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
  10 #include <linux/module.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/blkdev.h>
  13 #include <linux/namei.h>
  14 #include <linux/ctype.h>
  15 #include <linux/string.h>
  16 #include <linux/slab.h>
  17 #include <linux/interrupt.h>
  18 #include <linux/mutex.h>
  19 #include <linux/delay.h>
  20 #include <linux/atomic.h>
  21 #include <linux/blk-mq.h>
  22 #include <linux/mount.h>
  23 #include <linux/dax.h>
  24
  25 #define DM_MSG_PREFIX "table"
  26
  27 #define NODE_SIZE L1_CACHE_BYTES
  28 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
  29 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
  30
  31 /*
  32  * Similar to ceiling(log_size(n))
  33  */
  34 static unsigned int int_log(unsigned int n, unsigned int base)
  35 {
  36         int result = 0;
  37
  38         while (n > 1) {
  39                 n = dm_div_up(n, base);
  40                 result++;
  41         }
  42
  43         return result;
  44 }
  45
  46 /*
  47  * Calculate the index of the child node of the n'th node k'th key.
  48  */
  49 static inline unsigned int get_child(unsigned int n, unsigned int k)
  50 {
  51         return (n * CHILDREN_PER_NODE) + k;
  52 }
  53
  54 /*
  55  * Return the n'th node of level l from table t.
  56  */
  57 static inline sector_t *get_node(struct dm_table *t,
  58                                  unsigned int l, unsigned int n)
  59 {
  60         return t->index[l] + (n * KEYS_PER_NODE);
  61 }
  62
  63 /*
  64  * Return the highest key that you could lookup from the n'th
  65  * node on level l of the btree.
  66  */
  67 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
  68 {
  69         for (; l < t->depth - 1; l++)
  70                 n = get_child(n, CHILDREN_PER_NODE - 1);
  71
  72         if (n >= t->counts[l])
  73                 return (sector_t) - 1;
  74
  75         return get_node(t, l, n)[KEYS_PER_NODE - 1];
  76 }
  77
  78 /*
  79  * Fills in a level of the btree based on the highs of the level
  80  * below it.
  81  */
  82 static int setup_btree_index(unsigned int l, struct dm_table *t)
  83 {
  84         unsigned int n, k;
  85         sector_t *node;
  86
  87         for (n = 0U; n < t->counts[l]; n++) {
  88                 node = get_node(t, l, n);
  89
  90                 for (k = 0U; k < KEYS_PER_NODE; k++)
  91                         node[k] = high(t, l + 1, get_child(n, k));
  92         }
  93
  94         return 0;
  95 }
  96
  97 /*
  98  * highs, and targets are managed as dynamic arrays during a
  99  * table load.
 100  */
 101 static int alloc_targets(struct dm_table *t, unsigned int num)
 102 {
 103         sector_t *n_highs;
 104         struct dm_target *n_targets;
 105
 106         /*
 107          * Allocate both the target array and offset array at once.
 108          */
 109         n_highs = kvcalloc(num, sizeof(struct dm_target) + sizeof(sector_t),
 110                            GFP_KERNEL);
 111         if (!n_highs)
 112                 return -ENOMEM;
 113
 114         n_targets = (struct dm_target *) (n_highs + num);
 115
 116         memset(n_highs, -1, sizeof(*n_highs) * num);
 117         kvfree(t->highs);
 118
 119         t->num_allocated = num;
 120         t->highs = n_highs;
 121         t->targets = n_targets;
 122
 123         return 0;
 124 }
 125
 126 int dm_table_create(struct dm_table **result, fmode_t mode,
 127                     unsigned num_targets, struct mapped_device *md)
 128 {
 129         struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
 130
 131         if (!t)
 132                 return -ENOMEM;
 133
 134         INIT_LIST_HEAD(&t->devices);
 135
 136         if (!num_targets)
 137                 num_targets = KEYS_PER_NODE;
 138
 139         num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
 140
 141         if (!num_targets) {
 142                 kfree(t);
 143                 return -ENOMEM;
 144         }
 145
 146         if (alloc_targets(t, num_targets)) {
 147                 kfree(t);
 148                 return -ENOMEM;
 149         }
 150
 151         t->type = DM_TYPE_NONE;
 152         t->mode = mode;
 153         t->md = md;
 154         *result = t;
 155         return 0;
 156 }
 157
 158 static void free_devices(struct list_head *devices, struct mapped_device *md)
 159 {
 160         struct list_head *tmp, *next;
 161
 162         list_for_each_safe(tmp, next, devices) {
 163                 struct dm_dev_internal *dd =
 164                     list_entry(tmp, struct dm_dev_internal, list);
 165                 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
 166                        dm_device_name(md), dd->dm_dev->name);
 167                 dm_put_table_device(md, dd->dm_dev);
 168                 kfree(dd);
 169         }
 170 }
 171
 172 static void dm_table_destroy_keyslot_manager(struct dm_table *t);
 173
 174 void dm_table_destroy(struct dm_table *t)
 175 {
 176         unsigned int i;
 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 (i = 0; i < t->num_targets; i++) {
 187                 struct dm_target *tgt = t->targets + i;
 188
 189                 if (tgt->type->dtr)
 190                         tgt->type->dtr(tgt);
 191
 192                 dm_put_target_type(tgt->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_keyslot_manager(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 =
 230                 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
 231         unsigned short logical_block_size_sectors =
 232                 limits->logical_block_size >> SECTOR_SHIFT;
 233         char b[BDEVNAME_SIZE];
 234
 235         if (!dev_size)
 236                 return 0;
 237
 238         if ((start >= dev_size) || (start + len > dev_size)) {
 239                 DMWARN("%s: %s too small for target: "
 240                        "start=%llu, len=%llu, dev_size=%llu",
 241                        dm_device_name(ti->table->md), bdevname(bdev, b),
 242                        (unsigned long long)start,
 243                        (unsigned long long)len,
 244                        (unsigned long long)dev_size);
 245                 return 1;
 246         }
 247
 248         /*
 249          * If the target is mapped to zoned block device(s), check
 250          * that the zones are not partially mapped.
 251          */
 252         if (bdev_is_zoned(bdev)) {
 253                 unsigned int zone_sectors = bdev_zone_sectors(bdev);
 254
 255                 if (start & (zone_sectors - 1)) {
 256                         DMWARN("%s: start=%llu not aligned to h/w zone size %u of %s",
 257                                dm_device_name(ti->table->md),
 258                                (unsigned long long)start,
 259                                zone_sectors, bdevname(bdev, b));
 260                         return 1;
 261                 }
 262
 263                 /*
 264                  * Note: The last zone of a zoned block device may be smaller
 265                  * than other zones. So for a target mapping the end of a
 266                  * zoned block device with such a zone, len would not be zone
 267                  * aligned. We do not allow such last smaller zone to be part
 268                  * of the mapping here to ensure that mappings with multiple
 269                  * devices do not end up with a smaller zone in the middle of
 270                  * the sector range.
 271                  */
 272                 if (len & (zone_sectors - 1)) {
 273                         DMWARN("%s: len=%llu not aligned to h/w zone size %u of %s",
 274                                dm_device_name(ti->table->md),
 275                                (unsigned long long)len,
 276                                zone_sectors, bdevname(bdev, b));
 277                         return 1;
 278                 }
 279         }
 280
 281         if (logical_block_size_sectors <= 1)
 282                 return 0;
 283
 284         if (start & (logical_block_size_sectors - 1)) {
 285                 DMWARN("%s: start=%llu not aligned to h/w "
 286                        "logical block size %u of %s",
 287                        dm_device_name(ti->table->md),
 288                        (unsigned long long)start,
 289                        limits->logical_block_size, bdevname(bdev, b));
 290                 return 1;
 291         }
 292
 293         if (len & (logical_block_size_sectors - 1)) {
 294                 DMWARN("%s: len=%llu not aligned to h/w "
 295                        "logical block size %u of %s",
 296                        dm_device_name(ti->table->md),
 297                        (unsigned long long)len,
 298                        limits->logical_block_size, bdevname(bdev, b));
 299                 return 1;
 300         }
 301
 302         return 0;
 303 }
 304
 305 /*
 306  * This upgrades the mode on an already open dm_dev, being
 307  * careful to leave things as they were if we fail to reopen the
 308  * device and not to touch the existing bdev field in case
 309  * it is accessed concurrently.
 310  */
 311 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
 312                         struct mapped_device *md)
 313 {
 314         int r;
 315         struct dm_dev *old_dev, *new_dev;
 316
 317         old_dev = dd->dm_dev;
 318
 319         r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
 320                                 dd->dm_dev->mode | new_mode, &new_dev);
 321         if (r)
 322                 return r;
 323
 324         dd->dm_dev = new_dev;
 325         dm_put_table_device(md, old_dev);
 326
 327         return 0;
 328 }
 329
 330 /*
 331  * Convert the path to a device
 332  */
 333 dev_t dm_get_dev_t(const char *path)
 334 {
 335         dev_t dev;
 336
 337         if (lookup_bdev(path, &dev))
 338                 dev = name_to_dev_t(path);
 339         return dev;
 340 }
 341 EXPORT_SYMBOL_GPL(dm_get_dev_t);
 342
 343 /*
 344  * Add a device to the list, or just increment the usage count if
 345  * it's already present.
 346  */
 347 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
 348                   struct dm_dev **result)
 349 {
 350         int r;
 351         dev_t dev;
 352         unsigned int major, minor;
 353         char dummy;
 354         struct dm_dev_internal *dd;
 355         struct dm_table *t = ti->table;
 356
 357         BUG_ON(!t);
 358
 359         if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
 360                 /* Extract the major/minor numbers */
 361                 dev = MKDEV(major, minor);
 362                 if (MAJOR(dev) != major || MINOR(dev) != minor)
 363                         return -EOVERFLOW;
 364         } else {
 365                 dev = dm_get_dev_t(path);
 366                 if (!dev)
 367                         return -ENODEV;
 368         }
 369
 370         dd = find_device(&t->devices, dev);
 371         if (!dd) {
 372                 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
 373                 if (!dd)
 374                         return -ENOMEM;
 375
 376                 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
 377                         kfree(dd);
 378                         return r;
 379                 }
 380
 381                 refcount_set(&dd->count, 1);
 382                 list_add(&dd->list, &t->devices);
 383                 goto out;
 384
 385         } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
 386                 r = upgrade_mode(dd, mode, t->md);
 387                 if (r)
 388                         return r;
 389         }
 390         refcount_inc(&dd->count);
 391 out:
 392         *result = dd->dm_dev;
 393         return 0;
 394 }
 395 EXPORT_SYMBOL(dm_get_device);
 396
 397 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
 398                                 sector_t start, sector_t len, void *data)
 399 {
 400         struct queue_limits *limits = data;
 401         struct block_device *bdev = dev->bdev;
 402         struct request_queue *q = bdev_get_queue(bdev);
 403         char b[BDEVNAME_SIZE];
 404
 405         if (unlikely(!q)) {
 406                 DMWARN("%s: Cannot set limits for nonexistent device %s",
 407                        dm_device_name(ti->table->md), bdevname(bdev, b));
 408                 return 0;
 409         }
 410
 411         if (blk_stack_limits(limits, &q->limits,
 412                         get_start_sect(bdev) + start) < 0)
 413                 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
 414                        "physical_block_size=%u, logical_block_size=%u, "
 415                        "alignment_offset=%u, start=%llu",
 416                        dm_device_name(ti->table->md), bdevname(bdev, b),
 417                        q->limits.physical_block_size,
 418                        q->limits.logical_block_size,
 419                        q->limits.alignment_offset,
 420                        (unsigned long long) start << SECTOR_SHIFT);
 421         return 0;
 422 }
 423
 424 /*
 425  * Decrement a device's use count and remove it if necessary.
 426  */
 427 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
 428 {
 429         int found = 0;
 430         struct list_head *devices = &ti->table->devices;
 431         struct dm_dev_internal *dd;
 432
 433         list_for_each_entry(dd, devices, list) {
 434                 if (dd->dm_dev == d) {
 435                         found = 1;
 436                         break;
 437                 }
 438         }
 439         if (!found) {
 440                 DMWARN("%s: device %s not in table devices list",
 441                        dm_device_name(ti->table->md), d->name);
 442                 return;
 443         }
 444         if (refcount_dec_and_test(&dd->count)) {
 445                 dm_put_table_device(ti->table->md, d);
 446                 list_del(&dd->list);
 447                 kfree(dd);
 448         }
 449 }
 450 EXPORT_SYMBOL(dm_put_device);
 451
 452 /*
 453  * Checks to see if the target joins onto the end of the table.
 454  */
 455 static int adjoin(struct dm_table *table, struct dm_target *ti)
 456 {
 457         struct dm_target *prev;
 458
 459         if (!table->num_targets)
 460                 return !ti->begin;
 461
 462         prev = &table->targets[table->num_targets - 1];
 463         return (ti->begin == (prev->begin + prev->len));
 464 }
 465
 466 /*
 467  * Used to dynamically allocate the arg array.
 468  *
 469  * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
 470  * process messages even if some device is suspended. These messages have a
 471  * small fixed number of arguments.
 472  *
 473  * On the other hand, dm-switch needs to process bulk data using messages and
 474  * excessive use of GFP_NOIO could cause trouble.
 475  */
 476 static char **realloc_argv(unsigned *size, char **old_argv)
 477 {
 478         char **argv;
 479         unsigned new_size;
 480         gfp_t gfp;
 481
 482         if (*size) {
 483                 new_size = *size * 2;
 484                 gfp = GFP_KERNEL;
 485         } else {
 486                 new_size = 8;
 487                 gfp = GFP_NOIO;
 488         }
 489         argv = kmalloc_array(new_size, sizeof(*argv), gfp);
 490         if (argv && old_argv) {
 491                 memcpy(argv, old_argv, *size * sizeof(*argv));
 492                 *size = new_size;
 493         }
 494
 495         kfree(old_argv);
 496         return argv;
 497 }
 498
 499 /*
 500  * Destructively splits up the argument list to pass to ctr.
 501  */
 502 int dm_split_args(int *argc, char ***argvp, char *input)
 503 {
 504         char *start, *end = input, *out, **argv = NULL;
 505         unsigned array_size = 0;
 506
 507         *argc = 0;
 508
 509         if (!input) {
 510                 *argvp = NULL;
 511                 return 0;
 512         }
 513
 514         argv = realloc_argv(&array_size, argv);
 515         if (!argv)
 516                 return -ENOMEM;
 517
 518         while (1) {
 519                 /* Skip whitespace */
 520                 start = skip_spaces(end);
 521
 522                 if (!*start)
 523                         break;  /* success, we hit the end */
 524
 525                 /* 'out' is used to remove any back-quotes */
 526                 end = out = start;
 527                 while (*end) {
 528                         /* Everything apart from '\0' can be quoted */
 529                         if (*end == '\\' && *(end + 1)) {
 530                                 *out++ = *(end + 1);
 531                                 end += 2;
 532                                 continue;
 533                         }
 534
 535                         if (isspace(*end))
 536                                 break;  /* end of token */
 537
 538                         *out++ = *end++;
 539                 }
 540
 541                 /* have we already filled the array ? */
 542                 if ((*argc + 1) > array_size) {
 543                         argv = realloc_argv(&array_size, argv);
 544                         if (!argv)
 545                                 return -ENOMEM;
 546                 }
 547
 548                 /* we know this is whitespace */
 549                 if (*end)
 550                         end++;
 551
 552                 /* terminate the string and put it in the array */
 553                 *out = '\0';
 554                 argv[*argc] = start;
 555                 (*argc)++;
 556         }
 557
 558         *argvp = argv;
 559         return 0;
 560 }
 561
 562 /*
 563  * Impose necessary and sufficient conditions on a devices's table such
 564  * that any incoming bio which respects its logical_block_size can be
 565  * processed successfully.  If it falls across the boundary between
 566  * two or more targets, the size of each piece it gets split into must
 567  * be compatible with the logical_block_size of the target processing it.
 568  */
 569 static int validate_hardware_logical_block_alignment(struct dm_table *table,
 570                                                  struct queue_limits *limits)
 571 {
 572         /*
 573          * This function uses arithmetic modulo the logical_block_size
 574          * (in units of 512-byte sectors).
 575          */
 576         unsigned short device_logical_block_size_sects =
 577                 limits->logical_block_size >> SECTOR_SHIFT;
 578
 579         /*
 580          * Offset of the start of the next table entry, mod logical_block_size.
 581          */
 582         unsigned short next_target_start = 0;
 583
 584         /*
 585          * Given an aligned bio that extends beyond the end of a
 586          * target, how many sectors must the next target handle?
 587          */
 588         unsigned short remaining = 0;
 589
 590         struct dm_target *ti;
 591         struct queue_limits ti_limits;
 592         unsigned i;
 593
 594         /*
 595          * Check each entry in the table in turn.
 596          */
 597         for (i = 0; i < dm_table_get_num_targets(table); i++) {
 598                 ti = dm_table_get_target(table, i);
 599
 600                 blk_set_stacking_limits(&ti_limits);
 601
 602                 /* combine all target devices' limits */
 603                 if (ti->type->iterate_devices)
 604                         ti->type->iterate_devices(ti, dm_set_device_limits,
 605                                                   &ti_limits);
 606
 607                 /*
 608                  * If the remaining sectors fall entirely within this
 609                  * table entry are they compatible with its logical_block_size?
 610                  */
 611                 if (remaining < ti->len &&
 612                     remaining & ((ti_limits.logical_block_size >>
 613                                   SECTOR_SHIFT) - 1))
 614                         break;  /* Error */
 615
 616                 next_target_start =
 617                     (unsigned short) ((next_target_start + ti->len) &
 618                                       (device_logical_block_size_sects - 1));
 619                 remaining = next_target_start ?
 620                     device_logical_block_size_sects - next_target_start : 0;
 621         }
 622
 623         if (remaining) {
 624                 DMWARN("%s: table line %u (start sect %llu len %llu) "
 625                        "not aligned to h/w logical block size %u",
 626                        dm_device_name(table->md), i,
 627                        (unsigned long long) ti->begin,
 628                        (unsigned long long) ti->len,
 629                        limits->logical_block_size);
 630                 return -EINVAL;
 631         }
 632
 633         return 0;
 634 }
 635
 636 int dm_table_add_target(struct dm_table *t, const char *type,
 637                         sector_t start, sector_t len, char *params)
 638 {
 639         int r = -EINVAL, argc;
 640         char **argv;
 641         struct dm_target *tgt;
 642
 643         if (t->singleton) {
 644                 DMERR("%s: target type %s must appear alone in table",
 645                       dm_device_name(t->md), t->targets->type->name);
 646                 return -EINVAL;
 647         }
 648
 649         BUG_ON(t->num_targets >= t->num_allocated);
 650
 651         tgt = t->targets + t->num_targets;
 652         memset(tgt, 0, sizeof(*tgt));
 653
 654         if (!len) {
 655                 DMERR("%s: zero-length target", dm_device_name(t->md));
 656                 return -EINVAL;
 657         }
 658
 659         tgt->type = dm_get_target_type(type);
 660         if (!tgt->type) {
 661                 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
 662                 return -EINVAL;
 663         }
 664
 665         if (dm_target_needs_singleton(tgt->type)) {
 666                 if (t->num_targets) {
 667                         tgt->error = "singleton target type must appear alone in table";
 668                         goto bad;
 669                 }
 670                 t->singleton = true;
 671         }
 672
 673         if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
 674                 tgt->error = "target type may not be included in a read-only table";
 675                 goto bad;
 676         }
 677
 678         if (t->immutable_target_type) {
 679                 if (t->immutable_target_type != tgt->type) {
 680                         tgt->error = "immutable target type cannot be mixed with other target types";
 681                         goto bad;
 682                 }
 683         } else if (dm_target_is_immutable(tgt->type)) {
 684                 if (t->num_targets) {
 685                         tgt->error = "immutable target type cannot be mixed with other target types";
 686                         goto bad;
 687                 }
 688                 t->immutable_target_type = tgt->type;
 689         }
 690
 691         if (dm_target_has_integrity(tgt->type))
 692                 t->integrity_added = 1;
 693
 694         tgt->table = t;
 695         tgt->begin = start;
 696         tgt->len = len;
 697         tgt->error = "Unknown error";
 698
 699         /*
 700          * Does this target adjoin the previous one ?
 701          */
 702         if (!adjoin(t, tgt)) {
 703                 tgt->error = "Gap in table";
 704                 goto bad;
 705         }
 706
 707         r = dm_split_args(&argc, &argv, params);
 708         if (r) {
 709                 tgt->error = "couldn't split parameters (insufficient memory)";
 710                 goto bad;
 711         }
 712
 713         r = tgt->type->ctr(tgt, argc, argv);
 714         kfree(argv);
 715         if (r)
 716                 goto bad;
 717
 718         t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
 719
 720         if (!tgt->num_discard_bios && tgt->discards_supported)
 721                 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
 722                        dm_device_name(t->md), type);
 723
 724         return 0;
 725
 726  bad:
 727         DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
 728         dm_put_target_type(tgt->type);
 729         return r;
 730 }
 731
 732 /*
 733  * Target argument parsing helpers.
 734  */
 735 static 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
 754 int 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 }
 759 EXPORT_SYMBOL(dm_read_arg);
 760
 761 int 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 }
 766 EXPORT_SYMBOL(dm_read_arg_group);
 767
 768 const 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 }
 781 EXPORT_SYMBOL(dm_shift_arg);
 782
 783 void 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 }
 789 EXPORT_SYMBOL(dm_consume_args);
 790
 791 static 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
 797 static bool __table_type_request_based(enum dm_queue_mode table_type)
 798 {
 799         return table_type == DM_TYPE_REQUEST_BASED;
 800 }
 801
 802 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
 803 {
 804         t->type = type;
 805 }
 806 EXPORT_SYMBOL_GPL(dm_table_set_type);
 807
 808 /* validate the dax capability of the target device span */
 809 int device_not_dax_capable(struct dm_target *ti, struct dm_dev *dev,
 810                         sector_t start, sector_t len, void *data)
 811 {
 812         int blocksize = *(int *) data;
 813
 814         return !dax_supported(dev->dax_dev, dev->bdev, blocksize, start, len);
 815 }
 816
 817 /* Check devices support synchronous DAX */
 818 static int device_not_dax_synchronous_capable(struct dm_target *ti, struct dm_dev *dev,
 819                                               sector_t start, sector_t len, void *data)
 820 {
 821         return !dev->dax_dev || !dax_synchronous(dev->dax_dev);
 822 }
 823
 824 bool dm_table_supports_dax(struct dm_table *t,
 825                            iterate_devices_callout_fn iterate_fn, int *blocksize)
 826 {
 827         struct dm_target *ti;
 828         unsigned i;
 829
 830         /* Ensure that all targets support DAX. */
 831         for (i = 0; i < dm_table_get_num_targets(t); i++) {
 832                 ti = dm_table_get_target(t, i);
 833
 834                 if (!ti->type->direct_access)
 835                         return false;
 836
 837                 if (!ti->type->iterate_devices ||
 838                     ti->type->iterate_devices(ti, iterate_fn, blocksize))
 839                         return false;
 840         }
 841
 842         return true;
 843 }
 844
 845 static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev,
 846                                   sector_t start, sector_t len, void *data)
 847 {
 848         struct block_device *bdev = dev->bdev;
 849         struct request_queue *q = bdev_get_queue(bdev);
 850
 851         /* request-based cannot stack on partitions! */
 852         if (bdev_is_partition(bdev))
 853                 return false;
 854
 855         return queue_is_mq(q);
 856 }
 857
 858 static int dm_table_determine_type(struct dm_table *t)
 859 {
 860         unsigned i;
 861         unsigned bio_based = 0, request_based = 0, hybrid = 0;
 862         struct dm_target *tgt;
 863         struct list_head *devices = dm_table_get_devices(t);
 864         enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
 865         int page_size = PAGE_SIZE;
 866
 867         if (t->type != DM_TYPE_NONE) {
 868                 /* target already set the table's type */
 869                 if (t->type == DM_TYPE_BIO_BASED) {
 870                         /* possibly upgrade to a variant of bio-based */
 871                         goto verify_bio_based;
 872                 }
 873                 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
 874                 goto verify_rq_based;
 875         }
 876
 877         for (i = 0; i < t->num_targets; i++) {
 878                 tgt = t->targets + i;
 879                 if (dm_target_hybrid(tgt))
 880                         hybrid = 1;
 881                 else if (dm_target_request_based(tgt))
 882                         request_based = 1;
 883                 else
 884                         bio_based = 1;
 885
 886                 if (bio_based && request_based) {
 887                         DMERR("Inconsistent table: different target types"
 888                               " can't be mixed up");
 889                         return -EINVAL;
 890                 }
 891         }
 892
 893         if (hybrid && !bio_based && !request_based) {
 894                 /*
 895                  * The targets can work either way.
 896                  * Determine the type from the live device.
 897                  * Default to bio-based if device is new.
 898                  */
 899                 if (__table_type_request_based(live_md_type))
 900                         request_based = 1;
 901                 else
 902                         bio_based = 1;
 903         }
 904
 905         if (bio_based) {
 906 verify_bio_based:
 907                 /* We must use this table as bio-based */
 908                 t->type = DM_TYPE_BIO_BASED;
 909                 if (dm_table_supports_dax(t, device_not_dax_capable, &page_size) ||
 910                     (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
 911                         t->type = DM_TYPE_DAX_BIO_BASED;
 912                 }
 913                 return 0;
 914         }
 915
 916         BUG_ON(!request_based); /* No targets in this table */
 917
 918         t->type = DM_TYPE_REQUEST_BASED;
 919
 920 verify_rq_based:
 921         /*
 922          * Request-based dm supports only tables that have a single target now.
 923          * To support multiple targets, request splitting support is needed,
 924          * and that needs lots of changes in the block-layer.
 925          * (e.g. request completion process for partial completion.)
 926          */
 927         if (t->num_targets > 1) {
 928                 DMERR("request-based DM doesn't support multiple targets");
 929                 return -EINVAL;
 930         }
 931
 932         if (list_empty(devices)) {
 933                 int srcu_idx;
 934                 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
 935
 936                 /* inherit live table's type */
 937                 if (live_table)
 938                         t->type = live_table->type;
 939                 dm_put_live_table(t->md, srcu_idx);
 940                 return 0;
 941         }
 942
 943         tgt = dm_table_get_immutable_target(t);
 944         if (!tgt) {
 945                 DMERR("table load rejected: immutable target is required");
 946                 return -EINVAL;
 947         } else if (tgt->max_io_len) {
 948                 DMERR("table load rejected: immutable target that splits IO is not supported");
 949                 return -EINVAL;
 950         }
 951
 952         /* Non-request-stackable devices can't be used for request-based dm */
 953         if (!tgt->type->iterate_devices ||
 954             !tgt->type->iterate_devices(tgt, device_is_rq_stackable, NULL)) {
 955                 DMERR("table load rejected: including non-request-stackable devices");
 956                 return -EINVAL;
 957         }
 958
 959         return 0;
 960 }
 961
 962 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
 963 {
 964         return t->type;
 965 }
 966
 967 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
 968 {
 969         return t->immutable_target_type;
 970 }
 971
 972 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
 973 {
 974         /* Immutable target is implicitly a singleton */
 975         if (t->num_targets > 1 ||
 976             !dm_target_is_immutable(t->targets[0].type))
 977                 return NULL;
 978
 979         return t->targets;
 980 }
 981
 982 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
 983 {
 984         struct dm_target *ti;
 985         unsigned i;
 986
 987         for (i = 0; i < dm_table_get_num_targets(t); i++) {
 988                 ti = dm_table_get_target(t, i);
 989                 if (dm_target_is_wildcard(ti->type))
 990                         return ti;
 991         }
 992
 993         return NULL;
 994 }
 995
 996 bool dm_table_bio_based(struct dm_table *t)
 997 {
 998         return __table_type_bio_based(dm_table_get_type(t));
 999 }
1000
1001 bool dm_table_request_based(struct dm_table *t)
1002 {
1003         return __table_type_request_based(dm_table_get_type(t));
1004 }
1005
1006 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1007 {
1008         enum dm_queue_mode type = dm_table_get_type(t);
1009         unsigned per_io_data_size = 0;
1010         unsigned min_pool_size = 0;
1011         struct dm_target *ti;
1012         unsigned i;
1013
1014         if (unlikely(type == DM_TYPE_NONE)) {
1015                 DMWARN("no table type is set, can't allocate mempools");
1016                 return -EINVAL;
1017         }
1018
1019         if (__table_type_bio_based(type))
1020                 for (i = 0; i < t->num_targets; i++) {
1021                         ti = t->targets + i;
1022                         per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1023                         min_pool_size = max(min_pool_size, ti->num_flush_bios);
1024                 }
1025
1026         t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported,
1027                                            per_io_data_size, min_pool_size);
1028         if (!t->mempools)
1029                 return -ENOMEM;
1030
1031         return 0;
1032 }
1033
1034 void dm_table_free_md_mempools(struct dm_table *t)
1035 {
1036         dm_free_md_mempools(t->mempools);
1037         t->mempools = NULL;
1038 }
1039
1040 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1041 {
1042         return t->mempools;
1043 }
1044
1045 static int setup_indexes(struct dm_table *t)
1046 {
1047         int i;
1048         unsigned int total = 0;
1049         sector_t *indexes;
1050
1051         /* allocate the space for *all* the indexes */
1052         for (i = t->depth - 2; i >= 0; i--) {
1053                 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1054                 total += t->counts[i];
1055         }
1056
1057         indexes = kvcalloc(total, NODE_SIZE, GFP_KERNEL);
1058         if (!indexes)
1059                 return -ENOMEM;
1060
1061         /* set up internal nodes, bottom-up */
1062         for (i = t->depth - 2; i >= 0; i--) {
1063                 t->index[i] = indexes;
1064                 indexes += (KEYS_PER_NODE * t->counts[i]);
1065                 setup_btree_index(i, t);
1066         }
1067
1068         return 0;
1069 }
1070
1071 /*
1072  * Builds the btree to index the map.
1073  */
1074 static int dm_table_build_index(struct dm_table *t)
1075 {
1076         int r = 0;
1077         unsigned int leaf_nodes;
1078
1079         /* how many indexes will the btree have ? */
1080         leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1081         t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1082
1083         /* leaf layer has already been set up */
1084         t->counts[t->depth - 1] = leaf_nodes;
1085         t->index[t->depth - 1] = t->highs;
1086
1087         if (t->depth >= 2)
1088                 r = setup_indexes(t);
1089
1090         return r;
1091 }
1092
1093 static bool integrity_profile_exists(struct gendisk *disk)
1094 {
1095         return !!blk_get_integrity(disk);
1096 }
1097
1098 /*
1099  * Get a disk whose integrity profile reflects the table's profile.
1100  * Returns NULL if integrity support was inconsistent or unavailable.
1101  */
1102 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1103 {
1104         struct list_head *devices = dm_table_get_devices(t);
1105         struct dm_dev_internal *dd = NULL;
1106         struct gendisk *prev_disk = NULL, *template_disk = NULL;
1107         unsigned i;
1108
1109         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1110                 struct dm_target *ti = dm_table_get_target(t, i);
1111                 if (!dm_target_passes_integrity(ti->type))
1112                         goto no_integrity;
1113         }
1114
1115         list_for_each_entry(dd, devices, list) {
1116                 template_disk = dd->dm_dev->bdev->bd_disk;
1117                 if (!integrity_profile_exists(template_disk))
1118                         goto no_integrity;
1119                 else if (prev_disk &&
1120                          blk_integrity_compare(prev_disk, template_disk) < 0)
1121                         goto no_integrity;
1122                 prev_disk = template_disk;
1123         }
1124
1125         return template_disk;
1126
1127 no_integrity:
1128         if (prev_disk)
1129                 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1130                        dm_device_name(t->md),
1131                        prev_disk->disk_name,
1132                        template_disk->disk_name);
1133         return NULL;
1134 }
1135
1136 /*
1137  * Register the mapped device for blk_integrity support if the
1138  * underlying devices have an integrity profile.  But all devices may
1139  * not have matching profiles (checking all devices isn't reliable
1140  * during table load because this table may use other DM device(s) which
1141  * must be resumed before they will have an initialized integity
1142  * profile).  Consequently, stacked DM devices force a 2 stage integrity
1143  * profile validation: First pass during table load, final pass during
1144  * resume.
1145  */
1146 static int dm_table_register_integrity(struct dm_table *t)
1147 {
1148         struct mapped_device *md = t->md;
1149         struct gendisk *template_disk = NULL;
1150
1151         /* If target handles integrity itself do not register it here. */
1152         if (t->integrity_added)
1153                 return 0;
1154
1155         template_disk = dm_table_get_integrity_disk(t);
1156         if (!template_disk)
1157                 return 0;
1158
1159         if (!integrity_profile_exists(dm_disk(md))) {
1160                 t->integrity_supported = true;
1161                 /*
1162                  * Register integrity profile during table load; we can do
1163                  * this because the final profile must match during resume.
1164                  */
1165                 blk_integrity_register(dm_disk(md),
1166                                        blk_get_integrity(template_disk));
1167                 return 0;
1168         }
1169
1170         /*
1171          * If DM device already has an initialized integrity
1172          * profile the new profile should not conflict.
1173          */
1174         if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1175                 DMWARN("%s: conflict with existing integrity profile: "
1176                        "%s profile mismatch",
1177                        dm_device_name(t->md),
1178                        template_disk->disk_name);
1179                 return 1;
1180         }
1181
1182         /* Preserve existing integrity profile */
1183         t->integrity_supported = true;
1184         return 0;
1185 }
1186
1187 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
1188
1189 struct dm_keyslot_manager {
1190         struct blk_keyslot_manager ksm;
1191         struct mapped_device *md;
1192 };
1193
1194 struct dm_keyslot_evict_args {
1195         const struct blk_crypto_key *key;
1196         int err;
1197 };
1198
1199 static int dm_keyslot_evict_callback(struct dm_target *ti, struct dm_dev *dev,
1200                                      sector_t start, sector_t len, void *data)
1201 {
1202         struct dm_keyslot_evict_args *args = data;
1203         int err;
1204
1205         err = blk_crypto_evict_key(bdev_get_queue(dev->bdev), args->key);
1206         if (!args->err)
1207                 args->err = err;
1208         /* Always try to evict the key from all devices. */
1209         return 0;
1210 }
1211
1212 /*
1213  * When an inline encryption key is evicted from a device-mapper device, evict
1214  * it from all the underlying devices.
1215  */
1216 static int dm_keyslot_evict(struct blk_keyslot_manager *ksm,
1217                             const struct blk_crypto_key *key, unsigned int slot)
1218 {
1219         struct dm_keyslot_manager *dksm = container_of(ksm,
1220                                                        struct dm_keyslot_manager,
1221                                                        ksm);
1222         struct mapped_device *md = dksm->md;
1223         struct dm_keyslot_evict_args args = { key };
1224         struct dm_table *t;
1225         int srcu_idx;
1226         int i;
1227         struct dm_target *ti;
1228
1229         t = dm_get_live_table(md, &srcu_idx);
1230         if (!t)
1231                 return 0;
1232         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1233                 ti = dm_table_get_target(t, i);
1234                 if (!ti->type->iterate_devices)
1235                         continue;
1236                 ti->type->iterate_devices(ti, dm_keyslot_evict_callback, &args);
1237         }
1238         dm_put_live_table(md, srcu_idx);
1239         return args.err;
1240 }
1241
1242 static const struct blk_ksm_ll_ops dm_ksm_ll_ops = {
1243         .keyslot_evict = dm_keyslot_evict,
1244 };
1245
1246 static int device_intersect_crypto_modes(struct dm_target *ti,
1247                                          struct dm_dev *dev, sector_t start,
1248                                          sector_t len, void *data)
1249 {
1250         struct blk_keyslot_manager *parent = data;
1251         struct blk_keyslot_manager *child = bdev_get_queue(dev->bdev)->ksm;
1252
1253         blk_ksm_intersect_modes(parent, child);
1254         return 0;
1255 }
1256
1257 void dm_destroy_keyslot_manager(struct blk_keyslot_manager *ksm)
1258 {
1259         struct dm_keyslot_manager *dksm = container_of(ksm,
1260                                                        struct dm_keyslot_manager,
1261                                                        ksm);
1262
1263         if (!ksm)
1264                 return;
1265
1266         blk_ksm_destroy(ksm);
1267         kfree(dksm);
1268 }
1269
1270 static void dm_table_destroy_keyslot_manager(struct dm_table *t)
1271 {
1272         dm_destroy_keyslot_manager(t->ksm);
1273         t->ksm = NULL;
1274 }
1275
1276 /*
1277  * Constructs and initializes t->ksm with a keyslot manager that
1278  * represents the common set of crypto capabilities of the devices
1279  * described by the dm_table. However, if the constructed keyslot
1280  * manager does not support a superset of the crypto capabilities
1281  * supported by the current keyslot manager of the mapped_device,
1282  * it returns an error instead, since we don't support restricting
1283  * crypto capabilities on table changes. Finally, if the constructed
1284  * keyslot manager doesn't actually support any crypto modes at all,
1285  * it just returns NULL.
1286  */
1287 static int dm_table_construct_keyslot_manager(struct dm_table *t)
1288 {
1289         struct dm_keyslot_manager *dksm;
1290         struct blk_keyslot_manager *ksm;
1291         struct dm_target *ti;
1292         unsigned int i;
1293         bool ksm_is_empty = true;
1294
1295         dksm = kmalloc(sizeof(*dksm), GFP_KERNEL);
1296         if (!dksm)
1297                 return -ENOMEM;
1298         dksm->md = t->md;
1299
1300         ksm = &dksm->ksm;
1301         blk_ksm_init_passthrough(ksm);
1302         ksm->ksm_ll_ops = dm_ksm_ll_ops;
1303         ksm->max_dun_bytes_supported = UINT_MAX;
1304         memset(ksm->crypto_modes_supported, 0xFF,
1305                sizeof(ksm->crypto_modes_supported));
1306
1307         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1308                 ti = dm_table_get_target(t, i);
1309
1310                 if (!dm_target_passes_crypto(ti->type)) {
1311                         blk_ksm_intersect_modes(ksm, NULL);
1312                         break;
1313                 }
1314                 if (!ti->type->iterate_devices)
1315                         continue;
1316                 ti->type->iterate_devices(ti, device_intersect_crypto_modes,
1317                                           ksm);
1318         }
1319
1320         if (t->md->queue && !blk_ksm_is_superset(ksm, t->md->queue->ksm)) {
1321                 DMWARN("Inline encryption capabilities of new DM table were more restrictive than the old table's. This is not supported!");
1322                 dm_destroy_keyslot_manager(ksm);
1323                 return -EINVAL;
1324         }
1325
1326         /*
1327          * If the new KSM doesn't actually support any crypto modes, we may as
1328          * well represent it with a NULL ksm.
1329          */
1330         ksm_is_empty = true;
1331         for (i = 0; i < ARRAY_SIZE(ksm->crypto_modes_supported); i++) {
1332                 if (ksm->crypto_modes_supported[i]) {
1333                         ksm_is_empty = false;
1334                         break;
1335                 }
1336         }
1337
1338         if (ksm_is_empty) {
1339                 dm_destroy_keyslot_manager(ksm);
1340                 ksm = NULL;
1341         }
1342
1343         /*
1344          * t->ksm is only set temporarily while the table is being set
1345          * up, and it gets set to NULL after the capabilities have
1346          * been transferred to the request_queue.
1347          */
1348         t->ksm = ksm;
1349
1350         return 0;
1351 }
1352
1353 static void dm_update_keyslot_manager(struct request_queue *q,
1354                                       struct dm_table *t)
1355 {
1356         if (!t->ksm)
1357                 return;
1358
1359         /* Make the ksm less restrictive */
1360         if (!q->ksm) {
1361                 blk_ksm_register(t->ksm, q);
1362         } else {
1363                 blk_ksm_update_capabilities(q->ksm, t->ksm);
1364                 dm_destroy_keyslot_manager(t->ksm);
1365         }
1366         t->ksm = NULL;
1367 }
1368
1369 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1370
1371 static int dm_table_construct_keyslot_manager(struct dm_table *t)
1372 {
1373         return 0;
1374 }
1375
1376 void dm_destroy_keyslot_manager(struct blk_keyslot_manager *ksm)
1377 {
1378 }
1379
1380 static void dm_table_destroy_keyslot_manager(struct dm_table *t)
1381 {
1382 }
1383
1384 static void dm_update_keyslot_manager(struct request_queue *q,
1385                                       struct dm_table *t)
1386 {
1387 }
1388
1389 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1390
1391 /*
1392  * Prepares the table for use by building the indices,
1393  * setting the type, and allocating mempools.
1394  */
1395 int dm_table_complete(struct dm_table *t)
1396 {
1397         int r;
1398
1399         r = dm_table_determine_type(t);
1400         if (r) {
1401                 DMERR("unable to determine table type");
1402                 return r;
1403         }
1404
1405         r = dm_table_build_index(t);
1406         if (r) {
1407                 DMERR("unable to build btrees");
1408                 return r;
1409         }
1410
1411         r = dm_table_register_integrity(t);
1412         if (r) {
1413                 DMERR("could not register integrity profile.");
1414                 return r;
1415         }
1416
1417         r = dm_table_construct_keyslot_manager(t);
1418         if (r) {
1419                 DMERR("could not construct keyslot manager.");
1420                 return r;
1421         }
1422
1423         r = dm_table_alloc_md_mempools(t, t->md);
1424         if (r)
1425                 DMERR("unable to allocate mempools");
1426
1427         return r;
1428 }
1429
1430 static DEFINE_MUTEX(_event_lock);
1431 void dm_table_event_callback(struct dm_table *t,
1432                              void (*fn)(void *), void *context)
1433 {
1434         mutex_lock(&_event_lock);
1435         t->event_fn = fn;
1436         t->event_context = context;
1437         mutex_unlock(&_event_lock);
1438 }
1439
1440 void dm_table_event(struct dm_table *t)
1441 {
1442         mutex_lock(&_event_lock);
1443         if (t->event_fn)
1444                 t->event_fn(t->event_context);
1445         mutex_unlock(&_event_lock);
1446 }
1447 EXPORT_SYMBOL(dm_table_event);
1448
1449 inline sector_t dm_table_get_size(struct dm_table *t)
1450 {
1451         return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1452 }
1453 EXPORT_SYMBOL(dm_table_get_size);
1454
1455 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1456 {
1457         if (index >= t->num_targets)
1458                 return NULL;
1459
1460         return t->targets + index;
1461 }
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 /*
1490  * type->iterate_devices() should be called when the sanity check needs to
1491  * iterate and check all underlying data devices. iterate_devices() will
1492  * iterate all underlying data devices until it encounters a non-zero return
1493  * code, returned by whether the input iterate_devices_callout_fn, or
1494  * iterate_devices() itself internally.
1495  *
1496  * For some target type (e.g. dm-stripe), one call of iterate_devices() may
1497  * iterate multiple underlying devices internally, in which case a non-zero
1498  * return code returned by iterate_devices_callout_fn will stop the iteration
1499  * in advance.
1500  *
1501  * Cases requiring _any_ underlying device supporting some kind of attribute,
1502  * should use the iteration structure like dm_table_any_dev_attr(), or call
1503  * it directly. @func should handle semantics of positive examples, e.g.
1504  * capable of something.
1505  *
1506  * Cases requiring _all_ underlying devices supporting some kind of attribute,
1507  * should use the iteration structure like dm_table_supports_nowait() or
1508  * dm_table_supports_discards(). Or introduce dm_table_all_devs_attr() that
1509  * uses an @anti_func that handle semantics of counter examples, e.g. not
1510  * capable of something. So: return !dm_table_any_dev_attr(t, anti_func, data);
1511  */
1512 static bool dm_table_any_dev_attr(struct dm_table *t,
1513                                   iterate_devices_callout_fn func, void *data)
1514 {
1515         struct dm_target *ti;
1516         unsigned int i;
1517
1518         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1519                 ti = dm_table_get_target(t, i);
1520
1521                 if (ti->type->iterate_devices &&
1522                     ti->type->iterate_devices(ti, func, data))
1523                         return true;
1524         }
1525
1526         return false;
1527 }
1528
1529 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1530                         sector_t start, sector_t len, void *data)
1531 {
1532         unsigned *num_devices = data;
1533
1534         (*num_devices)++;
1535
1536         return 0;
1537 }
1538
1539 /*
1540  * Check whether a table has no data devices attached using each
1541  * target's iterate_devices method.
1542  * Returns false if the result is unknown because a target doesn't
1543  * support iterate_devices.
1544  */
1545 bool dm_table_has_no_data_devices(struct dm_table *table)
1546 {
1547         struct dm_target *ti;
1548         unsigned i, num_devices;
1549
1550         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1551                 ti = dm_table_get_target(table, i);
1552
1553                 if (!ti->type->iterate_devices)
1554                         return false;
1555
1556                 num_devices = 0;
1557                 ti->type->iterate_devices(ti, count_device, &num_devices);
1558                 if (num_devices)
1559                         return false;
1560         }
1561
1562         return true;
1563 }
1564
1565 static int device_not_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1566                                   sector_t start, sector_t len, void *data)
1567 {
1568         struct request_queue *q = bdev_get_queue(dev->bdev);
1569         enum blk_zoned_model *zoned_model = data;
1570
1571         return blk_queue_zoned_model(q) != *zoned_model;
1572 }
1573
1574 /*
1575  * Check the device zoned model based on the target feature flag. If the target
1576  * has the DM_TARGET_ZONED_HM feature flag set, host-managed zoned devices are
1577  * also accepted but all devices must have the same zoned model. If the target
1578  * has the DM_TARGET_MIXED_ZONED_MODEL feature set, the devices can have any
1579  * zoned model with all zoned devices having the same zone size.
1580  */
1581 static bool dm_table_supports_zoned_model(struct dm_table *t,
1582                                           enum blk_zoned_model zoned_model)
1583 {
1584         struct dm_target *ti;
1585         unsigned i;
1586
1587         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1588                 ti = dm_table_get_target(t, i);
1589
1590                 if (dm_target_supports_zoned_hm(ti->type)) {
1591                         if (!ti->type->iterate_devices ||
1592                             ti->type->iterate_devices(ti, device_not_zoned_model,
1593                                                       &zoned_model))
1594                                 return false;
1595                 } else if (!dm_target_supports_mixed_zoned_model(ti->type)) {
1596                         if (zoned_model == BLK_ZONED_HM)
1597                                 return false;
1598                 }
1599         }
1600
1601         return true;
1602 }
1603
1604 static int device_not_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1605                                            sector_t start, sector_t len, void *data)
1606 {
1607         struct request_queue *q = bdev_get_queue(dev->bdev);
1608         unsigned int *zone_sectors = data;
1609
1610         if (!blk_queue_is_zoned(q))
1611                 return 0;
1612
1613         return blk_queue_zone_sectors(q) != *zone_sectors;
1614 }
1615
1616 /*
1617  * Check consistency of zoned model and zone sectors across all targets. For
1618  * zone sectors, if the destination device is a zoned block device, it shall
1619  * have the specified zone_sectors.
1620  */
1621 static int validate_hardware_zoned_model(struct dm_table *table,
1622                                          enum blk_zoned_model zoned_model,
1623                                          unsigned int zone_sectors)
1624 {
1625         if (zoned_model == BLK_ZONED_NONE)
1626                 return 0;
1627
1628         if (!dm_table_supports_zoned_model(table, zoned_model)) {
1629                 DMERR("%s: zoned model is not consistent across all devices",
1630                       dm_device_name(table->md));
1631                 return -EINVAL;
1632         }
1633
1634         /* Check zone size validity and compatibility */
1635         if (!zone_sectors || !is_power_of_2(zone_sectors))
1636                 return -EINVAL;
1637
1638         if (dm_table_any_dev_attr(table, device_not_matches_zone_sectors, &zone_sectors)) {
1639                 DMERR("%s: zone sectors is not consistent across all zoned devices",
1640                       dm_device_name(table->md));
1641                 return -EINVAL;
1642         }
1643
1644         return 0;
1645 }
1646
1647 /*
1648  * Establish the new table's queue_limits and validate them.
1649  */
1650 int dm_calculate_queue_limits(struct dm_table *table,
1651                               struct queue_limits *limits)
1652 {
1653         struct dm_target *ti;
1654         struct queue_limits ti_limits;
1655         unsigned i;
1656         enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1657         unsigned int zone_sectors = 0;
1658
1659         blk_set_stacking_limits(limits);
1660
1661         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1662                 blk_set_stacking_limits(&ti_limits);
1663
1664                 ti = dm_table_get_target(table, i);
1665
1666                 if (!ti->type->iterate_devices)
1667                         goto combine_limits;
1668
1669                 /*
1670                  * Combine queue limits of all the devices this target uses.
1671                  */
1672                 ti->type->iterate_devices(ti, dm_set_device_limits,
1673                                           &ti_limits);
1674
1675                 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1676                         /*
1677                          * After stacking all limits, validate all devices
1678                          * in table support this zoned model and zone sectors.
1679                          */
1680                         zoned_model = ti_limits.zoned;
1681                         zone_sectors = ti_limits.chunk_sectors;
1682                 }
1683
1684                 /* Set I/O hints portion of queue limits */
1685                 if (ti->type->io_hints)
1686                         ti->type->io_hints(ti, &ti_limits);
1687
1688                 /*
1689                  * Check each device area is consistent with the target's
1690                  * overall queue limits.
1691                  */
1692                 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1693                                               &ti_limits))
1694                         return -EINVAL;
1695
1696 combine_limits:
1697                 /*
1698                  * Merge this target's queue limits into the overall limits
1699                  * for the table.
1700                  */
1701                 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1702                         DMWARN("%s: adding target device "
1703                                "(start sect %llu len %llu) "
1704                                "caused an alignment inconsistency",
1705                                dm_device_name(table->md),
1706                                (unsigned long long) ti->begin,
1707                                (unsigned long long) ti->len);
1708         }
1709
1710         /*
1711          * Verify that the zoned model and zone sectors, as determined before
1712          * any .io_hints override, are the same across all devices in the table.
1713          * - this is especially relevant if .io_hints is emulating a disk-managed
1714          *   zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1715          * BUT...
1716          */
1717         if (limits->zoned != BLK_ZONED_NONE) {
1718                 /*
1719                  * ...IF the above limits stacking determined a zoned model
1720                  * validate that all of the table's devices conform to it.
1721                  */
1722                 zoned_model = limits->zoned;
1723                 zone_sectors = limits->chunk_sectors;
1724         }
1725         if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1726                 return -EINVAL;
1727
1728         return validate_hardware_logical_block_alignment(table, limits);
1729 }
1730
1731 /*
1732  * Verify that all devices have an integrity profile that matches the
1733  * DM device's registered integrity profile.  If the profiles don't
1734  * match then unregister the DM device's integrity profile.
1735  */
1736 static void dm_table_verify_integrity(struct dm_table *t)
1737 {
1738         struct gendisk *template_disk = NULL;
1739
1740         if (t->integrity_added)
1741                 return;
1742
1743         if (t->integrity_supported) {
1744                 /*
1745                  * Verify that the original integrity profile
1746                  * matches all the devices in this table.
1747                  */
1748                 template_disk = dm_table_get_integrity_disk(t);
1749                 if (template_disk &&
1750                     blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1751                         return;
1752         }
1753
1754         if (integrity_profile_exists(dm_disk(t->md))) {
1755                 DMWARN("%s: unable to establish an integrity profile",
1756                        dm_device_name(t->md));
1757                 blk_integrity_unregister(dm_disk(t->md));
1758         }
1759 }
1760
1761 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1762                                 sector_t start, sector_t len, void *data)
1763 {
1764         unsigned long flush = (unsigned long) data;
1765         struct request_queue *q = bdev_get_queue(dev->bdev);
1766
1767         return (q->queue_flags & flush);
1768 }
1769
1770 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1771 {
1772         struct dm_target *ti;
1773         unsigned i;
1774
1775         /*
1776          * Require at least one underlying device to support flushes.
1777          * t->devices includes internal dm devices such as mirror logs
1778          * so we need to use iterate_devices here, which targets
1779          * supporting flushes must provide.
1780          */
1781         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1782                 ti = dm_table_get_target(t, i);
1783
1784                 if (!ti->num_flush_bios)
1785                         continue;
1786
1787                 if (ti->flush_supported)
1788                         return true;
1789
1790                 if (ti->type->iterate_devices &&
1791                     ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1792                         return true;
1793         }
1794
1795         return false;
1796 }
1797
1798 static int device_dax_write_cache_enabled(struct dm_target *ti,
1799                                           struct dm_dev *dev, sector_t start,
1800                                           sector_t len, void *data)
1801 {
1802         struct dax_device *dax_dev = dev->dax_dev;
1803
1804         if (!dax_dev)
1805                 return false;
1806
1807         if (dax_write_cache_enabled(dax_dev))
1808                 return true;
1809         return false;
1810 }
1811
1812 static int device_is_rotational(struct dm_target *ti, struct dm_dev *dev,
1813                                 sector_t start, sector_t len, void *data)
1814 {
1815         struct request_queue *q = bdev_get_queue(dev->bdev);
1816
1817         return !blk_queue_nonrot(q);
1818 }
1819
1820 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1821                              sector_t start, sector_t len, void *data)
1822 {
1823         struct request_queue *q = bdev_get_queue(dev->bdev);
1824
1825         return !blk_queue_add_random(q);
1826 }
1827
1828 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1829                                          sector_t start, sector_t len, void *data)
1830 {
1831         struct request_queue *q = bdev_get_queue(dev->bdev);
1832
1833         return !q->limits.max_write_same_sectors;
1834 }
1835
1836 static bool dm_table_supports_write_same(struct dm_table *t)
1837 {
1838         struct dm_target *ti;
1839         unsigned i;
1840
1841         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1842                 ti = dm_table_get_target(t, i);
1843
1844                 if (!ti->num_write_same_bios)
1845                         return false;
1846
1847                 if (!ti->type->iterate_devices ||
1848                     ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1849                         return false;
1850         }
1851
1852         return true;
1853 }
1854
1855 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1856                                            sector_t start, sector_t len, void *data)
1857 {
1858         struct request_queue *q = bdev_get_queue(dev->bdev);
1859
1860         return !q->limits.max_write_zeroes_sectors;
1861 }
1862
1863 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1864 {
1865         struct dm_target *ti;
1866         unsigned i = 0;
1867
1868         while (i < dm_table_get_num_targets(t)) {
1869                 ti = dm_table_get_target(t, i++);
1870
1871                 if (!ti->num_write_zeroes_bios)
1872                         return false;
1873
1874                 if (!ti->type->iterate_devices ||
1875                     ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1876                         return false;
1877         }
1878
1879         return true;
1880 }
1881
1882 static int device_not_nowait_capable(struct dm_target *ti, struct dm_dev *dev,
1883                                      sector_t start, sector_t len, void *data)
1884 {
1885         struct request_queue *q = bdev_get_queue(dev->bdev);
1886
1887         return !blk_queue_nowait(q);
1888 }
1889
1890 static bool dm_table_supports_nowait(struct dm_table *t)
1891 {
1892         struct dm_target *ti;
1893         unsigned i = 0;
1894
1895         while (i < dm_table_get_num_targets(t)) {
1896                 ti = dm_table_get_target(t, i++);
1897
1898                 if (!dm_target_supports_nowait(ti->type))
1899                         return false;
1900
1901                 if (!ti->type->iterate_devices ||
1902                     ti->type->iterate_devices(ti, device_not_nowait_capable, NULL))
1903                         return false;
1904         }
1905
1906         return true;
1907 }
1908
1909 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1910                                       sector_t start, sector_t len, void *data)
1911 {
1912         struct request_queue *q = bdev_get_queue(dev->bdev);
1913
1914         return !blk_queue_discard(q);
1915 }
1916
1917 static bool dm_table_supports_discards(struct dm_table *t)
1918 {
1919         struct dm_target *ti;
1920         unsigned i;
1921
1922         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1923                 ti = dm_table_get_target(t, i);
1924
1925                 if (!ti->num_discard_bios)
1926                         return false;
1927
1928                 /*
1929                  * Either the target provides discard support (as implied by setting
1930                  * 'discards_supported') or it relies on _all_ data devices having
1931                  * discard support.
1932                  */
1933                 if (!ti->discards_supported &&
1934                     (!ti->type->iterate_devices ||
1935                      ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1936                         return false;
1937         }
1938
1939         return true;
1940 }
1941
1942 static int device_not_secure_erase_capable(struct dm_target *ti,
1943                                            struct dm_dev *dev, sector_t start,
1944                                            sector_t len, void *data)
1945 {
1946         struct request_queue *q = bdev_get_queue(dev->bdev);
1947
1948         return !blk_queue_secure_erase(q);
1949 }
1950
1951 static bool dm_table_supports_secure_erase(struct dm_table *t)
1952 {
1953         struct dm_target *ti;
1954         unsigned int i;
1955
1956         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1957                 ti = dm_table_get_target(t, i);
1958
1959                 if (!ti->num_secure_erase_bios)
1960                         return false;
1961
1962                 if (!ti->type->iterate_devices ||
1963                     ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1964                         return false;
1965         }
1966
1967         return true;
1968 }
1969
1970 static int device_requires_stable_pages(struct dm_target *ti,
1971                                         struct dm_dev *dev, sector_t start,
1972                                         sector_t len, void *data)
1973 {
1974         struct request_queue *q = bdev_get_queue(dev->bdev);
1975
1976         return blk_queue_stable_writes(q);
1977 }
1978
1979 int dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1980                               struct queue_limits *limits)
1981 {
1982         bool wc = false, fua = false;
1983         int page_size = PAGE_SIZE;
1984         int r;
1985
1986         /*
1987          * Copy table's limits to the DM device's request_queue
1988          */
1989         q->limits = *limits;
1990
1991         if (dm_table_supports_nowait(t))
1992                 blk_queue_flag_set(QUEUE_FLAG_NOWAIT, q);
1993         else
1994                 blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, q);
1995
1996         if (!dm_table_supports_discards(t)) {
1997                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
1998                 /* Must also clear discard limits... */
1999                 q->limits.max_discard_sectors = 0;
2000                 q->limits.max_hw_discard_sectors = 0;
2001                 q->limits.discard_granularity = 0;
2002                 q->limits.discard_alignment = 0;
2003                 q->limits.discard_misaligned = 0;
2004         } else
2005                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
2006
2007         if (dm_table_supports_secure_erase(t))
2008                 blk_queue_flag_set(QUEUE_FLAG_SECERASE, q);
2009
2010         if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
2011                 wc = true;
2012                 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
2013                         fua = true;
2014         }
2015         blk_queue_write_cache(q, wc, fua);
2016
2017         if (dm_table_supports_dax(t, device_not_dax_capable, &page_size)) {
2018                 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
2019                 if (dm_table_supports_dax(t, device_not_dax_synchronous_capable, NULL))
2020                         set_dax_synchronous(t->md->dax_dev);
2021         }
2022         else
2023                 blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
2024
2025         if (dm_table_any_dev_attr(t, device_dax_write_cache_enabled, NULL))
2026                 dax_write_cache(t->md->dax_dev, true);
2027
2028         /* Ensure that all underlying devices are non-rotational. */
2029         if (dm_table_any_dev_attr(t, device_is_rotational, NULL))
2030                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
2031         else
2032                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2033
2034         if (!dm_table_supports_write_same(t))
2035                 q->limits.max_write_same_sectors = 0;
2036         if (!dm_table_supports_write_zeroes(t))
2037                 q->limits.max_write_zeroes_sectors = 0;
2038
2039         dm_table_verify_integrity(t);
2040
2041         /*
2042          * Some devices don't use blk_integrity but still want stable pages
2043          * because they do their own checksumming.
2044          * If any underlying device requires stable pages, a table must require
2045          * them as well.  Only targets that support iterate_devices are considered:
2046          * don't want error, zero, etc to require stable pages.
2047          */
2048         if (dm_table_any_dev_attr(t, device_requires_stable_pages, NULL))
2049                 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, q);
2050         else
2051                 blk_queue_flag_clear(QUEUE_FLAG_STABLE_WRITES, q);
2052
2053         /*
2054          * Determine whether or not this queue's I/O timings contribute
2055          * to the entropy pool, Only request-based targets use this.
2056          * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
2057          * have it set.
2058          */
2059         if (blk_queue_add_random(q) &&
2060             dm_table_any_dev_attr(t, device_is_not_random, NULL))
2061                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2062
2063         /*
2064          * For a zoned target, setup the zones related queue attributes
2065          * and resources necessary for zone append emulation if necessary.
2066          */
2067         if (blk_queue_is_zoned(q)) {
2068                 r = dm_set_zones_restrictions(t, q);
2069                 if (r)
2070                         return r;
2071         }
2072
2073         dm_update_keyslot_manager(q, t);
2074         disk_update_readahead(t->md->disk);
2075
2076         return 0;
2077 }
2078
2079 unsigned int dm_table_get_num_targets(struct dm_table *t)
2080 {
2081         return t->num_targets;
2082 }
2083
2084 struct list_head *dm_table_get_devices(struct dm_table *t)
2085 {
2086         return &t->devices;
2087 }
2088
2089 fmode_t dm_table_get_mode(struct dm_table *t)
2090 {
2091         return t->mode;
2092 }
2093 EXPORT_SYMBOL(dm_table_get_mode);
2094
2095 enum suspend_mode {
2096         PRESUSPEND,
2097         PRESUSPEND_UNDO,
2098         POSTSUSPEND,
2099 };
2100
2101 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
2102 {
2103         int i = t->num_targets;
2104         struct dm_target *ti = t->targets;
2105
2106         lockdep_assert_held(&t->md->suspend_lock);
2107
2108         while (i--) {
2109                 switch (mode) {
2110                 case PRESUSPEND:
2111                         if (ti->type->presuspend)
2112                                 ti->type->presuspend(ti);
2113                         break;
2114                 case PRESUSPEND_UNDO:
2115                         if (ti->type->presuspend_undo)
2116                                 ti->type->presuspend_undo(ti);
2117                         break;
2118                 case POSTSUSPEND:
2119                         if (ti->type->postsuspend)
2120                                 ti->type->postsuspend(ti);
2121                         break;
2122                 }
2123                 ti++;
2124         }
2125 }
2126
2127 void dm_table_presuspend_targets(struct dm_table *t)
2128 {
2129         if (!t)
2130                 return;
2131
2132         suspend_targets(t, PRESUSPEND);
2133 }
2134
2135 void dm_table_presuspend_undo_targets(struct dm_table *t)
2136 {
2137         if (!t)
2138                 return;
2139
2140         suspend_targets(t, PRESUSPEND_UNDO);
2141 }
2142
2143 void dm_table_postsuspend_targets(struct dm_table *t)
2144 {
2145         if (!t)
2146                 return;
2147
2148         suspend_targets(t, POSTSUSPEND);
2149 }
2150
2151 int dm_table_resume_targets(struct dm_table *t)
2152 {
2153         int i, r = 0;
2154
2155         lockdep_assert_held(&t->md->suspend_lock);
2156
2157         for (i = 0; i < t->num_targets; i++) {
2158                 struct dm_target *ti = t->targets + i;
2159
2160                 if (!ti->type->preresume)
2161                         continue;
2162
2163                 r = ti->type->preresume(ti);
2164                 if (r) {
2165                         DMERR("%s: %s: preresume failed, error = %d",
2166                               dm_device_name(t->md), ti->type->name, r);
2167                         return r;
2168                 }
2169         }
2170
2171         for (i = 0; i < t->num_targets; i++) {
2172                 struct dm_target *ti = t->targets + i;
2173
2174                 if (ti->type->resume)
2175                         ti->type->resume(ti);
2176         }
2177
2178         return 0;
2179 }
2180
2181 struct mapped_device *dm_table_get_md(struct dm_table *t)
2182 {
2183         return t->md;
2184 }
2185 EXPORT_SYMBOL(dm_table_get_md);
2186
2187 const char *dm_table_device_name(struct dm_table *t)
2188 {
2189         return dm_device_name(t->md);
2190 }
2191 EXPORT_SYMBOL_GPL(dm_table_device_name);
2192
2193 void dm_table_run_md_queue_async(struct dm_table *t)
2194 {
2195         if (!dm_table_request_based(t))
2196                 return;
2197
2198         if (t->md->queue)
2199                 blk_mq_run_hw_queues(t->md->queue, true);
2200 }
2201 EXPORT_SYMBOL(dm_table_run_md_queue_async);
2202