Linux Audio

Check our new training course

Real-Time Linux with PREEMPT_RT training

Feb 18-20, 2025
Register
Loading...
v3.15
 
   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
  53static void do_deferred_remove(struct work_struct *w);
  54
  55static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
  56
  57/*
  58 * For bio-based dm.
  59 * One of these is allocated per bio.
  60 */
  61struct dm_io {
  62	struct mapped_device *md;
  63	int error;
  64	atomic_t io_count;
  65	struct bio *bio;
  66	unsigned long start_time;
  67	spinlock_t endio_lock;
  68	struct dm_stats_aux stats_aux;
  69};
  70
  71/*
  72 * For request-based dm.
  73 * One of these is allocated per request.
  74 */
  75struct dm_rq_target_io {
  76	struct mapped_device *md;
  77	struct dm_target *ti;
  78	struct request *orig, clone;
  79	int error;
  80	union map_info info;
  81};
  82
  83/*
  84 * For request-based dm - the bio clones we allocate are embedded in these
  85 * structs.
  86 *
  87 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
  88 * the bioset is created - this means the bio has to come at the end of the
  89 * struct.
  90 */
  91struct dm_rq_clone_bio_info {
  92	struct bio *orig;
  93	struct dm_rq_target_io *tio;
  94	struct bio clone;
  95};
  96
  97union map_info *dm_get_rq_mapinfo(struct request *rq)
  98{
  99	if (rq && rq->end_io_data)
 100		return &((struct dm_rq_target_io *)rq->end_io_data)->info;
 101	return NULL;
 102}
 103EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
 104
 105#define MINOR_ALLOCED ((void *)-1)
 106
 107/*
 108 * Bits for the md->flags field.
 109 */
 110#define DMF_BLOCK_IO_FOR_SUSPEND 0
 111#define DMF_SUSPENDED 1
 112#define DMF_FROZEN 2
 113#define DMF_FREEING 3
 114#define DMF_DELETING 4
 115#define DMF_NOFLUSH_SUSPENDING 5
 116#define DMF_MERGE_IS_OPTIONAL 6
 117#define DMF_DEFERRED_REMOVE 7
 118
 119/*
 120 * A dummy definition to make RCU happy.
 121 * struct dm_table should never be dereferenced in this file.
 122 */
 123struct dm_table {
 124	int undefined__;
 125};
 126
 127/*
 128 * Work processed by per-device workqueue.
 129 */
 130struct mapped_device {
 131	struct srcu_struct io_barrier;
 132	struct mutex suspend_lock;
 133	atomic_t holders;
 134	atomic_t open_count;
 135
 136	/*
 137	 * The current mapping.
 138	 * Use dm_get_live_table{_fast} or take suspend_lock for
 139	 * dereference.
 140	 */
 141	struct dm_table *map;
 
 
 
 
 
 
 
 142
 143	unsigned long flags;
 144
 145	struct request_queue *queue;
 146	unsigned type;
 147	/* Protect queue and type against concurrent access. */
 148	struct mutex type_lock;
 149
 150	struct target_type *immutable_target_type;
 151
 152	struct gendisk *disk;
 153	char name[16];
 154
 155	void *interface_ptr;
 156
 157	/*
 158	 * A list of ios that arrived while we were suspended.
 159	 */
 160	atomic_t pending[2];
 161	wait_queue_head_t wait;
 162	struct work_struct work;
 163	struct bio_list deferred;
 164	spinlock_t deferred_lock;
 165
 166	/*
 167	 * Processing queue (flush)
 168	 */
 169	struct workqueue_struct *wq;
 
 
 
 170
 171	/*
 172	 * io objects are allocated from here.
 173	 */
 174	mempool_t *io_pool;
 175
 176	struct bio_set *bs;
 
 
 
 
 
 177
 178	/*
 179	 * Event handling.
 180	 */
 181	atomic_t event_nr;
 182	wait_queue_head_t eventq;
 183	atomic_t uevent_seq;
 184	struct list_head uevent_list;
 185	spinlock_t uevent_lock; /* Protect access to uevent_list */
 186
 187	/*
 188	 * freeze/thaw support require holding onto a super block
 189	 */
 190	struct super_block *frozen_sb;
 191	struct block_device *bdev;
 192
 193	/* forced geometry settings */
 194	struct hd_geometry geometry;
 195
 196	/* kobject and completion */
 197	struct dm_kobject_holder kobj_holder;
 
 
 
 198
 199	/* zero-length flush that will be cloned and submitted to targets */
 200	struct bio flush_bio;
 
 
 201
 202	struct dm_stats stats;
 
 
 
 203};
 204
 205/*
 206 * For mempools pre-allocation at the table loading time.
 207 */
 208struct dm_md_mempools {
 209	mempool_t *io_pool;
 210	struct bio_set *bs;
 211};
 212
 213#define RESERVED_BIO_BASED_IOS		16
 214#define RESERVED_REQUEST_BASED_IOS	256
 215#define RESERVED_MAX_IOS		1024
 216static struct kmem_cache *_io_cache;
 217static struct kmem_cache *_rq_tio_cache;
 218
 219/*
 220 * Bio-based DM's mempools' reserved IOs set by the user.
 221 */
 222static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
 
 
 
 
 
 
 
 
 223
 224/*
 225 * Request-based DM's mempools' reserved IOs set by the user.
 226 */
 227static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS;
 
 
 
 228
 229static unsigned __dm_get_reserved_ios(unsigned *reserved_ios,
 230				      unsigned def, unsigned max)
 231{
 232	unsigned ios = ACCESS_ONCE(*reserved_ios);
 233	unsigned modified_ios = 0;
 234
 235	if (!ios)
 236		modified_ios = def;
 237	else if (ios > max)
 238		modified_ios = max;
 239
 240	if (modified_ios) {
 241		(void)cmpxchg(reserved_ios, ios, modified_ios);
 242		ios = modified_ios;
 243	}
 244
 245	return ios;
 246}
 247
 248unsigned dm_get_reserved_bio_based_ios(void)
 249{
 250	return __dm_get_reserved_ios(&reserved_bio_based_ios,
 251				     RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS);
 252}
 253EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
 254
 255unsigned dm_get_reserved_rq_based_ios(void)
 256{
 257	return __dm_get_reserved_ios(&reserved_rq_based_ios,
 258				     RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS);
 259}
 260EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios);
 261
 262static int __init local_init(void)
 263{
 264	int r = -ENOMEM;
 265
 266	/* allocate a slab for the dm_ios */
 267	_io_cache = KMEM_CACHE(dm_io, 0);
 268	if (!_io_cache)
 269		return r;
 270
 271	_rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
 272	if (!_rq_tio_cache)
 273		goto out_free_io_cache;
 274
 275	r = dm_uevent_init();
 276	if (r)
 277		goto out_free_rq_tio_cache;
 
 
 
 
 
 
 278
 279	_major = major;
 280	r = register_blkdev(_major, _name);
 281	if (r < 0)
 282		goto out_uevent_exit;
 283
 284	if (!_major)
 285		_major = r;
 286
 287	return 0;
 288
 
 
 289out_uevent_exit:
 290	dm_uevent_exit();
 291out_free_rq_tio_cache:
 292	kmem_cache_destroy(_rq_tio_cache);
 293out_free_io_cache:
 294	kmem_cache_destroy(_io_cache);
 295
 296	return r;
 297}
 298
 299static void local_exit(void)
 300{
 301	flush_scheduled_work();
 302
 303	kmem_cache_destroy(_rq_tio_cache);
 304	kmem_cache_destroy(_io_cache);
 305	unregister_blkdev(_major, _name);
 306	dm_uevent_exit();
 307
 308	_major = 0;
 309
 310	DMINFO("cleaned up");
 311}
 312
 313static int (*_inits[])(void) __initdata = {
 314	local_init,
 315	dm_target_init,
 316	dm_linear_init,
 317	dm_stripe_init,
 318	dm_io_init,
 319	dm_kcopyd_init,
 320	dm_interface_init,
 321	dm_statistics_init,
 322};
 323
 324static void (*_exits[])(void) = {
 325	local_exit,
 326	dm_target_exit,
 327	dm_linear_exit,
 328	dm_stripe_exit,
 329	dm_io_exit,
 330	dm_kcopyd_exit,
 331	dm_interface_exit,
 332	dm_statistics_exit,
 333};
 334
 335static int __init dm_init(void)
 336{
 337	const int count = ARRAY_SIZE(_inits);
 338
 339	int r, i;
 340
 
 
 
 
 
 341	for (i = 0; i < count; i++) {
 342		r = _inits[i]();
 343		if (r)
 344			goto bad;
 345	}
 346
 347	return 0;
 348
 349      bad:
 350	while (i--)
 351		_exits[i]();
 352
 353	return r;
 354}
 355
 356static void __exit dm_exit(void)
 357{
 358	int i = ARRAY_SIZE(_exits);
 359
 360	while (i--)
 361		_exits[i]();
 362
 363	/*
 364	 * Should be empty by this point.
 365	 */
 366	idr_destroy(&_minor_idr);
 367}
 368
 369/*
 370 * Block device functions
 371 */
 372int dm_deleting_md(struct mapped_device *md)
 373{
 374	return test_bit(DMF_DELETING, &md->flags);
 375}
 376
 377static int dm_blk_open(struct block_device *bdev, fmode_t mode)
 378{
 379	struct mapped_device *md;
 380
 381	spin_lock(&_minor_lock);
 382
 383	md = bdev->bd_disk->private_data;
 384	if (!md)
 385		goto out;
 386
 387	if (test_bit(DMF_FREEING, &md->flags) ||
 388	    dm_deleting_md(md)) {
 389		md = NULL;
 390		goto out;
 391	}
 392
 393	dm_get(md);
 394	atomic_inc(&md->open_count);
 395
 396out:
 397	spin_unlock(&_minor_lock);
 398
 399	return md ? 0 : -ENXIO;
 400}
 401
 402static void dm_blk_close(struct gendisk *disk, fmode_t mode)
 403{
 404	struct mapped_device *md = disk->private_data;
 405
 406	spin_lock(&_minor_lock);
 407
 
 
 
 
 408	if (atomic_dec_and_test(&md->open_count) &&
 409	    (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
 410		schedule_work(&deferred_remove_work);
 411
 412	dm_put(md);
 413
 414	spin_unlock(&_minor_lock);
 415}
 416
 417int dm_open_count(struct mapped_device *md)
 418{
 419	return atomic_read(&md->open_count);
 420}
 421
 422/*
 423 * Guarantees nothing is using the device before it's deleted.
 424 */
 425int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
 426{
 427	int r = 0;
 428
 429	spin_lock(&_minor_lock);
 430
 431	if (dm_open_count(md)) {
 432		r = -EBUSY;
 433		if (mark_deferred)
 434			set_bit(DMF_DEFERRED_REMOVE, &md->flags);
 435	} else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
 436		r = -EEXIST;
 437	else
 438		set_bit(DMF_DELETING, &md->flags);
 439
 440	spin_unlock(&_minor_lock);
 441
 442	return r;
 443}
 444
 445int dm_cancel_deferred_remove(struct mapped_device *md)
 446{
 447	int r = 0;
 448
 449	spin_lock(&_minor_lock);
 450
 451	if (test_bit(DMF_DELETING, &md->flags))
 452		r = -EBUSY;
 453	else
 454		clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
 455
 456	spin_unlock(&_minor_lock);
 457
 458	return r;
 459}
 460
 461static void do_deferred_remove(struct work_struct *w)
 462{
 463	dm_deferred_remove();
 464}
 465
 466sector_t dm_get_size(struct mapped_device *md)
 467{
 468	return get_capacity(md->disk);
 469}
 470
 471struct request_queue *dm_get_md_queue(struct mapped_device *md)
 472{
 473	return md->queue;
 474}
 475
 476struct dm_stats *dm_get_stats(struct mapped_device *md)
 477{
 478	return &md->stats;
 479}
 480
 481static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 482{
 483	struct mapped_device *md = bdev->bd_disk->private_data;
 484
 485	return dm_get_geometry(md, geo);
 486}
 487
 488static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
 489			unsigned int cmd, unsigned long arg)
 490{
 491	struct mapped_device *md = bdev->bd_disk->private_data;
 492	int srcu_idx;
 493	struct dm_table *map;
 494	struct dm_target *tgt;
 495	int r = -ENOTTY;
 496
 497retry:
 498	map = dm_get_live_table(md, &srcu_idx);
 499
 500	if (!map || !dm_table_get_size(map))
 501		goto out;
 502
 503	/* We only support devices that have a single target */
 504	if (dm_table_get_num_targets(map) != 1)
 505		goto out;
 506
 507	tgt = dm_table_get_target(map, 0);
 
 
 508
 509	if (dm_suspended_md(md)) {
 510		r = -EAGAIN;
 511		goto out;
 
 
 
 
 
 512	}
 513
 514	if (tgt->type->ioctl)
 515		r = tgt->type->ioctl(tgt, cmd, arg);
 516
 517out:
 
 518	dm_put_live_table(md, srcu_idx);
 
 519
 520	if (r == -ENOTCONN) {
 521		msleep(10);
 522		goto retry;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 523	}
 524
 
 
 
 
 
 
 525	return r;
 526}
 527
 528static struct dm_io *alloc_io(struct mapped_device *md)
 529{
 530	return mempool_alloc(md->io_pool, GFP_NOIO);
 531}
 
 532
 533static void free_io(struct mapped_device *md, struct dm_io *io)
 534{
 535	mempool_free(io, md->io_pool);
 536}
 537
 538static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
 539{
 540	bio_put(&tio->clone);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 541}
 542
 543static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
 544					    gfp_t gfp_mask)
 545{
 546	return mempool_alloc(md->io_pool, gfp_mask);
 547}
 548
 549static void free_rq_tio(struct dm_rq_target_io *tio)
 550{
 551	mempool_free(tio, tio->md->io_pool);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 552}
 553
 554static int md_in_flight(struct mapped_device *md)
 555{
 556	return atomic_read(&md->pending[READ]) +
 557	       atomic_read(&md->pending[WRITE]);
 558}
 559
 560static void start_io_acct(struct dm_io *io)
 561{
 562	struct mapped_device *md = io->md;
 563	struct bio *bio = io->bio;
 564	int cpu;
 565	int rw = bio_data_dir(bio);
 566
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 567	io->start_time = jiffies;
 
 
 
 
 
 
 
 568
 569	cpu = part_stat_lock();
 570	part_round_stats(cpu, &dm_disk(md)->part0);
 571	part_stat_unlock();
 572	atomic_set(&dm_disk(md)->part0.in_flight[rw],
 573		atomic_inc_return(&md->pending[rw]));
 574
 575	if (unlikely(dm_stats_used(&md->stats)))
 576		dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
 577				    bio_sectors(bio), false, 0, &io->stats_aux);
 578}
 579
 580static void end_io_acct(struct dm_io *io)
 581{
 582	struct mapped_device *md = io->md;
 583	struct bio *bio = io->bio;
 584	unsigned long duration = jiffies - io->start_time;
 585	int pending, cpu;
 586	int rw = bio_data_dir(bio);
 587
 588	cpu = part_stat_lock();
 589	part_round_stats(cpu, &dm_disk(md)->part0);
 590	part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
 591	part_stat_unlock();
 592
 593	if (unlikely(dm_stats_used(&md->stats)))
 594		dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
 595				    bio_sectors(bio), true, duration, &io->stats_aux);
 596
 597	/*
 598	 * After this is decremented the bio must not be touched if it is
 599	 * a flush.
 600	 */
 601	pending = atomic_dec_return(&md->pending[rw]);
 602	atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
 603	pending += atomic_read(&md->pending[rw^0x1]);
 604
 605	/* nudge anyone waiting on suspend queue */
 606	if (!pending)
 607		wake_up(&md->wait);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 608}
 609
 610/*
 611 * Add the bio to the list of deferred io.
 612 */
 613static void queue_io(struct mapped_device *md, struct bio *bio)
 614{
 615	unsigned long flags;
 616
 617	spin_lock_irqsave(&md->deferred_lock, flags);
 618	bio_list_add(&md->deferred, bio);
 619	spin_unlock_irqrestore(&md->deferred_lock, flags);
 620	queue_work(md->wq, &md->work);
 621}
 622
 623/*
 624 * Everyone (including functions in this file), should use this
 625 * function to access the md->map field, and make sure they call
 626 * dm_put_live_table() when finished.
 627 */
 628struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
 
 629{
 630	*srcu_idx = srcu_read_lock(&md->io_barrier);
 631
 632	return srcu_dereference(md->map, &md->io_barrier);
 633}
 634
 635void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
 
 636{
 637	srcu_read_unlock(&md->io_barrier, srcu_idx);
 638}
 639
 640void dm_sync_table(struct mapped_device *md)
 641{
 642	synchronize_srcu(&md->io_barrier);
 643	synchronize_rcu_expedited();
 644}
 645
 646/*
 647 * A fast alternative to dm_get_live_table/dm_put_live_table.
 648 * The caller must not block between these two functions.
 649 */
 650static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
 651{
 652	rcu_read_lock();
 653	return rcu_dereference(md->map);
 654}
 655
 656static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
 657{
 658	rcu_read_unlock();
 659}
 660
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 661/*
 662 * Get the geometry associated with a dm device
 663 */
 664int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
 665{
 666	*geo = md->geometry;
 667
 668	return 0;
 669}
 670
 671/*
 672 * Set the geometry of a device.
 673 */
 674int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
 675{
 676	sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
 677
 678	if (geo->start > sz) {
 679		DMWARN("Start sector is beyond the geometry limits.");
 680		return -EINVAL;
 681	}
 682
 683	md->geometry = *geo;
 684
 685	return 0;
 686}
 687
 688/*-----------------------------------------------------------------
 689 * CRUD START:
 690 *   A more elegant soln is in the works that uses the queue
 691 *   merge fn, unfortunately there are a couple of changes to
 692 *   the block layer that I want to make for this.  So in the
 693 *   interests of getting something for people to use I give
 694 *   you this clearly demarcated crap.
 695 *---------------------------------------------------------------*/
 696
 697static int __noflush_suspending(struct mapped_device *md)
 698{
 699	return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
 700}
 701
 702/*
 703 * Decrements the number of outstanding ios that a bio has been
 704 * cloned into, completing the original io if necc.
 705 */
 706static void dec_pending(struct dm_io *io, int error)
 707{
 708	unsigned long flags;
 709	int io_error;
 710	struct bio *bio;
 711	struct mapped_device *md = io->md;
 712
 713	/* Push-back supersedes any I/O errors */
 714	if (unlikely(error)) {
 715		spin_lock_irqsave(&io->endio_lock, flags);
 716		if (!(io->error > 0 && __noflush_suspending(md)))
 717			io->error = error;
 718		spin_unlock_irqrestore(&io->endio_lock, flags);
 
 719	}
 
 720
 721	if (atomic_dec_and_test(&io->io_count)) {
 722		if (io->error == DM_ENDIO_REQUEUE) {
 723			/*
 724			 * Target requested pushing back the I/O.
 725			 */
 726			spin_lock_irqsave(&md->deferred_lock, flags);
 727			if (__noflush_suspending(md))
 728				bio_list_add_head(&md->deferred, io->bio);
 729			else
 730				/* noflush suspend was interrupted. */
 731				io->error = -EIO;
 732			spin_unlock_irqrestore(&md->deferred_lock, flags);
 733		}
 734
 735		io_error = io->error;
 736		bio = io->bio;
 737		end_io_acct(io);
 738		free_io(md, io);
 
 
 
 
 
 
 
 
 739
 740		if (io_error == DM_ENDIO_REQUEUE)
 741			return;
 742
 743		if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) {
 744			/*
 745			 * Preflush done for flush with data, reissue
 746			 * without REQ_FLUSH.
 
 747			 */
 748			bio->bi_rw &= ~REQ_FLUSH;
 749			queue_io(md, bio);
 750		} else {
 751			/* done with normal IO or empty flush */
 752			trace_block_bio_complete(md->queue, bio, io_error);
 753			bio_endio(bio, io_error);
 754		}
 755	}
 756}
 757
 758static void clone_endio(struct bio *bio, int error)
 759{
 760	int r = 0;
 761	struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
 762	struct dm_io *io = tio->io;
 763	struct mapped_device *md = tio->io->md;
 764	dm_endio_fn endio = tio->ti->type->end_io;
 765
 766	if (!bio_flagged(bio, BIO_UPTODATE) && !error)
 767		error = -EIO;
 768
 769	if (endio) {
 770		r = endio(tio->ti, bio, error);
 771		if (r < 0 || r == DM_ENDIO_REQUEUE)
 772			/*
 773			 * error and requeue request are handled
 774			 * in dec_pending().
 775			 */
 776			error = r;
 777		else if (r == DM_ENDIO_INCOMPLETE)
 778			/* The target will handle the io */
 779			return;
 780		else if (r) {
 781			DMWARN("unimplemented target endio return value: %d", r);
 782			BUG();
 783		}
 
 784	}
 785
 786	free_tio(md, tio);
 787	dec_pending(io, error);
 
 
 788}
 789
 790/*
 791 * Partial completion handling for request-based dm
 792 */
 793static void end_clone_bio(struct bio *clone, int error)
 794{
 795	struct dm_rq_clone_bio_info *info =
 796		container_of(clone, struct dm_rq_clone_bio_info, clone);
 797	struct dm_rq_target_io *tio = info->tio;
 798	struct bio *bio = info->orig;
 799	unsigned int nr_bytes = info->orig->bi_iter.bi_size;
 800
 801	bio_put(clone);
 802
 803	if (tio->error)
 804		/*
 805		 * An error has already been detected on the request.
 806		 * Once error occurred, just let clone->end_io() handle
 807		 * the remainder.
 808		 */
 809		return;
 810	else if (error) {
 
 
 
 
 811		/*
 812		 * Don't notice the error to the upper layer yet.
 813		 * The error handling decision is made by the target driver,
 814		 * when the request is completed.
 815		 */
 816		tio->error = error;
 817		return;
 818	}
 819
 820	/*
 821	 * I/O for the bio successfully completed.
 822	 * Notice the data completion to the upper layer.
 823	 */
 824
 825	/*
 826	 * bios are processed from the head of the list.
 827	 * So the completing bio should always be rq->bio.
 828	 * If it's not, something wrong is happening.
 829	 */
 830	if (tio->orig->bio != bio)
 831		DMERR("bio completion is going in the middle of the request");
 832
 833	/*
 834	 * Update the original request.
 835	 * Do not use blk_end_request() here, because it may complete
 836	 * the original request before the clone, and break the ordering.
 837	 */
 838	blk_update_request(tio->orig, 0, nr_bytes);
 839}
 840
 841/*
 842 * Don't touch any member of the md after calling this function because
 843 * the md may be freed in dm_put() at the end of this function.
 844 * Or do dm_get() before calling this function and dm_put() later.
 845 */
 846static void rq_completed(struct mapped_device *md, int rw, int run_queue)
 847{
 848	atomic_dec(&md->pending[rw]);
 849
 850	/* nudge anyone waiting on suspend queue */
 851	if (!md_in_flight(md))
 852		wake_up(&md->wait);
 853
 854	/*
 855	 * Run this off this callpath, as drivers could invoke end_io while
 856	 * inside their request_fn (and holding the queue lock). Calling
 857	 * back into ->request_fn() could deadlock attempting to grab the
 858	 * queue lock again.
 859	 */
 860	if (run_queue)
 861		blk_run_queue_async(md->queue);
 862
 863	/*
 864	 * dm_put() must be at the end of this function. See the comment above
 865	 */
 866	dm_put(md);
 
 
 
 
 
 
 
 
 
 867}
 868
 869static void free_rq_clone(struct request *clone)
 870{
 871	struct dm_rq_target_io *tio = clone->end_io_data;
 
 
 
 872
 873	blk_rq_unprep_clone(clone);
 874	free_rq_tio(tio);
 875}
 
 
 876
 877/*
 878 * Complete the clone and the original request.
 879 * Must be called without queue lock.
 880 */
 881static void dm_end_request(struct request *clone, int error)
 882{
 883	int rw = rq_data_dir(clone);
 884	struct dm_rq_target_io *tio = clone->end_io_data;
 885	struct mapped_device *md = tio->md;
 886	struct request *rq = tio->orig;
 887
 888	if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
 889		rq->errors = clone->errors;
 890		rq->resid_len = clone->resid_len;
 891
 892		if (rq->sense)
 893			/*
 894			 * We are using the sense buffer of the original
 895			 * request.
 896			 * So setting the length of the sense data is enough.
 897			 */
 898			rq->sense_len = clone->sense_len;
 899	}
 900
 901	free_rq_clone(clone);
 902	blk_end_request_all(rq, error);
 903	rq_completed(md, rw, true);
 904}
 905
 906static void dm_unprep_request(struct request *rq)
 
 
 
 
 
 
 
 
 907{
 908	struct request *clone = rq->special;
 909
 910	rq->special = NULL;
 911	rq->cmd_flags &= ~REQ_DONTPREP;
 
 
 
 
 
 
 
 
 
 
 912
 913	free_rq_clone(clone);
 914}
 915
 916/*
 917 * Requeue the original request of a clone.
 
 918 */
 919void dm_requeue_unmapped_request(struct request *clone)
 920{
 921	int rw = rq_data_dir(clone);
 922	struct dm_rq_target_io *tio = clone->end_io_data;
 923	struct mapped_device *md = tio->md;
 924	struct request *rq = tio->orig;
 925	struct request_queue *q = rq->q;
 926	unsigned long flags;
 927
 928	dm_unprep_request(rq);
 929
 930	spin_lock_irqsave(q->queue_lock, flags);
 931	blk_requeue_request(q, rq);
 932	spin_unlock_irqrestore(q->queue_lock, flags);
 933
 934	rq_completed(md, rw, 0);
 935}
 936EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
 937
 938static void __stop_queue(struct request_queue *q)
 939{
 940	blk_stop_queue(q);
 
 
 
 
 
 
 
 
 941}
 942
 943static void stop_queue(struct request_queue *q)
 944{
 945	unsigned long flags;
 
 946
 947	spin_lock_irqsave(q->queue_lock, flags);
 948	__stop_queue(q);
 949	spin_unlock_irqrestore(q->queue_lock, flags);
 950}
 951
 952static void __start_queue(struct request_queue *q)
 
 
 
 
 953{
 954	if (blk_queue_stopped(q))
 955		blk_start_queue(q);
 956}
 957
 958static void start_queue(struct request_queue *q)
 959{
 960	unsigned long flags;
 961
 962	spin_lock_irqsave(q->queue_lock, flags);
 963	__start_queue(q);
 964	spin_unlock_irqrestore(q->queue_lock, flags);
 965}
 966
 967static void dm_done(struct request *clone, int error, bool mapped)
 968{
 969	int r = error;
 970	struct dm_rq_target_io *tio = clone->end_io_data;
 971	dm_request_endio_fn rq_end_io = NULL;
 972
 973	if (tio->ti) {
 974		rq_end_io = tio->ti->type->rq_end_io;
 975
 976		if (mapped && rq_end_io)
 977			r = rq_end_io(tio->ti, clone, error, &tio->info);
 978	}
 979
 980	if (r <= 0)
 981		/* The target wants to complete the I/O */
 982		dm_end_request(clone, r);
 983	else if (r == DM_ENDIO_INCOMPLETE)
 984		/* The target will handle the I/O */
 985		return;
 986	else if (r == DM_ENDIO_REQUEUE)
 987		/* The target wants to requeue the I/O */
 988		dm_requeue_unmapped_request(clone);
 989	else {
 990		DMWARN("unimplemented target endio return value: %d", r);
 991		BUG();
 992	}
 993}
 994
 995/*
 996 * Request completion handler for request-based dm
 997 */
 998static void dm_softirq_done(struct request *rq)
 999{
1000	bool mapped = true;
1001	struct request *clone = rq->completion_data;
1002	struct dm_rq_target_io *tio = clone->end_io_data;
1003
1004	if (rq->cmd_flags & REQ_FAILED)
1005		mapped = false;
1006
1007	dm_done(clone, tio->error, mapped);
1008}
1009
1010/*
1011 * Complete the clone and the original request with the error status
1012 * through softirq context.
1013 */
1014static void dm_complete_request(struct request *clone, int error)
1015{
1016	struct dm_rq_target_io *tio = clone->end_io_data;
1017	struct request *rq = tio->orig;
 
 
 
 
1018
1019	tio->error = error;
1020	rq->completion_data = clone;
1021	blk_complete_request(rq);
1022}
 
 
 
 
1023
1024/*
1025 * Complete the not-mapped clone and the original request with the error status
1026 * through softirq context.
1027 * Target's rq_end_io() function isn't called.
1028 * This may be used when the target's map_rq() function fails.
1029 */
1030void dm_kill_unmapped_request(struct request *clone, int error)
1031{
1032	struct dm_rq_target_io *tio = clone->end_io_data;
1033	struct request *rq = tio->orig;
1034
1035	rq->cmd_flags |= REQ_FAILED;
1036	dm_complete_request(clone, error);
1037}
1038EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
1039
1040/*
1041 * Called with the queue lock held
1042 */
1043static void end_clone_request(struct request *clone, int error)
1044{
1045	/*
1046	 * For just cleaning up the information of the queue in which
1047	 * the clone was dispatched.
1048	 * The clone is *NOT* freed actually here because it is alloced from
1049	 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
1050	 */
1051	__blk_put_request(clone->q, clone);
 
 
 
 
 
 
 
 
 
 
 
 
 
1052
1053	/*
1054	 * Actual request completion is done in a softirq context which doesn't
1055	 * hold the queue lock.  Otherwise, deadlock could occur because:
1056	 *     - another request may be submitted by the upper level driver
1057	 *       of the stacking during the completion
1058	 *     - the submission which requires queue lock may be done
1059	 *       against this queue
1060	 */
1061	dm_complete_request(clone, error);
1062}
1063
1064/*
1065 * Return maximum size of I/O possible at the supplied sector up to the current
1066 * target boundary.
1067 */
1068static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
 
1069{
1070	sector_t target_offset = dm_target_offset(ti, sector);
1071
1072	return ti->len - target_offset;
1073}
1074
1075static sector_t max_io_len(sector_t sector, struct dm_target *ti)
 
 
1076{
1077	sector_t len = max_io_len_target_boundary(sector, ti);
1078	sector_t offset, max_len;
1079
1080	/*
1081	 * Does the target need to split even further?
 
 
 
1082	 */
1083	if (ti->max_io_len) {
1084		offset = dm_target_offset(ti, sector);
1085		if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1086			max_len = sector_div(offset, ti->max_io_len);
1087		else
1088			max_len = offset & (ti->max_io_len - 1);
1089		max_len = ti->max_io_len - max_len;
1090
1091		if (len > max_len)
1092			len = max_len;
1093	}
1094
1095	return len;
 
 
1096}
1097
1098int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1099{
1100	if (len > UINT_MAX) {
1101		DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1102		      (unsigned long long)len, UINT_MAX);
1103		ti->error = "Maximum size of target IO is too large";
1104		return -EINVAL;
1105	}
1106
1107	ti->max_io_len = (uint32_t) len;
1108
1109	return 0;
1110}
1111EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1112
1113static void __map_bio(struct dm_target_io *tio)
 
 
1114{
1115	int r;
1116	sector_t sector;
1117	struct mapped_device *md;
1118	struct bio *clone = &tio->clone;
1119	struct dm_target *ti = tio->ti;
1120
1121	clone->bi_end_io = clone_endio;
1122
1123	/*
1124	 * Map the clone.  If r == 0 we don't need to do
1125	 * anything, the target has assumed ownership of
1126	 * this io.
1127	 */
1128	atomic_inc(&tio->io->io_count);
1129	sector = clone->bi_iter.bi_sector;
1130	r = ti->type->map(ti, clone);
1131	if (r == DM_MAPIO_REMAPPED) {
1132		/* the bio has been remapped so dispatch it */
1133
1134		trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1135				      tio->io->bio->bi_bdev->bd_dev, sector);
1136
1137		generic_make_request(clone);
1138	} else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1139		/* error the io and bail out, or requeue it if needed */
1140		md = tio->io->md;
1141		dec_pending(tio->io, r);
1142		free_tio(md, tio);
1143	} else if (r) {
1144		DMWARN("unimplemented target map return value: %d", r);
1145		BUG();
1146	}
1147}
1148
1149struct clone_info {
1150	struct mapped_device *md;
1151	struct dm_table *map;
1152	struct bio *bio;
1153	struct dm_io *io;
1154	sector_t sector;
1155	sector_t sector_count;
1156};
1157
1158static void bio_setup_sector(struct bio *bio, sector_t sector, sector_t len)
1159{
1160	bio->bi_iter.bi_sector = sector;
1161	bio->bi_iter.bi_size = to_bytes(len);
1162}
1163
1164/*
1165 * Creates a bio that consists of range of complete bvecs.
1166 */
1167static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1168		      sector_t sector, unsigned len)
1169{
1170	struct bio *clone = &tio->clone;
 
 
 
 
1171
1172	__bio_clone_fast(clone, bio);
1173
1174	if (bio_integrity(bio))
1175		bio_integrity_clone(clone, bio, GFP_NOIO);
 
 
 
 
 
 
 
1176
1177	bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1178	clone->bi_iter.bi_size = to_bytes(len);
1179
1180	if (bio_integrity(bio))
1181		bio_integrity_trim(clone, 0, len);
1182}
1183
1184static struct dm_target_io *alloc_tio(struct clone_info *ci,
1185				      struct dm_target *ti, int nr_iovecs,
1186				      unsigned target_bio_nr)
1187{
1188	struct dm_target_io *tio;
1189	struct bio *clone;
 
 
 
1190
1191	clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs);
1192	tio = container_of(clone, struct dm_target_io, clone);
1193
1194	tio->io = ci->io;
1195	tio->ti = ti;
1196	tio->target_bio_nr = target_bio_nr;
 
 
 
 
 
 
 
 
 
1197
1198	return tio;
1199}
1200
1201static void __clone_and_map_simple_bio(struct clone_info *ci,
1202				       struct dm_target *ti,
1203				       unsigned target_bio_nr, sector_t len)
1204{
1205	struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs, target_bio_nr);
1206	struct bio *clone = &tio->clone;
 
 
 
1207
1208	/*
1209	 * Discard requests require the bio's inline iovecs be initialized.
1210	 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1211	 * and discard, so no need for concern about wasted bvec allocations.
1212	 */
1213	 __bio_clone_fast(clone, ci->bio);
1214	if (len)
1215		bio_setup_sector(clone, ci->sector, len);
1216
1217	__map_bio(tio);
 
 
 
1218}
1219
1220static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1221				  unsigned num_bios, sector_t len)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1222{
1223	unsigned target_bio_nr;
 
 
1224
1225	for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1226		__clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1227}
 
 
1228
1229static int __send_empty_flush(struct clone_info *ci)
1230{
1231	unsigned target_nr = 0;
1232	struct dm_target *ti;
1233
1234	BUG_ON(bio_has_data(ci->bio));
1235	while ((ti = dm_table_get_target(ci->map, target_nr++)))
1236		__send_duplicate_bios(ci, ti, ti->num_flush_bios, 0);
1237
1238	return 0;
 
 
 
 
 
 
1239}
 
1240
1241static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1242				     sector_t sector, unsigned len)
 
 
 
 
 
 
 
 
1243{
1244	struct bio *bio = ci->bio;
1245	struct dm_target_io *tio;
1246	unsigned target_bio_nr;
1247	unsigned num_target_bios = 1;
 
 
1248
1249	/*
1250	 * Does the target want to receive duplicate copies of the bio?
 
1251	 */
1252	if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1253		num_target_bios = ti->num_write_bios(ti, bio);
1254
1255	for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1256		tio = alloc_tio(ci, ti, 0, target_bio_nr);
1257		clone_bio(tio, bio, sector, len);
1258		__map_bio(tio);
1259	}
1260}
 
1261
1262typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1263
1264static unsigned get_num_discard_bios(struct dm_target *ti)
1265{
1266	return ti->num_discard_bios;
 
 
 
 
 
 
 
 
 
 
 
1267}
1268
1269static unsigned get_num_write_same_bios(struct dm_target *ti)
1270{
1271	return ti->num_write_same_bios;
1272}
 
 
 
1273
1274typedef bool (*is_split_required_fn)(struct dm_target *ti);
1275
1276static bool is_split_required_for_discard(struct dm_target *ti)
1277{
1278	return ti->split_discard_bios;
1279}
1280
1281static int __send_changing_extent_only(struct clone_info *ci,
1282				       get_num_bios_fn get_num_bios,
1283				       is_split_required_fn is_split_required)
1284{
1285	struct dm_target *ti;
1286	sector_t len;
1287	unsigned num_bios;
1288
1289	do {
1290		ti = dm_table_find_target(ci->map, ci->sector);
1291		if (!dm_target_is_valid(ti))
1292			return -EIO;
1293
 
1294		/*
1295		 * Even though the device advertised support for this type of
1296		 * request, that does not mean every target supports it, and
1297		 * reconfiguration might also have changed that since the
1298		 * check was performed.
1299		 */
1300		num_bios = get_num_bios ? get_num_bios(ti) : 0;
1301		if (!num_bios)
1302			return -EOPNOTSUPP;
1303
1304		if (is_split_required && !is_split_required(ti))
1305			len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1306		else
1307			len = min(ci->sector_count, max_io_len(ci->sector, ti));
1308
1309		__send_duplicate_bios(ci, ti, num_bios, len);
1310
1311		ci->sector += len;
1312	} while (ci->sector_count -= len);
 
 
 
 
1313
1314	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1315}
1316
1317static int __send_discard(struct clone_info *ci)
1318{
1319	return __send_changing_extent_only(ci, get_num_discard_bios,
1320					   is_split_required_for_discard);
1321}
1322
1323static int __send_write_same(struct clone_info *ci)
1324{
1325	return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
 
 
 
 
 
 
1326}
1327
1328/*
1329 * Select the correct strategy for processing a non-flush bio.
1330 */
1331static int __split_and_process_non_flush(struct clone_info *ci)
1332{
1333	struct bio *bio = ci->bio;
1334	struct dm_target *ti;
1335	unsigned len;
1336
1337	if (unlikely(bio->bi_rw & REQ_DISCARD))
1338		return __send_discard(ci);
1339	else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1340		return __send_write_same(ci);
1341
1342	ti = dm_table_find_target(ci->map, ci->sector);
1343	if (!dm_target_is_valid(ti))
1344		return -EIO;
1345
1346	len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
 
1347
1348	__clone_and_map_data_bio(ci, ti, ci->sector, len);
 
 
 
 
 
 
1349
1350	ci->sector += len;
1351	ci->sector_count -= len;
 
 
 
 
1352
1353	return 0;
 
 
1354}
1355
1356/*
1357 * Entry point to split a bio into clones and submit them to the targets.
1358 */
1359static void __split_and_process_bio(struct mapped_device *md,
1360				    struct dm_table *map, struct bio *bio)
1361{
1362	struct clone_info ci;
1363	int error = 0;
 
1364
1365	if (unlikely(!map)) {
1366		bio_io_error(bio);
1367		return;
1368	}
1369
1370	ci.map = map;
1371	ci.md = md;
1372	ci.io = alloc_io(md);
1373	ci.io->error = 0;
1374	atomic_set(&ci.io->io_count, 1);
1375	ci.io->bio = bio;
1376	ci.io->md = md;
1377	spin_lock_init(&ci.io->endio_lock);
1378	ci.sector = bio->bi_iter.bi_sector;
1379
1380	start_io_acct(ci.io);
1381
1382	if (bio->bi_rw & REQ_FLUSH) {
1383		ci.bio = &ci.md->flush_bio;
1384		ci.sector_count = 0;
1385		error = __send_empty_flush(&ci);
1386		/* dec_pending submits any data associated with flush */
1387	} else {
1388		ci.bio = bio;
1389		ci.sector_count = bio_sectors(bio);
1390		while (ci.sector_count && !error)
1391			error = __split_and_process_non_flush(&ci);
1392	}
1393
1394	/* drop the extra reference count */
1395	dec_pending(ci.io, error);
1396}
1397/*-----------------------------------------------------------------
1398 * CRUD END
1399 *---------------------------------------------------------------*/
1400
1401static int dm_merge_bvec(struct request_queue *q,
1402			 struct bvec_merge_data *bvm,
1403			 struct bio_vec *biovec)
1404{
1405	struct mapped_device *md = q->queuedata;
1406	struct dm_table *map = dm_get_live_table_fast(md);
1407	struct dm_target *ti;
1408	sector_t max_sectors;
1409	int max_size = 0;
1410
1411	if (unlikely(!map))
1412		goto out;
1413
1414	ti = dm_table_find_target(map, bvm->bi_sector);
1415	if (!dm_target_is_valid(ti))
1416		goto out;
1417
1418	/*
1419	 * Find maximum amount of I/O that won't need splitting
1420	 */
1421	max_sectors = min(max_io_len(bvm->bi_sector, ti),
1422			  (sector_t) BIO_MAX_SECTORS);
1423	max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1424	if (max_size < 0)
1425		max_size = 0;
 
 
 
 
 
 
 
1426
1427	/*
1428	 * merge_bvec_fn() returns number of bytes
1429	 * it can accept at this offset
1430	 * max is precomputed maximal io size
1431	 */
1432	if (max_size && ti->type->merge)
1433		max_size = ti->type->merge(ti, bvm, biovec, max_size);
1434	/*
1435	 * If the target doesn't support merge method and some of the devices
1436	 * provided their merge_bvec method (we know this by looking at
1437	 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1438	 * entries.  So always set max_size to 0, and the code below allows
1439	 * just one page.
1440	 */
1441	else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1442
1443		max_size = 0;
 
 
 
 
1444
1445out:
1446	dm_put_live_table_fast(md);
1447	/*
1448	 * Always allow an entire first page
 
1449	 */
1450	if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1451		max_size = biovec->bv_len;
1452
1453	return max_size;
1454}
1455
1456/*
1457 * The request function that just remaps the bio built up by
1458 * dm_merge_bvec.
1459 */
1460static void _dm_request(struct request_queue *q, struct bio *bio)
1461{
1462	int rw = bio_data_dir(bio);
1463	struct mapped_device *md = q->queuedata;
1464	int cpu;
1465	int srcu_idx;
1466	struct dm_table *map;
1467
1468	map = dm_get_live_table(md, &srcu_idx);
1469
1470	cpu = part_stat_lock();
1471	part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1472	part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1473	part_stat_unlock();
1474
1475	/* if we're suspended, we have to queue this io for later */
1476	if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1477		dm_put_live_table(md, srcu_idx);
1478
1479		if (bio_rw(bio) != READA)
1480			queue_io(md, bio);
1481		else
1482			bio_io_error(bio);
1483		return;
1484	}
 
1485
1486	__split_and_process_bio(md, map, bio);
1487	dm_put_live_table(md, srcu_idx);
1488	return;
1489}
1490
1491int dm_request_based(struct mapped_device *md)
1492{
1493	return blk_queue_stackable(md->queue);
1494}
1495
1496static void dm_request(struct request_queue *q, struct bio *bio)
1497{
1498	struct mapped_device *md = q->queuedata;
 
 
 
 
 
 
 
1499
1500	if (dm_request_based(md))
1501		blk_queue_bio(q, bio);
1502	else
1503		_dm_request(q, bio);
1504}
1505
1506void dm_dispatch_request(struct request *rq)
 
1507{
1508	int r;
1509
1510	if (blk_queue_io_stat(rq->q))
1511		rq->cmd_flags |= REQ_IO_STAT;
1512
1513	rq->start_time = jiffies;
1514	r = blk_insert_cloned_request(rq->q, rq);
1515	if (r)
1516		dm_complete_request(rq, r);
1517}
1518EXPORT_SYMBOL_GPL(dm_dispatch_request);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1519
1520static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1521				 void *data)
1522{
1523	struct dm_rq_target_io *tio = data;
1524	struct dm_rq_clone_bio_info *info =
1525		container_of(bio, struct dm_rq_clone_bio_info, clone);
 
 
1526
1527	info->orig = bio_orig;
1528	info->tio = tio;
1529	bio->bi_end_io = end_clone_bio;
1530
1531	return 0;
1532}
1533
1534static int setup_clone(struct request *clone, struct request *rq,
1535		       struct dm_rq_target_io *tio)
 
 
 
 
 
 
 
 
1536{
1537	int r;
1538
1539	r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1540			      dm_rq_bio_constructor, tio);
1541	if (r)
1542		return r;
1543
1544	clone->cmd = rq->cmd;
1545	clone->cmd_len = rq->cmd_len;
1546	clone->sense = rq->sense;
1547	clone->buffer = rq->buffer;
1548	clone->end_io = end_clone_request;
1549	clone->end_io_data = tio;
1550
1551	return 0;
1552}
1553
1554static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1555				gfp_t gfp_mask)
1556{
1557	struct request *clone;
1558	struct dm_rq_target_io *tio;
1559
1560	tio = alloc_rq_tio(md, gfp_mask);
1561	if (!tio)
1562		return NULL;
 
 
 
 
1563
1564	tio->md = md;
1565	tio->ti = NULL;
1566	tio->orig = rq;
1567	tio->error = 0;
1568	memset(&tio->info, 0, sizeof(tio->info));
1569
1570	clone = &tio->clone;
1571	if (setup_clone(clone, rq, tio)) {
1572		/* -ENOMEM */
1573		free_rq_tio(tio);
1574		return NULL;
1575	}
1576
1577	return clone;
1578}
1579
1580/*
1581 * Called with the queue lock held.
1582 */
1583static int dm_prep_fn(struct request_queue *q, struct request *rq)
1584{
1585	struct mapped_device *md = q->queuedata;
1586	struct request *clone;
1587
1588	if (unlikely(rq->special)) {
1589		DMWARN("Already has something in rq->special.");
1590		return BLKPREP_KILL;
1591	}
1592
1593	clone = clone_rq(rq, md, GFP_ATOMIC);
1594	if (!clone)
1595		return BLKPREP_DEFER;
1596
1597	rq->special = clone;
1598	rq->cmd_flags |= REQ_DONTPREP;
1599
1600	return BLKPREP_OK;
1601}
 
 
 
1602
1603/*
1604 * Returns:
1605 * 0  : the request has been processed (not requeued)
1606 * !0 : the request has been requeued
1607 */
1608static int map_request(struct dm_target *ti, struct request *clone,
1609		       struct mapped_device *md)
1610{
1611	int r, requeued = 0;
1612	struct dm_rq_target_io *tio = clone->end_io_data;
1613
1614	tio->ti = ti;
1615	r = ti->type->map_rq(ti, clone, &tio->info);
1616	switch (r) {
1617	case DM_MAPIO_SUBMITTED:
1618		/* The target has taken the I/O to submit by itself later */
1619		break;
1620	case DM_MAPIO_REMAPPED:
1621		/* The target has remapped the I/O so dispatch it */
1622		trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1623				     blk_rq_pos(tio->orig));
1624		dm_dispatch_request(clone);
1625		break;
1626	case DM_MAPIO_REQUEUE:
1627		/* The target wants to requeue the I/O */
1628		dm_requeue_unmapped_request(clone);
1629		requeued = 1;
1630		break;
1631	default:
1632		if (r > 0) {
1633			DMWARN("unimplemented target map return value: %d", r);
1634			BUG();
1635		}
1636
1637		/* The target wants to complete the I/O */
1638		dm_kill_unmapped_request(clone, r);
1639		break;
 
 
1640	}
 
1641
1642	return requeued;
 
 
 
1643}
1644
1645static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
 
1646{
1647	struct request *clone;
1648
1649	blk_start_request(orig);
1650	clone = orig->special;
1651	atomic_inc(&md->pending[rq_data_dir(clone)]);
1652
1653	/*
1654	 * Hold the md reference here for the in-flight I/O.
1655	 * We can't rely on the reference count by device opener,
1656	 * because the device may be closed during the request completion
1657	 * when all bios are completed.
1658	 * See the comment in rq_completed() too.
1659	 */
1660	dm_get(md);
1661
1662	return clone;
1663}
1664
1665/*
1666 * q->request_fn for request-based dm.
1667 * Called with the queue lock held.
1668 */
1669static void dm_request_fn(struct request_queue *q)
 
1670{
1671	struct mapped_device *md = q->queuedata;
1672	int srcu_idx;
1673	struct dm_table *map = dm_get_live_table(md, &srcu_idx);
1674	struct dm_target *ti;
1675	struct request *rq, *clone;
1676	sector_t pos;
1677
1678	/*
1679	 * For suspend, check blk_queue_stopped() and increment
1680	 * ->pending within a single queue_lock not to increment the
1681	 * number of in-flight I/Os after the queue is stopped in
1682	 * dm_suspend().
1683	 */
1684	while (!blk_queue_stopped(q)) {
1685		rq = blk_peek_request(q);
1686		if (!rq)
1687			goto delay_and_out;
1688
1689		/* always use block 0 to find the target for flushes for now */
1690		pos = 0;
1691		if (!(rq->cmd_flags & REQ_FLUSH))
1692			pos = blk_rq_pos(rq);
1693
1694		ti = dm_table_find_target(map, pos);
1695		if (!dm_target_is_valid(ti)) {
1696			/*
1697			 * Must perform setup, that dm_done() requires,
1698			 * before calling dm_kill_unmapped_request
1699			 */
1700			DMERR_LIMIT("request attempted access beyond the end of device");
1701			clone = dm_start_request(md, rq);
1702			dm_kill_unmapped_request(clone, -EIO);
1703			continue;
1704		}
 
 
 
 
1705
1706		if (ti->type->busy && ti->type->busy(ti))
1707			goto delay_and_out;
1708
1709		clone = dm_start_request(md, rq);
 
1710
1711		spin_unlock(q->queue_lock);
1712		if (map_request(ti, clone, md))
1713			goto requeued;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1714
1715		BUG_ON(!irqs_disabled());
1716		spin_lock(q->queue_lock);
1717	}
 
 
1718
1719	goto out;
1720
1721requeued:
1722	BUG_ON(!irqs_disabled());
1723	spin_lock(q->queue_lock);
 
 
 
 
 
 
 
 
1724
1725delay_and_out:
1726	blk_delay_queue(q, HZ / 10);
1727out:
1728	dm_put_live_table(md, srcu_idx);
1729}
1730
1731int dm_underlying_device_busy(struct request_queue *q)
 
1732{
1733	return blk_lld_busy(q);
1734}
1735EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1736
1737static int dm_lld_busy(struct request_queue *q)
1738{
1739	int r;
1740	struct mapped_device *md = q->queuedata;
1741	struct dm_table *map = dm_get_live_table_fast(md);
1742
1743	if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1744		r = 1;
1745	else
1746		r = dm_table_any_busy_target(map);
1747
1748	dm_put_live_table_fast(md);
1749
1750	return r;
 
1751}
1752
1753static int dm_any_congested(void *congested_data, int bdi_bits)
 
1754{
1755	int r = bdi_bits;
1756	struct mapped_device *md = congested_data;
1757	struct dm_table *map;
 
 
 
 
 
1758
1759	if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1760		map = dm_get_live_table_fast(md);
1761		if (map) {
 
 
 
 
 
 
 
 
 
 
 
 
 
1762			/*
1763			 * Request-based dm cares about only own queue for
1764			 * the query about congestion status of request_queue
1765			 */
1766			if (dm_request_based(md))
1767				r = md->queue->backing_dev_info.state &
1768				    bdi_bits;
1769			else
1770				r = dm_table_any_congested(map, bdi_bits);
1771		}
1772		dm_put_live_table_fast(md);
1773	}
1774
1775	return r;
 
 
 
 
 
 
 
1776}
1777
1778/*-----------------------------------------------------------------
 
1779 * An IDR is used to keep track of allocated minor numbers.
1780 *---------------------------------------------------------------*/
 
1781static void free_minor(int minor)
1782{
1783	spin_lock(&_minor_lock);
1784	idr_remove(&_minor_idr, minor);
1785	spin_unlock(&_minor_lock);
1786}
1787
1788/*
1789 * See if the device with a specific minor # is free.
1790 */
1791static int specific_minor(int minor)
1792{
1793	int r;
1794
1795	if (minor >= (1 << MINORBITS))
1796		return -EINVAL;
1797
1798	idr_preload(GFP_KERNEL);
1799	spin_lock(&_minor_lock);
1800
1801	r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1802
1803	spin_unlock(&_minor_lock);
1804	idr_preload_end();
1805	if (r < 0)
1806		return r == -ENOSPC ? -EBUSY : r;
1807	return 0;
1808}
1809
1810static int next_free_minor(int *minor)
1811{
1812	int r;
1813
1814	idr_preload(GFP_KERNEL);
1815	spin_lock(&_minor_lock);
1816
1817	r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1818
1819	spin_unlock(&_minor_lock);
1820	idr_preload_end();
1821	if (r < 0)
1822		return r;
1823	*minor = r;
1824	return 0;
1825}
1826
1827static const struct block_device_operations dm_blk_dops;
 
 
1828
1829static void dm_wq_work(struct work_struct *work);
1830
1831static void dm_init_md_queue(struct mapped_device *md)
 
1832{
1833	/*
1834	 * Request-based dm devices cannot be stacked on top of bio-based dm
1835	 * devices.  The type of this dm device has not been decided yet.
1836	 * The type is decided at the first table loading time.
1837	 * To prevent problematic device stacking, clear the queue flag
1838	 * for request stacking support until then.
1839	 *
1840	 * This queue is new, so no concurrency on the queue_flags.
1841	 */
1842	queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1843
1844	md->queue->queuedata = md;
1845	md->queue->backing_dev_info.congested_fn = dm_any_congested;
1846	md->queue->backing_dev_info.congested_data = md;
1847	blk_queue_make_request(md->queue, dm_request);
1848	blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1849	blk_queue_merge_bvec(md->queue, dm_merge_bvec);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1850}
1851
1852/*
1853 * Allocate and initialise a blank device with a given minor.
1854 */
1855static struct mapped_device *alloc_dev(int minor)
1856{
1857	int r;
1858	struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
 
1859	void *old_md;
1860
 
1861	if (!md) {
1862		DMWARN("unable to allocate device, out of memory.");
1863		return NULL;
1864	}
1865
1866	if (!try_module_get(THIS_MODULE))
1867		goto bad_module_get;
1868
1869	/* get a minor number for the dev */
1870	if (minor == DM_ANY_MINOR)
1871		r = next_free_minor(&minor);
1872	else
1873		r = specific_minor(minor);
1874	if (r < 0)
1875		goto bad_minor;
1876
1877	r = init_srcu_struct(&md->io_barrier);
1878	if (r < 0)
1879		goto bad_io_barrier;
1880
 
 
1881	md->type = DM_TYPE_NONE;
1882	mutex_init(&md->suspend_lock);
1883	mutex_init(&md->type_lock);
 
1884	spin_lock_init(&md->deferred_lock);
1885	atomic_set(&md->holders, 1);
1886	atomic_set(&md->open_count, 0);
1887	atomic_set(&md->event_nr, 0);
1888	atomic_set(&md->uevent_seq, 0);
1889	INIT_LIST_HEAD(&md->uevent_list);
 
1890	spin_lock_init(&md->uevent_lock);
1891
1892	md->queue = blk_alloc_queue(GFP_KERNEL);
1893	if (!md->queue)
1894		goto bad_queue;
1895
1896	dm_init_md_queue(md);
1897
1898	md->disk = alloc_disk(1);
1899	if (!md->disk)
1900		goto bad_disk;
1901
1902	atomic_set(&md->pending[0], 0);
1903	atomic_set(&md->pending[1], 0);
1904	init_waitqueue_head(&md->wait);
1905	INIT_WORK(&md->work, dm_wq_work);
 
1906	init_waitqueue_head(&md->eventq);
1907	init_completion(&md->kobj_holder.completion);
1908
 
 
 
 
 
1909	md->disk->major = _major;
1910	md->disk->first_minor = minor;
 
 
1911	md->disk->fops = &dm_blk_dops;
1912	md->disk->queue = md->queue;
1913	md->disk->private_data = md;
1914	sprintf(md->disk->disk_name, "dm-%d", minor);
1915	add_disk(md->disk);
 
 
 
 
 
 
 
 
 
 
 
 
1916	format_dev_t(md->name, MKDEV(_major, minor));
1917
1918	md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1919	if (!md->wq)
1920		goto bad_thread;
1921
1922	md->bdev = bdget_disk(md->disk, 0);
1923	if (!md->bdev)
1924		goto bad_bdev;
1925
1926	bio_init(&md->flush_bio);
1927	md->flush_bio.bi_bdev = md->bdev;
1928	md->flush_bio.bi_rw = WRITE_FLUSH;
1929
1930	dm_stats_init(&md->stats);
 
 
1931
1932	/* Populate the mapping, nobody knows we exist yet */
1933	spin_lock(&_minor_lock);
1934	old_md = idr_replace(&_minor_idr, md, minor);
1935	spin_unlock(&_minor_lock);
1936
1937	BUG_ON(old_md != MINOR_ALLOCED);
1938
1939	return md;
1940
1941bad_bdev:
1942	destroy_workqueue(md->wq);
1943bad_thread:
1944	del_gendisk(md->disk);
1945	put_disk(md->disk);
1946bad_disk:
1947	blk_cleanup_queue(md->queue);
1948bad_queue:
1949	cleanup_srcu_struct(&md->io_barrier);
1950bad_io_barrier:
1951	free_minor(minor);
1952bad_minor:
1953	module_put(THIS_MODULE);
1954bad_module_get:
1955	kfree(md);
1956	return NULL;
1957}
1958
1959static void unlock_fs(struct mapped_device *md);
1960
1961static void free_dev(struct mapped_device *md)
1962{
1963	int minor = MINOR(disk_devt(md->disk));
1964
1965	unlock_fs(md);
1966	bdput(md->bdev);
1967	destroy_workqueue(md->wq);
1968	if (md->io_pool)
1969		mempool_destroy(md->io_pool);
1970	if (md->bs)
1971		bioset_free(md->bs);
1972	blk_integrity_unregister(md->disk);
1973	del_gendisk(md->disk);
1974	cleanup_srcu_struct(&md->io_barrier);
1975	free_minor(minor);
1976
1977	spin_lock(&_minor_lock);
1978	md->disk->private_data = NULL;
1979	spin_unlock(&_minor_lock);
1980
1981	put_disk(md->disk);
1982	blk_cleanup_queue(md->queue);
1983	dm_stats_cleanup(&md->stats);
1984	module_put(THIS_MODULE);
1985	kfree(md);
1986}
1987
1988static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1989{
1990	struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1991
1992	if (md->io_pool && md->bs) {
1993		/* The md already has necessary mempools. */
1994		if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
1995			/*
1996			 * Reload bioset because front_pad may have changed
1997			 * because a different table was loaded.
1998			 */
1999			bioset_free(md->bs);
2000			md->bs = p->bs;
2001			p->bs = NULL;
2002		} else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) {
2003			/*
2004			 * There's no need to reload with request-based dm
2005			 * because the size of front_pad doesn't change.
2006			 * Note for future: If you are to reload bioset,
2007			 * prep-ed requests in the queue may refer
2008			 * to bio from the old bioset, so you must walk
2009			 * through the queue to unprep.
2010			 */
2011		}
2012		goto out;
2013	}
2014
2015	BUG_ON(!p || md->io_pool || md->bs);
2016
2017	md->io_pool = p->io_pool;
2018	p->io_pool = NULL;
2019	md->bs = p->bs;
2020	p->bs = NULL;
2021
2022out:
2023	/* mempool bind completed, now no need any mempools in the table */
2024	dm_table_free_md_mempools(t);
2025}
2026
2027/*
2028 * Bind a table to the device.
2029 */
2030static void event_callback(void *context)
2031{
2032	unsigned long flags;
2033	LIST_HEAD(uevents);
2034	struct mapped_device *md = (struct mapped_device *) context;
2035
2036	spin_lock_irqsave(&md->uevent_lock, flags);
2037	list_splice_init(&md->uevent_list, &uevents);
2038	spin_unlock_irqrestore(&md->uevent_lock, flags);
2039
2040	dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2041
2042	atomic_inc(&md->event_nr);
2043	wake_up(&md->eventq);
2044}
2045
2046/*
2047 * Protected by md->suspend_lock obtained by dm_swap_table().
2048 */
2049static void __set_size(struct mapped_device *md, sector_t size)
2050{
2051	set_capacity(md->disk, size);
2052
2053	i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2054}
2055
2056/*
2057 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2058 *
2059 * If this function returns 0, then the device is either a non-dm
2060 * device without a merge_bvec_fn, or it is a dm device that is
2061 * able to split any bios it receives that are too big.
2062 */
2063int dm_queue_merge_is_compulsory(struct request_queue *q)
2064{
2065	struct mapped_device *dev_md;
2066
2067	if (!q->merge_bvec_fn)
2068		return 0;
2069
2070	if (q->make_request_fn == dm_request) {
2071		dev_md = q->queuedata;
2072		if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2073			return 0;
2074	}
2075
2076	return 1;
2077}
2078
2079static int dm_device_merge_is_compulsory(struct dm_target *ti,
2080					 struct dm_dev *dev, sector_t start,
2081					 sector_t len, void *data)
2082{
2083	struct block_device *bdev = dev->bdev;
2084	struct request_queue *q = bdev_get_queue(bdev);
2085
2086	return dm_queue_merge_is_compulsory(q);
2087}
2088
2089/*
2090 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2091 * on the properties of the underlying devices.
2092 */
2093static int dm_table_merge_is_optional(struct dm_table *table)
2094{
2095	unsigned i = 0;
2096	struct dm_target *ti;
2097
2098	while (i < dm_table_get_num_targets(table)) {
2099		ti = dm_table_get_target(table, i++);
2100
2101		if (ti->type->iterate_devices &&
2102		    ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2103			return 0;
2104	}
2105
2106	return 1;
2107}
2108
2109/*
2110 * Returns old map, which caller must destroy.
2111 */
2112static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2113			       struct queue_limits *limits)
2114{
2115	struct dm_table *old_map;
2116	struct request_queue *q = md->queue;
2117	sector_t size;
2118	int merge_is_optional;
 
 
2119
2120	size = dm_table_get_size(t);
2121
2122	/*
2123	 * Wipe any geometry if the size of the table changed.
2124	 */
2125	if (size != dm_get_size(md))
2126		memset(&md->geometry, 0, sizeof(md->geometry));
2127
2128	__set_size(md, size);
2129
2130	dm_table_event_callback(t, event_callback, md);
2131
2132	/*
2133	 * The queue hasn't been stopped yet, if the old table type wasn't
2134	 * for request-based during suspension.  So stop it to prevent
2135	 * I/O mapping before resume.
2136	 * This must be done before setting the queue restrictions,
2137	 * because request-based dm may be run just after the setting.
2138	 */
2139	if (dm_table_request_based(t) && !blk_queue_stopped(q))
2140		stop_queue(q);
2141
2142	__bind_mempools(md, t);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2143
2144	merge_is_optional = dm_table_merge_is_optional(t);
 
 
 
 
2145
2146	old_map = md->map;
2147	rcu_assign_pointer(md->map, t);
2148	md->immutable_target_type = dm_table_get_immutable_target_type(t);
2149
2150	dm_table_set_restrictions(t, q, limits);
2151	if (merge_is_optional)
2152		set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2153	else
2154		clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2155	dm_sync_table(md);
2156
2157	return old_map;
2158}
2159
2160/*
2161 * Returns unbound table for the caller to free.
2162 */
2163static struct dm_table *__unbind(struct mapped_device *md)
2164{
2165	struct dm_table *map = md->map;
2166
2167	if (!map)
2168		return NULL;
2169
2170	dm_table_event_callback(map, NULL, NULL);
2171	RCU_INIT_POINTER(md->map, NULL);
2172	dm_sync_table(md);
2173
2174	return map;
2175}
2176
2177/*
2178 * Constructor for a new device.
2179 */
2180int dm_create(int minor, struct mapped_device **result)
2181{
2182	struct mapped_device *md;
2183
2184	md = alloc_dev(minor);
2185	if (!md)
2186		return -ENXIO;
2187
2188	dm_sysfs_init(md);
2189
2190	*result = md;
2191	return 0;
2192}
2193
2194/*
2195 * Functions to manage md->type.
2196 * All are required to hold md->type_lock.
2197 */
2198void dm_lock_md_type(struct mapped_device *md)
2199{
2200	mutex_lock(&md->type_lock);
2201}
2202
2203void dm_unlock_md_type(struct mapped_device *md)
2204{
2205	mutex_unlock(&md->type_lock);
2206}
2207
2208void dm_set_md_type(struct mapped_device *md, unsigned type)
2209{
2210	BUG_ON(!mutex_is_locked(&md->type_lock));
2211	md->type = type;
2212}
2213
2214unsigned dm_get_md_type(struct mapped_device *md)
2215{
2216	BUG_ON(!mutex_is_locked(&md->type_lock));
2217	return md->type;
2218}
2219
2220struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2221{
2222	return md->immutable_target_type;
2223}
2224
2225/*
2226 * The queue_limits are only valid as long as you have a reference
2227 * count on 'md'.
2228 */
2229struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2230{
2231	BUG_ON(!atomic_read(&md->holders));
2232	return &md->queue->limits;
2233}
2234EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2235
2236/*
2237 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2238 */
2239static int dm_init_request_based_queue(struct mapped_device *md)
2240{
2241	struct request_queue *q = NULL;
2242
2243	if (md->queue->elevator)
2244		return 1;
2245
2246	/* Fully initialize the queue */
2247	q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2248	if (!q)
2249		return 0;
2250
2251	md->queue = q;
2252	dm_init_md_queue(md);
2253	blk_queue_softirq_done(md->queue, dm_softirq_done);
2254	blk_queue_prep_rq(md->queue, dm_prep_fn);
2255	blk_queue_lld_busy(md->queue, dm_lld_busy);
 
 
 
 
 
 
 
 
 
 
 
 
2256
2257	elv_register_queue(md->queue);
 
 
 
 
 
 
 
2258
2259	return 1;
2260}
 
 
 
 
 
 
 
2261
2262/*
2263 * Setup the DM device's queue based on md's type
2264 */
2265int dm_setup_md_queue(struct mapped_device *md)
2266{
2267	if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2268	    !dm_init_request_based_queue(md)) {
2269		DMWARN("Cannot initialize queue for request-based mapped device");
2270		return -EINVAL;
2271	}
2272
 
 
 
 
 
2273	return 0;
 
 
 
 
 
 
 
 
2274}
2275
2276static struct mapped_device *dm_find_md(dev_t dev)
2277{
2278	struct mapped_device *md;
2279	unsigned minor = MINOR(dev);
2280
2281	if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2282		return NULL;
2283
2284	spin_lock(&_minor_lock);
2285
2286	md = idr_find(&_minor_idr, minor);
2287	if (md && (md == MINOR_ALLOCED ||
2288		   (MINOR(disk_devt(dm_disk(md))) != minor) ||
2289		   dm_deleting_md(md) ||
2290		   test_bit(DMF_FREEING, &md->flags))) {
2291		md = NULL;
2292		goto out;
2293	}
2294
2295out:
2296	spin_unlock(&_minor_lock);
2297
2298	return md;
2299}
2300
2301struct mapped_device *dm_get_md(dev_t dev)
2302{
2303	struct mapped_device *md = dm_find_md(dev);
2304
2305	if (md)
2306		dm_get(md);
2307
2308	return md;
2309}
2310EXPORT_SYMBOL_GPL(dm_get_md);
2311
2312void *dm_get_mdptr(struct mapped_device *md)
2313{
2314	return md->interface_ptr;
2315}
2316
2317void dm_set_mdptr(struct mapped_device *md, void *ptr)
2318{
2319	md->interface_ptr = ptr;
2320}
2321
2322void dm_get(struct mapped_device *md)
2323{
2324	atomic_inc(&md->holders);
2325	BUG_ON(test_bit(DMF_FREEING, &md->flags));
2326}
2327
 
 
 
 
 
 
 
 
 
 
 
 
 
2328const char *dm_device_name(struct mapped_device *md)
2329{
2330	return md->name;
2331}
2332EXPORT_SYMBOL_GPL(dm_device_name);
2333
2334static void __dm_destroy(struct mapped_device *md, bool wait)
2335{
2336	struct dm_table *map;
2337	int srcu_idx;
2338
2339	might_sleep();
2340
2341	spin_lock(&_minor_lock);
2342	map = dm_get_live_table(md, &srcu_idx);
2343	idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2344	set_bit(DMF_FREEING, &md->flags);
2345	spin_unlock(&_minor_lock);
2346
 
 
 
 
 
 
 
 
2347	if (!dm_suspended_md(md)) {
2348		dm_table_presuspend_targets(map);
 
 
2349		dm_table_postsuspend_targets(map);
2350	}
2351
2352	/* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2353	dm_put_live_table(md, srcu_idx);
 
2354
2355	/*
2356	 * Rare, but there may be I/O requests still going to complete,
2357	 * for example.  Wait for all references to disappear.
2358	 * No one should increment the reference count of the mapped_device,
2359	 * after the mapped_device state becomes DMF_FREEING.
2360	 */
2361	if (wait)
2362		while (atomic_read(&md->holders))
2363			msleep(1);
2364	else if (atomic_read(&md->holders))
2365		DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2366		       dm_device_name(md), atomic_read(&md->holders));
2367
2368	dm_sysfs_exit(md);
2369	dm_table_destroy(__unbind(md));
2370	free_dev(md);
2371}
2372
2373void dm_destroy(struct mapped_device *md)
2374{
2375	__dm_destroy(md, true);
2376}
2377
2378void dm_destroy_immediate(struct mapped_device *md)
2379{
2380	__dm_destroy(md, false);
2381}
2382
2383void dm_put(struct mapped_device *md)
2384{
2385	atomic_dec(&md->holders);
2386}
2387EXPORT_SYMBOL_GPL(dm_put);
2388
2389static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2390{
2391	int r = 0;
2392	DECLARE_WAITQUEUE(wait, current);
 
 
 
 
 
 
2393
2394	add_wait_queue(&md->wait, &wait);
 
 
 
2395
2396	while (1) {
2397		set_current_state(interruptible);
2398
2399		if (!md_in_flight(md))
2400			break;
2401
2402		if (interruptible == TASK_INTERRUPTIBLE &&
2403		    signal_pending(current)) {
2404			r = -EINTR;
2405			break;
2406		}
2407
2408		io_schedule();
2409	}
2410	set_current_state(TASK_RUNNING);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2411
2412	remove_wait_queue(&md->wait, &wait);
 
2413
2414	return r;
2415}
2416
2417/*
2418 * Process the deferred bios
2419 */
2420static void dm_wq_work(struct work_struct *work)
2421{
2422	struct mapped_device *md = container_of(work, struct mapped_device,
2423						work);
2424	struct bio *c;
2425	int srcu_idx;
2426	struct dm_table *map;
2427
2428	map = dm_get_live_table(md, &srcu_idx);
2429
2430	while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2431		spin_lock_irq(&md->deferred_lock);
2432		c = bio_list_pop(&md->deferred);
2433		spin_unlock_irq(&md->deferred_lock);
2434
2435		if (!c)
2436			break;
2437
2438		if (dm_request_based(md))
2439			generic_make_request(c);
2440		else
2441			__split_and_process_bio(md, map, c);
2442	}
2443
2444	dm_put_live_table(md, srcu_idx);
2445}
2446
2447static void dm_queue_flush(struct mapped_device *md)
2448{
2449	clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2450	smp_mb__after_clear_bit();
2451	queue_work(md->wq, &md->work);
2452}
2453
2454/*
2455 * Swap in a new table, returning the old one for the caller to destroy.
2456 */
2457struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2458{
2459	struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2460	struct queue_limits limits;
2461	int r;
2462
2463	mutex_lock(&md->suspend_lock);
2464
2465	/* device must be suspended */
2466	if (!dm_suspended_md(md))
2467		goto out;
2468
2469	/*
2470	 * If the new table has no data devices, retain the existing limits.
2471	 * This helps multipath with queue_if_no_path if all paths disappear,
2472	 * then new I/O is queued based on these limits, and then some paths
2473	 * reappear.
2474	 */
2475	if (dm_table_has_no_data_devices(table)) {
2476		live_map = dm_get_live_table_fast(md);
2477		if (live_map)
2478			limits = md->queue->limits;
2479		dm_put_live_table_fast(md);
2480	}
2481
2482	if (!live_map) {
2483		r = dm_calculate_queue_limits(table, &limits);
2484		if (r) {
2485			map = ERR_PTR(r);
2486			goto out;
2487		}
2488	}
2489
2490	map = __bind(md, table, &limits);
 
2491
2492out:
2493	mutex_unlock(&md->suspend_lock);
2494	return map;
2495}
2496
2497/*
2498 * Functions to lock and unlock any filesystem running on the
2499 * device.
2500 */
2501static int lock_fs(struct mapped_device *md)
2502{
2503	int r;
2504
2505	WARN_ON(md->frozen_sb);
2506
2507	md->frozen_sb = freeze_bdev(md->bdev);
2508	if (IS_ERR(md->frozen_sb)) {
2509		r = PTR_ERR(md->frozen_sb);
2510		md->frozen_sb = NULL;
2511		return r;
2512	}
2513
2514	set_bit(DMF_FROZEN, &md->flags);
2515
2516	return 0;
2517}
2518
2519static void unlock_fs(struct mapped_device *md)
2520{
2521	if (!test_bit(DMF_FROZEN, &md->flags))
2522		return;
2523
2524	thaw_bdev(md->bdev, md->frozen_sb);
2525	md->frozen_sb = NULL;
2526	clear_bit(DMF_FROZEN, &md->flags);
2527}
2528
2529/*
2530 * We need to be able to change a mapping table under a mounted
2531 * filesystem.  For example we might want to move some data in
2532 * the background.  Before the table can be swapped with
2533 * dm_bind_table, dm_suspend must be called to flush any in
2534 * flight bios and ensure that any further io gets deferred.
2535 */
2536/*
2537 * Suspend mechanism in request-based dm.
2538 *
2539 * 1. Flush all I/Os by lock_fs() if needed.
2540 * 2. Stop dispatching any I/O by stopping the request_queue.
2541 * 3. Wait for all in-flight I/Os to be completed or requeued.
2542 *
2543 * To abort suspend, start the request_queue.
2544 */
2545int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2546{
2547	struct dm_table *map = NULL;
2548	int r = 0;
2549	int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2550	int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2551
2552	mutex_lock(&md->suspend_lock);
2553
2554	if (dm_suspended_md(md)) {
2555		r = -EINVAL;
2556		goto out_unlock;
2557	}
2558
2559	map = md->map;
2560
2561	/*
2562	 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2563	 * This flag is cleared before dm_suspend returns.
2564	 */
2565	if (noflush)
2566		set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
 
 
2567
2568	/* This does not get reverted if there's an error later. */
 
 
 
2569	dm_table_presuspend_targets(map);
2570
2571	/*
2572	 * Flush I/O to the device.
2573	 * Any I/O submitted after lock_fs() may not be flushed.
2574	 * noflush takes precedence over do_lockfs.
2575	 * (lock_fs() flushes I/Os and waits for them to complete.)
2576	 */
2577	if (!noflush && do_lockfs) {
2578		r = lock_fs(md);
2579		if (r)
2580			goto out_unlock;
 
 
2581	}
2582
2583	/*
2584	 * Here we must make sure that no processes are submitting requests
2585	 * to target drivers i.e. no one may be executing
2586	 * __split_and_process_bio. This is called from dm_request and
2587	 * dm_wq_work.
2588	 *
2589	 * To get all processes out of __split_and_process_bio in dm_request,
2590	 * we take the write lock. To prevent any process from reentering
2591	 * __split_and_process_bio from dm_request and quiesce the thread
2592	 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2593	 * flush_workqueue(md->wq).
2594	 */
2595	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2596	synchronize_srcu(&md->io_barrier);
 
2597
2598	/*
2599	 * Stop md->queue before flushing md->wq in case request-based
2600	 * dm defers requests to md->wq from md->queue.
2601	 */
2602	if (dm_request_based(md))
2603		stop_queue(md->queue);
2604
2605	flush_workqueue(md->wq);
2606
2607	/*
2608	 * At this point no more requests are entering target request routines.
2609	 * We call dm_wait_for_completion to wait for all existing requests
2610	 * to finish.
2611	 */
2612	r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
 
 
2613
2614	if (noflush)
2615		clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2616	synchronize_srcu(&md->io_barrier);
 
2617
2618	/* were we interrupted ? */
2619	if (r < 0) {
2620		dm_queue_flush(md);
2621
2622		if (dm_request_based(md))
2623			start_queue(md->queue);
2624
2625		unlock_fs(md);
2626		goto out_unlock; /* pushback list is already flushed, so skip flush */
 
2627	}
2628
2629	/*
2630	 * If dm_wait_for_completion returned 0, the device is completely
2631	 * quiescent now. There is no request-processing activity. All new
2632	 * requests are being added to md->deferred list.
2633	 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2634
2635	set_bit(DMF_SUSPENDED, &md->flags);
 
2636
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2637	dm_table_postsuspend_targets(map);
 
2638
2639out_unlock:
2640	mutex_unlock(&md->suspend_lock);
2641	return r;
2642}
2643
2644int dm_resume(struct mapped_device *md)
2645{
2646	int r = -EINVAL;
2647	struct dm_table *map = NULL;
2648
2649	mutex_lock(&md->suspend_lock);
2650	if (!dm_suspended_md(md))
2651		goto out;
2652
2653	map = md->map;
2654	if (!map || !dm_table_get_size(map))
2655		goto out;
2656
2657	r = dm_table_resume_targets(map);
2658	if (r)
2659		goto out;
2660
2661	dm_queue_flush(md);
2662
2663	/*
2664	 * Flushing deferred I/Os must be done after targets are resumed
2665	 * so that mapping of targets can work correctly.
2666	 * Request-based dm is queueing the deferred I/Os in its request_queue.
2667	 */
2668	if (dm_request_based(md))
2669		start_queue(md->queue);
2670
2671	unlock_fs(md);
2672
2673	clear_bit(DMF_SUSPENDED, &md->flags);
 
 
 
 
 
 
 
 
 
 
2674
2675	r = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2676out:
2677	mutex_unlock(&md->suspend_lock);
2678
2679	return r;
2680}
2681
2682/*
2683 * Internal suspend/resume works like userspace-driven suspend. It waits
2684 * until all bios finish and prevents issuing new bios to the target drivers.
2685 * It may be used only from the kernel.
2686 *
2687 * Internal suspend holds md->suspend_lock, which prevents interaction with
2688 * userspace-driven suspend.
2689 */
2690
2691void dm_internal_suspend(struct mapped_device *md)
2692{
2693	mutex_lock(&md->suspend_lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2694	if (dm_suspended_md(md))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2695		return;
2696
2697	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2698	synchronize_srcu(&md->io_barrier);
2699	flush_workqueue(md->wq);
2700	dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2701}
 
2702
2703void dm_internal_resume(struct mapped_device *md)
2704{
2705	if (dm_suspended_md(md))
2706		goto done;
2707
2708	dm_queue_flush(md);
2709
2710done:
2711	mutex_unlock(&md->suspend_lock);
2712}
 
2713
2714/*-----------------------------------------------------------------
 
2715 * Event notification.
2716 *---------------------------------------------------------------*/
 
2717int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2718		       unsigned cookie)
2719{
 
 
2720	char udev_cookie[DM_COOKIE_LENGTH];
2721	char *envp[] = { udev_cookie, NULL };
2722
2723	if (!cookie)
2724		return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2725	else {
2726		snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2727			 DM_COOKIE_ENV_VAR_NAME, cookie);
2728		return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2729					  action, envp);
 
 
2730	}
 
 
 
 
 
 
 
 
2731}
2732
2733uint32_t dm_next_uevent_seq(struct mapped_device *md)
2734{
2735	return atomic_add_return(1, &md->uevent_seq);
2736}
2737
2738uint32_t dm_get_event_nr(struct mapped_device *md)
2739{
2740	return atomic_read(&md->event_nr);
2741}
2742
2743int dm_wait_event(struct mapped_device *md, int event_nr)
2744{
2745	return wait_event_interruptible(md->eventq,
2746			(event_nr != atomic_read(&md->event_nr)));
2747}
2748
2749void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2750{
2751	unsigned long flags;
2752
2753	spin_lock_irqsave(&md->uevent_lock, flags);
2754	list_add(elist, &md->uevent_list);
2755	spin_unlock_irqrestore(&md->uevent_lock, flags);
2756}
2757
2758/*
2759 * The gendisk is only valid as long as you have a reference
2760 * count on 'md'.
2761 */
2762struct gendisk *dm_disk(struct mapped_device *md)
2763{
2764	return md->disk;
2765}
 
2766
2767struct kobject *dm_kobject(struct mapped_device *md)
2768{
2769	return &md->kobj_holder.kobj;
2770}
2771
2772struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2773{
2774	struct mapped_device *md;
2775
2776	md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2777
2778	if (test_bit(DMF_FREEING, &md->flags) ||
2779	    dm_deleting_md(md))
2780		return NULL;
2781
 
2782	dm_get(md);
 
 
 
2783	return md;
2784}
2785
2786int dm_suspended_md(struct mapped_device *md)
2787{
2788	return test_bit(DMF_SUSPENDED, &md->flags);
2789}
2790
 
 
 
 
 
 
 
 
 
 
2791int dm_test_deferred_remove_flag(struct mapped_device *md)
2792{
2793	return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2794}
2795
2796int dm_suspended(struct dm_target *ti)
2797{
2798	return dm_suspended_md(dm_table_get_md(ti->table));
2799}
2800EXPORT_SYMBOL_GPL(dm_suspended);
2801
 
 
 
 
 
 
2802int dm_noflush_suspending(struct dm_target *ti)
2803{
2804	return __noflush_suspending(dm_table_get_md(ti->table));
2805}
2806EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2807
2808struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size)
2809{
2810	struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
2811	struct kmem_cache *cachep;
2812	unsigned int pool_size;
2813	unsigned int front_pad;
2814
2815	if (!pools)
2816		return NULL;
2817
2818	if (type == DM_TYPE_BIO_BASED) {
2819		cachep = _io_cache;
2820		pool_size = dm_get_reserved_bio_based_ios();
2821		front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2822	} else if (type == DM_TYPE_REQUEST_BASED) {
2823		cachep = _rq_tio_cache;
2824		pool_size = dm_get_reserved_rq_based_ios();
2825		front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2826		/* per_bio_data_size is not used. See __bind_mempools(). */
2827		WARN_ON(per_bio_data_size != 0);
2828	} else
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2829		goto out;
2830
2831	pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
2832	if (!pools->io_pool)
2833		goto out;
 
2834
2835	pools->bs = bioset_create(pool_size, front_pad);
2836	if (!pools->bs)
2837		goto out;
 
2838
2839	if (integrity && bioset_integrity_create(pools->bs, pool_size))
 
2840		goto out;
2841
2842	return pools;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2843
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2844out:
2845	dm_free_md_mempools(pools);
 
 
2846
2847	return NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2848}
2849
2850void dm_free_md_mempools(struct dm_md_mempools *pools)
2851{
2852	if (!pools)
2853		return;
 
 
2854
2855	if (pools->io_pool)
2856		mempool_destroy(pools->io_pool);
 
2857
2858	if (pools->bs)
2859		bioset_free(pools->bs);
2860
2861	kfree(pools);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2862}
2863
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2864static const struct block_device_operations dm_blk_dops = {
 
 
2865	.open = dm_blk_open,
2866	.release = dm_blk_close,
2867	.ioctl = dm_blk_ioctl,
2868	.getgeo = dm_blk_getgeo,
 
 
2869	.owner = THIS_MODULE
2870};
2871
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2872/*
2873 * module hooks
2874 */
2875module_init(dm_init);
2876module_exit(dm_exit);
2877
2878module_param(major, uint, 0);
2879MODULE_PARM_DESC(major, "The major number of the device mapper");
2880
2881module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2882MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2883
2884module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
2885MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
 
 
 
2886
2887MODULE_DESCRIPTION(DM_NAME " driver");
2888MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2889MODULE_LICENSE("GPL");
v6.9.4
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
   4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
   5 *
   6 * This file is released under the GPL.
   7 */
   8
   9#include "dm-core.h"
  10#include "dm-rq.h"
  11#include "dm-uevent.h"
  12#include "dm-ima.h"
  13
  14#include <linux/init.h>
  15#include <linux/module.h>
  16#include <linux/mutex.h>
  17#include <linux/sched/mm.h>
  18#include <linux/sched/signal.h>
  19#include <linux/blkpg.h>
  20#include <linux/bio.h>
  21#include <linux/mempool.h>
  22#include <linux/dax.h>
  23#include <linux/slab.h>
  24#include <linux/idr.h>
  25#include <linux/uio.h>
  26#include <linux/hdreg.h>
  27#include <linux/delay.h>
  28#include <linux/wait.h>
  29#include <linux/pr.h>
  30#include <linux/refcount.h>
  31#include <linux/part_stat.h>
  32#include <linux/blk-crypto.h>
  33#include <linux/blk-crypto-profile.h>
  34
  35#define DM_MSG_PREFIX "core"
  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
  44/*
  45 * For REQ_POLLED fs bio, this flag is set if we link mapped underlying
  46 * dm_io into one list, and reuse bio->bi_private as the list head. Before
  47 * ending this fs bio, we will recover its ->bi_private.
  48 */
  49#define REQ_DM_POLL_LIST	REQ_DRV
  50
  51static const char *_name = DM_NAME;
  52
  53static unsigned int major;
  54static unsigned int _major;
  55
  56static DEFINE_IDR(_minor_idr);
  57
  58static DEFINE_SPINLOCK(_minor_lock);
  59
  60static void do_deferred_remove(struct work_struct *w);
  61
  62static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
  63
  64static struct workqueue_struct *deferred_remove_workqueue;
 
 
 
 
 
 
 
 
 
 
 
 
  65
  66atomic_t dm_global_event_nr = ATOMIC_INIT(0);
  67DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
 
 
 
 
 
 
 
 
 
  68
  69void dm_issue_global_event(void)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  70{
  71	atomic_inc(&dm_global_event_nr);
  72	wake_up(&dm_global_eventq);
 
  73}
 
 
 
  74
  75DEFINE_STATIC_KEY_FALSE(stats_enabled);
  76DEFINE_STATIC_KEY_FALSE(swap_bios_enabled);
  77DEFINE_STATIC_KEY_FALSE(zoned_enabled);
 
 
 
 
 
 
 
 
  78
  79/*
  80 * One of these is allocated (on-stack) per original bio.
 
  81 */
  82struct clone_info {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  83	struct dm_table *map;
  84	struct bio *bio;
  85	struct dm_io *io;
  86	sector_t sector;
  87	unsigned int sector_count;
  88	bool is_abnormal_io:1;
  89	bool submit_as_polled:1;
  90};
  91
  92static inline struct dm_target_io *clone_to_tio(struct bio *clone)
  93{
  94	return container_of(clone, struct dm_target_io, clone);
  95}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  96
  97void *dm_per_bio_data(struct bio *bio, size_t data_size)
  98{
  99	if (!dm_tio_flagged(clone_to_tio(bio), DM_TIO_INSIDE_DM_IO))
 100		return (char *)bio - DM_TARGET_IO_BIO_OFFSET - data_size;
 101	return (char *)bio - DM_IO_BIO_OFFSET - data_size;
 102}
 103EXPORT_SYMBOL_GPL(dm_per_bio_data);
 104
 105struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
 106{
 107	struct dm_io *io = (struct dm_io *)((char *)data + data_size);
 
 108
 109	if (io->magic == DM_IO_MAGIC)
 110		return (struct bio *)((char *)io + DM_IO_BIO_OFFSET);
 111	BUG_ON(io->magic != DM_TIO_MAGIC);
 112	return (struct bio *)((char *)io + DM_TARGET_IO_BIO_OFFSET);
 113}
 114EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
 115
 116unsigned int dm_bio_get_target_bio_nr(const struct bio *bio)
 117{
 118	return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
 119}
 120EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
 
 
 
 121
 122#define MINOR_ALLOCED ((void *)-1)
 
 
 
 
 123
 124#define DM_NUMA_NODE NUMA_NO_NODE
 125static int dm_numa_node = DM_NUMA_NODE;
 126
 127#define DEFAULT_SWAP_BIOS	(8 * 1048576 / PAGE_SIZE)
 128static int swap_bios = DEFAULT_SWAP_BIOS;
 129static int get_swap_bios(void)
 130{
 131	int latch = READ_ONCE(swap_bios);
 132
 133	if (unlikely(latch <= 0))
 134		latch = DEFAULT_SWAP_BIOS;
 135	return latch;
 136}
 137
 138struct table_device {
 139	struct list_head list;
 140	refcount_t count;
 141	struct dm_dev dm_dev;
 142};
 143
 144/*
 145 * Bio-based DM's mempools' reserved IOs set by the user.
 146 */
 
 
 
 
 
 147#define RESERVED_BIO_BASED_IOS		16
 148static unsigned int reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
 
 
 
 149
 150static int __dm_get_module_param_int(int *module_param, int min, int max)
 151{
 152	int param = READ_ONCE(*module_param);
 153	int modified_param = 0;
 154	bool modified = true;
 155
 156	if (param < min)
 157		modified_param = min;
 158	else if (param > max)
 159		modified_param = max;
 160	else
 161		modified = false;
 162
 163	if (modified) {
 164		(void)cmpxchg(module_param, param, modified_param);
 165		param = modified_param;
 166	}
 167
 168	return param;
 169}
 170
 171unsigned int __dm_get_module_param(unsigned int *module_param, unsigned int def, unsigned int max)
 
 172{
 173	unsigned int param = READ_ONCE(*module_param);
 174	unsigned int modified_param = 0;
 175
 176	if (!param)
 177		modified_param = def;
 178	else if (param > max)
 179		modified_param = max;
 180
 181	if (modified_param) {
 182		(void)cmpxchg(module_param, param, modified_param);
 183		param = modified_param;
 184	}
 185
 186	return param;
 187}
 188
 189unsigned int dm_get_reserved_bio_based_ios(void)
 190{
 191	return __dm_get_module_param(&reserved_bio_based_ios,
 192				     RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
 193}
 194EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
 195
 196static unsigned int dm_get_numa_node(void)
 197{
 198	return __dm_get_module_param_int(&dm_numa_node,
 199					 DM_NUMA_NODE, num_online_nodes() - 1);
 200}
 
 201
 202static int __init local_init(void)
 203{
 204	int r;
 
 
 
 
 
 
 
 
 
 205
 206	r = dm_uevent_init();
 207	if (r)
 208		return r;
 209
 210	deferred_remove_workqueue = alloc_ordered_workqueue("kdmremove", 0);
 211	if (!deferred_remove_workqueue) {
 212		r = -ENOMEM;
 213		goto out_uevent_exit;
 214	}
 215
 216	_major = major;
 217	r = register_blkdev(_major, _name);
 218	if (r < 0)
 219		goto out_free_workqueue;
 220
 221	if (!_major)
 222		_major = r;
 223
 224	return 0;
 225
 226out_free_workqueue:
 227	destroy_workqueue(deferred_remove_workqueue);
 228out_uevent_exit:
 229	dm_uevent_exit();
 
 
 
 
 230
 231	return r;
 232}
 233
 234static void local_exit(void)
 235{
 236	destroy_workqueue(deferred_remove_workqueue);
 237
 
 
 238	unregister_blkdev(_major, _name);
 239	dm_uevent_exit();
 240
 241	_major = 0;
 242
 243	DMINFO("cleaned up");
 244}
 245
 246static int (*_inits[])(void) __initdata = {
 247	local_init,
 248	dm_target_init,
 249	dm_linear_init,
 250	dm_stripe_init,
 251	dm_io_init,
 252	dm_kcopyd_init,
 253	dm_interface_init,
 254	dm_statistics_init,
 255};
 256
 257static void (*_exits[])(void) = {
 258	local_exit,
 259	dm_target_exit,
 260	dm_linear_exit,
 261	dm_stripe_exit,
 262	dm_io_exit,
 263	dm_kcopyd_exit,
 264	dm_interface_exit,
 265	dm_statistics_exit,
 266};
 267
 268static int __init dm_init(void)
 269{
 270	const int count = ARRAY_SIZE(_inits);
 
 271	int r, i;
 272
 273#if (IS_ENABLED(CONFIG_IMA) && !IS_ENABLED(CONFIG_IMA_DISABLE_HTABLE))
 274	DMWARN("CONFIG_IMA_DISABLE_HTABLE is disabled."
 275	       " Duplicate IMA measurements will not be recorded in the IMA log.");
 276#endif
 277
 278	for (i = 0; i < count; i++) {
 279		r = _inits[i]();
 280		if (r)
 281			goto bad;
 282	}
 283
 284	return 0;
 285bad:
 
 286	while (i--)
 287		_exits[i]();
 288
 289	return r;
 290}
 291
 292static void __exit dm_exit(void)
 293{
 294	int i = ARRAY_SIZE(_exits);
 295
 296	while (i--)
 297		_exits[i]();
 298
 299	/*
 300	 * Should be empty by this point.
 301	 */
 302	idr_destroy(&_minor_idr);
 303}
 304
 305/*
 306 * Block device functions
 307 */
 308int dm_deleting_md(struct mapped_device *md)
 309{
 310	return test_bit(DMF_DELETING, &md->flags);
 311}
 312
 313static int dm_blk_open(struct gendisk *disk, blk_mode_t mode)
 314{
 315	struct mapped_device *md;
 316
 317	spin_lock(&_minor_lock);
 318
 319	md = disk->private_data;
 320	if (!md)
 321		goto out;
 322
 323	if (test_bit(DMF_FREEING, &md->flags) ||
 324	    dm_deleting_md(md)) {
 325		md = NULL;
 326		goto out;
 327	}
 328
 329	dm_get(md);
 330	atomic_inc(&md->open_count);
 
 331out:
 332	spin_unlock(&_minor_lock);
 333
 334	return md ? 0 : -ENXIO;
 335}
 336
 337static void dm_blk_close(struct gendisk *disk)
 338{
 339	struct mapped_device *md;
 340
 341	spin_lock(&_minor_lock);
 342
 343	md = disk->private_data;
 344	if (WARN_ON(!md))
 345		goto out;
 346
 347	if (atomic_dec_and_test(&md->open_count) &&
 348	    (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
 349		queue_work(deferred_remove_workqueue, &deferred_remove_work);
 350
 351	dm_put(md);
 352out:
 353	spin_unlock(&_minor_lock);
 354}
 355
 356int dm_open_count(struct mapped_device *md)
 357{
 358	return atomic_read(&md->open_count);
 359}
 360
 361/*
 362 * Guarantees nothing is using the device before it's deleted.
 363 */
 364int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
 365{
 366	int r = 0;
 367
 368	spin_lock(&_minor_lock);
 369
 370	if (dm_open_count(md)) {
 371		r = -EBUSY;
 372		if (mark_deferred)
 373			set_bit(DMF_DEFERRED_REMOVE, &md->flags);
 374	} else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
 375		r = -EEXIST;
 376	else
 377		set_bit(DMF_DELETING, &md->flags);
 378
 379	spin_unlock(&_minor_lock);
 380
 381	return r;
 382}
 383
 384int dm_cancel_deferred_remove(struct mapped_device *md)
 385{
 386	int r = 0;
 387
 388	spin_lock(&_minor_lock);
 389
 390	if (test_bit(DMF_DELETING, &md->flags))
 391		r = -EBUSY;
 392	else
 393		clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
 394
 395	spin_unlock(&_minor_lock);
 396
 397	return r;
 398}
 399
 400static void do_deferred_remove(struct work_struct *w)
 401{
 402	dm_deferred_remove();
 403}
 404
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 405static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 406{
 407	struct mapped_device *md = bdev->bd_disk->private_data;
 408
 409	return dm_get_geometry(md, geo);
 410}
 411
 412static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
 413			    struct block_device **bdev)
 414{
 415	struct dm_target *ti;
 
 416	struct dm_table *map;
 417	int r;
 
 418
 419retry:
 420	r = -ENOTTY;
 421	map = dm_get_live_table(md, srcu_idx);
 422	if (!map || !dm_table_get_size(map))
 423		return r;
 424
 425	/* We only support devices that have a single target */
 426	if (map->num_targets != 1)
 427		return r;
 428
 429	ti = dm_table_get_target(map, 0);
 430	if (!ti->type->prepare_ioctl)
 431		return r;
 432
 433	if (dm_suspended_md(md))
 434		return -EAGAIN;
 435
 436	r = ti->type->prepare_ioctl(ti, bdev);
 437	if (r == -ENOTCONN && !fatal_signal_pending(current)) {
 438		dm_put_live_table(md, *srcu_idx);
 439		fsleep(10000);
 440		goto retry;
 441	}
 442
 443	return r;
 444}
 445
 446static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
 447{
 448	dm_put_live_table(md, srcu_idx);
 449}
 450
 451static int dm_blk_ioctl(struct block_device *bdev, blk_mode_t mode,
 452			unsigned int cmd, unsigned long arg)
 453{
 454	struct mapped_device *md = bdev->bd_disk->private_data;
 455	int r, srcu_idx;
 456
 457	r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
 458	if (r < 0)
 459		goto out;
 460
 461	if (r > 0) {
 462		/*
 463		 * Target determined this ioctl is being issued against a
 464		 * subset of the parent bdev; require extra privileges.
 465		 */
 466		if (!capable(CAP_SYS_RAWIO)) {
 467			DMDEBUG_LIMIT(
 468	"%s: sending ioctl %x to DM device without required privilege.",
 469				current->comm, cmd);
 470			r = -ENOIOCTLCMD;
 471			goto out;
 472		}
 473	}
 474
 475	if (!bdev->bd_disk->fops->ioctl)
 476		r = -ENOTTY;
 477	else
 478		r = bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
 479out:
 480	dm_unprepare_ioctl(md, srcu_idx);
 481	return r;
 482}
 483
 484u64 dm_start_time_ns_from_clone(struct bio *bio)
 485{
 486	return jiffies_to_nsecs(clone_to_tio(bio)->io->start_time);
 487}
 488EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
 489
 490static inline bool bio_is_flush_with_data(struct bio *bio)
 491{
 492	return ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size);
 493}
 494
 495static inline unsigned int dm_io_sectors(struct dm_io *io, struct bio *bio)
 496{
 497	/*
 498	 * If REQ_PREFLUSH set, don't account payload, it will be
 499	 * submitted (and accounted) after this flush completes.
 500	 */
 501	if (bio_is_flush_with_data(bio))
 502		return 0;
 503	if (unlikely(dm_io_flagged(io, DM_IO_WAS_SPLIT)))
 504		return io->sectors;
 505	return bio_sectors(bio);
 506}
 507
 508static void dm_io_acct(struct dm_io *io, bool end)
 509{
 510	struct bio *bio = io->orig_bio;
 511
 512	if (dm_io_flagged(io, DM_IO_BLK_STAT)) {
 513		if (!end)
 514			bdev_start_io_acct(bio->bi_bdev, bio_op(bio),
 515					   io->start_time);
 516		else
 517			bdev_end_io_acct(bio->bi_bdev, bio_op(bio),
 518					 dm_io_sectors(io, bio),
 519					 io->start_time);
 520	}
 521
 522	if (static_branch_unlikely(&stats_enabled) &&
 523	    unlikely(dm_stats_used(&io->md->stats))) {
 524		sector_t sector;
 525
 526		if (unlikely(dm_io_flagged(io, DM_IO_WAS_SPLIT)))
 527			sector = bio_end_sector(bio) - io->sector_offset;
 528		else
 529			sector = bio->bi_iter.bi_sector;
 530
 531		dm_stats_account_io(&io->md->stats, bio_data_dir(bio),
 532				    sector, dm_io_sectors(io, bio),
 533				    end, io->start_time, &io->stats_aux);
 534	}
 535}
 536
 537static void __dm_start_io_acct(struct dm_io *io)
 
 538{
 539	dm_io_acct(io, false);
 540}
 541
 542static void dm_start_io_acct(struct dm_io *io, struct bio *clone)
 543{
 544	/*
 545	 * Ensure IO accounting is only ever started once.
 546	 */
 547	if (dm_io_flagged(io, DM_IO_ACCOUNTED))
 548		return;
 549
 550	/* Expect no possibility for race unless DM_TIO_IS_DUPLICATE_BIO. */
 551	if (!clone || likely(dm_tio_is_normal(clone_to_tio(clone)))) {
 552		dm_io_set_flag(io, DM_IO_ACCOUNTED);
 553	} else {
 554		unsigned long flags;
 555		/* Can afford locking given DM_TIO_IS_DUPLICATE_BIO */
 556		spin_lock_irqsave(&io->lock, flags);
 557		if (dm_io_flagged(io, DM_IO_ACCOUNTED)) {
 558			spin_unlock_irqrestore(&io->lock, flags);
 559			return;
 560		}
 561		dm_io_set_flag(io, DM_IO_ACCOUNTED);
 562		spin_unlock_irqrestore(&io->lock, flags);
 563	}
 564
 565	__dm_start_io_acct(io);
 566}
 567
 568static void dm_end_io_acct(struct dm_io *io)
 569{
 570	dm_io_acct(io, true);
 
 571}
 572
 573static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio, gfp_t gfp_mask)
 574{
 575	struct dm_io *io;
 576	struct dm_target_io *tio;
 577	struct bio *clone;
 
 578
 579	clone = bio_alloc_clone(NULL, bio, gfp_mask, &md->mempools->io_bs);
 580	if (unlikely(!clone))
 581		return NULL;
 582	tio = clone_to_tio(clone);
 583	tio->flags = 0;
 584	dm_tio_set_flag(tio, DM_TIO_INSIDE_DM_IO);
 585	tio->io = NULL;
 586
 587	io = container_of(tio, struct dm_io, tio);
 588	io->magic = DM_IO_MAGIC;
 589	io->status = BLK_STS_OK;
 590
 591	/* one ref is for submission, the other is for completion */
 592	atomic_set(&io->io_count, 2);
 593	this_cpu_inc(*md->pending_io);
 594	io->orig_bio = bio;
 595	io->md = md;
 596	spin_lock_init(&io->lock);
 597	io->start_time = jiffies;
 598	io->flags = 0;
 599	if (blk_queue_io_stat(md->queue))
 600		dm_io_set_flag(io, DM_IO_BLK_STAT);
 601
 602	if (static_branch_unlikely(&stats_enabled) &&
 603	    unlikely(dm_stats_used(&md->stats)))
 604		dm_stats_record_start(&md->stats, &io->stats_aux);
 605
 606	return io;
 
 
 
 
 
 
 
 
 607}
 608
 609static void free_io(struct dm_io *io)
 610{
 611	bio_put(&io->tio.clone);
 612}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 613
 614static struct bio *alloc_tio(struct clone_info *ci, struct dm_target *ti,
 615			     unsigned int target_bio_nr, unsigned int *len, gfp_t gfp_mask)
 616{
 617	struct mapped_device *md = ci->io->md;
 618	struct dm_target_io *tio;
 619	struct bio *clone;
 620
 621	if (!ci->io->tio.io) {
 622		/* the dm_target_io embedded in ci->io is available */
 623		tio = &ci->io->tio;
 624		/* alloc_io() already initialized embedded clone */
 625		clone = &tio->clone;
 626	} else {
 627		clone = bio_alloc_clone(NULL, ci->bio, gfp_mask,
 628					&md->mempools->bs);
 629		if (!clone)
 630			return NULL;
 631
 632		/* REQ_DM_POLL_LIST shouldn't be inherited */
 633		clone->bi_opf &= ~REQ_DM_POLL_LIST;
 634
 635		tio = clone_to_tio(clone);
 636		tio->flags = 0; /* also clears DM_TIO_INSIDE_DM_IO */
 637	}
 638
 639	tio->magic = DM_TIO_MAGIC;
 640	tio->io = ci->io;
 641	tio->ti = ti;
 642	tio->target_bio_nr = target_bio_nr;
 643	tio->len_ptr = len;
 644	tio->old_sector = 0;
 645
 646	/* Set default bdev, but target must bio_set_dev() before issuing IO */
 647	clone->bi_bdev = md->disk->part0;
 648	if (unlikely(ti->needs_bio_set_dev))
 649		bio_set_dev(clone, md->disk->part0);
 650
 651	if (len) {
 652		clone->bi_iter.bi_size = to_bytes(*len);
 653		if (bio_integrity(clone))
 654			bio_integrity_trim(clone);
 655	}
 656
 657	return clone;
 658}
 659
 660static void free_tio(struct bio *clone)
 661{
 662	if (dm_tio_flagged(clone_to_tio(clone), DM_TIO_INSIDE_DM_IO))
 663		return;
 664	bio_put(clone);
 665}
 666
 667/*
 668 * Add the bio to the list of deferred io.
 669 */
 670static void queue_io(struct mapped_device *md, struct bio *bio)
 671{
 672	unsigned long flags;
 673
 674	spin_lock_irqsave(&md->deferred_lock, flags);
 675	bio_list_add(&md->deferred, bio);
 676	spin_unlock_irqrestore(&md->deferred_lock, flags);
 677	queue_work(md->wq, &md->work);
 678}
 679
 680/*
 681 * Everyone (including functions in this file), should use this
 682 * function to access the md->map field, and make sure they call
 683 * dm_put_live_table() when finished.
 684 */
 685struct dm_table *dm_get_live_table(struct mapped_device *md,
 686				   int *srcu_idx) __acquires(md->io_barrier)
 687{
 688	*srcu_idx = srcu_read_lock(&md->io_barrier);
 689
 690	return srcu_dereference(md->map, &md->io_barrier);
 691}
 692
 693void dm_put_live_table(struct mapped_device *md,
 694		       int srcu_idx) __releases(md->io_barrier)
 695{
 696	srcu_read_unlock(&md->io_barrier, srcu_idx);
 697}
 698
 699void dm_sync_table(struct mapped_device *md)
 700{
 701	synchronize_srcu(&md->io_barrier);
 702	synchronize_rcu_expedited();
 703}
 704
 705/*
 706 * A fast alternative to dm_get_live_table/dm_put_live_table.
 707 * The caller must not block between these two functions.
 708 */
 709static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
 710{
 711	rcu_read_lock();
 712	return rcu_dereference(md->map);
 713}
 714
 715static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
 716{
 717	rcu_read_unlock();
 718}
 719
 720static char *_dm_claim_ptr = "I belong to device-mapper";
 721
 722/*
 723 * Open a table device so we can use it as a map destination.
 724 */
 725static struct table_device *open_table_device(struct mapped_device *md,
 726		dev_t dev, blk_mode_t mode)
 727{
 728	struct table_device *td;
 729	struct file *bdev_file;
 730	struct block_device *bdev;
 731	u64 part_off;
 732	int r;
 733
 734	td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
 735	if (!td)
 736		return ERR_PTR(-ENOMEM);
 737	refcount_set(&td->count, 1);
 738
 739	bdev_file = bdev_file_open_by_dev(dev, mode, _dm_claim_ptr, NULL);
 740	if (IS_ERR(bdev_file)) {
 741		r = PTR_ERR(bdev_file);
 742		goto out_free_td;
 743	}
 744
 745	bdev = file_bdev(bdev_file);
 746
 747	/*
 748	 * We can be called before the dm disk is added.  In that case we can't
 749	 * register the holder relation here.  It will be done once add_disk was
 750	 * called.
 751	 */
 752	if (md->disk->slave_dir) {
 753		r = bd_link_disk_holder(bdev, md->disk);
 754		if (r)
 755			goto out_blkdev_put;
 756	}
 757
 758	td->dm_dev.mode = mode;
 759	td->dm_dev.bdev = bdev;
 760	td->dm_dev.bdev_file = bdev_file;
 761	td->dm_dev.dax_dev = fs_dax_get_by_bdev(bdev, &part_off,
 762						NULL, NULL);
 763	format_dev_t(td->dm_dev.name, dev);
 764	list_add(&td->list, &md->table_devices);
 765	return td;
 766
 767out_blkdev_put:
 768	__fput_sync(bdev_file);
 769out_free_td:
 770	kfree(td);
 771	return ERR_PTR(r);
 772}
 773
 774/*
 775 * Close a table device that we've been using.
 776 */
 777static void close_table_device(struct table_device *td, struct mapped_device *md)
 778{
 779	if (md->disk->slave_dir)
 780		bd_unlink_disk_holder(td->dm_dev.bdev, md->disk);
 781
 782	/* Leverage async fput() if DMF_DEFERRED_REMOVE set */
 783	if (unlikely(test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
 784		fput(td->dm_dev.bdev_file);
 785	else
 786		__fput_sync(td->dm_dev.bdev_file);
 787
 788	put_dax(td->dm_dev.dax_dev);
 789	list_del(&td->list);
 790	kfree(td);
 791}
 792
 793static struct table_device *find_table_device(struct list_head *l, dev_t dev,
 794					      blk_mode_t mode)
 795{
 796	struct table_device *td;
 797
 798	list_for_each_entry(td, l, list)
 799		if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
 800			return td;
 801
 802	return NULL;
 803}
 804
 805int dm_get_table_device(struct mapped_device *md, dev_t dev, blk_mode_t mode,
 806			struct dm_dev **result)
 807{
 808	struct table_device *td;
 809
 810	mutex_lock(&md->table_devices_lock);
 811	td = find_table_device(&md->table_devices, dev, mode);
 812	if (!td) {
 813		td = open_table_device(md, dev, mode);
 814		if (IS_ERR(td)) {
 815			mutex_unlock(&md->table_devices_lock);
 816			return PTR_ERR(td);
 817		}
 818	} else {
 819		refcount_inc(&td->count);
 820	}
 821	mutex_unlock(&md->table_devices_lock);
 822
 823	*result = &td->dm_dev;
 824	return 0;
 825}
 826
 827void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
 828{
 829	struct table_device *td = container_of(d, struct table_device, dm_dev);
 830
 831	mutex_lock(&md->table_devices_lock);
 832	if (refcount_dec_and_test(&td->count))
 833		close_table_device(td, md);
 834	mutex_unlock(&md->table_devices_lock);
 835}
 836
 837/*
 838 * Get the geometry associated with a dm device
 839 */
 840int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
 841{
 842	*geo = md->geometry;
 843
 844	return 0;
 845}
 846
 847/*
 848 * Set the geometry of a device.
 849 */
 850int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
 851{
 852	sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
 853
 854	if (geo->start > sz) {
 855		DMERR("Start sector is beyond the geometry limits.");
 856		return -EINVAL;
 857	}
 858
 859	md->geometry = *geo;
 860
 861	return 0;
 862}
 863
 
 
 
 
 
 
 
 
 
 864static int __noflush_suspending(struct mapped_device *md)
 865{
 866	return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
 867}
 868
 869static void dm_requeue_add_io(struct dm_io *io, bool first_stage)
 
 
 
 
 870{
 
 
 
 871	struct mapped_device *md = io->md;
 872
 873	if (first_stage) {
 874		struct dm_io *next = md->requeue_list;
 875
 876		md->requeue_list = io;
 877		io->next = next;
 878	} else {
 879		bio_list_add_head(&md->deferred, io->orig_bio);
 880	}
 881}
 882
 883static void dm_kick_requeue(struct mapped_device *md, bool first_stage)
 884{
 885	if (first_stage)
 886		queue_work(md->wq, &md->requeue_work);
 887	else
 888		queue_work(md->wq, &md->work);
 889}
 
 
 
 
 
 
 890
 891/*
 892 * Return true if the dm_io's original bio is requeued.
 893 * io->status is updated with error if requeue disallowed.
 894 */
 895static bool dm_handle_requeue(struct dm_io *io, bool first_stage)
 896{
 897	struct bio *bio = io->orig_bio;
 898	bool handle_requeue = (io->status == BLK_STS_DM_REQUEUE);
 899	bool handle_polled_eagain = ((io->status == BLK_STS_AGAIN) &&
 900				     (bio->bi_opf & REQ_POLLED));
 901	struct mapped_device *md = io->md;
 902	bool requeued = false;
 903
 904	if (handle_requeue || handle_polled_eagain) {
 905		unsigned long flags;
 906
 907		if (bio->bi_opf & REQ_POLLED) {
 908			/*
 909			 * Upper layer won't help us poll split bio
 910			 * (io->orig_bio may only reflect a subset of the
 911			 * pre-split original) so clear REQ_POLLED.
 912			 */
 913			bio_clear_polled(bio);
 
 
 
 
 
 914		}
 
 
 915
 916		/*
 917		 * Target requested pushing back the I/O or
 918		 * polled IO hit BLK_STS_AGAIN.
 919		 */
 920		spin_lock_irqsave(&md->deferred_lock, flags);
 921		if ((__noflush_suspending(md) &&
 922		     !WARN_ON_ONCE(dm_is_zone_write(md, bio))) ||
 923		    handle_polled_eagain || first_stage) {
 924			dm_requeue_add_io(io, first_stage);
 925			requeued = true;
 926		} else {
 
 
 
 927			/*
 928			 * noflush suspend was interrupted or this is
 929			 * a write to a zoned target.
 930			 */
 931			io->status = BLK_STS_IOERR;
 
 
 
 
 
 
 932		}
 933		spin_unlock_irqrestore(&md->deferred_lock, flags);
 934	}
 935
 936	if (requeued)
 937		dm_kick_requeue(md, first_stage);
 938
 939	return requeued;
 940}
 941
 942static void __dm_io_complete(struct dm_io *io, bool first_stage)
 
 
 
 943{
 944	struct bio *bio = io->orig_bio;
 945	struct mapped_device *md = io->md;
 946	blk_status_t io_error;
 947	bool requeued;
 
 
 
 948
 949	requeued = dm_handle_requeue(io, first_stage);
 950	if (requeued && first_stage)
 
 
 
 
 951		return;
 952
 953	io_error = io->status;
 954	if (dm_io_flagged(io, DM_IO_ACCOUNTED))
 955		dm_end_io_acct(io);
 956	else if (!io_error) {
 957		/*
 958		 * Must handle target that DM_MAPIO_SUBMITTED only to
 959		 * then bio_endio() rather than dm_submit_bio_remap()
 
 960		 */
 961		__dm_start_io_acct(io);
 962		dm_end_io_acct(io);
 963	}
 964	free_io(io);
 965	smp_wmb();
 966	this_cpu_dec(*md->pending_io);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 967
 968	/* nudge anyone waiting on suspend queue */
 969	if (unlikely(wq_has_sleeper(&md->wait)))
 970		wake_up(&md->wait);
 971
 972	/* Return early if the original bio was requeued */
 973	if (requeued)
 974		return;
 
 
 
 
 
 975
 976	if (bio_is_flush_with_data(bio)) {
 977		/*
 978		 * Preflush done for flush with data, reissue
 979		 * without REQ_PREFLUSH.
 980		 */
 981		bio->bi_opf &= ~REQ_PREFLUSH;
 982		queue_io(md, bio);
 983	} else {
 984		/* done with normal IO or empty flush */
 985		if (io_error)
 986			bio->bi_status = io_error;
 987		bio_endio(bio);
 988	}
 989}
 990
 991static void dm_wq_requeue_work(struct work_struct *work)
 992{
 993	struct mapped_device *md = container_of(work, struct mapped_device,
 994						requeue_work);
 995	unsigned long flags;
 996	struct dm_io *io;
 997
 998	/* reuse deferred lock to simplify dm_handle_requeue */
 999	spin_lock_irqsave(&md->deferred_lock, flags);
1000	io = md->requeue_list;
1001	md->requeue_list = NULL;
1002	spin_unlock_irqrestore(&md->deferred_lock, flags);
1003
1004	while (io) {
1005		struct dm_io *next = io->next;
 
 
 
 
 
 
 
 
1006
1007		dm_io_rewind(io, &md->disk->bio_split);
 
 
1008
1009		io->next = NULL;
1010		__dm_io_complete(io, false);
1011		io = next;
1012		cond_resched();
 
 
 
1013	}
 
 
 
 
1014}
1015
1016/*
1017 * Two staged requeue:
1018 *
1019 * 1) io->orig_bio points to the real original bio, and the part mapped to
1020 *    this io must be requeued, instead of other parts of the original bio.
1021 *
1022 * 2) io->orig_bio points to new cloned bio which matches the requeued dm_io.
1023 */
1024static void dm_io_complete(struct dm_io *io)
1025{
1026	bool first_requeue;
1027
1028	/*
1029	 * Only dm_io that has been split needs two stage requeue, otherwise
1030	 * we may run into long bio clone chain during suspend and OOM could
1031	 * be triggered.
1032	 *
1033	 * Also flush data dm_io won't be marked as DM_IO_WAS_SPLIT, so they
1034	 * also aren't handled via the first stage requeue.
1035	 */
1036	if (dm_io_flagged(io, DM_IO_WAS_SPLIT))
1037		first_requeue = true;
1038	else
1039		first_requeue = false;
1040
1041	__dm_io_complete(io, first_requeue);
1042}
1043
1044/*
1045 * Decrements the number of outstanding ios that a bio has been
1046 * cloned into, completing the original io if necc.
1047 */
1048static inline void __dm_io_dec_pending(struct dm_io *io)
1049{
1050	if (atomic_dec_and_test(&io->io_count))
1051		dm_io_complete(io);
 
 
 
 
 
 
 
 
 
 
 
 
1052}
 
1053
1054static void dm_io_set_error(struct dm_io *io, blk_status_t error)
1055{
1056	unsigned long flags;
1057
1058	/* Push-back supersedes any I/O errors */
1059	spin_lock_irqsave(&io->lock, flags);
1060	if (!(io->status == BLK_STS_DM_REQUEUE &&
1061	      __noflush_suspending(io->md))) {
1062		io->status = error;
1063	}
1064	spin_unlock_irqrestore(&io->lock, flags);
1065}
1066
1067static void dm_io_dec_pending(struct dm_io *io, blk_status_t error)
1068{
1069	if (unlikely(error))
1070		dm_io_set_error(io, error);
1071
1072	__dm_io_dec_pending(io);
 
 
1073}
1074
1075/*
1076 * The queue_limits are only valid as long as you have a reference
1077 * count on 'md'. But _not_ imposing verification to avoid atomic_read(),
1078 */
1079static inline struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
1080{
1081	return &md->queue->limits;
 
1082}
1083
1084void disable_discard(struct mapped_device *md)
1085{
1086	struct queue_limits *limits = dm_get_queue_limits(md);
1087
1088	/* device doesn't really support DISCARD, disable it */
1089	limits->max_discard_sectors = 0;
 
1090}
1091
1092void disable_write_zeroes(struct mapped_device *md)
1093{
1094	struct queue_limits *limits = dm_get_queue_limits(md);
 
 
 
 
 
1095
1096	/* device doesn't really support WRITE ZEROES, disable it */
1097	limits->max_write_zeroes_sectors = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1098}
1099
1100static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
 
 
 
1101{
1102	return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
 
 
 
 
 
 
 
1103}
1104
1105static void clone_endio(struct bio *bio)
 
 
 
 
1106{
1107	blk_status_t error = bio->bi_status;
1108	struct dm_target_io *tio = clone_to_tio(bio);
1109	struct dm_target *ti = tio->ti;
1110	dm_endio_fn endio = ti->type->end_io;
1111	struct dm_io *io = tio->io;
1112	struct mapped_device *md = io->md;
1113
1114	if (unlikely(error == BLK_STS_TARGET)) {
1115		if (bio_op(bio) == REQ_OP_DISCARD &&
1116		    !bdev_max_discard_sectors(bio->bi_bdev))
1117			disable_discard(md);
1118		else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
1119			 !bdev_write_zeroes_sectors(bio->bi_bdev))
1120			disable_write_zeroes(md);
1121	}
1122
1123	if (static_branch_unlikely(&zoned_enabled) &&
1124	    unlikely(bdev_is_zoned(bio->bi_bdev)))
1125		dm_zone_endio(io, bio);
 
 
 
 
 
 
 
1126
1127	if (endio) {
1128		int r = endio(ti, bio, &error);
 
 
1129
1130		switch (r) {
1131		case DM_ENDIO_REQUEUE:
1132			if (static_branch_unlikely(&zoned_enabled)) {
1133				/*
1134				 * Requeuing writes to a sequential zone of a zoned
1135				 * target will break the sequential write pattern:
1136				 * fail such IO.
1137				 */
1138				if (WARN_ON_ONCE(dm_is_zone_write(md, bio)))
1139					error = BLK_STS_IOERR;
1140				else
1141					error = BLK_STS_DM_REQUEUE;
1142			} else
1143				error = BLK_STS_DM_REQUEUE;
1144			fallthrough;
1145		case DM_ENDIO_DONE:
1146			break;
1147		case DM_ENDIO_INCOMPLETE:
1148			/* The target will handle the io */
1149			return;
1150		default:
1151			DMCRIT("unimplemented target endio return value: %d", r);
1152			BUG();
1153		}
1154	}
1155
1156	if (static_branch_unlikely(&swap_bios_enabled) &&
1157	    unlikely(swap_bios_limit(ti, bio)))
1158		up(&md->swap_bios_semaphore);
1159
1160	free_tio(bio);
1161	dm_io_dec_pending(io, error);
 
 
 
1162}
1163
1164/*
1165 * Return maximum size of I/O possible at the supplied sector up to the current
1166 * target boundary.
1167 */
1168static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
1169						  sector_t target_offset)
1170{
 
 
1171	return ti->len - target_offset;
1172}
1173
1174static sector_t __max_io_len(struct dm_target *ti, sector_t sector,
1175			     unsigned int max_granularity,
1176			     unsigned int max_sectors)
1177{
1178	sector_t target_offset = dm_target_offset(ti, sector);
1179	sector_t len = max_io_len_target_boundary(ti, target_offset);
1180
1181	/*
1182	 * Does the target need to split IO even further?
1183	 * - varied (per target) IO splitting is a tenet of DM; this
1184	 *   explains why stacked chunk_sectors based splitting via
1185	 *   bio_split_to_limits() isn't possible here.
1186	 */
1187	if (!max_granularity)
1188		return len;
1189	return min_t(sector_t, len,
1190		min(max_sectors ? : queue_max_sectors(ti->table->md->queue),
1191		    blk_chunk_sectors_left(target_offset, max_granularity)));
1192}
 
 
 
 
 
1193
1194static inline sector_t max_io_len(struct dm_target *ti, sector_t sector)
1195{
1196	return __max_io_len(ti, sector, ti->max_io_len, 0);
1197}
1198
1199int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1200{
1201	if (len > UINT_MAX) {
1202		DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1203		      (unsigned long long)len, UINT_MAX);
1204		ti->error = "Maximum size of target IO is too large";
1205		return -EINVAL;
1206	}
1207
1208	ti->max_io_len = (uint32_t) len;
1209
1210	return 0;
1211}
1212EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1213
1214static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1215						sector_t sector, int *srcu_idx)
1216	__acquires(md->io_barrier)
1217{
1218	struct dm_table *map;
1219	struct dm_target *ti;
 
 
 
 
 
1220
1221	map = dm_get_live_table(md, srcu_idx);
1222	if (!map)
1223		return NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1224
1225	ti = dm_table_find_target(map, sector);
1226	if (!ti)
1227		return NULL;
 
 
 
 
 
1228
1229	return ti;
 
 
 
1230}
1231
1232static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1233		long nr_pages, enum dax_access_mode mode, void **kaddr,
1234		pfn_t *pfn)
 
 
1235{
1236	struct mapped_device *md = dax_get_private(dax_dev);
1237	sector_t sector = pgoff * PAGE_SECTORS;
1238	struct dm_target *ti;
1239	long len, ret = -EIO;
1240	int srcu_idx;
1241
1242	ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1243
1244	if (!ti)
1245		goto out;
1246	if (!ti->type->direct_access)
1247		goto out;
1248	len = max_io_len(ti, sector) / PAGE_SECTORS;
1249	if (len < 1)
1250		goto out;
1251	nr_pages = min(len, nr_pages);
1252	ret = ti->type->direct_access(ti, pgoff, nr_pages, mode, kaddr, pfn);
1253
1254 out:
1255	dm_put_live_table(md, srcu_idx);
1256
1257	return ret;
 
1258}
1259
1260static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1261				  size_t nr_pages)
 
1262{
1263	struct mapped_device *md = dax_get_private(dax_dev);
1264	sector_t sector = pgoff * PAGE_SECTORS;
1265	struct dm_target *ti;
1266	int ret = -EIO;
1267	int srcu_idx;
1268
1269	ti = dm_dax_get_live_target(md, sector, &srcu_idx);
 
1270
1271	if (!ti)
1272		goto out;
1273	if (WARN_ON(!ti->type->dax_zero_page_range)) {
1274		/*
1275		 * ->zero_page_range() is mandatory dax operation. If we are
1276		 *  here, something is wrong.
1277		 */
1278		goto out;
1279	}
1280	ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1281 out:
1282	dm_put_live_table(md, srcu_idx);
1283
1284	return ret;
1285}
1286
1287static size_t dm_dax_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff,
1288		void *addr, size_t bytes, struct iov_iter *i)
 
1289{
1290	struct mapped_device *md = dax_get_private(dax_dev);
1291	sector_t sector = pgoff * PAGE_SECTORS;
1292	struct dm_target *ti;
1293	int srcu_idx;
1294	long ret = 0;
1295
1296	ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1297	if (!ti || !ti->type->dax_recovery_write)
1298		goto out;
 
 
 
 
 
1299
1300	ret = ti->type->dax_recovery_write(ti, pgoff, addr, bytes, i);
1301out:
1302	dm_put_live_table(md, srcu_idx);
1303	return ret;
1304}
1305
1306/*
1307 * A target may call dm_accept_partial_bio only from the map routine.  It is
1308 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1309 * operations, REQ_OP_ZONE_APPEND (zone append writes) and any bio serviced by
1310 * __send_duplicate_bios().
1311 *
1312 * dm_accept_partial_bio informs the dm that the target only wants to process
1313 * additional n_sectors sectors of the bio and the rest of the data should be
1314 * sent in a next bio.
1315 *
1316 * A diagram that explains the arithmetics:
1317 * +--------------------+---------------+-------+
1318 * |         1          |       2       |   3   |
1319 * +--------------------+---------------+-------+
1320 *
1321 * <-------------- *tio->len_ptr --------------->
1322 *                      <----- bio_sectors ----->
1323 *                      <-- n_sectors -->
1324 *
1325 * Region 1 was already iterated over with bio_advance or similar function.
1326 *	(it may be empty if the target doesn't use bio_advance)
1327 * Region 2 is the remaining bio size that the target wants to process.
1328 *	(it may be empty if region 1 is non-empty, although there is no reason
1329 *	 to make it empty)
1330 * The target requires that region 3 is to be sent in the next bio.
1331 *
1332 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1333 * the partially processed part (the sum of regions 1+2) must be the same for all
1334 * copies of the bio.
1335 */
1336void dm_accept_partial_bio(struct bio *bio, unsigned int n_sectors)
1337{
1338	struct dm_target_io *tio = clone_to_tio(bio);
1339	struct dm_io *io = tio->io;
1340	unsigned int bio_sectors = bio_sectors(bio);
1341
1342	BUG_ON(dm_tio_flagged(tio, DM_TIO_IS_DUPLICATE_BIO));
1343	BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1344	BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1345	BUG_ON(bio_sectors > *tio->len_ptr);
1346	BUG_ON(n_sectors > bio_sectors);
1347
1348	*tio->len_ptr -= bio_sectors - n_sectors;
1349	bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
 
 
 
 
 
 
1350
1351	/*
1352	 * __split_and_process_bio() may have already saved mapped part
1353	 * for accounting but it is being reduced so update accordingly.
1354	 */
1355	dm_io_set_flag(io, DM_IO_WAS_SPLIT);
1356	io->sectors = n_sectors;
1357	io->sector_offset = bio_sectors(io->orig_bio);
1358}
1359EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1360
1361/*
1362 * @clone: clone bio that DM core passed to target's .map function
1363 * @tgt_clone: clone of @clone bio that target needs submitted
1364 *
1365 * Targets should use this interface to submit bios they take
1366 * ownership of when returning DM_MAPIO_SUBMITTED.
1367 *
1368 * Target should also enable ti->accounts_remapped_io
1369 */
1370void dm_submit_bio_remap(struct bio *clone, struct bio *tgt_clone)
1371{
1372	struct dm_target_io *tio = clone_to_tio(clone);
1373	struct dm_io *io = tio->io;
1374
1375	/* establish bio that will get submitted */
1376	if (!tgt_clone)
1377		tgt_clone = clone;
1378
1379	/*
1380	 * Account io->origin_bio to DM dev on behalf of target
1381	 * that took ownership of IO with DM_MAPIO_SUBMITTED.
1382	 */
1383	dm_start_io_acct(io, clone);
 
1384
1385	trace_block_bio_remap(tgt_clone, disk_devt(io->md->disk),
1386			      tio->old_sector);
1387	submit_bio_noacct(tgt_clone);
 
 
1388}
1389EXPORT_SYMBOL_GPL(dm_submit_bio_remap);
1390
1391static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
 
 
1392{
1393	mutex_lock(&md->swap_bios_lock);
1394	while (latch < md->swap_bios) {
1395		cond_resched();
1396		down(&md->swap_bios_semaphore);
1397		md->swap_bios--;
1398	}
1399	while (latch > md->swap_bios) {
1400		cond_resched();
1401		up(&md->swap_bios_semaphore);
1402		md->swap_bios++;
1403	}
1404	mutex_unlock(&md->swap_bios_lock);
1405}
1406
1407static void __map_bio(struct bio *clone)
1408{
1409	struct dm_target_io *tio = clone_to_tio(clone);
1410	struct dm_target *ti = tio->ti;
1411	struct dm_io *io = tio->io;
1412	struct mapped_device *md = io->md;
1413	int r;
1414
1415	clone->bi_end_io = clone_endio;
1416
1417	/*
1418	 * Map the clone.
1419	 */
1420	tio->old_sector = clone->bi_iter.bi_sector;
1421
1422	if (static_branch_unlikely(&swap_bios_enabled) &&
1423	    unlikely(swap_bios_limit(ti, clone))) {
1424		int latch = get_swap_bios();
 
 
 
 
1425
1426		if (unlikely(latch != md->swap_bios))
1427			__set_swap_bios_limit(md, latch);
1428		down(&md->swap_bios_semaphore);
1429	}
1430
1431	if (static_branch_unlikely(&zoned_enabled)) {
1432		/*
1433		 * Check if the IO needs a special mapping due to zone append
1434		 * emulation on zoned target. In this case, dm_zone_map_bio()
1435		 * calls the target map operation.
 
1436		 */
1437		if (unlikely(dm_emulate_zone_append(md)))
1438			r = dm_zone_map_bio(tio);
 
 
 
 
1439		else
1440			goto do_map;
1441	} else {
1442do_map:
1443		if (likely(ti->type->map == linear_map))
1444			r = linear_map(ti, clone);
1445		else if (ti->type->map == stripe_map)
1446			r = stripe_map(ti, clone);
1447		else
1448			r = ti->type->map(ti, clone);
1449	}
1450
1451	switch (r) {
1452	case DM_MAPIO_SUBMITTED:
1453		/* target has assumed ownership of this io */
1454		if (!ti->accounts_remapped_io)
1455			dm_start_io_acct(io, clone);
1456		break;
1457	case DM_MAPIO_REMAPPED:
1458		dm_submit_bio_remap(clone, NULL);
1459		break;
1460	case DM_MAPIO_KILL:
1461	case DM_MAPIO_REQUEUE:
1462		if (static_branch_unlikely(&swap_bios_enabled) &&
1463		    unlikely(swap_bios_limit(ti, clone)))
1464			up(&md->swap_bios_semaphore);
1465		free_tio(clone);
1466		if (r == DM_MAPIO_KILL)
1467			dm_io_dec_pending(io, BLK_STS_IOERR);
1468		else
1469			dm_io_dec_pending(io, BLK_STS_DM_REQUEUE);
1470		break;
1471	default:
1472		DMCRIT("unimplemented target map return value: %d", r);
1473		BUG();
1474	}
1475}
1476
1477static void setup_split_accounting(struct clone_info *ci, unsigned int len)
1478{
1479	struct dm_io *io = ci->io;
 
 
1480
1481	if (ci->sector_count > len) {
1482		/*
1483		 * Split needed, save the mapped part for accounting.
1484		 * NOTE: dm_accept_partial_bio() will update accordingly.
1485		 */
1486		dm_io_set_flag(io, DM_IO_WAS_SPLIT);
1487		io->sectors = len;
1488		io->sector_offset = bio_sectors(ci->bio);
1489	}
1490}
1491
1492static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1493				struct dm_target *ti, unsigned int num_bios,
1494				unsigned *len, gfp_t gfp_flag)
 
1495{
1496	struct bio *bio;
1497	int try = (gfp_flag & GFP_NOWAIT) ? 0 : 1;
 
 
 
 
 
 
 
 
 
 
1498
1499	for (; try < 2; try++) {
1500		int bio_nr;
1501
1502		if (try && num_bios > 1)
1503			mutex_lock(&ci->io->md->table_devices_lock);
1504		for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1505			bio = alloc_tio(ci, ti, bio_nr, len,
1506					try ? GFP_NOIO : GFP_NOWAIT);
1507			if (!bio)
1508				break;
1509
1510			bio_list_add(blist, bio);
1511		}
1512		if (try && num_bios > 1)
1513			mutex_unlock(&ci->io->md->table_devices_lock);
1514		if (bio_nr == num_bios)
1515			return;
1516
1517		while ((bio = bio_list_pop(blist)))
1518			free_tio(bio);
1519	}
1520}
1521
1522static unsigned int __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1523					  unsigned int num_bios, unsigned int *len,
1524					  gfp_t gfp_flag)
 
 
1525{
1526	struct bio_list blist = BIO_EMPTY_LIST;
1527	struct bio *clone;
1528	unsigned int ret = 0;
1529
1530	if (WARN_ON_ONCE(num_bios == 0)) /* num_bios = 0 is a bug in caller */
1531		return 0;
 
 
1532
1533	/* dm_accept_partial_bio() is not supported with shared tio->len_ptr */
1534	if (len)
1535		setup_split_accounting(ci, *len);
1536
1537	/*
1538	 * Using alloc_multiple_bios(), even if num_bios is 1, to consistently
1539	 * support allocating using GFP_NOWAIT with GFP_NOIO fallback.
1540	 */
1541	alloc_multiple_bios(&blist, ci, ti, num_bios, len, gfp_flag);
1542	while ((clone = bio_list_pop(&blist))) {
1543		if (num_bios > 1)
1544			dm_tio_set_flag(clone_to_tio(clone), DM_TIO_IS_DUPLICATE_BIO);
1545		__map_bio(clone);
1546		ret += 1;
 
 
 
 
 
 
 
 
1547	}
1548
1549	return ret;
 
1550}
 
 
 
1551
1552static void __send_empty_flush(struct clone_info *ci)
 
 
1553{
1554	struct dm_table *t = ci->map;
1555	struct bio flush_bio;
 
 
 
 
 
 
 
 
 
 
1556
1557	/*
1558	 * Use an on-stack bio for this, it's safe since we don't
1559	 * need to reference it after submit. It's just used as
1560	 * the basis for the clone(s).
1561	 */
1562	bio_init(&flush_bio, ci->io->md->disk->part0, NULL, 0,
1563		 REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC);
1564
1565	ci->bio = &flush_bio;
1566	ci->sector_count = 0;
1567	ci->io->tio.clone.bi_iter.bi_size = 0;
1568
1569	for (unsigned int i = 0; i < t->num_targets; i++) {
1570		unsigned int bios;
1571		struct dm_target *ti = dm_table_get_target(t, i);
1572
1573		if (unlikely(ti->num_flush_bios == 0))
1574			continue;
 
 
 
 
 
 
 
 
 
 
 
 
 
1575
1576		atomic_add(ti->num_flush_bios, &ci->io->io_count);
1577		bios = __send_duplicate_bios(ci, ti, ti->num_flush_bios,
1578					     NULL, GFP_NOWAIT);
1579		atomic_sub(ti->num_flush_bios - bios, &ci->io->io_count);
1580	}
1581
 
 
1582	/*
1583	 * alloc_io() takes one extra reference for submission, so the
1584	 * reference won't reach 0 without the following subtraction
1585	 */
1586	atomic_sub(1, &ci->io->io_count);
 
1587
1588	bio_uninit(ci->bio);
1589}
1590
1591static void __send_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1592			       unsigned int num_bios, unsigned int max_granularity,
1593			       unsigned int max_sectors)
 
 
1594{
1595	unsigned int len, bios;
 
 
 
 
 
 
1596
1597	len = min_t(sector_t, ci->sector_count,
1598		    __max_io_len(ti, ci->sector, max_granularity, max_sectors));
 
 
 
 
 
 
1599
1600	atomic_add(num_bios, &ci->io->io_count);
1601	bios = __send_duplicate_bios(ci, ti, num_bios, &len, GFP_NOIO);
1602	/*
1603	 * alloc_io() takes one extra reference for submission, so the
1604	 * reference won't reach 0 without the following (+1) subtraction
1605	 */
1606	atomic_sub(num_bios - bios + 1, &ci->io->io_count);
1607
1608	ci->sector += len;
1609	ci->sector_count -= len;
 
1610}
1611
1612static bool is_abnormal_io(struct bio *bio)
1613{
1614	enum req_op op = bio_op(bio);
 
1615
1616	if (op != REQ_OP_READ && op != REQ_OP_WRITE && op != REQ_OP_FLUSH) {
1617		switch (op) {
1618		case REQ_OP_DISCARD:
1619		case REQ_OP_SECURE_ERASE:
1620		case REQ_OP_WRITE_ZEROES:
1621			return true;
1622		default:
1623			break;
1624		}
1625	}
1626
1627	return false;
 
 
 
1628}
1629
1630static blk_status_t __process_abnormal_io(struct clone_info *ci,
1631					  struct dm_target *ti)
1632{
1633	unsigned int num_bios = 0;
1634	unsigned int max_granularity = 0;
1635	unsigned int max_sectors = 0;
1636	struct queue_limits *limits = dm_get_queue_limits(ti->table->md);
1637
1638	switch (bio_op(ci->bio)) {
1639	case REQ_OP_DISCARD:
1640		num_bios = ti->num_discard_bios;
1641		max_sectors = limits->max_discard_sectors;
1642		if (ti->max_discard_granularity)
1643			max_granularity = max_sectors;
1644		break;
1645	case REQ_OP_SECURE_ERASE:
1646		num_bios = ti->num_secure_erase_bios;
1647		max_sectors = limits->max_secure_erase_sectors;
1648		if (ti->max_secure_erase_granularity)
1649			max_granularity = max_sectors;
1650		break;
1651	case REQ_OP_WRITE_ZEROES:
1652		num_bios = ti->num_write_zeroes_bios;
1653		max_sectors = limits->max_write_zeroes_sectors;
1654		if (ti->max_write_zeroes_granularity)
1655			max_granularity = max_sectors;
1656		break;
1657	default:
1658		break;
1659	}
1660
1661	/*
1662	 * Even though the device advertised support for this type of
1663	 * request, that does not mean every target supports it, and
1664	 * reconfiguration might also have changed that since the
1665	 * check was performed.
1666	 */
1667	if (unlikely(!num_bios))
1668		return BLK_STS_NOTSUPP;
1669
1670	__send_abnormal_io(ci, ti, num_bios, max_granularity, max_sectors);
 
 
1671
1672	return BLK_STS_OK;
1673}
1674
1675/*
1676 * Reuse ->bi_private as dm_io list head for storing all dm_io instances
1677 * associated with this bio, and this bio's bi_private needs to be
1678 * stored in dm_io->data before the reuse.
1679 *
1680 * bio->bi_private is owned by fs or upper layer, so block layer won't
1681 * touch it after splitting. Meantime it won't be changed by anyone after
1682 * bio is submitted. So this reuse is safe.
1683 */
1684static inline struct dm_io **dm_poll_list_head(struct bio *bio)
1685{
1686	return (struct dm_io **)&bio->bi_private;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1687}
1688
1689static void dm_queue_poll_io(struct bio *bio, struct dm_io *io)
 
1690{
1691	struct dm_io **head = dm_poll_list_head(bio);
 
1692
1693	if (!(bio->bi_opf & REQ_DM_POLL_LIST)) {
1694		bio->bi_opf |= REQ_DM_POLL_LIST;
1695		/*
1696		 * Save .bi_private into dm_io, so that we can reuse
1697		 * .bi_private as dm_io list head for storing dm_io list
1698		 */
1699		io->data = bio->bi_private;
1700
1701		/* tell block layer to poll for completion */
1702		bio->bi_cookie = ~BLK_QC_T_NONE;
1703
1704		io->next = NULL;
1705	} else {
1706		/*
1707		 * bio recursed due to split, reuse original poll list,
1708		 * and save bio->bi_private too.
1709		 */
1710		io->data = (*head)->data;
1711		io->next = *head;
1712	}
1713
1714	*head = io;
1715}
1716
1717/*
1718 * Select the correct strategy for processing a non-flush bio.
1719 */
1720static blk_status_t __split_and_process_bio(struct clone_info *ci)
1721{
1722	struct bio *clone;
1723	struct dm_target *ti;
1724	unsigned int len;
 
 
 
 
1725
1726	ti = dm_table_find_target(ci->map, ci->sector);
1727	if (unlikely(!ti))
1728		return BLK_STS_IOERR;
1729
1730	if (unlikely(ci->is_abnormal_io))
1731		return __process_abnormal_io(ci, ti);
1732
1733	/*
1734	 * Only support bio polling for normal IO, and the target io is
1735	 * exactly inside the dm_io instance (verified in dm_poll_dm_io)
1736	 */
1737	ci->submit_as_polled = !!(ci->bio->bi_opf & REQ_POLLED);
1738
1739	len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1740	setup_split_accounting(ci, len);
 
 
 
 
 
 
 
 
1741
1742	if (unlikely(ci->bio->bi_opf & REQ_NOWAIT)) {
1743		if (unlikely(!dm_target_supports_nowait(ti->type)))
1744			return BLK_STS_NOTSUPP;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1745
1746		clone = alloc_tio(ci, ti, 0, &len, GFP_NOWAIT);
1747		if (unlikely(!clone))
1748			return BLK_STS_AGAIN;
1749	} else {
1750		clone = alloc_tio(ci, ti, 0, &len, GFP_NOIO);
1751	}
1752	__map_bio(clone);
1753
1754	ci->sector += len;
1755	ci->sector_count -= len;
1756
1757	return BLK_STS_OK;
1758}
1759
1760static void init_clone_info(struct clone_info *ci, struct dm_io *io,
1761			    struct dm_table *map, struct bio *bio, bool is_abnormal)
1762{
1763	ci->map = map;
1764	ci->io = io;
1765	ci->bio = bio;
1766	ci->is_abnormal_io = is_abnormal;
1767	ci->submit_as_polled = false;
1768	ci->sector = bio->bi_iter.bi_sector;
1769	ci->sector_count = bio_sectors(bio);
1770
1771	/* Shouldn't happen but sector_count was being set to 0 so... */
1772	if (static_branch_unlikely(&zoned_enabled) &&
1773	    WARN_ON_ONCE(op_is_zone_mgmt(bio_op(bio)) && ci->sector_count))
1774		ci->sector_count = 0;
 
 
 
 
1775}
1776
1777/*
1778 * Entry point to split a bio into clones and submit them to the targets.
 
1779 */
1780static void dm_split_and_process_bio(struct mapped_device *md,
1781				     struct dm_table *map, struct bio *bio)
1782{
1783	struct clone_info ci;
1784	struct dm_io *io;
1785	blk_status_t error = BLK_STS_OK;
1786	bool is_abnormal;
 
 
1787
1788	is_abnormal = is_abnormal_io(bio);
1789	if (unlikely(is_abnormal)) {
1790		/*
1791		 * Use bio_split_to_limits() for abnormal IO (e.g. discard, etc)
1792		 * otherwise associated queue_limits won't be imposed.
1793		 */
1794		bio = bio_split_to_limits(bio);
1795		if (!bio)
1796			return;
1797	}
 
 
 
 
 
1798
1799	/* Only support nowait for normal IO */
1800	if (unlikely(bio->bi_opf & REQ_NOWAIT) && !is_abnormal) {
1801		io = alloc_io(md, bio, GFP_NOWAIT);
1802		if (unlikely(!io)) {
1803			/* Unable to do anything without dm_io. */
1804			bio_wouldblock_error(bio);
1805			return;
 
 
 
1806		}
1807	} else {
1808		io = alloc_io(md, bio, GFP_NOIO);
1809	}
1810	init_clone_info(&ci, io, map, bio, is_abnormal);
1811
1812	if (bio->bi_opf & REQ_PREFLUSH) {
1813		__send_empty_flush(&ci);
1814		/* dm_io_complete submits any data associated with flush */
1815		goto out;
1816	}
1817
1818	error = __split_and_process_bio(&ci);
1819	if (error || !ci.sector_count)
1820		goto out;
1821	/*
1822	 * Remainder must be passed to submit_bio_noacct() so it gets handled
1823	 * *after* bios already submitted have been completely processed.
1824	 */
1825	bio_trim(bio, io->sectors, ci.sector_count);
1826	trace_block_split(bio, bio->bi_iter.bi_sector);
1827	bio_inc_remaining(bio);
1828	submit_bio_noacct(bio);
1829out:
1830	/*
1831	 * Drop the extra reference count for non-POLLED bio, and hold one
1832	 * reference for POLLED bio, which will be released in dm_poll_bio
1833	 *
1834	 * Add every dm_io instance into the dm_io list head which is stored
1835	 * in bio->bi_private, so that dm_poll_bio can poll them all.
1836	 */
1837	if (error || !ci.submit_as_polled) {
1838		/*
1839		 * In case of submission failure, the extra reference for
1840		 * submitting io isn't consumed yet
1841		 */
1842		if (error)
1843			atomic_dec(&io->io_count);
1844		dm_io_dec_pending(io, error);
1845	} else
1846		dm_queue_poll_io(bio, io);
1847}
1848
1849static void dm_submit_bio(struct bio *bio)
1850{
1851	struct mapped_device *md = bio->bi_bdev->bd_disk->private_data;
1852	int srcu_idx;
1853	struct dm_table *map;
1854
1855	map = dm_get_live_table(md, &srcu_idx);
1856
1857	/* If suspended, or map not yet available, queue this IO for later */
1858	if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) ||
1859	    unlikely(!map)) {
1860		if (bio->bi_opf & REQ_NOWAIT)
1861			bio_wouldblock_error(bio);
1862		else if (bio->bi_opf & REQ_RAHEAD)
1863			bio_io_error(bio);
1864		else
1865			queue_io(md, bio);
1866		goto out;
1867	}
1868
1869	dm_split_and_process_bio(md, map, bio);
 
1870out:
1871	dm_put_live_table(md, srcu_idx);
1872}
1873
1874static bool dm_poll_dm_io(struct dm_io *io, struct io_comp_batch *iob,
1875			  unsigned int flags)
1876{
1877	WARN_ON_ONCE(!dm_tio_is_normal(&io->tio));
 
 
1878
1879	/* don't poll if the mapped io is done */
1880	if (atomic_read(&io->io_count) > 1)
1881		bio_poll(&io->tio.clone, iob, flags);
 
 
 
 
 
 
 
 
 
1882
1883	/* bio_poll holds the last reference */
1884	return atomic_read(&io->io_count) == 1;
1885}
1886
1887static int dm_poll_bio(struct bio *bio, struct io_comp_batch *iob,
1888		       unsigned int flags)
1889{
1890	struct dm_io **head = dm_poll_list_head(bio);
1891	struct dm_io *list = *head;
1892	struct dm_io *tmp = NULL;
1893	struct dm_io *curr, *next;
1894
1895	/* Only poll normal bio which was marked as REQ_DM_POLL_LIST */
1896	if (!(bio->bi_opf & REQ_DM_POLL_LIST))
1897		return 0;
1898
1899	WARN_ON_ONCE(!list);
1900
1901	/*
1902	 * Restore .bi_private before possibly completing dm_io.
1903	 *
1904	 * bio_poll() is only possible once @bio has been completely
1905	 * submitted via submit_bio_noacct()'s depth-first submission.
1906	 * So there is no dm_queue_poll_io() race associated with
1907	 * clearing REQ_DM_POLL_LIST here.
1908	 */
1909	bio->bi_opf &= ~REQ_DM_POLL_LIST;
1910	bio->bi_private = list->data;
1911
1912	for (curr = list, next = curr->next; curr; curr = next, next =
1913			curr ? curr->next : NULL) {
1914		if (dm_poll_dm_io(curr, iob, flags)) {
1915			/*
1916			 * clone_endio() has already occurred, so no
1917			 * error handling is needed here.
1918			 */
1919			__dm_io_dec_pending(curr);
1920		} else {
1921			curr->next = tmp;
1922			tmp = curr;
 
1923		}
 
1924	}
1925
1926	/* Not done? */
1927	if (tmp) {
1928		bio->bi_opf |= REQ_DM_POLL_LIST;
1929		/* Reset bio->bi_private to dm_io list head */
1930		*head = tmp;
1931		return 0;
1932	}
1933	return 1;
1934}
1935
1936/*
1937 *---------------------------------------------------------------
1938 * An IDR is used to keep track of allocated minor numbers.
1939 *---------------------------------------------------------------
1940 */
1941static void free_minor(int minor)
1942{
1943	spin_lock(&_minor_lock);
1944	idr_remove(&_minor_idr, minor);
1945	spin_unlock(&_minor_lock);
1946}
1947
1948/*
1949 * See if the device with a specific minor # is free.
1950 */
1951static int specific_minor(int minor)
1952{
1953	int r;
1954
1955	if (minor >= (1 << MINORBITS))
1956		return -EINVAL;
1957
1958	idr_preload(GFP_KERNEL);
1959	spin_lock(&_minor_lock);
1960
1961	r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1962
1963	spin_unlock(&_minor_lock);
1964	idr_preload_end();
1965	if (r < 0)
1966		return r == -ENOSPC ? -EBUSY : r;
1967	return 0;
1968}
1969
1970static int next_free_minor(int *minor)
1971{
1972	int r;
1973
1974	idr_preload(GFP_KERNEL);
1975	spin_lock(&_minor_lock);
1976
1977	r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1978
1979	spin_unlock(&_minor_lock);
1980	idr_preload_end();
1981	if (r < 0)
1982		return r;
1983	*minor = r;
1984	return 0;
1985}
1986
1987static const struct block_device_operations dm_blk_dops;
1988static const struct block_device_operations dm_rq_blk_dops;
1989static const struct dax_operations dm_dax_ops;
1990
1991static void dm_wq_work(struct work_struct *work);
1992
1993#ifdef CONFIG_BLK_INLINE_ENCRYPTION
1994static void dm_queue_destroy_crypto_profile(struct request_queue *q)
1995{
1996	dm_destroy_crypto_profile(q->crypto_profile);
1997}
 
 
 
 
 
 
 
 
1998
1999#else /* CONFIG_BLK_INLINE_ENCRYPTION */
2000
2001static inline void dm_queue_destroy_crypto_profile(struct request_queue *q)
2002{
2003}
2004#endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
2005
2006static void cleanup_mapped_device(struct mapped_device *md)
2007{
2008	if (md->wq)
2009		destroy_workqueue(md->wq);
2010	dm_free_md_mempools(md->mempools);
2011
2012	if (md->dax_dev) {
2013		dax_remove_host(md->disk);
2014		kill_dax(md->dax_dev);
2015		put_dax(md->dax_dev);
2016		md->dax_dev = NULL;
2017	}
2018
2019	dm_cleanup_zoned_dev(md);
2020	if (md->disk) {
2021		spin_lock(&_minor_lock);
2022		md->disk->private_data = NULL;
2023		spin_unlock(&_minor_lock);
2024		if (dm_get_md_type(md) != DM_TYPE_NONE) {
2025			struct table_device *td;
2026
2027			dm_sysfs_exit(md);
2028			list_for_each_entry(td, &md->table_devices, list) {
2029				bd_unlink_disk_holder(td->dm_dev.bdev,
2030						      md->disk);
2031			}
2032
2033			/*
2034			 * Hold lock to make sure del_gendisk() won't concurrent
2035			 * with open/close_table_device().
2036			 */
2037			mutex_lock(&md->table_devices_lock);
2038			del_gendisk(md->disk);
2039			mutex_unlock(&md->table_devices_lock);
2040		}
2041		dm_queue_destroy_crypto_profile(md->queue);
2042		put_disk(md->disk);
2043	}
2044
2045	if (md->pending_io) {
2046		free_percpu(md->pending_io);
2047		md->pending_io = NULL;
2048	}
2049
2050	cleanup_srcu_struct(&md->io_barrier);
2051
2052	mutex_destroy(&md->suspend_lock);
2053	mutex_destroy(&md->type_lock);
2054	mutex_destroy(&md->table_devices_lock);
2055	mutex_destroy(&md->swap_bios_lock);
2056
2057	dm_mq_cleanup_mapped_device(md);
2058}
2059
2060/*
2061 * Allocate and initialise a blank device with a given minor.
2062 */
2063static struct mapped_device *alloc_dev(int minor)
2064{
2065	int r, numa_node_id = dm_get_numa_node();
2066	struct dax_device *dax_dev;
2067	struct mapped_device *md;
2068	void *old_md;
2069
2070	md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
2071	if (!md) {
2072		DMERR("unable to allocate device, out of memory.");
2073		return NULL;
2074	}
2075
2076	if (!try_module_get(THIS_MODULE))
2077		goto bad_module_get;
2078
2079	/* get a minor number for the dev */
2080	if (minor == DM_ANY_MINOR)
2081		r = next_free_minor(&minor);
2082	else
2083		r = specific_minor(minor);
2084	if (r < 0)
2085		goto bad_minor;
2086
2087	r = init_srcu_struct(&md->io_barrier);
2088	if (r < 0)
2089		goto bad_io_barrier;
2090
2091	md->numa_node_id = numa_node_id;
2092	md->init_tio_pdu = false;
2093	md->type = DM_TYPE_NONE;
2094	mutex_init(&md->suspend_lock);
2095	mutex_init(&md->type_lock);
2096	mutex_init(&md->table_devices_lock);
2097	spin_lock_init(&md->deferred_lock);
2098	atomic_set(&md->holders, 1);
2099	atomic_set(&md->open_count, 0);
2100	atomic_set(&md->event_nr, 0);
2101	atomic_set(&md->uevent_seq, 0);
2102	INIT_LIST_HEAD(&md->uevent_list);
2103	INIT_LIST_HEAD(&md->table_devices);
2104	spin_lock_init(&md->uevent_lock);
2105
2106	/*
2107	 * default to bio-based until DM table is loaded and md->type
2108	 * established. If request-based table is loaded: blk-mq will
2109	 * override accordingly.
2110	 */
2111	md->disk = blk_alloc_disk(NULL, md->numa_node_id);
2112	if (IS_ERR(md->disk))
2113		goto bad;
2114	md->queue = md->disk->queue;
2115
 
 
2116	init_waitqueue_head(&md->wait);
2117	INIT_WORK(&md->work, dm_wq_work);
2118	INIT_WORK(&md->requeue_work, dm_wq_requeue_work);
2119	init_waitqueue_head(&md->eventq);
2120	init_completion(&md->kobj_holder.completion);
2121
2122	md->requeue_list = NULL;
2123	md->swap_bios = get_swap_bios();
2124	sema_init(&md->swap_bios_semaphore, md->swap_bios);
2125	mutex_init(&md->swap_bios_lock);
2126
2127	md->disk->major = _major;
2128	md->disk->first_minor = minor;
2129	md->disk->minors = 1;
2130	md->disk->flags |= GENHD_FL_NO_PART;
2131	md->disk->fops = &dm_blk_dops;
 
2132	md->disk->private_data = md;
2133	sprintf(md->disk->disk_name, "dm-%d", minor);
2134
2135	dax_dev = alloc_dax(md, &dm_dax_ops);
2136	if (IS_ERR(dax_dev)) {
2137		if (PTR_ERR(dax_dev) != -EOPNOTSUPP)
2138			goto bad;
2139	} else {
2140		set_dax_nocache(dax_dev);
2141		set_dax_nomc(dax_dev);
2142		md->dax_dev = dax_dev;
2143		if (dax_add_host(dax_dev, md->disk))
2144			goto bad;
2145	}
2146
2147	format_dev_t(md->name, MKDEV(_major, minor));
2148
2149	md->wq = alloc_workqueue("kdmflush/%s", WQ_MEM_RECLAIM, 0, md->name);
2150	if (!md->wq)
2151		goto bad;
 
 
 
 
2152
2153	md->pending_io = alloc_percpu(unsigned long);
2154	if (!md->pending_io)
2155		goto bad;
2156
2157	r = dm_stats_init(&md->stats);
2158	if (r < 0)
2159		goto bad;
2160
2161	/* Populate the mapping, nobody knows we exist yet */
2162	spin_lock(&_minor_lock);
2163	old_md = idr_replace(&_minor_idr, md, minor);
2164	spin_unlock(&_minor_lock);
2165
2166	BUG_ON(old_md != MINOR_ALLOCED);
2167
2168	return md;
2169
2170bad:
2171	cleanup_mapped_device(md);
 
 
 
 
 
 
 
2172bad_io_barrier:
2173	free_minor(minor);
2174bad_minor:
2175	module_put(THIS_MODULE);
2176bad_module_get:
2177	kvfree(md);
2178	return NULL;
2179}
2180
2181static void unlock_fs(struct mapped_device *md);
2182
2183static void free_dev(struct mapped_device *md)
2184{
2185	int minor = MINOR(disk_devt(md->disk));
2186
2187	unlock_fs(md);
 
 
 
 
 
 
 
 
 
 
2188
2189	cleanup_mapped_device(md);
 
 
2190
2191	WARN_ON_ONCE(!list_empty(&md->table_devices));
 
2192	dm_stats_cleanup(&md->stats);
2193	free_minor(minor);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2194
2195	module_put(THIS_MODULE);
2196	kvfree(md);
 
2197}
2198
2199/*
2200 * Bind a table to the device.
2201 */
2202static void event_callback(void *context)
2203{
2204	unsigned long flags;
2205	LIST_HEAD(uevents);
2206	struct mapped_device *md = context;
2207
2208	spin_lock_irqsave(&md->uevent_lock, flags);
2209	list_splice_init(&md->uevent_list, &uevents);
2210	spin_unlock_irqrestore(&md->uevent_lock, flags);
2211
2212	dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2213
2214	atomic_inc(&md->event_nr);
2215	wake_up(&md->eventq);
2216	dm_issue_global_event();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2217}
2218
2219/*
2220 * Returns old map, which caller must destroy.
2221 */
2222static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2223			       struct queue_limits *limits)
2224{
2225	struct dm_table *old_map;
 
2226	sector_t size;
2227	int ret;
2228
2229	lockdep_assert_held(&md->suspend_lock);
2230
2231	size = dm_table_get_size(t);
2232
2233	/*
2234	 * Wipe any geometry if the size of the table changed.
2235	 */
2236	if (size != dm_get_size(md))
2237		memset(&md->geometry, 0, sizeof(md->geometry));
2238
2239	set_capacity(md->disk, size);
2240
2241	dm_table_event_callback(t, event_callback, md);
2242
2243	if (dm_table_request_based(t)) {
2244		/*
2245		 * Leverage the fact that request-based DM targets are
2246		 * immutable singletons - used to optimize dm_mq_queue_rq.
2247		 */
2248		md->immutable_target = dm_table_get_immutable_target(t);
 
 
 
2249
2250		/*
2251		 * There is no need to reload with request-based dm because the
2252		 * size of front_pad doesn't change.
2253		 *
2254		 * Note for future: If you are to reload bioset, prep-ed
2255		 * requests in the queue may refer to bio from the old bioset,
2256		 * so you must walk through the queue to unprep.
2257		 */
2258		if (!md->mempools) {
2259			md->mempools = t->mempools;
2260			t->mempools = NULL;
2261		}
2262	} else {
2263		/*
2264		 * The md may already have mempools that need changing.
2265		 * If so, reload bioset because front_pad may have changed
2266		 * because a different table was loaded.
2267		 */
2268		dm_free_md_mempools(md->mempools);
2269		md->mempools = t->mempools;
2270		t->mempools = NULL;
2271	}
2272
2273	ret = dm_table_set_restrictions(t, md->queue, limits);
2274	if (ret) {
2275		old_map = ERR_PTR(ret);
2276		goto out;
2277	}
2278
2279	old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2280	rcu_assign_pointer(md->map, (void *)t);
2281	md->immutable_target_type = dm_table_get_immutable_target_type(t);
2282
2283	if (old_map)
2284		dm_sync_table(md);
2285out:
 
 
 
 
2286	return old_map;
2287}
2288
2289/*
2290 * Returns unbound table for the caller to free.
2291 */
2292static struct dm_table *__unbind(struct mapped_device *md)
2293{
2294	struct dm_table *map = rcu_dereference_protected(md->map, 1);
2295
2296	if (!map)
2297		return NULL;
2298
2299	dm_table_event_callback(map, NULL, NULL);
2300	RCU_INIT_POINTER(md->map, NULL);
2301	dm_sync_table(md);
2302
2303	return map;
2304}
2305
2306/*
2307 * Constructor for a new device.
2308 */
2309int dm_create(int minor, struct mapped_device **result)
2310{
2311	struct mapped_device *md;
2312
2313	md = alloc_dev(minor);
2314	if (!md)
2315		return -ENXIO;
2316
2317	dm_ima_reset_data(md);
2318
2319	*result = md;
2320	return 0;
2321}
2322
2323/*
2324 * Functions to manage md->type.
2325 * All are required to hold md->type_lock.
2326 */
2327void dm_lock_md_type(struct mapped_device *md)
2328{
2329	mutex_lock(&md->type_lock);
2330}
2331
2332void dm_unlock_md_type(struct mapped_device *md)
2333{
2334	mutex_unlock(&md->type_lock);
2335}
2336
2337void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2338{
2339	BUG_ON(!mutex_is_locked(&md->type_lock));
2340	md->type = type;
2341}
2342
2343enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2344{
 
2345	return md->type;
2346}
2347
2348struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2349{
2350	return md->immutable_target_type;
2351}
2352
2353/*
2354 * Setup the DM device's queue based on md's type
 
 
 
 
 
 
 
 
 
 
 
2355 */
2356int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2357{
2358	enum dm_queue_mode type = dm_table_get_type(t);
2359	struct queue_limits limits;
2360	struct table_device *td;
2361	int r;
 
 
 
 
 
2362
2363	switch (type) {
2364	case DM_TYPE_REQUEST_BASED:
2365		md->disk->fops = &dm_rq_blk_dops;
2366		r = dm_mq_init_request_queue(md, t);
2367		if (r) {
2368			DMERR("Cannot initialize queue for request-based dm mapped device");
2369			return r;
2370		}
2371		break;
2372	case DM_TYPE_BIO_BASED:
2373	case DM_TYPE_DAX_BIO_BASED:
2374		blk_queue_flag_set(QUEUE_FLAG_IO_STAT, md->queue);
2375		break;
2376	case DM_TYPE_NONE:
2377		WARN_ON_ONCE(true);
2378		break;
2379	}
2380
2381	r = dm_calculate_queue_limits(t, &limits);
2382	if (r) {
2383		DMERR("Cannot calculate initial queue limits");
2384		return r;
2385	}
2386	r = dm_table_set_restrictions(t, md->queue, &limits);
2387	if (r)
2388		return r;
2389
2390	/*
2391	 * Hold lock to make sure add_disk() and del_gendisk() won't concurrent
2392	 * with open_table_device() and close_table_device().
2393	 */
2394	mutex_lock(&md->table_devices_lock);
2395	r = add_disk(md->disk);
2396	mutex_unlock(&md->table_devices_lock);
2397	if (r)
2398		return r;
2399
2400	/*
2401	 * Register the holder relationship for devices added before the disk
2402	 * was live.
2403	 */
2404	list_for_each_entry(td, &md->table_devices, list) {
2405		r = bd_link_disk_holder(td->dm_dev.bdev, md->disk);
2406		if (r)
2407			goto out_undo_holders;
 
2408	}
2409
2410	r = dm_sysfs_init(md);
2411	if (r)
2412		goto out_undo_holders;
2413
2414	md->type = type;
2415	return 0;
2416
2417out_undo_holders:
2418	list_for_each_entry_continue_reverse(td, &md->table_devices, list)
2419		bd_unlink_disk_holder(td->dm_dev.bdev, md->disk);
2420	mutex_lock(&md->table_devices_lock);
2421	del_gendisk(md->disk);
2422	mutex_unlock(&md->table_devices_lock);
2423	return r;
2424}
2425
2426struct mapped_device *dm_get_md(dev_t dev)
2427{
2428	struct mapped_device *md;
2429	unsigned int minor = MINOR(dev);
2430
2431	if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2432		return NULL;
2433
2434	spin_lock(&_minor_lock);
2435
2436	md = idr_find(&_minor_idr, minor);
2437	if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2438	    test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
 
 
2439		md = NULL;
2440		goto out;
2441	}
2442	dm_get(md);
2443out:
2444	spin_unlock(&_minor_lock);
2445
2446	return md;
2447}
 
 
 
 
 
 
 
 
 
 
2448EXPORT_SYMBOL_GPL(dm_get_md);
2449
2450void *dm_get_mdptr(struct mapped_device *md)
2451{
2452	return md->interface_ptr;
2453}
2454
2455void dm_set_mdptr(struct mapped_device *md, void *ptr)
2456{
2457	md->interface_ptr = ptr;
2458}
2459
2460void dm_get(struct mapped_device *md)
2461{
2462	atomic_inc(&md->holders);
2463	BUG_ON(test_bit(DMF_FREEING, &md->flags));
2464}
2465
2466int dm_hold(struct mapped_device *md)
2467{
2468	spin_lock(&_minor_lock);
2469	if (test_bit(DMF_FREEING, &md->flags)) {
2470		spin_unlock(&_minor_lock);
2471		return -EBUSY;
2472	}
2473	dm_get(md);
2474	spin_unlock(&_minor_lock);
2475	return 0;
2476}
2477EXPORT_SYMBOL_GPL(dm_hold);
2478
2479const char *dm_device_name(struct mapped_device *md)
2480{
2481	return md->name;
2482}
2483EXPORT_SYMBOL_GPL(dm_device_name);
2484
2485static void __dm_destroy(struct mapped_device *md, bool wait)
2486{
2487	struct dm_table *map;
2488	int srcu_idx;
2489
2490	might_sleep();
2491
2492	spin_lock(&_minor_lock);
 
2493	idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2494	set_bit(DMF_FREEING, &md->flags);
2495	spin_unlock(&_minor_lock);
2496
2497	blk_mark_disk_dead(md->disk);
2498
2499	/*
2500	 * Take suspend_lock so that presuspend and postsuspend methods
2501	 * do not race with internal suspend.
2502	 */
2503	mutex_lock(&md->suspend_lock);
2504	map = dm_get_live_table(md, &srcu_idx);
2505	if (!dm_suspended_md(md)) {
2506		dm_table_presuspend_targets(map);
2507		set_bit(DMF_SUSPENDED, &md->flags);
2508		set_bit(DMF_POST_SUSPENDING, &md->flags);
2509		dm_table_postsuspend_targets(map);
2510	}
2511	/* dm_put_live_table must be before fsleep, otherwise deadlock is possible */
 
2512	dm_put_live_table(md, srcu_idx);
2513	mutex_unlock(&md->suspend_lock);
2514
2515	/*
2516	 * Rare, but there may be I/O requests still going to complete,
2517	 * for example.  Wait for all references to disappear.
2518	 * No one should increment the reference count of the mapped_device,
2519	 * after the mapped_device state becomes DMF_FREEING.
2520	 */
2521	if (wait)
2522		while (atomic_read(&md->holders))
2523			fsleep(1000);
2524	else if (atomic_read(&md->holders))
2525		DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2526		       dm_device_name(md), atomic_read(&md->holders));
2527
 
2528	dm_table_destroy(__unbind(md));
2529	free_dev(md);
2530}
2531
2532void dm_destroy(struct mapped_device *md)
2533{
2534	__dm_destroy(md, true);
2535}
2536
2537void dm_destroy_immediate(struct mapped_device *md)
2538{
2539	__dm_destroy(md, false);
2540}
2541
2542void dm_put(struct mapped_device *md)
2543{
2544	atomic_dec(&md->holders);
2545}
2546EXPORT_SYMBOL_GPL(dm_put);
2547
2548static bool dm_in_flight_bios(struct mapped_device *md)
2549{
2550	int cpu;
2551	unsigned long sum = 0;
2552
2553	for_each_possible_cpu(cpu)
2554		sum += *per_cpu_ptr(md->pending_io, cpu);
2555
2556	return sum != 0;
2557}
2558
2559static int dm_wait_for_bios_completion(struct mapped_device *md, unsigned int task_state)
2560{
2561	int r = 0;
2562	DEFINE_WAIT(wait);
2563
2564	while (true) {
2565		prepare_to_wait(&md->wait, &wait, task_state);
2566
2567		if (!dm_in_flight_bios(md))
2568			break;
2569
2570		if (signal_pending_state(task_state, current)) {
 
2571			r = -EINTR;
2572			break;
2573		}
2574
2575		io_schedule();
2576	}
2577	finish_wait(&md->wait, &wait);
2578
2579	smp_rmb();
2580
2581	return r;
2582}
2583
2584static int dm_wait_for_completion(struct mapped_device *md, unsigned int task_state)
2585{
2586	int r = 0;
2587
2588	if (!queue_is_mq(md->queue))
2589		return dm_wait_for_bios_completion(md, task_state);
2590
2591	while (true) {
2592		if (!blk_mq_queue_inflight(md->queue))
2593			break;
2594
2595		if (signal_pending_state(task_state, current)) {
2596			r = -EINTR;
2597			break;
2598		}
2599
2600		fsleep(5000);
2601	}
2602
2603	return r;
2604}
2605
2606/*
2607 * Process the deferred bios
2608 */
2609static void dm_wq_work(struct work_struct *work)
2610{
2611	struct mapped_device *md = container_of(work, struct mapped_device, work);
2612	struct bio *bio;
 
 
 
 
 
2613
2614	while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2615		spin_lock_irq(&md->deferred_lock);
2616		bio = bio_list_pop(&md->deferred);
2617		spin_unlock_irq(&md->deferred_lock);
2618
2619		if (!bio)
2620			break;
2621
2622		submit_bio_noacct(bio);
2623		cond_resched();
 
 
2624	}
 
 
2625}
2626
2627static void dm_queue_flush(struct mapped_device *md)
2628{
2629	clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2630	smp_mb__after_atomic();
2631	queue_work(md->wq, &md->work);
2632}
2633
2634/*
2635 * Swap in a new table, returning the old one for the caller to destroy.
2636 */
2637struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2638{
2639	struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2640	struct queue_limits limits;
2641	int r;
2642
2643	mutex_lock(&md->suspend_lock);
2644
2645	/* device must be suspended */
2646	if (!dm_suspended_md(md))
2647		goto out;
2648
2649	/*
2650	 * If the new table has no data devices, retain the existing limits.
2651	 * This helps multipath with queue_if_no_path if all paths disappear,
2652	 * then new I/O is queued based on these limits, and then some paths
2653	 * reappear.
2654	 */
2655	if (dm_table_has_no_data_devices(table)) {
2656		live_map = dm_get_live_table_fast(md);
2657		if (live_map)
2658			limits = md->queue->limits;
2659		dm_put_live_table_fast(md);
2660	}
2661
2662	if (!live_map) {
2663		r = dm_calculate_queue_limits(table, &limits);
2664		if (r) {
2665			map = ERR_PTR(r);
2666			goto out;
2667		}
2668	}
2669
2670	map = __bind(md, table, &limits);
2671	dm_issue_global_event();
2672
2673out:
2674	mutex_unlock(&md->suspend_lock);
2675	return map;
2676}
2677
2678/*
2679 * Functions to lock and unlock any filesystem running on the
2680 * device.
2681 */
2682static int lock_fs(struct mapped_device *md)
2683{
2684	int r;
2685
2686	WARN_ON(test_bit(DMF_FROZEN, &md->flags));
2687
2688	r = bdev_freeze(md->disk->part0);
2689	if (!r)
2690		set_bit(DMF_FROZEN, &md->flags);
2691	return r;
 
 
 
 
 
 
2692}
2693
2694static void unlock_fs(struct mapped_device *md)
2695{
2696	if (!test_bit(DMF_FROZEN, &md->flags))
2697		return;
2698	bdev_thaw(md->disk->part0);
 
 
2699	clear_bit(DMF_FROZEN, &md->flags);
2700}
2701
2702/*
2703 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2704 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2705 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
 
 
 
 
 
2706 *
2707 * If __dm_suspend returns 0, the device is completely quiescent
2708 * now. There is no request-processing activity. All new requests
2709 * are being added to md->deferred list.
2710 */
2711static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2712			unsigned int suspend_flags, unsigned int task_state,
2713			int dmf_suspended_flag)
2714{
2715	bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2716	bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2717	int r;
 
 
 
 
 
 
 
 
2718
2719	lockdep_assert_held(&md->suspend_lock);
2720
2721	/*
2722	 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2723	 * This flag is cleared before dm_suspend returns.
2724	 */
2725	if (noflush)
2726		set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2727	else
2728		DMDEBUG("%s: suspending with flush", dm_device_name(md));
2729
2730	/*
2731	 * This gets reverted if there's an error later and the targets
2732	 * provide the .presuspend_undo hook.
2733	 */
2734	dm_table_presuspend_targets(map);
2735
2736	/*
2737	 * Flush I/O to the device.
2738	 * Any I/O submitted after lock_fs() may not be flushed.
2739	 * noflush takes precedence over do_lockfs.
2740	 * (lock_fs() flushes I/Os and waits for them to complete.)
2741	 */
2742	if (!noflush && do_lockfs) {
2743		r = lock_fs(md);
2744		if (r) {
2745			dm_table_presuspend_undo_targets(map);
2746			return r;
2747		}
2748	}
2749
2750	/*
2751	 * Here we must make sure that no processes are submitting requests
2752	 * to target drivers i.e. no one may be executing
2753	 * dm_split_and_process_bio from dm_submit_bio.
 
2754	 *
2755	 * To get all processes out of dm_split_and_process_bio in dm_submit_bio,
2756	 * we take the write lock. To prevent any process from reentering
2757	 * dm_split_and_process_bio from dm_submit_bio and quiesce the thread
2758	 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2759	 * flush_workqueue(md->wq).
2760	 */
2761	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2762	if (map)
2763		synchronize_srcu(&md->io_barrier);
2764
2765	/*
2766	 * Stop md->queue before flushing md->wq in case request-based
2767	 * dm defers requests to md->wq from md->queue.
2768	 */
2769	if (dm_request_based(md))
2770		dm_stop_queue(md->queue);
2771
2772	flush_workqueue(md->wq);
2773
2774	/*
2775	 * At this point no more requests are entering target request routines.
2776	 * We call dm_wait_for_completion to wait for all existing requests
2777	 * to finish.
2778	 */
2779	r = dm_wait_for_completion(md, task_state);
2780	if (!r)
2781		set_bit(dmf_suspended_flag, &md->flags);
2782
2783	if (noflush)
2784		clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2785	if (map)
2786		synchronize_srcu(&md->io_barrier);
2787
2788	/* were we interrupted ? */
2789	if (r < 0) {
2790		dm_queue_flush(md);
2791
2792		if (dm_request_based(md))
2793			dm_start_queue(md->queue);
2794
2795		unlock_fs(md);
2796		dm_table_presuspend_undo_targets(map);
2797		/* pushback list is already flushed, so skip flush */
2798	}
2799
2800	return r;
2801}
2802
2803/*
2804 * We need to be able to change a mapping table under a mounted
2805 * filesystem.  For example we might want to move some data in
2806 * the background.  Before the table can be swapped with
2807 * dm_bind_table, dm_suspend must be called to flush any in
2808 * flight bios and ensure that any further io gets deferred.
2809 */
2810/*
2811 * Suspend mechanism in request-based dm.
2812 *
2813 * 1. Flush all I/Os by lock_fs() if needed.
2814 * 2. Stop dispatching any I/O by stopping the request_queue.
2815 * 3. Wait for all in-flight I/Os to be completed or requeued.
2816 *
2817 * To abort suspend, start the request_queue.
2818 */
2819int dm_suspend(struct mapped_device *md, unsigned int suspend_flags)
2820{
2821	struct dm_table *map = NULL;
2822	int r = 0;
2823
2824retry:
2825	mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2826
2827	if (dm_suspended_md(md)) {
2828		r = -EINVAL;
2829		goto out_unlock;
2830	}
2831
2832	if (dm_suspended_internally_md(md)) {
2833		/* already internally suspended, wait for internal resume */
2834		mutex_unlock(&md->suspend_lock);
2835		r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2836		if (r)
2837			return r;
2838		goto retry;
2839	}
2840
2841	map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2842	if (!map) {
2843		/* avoid deadlock with fs/namespace.c:do_mount() */
2844		suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
2845	}
2846
2847	r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2848	if (r)
2849		goto out_unlock;
2850
2851	set_bit(DMF_POST_SUSPENDING, &md->flags);
2852	dm_table_postsuspend_targets(map);
2853	clear_bit(DMF_POST_SUSPENDING, &md->flags);
2854
2855out_unlock:
2856	mutex_unlock(&md->suspend_lock);
2857	return r;
2858}
2859
2860static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2861{
2862	if (map) {
2863		int r = dm_table_resume_targets(map);
2864
2865		if (r)
2866			return r;
2867	}
 
 
 
 
 
 
 
 
2868
2869	dm_queue_flush(md);
2870
2871	/*
2872	 * Flushing deferred I/Os must be done after targets are resumed
2873	 * so that mapping of targets can work correctly.
2874	 * Request-based dm is queueing the deferred I/Os in its request_queue.
2875	 */
2876	if (dm_request_based(md))
2877		dm_start_queue(md->queue);
2878
2879	unlock_fs(md);
2880
2881	return 0;
2882}
2883
2884int dm_resume(struct mapped_device *md)
2885{
2886	int r;
2887	struct dm_table *map = NULL;
2888
2889retry:
2890	r = -EINVAL;
2891	mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2892
2893	if (!dm_suspended_md(md))
2894		goto out;
2895
2896	if (dm_suspended_internally_md(md)) {
2897		/* already internally suspended, wait for internal resume */
2898		mutex_unlock(&md->suspend_lock);
2899		r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2900		if (r)
2901			return r;
2902		goto retry;
2903	}
2904
2905	map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2906	if (!map || !dm_table_get_size(map))
2907		goto out;
2908
2909	r = __dm_resume(md, map);
2910	if (r)
2911		goto out;
2912
2913	clear_bit(DMF_SUSPENDED, &md->flags);
2914out:
2915	mutex_unlock(&md->suspend_lock);
2916
2917	return r;
2918}
2919
2920/*
2921 * Internal suspend/resume works like userspace-driven suspend. It waits
2922 * until all bios finish and prevents issuing new bios to the target drivers.
2923 * It may be used only from the kernel.
 
 
 
2924 */
2925
2926static void __dm_internal_suspend(struct mapped_device *md, unsigned int suspend_flags)
2927{
2928	struct dm_table *map = NULL;
2929
2930	lockdep_assert_held(&md->suspend_lock);
2931
2932	if (md->internal_suspend_count++)
2933		return; /* nested internal suspend */
2934
2935	if (dm_suspended_md(md)) {
2936		set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2937		return; /* nest suspend */
2938	}
2939
2940	map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2941
2942	/*
2943	 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2944	 * supported.  Properly supporting a TASK_INTERRUPTIBLE internal suspend
2945	 * would require changing .presuspend to return an error -- avoid this
2946	 * until there is a need for more elaborate variants of internal suspend.
2947	 */
2948	(void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2949			    DMF_SUSPENDED_INTERNALLY);
2950
2951	set_bit(DMF_POST_SUSPENDING, &md->flags);
2952	dm_table_postsuspend_targets(map);
2953	clear_bit(DMF_POST_SUSPENDING, &md->flags);
2954}
2955
2956static void __dm_internal_resume(struct mapped_device *md)
2957{
2958	int r;
2959	struct dm_table *map;
2960
2961	BUG_ON(!md->internal_suspend_count);
2962
2963	if (--md->internal_suspend_count)
2964		return; /* resume from nested internal suspend */
2965
2966	if (dm_suspended_md(md))
2967		goto done; /* resume from nested suspend */
2968
2969	map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2970	r = __dm_resume(md, map);
2971	if (r) {
2972		/*
2973		 * If a preresume method of some target failed, we are in a
2974		 * tricky situation. We can't return an error to the caller. We
2975		 * can't fake success because then the "resume" and
2976		 * "postsuspend" methods would not be paired correctly, and it
2977		 * would break various targets, for example it would cause list
2978		 * corruption in the "origin" target.
2979		 *
2980		 * So, we fake normal suspend here, to make sure that the
2981		 * "resume" and "postsuspend" methods will be paired correctly.
2982		 */
2983		DMERR("Preresume method failed: %d", r);
2984		set_bit(DMF_SUSPENDED, &md->flags);
2985	}
2986done:
2987	clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2988	smp_mb__after_atomic();
2989	wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2990}
2991
2992void dm_internal_suspend_noflush(struct mapped_device *md)
2993{
2994	mutex_lock(&md->suspend_lock);
2995	__dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2996	mutex_unlock(&md->suspend_lock);
2997}
2998EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2999
3000void dm_internal_resume(struct mapped_device *md)
3001{
3002	mutex_lock(&md->suspend_lock);
3003	__dm_internal_resume(md);
3004	mutex_unlock(&md->suspend_lock);
3005}
3006EXPORT_SYMBOL_GPL(dm_internal_resume);
3007
3008/*
3009 * Fast variants of internal suspend/resume hold md->suspend_lock,
3010 * which prevents interaction with userspace-driven suspend.
3011 */
3012
3013void dm_internal_suspend_fast(struct mapped_device *md)
3014{
3015	mutex_lock(&md->suspend_lock);
3016	if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3017		return;
3018
3019	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
3020	synchronize_srcu(&md->io_barrier);
3021	flush_workqueue(md->wq);
3022	dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
3023}
3024EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
3025
3026void dm_internal_resume_fast(struct mapped_device *md)
3027{
3028	if (dm_suspended_md(md) || dm_suspended_internally_md(md))
3029		goto done;
3030
3031	dm_queue_flush(md);
3032
3033done:
3034	mutex_unlock(&md->suspend_lock);
3035}
3036EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
3037
3038/*
3039 *---------------------------------------------------------------
3040 * Event notification.
3041 *---------------------------------------------------------------
3042 */
3043int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
3044		      unsigned int cookie, bool need_resize_uevent)
3045{
3046	int r;
3047	unsigned int noio_flag;
3048	char udev_cookie[DM_COOKIE_LENGTH];
3049	char *envp[3] = { NULL, NULL, NULL };
3050	char **envpp = envp;
3051	if (cookie) {
 
 
3052		snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
3053			 DM_COOKIE_ENV_VAR_NAME, cookie);
3054		*envpp++ = udev_cookie;
3055	}
3056	if (need_resize_uevent) {
3057		*envpp++ = "RESIZE=1";
3058	}
3059
3060	noio_flag = memalloc_noio_save();
3061
3062	r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj, action, envp);
3063
3064	memalloc_noio_restore(noio_flag);
3065
3066	return r;
3067}
3068
3069uint32_t dm_next_uevent_seq(struct mapped_device *md)
3070{
3071	return atomic_add_return(1, &md->uevent_seq);
3072}
3073
3074uint32_t dm_get_event_nr(struct mapped_device *md)
3075{
3076	return atomic_read(&md->event_nr);
3077}
3078
3079int dm_wait_event(struct mapped_device *md, int event_nr)
3080{
3081	return wait_event_interruptible(md->eventq,
3082			(event_nr != atomic_read(&md->event_nr)));
3083}
3084
3085void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
3086{
3087	unsigned long flags;
3088
3089	spin_lock_irqsave(&md->uevent_lock, flags);
3090	list_add(elist, &md->uevent_list);
3091	spin_unlock_irqrestore(&md->uevent_lock, flags);
3092}
3093
3094/*
3095 * The gendisk is only valid as long as you have a reference
3096 * count on 'md'.
3097 */
3098struct gendisk *dm_disk(struct mapped_device *md)
3099{
3100	return md->disk;
3101}
3102EXPORT_SYMBOL_GPL(dm_disk);
3103
3104struct kobject *dm_kobject(struct mapped_device *md)
3105{
3106	return &md->kobj_holder.kobj;
3107}
3108
3109struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3110{
3111	struct mapped_device *md;
3112
3113	md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3114
3115	spin_lock(&_minor_lock);
3116	if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
3117		md = NULL;
3118		goto out;
3119	}
3120	dm_get(md);
3121out:
3122	spin_unlock(&_minor_lock);
3123
3124	return md;
3125}
3126
3127int dm_suspended_md(struct mapped_device *md)
3128{
3129	return test_bit(DMF_SUSPENDED, &md->flags);
3130}
3131
3132static int dm_post_suspending_md(struct mapped_device *md)
3133{
3134	return test_bit(DMF_POST_SUSPENDING, &md->flags);
3135}
3136
3137int dm_suspended_internally_md(struct mapped_device *md)
3138{
3139	return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3140}
3141
3142int dm_test_deferred_remove_flag(struct mapped_device *md)
3143{
3144	return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3145}
3146
3147int dm_suspended(struct dm_target *ti)
3148{
3149	return dm_suspended_md(ti->table->md);
3150}
3151EXPORT_SYMBOL_GPL(dm_suspended);
3152
3153int dm_post_suspending(struct dm_target *ti)
3154{
3155	return dm_post_suspending_md(ti->table->md);
3156}
3157EXPORT_SYMBOL_GPL(dm_post_suspending);
3158
3159int dm_noflush_suspending(struct dm_target *ti)
3160{
3161	return __noflush_suspending(ti->table->md);
3162}
3163EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3164
3165void dm_free_md_mempools(struct dm_md_mempools *pools)
3166{
 
 
 
 
 
3167	if (!pools)
3168		return;
3169
3170	bioset_exit(&pools->bs);
3171	bioset_exit(&pools->io_bs);
3172
3173	kfree(pools);
3174}
3175
3176struct dm_pr {
3177	u64	old_key;
3178	u64	new_key;
3179	u32	flags;
3180	bool	abort;
3181	bool	fail_early;
3182	int	ret;
3183	enum pr_type type;
3184	struct pr_keys *read_keys;
3185	struct pr_held_reservation *rsv;
3186};
3187
3188static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3189		      struct dm_pr *pr)
3190{
3191	struct mapped_device *md = bdev->bd_disk->private_data;
3192	struct dm_table *table;
3193	struct dm_target *ti;
3194	int ret = -ENOTTY, srcu_idx;
3195
3196	table = dm_get_live_table(md, &srcu_idx);
3197	if (!table || !dm_table_get_size(table))
3198		goto out;
3199
3200	/* We only support devices that have a single target */
3201	if (table->num_targets != 1)
3202		goto out;
3203	ti = dm_table_get_target(table, 0);
3204
3205	if (dm_suspended_md(md)) {
3206		ret = -EAGAIN;
3207		goto out;
3208	}
3209
3210	ret = -EINVAL;
3211	if (!ti->type->iterate_devices)
3212		goto out;
3213
3214	ti->type->iterate_devices(ti, fn, pr);
3215	ret = 0;
3216out:
3217	dm_put_live_table(md, srcu_idx);
3218	return ret;
3219}
3220
3221/*
3222 * For register / unregister we need to manually call out to every path.
3223 */
3224static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3225			    sector_t start, sector_t len, void *data)
3226{
3227	struct dm_pr *pr = data;
3228	const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3229	int ret;
3230
3231	if (!ops || !ops->pr_register) {
3232		pr->ret = -EOPNOTSUPP;
3233		return -1;
3234	}
3235
3236	ret = ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3237	if (!ret)
3238		return 0;
3239
3240	if (!pr->ret)
3241		pr->ret = ret;
3242
3243	if (pr->fail_early)
3244		return -1;
3245
3246	return 0;
3247}
3248
3249static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3250			  u32 flags)
3251{
3252	struct dm_pr pr = {
3253		.old_key	= old_key,
3254		.new_key	= new_key,
3255		.flags		= flags,
3256		.fail_early	= true,
3257		.ret		= 0,
3258	};
3259	int ret;
3260
3261	ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3262	if (ret) {
3263		/* Didn't even get to register a path */
3264		return ret;
3265	}
3266
3267	if (!pr.ret)
3268		return 0;
3269	ret = pr.ret;
3270
3271	if (!new_key)
3272		return ret;
3273
3274	/* unregister all paths if we failed to register any path */
3275	pr.old_key = new_key;
3276	pr.new_key = 0;
3277	pr.flags = 0;
3278	pr.fail_early = false;
3279	(void) dm_call_pr(bdev, __dm_pr_register, &pr);
3280	return ret;
3281}
3282
3283
3284static int __dm_pr_reserve(struct dm_target *ti, struct dm_dev *dev,
3285			   sector_t start, sector_t len, void *data)
3286{
3287	struct dm_pr *pr = data;
3288	const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3289
3290	if (!ops || !ops->pr_reserve) {
3291		pr->ret = -EOPNOTSUPP;
3292		return -1;
3293	}
3294
3295	pr->ret = ops->pr_reserve(dev->bdev, pr->old_key, pr->type, pr->flags);
3296	if (!pr->ret)
3297		return -1;
3298
3299	return 0;
3300}
3301
3302static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3303			 u32 flags)
3304{
3305	struct dm_pr pr = {
3306		.old_key	= key,
3307		.flags		= flags,
3308		.type		= type,
3309		.fail_early	= false,
3310		.ret		= 0,
3311	};
3312	int ret;
3313
3314	ret = dm_call_pr(bdev, __dm_pr_reserve, &pr);
3315	if (ret)
3316		return ret;
3317
3318	return pr.ret;
3319}
3320
3321/*
3322 * If there is a non-All Registrants type of reservation, the release must be
3323 * sent down the holding path. For the cases where there is no reservation or
3324 * the path is not the holder the device will also return success, so we must
3325 * try each path to make sure we got the correct path.
3326 */
3327static int __dm_pr_release(struct dm_target *ti, struct dm_dev *dev,
3328			   sector_t start, sector_t len, void *data)
3329{
3330	struct dm_pr *pr = data;
3331	const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3332
3333	if (!ops || !ops->pr_release) {
3334		pr->ret = -EOPNOTSUPP;
3335		return -1;
3336	}
3337
3338	pr->ret = ops->pr_release(dev->bdev, pr->old_key, pr->type);
3339	if (pr->ret)
3340		return -1;
3341
3342	return 0;
3343}
3344
3345static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3346{
3347	struct dm_pr pr = {
3348		.old_key	= key,
3349		.type		= type,
3350		.fail_early	= false,
3351	};
3352	int ret;
3353
3354	ret = dm_call_pr(bdev, __dm_pr_release, &pr);
3355	if (ret)
3356		return ret;
3357
3358	return pr.ret;
3359}
3360
3361static int __dm_pr_preempt(struct dm_target *ti, struct dm_dev *dev,
3362			   sector_t start, sector_t len, void *data)
3363{
3364	struct dm_pr *pr = data;
3365	const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3366
3367	if (!ops || !ops->pr_preempt) {
3368		pr->ret = -EOPNOTSUPP;
3369		return -1;
3370	}
3371
3372	pr->ret = ops->pr_preempt(dev->bdev, pr->old_key, pr->new_key, pr->type,
3373				  pr->abort);
3374	if (!pr->ret)
3375		return -1;
3376
3377	return 0;
3378}
3379
3380static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3381			 enum pr_type type, bool abort)
3382{
3383	struct dm_pr pr = {
3384		.new_key	= new_key,
3385		.old_key	= old_key,
3386		.type		= type,
3387		.fail_early	= false,
3388	};
3389	int ret;
3390
3391	ret = dm_call_pr(bdev, __dm_pr_preempt, &pr);
3392	if (ret)
3393		return ret;
3394
3395	return pr.ret;
3396}
3397
3398static int dm_pr_clear(struct block_device *bdev, u64 key)
3399{
3400	struct mapped_device *md = bdev->bd_disk->private_data;
3401	const struct pr_ops *ops;
3402	int r, srcu_idx;
3403
3404	r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3405	if (r < 0)
3406		goto out;
3407
3408	ops = bdev->bd_disk->fops->pr_ops;
3409	if (ops && ops->pr_clear)
3410		r = ops->pr_clear(bdev, key);
3411	else
3412		r = -EOPNOTSUPP;
3413out:
3414	dm_unprepare_ioctl(md, srcu_idx);
3415	return r;
3416}
3417
3418static int __dm_pr_read_keys(struct dm_target *ti, struct dm_dev *dev,
3419			     sector_t start, sector_t len, void *data)
3420{
3421	struct dm_pr *pr = data;
3422	const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3423
3424	if (!ops || !ops->pr_read_keys) {
3425		pr->ret = -EOPNOTSUPP;
3426		return -1;
3427	}
3428
3429	pr->ret = ops->pr_read_keys(dev->bdev, pr->read_keys);
3430	if (!pr->ret)
3431		return -1;
3432
3433	return 0;
3434}
3435
3436static int dm_pr_read_keys(struct block_device *bdev, struct pr_keys *keys)
3437{
3438	struct dm_pr pr = {
3439		.read_keys = keys,
3440	};
3441	int ret;
3442
3443	ret = dm_call_pr(bdev, __dm_pr_read_keys, &pr);
3444	if (ret)
3445		return ret;
3446
3447	return pr.ret;
3448}
3449
3450static int __dm_pr_read_reservation(struct dm_target *ti, struct dm_dev *dev,
3451				    sector_t start, sector_t len, void *data)
3452{
3453	struct dm_pr *pr = data;
3454	const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3455
3456	if (!ops || !ops->pr_read_reservation) {
3457		pr->ret = -EOPNOTSUPP;
3458		return -1;
3459	}
3460
3461	pr->ret = ops->pr_read_reservation(dev->bdev, pr->rsv);
3462	if (!pr->ret)
3463		return -1;
3464
3465	return 0;
3466}
3467
3468static int dm_pr_read_reservation(struct block_device *bdev,
3469				  struct pr_held_reservation *rsv)
3470{
3471	struct dm_pr pr = {
3472		.rsv = rsv,
3473	};
3474	int ret;
3475
3476	ret = dm_call_pr(bdev, __dm_pr_read_reservation, &pr);
3477	if (ret)
3478		return ret;
3479
3480	return pr.ret;
3481}
3482
3483static const struct pr_ops dm_pr_ops = {
3484	.pr_register	= dm_pr_register,
3485	.pr_reserve	= dm_pr_reserve,
3486	.pr_release	= dm_pr_release,
3487	.pr_preempt	= dm_pr_preempt,
3488	.pr_clear	= dm_pr_clear,
3489	.pr_read_keys	= dm_pr_read_keys,
3490	.pr_read_reservation = dm_pr_read_reservation,
3491};
3492
3493static const struct block_device_operations dm_blk_dops = {
3494	.submit_bio = dm_submit_bio,
3495	.poll_bio = dm_poll_bio,
3496	.open = dm_blk_open,
3497	.release = dm_blk_close,
3498	.ioctl = dm_blk_ioctl,
3499	.getgeo = dm_blk_getgeo,
3500	.report_zones = dm_blk_report_zones,
3501	.pr_ops = &dm_pr_ops,
3502	.owner = THIS_MODULE
3503};
3504
3505static const struct block_device_operations dm_rq_blk_dops = {
3506	.open = dm_blk_open,
3507	.release = dm_blk_close,
3508	.ioctl = dm_blk_ioctl,
3509	.getgeo = dm_blk_getgeo,
3510	.pr_ops = &dm_pr_ops,
3511	.owner = THIS_MODULE
3512};
3513
3514static const struct dax_operations dm_dax_ops = {
3515	.direct_access = dm_dax_direct_access,
3516	.zero_page_range = dm_dax_zero_page_range,
3517	.recovery_write = dm_dax_recovery_write,
3518};
3519
3520/*
3521 * module hooks
3522 */
3523module_init(dm_init);
3524module_exit(dm_exit);
3525
3526module_param(major, uint, 0);
3527MODULE_PARM_DESC(major, "The major number of the device mapper");
3528
3529module_param(reserved_bio_based_ios, uint, 0644);
3530MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3531
3532module_param(dm_numa_node, int, 0644);
3533MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3534
3535module_param(swap_bios, int, 0644);
3536MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3537
3538MODULE_DESCRIPTION(DM_NAME " driver");
3539MODULE_AUTHOR("Joe Thornber <dm-devel@lists.linux.dev>");
3540MODULE_LICENSE("GPL");