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 MAX_DEPTH 16
  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 struct dm_table {
  33         struct mapped_device *md;
  34         enum dm_queue_mode type;
  35
  36         /* btree table */
  37         unsigned int depth;
  38         unsigned int counts[MAX_DEPTH]; /* in nodes */
  39         sector_t *index[MAX_DEPTH];
  40
  41         unsigned int num_targets;
  42         unsigned int num_allocated;
  43         sector_t *highs;
  44         struct dm_target *targets;
  45
  46         struct target_type *immutable_target_type;
  47
  48         bool integrity_supported:1;
  49         bool singleton:1;
  50         bool all_blk_mq:1;
  51         unsigned integrity_added:1;
  52
  53         /*
  54          * Indicates the rw permissions for the new logical
  55          * device.  This should be a combination of FMODE_READ
  56          * and FMODE_WRITE.
  57          */
  58         fmode_t mode;
  59
  60         /* a list of devices used by this table */
  61         struct list_head devices;
  62
  63         /* events get handed up using this callback */
  64         void (*event_fn)(void *);
  65         void *event_context;
  66
  67         struct dm_md_mempools *mempools;
  68
  69         struct list_head target_callbacks;
  70 };
  71
  72 /*
  73  * Similar to ceiling(log_size(n))
  74  */
  75 static unsigned int int_log(unsigned int n, unsigned int base)
  76 {
  77         int result = 0;
  78
  79         while (n > 1) {
  80                 n = dm_div_up(n, base);
  81                 result++;
  82         }
  83
  84         return result;
  85 }
  86
  87 /*
  88  * Calculate the index of the child node of the n'th node k'th key.
  89  */
  90 static inline unsigned int get_child(unsigned int n, unsigned int k)
  91 {
  92         return (n * CHILDREN_PER_NODE) + k;
  93 }
  94
  95 /*
  96  * Return the n'th node of level l from table t.
  97  */
  98 static inline sector_t *get_node(struct dm_table *t,
  99                                  unsigned int l, unsigned int n)
 100 {
 101         return t->index[l] + (n * KEYS_PER_NODE);
 102 }
 103
 104 /*
 105  * Return the highest key that you could lookup from the n'th
 106  * node on level l of the btree.
 107  */
 108 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
 109 {
 110         for (; l < t->depth - 1; l++)
 111                 n = get_child(n, CHILDREN_PER_NODE - 1);
 112
 113         if (n >= t->counts[l])
 114                 return (sector_t) - 1;
 115
 116         return get_node(t, l, n)[KEYS_PER_NODE - 1];
 117 }
 118
 119 /*
 120  * Fills in a level of the btree based on the highs of the level
 121  * below it.
 122  */
 123 static int setup_btree_index(unsigned int l, struct dm_table *t)
 124 {
 125         unsigned int n, k;
 126         sector_t *node;
 127
 128         for (n = 0U; n < t->counts[l]; n++) {
 129                 node = get_node(t, l, n);
 130
 131                 for (k = 0U; k < KEYS_PER_NODE; k++)
 132                         node[k] = high(t, l + 1, get_child(n, k));
 133         }
 134
 135         return 0;
 136 }
 137
 138 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
 139 {
 140         unsigned long size;
 141         void *addr;
 142
 143         /*
 144          * Check that we're not going to overflow.
 145          */
 146         if (nmemb > (ULONG_MAX / elem_size))
 147                 return NULL;
 148
 149         size = nmemb * elem_size;
 150         addr = vzalloc(size);
 151
 152         return addr;
 153 }
 154 EXPORT_SYMBOL(dm_vcalloc);
 155
 156 /*
 157  * highs, and targets are managed as dynamic arrays during a
 158  * table load.
 159  */
 160 static int alloc_targets(struct dm_table *t, unsigned int num)
 161 {
 162         sector_t *n_highs;
 163         struct dm_target *n_targets;
 164
 165         /*
 166          * Allocate both the target array and offset array at once.
 167          * Append an empty entry to catch sectors beyond the end of
 168          * the device.
 169          */
 170         n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
 171                                           sizeof(sector_t));
 172         if (!n_highs)
 173                 return -ENOMEM;
 174
 175         n_targets = (struct dm_target *) (n_highs + num);
 176
 177         memset(n_highs, -1, sizeof(*n_highs) * num);
 178         vfree(t->highs);
 179
 180         t->num_allocated = num;
 181         t->highs = n_highs;
 182         t->targets = n_targets;
 183
 184         return 0;
 185 }
 186
 187 int dm_table_create(struct dm_table **result, fmode_t mode,
 188                     unsigned num_targets, struct mapped_device *md)
 189 {
 190         struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
 191
 192         if (!t)
 193                 return -ENOMEM;
 194
 195         INIT_LIST_HEAD(&t->devices);
 196         INIT_LIST_HEAD(&t->target_callbacks);
 197
 198         if (!num_targets)
 199                 num_targets = KEYS_PER_NODE;
 200
 201         num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
 202
 203         if (!num_targets) {
 204                 kfree(t);
 205                 return -ENOMEM;
 206         }
 207
 208         if (alloc_targets(t, num_targets)) {
 209                 kfree(t);
 210                 return -ENOMEM;
 211         }
 212
 213         t->type = DM_TYPE_NONE;
 214         t->mode = mode;
 215         t->md = md;
 216         *result = t;
 217         return 0;
 218 }
 219
 220 static void free_devices(struct list_head *devices, struct mapped_device *md)
 221 {
 222         struct list_head *tmp, *next;
 223
 224         list_for_each_safe(tmp, next, devices) {
 225                 struct dm_dev_internal *dd =
 226                     list_entry(tmp, struct dm_dev_internal, list);
 227                 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
 228                        dm_device_name(md), dd->dm_dev->name);
 229                 dm_put_table_device(md, dd->dm_dev);
 230                 kfree(dd);
 231         }
 232 }
 233
 234 void dm_table_destroy(struct dm_table *t)
 235 {
 236         unsigned int i;
 237
 238         if (!t)
 239                 return;
 240
 241         /* free the indexes */
 242         if (t->depth >= 2)
 243                 vfree(t->index[t->depth - 2]);
 244
 245         /* free the targets */
 246         for (i = 0; i < t->num_targets; i++) {
 247                 struct dm_target *tgt = t->targets + i;
 248
 249                 if (tgt->type->dtr)
 250                         tgt->type->dtr(tgt);
 251
 252                 dm_put_target_type(tgt->type);
 253         }
 254
 255         vfree(t->highs);
 256
 257         /* free the device list */
 258         free_devices(&t->devices, t->md);
 259
 260         dm_free_md_mempools(t->mempools);
 261
 262         kfree(t);
 263 }
 264
 265 /*
 266  * See if we've already got a device in the list.
 267  */
 268 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
 269 {
 270         struct dm_dev_internal *dd;
 271
 272         list_for_each_entry (dd, l, list)
 273                 if (dd->dm_dev->bdev->bd_dev == dev)
 274                         return dd;
 275
 276         return NULL;
 277 }
 278
 279 /*
 280  * If possible, this checks an area of a destination device is invalid.
 281  */
 282 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
 283                                   sector_t start, sector_t len, void *data)
 284 {
 285         struct request_queue *q;
 286         struct queue_limits *limits = data;
 287         struct block_device *bdev = dev->bdev;
 288         sector_t dev_size =
 289                 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
 290         unsigned short logical_block_size_sectors =
 291                 limits->logical_block_size >> SECTOR_SHIFT;
 292         char b[BDEVNAME_SIZE];
 293
 294         /*
 295          * Some devices exist without request functions,
 296          * such as loop devices not yet bound to backing files.
 297          * Forbid the use of such devices.
 298          */
 299         q = bdev_get_queue(bdev);
 300         if (!q || !q->make_request_fn) {
 301                 DMWARN("%s: %s is not yet initialised: "
 302                        "start=%llu, len=%llu, dev_size=%llu",
 303                        dm_device_name(ti->table->md), bdevname(bdev, b),
 304                        (unsigned long long)start,
 305                        (unsigned long long)len,
 306                        (unsigned long long)dev_size);
 307                 return 1;
 308         }
 309
 310         if (!dev_size)
 311                 return 0;
 312
 313         if ((start >= dev_size) || (start + len > dev_size)) {
 314                 DMWARN("%s: %s too small for target: "
 315                        "start=%llu, len=%llu, dev_size=%llu",
 316                        dm_device_name(ti->table->md), bdevname(bdev, b),
 317                        (unsigned long long)start,
 318                        (unsigned long long)len,
 319                        (unsigned long long)dev_size);
 320                 return 1;
 321         }
 322
 323         /*
 324          * If the target is mapped to zoned block device(s), check
 325          * that the zones are not partially mapped.
 326          */
 327         if (bdev_zoned_model(bdev) != BLK_ZONED_NONE) {
 328                 unsigned int zone_sectors = bdev_zone_sectors(bdev);
 329
 330                 if (start & (zone_sectors - 1)) {
 331                         DMWARN("%s: start=%llu not aligned to h/w zone size %u of %s",
 332                                dm_device_name(ti->table->md),
 333                                (unsigned long long)start,
 334                                zone_sectors, bdevname(bdev, b));
 335                         return 1;
 336                 }
 337
 338                 /*
 339                  * Note: The last zone of a zoned block device may be smaller
 340                  * than other zones. So for a target mapping the end of a
 341                  * zoned block device with such a zone, len would not be zone
 342                  * aligned. We do not allow such last smaller zone to be part
 343                  * of the mapping here to ensure that mappings with multiple
 344                  * devices do not end up with a smaller zone in the middle of
 345                  * the sector range.
 346                  */
 347                 if (len & (zone_sectors - 1)) {
 348                         DMWARN("%s: len=%llu not aligned to h/w zone size %u of %s",
 349                                dm_device_name(ti->table->md),
 350                                (unsigned long long)len,
 351                                zone_sectors, bdevname(bdev, b));
 352                         return 1;
 353                 }
 354         }
 355
 356         if (logical_block_size_sectors <= 1)
 357                 return 0;
 358
 359         if (start & (logical_block_size_sectors - 1)) {
 360                 DMWARN("%s: start=%llu not aligned to h/w "
 361                        "logical block size %u of %s",
 362                        dm_device_name(ti->table->md),
 363                        (unsigned long long)start,
 364                        limits->logical_block_size, bdevname(bdev, b));
 365                 return 1;
 366         }
 367
 368         if (len & (logical_block_size_sectors - 1)) {
 369                 DMWARN("%s: len=%llu not aligned to h/w "
 370                        "logical block size %u of %s",
 371                        dm_device_name(ti->table->md),
 372                        (unsigned long long)len,
 373                        limits->logical_block_size, bdevname(bdev, b));
 374                 return 1;
 375         }
 376
 377         return 0;
 378 }
 379
 380 /*
 381  * This upgrades the mode on an already open dm_dev, being
 382  * careful to leave things as they were if we fail to reopen the
 383  * device and not to touch the existing bdev field in case
 384  * it is accessed concurrently inside dm_table_any_congested().
 385  */
 386 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
 387                         struct mapped_device *md)
 388 {
 389         int r;
 390         struct dm_dev *old_dev, *new_dev;
 391
 392         old_dev = dd->dm_dev;
 393
 394         r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
 395                                 dd->dm_dev->mode | new_mode, &new_dev);
 396         if (r)
 397                 return r;
 398
 399         dd->dm_dev = new_dev;
 400         dm_put_table_device(md, old_dev);
 401
 402         return 0;
 403 }
 404
 405 /*
 406  * Convert the path to a device
 407  */
 408 dev_t dm_get_dev_t(const char *path)
 409 {
 410         dev_t dev;
 411         struct block_device *bdev;
 412
 413         bdev = lookup_bdev(path);
 414         if (IS_ERR(bdev))
 415                 dev = name_to_dev_t(path);
 416         else {
 417                 dev = bdev->bd_dev;
 418                 bdput(bdev);
 419         }
 420
 421         return dev;
 422 }
 423 EXPORT_SYMBOL_GPL(dm_get_dev_t);
 424
 425 /*
 426  * Add a device to the list, or just increment the usage count if
 427  * it's already present.
 428  */
 429 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
 430                   struct dm_dev **result)
 431 {
 432         int r;
 433         dev_t dev;
 434         unsigned int major, minor;
 435         char dummy;
 436         struct dm_dev_internal *dd;
 437         struct dm_table *t = ti->table;
 438
 439         BUG_ON(!t);
 440
 441         if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
 442                 /* Extract the major/minor numbers */
 443                 dev = MKDEV(major, minor);
 444                 if (MAJOR(dev) != major || MINOR(dev) != minor)
 445                         return -EOVERFLOW;
 446         } else {
 447                 dev = dm_get_dev_t(path);
 448                 if (!dev)
 449                         return -ENODEV;
 450         }
 451
 452         dd = find_device(&t->devices, dev);
 453         if (!dd) {
 454                 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
 455                 if (!dd)
 456                         return -ENOMEM;
 457
 458                 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
 459                         kfree(dd);
 460                         return r;
 461                 }
 462
 463                 refcount_set(&dd->count, 1);
 464                 list_add(&dd->list, &t->devices);
 465                 goto out;
 466
 467         } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
 468                 r = upgrade_mode(dd, mode, t->md);
 469                 if (r)
 470                         return r;
 471         }
 472         refcount_inc(&dd->count);
 473 out:
 474         *result = dd->dm_dev;
 475         return 0;
 476 }
 477 EXPORT_SYMBOL(dm_get_device);
 478
 479 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
 480                                 sector_t start, sector_t len, void *data)
 481 {
 482         struct queue_limits *limits = data;
 483         struct block_device *bdev = dev->bdev;
 484         struct request_queue *q = bdev_get_queue(bdev);
 485         char b[BDEVNAME_SIZE];
 486
 487         if (unlikely(!q)) {
 488                 DMWARN("%s: Cannot set limits for nonexistent device %s",
 489                        dm_device_name(ti->table->md), bdevname(bdev, b));
 490                 return 0;
 491         }
 492
 493         if (bdev_stack_limits(limits, bdev, start) < 0)
 494                 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
 495                        "physical_block_size=%u, logical_block_size=%u, "
 496                        "alignment_offset=%u, start=%llu",
 497                        dm_device_name(ti->table->md), bdevname(bdev, b),
 498                        q->limits.physical_block_size,
 499                        q->limits.logical_block_size,
 500                        q->limits.alignment_offset,
 501                        (unsigned long long) start << SECTOR_SHIFT);
 502
 503         limits->zoned = blk_queue_zoned_model(q);
 504
 505         return 0;
 506 }
 507
 508 /*
 509  * Decrement a device's use count and remove it if necessary.
 510  */
 511 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
 512 {
 513         int found = 0;
 514         struct list_head *devices = &ti->table->devices;
 515         struct dm_dev_internal *dd;
 516
 517         list_for_each_entry(dd, devices, list) {
 518                 if (dd->dm_dev == d) {
 519                         found = 1;
 520                         break;
 521                 }
 522         }
 523         if (!found) {
 524                 DMWARN("%s: device %s not in table devices list",
 525                        dm_device_name(ti->table->md), d->name);
 526                 return;
 527         }
 528         if (refcount_dec_and_test(&dd->count)) {
 529                 dm_put_table_device(ti->table->md, d);
 530                 list_del(&dd->list);
 531                 kfree(dd);
 532         }
 533 }
 534 EXPORT_SYMBOL(dm_put_device);
 535
 536 /*
 537  * Checks to see if the target joins onto the end of the table.
 538  */
 539 static int adjoin(struct dm_table *table, struct dm_target *ti)
 540 {
 541         struct dm_target *prev;
 542
 543         if (!table->num_targets)
 544                 return !ti->begin;
 545
 546         prev = &table->targets[table->num_targets - 1];
 547         return (ti->begin == (prev->begin + prev->len));
 548 }
 549
 550 /*
 551  * Used to dynamically allocate the arg array.
 552  *
 553  * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
 554  * process messages even if some device is suspended. These messages have a
 555  * small fixed number of arguments.
 556  *
 557  * On the other hand, dm-switch needs to process bulk data using messages and
 558  * excessive use of GFP_NOIO could cause trouble.
 559  */
 560 static char **realloc_argv(unsigned *size, char **old_argv)
 561 {
 562         char **argv;
 563         unsigned new_size;
 564         gfp_t gfp;
 565
 566         if (*size) {
 567                 new_size = *size * 2;
 568                 gfp = GFP_KERNEL;
 569         } else {
 570                 new_size = 8;
 571                 gfp = GFP_NOIO;
 572         }
 573         argv = kmalloc_array(new_size, sizeof(*argv), gfp);
 574         if (argv && old_argv) {
 575                 memcpy(argv, old_argv, *size * sizeof(*argv));
 576                 *size = new_size;
 577         }
 578
 579         kfree(old_argv);
 580         return argv;
 581 }
 582
 583 /*
 584  * Destructively splits up the argument list to pass to ctr.
 585  */
 586 int dm_split_args(int *argc, char ***argvp, char *input)
 587 {
 588         char *start, *end = input, *out, **argv = NULL;
 589         unsigned array_size = 0;
 590
 591         *argc = 0;
 592
 593         if (!input) {
 594                 *argvp = NULL;
 595                 return 0;
 596         }
 597
 598         argv = realloc_argv(&array_size, argv);
 599         if (!argv)
 600                 return -ENOMEM;
 601
 602         while (1) {
 603                 /* Skip whitespace */
 604                 start = skip_spaces(end);
 605
 606                 if (!*start)
 607                         break;  /* success, we hit the end */
 608
 609                 /* 'out' is used to remove any back-quotes */
 610                 end = out = start;
 611                 while (*end) {
 612                         /* Everything apart from '\0' can be quoted */
 613                         if (*end == '\\' && *(end + 1)) {
 614                                 *out++ = *(end + 1);
 615                                 end += 2;
 616                                 continue;
 617                         }
 618
 619                         if (isspace(*end))
 620                                 break;  /* end of token */
 621
 622                         *out++ = *end++;
 623                 }
 624
 625                 /* have we already filled the array ? */
 626                 if ((*argc + 1) > array_size) {
 627                         argv = realloc_argv(&array_size, argv);
 628                         if (!argv)
 629                                 return -ENOMEM;
 630                 }
 631
 632                 /* we know this is whitespace */
 633                 if (*end)
 634                         end++;
 635
 636                 /* terminate the string and put it in the array */
 637                 *out = '\0';
 638                 argv[*argc] = start;
 639                 (*argc)++;
 640         }
 641
 642         *argvp = argv;
 643         return 0;
 644 }
 645
 646 /*
 647  * Impose necessary and sufficient conditions on a devices's table such
 648  * that any incoming bio which respects its logical_block_size can be
 649  * processed successfully.  If it falls across the boundary between
 650  * two or more targets, the size of each piece it gets split into must
 651  * be compatible with the logical_block_size of the target processing it.
 652  */
 653 static int validate_hardware_logical_block_alignment(struct dm_table *table,
 654                                                  struct queue_limits *limits)
 655 {
 656         /*
 657          * This function uses arithmetic modulo the logical_block_size
 658          * (in units of 512-byte sectors).
 659          */
 660         unsigned short device_logical_block_size_sects =
 661                 limits->logical_block_size >> SECTOR_SHIFT;
 662
 663         /*
 664          * Offset of the start of the next table entry, mod logical_block_size.
 665          */
 666         unsigned short next_target_start = 0;
 667
 668         /*
 669          * Given an aligned bio that extends beyond the end of a
 670          * target, how many sectors must the next target handle?
 671          */
 672         unsigned short remaining = 0;
 673
 674         struct dm_target *uninitialized_var(ti);
 675         struct queue_limits ti_limits;
 676         unsigned i;
 677
 678         /*
 679          * Check each entry in the table in turn.
 680          */
 681         for (i = 0; i < dm_table_get_num_targets(table); i++) {
 682                 ti = dm_table_get_target(table, i);
 683
 684                 blk_set_stacking_limits(&ti_limits);
 685
 686                 /* combine all target devices' limits */
 687                 if (ti->type->iterate_devices)
 688                         ti->type->iterate_devices(ti, dm_set_device_limits,
 689                                                   &ti_limits);
 690
 691                 /*
 692                  * If the remaining sectors fall entirely within this
 693                  * table entry are they compatible with its logical_block_size?
 694                  */
 695                 if (remaining < ti->len &&
 696                     remaining & ((ti_limits.logical_block_size >>
 697                                   SECTOR_SHIFT) - 1))
 698                         break;  /* Error */
 699
 700                 next_target_start =
 701                     (unsigned short) ((next_target_start + ti->len) &
 702                                       (device_logical_block_size_sects - 1));
 703                 remaining = next_target_start ?
 704                     device_logical_block_size_sects - next_target_start : 0;
 705         }
 706
 707         if (remaining) {
 708                 DMWARN("%s: table line %u (start sect %llu len %llu) "
 709                        "not aligned to h/w logical block size %u",
 710                        dm_device_name(table->md), i,
 711                        (unsigned long long) ti->begin,
 712                        (unsigned long long) ti->len,
 713                        limits->logical_block_size);
 714                 return -EINVAL;
 715         }
 716
 717         return 0;
 718 }
 719
 720 int dm_table_add_target(struct dm_table *t, const char *type,
 721                         sector_t start, sector_t len, char *params)
 722 {
 723         int r = -EINVAL, argc;
 724         char **argv;
 725         struct dm_target *tgt;
 726
 727         if (t->singleton) {
 728                 DMERR("%s: target type %s must appear alone in table",
 729                       dm_device_name(t->md), t->targets->type->name);
 730                 return -EINVAL;
 731         }
 732
 733         BUG_ON(t->num_targets >= t->num_allocated);
 734
 735         tgt = t->targets + t->num_targets;
 736         memset(tgt, 0, sizeof(*tgt));
 737
 738         if (!len) {
 739                 DMERR("%s: zero-length target", dm_device_name(t->md));
 740                 return -EINVAL;
 741         }
 742
 743         tgt->type = dm_get_target_type(type);
 744         if (!tgt->type) {
 745                 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
 746                 return -EINVAL;
 747         }
 748
 749         if (dm_target_needs_singleton(tgt->type)) {
 750                 if (t->num_targets) {
 751                         tgt->error = "singleton target type must appear alone in table";
 752                         goto bad;
 753                 }
 754                 t->singleton = true;
 755         }
 756
 757         if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
 758                 tgt->error = "target type may not be included in a read-only table";
 759                 goto bad;
 760         }
 761
 762         if (t->immutable_target_type) {
 763                 if (t->immutable_target_type != tgt->type) {
 764                         tgt->error = "immutable target type cannot be mixed with other target types";
 765                         goto bad;
 766                 }
 767         } else if (dm_target_is_immutable(tgt->type)) {
 768                 if (t->num_targets) {
 769                         tgt->error = "immutable target type cannot be mixed with other target types";
 770                         goto bad;
 771                 }
 772                 t->immutable_target_type = tgt->type;
 773         }
 774
 775         if (dm_target_has_integrity(tgt->type))
 776                 t->integrity_added = 1;
 777
 778         tgt->table = t;
 779         tgt->begin = start;
 780         tgt->len = len;
 781         tgt->error = "Unknown error";
 782
 783         /*
 784          * Does this target adjoin the previous one ?
 785          */
 786         if (!adjoin(t, tgt)) {
 787                 tgt->error = "Gap in table";
 788                 goto bad;
 789         }
 790
 791         r = dm_split_args(&argc, &argv, params);
 792         if (r) {
 793                 tgt->error = "couldn't split parameters (insufficient memory)";
 794                 goto bad;
 795         }
 796
 797         r = tgt->type->ctr(tgt, argc, argv);
 798         kfree(argv);
 799         if (r)
 800                 goto bad;
 801
 802         t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
 803
 804         if (!tgt->num_discard_bios && tgt->discards_supported)
 805                 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
 806                        dm_device_name(t->md), type);
 807
 808         return 0;
 809
 810  bad:
 811         DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
 812         dm_put_target_type(tgt->type);
 813         return r;
 814 }
 815
 816 /*
 817  * Target argument parsing helpers.
 818  */
 819 static int validate_next_arg(const struct dm_arg *arg,
 820                              struct dm_arg_set *arg_set,
 821                              unsigned *value, char **error, unsigned grouped)
 822 {
 823         const char *arg_str = dm_shift_arg(arg_set);
 824         char dummy;
 825
 826         if (!arg_str ||
 827             (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
 828             (*value < arg->min) ||
 829             (*value > arg->max) ||
 830             (grouped && arg_set->argc < *value)) {
 831                 *error = arg->error;
 832                 return -EINVAL;
 833         }
 834
 835         return 0;
 836 }
 837
 838 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 839                 unsigned *value, char **error)
 840 {
 841         return validate_next_arg(arg, arg_set, value, error, 0);
 842 }
 843 EXPORT_SYMBOL(dm_read_arg);
 844
 845 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 846                       unsigned *value, char **error)
 847 {
 848         return validate_next_arg(arg, arg_set, value, error, 1);
 849 }
 850 EXPORT_SYMBOL(dm_read_arg_group);
 851
 852 const char *dm_shift_arg(struct dm_arg_set *as)
 853 {
 854         char *r;
 855
 856         if (as->argc) {
 857                 as->argc--;
 858                 r = *as->argv;
 859                 as->argv++;
 860                 return r;
 861         }
 862
 863         return NULL;
 864 }
 865 EXPORT_SYMBOL(dm_shift_arg);
 866
 867 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
 868 {
 869         BUG_ON(as->argc < num_args);
 870         as->argc -= num_args;
 871         as->argv += num_args;
 872 }
 873 EXPORT_SYMBOL(dm_consume_args);
 874
 875 static bool __table_type_bio_based(enum dm_queue_mode table_type)
 876 {
 877         return (table_type == DM_TYPE_BIO_BASED ||
 878                 table_type == DM_TYPE_DAX_BIO_BASED ||
 879                 table_type == DM_TYPE_NVME_BIO_BASED);
 880 }
 881
 882 static bool __table_type_request_based(enum dm_queue_mode table_type)
 883 {
 884         return (table_type == DM_TYPE_REQUEST_BASED ||
 885                 table_type == DM_TYPE_MQ_REQUEST_BASED);
 886 }
 887
 888 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
 889 {
 890         t->type = type;
 891 }
 892 EXPORT_SYMBOL_GPL(dm_table_set_type);
 893
 894 static int device_not_dax_capable(struct dm_target *ti, struct dm_dev *dev,
 895                                sector_t start, sector_t len, void *data)
 896 {
 897         return !bdev_dax_supported(dev->bdev, PAGE_SIZE);
 898 }
 899
 900 static bool dm_table_supports_dax(struct dm_table *t)
 901 {
 902         struct dm_target *ti;
 903         unsigned i;
 904
 905         /* Ensure that all targets support DAX. */
 906         for (i = 0; i < dm_table_get_num_targets(t); i++) {
 907                 ti = dm_table_get_target(t, i);
 908
 909                 if (!ti->type->direct_access)
 910                         return false;
 911
 912                 if (!ti->type->iterate_devices ||
 913                     ti->type->iterate_devices(ti, device_not_dax_capable, NULL))
 914                         return false;
 915         }
 916
 917         return true;
 918 }
 919
 920 static bool dm_table_does_not_support_partial_completion(struct dm_table *t);
 921
 922 struct verify_rq_based_data {
 923         unsigned sq_count;
 924         unsigned mq_count;
 925 };
 926
 927 static int device_is_rq_based(struct dm_target *ti, struct dm_dev *dev,
 928                               sector_t start, sector_t len, void *data)
 929 {
 930         struct request_queue *q = bdev_get_queue(dev->bdev);
 931         struct verify_rq_based_data *v = data;
 932
 933         if (q->mq_ops)
 934                 v->mq_count++;
 935         else
 936                 v->sq_count++;
 937
 938         return queue_is_rq_based(q);
 939 }
 940
 941 static int dm_table_determine_type(struct dm_table *t)
 942 {
 943         unsigned i;
 944         unsigned bio_based = 0, request_based = 0, hybrid = 0;
 945         struct verify_rq_based_data v = {.sq_count = 0, .mq_count = 0};
 946         struct dm_target *tgt;
 947         struct list_head *devices = dm_table_get_devices(t);
 948         enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
 949
 950         if (t->type != DM_TYPE_NONE) {
 951                 /* target already set the table's type */
 952                 if (t->type == DM_TYPE_BIO_BASED) {
 953                         /* possibly upgrade to a variant of bio-based */
 954                         goto verify_bio_based;
 955                 }
 956                 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
 957                 BUG_ON(t->type == DM_TYPE_NVME_BIO_BASED);
 958                 goto verify_rq_based;
 959         }
 960
 961         for (i = 0; i < t->num_targets; i++) {
 962                 tgt = t->targets + i;
 963                 if (dm_target_hybrid(tgt))
 964                         hybrid = 1;
 965                 else if (dm_target_request_based(tgt))
 966                         request_based = 1;
 967                 else
 968                         bio_based = 1;
 969
 970                 if (bio_based && request_based) {
 971                         DMERR("Inconsistent table: different target types"
 972                               " can't be mixed up");
 973                         return -EINVAL;
 974                 }
 975         }
 976
 977         if (hybrid && !bio_based && !request_based) {
 978                 /*
 979                  * The targets can work either way.
 980                  * Determine the type from the live device.
 981                  * Default to bio-based if device is new.
 982                  */
 983                 if (__table_type_request_based(live_md_type))
 984                         request_based = 1;
 985                 else
 986                         bio_based = 1;
 987         }
 988
 989         if (bio_based) {
 990 verify_bio_based:
 991                 /* We must use this table as bio-based */
 992                 t->type = DM_TYPE_BIO_BASED;
 993                 if (dm_table_supports_dax(t) ||
 994                     (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
 995                         t->type = DM_TYPE_DAX_BIO_BASED;
 996                 } else {
 997                         /* Check if upgrading to NVMe bio-based is valid or required */
 998                         tgt = dm_table_get_immutable_target(t);
 999                         if (tgt && !tgt->max_io_len && dm_table_does_not_support_partial_completion(t)) {
1000                                 t->type = DM_TYPE_NVME_BIO_BASED;
1001                                 goto verify_rq_based; /* must be stacked directly on NVMe (blk-mq) */
1002                         } else if (list_empty(devices) && live_md_type == DM_TYPE_NVME_BIO_BASED) {
1003                                 t->type = DM_TYPE_NVME_BIO_BASED;
1004                         }
1005                 }
1006                 return 0;
1007         }
1008
1009         BUG_ON(!request_based); /* No targets in this table */
1010
1011         /*
1012          * The only way to establish DM_TYPE_MQ_REQUEST_BASED is by
1013          * having a compatible target use dm_table_set_type.
1014          */
1015         t->type = DM_TYPE_REQUEST_BASED;
1016
1017 verify_rq_based:
1018         /*
1019          * Request-based dm supports only tables that have a single target now.
1020          * To support multiple targets, request splitting support is needed,
1021          * and that needs lots of changes in the block-layer.
1022          * (e.g. request completion process for partial completion.)
1023          */
1024         if (t->num_targets > 1) {
1025                 DMERR("%s DM doesn't support multiple targets",
1026                       t->type == DM_TYPE_NVME_BIO_BASED ? "nvme bio-based" : "request-based");
1027                 return -EINVAL;
1028         }
1029
1030         if (list_empty(devices)) {
1031                 int srcu_idx;
1032                 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
1033
1034                 /* inherit live table's type and all_blk_mq */
1035                 if (live_table) {
1036                         t->type = live_table->type;
1037                         t->all_blk_mq = live_table->all_blk_mq;
1038                 }
1039                 dm_put_live_table(t->md, srcu_idx);
1040                 return 0;
1041         }
1042
1043         tgt = dm_table_get_immutable_target(t);
1044         if (!tgt) {
1045                 DMERR("table load rejected: immutable target is required");
1046                 return -EINVAL;
1047         } else if (tgt->max_io_len) {
1048                 DMERR("table load rejected: immutable target that splits IO is not supported");
1049                 return -EINVAL;
1050         }
1051
1052         /* Non-request-stackable devices can't be used for request-based dm */
1053         if (!tgt->type->iterate_devices ||
1054             !tgt->type->iterate_devices(tgt, device_is_rq_based, &v)) {
1055                 DMERR("table load rejected: including non-request-stackable devices");
1056                 return -EINVAL;
1057         }
1058         if (v.sq_count && v.mq_count) {
1059                 DMERR("table load rejected: not all devices are blk-mq request-stackable");
1060                 return -EINVAL;
1061         }
1062         t->all_blk_mq = v.mq_count > 0;
1063
1064         if (!t->all_blk_mq &&
1065             (t->type == DM_TYPE_MQ_REQUEST_BASED || t->type == DM_TYPE_NVME_BIO_BASED)) {
1066                 DMERR("table load rejected: all devices are not blk-mq request-stackable");
1067                 return -EINVAL;
1068         }
1069
1070         return 0;
1071 }
1072
1073 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
1074 {
1075         return t->type;
1076 }
1077
1078 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
1079 {
1080         return t->immutable_target_type;
1081 }
1082
1083 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
1084 {
1085         /* Immutable target is implicitly a singleton */
1086         if (t->num_targets > 1 ||
1087             !dm_target_is_immutable(t->targets[0].type))
1088                 return NULL;
1089
1090         return t->targets;
1091 }
1092
1093 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1094 {
1095         struct dm_target *ti;
1096         unsigned i;
1097
1098         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1099                 ti = dm_table_get_target(t, i);
1100                 if (dm_target_is_wildcard(ti->type))
1101                         return ti;
1102         }
1103
1104         return NULL;
1105 }
1106
1107 bool dm_table_bio_based(struct dm_table *t)
1108 {
1109         return __table_type_bio_based(dm_table_get_type(t));
1110 }
1111
1112 bool dm_table_request_based(struct dm_table *t)
1113 {
1114         return __table_type_request_based(dm_table_get_type(t));
1115 }
1116
1117 bool dm_table_all_blk_mq_devices(struct dm_table *t)
1118 {
1119         return t->all_blk_mq;
1120 }
1121
1122 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1123 {
1124         enum dm_queue_mode type = dm_table_get_type(t);
1125         unsigned per_io_data_size = 0;
1126         unsigned min_pool_size = 0;
1127         struct dm_target *ti;
1128         unsigned i;
1129
1130         if (unlikely(type == DM_TYPE_NONE)) {
1131                 DMWARN("no table type is set, can't allocate mempools");
1132                 return -EINVAL;
1133         }
1134
1135         if (__table_type_bio_based(type))
1136                 for (i = 0; i < t->num_targets; i++) {
1137                         ti = t->targets + i;
1138                         per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1139                         min_pool_size = max(min_pool_size, ti->num_flush_bios);
1140                 }
1141
1142         t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported,
1143                                            per_io_data_size, min_pool_size);
1144         if (!t->mempools)
1145                 return -ENOMEM;
1146
1147         return 0;
1148 }
1149
1150 void dm_table_free_md_mempools(struct dm_table *t)
1151 {
1152         dm_free_md_mempools(t->mempools);
1153         t->mempools = NULL;
1154 }
1155
1156 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1157 {
1158         return t->mempools;
1159 }
1160
1161 static int setup_indexes(struct dm_table *t)
1162 {
1163         int i;
1164         unsigned int total = 0;
1165         sector_t *indexes;
1166
1167         /* allocate the space for *all* the indexes */
1168         for (i = t->depth - 2; i >= 0; i--) {
1169                 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1170                 total += t->counts[i];
1171         }
1172
1173         indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1174         if (!indexes)
1175                 return -ENOMEM;
1176
1177         /* set up internal nodes, bottom-up */
1178         for (i = t->depth - 2; i >= 0; i--) {
1179                 t->index[i] = indexes;
1180                 indexes += (KEYS_PER_NODE * t->counts[i]);
1181                 setup_btree_index(i, t);
1182         }
1183
1184         return 0;
1185 }
1186
1187 /*
1188  * Builds the btree to index the map.
1189  */
1190 static int dm_table_build_index(struct dm_table *t)
1191 {
1192         int r = 0;
1193         unsigned int leaf_nodes;
1194
1195         /* how many indexes will the btree have ? */
1196         leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1197         t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1198
1199         /* leaf layer has already been set up */
1200         t->counts[t->depth - 1] = leaf_nodes;
1201         t->index[t->depth - 1] = t->highs;
1202
1203         if (t->depth >= 2)
1204                 r = setup_indexes(t);
1205
1206         return r;
1207 }
1208
1209 static bool integrity_profile_exists(struct gendisk *disk)
1210 {
1211         return !!blk_get_integrity(disk);
1212 }
1213
1214 /*
1215  * Get a disk whose integrity profile reflects the table's profile.
1216  * Returns NULL if integrity support was inconsistent or unavailable.
1217  */
1218 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1219 {
1220         struct list_head *devices = dm_table_get_devices(t);
1221         struct dm_dev_internal *dd = NULL;
1222         struct gendisk *prev_disk = NULL, *template_disk = NULL;
1223         unsigned i;
1224
1225         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1226                 struct dm_target *ti = dm_table_get_target(t, i);
1227                 if (!dm_target_passes_integrity(ti->type))
1228                         goto no_integrity;
1229         }
1230
1231         list_for_each_entry(dd, devices, list) {
1232                 template_disk = dd->dm_dev->bdev->bd_disk;
1233                 if (!integrity_profile_exists(template_disk))
1234                         goto no_integrity;
1235                 else if (prev_disk &&
1236                          blk_integrity_compare(prev_disk, template_disk) < 0)
1237                         goto no_integrity;
1238                 prev_disk = template_disk;
1239         }
1240
1241         return template_disk;
1242
1243 no_integrity:
1244         if (prev_disk)
1245                 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1246                        dm_device_name(t->md),
1247                        prev_disk->disk_name,
1248                        template_disk->disk_name);
1249         return NULL;
1250 }
1251
1252 /*
1253  * Register the mapped device for blk_integrity support if the
1254  * underlying devices have an integrity profile.  But all devices may
1255  * not have matching profiles (checking all devices isn't reliable
1256  * during table load because this table may use other DM device(s) which
1257  * must be resumed before they will have an initialized integity
1258  * profile).  Consequently, stacked DM devices force a 2 stage integrity
1259  * profile validation: First pass during table load, final pass during
1260  * resume.
1261  */
1262 static int dm_table_register_integrity(struct dm_table *t)
1263 {
1264         struct mapped_device *md = t->md;
1265         struct gendisk *template_disk = NULL;
1266
1267         /* If target handles integrity itself do not register it here. */
1268         if (t->integrity_added)
1269                 return 0;
1270
1271         template_disk = dm_table_get_integrity_disk(t);
1272         if (!template_disk)
1273                 return 0;
1274
1275         if (!integrity_profile_exists(dm_disk(md))) {
1276                 t->integrity_supported = true;
1277                 /*
1278                  * Register integrity profile during table load; we can do
1279                  * this because the final profile must match during resume.
1280                  */
1281                 blk_integrity_register(dm_disk(md),
1282                                        blk_get_integrity(template_disk));
1283                 return 0;
1284         }
1285
1286         /*
1287          * If DM device already has an initialized integrity
1288          * profile the new profile should not conflict.
1289          */
1290         if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1291                 DMWARN("%s: conflict with existing integrity profile: "
1292                        "%s profile mismatch",
1293                        dm_device_name(t->md),
1294                        template_disk->disk_name);
1295                 return 1;
1296         }
1297
1298         /* Preserve existing integrity profile */
1299         t->integrity_supported = true;
1300         return 0;
1301 }
1302
1303 /*
1304  * Prepares the table for use by building the indices,
1305  * setting the type, and allocating mempools.
1306  */
1307 int dm_table_complete(struct dm_table *t)
1308 {
1309         int r;
1310
1311         r = dm_table_determine_type(t);
1312         if (r) {
1313                 DMERR("unable to determine table type");
1314                 return r;
1315         }
1316
1317         r = dm_table_build_index(t);
1318         if (r) {
1319                 DMERR("unable to build btrees");
1320                 return r;
1321         }
1322
1323         r = dm_table_register_integrity(t);
1324         if (r) {
1325                 DMERR("could not register integrity profile.");
1326                 return r;
1327         }
1328
1329         r = dm_table_alloc_md_mempools(t, t->md);
1330         if (r)
1331                 DMERR("unable to allocate mempools");
1332
1333         return r;
1334 }
1335
1336 static DEFINE_MUTEX(_event_lock);
1337 void dm_table_event_callback(struct dm_table *t,
1338                              void (*fn)(void *), void *context)
1339 {
1340         mutex_lock(&_event_lock);
1341         t->event_fn = fn;
1342         t->event_context = context;
1343         mutex_unlock(&_event_lock);
1344 }
1345
1346 void dm_table_event(struct dm_table *t)
1347 {
1348         mutex_lock(&_event_lock);
1349         if (t->event_fn)
1350                 t->event_fn(t->event_context);
1351         mutex_unlock(&_event_lock);
1352 }
1353 EXPORT_SYMBOL(dm_table_event);
1354
1355 inline sector_t dm_table_get_size(struct dm_table *t)
1356 {
1357         return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1358 }
1359 EXPORT_SYMBOL(dm_table_get_size);
1360
1361 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1362 {
1363         if (index >= t->num_targets)
1364                 return NULL;
1365
1366         return t->targets + index;
1367 }
1368
1369 /*
1370  * Search the btree for the correct target.
1371  *
1372  * Caller should check returned pointer with dm_target_is_valid()
1373  * to trap I/O beyond end of device.
1374  */
1375 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1376 {
1377         unsigned int l, n = 0, k = 0;
1378         sector_t *node;
1379
1380         if (unlikely(sector >= dm_table_get_size(t)))
1381                 return &t->targets[t->num_targets];
1382
1383         for (l = 0; l < t->depth; l++) {
1384                 n = get_child(n, k);
1385                 node = get_node(t, l, n);
1386
1387                 for (k = 0; k < KEYS_PER_NODE; k++)
1388                         if (node[k] >= sector)
1389                                 break;
1390         }
1391
1392         return &t->targets[(KEYS_PER_NODE * n) + k];
1393 }
1394
1395 /*
1396  * type->iterate_devices() should be called when the sanity check needs to
1397  * iterate and check all underlying data devices. iterate_devices() will
1398  * iterate all underlying data devices until it encounters a non-zero return
1399  * code, returned by whether the input iterate_devices_callout_fn, or
1400  * iterate_devices() itself internally.
1401  *
1402  * For some target type (e.g. dm-stripe), one call of iterate_devices() may
1403  * iterate multiple underlying devices internally, in which case a non-zero
1404  * return code returned by iterate_devices_callout_fn will stop the iteration
1405  * in advance.
1406  *
1407  * Cases requiring _any_ underlying device supporting some kind of attribute,
1408  * should use the iteration structure like dm_table_any_dev_attr(), or call
1409  * it directly. @func should handle semantics of positive examples, e.g.
1410  * capable of something.
1411  *
1412  * Cases requiring _all_ underlying devices supporting some kind of attribute,
1413  * should use the iteration structure like dm_table_supports_nowait() or
1414  * dm_table_supports_discards(). Or introduce dm_table_all_devs_attr() that
1415  * uses an @anti_func that handle semantics of counter examples, e.g. not
1416  * capable of something. So: return !dm_table_any_dev_attr(t, anti_func, data);
1417  */
1418 static bool dm_table_any_dev_attr(struct dm_table *t,
1419                                   iterate_devices_callout_fn func, void *data)
1420 {
1421         struct dm_target *ti;
1422         unsigned int i;
1423
1424         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1425                 ti = dm_table_get_target(t, i);
1426
1427                 if (ti->type->iterate_devices &&
1428                     ti->type->iterate_devices(ti, func, data))
1429                         return true;
1430         }
1431
1432         return false;
1433 }
1434
1435 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1436                         sector_t start, sector_t len, void *data)
1437 {
1438         unsigned *num_devices = data;
1439
1440         (*num_devices)++;
1441
1442         return 0;
1443 }
1444
1445 /*
1446  * Check whether a table has no data devices attached using each
1447  * target's iterate_devices method.
1448  * Returns false if the result is unknown because a target doesn't
1449  * support iterate_devices.
1450  */
1451 bool dm_table_has_no_data_devices(struct dm_table *table)
1452 {
1453         struct dm_target *ti;
1454         unsigned i, num_devices;
1455
1456         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1457                 ti = dm_table_get_target(table, i);
1458
1459                 if (!ti->type->iterate_devices)
1460                         return false;
1461
1462                 num_devices = 0;
1463                 ti->type->iterate_devices(ti, count_device, &num_devices);
1464                 if (num_devices)
1465                         return false;
1466         }
1467
1468         return true;
1469 }
1470
1471 static int device_not_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1472                                   sector_t start, sector_t len, void *data)
1473 {
1474         struct request_queue *q = bdev_get_queue(dev->bdev);
1475         enum blk_zoned_model *zoned_model = data;
1476
1477         return !q || blk_queue_zoned_model(q) != *zoned_model;
1478 }
1479
1480 static bool dm_table_supports_zoned_model(struct dm_table *t,
1481                                           enum blk_zoned_model zoned_model)
1482 {
1483         struct dm_target *ti;
1484         unsigned i;
1485
1486         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1487                 ti = dm_table_get_target(t, i);
1488
1489                 if (zoned_model == BLK_ZONED_HM &&
1490                     !dm_target_supports_zoned_hm(ti->type))
1491                         return false;
1492
1493                 if (!ti->type->iterate_devices ||
1494                     ti->type->iterate_devices(ti, device_not_zoned_model, &zoned_model))
1495                         return false;
1496         }
1497
1498         return true;
1499 }
1500
1501 static int device_not_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1502                                            sector_t start, sector_t len, void *data)
1503 {
1504         struct request_queue *q = bdev_get_queue(dev->bdev);
1505         unsigned int *zone_sectors = data;
1506
1507         return !q || blk_queue_zone_sectors(q) != *zone_sectors;
1508 }
1509
1510 static int validate_hardware_zoned_model(struct dm_table *table,
1511                                          enum blk_zoned_model zoned_model,
1512                                          unsigned int zone_sectors)
1513 {
1514         if (zoned_model == BLK_ZONED_NONE)
1515                 return 0;
1516
1517         if (!dm_table_supports_zoned_model(table, zoned_model)) {
1518                 DMERR("%s: zoned model is not consistent across all devices",
1519                       dm_device_name(table->md));
1520                 return -EINVAL;
1521         }
1522
1523         /* Check zone size validity and compatibility */
1524         if (!zone_sectors || !is_power_of_2(zone_sectors))
1525                 return -EINVAL;
1526
1527         if (dm_table_any_dev_attr(table, device_not_matches_zone_sectors, &zone_sectors)) {
1528                 DMERR("%s: zone sectors is not consistent across all devices",
1529                       dm_device_name(table->md));
1530                 return -EINVAL;
1531         }
1532
1533         return 0;
1534 }
1535
1536 /*
1537  * Establish the new table's queue_limits and validate them.
1538  */
1539 int dm_calculate_queue_limits(struct dm_table *table,
1540                               struct queue_limits *limits)
1541 {
1542         struct dm_target *ti;
1543         struct queue_limits ti_limits;
1544         unsigned i;
1545         enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1546         unsigned int zone_sectors = 0;
1547
1548         blk_set_stacking_limits(limits);
1549
1550         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1551                 blk_set_stacking_limits(&ti_limits);
1552
1553                 ti = dm_table_get_target(table, i);
1554
1555                 if (!ti->type->iterate_devices)
1556                         goto combine_limits;
1557
1558                 /*
1559                  * Combine queue limits of all the devices this target uses.
1560                  */
1561                 ti->type->iterate_devices(ti, dm_set_device_limits,
1562                                           &ti_limits);
1563
1564                 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1565                         /*
1566                          * After stacking all limits, validate all devices
1567                          * in table support this zoned model and zone sectors.
1568                          */
1569                         zoned_model = ti_limits.zoned;
1570                         zone_sectors = ti_limits.chunk_sectors;
1571                 }
1572
1573                 /* Set I/O hints portion of queue limits */
1574                 if (ti->type->io_hints)
1575                         ti->type->io_hints(ti, &ti_limits);
1576
1577                 /*
1578                  * Check each device area is consistent with the target's
1579                  * overall queue limits.
1580                  */
1581                 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1582                                               &ti_limits))
1583                         return -EINVAL;
1584
1585 combine_limits:
1586                 /*
1587                  * Merge this target's queue limits into the overall limits
1588                  * for the table.
1589                  */
1590                 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1591                         DMWARN("%s: adding target device "
1592                                "(start sect %llu len %llu) "
1593                                "caused an alignment inconsistency",
1594                                dm_device_name(table->md),
1595                                (unsigned long long) ti->begin,
1596                                (unsigned long long) ti->len);
1597
1598                 /*
1599                  * FIXME: this should likely be moved to blk_stack_limits(), would
1600                  * also eliminate limits->zoned stacking hack in dm_set_device_limits()
1601                  */
1602                 if (limits->zoned == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1603                         /*
1604                          * By default, the stacked limits zoned model is set to
1605                          * BLK_ZONED_NONE in blk_set_stacking_limits(). Update
1606                          * this model using the first target model reported
1607                          * that is not BLK_ZONED_NONE. This will be either the
1608                          * first target device zoned model or the model reported
1609                          * by the target .io_hints.
1610                          */
1611                         limits->zoned = ti_limits.zoned;
1612                 }
1613         }
1614
1615         /*
1616          * Verify that the zoned model and zone sectors, as determined before
1617          * any .io_hints override, are the same across all devices in the table.
1618          * - this is especially relevant if .io_hints is emulating a disk-managed
1619          *   zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1620          * BUT...
1621          */
1622         if (limits->zoned != BLK_ZONED_NONE) {
1623                 /*
1624                  * ...IF the above limits stacking determined a zoned model
1625                  * validate that all of the table's devices conform to it.
1626                  */
1627                 zoned_model = limits->zoned;
1628                 zone_sectors = limits->chunk_sectors;
1629         }
1630         if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1631                 return -EINVAL;
1632
1633         return validate_hardware_logical_block_alignment(table, limits);
1634 }
1635
1636 /*
1637  * Verify that all devices have an integrity profile that matches the
1638  * DM device's registered integrity profile.  If the profiles don't
1639  * match then unregister the DM device's integrity profile.
1640  */
1641 static void dm_table_verify_integrity(struct dm_table *t)
1642 {
1643         struct gendisk *template_disk = NULL;
1644
1645         if (t->integrity_added)
1646                 return;
1647
1648         if (t->integrity_supported) {
1649                 /*
1650                  * Verify that the original integrity profile
1651                  * matches all the devices in this table.
1652                  */
1653                 template_disk = dm_table_get_integrity_disk(t);
1654                 if (template_disk &&
1655                     blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1656                         return;
1657         }
1658
1659         if (integrity_profile_exists(dm_disk(t->md))) {
1660                 DMWARN("%s: unable to establish an integrity profile",
1661                        dm_device_name(t->md));
1662                 blk_integrity_unregister(dm_disk(t->md));
1663         }
1664 }
1665
1666 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1667                                 sector_t start, sector_t len, void *data)
1668 {
1669         unsigned long flush = (unsigned long) data;
1670         struct request_queue *q = bdev_get_queue(dev->bdev);
1671
1672         return q && (q->queue_flags & flush);
1673 }
1674
1675 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1676 {
1677         struct dm_target *ti;
1678         unsigned i;
1679
1680         /*
1681          * Require at least one underlying device to support flushes.
1682          * t->devices includes internal dm devices such as mirror logs
1683          * so we need to use iterate_devices here, which targets
1684          * supporting flushes must provide.
1685          */
1686         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1687                 ti = dm_table_get_target(t, i);
1688
1689                 if (!ti->num_flush_bios)
1690                         continue;
1691
1692                 if (ti->flush_supported)
1693                         return true;
1694
1695                 if (ti->type->iterate_devices &&
1696                     ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1697                         return true;
1698         }
1699
1700         return false;
1701 }
1702
1703 static int device_dax_write_cache_enabled(struct dm_target *ti,
1704                                           struct dm_dev *dev, sector_t start,
1705                                           sector_t len, void *data)
1706 {
1707         struct dax_device *dax_dev = dev->dax_dev;
1708
1709         if (!dax_dev)
1710                 return false;
1711
1712         if (dax_write_cache_enabled(dax_dev))
1713                 return true;
1714         return false;
1715 }
1716
1717 static int device_is_rotational(struct dm_target *ti, struct dm_dev *dev,
1718                                 sector_t start, sector_t len, void *data)
1719 {
1720         struct request_queue *q = bdev_get_queue(dev->bdev);
1721
1722         return q && !blk_queue_nonrot(q);
1723 }
1724
1725 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1726                              sector_t start, sector_t len, void *data)
1727 {
1728         struct request_queue *q = bdev_get_queue(dev->bdev);
1729
1730         return q && !blk_queue_add_random(q);
1731 }
1732
1733 static int queue_no_sg_merge(struct dm_target *ti, struct dm_dev *dev,
1734                              sector_t start, sector_t len, void *data)
1735 {
1736         struct request_queue *q = bdev_get_queue(dev->bdev);
1737
1738         return q && test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1739 }
1740
1741 static int device_is_partial_completion(struct dm_target *ti, struct dm_dev *dev,
1742                                         sector_t start, sector_t len, void *data)
1743 {
1744         char b[BDEVNAME_SIZE];
1745
1746         /* For now, NVMe devices are the only devices of this class */
1747         return (strncmp(bdevname(dev->bdev, b), "nvme", 4) != 0);
1748 }
1749
1750 static bool dm_table_does_not_support_partial_completion(struct dm_table *t)
1751 {
1752         return !dm_table_any_dev_attr(t, device_is_partial_completion, NULL);
1753 }
1754
1755 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1756                                          sector_t start, sector_t len, void *data)
1757 {
1758         struct request_queue *q = bdev_get_queue(dev->bdev);
1759
1760         return q && !q->limits.max_write_same_sectors;
1761 }
1762
1763 static bool dm_table_supports_write_same(struct dm_table *t)
1764 {
1765         struct dm_target *ti;
1766         unsigned 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_write_same_bios)
1772                         return false;
1773
1774                 if (!ti->type->iterate_devices ||
1775                     ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1776                         return false;
1777         }
1778
1779         return true;
1780 }
1781
1782 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1783                                            sector_t start, sector_t len, void *data)
1784 {
1785         struct request_queue *q = bdev_get_queue(dev->bdev);
1786
1787         return q && !q->limits.max_write_zeroes_sectors;
1788 }
1789
1790 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1791 {
1792         struct dm_target *ti;
1793         unsigned i = 0;
1794
1795         while (i < dm_table_get_num_targets(t)) {
1796                 ti = dm_table_get_target(t, i++);
1797
1798                 if (!ti->num_write_zeroes_bios)
1799                         return false;
1800
1801                 if (!ti->type->iterate_devices ||
1802                     ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1803                         return false;
1804         }
1805
1806         return true;
1807 }
1808
1809 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1810                                       sector_t start, sector_t len, void *data)
1811 {
1812         struct request_queue *q = bdev_get_queue(dev->bdev);
1813
1814         return q && !blk_queue_discard(q);
1815 }
1816
1817 static bool dm_table_supports_discards(struct dm_table *t)
1818 {
1819         struct dm_target *ti;
1820         unsigned i;
1821
1822         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1823                 ti = dm_table_get_target(t, i);
1824
1825                 if (!ti->num_discard_bios)
1826                         return false;
1827
1828                 /*
1829                  * Either the target provides discard support (as implied by setting
1830                  * 'discards_supported') or it relies on _all_ data devices having
1831                  * discard support.
1832                  */
1833                 if (!ti->discards_supported &&
1834                     (!ti->type->iterate_devices ||
1835                      ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1836                         return false;
1837         }
1838
1839         return true;
1840 }
1841
1842 static int device_not_secure_erase_capable(struct dm_target *ti,
1843                                            struct dm_dev *dev, sector_t start,
1844                                            sector_t len, void *data)
1845 {
1846         struct request_queue *q = bdev_get_queue(dev->bdev);
1847
1848         return q && !blk_queue_secure_erase(q);
1849 }
1850
1851 static bool dm_table_supports_secure_erase(struct dm_table *t)
1852 {
1853         struct dm_target *ti;
1854         unsigned int i;
1855
1856         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1857                 ti = dm_table_get_target(t, i);
1858
1859                 if (!ti->num_secure_erase_bios)
1860                         return false;
1861
1862                 if (!ti->type->iterate_devices ||
1863                     ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1864                         return false;
1865         }
1866
1867         return true;
1868 }
1869
1870 static int device_requires_stable_pages(struct dm_target *ti,
1871                                         struct dm_dev *dev, sector_t start,
1872                                         sector_t len, void *data)
1873 {
1874         struct request_queue *q = bdev_get_queue(dev->bdev);
1875
1876         return q && bdi_cap_stable_pages_required(q->backing_dev_info);
1877 }
1878
1879 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1880                                struct queue_limits *limits)
1881 {
1882         bool wc = false, fua = false;
1883
1884         /*
1885          * Copy table's limits to the DM device's request_queue
1886          */
1887         q->limits = *limits;
1888
1889         if (!dm_table_supports_discards(t)) {
1890                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
1891                 /* Must also clear discard limits... */
1892                 q->limits.max_discard_sectors = 0;
1893                 q->limits.max_hw_discard_sectors = 0;
1894                 q->limits.discard_granularity = 0;
1895                 q->limits.discard_alignment = 0;
1896                 q->limits.discard_misaligned = 0;
1897         } else
1898                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
1899
1900         if (dm_table_supports_secure_erase(t))
1901                 blk_queue_flag_set(QUEUE_FLAG_SECERASE, q);
1902
1903         if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1904                 wc = true;
1905                 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1906                         fua = true;
1907         }
1908         blk_queue_write_cache(q, wc, fua);
1909
1910         if (dm_table_supports_dax(t))
1911                 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
1912         else
1913                 blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
1914
1915         if (dm_table_any_dev_attr(t, device_dax_write_cache_enabled, NULL))
1916                 dax_write_cache(t->md->dax_dev, true);
1917
1918         /* Ensure that all underlying devices are non-rotational. */
1919         if (dm_table_any_dev_attr(t, device_is_rotational, NULL))
1920                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
1921         else
1922                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
1923
1924         if (!dm_table_supports_write_same(t))
1925                 q->limits.max_write_same_sectors = 0;
1926         if (!dm_table_supports_write_zeroes(t))
1927                 q->limits.max_write_zeroes_sectors = 0;
1928
1929         if (dm_table_any_dev_attr(t, queue_no_sg_merge, NULL))
1930                 blk_queue_flag_set(QUEUE_FLAG_NO_SG_MERGE, q);
1931         else
1932                 blk_queue_flag_clear(QUEUE_FLAG_NO_SG_MERGE, q);
1933
1934         dm_table_verify_integrity(t);
1935
1936         /*
1937          * Some devices don't use blk_integrity but still want stable pages
1938          * because they do their own checksumming.
1939          * If any underlying device requires stable pages, a table must require
1940          * them as well.  Only targets that support iterate_devices are considered:
1941          * don't want error, zero, etc to require stable pages.
1942          */
1943         if (dm_table_any_dev_attr(t, device_requires_stable_pages, NULL))
1944                 q->backing_dev_info->capabilities |= BDI_CAP_STABLE_WRITES;
1945         else
1946                 q->backing_dev_info->capabilities &= ~BDI_CAP_STABLE_WRITES;
1947
1948         /*
1949          * Determine whether or not this queue's I/O timings contribute
1950          * to the entropy pool, Only request-based targets use this.
1951          * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1952          * have it set.
1953          */
1954         if (blk_queue_add_random(q) &&
1955             dm_table_any_dev_attr(t, device_is_not_random, NULL))
1956                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
1957 }
1958
1959 unsigned int dm_table_get_num_targets(struct dm_table *t)
1960 {
1961         return t->num_targets;
1962 }
1963
1964 struct list_head *dm_table_get_devices(struct dm_table *t)
1965 {
1966         return &t->devices;
1967 }
1968
1969 fmode_t dm_table_get_mode(struct dm_table *t)
1970 {
1971         return t->mode;
1972 }
1973 EXPORT_SYMBOL(dm_table_get_mode);
1974
1975 enum suspend_mode {
1976         PRESUSPEND,
1977         PRESUSPEND_UNDO,
1978         POSTSUSPEND,
1979 };
1980
1981 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1982 {
1983         int i = t->num_targets;
1984         struct dm_target *ti = t->targets;
1985
1986         lockdep_assert_held(&t->md->suspend_lock);
1987
1988         while (i--) {
1989                 switch (mode) {
1990                 case PRESUSPEND:
1991                         if (ti->type->presuspend)
1992                                 ti->type->presuspend(ti);
1993                         break;
1994                 case PRESUSPEND_UNDO:
1995                         if (ti->type->presuspend_undo)
1996                                 ti->type->presuspend_undo(ti);
1997                         break;
1998                 case POSTSUSPEND:
1999                         if (ti->type->postsuspend)
2000                                 ti->type->postsuspend(ti);
2001                         break;
2002                 }
2003                 ti++;
2004         }
2005 }
2006
2007 void dm_table_presuspend_targets(struct dm_table *t)
2008 {
2009         if (!t)
2010                 return;
2011
2012         suspend_targets(t, PRESUSPEND);
2013 }
2014
2015 void dm_table_presuspend_undo_targets(struct dm_table *t)
2016 {
2017         if (!t)
2018                 return;
2019
2020         suspend_targets(t, PRESUSPEND_UNDO);
2021 }
2022
2023 void dm_table_postsuspend_targets(struct dm_table *t)
2024 {
2025         if (!t)
2026                 return;
2027
2028         suspend_targets(t, POSTSUSPEND);
2029 }
2030
2031 int dm_table_resume_targets(struct dm_table *t)
2032 {
2033         int i, r = 0;
2034
2035         lockdep_assert_held(&t->md->suspend_lock);
2036
2037         for (i = 0; i < t->num_targets; i++) {
2038                 struct dm_target *ti = t->targets + i;
2039
2040                 if (!ti->type->preresume)
2041                         continue;
2042
2043                 r = ti->type->preresume(ti);
2044                 if (r) {
2045                         DMERR("%s: %s: preresume failed, error = %d",
2046                               dm_device_name(t->md), ti->type->name, r);
2047                         return r;
2048                 }
2049         }
2050
2051         for (i = 0; i < t->num_targets; i++) {
2052                 struct dm_target *ti = t->targets + i;
2053
2054                 if (ti->type->resume)
2055                         ti->type->resume(ti);
2056         }
2057
2058         return 0;
2059 }
2060
2061 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
2062 {
2063         list_add(&cb->list, &t->target_callbacks);
2064 }
2065 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
2066
2067 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
2068 {
2069         struct dm_dev_internal *dd;
2070         struct list_head *devices = dm_table_get_devices(t);
2071         struct dm_target_callbacks *cb;
2072         int r = 0;
2073
2074         list_for_each_entry(dd, devices, list) {
2075                 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
2076                 char b[BDEVNAME_SIZE];
2077
2078                 if (likely(q))
2079                         r |= bdi_congested(q->backing_dev_info, bdi_bits);
2080                 else
2081                         DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
2082                                      dm_device_name(t->md),
2083                                      bdevname(dd->dm_dev->bdev, b));
2084         }
2085
2086         list_for_each_entry(cb, &t->target_callbacks, list)
2087                 if (cb->congested_fn)
2088                         r |= cb->congested_fn(cb, bdi_bits);
2089
2090         return r;
2091 }
2092
2093 struct mapped_device *dm_table_get_md(struct dm_table *t)
2094 {
2095         return t->md;
2096 }
2097 EXPORT_SYMBOL(dm_table_get_md);
2098
2099 void dm_table_run_md_queue_async(struct dm_table *t)
2100 {
2101         struct mapped_device *md;
2102         struct request_queue *queue;
2103         unsigned long flags;
2104
2105         if (!dm_table_request_based(t))
2106                 return;
2107
2108         md = dm_table_get_md(t);
2109         queue = dm_get_md_queue(md);
2110         if (queue) {
2111                 if (queue->mq_ops)
2112                         blk_mq_run_hw_queues(queue, true);
2113                 else {
2114                         spin_lock_irqsave(queue->queue_lock, flags);
2115                         blk_run_queue_async(queue);
2116                         spin_unlock_irqrestore(queue->queue_lock, flags);
2117                 }
2118         }
2119 }
2120 EXPORT_SYMBOL(dm_table_run_md_queue_async);
2121