dm-mpath.c 48.5 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7
/*
 * Copyright (C) 2003 Sistina Software Limited.
 * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved.
 *
 * This file is released under the GPL.
 */

8 9
#include <linux/device-mapper.h>

10
#include "dm-rq.h"
11
#include "dm-bio-record.h"
L
Linus Torvalds 已提交
12
#include "dm-path-selector.h"
M
Mike Anderson 已提交
13
#include "dm-uevent.h"
L
Linus Torvalds 已提交
14

15
#include <linux/blkdev.h>
L
Linus Torvalds 已提交
16 17 18 19 20 21 22 23
#include <linux/ctype.h>
#include <linux/init.h>
#include <linux/mempool.h>
#include <linux/module.h>
#include <linux/pagemap.h>
#include <linux/slab.h>
#include <linux/time.h>
#include <linux/workqueue.h>
24
#include <linux/delay.h>
25
#include <scsi/scsi_dh.h>
A
Arun Sharma 已提交
26
#include <linux/atomic.h>
27
#include <linux/blk-mq.h>
L
Linus Torvalds 已提交
28

29
#define DM_MSG_PREFIX "multipath"
30 31
#define DM_PG_INIT_DELAY_MSECS 2000
#define DM_PG_INIT_DELAY_DEFAULT ((unsigned) -1)
L
Linus Torvalds 已提交
32 33 34 35 36 37 38 39

/* Path properties */
struct pgpath {
	struct list_head list;

	struct priority_group *pg;	/* Owning PG */
	unsigned fail_count;		/* Cumulative failure count */

40
	struct dm_path path;
41
	struct delayed_work activate_path;
42 43

	bool is_active:1;		/* Path status */
L
Linus Torvalds 已提交
44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
};

#define path_to_pgpath(__pgp) container_of((__pgp), struct pgpath, path)

/*
 * Paths are grouped into Priority Groups and numbered from 1 upwards.
 * Each has a path selector which controls which path gets used.
 */
struct priority_group {
	struct list_head list;

	struct multipath *m;		/* Owning multipath instance */
	struct path_selector ps;

	unsigned pg_num;		/* Reference number */
	unsigned nr_pgpaths;		/* Number of paths in PG */
	struct list_head pgpaths;
61 62

	bool bypassed:1;		/* Temporarily bypass this PG? */
L
Linus Torvalds 已提交
63 64 65 66 67 68 69
};

/* Multipath context */
struct multipath {
	struct list_head list;
	struct dm_target *ti;

70
	const char *hw_handler_name;
71
	char *hw_handler_params;
72

73 74
	spinlock_t lock;

L
Linus Torvalds 已提交
75 76
	unsigned nr_priority_groups;
	struct list_head priority_groups;
77 78 79

	wait_queue_head_t pg_init_wait;	/* Wait for pg_init completion */

L
Linus Torvalds 已提交
80 81 82 83
	struct pgpath *current_pgpath;
	struct priority_group *current_pg;
	struct priority_group *next_pg;	/* Switch to this PG if set */

84
	unsigned long flags;		/* Multipath state flags */
85

D
Dave Wysochanski 已提交
86
	unsigned pg_init_retries;	/* Number of times to retry pg_init */
87
	unsigned pg_init_delay_msecs;	/* Number of msecs before pg_init retry */
L
Linus Torvalds 已提交
88

89 90 91 92
	atomic_t nr_valid_paths;	/* Total number of usable paths */
	atomic_t pg_init_in_progress;	/* Only one pg_init allowed at once */
	atomic_t pg_init_count;		/* Number of times pg_init called */

93
	enum dm_queue_mode queue_mode;
94

95
	struct mutex work_mutex;
96
	struct work_struct trigger_event;
97 98 99

	struct work_struct process_queued_bios;
	struct bio_list queued_bios;
L
Linus Torvalds 已提交
100 101 102
};

/*
103
 * Context information attached to each io we process.
L
Linus Torvalds 已提交
104
 */
A
Alasdair G Kergon 已提交
105
struct dm_mpath_io {
L
Linus Torvalds 已提交
106
	struct pgpath *pgpath;
107
	size_t nr_bytes;
L
Linus Torvalds 已提交
108 109 110 111
};

typedef int (*action_fn) (struct pgpath *pgpath);

112
static struct workqueue_struct *kmultipathd, *kmpath_handlerd;
D
David Howells 已提交
113
static void trigger_event(struct work_struct *work);
114 115
static void activate_or_offline_path(struct pgpath *pgpath);
static void activate_path_work(struct work_struct *work);
116
static void process_queued_bios(struct work_struct *work);
L
Linus Torvalds 已提交
117

118 119 120 121 122 123 124 125 126 127 128
/*-----------------------------------------------
 * Multipath state flags.
 *-----------------------------------------------*/

#define MPATHF_QUEUE_IO 0			/* Must we queue all I/O? */
#define MPATHF_QUEUE_IF_NO_PATH 1		/* Queue I/O if last path fails? */
#define MPATHF_SAVED_QUEUE_IF_NO_PATH 2		/* Saved state during suspension */
#define MPATHF_RETAIN_ATTACHED_HW_HANDLER 3	/* If there's already a hw_handler present, don't change it. */
#define MPATHF_PG_INIT_DISABLED 4		/* pg_init is not currently allowed */
#define MPATHF_PG_INIT_REQUIRED 5		/* pg_init needs calling? */
#define MPATHF_PG_INIT_DELAY_RETRY 6		/* Delay pg_init retry? */
L
Linus Torvalds 已提交
129 130 131 132 133 134 135

/*-----------------------------------------------
 * Allocation routines
 *-----------------------------------------------*/

static struct pgpath *alloc_pgpath(void)
{
M
Micha³ Miros³aw 已提交
136
	struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL);
L
Linus Torvalds 已提交
137

138
	if (pgpath) {
139
		pgpath->is_active = true;
140
		INIT_DELAYED_WORK(&pgpath->activate_path, activate_path_work);
141
	}
L
Linus Torvalds 已提交
142 143 144 145

	return pgpath;
}

A
Alasdair G Kergon 已提交
146
static void free_pgpath(struct pgpath *pgpath)
L
Linus Torvalds 已提交
147 148 149 150 151 152 153 154
{
	kfree(pgpath);
}

static struct priority_group *alloc_priority_group(void)
{
	struct priority_group *pg;

M
Micha³ Miros³aw 已提交
155
	pg = kzalloc(sizeof(*pg), GFP_KERNEL);
L
Linus Torvalds 已提交
156

M
Micha³ Miros³aw 已提交
157 158
	if (pg)
		INIT_LIST_HEAD(&pg->pgpaths);
L
Linus Torvalds 已提交
159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187

	return pg;
}

static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti)
{
	struct pgpath *pgpath, *tmp;

	list_for_each_entry_safe(pgpath, tmp, pgpaths, list) {
		list_del(&pgpath->list);
		dm_put_device(ti, pgpath->path.dev);
		free_pgpath(pgpath);
	}
}

static void free_priority_group(struct priority_group *pg,
				struct dm_target *ti)
{
	struct path_selector *ps = &pg->ps;

	if (ps->type) {
		ps->type->destroy(ps);
		dm_put_path_selector(ps->type);
	}

	free_pgpaths(&pg->pgpaths, ti);
	kfree(pg);
}

188
static struct multipath *alloc_multipath(struct dm_target *ti)
L
Linus Torvalds 已提交
189 190 191
{
	struct multipath *m;

M
Micha³ Miros³aw 已提交
192
	m = kzalloc(sizeof(*m), GFP_KERNEL);
L
Linus Torvalds 已提交
193 194 195
	if (m) {
		INIT_LIST_HEAD(&m->priority_groups);
		spin_lock_init(&m->lock);
196
		set_bit(MPATHF_QUEUE_IO, &m->flags);
197 198 199
		atomic_set(&m->nr_valid_paths, 0);
		atomic_set(&m->pg_init_in_progress, 0);
		atomic_set(&m->pg_init_count, 0);
200
		m->pg_init_delay_msecs = DM_PG_INIT_DELAY_DEFAULT;
D
David Howells 已提交
201
		INIT_WORK(&m->trigger_event, trigger_event);
202
		init_waitqueue_head(&m->pg_init_wait);
203
		mutex_init(&m->work_mutex);
204

205
		m->queue_mode = DM_TYPE_NONE;
206

M
Micha³ Miros³aw 已提交
207 208
		m->ti = ti;
		ti->private = m;
L
Linus Torvalds 已提交
209 210 211 212 213
	}

	return m;
}

214 215 216 217 218 219 220 221 222 223
static int alloc_multipath_stage2(struct dm_target *ti, struct multipath *m)
{
	if (m->queue_mode == DM_TYPE_NONE) {
		/*
		 * Default to request-based.
		 */
		if (dm_use_blk_mq(dm_table_get_md(ti->table)))
			m->queue_mode = DM_TYPE_MQ_REQUEST_BASED;
		else
			m->queue_mode = DM_TYPE_REQUEST_BASED;
224
	} else if (m->queue_mode == DM_TYPE_BIO_BASED) {
225 226 227 228 229 230 231 232 233 234 235 236 237
		INIT_WORK(&m->process_queued_bios, process_queued_bios);
		/*
		 * bio-based doesn't support any direct scsi_dh management;
		 * it just discovers if a scsi_dh is attached.
		 */
		set_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags);
	}

	dm_table_set_type(ti->table, m->queue_mode);

	return 0;
}

L
Linus Torvalds 已提交
238 239 240 241 242 243 244 245 246
static void free_multipath(struct multipath *m)
{
	struct priority_group *pg, *tmp;

	list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) {
		list_del(&pg->list);
		free_priority_group(pg, m->ti);
	}

247
	kfree(m->hw_handler_name);
248
	kfree(m->hw_handler_params);
L
Linus Torvalds 已提交
249 250 251
	kfree(m);
}

252 253 254 255 256
static struct dm_mpath_io *get_mpio(union map_info *info)
{
	return info->ptr;
}

257 258 259 260 261
static size_t multipath_per_bio_data_size(void)
{
	return sizeof(struct dm_mpath_io) + sizeof(struct dm_bio_details);
}

262 263
static struct dm_mpath_io *get_mpio_from_bio(struct bio *bio)
{
264
	return dm_per_bio_data(bio, multipath_per_bio_data_size());
265 266
}

267
static struct dm_bio_details *get_bio_details_from_bio(struct bio *bio)
268
{
269
	/* dm_bio_details is immediately after the dm_mpath_io in bio's per-bio-data */
270
	struct dm_mpath_io *mpio = get_mpio_from_bio(bio);
271 272 273 274 275 276 277 278 279 280
	void *bio_details = mpio + 1;

	return bio_details;
}

static void multipath_init_per_bio_data(struct bio *bio, struct dm_mpath_io **mpio_p,
					struct dm_bio_details **bio_details_p)
{
	struct dm_mpath_io *mpio = get_mpio_from_bio(bio);
	struct dm_bio_details *bio_details = get_bio_details_from_bio(bio);
281 282

	memset(mpio, 0, sizeof(*mpio));
283 284
	memset(bio_details, 0, sizeof(*bio_details));
	dm_bio_record(bio_details, bio);
285

286 287 288 289
	if (mpio_p)
		*mpio_p = mpio;
	if (bio_details_p)
		*bio_details_p = bio_details;
290 291
}

L
Linus Torvalds 已提交
292 293 294 295
/*-----------------------------------------------
 * Path selection
 *-----------------------------------------------*/

296
static int __pg_init_all_paths(struct multipath *m)
K
Kiyoshi Ueda 已提交
297 298
{
	struct pgpath *pgpath;
299
	unsigned long pg_init_delay = 0;
K
Kiyoshi Ueda 已提交
300

301 302
	lockdep_assert_held(&m->lock);

303
	if (atomic_read(&m->pg_init_in_progress) || test_bit(MPATHF_PG_INIT_DISABLED, &m->flags))
304
		return 0;
305

306
	atomic_inc(&m->pg_init_count);
307
	clear_bit(MPATHF_PG_INIT_REQUIRED, &m->flags);
308 309 310 311 312

	/* Check here to reset pg_init_required */
	if (!m->current_pg)
		return 0;

313
	if (test_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags))
314 315
		pg_init_delay = msecs_to_jiffies(m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT ?
						 m->pg_init_delay_msecs : DM_PG_INIT_DELAY_MSECS);
K
Kiyoshi Ueda 已提交
316 317 318 319
	list_for_each_entry(pgpath, &m->current_pg->pgpaths, list) {
		/* Skip failed paths */
		if (!pgpath->is_active)
			continue;
320 321
		if (queue_delayed_work(kmpath_handlerd, &pgpath->activate_path,
				       pg_init_delay))
322
			atomic_inc(&m->pg_init_in_progress);
K
Kiyoshi Ueda 已提交
323
	}
324
	return atomic_read(&m->pg_init_in_progress);
K
Kiyoshi Ueda 已提交
325 326
}

327
static int pg_init_all_paths(struct multipath *m)
L
Linus Torvalds 已提交
328
{
329
	int ret;
330 331 332
	unsigned long flags;

	spin_lock_irqsave(&m->lock, flags);
333
	ret = __pg_init_all_paths(m);
334
	spin_unlock_irqrestore(&m->lock, flags);
335 336

	return ret;
337 338 339 340 341
}

static void __switch_pg(struct multipath *m, struct priority_group *pg)
{
	m->current_pg = pg;
L
Linus Torvalds 已提交
342 343

	/* Must we initialise the PG first, and queue I/O till it's ready? */
344
	if (m->hw_handler_name) {
345 346
		set_bit(MPATHF_PG_INIT_REQUIRED, &m->flags);
		set_bit(MPATHF_QUEUE_IO, &m->flags);
L
Linus Torvalds 已提交
347
	} else {
348 349
		clear_bit(MPATHF_PG_INIT_REQUIRED, &m->flags);
		clear_bit(MPATHF_QUEUE_IO, &m->flags);
L
Linus Torvalds 已提交
350
	}
D
Dave Wysochanski 已提交
351

352
	atomic_set(&m->pg_init_count, 0);
L
Linus Torvalds 已提交
353 354
}

355 356 357
static struct pgpath *choose_path_in_pg(struct multipath *m,
					struct priority_group *pg,
					size_t nr_bytes)
L
Linus Torvalds 已提交
358
{
359
	unsigned long flags;
360
	struct dm_path *path;
361
	struct pgpath *pgpath;
L
Linus Torvalds 已提交
362

363
	path = pg->ps.type->select_path(&pg->ps, nr_bytes);
L
Linus Torvalds 已提交
364
	if (!path)
365
		return ERR_PTR(-ENXIO);
L
Linus Torvalds 已提交
366

367
	pgpath = path_to_pgpath(path);
L
Linus Torvalds 已提交
368

369 370 371 372 373 374 375
	if (unlikely(lockless_dereference(m->current_pg) != pg)) {
		/* Only update current_pgpath if pg changed */
		spin_lock_irqsave(&m->lock, flags);
		m->current_pgpath = pgpath;
		__switch_pg(m, pg);
		spin_unlock_irqrestore(&m->lock, flags);
	}
L
Linus Torvalds 已提交
376

377
	return pgpath;
L
Linus Torvalds 已提交
378 379
}

380
static struct pgpath *choose_pgpath(struct multipath *m, size_t nr_bytes)
L
Linus Torvalds 已提交
381
{
382
	unsigned long flags;
L
Linus Torvalds 已提交
383
	struct priority_group *pg;
384
	struct pgpath *pgpath;
385
	unsigned bypassed = 1;
L
Linus Torvalds 已提交
386

387
	if (!atomic_read(&m->nr_valid_paths)) {
388
		clear_bit(MPATHF_QUEUE_IO, &m->flags);
L
Linus Torvalds 已提交
389
		goto failed;
390
	}
L
Linus Torvalds 已提交
391 392

	/* Were we instructed to switch PG? */
393 394
	if (lockless_dereference(m->next_pg)) {
		spin_lock_irqsave(&m->lock, flags);
L
Linus Torvalds 已提交
395
		pg = m->next_pg;
396 397 398 399
		if (!pg) {
			spin_unlock_irqrestore(&m->lock, flags);
			goto check_current_pg;
		}
L
Linus Torvalds 已提交
400
		m->next_pg = NULL;
401 402 403 404
		spin_unlock_irqrestore(&m->lock, flags);
		pgpath = choose_path_in_pg(m, pg, nr_bytes);
		if (!IS_ERR_OR_NULL(pgpath))
			return pgpath;
L
Linus Torvalds 已提交
405 406 407
	}

	/* Don't change PG until it has no remaining paths */
408 409 410 411 412 413 414
check_current_pg:
	pg = lockless_dereference(m->current_pg);
	if (pg) {
		pgpath = choose_path_in_pg(m, pg, nr_bytes);
		if (!IS_ERR_OR_NULL(pgpath))
			return pgpath;
	}
L
Linus Torvalds 已提交
415 416 417 418

	/*
	 * Loop through priority groups until we find a valid path.
	 * First time we skip PGs marked 'bypassed'.
419 420
	 * Second time we only try the ones we skipped, but set
	 * pg_init_delay_retry so we do not hammer controllers.
L
Linus Torvalds 已提交
421 422 423
	 */
	do {
		list_for_each_entry(pg, &m->priority_groups, list) {
424
			if (pg->bypassed == !!bypassed)
L
Linus Torvalds 已提交
425
				continue;
426 427
			pgpath = choose_path_in_pg(m, pg, nr_bytes);
			if (!IS_ERR_OR_NULL(pgpath)) {
428
				if (!bypassed)
429
					set_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags);
430
				return pgpath;
431
			}
L
Linus Torvalds 已提交
432 433 434 435
		}
	} while (bypassed--);

failed:
436
	spin_lock_irqsave(&m->lock, flags);
L
Linus Torvalds 已提交
437 438
	m->current_pgpath = NULL;
	m->current_pg = NULL;
439 440 441
	spin_unlock_irqrestore(&m->lock, flags);

	return NULL;
L
Linus Torvalds 已提交
442 443
}

444
/*
445 446 447
 * dm_report_EIO() is a macro instead of a function to make pr_debug()
 * report the function name and line number of the function from which
 * it has been invoked.
448
 */
449
#define dm_report_EIO(m)						\
450
do {									\
451 452 453 454 455 456 457
	struct mapped_device *md = dm_table_get_md((m)->ti->table);	\
									\
	pr_debug("%s: returning EIO; QIFNP = %d; SQIFNP = %d; DNFS = %d\n", \
		 dm_device_name(md),					\
		 test_bit(MPATHF_QUEUE_IF_NO_PATH, &(m)->flags),	\
		 test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &(m)->flags),	\
		 dm_noflush_suspending((m)->ti));			\
458
} while (0)
459

H
Hannes Reinecke 已提交
460
/*
461
 * Map cloned requests (request-based multipath)
H
Hannes Reinecke 已提交
462
 */
463 464 465
static int multipath_clone_and_map(struct dm_target *ti, struct request *rq,
				   union map_info *map_context,
				   struct request **__clone)
L
Linus Torvalds 已提交
466
{
467
	struct multipath *m = ti->private;
468
	size_t nr_bytes = blk_rq_bytes(rq);
L
Linus Torvalds 已提交
469
	struct pgpath *pgpath;
470
	struct block_device *bdev;
471
	struct dm_mpath_io *mpio = get_mpio(map_context);
472
	struct request_queue *q;
473
	struct request *clone;
L
Linus Torvalds 已提交
474 475

	/* Do we need to select a new pgpath? */
476 477 478
	pgpath = lockless_dereference(m->current_pgpath);
	if (!pgpath || !test_bit(MPATHF_QUEUE_IO, &m->flags))
		pgpath = choose_pgpath(m, nr_bytes);
L
Linus Torvalds 已提交
479

480
	if (!pgpath) {
481
		if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))
482
			return DM_MAPIO_DELAY_REQUEUE;
483
		dm_report_EIO(m);	/* Failed */
484
		return DM_MAPIO_KILL;
485 486
	} else if (test_bit(MPATHF_QUEUE_IO, &m->flags) ||
		   test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) {
487 488
		if (pg_init_all_paths(m))
			return DM_MAPIO_DELAY_REQUEUE;
489
		return DM_MAPIO_REQUEUE;
490
	}
491

492
	memset(mpio, 0, sizeof(*mpio));
493 494 495
	mpio->pgpath = pgpath;
	mpio->nr_bytes = nr_bytes;

496
	bdev = pgpath->path.dev->bdev;
497 498
	q = bdev_get_queue(bdev);
	clone = blk_get_request(q, rq->cmd_flags | REQ_NOMERGE, GFP_ATOMIC);
499 500
	if (IS_ERR(clone)) {
		/* EBUSY, ENODEV or EWOULDBLOCK: requeue */
501 502 503 504 505 506 507 508
		bool queue_dying = blk_queue_dying(q);
		DMERR_LIMIT("blk_get_request() returned %ld%s - requeuing",
			    PTR_ERR(clone), queue_dying ? " (path offline)" : "");
		if (queue_dying) {
			atomic_inc(&m->pg_init_in_progress);
			activate_or_offline_path(pgpath);
			return DM_MAPIO_REQUEUE;
		}
509
		return DM_MAPIO_DELAY_REQUEUE;
510
	}
511 512 513 514
	clone->bio = clone->biotail = NULL;
	clone->rq_disk = bdev->bd_disk;
	clone->cmd_flags |= REQ_FAILFAST_TRANSPORT;
	*__clone = clone;
515

516 517 518 519
	if (pgpath->pg->ps.type->start_io)
		pgpath->pg->ps.type->start_io(&pgpath->pg->ps,
					      &pgpath->path,
					      nr_bytes);
520
	return DM_MAPIO_REMAPPED;
L
Linus Torvalds 已提交
521 522
}

523 524
static void multipath_release_clone(struct request *clone)
{
525
	blk_put_request(clone);
526 527
}

528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558
/*
 * Map cloned bios (bio-based multipath)
 */
static int __multipath_map_bio(struct multipath *m, struct bio *bio, struct dm_mpath_io *mpio)
{
	size_t nr_bytes = bio->bi_iter.bi_size;
	struct pgpath *pgpath;
	unsigned long flags;
	bool queue_io;

	/* Do we need to select a new pgpath? */
	pgpath = lockless_dereference(m->current_pgpath);
	queue_io = test_bit(MPATHF_QUEUE_IO, &m->flags);
	if (!pgpath || !queue_io)
		pgpath = choose_pgpath(m, nr_bytes);

	if ((pgpath && queue_io) ||
	    (!pgpath && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))) {
		/* Queue for the daemon to resubmit */
		spin_lock_irqsave(&m->lock, flags);
		bio_list_add(&m->queued_bios, bio);
		spin_unlock_irqrestore(&m->lock, flags);
		/* PG_INIT_REQUIRED cannot be set without QUEUE_IO */
		if (queue_io || test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags))
			pg_init_all_paths(m);
		else if (!queue_io)
			queue_work(kmultipathd, &m->process_queued_bios);
		return DM_MAPIO_SUBMITTED;
	}

	if (!pgpath) {
559 560
		if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))
			return DM_MAPIO_REQUEUE;
561
		dm_report_EIO(m);
562
		return DM_MAPIO_KILL;
563 564 565 566 567
	}

	mpio->pgpath = pgpath;
	mpio->nr_bytes = nr_bytes;

568
	bio->bi_status = 0;
569
	bio->bi_bdev = pgpath->path.dev->bdev;
J
Jens Axboe 已提交
570
	bio->bi_opf |= REQ_FAILFAST_TRANSPORT;
571 572 573 574 575 576 577 578 579 580 581

	if (pgpath->pg->ps.type->start_io)
		pgpath->pg->ps.type->start_io(&pgpath->pg->ps,
					      &pgpath->path,
					      nr_bytes);
	return DM_MAPIO_REMAPPED;
}

static int multipath_map_bio(struct dm_target *ti, struct bio *bio)
{
	struct multipath *m = ti->private;
582 583 584
	struct dm_mpath_io *mpio = NULL;

	multipath_init_per_bio_data(bio, &mpio, NULL);
585 586 587 588

	return __multipath_map_bio(m, bio, mpio);
}

589
static void process_queued_io_list(struct multipath *m)
590
{
591 592 593
	if (m->queue_mode == DM_TYPE_MQ_REQUEST_BASED)
		dm_mq_kick_requeue_list(dm_table_get_md(m->ti->table));
	else if (m->queue_mode == DM_TYPE_BIO_BASED)
594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623
		queue_work(kmultipathd, &m->process_queued_bios);
}

static void process_queued_bios(struct work_struct *work)
{
	int r;
	unsigned long flags;
	struct bio *bio;
	struct bio_list bios;
	struct blk_plug plug;
	struct multipath *m =
		container_of(work, struct multipath, process_queued_bios);

	bio_list_init(&bios);

	spin_lock_irqsave(&m->lock, flags);

	if (bio_list_empty(&m->queued_bios)) {
		spin_unlock_irqrestore(&m->lock, flags);
		return;
	}

	bio_list_merge(&bios, &m->queued_bios);
	bio_list_init(&m->queued_bios);

	spin_unlock_irqrestore(&m->lock, flags);

	blk_start_plug(&plug);
	while ((bio = bio_list_pop(&bios))) {
		r = __multipath_map_bio(m, bio, get_mpio_from_bio(bio));
624 625
		switch (r) {
		case DM_MAPIO_KILL:
626 627
			bio->bi_status = BLK_STS_IOERR;
			bio_endio(bio);
628
		case DM_MAPIO_REQUEUE:
629
			bio->bi_status = BLK_STS_DM_REQUEUE;
630
			bio_endio(bio);
631 632
			break;
		case DM_MAPIO_REMAPPED:
633
			generic_make_request(bio);
634 635
			break;
		}
636 637 638 639
	}
	blk_finish_plug(&plug);
}

640 641 642 643 644 645 646 647
static void assign_bit(bool value, long nr, unsigned long *addr)
{
	if (value)
		set_bit(nr, addr);
	else
		clear_bit(nr, addr);
}

L
Linus Torvalds 已提交
648 649 650
/*
 * If we run out of usable paths, should we queue I/O or error it?
 */
651 652
static int queue_if_no_path(struct multipath *m, bool queue_if_no_path,
			    bool save_old_value)
L
Linus Torvalds 已提交
653 654 655 656
{
	unsigned long flags;

	spin_lock_irqsave(&m->lock, flags);
657 658 659 660 661
	assign_bit((save_old_value && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) ||
		   (!save_old_value && queue_if_no_path),
		   MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags);
	assign_bit(queue_if_no_path || dm_noflush_suspending(m->ti),
		   MPATHF_QUEUE_IF_NO_PATH, &m->flags);
L
Linus Torvalds 已提交
662 663
	spin_unlock_irqrestore(&m->lock, flags);

664
	if (!queue_if_no_path) {
665
		dm_table_run_md_queue_async(m->ti->table);
666
		process_queued_io_list(m);
667
	}
668

L
Linus Torvalds 已提交
669 670 671 672 673 674 675
	return 0;
}

/*
 * An event is triggered whenever a path is taken out of use.
 * Includes path failure and PG bypass.
 */
D
David Howells 已提交
676
static void trigger_event(struct work_struct *work)
L
Linus Torvalds 已提交
677
{
D
David Howells 已提交
678 679
	struct multipath *m =
		container_of(work, struct multipath, trigger_event);
L
Linus Torvalds 已提交
680 681 682 683 684 685 686 687 688 689 690 691 692 693

	dm_table_event(m->ti->table);
}

/*-----------------------------------------------------------------
 * Constructor/argument parsing:
 * <#multipath feature args> [<arg>]*
 * <#hw_handler args> [hw_handler [<arg>]*]
 * <#priority groups>
 * <initial priority group>
 *     [<selector> <#selector args> [<arg>]*
 *      <#paths> <#per-path selector args>
 *         [<path> [<arg>]* ]+ ]+
 *---------------------------------------------------------------*/
694
static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg,
L
Linus Torvalds 已提交
695 696 697 698 699 700
			       struct dm_target *ti)
{
	int r;
	struct path_selector_type *pst;
	unsigned ps_argc;

701
	static struct dm_arg _args[] = {
702
		{0, 1024, "invalid number of path selector args"},
L
Linus Torvalds 已提交
703 704
	};

705
	pst = dm_get_path_selector(dm_shift_arg(as));
L
Linus Torvalds 已提交
706
	if (!pst) {
707
		ti->error = "unknown path selector type";
L
Linus Torvalds 已提交
708 709 710
		return -EINVAL;
	}

711
	r = dm_read_arg_group(_args, as, &ps_argc, &ti->error);
712 713
	if (r) {
		dm_put_path_selector(pst);
L
Linus Torvalds 已提交
714
		return -EINVAL;
715
	}
L
Linus Torvalds 已提交
716 717 718 719

	r = pst->create(&pg->ps, ps_argc, as->argv);
	if (r) {
		dm_put_path_selector(pst);
720
		ti->error = "path selector constructor failed";
L
Linus Torvalds 已提交
721 722 723 724
		return r;
	}

	pg->ps.type = pst;
725
	dm_consume_args(as, ps_argc);
L
Linus Torvalds 已提交
726 727 728 729

	return 0;
}

730
static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps,
L
Linus Torvalds 已提交
731 732 733 734
			       struct dm_target *ti)
{
	int r;
	struct pgpath *p;
735
	struct multipath *m = ti->private;
736 737
	struct request_queue *q = NULL;
	const char *attached_handler_name;
L
Linus Torvalds 已提交
738 739 740

	/* we need at least a path arg */
	if (as->argc < 1) {
741
		ti->error = "no device given";
742
		return ERR_PTR(-EINVAL);
L
Linus Torvalds 已提交
743 744 745 746
	}

	p = alloc_pgpath();
	if (!p)
747
		return ERR_PTR(-ENOMEM);
L
Linus Torvalds 已提交
748

749
	r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table),
750
			  &p->path.dev);
L
Linus Torvalds 已提交
751
	if (r) {
752
		ti->error = "error getting device";
L
Linus Torvalds 已提交
753 754 755
		goto bad;
	}

756
	if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags) || m->hw_handler_name)
757 758
		q = bdev_get_queue(p->path.dev->bdev);

759
	if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags)) {
760
retain:
761 762
		attached_handler_name = scsi_dh_attached_handler_name(q, GFP_KERNEL);
		if (attached_handler_name) {
763 764 765 766 767 768 769 770 771
			/*
			 * Clear any hw_handler_params associated with a
			 * handler that isn't already attached.
			 */
			if (m->hw_handler_name && strcmp(attached_handler_name, m->hw_handler_name)) {
				kfree(m->hw_handler_params);
				m->hw_handler_params = NULL;
			}

772 773 774 775 776 777 778 779 780 781
			/*
			 * Reset hw_handler_name to match the attached handler
			 *
			 * NB. This modifies the table line to show the actual
			 * handler instead of the original table passed in.
			 */
			kfree(m->hw_handler_name);
			m->hw_handler_name = attached_handler_name;
		}
	}
782

783
	if (m->hw_handler_name) {
784 785
		r = scsi_dh_attach(q, m->hw_handler_name);
		if (r == -EBUSY) {
786
			char b[BDEVNAME_SIZE];
787

788 789 790 791
			printk(KERN_INFO "dm-mpath: retaining handler on device %s\n",
				bdevname(p->path.dev->bdev, b));
			goto retain;
		}
792
		if (r < 0) {
793
			ti->error = "error attaching hardware handler";
794 795 796
			dm_put_device(ti, p->path.dev);
			goto bad;
		}
797 798 799 800 801 802 803 804 805 806

		if (m->hw_handler_params) {
			r = scsi_dh_set_params(q, m->hw_handler_params);
			if (r < 0) {
				ti->error = "unable to set hardware "
							"handler parameters";
				dm_put_device(ti, p->path.dev);
				goto bad;
			}
		}
807 808
	}

L
Linus Torvalds 已提交
809 810 811 812 813 814 815 816 817 818
	r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error);
	if (r) {
		dm_put_device(ti, p->path.dev);
		goto bad;
	}

	return p;

 bad:
	free_pgpath(p);
819
	return ERR_PTR(r);
L
Linus Torvalds 已提交
820 821
}

822
static struct priority_group *parse_priority_group(struct dm_arg_set *as,
M
Micha³ Miros³aw 已提交
823
						   struct multipath *m)
L
Linus Torvalds 已提交
824
{
825
	static struct dm_arg _args[] = {
826 827
		{1, 1024, "invalid number of paths"},
		{0, 1024, "invalid number of selector args"}
L
Linus Torvalds 已提交
828 829 830
	};

	int r;
831
	unsigned i, nr_selector_args, nr_args;
L
Linus Torvalds 已提交
832
	struct priority_group *pg;
M
Micha³ Miros³aw 已提交
833
	struct dm_target *ti = m->ti;
L
Linus Torvalds 已提交
834 835 836

	if (as->argc < 2) {
		as->argc = 0;
837 838
		ti->error = "not enough priority group arguments";
		return ERR_PTR(-EINVAL);
L
Linus Torvalds 已提交
839 840 841 842
	}

	pg = alloc_priority_group();
	if (!pg) {
843
		ti->error = "couldn't allocate priority group";
844
		return ERR_PTR(-ENOMEM);
L
Linus Torvalds 已提交
845 846 847 848 849 850 851 852 853 854
	}
	pg->m = m;

	r = parse_path_selector(as, pg, ti);
	if (r)
		goto bad;

	/*
	 * read the paths
	 */
855
	r = dm_read_arg(_args, as, &pg->nr_pgpaths, &ti->error);
L
Linus Torvalds 已提交
856 857 858
	if (r)
		goto bad;

859
	r = dm_read_arg(_args + 1, as, &nr_selector_args, &ti->error);
L
Linus Torvalds 已提交
860 861 862
	if (r)
		goto bad;

863
	nr_args = 1 + nr_selector_args;
L
Linus Torvalds 已提交
864 865
	for (i = 0; i < pg->nr_pgpaths; i++) {
		struct pgpath *pgpath;
866
		struct dm_arg_set path_args;
L
Linus Torvalds 已提交
867

868
		if (as->argc < nr_args) {
869
			ti->error = "not enough path parameters";
870
			r = -EINVAL;
L
Linus Torvalds 已提交
871
			goto bad;
872
		}
L
Linus Torvalds 已提交
873

874
		path_args.argc = nr_args;
L
Linus Torvalds 已提交
875 876 877
		path_args.argv = as->argv;

		pgpath = parse_path(&path_args, &pg->ps, ti);
878 879
		if (IS_ERR(pgpath)) {
			r = PTR_ERR(pgpath);
L
Linus Torvalds 已提交
880
			goto bad;
881
		}
L
Linus Torvalds 已提交
882 883 884

		pgpath->pg = pg;
		list_add_tail(&pgpath->list, &pg->pgpaths);
885
		dm_consume_args(as, nr_args);
L
Linus Torvalds 已提交
886 887 888 889 890 891
	}

	return pg;

 bad:
	free_priority_group(pg, ti);
892
	return ERR_PTR(r);
L
Linus Torvalds 已提交
893 894
}

895
static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m)
L
Linus Torvalds 已提交
896 897
{
	unsigned hw_argc;
898
	int ret;
M
Micha³ Miros³aw 已提交
899
	struct dm_target *ti = m->ti;
L
Linus Torvalds 已提交
900

901
	static struct dm_arg _args[] = {
902
		{0, 1024, "invalid number of hardware handler args"},
L
Linus Torvalds 已提交
903 904
	};

905
	if (dm_read_arg_group(_args, as, &hw_argc, &ti->error))
L
Linus Torvalds 已提交
906 907 908 909 910
		return -EINVAL;

	if (!hw_argc)
		return 0;

911
	if (m->queue_mode == DM_TYPE_BIO_BASED) {
912 913 914 915 916
		dm_consume_args(as, hw_argc);
		DMERR("bio-based multipath doesn't allow hardware handler args");
		return 0;
	}

917
	m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL);
918 919
	if (!m->hw_handler_name)
		return -EINVAL;
920

921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936
	if (hw_argc > 1) {
		char *p;
		int i, j, len = 4;

		for (i = 0; i <= hw_argc - 2; i++)
			len += strlen(as->argv[i]) + 1;
		p = m->hw_handler_params = kzalloc(len, GFP_KERNEL);
		if (!p) {
			ti->error = "memory allocation failed";
			ret = -ENOMEM;
			goto fail;
		}
		j = sprintf(p, "%d", hw_argc - 1);
		for (i = 0, p+=j+1; i <= hw_argc - 2; i++, p+=j+1)
			j = sprintf(p, "%s", as->argv[i]);
	}
937
	dm_consume_args(as, hw_argc - 1);
L
Linus Torvalds 已提交
938 939

	return 0;
940 941 942 943
fail:
	kfree(m->hw_handler_name);
	m->hw_handler_name = NULL;
	return ret;
L
Linus Torvalds 已提交
944 945
}

946
static int parse_features(struct dm_arg_set *as, struct multipath *m)
L
Linus Torvalds 已提交
947 948 949
{
	int r;
	unsigned argc;
M
Micha³ Miros³aw 已提交
950
	struct dm_target *ti = m->ti;
951
	const char *arg_name;
L
Linus Torvalds 已提交
952

953
	static struct dm_arg _args[] = {
954
		{0, 8, "invalid number of feature args"},
D
Dave Wysochanski 已提交
955
		{1, 50, "pg_init_retries must be between 1 and 50"},
956
		{0, 60000, "pg_init_delay_msecs must be between 0 and 60000"},
L
Linus Torvalds 已提交
957 958
	};

959
	r = dm_read_arg_group(_args, as, &argc, &ti->error);
L
Linus Torvalds 已提交
960 961 962 963 964 965
	if (r)
		return -EINVAL;

	if (!argc)
		return 0;

D
Dave Wysochanski 已提交
966
	do {
967
		arg_name = dm_shift_arg(as);
D
Dave Wysochanski 已提交
968 969
		argc--;

970
		if (!strcasecmp(arg_name, "queue_if_no_path")) {
971
			r = queue_if_no_path(m, true, false);
D
Dave Wysochanski 已提交
972 973 974
			continue;
		}

975
		if (!strcasecmp(arg_name, "retain_attached_hw_handler")) {
976
			set_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags);
977 978 979
			continue;
		}

980
		if (!strcasecmp(arg_name, "pg_init_retries") &&
D
Dave Wysochanski 已提交
981
		    (argc >= 1)) {
982
			r = dm_read_arg(_args + 1, as, &m->pg_init_retries, &ti->error);
D
Dave Wysochanski 已提交
983 984 985 986
			argc--;
			continue;
		}

987
		if (!strcasecmp(arg_name, "pg_init_delay_msecs") &&
988
		    (argc >= 1)) {
989
			r = dm_read_arg(_args + 2, as, &m->pg_init_delay_msecs, &ti->error);
990 991 992 993
			argc--;
			continue;
		}

994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011
		if (!strcasecmp(arg_name, "queue_mode") &&
		    (argc >= 1)) {
			const char *queue_mode_name = dm_shift_arg(as);

			if (!strcasecmp(queue_mode_name, "bio"))
				m->queue_mode = DM_TYPE_BIO_BASED;
			else if (!strcasecmp(queue_mode_name, "rq"))
				m->queue_mode = DM_TYPE_REQUEST_BASED;
			else if (!strcasecmp(queue_mode_name, "mq"))
				m->queue_mode = DM_TYPE_MQ_REQUEST_BASED;
			else {
				ti->error = "Unknown 'queue_mode' requested";
				r = -EINVAL;
			}
			argc--;
			continue;
		}

L
Linus Torvalds 已提交
1012
		ti->error = "Unrecognised multipath feature request";
D
Dave Wysochanski 已提交
1013 1014 1015 1016
		r = -EINVAL;
	} while (argc && !r);

	return r;
L
Linus Torvalds 已提交
1017 1018
}

1019
static int multipath_ctr(struct dm_target *ti, unsigned argc, char **argv)
L
Linus Torvalds 已提交
1020
{
1021 1022
	/* target arguments */
	static struct dm_arg _args[] = {
1023 1024
		{0, 1024, "invalid number of priority groups"},
		{0, 1024, "invalid initial priority group number"},
L
Linus Torvalds 已提交
1025 1026 1027 1028
	};

	int r;
	struct multipath *m;
1029
	struct dm_arg_set as;
L
Linus Torvalds 已提交
1030 1031 1032 1033 1034 1035
	unsigned pg_count = 0;
	unsigned next_pg_num;

	as.argc = argc;
	as.argv = argv;

1036
	m = alloc_multipath(ti);
L
Linus Torvalds 已提交
1037
	if (!m) {
1038
		ti->error = "can't allocate multipath";
L
Linus Torvalds 已提交
1039 1040 1041
		return -EINVAL;
	}

M
Micha³ Miros³aw 已提交
1042
	r = parse_features(&as, m);
L
Linus Torvalds 已提交
1043 1044 1045
	if (r)
		goto bad;

1046 1047 1048 1049
	r = alloc_multipath_stage2(ti, m);
	if (r)
		goto bad;

M
Micha³ Miros³aw 已提交
1050
	r = parse_hw_handler(&as, m);
L
Linus Torvalds 已提交
1051 1052 1053
	if (r)
		goto bad;

1054
	r = dm_read_arg(_args, &as, &m->nr_priority_groups, &ti->error);
L
Linus Torvalds 已提交
1055 1056 1057
	if (r)
		goto bad;

1058
	r = dm_read_arg(_args + 1, &as, &next_pg_num, &ti->error);
L
Linus Torvalds 已提交
1059 1060 1061
	if (r)
		goto bad;

1062 1063 1064 1065 1066 1067 1068
	if ((!m->nr_priority_groups && next_pg_num) ||
	    (m->nr_priority_groups && !next_pg_num)) {
		ti->error = "invalid initial priority group";
		r = -EINVAL;
		goto bad;
	}

L
Linus Torvalds 已提交
1069 1070 1071
	/* parse the priority groups */
	while (as.argc) {
		struct priority_group *pg;
1072
		unsigned nr_valid_paths = atomic_read(&m->nr_valid_paths);
L
Linus Torvalds 已提交
1073

M
Micha³ Miros³aw 已提交
1074
		pg = parse_priority_group(&as, m);
1075 1076
		if (IS_ERR(pg)) {
			r = PTR_ERR(pg);
L
Linus Torvalds 已提交
1077 1078 1079
			goto bad;
		}

1080 1081 1082
		nr_valid_paths += pg->nr_pgpaths;
		atomic_set(&m->nr_valid_paths, nr_valid_paths);

L
Linus Torvalds 已提交
1083 1084 1085 1086 1087 1088 1089 1090
		list_add_tail(&pg->list, &m->priority_groups);
		pg_count++;
		pg->pg_num = pg_count;
		if (!--next_pg_num)
			m->next_pg = pg;
	}

	if (pg_count != m->nr_priority_groups) {
1091
		ti->error = "priority group count mismatch";
L
Linus Torvalds 已提交
1092 1093 1094 1095
		r = -EINVAL;
		goto bad;
	}

1096 1097
	ti->num_flush_bios = 1;
	ti->num_discard_bios = 1;
1098
	ti->num_write_same_bios = 1;
1099
	ti->num_write_zeroes_bios = 1;
1100
	if (m->queue_mode == DM_TYPE_BIO_BASED)
1101
		ti->per_io_data_size = multipath_per_bio_data_size();
1102
	else
1103
		ti->per_io_data_size = sizeof(struct dm_mpath_io);
M
Mikulas Patocka 已提交
1104

L
Linus Torvalds 已提交
1105 1106 1107 1108 1109 1110 1111
	return 0;

 bad:
	free_multipath(m);
	return r;
}

1112 1113
static void multipath_wait_for_pg_init_completion(struct multipath *m)
{
1114
	DEFINE_WAIT(wait);
1115 1116

	while (1) {
1117
		prepare_to_wait(&m->pg_init_wait, &wait, TASK_UNINTERRUPTIBLE);
1118

1119
		if (!atomic_read(&m->pg_init_in_progress))
1120 1121 1122 1123
			break;

		io_schedule();
	}
1124
	finish_wait(&m->pg_init_wait, &wait);
1125 1126 1127
}

static void flush_multipath_work(struct multipath *m)
L
Linus Torvalds 已提交
1128
{
1129 1130
	set_bit(MPATHF_PG_INIT_DISABLED, &m->flags);
	smp_mb__after_atomic();
1131

1132
	flush_workqueue(kmpath_handlerd);
1133
	multipath_wait_for_pg_init_completion(m);
1134
	flush_workqueue(kmultipathd);
1135
	flush_work(&m->trigger_event);
1136

1137 1138
	clear_bit(MPATHF_PG_INIT_DISABLED, &m->flags);
	smp_mb__after_atomic();
1139 1140 1141 1142 1143 1144
}

static void multipath_dtr(struct dm_target *ti)
{
	struct multipath *m = ti->private;

1145
	flush_multipath_work(m);
L
Linus Torvalds 已提交
1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158
	free_multipath(m);
}

/*
 * Take a path out of use.
 */
static int fail_path(struct pgpath *pgpath)
{
	unsigned long flags;
	struct multipath *m = pgpath->pg->m;

	spin_lock_irqsave(&m->lock, flags);

1159
	if (!pgpath->is_active)
L
Linus Torvalds 已提交
1160 1161
		goto out;

1162
	DMWARN("Failing path %s.", pgpath->path.dev->name);
L
Linus Torvalds 已提交
1163 1164

	pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path);
1165
	pgpath->is_active = false;
L
Linus Torvalds 已提交
1166 1167
	pgpath->fail_count++;

1168
	atomic_dec(&m->nr_valid_paths);
L
Linus Torvalds 已提交
1169 1170 1171 1172

	if (pgpath == m->current_pgpath)
		m->current_pgpath = NULL;

M
Mike Anderson 已提交
1173
	dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti,
1174
		       pgpath->path.dev->name, atomic_read(&m->nr_valid_paths));
M
Mike Anderson 已提交
1175

1176
	schedule_work(&m->trigger_event);
L
Linus Torvalds 已提交
1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188

out:
	spin_unlock_irqrestore(&m->lock, flags);

	return 0;
}

/*
 * Reinstate a previously-failed path
 */
static int reinstate_path(struct pgpath *pgpath)
{
1189
	int r = 0, run_queue = 0;
L
Linus Torvalds 已提交
1190 1191
	unsigned long flags;
	struct multipath *m = pgpath->pg->m;
1192
	unsigned nr_valid_paths;
L
Linus Torvalds 已提交
1193 1194 1195

	spin_lock_irqsave(&m->lock, flags);

1196
	if (pgpath->is_active)
L
Linus Torvalds 已提交
1197 1198
		goto out;

1199
	DMWARN("Reinstating path %s.", pgpath->path.dev->name);
L
Linus Torvalds 已提交
1200 1201 1202 1203 1204

	r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path);
	if (r)
		goto out;

1205
	pgpath->is_active = true;
L
Linus Torvalds 已提交
1206

1207 1208
	nr_valid_paths = atomic_inc_return(&m->nr_valid_paths);
	if (nr_valid_paths == 1) {
1209
		m->current_pgpath = NULL;
1210
		run_queue = 1;
1211
	} else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) {
1212
		if (queue_work(kmpath_handlerd, &pgpath->activate_path.work))
1213
			atomic_inc(&m->pg_init_in_progress);
1214
	}
L
Linus Torvalds 已提交
1215

M
Mike Anderson 已提交
1216
	dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti,
1217
		       pgpath->path.dev->name, nr_valid_paths);
M
Mike Anderson 已提交
1218

1219
	schedule_work(&m->trigger_event);
L
Linus Torvalds 已提交
1220 1221 1222

out:
	spin_unlock_irqrestore(&m->lock, flags);
1223
	if (run_queue) {
1224
		dm_table_run_md_queue_async(m->ti->table);
1225
		process_queued_io_list(m);
1226
	}
L
Linus Torvalds 已提交
1227 1228 1229 1230 1231 1232 1233 1234 1235 1236

	return r;
}

/*
 * Fail or reinstate all paths that match the provided struct dm_dev.
 */
static int action_dev(struct multipath *m, struct dm_dev *dev,
		      action_fn action)
{
1237
	int r = -EINVAL;
L
Linus Torvalds 已提交
1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254
	struct pgpath *pgpath;
	struct priority_group *pg;

	list_for_each_entry(pg, &m->priority_groups, list) {
		list_for_each_entry(pgpath, &pg->pgpaths, list) {
			if (pgpath->path.dev == dev)
				r = action(pgpath);
		}
	}

	return r;
}

/*
 * Temporarily try to avoid having to use the specified PG
 */
static void bypass_pg(struct multipath *m, struct priority_group *pg,
1255
		      bool bypassed)
L
Linus Torvalds 已提交
1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266
{
	unsigned long flags;

	spin_lock_irqsave(&m->lock, flags);

	pg->bypassed = bypassed;
	m->current_pgpath = NULL;
	m->current_pg = NULL;

	spin_unlock_irqrestore(&m->lock, flags);

1267
	schedule_work(&m->trigger_event);
L
Linus Torvalds 已提交
1268 1269 1270 1271 1272 1273 1274 1275 1276 1277
}

/*
 * Switch to using the specified PG from the next I/O that gets mapped
 */
static int switch_pg_num(struct multipath *m, const char *pgstr)
{
	struct priority_group *pg;
	unsigned pgnum;
	unsigned long flags;
1278
	char dummy;
L
Linus Torvalds 已提交
1279

1280
	if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum ||
1281
	    !m->nr_priority_groups || (pgnum > m->nr_priority_groups)) {
L
Linus Torvalds 已提交
1282 1283 1284 1285 1286 1287
		DMWARN("invalid PG number supplied to switch_pg_num");
		return -EINVAL;
	}

	spin_lock_irqsave(&m->lock, flags);
	list_for_each_entry(pg, &m->priority_groups, list) {
1288
		pg->bypassed = false;
L
Linus Torvalds 已提交
1289 1290 1291 1292 1293 1294 1295 1296 1297
		if (--pgnum)
			continue;

		m->current_pgpath = NULL;
		m->current_pg = NULL;
		m->next_pg = pg;
	}
	spin_unlock_irqrestore(&m->lock, flags);

1298
	schedule_work(&m->trigger_event);
L
Linus Torvalds 已提交
1299 1300 1301 1302 1303 1304 1305
	return 0;
}

/*
 * Set/clear bypassed status of a PG.
 * PGs are numbered upwards from 1 in the order they were declared.
 */
1306
static int bypass_pg_num(struct multipath *m, const char *pgstr, bool bypassed)
L
Linus Torvalds 已提交
1307 1308 1309
{
	struct priority_group *pg;
	unsigned pgnum;
1310
	char dummy;
L
Linus Torvalds 已提交
1311

1312
	if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum ||
1313
	    !m->nr_priority_groups || (pgnum > m->nr_priority_groups)) {
L
Linus Torvalds 已提交
1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326
		DMWARN("invalid PG number supplied to bypass_pg");
		return -EINVAL;
	}

	list_for_each_entry(pg, &m->priority_groups, list) {
		if (!--pgnum)
			break;
	}

	bypass_pg(m, pg, bypassed);
	return 0;
}

D
Dave Wysochanski 已提交
1327 1328 1329
/*
 * Should we retry pg_init immediately?
 */
1330
static bool pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath)
D
Dave Wysochanski 已提交
1331 1332
{
	unsigned long flags;
1333
	bool limit_reached = false;
D
Dave Wysochanski 已提交
1334 1335 1336

	spin_lock_irqsave(&m->lock, flags);

1337 1338
	if (atomic_read(&m->pg_init_count) <= m->pg_init_retries &&
	    !test_bit(MPATHF_PG_INIT_DISABLED, &m->flags))
1339
		set_bit(MPATHF_PG_INIT_REQUIRED, &m->flags);
D
Dave Wysochanski 已提交
1340
	else
1341
		limit_reached = true;
D
Dave Wysochanski 已提交
1342 1343 1344 1345 1346 1347

	spin_unlock_irqrestore(&m->lock, flags);

	return limit_reached;
}

1348
static void pg_init_done(void *data, int errors)
1349
{
1350
	struct pgpath *pgpath = data;
1351 1352 1353
	struct priority_group *pg = pgpath->pg;
	struct multipath *m = pg->m;
	unsigned long flags;
1354
	bool delay_retry = false;
1355 1356 1357 1358 1359 1360 1361 1362 1363 1364

	/* device or driver problems */
	switch (errors) {
	case SCSI_DH_OK:
		break;
	case SCSI_DH_NOSYS:
		if (!m->hw_handler_name) {
			errors = 0;
			break;
		}
1365 1366
		DMERR("Could not failover the device: Handler scsi_dh_%s "
		      "Error %d.", m->hw_handler_name, errors);
1367 1368 1369 1370 1371 1372 1373 1374 1375 1376
		/*
		 * Fail path for now, so we do not ping pong
		 */
		fail_path(pgpath);
		break;
	case SCSI_DH_DEV_TEMP_BUSY:
		/*
		 * Probably doing something like FW upgrade on the
		 * controller so try the other pg.
		 */
1377
		bypass_pg(m, pg, true);
1378 1379
		break;
	case SCSI_DH_RETRY:
1380 1381
		/* Wait before retrying. */
		delay_retry = 1;
1382 1383 1384 1385 1386 1387
	case SCSI_DH_IMM_RETRY:
	case SCSI_DH_RES_TEMP_UNAVAIL:
		if (pg_init_limit_reached(m, pgpath))
			fail_path(pgpath);
		errors = 0;
		break;
1388
	case SCSI_DH_DEV_OFFLINED:
1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399
	default:
		/*
		 * We probably do not want to fail the path for a device
		 * error, but this is what the old dm did. In future
		 * patches we can do more advanced handling.
		 */
		fail_path(pgpath);
	}

	spin_lock_irqsave(&m->lock, flags);
	if (errors) {
1400 1401 1402 1403 1404
		if (pgpath == m->current_pgpath) {
			DMERR("Could not failover device. Error %d.", errors);
			m->current_pgpath = NULL;
			m->current_pg = NULL;
		}
1405
	} else if (!test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags))
1406
		pg->bypassed = false;
1407

1408
	if (atomic_dec_return(&m->pg_init_in_progress) > 0)
1409 1410 1411
		/* Activations of other paths are still on going */
		goto out;

1412 1413 1414 1415 1416 1417
	if (test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) {
		if (delay_retry)
			set_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags);
		else
			clear_bit(MPATHF_PG_INIT_DELAY_RETRY, &m->flags);

1418 1419 1420
		if (__pg_init_all_paths(m))
			goto out;
	}
1421
	clear_bit(MPATHF_QUEUE_IO, &m->flags);
1422

1423
	process_queued_io_list(m);
1424

1425 1426 1427 1428 1429
	/*
	 * Wake up any thread waiting to suspend.
	 */
	wake_up(&m->pg_init_wait);

1430
out:
1431 1432 1433
	spin_unlock_irqrestore(&m->lock, flags);
}

1434
static void activate_or_offline_path(struct pgpath *pgpath)
1435
{
1436
	struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev);
1437

1438 1439
	if (pgpath->is_active && !blk_queue_dying(q))
		scsi_dh_activate(q, pg_init_done, pgpath);
1440 1441
	else
		pg_init_done(pgpath, SCSI_DH_DEV_OFFLINED);
1442 1443
}

1444 1445 1446 1447 1448 1449 1450 1451
static void activate_path_work(struct work_struct *work)
{
	struct pgpath *pgpath =
		container_of(work, struct pgpath, activate_path.work);

	activate_or_offline_path(pgpath);
}

1452
static int noretry_error(blk_status_t error)
1453 1454
{
	switch (error) {
1455 1456 1457 1458 1459 1460
	case BLK_STS_NOTSUPP:
	case BLK_STS_NOSPC:
	case BLK_STS_TARGET:
	case BLK_STS_NEXUS:
	case BLK_STS_MEDIUM:
	case BLK_STS_RESOURCE:
1461 1462 1463 1464 1465 1466 1467
		return 1;
	}

	/* Anything else could be a path failure, so should be retried */
	return 0;
}

1468
static int multipath_end_io(struct dm_target *ti, struct request *clone,
1469
			    blk_status_t error, union map_info *map_context)
L
Linus Torvalds 已提交
1470
{
1471 1472
	struct dm_mpath_io *mpio = get_mpio(map_context);
	struct pgpath *pgpath = mpio->pgpath;
1473
	int r = DM_ENDIO_DONE;
1474

1475 1476 1477 1478 1479
	/*
	 * We don't queue any clone request inside the multipath target
	 * during end I/O handling, since those clone requests don't have
	 * bio clones.  If we queue them inside the multipath target,
	 * we need to make bio clones, that requires memory allocation.
1480
	 * (See drivers/md/dm-rq.c:end_clone_bio() about why the clone requests
1481 1482 1483 1484 1485
	 *  don't have bio clones.)
	 * Instead of queueing the clone request here, we queue the original
	 * request into dm core, which will remake a clone request and
	 * clone bios for it and resubmit it later.
	 */
1486 1487
	if (error && !noretry_error(error)) {
		struct multipath *m = ti->private;
L
Linus Torvalds 已提交
1488

1489
		r = DM_ENDIO_REQUEUE;
L
Linus Torvalds 已提交
1490

1491 1492
		if (pgpath)
			fail_path(pgpath);
L
Linus Torvalds 已提交
1493

1494
		if (atomic_read(&m->nr_valid_paths) == 0 &&
1495
		    !test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) {
1496
			if (error == BLK_STS_IOERR)
1497
				dm_report_EIO(m);
1498 1499 1500
			/* complete with the original error */
			r = DM_ENDIO_DONE;
		}
1501
	}
1502

L
Linus Torvalds 已提交
1503
	if (pgpath) {
1504 1505
		struct path_selector *ps = &pgpath->pg->ps;

L
Linus Torvalds 已提交
1506
		if (ps->type->end_io)
1507
			ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes);
L
Linus Torvalds 已提交
1508 1509
	}

1510
	return r;
L
Linus Torvalds 已提交
1511 1512
}

1513 1514
static int multipath_end_io_bio(struct dm_target *ti, struct bio *clone,
		blk_status_t *error)
1515
{
1516 1517 1518
	struct multipath *m = ti->private;
	struct dm_mpath_io *mpio = get_mpio_from_bio(clone);
	struct pgpath *pgpath = mpio->pgpath;
1519
	unsigned long flags;
1520
	int r = DM_ENDIO_DONE;
1521

1522
	if (!*error || noretry_error(*error))
1523
		goto done;
1524

1525 1526
	if (pgpath)
		fail_path(pgpath);
1527

1528
	if (atomic_read(&m->nr_valid_paths) == 0 &&
1529 1530
	    !test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) {
		dm_report_EIO(m);
1531
		*error = BLK_STS_IOERR;
1532
		goto done;
1533
	}
1534 1535

	/* Queue for the daemon to resubmit */
1536
	dm_bio_restore(get_bio_details_from_bio(clone), clone);
1537 1538 1539 1540 1541 1542 1543

	spin_lock_irqsave(&m->lock, flags);
	bio_list_add(&m->queued_bios, clone);
	spin_unlock_irqrestore(&m->lock, flags);
	if (!test_bit(MPATHF_QUEUE_IO, &m->flags))
		queue_work(kmultipathd, &m->process_queued_bios);

1544
	r = DM_ENDIO_INCOMPLETE;
1545
done:
1546
	if (pgpath) {
1547 1548
		struct path_selector *ps = &pgpath->pg->ps;

1549 1550 1551 1552
		if (ps->type->end_io)
			ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes);
	}

1553
	return r;
1554 1555
}

L
Linus Torvalds 已提交
1556 1557
/*
 * Suspend can't complete until all the I/O is processed so if
1558 1559 1560
 * the last path fails we must error any remaining I/O.
 * Note that if the freeze_bdev fails while suspending, the
 * queue_if_no_path state is lost - userspace should reset it.
L
Linus Torvalds 已提交
1561 1562 1563
 */
static void multipath_presuspend(struct dm_target *ti)
{
1564
	struct multipath *m = ti->private;
L
Linus Torvalds 已提交
1565

1566
	queue_if_no_path(m, false, true);
L
Linus Torvalds 已提交
1567 1568
}

1569 1570
static void multipath_postsuspend(struct dm_target *ti)
{
1571 1572 1573
	struct multipath *m = ti->private;

	mutex_lock(&m->work_mutex);
1574
	flush_multipath_work(m);
1575
	mutex_unlock(&m->work_mutex);
1576 1577
}

1578 1579 1580
/*
 * Restore the queue_if_no_path setting.
 */
L
Linus Torvalds 已提交
1581 1582
static void multipath_resume(struct dm_target *ti)
{
1583
	struct multipath *m = ti->private;
1584
	unsigned long flags;
L
Linus Torvalds 已提交
1585

1586
	spin_lock_irqsave(&m->lock, flags);
1587 1588
	assign_bit(test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags),
		   MPATHF_QUEUE_IF_NO_PATH, &m->flags);
1589
	spin_unlock_irqrestore(&m->lock, flags);
L
Linus Torvalds 已提交
1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607
}

/*
 * Info output has the following format:
 * num_multipath_feature_args [multipath_feature_args]*
 * num_handler_status_args [handler_status_args]*
 * num_groups init_group_number
 *            [A|D|E num_ps_status_args [ps_status_args]*
 *             num_paths num_selector_args
 *             [path_dev A|F fail_count [selector_args]* ]+ ]+
 *
 * Table output has the following format (identical to the constructor string):
 * num_feature_args [features_args]*
 * num_handler_args hw_handler [hw_handler_args]*
 * num_groups init_group_number
 *     [priority selector-name num_ps_args [ps_args]*
 *      num_paths num_selector_args [path_dev [selector_args]* ]+ ]+
 */
1608 1609
static void multipath_status(struct dm_target *ti, status_type_t type,
			     unsigned status_flags, char *result, unsigned maxlen)
L
Linus Torvalds 已提交
1610 1611 1612
{
	int sz = 0;
	unsigned long flags;
1613
	struct multipath *m = ti->private;
L
Linus Torvalds 已提交
1614 1615 1616 1617 1618 1619 1620 1621 1622
	struct priority_group *pg;
	struct pgpath *p;
	unsigned pg_num;
	char state;

	spin_lock_irqsave(&m->lock, flags);

	/* Features */
	if (type == STATUSTYPE_INFO)
1623 1624
		DMEMIT("2 %u %u ", test_bit(MPATHF_QUEUE_IO, &m->flags),
		       atomic_read(&m->pg_init_count));
D
Dave Wysochanski 已提交
1625
	else {
1626
		DMEMIT("%u ", test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) +
1627
			      (m->pg_init_retries > 0) * 2 +
1628
			      (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2 +
1629 1630 1631
			      test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags) +
			      (m->queue_mode != DM_TYPE_REQUEST_BASED) * 2);

1632
		if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))
D
Dave Wysochanski 已提交
1633 1634 1635
			DMEMIT("queue_if_no_path ");
		if (m->pg_init_retries)
			DMEMIT("pg_init_retries %u ", m->pg_init_retries);
1636 1637
		if (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT)
			DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs);
1638
		if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags))
1639
			DMEMIT("retain_attached_hw_handler ");
1640 1641 1642 1643 1644 1645 1646 1647
		if (m->queue_mode != DM_TYPE_REQUEST_BASED) {
			switch(m->queue_mode) {
			case DM_TYPE_BIO_BASED:
				DMEMIT("queue_mode bio ");
				break;
			case DM_TYPE_MQ_REQUEST_BASED:
				DMEMIT("queue_mode mq ");
				break;
1648 1649 1650
			default:
				WARN_ON_ONCE(true);
				break;
1651 1652
			}
		}
D
Dave Wysochanski 已提交
1653
	}
L
Linus Torvalds 已提交
1654

1655
	if (!m->hw_handler_name || type == STATUSTYPE_INFO)
L
Linus Torvalds 已提交
1656 1657
		DMEMIT("0 ");
	else
1658
		DMEMIT("1 %s ", m->hw_handler_name);
L
Linus Torvalds 已提交
1659 1660 1661 1662 1663 1664 1665 1666

	DMEMIT("%u ", m->nr_priority_groups);

	if (m->next_pg)
		pg_num = m->next_pg->pg_num;
	else if (m->current_pg)
		pg_num = m->current_pg->pg_num;
	else
1667
		pg_num = (m->nr_priority_groups ? 1 : 0);
L
Linus Torvalds 已提交
1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694

	DMEMIT("%u ", pg_num);

	switch (type) {
	case STATUSTYPE_INFO:
		list_for_each_entry(pg, &m->priority_groups, list) {
			if (pg->bypassed)
				state = 'D';	/* Disabled */
			else if (pg == m->current_pg)
				state = 'A';	/* Currently Active */
			else
				state = 'E';	/* Enabled */

			DMEMIT("%c ", state);

			if (pg->ps.type->status)
				sz += pg->ps.type->status(&pg->ps, NULL, type,
							  result + sz,
							  maxlen - sz);
			else
				DMEMIT("0 ");

			DMEMIT("%u %u ", pg->nr_pgpaths,
			       pg->ps.type->info_args);

			list_for_each_entry(p, &pg->pgpaths, list) {
				DMEMIT("%s %s %u ", p->path.dev->name,
1695
				       p->is_active ? "A" : "F",
L
Linus Torvalds 已提交
1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734
				       p->fail_count);
				if (pg->ps.type->status)
					sz += pg->ps.type->status(&pg->ps,
					      &p->path, type, result + sz,
					      maxlen - sz);
			}
		}
		break;

	case STATUSTYPE_TABLE:
		list_for_each_entry(pg, &m->priority_groups, list) {
			DMEMIT("%s ", pg->ps.type->name);

			if (pg->ps.type->status)
				sz += pg->ps.type->status(&pg->ps, NULL, type,
							  result + sz,
							  maxlen - sz);
			else
				DMEMIT("0 ");

			DMEMIT("%u %u ", pg->nr_pgpaths,
			       pg->ps.type->table_args);

			list_for_each_entry(p, &pg->pgpaths, list) {
				DMEMIT("%s ", p->path.dev->name);
				if (pg->ps.type->status)
					sz += pg->ps.type->status(&pg->ps,
					      &p->path, type, result + sz,
					      maxlen - sz);
			}
		}
		break;
	}

	spin_unlock_irqrestore(&m->lock, flags);
}

static int multipath_message(struct dm_target *ti, unsigned argc, char **argv)
{
1735
	int r = -EINVAL;
L
Linus Torvalds 已提交
1736
	struct dm_dev *dev;
1737
	struct multipath *m = ti->private;
L
Linus Torvalds 已提交
1738 1739
	action_fn action;

1740 1741
	mutex_lock(&m->work_mutex);

1742 1743 1744 1745 1746
	if (dm_suspended(ti)) {
		r = -EBUSY;
		goto out;
	}

L
Linus Torvalds 已提交
1747
	if (argc == 1) {
1748
		if (!strcasecmp(argv[0], "queue_if_no_path")) {
1749
			r = queue_if_no_path(m, true, false);
1750
			goto out;
1751
		} else if (!strcasecmp(argv[0], "fail_if_no_path")) {
1752
			r = queue_if_no_path(m, false, false);
1753 1754
			goto out;
		}
L
Linus Torvalds 已提交
1755 1756
	}

1757
	if (argc != 2) {
1758
		DMWARN("Invalid multipath message arguments. Expected 2 arguments, got %d.", argc);
1759 1760
		goto out;
	}
L
Linus Torvalds 已提交
1761

1762
	if (!strcasecmp(argv[0], "disable_group")) {
1763
		r = bypass_pg_num(m, argv[1], true);
1764
		goto out;
1765
	} else if (!strcasecmp(argv[0], "enable_group")) {
1766
		r = bypass_pg_num(m, argv[1], false);
1767
		goto out;
1768
	} else if (!strcasecmp(argv[0], "switch_group")) {
1769 1770
		r = switch_pg_num(m, argv[1]);
		goto out;
1771
	} else if (!strcasecmp(argv[0], "reinstate_path"))
L
Linus Torvalds 已提交
1772
		action = reinstate_path;
1773
	else if (!strcasecmp(argv[0], "fail_path"))
L
Linus Torvalds 已提交
1774
		action = fail_path;
1775
	else {
1776
		DMWARN("Unrecognised multipath message received: %s", argv[0]);
1777 1778
		goto out;
	}
L
Linus Torvalds 已提交
1779

1780
	r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev);
L
Linus Torvalds 已提交
1781
	if (r) {
1782
		DMWARN("message: error getting device %s",
L
Linus Torvalds 已提交
1783
		       argv[1]);
1784
		goto out;
L
Linus Torvalds 已提交
1785 1786 1787 1788 1789 1790
	}

	r = action_dev(m, dev, action);

	dm_put_device(ti, dev);

1791 1792
out:
	mutex_unlock(&m->work_mutex);
L
Linus Torvalds 已提交
1793 1794 1795
	return r;
}

C
Christoph Hellwig 已提交
1796 1797
static int multipath_prepare_ioctl(struct dm_target *ti,
		struct block_device **bdev, fmode_t *mode)
M
Milan Broz 已提交
1798
{
1799
	struct multipath *m = ti->private;
1800
	struct pgpath *current_pgpath;
1801 1802
	int r;

1803 1804 1805
	current_pgpath = lockless_dereference(m->current_pgpath);
	if (!current_pgpath)
		current_pgpath = choose_pgpath(m, 0);
M
Milan Broz 已提交
1806

1807
	if (current_pgpath) {
1808
		if (!test_bit(MPATHF_QUEUE_IO, &m->flags)) {
1809 1810
			*bdev = current_pgpath->path.dev->bdev;
			*mode = current_pgpath->path.dev->mode;
1811 1812 1813 1814 1815 1816 1817
			r = 0;
		} else {
			/* pg_init has not started or completed */
			r = -ENOTCONN;
		}
	} else {
		/* No path is available */
1818
		if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))
1819 1820 1821
			r = -ENOTCONN;
		else
			r = -EIO;
1822
	}
M
Milan Broz 已提交
1823

1824
	if (r == -ENOTCONN) {
1825
		if (!lockless_dereference(m->current_pg)) {
1826
			/* Path status changed, redo selection */
1827
			(void) choose_pgpath(m, 0);
1828
		}
1829
		if (test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags))
1830
			pg_init_all_paths(m);
1831
		dm_table_run_md_queue_async(m->ti->table);
1832
		process_queued_io_list(m);
1833
	}
1834

C
Christoph Hellwig 已提交
1835 1836 1837 1838 1839 1840
	/*
	 * Only pass ioctls through if the device sizes match exactly.
	 */
	if (!r && ti->len != i_size_read((*bdev)->bd_inode) >> SECTOR_SHIFT)
		return 1;
	return r;
M
Milan Broz 已提交
1841 1842
}

1843 1844 1845 1846 1847 1848 1849 1850 1851 1852
static int multipath_iterate_devices(struct dm_target *ti,
				     iterate_devices_callout_fn fn, void *data)
{
	struct multipath *m = ti->private;
	struct priority_group *pg;
	struct pgpath *p;
	int ret = 0;

	list_for_each_entry(pg, &m->priority_groups, list) {
		list_for_each_entry(p, &pg->pgpaths, list) {
1853
			ret = fn(ti, p->path.dev, ti->begin, ti->len, data);
1854 1855 1856 1857 1858 1859 1860 1861 1862
			if (ret)
				goto out;
		}
	}

out:
	return ret;
}

1863
static int pgpath_busy(struct pgpath *pgpath)
1864 1865 1866
{
	struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev);

1867
	return blk_lld_busy(q);
1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879
}

/*
 * We return "busy", only when we can map I/Os but underlying devices
 * are busy (so even if we map I/Os now, the I/Os will wait on
 * the underlying queue).
 * In other words, if we want to kill I/Os or queue them inside us
 * due to map unavailability, we don't return "busy".  Otherwise,
 * dm core won't give us the I/Os and we can't do what we want.
 */
static int multipath_busy(struct dm_target *ti)
{
1880
	bool busy = false, has_active = false;
1881
	struct multipath *m = ti->private;
1882
	struct priority_group *pg, *next_pg;
1883 1884
	struct pgpath *pgpath;

1885 1886
	/* pg_init in progress */
	if (atomic_read(&m->pg_init_in_progress))
1887 1888
		return true;

1889 1890 1891 1892
	/* no paths available, for blk-mq: rely on IO mapping to delay requeue */
	if (!atomic_read(&m->nr_valid_paths) && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))
		return (m->queue_mode != DM_TYPE_MQ_REQUEST_BASED);

1893
	/* Guess which priority_group will be used at next mapping time */
1894 1895 1896 1897 1898 1899
	pg = lockless_dereference(m->current_pg);
	next_pg = lockless_dereference(m->next_pg);
	if (unlikely(!lockless_dereference(m->current_pgpath) && next_pg))
		pg = next_pg;

	if (!pg) {
1900 1901
		/*
		 * We don't know which pg will be used at next mapping time.
1902
		 * We don't call choose_pgpath() here to avoid to trigger
1903 1904 1905 1906
		 * pg_init just by busy checking.
		 * So we don't know whether underlying devices we will be using
		 * at next mapping time are busy or not. Just try mapping.
		 */
1907 1908
		return busy;
	}
1909 1910 1911 1912 1913

	/*
	 * If there is one non-busy active path at least, the path selector
	 * will be able to select it. So we consider such a pg as not busy.
	 */
1914
	busy = true;
1915
	list_for_each_entry(pgpath, &pg->pgpaths, list) {
1916
		if (pgpath->is_active) {
1917
			has_active = true;
1918
			if (!pgpath_busy(pgpath)) {
1919
				busy = false;
1920 1921 1922
				break;
			}
		}
1923
	}
1924

1925
	if (!has_active) {
1926 1927 1928 1929 1930
		/*
		 * No active path in this pg, so this pg won't be used and
		 * the current_pg will be changed at next mapping time.
		 * We need to try mapping to determine it.
		 */
1931
		busy = false;
1932
	}
1933 1934 1935 1936

	return busy;
}

L
Linus Torvalds 已提交
1937 1938 1939 1940 1941
/*-----------------------------------------------------------------
 * Module setup
 *---------------------------------------------------------------*/
static struct target_type multipath_target = {
	.name = "multipath",
1942
	.version = {1, 12, 0},
M
Mike Snitzer 已提交
1943
	.features = DM_TARGET_SINGLETON | DM_TARGET_IMMUTABLE,
L
Linus Torvalds 已提交
1944 1945 1946
	.module = THIS_MODULE,
	.ctr = multipath_ctr,
	.dtr = multipath_dtr,
1947 1948
	.clone_and_map_rq = multipath_clone_and_map,
	.release_clone_rq = multipath_release_clone,
1949
	.rq_end_io = multipath_end_io,
1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961
	.map = multipath_map_bio,
	.end_io = multipath_end_io_bio,
	.presuspend = multipath_presuspend,
	.postsuspend = multipath_postsuspend,
	.resume = multipath_resume,
	.status = multipath_status,
	.message = multipath_message,
	.prepare_ioctl = multipath_prepare_ioctl,
	.iterate_devices = multipath_iterate_devices,
	.busy = multipath_busy,
};

L
Linus Torvalds 已提交
1962 1963 1964 1965 1966 1967
static int __init dm_multipath_init(void)
{
	int r;

	r = dm_register_target(&multipath_target);
	if (r < 0) {
1968
		DMERR("request-based register failed %d", r);
1969 1970
		r = -EINVAL;
		goto bad_register_target;
L
Linus Torvalds 已提交
1971 1972
	}

1973
	kmultipathd = alloc_workqueue("kmpathd", WQ_MEM_RECLAIM, 0);
1974
	if (!kmultipathd) {
1975
		DMERR("failed to create workqueue kmpathd");
1976 1977
		r = -ENOMEM;
		goto bad_alloc_kmultipathd;
1978 1979
	}

1980 1981 1982 1983 1984 1985
	/*
	 * A separate workqueue is used to handle the device handlers
	 * to avoid overloading existing workqueue. Overloading the
	 * old workqueue would also create a bottleneck in the
	 * path of the storage hardware device activation.
	 */
1986 1987
	kmpath_handlerd = alloc_ordered_workqueue("kmpath_handlerd",
						  WQ_MEM_RECLAIM);
1988 1989
	if (!kmpath_handlerd) {
		DMERR("failed to create workqueue kmpath_handlerd");
1990 1991
		r = -ENOMEM;
		goto bad_alloc_kmpath_handlerd;
1992 1993
	}

1994 1995 1996 1997 1998 1999 2000
	return 0;

bad_alloc_kmpath_handlerd:
	destroy_workqueue(kmultipathd);
bad_alloc_kmultipathd:
	dm_unregister_target(&multipath_target);
bad_register_target:
L
Linus Torvalds 已提交
2001 2002 2003 2004 2005
	return r;
}

static void __exit dm_multipath_exit(void)
{
2006
	destroy_workqueue(kmpath_handlerd);
2007 2008
	destroy_workqueue(kmultipathd);

2009
	dm_unregister_target(&multipath_target);
L
Linus Torvalds 已提交
2010 2011 2012 2013 2014 2015 2016 2017
}

module_init(dm_multipath_init);
module_exit(dm_multipath_exit);

MODULE_DESCRIPTION(DM_NAME " multipath target");
MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>");
MODULE_LICENSE("GPL");