1/*
   2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
   3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
   4 *
   5 * This file is released under the GPL.
   6 */
   7
   8#include "dm.h"
 
   9#include "dm-uevent.h"
  10
  11#include <linux/init.h>
  12#include <linux/module.h>
  13#include <linux/mutex.h>
  14#include <linux/moduleparam.h>
  15#include <linux/blkpg.h>
  16#include <linux/bio.h>
  17#include <linux/buffer_head.h>
  18#include <linux/mempool.h>
 
  19#include <linux/slab.h>
  20#include <linux/idr.h>
 
  21#include <linux/hdreg.h>
  22#include <linux/delay.h>
  23
  24#include <trace/events/block.h>
 
  25
  26#define DM_MSG_PREFIX "core"
  27
  28/*
  29 * Cookies are numeric values sent with CHANGE and REMOVE
  30 * uevents while resuming, removing or renaming the device.
  31 */
  32#define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
  33#define DM_COOKIE_LENGTH 24
  34
  35static const char *_name = DM_NAME;
  36
  37static unsigned int major = 0;
  38static unsigned int _major = 0;
  39
  40static DEFINE_IDR(_minor_idr);
  41
  42static DEFINE_SPINLOCK(_minor_lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  43/*
  44 * For bio-based dm.
  45 * One of these is allocated per bio.
  46 */
  47struct dm_io {
  48	struct mapped_device *md;
  49	int error;
  50	atomic_t io_count;
  51	struct bio *bio;
  52	unsigned long start_time;
  53	spinlock_t endio_lock;
 
  54};
  55
  56/*
  57 * For bio-based dm.
  58 * One of these is allocated per target within a bio.  Hopefully
  59 * this will be simplified out one day.
  60 */
 
  61struct dm_target_io {
 
  62	struct dm_io *io;
  63	struct dm_target *ti;
  64	union map_info info;
 
 
 
  65};
  66
  67/*
  68 * For request-based dm.
  69 * One of these is allocated per request.
  70 */
  71struct dm_rq_target_io {
 
 
  72	struct mapped_device *md;
  73	struct dm_target *ti;
  74	struct request *orig, clone;
  75	int error;
  76	union map_info info;
 
 
 
 
  77};
  78
  79/*
  80 * For request-based dm.
  81 * One of these is allocated per bio.
  82 */
  83struct dm_rq_clone_bio_info {
  84	struct bio *orig;
  85	struct dm_rq_target_io *tio;
  86};
  87
  88union map_info *dm_get_mapinfo(struct bio *bio)
  89{
  90	if (bio && bio->bi_private)
  91		return &((struct dm_target_io *)bio->bi_private)->info;
  92	return NULL;
 
 
  93}
 
  94
  95union map_info *dm_get_rq_mapinfo(struct request *rq)
  96{
  97	if (rq && rq->end_io_data)
  98		return &((struct dm_rq_target_io *)rq->end_io_data)->info;
  99	return NULL;
 100}
 101EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
 102
 103#define MINOR_ALLOCED ((void *)-1)
 104
 105/*
 106 * Bits for the md->flags field.
 107 */
 108#define DMF_BLOCK_IO_FOR_SUSPEND 0
 109#define DMF_SUSPENDED 1
 110#define DMF_FROZEN 2
 111#define DMF_FREEING 3
 112#define DMF_DELETING 4
 113#define DMF_NOFLUSH_SUSPENDING 5
 114#define DMF_MERGE_IS_OPTIONAL 6
 
 
 
 
 115
 116/*
 117 * Work processed by per-device workqueue.
 118 */
 119struct mapped_device {
 120	struct rw_semaphore io_lock;
 121	struct mutex suspend_lock;
 122	rwlock_t map_lock;
 123	atomic_t holders;
 124	atomic_t open_count;
 125
 126	unsigned long flags;
 127
 128	struct request_queue *queue;
 129	unsigned type;
 130	/* Protect queue and type against concurrent access. */
 131	struct mutex type_lock;
 132
 133	struct gendisk *disk;
 134	char name[16];
 135
 136	void *interface_ptr;
 137
 138	/*
 139	 * A list of ios that arrived while we were suspended.
 140	 */
 141	atomic_t pending[2];
 142	wait_queue_head_t wait;
 143	struct work_struct work;
 144	struct bio_list deferred;
 145	spinlock_t deferred_lock;
 146
 147	/*
 148	 * Processing queue (flush)
 149	 */
 150	struct workqueue_struct *wq;
 
 151
 152	/*
 153	 * The current mapping.
 154	 */
 155	struct dm_table *map;
 
 156
 157	/*
 158	 * io objects are allocated from here.
 159	 */
 160	mempool_t *io_pool;
 161	mempool_t *tio_pool;
 162
 163	struct bio_set *bs;
 
 
 
 
 
 164
 165	/*
 166	 * Event handling.
 167	 */
 168	atomic_t event_nr;
 169	wait_queue_head_t eventq;
 170	atomic_t uevent_seq;
 171	struct list_head uevent_list;
 172	spinlock_t uevent_lock; /* Protect access to uevent_list */
 173
 174	/*
 175	 * freeze/thaw support require holding onto a super block
 176	 */
 177	struct super_block *frozen_sb;
 178	struct block_device *bdev;
 179
 180	/* forced geometry settings */
 181	struct hd_geometry geometry;
 
 
 
 182
 183	/* For saving the address of __make_request for request based dm */
 184	make_request_fn *saved_make_request_fn;
 
 
 185
 186	/* sysfs handle */
 187	struct kobject kobj;
 
 
 188
 189	/* zero-length flush that will be cloned and submitted to targets */
 190	struct bio flush_bio;
 191};
 192
 193/*
 194 * For mempools pre-allocation at the table loading time.
 195 */
 196struct dm_md_mempools {
 197	mempool_t *io_pool;
 198	mempool_t *tio_pool;
 199	struct bio_set *bs;
 200};
 201
 202#define MIN_IOS 256
 203static struct kmem_cache *_io_cache;
 204static struct kmem_cache *_tio_cache;
 205static struct kmem_cache *_rq_tio_cache;
 206static struct kmem_cache *_rq_bio_info_cache;
 207
 208static int __init local_init(void)
 209{
 210	int r = -ENOMEM;
 211
 212	/* allocate a slab for the dm_ios */
 213	_io_cache = KMEM_CACHE(dm_io, 0);
 214	if (!_io_cache)
 215		return r;
 216
 217	/* allocate a slab for the target ios */
 218	_tio_cache = KMEM_CACHE(dm_target_io, 0);
 219	if (!_tio_cache)
 220		goto out_free_io_cache;
 221
 222	_rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
 223	if (!_rq_tio_cache)
 224		goto out_free_tio_cache;
 225
 226	_rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
 227	if (!_rq_bio_info_cache)
 228		goto out_free_rq_tio_cache;
 229
 230	r = dm_uevent_init();
 231	if (r)
 232		goto out_free_rq_bio_info_cache;
 
 
 
 
 
 
 233
 234	_major = major;
 235	r = register_blkdev(_major, _name);
 236	if (r < 0)
 237		goto out_uevent_exit;
 238
 239	if (!_major)
 240		_major = r;
 241
 242	return 0;
 243
 
 
 244out_uevent_exit:
 245	dm_uevent_exit();
 246out_free_rq_bio_info_cache:
 247	kmem_cache_destroy(_rq_bio_info_cache);
 248out_free_rq_tio_cache:
 249	kmem_cache_destroy(_rq_tio_cache);
 250out_free_tio_cache:
 251	kmem_cache_destroy(_tio_cache);
 252out_free_io_cache:
 253	kmem_cache_destroy(_io_cache);
 254
 255	return r;
 256}
 257
 258static void local_exit(void)
 259{
 260	kmem_cache_destroy(_rq_bio_info_cache);
 261	kmem_cache_destroy(_rq_tio_cache);
 262	kmem_cache_destroy(_tio_cache);
 263	kmem_cache_destroy(_io_cache);
 264	unregister_blkdev(_major, _name);
 265	dm_uevent_exit();
 266
 267	_major = 0;
 268
 269	DMINFO("cleaned up");
 270}
 271
 272static int (*_inits[])(void) __initdata = {
 273	local_init,
 274	dm_target_init,
 275	dm_linear_init,
 276	dm_stripe_init,
 277	dm_io_init,
 278	dm_kcopyd_init,
 279	dm_interface_init,
 
 280};
 281
 282static void (*_exits[])(void) = {
 283	local_exit,
 284	dm_target_exit,
 285	dm_linear_exit,
 286	dm_stripe_exit,
 287	dm_io_exit,
 288	dm_kcopyd_exit,
 289	dm_interface_exit,
 
 290};
 291
 292static int __init dm_init(void)
 293{
 294	const int count = ARRAY_SIZE(_inits);
 295
 296	int r, i;
 297
 298	for (i = 0; i < count; i++) {
 299		r = _inits[i]();
 300		if (r)
 301			goto bad;
 302	}
 303
 304	return 0;
 305
 306      bad:
 307	while (i--)
 308		_exits[i]();
 309
 310	return r;
 311}
 312
 313static void __exit dm_exit(void)
 314{
 315	int i = ARRAY_SIZE(_exits);
 316
 317	while (i--)
 318		_exits[i]();
 319
 320	/*
 321	 * Should be empty by this point.
 322	 */
 323	idr_remove_all(&_minor_idr);
 324	idr_destroy(&_minor_idr);
 325}
 326
 327/*
 328 * Block device functions
 329 */
 330int dm_deleting_md(struct mapped_device *md)
 331{
 332	return test_bit(DMF_DELETING, &md->flags);
 333}
 334
 335static int dm_blk_open(struct block_device *bdev, fmode_t mode)
 336{
 337	struct mapped_device *md;
 338
 339	spin_lock(&_minor_lock);
 340
 341	md = bdev->bd_disk->private_data;
 342	if (!md)
 343		goto out;
 344
 345	if (test_bit(DMF_FREEING, &md->flags) ||
 346	    dm_deleting_md(md)) {
 347		md = NULL;
 348		goto out;
 349	}
 350
 351	dm_get(md);
 352	atomic_inc(&md->open_count);
 353
 354out:
 355	spin_unlock(&_minor_lock);
 356
 357	return md ? 0 : -ENXIO;
 358}
 359
 360static int dm_blk_close(struct gendisk *disk, fmode_t mode)
 361{
 362	struct mapped_device *md = disk->private_data;
 363
 364	spin_lock(&_minor_lock);
 365
 366	atomic_dec(&md->open_count);
 367	dm_put(md);
 
 368
 369	spin_unlock(&_minor_lock);
 
 
 370
 371	return 0;
 
 
 372}
 373
 374int dm_open_count(struct mapped_device *md)
 375{
 376	return atomic_read(&md->open_count);
 377}
 378
 379/*
 380 * Guarantees nothing is using the device before it's deleted.
 381 */
 382int dm_lock_for_deletion(struct mapped_device *md)
 383{
 384	int r = 0;
 385
 386	spin_lock(&_minor_lock);
 387
 388	if (dm_open_count(md))
 389		r = -EBUSY;
 
 
 
 
 390	else
 391		set_bit(DMF_DELETING, &md->flags);
 392
 393	spin_unlock(&_minor_lock);
 394
 395	return r;
 396}
 397
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 398static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 399{
 400	struct mapped_device *md = bdev->bd_disk->private_data;
 401
 402	return dm_get_geometry(md, geo);
 403}
 404
 405static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
 406			unsigned int cmd, unsigned long arg)
 407{
 408	struct mapped_device *md = bdev->bd_disk->private_data;
 409	struct dm_table *map = dm_get_live_table(md);
 410	struct dm_target *tgt;
 411	int r = -ENOTTY;
 
 412
 413	if (!map || !dm_table_get_size(map))
 414		goto out;
 415
 416	/* We only support devices that have a single target */
 417	if (dm_table_get_num_targets(map) != 1)
 418		goto out;
 419
 420	tgt = dm_table_get_target(map, 0);
 
 
 
 
 421
 422	if (dm_suspended_md(md)) {
 423		r = -EAGAIN;
 
 
 
 
 
 
 424		goto out;
 425	}
 426
 427	if (tgt->type->ioctl)
 428		r = tgt->type->ioctl(tgt, cmd, arg);
 
 
 
 
 
 429
 430out:
 431	dm_table_put(map);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 432
 433	return r;
 434}
 435
 436static struct dm_io *alloc_io(struct mapped_device *md)
 
 437{
 438	return mempool_alloc(md->io_pool, GFP_NOIO);
 439}
 440
 441static void free_io(struct mapped_device *md, struct dm_io *io)
 
 442{
 443	mempool_free(io, md->io_pool);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 444}
 445
 446static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
 
 
 447{
 448	mempool_free(tio, md->tio_pool);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 449}
 450
 451static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
 452					    gfp_t gfp_mask)
 453{
 454	return mempool_alloc(md->tio_pool, gfp_mask);
 455}
 456
 457static void free_rq_tio(struct dm_rq_target_io *tio)
 
 458{
 459	mempool_free(tio, tio->md->tio_pool);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 460}
 461
 462static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md)
 463{
 464	return mempool_alloc(md->io_pool, GFP_ATOMIC);
 
 
 465}
 466
 467static void free_bio_info(struct dm_rq_clone_bio_info *info)
 468{
 469	mempool_free(info, info->tio->md->io_pool);
 
 
 
 
 
 
 
 
 
 470}
 471
 472static int md_in_flight(struct mapped_device *md)
 473{
 474	return atomic_read(&md->pending[READ]) +
 475	       atomic_read(&md->pending[WRITE]);
 
 
 476}
 477
 478static void start_io_acct(struct dm_io *io)
 479{
 480	struct mapped_device *md = io->md;
 481	int cpu;
 482	int rw = bio_data_dir(io->bio);
 483
 484	io->start_time = jiffies;
 485
 486	cpu = part_stat_lock();
 487	part_round_stats(cpu, &dm_disk(md)->part0);
 488	part_stat_unlock();
 489	atomic_set(&dm_disk(md)->part0.in_flight[rw],
 490		atomic_inc_return(&md->pending[rw]));
 
 
 491}
 492
 493static void end_io_acct(struct dm_io *io)
 494{
 495	struct mapped_device *md = io->md;
 496	struct bio *bio = io->bio;
 497	unsigned long duration = jiffies - io->start_time;
 498	int pending, cpu;
 499	int rw = bio_data_dir(bio);
 500
 501	cpu = part_stat_lock();
 502	part_round_stats(cpu, &dm_disk(md)->part0);
 503	part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
 504	part_stat_unlock();
 505
 506	/*
 507	 * After this is decremented the bio must not be touched if it is
 508	 * a flush.
 509	 */
 510	pending = atomic_dec_return(&md->pending[rw]);
 511	atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
 512	pending += atomic_read(&md->pending[rw^0x1]);
 513
 514	/* nudge anyone waiting on suspend queue */
 515	if (!pending)
 516		wake_up(&md->wait);
 517}
 518
 519/*
 520 * Add the bio to the list of deferred io.
 521 */
 522static void queue_io(struct mapped_device *md, struct bio *bio)
 523{
 524	unsigned long flags;
 525
 526	spin_lock_irqsave(&md->deferred_lock, flags);
 527	bio_list_add(&md->deferred, bio);
 528	spin_unlock_irqrestore(&md->deferred_lock, flags);
 529	queue_work(md->wq, &md->work);
 530}
 531
 532/*
 533 * Everyone (including functions in this file), should use this
 534 * function to access the md->map field, and make sure they call
 535 * dm_table_put() when finished.
 536 */
 537struct dm_table *dm_get_live_table(struct mapped_device *md)
 538{
 539	struct dm_table *t;
 540	unsigned long flags;
 
 
 
 
 
 
 
 
 
 
 
 
 
 541
 542	read_lock_irqsave(&md->map_lock, flags);
 543	t = md->map;
 544	if (t)
 545		dm_table_get(t);
 546	read_unlock_irqrestore(&md->map_lock, flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 547
 548	return t;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 549}
 550
 551/*
 552 * Get the geometry associated with a dm device
 553 */
 554int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
 555{
 556	*geo = md->geometry;
 557
 558	return 0;
 559}
 560
 561/*
 562 * Set the geometry of a device.
 563 */
 564int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
 565{
 566	sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
 567
 568	if (geo->start > sz) {
 569		DMWARN("Start sector is beyond the geometry limits.");
 570		return -EINVAL;
 571	}
 572
 573	md->geometry = *geo;
 574
 575	return 0;
 576}
 577
 578/*-----------------------------------------------------------------
 579 * CRUD START:
 580 *   A more elegant soln is in the works that uses the queue
 581 *   merge fn, unfortunately there are a couple of changes to
 582 *   the block layer that I want to make for this.  So in the
 583 *   interests of getting something for people to use I give
 584 *   you this clearly demarcated crap.
 585 *---------------------------------------------------------------*/
 586
 587static int __noflush_suspending(struct mapped_device *md)
 588{
 589	return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
 590}
 591
 592/*
 593 * Decrements the number of outstanding ios that a bio has been
 594 * cloned into, completing the original io if necc.
 595 */
 596static void dec_pending(struct dm_io *io, int error)
 597{
 598	unsigned long flags;
 599	int io_error;
 600	struct bio *bio;
 601	struct mapped_device *md = io->md;
 602
 603	/* Push-back supersedes any I/O errors */
 604	if (unlikely(error)) {
 605		spin_lock_irqsave(&io->endio_lock, flags);
 606		if (!(io->error > 0 && __noflush_suspending(md)))
 607			io->error = error;
 608		spin_unlock_irqrestore(&io->endio_lock, flags);
 609	}
 610
 611	if (atomic_dec_and_test(&io->io_count)) {
 612		if (io->error == DM_ENDIO_REQUEUE) {
 613			/*
 614			 * Target requested pushing back the I/O.
 615			 */
 616			spin_lock_irqsave(&md->deferred_lock, flags);
 617			if (__noflush_suspending(md))
 618				bio_list_add_head(&md->deferred, io->bio);
 
 619			else
 620				/* noflush suspend was interrupted. */
 621				io->error = -EIO;
 622			spin_unlock_irqrestore(&md->deferred_lock, flags);
 623		}
 624
 625		io_error = io->error;
 626		bio = io->bio;
 627		end_io_acct(io);
 628		free_io(md, io);
 629
 630		if (io_error == DM_ENDIO_REQUEUE)
 631			return;
 632
 633		if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) {
 634			/*
 635			 * Preflush done for flush with data, reissue
 636			 * without REQ_FLUSH.
 637			 */
 638			bio->bi_rw &= ~REQ_FLUSH;
 639			queue_io(md, bio);
 640		} else {
 641			/* done with normal IO or empty flush */
 642			trace_block_bio_complete(md->queue, bio, io_error);
 643			bio_endio(bio, io_error);
 
 644		}
 645	}
 646}
 647
 648static void clone_endio(struct bio *bio, int error)
 649{
 650	int r = 0;
 651	struct dm_target_io *tio = bio->bi_private;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 652	struct dm_io *io = tio->io;
 653	struct mapped_device *md = tio->io->md;
 654	dm_endio_fn endio = tio->ti->type->end_io;
 655
 656	if (!bio_flagged(bio, BIO_UPTODATE) && !error)
 657		error = -EIO;
 
 
 
 
 
 
 
 
 
 658
 659	if (endio) {
 660		r = endio(tio->ti, bio, error, &tio->info);
 661		if (r < 0 || r == DM_ENDIO_REQUEUE)
 662			/*
 663			 * error and requeue request are handled
 664			 * in dec_pending().
 665			 */
 666			error = r;
 667		else if (r == DM_ENDIO_INCOMPLETE)
 668			/* The target will handle the io */
 669			return;
 670		else if (r) {
 671			DMWARN("unimplemented target endio return value: %d", r);
 672			BUG();
 673		}
 674	}
 675
 676	/*
 677	 * Store md for cleanup instead of tio which is about to get freed.
 678	 */
 679	bio->bi_private = md->bs;
 680
 681	free_tio(md, tio);
 682	bio_put(bio);
 683	dec_pending(io, error);
 684}
 685
 686/*
 687 * Partial completion handling for request-based dm
 
 688 */
 689static void end_clone_bio(struct bio *clone, int error)
 690{
 691	struct dm_rq_clone_bio_info *info = clone->bi_private;
 692	struct dm_rq_target_io *tio = info->tio;
 693	struct bio *bio = info->orig;
 694	unsigned int nr_bytes = info->orig->bi_size;
 695
 696	bio_put(clone);
 697
 698	if (tio->error)
 699		/*
 700		 * An error has already been detected on the request.
 701		 * Once error occurred, just let clone->end_io() handle
 702		 * the remainder.
 703		 */
 704		return;
 705	else if (error) {
 706		/*
 707		 * Don't notice the error to the upper layer yet.
 708		 * The error handling decision is made by the target driver,
 709		 * when the request is completed.
 710		 */
 711		tio->error = error;
 712		return;
 713	}
 714
 715	/*
 716	 * I/O for the bio successfully completed.
 717	 * Notice the data completion to the upper layer.
 718	 */
 719
 720	/*
 721	 * bios are processed from the head of the list.
 722	 * So the completing bio should always be rq->bio.
 723	 * If it's not, something wrong is happening.
 724	 */
 725	if (tio->orig->bio != bio)
 726		DMERR("bio completion is going in the middle of the request");
 727
 728	/*
 729	 * Update the original request.
 730	 * Do not use blk_end_request() here, because it may complete
 731	 * the original request before the clone, and break the ordering.
 732	 */
 733	blk_update_request(tio->orig, 0, nr_bytes);
 734}
 735
 736/*
 737 * Don't touch any member of the md after calling this function because
 738 * the md may be freed in dm_put() at the end of this function.
 739 * Or do dm_get() before calling this function and dm_put() later.
 740 */
 741static void rq_completed(struct mapped_device *md, int rw, int run_queue)
 742{
 743	atomic_dec(&md->pending[rw]);
 744
 745	/* nudge anyone waiting on suspend queue */
 746	if (!md_in_flight(md))
 747		wake_up(&md->wait);
 748
 749	if (run_queue)
 750		blk_run_queue(md->queue);
 751
 752	/*
 753	 * dm_put() must be at the end of this function. See the comment above
 754	 */
 755	dm_put(md);
 756}
 757
 758static void free_rq_clone(struct request *clone)
 759{
 760	struct dm_rq_target_io *tio = clone->end_io_data;
 761
 762	blk_rq_unprep_clone(clone);
 763	free_rq_tio(tio);
 764}
 765
 766/*
 767 * Complete the clone and the original request.
 768 * Must be called without queue lock.
 769 */
 770static void dm_end_request(struct request *clone, int error)
 771{
 772	int rw = rq_data_dir(clone);
 773	struct dm_rq_target_io *tio = clone->end_io_data;
 774	struct mapped_device *md = tio->md;
 775	struct request *rq = tio->orig;
 776
 777	if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
 778		rq->errors = clone->errors;
 779		rq->resid_len = clone->resid_len;
 780
 781		if (rq->sense)
 782			/*
 783			 * We are using the sense buffer of the original
 784			 * request.
 785			 * So setting the length of the sense data is enough.
 786			 */
 787			rq->sense_len = clone->sense_len;
 788	}
 789
 790	free_rq_clone(clone);
 791	blk_end_request_all(rq, error);
 792	rq_completed(md, rw, true);
 793}
 794
 795static void dm_unprep_request(struct request *rq)
 796{
 797	struct request *clone = rq->special;
 
 
 
 
 
 798
 799	rq->special = NULL;
 800	rq->cmd_flags &= ~REQ_DONTPREP;
 801
 802	free_rq_clone(clone);
 803}
 
 804
 805/*
 806 * Requeue the original request of a clone.
 807 */
 808void dm_requeue_unmapped_request(struct request *clone)
 809{
 810	int rw = rq_data_dir(clone);
 811	struct dm_rq_target_io *tio = clone->end_io_data;
 812	struct mapped_device *md = tio->md;
 813	struct request *rq = tio->orig;
 814	struct request_queue *q = rq->q;
 815	unsigned long flags;
 816
 817	dm_unprep_request(rq);
 818
 819	spin_lock_irqsave(q->queue_lock, flags);
 820	blk_requeue_request(q, rq);
 821	spin_unlock_irqrestore(q->queue_lock, flags);
 822
 823	rq_completed(md, rw, 0);
 824}
 825EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
 826
 827static void __stop_queue(struct request_queue *q)
 828{
 829	blk_stop_queue(q);
 830}
 831
 832static void stop_queue(struct request_queue *q)
 
 833{
 834	unsigned long flags;
 
 
 
 
 835
 836	spin_lock_irqsave(q->queue_lock, flags);
 837	__stop_queue(q);
 838	spin_unlock_irqrestore(q->queue_lock, flags);
 839}
 840
 841static void __start_queue(struct request_queue *q)
 842{
 843	if (blk_queue_stopped(q))
 844		blk_start_queue(q);
 845}
 
 
 
 
 846
 847static void start_queue(struct request_queue *q)
 848{
 849	unsigned long flags;
 850
 851	spin_lock_irqsave(q->queue_lock, flags);
 852	__start_queue(q);
 853	spin_unlock_irqrestore(q->queue_lock, flags);
 854}
 855
 856static void dm_done(struct request *clone, int error, bool mapped)
 
 857{
 858	int r = error;
 859	struct dm_rq_target_io *tio = clone->end_io_data;
 860	dm_request_endio_fn rq_end_io = tio->ti->type->rq_end_io;
 861
 862	if (mapped && rq_end_io)
 863		r = rq_end_io(tio->ti, clone, error, &tio->info);
 864
 865	if (r <= 0)
 866		/* The target wants to complete the I/O */
 867		dm_end_request(clone, r);
 868	else if (r == DM_ENDIO_INCOMPLETE)
 869		/* The target will handle the I/O */
 870		return;
 871	else if (r == DM_ENDIO_REQUEUE)
 872		/* The target wants to requeue the I/O */
 873		dm_requeue_unmapped_request(clone);
 874	else {
 875		DMWARN("unimplemented target endio return value: %d", r);
 876		BUG();
 877	}
 878}
 879
 880/*
 881 * Request completion handler for request-based dm
 882 */
 883static void dm_softirq_done(struct request *rq)
 884{
 885	bool mapped = true;
 886	struct request *clone = rq->completion_data;
 887	struct dm_rq_target_io *tio = clone->end_io_data;
 888
 889	if (rq->cmd_flags & REQ_FAILED)
 890		mapped = false;
 891
 892	dm_done(clone, tio->error, mapped);
 893}
 894
 895/*
 896 * Complete the clone and the original request with the error status
 897 * through softirq context.
 898 */
 899static void dm_complete_request(struct request *clone, int error)
 900{
 901	struct dm_rq_target_io *tio = clone->end_io_data;
 902	struct request *rq = tio->orig;
 
 
 
 903
 904	tio->error = error;
 905	rq->completion_data = clone;
 906	blk_complete_request(rq);
 907}
 908
 909/*
 910 * Complete the not-mapped clone and the original request with the error status
 911 * through softirq context.
 912 * Target's rq_end_io() function isn't called.
 913 * This may be used when the target's map_rq() function fails.
 914 */
 915void dm_kill_unmapped_request(struct request *clone, int error)
 916{
 917	struct dm_rq_target_io *tio = clone->end_io_data;
 918	struct request *rq = tio->orig;
 919
 920	rq->cmd_flags |= REQ_FAILED;
 921	dm_complete_request(clone, error);
 922}
 923EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
 924
 925/*
 926 * Called with the queue lock held
 927 */
 928static void end_clone_request(struct request *clone, int error)
 929{
 930	/*
 931	 * For just cleaning up the information of the queue in which
 932	 * the clone was dispatched.
 933	 * The clone is *NOT* freed actually here because it is alloced from
 934	 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
 935	 */
 936	__blk_put_request(clone->q, clone);
 937
 938	/*
 939	 * Actual request completion is done in a softirq context which doesn't
 940	 * hold the queue lock.  Otherwise, deadlock could occur because:
 941	 *     - another request may be submitted by the upper level driver
 942	 *       of the stacking during the completion
 943	 *     - the submission which requires queue lock may be done
 944	 *       against this queue
 945	 */
 946	dm_complete_request(clone, error);
 947}
 948
 949/*
 950 * Return maximum size of I/O possible at the supplied sector up to the current
 951 * target boundary.
 952 */
 953static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
 954{
 955	sector_t target_offset = dm_target_offset(ti, sector);
 
 
 956
 957	return ti->len - target_offset;
 958}
 959
 960static sector_t max_io_len(sector_t sector, struct dm_target *ti)
 961{
 962	sector_t len = max_io_len_target_boundary(sector, ti);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 963
 964	/*
 965	 * Does the target need to split even further ?
 966	 */
 967	if (ti->split_io) {
 968		sector_t boundary;
 969		sector_t offset = dm_target_offset(ti, sector);
 970		boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
 971			   - offset;
 972		if (len > boundary)
 973			len = boundary;
 974	}
 975
 976	return len;
 
 
 
 977}
 
 978
 979static void __map_bio(struct dm_target *ti, struct bio *clone,
 980		      struct dm_target_io *tio)
 981{
 982	int r;
 983	sector_t sector;
 984	struct mapped_device *md;
 
 
 
 
 985
 986	clone->bi_end_io = clone_endio;
 987	clone->bi_private = tio;
 988
 989	/*
 990	 * Map the clone.  If r == 0 we don't need to do
 991	 * anything, the target has assumed ownership of
 992	 * this io.
 993	 */
 994	atomic_inc(&tio->io->io_count);
 995	sector = clone->bi_sector;
 996	r = ti->type->map(ti, clone, &tio->info);
 997	if (r == DM_MAPIO_REMAPPED) {
 998		/* the bio has been remapped so dispatch it */
 999
1000		trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1001				      tio->io->bio->bi_bdev->bd_dev, sector);
1002
1003		generic_make_request(clone);
1004	} else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1005		/* error the io and bail out, or requeue it if needed */
1006		md = tio->io->md;
1007		dec_pending(tio->io, r);
1008		/*
1009		 * Store bio_set for cleanup.
1010		 */
1011		clone->bi_private = md->bs;
1012		bio_put(clone);
1013		free_tio(md, tio);
1014	} else if (r) {
 
 
 
 
 
 
 
1015		DMWARN("unimplemented target map return value: %d", r);
1016		BUG();
1017	}
1018}
1019
1020struct clone_info {
1021	struct mapped_device *md;
1022	struct dm_table *map;
1023	struct bio *bio;
1024	struct dm_io *io;
1025	sector_t sector;
1026	sector_t sector_count;
1027	unsigned short idx;
1028};
1029
1030static void dm_bio_destructor(struct bio *bio)
1031{
1032	struct bio_set *bs = bio->bi_private;
1033
1034	bio_free(bio, bs);
1035}
1036
1037/*
1038 * Creates a little bio that just does part of a bvec.
1039 */
1040static struct bio *split_bvec(struct bio *bio, sector_t sector,
1041			      unsigned short idx, unsigned int offset,
1042			      unsigned int len, struct bio_set *bs)
1043{
1044	struct bio *clone;
1045	struct bio_vec *bv = bio->bi_io_vec + idx;
1046
1047	clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
1048	clone->bi_destructor = dm_bio_destructor;
1049	*clone->bi_io_vec = *bv;
1050
1051	clone->bi_sector = sector;
1052	clone->bi_bdev = bio->bi_bdev;
1053	clone->bi_rw = bio->bi_rw;
1054	clone->bi_vcnt = 1;
1055	clone->bi_size = to_bytes(len);
1056	clone->bi_io_vec->bv_offset = offset;
1057	clone->bi_io_vec->bv_len = clone->bi_size;
1058	clone->bi_flags |= 1 << BIO_CLONED;
1059
1060	if (bio_integrity(bio)) {
1061		bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1062		bio_integrity_trim(clone,
1063				   bio_sector_offset(bio, idx, offset), len);
1064	}
1065
1066	return clone;
1067}
1068
1069/*
1070 * Creates a bio that consists of range of complete bvecs.
1071 */
1072static struct bio *clone_bio(struct bio *bio, sector_t sector,
1073			     unsigned short idx, unsigned short bv_count,
1074			     unsigned int len, struct bio_set *bs)
1075{
1076	struct bio *clone;
1077
1078	clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
1079	__bio_clone(clone, bio);
1080	clone->bi_destructor = dm_bio_destructor;
1081	clone->bi_sector = sector;
1082	clone->bi_idx = idx;
1083	clone->bi_vcnt = idx + bv_count;
1084	clone->bi_size = to_bytes(len);
1085	clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1086
1087	if (bio_integrity(bio)) {
1088		bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1089
1090		if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1091			bio_integrity_trim(clone,
1092					   bio_sector_offset(bio, idx, 0), len);
1093	}
 
 
 
1094
1095	return clone;
1096}
 
 
1097
1098static struct dm_target_io *alloc_tio(struct clone_info *ci,
1099				      struct dm_target *ti)
1100{
1101	struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO);
1102
1103	tio->io = ci->io;
1104	tio->ti = ti;
1105	memset(&tio->info, 0, sizeof(tio->info));
1106
1107	return tio;
1108}
1109
1110static void __issue_target_request(struct clone_info *ci, struct dm_target *ti,
1111				   unsigned request_nr, sector_t len)
1112{
1113	struct dm_target_io *tio = alloc_tio(ci, ti);
1114	struct bio *clone;
1115
1116	tio->info.target_request_nr = request_nr;
 
1117
1118	/*
1119	 * Discard requests require the bio's inline iovecs be initialized.
1120	 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1121	 * and discard, so no need for concern about wasted bvec allocations.
1122	 */
1123	clone = bio_alloc_bioset(GFP_NOIO, ci->bio->bi_max_vecs, ci->md->bs);
1124	__bio_clone(clone, ci->bio);
1125	clone->bi_destructor = dm_bio_destructor;
1126	if (len) {
1127		clone->bi_sector = ci->sector;
1128		clone->bi_size = to_bytes(len);
1129	}
1130
1131	__map_bio(ti, clone, tio);
1132}
 
 
 
 
 
 
 
 
1133
1134static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti,
1135				    unsigned num_requests, sector_t len)
1136{
1137	unsigned request_nr;
 
 
1138
1139	for (request_nr = 0; request_nr < num_requests; request_nr++)
1140		__issue_target_request(ci, ti, request_nr, len);
 
 
 
1141}
1142
1143static int __clone_and_map_empty_flush(struct clone_info *ci)
 
1144{
1145	unsigned target_nr = 0;
1146	struct dm_target *ti;
1147
1148	BUG_ON(bio_has_data(ci->bio));
1149	while ((ti = dm_table_get_target(ci->map, target_nr++)))
1150		__issue_target_requests(ci, ti, ti->num_flush_requests, 0);
1151
1152	return 0;
 
 
 
 
1153}
1154
1155/*
1156 * Perform all io with a single clone.
1157 */
1158static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti)
1159{
1160	struct bio *clone, *bio = ci->bio;
 
1161	struct dm_target_io *tio;
1162
1163	tio = alloc_tio(ci, ti);
1164	clone = clone_bio(bio, ci->sector, ci->idx,
1165			  bio->bi_vcnt - ci->idx, ci->sector_count,
1166			  ci->md->bs);
1167	__map_bio(ti, clone, tio);
1168	ci->sector_count = 0;
1169}
1170
1171static int __clone_and_map_discard(struct clone_info *ci)
1172{
 
1173	struct dm_target *ti;
1174	sector_t len;
1175
1176	do {
1177		ti = dm_table_find_target(ci->map, ci->sector);
1178		if (!dm_target_is_valid(ti))
1179			return -EIO;
1180
1181		/*
1182		 * Even though the device advertised discard support,
1183		 * that does not mean every target supports it, and
1184		 * reconfiguration might also have changed that since the
1185		 * check was performed.
1186		 */
1187		if (!ti->num_discard_requests)
1188			return -EOPNOTSUPP;
1189
1190		len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
 
 
 
 
 
 
 
1191
1192		__issue_target_requests(ci, ti, ti->num_discard_requests, len);
 
 
1193
1194		ci->sector += len;
1195	} while (ci->sector_count -= len);
1196
1197	return 0;
1198}
1199
1200static int __clone_and_map(struct clone_info *ci)
 
1201{
1202	struct bio *clone, *bio = ci->bio;
1203	struct dm_target *ti;
1204	sector_t len = 0, max;
1205	struct dm_target_io *tio;
 
1206
1207	if (unlikely(bio->bi_rw & REQ_DISCARD))
1208		return __clone_and_map_discard(ci);
1209
1210	ti = dm_table_find_target(ci->map, ci->sector);
1211	if (!dm_target_is_valid(ti))
1212		return -EIO;
1213
1214	max = max_io_len(ci->sector, ti);
1215
1216	if (ci->sector_count <= max) {
1217		/*
1218		 * Optimise for the simple case where we can do all of
1219		 * the remaining io with a single clone.
1220		 */
1221		__clone_and_map_simple(ci, ti);
1222
1223	} else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1224		/*
1225		 * There are some bvecs that don't span targets.
1226		 * Do as many of these as possible.
1227		 */
1228		int i;
1229		sector_t remaining = max;
1230		sector_t bv_len;
1231
1232		for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1233			bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1234
1235			if (bv_len > remaining)
1236				break;
1237
1238			remaining -= bv_len;
1239			len += bv_len;
1240		}
1241
1242		tio = alloc_tio(ci, ti);
1243		clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
1244				  ci->md->bs);
1245		__map_bio(ti, clone, tio);
1246
1247		ci->sector += len;
1248		ci->sector_count -= len;
1249		ci->idx = i;
1250
1251	} else {
1252		/*
1253		 * Handle a bvec that must be split between two or more targets.
1254		 */
1255		struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1256		sector_t remaining = to_sector(bv->bv_len);
1257		unsigned int offset = 0;
1258
1259		do {
1260			if (offset) {
1261				ti = dm_table_find_target(ci->map, ci->sector);
1262				if (!dm_target_is_valid(ti))
1263					return -EIO;
1264
1265				max = max_io_len(ci->sector, ti);
1266			}
1267
1268			len = min(remaining, max);
1269
1270			tio = alloc_tio(ci, ti);
1271			clone = split_bvec(bio, ci->sector, ci->idx,
1272					   bv->bv_offset + offset, len,
1273					   ci->md->bs);
1274
1275			__map_bio(ti, clone, tio);
1276
1277			ci->sector += len;
1278			ci->sector_count -= len;
1279			offset += to_bytes(len);
1280		} while (remaining -= len);
1281
1282		ci->idx++;
1283	}
 
1284
1285	return 0;
1286}
1287
1288/*
1289 * Split the bio into several clones and submit it to targets.
1290 */
1291static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1292{
1293	struct clone_info ci;
1294	int error = 0;
1295
1296	ci.map = dm_get_live_table(md);
1297	if (unlikely(!ci.map)) {
1298		bio_io_error(bio);
1299		return;
1300	}
1301
1302	ci.md = md;
1303	ci.io = alloc_io(md);
1304	ci.io->error = 0;
1305	atomic_set(&ci.io->io_count, 1);
1306	ci.io->bio = bio;
1307	ci.io->md = md;
1308	spin_lock_init(&ci.io->endio_lock);
1309	ci.sector = bio->bi_sector;
1310	ci.idx = bio->bi_idx;
1311
1312	start_io_acct(ci.io);
1313	if (bio->bi_rw & REQ_FLUSH) {
1314		ci.bio = &ci.md->flush_bio;
1315		ci.sector_count = 0;
1316		error = __clone_and_map_empty_flush(&ci);
1317		/* dec_pending submits any data associated with flush */
1318	} else {
1319		ci.bio = bio;
1320		ci.sector_count = bio_sectors(bio);
1321		while (ci.sector_count && !error)
1322			error = __clone_and_map(&ci);
1323	}
1324
1325	/* drop the extra reference count */
1326	dec_pending(ci.io, error);
1327	dm_table_put(ci.map);
1328}
1329/*-----------------------------------------------------------------
1330 * CRUD END
1331 *---------------------------------------------------------------*/
1332
1333static int dm_merge_bvec(struct request_queue *q,
1334			 struct bvec_merge_data *bvm,
1335			 struct bio_vec *biovec)
1336{
1337	struct mapped_device *md = q->queuedata;
1338	struct dm_table *map = dm_get_live_table(md);
1339	struct dm_target *ti;
1340	sector_t max_sectors;
1341	int max_size = 0;
1342
1343	if (unlikely(!map))
1344		goto out;
 
 
1345
1346	ti = dm_table_find_target(map, bvm->bi_sector);
1347	if (!dm_target_is_valid(ti))
1348		goto out_table;
 
1349
1350	/*
1351	 * Find maximum amount of I/O that won't need splitting
1352	 */
1353	max_sectors = min(max_io_len(bvm->bi_sector, ti),
1354			  (sector_t) BIO_MAX_SECTORS);
1355	max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1356	if (max_size < 0)
1357		max_size = 0;
1358
1359	/*
1360	 * merge_bvec_fn() returns number of bytes
1361	 * it can accept at this offset
1362	 * max is precomputed maximal io size
 
1363	 */
1364	if (max_size && ti->type->merge)
1365		max_size = ti->type->merge(ti, bvm, biovec, max_size);
1366	/*
1367	 * If the target doesn't support merge method and some of the devices
1368	 * provided their merge_bvec method (we know this by looking at
1369	 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1370	 * entries.  So always set max_size to 0, and the code below allows
1371	 * just one page.
1372	 */
1373	else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1374
1375		max_size = 0;
1376
1377out_table:
1378	dm_table_put(map);
1379
1380out:
1381	/*
1382	 * Always allow an entire first page
1383	 */
1384	if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1385		max_size = biovec->bv_len;
1386
1387	return max_size;
1388}
1389
1390/*
1391 * The request function that just remaps the bio built up by
1392 * dm_merge_bvec.
1393 */
1394static int _dm_request(struct request_queue *q, struct bio *bio)
1395{
1396	int rw = bio_data_dir(bio);
1397	struct mapped_device *md = q->queuedata;
1398	int cpu;
1399
1400	down_read(&md->io_lock);
1401
1402	cpu = part_stat_lock();
1403	part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1404	part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1405	part_stat_unlock();
1406
1407	/* if we're suspended, we have to queue this io for later */
1408	if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1409		up_read(&md->io_lock);
1410
1411		if (bio_rw(bio) != READA)
1412			queue_io(md, bio);
1413		else
1414			bio_io_error(bio);
1415		return 0;
1416	}
1417
1418	__split_and_process_bio(md, bio);
1419	up_read(&md->io_lock);
1420	return 0;
1421}
1422
1423static int dm_make_request(struct request_queue *q, struct bio *bio)
1424{
1425	struct mapped_device *md = q->queuedata;
1426
1427	return md->saved_make_request_fn(q, bio); /* call __make_request() */
1428}
1429
1430static int dm_request_based(struct mapped_device *md)
1431{
1432	return blk_queue_stackable(md->queue);
1433}
1434
1435static int dm_request(struct request_queue *q, struct bio *bio)
1436{
1437	struct mapped_device *md = q->queuedata;
1438
1439	if (dm_request_based(md))
1440		return dm_make_request(q, bio);
1441
1442	return _dm_request(q, bio);
1443}
1444
1445void dm_dispatch_request(struct request *rq)
1446{
1447	int r;
1448
1449	if (blk_queue_io_stat(rq->q))
1450		rq->cmd_flags |= REQ_IO_STAT;
 
 
 
 
 
 
1451
1452	rq->start_time = jiffies;
1453	r = blk_insert_cloned_request(rq->q, rq);
1454	if (r)
1455		dm_complete_request(rq, r);
1456}
1457EXPORT_SYMBOL_GPL(dm_dispatch_request);
1458
1459static void dm_rq_bio_destructor(struct bio *bio)
 
1460{
1461	struct dm_rq_clone_bio_info *info = bio->bi_private;
1462	struct mapped_device *md = info->tio->md;
1463
1464	free_bio_info(info);
1465	bio_free(bio, md->bs);
 
 
 
 
 
 
 
 
 
 
1466}
1467
1468static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1469				 void *data)
 
 
1470{
1471	struct dm_rq_target_io *tio = data;
1472	struct mapped_device *md = tio->md;
1473	struct dm_rq_clone_bio_info *info = alloc_bio_info(md);
1474
1475	if (!info)
1476		return -ENOMEM;
1477
1478	info->orig = bio_orig;
1479	info->tio = tio;
1480	bio->bi_end_io = end_clone_bio;
1481	bio->bi_private = info;
1482	bio->bi_destructor = dm_rq_bio_destructor;
1483
1484	return 0;
1485}
1486
1487static int setup_clone(struct request *clone, struct request *rq,
1488		       struct dm_rq_target_io *tio)
1489{
1490	int r;
1491
1492	r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1493			      dm_rq_bio_constructor, tio);
1494	if (r)
1495		return r;
1496
1497	clone->cmd = rq->cmd;
1498	clone->cmd_len = rq->cmd_len;
1499	clone->sense = rq->sense;
1500	clone->buffer = rq->buffer;
1501	clone->end_io = end_clone_request;
1502	clone->end_io_data = tio;
1503
1504	return 0;
1505}
1506
1507static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1508				gfp_t gfp_mask)
1509{
1510	struct request *clone;
1511	struct dm_rq_target_io *tio;
1512
1513	tio = alloc_rq_tio(md, gfp_mask);
1514	if (!tio)
1515		return NULL;
1516
1517	tio->md = md;
1518	tio->ti = NULL;
1519	tio->orig = rq;
1520	tio->error = 0;
1521	memset(&tio->info, 0, sizeof(tio->info));
1522
1523	clone = &tio->clone;
1524	if (setup_clone(clone, rq, tio)) {
1525		/* -ENOMEM */
1526		free_rq_tio(tio);
1527		return NULL;
1528	}
1529
1530	return clone;
1531}
1532
 
 
 
1533/*
1534 * Called with the queue lock held.
1535 */
1536static int dm_prep_fn(struct request_queue *q, struct request *rq)
 
1537{
1538	struct mapped_device *md = q->queuedata;
1539	struct request *clone;
 
1540
1541	if (unlikely(rq->special)) {
1542		DMWARN("Already has something in rq->special.");
1543		return BLKPREP_KILL;
1544	}
1545
1546	clone = clone_rq(rq, md, GFP_ATOMIC);
1547	if (!clone)
1548		return BLKPREP_DEFER;
1549
1550	rq->special = clone;
1551	rq->cmd_flags |= REQ_DONTPREP;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1552
1553	return BLKPREP_OK;
 
 
1554}
1555
1556/*
1557 * Returns:
1558 * 0  : the request has been processed (not requeued)
1559 * !0 : the request has been requeued
1560 */
1561static int map_request(struct dm_target *ti, struct request *clone,
1562		       struct mapped_device *md)
1563{
1564	int r, requeued = 0;
1565	struct dm_rq_target_io *tio = clone->end_io_data;
 
1566
1567	/*
1568	 * Hold the md reference here for the in-flight I/O.
1569	 * We can't rely on the reference count by device opener,
1570	 * because the device may be closed during the request completion
1571	 * when all bios are completed.
1572	 * See the comment in rq_completed() too.
1573	 */
1574	dm_get(md);
1575
1576	tio->ti = ti;
1577	r = ti->type->map_rq(ti, clone, &tio->info);
1578	switch (r) {
1579	case DM_MAPIO_SUBMITTED:
1580		/* The target has taken the I/O to submit by itself later */
1581		break;
1582	case DM_MAPIO_REMAPPED:
1583		/* The target has remapped the I/O so dispatch it */
1584		trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1585				     blk_rq_pos(tio->orig));
1586		dm_dispatch_request(clone);
1587		break;
1588	case DM_MAPIO_REQUEUE:
1589		/* The target wants to requeue the I/O */
1590		dm_requeue_unmapped_request(clone);
1591		requeued = 1;
1592		break;
1593	default:
1594		if (r > 0) {
1595			DMWARN("unimplemented target map return value: %d", r);
1596			BUG();
1597		}
1598
1599		/* The target wants to complete the I/O */
1600		dm_kill_unmapped_request(clone, r);
1601		break;
1602	}
 
 
 
 
 
 
 
 
 
1603
1604	return requeued;
 
 
 
 
 
 
 
 
 
 
 
1605}
1606
1607/*
1608 * q->request_fn for request-based dm.
1609 * Called with the queue lock held.
1610 */
1611static void dm_request_fn(struct request_queue *q)
1612{
1613	struct mapped_device *md = q->queuedata;
1614	struct dm_table *map = dm_get_live_table(md);
1615	struct dm_target *ti;
1616	struct request *rq, *clone;
1617	sector_t pos;
1618
1619	/*
1620	 * For suspend, check blk_queue_stopped() and increment
1621	 * ->pending within a single queue_lock not to increment the
1622	 * number of in-flight I/Os after the queue is stopped in
1623	 * dm_suspend().
1624	 */
1625	while (!blk_queue_stopped(q)) {
1626		rq = blk_peek_request(q);
1627		if (!rq)
1628			goto delay_and_out;
1629
1630		/* always use block 0 to find the target for flushes for now */
1631		pos = 0;
1632		if (!(rq->cmd_flags & REQ_FLUSH))
1633			pos = blk_rq_pos(rq);
1634
1635		ti = dm_table_find_target(map, pos);
1636		BUG_ON(!dm_target_is_valid(ti));
1637
1638		if (ti->type->busy && ti->type->busy(ti))
1639			goto delay_and_out;
1640
1641		blk_start_request(rq);
1642		clone = rq->special;
1643		atomic_inc(&md->pending[rq_data_dir(clone)]);
1644
1645		spin_unlock(q->queue_lock);
1646		if (map_request(ti, clone, md))
1647			goto requeued;
1648
1649		BUG_ON(!irqs_disabled());
1650		spin_lock(q->queue_lock);
1651	}
 
1652
1653	goto out;
 
 
 
 
1654
1655requeued:
1656	BUG_ON(!irqs_disabled());
1657	spin_lock(q->queue_lock);
 
1658
1659delay_and_out:
1660	blk_delay_queue(q, HZ / 10);
1661out:
1662	dm_table_put(map);
 
 
 
1663
1664	return;
1665}
 
 
 
 
 
 
 
 
1666
1667int dm_underlying_device_busy(struct request_queue *q)
1668{
1669	return blk_lld_busy(q);
 
1670}
1671EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1672
1673static int dm_lld_busy(struct request_queue *q)
1674{
1675	int r;
1676	struct mapped_device *md = q->queuedata;
1677	struct dm_table *map = dm_get_live_table(md);
 
 
1678
1679	if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1680		r = 1;
1681	else
1682		r = dm_table_any_busy_target(map);
1683
1684	dm_table_put(map);
 
 
1685
1686	return r;
 
 
 
 
 
 
 
 
 
 
1687}
1688
1689static int dm_any_congested(void *congested_data, int bdi_bits)
1690{
1691	int r = bdi_bits;
1692	struct mapped_device *md = congested_data;
1693	struct dm_table *map;
1694
1695	if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1696		map = dm_get_live_table(md);
1697		if (map) {
1698			/*
1699			 * Request-based dm cares about only own queue for
1700			 * the query about congestion status of request_queue
1701			 */
1702			if (dm_request_based(md))
1703				r = md->queue->backing_dev_info.state &
1704				    bdi_bits;
1705			else
1706				r = dm_table_any_congested(map, bdi_bits);
1707
1708			dm_table_put(map);
1709		}
1710	}
1711
1712	return r;
1713}
1714
1715/*-----------------------------------------------------------------
1716 * An IDR is used to keep track of allocated minor numbers.
1717 *---------------------------------------------------------------*/
1718static void free_minor(int minor)
1719{
1720	spin_lock(&_minor_lock);
1721	idr_remove(&_minor_idr, minor);
1722	spin_unlock(&_minor_lock);
1723}
1724
1725/*
1726 * See if the device with a specific minor # is free.
1727 */
1728static int specific_minor(int minor)
1729{
1730	int r, m;
1731
1732	if (minor >= (1 << MINORBITS))
1733		return -EINVAL;
1734
1735	r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1736	if (!r)
1737		return -ENOMEM;
1738
1739	spin_lock(&_minor_lock);
1740
1741	if (idr_find(&_minor_idr, minor)) {
1742		r = -EBUSY;
1743		goto out;
1744	}
1745
1746	r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1747	if (r)
1748		goto out;
1749
1750	if (m != minor) {
1751		idr_remove(&_minor_idr, m);
1752		r = -EBUSY;
1753		goto out;
1754	}
1755
1756out:
1757	spin_unlock(&_minor_lock);
1758	return r;
 
 
 
1759}
1760
1761static int next_free_minor(int *minor)
1762{
1763	int r, m;
1764
1765	r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1766	if (!r)
1767		return -ENOMEM;
1768
 
1769	spin_lock(&_minor_lock);
1770
1771	r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1772	if (r)
1773		goto out;
1774
1775	if (m >= (1 << MINORBITS)) {
1776		idr_remove(&_minor_idr, m);
1777		r = -ENOSPC;
1778		goto out;
1779	}
1780
1781	*minor = m;
1782
1783out:
1784	spin_unlock(&_minor_lock);
1785	return r;
 
 
 
 
1786}
1787
1788static const struct block_device_operations dm_blk_dops;
 
1789
1790static void dm_wq_work(struct work_struct *work);
1791
1792static void dm_init_md_queue(struct mapped_device *md)
1793{
1794	/*
1795	 * Request-based dm devices cannot be stacked on top of bio-based dm
1796	 * devices.  The type of this dm device has not been decided yet.
1797	 * The type is decided at the first table loading time.
1798	 * To prevent problematic device stacking, clear the queue flag
1799	 * for request stacking support until then.
1800	 *
1801	 * This queue is new, so no concurrency on the queue_flags.
1802	 */
1803	queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
 
1804
1805	md->queue->queuedata = md;
1806	md->queue->backing_dev_info.congested_fn = dm_any_congested;
1807	md->queue->backing_dev_info.congested_data = md;
1808	blk_queue_make_request(md->queue, dm_request);
1809	blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1810	blk_queue_merge_bvec(md->queue, dm_merge_bvec);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1811}
1812
1813/*
1814 * Allocate and initialise a blank device with a given minor.
1815 */
1816static struct mapped_device *alloc_dev(int minor)
1817{
1818	int r;
1819	struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1820	void *old_md;
1821
 
1822	if (!md) {
1823		DMWARN("unable to allocate device, out of memory.");
1824		return NULL;
1825	}
1826
1827	if (!try_module_get(THIS_MODULE))
1828		goto bad_module_get;
1829
1830	/* get a minor number for the dev */
1831	if (minor == DM_ANY_MINOR)
1832		r = next_free_minor(&minor);
1833	else
1834		r = specific_minor(minor);
1835	if (r < 0)
1836		goto bad_minor;
1837
 
 
 
 
 
 
1838	md->type = DM_TYPE_NONE;
1839	init_rwsem(&md->io_lock);
1840	mutex_init(&md->suspend_lock);
1841	mutex_init(&md->type_lock);
 
1842	spin_lock_init(&md->deferred_lock);
1843	rwlock_init(&md->map_lock);
1844	atomic_set(&md->holders, 1);
1845	atomic_set(&md->open_count, 0);
1846	atomic_set(&md->event_nr, 0);
1847	atomic_set(&md->uevent_seq, 0);
1848	INIT_LIST_HEAD(&md->uevent_list);
 
1849	spin_lock_init(&md->uevent_lock);
1850
1851	md->queue = blk_alloc_queue(GFP_KERNEL);
1852	if (!md->queue)
1853		goto bad_queue;
1854
1855	dm_init_md_queue(md);
1856
1857	md->disk = alloc_disk(1);
1858	if (!md->disk)
1859		goto bad_disk;
1860
1861	atomic_set(&md->pending[0], 0);
1862	atomic_set(&md->pending[1], 0);
1863	init_waitqueue_head(&md->wait);
1864	INIT_WORK(&md->work, dm_wq_work);
1865	init_waitqueue_head(&md->eventq);
 
1866
1867	md->disk->major = _major;
1868	md->disk->first_minor = minor;
1869	md->disk->fops = &dm_blk_dops;
1870	md->disk->queue = md->queue;
1871	md->disk->private_data = md;
1872	sprintf(md->disk->disk_name, "dm-%d", minor);
1873	add_disk(md->disk);
 
 
 
 
 
 
 
 
1874	format_dev_t(md->name, MKDEV(_major, minor));
1875
1876	md->wq = alloc_workqueue("kdmflush",
1877				 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
1878	if (!md->wq)
1879		goto bad_thread;
1880
1881	md->bdev = bdget_disk(md->disk, 0);
1882	if (!md->bdev)
1883		goto bad_bdev;
1884
1885	bio_init(&md->flush_bio);
1886	md->flush_bio.bi_bdev = md->bdev;
1887	md->flush_bio.bi_rw = WRITE_FLUSH;
1888
1889	/* Populate the mapping, nobody knows we exist yet */
1890	spin_lock(&_minor_lock);
1891	old_md = idr_replace(&_minor_idr, md, minor);
1892	spin_unlock(&_minor_lock);
1893
1894	BUG_ON(old_md != MINOR_ALLOCED);
1895
1896	return md;
1897
1898bad_bdev:
1899	destroy_workqueue(md->wq);
1900bad_thread:
1901	del_gendisk(md->disk);
1902	put_disk(md->disk);
1903bad_disk:
1904	blk_cleanup_queue(md->queue);
1905bad_queue:
1906	free_minor(minor);
1907bad_minor:
1908	module_put(THIS_MODULE);
1909bad_module_get:
1910	kfree(md);
1911	return NULL;
1912}
1913
1914static void unlock_fs(struct mapped_device *md);
1915
1916static void free_dev(struct mapped_device *md)
1917{
1918	int minor = MINOR(disk_devt(md->disk));
1919
1920	unlock_fs(md);
1921	bdput(md->bdev);
1922	destroy_workqueue(md->wq);
1923	if (md->tio_pool)
1924		mempool_destroy(md->tio_pool);
1925	if (md->io_pool)
1926		mempool_destroy(md->io_pool);
1927	if (md->bs)
1928		bioset_free(md->bs);
1929	blk_integrity_unregister(md->disk);
1930	del_gendisk(md->disk);
1931	free_minor(minor);
1932
1933	spin_lock(&_minor_lock);
1934	md->disk->private_data = NULL;
1935	spin_unlock(&_minor_lock);
 
 
1936
1937	put_disk(md->disk);
1938	blk_cleanup_queue(md->queue);
1939	module_put(THIS_MODULE);
1940	kfree(md);
1941}
1942
1943static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1944{
1945	struct dm_md_mempools *p;
 
1946
1947	if (md->io_pool && md->tio_pool && md->bs)
1948		/* the md already has necessary mempools */
1949		goto out;
 
 
 
 
 
1950
1951	p = dm_table_get_md_mempools(t);
1952	BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
 
 
 
 
 
 
 
 
 
1953
1954	md->io_pool = p->io_pool;
1955	p->io_pool = NULL;
1956	md->tio_pool = p->tio_pool;
1957	p->tio_pool = NULL;
1958	md->bs = p->bs;
1959	p->bs = NULL;
1960
 
 
 
 
 
 
1961out:
1962	/* mempool bind completed, now no need any mempools in the table */
1963	dm_table_free_md_mempools(t);
 
1964}
1965
1966/*
1967 * Bind a table to the device.
1968 */
1969static void event_callback(void *context)
1970{
1971	unsigned long flags;
1972	LIST_HEAD(uevents);
1973	struct mapped_device *md = (struct mapped_device *) context;
1974
1975	spin_lock_irqsave(&md->uevent_lock, flags);
1976	list_splice_init(&md->uevent_list, &uevents);
1977	spin_unlock_irqrestore(&md->uevent_lock, flags);
1978
1979	dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1980
1981	atomic_inc(&md->event_nr);
1982	wake_up(&md->eventq);
 
1983}
1984
1985/*
1986 * Protected by md->suspend_lock obtained by dm_swap_table().
1987 */
1988static void __set_size(struct mapped_device *md, sector_t size)
1989{
 
 
1990	set_capacity(md->disk, size);
1991
1992	i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1993}
1994
1995/*
1996 * Return 1 if the queue has a compulsory merge_bvec_fn function.
1997 *
1998 * If this function returns 0, then the device is either a non-dm
1999 * device without a merge_bvec_fn, or it is a dm device that is
2000 * able to split any bios it receives that are too big.
2001 */
2002int dm_queue_merge_is_compulsory(struct request_queue *q)
2003{
2004	struct mapped_device *dev_md;
2005
2006	if (!q->merge_bvec_fn)
2007		return 0;
2008
2009	if (q->make_request_fn == dm_request) {
2010		dev_md = q->queuedata;
2011		if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2012			return 0;
2013	}
2014
2015	return 1;
2016}
2017
2018static int dm_device_merge_is_compulsory(struct dm_target *ti,
2019					 struct dm_dev *dev, sector_t start,
2020					 sector_t len, void *data)
2021{
2022	struct block_device *bdev = dev->bdev;
2023	struct request_queue *q = bdev_get_queue(bdev);
2024
2025	return dm_queue_merge_is_compulsory(q);
2026}
2027
2028/*
2029 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2030 * on the properties of the underlying devices.
2031 */
2032static int dm_table_merge_is_optional(struct dm_table *table)
2033{
2034	unsigned i = 0;
2035	struct dm_target *ti;
2036
2037	while (i < dm_table_get_num_targets(table)) {
2038		ti = dm_table_get_target(table, i++);
2039
2040		if (ti->type->iterate_devices &&
2041		    ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2042			return 0;
2043	}
2044
2045	return 1;
2046}
2047
2048/*
2049 * Returns old map, which caller must destroy.
2050 */
2051static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2052			       struct queue_limits *limits)
2053{
2054	struct dm_table *old_map;
2055	struct request_queue *q = md->queue;
 
2056	sector_t size;
2057	unsigned long flags;
2058	int merge_is_optional;
 
2059
2060	size = dm_table_get_size(t);
2061
2062	/*
2063	 * Wipe any geometry if the size of the table changed.
2064	 */
2065	if (size != get_capacity(md->disk))
2066		memset(&md->geometry, 0, sizeof(md->geometry));
2067
2068	__set_size(md, size);
2069
2070	dm_table_event_callback(t, event_callback, md);
2071
2072	/*
2073	 * The queue hasn't been stopped yet, if the old table type wasn't
2074	 * for request-based during suspension.  So stop it to prevent
2075	 * I/O mapping before resume.
2076	 * This must be done before setting the queue restrictions,
2077	 * because request-based dm may be run just after the setting.
2078	 */
2079	if (dm_table_request_based(t) && !blk_queue_stopped(q))
2080		stop_queue(q);
2081
2082	__bind_mempools(md, t);
 
 
 
 
 
 
 
 
2083
2084	merge_is_optional = dm_table_merge_is_optional(t);
 
 
 
 
 
 
 
 
2085
2086	write_lock_irqsave(&md->map_lock, flags);
2087	old_map = md->map;
2088	md->map = t;
2089	dm_table_set_restrictions(t, q, limits);
2090	if (merge_is_optional)
2091		set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2092	else
2093		clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2094	write_unlock_irqrestore(&md->map_lock, flags);
2095
 
2096	return old_map;
2097}
2098
2099/*
2100 * Returns unbound table for the caller to free.
2101 */
2102static struct dm_table *__unbind(struct mapped_device *md)
2103{
2104	struct dm_table *map = md->map;
2105	unsigned long flags;
2106
2107	if (!map)
2108		return NULL;
2109
2110	dm_table_event_callback(map, NULL, NULL);
2111	write_lock_irqsave(&md->map_lock, flags);
2112	md->map = NULL;
2113	write_unlock_irqrestore(&md->map_lock, flags);
2114
2115	return map;
2116}
2117
2118/*
2119 * Constructor for a new device.
2120 */
2121int dm_create(int minor, struct mapped_device **result)
2122{
 
2123	struct mapped_device *md;
2124
2125	md = alloc_dev(minor);
2126	if (!md)
2127		return -ENXIO;
2128
2129	dm_sysfs_init(md);
 
 
 
 
2130
2131	*result = md;
2132	return 0;
2133}
2134
2135/*
2136 * Functions to manage md->type.
2137 * All are required to hold md->type_lock.
2138 */
2139void dm_lock_md_type(struct mapped_device *md)
2140{
2141	mutex_lock(&md->type_lock);
2142}
2143
2144void dm_unlock_md_type(struct mapped_device *md)
2145{
2146	mutex_unlock(&md->type_lock);
2147}
2148
2149void dm_set_md_type(struct mapped_device *md, unsigned type)
2150{
 
2151	md->type = type;
2152}
2153
2154unsigned dm_get_md_type(struct mapped_device *md)
2155{
2156	return md->type;
2157}
2158
 
 
 
 
 
2159/*
2160 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
 
2161 */
2162static int dm_init_request_based_queue(struct mapped_device *md)
2163{
2164	struct request_queue *q = NULL;
2165
2166	if (md->queue->elevator)
2167		return 1;
2168
2169	/* Fully initialize the queue */
2170	q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2171	if (!q)
2172		return 0;
2173
2174	md->queue = q;
2175	md->saved_make_request_fn = md->queue->make_request_fn;
2176	dm_init_md_queue(md);
2177	blk_queue_softirq_done(md->queue, dm_softirq_done);
2178	blk_queue_prep_rq(md->queue, dm_prep_fn);
2179	blk_queue_lld_busy(md->queue, dm_lld_busy);
2180
2181	elv_register_queue(md->queue);
2182
2183	return 1;
2184}
 
2185
2186/*
2187 * Setup the DM device's queue based on md's type
2188 */
2189int dm_setup_md_queue(struct mapped_device *md)
2190{
2191	if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2192	    !dm_init_request_based_queue(md)) {
2193		DMWARN("Cannot initialize queue for request-based mapped device");
2194		return -EINVAL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2195	}
 
 
2196
2197	return 0;
2198}
2199
2200static struct mapped_device *dm_find_md(dev_t dev)
2201{
2202	struct mapped_device *md;
2203	unsigned minor = MINOR(dev);
2204
2205	if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2206		return NULL;
2207
2208	spin_lock(&_minor_lock);
2209
2210	md = idr_find(&_minor_idr, minor);
2211	if (md && (md == MINOR_ALLOCED ||
2212		   (MINOR(disk_devt(dm_disk(md))) != minor) ||
2213		   dm_deleting_md(md) ||
2214		   test_bit(DMF_FREEING, &md->flags))) {
2215		md = NULL;
2216		goto out;
2217	}
2218
2219out:
2220	spin_unlock(&_minor_lock);
2221
2222	return md;
2223}
2224
2225struct mapped_device *dm_get_md(dev_t dev)
2226{
2227	struct mapped_device *md = dm_find_md(dev);
2228
2229	if (md)
2230		dm_get(md);
2231
2232	return md;
2233}
2234
2235void *dm_get_mdptr(struct mapped_device *md)
2236{
2237	return md->interface_ptr;
2238}
2239
2240void dm_set_mdptr(struct mapped_device *md, void *ptr)
2241{
2242	md->interface_ptr = ptr;
2243}
2244
2245void dm_get(struct mapped_device *md)
2246{
2247	atomic_inc(&md->holders);
2248	BUG_ON(test_bit(DMF_FREEING, &md->flags));
2249}
2250
 
 
 
 
 
 
 
 
 
 
 
 
 
2251const char *dm_device_name(struct mapped_device *md)
2252{
2253	return md->name;
2254}
2255EXPORT_SYMBOL_GPL(dm_device_name);
2256
2257static void __dm_destroy(struct mapped_device *md, bool wait)
2258{
2259	struct dm_table *map;
 
2260
2261	might_sleep();
2262
2263	spin_lock(&_minor_lock);
2264	map = dm_get_live_table(md);
2265	idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2266	set_bit(DMF_FREEING, &md->flags);
2267	spin_unlock(&_minor_lock);
2268
 
 
 
 
 
 
 
 
2269	if (!dm_suspended_md(md)) {
2270		dm_table_presuspend_targets(map);
2271		dm_table_postsuspend_targets(map);
2272	}
 
 
 
2273
2274	/*
2275	 * Rare, but there may be I/O requests still going to complete,
2276	 * for example.  Wait for all references to disappear.
2277	 * No one should increment the reference count of the mapped_device,
2278	 * after the mapped_device state becomes DMF_FREEING.
2279	 */
2280	if (wait)
2281		while (atomic_read(&md->holders))
2282			msleep(1);
2283	else if (atomic_read(&md->holders))
2284		DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2285		       dm_device_name(md), atomic_read(&md->holders));
2286
2287	dm_sysfs_exit(md);
2288	dm_table_put(map);
2289	dm_table_destroy(__unbind(md));
2290	free_dev(md);
2291}
2292
2293void dm_destroy(struct mapped_device *md)
2294{
2295	__dm_destroy(md, true);
2296}
2297
2298void dm_destroy_immediate(struct mapped_device *md)
2299{
2300	__dm_destroy(md, false);
2301}
2302
2303void dm_put(struct mapped_device *md)
2304{
2305	atomic_dec(&md->holders);
2306}
2307EXPORT_SYMBOL_GPL(dm_put);
2308
2309static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2310{
2311	int r = 0;
2312	DECLARE_WAITQUEUE(wait, current);
2313
2314	add_wait_queue(&md->wait, &wait);
2315
2316	while (1) {
2317		set_current_state(interruptible);
2318
2319		smp_mb();
2320		if (!md_in_flight(md))
2321			break;
2322
2323		if (interruptible == TASK_INTERRUPTIBLE &&
2324		    signal_pending(current)) {
2325			r = -EINTR;
2326			break;
2327		}
2328
2329		io_schedule();
2330	}
2331	set_current_state(TASK_RUNNING);
2332
2333	remove_wait_queue(&md->wait, &wait);
2334
2335	return r;
2336}
2337
2338/*
2339 * Process the deferred bios
2340 */
2341static void dm_wq_work(struct work_struct *work)
2342{
2343	struct mapped_device *md = container_of(work, struct mapped_device,
2344						work);
2345	struct bio *c;
 
 
2346
2347	down_read(&md->io_lock);
2348
2349	while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2350		spin_lock_irq(&md->deferred_lock);
2351		c = bio_list_pop(&md->deferred);
2352		spin_unlock_irq(&md->deferred_lock);
2353
2354		if (!c)
2355			break;
2356
2357		up_read(&md->io_lock);
2358
2359		if (dm_request_based(md))
2360			generic_make_request(c);
2361		else
2362			__split_and_process_bio(md, c);
2363
2364		down_read(&md->io_lock);
2365	}
2366
2367	up_read(&md->io_lock);
2368}
2369
2370static void dm_queue_flush(struct mapped_device *md)
2371{
2372	clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2373	smp_mb__after_clear_bit();
2374	queue_work(md->wq, &md->work);
2375}
2376
2377/*
2378 * Swap in a new table, returning the old one for the caller to destroy.
2379 */
2380struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2381{
2382	struct dm_table *map = ERR_PTR(-EINVAL);
2383	struct queue_limits limits;
2384	int r;
2385
2386	mutex_lock(&md->suspend_lock);
2387
2388	/* device must be suspended */
2389	if (!dm_suspended_md(md))
2390		goto out;
2391
2392	r = dm_calculate_queue_limits(table, &limits);
2393	if (r) {
2394		map = ERR_PTR(r);
2395		goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2396	}
2397
2398	map = __bind(md, table, &limits);
 
2399
2400out:
2401	mutex_unlock(&md->suspend_lock);
2402	return map;
2403}
2404
2405/*
2406 * Functions to lock and unlock any filesystem running on the
2407 * device.
2408 */
2409static int lock_fs(struct mapped_device *md)
2410{
2411	int r;
2412
2413	WARN_ON(md->frozen_sb);
2414
2415	md->frozen_sb = freeze_bdev(md->bdev);
2416	if (IS_ERR(md->frozen_sb)) {
2417		r = PTR_ERR(md->frozen_sb);
2418		md->frozen_sb = NULL;
2419		return r;
2420	}
2421
2422	set_bit(DMF_FROZEN, &md->flags);
2423
2424	return 0;
2425}
2426
2427static void unlock_fs(struct mapped_device *md)
2428{
2429	if (!test_bit(DMF_FROZEN, &md->flags))
2430		return;
2431
2432	thaw_bdev(md->bdev, md->frozen_sb);
2433	md->frozen_sb = NULL;
2434	clear_bit(DMF_FROZEN, &md->flags);
2435}
2436
2437/*
2438 * We need to be able to change a mapping table under a mounted
2439 * filesystem.  For example we might want to move some data in
2440 * the background.  Before the table can be swapped with
2441 * dm_bind_table, dm_suspend must be called to flush any in
2442 * flight bios and ensure that any further io gets deferred.
2443 */
2444/*
2445 * Suspend mechanism in request-based dm.
2446 *
2447 * 1. Flush all I/Os by lock_fs() if needed.
2448 * 2. Stop dispatching any I/O by stopping the request_queue.
2449 * 3. Wait for all in-flight I/Os to be completed or requeued.
2450 *
2451 * To abort suspend, start the request_queue.
2452 */
2453int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
 
 
 
 
2454{
2455	struct dm_table *map = NULL;
2456	int r = 0;
2457	int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2458	int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2459
2460	mutex_lock(&md->suspend_lock);
2461
2462	if (dm_suspended_md(md)) {
2463		r = -EINVAL;
2464		goto out_unlock;
2465	}
2466
2467	map = dm_get_live_table(md);
2468
2469	/*
2470	 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2471	 * This flag is cleared before dm_suspend returns.
2472	 */
2473	if (noflush)
2474		set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
 
 
2475
2476	/* This does not get reverted if there's an error later. */
 
 
 
2477	dm_table_presuspend_targets(map);
2478
2479	/*
2480	 * Flush I/O to the device.
2481	 * Any I/O submitted after lock_fs() may not be flushed.
2482	 * noflush takes precedence over do_lockfs.
2483	 * (lock_fs() flushes I/Os and waits for them to complete.)
2484	 */
2485	if (!noflush && do_lockfs) {
2486		r = lock_fs(md);
2487		if (r)
2488			goto out;
 
 
2489	}
2490
2491	/*
2492	 * Here we must make sure that no processes are submitting requests
2493	 * to target drivers i.e. no one may be executing
2494	 * __split_and_process_bio. This is called from dm_request and
2495	 * dm_wq_work.
2496	 *
2497	 * To get all processes out of __split_and_process_bio in dm_request,
2498	 * we take the write lock. To prevent any process from reentering
2499	 * __split_and_process_bio from dm_request and quiesce the thread
2500	 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2501	 * flush_workqueue(md->wq).
2502	 */
2503	down_write(&md->io_lock);
2504	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2505	up_write(&md->io_lock);
 
2506
2507	/*
2508	 * Stop md->queue before flushing md->wq in case request-based
2509	 * dm defers requests to md->wq from md->queue.
2510	 */
2511	if (dm_request_based(md))
2512		stop_queue(md->queue);
2513
2514	flush_workqueue(md->wq);
2515
2516	/*
2517	 * At this point no more requests are entering target request routines.
2518	 * We call dm_wait_for_completion to wait for all existing requests
2519	 * to finish.
2520	 */
2521	r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
 
 
2522
2523	down_write(&md->io_lock);
2524	if (noflush)
2525		clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2526	up_write(&md->io_lock);
 
2527
2528	/* were we interrupted ? */
2529	if (r < 0) {
2530		dm_queue_flush(md);
2531
2532		if (dm_request_based(md))
2533			start_queue(md->queue);
2534
2535		unlock_fs(md);
2536		goto out; /* pushback list is already flushed, so skip flush */
 
2537	}
2538
2539	/*
2540	 * If dm_wait_for_completion returned 0, the device is completely
2541	 * quiescent now. There is no request-processing activity. All new
2542	 * requests are being added to md->deferred list.
2543	 */
2544
2545	set_bit(DMF_SUSPENDED, &md->flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2546
2547	dm_table_postsuspend_targets(map);
 
2548
2549out:
2550	dm_table_put(map);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2551
2552out_unlock:
2553	mutex_unlock(&md->suspend_lock);
2554	return r;
2555}
2556
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2557int dm_resume(struct mapped_device *md)
2558{
2559	int r = -EINVAL;
2560	struct dm_table *map = NULL;
2561
2562	mutex_lock(&md->suspend_lock);
 
 
 
2563	if (!dm_suspended_md(md))
2564		goto out;
2565
2566	map = dm_get_live_table(md);
 
 
 
 
 
 
 
 
 
2567	if (!map || !dm_table_get_size(map))
2568		goto out;
2569
2570	r = dm_table_resume_targets(map);
2571	if (r)
2572		goto out;
2573
2574	dm_queue_flush(md);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2575
2576	/*
2577	 * Flushing deferred I/Os must be done after targets are resumed
2578	 * so that mapping of targets can work correctly.
2579	 * Request-based dm is queueing the deferred I/Os in its request_queue.
 
2580	 */
2581	if (dm_request_based(md))
2582		start_queue(md->queue);
2583
2584	unlock_fs(md);
 
2585
2586	clear_bit(DMF_SUSPENDED, &md->flags);
 
 
2587
2588	r = 0;
2589out:
2590	dm_table_put(map);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2591	mutex_unlock(&md->suspend_lock);
 
 
2592
2593	return r;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2594}
 
2595
2596/*-----------------------------------------------------------------
2597 * Event notification.
2598 *---------------------------------------------------------------*/
2599int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2600		       unsigned cookie)
2601{
2602	char udev_cookie[DM_COOKIE_LENGTH];
2603	char *envp[] = { udev_cookie, NULL };
2604
2605	if (!cookie)
2606		return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2607	else {
2608		snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2609			 DM_COOKIE_ENV_VAR_NAME, cookie);
2610		return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2611					  action, envp);
2612	}
2613}
2614
2615uint32_t dm_next_uevent_seq(struct mapped_device *md)
2616{
2617	return atomic_add_return(1, &md->uevent_seq);
2618}
2619
2620uint32_t dm_get_event_nr(struct mapped_device *md)
2621{
2622	return atomic_read(&md->event_nr);
2623}
2624
2625int dm_wait_event(struct mapped_device *md, int event_nr)
2626{
2627	return wait_event_interruptible(md->eventq,
2628			(event_nr != atomic_read(&md->event_nr)));
2629}
2630
2631void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2632{
2633	unsigned long flags;
2634
2635	spin_lock_irqsave(&md->uevent_lock, flags);
2636	list_add(elist, &md->uevent_list);
2637	spin_unlock_irqrestore(&md->uevent_lock, flags);
2638}
2639
2640/*
2641 * The gendisk is only valid as long as you have a reference
2642 * count on 'md'.
2643 */
2644struct gendisk *dm_disk(struct mapped_device *md)
2645{
2646	return md->disk;
2647}
 
2648
2649struct kobject *dm_kobject(struct mapped_device *md)
2650{
2651	return &md->kobj;
2652}
2653
2654/*
2655 * struct mapped_device should not be exported outside of dm.c
2656 * so use this check to verify that kobj is part of md structure
2657 */
2658struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2659{
2660	struct mapped_device *md;
2661
2662	md = container_of(kobj, struct mapped_device, kobj);
2663	if (&md->kobj != kobj)
2664		return NULL;
2665
2666	if (test_bit(DMF_FREEING, &md->flags) ||
2667	    dm_deleting_md(md))
2668		return NULL;
2669
 
 
 
 
 
2670	dm_get(md);
 
 
 
2671	return md;
2672}
2673
2674int dm_suspended_md(struct mapped_device *md)
2675{
2676	return test_bit(DMF_SUSPENDED, &md->flags);
2677}
2678
 
 
 
 
 
 
 
 
 
 
2679int dm_suspended(struct dm_target *ti)
2680{
2681	return dm_suspended_md(dm_table_get_md(ti->table));
2682}
2683EXPORT_SYMBOL_GPL(dm_suspended);
2684
2685int dm_noflush_suspending(struct dm_target *ti)
2686{
2687	return __noflush_suspending(dm_table_get_md(ti->table));
2688}
2689EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2690
2691struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity)
2692{
2693	struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2694	unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS;
 
 
 
 
2695
2696	if (!pools)
2697		return NULL;
2698
2699	pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2700			 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2701			 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2702	if (!pools->io_pool)
2703		goto free_pools_and_out;
2704
2705	pools->tio_pool = (type == DM_TYPE_BIO_BASED) ?
2706			  mempool_create_slab_pool(MIN_IOS, _tio_cache) :
2707			  mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2708	if (!pools->tio_pool)
2709		goto free_io_pool_and_out;
2710
2711	pools->bs = bioset_create(pool_size, 0);
2712	if (!pools->bs)
2713		goto free_tio_pool_and_out;
2714
2715	if (integrity && bioset_integrity_create(pools->bs, pool_size))
2716		goto free_bioset_and_out;
2717
2718	return pools;
 
2719
2720free_bioset_and_out:
2721	bioset_free(pools->bs);
 
2722
2723free_tio_pool_and_out:
2724	mempool_destroy(pools->tio_pool);
2725
2726free_io_pool_and_out:
2727	mempool_destroy(pools->io_pool);
2728
2729free_pools_and_out:
2730	kfree(pools);
2731
2732	return NULL;
2733}
2734
2735void dm_free_md_mempools(struct dm_md_mempools *pools)
2736{
2737	if (!pools)
2738		return;
2739
2740	if (pools->io_pool)
2741		mempool_destroy(pools->io_pool);
2742
2743	if (pools->tio_pool)
2744		mempool_destroy(pools->tio_pool);
2745
2746	if (pools->bs)
2747		bioset_free(pools->bs);
 
 
 
 
2748
2749	kfree(pools);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2750}
2751
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2752static const struct block_device_operations dm_blk_dops = {
2753	.open = dm_blk_open,
2754	.release = dm_blk_close,
2755	.ioctl = dm_blk_ioctl,
2756	.getgeo = dm_blk_getgeo,
 
 
2757	.owner = THIS_MODULE
2758};
2759
2760EXPORT_SYMBOL(dm_get_mapinfo);
 
 
 
 
 
2761
2762/*
2763 * module hooks
2764 */
2765module_init(dm_init);
2766module_exit(dm_exit);
2767
2768module_param(major, uint, 0);
2769MODULE_PARM_DESC(major, "The major number of the device mapper");
 
 
 
 
 
 
 
2770MODULE_DESCRIPTION(DM_NAME " driver");
2771MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2772MODULE_LICENSE("GPL");