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/mempool.h>
  18#include <linux/slab.h>
  19#include <linux/idr.h>
  20#include <linux/hdreg.h>
  21#include <linux/delay.h>
  22
  23#include <trace/events/block.h>
  24
  25#define DM_MSG_PREFIX "core"
  26
  27#ifdef CONFIG_PRINTK
  28/*
  29 * ratelimit state to be used in DMXXX_LIMIT().
  30 */
  31DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
  32		       DEFAULT_RATELIMIT_INTERVAL,
  33		       DEFAULT_RATELIMIT_BURST);
  34EXPORT_SYMBOL(dm_ratelimit_state);
  35#endif
  36
  37/*
  38 * Cookies are numeric values sent with CHANGE and REMOVE
  39 * uevents while resuming, removing or renaming the device.
  40 */
  41#define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
  42#define DM_COOKIE_LENGTH 24
  43
  44static const char *_name = DM_NAME;
  45
  46static unsigned int major = 0;
  47static unsigned int _major = 0;
  48
  49static DEFINE_IDR(_minor_idr);
  50
  51static DEFINE_SPINLOCK(_minor_lock);
 
 
 
 
 
  52/*
  53 * For bio-based dm.
  54 * One of these is allocated per bio.
  55 */
  56struct dm_io {
  57	struct mapped_device *md;
  58	int error;
  59	atomic_t io_count;
  60	struct bio *bio;
  61	unsigned long start_time;
  62	spinlock_t endio_lock;
  63};
  64
  65/*
  66 * For bio-based dm.
  67 * One of these is allocated per target within a bio.  Hopefully
  68 * this will be simplified out one day.
  69 */
  70struct dm_target_io {
  71	struct dm_io *io;
  72	struct dm_target *ti;
  73	union map_info info;
  74};
  75
  76/*
  77 * For request-based dm.
  78 * One of these is allocated per request.
  79 */
  80struct dm_rq_target_io {
  81	struct mapped_device *md;
  82	struct dm_target *ti;
  83	struct request *orig, clone;
  84	int error;
  85	union map_info info;
  86};
  87
  88/*
  89 * For request-based dm.
  90 * One of these is allocated per bio.
 
 
 
 
  91 */
  92struct dm_rq_clone_bio_info {
  93	struct bio *orig;
  94	struct dm_rq_target_io *tio;
 
  95};
  96
  97union map_info *dm_get_mapinfo(struct bio *bio)
  98{
  99	if (bio && bio->bi_private)
 100		return &((struct dm_target_io *)bio->bi_private)->info;
 101	return NULL;
 102}
 103
 104union map_info *dm_get_rq_mapinfo(struct request *rq)
 105{
 106	if (rq && rq->end_io_data)
 107		return &((struct dm_rq_target_io *)rq->end_io_data)->info;
 108	return NULL;
 109}
 110EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
 111
 112#define MINOR_ALLOCED ((void *)-1)
 113
 114/*
 115 * Bits for the md->flags field.
 116 */
 117#define DMF_BLOCK_IO_FOR_SUSPEND 0
 118#define DMF_SUSPENDED 1
 119#define DMF_FROZEN 2
 120#define DMF_FREEING 3
 121#define DMF_DELETING 4
 122#define DMF_NOFLUSH_SUSPENDING 5
 123#define DMF_MERGE_IS_OPTIONAL 6
 
 
 
 
 
 
 
 
 
 124
 125/*
 126 * Work processed by per-device workqueue.
 127 */
 128struct mapped_device {
 129	struct rw_semaphore io_lock;
 130	struct mutex suspend_lock;
 131	rwlock_t map_lock;
 132	atomic_t holders;
 133	atomic_t open_count;
 134
 
 
 
 
 
 
 
 135	unsigned long flags;
 136
 137	struct request_queue *queue;
 138	unsigned type;
 139	/* Protect queue and type against concurrent access. */
 140	struct mutex type_lock;
 141
 142	struct target_type *immutable_target_type;
 143
 144	struct gendisk *disk;
 145	char name[16];
 146
 147	void *interface_ptr;
 148
 149	/*
 150	 * A list of ios that arrived while we were suspended.
 151	 */
 152	atomic_t pending[2];
 153	wait_queue_head_t wait;
 154	struct work_struct work;
 155	struct bio_list deferred;
 156	spinlock_t deferred_lock;
 157
 158	/*
 159	 * Processing queue (flush)
 160	 */
 161	struct workqueue_struct *wq;
 162
 163	/*
 164	 * The current mapping.
 165	 */
 166	struct dm_table *map;
 167
 168	/*
 169	 * io objects are allocated from here.
 170	 */
 171	mempool_t *io_pool;
 172	mempool_t *tio_pool;
 173
 174	struct bio_set *bs;
 175
 176	/*
 177	 * Event handling.
 178	 */
 179	atomic_t event_nr;
 180	wait_queue_head_t eventq;
 181	atomic_t uevent_seq;
 182	struct list_head uevent_list;
 183	spinlock_t uevent_lock; /* Protect access to uevent_list */
 184
 185	/*
 186	 * freeze/thaw support require holding onto a super block
 187	 */
 188	struct super_block *frozen_sb;
 189	struct block_device *bdev;
 190
 191	/* forced geometry settings */
 192	struct hd_geometry geometry;
 193
 194	/* sysfs handle */
 195	struct kobject kobj;
 196
 197	/* zero-length flush that will be cloned and submitted to targets */
 198	struct bio flush_bio;
 
 
 199};
 200
 201/*
 202 * For mempools pre-allocation at the table loading time.
 203 */
 204struct dm_md_mempools {
 205	mempool_t *io_pool;
 206	mempool_t *tio_pool;
 207	struct bio_set *bs;
 208};
 209
 210#define MIN_IOS 256
 
 
 211static struct kmem_cache *_io_cache;
 212static struct kmem_cache *_tio_cache;
 213static struct kmem_cache *_rq_tio_cache;
 214static struct kmem_cache *_rq_bio_info_cache;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 215
 216static int __init local_init(void)
 217{
 218	int r = -ENOMEM;
 219
 220	/* allocate a slab for the dm_ios */
 221	_io_cache = KMEM_CACHE(dm_io, 0);
 222	if (!_io_cache)
 223		return r;
 224
 225	/* allocate a slab for the target ios */
 226	_tio_cache = KMEM_CACHE(dm_target_io, 0);
 227	if (!_tio_cache)
 228		goto out_free_io_cache;
 229
 230	_rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
 231	if (!_rq_tio_cache)
 232		goto out_free_tio_cache;
 233
 234	_rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
 235	if (!_rq_bio_info_cache)
 236		goto out_free_rq_tio_cache;
 237
 238	r = dm_uevent_init();
 239	if (r)
 240		goto out_free_rq_bio_info_cache;
 241
 242	_major = major;
 243	r = register_blkdev(_major, _name);
 244	if (r < 0)
 245		goto out_uevent_exit;
 246
 247	if (!_major)
 248		_major = r;
 249
 250	return 0;
 251
 252out_uevent_exit:
 253	dm_uevent_exit();
 254out_free_rq_bio_info_cache:
 255	kmem_cache_destroy(_rq_bio_info_cache);
 256out_free_rq_tio_cache:
 257	kmem_cache_destroy(_rq_tio_cache);
 258out_free_tio_cache:
 259	kmem_cache_destroy(_tio_cache);
 260out_free_io_cache:
 261	kmem_cache_destroy(_io_cache);
 262
 263	return r;
 264}
 265
 266static void local_exit(void)
 267{
 268	kmem_cache_destroy(_rq_bio_info_cache);
 
 269	kmem_cache_destroy(_rq_tio_cache);
 270	kmem_cache_destroy(_tio_cache);
 271	kmem_cache_destroy(_io_cache);
 272	unregister_blkdev(_major, _name);
 273	dm_uevent_exit();
 274
 275	_major = 0;
 276
 277	DMINFO("cleaned up");
 278}
 279
 280static int (*_inits[])(void) __initdata = {
 281	local_init,
 282	dm_target_init,
 283	dm_linear_init,
 284	dm_stripe_init,
 285	dm_io_init,
 286	dm_kcopyd_init,
 287	dm_interface_init,
 
 288};
 289
 290static void (*_exits[])(void) = {
 291	local_exit,
 292	dm_target_exit,
 293	dm_linear_exit,
 294	dm_stripe_exit,
 295	dm_io_exit,
 296	dm_kcopyd_exit,
 297	dm_interface_exit,
 
 298};
 299
 300static int __init dm_init(void)
 301{
 302	const int count = ARRAY_SIZE(_inits);
 303
 304	int r, i;
 305
 306	for (i = 0; i < count; i++) {
 307		r = _inits[i]();
 308		if (r)
 309			goto bad;
 310	}
 311
 312	return 0;
 313
 314      bad:
 315	while (i--)
 316		_exits[i]();
 317
 318	return r;
 319}
 320
 321static void __exit dm_exit(void)
 322{
 323	int i = ARRAY_SIZE(_exits);
 324
 325	while (i--)
 326		_exits[i]();
 327
 328	/*
 329	 * Should be empty by this point.
 330	 */
 331	idr_remove_all(&_minor_idr);
 332	idr_destroy(&_minor_idr);
 333}
 334
 335/*
 336 * Block device functions
 337 */
 338int dm_deleting_md(struct mapped_device *md)
 339{
 340	return test_bit(DMF_DELETING, &md->flags);
 341}
 342
 343static int dm_blk_open(struct block_device *bdev, fmode_t mode)
 344{
 345	struct mapped_device *md;
 346
 347	spin_lock(&_minor_lock);
 348
 349	md = bdev->bd_disk->private_data;
 350	if (!md)
 351		goto out;
 352
 353	if (test_bit(DMF_FREEING, &md->flags) ||
 354	    dm_deleting_md(md)) {
 355		md = NULL;
 356		goto out;
 357	}
 358
 359	dm_get(md);
 360	atomic_inc(&md->open_count);
 361
 362out:
 363	spin_unlock(&_minor_lock);
 364
 365	return md ? 0 : -ENXIO;
 366}
 367
 368static int dm_blk_close(struct gendisk *disk, fmode_t mode)
 369{
 370	struct mapped_device *md = disk->private_data;
 371
 372	spin_lock(&_minor_lock);
 373
 374	atomic_dec(&md->open_count);
 
 
 
 375	dm_put(md);
 376
 377	spin_unlock(&_minor_lock);
 378
 379	return 0;
 380}
 381
 382int dm_open_count(struct mapped_device *md)
 383{
 384	return atomic_read(&md->open_count);
 385}
 386
 387/*
 388 * Guarantees nothing is using the device before it's deleted.
 389 */
 390int dm_lock_for_deletion(struct mapped_device *md)
 391{
 392	int r = 0;
 393
 394	spin_lock(&_minor_lock);
 395
 396	if (dm_open_count(md))
 397		r = -EBUSY;
 
 
 
 
 398	else
 399		set_bit(DMF_DELETING, &md->flags);
 400
 401	spin_unlock(&_minor_lock);
 402
 403	return r;
 404}
 405
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 406static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 407{
 408	struct mapped_device *md = bdev->bd_disk->private_data;
 409
 410	return dm_get_geometry(md, geo);
 411}
 412
 413static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
 414			unsigned int cmd, unsigned long arg)
 415{
 416	struct mapped_device *md = bdev->bd_disk->private_data;
 417	struct dm_table *map = dm_get_live_table(md);
 
 418	struct dm_target *tgt;
 419	int r = -ENOTTY;
 420
 
 
 
 421	if (!map || !dm_table_get_size(map))
 422		goto out;
 423
 424	/* We only support devices that have a single target */
 425	if (dm_table_get_num_targets(map) != 1)
 426		goto out;
 427
 428	tgt = dm_table_get_target(map, 0);
 429
 430	if (dm_suspended_md(md)) {
 431		r = -EAGAIN;
 432		goto out;
 433	}
 434
 435	if (tgt->type->ioctl)
 436		r = tgt->type->ioctl(tgt, cmd, arg);
 437
 438out:
 439	dm_table_put(map);
 
 
 
 
 
 440
 441	return r;
 442}
 443
 444static struct dm_io *alloc_io(struct mapped_device *md)
 445{
 446	return mempool_alloc(md->io_pool, GFP_NOIO);
 447}
 448
 449static void free_io(struct mapped_device *md, struct dm_io *io)
 450{
 451	mempool_free(io, md->io_pool);
 452}
 453
 454static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
 455{
 456	mempool_free(tio, md->tio_pool);
 457}
 458
 459static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
 460					    gfp_t gfp_mask)
 461{
 462	return mempool_alloc(md->tio_pool, gfp_mask);
 463}
 464
 465static void free_rq_tio(struct dm_rq_target_io *tio)
 466{
 467	mempool_free(tio, tio->md->tio_pool);
 468}
 469
 470static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md)
 471{
 472	return mempool_alloc(md->io_pool, GFP_ATOMIC);
 473}
 474
 475static void free_bio_info(struct dm_rq_clone_bio_info *info)
 476{
 477	mempool_free(info, info->tio->md->io_pool);
 478}
 479
 480static int md_in_flight(struct mapped_device *md)
 481{
 482	return atomic_read(&md->pending[READ]) +
 483	       atomic_read(&md->pending[WRITE]);
 484}
 485
 486static void start_io_acct(struct dm_io *io)
 487{
 488	struct mapped_device *md = io->md;
 
 489	int cpu;
 490	int rw = bio_data_dir(io->bio);
 491
 492	io->start_time = jiffies;
 493
 494	cpu = part_stat_lock();
 495	part_round_stats(cpu, &dm_disk(md)->part0);
 496	part_stat_unlock();
 497	atomic_set(&dm_disk(md)->part0.in_flight[rw],
 498		atomic_inc_return(&md->pending[rw]));
 
 
 
 
 499}
 500
 501static void end_io_acct(struct dm_io *io)
 502{
 503	struct mapped_device *md = io->md;
 504	struct bio *bio = io->bio;
 505	unsigned long duration = jiffies - io->start_time;
 506	int pending, cpu;
 507	int rw = bio_data_dir(bio);
 508
 509	cpu = part_stat_lock();
 510	part_round_stats(cpu, &dm_disk(md)->part0);
 511	part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
 512	part_stat_unlock();
 513
 
 
 
 
 514	/*
 515	 * After this is decremented the bio must not be touched if it is
 516	 * a flush.
 517	 */
 518	pending = atomic_dec_return(&md->pending[rw]);
 519	atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
 520	pending += atomic_read(&md->pending[rw^0x1]);
 521
 522	/* nudge anyone waiting on suspend queue */
 523	if (!pending)
 524		wake_up(&md->wait);
 525}
 526
 527/*
 528 * Add the bio to the list of deferred io.
 529 */
 530static void queue_io(struct mapped_device *md, struct bio *bio)
 531{
 532	unsigned long flags;
 533
 534	spin_lock_irqsave(&md->deferred_lock, flags);
 535	bio_list_add(&md->deferred, bio);
 536	spin_unlock_irqrestore(&md->deferred_lock, flags);
 537	queue_work(md->wq, &md->work);
 538}
 539
 540/*
 541 * Everyone (including functions in this file), should use this
 542 * function to access the md->map field, and make sure they call
 543 * dm_table_put() when finished.
 544 */
 545struct dm_table *dm_get_live_table(struct mapped_device *md)
 546{
 547	struct dm_table *t;
 548	unsigned long flags;
 
 
 549
 550	read_lock_irqsave(&md->map_lock, flags);
 551	t = md->map;
 552	if (t)
 553		dm_table_get(t);
 554	read_unlock_irqrestore(&md->map_lock, flags);
 555
 556	return t;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 557}
 558
 559/*
 560 * Get the geometry associated with a dm device
 561 */
 562int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
 563{
 564	*geo = md->geometry;
 565
 566	return 0;
 567}
 568
 569/*
 570 * Set the geometry of a device.
 571 */
 572int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
 573{
 574	sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
 575
 576	if (geo->start > sz) {
 577		DMWARN("Start sector is beyond the geometry limits.");
 578		return -EINVAL;
 579	}
 580
 581	md->geometry = *geo;
 582
 583	return 0;
 584}
 585
 586/*-----------------------------------------------------------------
 587 * CRUD START:
 588 *   A more elegant soln is in the works that uses the queue
 589 *   merge fn, unfortunately there are a couple of changes to
 590 *   the block layer that I want to make for this.  So in the
 591 *   interests of getting something for people to use I give
 592 *   you this clearly demarcated crap.
 593 *---------------------------------------------------------------*/
 594
 595static int __noflush_suspending(struct mapped_device *md)
 596{
 597	return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
 598}
 599
 600/*
 601 * Decrements the number of outstanding ios that a bio has been
 602 * cloned into, completing the original io if necc.
 603 */
 604static void dec_pending(struct dm_io *io, int error)
 605{
 606	unsigned long flags;
 607	int io_error;
 608	struct bio *bio;
 609	struct mapped_device *md = io->md;
 610
 611	/* Push-back supersedes any I/O errors */
 612	if (unlikely(error)) {
 613		spin_lock_irqsave(&io->endio_lock, flags);
 614		if (!(io->error > 0 && __noflush_suspending(md)))
 615			io->error = error;
 616		spin_unlock_irqrestore(&io->endio_lock, flags);
 617	}
 618
 619	if (atomic_dec_and_test(&io->io_count)) {
 620		if (io->error == DM_ENDIO_REQUEUE) {
 621			/*
 622			 * Target requested pushing back the I/O.
 623			 */
 624			spin_lock_irqsave(&md->deferred_lock, flags);
 625			if (__noflush_suspending(md))
 626				bio_list_add_head(&md->deferred, io->bio);
 627			else
 628				/* noflush suspend was interrupted. */
 629				io->error = -EIO;
 630			spin_unlock_irqrestore(&md->deferred_lock, flags);
 631		}
 632
 633		io_error = io->error;
 634		bio = io->bio;
 635		end_io_acct(io);
 636		free_io(md, io);
 637
 638		if (io_error == DM_ENDIO_REQUEUE)
 639			return;
 640
 641		if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) {
 642			/*
 643			 * Preflush done for flush with data, reissue
 644			 * without REQ_FLUSH.
 645			 */
 646			bio->bi_rw &= ~REQ_FLUSH;
 647			queue_io(md, bio);
 648		} else {
 649			/* done with normal IO or empty flush */
 650			trace_block_bio_complete(md->queue, bio, io_error);
 651			bio_endio(bio, io_error);
 652		}
 653	}
 654}
 655
 656static void clone_endio(struct bio *bio, int error)
 657{
 658	int r = 0;
 659	struct dm_target_io *tio = bio->bi_private;
 660	struct dm_io *io = tio->io;
 661	struct mapped_device *md = tio->io->md;
 662	dm_endio_fn endio = tio->ti->type->end_io;
 663
 664	if (!bio_flagged(bio, BIO_UPTODATE) && !error)
 665		error = -EIO;
 666
 667	if (endio) {
 668		r = endio(tio->ti, bio, error, &tio->info);
 669		if (r < 0 || r == DM_ENDIO_REQUEUE)
 670			/*
 671			 * error and requeue request are handled
 672			 * in dec_pending().
 673			 */
 674			error = r;
 675		else if (r == DM_ENDIO_INCOMPLETE)
 676			/* The target will handle the io */
 677			return;
 678		else if (r) {
 679			DMWARN("unimplemented target endio return value: %d", r);
 680			BUG();
 681		}
 682	}
 683
 684	/*
 685	 * Store md for cleanup instead of tio which is about to get freed.
 686	 */
 687	bio->bi_private = md->bs;
 688
 689	free_tio(md, tio);
 690	bio_put(bio);
 691	dec_pending(io, error);
 692}
 693
 694/*
 695 * Partial completion handling for request-based dm
 696 */
 697static void end_clone_bio(struct bio *clone, int error)
 698{
 699	struct dm_rq_clone_bio_info *info = clone->bi_private;
 
 700	struct dm_rq_target_io *tio = info->tio;
 701	struct bio *bio = info->orig;
 702	unsigned int nr_bytes = info->orig->bi_size;
 703
 704	bio_put(clone);
 705
 706	if (tio->error)
 707		/*
 708		 * An error has already been detected on the request.
 709		 * Once error occurred, just let clone->end_io() handle
 710		 * the remainder.
 711		 */
 712		return;
 713	else if (error) {
 714		/*
 715		 * Don't notice the error to the upper layer yet.
 716		 * The error handling decision is made by the target driver,
 717		 * when the request is completed.
 718		 */
 719		tio->error = error;
 720		return;
 721	}
 722
 723	/*
 724	 * I/O for the bio successfully completed.
 725	 * Notice the data completion to the upper layer.
 726	 */
 727
 728	/*
 729	 * bios are processed from the head of the list.
 730	 * So the completing bio should always be rq->bio.
 731	 * If it's not, something wrong is happening.
 732	 */
 733	if (tio->orig->bio != bio)
 734		DMERR("bio completion is going in the middle of the request");
 735
 736	/*
 737	 * Update the original request.
 738	 * Do not use blk_end_request() here, because it may complete
 739	 * the original request before the clone, and break the ordering.
 740	 */
 741	blk_update_request(tio->orig, 0, nr_bytes);
 742}
 743
 744/*
 745 * Don't touch any member of the md after calling this function because
 746 * the md may be freed in dm_put() at the end of this function.
 747 * Or do dm_get() before calling this function and dm_put() later.
 748 */
 749static void rq_completed(struct mapped_device *md, int rw, int run_queue)
 750{
 751	atomic_dec(&md->pending[rw]);
 752
 753	/* nudge anyone waiting on suspend queue */
 754	if (!md_in_flight(md))
 755		wake_up(&md->wait);
 756
 
 
 
 
 
 
 757	if (run_queue)
 758		blk_run_queue(md->queue);
 759
 760	/*
 761	 * dm_put() must be at the end of this function. See the comment above
 762	 */
 763	dm_put(md);
 764}
 765
 766static void free_rq_clone(struct request *clone)
 767{
 768	struct dm_rq_target_io *tio = clone->end_io_data;
 769
 770	blk_rq_unprep_clone(clone);
 771	free_rq_tio(tio);
 772}
 773
 774/*
 775 * Complete the clone and the original request.
 776 * Must be called without queue lock.
 777 */
 778static void dm_end_request(struct request *clone, int error)
 779{
 780	int rw = rq_data_dir(clone);
 781	struct dm_rq_target_io *tio = clone->end_io_data;
 782	struct mapped_device *md = tio->md;
 783	struct request *rq = tio->orig;
 784
 785	if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
 786		rq->errors = clone->errors;
 787		rq->resid_len = clone->resid_len;
 788
 789		if (rq->sense)
 790			/*
 791			 * We are using the sense buffer of the original
 792			 * request.
 793			 * So setting the length of the sense data is enough.
 794			 */
 795			rq->sense_len = clone->sense_len;
 796	}
 797
 798	free_rq_clone(clone);
 799	blk_end_request_all(rq, error);
 800	rq_completed(md, rw, true);
 801}
 802
 803static void dm_unprep_request(struct request *rq)
 804{
 805	struct request *clone = rq->special;
 806
 807	rq->special = NULL;
 808	rq->cmd_flags &= ~REQ_DONTPREP;
 809
 810	free_rq_clone(clone);
 811}
 812
 813/*
 814 * Requeue the original request of a clone.
 815 */
 816void dm_requeue_unmapped_request(struct request *clone)
 817{
 818	int rw = rq_data_dir(clone);
 819	struct dm_rq_target_io *tio = clone->end_io_data;
 820	struct mapped_device *md = tio->md;
 821	struct request *rq = tio->orig;
 822	struct request_queue *q = rq->q;
 823	unsigned long flags;
 824
 825	dm_unprep_request(rq);
 826
 827	spin_lock_irqsave(q->queue_lock, flags);
 828	blk_requeue_request(q, rq);
 829	spin_unlock_irqrestore(q->queue_lock, flags);
 830
 831	rq_completed(md, rw, 0);
 832}
 833EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
 834
 835static void __stop_queue(struct request_queue *q)
 836{
 837	blk_stop_queue(q);
 838}
 839
 840static void stop_queue(struct request_queue *q)
 841{
 842	unsigned long flags;
 843
 844	spin_lock_irqsave(q->queue_lock, flags);
 845	__stop_queue(q);
 846	spin_unlock_irqrestore(q->queue_lock, flags);
 847}
 848
 849static void __start_queue(struct request_queue *q)
 850{
 851	if (blk_queue_stopped(q))
 852		blk_start_queue(q);
 853}
 854
 855static void start_queue(struct request_queue *q)
 856{
 857	unsigned long flags;
 858
 859	spin_lock_irqsave(q->queue_lock, flags);
 860	__start_queue(q);
 861	spin_unlock_irqrestore(q->queue_lock, flags);
 862}
 863
 864static void dm_done(struct request *clone, int error, bool mapped)
 865{
 866	int r = error;
 867	struct dm_rq_target_io *tio = clone->end_io_data;
 868	dm_request_endio_fn rq_end_io = NULL;
 869
 870	if (tio->ti) {
 871		rq_end_io = tio->ti->type->rq_end_io;
 872
 873		if (mapped && rq_end_io)
 874			r = rq_end_io(tio->ti, clone, error, &tio->info);
 875	}
 876
 877	if (r <= 0)
 878		/* The target wants to complete the I/O */
 879		dm_end_request(clone, r);
 880	else if (r == DM_ENDIO_INCOMPLETE)
 881		/* The target will handle the I/O */
 882		return;
 883	else if (r == DM_ENDIO_REQUEUE)
 884		/* The target wants to requeue the I/O */
 885		dm_requeue_unmapped_request(clone);
 886	else {
 887		DMWARN("unimplemented target endio return value: %d", r);
 888		BUG();
 889	}
 890}
 891
 892/*
 893 * Request completion handler for request-based dm
 894 */
 895static void dm_softirq_done(struct request *rq)
 896{
 897	bool mapped = true;
 898	struct request *clone = rq->completion_data;
 899	struct dm_rq_target_io *tio = clone->end_io_data;
 900
 901	if (rq->cmd_flags & REQ_FAILED)
 902		mapped = false;
 903
 904	dm_done(clone, tio->error, mapped);
 905}
 906
 907/*
 908 * Complete the clone and the original request with the error status
 909 * through softirq context.
 910 */
 911static void dm_complete_request(struct request *clone, int error)
 912{
 913	struct dm_rq_target_io *tio = clone->end_io_data;
 914	struct request *rq = tio->orig;
 915
 916	tio->error = error;
 917	rq->completion_data = clone;
 918	blk_complete_request(rq);
 919}
 920
 921/*
 922 * Complete the not-mapped clone and the original request with the error status
 923 * through softirq context.
 924 * Target's rq_end_io() function isn't called.
 925 * This may be used when the target's map_rq() function fails.
 926 */
 927void dm_kill_unmapped_request(struct request *clone, int error)
 928{
 929	struct dm_rq_target_io *tio = clone->end_io_data;
 930	struct request *rq = tio->orig;
 931
 932	rq->cmd_flags |= REQ_FAILED;
 933	dm_complete_request(clone, error);
 934}
 935EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
 936
 937/*
 938 * Called with the queue lock held
 939 */
 940static void end_clone_request(struct request *clone, int error)
 941{
 942	/*
 943	 * For just cleaning up the information of the queue in which
 944	 * the clone was dispatched.
 945	 * The clone is *NOT* freed actually here because it is alloced from
 946	 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
 947	 */
 948	__blk_put_request(clone->q, clone);
 949
 950	/*
 951	 * Actual request completion is done in a softirq context which doesn't
 952	 * hold the queue lock.  Otherwise, deadlock could occur because:
 953	 *     - another request may be submitted by the upper level driver
 954	 *       of the stacking during the completion
 955	 *     - the submission which requires queue lock may be done
 956	 *       against this queue
 957	 */
 958	dm_complete_request(clone, error);
 959}
 960
 961/*
 962 * Return maximum size of I/O possible at the supplied sector up to the current
 963 * target boundary.
 964 */
 965static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
 966{
 967	sector_t target_offset = dm_target_offset(ti, sector);
 968
 969	return ti->len - target_offset;
 970}
 971
 972static sector_t max_io_len(sector_t sector, struct dm_target *ti)
 973{
 974	sector_t len = max_io_len_target_boundary(sector, ti);
 
 975
 976	/*
 977	 * Does the target need to split even further ?
 978	 */
 979	if (ti->split_io) {
 980		sector_t boundary;
 981		sector_t offset = dm_target_offset(ti, sector);
 982		boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
 983			   - offset;
 984		if (len > boundary)
 985			len = boundary;
 
 
 
 986	}
 987
 988	return len;
 989}
 990
 991static void __map_bio(struct dm_target *ti, struct bio *clone,
 992		      struct dm_target_io *tio)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 993{
 994	int r;
 995	sector_t sector;
 996	struct mapped_device *md;
 
 
 997
 998	clone->bi_end_io = clone_endio;
 999	clone->bi_private = tio;
1000
1001	/*
1002	 * Map the clone.  If r == 0 we don't need to do
1003	 * anything, the target has assumed ownership of
1004	 * this io.
1005	 */
1006	atomic_inc(&tio->io->io_count);
1007	sector = clone->bi_sector;
1008	r = ti->type->map(ti, clone, &tio->info);
1009	if (r == DM_MAPIO_REMAPPED) {
1010		/* the bio has been remapped so dispatch it */
1011
1012		trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1013				      tio->io->bio->bi_bdev->bd_dev, sector);
1014
1015		generic_make_request(clone);
1016	} else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1017		/* error the io and bail out, or requeue it if needed */
1018		md = tio->io->md;
1019		dec_pending(tio->io, r);
1020		/*
1021		 * Store bio_set for cleanup.
1022		 */
1023		clone->bi_end_io = NULL;
1024		clone->bi_private = md->bs;
1025		bio_put(clone);
1026		free_tio(md, tio);
1027	} else if (r) {
1028		DMWARN("unimplemented target map return value: %d", r);
1029		BUG();
1030	}
1031}
1032
1033struct clone_info {
1034	struct mapped_device *md;
1035	struct dm_table *map;
1036	struct bio *bio;
1037	struct dm_io *io;
1038	sector_t sector;
1039	sector_t sector_count;
1040	unsigned short idx;
1041};
1042
1043static void dm_bio_destructor(struct bio *bio)
1044{
1045	struct bio_set *bs = bio->bi_private;
1046
1047	bio_free(bio, bs);
1048}
1049
1050/*
1051 * Creates a little bio that just does part of a bvec.
1052 */
1053static struct bio *split_bvec(struct bio *bio, sector_t sector,
1054			      unsigned short idx, unsigned int offset,
1055			      unsigned int len, struct bio_set *bs)
1056{
1057	struct bio *clone;
1058	struct bio_vec *bv = bio->bi_io_vec + idx;
1059
1060	clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
1061	clone->bi_destructor = dm_bio_destructor;
1062	*clone->bi_io_vec = *bv;
1063
1064	clone->bi_sector = sector;
1065	clone->bi_bdev = bio->bi_bdev;
1066	clone->bi_rw = bio->bi_rw;
1067	clone->bi_vcnt = 1;
1068	clone->bi_size = to_bytes(len);
1069	clone->bi_io_vec->bv_offset = offset;
1070	clone->bi_io_vec->bv_len = clone->bi_size;
1071	clone->bi_flags |= 1 << BIO_CLONED;
1072
1073	if (bio_integrity(bio)) {
1074		bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1075		bio_integrity_trim(clone,
1076				   bio_sector_offset(bio, idx, offset), len);
1077	}
1078
1079	return clone;
1080}
1081
1082/*
1083 * Creates a bio that consists of range of complete bvecs.
1084 */
1085static struct bio *clone_bio(struct bio *bio, sector_t sector,
1086			     unsigned short idx, unsigned short bv_count,
1087			     unsigned int len, struct bio_set *bs)
1088{
1089	struct bio *clone;
1090
1091	clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
1092	__bio_clone(clone, bio);
1093	clone->bi_destructor = dm_bio_destructor;
1094	clone->bi_sector = sector;
1095	clone->bi_idx = idx;
1096	clone->bi_vcnt = idx + bv_count;
1097	clone->bi_size = to_bytes(len);
1098	clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1099
1100	if (bio_integrity(bio)) {
1101		bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1102
1103		if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1104			bio_integrity_trim(clone,
1105					   bio_sector_offset(bio, idx, 0), len);
1106	}
1107
1108	return clone;
1109}
1110
1111static struct dm_target_io *alloc_tio(struct clone_info *ci,
1112				      struct dm_target *ti)
 
1113{
1114	struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO);
 
 
 
 
1115
1116	tio->io = ci->io;
1117	tio->ti = ti;
1118	memset(&tio->info, 0, sizeof(tio->info));
1119
1120	return tio;
1121}
1122
1123static void __issue_target_request(struct clone_info *ci, struct dm_target *ti,
1124				   unsigned request_nr, sector_t len)
 
1125{
1126	struct dm_target_io *tio = alloc_tio(ci, ti);
1127	struct bio *clone;
1128
1129	tio->info.target_request_nr = request_nr;
1130
1131	/*
1132	 * Discard requests require the bio's inline iovecs be initialized.
1133	 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1134	 * and discard, so no need for concern about wasted bvec allocations.
1135	 */
1136	clone = bio_alloc_bioset(GFP_NOIO, ci->bio->bi_max_vecs, ci->md->bs);
1137	__bio_clone(clone, ci->bio);
1138	clone->bi_destructor = dm_bio_destructor;
1139	if (len) {
1140		clone->bi_sector = ci->sector;
1141		clone->bi_size = to_bytes(len);
1142	}
1143
1144	__map_bio(ti, clone, tio);
1145}
1146
1147static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti,
1148				    unsigned num_requests, sector_t len)
1149{
1150	unsigned request_nr;
1151
1152	for (request_nr = 0; request_nr < num_requests; request_nr++)
1153		__issue_target_request(ci, ti, request_nr, len);
1154}
1155
1156static int __clone_and_map_empty_flush(struct clone_info *ci)
1157{
1158	unsigned target_nr = 0;
1159	struct dm_target *ti;
1160
1161	BUG_ON(bio_has_data(ci->bio));
1162	while ((ti = dm_table_get_target(ci->map, target_nr++)))
1163		__issue_target_requests(ci, ti, ti->num_flush_requests, 0);
1164
1165	return 0;
1166}
1167
1168/*
1169 * Perform all io with a single clone.
1170 */
1171static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti)
1172{
1173	struct bio *clone, *bio = ci->bio;
1174	struct dm_target_io *tio;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1175
1176	tio = alloc_tio(ci, ti);
1177	clone = clone_bio(bio, ci->sector, ci->idx,
1178			  bio->bi_vcnt - ci->idx, ci->sector_count,
1179			  ci->md->bs);
1180	__map_bio(ti, clone, tio);
1181	ci->sector_count = 0;
1182}
1183
1184static int __clone_and_map_discard(struct clone_info *ci)
 
 
1185{
1186	struct dm_target *ti;
1187	sector_t len;
 
1188
1189	do {
1190		ti = dm_table_find_target(ci->map, ci->sector);
1191		if (!dm_target_is_valid(ti))
1192			return -EIO;
1193
1194		/*
1195		 * Even though the device advertised discard support,
1196		 * that does not mean every target supports it, and
1197		 * reconfiguration might also have changed that since the
1198		 * check was performed.
1199		 */
1200		if (!ti->num_discard_requests)
 
1201			return -EOPNOTSUPP;
1202
1203		len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
 
 
 
1204
1205		__issue_target_requests(ci, ti, ti->num_discard_requests, len);
1206
1207		ci->sector += len;
1208	} while (ci->sector_count -= len);
1209
1210	return 0;
1211}
1212
1213static int __clone_and_map(struct clone_info *ci)
1214{
1215	struct bio *clone, *bio = ci->bio;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1216	struct dm_target *ti;
1217	sector_t len = 0, max;
1218	struct dm_target_io *tio;
1219
1220	if (unlikely(bio->bi_rw & REQ_DISCARD))
1221		return __clone_and_map_discard(ci);
 
 
1222
1223	ti = dm_table_find_target(ci->map, ci->sector);
1224	if (!dm_target_is_valid(ti))
1225		return -EIO;
1226
1227	max = max_io_len(ci->sector, ti);
1228
1229	if (ci->sector_count <= max) {
1230		/*
1231		 * Optimise for the simple case where we can do all of
1232		 * the remaining io with a single clone.
1233		 */
1234		__clone_and_map_simple(ci, ti);
1235
1236	} else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1237		/*
1238		 * There are some bvecs that don't span targets.
1239		 * Do as many of these as possible.
1240		 */
1241		int i;
1242		sector_t remaining = max;
1243		sector_t bv_len;
1244
1245		for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1246			bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1247
1248			if (bv_len > remaining)
1249				break;
1250
1251			remaining -= bv_len;
1252			len += bv_len;
1253		}
1254
1255		tio = alloc_tio(ci, ti);
1256		clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
1257				  ci->md->bs);
1258		__map_bio(ti, clone, tio);
1259
1260		ci->sector += len;
1261		ci->sector_count -= len;
1262		ci->idx = i;
1263
1264	} else {
1265		/*
1266		 * Handle a bvec that must be split between two or more targets.
1267		 */
1268		struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1269		sector_t remaining = to_sector(bv->bv_len);
1270		unsigned int offset = 0;
1271
1272		do {
1273			if (offset) {
1274				ti = dm_table_find_target(ci->map, ci->sector);
1275				if (!dm_target_is_valid(ti))
1276					return -EIO;
1277
1278				max = max_io_len(ci->sector, ti);
1279			}
1280
1281			len = min(remaining, max);
1282
1283			tio = alloc_tio(ci, ti);
1284			clone = split_bvec(bio, ci->sector, ci->idx,
1285					   bv->bv_offset + offset, len,
1286					   ci->md->bs);
1287
1288			__map_bio(ti, clone, tio);
1289
1290			ci->sector += len;
1291			ci->sector_count -= len;
1292			offset += to_bytes(len);
1293		} while (remaining -= len);
1294
1295		ci->idx++;
1296	}
1297
1298	return 0;
1299}
1300
1301/*
1302 * Split the bio into several clones and submit it to targets.
1303 */
1304static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
 
1305{
1306	struct clone_info ci;
1307	int error = 0;
1308
1309	ci.map = dm_get_live_table(md);
1310	if (unlikely(!ci.map)) {
1311		bio_io_error(bio);
1312		return;
1313	}
1314
 
1315	ci.md = md;
1316	ci.io = alloc_io(md);
1317	ci.io->error = 0;
1318	atomic_set(&ci.io->io_count, 1);
1319	ci.io->bio = bio;
1320	ci.io->md = md;
1321	spin_lock_init(&ci.io->endio_lock);
1322	ci.sector = bio->bi_sector;
1323	ci.idx = bio->bi_idx;
1324
1325	start_io_acct(ci.io);
 
1326	if (bio->bi_rw & REQ_FLUSH) {
1327		ci.bio = &ci.md->flush_bio;
1328		ci.sector_count = 0;
1329		error = __clone_and_map_empty_flush(&ci);
1330		/* dec_pending submits any data associated with flush */
1331	} else {
1332		ci.bio = bio;
1333		ci.sector_count = bio_sectors(bio);
1334		while (ci.sector_count && !error)
1335			error = __clone_and_map(&ci);
1336	}
1337
1338	/* drop the extra reference count */
1339	dec_pending(ci.io, error);
1340	dm_table_put(ci.map);
1341}
1342/*-----------------------------------------------------------------
1343 * CRUD END
1344 *---------------------------------------------------------------*/
1345
1346static int dm_merge_bvec(struct request_queue *q,
1347			 struct bvec_merge_data *bvm,
1348			 struct bio_vec *biovec)
1349{
1350	struct mapped_device *md = q->queuedata;
1351	struct dm_table *map = dm_get_live_table(md);
1352	struct dm_target *ti;
1353	sector_t max_sectors;
1354	int max_size = 0;
1355
1356	if (unlikely(!map))
1357		goto out;
1358
1359	ti = dm_table_find_target(map, bvm->bi_sector);
1360	if (!dm_target_is_valid(ti))
1361		goto out_table;
1362
1363	/*
1364	 * Find maximum amount of I/O that won't need splitting
1365	 */
1366	max_sectors = min(max_io_len(bvm->bi_sector, ti),
1367			  (sector_t) BIO_MAX_SECTORS);
1368	max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1369	if (max_size < 0)
1370		max_size = 0;
1371
1372	/*
1373	 * merge_bvec_fn() returns number of bytes
1374	 * it can accept at this offset
1375	 * max is precomputed maximal io size
1376	 */
1377	if (max_size && ti->type->merge)
1378		max_size = ti->type->merge(ti, bvm, biovec, max_size);
1379	/*
1380	 * If the target doesn't support merge method and some of the devices
1381	 * provided their merge_bvec method (we know this by looking at
1382	 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1383	 * entries.  So always set max_size to 0, and the code below allows
1384	 * just one page.
1385	 */
1386	else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1387
1388		max_size = 0;
1389
1390out_table:
1391	dm_table_put(map);
1392
1393out:
 
1394	/*
1395	 * Always allow an entire first page
1396	 */
1397	if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1398		max_size = biovec->bv_len;
1399
1400	return max_size;
1401}
1402
1403/*
1404 * The request function that just remaps the bio built up by
1405 * dm_merge_bvec.
1406 */
1407static void _dm_request(struct request_queue *q, struct bio *bio)
1408{
1409	int rw = bio_data_dir(bio);
1410	struct mapped_device *md = q->queuedata;
1411	int cpu;
 
 
1412
1413	down_read(&md->io_lock);
1414
1415	cpu = part_stat_lock();
1416	part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1417	part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1418	part_stat_unlock();
1419
1420	/* if we're suspended, we have to queue this io for later */
1421	if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1422		up_read(&md->io_lock);
1423
1424		if (bio_rw(bio) != READA)
1425			queue_io(md, bio);
1426		else
1427			bio_io_error(bio);
1428		return;
1429	}
1430
1431	__split_and_process_bio(md, bio);
1432	up_read(&md->io_lock);
1433	return;
1434}
1435
1436static int dm_request_based(struct mapped_device *md)
1437{
1438	return blk_queue_stackable(md->queue);
1439}
1440
1441static void dm_request(struct request_queue *q, struct bio *bio)
1442{
1443	struct mapped_device *md = q->queuedata;
1444
1445	if (dm_request_based(md))
1446		blk_queue_bio(q, bio);
1447	else
1448		_dm_request(q, bio);
1449}
1450
1451void dm_dispatch_request(struct request *rq)
1452{
1453	int r;
1454
1455	if (blk_queue_io_stat(rq->q))
1456		rq->cmd_flags |= REQ_IO_STAT;
1457
1458	rq->start_time = jiffies;
1459	r = blk_insert_cloned_request(rq->q, rq);
1460	if (r)
1461		dm_complete_request(rq, r);
1462}
1463EXPORT_SYMBOL_GPL(dm_dispatch_request);
1464
1465static void dm_rq_bio_destructor(struct bio *bio)
1466{
1467	struct dm_rq_clone_bio_info *info = bio->bi_private;
1468	struct mapped_device *md = info->tio->md;
1469
1470	free_bio_info(info);
1471	bio_free(bio, md->bs);
1472}
1473
1474static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1475				 void *data)
1476{
1477	struct dm_rq_target_io *tio = data;
1478	struct mapped_device *md = tio->md;
1479	struct dm_rq_clone_bio_info *info = alloc_bio_info(md);
1480
1481	if (!info)
1482		return -ENOMEM;
1483
1484	info->orig = bio_orig;
1485	info->tio = tio;
1486	bio->bi_end_io = end_clone_bio;
1487	bio->bi_private = info;
1488	bio->bi_destructor = dm_rq_bio_destructor;
1489
1490	return 0;
1491}
1492
1493static int setup_clone(struct request *clone, struct request *rq,
1494		       struct dm_rq_target_io *tio)
1495{
1496	int r;
1497
1498	r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1499			      dm_rq_bio_constructor, tio);
1500	if (r)
1501		return r;
1502
1503	clone->cmd = rq->cmd;
1504	clone->cmd_len = rq->cmd_len;
1505	clone->sense = rq->sense;
1506	clone->buffer = rq->buffer;
1507	clone->end_io = end_clone_request;
1508	clone->end_io_data = tio;
1509
1510	return 0;
1511}
1512
1513static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1514				gfp_t gfp_mask)
1515{
1516	struct request *clone;
1517	struct dm_rq_target_io *tio;
1518
1519	tio = alloc_rq_tio(md, gfp_mask);
1520	if (!tio)
1521		return NULL;
1522
1523	tio->md = md;
1524	tio->ti = NULL;
1525	tio->orig = rq;
1526	tio->error = 0;
1527	memset(&tio->info, 0, sizeof(tio->info));
1528
1529	clone = &tio->clone;
1530	if (setup_clone(clone, rq, tio)) {
1531		/* -ENOMEM */
1532		free_rq_tio(tio);
1533		return NULL;
1534	}
1535
1536	return clone;
1537}
1538
1539/*
1540 * Called with the queue lock held.
1541 */
1542static int dm_prep_fn(struct request_queue *q, struct request *rq)
1543{
1544	struct mapped_device *md = q->queuedata;
1545	struct request *clone;
1546
1547	if (unlikely(rq->special)) {
1548		DMWARN("Already has something in rq->special.");
1549		return BLKPREP_KILL;
1550	}
1551
1552	clone = clone_rq(rq, md, GFP_ATOMIC);
1553	if (!clone)
1554		return BLKPREP_DEFER;
1555
1556	rq->special = clone;
1557	rq->cmd_flags |= REQ_DONTPREP;
1558
1559	return BLKPREP_OK;
1560}
1561
1562/*
1563 * Returns:
1564 * 0  : the request has been processed (not requeued)
1565 * !0 : the request has been requeued
1566 */
1567static int map_request(struct dm_target *ti, struct request *clone,
1568		       struct mapped_device *md)
1569{
1570	int r, requeued = 0;
1571	struct dm_rq_target_io *tio = clone->end_io_data;
1572
1573	tio->ti = ti;
1574	r = ti->type->map_rq(ti, clone, &tio->info);
1575	switch (r) {
1576	case DM_MAPIO_SUBMITTED:
1577		/* The target has taken the I/O to submit by itself later */
1578		break;
1579	case DM_MAPIO_REMAPPED:
1580		/* The target has remapped the I/O so dispatch it */
1581		trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1582				     blk_rq_pos(tio->orig));
1583		dm_dispatch_request(clone);
1584		break;
1585	case DM_MAPIO_REQUEUE:
1586		/* The target wants to requeue the I/O */
1587		dm_requeue_unmapped_request(clone);
1588		requeued = 1;
1589		break;
1590	default:
1591		if (r > 0) {
1592			DMWARN("unimplemented target map return value: %d", r);
1593			BUG();
1594		}
1595
1596		/* The target wants to complete the I/O */
1597		dm_kill_unmapped_request(clone, r);
1598		break;
1599	}
1600
1601	return requeued;
1602}
1603
1604static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1605{
1606	struct request *clone;
1607
1608	blk_start_request(orig);
1609	clone = orig->special;
1610	atomic_inc(&md->pending[rq_data_dir(clone)]);
1611
1612	/*
1613	 * Hold the md reference here for the in-flight I/O.
1614	 * We can't rely on the reference count by device opener,
1615	 * because the device may be closed during the request completion
1616	 * when all bios are completed.
1617	 * See the comment in rq_completed() too.
1618	 */
1619	dm_get(md);
1620
1621	return clone;
1622}
1623
1624/*
1625 * q->request_fn for request-based dm.
1626 * Called with the queue lock held.
1627 */
1628static void dm_request_fn(struct request_queue *q)
1629{
1630	struct mapped_device *md = q->queuedata;
1631	struct dm_table *map = dm_get_live_table(md);
 
1632	struct dm_target *ti;
1633	struct request *rq, *clone;
1634	sector_t pos;
1635
1636	/*
1637	 * For suspend, check blk_queue_stopped() and increment
1638	 * ->pending within a single queue_lock not to increment the
1639	 * number of in-flight I/Os after the queue is stopped in
1640	 * dm_suspend().
1641	 */
1642	while (!blk_queue_stopped(q)) {
1643		rq = blk_peek_request(q);
1644		if (!rq)
1645			goto delay_and_out;
1646
1647		/* always use block 0 to find the target for flushes for now */
1648		pos = 0;
1649		if (!(rq->cmd_flags & REQ_FLUSH))
1650			pos = blk_rq_pos(rq);
1651
1652		ti = dm_table_find_target(map, pos);
1653		if (!dm_target_is_valid(ti)) {
1654			/*
1655			 * Must perform setup, that dm_done() requires,
1656			 * before calling dm_kill_unmapped_request
1657			 */
1658			DMERR_LIMIT("request attempted access beyond the end of device");
1659			clone = dm_start_request(md, rq);
1660			dm_kill_unmapped_request(clone, -EIO);
1661			continue;
1662		}
1663
1664		if (ti->type->busy && ti->type->busy(ti))
1665			goto delay_and_out;
1666
1667		clone = dm_start_request(md, rq);
1668
1669		spin_unlock(q->queue_lock);
1670		if (map_request(ti, clone, md))
1671			goto requeued;
1672
1673		BUG_ON(!irqs_disabled());
1674		spin_lock(q->queue_lock);
1675	}
1676
1677	goto out;
1678
1679requeued:
1680	BUG_ON(!irqs_disabled());
1681	spin_lock(q->queue_lock);
1682
1683delay_and_out:
1684	blk_delay_queue(q, HZ / 10);
1685out:
1686	dm_table_put(map);
1687}
1688
1689int dm_underlying_device_busy(struct request_queue *q)
1690{
1691	return blk_lld_busy(q);
1692}
1693EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1694
1695static int dm_lld_busy(struct request_queue *q)
1696{
1697	int r;
1698	struct mapped_device *md = q->queuedata;
1699	struct dm_table *map = dm_get_live_table(md);
1700
1701	if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1702		r = 1;
1703	else
1704		r = dm_table_any_busy_target(map);
1705
1706	dm_table_put(map);
1707
1708	return r;
1709}
1710
1711static int dm_any_congested(void *congested_data, int bdi_bits)
1712{
1713	int r = bdi_bits;
1714	struct mapped_device *md = congested_data;
1715	struct dm_table *map;
1716
1717	if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1718		map = dm_get_live_table(md);
1719		if (map) {
1720			/*
1721			 * Request-based dm cares about only own queue for
1722			 * the query about congestion status of request_queue
1723			 */
1724			if (dm_request_based(md))
1725				r = md->queue->backing_dev_info.state &
1726				    bdi_bits;
1727			else
1728				r = dm_table_any_congested(map, bdi_bits);
1729
1730			dm_table_put(map);
1731		}
 
1732	}
1733
1734	return r;
1735}
1736
1737/*-----------------------------------------------------------------
1738 * An IDR is used to keep track of allocated minor numbers.
1739 *---------------------------------------------------------------*/
1740static void free_minor(int minor)
1741{
1742	spin_lock(&_minor_lock);
1743	idr_remove(&_minor_idr, minor);
1744	spin_unlock(&_minor_lock);
1745}
1746
1747/*
1748 * See if the device with a specific minor # is free.
1749 */
1750static int specific_minor(int minor)
1751{
1752	int r, m;
1753
1754	if (minor >= (1 << MINORBITS))
1755		return -EINVAL;
1756
1757	r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1758	if (!r)
1759		return -ENOMEM;
1760
1761	spin_lock(&_minor_lock);
1762
1763	if (idr_find(&_minor_idr, minor)) {
1764		r = -EBUSY;
1765		goto out;
1766	}
1767
1768	r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1769	if (r)
1770		goto out;
1771
1772	if (m != minor) {
1773		idr_remove(&_minor_idr, m);
1774		r = -EBUSY;
1775		goto out;
1776	}
1777
1778out:
1779	spin_unlock(&_minor_lock);
1780	return r;
 
 
 
1781}
1782
1783static int next_free_minor(int *minor)
1784{
1785	int r, m;
1786
1787	r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1788	if (!r)
1789		return -ENOMEM;
1790
 
1791	spin_lock(&_minor_lock);
1792
1793	r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1794	if (r)
1795		goto out;
1796
1797	if (m >= (1 << MINORBITS)) {
1798		idr_remove(&_minor_idr, m);
1799		r = -ENOSPC;
1800		goto out;
1801	}
1802
1803	*minor = m;
1804
1805out:
1806	spin_unlock(&_minor_lock);
1807	return r;
 
 
 
 
1808}
1809
1810static const struct block_device_operations dm_blk_dops;
1811
1812static void dm_wq_work(struct work_struct *work);
1813
1814static void dm_init_md_queue(struct mapped_device *md)
1815{
1816	/*
1817	 * Request-based dm devices cannot be stacked on top of bio-based dm
1818	 * devices.  The type of this dm device has not been decided yet.
1819	 * The type is decided at the first table loading time.
1820	 * To prevent problematic device stacking, clear the queue flag
1821	 * for request stacking support until then.
1822	 *
1823	 * This queue is new, so no concurrency on the queue_flags.
1824	 */
1825	queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1826
1827	md->queue->queuedata = md;
1828	md->queue->backing_dev_info.congested_fn = dm_any_congested;
1829	md->queue->backing_dev_info.congested_data = md;
1830	blk_queue_make_request(md->queue, dm_request);
1831	blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1832	blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1833}
1834
1835/*
1836 * Allocate and initialise a blank device with a given minor.
1837 */
1838static struct mapped_device *alloc_dev(int minor)
1839{
1840	int r;
1841	struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1842	void *old_md;
1843
1844	if (!md) {
1845		DMWARN("unable to allocate device, out of memory.");
1846		return NULL;
1847	}
1848
1849	if (!try_module_get(THIS_MODULE))
1850		goto bad_module_get;
1851
1852	/* get a minor number for the dev */
1853	if (minor == DM_ANY_MINOR)
1854		r = next_free_minor(&minor);
1855	else
1856		r = specific_minor(minor);
1857	if (r < 0)
1858		goto bad_minor;
1859
 
 
 
 
1860	md->type = DM_TYPE_NONE;
1861	init_rwsem(&md->io_lock);
1862	mutex_init(&md->suspend_lock);
1863	mutex_init(&md->type_lock);
1864	spin_lock_init(&md->deferred_lock);
1865	rwlock_init(&md->map_lock);
1866	atomic_set(&md->holders, 1);
1867	atomic_set(&md->open_count, 0);
1868	atomic_set(&md->event_nr, 0);
1869	atomic_set(&md->uevent_seq, 0);
1870	INIT_LIST_HEAD(&md->uevent_list);
1871	spin_lock_init(&md->uevent_lock);
1872
1873	md->queue = blk_alloc_queue(GFP_KERNEL);
1874	if (!md->queue)
1875		goto bad_queue;
1876
1877	dm_init_md_queue(md);
1878
1879	md->disk = alloc_disk(1);
1880	if (!md->disk)
1881		goto bad_disk;
1882
1883	atomic_set(&md->pending[0], 0);
1884	atomic_set(&md->pending[1], 0);
1885	init_waitqueue_head(&md->wait);
1886	INIT_WORK(&md->work, dm_wq_work);
1887	init_waitqueue_head(&md->eventq);
 
1888
1889	md->disk->major = _major;
1890	md->disk->first_minor = minor;
1891	md->disk->fops = &dm_blk_dops;
1892	md->disk->queue = md->queue;
1893	md->disk->private_data = md;
1894	sprintf(md->disk->disk_name, "dm-%d", minor);
1895	add_disk(md->disk);
1896	format_dev_t(md->name, MKDEV(_major, minor));
1897
1898	md->wq = alloc_workqueue("kdmflush",
1899				 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
1900	if (!md->wq)
1901		goto bad_thread;
1902
1903	md->bdev = bdget_disk(md->disk, 0);
1904	if (!md->bdev)
1905		goto bad_bdev;
1906
1907	bio_init(&md->flush_bio);
1908	md->flush_bio.bi_bdev = md->bdev;
1909	md->flush_bio.bi_rw = WRITE_FLUSH;
1910
 
 
1911	/* Populate the mapping, nobody knows we exist yet */
1912	spin_lock(&_minor_lock);
1913	old_md = idr_replace(&_minor_idr, md, minor);
1914	spin_unlock(&_minor_lock);
1915
1916	BUG_ON(old_md != MINOR_ALLOCED);
1917
1918	return md;
1919
1920bad_bdev:
1921	destroy_workqueue(md->wq);
1922bad_thread:
1923	del_gendisk(md->disk);
1924	put_disk(md->disk);
1925bad_disk:
1926	blk_cleanup_queue(md->queue);
1927bad_queue:
 
 
1928	free_minor(minor);
1929bad_minor:
1930	module_put(THIS_MODULE);
1931bad_module_get:
1932	kfree(md);
1933	return NULL;
1934}
1935
1936static void unlock_fs(struct mapped_device *md);
1937
1938static void free_dev(struct mapped_device *md)
1939{
1940	int minor = MINOR(disk_devt(md->disk));
1941
1942	unlock_fs(md);
1943	bdput(md->bdev);
1944	destroy_workqueue(md->wq);
1945	if (md->tio_pool)
1946		mempool_destroy(md->tio_pool);
1947	if (md->io_pool)
1948		mempool_destroy(md->io_pool);
1949	if (md->bs)
1950		bioset_free(md->bs);
1951	blk_integrity_unregister(md->disk);
1952	del_gendisk(md->disk);
 
1953	free_minor(minor);
1954
1955	spin_lock(&_minor_lock);
1956	md->disk->private_data = NULL;
1957	spin_unlock(&_minor_lock);
1958
1959	put_disk(md->disk);
1960	blk_cleanup_queue(md->queue);
 
1961	module_put(THIS_MODULE);
1962	kfree(md);
1963}
1964
1965static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1966{
1967	struct dm_md_mempools *p;
1968
1969	if (md->io_pool && md->tio_pool && md->bs)
1970		/* the md already has necessary mempools */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1971		goto out;
 
1972
1973	p = dm_table_get_md_mempools(t);
1974	BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
1975
1976	md->io_pool = p->io_pool;
1977	p->io_pool = NULL;
1978	md->tio_pool = p->tio_pool;
1979	p->tio_pool = NULL;
1980	md->bs = p->bs;
1981	p->bs = NULL;
1982
1983out:
1984	/* mempool bind completed, now no need any mempools in the table */
1985	dm_table_free_md_mempools(t);
1986}
1987
1988/*
1989 * Bind a table to the device.
1990 */
1991static void event_callback(void *context)
1992{
1993	unsigned long flags;
1994	LIST_HEAD(uevents);
1995	struct mapped_device *md = (struct mapped_device *) context;
1996
1997	spin_lock_irqsave(&md->uevent_lock, flags);
1998	list_splice_init(&md->uevent_list, &uevents);
1999	spin_unlock_irqrestore(&md->uevent_lock, flags);
2000
2001	dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2002
2003	atomic_inc(&md->event_nr);
2004	wake_up(&md->eventq);
2005}
2006
2007/*
2008 * Protected by md->suspend_lock obtained by dm_swap_table().
2009 */
2010static void __set_size(struct mapped_device *md, sector_t size)
2011{
2012	set_capacity(md->disk, size);
2013
2014	i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2015}
2016
2017/*
2018 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2019 *
2020 * If this function returns 0, then the device is either a non-dm
2021 * device without a merge_bvec_fn, or it is a dm device that is
2022 * able to split any bios it receives that are too big.
2023 */
2024int dm_queue_merge_is_compulsory(struct request_queue *q)
2025{
2026	struct mapped_device *dev_md;
2027
2028	if (!q->merge_bvec_fn)
2029		return 0;
2030
2031	if (q->make_request_fn == dm_request) {
2032		dev_md = q->queuedata;
2033		if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2034			return 0;
2035	}
2036
2037	return 1;
2038}
2039
2040static int dm_device_merge_is_compulsory(struct dm_target *ti,
2041					 struct dm_dev *dev, sector_t start,
2042					 sector_t len, void *data)
2043{
2044	struct block_device *bdev = dev->bdev;
2045	struct request_queue *q = bdev_get_queue(bdev);
2046
2047	return dm_queue_merge_is_compulsory(q);
2048}
2049
2050/*
2051 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2052 * on the properties of the underlying devices.
2053 */
2054static int dm_table_merge_is_optional(struct dm_table *table)
2055{
2056	unsigned i = 0;
2057	struct dm_target *ti;
2058
2059	while (i < dm_table_get_num_targets(table)) {
2060		ti = dm_table_get_target(table, i++);
2061
2062		if (ti->type->iterate_devices &&
2063		    ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2064			return 0;
2065	}
2066
2067	return 1;
2068}
2069
2070/*
2071 * Returns old map, which caller must destroy.
2072 */
2073static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2074			       struct queue_limits *limits)
2075{
2076	struct dm_table *old_map;
2077	struct request_queue *q = md->queue;
2078	sector_t size;
2079	unsigned long flags;
2080	int merge_is_optional;
2081
2082	size = dm_table_get_size(t);
2083
2084	/*
2085	 * Wipe any geometry if the size of the table changed.
2086	 */
2087	if (size != get_capacity(md->disk))
2088		memset(&md->geometry, 0, sizeof(md->geometry));
2089
2090	__set_size(md, size);
2091
2092	dm_table_event_callback(t, event_callback, md);
2093
2094	/*
2095	 * The queue hasn't been stopped yet, if the old table type wasn't
2096	 * for request-based during suspension.  So stop it to prevent
2097	 * I/O mapping before resume.
2098	 * This must be done before setting the queue restrictions,
2099	 * because request-based dm may be run just after the setting.
2100	 */
2101	if (dm_table_request_based(t) && !blk_queue_stopped(q))
2102		stop_queue(q);
2103
2104	__bind_mempools(md, t);
2105
2106	merge_is_optional = dm_table_merge_is_optional(t);
2107
2108	write_lock_irqsave(&md->map_lock, flags);
2109	old_map = md->map;
2110	md->map = t;
2111	md->immutable_target_type = dm_table_get_immutable_target_type(t);
2112
2113	dm_table_set_restrictions(t, q, limits);
2114	if (merge_is_optional)
2115		set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2116	else
2117		clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2118	write_unlock_irqrestore(&md->map_lock, flags);
2119
2120	return old_map;
2121}
2122
2123/*
2124 * Returns unbound table for the caller to free.
2125 */
2126static struct dm_table *__unbind(struct mapped_device *md)
2127{
2128	struct dm_table *map = md->map;
2129	unsigned long flags;
2130
2131	if (!map)
2132		return NULL;
2133
2134	dm_table_event_callback(map, NULL, NULL);
2135	write_lock_irqsave(&md->map_lock, flags);
2136	md->map = NULL;
2137	write_unlock_irqrestore(&md->map_lock, flags);
2138
2139	return map;
2140}
2141
2142/*
2143 * Constructor for a new device.
2144 */
2145int dm_create(int minor, struct mapped_device **result)
2146{
2147	struct mapped_device *md;
2148
2149	md = alloc_dev(minor);
2150	if (!md)
2151		return -ENXIO;
2152
2153	dm_sysfs_init(md);
2154
2155	*result = md;
2156	return 0;
2157}
2158
2159/*
2160 * Functions to manage md->type.
2161 * All are required to hold md->type_lock.
2162 */
2163void dm_lock_md_type(struct mapped_device *md)
2164{
2165	mutex_lock(&md->type_lock);
2166}
2167
2168void dm_unlock_md_type(struct mapped_device *md)
2169{
2170	mutex_unlock(&md->type_lock);
2171}
2172
2173void dm_set_md_type(struct mapped_device *md, unsigned type)
2174{
 
2175	md->type = type;
2176}
2177
2178unsigned dm_get_md_type(struct mapped_device *md)
2179{
 
2180	return md->type;
2181}
2182
2183struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2184{
2185	return md->immutable_target_type;
2186}
2187
2188/*
 
 
 
 
 
 
 
 
 
 
 
2189 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2190 */
2191static int dm_init_request_based_queue(struct mapped_device *md)
2192{
2193	struct request_queue *q = NULL;
2194
2195	if (md->queue->elevator)
2196		return 1;
2197
2198	/* Fully initialize the queue */
2199	q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2200	if (!q)
2201		return 0;
2202
2203	md->queue = q;
2204	dm_init_md_queue(md);
2205	blk_queue_softirq_done(md->queue, dm_softirq_done);
2206	blk_queue_prep_rq(md->queue, dm_prep_fn);
2207	blk_queue_lld_busy(md->queue, dm_lld_busy);
2208
2209	elv_register_queue(md->queue);
2210
2211	return 1;
2212}
2213
2214/*
2215 * Setup the DM device's queue based on md's type
2216 */
2217int dm_setup_md_queue(struct mapped_device *md)
2218{
2219	if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2220	    !dm_init_request_based_queue(md)) {
2221		DMWARN("Cannot initialize queue for request-based mapped device");
2222		return -EINVAL;
2223	}
2224
2225	return 0;
2226}
2227
2228static struct mapped_device *dm_find_md(dev_t dev)
2229{
2230	struct mapped_device *md;
2231	unsigned minor = MINOR(dev);
2232
2233	if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2234		return NULL;
2235
2236	spin_lock(&_minor_lock);
2237
2238	md = idr_find(&_minor_idr, minor);
2239	if (md && (md == MINOR_ALLOCED ||
2240		   (MINOR(disk_devt(dm_disk(md))) != minor) ||
2241		   dm_deleting_md(md) ||
2242		   test_bit(DMF_FREEING, &md->flags))) {
2243		md = NULL;
2244		goto out;
2245	}
2246
2247out:
2248	spin_unlock(&_minor_lock);
2249
2250	return md;
2251}
2252
2253struct mapped_device *dm_get_md(dev_t dev)
2254{
2255	struct mapped_device *md = dm_find_md(dev);
2256
2257	if (md)
2258		dm_get(md);
2259
2260	return md;
2261}
2262EXPORT_SYMBOL_GPL(dm_get_md);
2263
2264void *dm_get_mdptr(struct mapped_device *md)
2265{
2266	return md->interface_ptr;
2267}
2268
2269void dm_set_mdptr(struct mapped_device *md, void *ptr)
2270{
2271	md->interface_ptr = ptr;
2272}
2273
2274void dm_get(struct mapped_device *md)
2275{
2276	atomic_inc(&md->holders);
2277	BUG_ON(test_bit(DMF_FREEING, &md->flags));
2278}
2279
2280const char *dm_device_name(struct mapped_device *md)
2281{
2282	return md->name;
2283}
2284EXPORT_SYMBOL_GPL(dm_device_name);
2285
2286static void __dm_destroy(struct mapped_device *md, bool wait)
2287{
2288	struct dm_table *map;
 
2289
2290	might_sleep();
2291
2292	spin_lock(&_minor_lock);
2293	map = dm_get_live_table(md);
2294	idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2295	set_bit(DMF_FREEING, &md->flags);
2296	spin_unlock(&_minor_lock);
2297
2298	if (!dm_suspended_md(md)) {
2299		dm_table_presuspend_targets(map);
2300		dm_table_postsuspend_targets(map);
2301	}
2302
 
 
 
2303	/*
2304	 * Rare, but there may be I/O requests still going to complete,
2305	 * for example.  Wait for all references to disappear.
2306	 * No one should increment the reference count of the mapped_device,
2307	 * after the mapped_device state becomes DMF_FREEING.
2308	 */
2309	if (wait)
2310		while (atomic_read(&md->holders))
2311			msleep(1);
2312	else if (atomic_read(&md->holders))
2313		DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2314		       dm_device_name(md), atomic_read(&md->holders));
2315
2316	dm_sysfs_exit(md);
2317	dm_table_put(map);
2318	dm_table_destroy(__unbind(md));
2319	free_dev(md);
2320}
2321
2322void dm_destroy(struct mapped_device *md)
2323{
2324	__dm_destroy(md, true);
2325}
2326
2327void dm_destroy_immediate(struct mapped_device *md)
2328{
2329	__dm_destroy(md, false);
2330}
2331
2332void dm_put(struct mapped_device *md)
2333{
2334	atomic_dec(&md->holders);
2335}
2336EXPORT_SYMBOL_GPL(dm_put);
2337
2338static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2339{
2340	int r = 0;
2341	DECLARE_WAITQUEUE(wait, current);
2342
2343	add_wait_queue(&md->wait, &wait);
2344
2345	while (1) {
2346		set_current_state(interruptible);
2347
2348		if (!md_in_flight(md))
2349			break;
2350
2351		if (interruptible == TASK_INTERRUPTIBLE &&
2352		    signal_pending(current)) {
2353			r = -EINTR;
2354			break;
2355		}
2356
2357		io_schedule();
2358	}
2359	set_current_state(TASK_RUNNING);
2360
2361	remove_wait_queue(&md->wait, &wait);
2362
2363	return r;
2364}
2365
2366/*
2367 * Process the deferred bios
2368 */
2369static void dm_wq_work(struct work_struct *work)
2370{
2371	struct mapped_device *md = container_of(work, struct mapped_device,
2372						work);
2373	struct bio *c;
 
 
2374
2375	down_read(&md->io_lock);
2376
2377	while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2378		spin_lock_irq(&md->deferred_lock);
2379		c = bio_list_pop(&md->deferred);
2380		spin_unlock_irq(&md->deferred_lock);
2381
2382		if (!c)
2383			break;
2384
2385		up_read(&md->io_lock);
2386
2387		if (dm_request_based(md))
2388			generic_make_request(c);
2389		else
2390			__split_and_process_bio(md, c);
2391
2392		down_read(&md->io_lock);
2393	}
2394
2395	up_read(&md->io_lock);
2396}
2397
2398static void dm_queue_flush(struct mapped_device *md)
2399{
2400	clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2401	smp_mb__after_clear_bit();
2402	queue_work(md->wq, &md->work);
2403}
2404
2405/*
2406 * Swap in a new table, returning the old one for the caller to destroy.
2407 */
2408struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2409{
2410	struct dm_table *map = ERR_PTR(-EINVAL);
2411	struct queue_limits limits;
2412	int r;
2413
2414	mutex_lock(&md->suspend_lock);
2415
2416	/* device must be suspended */
2417	if (!dm_suspended_md(md))
2418		goto out;
2419
2420	r = dm_calculate_queue_limits(table, &limits);
2421	if (r) {
2422		map = ERR_PTR(r);
2423		goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2424	}
2425
2426	map = __bind(md, table, &limits);
2427
2428out:
2429	mutex_unlock(&md->suspend_lock);
2430	return map;
2431}
2432
2433/*
2434 * Functions to lock and unlock any filesystem running on the
2435 * device.
2436 */
2437static int lock_fs(struct mapped_device *md)
2438{
2439	int r;
2440
2441	WARN_ON(md->frozen_sb);
2442
2443	md->frozen_sb = freeze_bdev(md->bdev);
2444	if (IS_ERR(md->frozen_sb)) {
2445		r = PTR_ERR(md->frozen_sb);
2446		md->frozen_sb = NULL;
2447		return r;
2448	}
2449
2450	set_bit(DMF_FROZEN, &md->flags);
2451
2452	return 0;
2453}
2454
2455static void unlock_fs(struct mapped_device *md)
2456{
2457	if (!test_bit(DMF_FROZEN, &md->flags))
2458		return;
2459
2460	thaw_bdev(md->bdev, md->frozen_sb);
2461	md->frozen_sb = NULL;
2462	clear_bit(DMF_FROZEN, &md->flags);
2463}
2464
2465/*
2466 * We need to be able to change a mapping table under a mounted
2467 * filesystem.  For example we might want to move some data in
2468 * the background.  Before the table can be swapped with
2469 * dm_bind_table, dm_suspend must be called to flush any in
2470 * flight bios and ensure that any further io gets deferred.
2471 */
2472/*
2473 * Suspend mechanism in request-based dm.
2474 *
2475 * 1. Flush all I/Os by lock_fs() if needed.
2476 * 2. Stop dispatching any I/O by stopping the request_queue.
2477 * 3. Wait for all in-flight I/Os to be completed or requeued.
2478 *
2479 * To abort suspend, start the request_queue.
2480 */
2481int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2482{
2483	struct dm_table *map = NULL;
2484	int r = 0;
2485	int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2486	int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2487
2488	mutex_lock(&md->suspend_lock);
2489
2490	if (dm_suspended_md(md)) {
2491		r = -EINVAL;
2492		goto out_unlock;
2493	}
2494
2495	map = dm_get_live_table(md);
2496
2497	/*
2498	 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2499	 * This flag is cleared before dm_suspend returns.
2500	 */
2501	if (noflush)
2502		set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2503
2504	/* This does not get reverted if there's an error later. */
2505	dm_table_presuspend_targets(map);
2506
2507	/*
2508	 * Flush I/O to the device.
2509	 * Any I/O submitted after lock_fs() may not be flushed.
2510	 * noflush takes precedence over do_lockfs.
2511	 * (lock_fs() flushes I/Os and waits for them to complete.)
2512	 */
2513	if (!noflush && do_lockfs) {
2514		r = lock_fs(md);
2515		if (r)
2516			goto out;
2517	}
2518
2519	/*
2520	 * Here we must make sure that no processes are submitting requests
2521	 * to target drivers i.e. no one may be executing
2522	 * __split_and_process_bio. This is called from dm_request and
2523	 * dm_wq_work.
2524	 *
2525	 * To get all processes out of __split_and_process_bio in dm_request,
2526	 * we take the write lock. To prevent any process from reentering
2527	 * __split_and_process_bio from dm_request and quiesce the thread
2528	 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2529	 * flush_workqueue(md->wq).
2530	 */
2531	down_write(&md->io_lock);
2532	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2533	up_write(&md->io_lock);
2534
2535	/*
2536	 * Stop md->queue before flushing md->wq in case request-based
2537	 * dm defers requests to md->wq from md->queue.
2538	 */
2539	if (dm_request_based(md))
2540		stop_queue(md->queue);
2541
2542	flush_workqueue(md->wq);
2543
2544	/*
2545	 * At this point no more requests are entering target request routines.
2546	 * We call dm_wait_for_completion to wait for all existing requests
2547	 * to finish.
2548	 */
2549	r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2550
2551	down_write(&md->io_lock);
2552	if (noflush)
2553		clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2554	up_write(&md->io_lock);
2555
2556	/* were we interrupted ? */
2557	if (r < 0) {
2558		dm_queue_flush(md);
2559
2560		if (dm_request_based(md))
2561			start_queue(md->queue);
2562
2563		unlock_fs(md);
2564		goto out; /* pushback list is already flushed, so skip flush */
2565	}
2566
2567	/*
2568	 * If dm_wait_for_completion returned 0, the device is completely
2569	 * quiescent now. There is no request-processing activity. All new
2570	 * requests are being added to md->deferred list.
2571	 */
2572
2573	set_bit(DMF_SUSPENDED, &md->flags);
2574
2575	dm_table_postsuspend_targets(map);
2576
2577out:
2578	dm_table_put(map);
2579
2580out_unlock:
2581	mutex_unlock(&md->suspend_lock);
2582	return r;
2583}
2584
2585int dm_resume(struct mapped_device *md)
2586{
2587	int r = -EINVAL;
2588	struct dm_table *map = NULL;
2589
2590	mutex_lock(&md->suspend_lock);
2591	if (!dm_suspended_md(md))
2592		goto out;
2593
2594	map = dm_get_live_table(md);
2595	if (!map || !dm_table_get_size(map))
2596		goto out;
2597
2598	r = dm_table_resume_targets(map);
2599	if (r)
2600		goto out;
2601
2602	dm_queue_flush(md);
2603
2604	/*
2605	 * Flushing deferred I/Os must be done after targets are resumed
2606	 * so that mapping of targets can work correctly.
2607	 * Request-based dm is queueing the deferred I/Os in its request_queue.
2608	 */
2609	if (dm_request_based(md))
2610		start_queue(md->queue);
2611
2612	unlock_fs(md);
2613
2614	clear_bit(DMF_SUSPENDED, &md->flags);
2615
2616	r = 0;
2617out:
2618	dm_table_put(map);
2619	mutex_unlock(&md->suspend_lock);
2620
2621	return r;
2622}
2623
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2624/*-----------------------------------------------------------------
2625 * Event notification.
2626 *---------------------------------------------------------------*/
2627int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2628		       unsigned cookie)
2629{
2630	char udev_cookie[DM_COOKIE_LENGTH];
2631	char *envp[] = { udev_cookie, NULL };
2632
2633	if (!cookie)
2634		return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2635	else {
2636		snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2637			 DM_COOKIE_ENV_VAR_NAME, cookie);
2638		return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2639					  action, envp);
2640	}
2641}
2642
2643uint32_t dm_next_uevent_seq(struct mapped_device *md)
2644{
2645	return atomic_add_return(1, &md->uevent_seq);
2646}
2647
2648uint32_t dm_get_event_nr(struct mapped_device *md)
2649{
2650	return atomic_read(&md->event_nr);
2651}
2652
2653int dm_wait_event(struct mapped_device *md, int event_nr)
2654{
2655	return wait_event_interruptible(md->eventq,
2656			(event_nr != atomic_read(&md->event_nr)));
2657}
2658
2659void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2660{
2661	unsigned long flags;
2662
2663	spin_lock_irqsave(&md->uevent_lock, flags);
2664	list_add(elist, &md->uevent_list);
2665	spin_unlock_irqrestore(&md->uevent_lock, flags);
2666}
2667
2668/*
2669 * The gendisk is only valid as long as you have a reference
2670 * count on 'md'.
2671 */
2672struct gendisk *dm_disk(struct mapped_device *md)
2673{
2674	return md->disk;
2675}
2676
2677struct kobject *dm_kobject(struct mapped_device *md)
2678{
2679	return &md->kobj;
2680}
2681
2682/*
2683 * struct mapped_device should not be exported outside of dm.c
2684 * so use this check to verify that kobj is part of md structure
2685 */
2686struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2687{
2688	struct mapped_device *md;
2689
2690	md = container_of(kobj, struct mapped_device, kobj);
2691	if (&md->kobj != kobj)
2692		return NULL;
2693
2694	if (test_bit(DMF_FREEING, &md->flags) ||
2695	    dm_deleting_md(md))
2696		return NULL;
2697
2698	dm_get(md);
2699	return md;
2700}
2701
2702int dm_suspended_md(struct mapped_device *md)
2703{
2704	return test_bit(DMF_SUSPENDED, &md->flags);
2705}
2706
 
 
 
 
 
2707int dm_suspended(struct dm_target *ti)
2708{
2709	return dm_suspended_md(dm_table_get_md(ti->table));
2710}
2711EXPORT_SYMBOL_GPL(dm_suspended);
2712
2713int dm_noflush_suspending(struct dm_target *ti)
2714{
2715	return __noflush_suspending(dm_table_get_md(ti->table));
2716}
2717EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2718
2719struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity)
2720{
2721	struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2722	unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS;
 
 
2723
2724	if (!pools)
2725		return NULL;
2726
2727	pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2728			 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2729			 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2730	if (!pools->io_pool)
2731		goto free_pools_and_out;
 
 
 
 
 
 
 
2732
2733	pools->tio_pool = (type == DM_TYPE_BIO_BASED) ?
2734			  mempool_create_slab_pool(MIN_IOS, _tio_cache) :
2735			  mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2736	if (!pools->tio_pool)
2737		goto free_io_pool_and_out;
2738
2739	pools->bs = bioset_create(pool_size, 0);
2740	if (!pools->bs)
2741		goto free_tio_pool_and_out;
2742
2743	if (integrity && bioset_integrity_create(pools->bs, pool_size))
2744		goto free_bioset_and_out;
2745
2746	return pools;
2747
2748free_bioset_and_out:
2749	bioset_free(pools->bs);
2750
2751free_tio_pool_and_out:
2752	mempool_destroy(pools->tio_pool);
2753
2754free_io_pool_and_out:
2755	mempool_destroy(pools->io_pool);
2756
2757free_pools_and_out:
2758	kfree(pools);
2759
2760	return NULL;
2761}
2762
2763void dm_free_md_mempools(struct dm_md_mempools *pools)
2764{
2765	if (!pools)
2766		return;
2767
2768	if (pools->io_pool)
2769		mempool_destroy(pools->io_pool);
2770
2771	if (pools->tio_pool)
2772		mempool_destroy(pools->tio_pool);
2773
2774	if (pools->bs)
2775		bioset_free(pools->bs);
2776
2777	kfree(pools);
2778}
2779
2780static const struct block_device_operations dm_blk_dops = {
2781	.open = dm_blk_open,
2782	.release = dm_blk_close,
2783	.ioctl = dm_blk_ioctl,
2784	.getgeo = dm_blk_getgeo,
2785	.owner = THIS_MODULE
2786};
2787
2788EXPORT_SYMBOL(dm_get_mapinfo);
2789
2790/*
2791 * module hooks
2792 */
2793module_init(dm_init);
2794module_exit(dm_exit);
2795
2796module_param(major, uint, 0);
2797MODULE_PARM_DESC(major, "The major number of the device mapper");
 
 
 
 
 
 
 
2798MODULE_DESCRIPTION(DM_NAME " driver");
2799MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2800MODULE_LICENSE("GPL");