needed here
			 * as this is synchronous in the raid1d thread
			 * which is the thread that might remove
			 * a device.  If raid1d ever becomes multi-threaded....
			 */
2080 2081 2082
			sector_t first_bad;
			int bad_sectors;

2083 2084
			rdev = conf->mirrors[d].rdev;
			if (rdev &&
2085 2086 2087
			    (test_bit(In_sync, &rdev->flags) ||
			     (!test_bit(Faulty, &rdev->flags) &&
			      rdev->recovery_offset >= sect + s)) &&
2088 2089
			    is_badblock(rdev, sect, s,
					&first_bad, &bad_sectors) == 0 &&
J
Jonathan Brassow 已提交
2090
			    sync_page_io(rdev, sect, s<<9,
M
Mike Christie 已提交
2091
					 conf->tmppage, REQ_OP_READ, 0, false))
2092 2093 2094
				success = 1;
			else {
				d++;
2095
				if (d == conf->raid_disks * 2)
2096 2097 2098 2099 2100
					d = 0;
			}
		} while (!success && d != read_disk);

		if (!success) {
2101
			/* Cannot read from anywhere - mark it bad */
2102
			struct md_rdev *rdev = conf->mirrors[read_disk].rdev;
2103 2104
			if (!rdev_set_badblocks(rdev, sect, s, 0))
				md_error(mddev, rdev);
2105 2106 2107 2108 2109 2110
			break;
		}
		/* write it back and re-read */
		start = d;
		while (d != read_disk) {
			if (d==0)
2111
				d = conf->raid_disks * 2;
2112 2113 2114
			d--;
			rdev = conf->mirrors[d].rdev;
			if (rdev &&
2115
			    !test_bit(Faulty, &rdev->flags))
2116 2117
				r1_sync_page_io(rdev, sect, s,
						conf->tmppage, WRITE);
2118 2119 2120 2121 2122
		}
		d = start;
		while (d != read_disk) {
			char b[BDEVNAME_SIZE];
			if (d==0)
2123
				d = conf->raid_disks * 2;
2124 2125 2126
			d--;
			rdev = conf->mirrors[d].rdev;
			if (rdev &&
2127
			    !test_bit(Faulty, &rdev->flags)) {
2128 2129
				if (r1_sync_page_io(rdev, sect, s,
						    conf->tmppage, READ)) {
2130 2131
					atomic_add(s, &rdev->corrected_errors);
					printk(KERN_INFO
N
NeilBrown 已提交
2132
					       "md/raid1:%s: read error corrected "
2133 2134
					       "(%d sectors at %llu on %s)\n",
					       mdname(mddev), s,
2135 2136
					       (unsigned long long)(sect +
					           rdev->data_offset),
2137 2138 2139 2140 2141 2142 2143 2144 2145
					       bdevname(rdev->bdev, b));
				}
			}
		}
		sectors -= s;
		sect += s;
	}
}

2146
static int narrow_write_error(struct r1bio *r1_bio, int i)
2147
{
2148
	struct mddev *mddev = r1_bio->mddev;
2149
	struct r1conf *conf = mddev->private;
2150
	struct md_rdev *rdev = conf->mirrors[i].rdev;
2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171

	/* bio has the data to be written to device 'i' where
	 * we just recently had a write error.
	 * We repeatedly clone the bio and trim down to one block,
	 * then try the write.  Where the write fails we record
	 * a bad block.
	 * It is conceivable that the bio doesn't exactly align with
	 * blocks.  We must handle this somehow.
	 *
	 * We currently own a reference on the rdev.
	 */

	int block_sectors;
	sector_t sector;
	int sectors;
	int sect_to_write = r1_bio->sectors;
	int ok = 1;

	if (rdev->badblocks.shift < 0)
		return 0;

2172 2173
	block_sectors = roundup(1 << rdev->badblocks.shift,
				bdev_logical_block_size(rdev->bdev) >> 9);
2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184
	sector = r1_bio->sector;
	sectors = ((sector + block_sectors)
		   & ~(sector_t)(block_sectors - 1))
		- sector;

	while (sect_to_write) {
		struct bio *wbio;
		if (sectors > sect_to_write)
			sectors = sect_to_write;
		/* Write at 'sector' for 'sectors'*/

2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201
		if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
			unsigned vcnt = r1_bio->behind_page_count;
			struct bio_vec *vec = r1_bio->behind_bvecs;

			while (!vec->bv_page) {
				vec++;
				vcnt--;
			}

			wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev);
			memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec));

			wbio->bi_vcnt = vcnt;
		} else {
			wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
		}

M
Mike Christie 已提交
2202
		bio_set_op_attrs(wbio, REQ_OP_WRITE, 0);
2203 2204
		wbio->bi_iter.bi_sector = r1_bio->sector;
		wbio->bi_iter.bi_size = r1_bio->sectors << 9;
2205

2206
		bio_trim(wbio, sector - r1_bio->sector, sectors);
2207
		wbio->bi_iter.bi_sector += rdev->data_offset;
2208
		wbio->bi_bdev = rdev->bdev;
2209 2210

		if (submit_bio_wait(wbio) < 0)
2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223
			/* failure! */
			ok = rdev_set_badblocks(rdev, sector,
						sectors, 0)
				&& ok;

		bio_put(wbio);
		sect_to_write -= sectors;
		sector += sectors;
		sectors = block_sectors;
	}
	return ok;
}

2224
static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
2225 2226 2227
{
	int m;
	int s = r1_bio->sectors;
2228
	for (m = 0; m < conf->raid_disks * 2 ; m++) {
2229
		struct md_rdev *rdev = conf->mirrors[m].rdev;
2230 2231 2232
		struct bio *bio = r1_bio->bios[m];
		if (bio->bi_end_io == NULL)
			continue;
2233
		if (!bio->bi_error &&
2234
		    test_bit(R1BIO_MadeGood, &r1_bio->state)) {
2235
			rdev_clear_badblocks(rdev, r1_bio->sector, s, 0);
2236
		}
2237
		if (bio->bi_error &&
2238 2239 2240 2241 2242 2243 2244 2245 2246
		    test_bit(R1BIO_WriteError, &r1_bio->state)) {
			if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0))
				md_error(conf->mddev, rdev);
		}
	}
	put_buf(r1_bio);
	md_done_sync(conf->mddev, s, 1);
}

2247
static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
2248 2249
{
	int m;
2250
	bool fail = false;
2251
	for (m = 0; m < conf->raid_disks * 2 ; m++)
2252
		if (r1_bio->bios[m] == IO_MADE_GOOD) {
2253
			struct md_rdev *rdev = conf->mirrors[m].rdev;
2254 2255
			rdev_clear_badblocks(rdev,
					     r1_bio->sector,
2256
					     r1_bio->sectors, 0);
2257 2258 2259 2260 2261 2262
			rdev_dec_pending(rdev, conf->mddev);
		} else if (r1_bio->bios[m] != NULL) {
			/* This drive got a write error.  We need to
			 * narrow down and record precise write
			 * errors.
			 */
2263
			fail = true;
2264 2265 2266 2267 2268 2269 2270 2271 2272
			if (!narrow_write_error(r1_bio, m)) {
				md_error(conf->mddev,
					 conf->mirrors[m].rdev);
				/* an I/O failed, we can't clear the bitmap */
				set_bit(R1BIO_Degraded, &r1_bio->state);
			}
			rdev_dec_pending(conf->mirrors[m].rdev,
					 conf->mddev);
		}
2273 2274 2275
	if (fail) {
		spin_lock_irq(&conf->device_lock);
		list_add(&r1_bio->retry_list, &conf->bio_end_io_list);
2276
		conf->nr_queued++;
2277 2278
		spin_unlock_irq(&conf->device_lock);
		md_wakeup_thread(conf->mddev->thread);
2279 2280 2281
	} else {
		if (test_bit(R1BIO_WriteError, &r1_bio->state))
			close_write(r1_bio);
2282
		raid_end_bio_io(r1_bio);
2283
	}
2284 2285
}

2286
static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio)
2287 2288 2289
{
	int disk;
	int max_sectors;
2290
	struct mddev *mddev = conf->mddev;
2291 2292
	struct bio *bio;
	char b[BDEVNAME_SIZE];
2293
	struct md_rdev *rdev;
2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304

	clear_bit(R1BIO_ReadError, &r1_bio->state);
	/* we got a read error. Maybe the drive is bad.  Maybe just
	 * the block and we can fix it.
	 * We freeze all other IO, and try reading the block from
	 * other devices.  When we find one, we re-write
	 * and check it that fixes the read error.
	 * This is all done synchronously while the array is
	 * frozen
	 */
	if (mddev->ro == 0) {
2305
		freeze_array(conf, 1);
2306 2307 2308 2309 2310
		fix_read_error(conf, r1_bio->read_disk,
			       r1_bio->sector, r1_bio->sectors);
		unfreeze_array(conf);
	} else
		md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
2311
	rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev);
2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331

	bio = r1_bio->bios[r1_bio->read_disk];
	bdevname(bio->bi_bdev, b);
read_more:
	disk = read_balance(conf, r1_bio, &max_sectors);
	if (disk == -1) {
		printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O"
		       " read error for block %llu\n",
		       mdname(mddev), b, (unsigned long long)r1_bio->sector);
		raid_end_bio_io(r1_bio);
	} else {
		const unsigned long do_sync
			= r1_bio->master_bio->bi_rw & REQ_SYNC;
		if (bio) {
			r1_bio->bios[r1_bio->read_disk] =
				mddev->ro ? IO_BLOCKED : NULL;
			bio_put(bio);
		}
		r1_bio->read_disk = disk;
		bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
2332 2333
		bio_trim(bio, r1_bio->sector - bio->bi_iter.bi_sector,
			 max_sectors);
2334 2335 2336 2337 2338 2339 2340 2341
		r1_bio->bios[r1_bio->read_disk] = bio;
		rdev = conf->mirrors[disk].rdev;
		printk_ratelimited(KERN_ERR
				   "md/raid1:%s: redirecting sector %llu"
				   " to other mirror: %s\n",
				   mdname(mddev),
				   (unsigned long long)r1_bio->sector,
				   bdevname(rdev->bdev, b));
2342
		bio->bi_iter.bi_sector = r1_bio->sector + rdev->data_offset;
2343 2344
		bio->bi_bdev = rdev->bdev;
		bio->bi_end_io = raid1_end_read_request;
M
Mike Christie 已提交
2345
		bio_set_op_attrs(bio, REQ_OP_READ, do_sync);
2346 2347 2348 2349 2350
		bio->bi_private = r1_bio;
		if (max_sectors < r1_bio->sectors) {
			/* Drat - have to split this up more */
			struct bio *mbio = r1_bio->master_bio;
			int sectors_handled = (r1_bio->sector + max_sectors
2351
					       - mbio->bi_iter.bi_sector);
2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364
			r1_bio->sectors = max_sectors;
			spin_lock_irq(&conf->device_lock);
			if (mbio->bi_phys_segments == 0)
				mbio->bi_phys_segments = 2;
			else
				mbio->bi_phys_segments++;
			spin_unlock_irq(&conf->device_lock);
			generic_make_request(bio);
			bio = NULL;

			r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);

			r1_bio->master_bio = mbio;
2365
			r1_bio->sectors = bio_sectors(mbio) - sectors_handled;
2366 2367 2368
			r1_bio->state = 0;
			set_bit(R1BIO_ReadError, &r1_bio->state);
			r1_bio->mddev = mddev;
2369 2370
			r1_bio->sector = mbio->bi_iter.bi_sector +
				sectors_handled;
2371 2372 2373 2374 2375 2376 2377

			goto read_more;
		} else
			generic_make_request(bio);
	}
}

S
Shaohua Li 已提交
2378
static void raid1d(struct md_thread *thread)
L
Linus Torvalds 已提交
2379
{
S
Shaohua Li 已提交
2380
	struct mddev *mddev = thread->mddev;
2381
	struct r1bio *r1_bio;
L
Linus Torvalds 已提交
2382
	unsigned long flags;
2383
	struct r1conf *conf = mddev->private;
L
Linus Torvalds 已提交
2384
	struct list_head *head = &conf->retry_list;
2385
	struct blk_plug plug;
L
Linus Torvalds 已提交
2386 2387

	md_check_recovery(mddev);
2388

2389 2390 2391 2392 2393
	if (!list_empty_careful(&conf->bio_end_io_list) &&
	    !test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
		LIST_HEAD(tmp);
		spin_lock_irqsave(&conf->device_lock, flags);
		if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
2394 2395 2396 2397
			while (!list_empty(&conf->bio_end_io_list)) {
				list_move(conf->bio_end_io_list.prev, &tmp);
				conf->nr_queued--;
			}
2398 2399 2400
		}
		spin_unlock_irqrestore(&conf->device_lock, flags);
		while (!list_empty(&tmp)) {
2401 2402
			r1_bio = list_first_entry(&tmp, struct r1bio,
						  retry_list);
2403
			list_del(&r1_bio->retry_list);
2404 2405 2406 2407
			if (mddev->degraded)
				set_bit(R1BIO_Degraded, &r1_bio->state);
			if (test_bit(R1BIO_WriteError, &r1_bio->state))
				close_write(r1_bio);
2408 2409 2410 2411
			raid_end_bio_io(r1_bio);
		}
	}

2412
	blk_start_plug(&plug);
L
Linus Torvalds 已提交
2413
	for (;;) {
2414

2415
		flush_pending_writes(conf);
2416

2417 2418 2419
		spin_lock_irqsave(&conf->device_lock, flags);
		if (list_empty(head)) {
			spin_unlock_irqrestore(&conf->device_lock, flags);
L
Linus Torvalds 已提交
2420
			break;
2421
		}
2422
		r1_bio = list_entry(head->prev, struct r1bio, retry_list);
L
Linus Torvalds 已提交
2423
		list_del(head->prev);
2424
		conf->nr_queued--;
L
Linus Torvalds 已提交
2425 2426 2427
		spin_unlock_irqrestore(&conf->device_lock, flags);

		mddev = r1_bio->mddev;
2428
		conf = mddev->private;
2429
		if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
2430
			if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2431 2432 2433
			    test_bit(R1BIO_WriteError, &r1_bio->state))
				handle_sync_write_finished(conf, r1_bio);
			else
2434
				sync_request_write(mddev, r1_bio);
2435
		} else if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2436 2437 2438 2439 2440
			   test_bit(R1BIO_WriteError, &r1_bio->state))
			handle_write_finished(conf, r1_bio);
		else if (test_bit(R1BIO_ReadError, &r1_bio->state))
			handle_read_error(conf, r1_bio);
		else
2441 2442 2443 2444
			/* just a partial read to be scheduled from separate
			 * context
			 */
			generic_make_request(r1_bio->bios[r1_bio->read_disk]);
2445

N
NeilBrown 已提交
2446
		cond_resched();
2447 2448
		if (mddev->flags & ~(1<<MD_CHANGE_PENDING))
			md_check_recovery(mddev);
L
Linus Torvalds 已提交
2449
	}
2450
	blk_finish_plug(&plug);
L
Linus Torvalds 已提交
2451 2452
}

2453
static int init_resync(struct r1conf *conf)
L
Linus Torvalds 已提交
2454 2455 2456 2457
{
	int buffs;

	buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
2458
	BUG_ON(conf->r1buf_pool);
L
Linus Torvalds 已提交
2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476
	conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
					  conf->poolinfo);
	if (!conf->r1buf_pool)
		return -ENOMEM;
	conf->next_resync = 0;
	return 0;
}

/*
 * perform a "sync" on one "block"
 *
 * We need to make sure that no normal I/O request - particularly write
 * requests - conflict with active sync requests.
 *
 * This is achieved by tracking pending requests and a 'barrier' concept
 * that can be installed to exclude normal IO requests.
 */

S
Shaohua Li 已提交
2477 2478
static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr,
				   int *skipped)
L
Linus Torvalds 已提交
2479
{
2480
	struct r1conf *conf = mddev->private;
2481
	struct r1bio *r1_bio;
L
Linus Torvalds 已提交
2482 2483
	struct bio *bio;
	sector_t max_sector, nr_sectors;
2484
	int disk = -1;
L
Linus Torvalds 已提交
2485
	int i;
2486 2487
	int wonly = -1;
	int write_targets = 0, read_targets = 0;
N
NeilBrown 已提交
2488
	sector_t sync_blocks;
2489
	int still_degraded = 0;
2490 2491
	int good_sectors = RESYNC_SECTORS;
	int min_bad = 0; /* number of sectors that are bad in all devices */
L
Linus Torvalds 已提交
2492 2493 2494

	if (!conf->r1buf_pool)
		if (init_resync(conf))
2495
			return 0;
L
Linus Torvalds 已提交
2496

A
Andre Noll 已提交
2497
	max_sector = mddev->dev_sectors;
L
Linus Torvalds 已提交
2498
	if (sector_nr >= max_sector) {
2499 2500 2501 2502 2503
		/* If we aborted, we need to abort the
		 * sync on the 'current' bitmap chunk (there will
		 * only be one in raid1 resync.
		 * We can find the current addess in mddev->curr_resync
		 */
2504 2505
		if (mddev->curr_resync < max_sector) /* aborted */
			bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
2506
						&sync_blocks, 1);
2507
		else /* completed sync */
2508
			conf->fullsync = 0;
2509 2510

		bitmap_close_sync(mddev->bitmap);
L
Linus Torvalds 已提交
2511
		close_sync(conf);
2512 2513 2514 2515 2516

		if (mddev_is_clustered(mddev)) {
			conf->cluster_sync_low = 0;
			conf->cluster_sync_high = 0;
		}
L
Linus Torvalds 已提交
2517 2518 2519
		return 0;
	}

2520 2521
	if (mddev->bitmap == NULL &&
	    mddev->recovery_cp == MaxSector &&
2522
	    !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
2523 2524 2525 2526
	    conf->fullsync == 0) {
		*skipped = 1;
		return max_sector - sector_nr;
	}
2527 2528 2529
	/* before building a request, check if we can skip these blocks..
	 * This call the bitmap_start_sync doesn't actually record anything
	 */
2530
	if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
2531
	    !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2532 2533 2534 2535
		/* We can skip this block, and probably several more */
		*skipped = 1;
		return sync_blocks;
	}
2536

2537 2538 2539 2540 2541 2542
	/* we are incrementing sector_nr below. To be safe, we check against
	 * sector_nr + two times RESYNC_SECTORS
	 */

	bitmap_cond_end_sync(mddev->bitmap, sector_nr,
		mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high));
2543
	r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
2544

2545
	raise_barrier(conf, sector_nr);
L
Linus Torvalds 已提交
2546

2547
	rcu_read_lock();
L
Linus Torvalds 已提交
2548
	/*
2549 2550 2551 2552 2553 2554
	 * If we get a correctably read error during resync or recovery,
	 * we might want to read from a different device.  So we
	 * flag all drives that could conceivably be read from for READ,
	 * and any others (which will be non-In_sync devices) for WRITE.
	 * If a read fails, we try reading from something else for which READ
	 * is OK.
L
Linus Torvalds 已提交
2555 2556 2557 2558
	 */

	r1_bio->mddev = mddev;
	r1_bio->sector = sector_nr;
2559
	r1_bio->state = 0;
L
Linus Torvalds 已提交
2560 2561
	set_bit(R1BIO_IsSync, &r1_bio->state);

2562
	for (i = 0; i < conf->raid_disks * 2; i++) {
2563
		struct md_rdev *rdev;
L
Linus Torvalds 已提交
2564
		bio = r1_bio->bios[i];
K
Kent Overstreet 已提交
2565
		bio_reset(bio);
L
Linus Torvalds 已提交
2566

2567 2568
		rdev = rcu_dereference(conf->mirrors[i].rdev);
		if (rdev == NULL ||
2569
		    test_bit(Faulty, &rdev->flags)) {
2570 2571
			if (i < conf->raid_disks)
				still_degraded = 1;
2572
		} else if (!test_bit(In_sync, &rdev->flags)) {
M
Mike Christie 已提交
2573
			bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
L
Linus Torvalds 已提交
2574 2575
			bio->bi_end_io = end_sync_write;
			write_targets ++;
2576 2577
		} else {
			/* may need to read from here */
2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599
			sector_t first_bad = MaxSector;
			int bad_sectors;

			if (is_badblock(rdev, sector_nr, good_sectors,
					&first_bad, &bad_sectors)) {
				if (first_bad > sector_nr)
					good_sectors = first_bad - sector_nr;
				else {
					bad_sectors -= (sector_nr - first_bad);
					if (min_bad == 0 ||
					    min_bad > bad_sectors)
						min_bad = bad_sectors;
				}
			}
			if (sector_nr < first_bad) {
				if (test_bit(WriteMostly, &rdev->flags)) {
					if (wonly < 0)
						wonly = i;
				} else {
					if (disk < 0)
						disk = i;
				}
M
Mike Christie 已提交
2600
				bio_set_op_attrs(bio, REQ_OP_READ, 0);
2601 2602
				bio->bi_end_io = end_sync_read;
				read_targets++;
2603 2604 2605 2606 2607 2608 2609 2610 2611
			} else if (!test_bit(WriteErrorSeen, &rdev->flags) &&
				test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
				!test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
				/*
				 * The device is suitable for reading (InSync),
				 * but has bad block(s) here. Let's try to correct them,
				 * if we are doing resync or repair. Otherwise, leave
				 * this device alone for this sync request.
				 */
M
Mike Christie 已提交
2612
				bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
2613 2614
				bio->bi_end_io = end_sync_write;
				write_targets++;
2615 2616
			}
		}
2617 2618
		if (bio->bi_end_io) {
			atomic_inc(&rdev->nr_pending);
2619
			bio->bi_iter.bi_sector = sector_nr + rdev->data_offset;
2620 2621 2622
			bio->bi_bdev = rdev->bdev;
			bio->bi_private = r1_bio;
		}
L
Linus Torvalds 已提交
2623
	}
2624 2625 2626 2627
	rcu_read_unlock();
	if (disk < 0)
		disk = wonly;
	r1_bio->read_disk = disk;
2628

2629 2630 2631 2632 2633
	if (read_targets == 0 && min_bad > 0) {
		/* These sectors are bad on all InSync devices, so we
		 * need to mark them bad on all write targets
		 */
		int ok = 1;
2634
		for (i = 0 ; i < conf->raid_disks * 2 ; i++)
2635
			if (r1_bio->bios[i]->bi_end_io == end_sync_write) {
2636
				struct md_rdev *rdev = conf->mirrors[i].rdev;
2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663
				ok = rdev_set_badblocks(rdev, sector_nr,
							min_bad, 0
					) && ok;
			}
		set_bit(MD_CHANGE_DEVS, &mddev->flags);
		*skipped = 1;
		put_buf(r1_bio);

		if (!ok) {
			/* Cannot record the badblocks, so need to
			 * abort the resync.
			 * If there are multiple read targets, could just
			 * fail the really bad ones ???
			 */
			conf->recovery_disabled = mddev->recovery_disabled;
			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
			return 0;
		} else
			return min_bad;

	}
	if (min_bad > 0 && min_bad < good_sectors) {
		/* only resync enough to reach the next bad->good
		 * transition */
		good_sectors = min_bad;
	}

2664 2665 2666 2667 2668
	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
		/* extra read targets are also write targets */
		write_targets += read_targets-1;

	if (write_targets == 0 || read_targets == 0) {
L
Linus Torvalds 已提交
2669 2670 2671
		/* There is nowhere to write, so all non-sync
		 * drives must be failed - so we are finished
		 */
2672 2673 2674 2675
		sector_t rv;
		if (min_bad > 0)
			max_sector = sector_nr + min_bad;
		rv = max_sector - sector_nr;
2676
		*skipped = 1;
L
Linus Torvalds 已提交
2677 2678 2679 2680
		put_buf(r1_bio);
		return rv;
	}

2681 2682
	if (max_sector > mddev->resync_max)
		max_sector = mddev->resync_max; /* Don't do IO beyond here */
2683 2684
	if (max_sector > sector_nr + good_sectors)
		max_sector = sector_nr + good_sectors;
L
Linus Torvalds 已提交
2685
	nr_sectors = 0;
2686
	sync_blocks = 0;
L
Linus Torvalds 已提交
2687 2688 2689 2690 2691 2692 2693
	do {
		struct page *page;
		int len = PAGE_SIZE;
		if (sector_nr + (len>>9) > max_sector)
			len = (max_sector - sector_nr) << 9;
		if (len == 0)
			break;
2694 2695
		if (sync_blocks == 0) {
			if (!bitmap_start_sync(mddev->bitmap, sector_nr,
2696 2697 2698
					       &sync_blocks, still_degraded) &&
			    !conf->fullsync &&
			    !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2699
				break;
2700
			if ((len >> 9) > sync_blocks)
2701
				len = sync_blocks<<9;
2702
		}
2703

2704
		for (i = 0 ; i < conf->raid_disks * 2; i++) {
L
Linus Torvalds 已提交
2705 2706
			bio = r1_bio->bios[i];
			if (bio->bi_end_io) {
2707
				page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
L
Linus Torvalds 已提交
2708 2709
				if (bio_add_page(bio, page, len, 0) == 0) {
					/* stop here */
2710
					bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
L
Linus Torvalds 已提交
2711 2712 2713
					while (i > 0) {
						i--;
						bio = r1_bio->bios[i];
2714 2715
						if (bio->bi_end_io==NULL)
							continue;
L
Linus Torvalds 已提交
2716 2717
						/* remove last page from this bio */
						bio->bi_vcnt--;
2718
						bio->bi_iter.bi_size -= len;
2719
						bio_clear_flag(bio, BIO_SEG_VALID);
L
Linus Torvalds 已提交
2720 2721 2722 2723 2724 2725 2726
					}
					goto bio_full;
				}
			}
		}
		nr_sectors += len>>9;
		sector_nr += len>>9;
2727
		sync_blocks -= (len>>9);
L
Linus Torvalds 已提交
2728 2729 2730 2731
	} while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
 bio_full:
	r1_bio->sectors = nr_sectors;

2732 2733 2734 2735 2736 2737 2738 2739 2740 2741
	if (mddev_is_clustered(mddev) &&
			conf->cluster_sync_high < sector_nr + nr_sectors) {
		conf->cluster_sync_low = mddev->curr_resync_completed;
		conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS;
		/* Send resync message */
		md_cluster_ops->resync_info_update(mddev,
				conf->cluster_sync_low,
				conf->cluster_sync_high);
	}

2742 2743 2744 2745 2746
	/* For a user-requested sync, we read all readable devices and do a
	 * compare
	 */
	if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
		atomic_set(&r1_bio->remaining, read_targets);
2747
		for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) {
2748 2749
			bio = r1_bio->bios[i];
			if (bio->bi_end_io == end_sync_read) {
2750
				read_targets--;
2751
				md_sync_acct(bio->bi_bdev, nr_sectors);
2752 2753 2754 2755 2756 2757
				generic_make_request(bio);
			}
		}
	} else {
		atomic_set(&r1_bio->remaining, 1);
		bio = r1_bio->bios[r1_bio->read_disk];
2758
		md_sync_acct(bio->bi_bdev, nr_sectors);
2759
		generic_make_request(bio);
L
Linus Torvalds 已提交
2760

2761
	}
L
Linus Torvalds 已提交
2762 2763 2764
	return nr_sectors;
}

2765
static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks)
2766 2767 2768 2769 2770 2771 2772
{
	if (sectors)
		return sectors;

	return mddev->dev_sectors;
}

2773
static struct r1conf *setup_conf(struct mddev *mddev)
L
Linus Torvalds 已提交
2774
{
2775
	struct r1conf *conf;
2776
	int i;
2777
	struct raid1_info *disk;
2778
	struct md_rdev *rdev;
2779
	int err = -ENOMEM;
L
Linus Torvalds 已提交
2780

2781
	conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL);
L
Linus Torvalds 已提交
2782
	if (!conf)
2783
		goto abort;
L
Linus Torvalds 已提交
2784

2785
	conf->mirrors = kzalloc(sizeof(struct raid1_info)
2786
				* mddev->raid_disks * 2,
L
Linus Torvalds 已提交
2787 2788
				 GFP_KERNEL);
	if (!conf->mirrors)
2789
		goto abort;
L
Linus Torvalds 已提交
2790

2791 2792
	conf->tmppage = alloc_page(GFP_KERNEL);
	if (!conf->tmppage)
2793
		goto abort;
2794

2795
	conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
L
Linus Torvalds 已提交
2796
	if (!conf->poolinfo)
2797
		goto abort;
2798
	conf->poolinfo->raid_disks = mddev->raid_disks * 2;
L
Linus Torvalds 已提交
2799 2800 2801 2802
	conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
					  r1bio_pool_free,
					  conf->poolinfo);
	if (!conf->r1bio_pool)
2803 2804
		goto abort;

2805
	conf->poolinfo->mddev = mddev;
L
Linus Torvalds 已提交
2806

2807
	err = -EINVAL;
2808
	spin_lock_init(&conf->device_lock);
N
NeilBrown 已提交
2809
	rdev_for_each(rdev, mddev) {
2810
		struct request_queue *q;
2811
		int disk_idx = rdev->raid_disk;
L
Linus Torvalds 已提交
2812 2813 2814
		if (disk_idx >= mddev->raid_disks
		    || disk_idx < 0)
			continue;
2815
		if (test_bit(Replacement, &rdev->flags))
2816
			disk = conf->mirrors + mddev->raid_disks + disk_idx;
2817 2818
		else
			disk = conf->mirrors + disk_idx;
L
Linus Torvalds 已提交
2819

2820 2821
		if (disk->rdev)
			goto abort;
L
Linus Torvalds 已提交
2822
		disk->rdev = rdev;
2823
		q = bdev_get_queue(rdev->bdev);
L
Linus Torvalds 已提交
2824 2825

		disk->head_position = 0;
2826
		disk->seq_start = MaxSector;
L
Linus Torvalds 已提交
2827 2828 2829 2830
	}
	conf->raid_disks = mddev->raid_disks;
	conf->mddev = mddev;
	INIT_LIST_HEAD(&conf->retry_list);
2831
	INIT_LIST_HEAD(&conf->bio_end_io_list);
L
Linus Torvalds 已提交
2832 2833

	spin_lock_init(&conf->resync_lock);
2834
	init_waitqueue_head(&conf->wait_barrier);
L
Linus Torvalds 已提交
2835

2836
	bio_list_init(&conf->pending_bio_list);
2837
	conf->pending_count = 0;
2838
	conf->recovery_disabled = mddev->recovery_disabled - 1;
2839

2840 2841 2842
	conf->start_next_window = MaxSector;
	conf->current_window_requests = conf->next_window_requests = 0;

2843
	err = -EIO;
2844
	for (i = 0; i < conf->raid_disks * 2; i++) {
L
Linus Torvalds 已提交
2845 2846 2847

		disk = conf->mirrors + i;

2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862
		if (i < conf->raid_disks &&
		    disk[conf->raid_disks].rdev) {
			/* This slot has a replacement. */
			if (!disk->rdev) {
				/* No original, just make the replacement
				 * a recovering spare
				 */
				disk->rdev =
					disk[conf->raid_disks].rdev;
				disk[conf->raid_disks].rdev = NULL;
			} else if (!test_bit(In_sync, &disk->rdev->flags))
				/* Original is not in_sync - bad */
				goto abort;
		}

2863 2864
		if (!disk->rdev ||
		    !test_bit(In_sync, &disk->rdev->flags)) {
L
Linus Torvalds 已提交
2865
			disk->head_position = 0;
2866 2867
			if (disk->rdev &&
			    (disk->rdev->saved_raid_disk < 0))
2868
				conf->fullsync = 1;
2869
		}
L
Linus Torvalds 已提交
2870
	}
2871 2872

	err = -ENOMEM;
2873
	conf->thread = md_register_thread(raid1d, mddev, "raid1");
2874 2875
	if (!conf->thread) {
		printk(KERN_ERR
N
NeilBrown 已提交
2876
		       "md/raid1:%s: couldn't allocate thread\n",
2877 2878
		       mdname(mddev));
		goto abort;
2879
	}
L
Linus Torvalds 已提交
2880

2881 2882 2883 2884
	return conf;

 abort:
	if (conf) {
2885
		mempool_destroy(conf->r1bio_pool);
2886 2887 2888 2889 2890 2891 2892 2893
		kfree(conf->mirrors);
		safe_put_page(conf->tmppage);
		kfree(conf->poolinfo);
		kfree(conf);
	}
	return ERR_PTR(err);
}

N
NeilBrown 已提交
2894
static void raid1_free(struct mddev *mddev, void *priv);
S
Shaohua Li 已提交
2895
static int raid1_run(struct mddev *mddev)
2896
{
2897
	struct r1conf *conf;
2898
	int i;
2899
	struct md_rdev *rdev;
2900
	int ret;
S
Shaohua Li 已提交
2901
	bool discard_supported = false;
2902 2903

	if (mddev->level != 1) {
N
NeilBrown 已提交
2904
		printk(KERN_ERR "md/raid1:%s: raid level not set to mirroring (%d)\n",
2905 2906 2907 2908
		       mdname(mddev), mddev->level);
		return -EIO;
	}
	if (mddev->reshape_position != MaxSector) {
N
NeilBrown 已提交
2909
		printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n",
2910 2911 2912
		       mdname(mddev));
		return -EIO;
	}
L
Linus Torvalds 已提交
2913
	/*
2914 2915
	 * copy the already verified devices into our private RAID1
	 * bookkeeping area. [whatever we allocate in run(),
N
NeilBrown 已提交
2916
	 * should be freed in raid1_free()]
L
Linus Torvalds 已提交
2917
	 */
2918 2919 2920 2921
	if (mddev->private == NULL)
		conf = setup_conf(mddev);
	else
		conf = mddev->private;
L
Linus Torvalds 已提交
2922

2923 2924
	if (IS_ERR(conf))
		return PTR_ERR(conf);
L
Linus Torvalds 已提交
2925

2926
	if (mddev->queue)
2927 2928
		blk_queue_max_write_same_sectors(mddev->queue, 0);

N
NeilBrown 已提交
2929
	rdev_for_each(rdev, mddev) {
2930 2931
		if (!mddev->gendisk)
			continue;
2932 2933
		disk_stack_limits(mddev->gendisk, rdev->bdev,
				  rdev->data_offset << 9);
S
Shaohua Li 已提交
2934 2935
		if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
			discard_supported = true;
L
Linus Torvalds 已提交
2936
	}
2937

2938 2939 2940 2941 2942 2943 2944 2945 2946 2947
	mddev->degraded = 0;
	for (i=0; i < conf->raid_disks; i++)
		if (conf->mirrors[i].rdev == NULL ||
		    !test_bit(In_sync, &conf->mirrors[i].rdev->flags) ||
		    test_bit(Faulty, &conf->mirrors[i].rdev->flags))
			mddev->degraded++;

	if (conf->raid_disks - mddev->degraded == 1)
		mddev->recovery_cp = MaxSector;

2948
	if (mddev->recovery_cp != MaxSector)
N
NeilBrown 已提交
2949
		printk(KERN_NOTICE "md/raid1:%s: not clean"
2950 2951
		       " -- starting background reconstruction\n",
		       mdname(mddev));
2952
	printk(KERN_INFO
N
NeilBrown 已提交
2953
		"md/raid1:%s: active with %d out of %d mirrors\n",
2954
		mdname(mddev), mddev->raid_disks - mddev->degraded,
L
Linus Torvalds 已提交
2955
		mddev->raid_disks);
2956

L
Linus Torvalds 已提交
2957 2958 2959
	/*
	 * Ok, everything is just fine now
	 */
2960 2961 2962 2963
	mddev->thread = conf->thread;
	conf->thread = NULL;
	mddev->private = conf;

2964
	md_set_array_sectors(mddev, raid1_size(mddev, 0, 0));
L
Linus Torvalds 已提交
2965

2966
	if (mddev->queue) {
S
Shaohua Li 已提交
2967 2968 2969 2970 2971 2972
		if (discard_supported)
			queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
						mddev->queue);
		else
			queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
						  mddev->queue);
2973
	}
2974 2975

	ret =  md_integrity_register(mddev);
2976 2977
	if (ret) {
		md_unregister_thread(&mddev->thread);
N
NeilBrown 已提交
2978
		raid1_free(mddev, conf);
2979
	}
2980
	return ret;
L
Linus Torvalds 已提交
2981 2982
}

N
NeilBrown 已提交
2983
static void raid1_free(struct mddev *mddev, void *priv)
L
Linus Torvalds 已提交
2984
{
N
NeilBrown 已提交
2985
	struct r1conf *conf = priv;
2986

2987
	mempool_destroy(conf->r1bio_pool);
2988
	kfree(conf->mirrors);
2989
	safe_put_page(conf->tmppage);
2990
	kfree(conf->poolinfo);
L
Linus Torvalds 已提交
2991 2992 2993
	kfree(conf);
}

2994
static int raid1_resize(struct mddev *mddev, sector_t sectors)
L
Linus Torvalds 已提交
2995 2996 2997 2998 2999 3000 3001 3002
{
	/* no resync is happening, and there is enough space
	 * on all devices, so we can resize.
	 * We need to make sure resync covers any new space.
	 * If the array is shrinking we should possibly wait until
	 * any io in the removed space completes, but it hardly seems
	 * worth it.
	 */
3003 3004 3005
	sector_t newsize = raid1_size(mddev, sectors, 0);
	if (mddev->external_size &&
	    mddev->array_sectors > newsize)
D
Dan Williams 已提交
3006
		return -EINVAL;
3007 3008 3009 3010 3011 3012
	if (mddev->bitmap) {
		int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0);
		if (ret)
			return ret;
	}
	md_set_array_sectors(mddev, newsize);
3013
	set_capacity(mddev->gendisk, mddev->array_sectors);
3014
	revalidate_disk(mddev->gendisk);
D
Dan Williams 已提交
3015
	if (sectors > mddev->dev_sectors &&
3016
	    mddev->recovery_cp > mddev->dev_sectors) {
A
Andre Noll 已提交
3017
		mddev->recovery_cp = mddev->dev_sectors;
L
Linus Torvalds 已提交
3018 3019
		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
	}
D
Dan Williams 已提交
3020
	mddev->dev_sectors = sectors;
3021
	mddev->resync_max_sectors = sectors;
L
Linus Torvalds 已提交
3022 3023 3024
	return 0;
}

3025
static int raid1_reshape(struct mddev *mddev)
L
Linus Torvalds 已提交
3026 3027 3028 3029 3030 3031 3032 3033
{
	/* We need to:
	 * 1/ resize the r1bio_pool
	 * 2/ resize conf->mirrors
	 *
	 * We allocate a new r1bio_pool if we can.
	 * Then raise a device barrier and wait until all IO stops.
	 * Then resize conf->mirrors and swap in the new r1bio pool.
3034 3035 3036
	 *
	 * At the same time, we "pack" the devices so that all the missing
	 * devices have the higher raid_disk numbers.
L
Linus Torvalds 已提交
3037 3038 3039
	 */
	mempool_t *newpool, *oldpool;
	struct pool_info *newpoolinfo;
3040
	struct raid1_info *newmirrors;
3041
	struct r1conf *conf = mddev->private;
3042
	int cnt, raid_disks;
3043
	unsigned long flags;
3044
	int d, d2, err;
L
Linus Torvalds 已提交
3045

3046
	/* Cannot change chunk_size, layout, or level */
3047
	if (mddev->chunk_sectors != mddev->new_chunk_sectors ||
3048 3049
	    mddev->layout != mddev->new_layout ||
	    mddev->level != mddev->new_level) {
3050
		mddev->new_chunk_sectors = mddev->chunk_sectors;
3051 3052 3053 3054 3055
		mddev->new_layout = mddev->layout;
		mddev->new_level = mddev->level;
		return -EINVAL;
	}

3056 3057 3058 3059 3060
	if (!mddev_is_clustered(mddev)) {
		err = md_allow_write(mddev);
		if (err)
			return err;
	}
3061

3062 3063
	raid_disks = mddev->raid_disks + mddev->delta_disks;

3064 3065 3066 3067 3068 3069
	if (raid_disks < conf->raid_disks) {
		cnt=0;
		for (d= 0; d < conf->raid_disks; d++)
			if (conf->mirrors[d].rdev)
				cnt++;
		if (cnt > raid_disks)
L
Linus Torvalds 已提交
3070
			return -EBUSY;
3071
	}
L
Linus Torvalds 已提交
3072 3073 3074 3075 3076

	newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
	if (!newpoolinfo)
		return -ENOMEM;
	newpoolinfo->mddev = mddev;
3077
	newpoolinfo->raid_disks = raid_disks * 2;
L
Linus Torvalds 已提交
3078 3079 3080 3081 3082 3083 3084

	newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
				 r1bio_pool_free, newpoolinfo);
	if (!newpool) {
		kfree(newpoolinfo);
		return -ENOMEM;
	}
3085
	newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2,
3086
			     GFP_KERNEL);
L
Linus Torvalds 已提交
3087 3088 3089 3090 3091 3092
	if (!newmirrors) {
		kfree(newpoolinfo);
		mempool_destroy(newpool);
		return -ENOMEM;
	}

3093
	freeze_array(conf, 0);
L
Linus Torvalds 已提交
3094 3095 3096 3097

	/* ok, everything is stopped */
	oldpool = conf->r1bio_pool;
	conf->r1bio_pool = newpool;
3098

3099
	for (d = d2 = 0; d < conf->raid_disks; d++) {
3100
		struct md_rdev *rdev = conf->mirrors[d].rdev;
3101
		if (rdev && rdev->raid_disk != d2) {
3102
			sysfs_unlink_rdev(mddev, rdev);
3103
			rdev->raid_disk = d2;
3104 3105
			sysfs_unlink_rdev(mddev, rdev);
			if (sysfs_link_rdev(mddev, rdev))
3106
				printk(KERN_WARNING
3107 3108
				       "md/raid1:%s: cannot register rd%d\n",
				       mdname(mddev), rdev->raid_disk);
3109
		}
3110 3111 3112
		if (rdev)
			newmirrors[d2++].rdev = rdev;
	}
L
Linus Torvalds 已提交
3113 3114 3115 3116 3117
	kfree(conf->mirrors);
	conf->mirrors = newmirrors;
	kfree(conf->poolinfo);
	conf->poolinfo = newpoolinfo;

3118
	spin_lock_irqsave(&conf->device_lock, flags);
L
Linus Torvalds 已提交
3119
	mddev->degraded += (raid_disks - conf->raid_disks);
3120
	spin_unlock_irqrestore(&conf->device_lock, flags);
L
Linus Torvalds 已提交
3121
	conf->raid_disks = mddev->raid_disks = raid_disks;
3122
	mddev->delta_disks = 0;
L
Linus Torvalds 已提交
3123

3124
	unfreeze_array(conf);
L
Linus Torvalds 已提交
3125

3126
	set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
L
Linus Torvalds 已提交
3127 3128 3129 3130 3131 3132 3133
	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
	md_wakeup_thread(mddev->thread);

	mempool_destroy(oldpool);
	return 0;
}

3134
static void raid1_quiesce(struct mddev *mddev, int state)
3135
{
3136
	struct r1conf *conf = mddev->private;
3137 3138

	switch(state) {
3139 3140 3141
	case 2: /* wake for suspend */
		wake_up(&conf->wait_barrier);
		break;
3142
	case 1:
3143
		freeze_array(conf, 0);
3144
		break;
3145
	case 0:
3146
		unfreeze_array(conf);
3147 3148 3149 3150
		break;
	}
}

3151
static void *raid1_takeover(struct mddev *mddev)
3152 3153 3154 3155 3156
{
	/* raid1 can take over:
	 *  raid5 with 2 devices, any layout or chunk size
	 */
	if (mddev->level == 5 && mddev->raid_disks == 2) {
3157
		struct r1conf *conf;
3158 3159 3160 3161 3162
		mddev->new_level = 1;
		mddev->new_layout = 0;
		mddev->new_chunk_sectors = 0;
		conf = setup_conf(mddev);
		if (!IS_ERR(conf))
3163 3164
			/* Array must appear to be quiesced */
			conf->array_frozen = 1;
3165 3166 3167 3168
		return conf;
	}
	return ERR_PTR(-EINVAL);
}
L
Linus Torvalds 已提交
3169

3170
static struct md_personality raid1_personality =
L
Linus Torvalds 已提交
3171 3172
{
	.name		= "raid1",
3173
	.level		= 1,
L
Linus Torvalds 已提交
3174
	.owner		= THIS_MODULE,
S
Shaohua Li 已提交
3175 3176
	.make_request	= raid1_make_request,
	.run		= raid1_run,
N
NeilBrown 已提交
3177
	.free		= raid1_free,
S
Shaohua Li 已提交
3178 3179
	.status		= raid1_status,
	.error_handler	= raid1_error,
L
Linus Torvalds 已提交
3180 3181 3182
	.hot_add_disk	= raid1_add_disk,
	.hot_remove_disk= raid1_remove_disk,
	.spare_active	= raid1_spare_active,
S
Shaohua Li 已提交
3183
	.sync_request	= raid1_sync_request,
L
Linus Torvalds 已提交
3184
	.resize		= raid1_resize,
3185
	.size		= raid1_size,
3186
	.check_reshape	= raid1_reshape,
3187
	.quiesce	= raid1_quiesce,
3188
	.takeover	= raid1_takeover,
3189
	.congested	= raid1_congested,
L
Linus Torvalds 已提交
3190 3191 3192 3193
};

static int __init raid_init(void)
{
3194
	return register_md_personality(&raid1_personality);
L
Linus Torvalds 已提交
3195 3196 3197 3198
}

static void raid_exit(void)
{
3199
	unregister_md_personality(&raid1_personality);
L
Linus Torvalds 已提交
3200 3201 3202 3203 3204
}

module_init(raid_init);
module_exit(raid_exit);
MODULE_LICENSE("GPL");
3205
MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
L
Linus Torvalds 已提交
3206
MODULE_ALIAS("md-personality-3"); /* RAID1 */
3207
MODULE_ALIAS("md-raid1");
3208
MODULE_ALIAS("md-level-1");
3209 3210

module_param(max_queued_requests, int, S_IRUGO|S_IWUSR);