dm-mpath.c 39.9 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.h"
L
Linus Torvalds 已提交
11
#include "dm-path-selector.h"
M
Mike Anderson 已提交
12
#include "dm-uevent.h"
L
Linus Torvalds 已提交
13 14 15 16 17 18 19 20 21

#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>
22
#include <linux/delay.h>
23
#include <scsi/scsi_dh.h>
A
Arun Sharma 已提交
24
#include <linux/atomic.h>
L
Linus Torvalds 已提交
25

26
#define DM_MSG_PREFIX "multipath"
27 28
#define DM_PG_INIT_DELAY_MSECS 2000
#define DM_PG_INIT_DELAY_DEFAULT ((unsigned) -1)
L
Linus Torvalds 已提交
29 30 31 32 33 34

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

	struct priority_group *pg;	/* Owning PG */
35
	unsigned is_active;		/* Path status */
L
Linus Torvalds 已提交
36 37
	unsigned fail_count;		/* Cumulative failure count */

38
	struct dm_path path;
39
	struct delayed_work activate_path;
L
Linus Torvalds 已提交
40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
};

#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 bypassed;		/* Temporarily bypass this PG? */

	unsigned nr_pgpaths;		/* Number of paths in PG */
	struct list_head pgpaths;
};

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

66
	const char *hw_handler_name;
67
	char *hw_handler_params;
68

69 70
	spinlock_t lock;

L
Linus Torvalds 已提交
71 72
	unsigned nr_priority_groups;
	struct list_head priority_groups;
73 74 75

	wait_queue_head_t pg_init_wait;	/* Wait for pg_init completion */

L
Linus Torvalds 已提交
76
	unsigned pg_init_required;	/* pg_init needs calling? */
77
	unsigned pg_init_in_progress;	/* Only one pg_init allowed at once */
78
	unsigned pg_init_delay_retry;	/* Delay pg_init retry? */
L
Linus Torvalds 已提交
79 80 81 82 83 84 85

	unsigned nr_valid_paths;	/* Total number of usable paths */
	struct pgpath *current_pgpath;
	struct priority_group *current_pg;
	struct priority_group *next_pg;	/* Switch to this PG if set */
	unsigned repeat_count;		/* I/Os left before calling PS again */

86 87 88
	unsigned queue_io:1;		/* Must we queue all I/O? */
	unsigned queue_if_no_path:1;	/* Queue I/O if last path fails? */
	unsigned saved_queue_if_no_path:1; /* Saved state during suspension */
89
	unsigned retain_attached_hw_handler:1; /* If there's already a hw_handler present, don't change it. */
90
	unsigned pg_init_disabled:1;	/* pg_init is not currently allowed */
91

D
Dave Wysochanski 已提交
92 93
	unsigned pg_init_retries;	/* Number of times to retry pg_init */
	unsigned pg_init_count;		/* Number of times pg_init called */
94
	unsigned pg_init_delay_msecs;	/* Number of msecs before pg_init retry */
L
Linus Torvalds 已提交
95 96 97 98

	struct work_struct trigger_event;

	/*
A
Alasdair G Kergon 已提交
99
	 * We must use a mempool of dm_mpath_io structs so that we
L
Linus Torvalds 已提交
100 101 102
	 * can resubmit bios on error.
	 */
	mempool_t *mpio_pool;
103 104

	struct mutex work_mutex;
L
Linus Torvalds 已提交
105 106 107 108 109
};

/*
 * Context information attached to each bio we process.
 */
A
Alasdair G Kergon 已提交
110
struct dm_mpath_io {
L
Linus Torvalds 已提交
111
	struct pgpath *pgpath;
112
	size_t nr_bytes;
L
Linus Torvalds 已提交
113 114 115 116
};

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

117
static struct kmem_cache *_mpio_cache;
L
Linus Torvalds 已提交
118

119
static struct workqueue_struct *kmultipathd, *kmpath_handlerd;
D
David Howells 已提交
120
static void trigger_event(struct work_struct *work);
121
static void activate_path(struct work_struct *work);
122
static int __pgpath_busy(struct pgpath *pgpath);
L
Linus Torvalds 已提交
123 124 125 126 127 128 129 130


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

static struct pgpath *alloc_pgpath(void)
{
M
Micha³ Miros³aw 已提交
131
	struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL);
L
Linus Torvalds 已提交
132

133
	if (pgpath) {
134
		pgpath->is_active = 1;
135
		INIT_DELAYED_WORK(&pgpath->activate_path, activate_path);
136
	}
L
Linus Torvalds 已提交
137 138 139 140

	return pgpath;
}

A
Alasdair G Kergon 已提交
141
static void free_pgpath(struct pgpath *pgpath)
L
Linus Torvalds 已提交
142 143 144 145 146 147 148 149
{
	kfree(pgpath);
}

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

M
Micha³ Miros³aw 已提交
150
	pg = kzalloc(sizeof(*pg), GFP_KERNEL);
L
Linus Torvalds 已提交
151

M
Micha³ Miros³aw 已提交
152 153
	if (pg)
		INIT_LIST_HEAD(&pg->pgpaths);
L
Linus Torvalds 已提交
154 155 156 157 158 159 160

	return pg;
}

static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti)
{
	struct pgpath *pgpath, *tmp;
161
	struct multipath *m = ti->private;
L
Linus Torvalds 已提交
162 163 164

	list_for_each_entry_safe(pgpath, tmp, pgpaths, list) {
		list_del(&pgpath->list);
165 166
		if (m->hw_handler_name)
			scsi_dh_detach(bdev_get_queue(pgpath->path.dev->bdev));
L
Linus Torvalds 已提交
167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185
		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);
}

M
Micha³ Miros³aw 已提交
186
static struct multipath *alloc_multipath(struct dm_target *ti)
L
Linus Torvalds 已提交
187 188
{
	struct multipath *m;
189
	unsigned min_ios = dm_get_reserved_rq_based_ios();
L
Linus Torvalds 已提交
190

M
Micha³ Miros³aw 已提交
191
	m = kzalloc(sizeof(*m), GFP_KERNEL);
L
Linus Torvalds 已提交
192 193 194 195
	if (m) {
		INIT_LIST_HEAD(&m->priority_groups);
		spin_lock_init(&m->lock);
		m->queue_io = 1;
196
		m->pg_init_delay_msecs = DM_PG_INIT_DELAY_DEFAULT;
D
David Howells 已提交
197
		INIT_WORK(&m->trigger_event, trigger_event);
198
		init_waitqueue_head(&m->pg_init_wait);
199
		mutex_init(&m->work_mutex);
200
		m->mpio_pool = mempool_create_slab_pool(min_ios, _mpio_cache);
L
Linus Torvalds 已提交
201 202 203 204
		if (!m->mpio_pool) {
			kfree(m);
			return NULL;
		}
M
Micha³ Miros³aw 已提交
205 206
		m->ti = ti;
		ti->private = m;
L
Linus Torvalds 已提交
207 208 209 210 211 212 213 214 215 216 217 218 219 220
	}

	return m;
}

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);
	}

221
	kfree(m->hw_handler_name);
222
	kfree(m->hw_handler_params);
L
Linus Torvalds 已提交
223 224 225 226
	mempool_destroy(m->mpio_pool);
	kfree(m);
}

227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247
static int set_mapinfo(struct multipath *m, union map_info *info)
{
	struct dm_mpath_io *mpio;

	mpio = mempool_alloc(m->mpio_pool, GFP_ATOMIC);
	if (!mpio)
		return -ENOMEM;

	memset(mpio, 0, sizeof(*mpio));
	info->ptr = mpio;

	return 0;
}

static void clear_mapinfo(struct multipath *m, union map_info *info)
{
	struct dm_mpath_io *mpio = info->ptr;

	info->ptr = NULL;
	mempool_free(mpio, m->mpio_pool);
}
L
Linus Torvalds 已提交
248 249 250 251 252

/*-----------------------------------------------
 * Path selection
 *-----------------------------------------------*/

253
static int __pg_init_all_paths(struct multipath *m)
K
Kiyoshi Ueda 已提交
254 255
{
	struct pgpath *pgpath;
256
	unsigned long pg_init_delay = 0;
K
Kiyoshi Ueda 已提交
257

258
	if (m->pg_init_in_progress || m->pg_init_disabled)
259
		return 0;
260

K
Kiyoshi Ueda 已提交
261 262
	m->pg_init_count++;
	m->pg_init_required = 0;
263 264 265 266 267

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

268 269 270
	if (m->pg_init_delay_retry)
		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
271 272 273 274
	list_for_each_entry(pgpath, &m->current_pg->pgpaths, list) {
		/* Skip failed paths */
		if (!pgpath->is_active)
			continue;
275 276
		if (queue_delayed_work(kmpath_handlerd, &pgpath->activate_path,
				       pg_init_delay))
K
Kiyoshi Ueda 已提交
277 278
			m->pg_init_in_progress++;
	}
279
	return m->pg_init_in_progress;
K
Kiyoshi Ueda 已提交
280 281
}

L
Linus Torvalds 已提交
282 283 284 285 286
static void __switch_pg(struct multipath *m, struct pgpath *pgpath)
{
	m->current_pg = pgpath->pg;

	/* Must we initialise the PG first, and queue I/O till it's ready? */
287
	if (m->hw_handler_name) {
L
Linus Torvalds 已提交
288 289 290 291 292 293
		m->pg_init_required = 1;
		m->queue_io = 1;
	} else {
		m->pg_init_required = 0;
		m->queue_io = 0;
	}
D
Dave Wysochanski 已提交
294 295

	m->pg_init_count = 0;
L
Linus Torvalds 已提交
296 297
}

298 299
static int __choose_path_in_pg(struct multipath *m, struct priority_group *pg,
			       size_t nr_bytes)
L
Linus Torvalds 已提交
300
{
301
	struct dm_path *path;
L
Linus Torvalds 已提交
302

303
	path = pg->ps.type->select_path(&pg->ps, &m->repeat_count, nr_bytes);
L
Linus Torvalds 已提交
304 305 306 307 308 309 310 311 312 313 314
	if (!path)
		return -ENXIO;

	m->current_pgpath = path_to_pgpath(path);

	if (m->current_pg != pg)
		__switch_pg(m, m->current_pgpath);

	return 0;
}

315
static void __choose_pgpath(struct multipath *m, size_t nr_bytes)
L
Linus Torvalds 已提交
316 317 318 319
{
	struct priority_group *pg;
	unsigned bypassed = 1;

320 321
	if (!m->nr_valid_paths) {
		m->queue_io = 0;
L
Linus Torvalds 已提交
322
		goto failed;
323
	}
L
Linus Torvalds 已提交
324 325 326 327 328

	/* Were we instructed to switch PG? */
	if (m->next_pg) {
		pg = m->next_pg;
		m->next_pg = NULL;
329
		if (!__choose_path_in_pg(m, pg, nr_bytes))
L
Linus Torvalds 已提交
330 331 332 333
			return;
	}

	/* Don't change PG until it has no remaining paths */
334
	if (m->current_pg && !__choose_path_in_pg(m, m->current_pg, nr_bytes))
L
Linus Torvalds 已提交
335 336 337 338 339
		return;

	/*
	 * Loop through priority groups until we find a valid path.
	 * First time we skip PGs marked 'bypassed'.
340 341
	 * 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
342 343 344 345 346
	 */
	do {
		list_for_each_entry(pg, &m->priority_groups, list) {
			if (pg->bypassed == bypassed)
				continue;
347 348 349
			if (!__choose_path_in_pg(m, pg, nr_bytes)) {
				if (!bypassed)
					m->pg_init_delay_retry = 1;
L
Linus Torvalds 已提交
350
				return;
351
			}
L
Linus Torvalds 已提交
352 353 354 355 356 357 358 359
		}
	} while (bypassed--);

failed:
	m->current_pgpath = NULL;
	m->current_pg = NULL;
}

360 361 362 363 364 365 366 367 368 369 370 371 372
/*
 * Check whether bios must be queued in the device-mapper core rather
 * than here in the target.
 *
 * m->lock must be held on entry.
 *
 * If m->queue_if_no_path and m->saved_queue_if_no_path hold the
 * same value then we are not between multipath_presuspend()
 * and multipath_resume() calls and we have no need to check
 * for the DMF_NOFLUSH_SUSPENDING flag.
 */
static int __must_push_back(struct multipath *m)
{
373 374 375
	return (m->queue_if_no_path ||
		(m->queue_if_no_path != m->saved_queue_if_no_path &&
		 dm_noflush_suspending(m->ti)));
376 377
}

H
Hannes Reinecke 已提交
378 379 380 381 382
/*
 * Map cloned requests
 */
static int multipath_map(struct dm_target *ti, struct request *clone,
			 union map_info *map_context)
L
Linus Torvalds 已提交
383
{
H
Hannes Reinecke 已提交
384
	struct multipath *m = (struct multipath *) ti->private;
385
	int r = DM_MAPIO_REQUEUE;
386
	size_t nr_bytes = blk_rq_bytes(clone);
L
Linus Torvalds 已提交
387 388
	unsigned long flags;
	struct pgpath *pgpath;
389
	struct block_device *bdev;
390
	struct dm_mpath_io *mpio;
L
Linus Torvalds 已提交
391 392 393 394 395 396

	spin_lock_irqsave(&m->lock, flags);

	/* Do we need to select a new pgpath? */
	if (!m->current_pgpath ||
	    (!m->queue_io && (m->repeat_count && --m->repeat_count == 0)))
397
		__choose_pgpath(m, nr_bytes);
L
Linus Torvalds 已提交
398 399 400

	pgpath = m->current_pgpath;

401 402 403 404
	if (!pgpath) {
		if (!__must_push_back(m))
			r = -EIO;	/* Failed */
		goto out_unlock;
405
	} else if (m->queue_io || m->pg_init_required) {
406
		__pg_init_all_paths(m);
407 408
		goto out_unlock;
	}
409

410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425
	if (set_mapinfo(m, map_context) < 0)
		/* ENOMEM, requeue */
		goto out_unlock;

	bdev = pgpath->path.dev->bdev;
	clone->q = bdev_get_queue(bdev);
	clone->rq_disk = bdev->bd_disk;
	clone->cmd_flags |= REQ_FAILFAST_TRANSPORT;
	mpio = map_context->ptr;
	mpio->pgpath = pgpath;
	mpio->nr_bytes = nr_bytes;
	if (pgpath->pg->ps.type->start_io)
		pgpath->pg->ps.type->start_io(&pgpath->pg->ps,
					      &pgpath->path,
					      nr_bytes);
	r = DM_MAPIO_REMAPPED;
L
Linus Torvalds 已提交
426

427
out_unlock:
L
Linus Torvalds 已提交
428 429 430 431 432 433 434 435
	spin_unlock_irqrestore(&m->lock, flags);

	return r;
}

/*
 * If we run out of usable paths, should we queue I/O or error it?
 */
436 437
static int queue_if_no_path(struct multipath *m, unsigned queue_if_no_path,
			    unsigned save_old_value)
L
Linus Torvalds 已提交
438 439 440 441 442
{
	unsigned long flags;

	spin_lock_irqsave(&m->lock, flags);

443 444 445 446
	if (save_old_value)
		m->saved_queue_if_no_path = m->queue_if_no_path;
	else
		m->saved_queue_if_no_path = queue_if_no_path;
L
Linus Torvalds 已提交
447 448 449
	m->queue_if_no_path = queue_if_no_path;
	spin_unlock_irqrestore(&m->lock, flags);

450 451 452
	if (!queue_if_no_path)
		dm_table_run_md_queue_async(m->ti->table);

L
Linus Torvalds 已提交
453 454 455 456 457 458 459
	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
460
static void trigger_event(struct work_struct *work)
L
Linus Torvalds 已提交
461
{
D
David Howells 已提交
462 463
	struct multipath *m =
		container_of(work, struct multipath, trigger_event);
L
Linus Torvalds 已提交
464 465 466 467 468 469 470 471 472 473 474 475 476 477

	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>]* ]+ ]+
 *---------------------------------------------------------------*/
478
static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg,
L
Linus Torvalds 已提交
479 480 481 482 483 484
			       struct dm_target *ti)
{
	int r;
	struct path_selector_type *pst;
	unsigned ps_argc;

485
	static struct dm_arg _args[] = {
486
		{0, 1024, "invalid number of path selector args"},
L
Linus Torvalds 已提交
487 488
	};

489
	pst = dm_get_path_selector(dm_shift_arg(as));
L
Linus Torvalds 已提交
490
	if (!pst) {
491
		ti->error = "unknown path selector type";
L
Linus Torvalds 已提交
492 493 494
		return -EINVAL;
	}

495
	r = dm_read_arg_group(_args, as, &ps_argc, &ti->error);
496 497
	if (r) {
		dm_put_path_selector(pst);
L
Linus Torvalds 已提交
498
		return -EINVAL;
499
	}
L
Linus Torvalds 已提交
500 501 502 503

	r = pst->create(&pg->ps, ps_argc, as->argv);
	if (r) {
		dm_put_path_selector(pst);
504
		ti->error = "path selector constructor failed";
L
Linus Torvalds 已提交
505 506 507 508
		return r;
	}

	pg->ps.type = pst;
509
	dm_consume_args(as, ps_argc);
L
Linus Torvalds 已提交
510 511 512 513

	return 0;
}

514
static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps,
L
Linus Torvalds 已提交
515 516 517 518
			       struct dm_target *ti)
{
	int r;
	struct pgpath *p;
519
	struct multipath *m = ti->private;
520 521
	struct request_queue *q = NULL;
	const char *attached_handler_name;
L
Linus Torvalds 已提交
522 523 524

	/* we need at least a path arg */
	if (as->argc < 1) {
525
		ti->error = "no device given";
526
		return ERR_PTR(-EINVAL);
L
Linus Torvalds 已提交
527 528 529 530
	}

	p = alloc_pgpath();
	if (!p)
531
		return ERR_PTR(-ENOMEM);
L
Linus Torvalds 已提交
532

533
	r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table),
534
			  &p->path.dev);
L
Linus Torvalds 已提交
535
	if (r) {
536
		ti->error = "error getting device";
L
Linus Torvalds 已提交
537 538 539
		goto bad;
	}

540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560
	if (m->retain_attached_hw_handler || m->hw_handler_name)
		q = bdev_get_queue(p->path.dev->bdev);

	if (m->retain_attached_hw_handler) {
		attached_handler_name = scsi_dh_attached_handler_name(q, GFP_KERNEL);
		if (attached_handler_name) {
			/*
			 * Reset hw_handler_name to match the attached handler
			 * and clear any hw_handler_params associated with the
			 * ignored 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;

			kfree(m->hw_handler_params);
			m->hw_handler_params = NULL;
		}
	}
561

562 563 564 565 566
	if (m->hw_handler_name) {
		/*
		 * Increments scsi_dh reference, even when using an
		 * already-attached handler.
		 */
567 568 569
		r = scsi_dh_attach(q, m->hw_handler_name);
		if (r == -EBUSY) {
			/*
570
			 * Already attached to different hw_handler:
571 572 573 574 575 576
			 * try to reattach with correct one.
			 */
			scsi_dh_detach(q);
			r = scsi_dh_attach(q, m->hw_handler_name);
		}

577
		if (r < 0) {
578
			ti->error = "error attaching hardware handler";
579 580 581
			dm_put_device(ti, p->path.dev);
			goto bad;
		}
582 583 584 585 586 587 588 589 590 591 592

		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";
				scsi_dh_detach(q);
				dm_put_device(ti, p->path.dev);
				goto bad;
			}
		}
593 594
	}

L
Linus Torvalds 已提交
595 596 597 598 599 600 601 602 603 604
	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);
605
	return ERR_PTR(r);
L
Linus Torvalds 已提交
606 607
}

608
static struct priority_group *parse_priority_group(struct dm_arg_set *as,
M
Micha³ Miros³aw 已提交
609
						   struct multipath *m)
L
Linus Torvalds 已提交
610
{
611
	static struct dm_arg _args[] = {
612 613
		{1, 1024, "invalid number of paths"},
		{0, 1024, "invalid number of selector args"}
L
Linus Torvalds 已提交
614 615 616
	};

	int r;
617
	unsigned i, nr_selector_args, nr_args;
L
Linus Torvalds 已提交
618
	struct priority_group *pg;
M
Micha³ Miros³aw 已提交
619
	struct dm_target *ti = m->ti;
L
Linus Torvalds 已提交
620 621 622

	if (as->argc < 2) {
		as->argc = 0;
623 624
		ti->error = "not enough priority group arguments";
		return ERR_PTR(-EINVAL);
L
Linus Torvalds 已提交
625 626 627 628
	}

	pg = alloc_priority_group();
	if (!pg) {
629
		ti->error = "couldn't allocate priority group";
630
		return ERR_PTR(-ENOMEM);
L
Linus Torvalds 已提交
631 632 633 634 635 636 637 638 639 640
	}
	pg->m = m;

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

	/*
	 * read the paths
	 */
641
	r = dm_read_arg(_args, as, &pg->nr_pgpaths, &ti->error);
L
Linus Torvalds 已提交
642 643 644
	if (r)
		goto bad;

645
	r = dm_read_arg(_args + 1, as, &nr_selector_args, &ti->error);
L
Linus Torvalds 已提交
646 647 648
	if (r)
		goto bad;

649
	nr_args = 1 + nr_selector_args;
L
Linus Torvalds 已提交
650 651
	for (i = 0; i < pg->nr_pgpaths; i++) {
		struct pgpath *pgpath;
652
		struct dm_arg_set path_args;
L
Linus Torvalds 已提交
653

654
		if (as->argc < nr_args) {
655
			ti->error = "not enough path parameters";
656
			r = -EINVAL;
L
Linus Torvalds 已提交
657
			goto bad;
658
		}
L
Linus Torvalds 已提交
659

660
		path_args.argc = nr_args;
L
Linus Torvalds 已提交
661 662 663
		path_args.argv = as->argv;

		pgpath = parse_path(&path_args, &pg->ps, ti);
664 665
		if (IS_ERR(pgpath)) {
			r = PTR_ERR(pgpath);
L
Linus Torvalds 已提交
666
			goto bad;
667
		}
L
Linus Torvalds 已提交
668 669 670

		pgpath->pg = pg;
		list_add_tail(&pgpath->list, &pg->pgpaths);
671
		dm_consume_args(as, nr_args);
L
Linus Torvalds 已提交
672 673 674 675 676 677
	}

	return pg;

 bad:
	free_priority_group(pg, ti);
678
	return ERR_PTR(r);
L
Linus Torvalds 已提交
679 680
}

681
static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m)
L
Linus Torvalds 已提交
682 683
{
	unsigned hw_argc;
684
	int ret;
M
Micha³ Miros³aw 已提交
685
	struct dm_target *ti = m->ti;
L
Linus Torvalds 已提交
686

687
	static struct dm_arg _args[] = {
688
		{0, 1024, "invalid number of hardware handler args"},
L
Linus Torvalds 已提交
689 690
	};

691
	if (dm_read_arg_group(_args, as, &hw_argc, &ti->error))
L
Linus Torvalds 已提交
692 693 694 695 696
		return -EINVAL;

	if (!hw_argc)
		return 0;

697
	m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL);
698 699
	if (!try_then_request_module(scsi_dh_handler_exist(m->hw_handler_name),
				     "scsi_dh_%s", m->hw_handler_name)) {
700
		ti->error = "unknown hardware handler type";
701 702
		ret = -EINVAL;
		goto fail;
L
Linus Torvalds 已提交
703
	}
704

705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720
	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]);
	}
721
	dm_consume_args(as, hw_argc - 1);
L
Linus Torvalds 已提交
722 723

	return 0;
724 725 726 727
fail:
	kfree(m->hw_handler_name);
	m->hw_handler_name = NULL;
	return ret;
L
Linus Torvalds 已提交
728 729
}

730
static int parse_features(struct dm_arg_set *as, struct multipath *m)
L
Linus Torvalds 已提交
731 732 733
{
	int r;
	unsigned argc;
M
Micha³ Miros³aw 已提交
734
	struct dm_target *ti = m->ti;
735
	const char *arg_name;
L
Linus Torvalds 已提交
736

737
	static struct dm_arg _args[] = {
738
		{0, 6, "invalid number of feature args"},
D
Dave Wysochanski 已提交
739
		{1, 50, "pg_init_retries must be between 1 and 50"},
740
		{0, 60000, "pg_init_delay_msecs must be between 0 and 60000"},
L
Linus Torvalds 已提交
741 742
	};

743
	r = dm_read_arg_group(_args, as, &argc, &ti->error);
L
Linus Torvalds 已提交
744 745 746 747 748 749
	if (r)
		return -EINVAL;

	if (!argc)
		return 0;

D
Dave Wysochanski 已提交
750
	do {
751
		arg_name = dm_shift_arg(as);
D
Dave Wysochanski 已提交
752 753
		argc--;

754
		if (!strcasecmp(arg_name, "queue_if_no_path")) {
D
Dave Wysochanski 已提交
755 756 757 758
			r = queue_if_no_path(m, 1, 0);
			continue;
		}

759 760 761 762 763
		if (!strcasecmp(arg_name, "retain_attached_hw_handler")) {
			m->retain_attached_hw_handler = 1;
			continue;
		}

764
		if (!strcasecmp(arg_name, "pg_init_retries") &&
D
Dave Wysochanski 已提交
765
		    (argc >= 1)) {
766
			r = dm_read_arg(_args + 1, as, &m->pg_init_retries, &ti->error);
D
Dave Wysochanski 已提交
767 768 769 770
			argc--;
			continue;
		}

771
		if (!strcasecmp(arg_name, "pg_init_delay_msecs") &&
772
		    (argc >= 1)) {
773
			r = dm_read_arg(_args + 2, as, &m->pg_init_delay_msecs, &ti->error);
774 775 776 777
			argc--;
			continue;
		}

L
Linus Torvalds 已提交
778
		ti->error = "Unrecognised multipath feature request";
D
Dave Wysochanski 已提交
779 780 781 782
		r = -EINVAL;
	} while (argc && !r);

	return r;
L
Linus Torvalds 已提交
783 784 785 786 787
}

static int multipath_ctr(struct dm_target *ti, unsigned int argc,
			 char **argv)
{
788 789
	/* target arguments */
	static struct dm_arg _args[] = {
790 791
		{0, 1024, "invalid number of priority groups"},
		{0, 1024, "invalid initial priority group number"},
L
Linus Torvalds 已提交
792 793 794 795
	};

	int r;
	struct multipath *m;
796
	struct dm_arg_set as;
L
Linus Torvalds 已提交
797 798 799 800 801 802
	unsigned pg_count = 0;
	unsigned next_pg_num;

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

M
Micha³ Miros³aw 已提交
803
	m = alloc_multipath(ti);
L
Linus Torvalds 已提交
804
	if (!m) {
805
		ti->error = "can't allocate multipath";
L
Linus Torvalds 已提交
806 807 808
		return -EINVAL;
	}

M
Micha³ Miros³aw 已提交
809
	r = parse_features(&as, m);
L
Linus Torvalds 已提交
810 811 812
	if (r)
		goto bad;

M
Micha³ Miros³aw 已提交
813
	r = parse_hw_handler(&as, m);
L
Linus Torvalds 已提交
814 815 816
	if (r)
		goto bad;

817
	r = dm_read_arg(_args, &as, &m->nr_priority_groups, &ti->error);
L
Linus Torvalds 已提交
818 819 820
	if (r)
		goto bad;

821
	r = dm_read_arg(_args + 1, &as, &next_pg_num, &ti->error);
L
Linus Torvalds 已提交
822 823 824
	if (r)
		goto bad;

825 826 827 828 829 830 831
	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
832 833 834 835
	/* parse the priority groups */
	while (as.argc) {
		struct priority_group *pg;

M
Micha³ Miros³aw 已提交
836
		pg = parse_priority_group(&as, m);
837 838
		if (IS_ERR(pg)) {
			r = PTR_ERR(pg);
L
Linus Torvalds 已提交
839 840 841 842 843 844 845 846 847 848 849 850
			goto bad;
		}

		m->nr_valid_paths += pg->nr_pgpaths;
		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) {
851
		ti->error = "priority group count mismatch";
L
Linus Torvalds 已提交
852 853 854 855
		r = -EINVAL;
		goto bad;
	}

856 857
	ti->num_flush_bios = 1;
	ti->num_discard_bios = 1;
858
	ti->num_write_same_bios = 1;
M
Mikulas Patocka 已提交
859

L
Linus Torvalds 已提交
860 861 862 863 864 865 866
	return 0;

 bad:
	free_multipath(m);
	return r;
}

867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891
static void multipath_wait_for_pg_init_completion(struct multipath *m)
{
	DECLARE_WAITQUEUE(wait, current);
	unsigned long flags;

	add_wait_queue(&m->pg_init_wait, &wait);

	while (1) {
		set_current_state(TASK_UNINTERRUPTIBLE);

		spin_lock_irqsave(&m->lock, flags);
		if (!m->pg_init_in_progress) {
			spin_unlock_irqrestore(&m->lock, flags);
			break;
		}
		spin_unlock_irqrestore(&m->lock, flags);

		io_schedule();
	}
	set_current_state(TASK_RUNNING);

	remove_wait_queue(&m->pg_init_wait, &wait);
}

static void flush_multipath_work(struct multipath *m)
L
Linus Torvalds 已提交
892
{
893 894 895 896 897 898
	unsigned long flags;

	spin_lock_irqsave(&m->lock, flags);
	m->pg_init_disabled = 1;
	spin_unlock_irqrestore(&m->lock, flags);

899
	flush_workqueue(kmpath_handlerd);
900
	multipath_wait_for_pg_init_completion(m);
901
	flush_workqueue(kmultipathd);
902
	flush_work(&m->trigger_event);
903 904 905 906

	spin_lock_irqsave(&m->lock, flags);
	m->pg_init_disabled = 0;
	spin_unlock_irqrestore(&m->lock, flags);
907 908 909 910 911 912
}

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

913
	flush_multipath_work(m);
L
Linus Torvalds 已提交
914 915 916 917 918 919 920 921 922 923 924 925 926
	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);

927
	if (!pgpath->is_active)
L
Linus Torvalds 已提交
928 929
		goto out;

930
	DMWARN("Failing path %s.", pgpath->path.dev->name);
L
Linus Torvalds 已提交
931 932

	pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path);
933
	pgpath->is_active = 0;
L
Linus Torvalds 已提交
934 935 936 937 938 939 940
	pgpath->fail_count++;

	m->nr_valid_paths--;

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

M
Mike Anderson 已提交
941 942 943
	dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti,
		      pgpath->path.dev->name, m->nr_valid_paths);

944
	schedule_work(&m->trigger_event);
L
Linus Torvalds 已提交
945 946 947 948 949 950 951 952 953 954 955 956

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

	return 0;
}

/*
 * Reinstate a previously-failed path
 */
static int reinstate_path(struct pgpath *pgpath)
{
957
	int r = 0, run_queue = 0;
L
Linus Torvalds 已提交
958 959 960 961 962
	unsigned long flags;
	struct multipath *m = pgpath->pg->m;

	spin_lock_irqsave(&m->lock, flags);

963
	if (pgpath->is_active)
L
Linus Torvalds 已提交
964 965
		goto out;

966
	if (!pgpath->pg->ps.type->reinstate_path) {
L
Linus Torvalds 已提交
967 968 969 970 971 972 973 974 975 976
		DMWARN("Reinstate path not supported by path selector %s",
		       pgpath->pg->ps.type->name);
		r = -EINVAL;
		goto out;
	}

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

977
	pgpath->is_active = 1;
L
Linus Torvalds 已提交
978

979
	if (!m->nr_valid_paths++) {
980
		m->current_pgpath = NULL;
981
		run_queue = 1;
982
	} else if (m->hw_handler_name && (m->current_pg == pgpath->pg)) {
983
		if (queue_work(kmpath_handlerd, &pgpath->activate_path.work))
984 985
			m->pg_init_in_progress++;
	}
L
Linus Torvalds 已提交
986

M
Mike Anderson 已提交
987 988 989
	dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti,
		      pgpath->path.dev->name, m->nr_valid_paths);

990
	schedule_work(&m->trigger_event);
L
Linus Torvalds 已提交
991 992 993

out:
	spin_unlock_irqrestore(&m->lock, flags);
994 995
	if (run_queue)
		dm_table_run_md_queue_async(m->ti->table);
L
Linus Torvalds 已提交
996 997 998 999 1000 1001 1002 1003 1004 1005

	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)
{
1006
	int r = -EINVAL;
L
Linus Torvalds 已提交
1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035
	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,
		      int bypassed)
{
	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);

1036
	schedule_work(&m->trigger_event);
L
Linus Torvalds 已提交
1037 1038 1039 1040 1041 1042 1043 1044 1045 1046
}

/*
 * 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;
1047
	char dummy;
L
Linus Torvalds 已提交
1048

1049
	if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum ||
L
Linus Torvalds 已提交
1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066
	    (pgnum > m->nr_priority_groups)) {
		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) {
		pg->bypassed = 0;
		if (--pgnum)
			continue;

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

1067
	schedule_work(&m->trigger_event);
L
Linus Torvalds 已提交
1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078
	return 0;
}

/*
 * Set/clear bypassed status of a PG.
 * PGs are numbered upwards from 1 in the order they were declared.
 */
static int bypass_pg_num(struct multipath *m, const char *pgstr, int bypassed)
{
	struct priority_group *pg;
	unsigned pgnum;
1079
	char dummy;
L
Linus Torvalds 已提交
1080

1081
	if (!pgstr || (sscanf(pgstr, "%u%c", &pgnum, &dummy) != 1) || !pgnum ||
L
Linus Torvalds 已提交
1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095
	    (pgnum > m->nr_priority_groups)) {
		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
1096 1097 1098 1099 1100 1101 1102 1103 1104 1105
/*
 * Should we retry pg_init immediately?
 */
static int pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath)
{
	unsigned long flags;
	int limit_reached = 0;

	spin_lock_irqsave(&m->lock, flags);

1106
	if (m->pg_init_count <= m->pg_init_retries && !m->pg_init_disabled)
D
Dave Wysochanski 已提交
1107 1108 1109 1110 1111 1112 1113 1114 1115
		m->pg_init_required = 1;
	else
		limit_reached = 1;

	spin_unlock_irqrestore(&m->lock, flags);

	return limit_reached;
}

1116
static void pg_init_done(void *data, int errors)
1117
{
1118
	struct pgpath *pgpath = data;
1119 1120 1121
	struct priority_group *pg = pgpath->pg;
	struct multipath *m = pg->m;
	unsigned long flags;
1122
	unsigned delay_retry = 0;
1123 1124 1125 1126 1127 1128 1129 1130 1131 1132

	/* device or driver problems */
	switch (errors) {
	case SCSI_DH_OK:
		break;
	case SCSI_DH_NOSYS:
		if (!m->hw_handler_name) {
			errors = 0;
			break;
		}
1133 1134
		DMERR("Could not failover the device: Handler scsi_dh_%s "
		      "Error %d.", m->hw_handler_name, errors);
1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147
		/*
		 * 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.
		 */
		bypass_pg(m, pg, 1);
		break;
	case SCSI_DH_RETRY:
1148 1149
		/* Wait before retrying. */
		delay_retry = 1;
1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166
	case SCSI_DH_IMM_RETRY:
	case SCSI_DH_RES_TEMP_UNAVAIL:
		if (pg_init_limit_reached(m, pgpath))
			fail_path(pgpath);
		errors = 0;
		break;
	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) {
1167 1168 1169 1170 1171
		if (pgpath == m->current_pgpath) {
			DMERR("Could not failover device. Error %d.", errors);
			m->current_pgpath = NULL;
			m->current_pg = NULL;
		}
1172
	} else if (!m->pg_init_required)
1173 1174
		pg->bypassed = 0;

1175 1176 1177 1178
	if (--m->pg_init_in_progress)
		/* Activations of other paths are still on going */
		goto out;

1179 1180 1181 1182 1183 1184
	if (m->pg_init_required) {
		m->pg_init_delay_retry = delay_retry;
		if (__pg_init_all_paths(m))
			goto out;
	}
	m->queue_io = 0;
1185

1186 1187 1188 1189 1190
	/*
	 * Wake up any thread waiting to suspend.
	 */
	wake_up(&m->pg_init_wait);

1191
out:
1192 1193 1194
	spin_unlock_irqrestore(&m->lock, flags);
}

1195 1196
static void activate_path(struct work_struct *work)
{
1197
	struct pgpath *pgpath =
1198
		container_of(work, struct pgpath, activate_path.work);
1199

1200 1201 1202 1203 1204
	if (pgpath->is_active)
		scsi_dh_activate(bdev_get_queue(pgpath->path.dev->bdev),
				 pg_init_done, pgpath);
	else
		pg_init_done(pgpath, SCSI_DH_DEV_OFFLINED);
1205 1206
}

1207 1208 1209 1210 1211 1212 1213
static int noretry_error(int error)
{
	switch (error) {
	case -EOPNOTSUPP:
	case -EREMOTEIO:
	case -EILSEQ:
	case -ENODATA:
1214
	case -ENOSPC:
1215 1216 1217 1218 1219 1220 1221
		return 1;
	}

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

L
Linus Torvalds 已提交
1222 1223 1224
/*
 * end_io handling
 */
1225
static int do_end_io(struct multipath *m, struct request *clone,
A
Alasdair G Kergon 已提交
1226
		     int error, struct dm_mpath_io *mpio)
L
Linus Torvalds 已提交
1227
{
1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239
	/*
	 * 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.
	 * (See drivers/md/dm.c:end_clone_bio() about why the clone requests
	 *  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.
	 */
	int r = DM_ENDIO_REQUEUE;
1240
	unsigned long flags;
L
Linus Torvalds 已提交
1241

1242
	if (!error && !clone->errors)
L
Linus Torvalds 已提交
1243 1244
		return 0;	/* I/O complete */

1245
	if (noretry_error(error))
M
Mike Snitzer 已提交
1246 1247
		return error;

1248 1249
	if (mpio->pgpath)
		fail_path(mpio->pgpath);
L
Linus Torvalds 已提交
1250

1251
	spin_lock_irqsave(&m->lock, flags);
1252 1253 1254 1255 1256 1257 1258 1259 1260
	if (!m->nr_valid_paths) {
		if (!m->queue_if_no_path) {
			if (!__must_push_back(m))
				r = -EIO;
		} else {
			if (error == -EBADE)
				r = error;
		}
	}
1261
	spin_unlock_irqrestore(&m->lock, flags);
L
Linus Torvalds 已提交
1262

1263
	return r;
L
Linus Torvalds 已提交
1264 1265
}

1266
static int multipath_end_io(struct dm_target *ti, struct request *clone,
L
Linus Torvalds 已提交
1267 1268
			    int error, union map_info *map_context)
{
A
Alasdair G Kergon 已提交
1269 1270
	struct multipath *m = ti->private;
	struct dm_mpath_io *mpio = map_context->ptr;
1271
	struct pgpath *pgpath;
L
Linus Torvalds 已提交
1272 1273 1274
	struct path_selector *ps;
	int r;

1275 1276
	BUG_ON(!mpio);

1277
	r  = do_end_io(m, clone, error, mpio);
1278
	pgpath = mpio->pgpath;
L
Linus Torvalds 已提交
1279 1280 1281
	if (pgpath) {
		ps = &pgpath->pg->ps;
		if (ps->type->end_io)
1282
			ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes);
L
Linus Torvalds 已提交
1283
	}
1284
	clear_mapinfo(m, map_context);
L
Linus Torvalds 已提交
1285 1286 1287 1288 1289 1290

	return r;
}

/*
 * Suspend can't complete until all the I/O is processed so if
1291 1292 1293
 * 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
1294 1295 1296 1297 1298
 */
static void multipath_presuspend(struct dm_target *ti)
{
	struct multipath *m = (struct multipath *) ti->private;

1299
	queue_if_no_path(m, 0, 1);
L
Linus Torvalds 已提交
1300 1301
}

1302 1303
static void multipath_postsuspend(struct dm_target *ti)
{
1304 1305 1306
	struct multipath *m = ti->private;

	mutex_lock(&m->work_mutex);
1307
	flush_multipath_work(m);
1308
	mutex_unlock(&m->work_mutex);
1309 1310
}

1311 1312 1313
/*
 * Restore the queue_if_no_path setting.
 */
L
Linus Torvalds 已提交
1314 1315 1316 1317 1318 1319
static void multipath_resume(struct dm_target *ti)
{
	struct multipath *m = (struct multipath *) ti->private;
	unsigned long flags;

	spin_lock_irqsave(&m->lock, flags);
1320
	m->queue_if_no_path = m->saved_queue_if_no_path;
L
Linus Torvalds 已提交
1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339
	spin_unlock_irqrestore(&m->lock, flags);
}

/*
 * 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]* ]+ ]+
 */
1340 1341
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
1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354
{
	int sz = 0;
	unsigned long flags;
	struct multipath *m = (struct multipath *) ti->private;
	struct priority_group *pg;
	struct pgpath *p;
	unsigned pg_num;
	char state;

	spin_lock_irqsave(&m->lock, flags);

	/* Features */
	if (type == STATUSTYPE_INFO)
1355
		DMEMIT("2 %u %u ", m->queue_io, m->pg_init_count);
D
Dave Wysochanski 已提交
1356 1357
	else {
		DMEMIT("%u ", m->queue_if_no_path +
1358
			      (m->pg_init_retries > 0) * 2 +
1359 1360
			      (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2 +
			      m->retain_attached_hw_handler);
D
Dave Wysochanski 已提交
1361 1362 1363 1364
		if (m->queue_if_no_path)
			DMEMIT("queue_if_no_path ");
		if (m->pg_init_retries)
			DMEMIT("pg_init_retries %u ", m->pg_init_retries);
1365 1366
		if (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT)
			DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs);
1367 1368
		if (m->retain_attached_hw_handler)
			DMEMIT("retain_attached_hw_handler ");
D
Dave Wysochanski 已提交
1369
	}
L
Linus Torvalds 已提交
1370

1371
	if (!m->hw_handler_name || type == STATUSTYPE_INFO)
L
Linus Torvalds 已提交
1372 1373
		DMEMIT("0 ");
	else
1374
		DMEMIT("1 %s ", m->hw_handler_name);
L
Linus Torvalds 已提交
1375 1376 1377 1378 1379 1380 1381 1382

	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
1383
		pg_num = (m->nr_priority_groups ? 1 : 0);
L
Linus Torvalds 已提交
1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410

	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,
1411
				       p->is_active ? "A" : "F",
L
Linus Torvalds 已提交
1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450
				       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)
{
1451
	int r = -EINVAL;
L
Linus Torvalds 已提交
1452 1453 1454 1455
	struct dm_dev *dev;
	struct multipath *m = (struct multipath *) ti->private;
	action_fn action;

1456 1457
	mutex_lock(&m->work_mutex);

1458 1459 1460 1461 1462
	if (dm_suspended(ti)) {
		r = -EBUSY;
		goto out;
	}

L
Linus Torvalds 已提交
1463
	if (argc == 1) {
1464
		if (!strcasecmp(argv[0], "queue_if_no_path")) {
1465 1466
			r = queue_if_no_path(m, 1, 0);
			goto out;
1467
		} else if (!strcasecmp(argv[0], "fail_if_no_path")) {
1468 1469 1470
			r = queue_if_no_path(m, 0, 0);
			goto out;
		}
L
Linus Torvalds 已提交
1471 1472
	}

1473
	if (argc != 2) {
1474
		DMWARN("Invalid multipath message arguments. Expected 2 arguments, got %d.", argc);
1475 1476
		goto out;
	}
L
Linus Torvalds 已提交
1477

1478
	if (!strcasecmp(argv[0], "disable_group")) {
1479 1480
		r = bypass_pg_num(m, argv[1], 1);
		goto out;
1481
	} else if (!strcasecmp(argv[0], "enable_group")) {
1482 1483
		r = bypass_pg_num(m, argv[1], 0);
		goto out;
1484
	} else if (!strcasecmp(argv[0], "switch_group")) {
1485 1486
		r = switch_pg_num(m, argv[1]);
		goto out;
1487
	} else if (!strcasecmp(argv[0], "reinstate_path"))
L
Linus Torvalds 已提交
1488
		action = reinstate_path;
1489
	else if (!strcasecmp(argv[0], "fail_path"))
L
Linus Torvalds 已提交
1490
		action = fail_path;
1491
	else {
1492
		DMWARN("Unrecognised multipath message received: %s", argv[0]);
1493 1494
		goto out;
	}
L
Linus Torvalds 已提交
1495

1496
	r = dm_get_device(ti, argv[1], dm_table_get_mode(ti->table), &dev);
L
Linus Torvalds 已提交
1497
	if (r) {
1498
		DMWARN("message: error getting device %s",
L
Linus Torvalds 已提交
1499
		       argv[1]);
1500
		goto out;
L
Linus Torvalds 已提交
1501 1502 1503 1504 1505 1506
	}

	r = action_dev(m, dev, action);

	dm_put_device(ti, dev);

1507 1508
out:
	mutex_unlock(&m->work_mutex);
L
Linus Torvalds 已提交
1509 1510 1511
	return r;
}

1512
static int multipath_ioctl(struct dm_target *ti, unsigned int cmd,
M
Milan Broz 已提交
1513 1514
			   unsigned long arg)
{
1515
	struct multipath *m = ti->private;
1516
	struct pgpath *pgpath;
1517 1518
	struct block_device *bdev;
	fmode_t mode;
M
Milan Broz 已提交
1519
	unsigned long flags;
1520 1521 1522 1523 1524
	int r;

	bdev = NULL;
	mode = 0;
	r = 0;
M
Milan Broz 已提交
1525 1526 1527 1528

	spin_lock_irqsave(&m->lock, flags);

	if (!m->current_pgpath)
1529
		__choose_pgpath(m, 0);
M
Milan Broz 已提交
1530

1531 1532 1533 1534 1535
	pgpath = m->current_pgpath;

	if (pgpath) {
		bdev = pgpath->path.dev->bdev;
		mode = pgpath->path.dev->mode;
1536
	}
M
Milan Broz 已提交
1537

1538
	if ((pgpath && m->queue_io) || (!pgpath && m->queue_if_no_path))
1539
		r = -ENOTCONN;
M
Milan Broz 已提交
1540 1541 1542 1543 1544
	else if (!bdev)
		r = -EIO;

	spin_unlock_irqrestore(&m->lock, flags);

1545 1546 1547
	/*
	 * Only pass ioctls through if the device sizes match exactly.
	 */
1548 1549 1550 1551 1552
	if (!bdev || ti->len != i_size_read(bdev->bd_inode) >> SECTOR_SHIFT) {
		int err = scsi_verify_blk_ioctl(NULL, cmd);
		if (err)
			r = err;
	}
1553

1554 1555 1556 1557 1558 1559 1560 1561
	if (r == -ENOTCONN && !fatal_signal_pending(current)) {
		spin_lock_irqsave(&m->lock, flags);
		if (!m->current_pg) {
			/* Path status changed, redo selection */
			__choose_pgpath(m, 0);
		}
		if (m->pg_init_required)
			__pg_init_all_paths(m);
1562
		spin_unlock_irqrestore(&m->lock, flags);
1563
		dm_table_run_md_queue_async(m->ti->table);
1564
	}
1565

1566
	return r ? : __blkdev_driver_ioctl(bdev, mode, cmd, arg);
M
Milan Broz 已提交
1567 1568
}

1569 1570 1571 1572 1573 1574 1575 1576 1577 1578
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) {
1579
			ret = fn(ti, p->path.dev, ti->begin, ti->len, data);
1580 1581 1582 1583 1584 1585 1586 1587 1588
			if (ret)
				goto out;
		}
	}

out:
	return ret;
}

1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613
static int __pgpath_busy(struct pgpath *pgpath)
{
	struct request_queue *q = bdev_get_queue(pgpath->path.dev->bdev);

	return dm_underlying_device_busy(q);
}

/*
 * 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)
{
	int busy = 0, has_active = 0;
	struct multipath *m = ti->private;
	struct priority_group *pg;
	struct pgpath *pgpath;
	unsigned long flags;

	spin_lock_irqsave(&m->lock, flags);

1614 1615 1616
	/* pg_init in progress or no paths available */
	if (m->pg_init_in_progress ||
	    (!m->nr_valid_paths && m->queue_if_no_path)) {
1617 1618 1619
		busy = 1;
		goto out;
	}
1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663
	/* Guess which priority_group will be used at next mapping time */
	if (unlikely(!m->current_pgpath && m->next_pg))
		pg = m->next_pg;
	else if (likely(m->current_pg))
		pg = m->current_pg;
	else
		/*
		 * We don't know which pg will be used at next mapping time.
		 * We don't call __choose_pgpath() here to avoid to trigger
		 * 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.
		 */
		goto out;

	/*
	 * 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.
	 */
	busy = 1;
	list_for_each_entry(pgpath, &pg->pgpaths, list)
		if (pgpath->is_active) {
			has_active = 1;

			if (!__pgpath_busy(pgpath)) {
				busy = 0;
				break;
			}
		}

	if (!has_active)
		/*
		 * 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.
		 */
		busy = 0;

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

	return busy;
}

L
Linus Torvalds 已提交
1664 1665 1666 1667 1668
/*-----------------------------------------------------------------
 * Module setup
 *---------------------------------------------------------------*/
static struct target_type multipath_target = {
	.name = "multipath",
1669
	.version = {1, 7, 0},
L
Linus Torvalds 已提交
1670 1671 1672
	.module = THIS_MODULE,
	.ctr = multipath_ctr,
	.dtr = multipath_dtr,
1673 1674
	.map_rq = multipath_map,
	.rq_end_io = multipath_end_io,
L
Linus Torvalds 已提交
1675
	.presuspend = multipath_presuspend,
1676
	.postsuspend = multipath_postsuspend,
L
Linus Torvalds 已提交
1677 1678 1679
	.resume = multipath_resume,
	.status = multipath_status,
	.message = multipath_message,
M
Milan Broz 已提交
1680
	.ioctl  = multipath_ioctl,
1681
	.iterate_devices = multipath_iterate_devices,
1682
	.busy = multipath_busy,
L
Linus Torvalds 已提交
1683 1684 1685 1686 1687 1688 1689
};

static int __init dm_multipath_init(void)
{
	int r;

	/* allocate a slab for the dm_ios */
A
Alasdair G Kergon 已提交
1690
	_mpio_cache = KMEM_CACHE(dm_mpath_io, 0);
L
Linus Torvalds 已提交
1691 1692 1693 1694 1695
	if (!_mpio_cache)
		return -ENOMEM;

	r = dm_register_target(&multipath_target);
	if (r < 0) {
1696
		DMERR("register failed %d", r);
L
Linus Torvalds 已提交
1697 1698 1699 1700
		kmem_cache_destroy(_mpio_cache);
		return -EINVAL;
	}

1701
	kmultipathd = alloc_workqueue("kmpathd", WQ_MEM_RECLAIM, 0);
1702
	if (!kmultipathd) {
1703
		DMERR("failed to create workqueue kmpathd");
1704 1705 1706 1707 1708
		dm_unregister_target(&multipath_target);
		kmem_cache_destroy(_mpio_cache);
		return -ENOMEM;
	}

1709 1710 1711 1712 1713 1714
	/*
	 * 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.
	 */
1715 1716
	kmpath_handlerd = alloc_ordered_workqueue("kmpath_handlerd",
						  WQ_MEM_RECLAIM);
1717 1718 1719 1720 1721 1722 1723 1724
	if (!kmpath_handlerd) {
		DMERR("failed to create workqueue kmpath_handlerd");
		destroy_workqueue(kmultipathd);
		dm_unregister_target(&multipath_target);
		kmem_cache_destroy(_mpio_cache);
		return -ENOMEM;
	}

1725
	DMINFO("version %u.%u.%u loaded",
L
Linus Torvalds 已提交
1726 1727 1728 1729 1730 1731 1732 1733
	       multipath_target.version[0], multipath_target.version[1],
	       multipath_target.version[2]);

	return r;
}

static void __exit dm_multipath_exit(void)
{
1734
	destroy_workqueue(kmpath_handlerd);
1735 1736
	destroy_workqueue(kmultipathd);

1737
	dm_unregister_target(&multipath_target);
L
Linus Torvalds 已提交
1738 1739 1740 1741 1742 1743 1744 1745 1746
	kmem_cache_destroy(_mpio_cache);
}

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");