Linux Audio

Check our new training course

Linux kernel drivers training

May 6-19, 2025
Register
Loading...
v5.9
   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-core.h"
   9#include "dm-rq.h"
  10#include "dm-uevent.h"
  11
  12#include <linux/init.h>
  13#include <linux/module.h>
  14#include <linux/mutex.h>
  15#include <linux/sched/mm.h>
  16#include <linux/sched/signal.h>
  17#include <linux/blkpg.h>
  18#include <linux/bio.h>
  19#include <linux/mempool.h>
  20#include <linux/dax.h>
  21#include <linux/slab.h>
  22#include <linux/idr.h>
  23#include <linux/uio.h>
  24#include <linux/hdreg.h>
  25#include <linux/delay.h>
  26#include <linux/wait.h>
  27#include <linux/pr.h>
  28#include <linux/refcount.h>
  29#include <linux/part_stat.h>
  30#include <linux/blk-crypto.h>
  31
  32#define DM_MSG_PREFIX "core"
  33
 
 
 
 
 
 
 
 
 
 
  34/*
  35 * Cookies are numeric values sent with CHANGE and REMOVE
  36 * uevents while resuming, removing or renaming the device.
  37 */
  38#define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
  39#define DM_COOKIE_LENGTH 24
  40
  41static const char *_name = DM_NAME;
  42
  43static unsigned int major = 0;
  44static unsigned int _major = 0;
  45
  46static DEFINE_IDR(_minor_idr);
  47
  48static DEFINE_SPINLOCK(_minor_lock);
  49
  50static void do_deferred_remove(struct work_struct *w);
  51
  52static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
  53
  54static struct workqueue_struct *deferred_remove_workqueue;
  55
  56atomic_t dm_global_event_nr = ATOMIC_INIT(0);
  57DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
  58
  59void dm_issue_global_event(void)
  60{
  61	atomic_inc(&dm_global_event_nr);
  62	wake_up(&dm_global_eventq);
  63}
  64
  65/*
  66 * One of these is allocated (on-stack) per original bio.
  67 */
  68struct clone_info {
  69	struct dm_table *map;
  70	struct bio *bio;
  71	struct dm_io *io;
  72	sector_t sector;
  73	unsigned sector_count;
  74};
  75
  76/*
  77 * One of these is allocated per clone bio.
  78 */
  79#define DM_TIO_MAGIC 7282014
  80struct dm_target_io {
  81	unsigned magic;
  82	struct dm_io *io;
  83	struct dm_target *ti;
  84	unsigned target_bio_nr;
  85	unsigned *len_ptr;
  86	bool inside_dm_io;
  87	struct bio clone;
  88};
  89
  90/*
  91 * One of these is allocated per original bio.
  92 * It contains the first clone used for that original.
  93 */
  94#define DM_IO_MAGIC 5191977
  95struct dm_io {
  96	unsigned magic;
  97	struct mapped_device *md;
  98	blk_status_t status;
  99	atomic_t io_count;
 100	struct bio *orig_bio;
 101	unsigned long start_time;
 102	spinlock_t endio_lock;
 103	struct dm_stats_aux stats_aux;
 104	/* last member of dm_target_io is 'struct bio' */
 105	struct dm_target_io tio;
 106};
 107
 108void *dm_per_bio_data(struct bio *bio, size_t data_size)
 109{
 110	struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
 111	if (!tio->inside_dm_io)
 112		return (char *)bio - offsetof(struct dm_target_io, clone) - data_size;
 113	return (char *)bio - offsetof(struct dm_target_io, clone) - offsetof(struct dm_io, tio) - data_size;
 114}
 115EXPORT_SYMBOL_GPL(dm_per_bio_data);
 116
 117struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
 118{
 119	struct dm_io *io = (struct dm_io *)((char *)data + data_size);
 120	if (io->magic == DM_IO_MAGIC)
 121		return (struct bio *)((char *)io + offsetof(struct dm_io, tio) + offsetof(struct dm_target_io, clone));
 122	BUG_ON(io->magic != DM_TIO_MAGIC);
 123	return (struct bio *)((char *)io + offsetof(struct dm_target_io, clone));
 124}
 125EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
 126
 127unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
 128{
 129	return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
 130}
 131EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
 132
 133#define MINOR_ALLOCED ((void *)-1)
 134
 135/*
 136 * Bits for the md->flags field.
 137 */
 138#define DMF_BLOCK_IO_FOR_SUSPEND 0
 139#define DMF_SUSPENDED 1
 140#define DMF_FROZEN 2
 141#define DMF_FREEING 3
 142#define DMF_DELETING 4
 143#define DMF_NOFLUSH_SUSPENDING 5
 144#define DMF_DEFERRED_REMOVE 6
 145#define DMF_SUSPENDED_INTERNALLY 7
 146#define DMF_POST_SUSPENDING 8
 147
 148#define DM_NUMA_NODE NUMA_NO_NODE
 149static int dm_numa_node = DM_NUMA_NODE;
 150
 151/*
 152 * For mempools pre-allocation at the table loading time.
 153 */
 154struct dm_md_mempools {
 155	struct bio_set bs;
 156	struct bio_set io_bs;
 
 157};
 158
 159struct table_device {
 160	struct list_head list;
 161	refcount_t count;
 162	struct dm_dev dm_dev;
 163};
 164
 
 
 
 
 165/*
 166 * Bio-based DM's mempools' reserved IOs set by the user.
 167 */
 168#define RESERVED_BIO_BASED_IOS		16
 169static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
 170
 171static int __dm_get_module_param_int(int *module_param, int min, int max)
 172{
 173	int param = READ_ONCE(*module_param);
 174	int modified_param = 0;
 175	bool modified = true;
 176
 177	if (param < min)
 178		modified_param = min;
 179	else if (param > max)
 180		modified_param = max;
 181	else
 182		modified = false;
 183
 184	if (modified) {
 185		(void)cmpxchg(module_param, param, modified_param);
 186		param = modified_param;
 187	}
 188
 189	return param;
 190}
 191
 192unsigned __dm_get_module_param(unsigned *module_param,
 193			       unsigned def, unsigned max)
 194{
 195	unsigned param = READ_ONCE(*module_param);
 196	unsigned modified_param = 0;
 197
 198	if (!param)
 199		modified_param = def;
 200	else if (param > max)
 201		modified_param = max;
 202
 203	if (modified_param) {
 204		(void)cmpxchg(module_param, param, modified_param);
 205		param = modified_param;
 206	}
 207
 208	return param;
 209}
 210
 211unsigned dm_get_reserved_bio_based_ios(void)
 212{
 213	return __dm_get_module_param(&reserved_bio_based_ios,
 214				     RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
 215}
 216EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
 217
 218static unsigned dm_get_numa_node(void)
 219{
 220	return __dm_get_module_param_int(&dm_numa_node,
 221					 DM_NUMA_NODE, num_online_nodes() - 1);
 222}
 223
 224static int __init local_init(void)
 225{
 226	int r;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 227
 228	r = dm_uevent_init();
 229	if (r)
 230		return r;
 231
 232	deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
 233	if (!deferred_remove_workqueue) {
 234		r = -ENOMEM;
 235		goto out_uevent_exit;
 236	}
 237
 238	_major = major;
 239	r = register_blkdev(_major, _name);
 240	if (r < 0)
 241		goto out_free_workqueue;
 242
 243	if (!_major)
 244		_major = r;
 245
 246	return 0;
 247
 248out_free_workqueue:
 249	destroy_workqueue(deferred_remove_workqueue);
 250out_uevent_exit:
 251	dm_uevent_exit();
 
 
 
 
 
 
 252
 253	return r;
 254}
 255
 256static void local_exit(void)
 257{
 258	flush_scheduled_work();
 259	destroy_workqueue(deferred_remove_workqueue);
 260
 
 
 
 261	unregister_blkdev(_major, _name);
 262	dm_uevent_exit();
 263
 264	_major = 0;
 265
 266	DMINFO("cleaned up");
 267}
 268
 269static int (*_inits[])(void) __initdata = {
 270	local_init,
 271	dm_target_init,
 272	dm_linear_init,
 273	dm_stripe_init,
 274	dm_io_init,
 275	dm_kcopyd_init,
 276	dm_interface_init,
 277	dm_statistics_init,
 278};
 279
 280static void (*_exits[])(void) = {
 281	local_exit,
 282	dm_target_exit,
 283	dm_linear_exit,
 284	dm_stripe_exit,
 285	dm_io_exit,
 286	dm_kcopyd_exit,
 287	dm_interface_exit,
 288	dm_statistics_exit,
 289};
 290
 291static int __init dm_init(void)
 292{
 293	const int count = ARRAY_SIZE(_inits);
 294
 295	int r, i;
 296
 297	for (i = 0; i < count; i++) {
 298		r = _inits[i]();
 299		if (r)
 300			goto bad;
 301	}
 302
 303	return 0;
 304
 305      bad:
 306	while (i--)
 307		_exits[i]();
 308
 309	return r;
 310}
 311
 312static void __exit dm_exit(void)
 313{
 314	int i = ARRAY_SIZE(_exits);
 315
 316	while (i--)
 317		_exits[i]();
 318
 319	/*
 320	 * Should be empty by this point.
 321	 */
 322	idr_destroy(&_minor_idr);
 323}
 324
 325/*
 326 * Block device functions
 327 */
 328int dm_deleting_md(struct mapped_device *md)
 329{
 330	return test_bit(DMF_DELETING, &md->flags);
 331}
 332
 333static int dm_blk_open(struct block_device *bdev, fmode_t mode)
 334{
 335	struct mapped_device *md;
 336
 337	spin_lock(&_minor_lock);
 338
 339	md = bdev->bd_disk->private_data;
 340	if (!md)
 341		goto out;
 342
 343	if (test_bit(DMF_FREEING, &md->flags) ||
 344	    dm_deleting_md(md)) {
 345		md = NULL;
 346		goto out;
 347	}
 348
 349	dm_get(md);
 350	atomic_inc(&md->open_count);
 351out:
 352	spin_unlock(&_minor_lock);
 353
 354	return md ? 0 : -ENXIO;
 355}
 356
 357static void dm_blk_close(struct gendisk *disk, fmode_t mode)
 358{
 359	struct mapped_device *md;
 360
 361	spin_lock(&_minor_lock);
 362
 363	md = disk->private_data;
 364	if (WARN_ON(!md))
 365		goto out;
 366
 367	if (atomic_dec_and_test(&md->open_count) &&
 368	    (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
 369		queue_work(deferred_remove_workqueue, &deferred_remove_work);
 370
 371	dm_put(md);
 372out:
 373	spin_unlock(&_minor_lock);
 374}
 375
 376int dm_open_count(struct mapped_device *md)
 377{
 378	return atomic_read(&md->open_count);
 379}
 380
 381/*
 382 * Guarantees nothing is using the device before it's deleted.
 383 */
 384int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
 385{
 386	int r = 0;
 387
 388	spin_lock(&_minor_lock);
 389
 390	if (dm_open_count(md)) {
 391		r = -EBUSY;
 392		if (mark_deferred)
 393			set_bit(DMF_DEFERRED_REMOVE, &md->flags);
 394	} else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
 395		r = -EEXIST;
 396	else
 397		set_bit(DMF_DELETING, &md->flags);
 398
 399	spin_unlock(&_minor_lock);
 400
 401	return r;
 402}
 403
 404int dm_cancel_deferred_remove(struct mapped_device *md)
 405{
 406	int r = 0;
 407
 408	spin_lock(&_minor_lock);
 409
 410	if (test_bit(DMF_DELETING, &md->flags))
 411		r = -EBUSY;
 412	else
 413		clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
 414
 415	spin_unlock(&_minor_lock);
 416
 417	return r;
 418}
 419
 420static void do_deferred_remove(struct work_struct *w)
 421{
 422	dm_deferred_remove();
 423}
 424
 425sector_t dm_get_size(struct mapped_device *md)
 426{
 427	return get_capacity(md->disk);
 428}
 429
 430struct request_queue *dm_get_md_queue(struct mapped_device *md)
 431{
 432	return md->queue;
 433}
 434
 435struct dm_stats *dm_get_stats(struct mapped_device *md)
 436{
 437	return &md->stats;
 438}
 439
 440static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 441{
 442	struct mapped_device *md = bdev->bd_disk->private_data;
 443
 444	return dm_get_geometry(md, geo);
 445}
 446
 447#ifdef CONFIG_BLK_DEV_ZONED
 448int dm_report_zones_cb(struct blk_zone *zone, unsigned int idx, void *data)
 449{
 450	struct dm_report_zones_args *args = data;
 451	sector_t sector_diff = args->tgt->begin - args->start;
 452
 453	/*
 454	 * Ignore zones beyond the target range.
 455	 */
 456	if (zone->start >= args->start + args->tgt->len)
 457		return 0;
 458
 459	/*
 460	 * Remap the start sector and write pointer position of the zone
 461	 * to match its position in the target range.
 462	 */
 463	zone->start += sector_diff;
 464	if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
 465		if (zone->cond == BLK_ZONE_COND_FULL)
 466			zone->wp = zone->start + zone->len;
 467		else if (zone->cond == BLK_ZONE_COND_EMPTY)
 468			zone->wp = zone->start;
 469		else
 470			zone->wp += sector_diff;
 471	}
 472
 473	args->next_sector = zone->start + zone->len;
 474	return args->orig_cb(zone, args->zone_idx++, args->orig_data);
 475}
 476EXPORT_SYMBOL_GPL(dm_report_zones_cb);
 477
 478static int dm_blk_report_zones(struct gendisk *disk, sector_t sector,
 479		unsigned int nr_zones, report_zones_cb cb, void *data)
 480{
 481	struct mapped_device *md = disk->private_data;
 482	struct dm_table *map;
 483	int srcu_idx, ret;
 484	struct dm_report_zones_args args = {
 485		.next_sector = sector,
 486		.orig_data = data,
 487		.orig_cb = cb,
 488	};
 489
 490	if (dm_suspended_md(md))
 491		return -EAGAIN;
 492
 493	map = dm_get_live_table(md, &srcu_idx);
 494	if (!map)
 495		return -EIO;
 496
 497	do {
 498		struct dm_target *tgt;
 499
 500		tgt = dm_table_find_target(map, args.next_sector);
 501		if (WARN_ON_ONCE(!tgt->type->report_zones)) {
 502			ret = -EIO;
 503			goto out;
 504		}
 505
 506		args.tgt = tgt;
 507		ret = tgt->type->report_zones(tgt, &args,
 508					      nr_zones - args.zone_idx);
 509		if (ret < 0)
 510			goto out;
 511	} while (args.zone_idx < nr_zones &&
 512		 args.next_sector < get_capacity(disk));
 513
 514	ret = args.zone_idx;
 515out:
 516	dm_put_live_table(md, srcu_idx);
 517	return ret;
 518}
 519#else
 520#define dm_blk_report_zones		NULL
 521#endif /* CONFIG_BLK_DEV_ZONED */
 522
 523static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
 524			    struct block_device **bdev)
 525	__acquires(md->io_barrier)
 526{
 527	struct dm_target *tgt;
 528	struct dm_table *map;
 529	int r;
 530
 531retry:
 532	r = -ENOTTY;
 533	map = dm_get_live_table(md, srcu_idx);
 534	if (!map || !dm_table_get_size(map))
 535		return r;
 536
 537	/* We only support devices that have a single target */
 538	if (dm_table_get_num_targets(map) != 1)
 539		return r;
 540
 541	tgt = dm_table_get_target(map, 0);
 542	if (!tgt->type->prepare_ioctl)
 543		return r;
 544
 545	if (dm_suspended_md(md))
 546		return -EAGAIN;
 
 
 547
 548	r = tgt->type->prepare_ioctl(tgt, bdev);
 
 
 
 
 
 
 
 
 
 549	if (r == -ENOTCONN && !fatal_signal_pending(current)) {
 550		dm_put_live_table(md, *srcu_idx);
 551		msleep(10);
 552		goto retry;
 553	}
 554
 555	return r;
 556}
 557
 558static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
 559	__releases(md->io_barrier)
 560{
 561	dm_put_live_table(md, srcu_idx);
 562}
 563
 564static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
 565			unsigned int cmd, unsigned long arg)
 566{
 567	struct mapped_device *md = bdev->bd_disk->private_data;
 568	int r, srcu_idx;
 569
 570	r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
 571	if (r < 0)
 572		goto out;
 573
 574	if (r > 0) {
 575		/*
 576		 * Target determined this ioctl is being issued against a
 577		 * subset of the parent bdev; require extra privileges.
 
 578		 */
 579		if (!capable(CAP_SYS_RAWIO)) {
 580			DMWARN_LIMIT(
 581	"%s: sending ioctl %x to DM device without required privilege.",
 582				current->comm, cmd);
 583			r = -ENOIOCTLCMD;
 584			goto out;
 585		}
 586	}
 587
 588	r =  __blkdev_driver_ioctl(bdev, mode, cmd, arg);
 589out:
 590	dm_unprepare_ioctl(md, srcu_idx);
 591	return r;
 592}
 593
 594static void start_io_acct(struct dm_io *io);
 595
 596static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
 597{
 598	struct dm_io *io;
 599	struct dm_target_io *tio;
 600	struct bio *clone;
 601
 602	clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
 603	if (!clone)
 604		return NULL;
 605
 606	tio = container_of(clone, struct dm_target_io, clone);
 607	tio->inside_dm_io = true;
 608	tio->io = NULL;
 609
 610	io = container_of(tio, struct dm_io, tio);
 611	io->magic = DM_IO_MAGIC;
 612	io->status = 0;
 613	atomic_set(&io->io_count, 1);
 614	io->orig_bio = bio;
 615	io->md = md;
 616	spin_lock_init(&io->endio_lock);
 617
 618	start_io_acct(io);
 619
 620	return io;
 621}
 622
 623static void free_io(struct mapped_device *md, struct dm_io *io)
 624{
 625	bio_put(&io->tio.clone);
 626}
 627
 628static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
 629				      unsigned target_bio_nr, gfp_t gfp_mask)
 630{
 631	struct dm_target_io *tio;
 632
 633	if (!ci->io->tio.io) {
 634		/* the dm_target_io embedded in ci->io is available */
 635		tio = &ci->io->tio;
 636	} else {
 637		struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
 638		if (!clone)
 639			return NULL;
 640
 641		tio = container_of(clone, struct dm_target_io, clone);
 642		tio->inside_dm_io = false;
 643	}
 644
 645	tio->magic = DM_TIO_MAGIC;
 646	tio->io = ci->io;
 647	tio->ti = ti;
 648	tio->target_bio_nr = target_bio_nr;
 649
 650	return tio;
 651}
 652
 653static void free_tio(struct dm_target_io *tio)
 654{
 655	if (tio->inside_dm_io)
 656		return;
 657	bio_put(&tio->clone);
 658}
 659
 660u64 dm_start_time_ns_from_clone(struct bio *bio)
 661{
 662	struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
 663	struct dm_io *io = tio->io;
 664
 665	return jiffies_to_nsecs(io->start_time);
 666}
 667EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
 668
 669static void start_io_acct(struct dm_io *io)
 670{
 671	struct mapped_device *md = io->md;
 672	struct bio *bio = io->orig_bio;
 
 
 
 
 
 
 
 
 
 
 673
 674	io->start_time = bio_start_io_acct(bio);
 675	if (unlikely(dm_stats_used(&md->stats)))
 676		dm_stats_account_io(&md->stats, bio_data_dir(bio),
 677				    bio->bi_iter.bi_sector, bio_sectors(bio),
 678				    false, 0, &io->stats_aux);
 679}
 680
 681static void end_io_acct(struct dm_io *io)
 682{
 683	struct mapped_device *md = io->md;
 684	struct bio *bio = io->orig_bio;
 685	unsigned long duration = jiffies - io->start_time;
 
 
 686
 687	bio_end_io_acct(bio, io->start_time);
 688
 689	if (unlikely(dm_stats_used(&md->stats)))
 690		dm_stats_account_io(&md->stats, bio_data_dir(bio),
 691				    bio->bi_iter.bi_sector, bio_sectors(bio),
 692				    true, duration, &io->stats_aux);
 693
 
 
 
 
 
 
 
 
 694	/* nudge anyone waiting on suspend queue */
 695	if (unlikely(wq_has_sleeper(&md->wait)))
 696		wake_up(&md->wait);
 697}
 698
 699/*
 700 * Add the bio to the list of deferred io.
 701 */
 702static void queue_io(struct mapped_device *md, struct bio *bio)
 703{
 704	unsigned long flags;
 705
 706	spin_lock_irqsave(&md->deferred_lock, flags);
 707	bio_list_add(&md->deferred, bio);
 708	spin_unlock_irqrestore(&md->deferred_lock, flags);
 709	queue_work(md->wq, &md->work);
 710}
 711
 712/*
 713 * Everyone (including functions in this file), should use this
 714 * function to access the md->map field, and make sure they call
 715 * dm_put_live_table() when finished.
 716 */
 717struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
 718{
 719	*srcu_idx = srcu_read_lock(&md->io_barrier);
 720
 721	return srcu_dereference(md->map, &md->io_barrier);
 722}
 723
 724void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
 725{
 726	srcu_read_unlock(&md->io_barrier, srcu_idx);
 727}
 728
 729void dm_sync_table(struct mapped_device *md)
 730{
 731	synchronize_srcu(&md->io_barrier);
 732	synchronize_rcu_expedited();
 733}
 734
 735/*
 736 * A fast alternative to dm_get_live_table/dm_put_live_table.
 737 * The caller must not block between these two functions.
 738 */
 739static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
 740{
 741	rcu_read_lock();
 742	return rcu_dereference(md->map);
 743}
 744
 745static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
 746{
 747	rcu_read_unlock();
 748}
 749
 750static char *_dm_claim_ptr = "I belong to device-mapper";
 751
 752/*
 753 * Open a table device so we can use it as a map destination.
 754 */
 755static int open_table_device(struct table_device *td, dev_t dev,
 756			     struct mapped_device *md)
 757{
 
 758	struct block_device *bdev;
 759
 760	int r;
 761
 762	BUG_ON(td->dm_dev.bdev);
 763
 764	bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
 765	if (IS_ERR(bdev))
 766		return PTR_ERR(bdev);
 767
 768	r = bd_link_disk_holder(bdev, dm_disk(md));
 769	if (r) {
 770		blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
 771		return r;
 772	}
 773
 774	td->dm_dev.bdev = bdev;
 775	td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
 776	return 0;
 777}
 778
 779/*
 780 * Close a table device that we've been using.
 781 */
 782static void close_table_device(struct table_device *td, struct mapped_device *md)
 783{
 784	if (!td->dm_dev.bdev)
 785		return;
 786
 787	bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
 788	blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
 789	put_dax(td->dm_dev.dax_dev);
 790	td->dm_dev.bdev = NULL;
 791	td->dm_dev.dax_dev = NULL;
 792}
 793
 794static struct table_device *find_table_device(struct list_head *l, dev_t dev,
 795					      fmode_t mode)
 796{
 797	struct table_device *td;
 798
 799	list_for_each_entry(td, l, list)
 800		if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
 801			return td;
 802
 803	return NULL;
 804}
 805
 806int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
 807			struct dm_dev **result)
 808{
 809	int r;
 810	struct table_device *td;
 811
 812	mutex_lock(&md->table_devices_lock);
 813	td = find_table_device(&md->table_devices, dev, mode);
 814	if (!td) {
 815		td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
 816		if (!td) {
 817			mutex_unlock(&md->table_devices_lock);
 818			return -ENOMEM;
 819		}
 820
 821		td->dm_dev.mode = mode;
 822		td->dm_dev.bdev = NULL;
 823
 824		if ((r = open_table_device(td, dev, md))) {
 825			mutex_unlock(&md->table_devices_lock);
 826			kfree(td);
 827			return r;
 828		}
 829
 830		format_dev_t(td->dm_dev.name, dev);
 831
 832		refcount_set(&td->count, 1);
 833		list_add(&td->list, &md->table_devices);
 834	} else {
 835		refcount_inc(&td->count);
 836	}
 
 837	mutex_unlock(&md->table_devices_lock);
 838
 839	*result = &td->dm_dev;
 840	return 0;
 841}
 842EXPORT_SYMBOL_GPL(dm_get_table_device);
 843
 844void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
 845{
 846	struct table_device *td = container_of(d, struct table_device, dm_dev);
 847
 848	mutex_lock(&md->table_devices_lock);
 849	if (refcount_dec_and_test(&td->count)) {
 850		close_table_device(td, md);
 851		list_del(&td->list);
 852		kfree(td);
 853	}
 854	mutex_unlock(&md->table_devices_lock);
 855}
 856EXPORT_SYMBOL(dm_put_table_device);
 857
 858static void free_table_devices(struct list_head *devices)
 859{
 860	struct list_head *tmp, *next;
 861
 862	list_for_each_safe(tmp, next, devices) {
 863		struct table_device *td = list_entry(tmp, struct table_device, list);
 864
 865		DMWARN("dm_destroy: %s still exists with %d references",
 866		       td->dm_dev.name, refcount_read(&td->count));
 867		kfree(td);
 868	}
 869}
 870
 871/*
 872 * Get the geometry associated with a dm device
 873 */
 874int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
 875{
 876	*geo = md->geometry;
 877
 878	return 0;
 879}
 880
 881/*
 882 * Set the geometry of a device.
 883 */
 884int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
 885{
 886	sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
 887
 888	if (geo->start > sz) {
 889		DMWARN("Start sector is beyond the geometry limits.");
 890		return -EINVAL;
 891	}
 892
 893	md->geometry = *geo;
 894
 895	return 0;
 896}
 897
 
 
 
 
 
 
 
 
 
 898static int __noflush_suspending(struct mapped_device *md)
 899{
 900	return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
 901}
 902
 903/*
 904 * Decrements the number of outstanding ios that a bio has been
 905 * cloned into, completing the original io if necc.
 906 */
 907static void dec_pending(struct dm_io *io, blk_status_t error)
 908{
 909	unsigned long flags;
 910	blk_status_t io_error;
 911	struct bio *bio;
 912	struct mapped_device *md = io->md;
 913
 914	/* Push-back supersedes any I/O errors */
 915	if (unlikely(error)) {
 916		spin_lock_irqsave(&io->endio_lock, flags);
 917		if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
 918			io->status = error;
 919		spin_unlock_irqrestore(&io->endio_lock, flags);
 920	}
 921
 922	if (atomic_dec_and_test(&io->io_count)) {
 923		if (io->status == BLK_STS_DM_REQUEUE) {
 924			/*
 925			 * Target requested pushing back the I/O.
 926			 */
 927			spin_lock_irqsave(&md->deferred_lock, flags);
 928			if (__noflush_suspending(md))
 929				/* NOTE early return due to BLK_STS_DM_REQUEUE below */
 930				bio_list_add_head(&md->deferred, io->orig_bio);
 931			else
 932				/* noflush suspend was interrupted. */
 933				io->status = BLK_STS_IOERR;
 934			spin_unlock_irqrestore(&md->deferred_lock, flags);
 935		}
 936
 937		io_error = io->status;
 938		bio = io->orig_bio;
 939		end_io_acct(io);
 940		free_io(md, io);
 941
 942		if (io_error == BLK_STS_DM_REQUEUE)
 943			return;
 944
 945		if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
 946			/*
 947			 * Preflush done for flush with data, reissue
 948			 * without REQ_PREFLUSH.
 949			 */
 950			bio->bi_opf &= ~REQ_PREFLUSH;
 951			queue_io(md, bio);
 952		} else {
 953			/* done with normal IO or empty flush */
 954			if (io_error)
 955				bio->bi_status = io_error;
 956			bio_endio(bio);
 957		}
 958	}
 959}
 960
 961void disable_discard(struct mapped_device *md)
 962{
 963	struct queue_limits *limits = dm_get_queue_limits(md);
 964
 965	/* device doesn't really support DISCARD, disable it */
 966	limits->max_discard_sectors = 0;
 967	blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
 968}
 969
 970void disable_write_same(struct mapped_device *md)
 971{
 972	struct queue_limits *limits = dm_get_queue_limits(md);
 973
 974	/* device doesn't really support WRITE SAME, disable it */
 975	limits->max_write_same_sectors = 0;
 976}
 977
 978void disable_write_zeroes(struct mapped_device *md)
 979{
 980	struct queue_limits *limits = dm_get_queue_limits(md);
 981
 982	/* device doesn't really support WRITE ZEROES, disable it */
 983	limits->max_write_zeroes_sectors = 0;
 984}
 985
 986static void clone_endio(struct bio *bio)
 987{
 988	blk_status_t error = bio->bi_status;
 
 989	struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
 990	struct dm_io *io = tio->io;
 991	struct mapped_device *md = tio->io->md;
 992	dm_endio_fn endio = tio->ti->type->end_io;
 993	struct bio *orig_bio = io->orig_bio;
 994
 995	if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) {
 996		if (bio_op(bio) == REQ_OP_DISCARD &&
 997		    !bio->bi_disk->queue->limits.max_discard_sectors)
 998			disable_discard(md);
 999		else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
1000			 !bio->bi_disk->queue->limits.max_write_same_sectors)
1001			disable_write_same(md);
1002		else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
1003			 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
1004			disable_write_zeroes(md);
1005	}
1006
1007	/*
1008	 * For zone-append bios get offset in zone of the written
1009	 * sector and add that to the original bio sector pos.
1010	 */
1011	if (bio_op(orig_bio) == REQ_OP_ZONE_APPEND) {
1012		sector_t written_sector = bio->bi_iter.bi_sector;
1013		struct request_queue *q = orig_bio->bi_disk->queue;
1014		u64 mask = (u64)blk_queue_zone_sectors(q) - 1;
1015
1016		orig_bio->bi_iter.bi_sector += written_sector & mask;
1017	}
1018
1019	if (endio) {
1020		int r = endio(tio->ti, bio, &error);
1021		switch (r) {
1022		case DM_ENDIO_REQUEUE:
1023			error = BLK_STS_DM_REQUEUE;
1024			fallthrough;
1025		case DM_ENDIO_DONE:
1026			break;
1027		case DM_ENDIO_INCOMPLETE:
1028			/* The target will handle the io */
1029			return;
1030		default:
1031			DMWARN("unimplemented target endio return value: %d", r);
1032			BUG();
1033		}
1034	}
1035
 
 
 
 
1036	free_tio(tio);
1037	dec_pending(io, error);
1038}
1039
1040/*
1041 * Return maximum size of I/O possible at the supplied sector up to the current
1042 * target boundary.
1043 */
1044static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1045{
1046	sector_t target_offset = dm_target_offset(ti, sector);
1047
1048	return ti->len - target_offset;
1049}
1050
1051static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1052{
1053	sector_t len = max_io_len_target_boundary(sector, ti);
1054	sector_t offset, max_len;
1055
1056	/*
1057	 * Does the target need to split even further?
1058	 */
1059	if (ti->max_io_len) {
1060		offset = dm_target_offset(ti, sector);
1061		if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1062			max_len = sector_div(offset, ti->max_io_len);
1063		else
1064			max_len = offset & (ti->max_io_len - 1);
1065		max_len = ti->max_io_len - max_len;
1066
1067		if (len > max_len)
1068			len = max_len;
1069	}
1070
1071	return len;
1072}
1073
1074int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1075{
1076	if (len > UINT_MAX) {
1077		DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1078		      (unsigned long long)len, UINT_MAX);
1079		ti->error = "Maximum size of target IO is too large";
1080		return -EINVAL;
1081	}
1082
1083	ti->max_io_len = (uint32_t) len;
1084
1085	return 0;
1086}
1087EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1088
1089static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1090						sector_t sector, int *srcu_idx)
1091	__acquires(md->io_barrier)
1092{
 
1093	struct dm_table *map;
1094	struct dm_target *ti;
1095
1096	map = dm_get_live_table(md, srcu_idx);
1097	if (!map)
1098		return NULL;
1099
1100	ti = dm_table_find_target(map, sector);
1101	if (!ti)
1102		return NULL;
1103
1104	return ti;
1105}
1106
1107static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1108				 long nr_pages, void **kaddr, pfn_t *pfn)
1109{
1110	struct mapped_device *md = dax_get_private(dax_dev);
1111	sector_t sector = pgoff * PAGE_SECTORS;
1112	struct dm_target *ti;
1113	long len, ret = -EIO;
1114	int srcu_idx;
1115
1116	ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1117
1118	if (!ti)
1119		goto out;
1120	if (!ti->type->direct_access)
1121		goto out;
1122	len = max_io_len(sector, ti) / PAGE_SECTORS;
1123	if (len < 1)
1124		goto out;
1125	nr_pages = min(len, nr_pages);
1126	ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1127
1128 out:
1129	dm_put_live_table(md, srcu_idx);
1130
1131	return ret;
1132}
1133
1134static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
1135		int blocksize, sector_t start, sector_t len)
1136{
1137	struct mapped_device *md = dax_get_private(dax_dev);
1138	struct dm_table *map;
1139	bool ret = false;
1140	int srcu_idx;
 
1141
1142	map = dm_get_live_table(md, &srcu_idx);
1143	if (!map)
1144		goto out;
1145
1146	ret = dm_table_supports_dax(map, device_supports_dax, &blocksize);
1147
1148out:
1149	dm_put_live_table(md, srcu_idx);
1150
1151	return ret;
1152}
1153
1154static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1155				    void *addr, size_t bytes, struct iov_iter *i)
1156{
1157	struct mapped_device *md = dax_get_private(dax_dev);
1158	sector_t sector = pgoff * PAGE_SECTORS;
1159	struct dm_target *ti;
1160	long ret = 0;
1161	int srcu_idx;
1162
1163	ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1164
1165	if (!ti)
1166		goto out;
1167	if (!ti->type->dax_copy_from_iter) {
1168		ret = copy_from_iter(addr, bytes, i);
1169		goto out;
1170	}
1171	ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1172 out:
1173	dm_put_live_table(md, srcu_idx);
1174
1175	return ret;
1176}
1177
1178static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1179		void *addr, size_t bytes, struct iov_iter *i)
1180{
1181	struct mapped_device *md = dax_get_private(dax_dev);
1182	sector_t sector = pgoff * PAGE_SECTORS;
1183	struct dm_target *ti;
1184	long ret = 0;
1185	int srcu_idx;
1186
1187	ti = dm_dax_get_live_target(md, sector, &srcu_idx);
 
1188
1189	if (!ti)
1190		goto out;
1191	if (!ti->type->dax_copy_to_iter) {
1192		ret = copy_to_iter(addr, bytes, i);
1193		goto out;
1194	}
1195	ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1196 out:
1197	dm_put_live_table(md, srcu_idx);
1198
1199	return ret;
1200}
1201
1202static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1203				  size_t nr_pages)
1204{
1205	struct mapped_device *md = dax_get_private(dax_dev);
1206	sector_t sector = pgoff * PAGE_SECTORS;
1207	struct dm_target *ti;
1208	int ret = -EIO;
1209	int srcu_idx;
1210
1211	ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1212
1213	if (!ti)
1214		goto out;
1215	if (WARN_ON(!ti->type->dax_zero_page_range)) {
1216		/*
1217		 * ->zero_page_range() is mandatory dax operation. If we are
1218		 *  here, something is wrong.
1219		 */
1220		dm_put_live_table(md, srcu_idx);
1221		goto out;
1222	}
1223	ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1224
1225 out:
1226	dm_put_live_table(md, srcu_idx);
1227
1228	return ret;
1229}
1230
1231/*
1232 * A target may call dm_accept_partial_bio only from the map routine.  It is
1233 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_RESET,
1234 * REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE and REQ_OP_ZONE_FINISH.
1235 *
1236 * dm_accept_partial_bio informs the dm that the target only wants to process
1237 * additional n_sectors sectors of the bio and the rest of the data should be
1238 * sent in a next bio.
1239 *
1240 * A diagram that explains the arithmetics:
1241 * +--------------------+---------------+-------+
1242 * |         1          |       2       |   3   |
1243 * +--------------------+---------------+-------+
1244 *
1245 * <-------------- *tio->len_ptr --------------->
1246 *                      <------- bi_size ------->
1247 *                      <-- n_sectors -->
1248 *
1249 * Region 1 was already iterated over with bio_advance or similar function.
1250 *	(it may be empty if the target doesn't use bio_advance)
1251 * Region 2 is the remaining bio size that the target wants to process.
1252 *	(it may be empty if region 1 is non-empty, although there is no reason
1253 *	 to make it empty)
1254 * The target requires that region 3 is to be sent in the next bio.
1255 *
1256 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1257 * the partially processed part (the sum of regions 1+2) must be the same for all
1258 * copies of the bio.
1259 */
1260void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1261{
1262	struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1263	unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1264	BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1265	BUG_ON(bi_size > *tio->len_ptr);
1266	BUG_ON(n_sectors > bi_size);
1267	*tio->len_ptr -= bi_size - n_sectors;
1268	bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1269}
1270EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1271
1272static blk_qc_t __map_bio(struct dm_target_io *tio)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1273{
1274	int r;
1275	sector_t sector;
 
1276	struct bio *clone = &tio->clone;
1277	struct dm_io *io = tio->io;
1278	struct dm_target *ti = tio->ti;
1279	blk_qc_t ret = BLK_QC_T_NONE;
1280
1281	clone->bi_end_io = clone_endio;
1282
1283	/*
1284	 * Map the clone.  If r == 0 we don't need to do
1285	 * anything, the target has assumed ownership of
1286	 * this io.
1287	 */
1288	atomic_inc(&io->io_count);
1289	sector = clone->bi_iter.bi_sector;
1290
 
1291	r = ti->type->map(ti, clone);
1292	switch (r) {
1293	case DM_MAPIO_SUBMITTED:
1294		break;
1295	case DM_MAPIO_REMAPPED:
1296		/* the bio has been remapped so dispatch it */
1297		trace_block_bio_remap(clone->bi_disk->queue, clone,
1298				      bio_dev(io->orig_bio), sector);
1299		ret = submit_bio_noacct(clone);
1300		break;
1301	case DM_MAPIO_KILL:
1302		free_tio(tio);
1303		dec_pending(io, BLK_STS_IOERR);
1304		break;
1305	case DM_MAPIO_REQUEUE:
1306		free_tio(tio);
1307		dec_pending(io, BLK_STS_DM_REQUEUE);
1308		break;
1309	default:
1310		DMWARN("unimplemented target map return value: %d", r);
1311		BUG();
1312	}
1313
1314	return ret;
1315}
1316
 
 
 
 
 
 
 
 
 
1317static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1318{
1319	bio->bi_iter.bi_sector = sector;
1320	bio->bi_iter.bi_size = to_bytes(len);
1321}
1322
1323/*
1324 * Creates a bio that consists of range of complete bvecs.
1325 */
1326static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1327		     sector_t sector, unsigned len)
1328{
1329	struct bio *clone = &tio->clone;
1330
1331	__bio_clone_fast(clone, bio);
1332
1333	bio_crypt_clone(clone, bio, GFP_NOIO);
1334
1335	if (bio_integrity(bio)) {
1336		int r;
1337
1338		if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1339			     !dm_target_passes_integrity(tio->ti->type))) {
1340			DMWARN("%s: the target %s doesn't support integrity data.",
1341				dm_device_name(tio->io->md),
1342				tio->ti->type->name);
1343			return -EIO;
1344		}
1345
1346		r = bio_integrity_clone(clone, bio, GFP_NOIO);
1347		if (r < 0)
1348			return r;
1349	}
1350
1351	bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1352	clone->bi_iter.bi_size = to_bytes(len);
1353
1354	if (bio_integrity(bio))
1355		bio_integrity_trim(clone);
1356
1357	return 0;
1358}
1359
1360static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1361				struct dm_target *ti, unsigned num_bios)
 
1362{
1363	struct dm_target_io *tio;
1364	int try;
1365
1366	if (!num_bios)
1367		return;
1368
1369	if (num_bios == 1) {
1370		tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1371		bio_list_add(blist, &tio->clone);
1372		return;
1373	}
1374
1375	for (try = 0; try < 2; try++) {
1376		int bio_nr;
1377		struct bio *bio;
1378
1379		if (try)
1380			mutex_lock(&ci->io->md->table_devices_lock);
1381		for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1382			tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1383			if (!tio)
1384				break;
1385
1386			bio_list_add(blist, &tio->clone);
1387		}
1388		if (try)
1389			mutex_unlock(&ci->io->md->table_devices_lock);
1390		if (bio_nr == num_bios)
1391			return;
1392
1393		while ((bio = bio_list_pop(blist))) {
1394			tio = container_of(bio, struct dm_target_io, clone);
1395			free_tio(tio);
1396		}
1397	}
1398}
1399
1400static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1401					   struct dm_target_io *tio, unsigned *len)
 
1402{
 
1403	struct bio *clone = &tio->clone;
1404
1405	tio->len_ptr = len;
1406
1407	__bio_clone_fast(clone, ci->bio);
1408	if (len)
1409		bio_setup_sector(clone, ci->sector, *len);
1410
1411	return __map_bio(tio);
1412}
1413
1414static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1415				  unsigned num_bios, unsigned *len)
1416{
1417	struct bio_list blist = BIO_EMPTY_LIST;
1418	struct bio *bio;
1419	struct dm_target_io *tio;
1420
1421	alloc_multiple_bios(&blist, ci, ti, num_bios);
1422
1423	while ((bio = bio_list_pop(&blist))) {
1424		tio = container_of(bio, struct dm_target_io, clone);
1425		(void) __clone_and_map_simple_bio(ci, tio, len);
1426	}
1427}
1428
1429static int __send_empty_flush(struct clone_info *ci)
1430{
1431	unsigned target_nr = 0;
1432	struct dm_target *ti;
1433
1434	/*
1435	 * Empty flush uses a statically initialized bio, as the base for
1436	 * cloning.  However, blkg association requires that a bdev is
1437	 * associated with a gendisk, which doesn't happen until the bdev is
1438	 * opened.  So, blkg association is done at issue time of the flush
1439	 * rather than when the device is created in alloc_dev().
1440	 */
1441	bio_set_dev(ci->bio, ci->io->md->bdev);
1442
1443	BUG_ON(bio_has_data(ci->bio));
1444	while ((ti = dm_table_get_target(ci->map, target_nr++)))
1445		__send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
 
1446	return 0;
1447}
1448
1449static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1450				    sector_t sector, unsigned *len)
1451{
1452	struct bio *bio = ci->bio;
1453	struct dm_target_io *tio;
1454	int r;
 
 
1455
1456	tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1457	tio->len_ptr = len;
1458	r = clone_bio(tio, bio, sector, *len);
1459	if (r < 0) {
1460		free_tio(tio);
1461		return r;
 
 
 
 
 
 
 
 
 
1462	}
1463	(void) __map_bio(tio);
1464
1465	return 0;
1466}
1467
1468typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1469
1470static unsigned get_num_discard_bios(struct dm_target *ti)
1471{
1472	return ti->num_discard_bios;
1473}
1474
1475static unsigned get_num_secure_erase_bios(struct dm_target *ti)
1476{
1477	return ti->num_secure_erase_bios;
1478}
1479
1480static unsigned get_num_write_same_bios(struct dm_target *ti)
1481{
1482	return ti->num_write_same_bios;
1483}
1484
1485static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
 
 
1486{
1487	return ti->num_write_zeroes_bios;
1488}
1489
1490static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1491				       unsigned num_bios)
 
1492{
 
1493	unsigned len;
 
1494
1495	/*
1496	 * Even though the device advertised support for this type of
1497	 * request, that does not mean every target supports it, and
1498	 * reconfiguration might also have changed that since the
1499	 * check was performed.
1500	 */
1501	if (!num_bios)
1502		return -EOPNOTSUPP;
1503
1504	len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1505
1506	__send_duplicate_bios(ci, ti, num_bios, &len);
1507
1508	ci->sector += len;
1509	ci->sector_count -= len;
1510
1511	return 0;
1512}
 
 
 
 
 
 
 
1513
1514static int __send_discard(struct clone_info *ci, struct dm_target *ti)
1515{
1516	return __send_changing_extent_only(ci, ti, get_num_discard_bios(ti));
1517}
1518
1519static int __send_secure_erase(struct clone_info *ci, struct dm_target *ti)
1520{
1521	return __send_changing_extent_only(ci, ti, get_num_secure_erase_bios(ti));
1522}
1523
1524static int __send_write_same(struct clone_info *ci, struct dm_target *ti)
1525{
1526	return __send_changing_extent_only(ci, ti, get_num_write_same_bios(ti));
1527}
1528
1529static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti)
1530{
1531	return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios(ti));
1532}
1533
1534static bool is_abnormal_io(struct bio *bio)
1535{
1536	bool r = false;
1537
1538	switch (bio_op(bio)) {
1539	case REQ_OP_DISCARD:
1540	case REQ_OP_SECURE_ERASE:
1541	case REQ_OP_WRITE_SAME:
1542	case REQ_OP_WRITE_ZEROES:
1543		r = true;
1544		break;
1545	}
1546
1547	return r;
1548}
1549
1550static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1551				  int *result)
1552{
1553	struct bio *bio = ci->bio;
1554
1555	if (bio_op(bio) == REQ_OP_DISCARD)
1556		*result = __send_discard(ci, ti);
1557	else if (bio_op(bio) == REQ_OP_SECURE_ERASE)
1558		*result = __send_secure_erase(ci, ti);
1559	else if (bio_op(bio) == REQ_OP_WRITE_SAME)
1560		*result = __send_write_same(ci, ti);
1561	else if (bio_op(bio) == REQ_OP_WRITE_ZEROES)
1562		*result = __send_write_zeroes(ci, ti);
1563	else
1564		return false;
1565
1566	return true;
1567}
1568
1569/*
1570 * Select the correct strategy for processing a non-flush bio.
1571 */
1572static int __split_and_process_non_flush(struct clone_info *ci)
1573{
 
1574	struct dm_target *ti;
1575	unsigned len;
1576	int r;
1577
 
 
 
 
 
1578	ti = dm_table_find_target(ci->map, ci->sector);
1579	if (!ti)
1580		return -EIO;
1581
1582	if (__process_abnormal_io(ci, ti, &r))
1583		return r;
1584
1585	len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1586
1587	r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1588	if (r < 0)
1589		return r;
1590
1591	ci->sector += len;
1592	ci->sector_count -= len;
1593
1594	return 0;
1595}
1596
1597static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1598			    struct dm_table *map, struct bio *bio)
1599{
1600	ci->map = map;
1601	ci->io = alloc_io(md, bio);
1602	ci->sector = bio->bi_iter.bi_sector;
1603}
1604
1605#define __dm_part_stat_sub(part, field, subnd)	\
1606	(part_stat_get(part, field) -= (subnd))
1607
1608/*
1609 * Entry point to split a bio into clones and submit them to the targets.
1610 */
1611static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1612					struct dm_table *map, struct bio *bio)
1613{
1614	struct clone_info ci;
1615	blk_qc_t ret = BLK_QC_T_NONE;
1616	int error = 0;
1617
1618	init_clone_info(&ci, md, map, bio);
 
 
 
1619
1620	if (bio->bi_opf & REQ_PREFLUSH) {
1621		struct bio flush_bio;
 
 
 
 
 
 
 
1622
1623		/*
1624		 * Use an on-stack bio for this, it's safe since we don't
1625		 * need to reference it after submit. It's just used as
1626		 * the basis for the clone(s).
1627		 */
1628		bio_init(&flush_bio, NULL, 0);
1629		flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1630		ci.bio = &flush_bio;
1631		ci.sector_count = 0;
1632		error = __send_empty_flush(&ci);
1633		bio_uninit(ci.bio);
1634		/* dec_pending submits any data associated with flush */
1635	} else if (op_is_zone_mgmt(bio_op(bio))) {
1636		ci.bio = bio;
1637		ci.sector_count = 0;
1638		error = __split_and_process_non_flush(&ci);
1639	} else {
1640		ci.bio = bio;
1641		ci.sector_count = bio_sectors(bio);
1642		while (ci.sector_count && !error) {
1643			error = __split_and_process_non_flush(&ci);
1644			if (current->bio_list && ci.sector_count && !error) {
1645				/*
1646				 * Remainder must be passed to submit_bio_noacct()
1647				 * so that it gets handled *after* bios already submitted
1648				 * have been completely processed.
1649				 * We take a clone of the original to store in
1650				 * ci.io->orig_bio to be used by end_io_acct() and
1651				 * for dec_pending to use for completion handling.
1652				 */
1653				struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1654							  GFP_NOIO, &md->queue->bio_split);
1655				ci.io->orig_bio = b;
1656
1657				/*
1658				 * Adjust IO stats for each split, otherwise upon queue
1659				 * reentry there will be redundant IO accounting.
1660				 * NOTE: this is a stop-gap fix, a proper fix involves
1661				 * significant refactoring of DM core's bio splitting
1662				 * (by eliminating DM's splitting and just using bio_split)
1663				 */
1664				part_stat_lock();
1665				__dm_part_stat_sub(&dm_disk(md)->part0,
1666						   sectors[op_stat_group(bio_op(bio))], ci.sector_count);
1667				part_stat_unlock();
1668
1669				bio_chain(b, bio);
1670				trace_block_split(md->queue, b, bio->bi_iter.bi_sector);
1671				ret = submit_bio_noacct(bio);
1672				break;
1673			}
1674		}
1675	}
1676
1677	/* drop the extra reference count */
1678	dec_pending(ci.io, errno_to_blk_status(error));
1679	return ret;
1680}
 
 
 
1681
1682/*
1683 * Optimized variant of __split_and_process_bio that leverages the
1684 * fact that targets that use it do _not_ have a need to split bios.
1685 */
1686static blk_qc_t __process_bio(struct mapped_device *md, struct dm_table *map,
1687			      struct bio *bio, struct dm_target *ti)
1688{
1689	struct clone_info ci;
1690	blk_qc_t ret = BLK_QC_T_NONE;
1691	int error = 0;
1692
1693	init_clone_info(&ci, md, map, bio);
1694
1695	if (bio->bi_opf & REQ_PREFLUSH) {
1696		struct bio flush_bio;
1697
1698		/*
1699		 * Use an on-stack bio for this, it's safe since we don't
1700		 * need to reference it after submit. It's just used as
1701		 * the basis for the clone(s).
1702		 */
1703		bio_init(&flush_bio, NULL, 0);
1704		flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1705		ci.bio = &flush_bio;
1706		ci.sector_count = 0;
1707		error = __send_empty_flush(&ci);
1708		bio_uninit(ci.bio);
1709		/* dec_pending submits any data associated with flush */
1710	} else {
1711		struct dm_target_io *tio;
1712
1713		ci.bio = bio;
1714		ci.sector_count = bio_sectors(bio);
1715		if (__process_abnormal_io(&ci, ti, &error))
1716			goto out;
1717
1718		tio = alloc_tio(&ci, ti, 0, GFP_NOIO);
1719		ret = __clone_and_map_simple_bio(&ci, tio, NULL);
1720	}
1721out:
1722	/* drop the extra reference count */
1723	dec_pending(ci.io, errno_to_blk_status(error));
1724	return ret;
1725}
1726
1727static blk_qc_t dm_process_bio(struct mapped_device *md,
1728			       struct dm_table *map, struct bio *bio)
1729{
1730	blk_qc_t ret = BLK_QC_T_NONE;
1731	struct dm_target *ti = md->immutable_target;
1732
1733	if (unlikely(!map)) {
1734		bio_io_error(bio);
1735		return ret;
1736	}
1737
1738	if (!ti) {
1739		ti = dm_table_find_target(map, bio->bi_iter.bi_sector);
1740		if (unlikely(!ti)) {
1741			bio_io_error(bio);
1742			return ret;
1743		}
1744	}
1745
1746	/*
1747	 * If in ->submit_bio we need to use blk_queue_split(), otherwise
1748	 * queue_limits for abnormal requests (e.g. discard, writesame, etc)
1749	 * won't be imposed.
1750	 * If called from dm_wq_work() for deferred bio processing, bio
1751	 * was already handled by following code with previous ->submit_bio.
1752	 */
1753	if (current->bio_list) {
1754		if (is_abnormal_io(bio))
1755			blk_queue_split(&bio);
1756		/* regular IO is split by __split_and_process_bio */
1757	}
1758
1759	if (dm_get_md_type(md) == DM_TYPE_NVME_BIO_BASED)
1760		return __process_bio(md, map, bio, ti);
1761	return __split_and_process_bio(md, map, bio);
1762}
1763
1764static blk_qc_t dm_submit_bio(struct bio *bio)
1765{
1766	struct mapped_device *md = bio->bi_disk->private_data;
1767	blk_qc_t ret = BLK_QC_T_NONE;
1768	int srcu_idx;
1769	struct dm_table *map;
1770
1771	if (dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) {
1772		/*
1773		 * We are called with a live reference on q_usage_counter, but
1774		 * that one will be released as soon as we return.  Grab an
1775		 * extra one as blk_mq_submit_bio expects to be able to consume
1776		 * a reference (which lives until the request is freed in case a
1777		 * request is allocated).
1778		 */
1779		percpu_ref_get(&bio->bi_disk->queue->q_usage_counter);
1780		return blk_mq_submit_bio(bio);
1781	}
1782
1783	map = dm_get_live_table(md, &srcu_idx);
1784
 
 
1785	/* if we're suspended, we have to queue this io for later */
1786	if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1787		dm_put_live_table(md, srcu_idx);
1788
1789		if (!(bio->bi_opf & REQ_RAHEAD))
1790			queue_io(md, bio);
1791		else
1792			bio_io_error(bio);
1793		return ret;
1794	}
1795
1796	ret = dm_process_bio(md, map, bio);
1797
1798	dm_put_live_table(md, srcu_idx);
1799	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1800}
1801
1802/*-----------------------------------------------------------------
1803 * An IDR is used to keep track of allocated minor numbers.
1804 *---------------------------------------------------------------*/
1805static void free_minor(int minor)
1806{
1807	spin_lock(&_minor_lock);
1808	idr_remove(&_minor_idr, minor);
1809	spin_unlock(&_minor_lock);
1810}
1811
1812/*
1813 * See if the device with a specific minor # is free.
1814 */
1815static int specific_minor(int minor)
1816{
1817	int r;
1818
1819	if (minor >= (1 << MINORBITS))
1820		return -EINVAL;
1821
1822	idr_preload(GFP_KERNEL);
1823	spin_lock(&_minor_lock);
1824
1825	r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1826
1827	spin_unlock(&_minor_lock);
1828	idr_preload_end();
1829	if (r < 0)
1830		return r == -ENOSPC ? -EBUSY : r;
1831	return 0;
1832}
1833
1834static int next_free_minor(int *minor)
1835{
1836	int r;
1837
1838	idr_preload(GFP_KERNEL);
1839	spin_lock(&_minor_lock);
1840
1841	r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1842
1843	spin_unlock(&_minor_lock);
1844	idr_preload_end();
1845	if (r < 0)
1846		return r;
1847	*minor = r;
1848	return 0;
1849}
1850
1851static const struct block_device_operations dm_blk_dops;
1852static const struct dax_operations dm_dax_ops;
1853
1854static void dm_wq_work(struct work_struct *work);
1855
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1856static void cleanup_mapped_device(struct mapped_device *md)
1857{
1858	if (md->wq)
1859		destroy_workqueue(md->wq);
1860	bioset_exit(&md->bs);
1861	bioset_exit(&md->io_bs);
1862
1863	if (md->dax_dev) {
1864		kill_dax(md->dax_dev);
1865		put_dax(md->dax_dev);
1866		md->dax_dev = NULL;
1867	}
1868
1869	if (md->disk) {
1870		spin_lock(&_minor_lock);
1871		md->disk->private_data = NULL;
1872		spin_unlock(&_minor_lock);
1873		del_gendisk(md->disk);
1874		put_disk(md->disk);
1875	}
1876
1877	if (md->queue)
1878		blk_cleanup_queue(md->queue);
1879
1880	cleanup_srcu_struct(&md->io_barrier);
1881
1882	if (md->bdev) {
1883		bdput(md->bdev);
1884		md->bdev = NULL;
1885	}
1886
1887	mutex_destroy(&md->suspend_lock);
1888	mutex_destroy(&md->type_lock);
1889	mutex_destroy(&md->table_devices_lock);
1890
1891	dm_mq_cleanup_mapped_device(md);
1892}
1893
1894/*
1895 * Allocate and initialise a blank device with a given minor.
1896 */
1897static struct mapped_device *alloc_dev(int minor)
1898{
1899	int r, numa_node_id = dm_get_numa_node();
1900	struct mapped_device *md;
1901	void *old_md;
1902
1903	md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1904	if (!md) {
1905		DMWARN("unable to allocate device, out of memory.");
1906		return NULL;
1907	}
1908
1909	if (!try_module_get(THIS_MODULE))
1910		goto bad_module_get;
1911
1912	/* get a minor number for the dev */
1913	if (minor == DM_ANY_MINOR)
1914		r = next_free_minor(&minor);
1915	else
1916		r = specific_minor(minor);
1917	if (r < 0)
1918		goto bad_minor;
1919
1920	r = init_srcu_struct(&md->io_barrier);
1921	if (r < 0)
1922		goto bad_io_barrier;
1923
1924	md->numa_node_id = numa_node_id;
 
1925	md->init_tio_pdu = false;
1926	md->type = DM_TYPE_NONE;
1927	mutex_init(&md->suspend_lock);
1928	mutex_init(&md->type_lock);
1929	mutex_init(&md->table_devices_lock);
1930	spin_lock_init(&md->deferred_lock);
1931	atomic_set(&md->holders, 1);
1932	atomic_set(&md->open_count, 0);
1933	atomic_set(&md->event_nr, 0);
1934	atomic_set(&md->uevent_seq, 0);
1935	INIT_LIST_HEAD(&md->uevent_list);
1936	INIT_LIST_HEAD(&md->table_devices);
1937	spin_lock_init(&md->uevent_lock);
1938
1939	/*
1940	 * default to bio-based until DM table is loaded and md->type
1941	 * established. If request-based table is loaded: blk-mq will
1942	 * override accordingly.
1943	 */
1944	md->queue = blk_alloc_queue(numa_node_id);
1945	if (!md->queue)
1946		goto bad;
1947
1948	md->disk = alloc_disk_node(1, md->numa_node_id);
 
 
1949	if (!md->disk)
1950		goto bad;
1951
 
 
1952	init_waitqueue_head(&md->wait);
1953	INIT_WORK(&md->work, dm_wq_work);
1954	init_waitqueue_head(&md->eventq);
1955	init_completion(&md->kobj_holder.completion);
 
1956
1957	md->disk->major = _major;
1958	md->disk->first_minor = minor;
1959	md->disk->fops = &dm_blk_dops;
1960	md->disk->queue = md->queue;
1961	md->disk->private_data = md;
1962	sprintf(md->disk->disk_name, "dm-%d", minor);
1963
1964	if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1965		md->dax_dev = alloc_dax(md, md->disk->disk_name,
1966					&dm_dax_ops, 0);
1967		if (IS_ERR(md->dax_dev))
1968			goto bad;
1969	}
1970
1971	add_disk_no_queue_reg(md->disk);
1972	format_dev_t(md->name, MKDEV(_major, minor));
1973
1974	md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1975	if (!md->wq)
1976		goto bad;
1977
1978	md->bdev = bdget_disk(md->disk, 0);
1979	if (!md->bdev)
1980		goto bad;
1981
 
 
 
 
1982	dm_stats_init(&md->stats);
1983
1984	/* Populate the mapping, nobody knows we exist yet */
1985	spin_lock(&_minor_lock);
1986	old_md = idr_replace(&_minor_idr, md, minor);
1987	spin_unlock(&_minor_lock);
1988
1989	BUG_ON(old_md != MINOR_ALLOCED);
1990
1991	return md;
1992
1993bad:
1994	cleanup_mapped_device(md);
1995bad_io_barrier:
1996	free_minor(minor);
1997bad_minor:
1998	module_put(THIS_MODULE);
1999bad_module_get:
2000	kvfree(md);
2001	return NULL;
2002}
2003
2004static void unlock_fs(struct mapped_device *md);
2005
2006static void free_dev(struct mapped_device *md)
2007{
2008	int minor = MINOR(disk_devt(md->disk));
2009
2010	unlock_fs(md);
2011
2012	cleanup_mapped_device(md);
2013
2014	free_table_devices(&md->table_devices);
2015	dm_stats_cleanup(&md->stats);
2016	free_minor(minor);
2017
2018	module_put(THIS_MODULE);
2019	kvfree(md);
2020}
2021
2022static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
2023{
2024	struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2025	int ret = 0;
2026
2027	if (dm_table_bio_based(t)) {
2028		/*
2029		 * The md may already have mempools that need changing.
2030		 * If so, reload bioset because front_pad may have changed
2031		 * because a different table was loaded.
2032		 */
2033		bioset_exit(&md->bs);
2034		bioset_exit(&md->io_bs);
2035
2036	} else if (bioset_initialized(&md->bs)) {
 
2037		/*
2038		 * There's no need to reload with request-based dm
2039		 * because the size of front_pad doesn't change.
2040		 * Note for future: If you are to reload bioset,
2041		 * prep-ed requests in the queue may refer
2042		 * to bio from the old bioset, so you must walk
2043		 * through the queue to unprep.
2044		 */
2045		goto out;
2046	}
2047
2048	BUG_ON(!p ||
2049	       bioset_initialized(&md->bs) ||
2050	       bioset_initialized(&md->io_bs));
 
 
 
 
 
2051
2052	ret = bioset_init_from_src(&md->bs, &p->bs);
2053	if (ret)
2054		goto out;
2055	ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
2056	if (ret)
2057		bioset_exit(&md->bs);
2058out:
2059	/* mempool bind completed, no longer need any mempools in the table */
2060	dm_table_free_md_mempools(t);
2061	return ret;
2062}
2063
2064/*
2065 * Bind a table to the device.
2066 */
2067static void event_callback(void *context)
2068{
2069	unsigned long flags;
2070	LIST_HEAD(uevents);
2071	struct mapped_device *md = (struct mapped_device *) context;
2072
2073	spin_lock_irqsave(&md->uevent_lock, flags);
2074	list_splice_init(&md->uevent_list, &uevents);
2075	spin_unlock_irqrestore(&md->uevent_lock, flags);
2076
2077	dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2078
2079	atomic_inc(&md->event_nr);
2080	wake_up(&md->eventq);
2081	dm_issue_global_event();
2082}
2083
2084/*
2085 * Protected by md->suspend_lock obtained by dm_swap_table().
2086 */
2087static void __set_size(struct mapped_device *md, sector_t size)
2088{
2089	lockdep_assert_held(&md->suspend_lock);
2090
2091	set_capacity(md->disk, size);
2092
2093	i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2094}
2095
2096/*
2097 * Returns old map, which caller must destroy.
2098 */
2099static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2100			       struct queue_limits *limits)
2101{
2102	struct dm_table *old_map;
2103	struct request_queue *q = md->queue;
2104	bool request_based = dm_table_request_based(t);
2105	sector_t size;
2106	int ret;
2107
2108	lockdep_assert_held(&md->suspend_lock);
2109
2110	size = dm_table_get_size(t);
2111
2112	/*
2113	 * Wipe any geometry if the size of the table changed.
2114	 */
2115	if (size != dm_get_size(md))
2116		memset(&md->geometry, 0, sizeof(md->geometry));
2117
2118	__set_size(md, size);
2119
2120	dm_table_event_callback(t, event_callback, md);
2121
2122	/*
2123	 * The queue hasn't been stopped yet, if the old table type wasn't
2124	 * for request-based during suspension.  So stop it to prevent
2125	 * I/O mapping before resume.
2126	 * This must be done before setting the queue restrictions,
2127	 * because request-based dm may be run just after the setting.
2128	 */
2129	if (request_based)
2130		dm_stop_queue(q);
2131
2132	if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) {
2133		/*
2134		 * Leverage the fact that request-based DM targets and
2135		 * NVMe bio based targets are immutable singletons
2136		 * - used to optimize both dm_request_fn and dm_mq_queue_rq;
2137		 *   and __process_bio.
2138		 */
2139		md->immutable_target = dm_table_get_immutable_target(t);
2140	}
2141
2142	ret = __bind_mempools(md, t);
2143	if (ret) {
2144		old_map = ERR_PTR(ret);
2145		goto out;
2146	}
2147
2148	old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2149	rcu_assign_pointer(md->map, (void *)t);
2150	md->immutable_target_type = dm_table_get_immutable_target_type(t);
2151
2152	dm_table_set_restrictions(t, q, limits);
2153	if (old_map)
2154		dm_sync_table(md);
2155
2156out:
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 = rcu_dereference_protected(md->map, 1);
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	int r;
2183	struct mapped_device *md;
2184
2185	md = alloc_dev(minor);
2186	if (!md)
2187		return -ENXIO;
2188
2189	r = dm_sysfs_init(md);
2190	if (r) {
2191		free_dev(md);
2192		return r;
2193	}
2194
2195	*result = md;
2196	return 0;
2197}
2198
2199/*
2200 * Functions to manage md->type.
2201 * All are required to hold md->type_lock.
2202 */
2203void dm_lock_md_type(struct mapped_device *md)
2204{
2205	mutex_lock(&md->type_lock);
2206}
2207
2208void dm_unlock_md_type(struct mapped_device *md)
2209{
2210	mutex_unlock(&md->type_lock);
2211}
2212
2213void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2214{
2215	BUG_ON(!mutex_is_locked(&md->type_lock));
2216	md->type = type;
2217}
2218
2219enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2220{
2221	return md->type;
2222}
2223
2224struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2225{
2226	return md->immutable_target_type;
2227}
2228
2229/*
2230 * The queue_limits are only valid as long as you have a reference
2231 * count on 'md'.
2232 */
2233struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2234{
2235	BUG_ON(!atomic_read(&md->holders));
2236	return &md->queue->limits;
2237}
2238EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2239
2240/*
2241 * Setup the DM device's queue based on md's type
2242 */
2243int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2244{
2245	int r;
2246	struct queue_limits limits;
2247	enum dm_queue_mode type = dm_get_md_type(md);
2248
2249	switch (type) {
2250	case DM_TYPE_REQUEST_BASED:
 
 
 
 
 
 
 
2251		r = dm_mq_init_request_queue(md, t);
2252		if (r) {
2253			DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2254			return r;
2255		}
2256		break;
2257	case DM_TYPE_BIO_BASED:
2258	case DM_TYPE_DAX_BIO_BASED:
2259	case DM_TYPE_NVME_BIO_BASED:
2260		break;
2261	case DM_TYPE_NONE:
2262		WARN_ON_ONCE(true);
2263		break;
2264	}
 
 
2265
2266	r = dm_calculate_queue_limits(t, &limits);
2267	if (r) {
2268		DMERR("Cannot calculate initial queue limits");
2269		return r;
2270	}
2271	dm_table_set_restrictions(t, md->queue, &limits);
2272	blk_register_queue(md->disk);
2273
2274	return 0;
2275}
2276
2277struct mapped_device *dm_get_md(dev_t dev)
2278{
2279	struct mapped_device *md;
2280	unsigned minor = MINOR(dev);
2281
2282	if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2283		return NULL;
2284
2285	spin_lock(&_minor_lock);
2286
2287	md = idr_find(&_minor_idr, minor);
2288	if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2289	    test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2290		md = NULL;
2291		goto out;
 
 
 
 
 
2292	}
2293	dm_get(md);
2294out:
2295	spin_unlock(&_minor_lock);
2296
2297	return md;
2298}
2299EXPORT_SYMBOL_GPL(dm_get_md);
2300
2301void *dm_get_mdptr(struct mapped_device *md)
2302{
2303	return md->interface_ptr;
2304}
2305
2306void dm_set_mdptr(struct mapped_device *md, void *ptr)
2307{
2308	md->interface_ptr = ptr;
2309}
2310
2311void dm_get(struct mapped_device *md)
2312{
2313	atomic_inc(&md->holders);
2314	BUG_ON(test_bit(DMF_FREEING, &md->flags));
2315}
2316
2317int dm_hold(struct mapped_device *md)
2318{
2319	spin_lock(&_minor_lock);
2320	if (test_bit(DMF_FREEING, &md->flags)) {
2321		spin_unlock(&_minor_lock);
2322		return -EBUSY;
2323	}
2324	dm_get(md);
2325	spin_unlock(&_minor_lock);
2326	return 0;
2327}
2328EXPORT_SYMBOL_GPL(dm_hold);
2329
2330const char *dm_device_name(struct mapped_device *md)
2331{
2332	return md->name;
2333}
2334EXPORT_SYMBOL_GPL(dm_device_name);
2335
2336static void __dm_destroy(struct mapped_device *md, bool wait)
2337{
 
2338	struct dm_table *map;
2339	int srcu_idx;
2340
2341	might_sleep();
2342
2343	spin_lock(&_minor_lock);
2344	idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2345	set_bit(DMF_FREEING, &md->flags);
2346	spin_unlock(&_minor_lock);
2347
2348	blk_set_queue_dying(md->queue);
 
 
 
2349
2350	/*
2351	 * Take suspend_lock so that presuspend and postsuspend methods
2352	 * do not race with internal suspend.
2353	 */
2354	mutex_lock(&md->suspend_lock);
2355	map = dm_get_live_table(md, &srcu_idx);
2356	if (!dm_suspended_md(md)) {
2357		dm_table_presuspend_targets(map);
2358		set_bit(DMF_SUSPENDED, &md->flags);
2359		set_bit(DMF_POST_SUSPENDING, &md->flags);
2360		dm_table_postsuspend_targets(map);
2361	}
2362	/* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2363	dm_put_live_table(md, srcu_idx);
2364	mutex_unlock(&md->suspend_lock);
2365
2366	/*
2367	 * Rare, but there may be I/O requests still going to complete,
2368	 * for example.  Wait for all references to disappear.
2369	 * No one should increment the reference count of the mapped_device,
2370	 * after the mapped_device state becomes DMF_FREEING.
2371	 */
2372	if (wait)
2373		while (atomic_read(&md->holders))
2374			msleep(1);
2375	else if (atomic_read(&md->holders))
2376		DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2377		       dm_device_name(md), atomic_read(&md->holders));
2378
2379	dm_sysfs_exit(md);
2380	dm_table_destroy(__unbind(md));
2381	free_dev(md);
2382}
2383
2384void dm_destroy(struct mapped_device *md)
2385{
2386	__dm_destroy(md, true);
2387}
2388
2389void dm_destroy_immediate(struct mapped_device *md)
2390{
2391	__dm_destroy(md, false);
2392}
2393
2394void dm_put(struct mapped_device *md)
2395{
2396	atomic_dec(&md->holders);
2397}
2398EXPORT_SYMBOL_GPL(dm_put);
2399
2400static bool md_in_flight_bios(struct mapped_device *md)
2401{
2402	int cpu;
2403	struct hd_struct *part = &dm_disk(md)->part0;
2404	long sum = 0;
2405
2406	for_each_possible_cpu(cpu) {
2407		sum += part_stat_local_read_cpu(part, in_flight[0], cpu);
2408		sum += part_stat_local_read_cpu(part, in_flight[1], cpu);
2409	}
2410
2411	return sum != 0;
2412}
2413
2414static int dm_wait_for_bios_completion(struct mapped_device *md, long task_state)
2415{
2416	int r = 0;
2417	DEFINE_WAIT(wait);
2418
2419	while (true) {
2420		prepare_to_wait(&md->wait, &wait, task_state);
2421
2422		if (!md_in_flight_bios(md))
2423			break;
2424
2425		if (signal_pending_state(task_state, current)) {
2426			r = -EINTR;
2427			break;
2428		}
2429
2430		io_schedule();
2431	}
2432	finish_wait(&md->wait, &wait);
2433
2434	return r;
2435}
2436
2437static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2438{
2439	int r = 0;
2440
2441	if (!queue_is_mq(md->queue))
2442		return dm_wait_for_bios_completion(md, task_state);
2443
2444	while (true) {
2445		if (!blk_mq_queue_inflight(md->queue))
2446			break;
2447
2448		if (signal_pending_state(task_state, current)) {
2449			r = -EINTR;
2450			break;
2451		}
2452
2453		msleep(5);
2454	}
2455
2456	return r;
2457}
2458
2459/*
2460 * Process the deferred bios
2461 */
2462static void dm_wq_work(struct work_struct *work)
2463{
2464	struct mapped_device *md = container_of(work, struct mapped_device,
2465						work);
2466	struct bio *c;
2467	int srcu_idx;
2468	struct dm_table *map;
2469
2470	map = dm_get_live_table(md, &srcu_idx);
2471
2472	while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2473		spin_lock_irq(&md->deferred_lock);
2474		c = bio_list_pop(&md->deferred);
2475		spin_unlock_irq(&md->deferred_lock);
2476
2477		if (!c)
2478			break;
2479
2480		if (dm_request_based(md))
2481			(void) submit_bio_noacct(c);
2482		else
2483			(void) dm_process_bio(md, map, c);
2484	}
2485
2486	dm_put_live_table(md, srcu_idx);
2487}
2488
2489static void dm_queue_flush(struct mapped_device *md)
2490{
2491	clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2492	smp_mb__after_atomic();
2493	queue_work(md->wq, &md->work);
2494}
2495
2496/*
2497 * Swap in a new table, returning the old one for the caller to destroy.
2498 */
2499struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2500{
2501	struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2502	struct queue_limits limits;
2503	int r;
2504
2505	mutex_lock(&md->suspend_lock);
2506
2507	/* device must be suspended */
2508	if (!dm_suspended_md(md))
2509		goto out;
2510
2511	/*
2512	 * If the new table has no data devices, retain the existing limits.
2513	 * This helps multipath with queue_if_no_path if all paths disappear,
2514	 * then new I/O is queued based on these limits, and then some paths
2515	 * reappear.
2516	 */
2517	if (dm_table_has_no_data_devices(table)) {
2518		live_map = dm_get_live_table_fast(md);
2519		if (live_map)
2520			limits = md->queue->limits;
2521		dm_put_live_table_fast(md);
2522	}
2523
2524	if (!live_map) {
2525		r = dm_calculate_queue_limits(table, &limits);
2526		if (r) {
2527			map = ERR_PTR(r);
2528			goto out;
2529		}
2530	}
2531
2532	map = __bind(md, table, &limits);
2533	dm_issue_global_event();
2534
2535out:
2536	mutex_unlock(&md->suspend_lock);
2537	return map;
2538}
2539
2540/*
2541 * Functions to lock and unlock any filesystem running on the
2542 * device.
2543 */
2544static int lock_fs(struct mapped_device *md)
2545{
2546	int r;
2547
2548	WARN_ON(md->frozen_sb);
2549
2550	md->frozen_sb = freeze_bdev(md->bdev);
2551	if (IS_ERR(md->frozen_sb)) {
2552		r = PTR_ERR(md->frozen_sb);
2553		md->frozen_sb = NULL;
2554		return r;
2555	}
2556
2557	set_bit(DMF_FROZEN, &md->flags);
2558
2559	return 0;
2560}
2561
2562static void unlock_fs(struct mapped_device *md)
2563{
2564	if (!test_bit(DMF_FROZEN, &md->flags))
2565		return;
2566
2567	thaw_bdev(md->bdev, md->frozen_sb);
2568	md->frozen_sb = NULL;
2569	clear_bit(DMF_FROZEN, &md->flags);
2570}
2571
2572/*
2573 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2574 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2575 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2576 *
2577 * If __dm_suspend returns 0, the device is completely quiescent
2578 * now. There is no request-processing activity. All new requests
2579 * are being added to md->deferred list.
 
 
2580 */
2581static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2582			unsigned suspend_flags, long task_state,
2583			int dmf_suspended_flag)
2584{
2585	bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2586	bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2587	int r;
2588
2589	lockdep_assert_held(&md->suspend_lock);
2590
2591	/*
2592	 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2593	 * This flag is cleared before dm_suspend returns.
2594	 */
2595	if (noflush)
2596		set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2597	else
2598		DMDEBUG("%s: suspending with flush", dm_device_name(md));
2599
2600	/*
2601	 * This gets reverted if there's an error later and the targets
2602	 * provide the .presuspend_undo hook.
2603	 */
2604	dm_table_presuspend_targets(map);
2605
2606	/*
2607	 * Flush I/O to the device.
2608	 * Any I/O submitted after lock_fs() may not be flushed.
2609	 * noflush takes precedence over do_lockfs.
2610	 * (lock_fs() flushes I/Os and waits for them to complete.)
2611	 */
2612	if (!noflush && do_lockfs) {
2613		r = lock_fs(md);
2614		if (r) {
2615			dm_table_presuspend_undo_targets(map);
2616			return r;
2617		}
2618	}
2619
2620	/*
2621	 * Here we must make sure that no processes are submitting requests
2622	 * to target drivers i.e. no one may be executing
2623	 * __split_and_process_bio. This is called from dm_request and
2624	 * dm_wq_work.
2625	 *
2626	 * To get all processes out of __split_and_process_bio in dm_request,
2627	 * we take the write lock. To prevent any process from reentering
2628	 * __split_and_process_bio from dm_request and quiesce the thread
2629	 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2630	 * flush_workqueue(md->wq).
2631	 */
2632	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2633	if (map)
2634		synchronize_srcu(&md->io_barrier);
2635
2636	/*
2637	 * Stop md->queue before flushing md->wq in case request-based
2638	 * dm defers requests to md->wq from md->queue.
2639	 */
2640	if (dm_request_based(md))
2641		dm_stop_queue(md->queue);
 
 
 
2642
2643	flush_workqueue(md->wq);
2644
2645	/*
2646	 * At this point no more requests are entering target request routines.
2647	 * We call dm_wait_for_completion to wait for all existing requests
2648	 * to finish.
2649	 */
2650	r = dm_wait_for_completion(md, task_state);
2651	if (!r)
2652		set_bit(dmf_suspended_flag, &md->flags);
2653
2654	if (noflush)
2655		clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2656	if (map)
2657		synchronize_srcu(&md->io_barrier);
2658
2659	/* were we interrupted ? */
2660	if (r < 0) {
2661		dm_queue_flush(md);
2662
2663		if (dm_request_based(md))
2664			dm_start_queue(md->queue);
2665
2666		unlock_fs(md);
2667		dm_table_presuspend_undo_targets(map);
2668		/* pushback list is already flushed, so skip flush */
2669	}
2670
2671	return r;
2672}
2673
2674/*
2675 * We need to be able to change a mapping table under a mounted
2676 * filesystem.  For example we might want to move some data in
2677 * the background.  Before the table can be swapped with
2678 * dm_bind_table, dm_suspend must be called to flush any in
2679 * flight bios and ensure that any further io gets deferred.
2680 */
2681/*
2682 * Suspend mechanism in request-based dm.
2683 *
2684 * 1. Flush all I/Os by lock_fs() if needed.
2685 * 2. Stop dispatching any I/O by stopping the request_queue.
2686 * 3. Wait for all in-flight I/Os to be completed or requeued.
2687 *
2688 * To abort suspend, start the request_queue.
2689 */
2690int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2691{
2692	struct dm_table *map = NULL;
2693	int r = 0;
2694
2695retry:
2696	mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2697
2698	if (dm_suspended_md(md)) {
2699		r = -EINVAL;
2700		goto out_unlock;
2701	}
2702
2703	if (dm_suspended_internally_md(md)) {
2704		/* already internally suspended, wait for internal resume */
2705		mutex_unlock(&md->suspend_lock);
2706		r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2707		if (r)
2708			return r;
2709		goto retry;
2710	}
2711
2712	map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2713
2714	r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2715	if (r)
2716		goto out_unlock;
2717
2718	set_bit(DMF_POST_SUSPENDING, &md->flags);
2719	dm_table_postsuspend_targets(map);
2720	clear_bit(DMF_POST_SUSPENDING, &md->flags);
2721
2722out_unlock:
2723	mutex_unlock(&md->suspend_lock);
2724	return r;
2725}
2726
2727static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2728{
2729	if (map) {
2730		int r = dm_table_resume_targets(map);
2731		if (r)
2732			return r;
2733	}
2734
2735	dm_queue_flush(md);
2736
2737	/*
2738	 * Flushing deferred I/Os must be done after targets are resumed
2739	 * so that mapping of targets can work correctly.
2740	 * Request-based dm is queueing the deferred I/Os in its request_queue.
2741	 */
2742	if (dm_request_based(md))
2743		dm_start_queue(md->queue);
2744
2745	unlock_fs(md);
2746
2747	return 0;
2748}
2749
2750int dm_resume(struct mapped_device *md)
2751{
2752	int r;
2753	struct dm_table *map = NULL;
2754
2755retry:
2756	r = -EINVAL;
2757	mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2758
2759	if (!dm_suspended_md(md))
2760		goto out;
2761
2762	if (dm_suspended_internally_md(md)) {
2763		/* already internally suspended, wait for internal resume */
2764		mutex_unlock(&md->suspend_lock);
2765		r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2766		if (r)
2767			return r;
2768		goto retry;
2769	}
2770
2771	map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2772	if (!map || !dm_table_get_size(map))
2773		goto out;
2774
2775	r = __dm_resume(md, map);
2776	if (r)
2777		goto out;
2778
2779	clear_bit(DMF_SUSPENDED, &md->flags);
2780out:
2781	mutex_unlock(&md->suspend_lock);
2782
2783	return r;
2784}
2785
2786/*
2787 * Internal suspend/resume works like userspace-driven suspend. It waits
2788 * until all bios finish and prevents issuing new bios to the target drivers.
2789 * It may be used only from the kernel.
2790 */
2791
2792static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2793{
2794	struct dm_table *map = NULL;
2795
2796	lockdep_assert_held(&md->suspend_lock);
2797
2798	if (md->internal_suspend_count++)
2799		return; /* nested internal suspend */
2800
2801	if (dm_suspended_md(md)) {
2802		set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2803		return; /* nest suspend */
2804	}
2805
2806	map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2807
2808	/*
2809	 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2810	 * supported.  Properly supporting a TASK_INTERRUPTIBLE internal suspend
2811	 * would require changing .presuspend to return an error -- avoid this
2812	 * until there is a need for more elaborate variants of internal suspend.
2813	 */
2814	(void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2815			    DMF_SUSPENDED_INTERNALLY);
2816
2817	set_bit(DMF_POST_SUSPENDING, &md->flags);
2818	dm_table_postsuspend_targets(map);
2819	clear_bit(DMF_POST_SUSPENDING, &md->flags);
2820}
2821
2822static void __dm_internal_resume(struct mapped_device *md)
2823{
2824	BUG_ON(!md->internal_suspend_count);
2825
2826	if (--md->internal_suspend_count)
2827		return; /* resume from nested internal suspend */
2828
2829	if (dm_suspended_md(md))
2830		goto done; /* resume from nested suspend */
2831
2832	/*
2833	 * NOTE: existing callers don't need to call dm_table_resume_targets
2834	 * (which may fail -- so best to avoid it for now by passing NULL map)
2835	 */
2836	(void) __dm_resume(md, NULL);
2837
2838done:
2839	clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2840	smp_mb__after_atomic();
2841	wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2842}
2843
2844void dm_internal_suspend_noflush(struct mapped_device *md)
2845{
2846	mutex_lock(&md->suspend_lock);
2847	__dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2848	mutex_unlock(&md->suspend_lock);
2849}
2850EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2851
2852void dm_internal_resume(struct mapped_device *md)
2853{
2854	mutex_lock(&md->suspend_lock);
2855	__dm_internal_resume(md);
2856	mutex_unlock(&md->suspend_lock);
2857}
2858EXPORT_SYMBOL_GPL(dm_internal_resume);
2859
2860/*
2861 * Fast variants of internal suspend/resume hold md->suspend_lock,
2862 * which prevents interaction with userspace-driven suspend.
2863 */
2864
2865void dm_internal_suspend_fast(struct mapped_device *md)
2866{
2867	mutex_lock(&md->suspend_lock);
2868	if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2869		return;
2870
2871	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2872	synchronize_srcu(&md->io_barrier);
2873	flush_workqueue(md->wq);
2874	dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2875}
2876EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2877
2878void dm_internal_resume_fast(struct mapped_device *md)
2879{
2880	if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2881		goto done;
2882
2883	dm_queue_flush(md);
2884
2885done:
2886	mutex_unlock(&md->suspend_lock);
2887}
2888EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2889
2890/*-----------------------------------------------------------------
2891 * Event notification.
2892 *---------------------------------------------------------------*/
2893int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2894		       unsigned cookie)
2895{
2896	int r;
2897	unsigned noio_flag;
2898	char udev_cookie[DM_COOKIE_LENGTH];
2899	char *envp[] = { udev_cookie, NULL };
2900
2901	noio_flag = memalloc_noio_save();
2902
2903	if (!cookie)
2904		r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2905	else {
2906		snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2907			 DM_COOKIE_ENV_VAR_NAME, cookie);
2908		r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2909				       action, envp);
2910	}
2911
2912	memalloc_noio_restore(noio_flag);
2913
2914	return r;
2915}
2916
2917uint32_t dm_next_uevent_seq(struct mapped_device *md)
2918{
2919	return atomic_add_return(1, &md->uevent_seq);
2920}
2921
2922uint32_t dm_get_event_nr(struct mapped_device *md)
2923{
2924	return atomic_read(&md->event_nr);
2925}
2926
2927int dm_wait_event(struct mapped_device *md, int event_nr)
2928{
2929	return wait_event_interruptible(md->eventq,
2930			(event_nr != atomic_read(&md->event_nr)));
2931}
2932
2933void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2934{
2935	unsigned long flags;
2936
2937	spin_lock_irqsave(&md->uevent_lock, flags);
2938	list_add(elist, &md->uevent_list);
2939	spin_unlock_irqrestore(&md->uevent_lock, flags);
2940}
2941
2942/*
2943 * The gendisk is only valid as long as you have a reference
2944 * count on 'md'.
2945 */
2946struct gendisk *dm_disk(struct mapped_device *md)
2947{
2948	return md->disk;
2949}
2950EXPORT_SYMBOL_GPL(dm_disk);
2951
2952struct kobject *dm_kobject(struct mapped_device *md)
2953{
2954	return &md->kobj_holder.kobj;
2955}
2956
2957struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2958{
2959	struct mapped_device *md;
2960
2961	md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2962
2963	spin_lock(&_minor_lock);
2964	if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2965		md = NULL;
2966		goto out;
2967	}
2968	dm_get(md);
2969out:
2970	spin_unlock(&_minor_lock);
2971
 
2972	return md;
2973}
2974
2975int dm_suspended_md(struct mapped_device *md)
2976{
2977	return test_bit(DMF_SUSPENDED, &md->flags);
2978}
2979
2980static int dm_post_suspending_md(struct mapped_device *md)
2981{
2982	return test_bit(DMF_POST_SUSPENDING, &md->flags);
2983}
2984
2985int dm_suspended_internally_md(struct mapped_device *md)
2986{
2987	return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2988}
2989
2990int dm_test_deferred_remove_flag(struct mapped_device *md)
2991{
2992	return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2993}
2994
2995int dm_suspended(struct dm_target *ti)
2996{
2997	return dm_suspended_md(dm_table_get_md(ti->table));
2998}
2999EXPORT_SYMBOL_GPL(dm_suspended);
3000
3001int dm_post_suspending(struct dm_target *ti)
3002{
3003	return dm_post_suspending_md(dm_table_get_md(ti->table));
3004}
3005EXPORT_SYMBOL_GPL(dm_post_suspending);
3006
3007int dm_noflush_suspending(struct dm_target *ti)
3008{
3009	return __noflush_suspending(dm_table_get_md(ti->table));
3010}
3011EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3012
3013struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
3014					    unsigned integrity, unsigned per_io_data_size,
3015					    unsigned min_pool_size)
3016{
3017	struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
 
3018	unsigned int pool_size = 0;
3019	unsigned int front_pad, io_front_pad;
3020	int ret;
3021
3022	if (!pools)
3023		return NULL;
3024
3025	switch (type) {
3026	case DM_TYPE_BIO_BASED:
3027	case DM_TYPE_DAX_BIO_BASED:
3028	case DM_TYPE_NVME_BIO_BASED:
3029		pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
3030		front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
3031		io_front_pad = roundup(front_pad,  __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
3032		ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
3033		if (ret)
3034			goto out;
3035		if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
3036			goto out;
3037		break;
3038	case DM_TYPE_REQUEST_BASED:
3039		pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
 
 
 
 
 
 
 
 
3040		front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
3041		/* per_io_data_size is used for blk-mq pdu at queue allocation */
3042		break;
3043	default:
3044		BUG();
3045	}
3046
3047	ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
3048	if (ret)
 
 
 
 
 
 
3049		goto out;
3050
3051	if (integrity && bioset_integrity_create(&pools->bs, pool_size))
3052		goto out;
3053
3054	return pools;
3055
3056out:
3057	dm_free_md_mempools(pools);
3058
3059	return NULL;
3060}
3061
3062void dm_free_md_mempools(struct dm_md_mempools *pools)
3063{
3064	if (!pools)
3065		return;
3066
3067	bioset_exit(&pools->bs);
3068	bioset_exit(&pools->io_bs);
 
 
 
3069
3070	kfree(pools);
3071}
3072
3073struct dm_pr {
3074	u64	old_key;
3075	u64	new_key;
3076	u32	flags;
3077	bool	fail_early;
3078};
3079
3080static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3081		      void *data)
3082{
3083	struct mapped_device *md = bdev->bd_disk->private_data;
3084	struct dm_table *table;
3085	struct dm_target *ti;
3086	int ret = -ENOTTY, srcu_idx;
3087
3088	table = dm_get_live_table(md, &srcu_idx);
3089	if (!table || !dm_table_get_size(table))
3090		goto out;
3091
3092	/* We only support devices that have a single target */
3093	if (dm_table_get_num_targets(table) != 1)
3094		goto out;
3095	ti = dm_table_get_target(table, 0);
3096
3097	ret = -EINVAL;
3098	if (!ti->type->iterate_devices)
3099		goto out;
3100
3101	ret = ti->type->iterate_devices(ti, fn, data);
3102out:
3103	dm_put_live_table(md, srcu_idx);
3104	return ret;
3105}
3106
3107/*
3108 * For register / unregister we need to manually call out to every path.
3109 */
3110static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3111			    sector_t start, sector_t len, void *data)
3112{
3113	struct dm_pr *pr = data;
3114	const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3115
3116	if (!ops || !ops->pr_register)
3117		return -EOPNOTSUPP;
3118	return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3119}
3120
3121static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3122			  u32 flags)
3123{
3124	struct dm_pr pr = {
3125		.old_key	= old_key,
3126		.new_key	= new_key,
3127		.flags		= flags,
3128		.fail_early	= true,
3129	};
3130	int ret;
3131
3132	ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3133	if (ret && new_key) {
3134		/* unregister all paths if we failed to register any path */
3135		pr.old_key = new_key;
3136		pr.new_key = 0;
3137		pr.flags = 0;
3138		pr.fail_early = false;
3139		dm_call_pr(bdev, __dm_pr_register, &pr);
3140	}
3141
3142	return ret;
3143}
3144
3145static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3146			 u32 flags)
3147{
3148	struct mapped_device *md = bdev->bd_disk->private_data;
3149	const struct pr_ops *ops;
3150	int r, srcu_idx;
 
3151
3152	r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3153	if (r < 0)
3154		goto out;
3155
3156	ops = bdev->bd_disk->fops->pr_ops;
3157	if (ops && ops->pr_reserve)
3158		r = ops->pr_reserve(bdev, key, type, flags);
3159	else
3160		r = -EOPNOTSUPP;
3161out:
3162	dm_unprepare_ioctl(md, srcu_idx);
3163	return r;
3164}
3165
3166static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3167{
3168	struct mapped_device *md = bdev->bd_disk->private_data;
3169	const struct pr_ops *ops;
3170	int r, srcu_idx;
 
3171
3172	r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3173	if (r < 0)
3174		goto out;
3175
3176	ops = bdev->bd_disk->fops->pr_ops;
3177	if (ops && ops->pr_release)
3178		r = ops->pr_release(bdev, key, type);
3179	else
3180		r = -EOPNOTSUPP;
3181out:
3182	dm_unprepare_ioctl(md, srcu_idx);
3183	return r;
3184}
3185
3186static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3187			 enum pr_type type, bool abort)
3188{
3189	struct mapped_device *md = bdev->bd_disk->private_data;
3190	const struct pr_ops *ops;
3191	int r, srcu_idx;
 
3192
3193	r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3194	if (r < 0)
3195		goto out;
3196
3197	ops = bdev->bd_disk->fops->pr_ops;
3198	if (ops && ops->pr_preempt)
3199		r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3200	else
3201		r = -EOPNOTSUPP;
3202out:
3203	dm_unprepare_ioctl(md, srcu_idx);
3204	return r;
3205}
3206
3207static int dm_pr_clear(struct block_device *bdev, u64 key)
3208{
3209	struct mapped_device *md = bdev->bd_disk->private_data;
3210	const struct pr_ops *ops;
3211	int r, srcu_idx;
 
3212
3213	r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3214	if (r < 0)
3215		goto out;
3216
3217	ops = bdev->bd_disk->fops->pr_ops;
3218	if (ops && ops->pr_clear)
3219		r = ops->pr_clear(bdev, key);
3220	else
3221		r = -EOPNOTSUPP;
3222out:
3223	dm_unprepare_ioctl(md, srcu_idx);
3224	return r;
3225}
3226
3227static const struct pr_ops dm_pr_ops = {
3228	.pr_register	= dm_pr_register,
3229	.pr_reserve	= dm_pr_reserve,
3230	.pr_release	= dm_pr_release,
3231	.pr_preempt	= dm_pr_preempt,
3232	.pr_clear	= dm_pr_clear,
3233};
3234
3235static const struct block_device_operations dm_blk_dops = {
3236	.submit_bio = dm_submit_bio,
3237	.open = dm_blk_open,
3238	.release = dm_blk_close,
3239	.ioctl = dm_blk_ioctl,
 
3240	.getgeo = dm_blk_getgeo,
3241	.report_zones = dm_blk_report_zones,
3242	.pr_ops = &dm_pr_ops,
3243	.owner = THIS_MODULE
3244};
3245
3246static const struct dax_operations dm_dax_ops = {
3247	.direct_access = dm_dax_direct_access,
3248	.dax_supported = dm_dax_supported,
3249	.copy_from_iter = dm_dax_copy_from_iter,
3250	.copy_to_iter = dm_dax_copy_to_iter,
3251	.zero_page_range = dm_dax_zero_page_range,
3252};
3253
3254/*
3255 * module hooks
3256 */
3257module_init(dm_init);
3258module_exit(dm_exit);
3259
3260module_param(major, uint, 0);
3261MODULE_PARM_DESC(major, "The major number of the device mapper");
3262
3263module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3264MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3265
3266module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3267MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3268
3269MODULE_DESCRIPTION(DM_NAME " driver");
3270MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3271MODULE_LICENSE("GPL");
v4.10.11
   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-core.h"
   9#include "dm-rq.h"
  10#include "dm-uevent.h"
  11
  12#include <linux/init.h>
  13#include <linux/module.h>
  14#include <linux/mutex.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#include <linux/wait.h>
  23#include <linux/pr.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
  57static struct workqueue_struct *deferred_remove_workqueue;
  58
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  59/*
  60 * One of these is allocated per bio.
 
  61 */
 
  62struct dm_io {
 
  63	struct mapped_device *md;
  64	int error;
  65	atomic_t io_count;
  66	struct bio *bio;
  67	unsigned long start_time;
  68	spinlock_t endio_lock;
  69	struct dm_stats_aux stats_aux;
 
 
  70};
  71
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  72#define MINOR_ALLOCED ((void *)-1)
  73
  74/*
  75 * Bits for the md->flags field.
  76 */
  77#define DMF_BLOCK_IO_FOR_SUSPEND 0
  78#define DMF_SUSPENDED 1
  79#define DMF_FROZEN 2
  80#define DMF_FREEING 3
  81#define DMF_DELETING 4
  82#define DMF_NOFLUSH_SUSPENDING 5
  83#define DMF_DEFERRED_REMOVE 6
  84#define DMF_SUSPENDED_INTERNALLY 7
 
  85
  86#define DM_NUMA_NODE NUMA_NO_NODE
  87static int dm_numa_node = DM_NUMA_NODE;
  88
  89/*
  90 * For mempools pre-allocation at the table loading time.
  91 */
  92struct dm_md_mempools {
  93	mempool_t *io_pool;
  94	mempool_t *rq_pool;
  95	struct bio_set *bs;
  96};
  97
  98struct table_device {
  99	struct list_head list;
 100	atomic_t count;
 101	struct dm_dev dm_dev;
 102};
 103
 104static struct kmem_cache *_io_cache;
 105static struct kmem_cache *_rq_tio_cache;
 106static struct kmem_cache *_rq_cache;
 107
 108/*
 109 * Bio-based DM's mempools' reserved IOs set by the user.
 110 */
 111#define RESERVED_BIO_BASED_IOS		16
 112static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
 113
 114static int __dm_get_module_param_int(int *module_param, int min, int max)
 115{
 116	int param = ACCESS_ONCE(*module_param);
 117	int modified_param = 0;
 118	bool modified = true;
 119
 120	if (param < min)
 121		modified_param = min;
 122	else if (param > max)
 123		modified_param = max;
 124	else
 125		modified = false;
 126
 127	if (modified) {
 128		(void)cmpxchg(module_param, param, modified_param);
 129		param = modified_param;
 130	}
 131
 132	return param;
 133}
 134
 135unsigned __dm_get_module_param(unsigned *module_param,
 136			       unsigned def, unsigned max)
 137{
 138	unsigned param = ACCESS_ONCE(*module_param);
 139	unsigned modified_param = 0;
 140
 141	if (!param)
 142		modified_param = def;
 143	else if (param > max)
 144		modified_param = max;
 145
 146	if (modified_param) {
 147		(void)cmpxchg(module_param, param, modified_param);
 148		param = modified_param;
 149	}
 150
 151	return param;
 152}
 153
 154unsigned dm_get_reserved_bio_based_ios(void)
 155{
 156	return __dm_get_module_param(&reserved_bio_based_ios,
 157				     RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
 158}
 159EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
 160
 161static unsigned dm_get_numa_node(void)
 162{
 163	return __dm_get_module_param_int(&dm_numa_node,
 164					 DM_NUMA_NODE, num_online_nodes() - 1);
 165}
 166
 167static int __init local_init(void)
 168{
 169	int r = -ENOMEM;
 170
 171	/* allocate a slab for the dm_ios */
 172	_io_cache = KMEM_CACHE(dm_io, 0);
 173	if (!_io_cache)
 174		return r;
 175
 176	_rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
 177	if (!_rq_tio_cache)
 178		goto out_free_io_cache;
 179
 180	_rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request),
 181				      __alignof__(struct request), 0, NULL);
 182	if (!_rq_cache)
 183		goto out_free_rq_tio_cache;
 184
 185	r = dm_uevent_init();
 186	if (r)
 187		goto out_free_rq_cache;
 188
 189	deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
 190	if (!deferred_remove_workqueue) {
 191		r = -ENOMEM;
 192		goto out_uevent_exit;
 193	}
 194
 195	_major = major;
 196	r = register_blkdev(_major, _name);
 197	if (r < 0)
 198		goto out_free_workqueue;
 199
 200	if (!_major)
 201		_major = r;
 202
 203	return 0;
 204
 205out_free_workqueue:
 206	destroy_workqueue(deferred_remove_workqueue);
 207out_uevent_exit:
 208	dm_uevent_exit();
 209out_free_rq_cache:
 210	kmem_cache_destroy(_rq_cache);
 211out_free_rq_tio_cache:
 212	kmem_cache_destroy(_rq_tio_cache);
 213out_free_io_cache:
 214	kmem_cache_destroy(_io_cache);
 215
 216	return r;
 217}
 218
 219static void local_exit(void)
 220{
 221	flush_scheduled_work();
 222	destroy_workqueue(deferred_remove_workqueue);
 223
 224	kmem_cache_destroy(_rq_cache);
 225	kmem_cache_destroy(_rq_tio_cache);
 226	kmem_cache_destroy(_io_cache);
 227	unregister_blkdev(_major, _name);
 228	dm_uevent_exit();
 229
 230	_major = 0;
 231
 232	DMINFO("cleaned up");
 233}
 234
 235static int (*_inits[])(void) __initdata = {
 236	local_init,
 237	dm_target_init,
 238	dm_linear_init,
 239	dm_stripe_init,
 240	dm_io_init,
 241	dm_kcopyd_init,
 242	dm_interface_init,
 243	dm_statistics_init,
 244};
 245
 246static void (*_exits[])(void) = {
 247	local_exit,
 248	dm_target_exit,
 249	dm_linear_exit,
 250	dm_stripe_exit,
 251	dm_io_exit,
 252	dm_kcopyd_exit,
 253	dm_interface_exit,
 254	dm_statistics_exit,
 255};
 256
 257static int __init dm_init(void)
 258{
 259	const int count = ARRAY_SIZE(_inits);
 260
 261	int r, i;
 262
 263	for (i = 0; i < count; i++) {
 264		r = _inits[i]();
 265		if (r)
 266			goto bad;
 267	}
 268
 269	return 0;
 270
 271      bad:
 272	while (i--)
 273		_exits[i]();
 274
 275	return r;
 276}
 277
 278static void __exit dm_exit(void)
 279{
 280	int i = ARRAY_SIZE(_exits);
 281
 282	while (i--)
 283		_exits[i]();
 284
 285	/*
 286	 * Should be empty by this point.
 287	 */
 288	idr_destroy(&_minor_idr);
 289}
 290
 291/*
 292 * Block device functions
 293 */
 294int dm_deleting_md(struct mapped_device *md)
 295{
 296	return test_bit(DMF_DELETING, &md->flags);
 297}
 298
 299static int dm_blk_open(struct block_device *bdev, fmode_t mode)
 300{
 301	struct mapped_device *md;
 302
 303	spin_lock(&_minor_lock);
 304
 305	md = bdev->bd_disk->private_data;
 306	if (!md)
 307		goto out;
 308
 309	if (test_bit(DMF_FREEING, &md->flags) ||
 310	    dm_deleting_md(md)) {
 311		md = NULL;
 312		goto out;
 313	}
 314
 315	dm_get(md);
 316	atomic_inc(&md->open_count);
 317out:
 318	spin_unlock(&_minor_lock);
 319
 320	return md ? 0 : -ENXIO;
 321}
 322
 323static void dm_blk_close(struct gendisk *disk, fmode_t mode)
 324{
 325	struct mapped_device *md;
 326
 327	spin_lock(&_minor_lock);
 328
 329	md = disk->private_data;
 330	if (WARN_ON(!md))
 331		goto out;
 332
 333	if (atomic_dec_and_test(&md->open_count) &&
 334	    (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
 335		queue_work(deferred_remove_workqueue, &deferred_remove_work);
 336
 337	dm_put(md);
 338out:
 339	spin_unlock(&_minor_lock);
 340}
 341
 342int dm_open_count(struct mapped_device *md)
 343{
 344	return atomic_read(&md->open_count);
 345}
 346
 347/*
 348 * Guarantees nothing is using the device before it's deleted.
 349 */
 350int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
 351{
 352	int r = 0;
 353
 354	spin_lock(&_minor_lock);
 355
 356	if (dm_open_count(md)) {
 357		r = -EBUSY;
 358		if (mark_deferred)
 359			set_bit(DMF_DEFERRED_REMOVE, &md->flags);
 360	} else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
 361		r = -EEXIST;
 362	else
 363		set_bit(DMF_DELETING, &md->flags);
 364
 365	spin_unlock(&_minor_lock);
 366
 367	return r;
 368}
 369
 370int dm_cancel_deferred_remove(struct mapped_device *md)
 371{
 372	int r = 0;
 373
 374	spin_lock(&_minor_lock);
 375
 376	if (test_bit(DMF_DELETING, &md->flags))
 377		r = -EBUSY;
 378	else
 379		clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
 380
 381	spin_unlock(&_minor_lock);
 382
 383	return r;
 384}
 385
 386static void do_deferred_remove(struct work_struct *w)
 387{
 388	dm_deferred_remove();
 389}
 390
 391sector_t dm_get_size(struct mapped_device *md)
 392{
 393	return get_capacity(md->disk);
 394}
 395
 396struct request_queue *dm_get_md_queue(struct mapped_device *md)
 397{
 398	return md->queue;
 399}
 400
 401struct dm_stats *dm_get_stats(struct mapped_device *md)
 402{
 403	return &md->stats;
 404}
 405
 406static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 407{
 408	struct mapped_device *md = bdev->bd_disk->private_data;
 409
 410	return dm_get_geometry(md, geo);
 411}
 412
 413static int dm_grab_bdev_for_ioctl(struct mapped_device *md,
 414				  struct block_device **bdev,
 415				  fmode_t *mode)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 416{
 417	struct dm_target *tgt;
 418	struct dm_table *map;
 419	int srcu_idx, r;
 420
 421retry:
 422	r = -ENOTTY;
 423	map = dm_get_live_table(md, &srcu_idx);
 424	if (!map || !dm_table_get_size(map))
 425		goto out;
 426
 427	/* We only support devices that have a single target */
 428	if (dm_table_get_num_targets(map) != 1)
 429		goto out;
 430
 431	tgt = dm_table_get_target(map, 0);
 432	if (!tgt->type->prepare_ioctl)
 433		goto out;
 434
 435	if (dm_suspended_md(md)) {
 436		r = -EAGAIN;
 437		goto out;
 438	}
 439
 440	r = tgt->type->prepare_ioctl(tgt, bdev, mode);
 441	if (r < 0)
 442		goto out;
 443
 444	bdgrab(*bdev);
 445	dm_put_live_table(md, srcu_idx);
 446	return r;
 447
 448out:
 449	dm_put_live_table(md, srcu_idx);
 450	if (r == -ENOTCONN && !fatal_signal_pending(current)) {
 
 451		msleep(10);
 452		goto retry;
 453	}
 
 454	return r;
 455}
 456
 
 
 
 
 
 
 457static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
 458			unsigned int cmd, unsigned long arg)
 459{
 460	struct mapped_device *md = bdev->bd_disk->private_data;
 461	int r;
 462
 463	r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
 464	if (r < 0)
 465		return r;
 466
 467	if (r > 0) {
 468		/*
 469		 * Target determined this ioctl is being issued against
 470		 * a logical partition of the parent bdev; so extra
 471		 * validation is needed.
 472		 */
 473		r = scsi_verify_blk_ioctl(NULL, cmd);
 474		if (r)
 
 
 
 475			goto out;
 
 476	}
 477
 478	r =  __blkdev_driver_ioctl(bdev, mode, cmd, arg);
 479out:
 480	bdput(bdev);
 481	return r;
 482}
 483
 484static struct dm_io *alloc_io(struct mapped_device *md)
 
 
 485{
 486	return mempool_alloc(md->io_pool, GFP_NOIO);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 487}
 488
 489static void free_io(struct mapped_device *md, struct dm_io *io)
 490{
 491	mempool_free(io, md->io_pool);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 492}
 493
 494static void free_tio(struct dm_target_io *tio)
 495{
 
 
 496	bio_put(&tio->clone);
 497}
 498
 499int md_in_flight(struct mapped_device *md)
 500{
 501	return atomic_read(&md->pending[READ]) +
 502	       atomic_read(&md->pending[WRITE]);
 
 
 503}
 
 504
 505static void start_io_acct(struct dm_io *io)
 506{
 507	struct mapped_device *md = io->md;
 508	struct bio *bio = io->bio;
 509	int cpu;
 510	int rw = bio_data_dir(bio);
 511
 512	io->start_time = jiffies;
 513
 514	cpu = part_stat_lock();
 515	part_round_stats(cpu, &dm_disk(md)->part0);
 516	part_stat_unlock();
 517	atomic_set(&dm_disk(md)->part0.in_flight[rw],
 518		atomic_inc_return(&md->pending[rw]));
 519
 
 520	if (unlikely(dm_stats_used(&md->stats)))
 521		dm_stats_account_io(&md->stats, bio_data_dir(bio),
 522				    bio->bi_iter.bi_sector, bio_sectors(bio),
 523				    false, 0, &io->stats_aux);
 524}
 525
 526static void end_io_acct(struct dm_io *io)
 527{
 528	struct mapped_device *md = io->md;
 529	struct bio *bio = io->bio;
 530	unsigned long duration = jiffies - io->start_time;
 531	int pending;
 532	int rw = bio_data_dir(bio);
 533
 534	generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time);
 535
 536	if (unlikely(dm_stats_used(&md->stats)))
 537		dm_stats_account_io(&md->stats, bio_data_dir(bio),
 538				    bio->bi_iter.bi_sector, bio_sectors(bio),
 539				    true, duration, &io->stats_aux);
 540
 541	/*
 542	 * After this is decremented the bio must not be touched if it is
 543	 * a flush.
 544	 */
 545	pending = atomic_dec_return(&md->pending[rw]);
 546	atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
 547	pending += atomic_read(&md->pending[rw^0x1]);
 548
 549	/* nudge anyone waiting on suspend queue */
 550	if (!pending)
 551		wake_up(&md->wait);
 552}
 553
 554/*
 555 * Add the bio to the list of deferred io.
 556 */
 557static void queue_io(struct mapped_device *md, struct bio *bio)
 558{
 559	unsigned long flags;
 560
 561	spin_lock_irqsave(&md->deferred_lock, flags);
 562	bio_list_add(&md->deferred, bio);
 563	spin_unlock_irqrestore(&md->deferred_lock, flags);
 564	queue_work(md->wq, &md->work);
 565}
 566
 567/*
 568 * Everyone (including functions in this file), should use this
 569 * function to access the md->map field, and make sure they call
 570 * dm_put_live_table() when finished.
 571 */
 572struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
 573{
 574	*srcu_idx = srcu_read_lock(&md->io_barrier);
 575
 576	return srcu_dereference(md->map, &md->io_barrier);
 577}
 578
 579void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
 580{
 581	srcu_read_unlock(&md->io_barrier, srcu_idx);
 582}
 583
 584void dm_sync_table(struct mapped_device *md)
 585{
 586	synchronize_srcu(&md->io_barrier);
 587	synchronize_rcu_expedited();
 588}
 589
 590/*
 591 * A fast alternative to dm_get_live_table/dm_put_live_table.
 592 * The caller must not block between these two functions.
 593 */
 594static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
 595{
 596	rcu_read_lock();
 597	return rcu_dereference(md->map);
 598}
 599
 600static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
 601{
 602	rcu_read_unlock();
 603}
 604
 
 
 605/*
 606 * Open a table device so we can use it as a map destination.
 607 */
 608static int open_table_device(struct table_device *td, dev_t dev,
 609			     struct mapped_device *md)
 610{
 611	static char *_claim_ptr = "I belong to device-mapper";
 612	struct block_device *bdev;
 613
 614	int r;
 615
 616	BUG_ON(td->dm_dev.bdev);
 617
 618	bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr);
 619	if (IS_ERR(bdev))
 620		return PTR_ERR(bdev);
 621
 622	r = bd_link_disk_holder(bdev, dm_disk(md));
 623	if (r) {
 624		blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
 625		return r;
 626	}
 627
 628	td->dm_dev.bdev = bdev;
 
 629	return 0;
 630}
 631
 632/*
 633 * Close a table device that we've been using.
 634 */
 635static void close_table_device(struct table_device *td, struct mapped_device *md)
 636{
 637	if (!td->dm_dev.bdev)
 638		return;
 639
 640	bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
 641	blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
 
 642	td->dm_dev.bdev = NULL;
 
 643}
 644
 645static struct table_device *find_table_device(struct list_head *l, dev_t dev,
 646					      fmode_t mode) {
 
 647	struct table_device *td;
 648
 649	list_for_each_entry(td, l, list)
 650		if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
 651			return td;
 652
 653	return NULL;
 654}
 655
 656int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
 657			struct dm_dev **result) {
 
 658	int r;
 659	struct table_device *td;
 660
 661	mutex_lock(&md->table_devices_lock);
 662	td = find_table_device(&md->table_devices, dev, mode);
 663	if (!td) {
 664		td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
 665		if (!td) {
 666			mutex_unlock(&md->table_devices_lock);
 667			return -ENOMEM;
 668		}
 669
 670		td->dm_dev.mode = mode;
 671		td->dm_dev.bdev = NULL;
 672
 673		if ((r = open_table_device(td, dev, md))) {
 674			mutex_unlock(&md->table_devices_lock);
 675			kfree(td);
 676			return r;
 677		}
 678
 679		format_dev_t(td->dm_dev.name, dev);
 680
 681		atomic_set(&td->count, 0);
 682		list_add(&td->list, &md->table_devices);
 
 
 683	}
 684	atomic_inc(&td->count);
 685	mutex_unlock(&md->table_devices_lock);
 686
 687	*result = &td->dm_dev;
 688	return 0;
 689}
 690EXPORT_SYMBOL_GPL(dm_get_table_device);
 691
 692void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
 693{
 694	struct table_device *td = container_of(d, struct table_device, dm_dev);
 695
 696	mutex_lock(&md->table_devices_lock);
 697	if (atomic_dec_and_test(&td->count)) {
 698		close_table_device(td, md);
 699		list_del(&td->list);
 700		kfree(td);
 701	}
 702	mutex_unlock(&md->table_devices_lock);
 703}
 704EXPORT_SYMBOL(dm_put_table_device);
 705
 706static void free_table_devices(struct list_head *devices)
 707{
 708	struct list_head *tmp, *next;
 709
 710	list_for_each_safe(tmp, next, devices) {
 711		struct table_device *td = list_entry(tmp, struct table_device, list);
 712
 713		DMWARN("dm_destroy: %s still exists with %d references",
 714		       td->dm_dev.name, atomic_read(&td->count));
 715		kfree(td);
 716	}
 717}
 718
 719/*
 720 * Get the geometry associated with a dm device
 721 */
 722int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
 723{
 724	*geo = md->geometry;
 725
 726	return 0;
 727}
 728
 729/*
 730 * Set the geometry of a device.
 731 */
 732int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
 733{
 734	sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
 735
 736	if (geo->start > sz) {
 737		DMWARN("Start sector is beyond the geometry limits.");
 738		return -EINVAL;
 739	}
 740
 741	md->geometry = *geo;
 742
 743	return 0;
 744}
 745
 746/*-----------------------------------------------------------------
 747 * CRUD START:
 748 *   A more elegant soln is in the works that uses the queue
 749 *   merge fn, unfortunately there are a couple of changes to
 750 *   the block layer that I want to make for this.  So in the
 751 *   interests of getting something for people to use I give
 752 *   you this clearly demarcated crap.
 753 *---------------------------------------------------------------*/
 754
 755static int __noflush_suspending(struct mapped_device *md)
 756{
 757	return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
 758}
 759
 760/*
 761 * Decrements the number of outstanding ios that a bio has been
 762 * cloned into, completing the original io if necc.
 763 */
 764static void dec_pending(struct dm_io *io, int error)
 765{
 766	unsigned long flags;
 767	int io_error;
 768	struct bio *bio;
 769	struct mapped_device *md = io->md;
 770
 771	/* Push-back supersedes any I/O errors */
 772	if (unlikely(error)) {
 773		spin_lock_irqsave(&io->endio_lock, flags);
 774		if (!(io->error > 0 && __noflush_suspending(md)))
 775			io->error = error;
 776		spin_unlock_irqrestore(&io->endio_lock, flags);
 777	}
 778
 779	if (atomic_dec_and_test(&io->io_count)) {
 780		if (io->error == DM_ENDIO_REQUEUE) {
 781			/*
 782			 * Target requested pushing back the I/O.
 783			 */
 784			spin_lock_irqsave(&md->deferred_lock, flags);
 785			if (__noflush_suspending(md))
 786				bio_list_add_head(&md->deferred, io->bio);
 
 787			else
 788				/* noflush suspend was interrupted. */
 789				io->error = -EIO;
 790			spin_unlock_irqrestore(&md->deferred_lock, flags);
 791		}
 792
 793		io_error = io->error;
 794		bio = io->bio;
 795		end_io_acct(io);
 796		free_io(md, io);
 797
 798		if (io_error == DM_ENDIO_REQUEUE)
 799			return;
 800
 801		if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
 802			/*
 803			 * Preflush done for flush with data, reissue
 804			 * without REQ_PREFLUSH.
 805			 */
 806			bio->bi_opf &= ~REQ_PREFLUSH;
 807			queue_io(md, bio);
 808		} else {
 809			/* done with normal IO or empty flush */
 810			trace_block_bio_complete(md->queue, bio, io_error);
 811			bio->bi_error = io_error;
 812			bio_endio(bio);
 813		}
 814	}
 815}
 816
 
 
 
 
 
 
 
 
 
 817void disable_write_same(struct mapped_device *md)
 818{
 819	struct queue_limits *limits = dm_get_queue_limits(md);
 820
 821	/* device doesn't really support WRITE SAME, disable it */
 822	limits->max_write_same_sectors = 0;
 823}
 824
 
 
 
 
 
 
 
 
 825static void clone_endio(struct bio *bio)
 826{
 827	int error = bio->bi_error;
 828	int r = error;
 829	struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
 830	struct dm_io *io = tio->io;
 831	struct mapped_device *md = tio->io->md;
 832	dm_endio_fn endio = tio->ti->type->end_io;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 833
 834	if (endio) {
 835		r = endio(tio->ti, bio, error);
 836		if (r < 0 || r == DM_ENDIO_REQUEUE)
 837			/*
 838			 * error and requeue request are handled
 839			 * in dec_pending().
 840			 */
 841			error = r;
 842		else if (r == DM_ENDIO_INCOMPLETE)
 843			/* The target will handle the io */
 844			return;
 845		else if (r) {
 846			DMWARN("unimplemented target endio return value: %d", r);
 847			BUG();
 848		}
 849	}
 850
 851	if (unlikely(r == -EREMOTEIO && (bio_op(bio) == REQ_OP_WRITE_SAME) &&
 852		     !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors))
 853		disable_write_same(md);
 854
 855	free_tio(tio);
 856	dec_pending(io, error);
 857}
 858
 859/*
 860 * Return maximum size of I/O possible at the supplied sector up to the current
 861 * target boundary.
 862 */
 863static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
 864{
 865	sector_t target_offset = dm_target_offset(ti, sector);
 866
 867	return ti->len - target_offset;
 868}
 869
 870static sector_t max_io_len(sector_t sector, struct dm_target *ti)
 871{
 872	sector_t len = max_io_len_target_boundary(sector, ti);
 873	sector_t offset, max_len;
 874
 875	/*
 876	 * Does the target need to split even further?
 877	 */
 878	if (ti->max_io_len) {
 879		offset = dm_target_offset(ti, sector);
 880		if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
 881			max_len = sector_div(offset, ti->max_io_len);
 882		else
 883			max_len = offset & (ti->max_io_len - 1);
 884		max_len = ti->max_io_len - max_len;
 885
 886		if (len > max_len)
 887			len = max_len;
 888	}
 889
 890	return len;
 891}
 892
 893int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
 894{
 895	if (len > UINT_MAX) {
 896		DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
 897		      (unsigned long long)len, UINT_MAX);
 898		ti->error = "Maximum size of target IO is too large";
 899		return -EINVAL;
 900	}
 901
 902	ti->max_io_len = (uint32_t) len;
 903
 904	return 0;
 905}
 906EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
 907
 908static long dm_blk_direct_access(struct block_device *bdev, sector_t sector,
 909				 void **kaddr, pfn_t *pfn, long size)
 
 910{
 911	struct mapped_device *md = bdev->bd_disk->private_data;
 912	struct dm_table *map;
 913	struct dm_target *ti;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 914	int srcu_idx;
 915	long len, ret = -EIO;
 916
 917	map = dm_get_live_table(md, &srcu_idx);
 918	if (!map)
 919		goto out;
 920
 921	ti = dm_table_find_target(map, sector);
 922	if (!dm_target_is_valid(ti))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 923		goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 924
 925	len = max_io_len(sector, ti) << SECTOR_SHIFT;
 926	size = min(len, size);
 927
 928	if (ti->type->direct_access)
 929		ret = ti->type->direct_access(ti, sector, kaddr, pfn, size);
 930out:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 931	dm_put_live_table(md, srcu_idx);
 932	return min(ret, size);
 
 933}
 934
 935/*
 936 * A target may call dm_accept_partial_bio only from the map routine.  It is
 937 * allowed for all bio types except REQ_PREFLUSH.
 
 938 *
 939 * dm_accept_partial_bio informs the dm that the target only wants to process
 940 * additional n_sectors sectors of the bio and the rest of the data should be
 941 * sent in a next bio.
 942 *
 943 * A diagram that explains the arithmetics:
 944 * +--------------------+---------------+-------+
 945 * |         1          |       2       |   3   |
 946 * +--------------------+---------------+-------+
 947 *
 948 * <-------------- *tio->len_ptr --------------->
 949 *                      <------- bi_size ------->
 950 *                      <-- n_sectors -->
 951 *
 952 * Region 1 was already iterated over with bio_advance or similar function.
 953 *	(it may be empty if the target doesn't use bio_advance)
 954 * Region 2 is the remaining bio size that the target wants to process.
 955 *	(it may be empty if region 1 is non-empty, although there is no reason
 956 *	 to make it empty)
 957 * The target requires that region 3 is to be sent in the next bio.
 958 *
 959 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
 960 * the partially processed part (the sum of regions 1+2) must be the same for all
 961 * copies of the bio.
 962 */
 963void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
 964{
 965	struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
 966	unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
 967	BUG_ON(bio->bi_opf & REQ_PREFLUSH);
 968	BUG_ON(bi_size > *tio->len_ptr);
 969	BUG_ON(n_sectors > bi_size);
 970	*tio->len_ptr -= bi_size - n_sectors;
 971	bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
 972}
 973EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
 974
 975/*
 976 * Flush current->bio_list when the target map method blocks.
 977 * This fixes deadlocks in snapshot and possibly in other targets.
 978 */
 979struct dm_offload {
 980	struct blk_plug plug;
 981	struct blk_plug_cb cb;
 982};
 983
 984static void flush_current_bio_list(struct blk_plug_cb *cb, bool from_schedule)
 985{
 986	struct dm_offload *o = container_of(cb, struct dm_offload, cb);
 987	struct bio_list list;
 988	struct bio *bio;
 989	int i;
 990
 991	INIT_LIST_HEAD(&o->cb.list);
 992
 993	if (unlikely(!current->bio_list))
 994		return;
 995
 996	for (i = 0; i < 2; i++) {
 997		list = current->bio_list[i];
 998		bio_list_init(&current->bio_list[i]);
 999
1000		while ((bio = bio_list_pop(&list))) {
1001			struct bio_set *bs = bio->bi_pool;
1002			if (unlikely(!bs) || bs == fs_bio_set) {
1003				bio_list_add(&current->bio_list[i], bio);
1004				continue;
1005			}
1006
1007			spin_lock(&bs->rescue_lock);
1008			bio_list_add(&bs->rescue_list, bio);
1009			queue_work(bs->rescue_workqueue, &bs->rescue_work);
1010			spin_unlock(&bs->rescue_lock);
1011		}
1012	}
1013}
1014
1015static void dm_offload_start(struct dm_offload *o)
1016{
1017	blk_start_plug(&o->plug);
1018	o->cb.callback = flush_current_bio_list;
1019	list_add(&o->cb.list, &current->plug->cb_list);
1020}
1021
1022static void dm_offload_end(struct dm_offload *o)
1023{
1024	list_del(&o->cb.list);
1025	blk_finish_plug(&o->plug);
1026}
1027
1028static void __map_bio(struct dm_target_io *tio)
1029{
1030	int r;
1031	sector_t sector;
1032	struct dm_offload o;
1033	struct bio *clone = &tio->clone;
 
1034	struct dm_target *ti = tio->ti;
 
1035
1036	clone->bi_end_io = clone_endio;
1037
1038	/*
1039	 * Map the clone.  If r == 0 we don't need to do
1040	 * anything, the target has assumed ownership of
1041	 * this io.
1042	 */
1043	atomic_inc(&tio->io->io_count);
1044	sector = clone->bi_iter.bi_sector;
1045
1046	dm_offload_start(&o);
1047	r = ti->type->map(ti, clone);
1048	dm_offload_end(&o);
1049
1050	if (r == DM_MAPIO_REMAPPED) {
 
1051		/* the bio has been remapped so dispatch it */
1052
1053		trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1054				      tio->io->bio->bi_bdev->bd_dev, sector);
1055
1056		generic_make_request(clone);
1057	} else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1058		/* error the io and bail out, or requeue it if needed */
1059		dec_pending(tio->io, r);
 
1060		free_tio(tio);
1061	} else if (r != DM_MAPIO_SUBMITTED) {
 
 
1062		DMWARN("unimplemented target map return value: %d", r);
1063		BUG();
1064	}
 
 
1065}
1066
1067struct clone_info {
1068	struct mapped_device *md;
1069	struct dm_table *map;
1070	struct bio *bio;
1071	struct dm_io *io;
1072	sector_t sector;
1073	unsigned sector_count;
1074};
1075
1076static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1077{
1078	bio->bi_iter.bi_sector = sector;
1079	bio->bi_iter.bi_size = to_bytes(len);
1080}
1081
1082/*
1083 * Creates a bio that consists of range of complete bvecs.
1084 */
1085static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1086		     sector_t sector, unsigned len)
1087{
1088	struct bio *clone = &tio->clone;
1089
1090	__bio_clone_fast(clone, bio);
1091
 
 
1092	if (bio_integrity(bio)) {
1093		int r = bio_integrity_clone(clone, bio, GFP_NOIO);
 
 
 
 
 
 
 
 
 
 
1094		if (r < 0)
1095			return r;
1096	}
1097
1098	bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1099	clone->bi_iter.bi_size = to_bytes(len);
1100
1101	if (bio_integrity(bio))
1102		bio_integrity_trim(clone, 0, len);
1103
1104	return 0;
1105}
1106
1107static struct dm_target_io *alloc_tio(struct clone_info *ci,
1108				      struct dm_target *ti,
1109				      unsigned target_bio_nr)
1110{
1111	struct dm_target_io *tio;
1112	struct bio *clone;
 
 
 
 
 
 
 
 
 
1113
1114	clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1115	tio = container_of(clone, struct dm_target_io, clone);
 
 
 
 
 
 
 
 
1116
1117	tio->io = ci->io;
1118	tio->ti = ti;
1119	tio->target_bio_nr = target_bio_nr;
 
 
 
1120
1121	return tio;
 
 
 
 
1122}
1123
1124static void __clone_and_map_simple_bio(struct clone_info *ci,
1125				       struct dm_target *ti,
1126				       unsigned target_bio_nr, unsigned *len)
1127{
1128	struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1129	struct bio *clone = &tio->clone;
1130
1131	tio->len_ptr = len;
1132
1133	__bio_clone_fast(clone, ci->bio);
1134	if (len)
1135		bio_setup_sector(clone, ci->sector, *len);
1136
1137	__map_bio(tio);
1138}
1139
1140static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1141				  unsigned num_bios, unsigned *len)
1142{
1143	unsigned target_bio_nr;
 
 
 
 
1144
1145	for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1146		__clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
 
 
1147}
1148
1149static int __send_empty_flush(struct clone_info *ci)
1150{
1151	unsigned target_nr = 0;
1152	struct dm_target *ti;
1153
 
 
 
 
 
 
 
 
 
1154	BUG_ON(bio_has_data(ci->bio));
1155	while ((ti = dm_table_get_target(ci->map, target_nr++)))
1156		__send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1157
1158	return 0;
1159}
1160
1161static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1162				     sector_t sector, unsigned *len)
1163{
1164	struct bio *bio = ci->bio;
1165	struct dm_target_io *tio;
1166	unsigned target_bio_nr;
1167	unsigned num_target_bios = 1;
1168	int r = 0;
1169
1170	/*
1171	 * Does the target want to receive duplicate copies of the bio?
1172	 */
1173	if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1174		num_target_bios = ti->num_write_bios(ti, bio);
1175
1176	for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1177		tio = alloc_tio(ci, ti, target_bio_nr);
1178		tio->len_ptr = len;
1179		r = clone_bio(tio, bio, sector, *len);
1180		if (r < 0) {
1181			free_tio(tio);
1182			break;
1183		}
1184		__map_bio(tio);
1185	}
 
1186
1187	return r;
1188}
1189
1190typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1191
1192static unsigned get_num_discard_bios(struct dm_target *ti)
1193{
1194	return ti->num_discard_bios;
1195}
1196
 
 
 
 
 
1197static unsigned get_num_write_same_bios(struct dm_target *ti)
1198{
1199	return ti->num_write_same_bios;
1200}
1201
1202typedef bool (*is_split_required_fn)(struct dm_target *ti);
1203
1204static bool is_split_required_for_discard(struct dm_target *ti)
1205{
1206	return ti->split_discard_bios;
1207}
1208
1209static int __send_changing_extent_only(struct clone_info *ci,
1210				       get_num_bios_fn get_num_bios,
1211				       is_split_required_fn is_split_required)
1212{
1213	struct dm_target *ti;
1214	unsigned len;
1215	unsigned num_bios;
1216
1217	do {
1218		ti = dm_table_find_target(ci->map, ci->sector);
1219		if (!dm_target_is_valid(ti))
1220			return -EIO;
 
 
 
 
 
 
 
 
 
 
 
1221
1222		/*
1223		 * Even though the device advertised support for this type of
1224		 * request, that does not mean every target supports it, and
1225		 * reconfiguration might also have changed that since the
1226		 * check was performed.
1227		 */
1228		num_bios = get_num_bios ? get_num_bios(ti) : 0;
1229		if (!num_bios)
1230			return -EOPNOTSUPP;
1231
1232		if (is_split_required && !is_split_required(ti))
1233			len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1234		else
1235			len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1236
1237		__send_duplicate_bios(ci, ti, num_bios, &len);
 
 
 
1238
1239		ci->sector += len;
1240	} while (ci->sector_count -= len);
 
 
1241
1242	return 0;
 
 
1243}
1244
1245static int __send_discard(struct clone_info *ci)
1246{
1247	return __send_changing_extent_only(ci, get_num_discard_bios,
1248					   is_split_required_for_discard);
 
 
 
 
 
 
 
 
 
 
1249}
1250
1251static int __send_write_same(struct clone_info *ci)
 
1252{
1253	return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
 
 
 
 
 
 
 
 
 
 
 
 
 
1254}
1255
1256/*
1257 * Select the correct strategy for processing a non-flush bio.
1258 */
1259static int __split_and_process_non_flush(struct clone_info *ci)
1260{
1261	struct bio *bio = ci->bio;
1262	struct dm_target *ti;
1263	unsigned len;
1264	int r;
1265
1266	if (unlikely(bio_op(bio) == REQ_OP_DISCARD))
1267		return __send_discard(ci);
1268	else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1269		return __send_write_same(ci);
1270
1271	ti = dm_table_find_target(ci->map, ci->sector);
1272	if (!dm_target_is_valid(ti))
1273		return -EIO;
1274
 
 
 
1275	len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1276
1277	r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1278	if (r < 0)
1279		return r;
1280
1281	ci->sector += len;
1282	ci->sector_count -= len;
1283
1284	return 0;
1285}
1286
 
 
 
 
 
 
 
 
 
 
 
1287/*
1288 * Entry point to split a bio into clones and submit them to the targets.
1289 */
1290static void __split_and_process_bio(struct mapped_device *md,
1291				    struct dm_table *map, struct bio *bio)
1292{
1293	struct clone_info ci;
 
1294	int error = 0;
1295
1296	if (unlikely(!map)) {
1297		bio_io_error(bio);
1298		return;
1299	}
1300
1301	ci.map = map;
1302	ci.md = md;
1303	ci.io = alloc_io(md);
1304	ci.io->error = 0;
1305	atomic_set(&ci.io->io_count, 1);
1306	ci.io->bio = bio;
1307	ci.io->md = md;
1308	spin_lock_init(&ci.io->endio_lock);
1309	ci.sector = bio->bi_iter.bi_sector;
1310
1311	start_io_acct(ci.io);
1312
1313	if (bio->bi_opf & REQ_PREFLUSH) {
1314		ci.bio = &ci.md->flush_bio;
 
 
 
 
1315		ci.sector_count = 0;
1316		error = __send_empty_flush(&ci);
 
1317		/* dec_pending submits any data associated with flush */
 
 
 
 
1318	} else {
1319		ci.bio = bio;
1320		ci.sector_count = bio_sectors(bio);
1321		while (ci.sector_count && !error)
1322			error = __split_and_process_non_flush(&ci);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1323	}
1324
1325	/* drop the extra reference count */
1326	dec_pending(ci.io, error);
 
1327}
1328/*-----------------------------------------------------------------
1329 * CRUD END
1330 *---------------------------------------------------------------*/
1331
1332/*
1333 * The request function that just remaps the bio built up by
1334 * dm_merge_bvec.
1335 */
1336static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1337{
1338	int rw = bio_data_dir(bio);
1339	struct mapped_device *md = q->queuedata;
1340	int srcu_idx;
1341	struct dm_table *map;
1342
 
 
 
 
 
 
 
 
 
 
 
 
1343	map = dm_get_live_table(md, &srcu_idx);
1344
1345	generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
1346
1347	/* if we're suspended, we have to queue this io for later */
1348	if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1349		dm_put_live_table(md, srcu_idx);
1350
1351		if (!(bio->bi_opf & REQ_RAHEAD))
1352			queue_io(md, bio);
1353		else
1354			bio_io_error(bio);
1355		return BLK_QC_T_NONE;
1356	}
1357
1358	__split_and_process_bio(md, map, bio);
 
1359	dm_put_live_table(md, srcu_idx);
1360	return BLK_QC_T_NONE;
1361}
1362
1363static int dm_any_congested(void *congested_data, int bdi_bits)
1364{
1365	int r = bdi_bits;
1366	struct mapped_device *md = congested_data;
1367	struct dm_table *map;
1368
1369	if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1370		if (dm_request_based(md)) {
1371			/*
1372			 * With request-based DM we only need to check the
1373			 * top-level queue for congestion.
1374			 */
1375			r = md->queue->backing_dev_info.wb.state & bdi_bits;
1376		} else {
1377			map = dm_get_live_table_fast(md);
1378			if (map)
1379				r = dm_table_any_congested(map, bdi_bits);
1380			dm_put_live_table_fast(md);
1381		}
1382	}
1383
1384	return r;
1385}
1386
1387/*-----------------------------------------------------------------
1388 * An IDR is used to keep track of allocated minor numbers.
1389 *---------------------------------------------------------------*/
1390static void free_minor(int minor)
1391{
1392	spin_lock(&_minor_lock);
1393	idr_remove(&_minor_idr, minor);
1394	spin_unlock(&_minor_lock);
1395}
1396
1397/*
1398 * See if the device with a specific minor # is free.
1399 */
1400static int specific_minor(int minor)
1401{
1402	int r;
1403
1404	if (minor >= (1 << MINORBITS))
1405		return -EINVAL;
1406
1407	idr_preload(GFP_KERNEL);
1408	spin_lock(&_minor_lock);
1409
1410	r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1411
1412	spin_unlock(&_minor_lock);
1413	idr_preload_end();
1414	if (r < 0)
1415		return r == -ENOSPC ? -EBUSY : r;
1416	return 0;
1417}
1418
1419static int next_free_minor(int *minor)
1420{
1421	int r;
1422
1423	idr_preload(GFP_KERNEL);
1424	spin_lock(&_minor_lock);
1425
1426	r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1427
1428	spin_unlock(&_minor_lock);
1429	idr_preload_end();
1430	if (r < 0)
1431		return r;
1432	*minor = r;
1433	return 0;
1434}
1435
1436static const struct block_device_operations dm_blk_dops;
 
1437
1438static void dm_wq_work(struct work_struct *work);
1439
1440void dm_init_md_queue(struct mapped_device *md)
1441{
1442	/*
1443	 * Request-based dm devices cannot be stacked on top of bio-based dm
1444	 * devices.  The type of this dm device may not have been decided yet.
1445	 * The type is decided at the first table loading time.
1446	 * To prevent problematic device stacking, clear the queue flag
1447	 * for request stacking support until then.
1448	 *
1449	 * This queue is new, so no concurrency on the queue_flags.
1450	 */
1451	queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1452
1453	/*
1454	 * Initialize data that will only be used by a non-blk-mq DM queue
1455	 * - must do so here (in alloc_dev callchain) before queue is used
1456	 */
1457	md->queue->queuedata = md;
1458	md->queue->backing_dev_info.congested_data = md;
1459}
1460
1461void dm_init_normal_md_queue(struct mapped_device *md)
1462{
1463	md->use_blk_mq = false;
1464	dm_init_md_queue(md);
1465
1466	/*
1467	 * Initialize aspects of queue that aren't relevant for blk-mq
1468	 */
1469	md->queue->backing_dev_info.congested_fn = dm_any_congested;
1470	blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1471}
1472
1473static void cleanup_mapped_device(struct mapped_device *md)
1474{
1475	if (md->wq)
1476		destroy_workqueue(md->wq);
1477	if (md->kworker_task)
1478		kthread_stop(md->kworker_task);
1479	mempool_destroy(md->io_pool);
1480	mempool_destroy(md->rq_pool);
1481	if (md->bs)
1482		bioset_free(md->bs);
 
 
1483
1484	if (md->disk) {
1485		spin_lock(&_minor_lock);
1486		md->disk->private_data = NULL;
1487		spin_unlock(&_minor_lock);
1488		del_gendisk(md->disk);
1489		put_disk(md->disk);
1490	}
1491
1492	if (md->queue)
1493		blk_cleanup_queue(md->queue);
1494
1495	cleanup_srcu_struct(&md->io_barrier);
1496
1497	if (md->bdev) {
1498		bdput(md->bdev);
1499		md->bdev = NULL;
1500	}
1501
 
 
 
 
1502	dm_mq_cleanup_mapped_device(md);
1503}
1504
1505/*
1506 * Allocate and initialise a blank device with a given minor.
1507 */
1508static struct mapped_device *alloc_dev(int minor)
1509{
1510	int r, numa_node_id = dm_get_numa_node();
1511	struct mapped_device *md;
1512	void *old_md;
1513
1514	md = kzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1515	if (!md) {
1516		DMWARN("unable to allocate device, out of memory.");
1517		return NULL;
1518	}
1519
1520	if (!try_module_get(THIS_MODULE))
1521		goto bad_module_get;
1522
1523	/* get a minor number for the dev */
1524	if (minor == DM_ANY_MINOR)
1525		r = next_free_minor(&minor);
1526	else
1527		r = specific_minor(minor);
1528	if (r < 0)
1529		goto bad_minor;
1530
1531	r = init_srcu_struct(&md->io_barrier);
1532	if (r < 0)
1533		goto bad_io_barrier;
1534
1535	md->numa_node_id = numa_node_id;
1536	md->use_blk_mq = dm_use_blk_mq_default();
1537	md->init_tio_pdu = false;
1538	md->type = DM_TYPE_NONE;
1539	mutex_init(&md->suspend_lock);
1540	mutex_init(&md->type_lock);
1541	mutex_init(&md->table_devices_lock);
1542	spin_lock_init(&md->deferred_lock);
1543	atomic_set(&md->holders, 1);
1544	atomic_set(&md->open_count, 0);
1545	atomic_set(&md->event_nr, 0);
1546	atomic_set(&md->uevent_seq, 0);
1547	INIT_LIST_HEAD(&md->uevent_list);
1548	INIT_LIST_HEAD(&md->table_devices);
1549	spin_lock_init(&md->uevent_lock);
1550
1551	md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id);
 
 
 
 
 
1552	if (!md->queue)
1553		goto bad;
1554
1555	dm_init_md_queue(md);
1556
1557	md->disk = alloc_disk_node(1, numa_node_id);
1558	if (!md->disk)
1559		goto bad;
1560
1561	atomic_set(&md->pending[0], 0);
1562	atomic_set(&md->pending[1], 0);
1563	init_waitqueue_head(&md->wait);
1564	INIT_WORK(&md->work, dm_wq_work);
1565	init_waitqueue_head(&md->eventq);
1566	init_completion(&md->kobj_holder.completion);
1567	md->kworker_task = NULL;
1568
1569	md->disk->major = _major;
1570	md->disk->first_minor = minor;
1571	md->disk->fops = &dm_blk_dops;
1572	md->disk->queue = md->queue;
1573	md->disk->private_data = md;
1574	sprintf(md->disk->disk_name, "dm-%d", minor);
1575	add_disk(md->disk);
 
 
 
 
 
 
 
 
1576	format_dev_t(md->name, MKDEV(_major, minor));
1577
1578	md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1579	if (!md->wq)
1580		goto bad;
1581
1582	md->bdev = bdget_disk(md->disk, 0);
1583	if (!md->bdev)
1584		goto bad;
1585
1586	bio_init(&md->flush_bio, NULL, 0);
1587	md->flush_bio.bi_bdev = md->bdev;
1588	md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
1589
1590	dm_stats_init(&md->stats);
1591
1592	/* Populate the mapping, nobody knows we exist yet */
1593	spin_lock(&_minor_lock);
1594	old_md = idr_replace(&_minor_idr, md, minor);
1595	spin_unlock(&_minor_lock);
1596
1597	BUG_ON(old_md != MINOR_ALLOCED);
1598
1599	return md;
1600
1601bad:
1602	cleanup_mapped_device(md);
1603bad_io_barrier:
1604	free_minor(minor);
1605bad_minor:
1606	module_put(THIS_MODULE);
1607bad_module_get:
1608	kfree(md);
1609	return NULL;
1610}
1611
1612static void unlock_fs(struct mapped_device *md);
1613
1614static void free_dev(struct mapped_device *md)
1615{
1616	int minor = MINOR(disk_devt(md->disk));
1617
1618	unlock_fs(md);
1619
1620	cleanup_mapped_device(md);
1621
1622	free_table_devices(&md->table_devices);
1623	dm_stats_cleanup(&md->stats);
1624	free_minor(minor);
1625
1626	module_put(THIS_MODULE);
1627	kfree(md);
1628}
1629
1630static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1631{
1632	struct dm_md_mempools *p = dm_table_get_md_mempools(t);
 
1633
1634	if (md->bs) {
1635		/* The md already has necessary mempools. */
1636		if (dm_table_bio_based(t)) {
1637			/*
1638			 * Reload bioset because front_pad may have changed
1639			 * because a different table was loaded.
1640			 */
1641			bioset_free(md->bs);
1642			md->bs = p->bs;
1643			p->bs = NULL;
1644		}
1645		/*
1646		 * There's no need to reload with request-based dm
1647		 * because the size of front_pad doesn't change.
1648		 * Note for future: If you are to reload bioset,
1649		 * prep-ed requests in the queue may refer
1650		 * to bio from the old bioset, so you must walk
1651		 * through the queue to unprep.
1652		 */
1653		goto out;
1654	}
1655
1656	BUG_ON(!p || md->io_pool || md->rq_pool || md->bs);
1657
1658	md->io_pool = p->io_pool;
1659	p->io_pool = NULL;
1660	md->rq_pool = p->rq_pool;
1661	p->rq_pool = NULL;
1662	md->bs = p->bs;
1663	p->bs = NULL;
1664
 
 
 
 
 
 
1665out:
1666	/* mempool bind completed, no longer need any mempools in the table */
1667	dm_table_free_md_mempools(t);
 
1668}
1669
1670/*
1671 * Bind a table to the device.
1672 */
1673static void event_callback(void *context)
1674{
1675	unsigned long flags;
1676	LIST_HEAD(uevents);
1677	struct mapped_device *md = (struct mapped_device *) context;
1678
1679	spin_lock_irqsave(&md->uevent_lock, flags);
1680	list_splice_init(&md->uevent_list, &uevents);
1681	spin_unlock_irqrestore(&md->uevent_lock, flags);
1682
1683	dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1684
1685	atomic_inc(&md->event_nr);
1686	wake_up(&md->eventq);
 
1687}
1688
1689/*
1690 * Protected by md->suspend_lock obtained by dm_swap_table().
1691 */
1692static void __set_size(struct mapped_device *md, sector_t size)
1693{
 
 
1694	set_capacity(md->disk, size);
1695
1696	i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1697}
1698
1699/*
1700 * Returns old map, which caller must destroy.
1701 */
1702static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1703			       struct queue_limits *limits)
1704{
1705	struct dm_table *old_map;
1706	struct request_queue *q = md->queue;
 
1707	sector_t size;
 
1708
1709	lockdep_assert_held(&md->suspend_lock);
1710
1711	size = dm_table_get_size(t);
1712
1713	/*
1714	 * Wipe any geometry if the size of the table changed.
1715	 */
1716	if (size != dm_get_size(md))
1717		memset(&md->geometry, 0, sizeof(md->geometry));
1718
1719	__set_size(md, size);
1720
1721	dm_table_event_callback(t, event_callback, md);
1722
1723	/*
1724	 * The queue hasn't been stopped yet, if the old table type wasn't
1725	 * for request-based during suspension.  So stop it to prevent
1726	 * I/O mapping before resume.
1727	 * This must be done before setting the queue restrictions,
1728	 * because request-based dm may be run just after the setting.
1729	 */
1730	if (dm_table_request_based(t)) {
1731		dm_stop_queue(q);
 
 
1732		/*
1733		 * Leverage the fact that request-based DM targets are
1734		 * immutable singletons and establish md->immutable_target
1735		 * - used to optimize both dm_request_fn and dm_mq_queue_rq
 
1736		 */
1737		md->immutable_target = dm_table_get_immutable_target(t);
1738	}
1739
1740	__bind_mempools(md, t);
 
 
 
 
1741
1742	old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
1743	rcu_assign_pointer(md->map, (void *)t);
1744	md->immutable_target_type = dm_table_get_immutable_target_type(t);
1745
1746	dm_table_set_restrictions(t, q, limits);
1747	if (old_map)
1748		dm_sync_table(md);
1749
 
1750	return old_map;
1751}
1752
1753/*
1754 * Returns unbound table for the caller to free.
1755 */
1756static struct dm_table *__unbind(struct mapped_device *md)
1757{
1758	struct dm_table *map = rcu_dereference_protected(md->map, 1);
1759
1760	if (!map)
1761		return NULL;
1762
1763	dm_table_event_callback(map, NULL, NULL);
1764	RCU_INIT_POINTER(md->map, NULL);
1765	dm_sync_table(md);
1766
1767	return map;
1768}
1769
1770/*
1771 * Constructor for a new device.
1772 */
1773int dm_create(int minor, struct mapped_device **result)
1774{
 
1775	struct mapped_device *md;
1776
1777	md = alloc_dev(minor);
1778	if (!md)
1779		return -ENXIO;
1780
1781	dm_sysfs_init(md);
 
 
 
 
1782
1783	*result = md;
1784	return 0;
1785}
1786
1787/*
1788 * Functions to manage md->type.
1789 * All are required to hold md->type_lock.
1790 */
1791void dm_lock_md_type(struct mapped_device *md)
1792{
1793	mutex_lock(&md->type_lock);
1794}
1795
1796void dm_unlock_md_type(struct mapped_device *md)
1797{
1798	mutex_unlock(&md->type_lock);
1799}
1800
1801void dm_set_md_type(struct mapped_device *md, unsigned type)
1802{
1803	BUG_ON(!mutex_is_locked(&md->type_lock));
1804	md->type = type;
1805}
1806
1807unsigned dm_get_md_type(struct mapped_device *md)
1808{
1809	return md->type;
1810}
1811
1812struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
1813{
1814	return md->immutable_target_type;
1815}
1816
1817/*
1818 * The queue_limits are only valid as long as you have a reference
1819 * count on 'md'.
1820 */
1821struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
1822{
1823	BUG_ON(!atomic_read(&md->holders));
1824	return &md->queue->limits;
1825}
1826EXPORT_SYMBOL_GPL(dm_get_queue_limits);
1827
1828/*
1829 * Setup the DM device's queue based on md's type
1830 */
1831int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
1832{
1833	int r;
1834	unsigned type = dm_get_md_type(md);
 
1835
1836	switch (type) {
1837	case DM_TYPE_REQUEST_BASED:
1838		r = dm_old_init_request_queue(md);
1839		if (r) {
1840			DMERR("Cannot initialize queue for request-based mapped device");
1841			return r;
1842		}
1843		break;
1844	case DM_TYPE_MQ_REQUEST_BASED:
1845		r = dm_mq_init_request_queue(md, t);
1846		if (r) {
1847			DMERR("Cannot initialize queue for request-based dm-mq mapped device");
1848			return r;
1849		}
1850		break;
1851	case DM_TYPE_BIO_BASED:
1852	case DM_TYPE_DAX_BIO_BASED:
1853		dm_init_normal_md_queue(md);
1854		blk_queue_make_request(md->queue, dm_make_request);
1855		/*
1856		 * DM handles splitting bios as needed.  Free the bio_split bioset
1857		 * since it won't be used (saves 1 process per bio-based DM device).
1858		 */
1859		bioset_free(md->queue->bio_split);
1860		md->queue->bio_split = NULL;
1861
1862		if (type == DM_TYPE_DAX_BIO_BASED)
1863			queue_flag_set_unlocked(QUEUE_FLAG_DAX, md->queue);
1864		break;
 
1865	}
 
 
1866
1867	return 0;
1868}
1869
1870struct mapped_device *dm_get_md(dev_t dev)
1871{
1872	struct mapped_device *md;
1873	unsigned minor = MINOR(dev);
1874
1875	if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1876		return NULL;
1877
1878	spin_lock(&_minor_lock);
1879
1880	md = idr_find(&_minor_idr, minor);
1881	if (md) {
1882		if ((md == MINOR_ALLOCED ||
1883		     (MINOR(disk_devt(dm_disk(md))) != minor) ||
1884		     dm_deleting_md(md) ||
1885		     test_bit(DMF_FREEING, &md->flags))) {
1886			md = NULL;
1887			goto out;
1888		}
1889		dm_get(md);
1890	}
1891
1892out:
1893	spin_unlock(&_minor_lock);
1894
1895	return md;
1896}
1897EXPORT_SYMBOL_GPL(dm_get_md);
1898
1899void *dm_get_mdptr(struct mapped_device *md)
1900{
1901	return md->interface_ptr;
1902}
1903
1904void dm_set_mdptr(struct mapped_device *md, void *ptr)
1905{
1906	md->interface_ptr = ptr;
1907}
1908
1909void dm_get(struct mapped_device *md)
1910{
1911	atomic_inc(&md->holders);
1912	BUG_ON(test_bit(DMF_FREEING, &md->flags));
1913}
1914
1915int dm_hold(struct mapped_device *md)
1916{
1917	spin_lock(&_minor_lock);
1918	if (test_bit(DMF_FREEING, &md->flags)) {
1919		spin_unlock(&_minor_lock);
1920		return -EBUSY;
1921	}
1922	dm_get(md);
1923	spin_unlock(&_minor_lock);
1924	return 0;
1925}
1926EXPORT_SYMBOL_GPL(dm_hold);
1927
1928const char *dm_device_name(struct mapped_device *md)
1929{
1930	return md->name;
1931}
1932EXPORT_SYMBOL_GPL(dm_device_name);
1933
1934static void __dm_destroy(struct mapped_device *md, bool wait)
1935{
1936	struct request_queue *q = dm_get_md_queue(md);
1937	struct dm_table *map;
1938	int srcu_idx;
1939
1940	might_sleep();
1941
1942	spin_lock(&_minor_lock);
1943	idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
1944	set_bit(DMF_FREEING, &md->flags);
1945	spin_unlock(&_minor_lock);
1946
1947	blk_set_queue_dying(q);
1948
1949	if (dm_request_based(md) && md->kworker_task)
1950		kthread_flush_worker(&md->kworker);
1951
1952	/*
1953	 * Take suspend_lock so that presuspend and postsuspend methods
1954	 * do not race with internal suspend.
1955	 */
1956	mutex_lock(&md->suspend_lock);
1957	map = dm_get_live_table(md, &srcu_idx);
1958	if (!dm_suspended_md(md)) {
1959		dm_table_presuspend_targets(map);
 
 
1960		dm_table_postsuspend_targets(map);
1961	}
1962	/* dm_put_live_table must be before msleep, otherwise deadlock is possible */
1963	dm_put_live_table(md, srcu_idx);
1964	mutex_unlock(&md->suspend_lock);
1965
1966	/*
1967	 * Rare, but there may be I/O requests still going to complete,
1968	 * for example.  Wait for all references to disappear.
1969	 * No one should increment the reference count of the mapped_device,
1970	 * after the mapped_device state becomes DMF_FREEING.
1971	 */
1972	if (wait)
1973		while (atomic_read(&md->holders))
1974			msleep(1);
1975	else if (atomic_read(&md->holders))
1976		DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
1977		       dm_device_name(md), atomic_read(&md->holders));
1978
1979	dm_sysfs_exit(md);
1980	dm_table_destroy(__unbind(md));
1981	free_dev(md);
1982}
1983
1984void dm_destroy(struct mapped_device *md)
1985{
1986	__dm_destroy(md, true);
1987}
1988
1989void dm_destroy_immediate(struct mapped_device *md)
1990{
1991	__dm_destroy(md, false);
1992}
1993
1994void dm_put(struct mapped_device *md)
1995{
1996	atomic_dec(&md->holders);
1997}
1998EXPORT_SYMBOL_GPL(dm_put);
1999
2000static int dm_wait_for_completion(struct mapped_device *md, long task_state)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2001{
2002	int r = 0;
2003	DEFINE_WAIT(wait);
2004
2005	while (1) {
2006		prepare_to_wait(&md->wait, &wait, task_state);
2007
2008		if (!md_in_flight(md))
2009			break;
2010
2011		if (signal_pending_state(task_state, current)) {
2012			r = -EINTR;
2013			break;
2014		}
2015
2016		io_schedule();
2017	}
2018	finish_wait(&md->wait, &wait);
2019
2020	return r;
2021}
2022
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2023/*
2024 * Process the deferred bios
2025 */
2026static void dm_wq_work(struct work_struct *work)
2027{
2028	struct mapped_device *md = container_of(work, struct mapped_device,
2029						work);
2030	struct bio *c;
2031	int srcu_idx;
2032	struct dm_table *map;
2033
2034	map = dm_get_live_table(md, &srcu_idx);
2035
2036	while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2037		spin_lock_irq(&md->deferred_lock);
2038		c = bio_list_pop(&md->deferred);
2039		spin_unlock_irq(&md->deferred_lock);
2040
2041		if (!c)
2042			break;
2043
2044		if (dm_request_based(md))
2045			generic_make_request(c);
2046		else
2047			__split_and_process_bio(md, map, c);
2048	}
2049
2050	dm_put_live_table(md, srcu_idx);
2051}
2052
2053static void dm_queue_flush(struct mapped_device *md)
2054{
2055	clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2056	smp_mb__after_atomic();
2057	queue_work(md->wq, &md->work);
2058}
2059
2060/*
2061 * Swap in a new table, returning the old one for the caller to destroy.
2062 */
2063struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2064{
2065	struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2066	struct queue_limits limits;
2067	int r;
2068
2069	mutex_lock(&md->suspend_lock);
2070
2071	/* device must be suspended */
2072	if (!dm_suspended_md(md))
2073		goto out;
2074
2075	/*
2076	 * If the new table has no data devices, retain the existing limits.
2077	 * This helps multipath with queue_if_no_path if all paths disappear,
2078	 * then new I/O is queued based on these limits, and then some paths
2079	 * reappear.
2080	 */
2081	if (dm_table_has_no_data_devices(table)) {
2082		live_map = dm_get_live_table_fast(md);
2083		if (live_map)
2084			limits = md->queue->limits;
2085		dm_put_live_table_fast(md);
2086	}
2087
2088	if (!live_map) {
2089		r = dm_calculate_queue_limits(table, &limits);
2090		if (r) {
2091			map = ERR_PTR(r);
2092			goto out;
2093		}
2094	}
2095
2096	map = __bind(md, table, &limits);
 
2097
2098out:
2099	mutex_unlock(&md->suspend_lock);
2100	return map;
2101}
2102
2103/*
2104 * Functions to lock and unlock any filesystem running on the
2105 * device.
2106 */
2107static int lock_fs(struct mapped_device *md)
2108{
2109	int r;
2110
2111	WARN_ON(md->frozen_sb);
2112
2113	md->frozen_sb = freeze_bdev(md->bdev);
2114	if (IS_ERR(md->frozen_sb)) {
2115		r = PTR_ERR(md->frozen_sb);
2116		md->frozen_sb = NULL;
2117		return r;
2118	}
2119
2120	set_bit(DMF_FROZEN, &md->flags);
2121
2122	return 0;
2123}
2124
2125static void unlock_fs(struct mapped_device *md)
2126{
2127	if (!test_bit(DMF_FROZEN, &md->flags))
2128		return;
2129
2130	thaw_bdev(md->bdev, md->frozen_sb);
2131	md->frozen_sb = NULL;
2132	clear_bit(DMF_FROZEN, &md->flags);
2133}
2134
2135/*
2136 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2137 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2138 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2139 *
2140 * If __dm_suspend returns 0, the device is completely quiescent
2141 * now. There is no request-processing activity. All new requests
2142 * are being added to md->deferred list.
2143 *
2144 * Caller must hold md->suspend_lock
2145 */
2146static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2147			unsigned suspend_flags, long task_state,
2148			int dmf_suspended_flag)
2149{
2150	bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2151	bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2152	int r;
2153
2154	lockdep_assert_held(&md->suspend_lock);
2155
2156	/*
2157	 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2158	 * This flag is cleared before dm_suspend returns.
2159	 */
2160	if (noflush)
2161		set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
 
 
2162
2163	/*
2164	 * This gets reverted if there's an error later and the targets
2165	 * provide the .presuspend_undo hook.
2166	 */
2167	dm_table_presuspend_targets(map);
2168
2169	/*
2170	 * Flush I/O to the device.
2171	 * Any I/O submitted after lock_fs() may not be flushed.
2172	 * noflush takes precedence over do_lockfs.
2173	 * (lock_fs() flushes I/Os and waits for them to complete.)
2174	 */
2175	if (!noflush && do_lockfs) {
2176		r = lock_fs(md);
2177		if (r) {
2178			dm_table_presuspend_undo_targets(map);
2179			return r;
2180		}
2181	}
2182
2183	/*
2184	 * Here we must make sure that no processes are submitting requests
2185	 * to target drivers i.e. no one may be executing
2186	 * __split_and_process_bio. This is called from dm_request and
2187	 * dm_wq_work.
2188	 *
2189	 * To get all processes out of __split_and_process_bio in dm_request,
2190	 * we take the write lock. To prevent any process from reentering
2191	 * __split_and_process_bio from dm_request and quiesce the thread
2192	 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2193	 * flush_workqueue(md->wq).
2194	 */
2195	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2196	if (map)
2197		synchronize_srcu(&md->io_barrier);
2198
2199	/*
2200	 * Stop md->queue before flushing md->wq in case request-based
2201	 * dm defers requests to md->wq from md->queue.
2202	 */
2203	if (dm_request_based(md)) {
2204		dm_stop_queue(md->queue);
2205		if (md->kworker_task)
2206			kthread_flush_worker(&md->kworker);
2207	}
2208
2209	flush_workqueue(md->wq);
2210
2211	/*
2212	 * At this point no more requests are entering target request routines.
2213	 * We call dm_wait_for_completion to wait for all existing requests
2214	 * to finish.
2215	 */
2216	r = dm_wait_for_completion(md, task_state);
2217	if (!r)
2218		set_bit(dmf_suspended_flag, &md->flags);
2219
2220	if (noflush)
2221		clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2222	if (map)
2223		synchronize_srcu(&md->io_barrier);
2224
2225	/* were we interrupted ? */
2226	if (r < 0) {
2227		dm_queue_flush(md);
2228
2229		if (dm_request_based(md))
2230			dm_start_queue(md->queue);
2231
2232		unlock_fs(md);
2233		dm_table_presuspend_undo_targets(map);
2234		/* pushback list is already flushed, so skip flush */
2235	}
2236
2237	return r;
2238}
2239
2240/*
2241 * We need to be able to change a mapping table under a mounted
2242 * filesystem.  For example we might want to move some data in
2243 * the background.  Before the table can be swapped with
2244 * dm_bind_table, dm_suspend must be called to flush any in
2245 * flight bios and ensure that any further io gets deferred.
2246 */
2247/*
2248 * Suspend mechanism in request-based dm.
2249 *
2250 * 1. Flush all I/Os by lock_fs() if needed.
2251 * 2. Stop dispatching any I/O by stopping the request_queue.
2252 * 3. Wait for all in-flight I/Os to be completed or requeued.
2253 *
2254 * To abort suspend, start the request_queue.
2255 */
2256int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2257{
2258	struct dm_table *map = NULL;
2259	int r = 0;
2260
2261retry:
2262	mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2263
2264	if (dm_suspended_md(md)) {
2265		r = -EINVAL;
2266		goto out_unlock;
2267	}
2268
2269	if (dm_suspended_internally_md(md)) {
2270		/* already internally suspended, wait for internal resume */
2271		mutex_unlock(&md->suspend_lock);
2272		r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2273		if (r)
2274			return r;
2275		goto retry;
2276	}
2277
2278	map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2279
2280	r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2281	if (r)
2282		goto out_unlock;
2283
 
2284	dm_table_postsuspend_targets(map);
 
2285
2286out_unlock:
2287	mutex_unlock(&md->suspend_lock);
2288	return r;
2289}
2290
2291static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2292{
2293	if (map) {
2294		int r = dm_table_resume_targets(map);
2295		if (r)
2296			return r;
2297	}
2298
2299	dm_queue_flush(md);
2300
2301	/*
2302	 * Flushing deferred I/Os must be done after targets are resumed
2303	 * so that mapping of targets can work correctly.
2304	 * Request-based dm is queueing the deferred I/Os in its request_queue.
2305	 */
2306	if (dm_request_based(md))
2307		dm_start_queue(md->queue);
2308
2309	unlock_fs(md);
2310
2311	return 0;
2312}
2313
2314int dm_resume(struct mapped_device *md)
2315{
2316	int r;
2317	struct dm_table *map = NULL;
2318
2319retry:
2320	r = -EINVAL;
2321	mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2322
2323	if (!dm_suspended_md(md))
2324		goto out;
2325
2326	if (dm_suspended_internally_md(md)) {
2327		/* already internally suspended, wait for internal resume */
2328		mutex_unlock(&md->suspend_lock);
2329		r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2330		if (r)
2331			return r;
2332		goto retry;
2333	}
2334
2335	map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2336	if (!map || !dm_table_get_size(map))
2337		goto out;
2338
2339	r = __dm_resume(md, map);
2340	if (r)
2341		goto out;
2342
2343	clear_bit(DMF_SUSPENDED, &md->flags);
2344out:
2345	mutex_unlock(&md->suspend_lock);
2346
2347	return r;
2348}
2349
2350/*
2351 * Internal suspend/resume works like userspace-driven suspend. It waits
2352 * until all bios finish and prevents issuing new bios to the target drivers.
2353 * It may be used only from the kernel.
2354 */
2355
2356static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2357{
2358	struct dm_table *map = NULL;
2359
 
 
2360	if (md->internal_suspend_count++)
2361		return; /* nested internal suspend */
2362
2363	if (dm_suspended_md(md)) {
2364		set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2365		return; /* nest suspend */
2366	}
2367
2368	map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2369
2370	/*
2371	 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2372	 * supported.  Properly supporting a TASK_INTERRUPTIBLE internal suspend
2373	 * would require changing .presuspend to return an error -- avoid this
2374	 * until there is a need for more elaborate variants of internal suspend.
2375	 */
2376	(void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2377			    DMF_SUSPENDED_INTERNALLY);
2378
 
2379	dm_table_postsuspend_targets(map);
 
2380}
2381
2382static void __dm_internal_resume(struct mapped_device *md)
2383{
2384	BUG_ON(!md->internal_suspend_count);
2385
2386	if (--md->internal_suspend_count)
2387		return; /* resume from nested internal suspend */
2388
2389	if (dm_suspended_md(md))
2390		goto done; /* resume from nested suspend */
2391
2392	/*
2393	 * NOTE: existing callers don't need to call dm_table_resume_targets
2394	 * (which may fail -- so best to avoid it for now by passing NULL map)
2395	 */
2396	(void) __dm_resume(md, NULL);
2397
2398done:
2399	clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2400	smp_mb__after_atomic();
2401	wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2402}
2403
2404void dm_internal_suspend_noflush(struct mapped_device *md)
2405{
2406	mutex_lock(&md->suspend_lock);
2407	__dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2408	mutex_unlock(&md->suspend_lock);
2409}
2410EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2411
2412void dm_internal_resume(struct mapped_device *md)
2413{
2414	mutex_lock(&md->suspend_lock);
2415	__dm_internal_resume(md);
2416	mutex_unlock(&md->suspend_lock);
2417}
2418EXPORT_SYMBOL_GPL(dm_internal_resume);
2419
2420/*
2421 * Fast variants of internal suspend/resume hold md->suspend_lock,
2422 * which prevents interaction with userspace-driven suspend.
2423 */
2424
2425void dm_internal_suspend_fast(struct mapped_device *md)
2426{
2427	mutex_lock(&md->suspend_lock);
2428	if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2429		return;
2430
2431	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2432	synchronize_srcu(&md->io_barrier);
2433	flush_workqueue(md->wq);
2434	dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2435}
2436EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2437
2438void dm_internal_resume_fast(struct mapped_device *md)
2439{
2440	if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2441		goto done;
2442
2443	dm_queue_flush(md);
2444
2445done:
2446	mutex_unlock(&md->suspend_lock);
2447}
2448EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2449
2450/*-----------------------------------------------------------------
2451 * Event notification.
2452 *---------------------------------------------------------------*/
2453int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2454		       unsigned cookie)
2455{
 
 
2456	char udev_cookie[DM_COOKIE_LENGTH];
2457	char *envp[] = { udev_cookie, NULL };
2458
 
 
2459	if (!cookie)
2460		return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2461	else {
2462		snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2463			 DM_COOKIE_ENV_VAR_NAME, cookie);
2464		return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2465					  action, envp);
2466	}
 
 
 
 
2467}
2468
2469uint32_t dm_next_uevent_seq(struct mapped_device *md)
2470{
2471	return atomic_add_return(1, &md->uevent_seq);
2472}
2473
2474uint32_t dm_get_event_nr(struct mapped_device *md)
2475{
2476	return atomic_read(&md->event_nr);
2477}
2478
2479int dm_wait_event(struct mapped_device *md, int event_nr)
2480{
2481	return wait_event_interruptible(md->eventq,
2482			(event_nr != atomic_read(&md->event_nr)));
2483}
2484
2485void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2486{
2487	unsigned long flags;
2488
2489	spin_lock_irqsave(&md->uevent_lock, flags);
2490	list_add(elist, &md->uevent_list);
2491	spin_unlock_irqrestore(&md->uevent_lock, flags);
2492}
2493
2494/*
2495 * The gendisk is only valid as long as you have a reference
2496 * count on 'md'.
2497 */
2498struct gendisk *dm_disk(struct mapped_device *md)
2499{
2500	return md->disk;
2501}
2502EXPORT_SYMBOL_GPL(dm_disk);
2503
2504struct kobject *dm_kobject(struct mapped_device *md)
2505{
2506	return &md->kobj_holder.kobj;
2507}
2508
2509struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2510{
2511	struct mapped_device *md;
2512
2513	md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2514
2515	if (test_bit(DMF_FREEING, &md->flags) ||
2516	    dm_deleting_md(md))
2517		return NULL;
 
 
 
 
 
2518
2519	dm_get(md);
2520	return md;
2521}
2522
2523int dm_suspended_md(struct mapped_device *md)
2524{
2525	return test_bit(DMF_SUSPENDED, &md->flags);
2526}
2527
 
 
 
 
 
2528int dm_suspended_internally_md(struct mapped_device *md)
2529{
2530	return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2531}
2532
2533int dm_test_deferred_remove_flag(struct mapped_device *md)
2534{
2535	return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2536}
2537
2538int dm_suspended(struct dm_target *ti)
2539{
2540	return dm_suspended_md(dm_table_get_md(ti->table));
2541}
2542EXPORT_SYMBOL_GPL(dm_suspended);
2543
 
 
 
 
 
 
2544int dm_noflush_suspending(struct dm_target *ti)
2545{
2546	return __noflush_suspending(dm_table_get_md(ti->table));
2547}
2548EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2549
2550struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type,
2551					    unsigned integrity, unsigned per_io_data_size)
 
2552{
2553	struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2554	struct kmem_cache *cachep = NULL;
2555	unsigned int pool_size = 0;
2556	unsigned int front_pad;
 
2557
2558	if (!pools)
2559		return NULL;
2560
2561	switch (type) {
2562	case DM_TYPE_BIO_BASED:
2563	case DM_TYPE_DAX_BIO_BASED:
2564		cachep = _io_cache;
2565		pool_size = dm_get_reserved_bio_based_ios();
2566		front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
 
 
 
 
 
 
2567		break;
2568	case DM_TYPE_REQUEST_BASED:
2569		cachep = _rq_tio_cache;
2570		pool_size = dm_get_reserved_rq_based_ios();
2571		pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache);
2572		if (!pools->rq_pool)
2573			goto out;
2574		/* fall through to setup remaining rq-based pools */
2575	case DM_TYPE_MQ_REQUEST_BASED:
2576		if (!pool_size)
2577			pool_size = dm_get_reserved_rq_based_ios();
2578		front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2579		/* per_io_data_size is used for blk-mq pdu at queue allocation */
2580		break;
2581	default:
2582		BUG();
2583	}
2584
2585	if (cachep) {
2586		pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
2587		if (!pools->io_pool)
2588			goto out;
2589	}
2590
2591	pools->bs = bioset_create_nobvec(pool_size, front_pad);
2592	if (!pools->bs)
2593		goto out;
2594
2595	if (integrity && bioset_integrity_create(pools->bs, pool_size))
2596		goto out;
2597
2598	return pools;
2599
2600out:
2601	dm_free_md_mempools(pools);
2602
2603	return NULL;
2604}
2605
2606void dm_free_md_mempools(struct dm_md_mempools *pools)
2607{
2608	if (!pools)
2609		return;
2610
2611	mempool_destroy(pools->io_pool);
2612	mempool_destroy(pools->rq_pool);
2613
2614	if (pools->bs)
2615		bioset_free(pools->bs);
2616
2617	kfree(pools);
2618}
2619
2620struct dm_pr {
2621	u64	old_key;
2622	u64	new_key;
2623	u32	flags;
2624	bool	fail_early;
2625};
2626
2627static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2628		      void *data)
2629{
2630	struct mapped_device *md = bdev->bd_disk->private_data;
2631	struct dm_table *table;
2632	struct dm_target *ti;
2633	int ret = -ENOTTY, srcu_idx;
2634
2635	table = dm_get_live_table(md, &srcu_idx);
2636	if (!table || !dm_table_get_size(table))
2637		goto out;
2638
2639	/* We only support devices that have a single target */
2640	if (dm_table_get_num_targets(table) != 1)
2641		goto out;
2642	ti = dm_table_get_target(table, 0);
2643
2644	ret = -EINVAL;
2645	if (!ti->type->iterate_devices)
2646		goto out;
2647
2648	ret = ti->type->iterate_devices(ti, fn, data);
2649out:
2650	dm_put_live_table(md, srcu_idx);
2651	return ret;
2652}
2653
2654/*
2655 * For register / unregister we need to manually call out to every path.
2656 */
2657static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2658			    sector_t start, sector_t len, void *data)
2659{
2660	struct dm_pr *pr = data;
2661	const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2662
2663	if (!ops || !ops->pr_register)
2664		return -EOPNOTSUPP;
2665	return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2666}
2667
2668static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2669			  u32 flags)
2670{
2671	struct dm_pr pr = {
2672		.old_key	= old_key,
2673		.new_key	= new_key,
2674		.flags		= flags,
2675		.fail_early	= true,
2676	};
2677	int ret;
2678
2679	ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2680	if (ret && new_key) {
2681		/* unregister all paths if we failed to register any path */
2682		pr.old_key = new_key;
2683		pr.new_key = 0;
2684		pr.flags = 0;
2685		pr.fail_early = false;
2686		dm_call_pr(bdev, __dm_pr_register, &pr);
2687	}
2688
2689	return ret;
2690}
2691
2692static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2693			 u32 flags)
2694{
2695	struct mapped_device *md = bdev->bd_disk->private_data;
2696	const struct pr_ops *ops;
2697	fmode_t mode;
2698	int r;
2699
2700	r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2701	if (r < 0)
2702		return r;
2703
2704	ops = bdev->bd_disk->fops->pr_ops;
2705	if (ops && ops->pr_reserve)
2706		r = ops->pr_reserve(bdev, key, type, flags);
2707	else
2708		r = -EOPNOTSUPP;
2709
2710	bdput(bdev);
2711	return r;
2712}
2713
2714static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2715{
2716	struct mapped_device *md = bdev->bd_disk->private_data;
2717	const struct pr_ops *ops;
2718	fmode_t mode;
2719	int r;
2720
2721	r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2722	if (r < 0)
2723		return r;
2724
2725	ops = bdev->bd_disk->fops->pr_ops;
2726	if (ops && ops->pr_release)
2727		r = ops->pr_release(bdev, key, type);
2728	else
2729		r = -EOPNOTSUPP;
2730
2731	bdput(bdev);
2732	return r;
2733}
2734
2735static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2736			 enum pr_type type, bool abort)
2737{
2738	struct mapped_device *md = bdev->bd_disk->private_data;
2739	const struct pr_ops *ops;
2740	fmode_t mode;
2741	int r;
2742
2743	r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2744	if (r < 0)
2745		return r;
2746
2747	ops = bdev->bd_disk->fops->pr_ops;
2748	if (ops && ops->pr_preempt)
2749		r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
2750	else
2751		r = -EOPNOTSUPP;
2752
2753	bdput(bdev);
2754	return r;
2755}
2756
2757static int dm_pr_clear(struct block_device *bdev, u64 key)
2758{
2759	struct mapped_device *md = bdev->bd_disk->private_data;
2760	const struct pr_ops *ops;
2761	fmode_t mode;
2762	int r;
2763
2764	r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2765	if (r < 0)
2766		return r;
2767
2768	ops = bdev->bd_disk->fops->pr_ops;
2769	if (ops && ops->pr_clear)
2770		r = ops->pr_clear(bdev, key);
2771	else
2772		r = -EOPNOTSUPP;
2773
2774	bdput(bdev);
2775	return r;
2776}
2777
2778static const struct pr_ops dm_pr_ops = {
2779	.pr_register	= dm_pr_register,
2780	.pr_reserve	= dm_pr_reserve,
2781	.pr_release	= dm_pr_release,
2782	.pr_preempt	= dm_pr_preempt,
2783	.pr_clear	= dm_pr_clear,
2784};
2785
2786static const struct block_device_operations dm_blk_dops = {
 
2787	.open = dm_blk_open,
2788	.release = dm_blk_close,
2789	.ioctl = dm_blk_ioctl,
2790	.direct_access = dm_blk_direct_access,
2791	.getgeo = dm_blk_getgeo,
 
2792	.pr_ops = &dm_pr_ops,
2793	.owner = THIS_MODULE
 
 
 
 
 
 
 
 
2794};
2795
2796/*
2797 * module hooks
2798 */
2799module_init(dm_init);
2800module_exit(dm_exit);
2801
2802module_param(major, uint, 0);
2803MODULE_PARM_DESC(major, "The major number of the device mapper");
2804
2805module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2806MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2807
2808module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
2809MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
2810
2811MODULE_DESCRIPTION(DM_NAME " driver");
2812MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2813MODULE_LICENSE("GPL");