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