drivers/md/dm-table.c

   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  */
  36 static 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  */
  51 static 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  */
  59 static 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  */
  69 static 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  */
  84 static 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  */
 103 static 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
 128 int dm_table_create(struct dm_table **result, fmode_t mode,
 129                     unsigned int 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
 160 static 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
 174 static void dm_table_destroy_crypto_profile(struct dm_table *t);
 175
 176 void 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  */
 210 static 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  */
 224 static 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  */
 309 static 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  */
 331 dev_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 }
 339 EXPORT_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  */
 345 int 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);
 389 out:
 390         *result = dd->dm_dev;
 391         return 0;
 392 }
 393 EXPORT_SYMBOL(dm_get_device);
 394
 395 static 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  */
 424 void 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 }
 447 EXPORT_SYMBOL(dm_put_device);
 448
 449 /*
 450  * Checks to see if the target joins onto the end of the table.
 451  */
 452 static 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  */
 473 static char **realloc_argv(unsigned int *size, char **old_argv)
 474 {
 475         char **argv;
 476         unsigned int 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  */
 499 int dm_split_args(int *argc, char ***argvp, char *input)
 500 {
 501         char *start, *end = input, *out, **argv = NULL;
 502         unsigned int 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  */
 566 static 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
 633 int 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  */
 735 static int validate_next_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 736                              unsigned int *value, char **error, unsigned int grouped)
 737 {
 738         const char *arg_str = dm_shift_arg(arg_set);
 739         char dummy;
 740
 741         if (!arg_str ||
 742             (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
 743             (*value < arg->min) ||
 744             (*value > arg->max) ||
 745             (grouped && arg_set->argc < *value)) {
 746                 *error = arg->error;
 747                 return -EINVAL;
 748         }
 749
 750         return 0;
 751 }
 752
 753 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 754                 unsigned int *value, char **error)
 755 {
 756         return validate_next_arg(arg, arg_set, value, error, 0);
 757 }
 758 EXPORT_SYMBOL(dm_read_arg);
 759
 760 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 761                       unsigned int *value, char **error)
 762 {
 763         return validate_next_arg(arg, arg_set, value, error, 1);
 764 }
 765 EXPORT_SYMBOL(dm_read_arg_group);
 766
 767 const char *dm_shift_arg(struct dm_arg_set *as)
 768 {
 769         char *r;
 770
 771         if (as->argc) {
 772                 as->argc--;
 773                 r = *as->argv;
 774                 as->argv++;
 775                 return r;
 776         }
 777
 778         return NULL;
 779 }
 780 EXPORT_SYMBOL(dm_shift_arg);
 781
 782 void dm_consume_args(struct dm_arg_set *as, unsigned int num_args)
 783 {
 784         BUG_ON(as->argc < num_args);
 785         as->argc -= num_args;
 786         as->argv += num_args;
 787 }
 788 EXPORT_SYMBOL(dm_consume_args);
 789
 790 static bool __table_type_bio_based(enum dm_queue_mode table_type)
 791 {
 792         return (table_type == DM_TYPE_BIO_BASED ||
 793                 table_type == DM_TYPE_DAX_BIO_BASED);
 794 }
 795
 796 static bool __table_type_request_based(enum dm_queue_mode table_type)
 797 {
 798         return table_type == DM_TYPE_REQUEST_BASED;
 799 }
 800
 801 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
 802 {
 803         t->type = type;
 804 }
 805 EXPORT_SYMBOL_GPL(dm_table_set_type);
 806
 807 /* validate the dax capability of the target device span */
 808 static int device_not_dax_capable(struct dm_target *ti, struct dm_dev *dev,
 809                         sector_t start, sector_t len, void *data)
 810 {
 811         if (dev->dax_dev)
 812                 return false;
 813
 814         DMDEBUG("%pg: error: dax unsupported by block device", dev->bdev);
 815         return true;
 816 }
 817
 818 /* Check devices support synchronous DAX */
 819 static int device_not_dax_synchronous_capable(struct dm_target *ti, struct dm_dev *dev,
 820                                               sector_t start, sector_t len, void *data)
 821 {
 822         return !dev->dax_dev || !dax_synchronous(dev->dax_dev);
 823 }
 824
 825 static bool dm_table_supports_dax(struct dm_table *t,
 826                                   iterate_devices_callout_fn iterate_fn)
 827 {
 828         /* Ensure that all targets support DAX. */
 829         for (unsigned int i = 0; i < t->num_targets; i++) {
 830                 struct dm_target *ti = dm_table_get_target(t, i);
 831
 832                 if (!ti->type->direct_access)
 833                         return false;
 834
 835                 if (!ti->type->iterate_devices ||
 836                     ti->type->iterate_devices(ti, iterate_fn, NULL))
 837                         return false;
 838         }
 839
 840         return true;
 841 }
 842
 843 static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev,
 844                                   sector_t start, sector_t len, void *data)
 845 {
 846         struct block_device *bdev = dev->bdev;
 847         struct request_queue *q = bdev_get_queue(bdev);
 848
 849         /* request-based cannot stack on partitions! */
 850         if (bdev_is_partition(bdev))
 851                 return false;
 852
 853         return queue_is_mq(q);
 854 }
 855
 856 static int dm_table_determine_type(struct dm_table *t)
 857 {
 858         unsigned int bio_based = 0, request_based = 0, hybrid = 0;
 859         struct dm_target *ti;
 860         struct list_head *devices = dm_table_get_devices(t);
 861         enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
 862
 863         if (t->type != DM_TYPE_NONE) {
 864                 /* target already set the table's type */
 865                 if (t->type == DM_TYPE_BIO_BASED) {
 866                         /* possibly upgrade to a variant of bio-based */
 867                         goto verify_bio_based;
 868                 }
 869                 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
 870                 goto verify_rq_based;
 871         }
 872
 873         for (unsigned int i = 0; i < t->num_targets; i++) {
 874                 ti = dm_table_get_target(t, i);
 875                 if (dm_target_hybrid(ti))
 876                         hybrid = 1;
 877                 else if (dm_target_request_based(ti))
 878                         request_based = 1;
 879                 else
 880                         bio_based = 1;
 881
 882                 if (bio_based && request_based) {
 883                         DMERR("Inconsistent table: different target types"
 884                               " can't be mixed up");
 885                         return -EINVAL;
 886                 }
 887         }
 888
 889         if (hybrid && !bio_based && !request_based) {
 890                 /*
 891                  * The targets can work either way.
 892                  * Determine the type from the live device.
 893                  * Default to bio-based if device is new.
 894                  */
 895                 if (__table_type_request_based(live_md_type))
 896                         request_based = 1;
 897                 else
 898                         bio_based = 1;
 899         }
 900
 901         if (bio_based) {
 902 verify_bio_based:
 903                 /* We must use this table as bio-based */
 904                 t->type = DM_TYPE_BIO_BASED;
 905                 if (dm_table_supports_dax(t, device_not_dax_capable) ||
 906                     (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
 907                         t->type = DM_TYPE_DAX_BIO_BASED;
 908                 }
 909                 return 0;
 910         }
 911
 912         BUG_ON(!request_based); /* No targets in this table */
 913
 914         t->type = DM_TYPE_REQUEST_BASED;
 915
 916 verify_rq_based:
 917         /*
 918          * Request-based dm supports only tables that have a single target now.
 919          * To support multiple targets, request splitting support is needed,
 920          * and that needs lots of changes in the block-layer.
 921          * (e.g. request completion process for partial completion.)
 922          */
 923         if (t->num_targets > 1) {
 924                 DMERR("request-based DM doesn't support multiple targets");
 925                 return -EINVAL;
 926         }
 927
 928         if (list_empty(devices)) {
 929                 int srcu_idx;
 930                 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
 931
 932                 /* inherit live table's type */
 933                 if (live_table)
 934                         t->type = live_table->type;
 935                 dm_put_live_table(t->md, srcu_idx);
 936                 return 0;
 937         }
 938
 939         ti = dm_table_get_immutable_target(t);
 940         if (!ti) {
 941                 DMERR("table load rejected: immutable target is required");
 942                 return -EINVAL;
 943         } else if (ti->max_io_len) {
 944                 DMERR("table load rejected: immutable target that splits IO is not supported");
 945                 return -EINVAL;
 946         }
 947
 948         /* Non-request-stackable devices can't be used for request-based dm */
 949         if (!ti->type->iterate_devices ||
 950             !ti->type->iterate_devices(ti, device_is_rq_stackable, NULL)) {
 951                 DMERR("table load rejected: including non-request-stackable devices");
 952                 return -EINVAL;
 953         }
 954
 955         return 0;
 956 }
 957
 958 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
 959 {
 960         return t->type;
 961 }
 962
 963 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
 964 {
 965         return t->immutable_target_type;
 966 }
 967
 968 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
 969 {
 970         /* Immutable target is implicitly a singleton */
 971         if (t->num_targets > 1 ||
 972             !dm_target_is_immutable(t->targets[0].type))
 973                 return NULL;
 974
 975         return t->targets;
 976 }
 977
 978 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
 979 {
 980         for (unsigned int i = 0; i < t->num_targets; i++) {
 981                 struct dm_target *ti = dm_table_get_target(t, i);
 982
 983                 if (dm_target_is_wildcard(ti->type))
 984                         return ti;
 985         }
 986
 987         return NULL;
 988 }
 989
 990 bool dm_table_bio_based(struct dm_table *t)
 991 {
 992         return __table_type_bio_based(dm_table_get_type(t));
 993 }
 994
 995 bool dm_table_request_based(struct dm_table *t)
 996 {
 997         return __table_type_request_based(dm_table_get_type(t));
 998 }
 999
1000 static bool dm_table_supports_poll(struct dm_table *t);
1001
1002 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1003 {
1004         enum dm_queue_mode type = dm_table_get_type(t);
1005         unsigned int per_io_data_size = 0, front_pad, io_front_pad;
1006         unsigned int min_pool_size = 0, pool_size;
1007         struct dm_md_mempools *pools;
1008
1009         if (unlikely(type == DM_TYPE_NONE)) {
1010                 DMERR("no table type is set, can't allocate mempools");
1011                 return -EINVAL;
1012         }
1013
1014         pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
1015         if (!pools)
1016                 return -ENOMEM;
1017
1018         if (type == DM_TYPE_REQUEST_BASED) {
1019                 pool_size = dm_get_reserved_rq_based_ios();
1020                 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
1021                 goto init_bs;
1022         }
1023
1024         for (unsigned int i = 0; i < t->num_targets; i++) {
1025                 struct dm_target *ti = dm_table_get_target(t, i);
1026
1027                 per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1028                 min_pool_size = max(min_pool_size, ti->num_flush_bios);
1029         }
1030         pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
1031         front_pad = roundup(per_io_data_size,
1032                 __alignof__(struct dm_target_io)) + DM_TARGET_IO_BIO_OFFSET;
1033
1034         io_front_pad = roundup(per_io_data_size,
1035                 __alignof__(struct dm_io)) + DM_IO_BIO_OFFSET;
1036         if (bioset_init(&pools->io_bs, pool_size, io_front_pad,
1037                         dm_table_supports_poll(t) ? BIOSET_PERCPU_CACHE : 0))
1038                 goto out_free_pools;
1039         if (t->integrity_supported &&
1040             bioset_integrity_create(&pools->io_bs, pool_size))
1041                 goto out_free_pools;
1042 init_bs:
1043         if (bioset_init(&pools->bs, pool_size, front_pad, 0))
1044                 goto out_free_pools;
1045         if (t->integrity_supported &&
1046             bioset_integrity_create(&pools->bs, pool_size))
1047                 goto out_free_pools;
1048
1049         t->mempools = pools;
1050         return 0;
1051
1052 out_free_pools:
1053         dm_free_md_mempools(pools);
1054         return -ENOMEM;
1055 }
1056
1057 static int setup_indexes(struct dm_table *t)
1058 {
1059         int i;
1060         unsigned int total = 0;
1061         sector_t *indexes;
1062
1063         /* allocate the space for *all* the indexes */
1064         for (i = t->depth - 2; i >= 0; i--) {
1065                 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1066                 total += t->counts[i];
1067         }
1068
1069         indexes = kvcalloc(total, NODE_SIZE, GFP_KERNEL);
1070         if (!indexes)
1071                 return -ENOMEM;
1072
1073         /* set up internal nodes, bottom-up */
1074         for (i = t->depth - 2; i >= 0; i--) {
1075                 t->index[i] = indexes;
1076                 indexes += (KEYS_PER_NODE * t->counts[i]);
1077                 setup_btree_index(i, t);
1078         }
1079
1080         return 0;
1081 }
1082
1083 /*
1084  * Builds the btree to index the map.
1085  */
1086 static int dm_table_build_index(struct dm_table *t)
1087 {
1088         int r = 0;
1089         unsigned int leaf_nodes;
1090
1091         /* how many indexes will the btree have ? */
1092         leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1093         t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1094
1095         /* leaf layer has already been set up */
1096         t->counts[t->depth - 1] = leaf_nodes;
1097         t->index[t->depth - 1] = t->highs;
1098
1099         if (t->depth >= 2)
1100                 r = setup_indexes(t);
1101
1102         return r;
1103 }
1104
1105 static bool integrity_profile_exists(struct gendisk *disk)
1106 {
1107         return !!blk_get_integrity(disk);
1108 }
1109
1110 /*
1111  * Get a disk whose integrity profile reflects the table's profile.
1112  * Returns NULL if integrity support was inconsistent or unavailable.
1113  */
1114 static struct gendisk *dm_table_get_integrity_disk(struct dm_table *t)
1115 {
1116         struct list_head *devices = dm_table_get_devices(t);
1117         struct dm_dev_internal *dd = NULL;
1118         struct gendisk *prev_disk = NULL, *template_disk = NULL;
1119
1120         for (unsigned int i = 0; i < t->num_targets; i++) {
1121                 struct dm_target *ti = dm_table_get_target(t, i);
1122
1123                 if (!dm_target_passes_integrity(ti->type))
1124                         goto no_integrity;
1125         }
1126
1127         list_for_each_entry(dd, devices, list) {
1128                 template_disk = dd->dm_dev->bdev->bd_disk;
1129                 if (!integrity_profile_exists(template_disk))
1130                         goto no_integrity;
1131                 else if (prev_disk &&
1132                          blk_integrity_compare(prev_disk, template_disk) < 0)
1133                         goto no_integrity;
1134                 prev_disk = template_disk;
1135         }
1136
1137         return template_disk;
1138
1139 no_integrity:
1140         if (prev_disk)
1141                 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1142                        dm_device_name(t->md),
1143                        prev_disk->disk_name,
1144                        template_disk->disk_name);
1145         return NULL;
1146 }
1147
1148 /*
1149  * Register the mapped device for blk_integrity support if the
1150  * underlying devices have an integrity profile.  But all devices may
1151  * not have matching profiles (checking all devices isn't reliable
1152  * during table load because this table may use other DM device(s) which
1153  * must be resumed before they will have an initialized integity
1154  * profile).  Consequently, stacked DM devices force a 2 stage integrity
1155  * profile validation: First pass during table load, final pass during
1156  * resume.
1157  */
1158 static int dm_table_register_integrity(struct dm_table *t)
1159 {
1160         struct mapped_device *md = t->md;
1161         struct gendisk *template_disk = NULL;
1162
1163         /* If target handles integrity itself do not register it here. */
1164         if (t->integrity_added)
1165                 return 0;
1166
1167         template_disk = dm_table_get_integrity_disk(t);
1168         if (!template_disk)
1169                 return 0;
1170
1171         if (!integrity_profile_exists(dm_disk(md))) {
1172                 t->integrity_supported = true;
1173                 /*
1174                  * Register integrity profile during table load; we can do
1175                  * this because the final profile must match during resume.
1176                  */
1177                 blk_integrity_register(dm_disk(md),
1178                                        blk_get_integrity(template_disk));
1179                 return 0;
1180         }
1181
1182         /*
1183          * If DM device already has an initialized integrity
1184          * profile the new profile should not conflict.
1185          */
1186         if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1187                 DMERR("%s: conflict with existing integrity profile: "
1188                       "%s profile mismatch",
1189                       dm_device_name(t->md),
1190                       template_disk->disk_name);
1191                 return 1;
1192         }
1193
1194         /* Preserve existing integrity profile */
1195         t->integrity_supported = true;
1196         return 0;
1197 }
1198
1199 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
1200
1201 struct dm_crypto_profile {
1202         struct blk_crypto_profile profile;
1203         struct mapped_device *md;
1204 };
1205
1206 struct dm_keyslot_evict_args {
1207         const struct blk_crypto_key *key;
1208         int err;
1209 };
1210
1211 static int dm_keyslot_evict_callback(struct dm_target *ti, struct dm_dev *dev,
1212                                      sector_t start, sector_t len, void *data)
1213 {
1214         struct dm_keyslot_evict_args *args = data;
1215         int err;
1216
1217         err = blk_crypto_evict_key(bdev_get_queue(dev->bdev), args->key);
1218         if (!args->err)
1219                 args->err = err;
1220         /* Always try to evict the key from all devices. */
1221         return 0;
1222 }
1223
1224 /*
1225  * When an inline encryption key is evicted from a device-mapper device, evict
1226  * it from all the underlying devices.
1227  */
1228 static int dm_keyslot_evict(struct blk_crypto_profile *profile,
1229                             const struct blk_crypto_key *key, unsigned int slot)
1230 {
1231         struct mapped_device *md =
1232                 container_of(profile, struct dm_crypto_profile, profile)->md;
1233         struct dm_keyslot_evict_args args = { key };
1234         struct dm_table *t;
1235         int srcu_idx;
1236
1237         t = dm_get_live_table(md, &srcu_idx);
1238         if (!t)
1239                 return 0;
1240
1241         for (unsigned int i = 0; i < t->num_targets; i++) {
1242                 struct dm_target *ti = dm_table_get_target(t, i);
1243
1244                 if (!ti->type->iterate_devices)
1245                         continue;
1246                 ti->type->iterate_devices(ti, dm_keyslot_evict_callback, &args);
1247         }
1248
1249         dm_put_live_table(md, srcu_idx);
1250         return args.err;
1251 }
1252
1253 static int
1254 device_intersect_crypto_capabilities(struct dm_target *ti, struct dm_dev *dev,
1255                                      sector_t start, sector_t len, void *data)
1256 {
1257         struct blk_crypto_profile *parent = data;
1258         struct blk_crypto_profile *child =
1259                 bdev_get_queue(dev->bdev)->crypto_profile;
1260
1261         blk_crypto_intersect_capabilities(parent, child);
1262         return 0;
1263 }
1264
1265 void dm_destroy_crypto_profile(struct blk_crypto_profile *profile)
1266 {
1267         struct dm_crypto_profile *dmcp = container_of(profile,
1268                                                       struct dm_crypto_profile,
1269                                                       profile);
1270
1271         if (!profile)
1272                 return;
1273
1274         blk_crypto_profile_destroy(profile);
1275         kfree(dmcp);
1276 }
1277
1278 static void dm_table_destroy_crypto_profile(struct dm_table *t)
1279 {
1280         dm_destroy_crypto_profile(t->crypto_profile);
1281         t->crypto_profile = NULL;
1282 }
1283
1284 /*
1285  * Constructs and initializes t->crypto_profile with a crypto profile that
1286  * represents the common set of crypto capabilities of the devices described by
1287  * the dm_table.  However, if the constructed crypto profile doesn't support all
1288  * crypto capabilities that are supported by the current mapped_device, it
1289  * returns an error instead, since we don't support removing crypto capabilities
1290  * on table changes.  Finally, if the constructed crypto profile is "empty" (has
1291  * no crypto capabilities at all), it just sets t->crypto_profile to NULL.
1292  */
1293 static int dm_table_construct_crypto_profile(struct dm_table *t)
1294 {
1295         struct dm_crypto_profile *dmcp;
1296         struct blk_crypto_profile *profile;
1297         unsigned int i;
1298         bool empty_profile = true;
1299
1300         dmcp = kmalloc(sizeof(*dmcp), GFP_KERNEL);
1301         if (!dmcp)
1302                 return -ENOMEM;
1303         dmcp->md = t->md;
1304
1305         profile = &dmcp->profile;
1306         blk_crypto_profile_init(profile, 0);
1307         profile->ll_ops.keyslot_evict = dm_keyslot_evict;
1308         profile->max_dun_bytes_supported = UINT_MAX;
1309         memset(profile->modes_supported, 0xFF,
1310                sizeof(profile->modes_supported));
1311
1312         for (i = 0; i < t->num_targets; i++) {
1313                 struct dm_target *ti = dm_table_get_target(t, i);
1314
1315                 if (!dm_target_passes_crypto(ti->type)) {
1316                         blk_crypto_intersect_capabilities(profile, NULL);
1317                         break;
1318                 }
1319                 if (!ti->type->iterate_devices)
1320                         continue;
1321                 ti->type->iterate_devices(ti,
1322                                           device_intersect_crypto_capabilities,
1323                                           profile);
1324         }
1325
1326         if (t->md->queue &&
1327             !blk_crypto_has_capabilities(profile,
1328                                          t->md->queue->crypto_profile)) {
1329                 DMERR("Inline encryption capabilities of new DM table were more restrictive than the old table's. This is not supported!");
1330                 dm_destroy_crypto_profile(profile);
1331                 return -EINVAL;
1332         }
1333
1334         /*
1335          * If the new profile doesn't actually support any crypto capabilities,
1336          * we may as well represent it with a NULL profile.
1337          */
1338         for (i = 0; i < ARRAY_SIZE(profile->modes_supported); i++) {
1339                 if (profile->modes_supported[i]) {
1340                         empty_profile = false;
1341                         break;
1342                 }
1343         }
1344
1345         if (empty_profile) {
1346                 dm_destroy_crypto_profile(profile);
1347                 profile = NULL;
1348         }
1349
1350         /*
1351          * t->crypto_profile is only set temporarily while the table is being
1352          * set up, and it gets set to NULL after the profile has been
1353          * transferred to the request_queue.
1354          */
1355         t->crypto_profile = profile;
1356
1357         return 0;
1358 }
1359
1360 static void dm_update_crypto_profile(struct request_queue *q,
1361                                      struct dm_table *t)
1362 {
1363         if (!t->crypto_profile)
1364                 return;
1365
1366         /* Make the crypto profile less restrictive. */
1367         if (!q->crypto_profile) {
1368                 blk_crypto_register(t->crypto_profile, q);
1369         } else {
1370                 blk_crypto_update_capabilities(q->crypto_profile,
1371                                                t->crypto_profile);
1372                 dm_destroy_crypto_profile(t->crypto_profile);
1373         }
1374         t->crypto_profile = NULL;
1375 }
1376
1377 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1378
1379 static int dm_table_construct_crypto_profile(struct dm_table *t)
1380 {
1381         return 0;
1382 }
1383
1384 void dm_destroy_crypto_profile(struct blk_crypto_profile *profile)
1385 {
1386 }
1387
1388 static void dm_table_destroy_crypto_profile(struct dm_table *t)
1389 {
1390 }
1391
1392 static void dm_update_crypto_profile(struct request_queue *q,
1393                                      struct dm_table *t)
1394 {
1395 }
1396
1397 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1398
1399 /*
1400  * Prepares the table for use by building the indices,
1401  * setting the type, and allocating mempools.
1402  */
1403 int dm_table_complete(struct dm_table *t)
1404 {
1405         int r;
1406
1407         r = dm_table_determine_type(t);
1408         if (r) {
1409                 DMERR("unable to determine table type");
1410                 return r;
1411         }
1412
1413         r = dm_table_build_index(t);
1414         if (r) {
1415                 DMERR("unable to build btrees");
1416                 return r;
1417         }
1418
1419         r = dm_table_register_integrity(t);
1420         if (r) {
1421                 DMERR("could not register integrity profile.");
1422                 return r;
1423         }
1424
1425         r = dm_table_construct_crypto_profile(t);
1426         if (r) {
1427                 DMERR("could not construct crypto profile.");
1428                 return r;
1429         }
1430
1431         r = dm_table_alloc_md_mempools(t, t->md);
1432         if (r)
1433                 DMERR("unable to allocate mempools");
1434
1435         return r;
1436 }
1437
1438 static DEFINE_MUTEX(_event_lock);
1439 void dm_table_event_callback(struct dm_table *t,
1440                              void (*fn)(void *), void *context)
1441 {
1442         mutex_lock(&_event_lock);
1443         t->event_fn = fn;
1444         t->event_context = context;
1445         mutex_unlock(&_event_lock);
1446 }
1447
1448 void dm_table_event(struct dm_table *t)
1449 {
1450         mutex_lock(&_event_lock);
1451         if (t->event_fn)
1452                 t->event_fn(t->event_context);
1453         mutex_unlock(&_event_lock);
1454 }
1455 EXPORT_SYMBOL(dm_table_event);
1456
1457 inline sector_t dm_table_get_size(struct dm_table *t)
1458 {
1459         return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1460 }
1461 EXPORT_SYMBOL(dm_table_get_size);
1462
1463 /*
1464  * Search the btree for the correct target.
1465  *
1466  * Caller should check returned pointer for NULL
1467  * to trap I/O beyond end of device.
1468  */
1469 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1470 {
1471         unsigned int l, n = 0, k = 0;
1472         sector_t *node;
1473
1474         if (unlikely(sector >= dm_table_get_size(t)))
1475                 return NULL;
1476
1477         for (l = 0; l < t->depth; l++) {
1478                 n = get_child(n, k);
1479                 node = get_node(t, l, n);
1480
1481                 for (k = 0; k < KEYS_PER_NODE; k++)
1482                         if (node[k] >= sector)
1483                                 break;
1484         }
1485
1486         return &t->targets[(KEYS_PER_NODE * n) + k];
1487 }
1488
1489 static int device_not_poll_capable(struct dm_target *ti, struct dm_dev *dev,
1490                                    sector_t start, sector_t len, void *data)
1491 {
1492         struct request_queue *q = bdev_get_queue(dev->bdev);
1493
1494         return !test_bit(QUEUE_FLAG_POLL, &q->queue_flags);
1495 }
1496
1497 /*
1498  * type->iterate_devices() should be called when the sanity check needs to
1499  * iterate and check all underlying data devices. iterate_devices() will
1500  * iterate all underlying data devices until it encounters a non-zero return
1501  * code, returned by whether the input iterate_devices_callout_fn, or
1502  * iterate_devices() itself internally.
1503  *
1504  * For some target type (e.g. dm-stripe), one call of iterate_devices() may
1505  * iterate multiple underlying devices internally, in which case a non-zero
1506  * return code returned by iterate_devices_callout_fn will stop the iteration
1507  * in advance.
1508  *
1509  * Cases requiring _any_ underlying device supporting some kind of attribute,
1510  * should use the iteration structure like dm_table_any_dev_attr(), or call
1511  * it directly. @func should handle semantics of positive examples, e.g.
1512  * capable of something.
1513  *
1514  * Cases requiring _all_ underlying devices supporting some kind of attribute,
1515  * should use the iteration structure like dm_table_supports_nowait() or
1516  * dm_table_supports_discards(). Or introduce dm_table_all_devs_attr() that
1517  * uses an @anti_func that handle semantics of counter examples, e.g. not
1518  * capable of something. So: return !dm_table_any_dev_attr(t, anti_func, data);
1519  */
1520 static bool dm_table_any_dev_attr(struct dm_table *t,
1521                                   iterate_devices_callout_fn func, void *data)
1522 {
1523         for (unsigned int i = 0; i < t->num_targets; i++) {
1524                 struct dm_target *ti = dm_table_get_target(t, i);
1525
1526                 if (ti->type->iterate_devices &&
1527                     ti->type->iterate_devices(ti, func, data))
1528                         return true;
1529         }
1530
1531         return false;
1532 }
1533
1534 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1535                         sector_t start, sector_t len, void *data)
1536 {
1537         unsigned int *num_devices = data;
1538
1539         (*num_devices)++;
1540
1541         return 0;
1542 }
1543
1544 static bool dm_table_supports_poll(struct dm_table *t)
1545 {
1546         for (unsigned int i = 0; i < t->num_targets; i++) {
1547                 struct dm_target *ti = dm_table_get_target(t, i);
1548
1549                 if (!ti->type->iterate_devices ||
1550                     ti->type->iterate_devices(ti, device_not_poll_capable, NULL))
1551                         return false;
1552         }
1553
1554         return true;
1555 }
1556
1557 /*
1558  * Check whether a table has no data devices attached using each
1559  * target's iterate_devices method.
1560  * Returns false if the result is unknown because a target doesn't
1561  * support iterate_devices.
1562  */
1563 bool dm_table_has_no_data_devices(struct dm_table *t)
1564 {
1565         for (unsigned int i = 0; i < t->num_targets; i++) {
1566                 struct dm_target *ti = dm_table_get_target(t, i);
1567                 unsigned int num_devices = 0;
1568
1569                 if (!ti->type->iterate_devices)
1570                         return false;
1571
1572                 ti->type->iterate_devices(ti, count_device, &num_devices);
1573                 if (num_devices)
1574                         return false;
1575         }
1576
1577         return true;
1578 }
1579
1580 static int device_not_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1581                                   sector_t start, sector_t len, void *data)
1582 {
1583         struct request_queue *q = bdev_get_queue(dev->bdev);
1584         enum blk_zoned_model *zoned_model = data;
1585
1586         return blk_queue_zoned_model(q) != *zoned_model;
1587 }
1588
1589 /*
1590  * Check the device zoned model based on the target feature flag. If the target
1591  * has the DM_TARGET_ZONED_HM feature flag set, host-managed zoned devices are
1592  * also accepted but all devices must have the same zoned model. If the target
1593  * has the DM_TARGET_MIXED_ZONED_MODEL feature set, the devices can have any
1594  * zoned model with all zoned devices having the same zone size.
1595  */
1596 static bool dm_table_supports_zoned_model(struct dm_table *t,
1597                                           enum blk_zoned_model zoned_model)
1598 {
1599         for (unsigned int i = 0; i < t->num_targets; i++) {
1600                 struct dm_target *ti = dm_table_get_target(t, i);
1601
1602                 if (dm_target_supports_zoned_hm(ti->type)) {
1603                         if (!ti->type->iterate_devices ||
1604                             ti->type->iterate_devices(ti, device_not_zoned_model,
1605                                                       &zoned_model))
1606                                 return false;
1607                 } else if (!dm_target_supports_mixed_zoned_model(ti->type)) {
1608                         if (zoned_model == BLK_ZONED_HM)
1609                                 return false;
1610                 }
1611         }
1612
1613         return true;
1614 }
1615
1616 static int device_not_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1617                                            sector_t start, sector_t len, void *data)
1618 {
1619         unsigned int *zone_sectors = data;
1620
1621         if (!bdev_is_zoned(dev->bdev))
1622                 return 0;
1623         return bdev_zone_sectors(dev->bdev) != *zone_sectors;
1624 }
1625
1626 /*
1627  * Check consistency of zoned model and zone sectors across all targets. For
1628  * zone sectors, if the destination device is a zoned block device, it shall
1629  * have the specified zone_sectors.
1630  */
1631 static int validate_hardware_zoned_model(struct dm_table *t,
1632                                          enum blk_zoned_model zoned_model,
1633                                          unsigned int zone_sectors)
1634 {
1635         if (zoned_model == BLK_ZONED_NONE)
1636                 return 0;
1637
1638         if (!dm_table_supports_zoned_model(t, zoned_model)) {
1639                 DMERR("%s: zoned model is not consistent across all devices",
1640                       dm_device_name(t->md));
1641                 return -EINVAL;
1642         }
1643
1644         /* Check zone size validity and compatibility */
1645         if (!zone_sectors || !is_power_of_2(zone_sectors))
1646                 return -EINVAL;
1647
1648         if (dm_table_any_dev_attr(t, device_not_matches_zone_sectors, &zone_sectors)) {
1649                 DMERR("%s: zone sectors is not consistent across all zoned devices",
1650                       dm_device_name(t->md));
1651                 return -EINVAL;
1652         }
1653
1654         return 0;
1655 }
1656
1657 /*
1658  * Establish the new table's queue_limits and validate them.
1659  */
1660 int dm_calculate_queue_limits(struct dm_table *t,
1661                               struct queue_limits *limits)
1662 {
1663         struct queue_limits ti_limits;
1664         enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1665         unsigned int zone_sectors = 0;
1666
1667         blk_set_stacking_limits(limits);
1668
1669         for (unsigned int i = 0; i < t->num_targets; i++) {
1670                 struct dm_target *ti = dm_table_get_target(t, i);
1671
1672                 blk_set_stacking_limits(&ti_limits);
1673
1674                 if (!ti->type->iterate_devices)
1675                         goto combine_limits;
1676
1677                 /*
1678                  * Combine queue limits of all the devices this target uses.
1679                  */
1680                 ti->type->iterate_devices(ti, dm_set_device_limits,
1681                                           &ti_limits);
1682
1683                 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1684                         /*
1685                          * After stacking all limits, validate all devices
1686                          * in table support this zoned model and zone sectors.
1687                          */
1688                         zoned_model = ti_limits.zoned;
1689                         zone_sectors = ti_limits.chunk_sectors;
1690                 }
1691
1692                 /* Set I/O hints portion of queue limits */
1693                 if (ti->type->io_hints)
1694                         ti->type->io_hints(ti, &ti_limits);
1695
1696                 /*
1697                  * Check each device area is consistent with the target's
1698                  * overall queue limits.
1699                  */
1700                 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1701                                               &ti_limits))
1702                         return -EINVAL;
1703
1704 combine_limits:
1705                 /*
1706                  * Merge this target's queue limits into the overall limits
1707                  * for the table.
1708                  */
1709                 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1710                         DMWARN("%s: adding target device "
1711                                "(start sect %llu len %llu) "
1712                                "caused an alignment inconsistency",
1713                                dm_device_name(t->md),
1714                                (unsigned long long) ti->begin,
1715                                (unsigned long long) ti->len);
1716         }
1717
1718         /*
1719          * Verify that the zoned model and zone sectors, as determined before
1720          * any .io_hints override, are the same across all devices in the table.
1721          * - this is especially relevant if .io_hints is emulating a disk-managed
1722          *   zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1723          * BUT...
1724          */
1725         if (limits->zoned != BLK_ZONED_NONE) {
1726                 /*
1727                  * ...IF the above limits stacking determined a zoned model
1728                  * validate that all of the table's devices conform to it.
1729                  */
1730                 zoned_model = limits->zoned;
1731                 zone_sectors = limits->chunk_sectors;
1732         }
1733         if (validate_hardware_zoned_model(t, zoned_model, zone_sectors))
1734                 return -EINVAL;
1735
1736         return validate_hardware_logical_block_alignment(t, limits);
1737 }
1738
1739 /*
1740  * Verify that all devices have an integrity profile that matches the
1741  * DM device's registered integrity profile.  If the profiles don't
1742  * match then unregister the DM device's integrity profile.
1743  */
1744 static void dm_table_verify_integrity(struct dm_table *t)
1745 {
1746         struct gendisk *template_disk = NULL;
1747
1748         if (t->integrity_added)
1749                 return;
1750
1751         if (t->integrity_supported) {
1752                 /*
1753                  * Verify that the original integrity profile
1754                  * matches all the devices in this table.
1755                  */
1756                 template_disk = dm_table_get_integrity_disk(t);
1757                 if (template_disk &&
1758                     blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1759                         return;
1760         }
1761
1762         if (integrity_profile_exists(dm_disk(t->md))) {
1763                 DMWARN("%s: unable to establish an integrity profile",
1764                        dm_device_name(t->md));
1765                 blk_integrity_unregister(dm_disk(t->md));
1766         }
1767 }
1768
1769 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1770                                 sector_t start, sector_t len, void *data)
1771 {
1772         unsigned long flush = (unsigned long) data;
1773         struct request_queue *q = bdev_get_queue(dev->bdev);
1774
1775         return (q->queue_flags & flush);
1776 }
1777
1778 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1779 {
1780         /*
1781          * Require at least one underlying device to support flushes.
1782          * t->devices includes internal dm devices such as mirror logs
1783          * so we need to use iterate_devices here, which targets
1784          * supporting flushes must provide.
1785          */
1786         for (unsigned int i = 0; i < t->num_targets; i++) {
1787                 struct dm_target *ti = dm_table_get_target(t, i);
1788
1789                 if (!ti->num_flush_bios)
1790                         continue;
1791
1792                 if (ti->flush_supported)
1793                         return true;
1794
1795                 if (ti->type->iterate_devices &&
1796                     ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1797                         return true;
1798         }
1799
1800         return false;
1801 }
1802
1803 static int device_dax_write_cache_enabled(struct dm_target *ti,
1804                                           struct dm_dev *dev, sector_t start,
1805                                           sector_t len, void *data)
1806 {
1807         struct dax_device *dax_dev = dev->dax_dev;
1808
1809         if (!dax_dev)
1810                 return false;
1811
1812         if (dax_write_cache_enabled(dax_dev))
1813                 return true;
1814         return false;
1815 }
1816
1817 static int device_is_rotational(struct dm_target *ti, struct dm_dev *dev,
1818                                 sector_t start, sector_t len, void *data)
1819 {
1820         return !bdev_nonrot(dev->bdev);
1821 }
1822
1823 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1824                              sector_t start, sector_t len, void *data)
1825 {
1826         struct request_queue *q = bdev_get_queue(dev->bdev);
1827
1828         return !blk_queue_add_random(q);
1829 }
1830
1831 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1832                                            sector_t start, sector_t len, void *data)
1833 {
1834         struct request_queue *q = bdev_get_queue(dev->bdev);
1835
1836         return !q->limits.max_write_zeroes_sectors;
1837 }
1838
1839 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1840 {
1841         for (unsigned int i = 0; i < t->num_targets; i++) {
1842                 struct dm_target *ti = dm_table_get_target(t, i);
1843
1844                 if (!ti->num_write_zeroes_bios)
1845                         return false;
1846
1847                 if (!ti->type->iterate_devices ||
1848                     ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1849                         return false;
1850         }
1851
1852         return true;
1853 }
1854
1855 static int device_not_nowait_capable(struct dm_target *ti, struct dm_dev *dev,
1856                                      sector_t start, sector_t len, void *data)
1857 {
1858         return !bdev_nowait(dev->bdev);
1859 }
1860
1861 static bool dm_table_supports_nowait(struct dm_table *t)
1862 {
1863         for (unsigned int i = 0; i < t->num_targets; i++) {
1864                 struct dm_target *ti = dm_table_get_target(t, i);
1865
1866                 if (!dm_target_supports_nowait(ti->type))
1867                         return false;
1868
1869                 if (!ti->type->iterate_devices ||
1870                     ti->type->iterate_devices(ti, device_not_nowait_capable, NULL))
1871                         return false;
1872         }
1873
1874         return true;
1875 }
1876
1877 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1878                                       sector_t start, sector_t len, void *data)
1879 {
1880         return !bdev_max_discard_sectors(dev->bdev);
1881 }
1882
1883 static bool dm_table_supports_discards(struct dm_table *t)
1884 {
1885         for (unsigned int i = 0; i < t->num_targets; i++) {
1886                 struct dm_target *ti = dm_table_get_target(t, i);
1887
1888                 if (!ti->num_discard_bios)
1889                         return false;
1890
1891                 /*
1892                  * Either the target provides discard support (as implied by setting
1893                  * 'discards_supported') or it relies on _all_ data devices having
1894                  * discard support.
1895                  */
1896                 if (!ti->discards_supported &&
1897                     (!ti->type->iterate_devices ||
1898                      ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1899                         return false;
1900         }
1901
1902         return true;
1903 }
1904
1905 static int device_not_secure_erase_capable(struct dm_target *ti,
1906                                            struct dm_dev *dev, sector_t start,
1907                                            sector_t len, void *data)
1908 {
1909         return !bdev_max_secure_erase_sectors(dev->bdev);
1910 }
1911
1912 static bool dm_table_supports_secure_erase(struct dm_table *t)
1913 {
1914         for (unsigned int i = 0; i < t->num_targets; i++) {
1915                 struct dm_target *ti = dm_table_get_target(t, i);
1916
1917                 if (!ti->num_secure_erase_bios)
1918                         return false;
1919
1920                 if (!ti->type->iterate_devices ||
1921                     ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1922                         return false;
1923         }
1924
1925         return true;
1926 }
1927
1928 static int device_requires_stable_pages(struct dm_target *ti,
1929                                         struct dm_dev *dev, sector_t start,
1930                                         sector_t len, void *data)
1931 {
1932         return bdev_stable_writes(dev->bdev);
1933 }
1934
1935 int dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1936                               struct queue_limits *limits)
1937 {
1938         bool wc = false, fua = false;
1939         int r;
1940
1941         /*
1942          * Copy table's limits to the DM device's request_queue
1943          */
1944         q->limits = *limits;
1945
1946         if (dm_table_supports_nowait(t))
1947                 blk_queue_flag_set(QUEUE_FLAG_NOWAIT, q);
1948         else
1949                 blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, q);
1950
1951         if (!dm_table_supports_discards(t)) {
1952                 q->limits.max_discard_sectors = 0;
1953                 q->limits.max_hw_discard_sectors = 0;
1954                 q->limits.discard_granularity = 0;
1955                 q->limits.discard_alignment = 0;
1956                 q->limits.discard_misaligned = 0;
1957         }
1958
1959         if (!dm_table_supports_secure_erase(t))
1960                 q->limits.max_secure_erase_sectors = 0;
1961
1962         if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1963                 wc = true;
1964                 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1965                         fua = true;
1966         }
1967         blk_queue_write_cache(q, wc, fua);
1968
1969         if (dm_table_supports_dax(t, device_not_dax_capable)) {
1970                 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
1971                 if (dm_table_supports_dax(t, device_not_dax_synchronous_capable))
1972                         set_dax_synchronous(t->md->dax_dev);
1973         }
1974         else
1975                 blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
1976
1977         if (dm_table_any_dev_attr(t, device_dax_write_cache_enabled, NULL))
1978                 dax_write_cache(t->md->dax_dev, true);
1979
1980         /* Ensure that all underlying devices are non-rotational. */
1981         if (dm_table_any_dev_attr(t, device_is_rotational, NULL))
1982                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
1983         else
1984                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
1985
1986         if (!dm_table_supports_write_zeroes(t))
1987                 q->limits.max_write_zeroes_sectors = 0;
1988
1989         dm_table_verify_integrity(t);
1990
1991         /*
1992          * Some devices don't use blk_integrity but still want stable pages
1993          * because they do their own checksumming.
1994          * If any underlying device requires stable pages, a table must require
1995          * them as well.  Only targets that support iterate_devices are considered:
1996          * don't want error, zero, etc to require stable pages.
1997          */
1998         if (dm_table_any_dev_attr(t, device_requires_stable_pages, NULL))
1999                 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, q);
2000         else
2001                 blk_queue_flag_clear(QUEUE_FLAG_STABLE_WRITES, q);
2002
2003         /*
2004          * Determine whether or not this queue's I/O timings contribute
2005          * to the entropy pool, Only request-based targets use this.
2006          * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
2007          * have it set.
2008          */
2009         if (blk_queue_add_random(q) &&
2010             dm_table_any_dev_attr(t, device_is_not_random, NULL))
2011                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2012
2013         /*
2014          * For a zoned target, setup the zones related queue attributes
2015          * and resources necessary for zone append emulation if necessary.
2016          */
2017         if (blk_queue_is_zoned(q)) {
2018                 r = dm_set_zones_restrictions(t, q);
2019                 if (r)
2020                         return r;
2021                 if (!static_key_enabled(&zoned_enabled.key))
2022                         static_branch_enable(&zoned_enabled);
2023         }
2024
2025         dm_update_crypto_profile(q, t);
2026         disk_update_readahead(t->md->disk);
2027
2028         /*
2029          * Check for request-based device is left to
2030          * dm_mq_init_request_queue()->blk_mq_init_allocated_queue().
2031          *
2032          * For bio-based device, only set QUEUE_FLAG_POLL when all
2033          * underlying devices supporting polling.
2034          */
2035         if (__table_type_bio_based(t->type)) {
2036                 if (dm_table_supports_poll(t))
2037                         blk_queue_flag_set(QUEUE_FLAG_POLL, q);
2038                 else
2039                         blk_queue_flag_clear(QUEUE_FLAG_POLL, q);
2040         }
2041
2042         return 0;
2043 }
2044
2045 struct list_head *dm_table_get_devices(struct dm_table *t)
2046 {
2047         return &t->devices;
2048 }
2049
2050 fmode_t dm_table_get_mode(struct dm_table *t)
2051 {
2052         return t->mode;
2053 }
2054 EXPORT_SYMBOL(dm_table_get_mode);
2055
2056 enum suspend_mode {
2057         PRESUSPEND,
2058         PRESUSPEND_UNDO,
2059         POSTSUSPEND,
2060 };
2061
2062 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
2063 {
2064         lockdep_assert_held(&t->md->suspend_lock);
2065
2066         for (unsigned int i = 0; i < t->num_targets; i++) {
2067                 struct dm_target *ti = dm_table_get_target(t, i);
2068
2069                 switch (mode) {
2070                 case PRESUSPEND:
2071                         if (ti->type->presuspend)
2072                                 ti->type->presuspend(ti);
2073                         break;
2074                 case PRESUSPEND_UNDO:
2075                         if (ti->type->presuspend_undo)
2076                                 ti->type->presuspend_undo(ti);
2077                         break;
2078                 case POSTSUSPEND:
2079                         if (ti->type->postsuspend)
2080                                 ti->type->postsuspend(ti);
2081                         break;
2082                 }
2083         }
2084 }
2085
2086 void dm_table_presuspend_targets(struct dm_table *t)
2087 {
2088         if (!t)
2089                 return;
2090
2091         suspend_targets(t, PRESUSPEND);
2092 }
2093
2094 void dm_table_presuspend_undo_targets(struct dm_table *t)
2095 {
2096         if (!t)
2097                 return;
2098
2099         suspend_targets(t, PRESUSPEND_UNDO);
2100 }
2101
2102 void dm_table_postsuspend_targets(struct dm_table *t)
2103 {
2104         if (!t)
2105                 return;
2106
2107         suspend_targets(t, POSTSUSPEND);
2108 }
2109
2110 int dm_table_resume_targets(struct dm_table *t)
2111 {
2112         unsigned int i;
2113         int r = 0;
2114
2115         lockdep_assert_held(&t->md->suspend_lock);
2116
2117         for (i = 0; i < t->num_targets; i++) {
2118                 struct dm_target *ti = dm_table_get_target(t, i);
2119
2120                 if (!ti->type->preresume)
2121                         continue;
2122
2123                 r = ti->type->preresume(ti);
2124                 if (r) {
2125                         DMERR("%s: %s: preresume failed, error = %d",
2126                               dm_device_name(t->md), ti->type->name, r);
2127                         return r;
2128                 }
2129         }
2130
2131         for (i = 0; i < t->num_targets; i++) {
2132                 struct dm_target *ti = dm_table_get_target(t, i);
2133
2134                 if (ti->type->resume)
2135                         ti->type->resume(ti);
2136         }
2137
2138         return 0;
2139 }
2140
2141 struct mapped_device *dm_table_get_md(struct dm_table *t)
2142 {
2143         return t->md;
2144 }
2145 EXPORT_SYMBOL(dm_table_get_md);
2146
2147 const char *dm_table_device_name(struct dm_table *t)
2148 {
2149         return dm_device_name(t->md);
2150 }
2151 EXPORT_SYMBOL_GPL(dm_table_device_name);
2152
2153 void dm_table_run_md_queue_async(struct dm_table *t)
2154 {
2155         if (!dm_table_request_based(t))
2156                 return;
2157
2158         if (t->md->queue)
2159                 blk_mq_run_hw_queues(t->md->queue, true);
2160 }
2161 EXPORT_SYMBOL(dm_table_run_md_queue_async);
2162