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                 atomic_set(&dd->count, 0);
 464                 list_add(&dd->list, &t->devices);
 465
 466         } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
 467                 r = upgrade_mode(dd, mode, t->md);
 468                 if (r)
 469                         return r;
 470         }
 471         atomic_inc(&dd->count);
 472
 473         *result = dd->dm_dev;
 474         return 0;
 475 }
 476 EXPORT_SYMBOL(dm_get_device);
 477
 478 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
 479                                 sector_t start, sector_t len, void *data)
 480 {
 481         struct queue_limits *limits = data;
 482         struct block_device *bdev = dev->bdev;
 483         struct request_queue *q = bdev_get_queue(bdev);
 484         char b[BDEVNAME_SIZE];
 485
 486         if (unlikely(!q)) {
 487                 DMWARN("%s: Cannot set limits for nonexistent device %s",
 488                        dm_device_name(ti->table->md), bdevname(bdev, b));
 489                 return 0;
 490         }
 491
 492         if (bdev_stack_limits(limits, bdev, start) < 0)
 493                 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
 494                        "physical_block_size=%u, logical_block_size=%u, "
 495                        "alignment_offset=%u, start=%llu",
 496                        dm_device_name(ti->table->md), bdevname(bdev, b),
 497                        q->limits.physical_block_size,
 498                        q->limits.logical_block_size,
 499                        q->limits.alignment_offset,
 500                        (unsigned long long) start << SECTOR_SHIFT);
 501
 502         limits->zoned = blk_queue_zoned_model(q);
 503
 504         return 0;
 505 }
 506
 507 /*
 508  * Decrement a device's use count and remove it if necessary.
 509  */
 510 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
 511 {
 512         int found = 0;
 513         struct list_head *devices = &ti->table->devices;
 514         struct dm_dev_internal *dd;
 515
 516         list_for_each_entry(dd, devices, list) {
 517                 if (dd->dm_dev == d) {
 518                         found = 1;
 519                         break;
 520                 }
 521         }
 522         if (!found) {
 523                 DMWARN("%s: device %s not in table devices list",
 524                        dm_device_name(ti->table->md), d->name);
 525                 return;
 526         }
 527         if (atomic_dec_and_test(&dd->count)) {
 528                 dm_put_table_device(ti->table->md, d);
 529                 list_del(&dd->list);
 530                 kfree(dd);
 531         }
 532 }
 533 EXPORT_SYMBOL(dm_put_device);
 534
 535 /*
 536  * Checks to see if the target joins onto the end of the table.
 537  */
 538 static int adjoin(struct dm_table *table, struct dm_target *ti)
 539 {
 540         struct dm_target *prev;
 541
 542         if (!table->num_targets)
 543                 return !ti->begin;
 544
 545         prev = &table->targets[table->num_targets - 1];
 546         return (ti->begin == (prev->begin + prev->len));
 547 }
 548
 549 /*
 550  * Used to dynamically allocate the arg array.
 551  *
 552  * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
 553  * process messages even if some device is suspended. These messages have a
 554  * small fixed number of arguments.
 555  *
 556  * On the other hand, dm-switch needs to process bulk data using messages and
 557  * excessive use of GFP_NOIO could cause trouble.
 558  */
 559 static char **realloc_argv(unsigned *size, char **old_argv)
 560 {
 561         char **argv;
 562         unsigned new_size;
 563         gfp_t gfp;
 564
 565         if (*size) {
 566                 new_size = *size * 2;
 567                 gfp = GFP_KERNEL;
 568         } else {
 569                 new_size = 8;
 570                 gfp = GFP_NOIO;
 571         }
 572         argv = kmalloc(new_size * sizeof(*argv), gfp);
 573         if (argv) {
 574                 memcpy(argv, old_argv, *size * sizeof(*argv));
 575                 *size = new_size;
 576         }
 577
 578         kfree(old_argv);
 579         return argv;
 580 }
 581
 582 /*
 583  * Destructively splits up the argument list to pass to ctr.
 584  */
 585 int dm_split_args(int *argc, char ***argvp, char *input)
 586 {
 587         char *start, *end = input, *out, **argv = NULL;
 588         unsigned array_size = 0;
 589
 590         *argc = 0;
 591
 592         if (!input) {
 593                 *argvp = NULL;
 594                 return 0;
 595         }
 596
 597         argv = realloc_argv(&array_size, argv);
 598         if (!argv)
 599                 return -ENOMEM;
 600
 601         while (1) {
 602                 /* Skip whitespace */
 603                 start = skip_spaces(end);
 604
 605                 if (!*start)
 606                         break;  /* success, we hit the end */
 607
 608                 /* 'out' is used to remove any back-quotes */
 609                 end = out = start;
 610                 while (*end) {
 611                         /* Everything apart from '\0' can be quoted */
 612                         if (*end == '\\' && *(end + 1)) {
 613                                 *out++ = *(end + 1);
 614                                 end += 2;
 615                                 continue;
 616                         }
 617
 618                         if (isspace(*end))
 619                                 break;  /* end of token */
 620
 621                         *out++ = *end++;
 622                 }
 623
 624                 /* have we already filled the array ? */
 625                 if ((*argc + 1) > array_size) {
 626                         argv = realloc_argv(&array_size, argv);
 627                         if (!argv)
 628                                 return -ENOMEM;
 629                 }
 630
 631                 /* we know this is whitespace */
 632                 if (*end)
 633                         end++;
 634
 635                 /* terminate the string and put it in the array */
 636                 *out = '\0';
 637                 argv[*argc] = start;
 638                 (*argc)++;
 639         }
 640
 641         *argvp = argv;
 642         return 0;
 643 }
 644
 645 /*
 646  * Impose necessary and sufficient conditions on a devices's table such
 647  * that any incoming bio which respects its logical_block_size can be
 648  * processed successfully.  If it falls across the boundary between
 649  * two or more targets, the size of each piece it gets split into must
 650  * be compatible with the logical_block_size of the target processing it.
 651  */
 652 static int validate_hardware_logical_block_alignment(struct dm_table *table,
 653                                                  struct queue_limits *limits)
 654 {
 655         /*
 656          * This function uses arithmetic modulo the logical_block_size
 657          * (in units of 512-byte sectors).
 658          */
 659         unsigned short device_logical_block_size_sects =
 660                 limits->logical_block_size >> SECTOR_SHIFT;
 661
 662         /*
 663          * Offset of the start of the next table entry, mod logical_block_size.
 664          */
 665         unsigned short next_target_start = 0;
 666
 667         /*
 668          * Given an aligned bio that extends beyond the end of a
 669          * target, how many sectors must the next target handle?
 670          */
 671         unsigned short remaining = 0;
 672
 673         struct dm_target *uninitialized_var(ti);
 674         struct queue_limits ti_limits;
 675         unsigned i;
 676
 677         /*
 678          * Check each entry in the table in turn.
 679          */
 680         for (i = 0; i < dm_table_get_num_targets(table); i++) {
 681                 ti = dm_table_get_target(table, i);
 682
 683                 blk_set_stacking_limits(&ti_limits);
 684
 685                 /* combine all target devices' limits */
 686                 if (ti->type->iterate_devices)
 687                         ti->type->iterate_devices(ti, dm_set_device_limits,
 688                                                   &ti_limits);
 689
 690                 /*
 691                  * If the remaining sectors fall entirely within this
 692                  * table entry are they compatible with its logical_block_size?
 693                  */
 694                 if (remaining < ti->len &&
 695                     remaining & ((ti_limits.logical_block_size >>
 696                                   SECTOR_SHIFT) - 1))
 697                         break;  /* Error */
 698
 699                 next_target_start =
 700                     (unsigned short) ((next_target_start + ti->len) &
 701                                       (device_logical_block_size_sects - 1));
 702                 remaining = next_target_start ?
 703                     device_logical_block_size_sects - next_target_start : 0;
 704         }
 705
 706         if (remaining) {
 707                 DMWARN("%s: table line %u (start sect %llu len %llu) "
 708                        "not aligned to h/w logical block size %u",
 709                        dm_device_name(table->md), i,
 710                        (unsigned long long) ti->begin,
 711                        (unsigned long long) ti->len,
 712                        limits->logical_block_size);
 713                 return -EINVAL;
 714         }
 715
 716         return 0;
 717 }
 718
 719 int dm_table_add_target(struct dm_table *t, const char *type,
 720                         sector_t start, sector_t len, char *params)
 721 {
 722         int r = -EINVAL, argc;
 723         char **argv;
 724         struct dm_target *tgt;
 725
 726         if (t->singleton) {
 727                 DMERR("%s: target type %s must appear alone in table",
 728                       dm_device_name(t->md), t->targets->type->name);
 729                 return -EINVAL;
 730         }
 731
 732         BUG_ON(t->num_targets >= t->num_allocated);
 733
 734         tgt = t->targets + t->num_targets;
 735         memset(tgt, 0, sizeof(*tgt));
 736
 737         if (!len) {
 738                 DMERR("%s: zero-length target", dm_device_name(t->md));
 739                 return -EINVAL;
 740         }
 741
 742         tgt->type = dm_get_target_type(type);
 743         if (!tgt->type) {
 744                 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
 745                 return -EINVAL;
 746         }
 747
 748         if (dm_target_needs_singleton(tgt->type)) {
 749                 if (t->num_targets) {
 750                         tgt->error = "singleton target type must appear alone in table";
 751                         goto bad;
 752                 }
 753                 t->singleton = true;
 754         }
 755
 756         if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
 757                 tgt->error = "target type may not be included in a read-only table";
 758                 goto bad;
 759         }
 760
 761         if (t->immutable_target_type) {
 762                 if (t->immutable_target_type != tgt->type) {
 763                         tgt->error = "immutable target type cannot be mixed with other target types";
 764                         goto bad;
 765                 }
 766         } else if (dm_target_is_immutable(tgt->type)) {
 767                 if (t->num_targets) {
 768                         tgt->error = "immutable target type cannot be mixed with other target types";
 769                         goto bad;
 770                 }
 771                 t->immutable_target_type = tgt->type;
 772         }
 773
 774         if (dm_target_has_integrity(tgt->type))
 775                 t->integrity_added = 1;
 776
 777         tgt->table = t;
 778         tgt->begin = start;
 779         tgt->len = len;
 780         tgt->error = "Unknown error";
 781
 782         /*
 783          * Does this target adjoin the previous one ?
 784          */
 785         if (!adjoin(t, tgt)) {
 786                 tgt->error = "Gap in table";
 787                 goto bad;
 788         }
 789
 790         r = dm_split_args(&argc, &argv, params);
 791         if (r) {
 792                 tgt->error = "couldn't split parameters (insufficient memory)";
 793                 goto bad;
 794         }
 795
 796         r = tgt->type->ctr(tgt, argc, argv);
 797         kfree(argv);
 798         if (r)
 799                 goto bad;
 800
 801         t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
 802
 803         if (!tgt->num_discard_bios && tgt->discards_supported)
 804                 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
 805                        dm_device_name(t->md), type);
 806
 807         return 0;
 808
 809  bad:
 810         DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
 811         dm_put_target_type(tgt->type);
 812         return r;
 813 }
 814
 815 /*
 816  * Target argument parsing helpers.
 817  */
 818 static int validate_next_arg(const struct dm_arg *arg,
 819                              struct dm_arg_set *arg_set,
 820                              unsigned *value, char **error, unsigned grouped)
 821 {
 822         const char *arg_str = dm_shift_arg(arg_set);
 823         char dummy;
 824
 825         if (!arg_str ||
 826             (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
 827             (*value < arg->min) ||
 828             (*value > arg->max) ||
 829             (grouped && arg_set->argc < *value)) {
 830                 *error = arg->error;
 831                 return -EINVAL;
 832         }
 833
 834         return 0;
 835 }
 836
 837 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 838                 unsigned *value, char **error)
 839 {
 840         return validate_next_arg(arg, arg_set, value, error, 0);
 841 }
 842 EXPORT_SYMBOL(dm_read_arg);
 843
 844 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 845                       unsigned *value, char **error)
 846 {
 847         return validate_next_arg(arg, arg_set, value, error, 1);
 848 }
 849 EXPORT_SYMBOL(dm_read_arg_group);
 850
 851 const char *dm_shift_arg(struct dm_arg_set *as)
 852 {
 853         char *r;
 854
 855         if (as->argc) {
 856                 as->argc--;
 857                 r = *as->argv;
 858                 as->argv++;
 859                 return r;
 860         }
 861
 862         return NULL;
 863 }
 864 EXPORT_SYMBOL(dm_shift_arg);
 865
 866 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
 867 {
 868         BUG_ON(as->argc < num_args);
 869         as->argc -= num_args;
 870         as->argv += num_args;
 871 }
 872 EXPORT_SYMBOL(dm_consume_args);
 873
 874 static bool __table_type_bio_based(enum dm_queue_mode table_type)
 875 {
 876         return (table_type == DM_TYPE_BIO_BASED ||
 877                 table_type == DM_TYPE_DAX_BIO_BASED);
 878 }
 879
 880 static bool __table_type_request_based(enum dm_queue_mode table_type)
 881 {
 882         return (table_type == DM_TYPE_REQUEST_BASED ||
 883                 table_type == DM_TYPE_MQ_REQUEST_BASED);
 884 }
 885
 886 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
 887 {
 888         t->type = type;
 889 }
 890 EXPORT_SYMBOL_GPL(dm_table_set_type);
 891
 892 static int device_not_dax_capable(struct dm_target *ti, struct dm_dev *dev,
 893                                sector_t start, sector_t len, void *data)
 894 {
 895         return !bdev_dax_supported(dev->bdev, PAGE_SIZE);
 896 }
 897
 898 static bool dm_table_supports_dax(struct dm_table *t)
 899 {
 900         struct dm_target *ti;
 901         unsigned i;
 902
 903         /* Ensure that all targets support DAX. */
 904         for (i = 0; i < dm_table_get_num_targets(t); i++) {
 905                 ti = dm_table_get_target(t, i);
 906
 907                 if (!ti->type->direct_access)
 908                         return false;
 909
 910                 if (!ti->type->iterate_devices ||
 911                     ti->type->iterate_devices(ti, device_not_dax_capable, NULL))
 912                         return false;
 913         }
 914
 915         return true;
 916 }
 917
 918 static int dm_table_determine_type(struct dm_table *t)
 919 {
 920         unsigned i;
 921         unsigned bio_based = 0, request_based = 0, hybrid = 0;
 922         unsigned sq_count = 0, mq_count = 0;
 923         struct dm_target *tgt;
 924         struct dm_dev_internal *dd;
 925         struct list_head *devices = dm_table_get_devices(t);
 926         enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
 927
 928         if (t->type != DM_TYPE_NONE) {
 929                 /* target already set the table's type */
 930                 if (t->type == DM_TYPE_BIO_BASED)
 931                         return 0;
 932                 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
 933                 goto verify_rq_based;
 934         }
 935
 936         for (i = 0; i < t->num_targets; i++) {
 937                 tgt = t->targets + i;
 938                 if (dm_target_hybrid(tgt))
 939                         hybrid = 1;
 940                 else if (dm_target_request_based(tgt))
 941                         request_based = 1;
 942                 else
 943                         bio_based = 1;
 944
 945                 if (bio_based && request_based) {
 946                         DMWARN("Inconsistent table: different target types"
 947                                " can't be mixed up");
 948                         return -EINVAL;
 949                 }
 950         }
 951
 952         if (hybrid && !bio_based && !request_based) {
 953                 /*
 954                  * The targets can work either way.
 955                  * Determine the type from the live device.
 956                  * Default to bio-based if device is new.
 957                  */
 958                 if (__table_type_request_based(live_md_type))
 959                         request_based = 1;
 960                 else
 961                         bio_based = 1;
 962         }
 963
 964         if (bio_based) {
 965                 /* We must use this table as bio-based */
 966                 t->type = DM_TYPE_BIO_BASED;
 967                 if (dm_table_supports_dax(t) ||
 968                     (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED))
 969                         t->type = DM_TYPE_DAX_BIO_BASED;
 970                 return 0;
 971         }
 972
 973         BUG_ON(!request_based); /* No targets in this table */
 974
 975         /*
 976          * The only way to establish DM_TYPE_MQ_REQUEST_BASED is by
 977          * having a compatible target use dm_table_set_type.
 978          */
 979         t->type = DM_TYPE_REQUEST_BASED;
 980
 981 verify_rq_based:
 982         /*
 983          * Request-based dm supports only tables that have a single target now.
 984          * To support multiple targets, request splitting support is needed,
 985          * and that needs lots of changes in the block-layer.
 986          * (e.g. request completion process for partial completion.)
 987          */
 988         if (t->num_targets > 1) {
 989                 DMWARN("Request-based dm doesn't support multiple targets yet");
 990                 return -EINVAL;
 991         }
 992
 993         if (list_empty(devices)) {
 994                 int srcu_idx;
 995                 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
 996
 997                 /* inherit live table's type and all_blk_mq */
 998                 if (live_table) {
 999                         t->type = live_table->type;
1000                         t->all_blk_mq = live_table->all_blk_mq;
1001                 }
1002                 dm_put_live_table(t->md, srcu_idx);
1003                 return 0;
1004         }
1005
1006         /* Non-request-stackable devices can't be used for request-based dm */
1007         list_for_each_entry(dd, devices, list) {
1008                 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1009
1010                 if (!blk_queue_stackable(q)) {
1011                         DMERR("table load rejected: including"
1012                               " non-request-stackable devices");
1013                         return -EINVAL;
1014                 }
1015
1016                 if (q->mq_ops)
1017                         mq_count++;
1018                 else
1019                         sq_count++;
1020         }
1021         if (sq_count && mq_count) {
1022                 DMERR("table load rejected: not all devices are blk-mq request-stackable");
1023                 return -EINVAL;
1024         }
1025         t->all_blk_mq = mq_count > 0;
1026
1027         if (t->type == DM_TYPE_MQ_REQUEST_BASED && !t->all_blk_mq) {
1028                 DMERR("table load rejected: all devices are not blk-mq request-stackable");
1029                 return -EINVAL;
1030         }
1031
1032         return 0;
1033 }
1034
1035 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
1036 {
1037         return t->type;
1038 }
1039
1040 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
1041 {
1042         return t->immutable_target_type;
1043 }
1044
1045 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
1046 {
1047         /* Immutable target is implicitly a singleton */
1048         if (t->num_targets > 1 ||
1049             !dm_target_is_immutable(t->targets[0].type))
1050                 return NULL;
1051
1052         return t->targets;
1053 }
1054
1055 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1056 {
1057         struct dm_target *ti;
1058         unsigned i;
1059
1060         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1061                 ti = dm_table_get_target(t, i);
1062                 if (dm_target_is_wildcard(ti->type))
1063                         return ti;
1064         }
1065
1066         return NULL;
1067 }
1068
1069 bool dm_table_bio_based(struct dm_table *t)
1070 {
1071         return __table_type_bio_based(dm_table_get_type(t));
1072 }
1073
1074 bool dm_table_request_based(struct dm_table *t)
1075 {
1076         return __table_type_request_based(dm_table_get_type(t));
1077 }
1078
1079 bool dm_table_all_blk_mq_devices(struct dm_table *t)
1080 {
1081         return t->all_blk_mq;
1082 }
1083
1084 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1085 {
1086         enum dm_queue_mode type = dm_table_get_type(t);
1087         unsigned per_io_data_size = 0;
1088         struct dm_target *tgt;
1089         unsigned i;
1090
1091         if (unlikely(type == DM_TYPE_NONE)) {
1092                 DMWARN("no table type is set, can't allocate mempools");
1093                 return -EINVAL;
1094         }
1095
1096         if (__table_type_bio_based(type))
1097                 for (i = 0; i < t->num_targets; i++) {
1098                         tgt = t->targets + i;
1099                         per_io_data_size = max(per_io_data_size, tgt->per_io_data_size);
1100                 }
1101
1102         t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported, per_io_data_size);
1103         if (!t->mempools)
1104                 return -ENOMEM;
1105
1106         return 0;
1107 }
1108
1109 void dm_table_free_md_mempools(struct dm_table *t)
1110 {
1111         dm_free_md_mempools(t->mempools);
1112         t->mempools = NULL;
1113 }
1114
1115 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1116 {
1117         return t->mempools;
1118 }
1119
1120 static int setup_indexes(struct dm_table *t)
1121 {
1122         int i;
1123         unsigned int total = 0;
1124         sector_t *indexes;
1125
1126         /* allocate the space for *all* the indexes */
1127         for (i = t->depth - 2; i >= 0; i--) {
1128                 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1129                 total += t->counts[i];
1130         }
1131
1132         indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1133         if (!indexes)
1134                 return -ENOMEM;
1135
1136         /* set up internal nodes, bottom-up */
1137         for (i = t->depth - 2; i >= 0; i--) {
1138                 t->index[i] = indexes;
1139                 indexes += (KEYS_PER_NODE * t->counts[i]);
1140                 setup_btree_index(i, t);
1141         }
1142
1143         return 0;
1144 }
1145
1146 /*
1147  * Builds the btree to index the map.
1148  */
1149 static int dm_table_build_index(struct dm_table *t)
1150 {
1151         int r = 0;
1152         unsigned int leaf_nodes;
1153
1154         /* how many indexes will the btree have ? */
1155         leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1156         t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1157
1158         /* leaf layer has already been set up */
1159         t->counts[t->depth - 1] = leaf_nodes;
1160         t->index[t->depth - 1] = t->highs;
1161
1162         if (t->depth >= 2)
1163                 r = setup_indexes(t);
1164
1165         return r;
1166 }
1167
1168 static bool integrity_profile_exists(struct gendisk *disk)
1169 {
1170         return !!blk_get_integrity(disk);
1171 }
1172
1173 /*
1174  * Get a disk whose integrity profile reflects the table's profile.
1175  * Returns NULL if integrity support was inconsistent or unavailable.
1176  */
1177 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1178 {
1179         struct list_head *devices = dm_table_get_devices(t);
1180         struct dm_dev_internal *dd = NULL;
1181         struct gendisk *prev_disk = NULL, *template_disk = NULL;
1182         unsigned i;
1183
1184         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1185                 struct dm_target *ti = dm_table_get_target(t, i);
1186                 if (!dm_target_passes_integrity(ti->type))
1187                         goto no_integrity;
1188         }
1189
1190         list_for_each_entry(dd, devices, list) {
1191                 template_disk = dd->dm_dev->bdev->bd_disk;
1192                 if (!integrity_profile_exists(template_disk))
1193                         goto no_integrity;
1194                 else if (prev_disk &&
1195                          blk_integrity_compare(prev_disk, template_disk) < 0)
1196                         goto no_integrity;
1197                 prev_disk = template_disk;
1198         }
1199
1200         return template_disk;
1201
1202 no_integrity:
1203         if (prev_disk)
1204                 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1205                        dm_device_name(t->md),
1206                        prev_disk->disk_name,
1207                        template_disk->disk_name);
1208         return NULL;
1209 }
1210
1211 /*
1212  * Register the mapped device for blk_integrity support if the
1213  * underlying devices have an integrity profile.  But all devices may
1214  * not have matching profiles (checking all devices isn't reliable
1215  * during table load because this table may use other DM device(s) which
1216  * must be resumed before they will have an initialized integity
1217  * profile).  Consequently, stacked DM devices force a 2 stage integrity
1218  * profile validation: First pass during table load, final pass during
1219  * resume.
1220  */
1221 static int dm_table_register_integrity(struct dm_table *t)
1222 {
1223         struct mapped_device *md = t->md;
1224         struct gendisk *template_disk = NULL;
1225
1226         /* If target handles integrity itself do not register it here. */
1227         if (t->integrity_added)
1228                 return 0;
1229
1230         template_disk = dm_table_get_integrity_disk(t);
1231         if (!template_disk)
1232                 return 0;
1233
1234         if (!integrity_profile_exists(dm_disk(md))) {
1235                 t->integrity_supported = true;
1236                 /*
1237                  * Register integrity profile during table load; we can do
1238                  * this because the final profile must match during resume.
1239                  */
1240                 blk_integrity_register(dm_disk(md),
1241                                        blk_get_integrity(template_disk));
1242                 return 0;
1243         }
1244
1245         /*
1246          * If DM device already has an initialized integrity
1247          * profile the new profile should not conflict.
1248          */
1249         if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1250                 DMWARN("%s: conflict with existing integrity profile: "
1251                        "%s profile mismatch",
1252                        dm_device_name(t->md),
1253                        template_disk->disk_name);
1254                 return 1;
1255         }
1256
1257         /* Preserve existing integrity profile */
1258         t->integrity_supported = true;
1259         return 0;
1260 }
1261
1262 /*
1263  * Prepares the table for use by building the indices,
1264  * setting the type, and allocating mempools.
1265  */
1266 int dm_table_complete(struct dm_table *t)
1267 {
1268         int r;
1269
1270         r = dm_table_determine_type(t);
1271         if (r) {
1272                 DMERR("unable to determine table type");
1273                 return r;
1274         }
1275
1276         r = dm_table_build_index(t);
1277         if (r) {
1278                 DMERR("unable to build btrees");
1279                 return r;
1280         }
1281
1282         r = dm_table_register_integrity(t);
1283         if (r) {
1284                 DMERR("could not register integrity profile.");
1285                 return r;
1286         }
1287
1288         r = dm_table_alloc_md_mempools(t, t->md);
1289         if (r)
1290                 DMERR("unable to allocate mempools");
1291
1292         return r;
1293 }
1294
1295 static DEFINE_MUTEX(_event_lock);
1296 void dm_table_event_callback(struct dm_table *t,
1297                              void (*fn)(void *), void *context)
1298 {
1299         mutex_lock(&_event_lock);
1300         t->event_fn = fn;
1301         t->event_context = context;
1302         mutex_unlock(&_event_lock);
1303 }
1304
1305 void dm_table_event(struct dm_table *t)
1306 {
1307         mutex_lock(&_event_lock);
1308         if (t->event_fn)
1309                 t->event_fn(t->event_context);
1310         mutex_unlock(&_event_lock);
1311 }
1312 EXPORT_SYMBOL(dm_table_event);
1313
1314 inline sector_t dm_table_get_size(struct dm_table *t)
1315 {
1316         return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1317 }
1318 EXPORT_SYMBOL(dm_table_get_size);
1319
1320 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1321 {
1322         if (index >= t->num_targets)
1323                 return NULL;
1324
1325         return t->targets + index;
1326 }
1327
1328 /*
1329  * Search the btree for the correct target.
1330  *
1331  * Caller should check returned pointer with dm_target_is_valid()
1332  * to trap I/O beyond end of device.
1333  */
1334 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1335 {
1336         unsigned int l, n = 0, k = 0;
1337         sector_t *node;
1338
1339         if (unlikely(sector >= dm_table_get_size(t)))
1340                 return &t->targets[t->num_targets];
1341
1342         for (l = 0; l < t->depth; l++) {
1343                 n = get_child(n, k);
1344                 node = get_node(t, l, n);
1345
1346                 for (k = 0; k < KEYS_PER_NODE; k++)
1347                         if (node[k] >= sector)
1348                                 break;
1349         }
1350
1351         return &t->targets[(KEYS_PER_NODE * n) + k];
1352 }
1353
1354 /*
1355  * type->iterate_devices() should be called when the sanity check needs to
1356  * iterate and check all underlying data devices. iterate_devices() will
1357  * iterate all underlying data devices until it encounters a non-zero return
1358  * code, returned by whether the input iterate_devices_callout_fn, or
1359  * iterate_devices() itself internally.
1360  *
1361  * For some target type (e.g. dm-stripe), one call of iterate_devices() may
1362  * iterate multiple underlying devices internally, in which case a non-zero
1363  * return code returned by iterate_devices_callout_fn will stop the iteration
1364  * in advance.
1365  *
1366  * Cases requiring _any_ underlying device supporting some kind of attribute,
1367  * should use the iteration structure like dm_table_any_dev_attr(), or call
1368  * it directly. @func should handle semantics of positive examples, e.g.
1369  * capable of something.
1370  *
1371  * Cases requiring _all_ underlying devices supporting some kind of attribute,
1372  * should use the iteration structure like dm_table_supports_nowait() or
1373  * dm_table_supports_discards(). Or introduce dm_table_all_devs_attr() that
1374  * uses an @anti_func that handle semantics of counter examples, e.g. not
1375  * capable of something. So: return !dm_table_any_dev_attr(t, anti_func, data);
1376  */
1377 static bool dm_table_any_dev_attr(struct dm_table *t,
1378                                   iterate_devices_callout_fn func, void *data)
1379 {
1380         struct dm_target *ti;
1381         unsigned int i;
1382
1383         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1384                 ti = dm_table_get_target(t, i);
1385
1386                 if (ti->type->iterate_devices &&
1387                     ti->type->iterate_devices(ti, func, data))
1388                         return true;
1389         }
1390
1391         return false;
1392 }
1393
1394 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1395                         sector_t start, sector_t len, void *data)
1396 {
1397         unsigned *num_devices = data;
1398
1399         (*num_devices)++;
1400
1401         return 0;
1402 }
1403
1404 /*
1405  * Check whether a table has no data devices attached using each
1406  * target's iterate_devices method.
1407  * Returns false if the result is unknown because a target doesn't
1408  * support iterate_devices.
1409  */
1410 bool dm_table_has_no_data_devices(struct dm_table *table)
1411 {
1412         struct dm_target *ti;
1413         unsigned i, num_devices;
1414
1415         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1416                 ti = dm_table_get_target(table, i);
1417
1418                 if (!ti->type->iterate_devices)
1419                         return false;
1420
1421                 num_devices = 0;
1422                 ti->type->iterate_devices(ti, count_device, &num_devices);
1423                 if (num_devices)
1424                         return false;
1425         }
1426
1427         return true;
1428 }
1429
1430 static int device_not_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1431                                   sector_t start, sector_t len, void *data)
1432 {
1433         struct request_queue *q = bdev_get_queue(dev->bdev);
1434         enum blk_zoned_model *zoned_model = data;
1435
1436         return !q || blk_queue_zoned_model(q) != *zoned_model;
1437 }
1438
1439 static bool dm_table_supports_zoned_model(struct dm_table *t,
1440                                           enum blk_zoned_model zoned_model)
1441 {
1442         struct dm_target *ti;
1443         unsigned i;
1444
1445         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1446                 ti = dm_table_get_target(t, i);
1447
1448                 if (zoned_model == BLK_ZONED_HM &&
1449                     !dm_target_supports_zoned_hm(ti->type))
1450                         return false;
1451
1452                 if (!ti->type->iterate_devices ||
1453                     ti->type->iterate_devices(ti, device_not_zoned_model, &zoned_model))
1454                         return false;
1455         }
1456
1457         return true;
1458 }
1459
1460 static int device_not_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1461                                            sector_t start, sector_t len, void *data)
1462 {
1463         struct request_queue *q = bdev_get_queue(dev->bdev);
1464         unsigned int *zone_sectors = data;
1465
1466         return !q || blk_queue_zone_sectors(q) != *zone_sectors;
1467 }
1468
1469 static int validate_hardware_zoned_model(struct dm_table *table,
1470                                          enum blk_zoned_model zoned_model,
1471                                          unsigned int zone_sectors)
1472 {
1473         if (zoned_model == BLK_ZONED_NONE)
1474                 return 0;
1475
1476         if (!dm_table_supports_zoned_model(table, zoned_model)) {
1477                 DMERR("%s: zoned model is not consistent across all devices",
1478                       dm_device_name(table->md));
1479                 return -EINVAL;
1480         }
1481
1482         /* Check zone size validity and compatibility */
1483         if (!zone_sectors || !is_power_of_2(zone_sectors))
1484                 return -EINVAL;
1485
1486         if (dm_table_any_dev_attr(table, device_not_matches_zone_sectors, &zone_sectors)) {
1487                 DMERR("%s: zone sectors is not consistent across all devices",
1488                       dm_device_name(table->md));
1489                 return -EINVAL;
1490         }
1491
1492         return 0;
1493 }
1494
1495 /*
1496  * Establish the new table's queue_limits and validate them.
1497  */
1498 int dm_calculate_queue_limits(struct dm_table *table,
1499                               struct queue_limits *limits)
1500 {
1501         struct dm_target *ti;
1502         struct queue_limits ti_limits;
1503         unsigned i;
1504         enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1505         unsigned int zone_sectors = 0;
1506
1507         blk_set_stacking_limits(limits);
1508
1509         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1510                 blk_set_stacking_limits(&ti_limits);
1511
1512                 ti = dm_table_get_target(table, i);
1513
1514                 if (!ti->type->iterate_devices)
1515                         goto combine_limits;
1516
1517                 /*
1518                  * Combine queue limits of all the devices this target uses.
1519                  */
1520                 ti->type->iterate_devices(ti, dm_set_device_limits,
1521                                           &ti_limits);
1522
1523                 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1524                         /*
1525                          * After stacking all limits, validate all devices
1526                          * in table support this zoned model and zone sectors.
1527                          */
1528                         zoned_model = ti_limits.zoned;
1529                         zone_sectors = ti_limits.chunk_sectors;
1530                 }
1531
1532                 /* Set I/O hints portion of queue limits */
1533                 if (ti->type->io_hints)
1534                         ti->type->io_hints(ti, &ti_limits);
1535
1536                 /*
1537                  * Check each device area is consistent with the target's
1538                  * overall queue limits.
1539                  */
1540                 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1541                                               &ti_limits))
1542                         return -EINVAL;
1543
1544 combine_limits:
1545                 /*
1546                  * Merge this target's queue limits into the overall limits
1547                  * for the table.
1548                  */
1549                 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1550                         DMWARN("%s: adding target device "
1551                                "(start sect %llu len %llu) "
1552                                "caused an alignment inconsistency",
1553                                dm_device_name(table->md),
1554                                (unsigned long long) ti->begin,
1555                                (unsigned long long) ti->len);
1556
1557                 /*
1558                  * FIXME: this should likely be moved to blk_stack_limits(), would
1559                  * also eliminate limits->zoned stacking hack in dm_set_device_limits()
1560                  */
1561                 if (limits->zoned == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1562                         /*
1563                          * By default, the stacked limits zoned model is set to
1564                          * BLK_ZONED_NONE in blk_set_stacking_limits(). Update
1565                          * this model using the first target model reported
1566                          * that is not BLK_ZONED_NONE. This will be either the
1567                          * first target device zoned model or the model reported
1568                          * by the target .io_hints.
1569                          */
1570                         limits->zoned = ti_limits.zoned;
1571                 }
1572         }
1573
1574         /*
1575          * Verify that the zoned model and zone sectors, as determined before
1576          * any .io_hints override, are the same across all devices in the table.
1577          * - this is especially relevant if .io_hints is emulating a disk-managed
1578          *   zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1579          * BUT...
1580          */
1581         if (limits->zoned != BLK_ZONED_NONE) {
1582                 /*
1583                  * ...IF the above limits stacking determined a zoned model
1584                  * validate that all of the table's devices conform to it.
1585                  */
1586                 zoned_model = limits->zoned;
1587                 zone_sectors = limits->chunk_sectors;
1588         }
1589         if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1590                 return -EINVAL;
1591
1592         return validate_hardware_logical_block_alignment(table, limits);
1593 }
1594
1595 /*
1596  * Verify that all devices have an integrity profile that matches the
1597  * DM device's registered integrity profile.  If the profiles don't
1598  * match then unregister the DM device's integrity profile.
1599  */
1600 static void dm_table_verify_integrity(struct dm_table *t)
1601 {
1602         struct gendisk *template_disk = NULL;
1603
1604         if (t->integrity_added)
1605                 return;
1606
1607         if (t->integrity_supported) {
1608                 /*
1609                  * Verify that the original integrity profile
1610                  * matches all the devices in this table.
1611                  */
1612                 template_disk = dm_table_get_integrity_disk(t);
1613                 if (template_disk &&
1614                     blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1615                         return;
1616         }
1617
1618         if (integrity_profile_exists(dm_disk(t->md))) {
1619                 DMWARN("%s: unable to establish an integrity profile",
1620                        dm_device_name(t->md));
1621                 blk_integrity_unregister(dm_disk(t->md));
1622         }
1623 }
1624
1625 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1626                                 sector_t start, sector_t len, void *data)
1627 {
1628         unsigned long flush = (unsigned long) data;
1629         struct request_queue *q = bdev_get_queue(dev->bdev);
1630
1631         return q && (q->queue_flags & flush);
1632 }
1633
1634 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1635 {
1636         struct dm_target *ti;
1637         unsigned i;
1638
1639         /*
1640          * Require at least one underlying device to support flushes.
1641          * t->devices includes internal dm devices such as mirror logs
1642          * so we need to use iterate_devices here, which targets
1643          * supporting flushes must provide.
1644          */
1645         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1646                 ti = dm_table_get_target(t, i);
1647
1648                 if (!ti->num_flush_bios)
1649                         continue;
1650
1651                 if (ti->flush_supported)
1652                         return true;
1653
1654                 if (ti->type->iterate_devices &&
1655                     ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1656                         return true;
1657         }
1658
1659         return false;
1660 }
1661
1662 static int device_dax_write_cache_enabled(struct dm_target *ti,
1663                                           struct dm_dev *dev, sector_t start,
1664                                           sector_t len, void *data)
1665 {
1666         struct dax_device *dax_dev = dev->dax_dev;
1667
1668         if (!dax_dev)
1669                 return false;
1670
1671         if (dax_write_cache_enabled(dax_dev))
1672                 return true;
1673         return false;
1674 }
1675
1676 static int device_is_rotational(struct dm_target *ti, struct dm_dev *dev,
1677                                 sector_t start, sector_t len, void *data)
1678 {
1679         struct request_queue *q = bdev_get_queue(dev->bdev);
1680
1681         return q && !blk_queue_nonrot(q);
1682 }
1683
1684 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1685                              sector_t start, sector_t len, void *data)
1686 {
1687         struct request_queue *q = bdev_get_queue(dev->bdev);
1688
1689         return q && !blk_queue_add_random(q);
1690 }
1691
1692 static int queue_no_sg_merge(struct dm_target *ti, struct dm_dev *dev,
1693                              sector_t start, sector_t len, void *data)
1694 {
1695         struct request_queue *q = bdev_get_queue(dev->bdev);
1696
1697         return q && test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1698 }
1699
1700 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1701                                          sector_t start, sector_t len, void *data)
1702 {
1703         struct request_queue *q = bdev_get_queue(dev->bdev);
1704
1705         return q && !q->limits.max_write_same_sectors;
1706 }
1707
1708 static bool dm_table_supports_write_same(struct dm_table *t)
1709 {
1710         struct dm_target *ti;
1711         unsigned i;
1712
1713         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1714                 ti = dm_table_get_target(t, i);
1715
1716                 if (!ti->num_write_same_bios)
1717                         return false;
1718
1719                 if (!ti->type->iterate_devices ||
1720                     ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1721                         return false;
1722         }
1723
1724         return true;
1725 }
1726
1727 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1728                                            sector_t start, sector_t len, void *data)
1729 {
1730         struct request_queue *q = bdev_get_queue(dev->bdev);
1731
1732         return q && !q->limits.max_write_zeroes_sectors;
1733 }
1734
1735 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1736 {
1737         struct dm_target *ti;
1738         unsigned i = 0;
1739
1740         while (i < dm_table_get_num_targets(t)) {
1741                 ti = dm_table_get_target(t, i++);
1742
1743                 if (!ti->num_write_zeroes_bios)
1744                         return false;
1745
1746                 if (!ti->type->iterate_devices ||
1747                     ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1748                         return false;
1749         }
1750
1751         return true;
1752 }
1753
1754 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1755                                       sector_t start, sector_t len, void *data)
1756 {
1757         struct request_queue *q = bdev_get_queue(dev->bdev);
1758
1759         return q && !blk_queue_discard(q);
1760 }
1761
1762 static bool dm_table_supports_discards(struct dm_table *t)
1763 {
1764         struct dm_target *ti;
1765         unsigned i;
1766
1767         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1768                 ti = dm_table_get_target(t, i);
1769
1770                 if (!ti->num_discard_bios)
1771                         return false;
1772
1773                 /*
1774                  * Either the target provides discard support (as implied by setting
1775                  * 'discards_supported') or it relies on _all_ data devices having
1776                  * discard support.
1777                  */
1778                 if (!ti->discards_supported &&
1779                     (!ti->type->iterate_devices ||
1780                      ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1781                         return false;
1782         }
1783
1784         return true;
1785 }
1786
1787 static int device_requires_stable_pages(struct dm_target *ti,
1788                                         struct dm_dev *dev, sector_t start,
1789                                         sector_t len, void *data)
1790 {
1791         struct request_queue *q = bdev_get_queue(dev->bdev);
1792
1793         return q && bdi_cap_stable_pages_required(q->backing_dev_info);
1794 }
1795
1796 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1797                                struct queue_limits *limits)
1798 {
1799         bool wc = false, fua = false;
1800
1801         /*
1802          * Copy table's limits to the DM device's request_queue
1803          */
1804         q->limits = *limits;
1805
1806         if (!dm_table_supports_discards(t))
1807                 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1808         else
1809                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1810
1811         if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1812                 wc = true;
1813                 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1814                         fua = true;
1815         }
1816         blk_queue_write_cache(q, wc, fua);
1817
1818         if (dm_table_supports_dax(t))
1819                 queue_flag_set_unlocked(QUEUE_FLAG_DAX, q);
1820         else
1821                 queue_flag_clear_unlocked(QUEUE_FLAG_DAX, q);
1822
1823         if (dm_table_any_dev_attr(t, device_dax_write_cache_enabled, NULL))
1824                 dax_write_cache(t->md->dax_dev, true);
1825
1826         /* Ensure that all underlying devices are non-rotational. */
1827         if (dm_table_any_dev_attr(t, device_is_rotational, NULL))
1828                 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1829         else
1830                 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1831
1832         if (!dm_table_supports_write_same(t))
1833                 q->limits.max_write_same_sectors = 0;
1834         if (!dm_table_supports_write_zeroes(t))
1835                 q->limits.max_write_zeroes_sectors = 0;
1836
1837         if (dm_table_any_dev_attr(t, queue_no_sg_merge, NULL))
1838                 queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1839         else
1840                 queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1841
1842         dm_table_verify_integrity(t);
1843
1844         /*
1845          * Some devices don't use blk_integrity but still want stable pages
1846          * because they do their own checksumming.
1847          * If any underlying device requires stable pages, a table must require
1848          * them as well.  Only targets that support iterate_devices are considered:
1849          * don't want error, zero, etc to require stable pages.
1850          */
1851         if (dm_table_any_dev_attr(t, device_requires_stable_pages, NULL))
1852                 q->backing_dev_info->capabilities |= BDI_CAP_STABLE_WRITES;
1853         else
1854                 q->backing_dev_info->capabilities &= ~BDI_CAP_STABLE_WRITES;
1855
1856         /*
1857          * Determine whether or not this queue's I/O timings contribute
1858          * to the entropy pool, Only request-based targets use this.
1859          * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1860          * have it set.
1861          */
1862         if (blk_queue_add_random(q) &&
1863             dm_table_any_dev_attr(t, device_is_not_random, NULL))
1864                 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1865
1866         /*
1867          * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1868          * visible to other CPUs because, once the flag is set, incoming bios
1869          * are processed by request-based dm, which refers to the queue
1870          * settings.
1871          * Until the flag set, bios are passed to bio-based dm and queued to
1872          * md->deferred where queue settings are not needed yet.
1873          * Those bios are passed to request-based dm at the resume time.
1874          */
1875         smp_mb();
1876         if (dm_table_request_based(t))
1877                 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1878 }
1879
1880 unsigned int dm_table_get_num_targets(struct dm_table *t)
1881 {
1882         return t->num_targets;
1883 }
1884
1885 struct list_head *dm_table_get_devices(struct dm_table *t)
1886 {
1887         return &t->devices;
1888 }
1889
1890 fmode_t dm_table_get_mode(struct dm_table *t)
1891 {
1892         return t->mode;
1893 }
1894 EXPORT_SYMBOL(dm_table_get_mode);
1895
1896 enum suspend_mode {
1897         PRESUSPEND,
1898         PRESUSPEND_UNDO,
1899         POSTSUSPEND,
1900 };
1901
1902 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1903 {
1904         int i = t->num_targets;
1905         struct dm_target *ti = t->targets;
1906
1907         lockdep_assert_held(&t->md->suspend_lock);
1908
1909         while (i--) {
1910                 switch (mode) {
1911                 case PRESUSPEND:
1912                         if (ti->type->presuspend)
1913                                 ti->type->presuspend(ti);
1914                         break;
1915                 case PRESUSPEND_UNDO:
1916                         if (ti->type->presuspend_undo)
1917                                 ti->type->presuspend_undo(ti);
1918                         break;
1919                 case POSTSUSPEND:
1920                         if (ti->type->postsuspend)
1921                                 ti->type->postsuspend(ti);
1922                         break;
1923                 }
1924                 ti++;
1925         }
1926 }
1927
1928 void dm_table_presuspend_targets(struct dm_table *t)
1929 {
1930         if (!t)
1931                 return;
1932
1933         suspend_targets(t, PRESUSPEND);
1934 }
1935
1936 void dm_table_presuspend_undo_targets(struct dm_table *t)
1937 {
1938         if (!t)
1939                 return;
1940
1941         suspend_targets(t, PRESUSPEND_UNDO);
1942 }
1943
1944 void dm_table_postsuspend_targets(struct dm_table *t)
1945 {
1946         if (!t)
1947                 return;
1948
1949         suspend_targets(t, POSTSUSPEND);
1950 }
1951
1952 int dm_table_resume_targets(struct dm_table *t)
1953 {
1954         int i, r = 0;
1955
1956         lockdep_assert_held(&t->md->suspend_lock);
1957
1958         for (i = 0; i < t->num_targets; i++) {
1959                 struct dm_target *ti = t->targets + i;
1960
1961                 if (!ti->type->preresume)
1962                         continue;
1963
1964                 r = ti->type->preresume(ti);
1965                 if (r) {
1966                         DMERR("%s: %s: preresume failed, error = %d",
1967                               dm_device_name(t->md), ti->type->name, r);
1968                         return r;
1969                 }
1970         }
1971
1972         for (i = 0; i < t->num_targets; i++) {
1973                 struct dm_target *ti = t->targets + i;
1974
1975                 if (ti->type->resume)
1976                         ti->type->resume(ti);
1977         }
1978
1979         return 0;
1980 }
1981
1982 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1983 {
1984         list_add(&cb->list, &t->target_callbacks);
1985 }
1986 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1987
1988 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1989 {
1990         struct dm_dev_internal *dd;
1991         struct list_head *devices = dm_table_get_devices(t);
1992         struct dm_target_callbacks *cb;
1993         int r = 0;
1994
1995         list_for_each_entry(dd, devices, list) {
1996                 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1997                 char b[BDEVNAME_SIZE];
1998
1999                 if (likely(q))
2000                         r |= bdi_congested(q->backing_dev_info, bdi_bits);
2001                 else
2002                         DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
2003                                      dm_device_name(t->md),
2004                                      bdevname(dd->dm_dev->bdev, b));
2005         }
2006
2007         list_for_each_entry(cb, &t->target_callbacks, list)
2008                 if (cb->congested_fn)
2009                         r |= cb->congested_fn(cb, bdi_bits);
2010
2011         return r;
2012 }
2013
2014 struct mapped_device *dm_table_get_md(struct dm_table *t)
2015 {
2016         return t->md;
2017 }
2018 EXPORT_SYMBOL(dm_table_get_md);
2019
2020 void dm_table_run_md_queue_async(struct dm_table *t)
2021 {
2022         struct mapped_device *md;
2023         struct request_queue *queue;
2024         unsigned long flags;
2025
2026         if (!dm_table_request_based(t))
2027                 return;
2028
2029         md = dm_table_get_md(t);
2030         queue = dm_get_md_queue(md);
2031         if (queue) {
2032                 if (queue->mq_ops)
2033                         blk_mq_run_hw_queues(queue, true);
2034                 else {
2035                         spin_lock_irqsave(queue->queue_lock, flags);
2036                         blk_run_queue_async(queue);
2037                         spin_unlock_irqrestore(queue->queue_lock, flags);
2038                 }
2039         }
2040 }
2041 EXPORT_SYMBOL(dm_table_run_md_queue_async);
2042