1/*
   2 * Copyright (C) 2001 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
  11#include <linux/module.h>
  12#include <linux/vmalloc.h>
  13#include <linux/blkdev.h>
  14#include <linux/blk-integrity.h>
  15#include <linux/namei.h>
  16#include <linux/ctype.h>
  17#include <linux/string.h>
  18#include <linux/slab.h>
  19#include <linux/interrupt.h>
  20#include <linux/mutex.h>
  21#include <linux/delay.h>
  22#include <linux/atomic.h>
  23#include <linux/blk-mq.h>
  24#include <linux/mount.h>
  25#include <linux/dax.h>
  26
  27#define DM_MSG_PREFIX "table"
  28
  29#define NODE_SIZE L1_CACHE_BYTES
  30#define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
  31#define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
  32
  33/*
  34 * Similar to ceiling(log_size(n))
  35 */
  36static unsigned int int_log(unsigned int n, unsigned int base)
  37{
  38	int result = 0;
  39
  40	while (n > 1) {
  41		n = dm_div_up(n, base);
  42		result++;
  43	}
  44
  45	return result;
  46}
  47
  48/*
  49 * Calculate the index of the child node of the n'th node k'th key.
  50 */
  51static inline unsigned int get_child(unsigned int n, unsigned int k)
  52{
  53	return (n * CHILDREN_PER_NODE) + k;
  54}
  55
  56/*
  57 * Return the n'th node of level l from table t.
  58 */
  59static inline sector_t *get_node(struct dm_table *t,
  60				 unsigned int l, unsigned int n)
  61{
  62	return t->index[l] + (n * KEYS_PER_NODE);
  63}
  64
  65/*
  66 * Return the highest key that you could lookup from the n'th
  67 * node on level l of the btree.
  68 */
  69static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
  70{
  71	for (; l < t->depth - 1; l++)
  72		n = get_child(n, CHILDREN_PER_NODE - 1);
  73
  74	if (n >= t->counts[l])
  75		return (sector_t) - 1;
  76
  77	return get_node(t, l, n)[KEYS_PER_NODE - 1];
  78}
  79
  80/*
  81 * Fills in a level of the btree based on the highs of the level
  82 * below it.
  83 */
  84static int setup_btree_index(unsigned int l, struct dm_table *t)
  85{
  86	unsigned int n, k;
  87	sector_t *node;
  88
  89	for (n = 0U; n < t->counts[l]; n++) {
  90		node = get_node(t, l, n);
  91
  92		for (k = 0U; k < KEYS_PER_NODE; k++)
  93			node[k] = high(t, l + 1, get_child(n, k));
  94	}
  95
  96	return 0;
  97}
  98
  99/*
 100 * highs, and targets are managed as dynamic arrays during a
 101 * table load.
 102 */
 103static int alloc_targets(struct dm_table *t, unsigned int num)
 104{
 105	sector_t *n_highs;
 106	struct dm_target *n_targets;
 107
 108	/*
 109	 * Allocate both the target array and offset array at once.
 110	 */
 111	n_highs = kvcalloc(num, sizeof(struct dm_target) + sizeof(sector_t),
 112			   GFP_KERNEL);
 113	if (!n_highs)
 114		return -ENOMEM;
 115
 116	n_targets = (struct dm_target *) (n_highs + num);
 117
 118	memset(n_highs, -1, sizeof(*n_highs) * num);
 119	kvfree(t->highs);
 120
 121	t->num_allocated = num;
 122	t->highs = n_highs;
 123	t->targets = n_targets;
 124
 125	return 0;
 126}
 127
 128int dm_table_create(struct dm_table **result, fmode_t mode,
 129		    unsigned num_targets, struct mapped_device *md)
 130{
 131	struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
 
 
 
 
 
 132
 133	if (!t)
 134		return -ENOMEM;
 135
 136	INIT_LIST_HEAD(&t->devices);
 
 137
 138	if (!num_targets)
 139		num_targets = KEYS_PER_NODE;
 140
 141	num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
 142
 143	if (!num_targets) {
 144		kfree(t);
 145		return -ENOMEM;
 146	}
 147
 148	if (alloc_targets(t, num_targets)) {
 149		kfree(t);
 150		return -ENOMEM;
 151	}
 152
 153	t->type = DM_TYPE_NONE;
 154	t->mode = mode;
 155	t->md = md;
 
 156	*result = t;
 157	return 0;
 158}
 159
 160static void free_devices(struct list_head *devices, struct mapped_device *md)
 161{
 162	struct list_head *tmp, *next;
 163
 164	list_for_each_safe(tmp, next, devices) {
 165		struct dm_dev_internal *dd =
 166		    list_entry(tmp, struct dm_dev_internal, list);
 167		DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
 168		       dm_device_name(md), dd->dm_dev->name);
 169		dm_put_table_device(md, dd->dm_dev);
 170		kfree(dd);
 171	}
 172}
 173
 174static void dm_table_destroy_crypto_profile(struct dm_table *t);
 175
 176void dm_table_destroy(struct dm_table *t)
 177{
 178	if (!t)
 179		return;
 180
 181	/* free the indexes */
 182	if (t->depth >= 2)
 183		kvfree(t->index[t->depth - 2]);
 184
 185	/* free the targets */
 186	for (unsigned int i = 0; i < t->num_targets; i++) {
 187		struct dm_target *ti = dm_table_get_target(t, i);
 188
 189		if (ti->type->dtr)
 190			ti->type->dtr(ti);
 191
 192		dm_put_target_type(ti->type);
 193	}
 194
 195	kvfree(t->highs);
 196
 197	/* free the device list */
 198	free_devices(&t->devices, t->md);
 199
 200	dm_free_md_mempools(t->mempools);
 201
 202	dm_table_destroy_crypto_profile(t);
 203
 204	kfree(t);
 205}
 206
 207/*
 208 * See if we've already got a device in the list.
 209 */
 210static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
 211{
 212	struct dm_dev_internal *dd;
 213
 214	list_for_each_entry (dd, l, list)
 215		if (dd->dm_dev->bdev->bd_dev == dev)
 216			return dd;
 217
 218	return NULL;
 219}
 220
 221/*
 222 * If possible, this checks an area of a destination device is invalid.
 223 */
 224static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
 225				  sector_t start, sector_t len, void *data)
 226{
 227	struct queue_limits *limits = data;
 228	struct block_device *bdev = dev->bdev;
 229	sector_t dev_size = bdev_nr_sectors(bdev);
 230	unsigned short logical_block_size_sectors =
 231		limits->logical_block_size >> SECTOR_SHIFT;
 232
 233	if (!dev_size)
 234		return 0;
 235
 236	if ((start >= dev_size) || (start + len > dev_size)) {
 237		DMERR("%s: %pg too small for target: "
 238		      "start=%llu, len=%llu, dev_size=%llu",
 239		      dm_device_name(ti->table->md), bdev,
 240		      (unsigned long long)start,
 241		      (unsigned long long)len,
 242		      (unsigned long long)dev_size);
 243		return 1;
 244	}
 245
 246	/*
 247	 * If the target is mapped to zoned block device(s), check
 248	 * that the zones are not partially mapped.
 249	 */
 250	if (bdev_is_zoned(bdev)) {
 251		unsigned int zone_sectors = bdev_zone_sectors(bdev);
 252
 253		if (start & (zone_sectors - 1)) {
 254			DMERR("%s: start=%llu not aligned to h/w zone size %u of %pg",
 255			      dm_device_name(ti->table->md),
 256			      (unsigned long long)start,
 257			      zone_sectors, bdev);
 258			return 1;
 259		}
 260
 261		/*
 262		 * Note: The last zone of a zoned block device may be smaller
 263		 * than other zones. So for a target mapping the end of a
 264		 * zoned block device with such a zone, len would not be zone
 265		 * aligned. We do not allow such last smaller zone to be part
 266		 * of the mapping here to ensure that mappings with multiple
 267		 * devices do not end up with a smaller zone in the middle of
 268		 * the sector range.
 269		 */
 270		if (len & (zone_sectors - 1)) {
 271			DMERR("%s: len=%llu not aligned to h/w zone size %u of %pg",
 272			      dm_device_name(ti->table->md),
 273			      (unsigned long long)len,
 274			      zone_sectors, bdev);
 275			return 1;
 276		}
 277	}
 278
 279	if (logical_block_size_sectors <= 1)
 280		return 0;
 281
 282	if (start & (logical_block_size_sectors - 1)) {
 283		DMERR("%s: start=%llu not aligned to h/w "
 284		      "logical block size %u of %pg",
 285		      dm_device_name(ti->table->md),
 286		      (unsigned long long)start,
 287		      limits->logical_block_size, bdev);
 288		return 1;
 289	}
 290
 291	if (len & (logical_block_size_sectors - 1)) {
 292		DMERR("%s: len=%llu not aligned to h/w "
 293		      "logical block size %u of %pg",
 294		      dm_device_name(ti->table->md),
 295		      (unsigned long long)len,
 296		      limits->logical_block_size, bdev);
 297		return 1;
 298	}
 299
 300	return 0;
 301}
 302
 303/*
 304 * This upgrades the mode on an already open dm_dev, being
 305 * careful to leave things as they were if we fail to reopen the
 306 * device and not to touch the existing bdev field in case
 307 * it is accessed concurrently.
 308 */
 309static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
 310			struct mapped_device *md)
 311{
 312	int r;
 313	struct dm_dev *old_dev, *new_dev;
 314
 315	old_dev = dd->dm_dev;
 316
 317	r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
 318				dd->dm_dev->mode | new_mode, &new_dev);
 319	if (r)
 320		return r;
 321
 322	dd->dm_dev = new_dev;
 323	dm_put_table_device(md, old_dev);
 324
 325	return 0;
 326}
 327
 328/*
 329 * Convert the path to a device
 
 330 */
 331dev_t dm_get_dev_t(const char *path)
 332{
 
 333	dev_t dev;
 
 
 334
 335	if (lookup_bdev(path, &dev))
 336		dev = name_to_dev_t(path);
 337	return dev;
 
 
 
 
 
 
 
 
 
 
 
 
 
 338}
 339EXPORT_SYMBOL_GPL(dm_get_dev_t);
 340
 341/*
 342 * Add a device to the list, or just increment the usage count if
 343 * it's already present.
 344 */
 345int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
 346		  struct dm_dev **result)
 347{
 348	int r;
 349	dev_t dev;
 350	unsigned int major, minor;
 351	char dummy;
 352	struct dm_dev_internal *dd;
 353	struct dm_table *t = ti->table;
 354
 355	BUG_ON(!t);
 356
 357	if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
 358		/* Extract the major/minor numbers */
 359		dev = MKDEV(major, minor);
 360		if (MAJOR(dev) != major || MINOR(dev) != minor)
 361			return -EOVERFLOW;
 362	} else {
 363		dev = dm_get_dev_t(path);
 364		if (!dev)
 365			return -ENODEV;
 366	}
 367
 368	dd = find_device(&t->devices, dev);
 369	if (!dd) {
 370		dd = kmalloc(sizeof(*dd), GFP_KERNEL);
 371		if (!dd)
 372			return -ENOMEM;
 
 
 373
 374		if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
 
 375			kfree(dd);
 376			return r;
 377		}
 378
 379		refcount_set(&dd->count, 1);
 380		list_add(&dd->list, &t->devices);
 381		goto out;
 382
 383	} else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
 384		r = upgrade_mode(dd, mode, t->md);
 385		if (r)
 386			return r;
 387	}
 388	refcount_inc(&dd->count);
 389out:
 
 390	*result = dd->dm_dev;
 391	return 0;
 
 
 
 
 392}
 393EXPORT_SYMBOL(dm_get_device);
 394
 395static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
 396				sector_t start, sector_t len, void *data)
 397{
 398	struct queue_limits *limits = data;
 399	struct block_device *bdev = dev->bdev;
 400	struct request_queue *q = bdev_get_queue(bdev);
 401
 402	if (unlikely(!q)) {
 403		DMWARN("%s: Cannot set limits for nonexistent device %pg",
 404		       dm_device_name(ti->table->md), bdev);
 405		return 0;
 406	}
 407
 408	if (blk_stack_limits(limits, &q->limits,
 409			get_start_sect(bdev) + start) < 0)
 410		DMWARN("%s: adding target device %pg caused an alignment inconsistency: "
 411		       "physical_block_size=%u, logical_block_size=%u, "
 412		       "alignment_offset=%u, start=%llu",
 413		       dm_device_name(ti->table->md), bdev,
 414		       q->limits.physical_block_size,
 415		       q->limits.logical_block_size,
 416		       q->limits.alignment_offset,
 417		       (unsigned long long) start << SECTOR_SHIFT);
 
 
 
 
 
 
 
 418	return 0;
 419}
 420
 421/*
 422 * Decrement a device's use count and remove it if necessary.
 423 */
 424void dm_put_device(struct dm_target *ti, struct dm_dev *d)
 425{
 426	int found = 0;
 427	struct list_head *devices = &ti->table->devices;
 
 428	struct dm_dev_internal *dd;
 429
 
 
 430	list_for_each_entry(dd, devices, list) {
 431		if (dd->dm_dev == d) {
 432			found = 1;
 433			break;
 434		}
 435	}
 436	if (!found) {
 437		DMERR("%s: device %s not in table devices list",
 438		      dm_device_name(ti->table->md), d->name);
 439		return;
 440	}
 441	if (refcount_dec_and_test(&dd->count)) {
 442		dm_put_table_device(ti->table->md, d);
 443		list_del(&dd->list);
 444		kfree(dd);
 445	}
 
 
 
 446}
 447EXPORT_SYMBOL(dm_put_device);
 448
 449/*
 450 * Checks to see if the target joins onto the end of the table.
 451 */
 452static int adjoin(struct dm_table *t, struct dm_target *ti)
 453{
 454	struct dm_target *prev;
 455
 456	if (!t->num_targets)
 457		return !ti->begin;
 458
 459	prev = &t->targets[t->num_targets - 1];
 460	return (ti->begin == (prev->begin + prev->len));
 461}
 462
 463/*
 464 * Used to dynamically allocate the arg array.
 465 *
 466 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
 467 * process messages even if some device is suspended. These messages have a
 468 * small fixed number of arguments.
 469 *
 470 * On the other hand, dm-switch needs to process bulk data using messages and
 471 * excessive use of GFP_NOIO could cause trouble.
 472 */
 473static char **realloc_argv(unsigned *size, char **old_argv)
 474{
 475	char **argv;
 476	unsigned new_size;
 477	gfp_t gfp;
 478
 479	if (*size) {
 480		new_size = *size * 2;
 481		gfp = GFP_KERNEL;
 482	} else {
 483		new_size = 8;
 484		gfp = GFP_NOIO;
 485	}
 486	argv = kmalloc_array(new_size, sizeof(*argv), gfp);
 487	if (argv && old_argv) {
 488		memcpy(argv, old_argv, *size * sizeof(*argv));
 489		*size = new_size;
 490	}
 491
 492	kfree(old_argv);
 493	return argv;
 494}
 495
 496/*
 497 * Destructively splits up the argument list to pass to ctr.
 498 */
 499int dm_split_args(int *argc, char ***argvp, char *input)
 500{
 501	char *start, *end = input, *out, **argv = NULL;
 502	unsigned array_size = 0;
 503
 504	*argc = 0;
 505
 506	if (!input) {
 507		*argvp = NULL;
 508		return 0;
 509	}
 510
 511	argv = realloc_argv(&array_size, argv);
 512	if (!argv)
 513		return -ENOMEM;
 514
 515	while (1) {
 516		/* Skip whitespace */
 517		start = skip_spaces(end);
 518
 519		if (!*start)
 520			break;	/* success, we hit the end */
 521
 522		/* 'out' is used to remove any back-quotes */
 523		end = out = start;
 524		while (*end) {
 525			/* Everything apart from '\0' can be quoted */
 526			if (*end == '\\' && *(end + 1)) {
 527				*out++ = *(end + 1);
 528				end += 2;
 529				continue;
 530			}
 531
 532			if (isspace(*end))
 533				break;	/* end of token */
 534
 535			*out++ = *end++;
 536		}
 537
 538		/* have we already filled the array ? */
 539		if ((*argc + 1) > array_size) {
 540			argv = realloc_argv(&array_size, argv);
 541			if (!argv)
 542				return -ENOMEM;
 543		}
 544
 545		/* we know this is whitespace */
 546		if (*end)
 547			end++;
 548
 549		/* terminate the string and put it in the array */
 550		*out = '\0';
 551		argv[*argc] = start;
 552		(*argc)++;
 553	}
 554
 555	*argvp = argv;
 556	return 0;
 557}
 558
 
 
 
 
 
 
 559/*
 560 * Impose necessary and sufficient conditions on a devices's table such
 561 * that any incoming bio which respects its logical_block_size can be
 562 * processed successfully.  If it falls across the boundary between
 563 * two or more targets, the size of each piece it gets split into must
 564 * be compatible with the logical_block_size of the target processing it.
 565 */
 566static int validate_hardware_logical_block_alignment(struct dm_table *t,
 567						     struct queue_limits *limits)
 568{
 569	/*
 570	 * This function uses arithmetic modulo the logical_block_size
 571	 * (in units of 512-byte sectors).
 572	 */
 573	unsigned short device_logical_block_size_sects =
 574		limits->logical_block_size >> SECTOR_SHIFT;
 575
 576	/*
 577	 * Offset of the start of the next table entry, mod logical_block_size.
 578	 */
 579	unsigned short next_target_start = 0;
 580
 581	/*
 582	 * Given an aligned bio that extends beyond the end of a
 583	 * target, how many sectors must the next target handle?
 584	 */
 585	unsigned short remaining = 0;
 586
 587	struct dm_target *ti;
 588	struct queue_limits ti_limits;
 589	unsigned int i;
 590
 591	/*
 592	 * Check each entry in the table in turn.
 593	 */
 594	for (i = 0; i < t->num_targets; i++) {
 595		ti = dm_table_get_target(t, i);
 596
 597		blk_set_stacking_limits(&ti_limits);
 598
 599		/* combine all target devices' limits */
 600		if (ti->type->iterate_devices)
 601			ti->type->iterate_devices(ti, dm_set_device_limits,
 602						  &ti_limits);
 603
 604		/*
 605		 * If the remaining sectors fall entirely within this
 606		 * table entry are they compatible with its logical_block_size?
 607		 */
 608		if (remaining < ti->len &&
 609		    remaining & ((ti_limits.logical_block_size >>
 610				  SECTOR_SHIFT) - 1))
 611			break;	/* Error */
 612
 613		next_target_start =
 614		    (unsigned short) ((next_target_start + ti->len) &
 615				      (device_logical_block_size_sects - 1));
 616		remaining = next_target_start ?
 617		    device_logical_block_size_sects - next_target_start : 0;
 618	}
 619
 620	if (remaining) {
 621		DMERR("%s: table line %u (start sect %llu len %llu) "
 622		      "not aligned to h/w logical block size %u",
 623		      dm_device_name(t->md), i,
 624		      (unsigned long long) ti->begin,
 625		      (unsigned long long) ti->len,
 626		      limits->logical_block_size);
 627		return -EINVAL;
 628	}
 629
 630	return 0;
 631}
 632
 633int dm_table_add_target(struct dm_table *t, const char *type,
 634			sector_t start, sector_t len, char *params)
 635{
 636	int r = -EINVAL, argc;
 637	char **argv;
 638	struct dm_target *ti;
 639
 640	if (t->singleton) {
 641		DMERR("%s: target type %s must appear alone in table",
 642		      dm_device_name(t->md), t->targets->type->name);
 643		return -EINVAL;
 644	}
 645
 646	BUG_ON(t->num_targets >= t->num_allocated);
 647
 648	ti = t->targets + t->num_targets;
 649	memset(ti, 0, sizeof(*ti));
 650
 651	if (!len) {
 652		DMERR("%s: zero-length target", dm_device_name(t->md));
 653		return -EINVAL;
 654	}
 655
 656	ti->type = dm_get_target_type(type);
 657	if (!ti->type) {
 658		DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
 659		return -EINVAL;
 660	}
 661
 662	if (dm_target_needs_singleton(ti->type)) {
 663		if (t->num_targets) {
 664			ti->error = "singleton target type must appear alone in table";
 665			goto bad;
 666		}
 667		t->singleton = true;
 668	}
 669
 670	if (dm_target_always_writeable(ti->type) && !(t->mode & FMODE_WRITE)) {
 
 671		ti->error = "target type may not be included in a read-only table";
 672		goto bad;
 673	}
 674
 675	if (t->immutable_target_type) {
 676		if (t->immutable_target_type != ti->type) {
 677			ti->error = "immutable target type cannot be mixed with other target types";
 678			goto bad;
 679		}
 680	} else if (dm_target_is_immutable(ti->type)) {
 681		if (t->num_targets) {
 682			ti->error = "immutable target type cannot be mixed with other target types";
 683			goto bad;
 684		}
 685		t->immutable_target_type = ti->type;
 686	}
 687
 688	if (dm_target_has_integrity(ti->type))
 689		t->integrity_added = 1;
 690
 691	ti->table = t;
 692	ti->begin = start;
 693	ti->len = len;
 694	ti->error = "Unknown error";
 695
 696	/*
 697	 * Does this target adjoin the previous one ?
 698	 */
 699	if (!adjoin(t, ti)) {
 700		ti->error = "Gap in table";
 701		goto bad;
 702	}
 703
 704	r = dm_split_args(&argc, &argv, params);
 705	if (r) {
 706		ti->error = "couldn't split parameters";
 707		goto bad;
 708	}
 709
 710	r = ti->type->ctr(ti, argc, argv);
 711	kfree(argv);
 712	if (r)
 713		goto bad;
 714
 715	t->highs[t->num_targets++] = ti->begin + ti->len - 1;
 716
 717	if (!ti->num_discard_bios && ti->discards_supported)
 718		DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
 719		       dm_device_name(t->md), type);
 720
 721	if (ti->limit_swap_bios && !static_key_enabled(&swap_bios_enabled.key))
 722		static_branch_enable(&swap_bios_enabled);
 723
 
 
 
 724	return 0;
 725
 726 bad:
 727	DMERR("%s: %s: %s (%pe)", dm_device_name(t->md), type, ti->error, ERR_PTR(r));
 728	dm_put_target_type(ti->type);
 729	return r;
 730}
 731
 732/*
 733 * Target argument parsing helpers.
 734 */
 735static int validate_next_arg(const struct dm_arg *arg,
 736			     struct dm_arg_set *arg_set,
 737			     unsigned *value, char **error, unsigned grouped)
 738{
 739	const char *arg_str = dm_shift_arg(arg_set);
 740	char dummy;
 741
 742	if (!arg_str ||
 743	    (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
 744	    (*value < arg->min) ||
 745	    (*value > arg->max) ||
 746	    (grouped && arg_set->argc < *value)) {
 747		*error = arg->error;
 748		return -EINVAL;
 749	}
 750
 751	return 0;
 752}
 753
 754int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 755		unsigned *value, char **error)
 756{
 757	return validate_next_arg(arg, arg_set, value, error, 0);
 758}
 759EXPORT_SYMBOL(dm_read_arg);
 760
 761int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 762		      unsigned *value, char **error)
 763{
 764	return validate_next_arg(arg, arg_set, value, error, 1);
 765}
 766EXPORT_SYMBOL(dm_read_arg_group);
 767
 768const char *dm_shift_arg(struct dm_arg_set *as)
 769{
 770	char *r;
 771
 772	if (as->argc) {
 773		as->argc--;
 774		r = *as->argv;
 775		as->argv++;
 776		return r;
 777	}
 778
 779	return NULL;
 780}
 781EXPORT_SYMBOL(dm_shift_arg);
 782
 783void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
 784{
 785	BUG_ON(as->argc < num_args);
 786	as->argc -= num_args;
 787	as->argv += num_args;
 788}
 789EXPORT_SYMBOL(dm_consume_args);
 790
 791static bool __table_type_bio_based(enum dm_queue_mode table_type)
 792{
 793	return (table_type == DM_TYPE_BIO_BASED ||
 794		table_type == DM_TYPE_DAX_BIO_BASED);
 795}
 796
 797static bool __table_type_request_based(enum dm_queue_mode table_type)
 798{
 799	return table_type == DM_TYPE_REQUEST_BASED;
 800}
 801
 802void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
 803{
 804	t->type = type;
 805}
 806EXPORT_SYMBOL_GPL(dm_table_set_type);
 807
 808/* validate the dax capability of the target device span */
 809static int device_not_dax_capable(struct dm_target *ti, struct dm_dev *dev,
 810			sector_t start, sector_t len, void *data)
 811{
 812	if (dev->dax_dev)
 813		return false;
 814
 815	DMDEBUG("%pg: error: dax unsupported by block device", dev->bdev);
 816	return true;
 817}
 818
 819/* Check devices support synchronous DAX */
 820static int device_not_dax_synchronous_capable(struct dm_target *ti, struct dm_dev *dev,
 821					      sector_t start, sector_t len, void *data)
 822{
 823	return !dev->dax_dev || !dax_synchronous(dev->dax_dev);
 824}
 825
 826static bool dm_table_supports_dax(struct dm_table *t,
 827				  iterate_devices_callout_fn iterate_fn)
 828{
 829	/* Ensure that all targets support DAX. */
 830	for (unsigned int i = 0; i < t->num_targets; i++) {
 831		struct dm_target *ti = dm_table_get_target(t, i);
 832
 833		if (!ti->type->direct_access)
 834			return false;
 835
 836		if (!ti->type->iterate_devices ||
 
 837		    ti->type->iterate_devices(ti, iterate_fn, NULL))
 838			return false;
 839	}
 840
 841	return true;
 842}
 843
 844static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev,
 845				  sector_t start, sector_t len, void *data)
 846{
 847	struct block_device *bdev = dev->bdev;
 848	struct request_queue *q = bdev_get_queue(bdev);
 849
 850	/* request-based cannot stack on partitions! */
 851	if (bdev_is_partition(bdev))
 852		return false;
 853
 854	return queue_is_mq(q);
 855}
 856
 857static int dm_table_determine_type(struct dm_table *t)
 858{
 859	unsigned bio_based = 0, request_based = 0, hybrid = 0;
 860	struct dm_target *ti;
 861	struct list_head *devices = dm_table_get_devices(t);
 862	enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
 863
 864	if (t->type != DM_TYPE_NONE) {
 865		/* target already set the table's type */
 866		if (t->type == DM_TYPE_BIO_BASED) {
 867			/* possibly upgrade to a variant of bio-based */
 868			goto verify_bio_based;
 869		}
 870		BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
 871		goto verify_rq_based;
 872	}
 873
 874	for (unsigned int i = 0; i < t->num_targets; i++) {
 875		ti = dm_table_get_target(t, i);
 876		if (dm_target_hybrid(ti))
 877			hybrid = 1;
 878		else if (dm_target_request_based(ti))
 879			request_based = 1;
 880		else
 881			bio_based = 1;
 882
 883		if (bio_based && request_based) {
 884			DMERR("Inconsistent table: different target types"
 885			      " can't be mixed up");
 886			return -EINVAL;
 887		}
 888	}
 889
 890	if (hybrid && !bio_based && !request_based) {
 891		/*
 892		 * The targets can work either way.
 893		 * Determine the type from the live device.
 894		 * Default to bio-based if device is new.
 895		 */
 896		if (__table_type_request_based(live_md_type))
 897			request_based = 1;
 898		else
 899			bio_based = 1;
 900	}
 901
 902	if (bio_based) {
 903verify_bio_based:
 904		/* We must use this table as bio-based */
 905		t->type = DM_TYPE_BIO_BASED;
 906		if (dm_table_supports_dax(t, device_not_dax_capable) ||
 907		    (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
 908			t->type = DM_TYPE_DAX_BIO_BASED;
 909		}
 910		return 0;
 911	}
 912
 913	BUG_ON(!request_based); /* No targets in this table */
 914
 915	t->type = DM_TYPE_REQUEST_BASED;
 916
 917verify_rq_based:
 918	/*
 919	 * Request-based dm supports only tables that have a single target now.
 920	 * To support multiple targets, request splitting support is needed,
 921	 * and that needs lots of changes in the block-layer.
 922	 * (e.g. request completion process for partial completion.)
 923	 */
 924	if (t->num_targets > 1) {
 925		DMERR("request-based DM doesn't support multiple targets");
 926		return -EINVAL;
 927	}
 928
 929	if (list_empty(devices)) {
 930		int srcu_idx;
 931		struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
 932
 933		/* inherit live table's type */
 934		if (live_table)
 935			t->type = live_table->type;
 936		dm_put_live_table(t->md, srcu_idx);
 937		return 0;
 938	}
 939
 940	ti = dm_table_get_immutable_target(t);
 941	if (!ti) {
 942		DMERR("table load rejected: immutable target is required");
 943		return -EINVAL;
 944	} else if (ti->max_io_len) {
 945		DMERR("table load rejected: immutable target that splits IO is not supported");
 946		return -EINVAL;
 947	}
 948
 949	/* Non-request-stackable devices can't be used for request-based dm */
 950	if (!ti->type->iterate_devices ||
 951	    !ti->type->iterate_devices(ti, device_is_rq_stackable, NULL)) {
 952		DMERR("table load rejected: including non-request-stackable devices");
 953		return -EINVAL;
 954	}
 955
 956	return 0;
 957}
 958
 959enum dm_queue_mode dm_table_get_type(struct dm_table *t)
 960{
 961	return t->type;
 962}
 963
 964struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
 965{
 966	return t->immutable_target_type;
 967}
 968
 969struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
 970{
 971	/* Immutable target is implicitly a singleton */
 972	if (t->num_targets > 1 ||
 973	    !dm_target_is_immutable(t->targets[0].type))
 974		return NULL;
 975
 976	return t->targets;
 977}
 978
 979struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
 980{
 981	for (unsigned int i = 0; i < t->num_targets; i++) {
 982		struct dm_target *ti = dm_table_get_target(t, i);
 983
 984		if (dm_target_is_wildcard(ti->type))
 985			return ti;
 986	}
 987
 988	return NULL;
 989}
 990
 991bool dm_table_bio_based(struct dm_table *t)
 992{
 993	return __table_type_bio_based(dm_table_get_type(t));
 994}
 995
 996bool dm_table_request_based(struct dm_table *t)
 997{
 998	return __table_type_request_based(dm_table_get_type(t));
 999}
1000
1001static bool dm_table_supports_poll(struct dm_table *t);
1002
1003static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1004{
1005	enum dm_queue_mode type = dm_table_get_type(t);
1006	unsigned int per_io_data_size = 0, front_pad, io_front_pad;
1007	unsigned int min_pool_size = 0, pool_size;
1008	struct dm_md_mempools *pools;
 
 
1009
1010	if (unlikely(type == DM_TYPE_NONE)) {
1011		DMERR("no table type is set, can't allocate mempools");
1012		return -EINVAL;
1013	}
1014
1015	pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
1016	if (!pools)
1017		return -ENOMEM;
1018
1019	if (type == DM_TYPE_REQUEST_BASED) {
1020		pool_size = dm_get_reserved_rq_based_ios();
1021		front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
1022		goto init_bs;
1023	}
1024
 
 
 
1025	for (unsigned int i = 0; i < t->num_targets; i++) {
1026		struct dm_target *ti = dm_table_get_target(t, i);
1027
1028		per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1029		min_pool_size = max(min_pool_size, ti->num_flush_bios);
 
 
1030	}
1031	pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
1032	front_pad = roundup(per_io_data_size,
1033		__alignof__(struct dm_target_io)) + DM_TARGET_IO_BIO_OFFSET;
1034
1035	io_front_pad = roundup(per_io_data_size,
1036		__alignof__(struct dm_io)) + DM_IO_BIO_OFFSET;
1037	if (bioset_init(&pools->io_bs, pool_size, io_front_pad,
1038			dm_table_supports_poll(t) ? BIOSET_PERCPU_CACHE : 0))
1039		goto out_free_pools;
1040	if (t->integrity_supported &&
1041	    bioset_integrity_create(&pools->io_bs, pool_size))
1042		goto out_free_pools;
1043init_bs:
1044	if (bioset_init(&pools->bs, pool_size, front_pad, 0))
1045		goto out_free_pools;
1046	if (t->integrity_supported &&
1047	    bioset_integrity_create(&pools->bs, pool_size))
1048		goto out_free_pools;
1049
1050	t->mempools = pools;
1051	return 0;
1052
1053out_free_pools:
1054	dm_free_md_mempools(pools);
1055	return -ENOMEM;
1056}
1057
1058static int setup_indexes(struct dm_table *t)
1059{
1060	int i;
1061	unsigned int total = 0;
1062	sector_t *indexes;
1063
1064	/* allocate the space for *all* the indexes */
1065	for (i = t->depth - 2; i >= 0; i--) {
1066		t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1067		total += t->counts[i];
1068	}
1069
1070	indexes = kvcalloc(total, NODE_SIZE, GFP_KERNEL);
1071	if (!indexes)
1072		return -ENOMEM;
1073
1074	/* set up internal nodes, bottom-up */
1075	for (i = t->depth - 2; i >= 0; i--) {
1076		t->index[i] = indexes;
1077		indexes += (KEYS_PER_NODE * t->counts[i]);
1078		setup_btree_index(i, t);
1079	}
1080
1081	return 0;
1082}
1083
1084/*
1085 * Builds the btree to index the map.
1086 */
1087static int dm_table_build_index(struct dm_table *t)
1088{
1089	int r = 0;
1090	unsigned int leaf_nodes;
1091
1092	/* how many indexes will the btree have ? */
1093	leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1094	t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1095
1096	/* leaf layer has already been set up */
1097	t->counts[t->depth - 1] = leaf_nodes;
1098	t->index[t->depth - 1] = t->highs;
1099
1100	if (t->depth >= 2)
1101		r = setup_indexes(t);
1102
1103	return r;
1104}
1105
1106static bool integrity_profile_exists(struct gendisk *disk)
1107{
1108	return !!blk_get_integrity(disk);
1109}
1110
1111/*
1112 * Get a disk whose integrity profile reflects the table's profile.
1113 * Returns NULL if integrity support was inconsistent or unavailable.
1114 */
1115static struct gendisk *dm_table_get_integrity_disk(struct dm_table *t)
1116{
1117	struct list_head *devices = dm_table_get_devices(t);
1118	struct dm_dev_internal *dd = NULL;
1119	struct gendisk *prev_disk = NULL, *template_disk = NULL;
1120
1121	for (unsigned int i = 0; i < t->num_targets; i++) {
1122		struct dm_target *ti = dm_table_get_target(t, i);
1123
1124		if (!dm_target_passes_integrity(ti->type))
1125			goto no_integrity;
1126	}
1127
1128	list_for_each_entry(dd, devices, list) {
1129		template_disk = dd->dm_dev->bdev->bd_disk;
1130		if (!integrity_profile_exists(template_disk))
1131			goto no_integrity;
1132		else if (prev_disk &&
1133			 blk_integrity_compare(prev_disk, template_disk) < 0)
1134			goto no_integrity;
1135		prev_disk = template_disk;
1136	}
1137
1138	return template_disk;
1139
1140no_integrity:
1141	if (prev_disk)
1142		DMWARN("%s: integrity not set: %s and %s profile mismatch",
1143		       dm_device_name(t->md),
1144		       prev_disk->disk_name,
1145		       template_disk->disk_name);
1146	return NULL;
1147}
1148
1149/*
1150 * Register the mapped device for blk_integrity support if the
1151 * underlying devices have an integrity profile.  But all devices may
1152 * not have matching profiles (checking all devices isn't reliable
1153 * during table load because this table may use other DM device(s) which
1154 * must be resumed before they will have an initialized integity
1155 * profile).  Consequently, stacked DM devices force a 2 stage integrity
1156 * profile validation: First pass during table load, final pass during
1157 * resume.
1158 */
1159static int dm_table_register_integrity(struct dm_table *t)
1160{
1161	struct mapped_device *md = t->md;
1162	struct gendisk *template_disk = NULL;
1163
1164	/* If target handles integrity itself do not register it here. */
1165	if (t->integrity_added)
1166		return 0;
1167
1168	template_disk = dm_table_get_integrity_disk(t);
1169	if (!template_disk)
1170		return 0;
1171
1172	if (!integrity_profile_exists(dm_disk(md))) {
1173		t->integrity_supported = true;
1174		/*
1175		 * Register integrity profile during table load; we can do
1176		 * this because the final profile must match during resume.
1177		 */
1178		blk_integrity_register(dm_disk(md),
1179				       blk_get_integrity(template_disk));
1180		return 0;
1181	}
1182
1183	/*
1184	 * If DM device already has an initialized integrity
1185	 * profile the new profile should not conflict.
1186	 */
1187	if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1188		DMERR("%s: conflict with existing integrity profile: "
1189		      "%s profile mismatch",
1190		      dm_device_name(t->md),
1191		      template_disk->disk_name);
1192		return 1;
1193	}
1194
1195	/* Preserve existing integrity profile */
1196	t->integrity_supported = true;
1197	return 0;
1198}
1199
1200#ifdef CONFIG_BLK_INLINE_ENCRYPTION
1201
1202struct dm_crypto_profile {
1203	struct blk_crypto_profile profile;
1204	struct mapped_device *md;
1205};
1206
1207struct dm_keyslot_evict_args {
1208	const struct blk_crypto_key *key;
1209	int err;
1210};
1211
1212static int dm_keyslot_evict_callback(struct dm_target *ti, struct dm_dev *dev,
1213				     sector_t start, sector_t len, void *data)
1214{
1215	struct dm_keyslot_evict_args *args = data;
1216	int err;
1217
1218	err = blk_crypto_evict_key(dev->bdev, args->key);
1219	if (!args->err)
1220		args->err = err;
1221	/* Always try to evict the key from all devices. */
1222	return 0;
1223}
1224
1225/*
1226 * When an inline encryption key is evicted from a device-mapper device, evict
1227 * it from all the underlying devices.
1228 */
1229static int dm_keyslot_evict(struct blk_crypto_profile *profile,
1230			    const struct blk_crypto_key *key, unsigned int slot)
1231{
1232	struct mapped_device *md =
1233		container_of(profile, struct dm_crypto_profile, profile)->md;
1234	struct dm_keyslot_evict_args args = { key };
1235	struct dm_table *t;
1236	int srcu_idx;
1237
1238	t = dm_get_live_table(md, &srcu_idx);
1239	if (!t)
1240		return 0;
1241
1242	for (unsigned int i = 0; i < t->num_targets; i++) {
1243		struct dm_target *ti = dm_table_get_target(t, i);
1244
1245		if (!ti->type->iterate_devices)
1246			continue;
1247		ti->type->iterate_devices(ti, dm_keyslot_evict_callback, &args);
 
1248	}
1249
1250	dm_put_live_table(md, srcu_idx);
1251	return args.err;
1252}
1253
1254static int
1255device_intersect_crypto_capabilities(struct dm_target *ti, struct dm_dev *dev,
1256				     sector_t start, sector_t len, void *data)
1257{
1258	struct blk_crypto_profile *parent = data;
1259	struct blk_crypto_profile *child =
1260		bdev_get_queue(dev->bdev)->crypto_profile;
1261
1262	blk_crypto_intersect_capabilities(parent, child);
1263	return 0;
1264}
1265
1266void dm_destroy_crypto_profile(struct blk_crypto_profile *profile)
1267{
1268	struct dm_crypto_profile *dmcp = container_of(profile,
1269						      struct dm_crypto_profile,
1270						      profile);
1271
1272	if (!profile)
1273		return;
1274
1275	blk_crypto_profile_destroy(profile);
1276	kfree(dmcp);
1277}
1278
1279static void dm_table_destroy_crypto_profile(struct dm_table *t)
1280{
1281	dm_destroy_crypto_profile(t->crypto_profile);
1282	t->crypto_profile = NULL;
1283}
1284
1285/*
1286 * Constructs and initializes t->crypto_profile with a crypto profile that
1287 * represents the common set of crypto capabilities of the devices described by
1288 * the dm_table.  However, if the constructed crypto profile doesn't support all
1289 * crypto capabilities that are supported by the current mapped_device, it
1290 * returns an error instead, since we don't support removing crypto capabilities
1291 * on table changes.  Finally, if the constructed crypto profile is "empty" (has
1292 * no crypto capabilities at all), it just sets t->crypto_profile to NULL.
1293 */
1294static int dm_table_construct_crypto_profile(struct dm_table *t)
1295{
1296	struct dm_crypto_profile *dmcp;
1297	struct blk_crypto_profile *profile;
1298	unsigned int i;
1299	bool empty_profile = true;
1300
1301	dmcp = kmalloc(sizeof(*dmcp), GFP_KERNEL);
1302	if (!dmcp)
1303		return -ENOMEM;
1304	dmcp->md = t->md;
1305
1306	profile = &dmcp->profile;
1307	blk_crypto_profile_init(profile, 0);
1308	profile->ll_ops.keyslot_evict = dm_keyslot_evict;
1309	profile->max_dun_bytes_supported = UINT_MAX;
1310	memset(profile->modes_supported, 0xFF,
1311	       sizeof(profile->modes_supported));
1312
1313	for (i = 0; i < t->num_targets; i++) {
1314		struct dm_target *ti = dm_table_get_target(t, i);
1315
1316		if (!dm_target_passes_crypto(ti->type)) {
1317			blk_crypto_intersect_capabilities(profile, NULL);
1318			break;
1319		}
1320		if (!ti->type->iterate_devices)
1321			continue;
1322		ti->type->iterate_devices(ti,
1323					  device_intersect_crypto_capabilities,
1324					  profile);
1325	}
1326
1327	if (t->md->queue &&
1328	    !blk_crypto_has_capabilities(profile,
1329					 t->md->queue->crypto_profile)) {
1330		DMERR("Inline encryption capabilities of new DM table were more restrictive than the old table's. This is not supported!");
1331		dm_destroy_crypto_profile(profile);
1332		return -EINVAL;
1333	}
1334
1335	/*
1336	 * If the new profile doesn't actually support any crypto capabilities,
1337	 * we may as well represent it with a NULL profile.
1338	 */
1339	for (i = 0; i < ARRAY_SIZE(profile->modes_supported); i++) {
1340		if (profile->modes_supported[i]) {
1341			empty_profile = false;
1342			break;
1343		}
1344	}
1345
1346	if (empty_profile) {
1347		dm_destroy_crypto_profile(profile);
1348		profile = NULL;
1349	}
1350
1351	/*
1352	 * t->crypto_profile is only set temporarily while the table is being
1353	 * set up, and it gets set to NULL after the profile has been
1354	 * transferred to the request_queue.
1355	 */
1356	t->crypto_profile = profile;
1357
1358	return 0;
1359}
1360
1361static void dm_update_crypto_profile(struct request_queue *q,
1362				     struct dm_table *t)
1363{
1364	if (!t->crypto_profile)
1365		return;
1366
1367	/* Make the crypto profile less restrictive. */
1368	if (!q->crypto_profile) {
1369		blk_crypto_register(t->crypto_profile, q);
1370	} else {
1371		blk_crypto_update_capabilities(q->crypto_profile,
1372					       t->crypto_profile);
1373		dm_destroy_crypto_profile(t->crypto_profile);
1374	}
1375	t->crypto_profile = NULL;
1376}
1377
1378#else /* CONFIG_BLK_INLINE_ENCRYPTION */
1379
1380static int dm_table_construct_crypto_profile(struct dm_table *t)
1381{
1382	return 0;
1383}
1384
1385void dm_destroy_crypto_profile(struct blk_crypto_profile *profile)
1386{
1387}
1388
1389static void dm_table_destroy_crypto_profile(struct dm_table *t)
1390{
1391}
1392
1393static void dm_update_crypto_profile(struct request_queue *q,
1394				     struct dm_table *t)
1395{
1396}
1397
1398#endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1399
1400/*
1401 * Prepares the table for use by building the indices,
1402 * setting the type, and allocating mempools.
1403 */
1404int dm_table_complete(struct dm_table *t)
1405{
1406	int r;
1407
1408	r = dm_table_determine_type(t);
1409	if (r) {
1410		DMERR("unable to determine table type");
1411		return r;
1412	}
1413
1414	r = dm_table_build_index(t);
1415	if (r) {
1416		DMERR("unable to build btrees");
1417		return r;
1418	}
1419
1420	r = dm_table_register_integrity(t);
1421	if (r) {
1422		DMERR("could not register integrity profile.");
1423		return r;
1424	}
1425
1426	r = dm_table_construct_crypto_profile(t);
1427	if (r) {
1428		DMERR("could not construct crypto profile.");
1429		return r;
1430	}
1431
1432	r = dm_table_alloc_md_mempools(t, t->md);
1433	if (r)
1434		DMERR("unable to allocate mempools");
1435
1436	return r;
1437}
1438
1439static DEFINE_MUTEX(_event_lock);
1440void dm_table_event_callback(struct dm_table *t,
1441			     void (*fn)(void *), void *context)
1442{
1443	mutex_lock(&_event_lock);
1444	t->event_fn = fn;
1445	t->event_context = context;
1446	mutex_unlock(&_event_lock);
1447}
1448
1449void dm_table_event(struct dm_table *t)
1450{
1451	mutex_lock(&_event_lock);
1452	if (t->event_fn)
1453		t->event_fn(t->event_context);
1454	mutex_unlock(&_event_lock);
1455}
1456EXPORT_SYMBOL(dm_table_event);
1457
1458inline sector_t dm_table_get_size(struct dm_table *t)
1459{
1460	return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1461}
1462EXPORT_SYMBOL(dm_table_get_size);
1463
1464/*
1465 * Search the btree for the correct target.
1466 *
1467 * Caller should check returned pointer for NULL
1468 * to trap I/O beyond end of device.
1469 */
1470struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1471{
1472	unsigned int l, n = 0, k = 0;
1473	sector_t *node;
1474
1475	if (unlikely(sector >= dm_table_get_size(t)))
1476		return NULL;
1477
1478	for (l = 0; l < t->depth; l++) {
1479		n = get_child(n, k);
1480		node = get_node(t, l, n);
1481
1482		for (k = 0; k < KEYS_PER_NODE; k++)
1483			if (node[k] >= sector)
1484				break;
1485	}
1486
1487	return &t->targets[(KEYS_PER_NODE * n) + k];
1488}
1489
1490static int device_not_poll_capable(struct dm_target *ti, struct dm_dev *dev,
1491				   sector_t start, sector_t len, void *data)
1492{
1493	struct request_queue *q = bdev_get_queue(dev->bdev);
1494
1495	return !test_bit(QUEUE_FLAG_POLL, &q->queue_flags);
1496}
1497
1498/*
1499 * type->iterate_devices() should be called when the sanity check needs to
1500 * iterate and check all underlying data devices. iterate_devices() will
1501 * iterate all underlying data devices until it encounters a non-zero return
1502 * code, returned by whether the input iterate_devices_callout_fn, or
1503 * iterate_devices() itself internally.
1504 *
1505 * For some target type (e.g. dm-stripe), one call of iterate_devices() may
1506 * iterate multiple underlying devices internally, in which case a non-zero
1507 * return code returned by iterate_devices_callout_fn will stop the iteration
1508 * in advance.
1509 *
1510 * Cases requiring _any_ underlying device supporting some kind of attribute,
1511 * should use the iteration structure like dm_table_any_dev_attr(), or call
1512 * it directly. @func should handle semantics of positive examples, e.g.
1513 * capable of something.
1514 *
1515 * Cases requiring _all_ underlying devices supporting some kind of attribute,
1516 * should use the iteration structure like dm_table_supports_nowait() or
1517 * dm_table_supports_discards(). Or introduce dm_table_all_devs_attr() that
1518 * uses an @anti_func that handle semantics of counter examples, e.g. not
1519 * capable of something. So: return !dm_table_any_dev_attr(t, anti_func, data);
1520 */
1521static bool dm_table_any_dev_attr(struct dm_table *t,
1522				  iterate_devices_callout_fn func, void *data)
1523{
1524	for (unsigned int i = 0; i < t->num_targets; i++) {
1525		struct dm_target *ti = dm_table_get_target(t, i);
1526
1527		if (ti->type->iterate_devices &&
1528		    ti->type->iterate_devices(ti, func, data))
1529			return true;
1530        }
1531
1532	return false;
1533}
1534
1535static int count_device(struct dm_target *ti, struct dm_dev *dev,
1536			sector_t start, sector_t len, void *data)
1537{
1538	unsigned *num_devices = data;
1539
1540	(*num_devices)++;
1541
1542	return 0;
1543}
1544
1545static bool dm_table_supports_poll(struct dm_table *t)
1546{
1547	for (unsigned int i = 0; i < t->num_targets; i++) {
1548		struct dm_target *ti = dm_table_get_target(t, i);
1549
1550		if (!ti->type->iterate_devices ||
1551		    ti->type->iterate_devices(ti, device_not_poll_capable, NULL))
1552			return false;
1553	}
1554
1555	return true;
1556}
1557
1558/*
1559 * Check whether a table has no data devices attached using each
1560 * target's iterate_devices method.
1561 * Returns false if the result is unknown because a target doesn't
1562 * support iterate_devices.
1563 */
1564bool dm_table_has_no_data_devices(struct dm_table *t)
1565{
1566	for (unsigned int i = 0; i < t->num_targets; i++) {
1567		struct dm_target *ti = dm_table_get_target(t, i);
1568		unsigned num_devices = 0;
1569
1570		if (!ti->type->iterate_devices)
1571			return false;
1572
1573		ti->type->iterate_devices(ti, count_device, &num_devices);
1574		if (num_devices)
1575			return false;
1576	}
1577
1578	return true;
1579}
1580
1581static int device_not_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1582				  sector_t start, sector_t len, void *data)
1583{
1584	struct request_queue *q = bdev_get_queue(dev->bdev);
1585	enum blk_zoned_model *zoned_model = data;
1586
1587	return blk_queue_zoned_model(q) != *zoned_model;
 
 
 
 
 
 
1588}
1589
1590/*
1591 * Check the device zoned model based on the target feature flag. If the target
1592 * has the DM_TARGET_ZONED_HM feature flag set, host-managed zoned devices are
1593 * also accepted but all devices must have the same zoned model. If the target
1594 * has the DM_TARGET_MIXED_ZONED_MODEL feature set, the devices can have any
1595 * zoned model with all zoned devices having the same zone size.
1596 */
1597static bool dm_table_supports_zoned_model(struct dm_table *t,
1598					  enum blk_zoned_model zoned_model)
1599{
1600	for (unsigned int i = 0; i < t->num_targets; i++) {
1601		struct dm_target *ti = dm_table_get_target(t, i);
1602
 
 
 
 
 
 
 
 
 
 
 
 
1603		if (dm_target_supports_zoned_hm(ti->type)) {
1604			if (!ti->type->iterate_devices ||
1605			    ti->type->iterate_devices(ti, device_not_zoned_model,
1606						      &zoned_model))
1607				return false;
1608		} else if (!dm_target_supports_mixed_zoned_model(ti->type)) {
1609			if (zoned_model == BLK_ZONED_HM)
1610				return false;
1611		}
1612	}
1613
1614	return true;
1615}
1616
1617static int device_not_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1618					   sector_t start, sector_t len, void *data)
1619{
1620	unsigned int *zone_sectors = data;
1621
1622	if (!bdev_is_zoned(dev->bdev))
1623		return 0;
1624	return bdev_zone_sectors(dev->bdev) != *zone_sectors;
1625}
1626
1627/*
1628 * Check consistency of zoned model and zone sectors across all targets. For
1629 * zone sectors, if the destination device is a zoned block device, it shall
1630 * have the specified zone_sectors.
1631 */
1632static int validate_hardware_zoned_model(struct dm_table *t,
1633					 enum blk_zoned_model zoned_model,
1634					 unsigned int zone_sectors)
1635{
1636	if (zoned_model == BLK_ZONED_NONE)
1637		return 0;
1638
1639	if (!dm_table_supports_zoned_model(t, zoned_model)) {
1640		DMERR("%s: zoned model is not consistent across all devices",
1641		      dm_device_name(t->md));
1642		return -EINVAL;
1643	}
1644
1645	/* Check zone size validity and compatibility */
1646	if (!zone_sectors || !is_power_of_2(zone_sectors))
1647		return -EINVAL;
1648
1649	if (dm_table_any_dev_attr(t, device_not_matches_zone_sectors, &zone_sectors)) {
1650		DMERR("%s: zone sectors is not consistent across all zoned devices",
1651		      dm_device_name(t->md));
1652		return -EINVAL;
1653	}
1654
1655	return 0;
1656}
1657
1658/*
1659 * Establish the new table's queue_limits and validate them.
1660 */
1661int dm_calculate_queue_limits(struct dm_table *t,
1662			      struct queue_limits *limits)
1663{
1664	struct queue_limits ti_limits;
1665	enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1666	unsigned int zone_sectors = 0;
 
1667
1668	blk_set_stacking_limits(limits);
1669
 
1670	for (unsigned int i = 0; i < t->num_targets; i++) {
1671		struct dm_target *ti = dm_table_get_target(t, i);
1672
1673		blk_set_stacking_limits(&ti_limits);
 
 
1674
1675		if (!ti->type->iterate_devices)
 
 
 
 
 
 
 
 
1676			goto combine_limits;
 
1677
1678		/*
1679		 * Combine queue limits of all the devices this target uses.
1680		 */
1681		ti->type->iterate_devices(ti, dm_set_device_limits,
1682					  &ti_limits);
1683
1684		if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1685			/*
1686			 * After stacking all limits, validate all devices
1687			 * in table support this zoned model and zone sectors.
1688			 */
1689			zoned_model = ti_limits.zoned;
1690			zone_sectors = ti_limits.chunk_sectors;
1691		}
1692
1693		/* Set I/O hints portion of queue limits */
1694		if (ti->type->io_hints)
1695			ti->type->io_hints(ti, &ti_limits);
1696
1697		/*
1698		 * Check each device area is consistent with the target's
1699		 * overall queue limits.
1700		 */
1701		if (ti->type->iterate_devices(ti, device_area_is_invalid,
1702					      &ti_limits))
1703			return -EINVAL;
1704
1705combine_limits:
1706		/*
1707		 * Merge this target's queue limits into the overall limits
1708		 * for the table.
1709		 */
1710		if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1711			DMWARN("%s: adding target device "
1712			       "(start sect %llu len %llu) "
1713			       "caused an alignment inconsistency",
1714			       dm_device_name(t->md),
1715			       (unsigned long long) ti->begin,
1716			       (unsigned long long) ti->len);
 
 
 
 
 
 
 
 
 
 
 
 
1717	}
1718
1719	/*
1720	 * Verify that the zoned model and zone sectors, as determined before
1721	 * any .io_hints override, are the same across all devices in the table.
1722	 * - this is especially relevant if .io_hints is emulating a disk-managed
1723	 *   zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1724	 * BUT...
1725	 */
1726	if (limits->zoned != BLK_ZONED_NONE) {
1727		/*
1728		 * ...IF the above limits stacking determined a zoned model
1729		 * validate that all of the table's devices conform to it.
1730		 */
1731		zoned_model = limits->zoned;
1732		zone_sectors = limits->chunk_sectors;
1733	}
1734	if (validate_hardware_zoned_model(t, zoned_model, zone_sectors))
1735		return -EINVAL;
1736
1737	return validate_hardware_logical_block_alignment(t, limits);
1738}
1739
1740/*
1741 * Verify that all devices have an integrity profile that matches the
1742 * DM device's registered integrity profile.  If the profiles don't
1743 * match then unregister the DM device's integrity profile.
1744 */
1745static void dm_table_verify_integrity(struct dm_table *t)
1746{
1747	struct gendisk *template_disk = NULL;
1748
1749	if (t->integrity_added)
1750		return;
1751
1752	if (t->integrity_supported) {
1753		/*
1754		 * Verify that the original integrity profile
1755		 * matches all the devices in this table.
1756		 */
1757		template_disk = dm_table_get_integrity_disk(t);
1758		if (template_disk &&
1759		    blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1760			return;
1761	}
1762
1763	if (integrity_profile_exists(dm_disk(t->md))) {
1764		DMWARN("%s: unable to establish an integrity profile",
1765		       dm_device_name(t->md));
1766		blk_integrity_unregister(dm_disk(t->md));
1767	}
1768}
1769
1770static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1771				sector_t start, sector_t len, void *data)
1772{
1773	unsigned long flush = (unsigned long) data;
1774	struct request_queue *q = bdev_get_queue(dev->bdev);
1775
1776	return (q->queue_flags & flush);
1777}
1778
1779static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1780{
1781	/*
1782	 * Require at least one underlying device to support flushes.
1783	 * t->devices includes internal dm devices such as mirror logs
1784	 * so we need to use iterate_devices here, which targets
1785	 * supporting flushes must provide.
1786	 */
1787	for (unsigned int i = 0; i < t->num_targets; i++) {
1788		struct dm_target *ti = dm_table_get_target(t, i);
1789
1790		if (!ti->num_flush_bios)
1791			continue;
1792
1793		if (ti->flush_supported)
1794			return true;
1795
1796		if (ti->type->iterate_devices &&
1797		    ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1798			return true;
1799	}
1800
1801	return false;
1802}
1803
1804static int device_dax_write_cache_enabled(struct dm_target *ti,
1805					  struct dm_dev *dev, sector_t start,
1806					  sector_t len, void *data)
1807{
1808	struct dax_device *dax_dev = dev->dax_dev;
1809
1810	if (!dax_dev)
1811		return false;
1812
1813	if (dax_write_cache_enabled(dax_dev))
1814		return true;
1815	return false;
1816}
1817
1818static int device_is_rotational(struct dm_target *ti, struct dm_dev *dev,
1819				sector_t start, sector_t len, void *data)
1820{
1821	return !bdev_nonrot(dev->bdev);
1822}
1823
1824static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1825			     sector_t start, sector_t len, void *data)
1826{
1827	struct request_queue *q = bdev_get_queue(dev->bdev);
1828
1829	return !blk_queue_add_random(q);
1830}
1831
1832static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1833					   sector_t start, sector_t len, void *data)
1834{
1835	struct request_queue *q = bdev_get_queue(dev->bdev);
1836
1837	return !q->limits.max_write_zeroes_sectors;
1838}
1839
1840static bool dm_table_supports_write_zeroes(struct dm_table *t)
1841{
1842	for (unsigned int i = 0; i < t->num_targets; i++) {
1843		struct dm_target *ti = dm_table_get_target(t, i);
1844
1845		if (!ti->num_write_zeroes_bios)
1846			return false;
1847
1848		if (!ti->type->iterate_devices ||
1849		    ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1850			return false;
1851	}
1852
1853	return true;
1854}
1855
1856static int device_not_nowait_capable(struct dm_target *ti, struct dm_dev *dev,
1857				     sector_t start, sector_t len, void *data)
1858{
1859	return !bdev_nowait(dev->bdev);
1860}
1861
1862static bool dm_table_supports_nowait(struct dm_table *t)
1863{
1864	for (unsigned int i = 0; i < t->num_targets; i++) {
1865		struct dm_target *ti = dm_table_get_target(t, i);
1866
1867		if (!dm_target_supports_nowait(ti->type))
1868			return false;
1869
1870		if (!ti->type->iterate_devices ||
1871		    ti->type->iterate_devices(ti, device_not_nowait_capable, NULL))
1872			return false;
1873	}
1874
1875	return true;
1876}
1877
1878static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1879				      sector_t start, sector_t len, void *data)
1880{
1881	return !bdev_max_discard_sectors(dev->bdev);
1882}
1883
1884static bool dm_table_supports_discards(struct dm_table *t)
1885{
1886	for (unsigned int i = 0; i < t->num_targets; i++) {
1887		struct dm_target *ti = dm_table_get_target(t, i);
1888
1889		if (!ti->num_discard_bios)
1890			return false;
1891
1892		/*
1893		 * Either the target provides discard support (as implied by setting
1894		 * 'discards_supported') or it relies on _all_ data devices having
1895		 * discard support.
1896		 */
1897		if (!ti->discards_supported &&
1898		    (!ti->type->iterate_devices ||
1899		     ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1900			return false;
1901	}
1902
1903	return true;
1904}
1905
1906static int device_not_secure_erase_capable(struct dm_target *ti,
1907					   struct dm_dev *dev, sector_t start,
1908					   sector_t len, void *data)
1909{
1910	return !bdev_max_secure_erase_sectors(dev->bdev);
1911}
1912
1913static bool dm_table_supports_secure_erase(struct dm_table *t)
1914{
1915	for (unsigned int i = 0; i < t->num_targets; i++) {
1916		struct dm_target *ti = dm_table_get_target(t, i);
1917
1918		if (!ti->num_secure_erase_bios)
1919			return false;
1920
1921		if (!ti->type->iterate_devices ||
1922		    ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1923			return false;
1924	}
1925
1926	return true;
1927}
1928
1929static int device_requires_stable_pages(struct dm_target *ti,
1930					struct dm_dev *dev, sector_t start,
1931					sector_t len, void *data)
1932{
1933	return bdev_stable_writes(dev->bdev);
1934}
1935
1936int dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1937			      struct queue_limits *limits)
1938{
1939	bool wc = false, fua = false;
1940	int r;
1941
 
 
 
1942	/*
1943	 * Copy table's limits to the DM device's request_queue
 
1944	 */
1945	q->limits = *limits;
1946
1947	if (dm_table_supports_nowait(t))
1948		blk_queue_flag_set(QUEUE_FLAG_NOWAIT, q);
1949	else
1950		blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, q);
1951
1952	if (!dm_table_supports_discards(t)) {
1953		q->limits.max_discard_sectors = 0;
1954		q->limits.max_hw_discard_sectors = 0;
1955		q->limits.discard_granularity = 0;
1956		q->limits.discard_alignment = 0;
1957		q->limits.discard_misaligned = 0;
1958	}
1959
 
 
 
1960	if (!dm_table_supports_secure_erase(t))
1961		q->limits.max_secure_erase_sectors = 0;
1962
1963	if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1964		wc = true;
1965		if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1966			fua = true;
1967	}
1968	blk_queue_write_cache(q, wc, fua);
1969
1970	if (dm_table_supports_dax(t, device_not_dax_capable)) {
1971		blk_queue_flag_set(QUEUE_FLAG_DAX, q);
1972		if (dm_table_supports_dax(t, device_not_dax_synchronous_capable))
1973			set_dax_synchronous(t->md->dax_dev);
1974	}
1975	else
1976		blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
1977
1978	if (dm_table_any_dev_attr(t, device_dax_write_cache_enabled, NULL))
1979		dax_write_cache(t->md->dax_dev, true);
1980
1981	/* Ensure that all underlying devices are non-rotational. */
1982	if (dm_table_any_dev_attr(t, device_is_rotational, NULL))
1983		blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
1984	else
1985		blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
1986
1987	if (!dm_table_supports_write_zeroes(t))
1988		q->limits.max_write_zeroes_sectors = 0;
1989
1990	dm_table_verify_integrity(t);
1991
1992	/*
1993	 * Some devices don't use blk_integrity but still want stable pages
1994	 * because they do their own checksumming.
1995	 * If any underlying device requires stable pages, a table must require
1996	 * them as well.  Only targets that support iterate_devices are considered:
1997	 * don't want error, zero, etc to require stable pages.
1998	 */
1999	if (dm_table_any_dev_attr(t, device_requires_stable_pages, NULL))
2000		blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, q);
2001	else
2002		blk_queue_flag_clear(QUEUE_FLAG_STABLE_WRITES, q);
2003
2004	/*
2005	 * Determine whether or not this queue's I/O timings contribute
2006	 * to the entropy pool, Only request-based targets use this.
2007	 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
2008	 * have it set.
2009	 */
2010	if (blk_queue_add_random(q) &&
2011	    dm_table_any_dev_attr(t, device_is_not_random, NULL))
2012		blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2013
2014	/*
2015	 * For a zoned target, setup the zones related queue attributes
2016	 * and resources necessary for zone append emulation if necessary.
2017	 */
2018	if (blk_queue_is_zoned(q)) {
2019		r = dm_set_zones_restrictions(t, q);
2020		if (r)
2021			return r;
2022		if (!static_key_enabled(&zoned_enabled.key))
2023			static_branch_enable(&zoned_enabled);
2024	}
2025
2026	dm_update_crypto_profile(q, t);
2027	disk_update_readahead(t->md->disk);
 
2028
2029	/*
2030	 * Check for request-based device is left to
2031	 * dm_mq_init_request_queue()->blk_mq_init_allocated_queue().
2032	 *
2033	 * For bio-based device, only set QUEUE_FLAG_POLL when all
2034	 * underlying devices supporting polling.
2035	 */
2036	if (__table_type_bio_based(t->type)) {
2037		if (dm_table_supports_poll(t))
2038			blk_queue_flag_set(QUEUE_FLAG_POLL, q);
2039		else
2040			blk_queue_flag_clear(QUEUE_FLAG_POLL, q);
2041	}
2042
 
2043	return 0;
2044}
2045
2046struct list_head *dm_table_get_devices(struct dm_table *t)
2047{
2048	return &t->devices;
2049}
2050
2051fmode_t dm_table_get_mode(struct dm_table *t)
2052{
2053	return t->mode;
2054}
2055EXPORT_SYMBOL(dm_table_get_mode);
2056
2057enum suspend_mode {
2058	PRESUSPEND,
2059	PRESUSPEND_UNDO,
2060	POSTSUSPEND,
2061};
2062
2063static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
2064{
2065	lockdep_assert_held(&t->md->suspend_lock);
2066
2067	for (unsigned int i = 0; i < t->num_targets; i++) {
2068		struct dm_target *ti = dm_table_get_target(t, i);
2069
2070		switch (mode) {
2071		case PRESUSPEND:
2072			if (ti->type->presuspend)
2073				ti->type->presuspend(ti);
2074			break;
2075		case PRESUSPEND_UNDO:
2076			if (ti->type->presuspend_undo)
2077				ti->type->presuspend_undo(ti);
2078			break;
2079		case POSTSUSPEND:
2080			if (ti->type->postsuspend)
2081				ti->type->postsuspend(ti);
2082			break;
2083		}
2084	}
2085}
2086
2087void dm_table_presuspend_targets(struct dm_table *t)
2088{
2089	if (!t)
2090		return;
2091
2092	suspend_targets(t, PRESUSPEND);
2093}
2094
2095void dm_table_presuspend_undo_targets(struct dm_table *t)
2096{
2097	if (!t)
2098		return;
2099
2100	suspend_targets(t, PRESUSPEND_UNDO);
2101}
2102
2103void dm_table_postsuspend_targets(struct dm_table *t)
2104{
2105	if (!t)
2106		return;
2107
2108	suspend_targets(t, POSTSUSPEND);
2109}
2110
2111int dm_table_resume_targets(struct dm_table *t)
2112{
2113	unsigned int i;
2114	int r = 0;
2115
2116	lockdep_assert_held(&t->md->suspend_lock);
2117
2118	for (i = 0; i < t->num_targets; i++) {
2119		struct dm_target *ti = dm_table_get_target(t, i);
2120
2121		if (!ti->type->preresume)
2122			continue;
2123
2124		r = ti->type->preresume(ti);
2125		if (r) {
2126			DMERR("%s: %s: preresume failed, error = %d",
2127			      dm_device_name(t->md), ti->type->name, r);
2128			return r;
2129		}
2130	}
2131
2132	for (i = 0; i < t->num_targets; i++) {
2133		struct dm_target *ti = dm_table_get_target(t, i);
2134
2135		if (ti->type->resume)
2136			ti->type->resume(ti);
2137	}
2138
2139	return 0;
2140}
2141
2142struct mapped_device *dm_table_get_md(struct dm_table *t)
2143{
2144	return t->md;
2145}
2146EXPORT_SYMBOL(dm_table_get_md);
2147
2148const char *dm_table_device_name(struct dm_table *t)
2149{
2150	return dm_device_name(t->md);
2151}
2152EXPORT_SYMBOL_GPL(dm_table_device_name);
2153
2154void dm_table_run_md_queue_async(struct dm_table *t)
2155{
2156	if (!dm_table_request_based(t))
2157		return;
2158
2159	if (t->md->queue)
2160		blk_mq_run_hw_queues(t->md->queue, true);
2161}
2162EXPORT_SYMBOL(dm_table_run_md_queue_async);
2163