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         unsigned integrity_added:1;
  51
  52         /*
  53          * Indicates the rw permissions for the new logical
  54          * device.  This should be a combination of FMODE_READ
  55          * and FMODE_WRITE.
  56          */
  57         fmode_t mode;
  58
  59         /* a list of devices used by this table */
  60         struct list_head devices;
  61
  62         /* events get handed up using this callback */
  63         void (*event_fn)(void *);
  64         void *event_context;
  65
  66         struct dm_md_mempools *mempools;
  67
  68         struct list_head target_callbacks;
  69 };
  70
  71 /*
  72  * Similar to ceiling(log_size(n))
  73  */
  74 static unsigned int int_log(unsigned int n, unsigned int base)
  75 {
  76         int result = 0;
  77
  78         while (n > 1) {
  79                 n = dm_div_up(n, base);
  80                 result++;
  81         }
  82
  83         return result;
  84 }
  85
  86 /*
  87  * Calculate the index of the child node of the n'th node k'th key.
  88  */
  89 static inline unsigned int get_child(unsigned int n, unsigned int k)
  90 {
  91         return (n * CHILDREN_PER_NODE) + k;
  92 }
  93
  94 /*
  95  * Return the n'th node of level l from table t.
  96  */
  97 static inline sector_t *get_node(struct dm_table *t,
  98                                  unsigned int l, unsigned int n)
  99 {
 100         return t->index[l] + (n * KEYS_PER_NODE);
 101 }
 102
 103 /*
 104  * Return the highest key that you could lookup from the n'th
 105  * node on level l of the btree.
 106  */
 107 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
 108 {
 109         for (; l < t->depth - 1; l++)
 110                 n = get_child(n, CHILDREN_PER_NODE - 1);
 111
 112         if (n >= t->counts[l])
 113                 return (sector_t) - 1;
 114
 115         return get_node(t, l, n)[KEYS_PER_NODE - 1];
 116 }
 117
 118 /*
 119  * Fills in a level of the btree based on the highs of the level
 120  * below it.
 121  */
 122 static int setup_btree_index(unsigned int l, struct dm_table *t)
 123 {
 124         unsigned int n, k;
 125         sector_t *node;
 126
 127         for (n = 0U; n < t->counts[l]; n++) {
 128                 node = get_node(t, l, n);
 129
 130                 for (k = 0U; k < KEYS_PER_NODE; k++)
 131                         node[k] = high(t, l + 1, get_child(n, k));
 132         }
 133
 134         return 0;
 135 }
 136
 137 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
 138 {
 139         unsigned long size;
 140         void *addr;
 141
 142         /*
 143          * Check that we're not going to overflow.
 144          */
 145         if (nmemb > (ULONG_MAX / elem_size))
 146                 return NULL;
 147
 148         size = nmemb * elem_size;
 149         addr = vzalloc(size);
 150
 151         return addr;
 152 }
 153 EXPORT_SYMBOL(dm_vcalloc);
 154
 155 /*
 156  * highs, and targets are managed as dynamic arrays during a
 157  * table load.
 158  */
 159 static int alloc_targets(struct dm_table *t, unsigned int num)
 160 {
 161         sector_t *n_highs;
 162         struct dm_target *n_targets;
 163
 164         /*
 165          * Allocate both the target array and offset array at once.
 166          */
 167         n_highs = (sector_t *) dm_vcalloc(num, sizeof(struct dm_target) +
 168                                           sizeof(sector_t));
 169         if (!n_highs)
 170                 return -ENOMEM;
 171
 172         n_targets = (struct dm_target *) (n_highs + num);
 173
 174         memset(n_highs, -1, sizeof(*n_highs) * num);
 175         vfree(t->highs);
 176
 177         t->num_allocated = num;
 178         t->highs = n_highs;
 179         t->targets = n_targets;
 180
 181         return 0;
 182 }
 183
 184 int dm_table_create(struct dm_table **result, fmode_t mode,
 185                     unsigned num_targets, struct mapped_device *md)
 186 {
 187         struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
 188
 189         if (!t)
 190                 return -ENOMEM;
 191
 192         INIT_LIST_HEAD(&t->devices);
 193         INIT_LIST_HEAD(&t->target_callbacks);
 194
 195         if (!num_targets)
 196                 num_targets = KEYS_PER_NODE;
 197
 198         num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
 199
 200         if (!num_targets) {
 201                 kfree(t);
 202                 return -ENOMEM;
 203         }
 204
 205         if (alloc_targets(t, num_targets)) {
 206                 kfree(t);
 207                 return -ENOMEM;
 208         }
 209
 210         t->type = DM_TYPE_NONE;
 211         t->mode = mode;
 212         t->md = md;
 213         *result = t;
 214         return 0;
 215 }
 216
 217 static void free_devices(struct list_head *devices, struct mapped_device *md)
 218 {
 219         struct list_head *tmp, *next;
 220
 221         list_for_each_safe(tmp, next, devices) {
 222                 struct dm_dev_internal *dd =
 223                     list_entry(tmp, struct dm_dev_internal, list);
 224                 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
 225                        dm_device_name(md), dd->dm_dev->name);
 226                 dm_put_table_device(md, dd->dm_dev);
 227                 kfree(dd);
 228         }
 229 }
 230
 231 void dm_table_destroy(struct dm_table *t)
 232 {
 233         unsigned int i;
 234
 235         if (!t)
 236                 return;
 237
 238         /* free the indexes */
 239         if (t->depth >= 2)
 240                 vfree(t->index[t->depth - 2]);
 241
 242         /* free the targets */
 243         for (i = 0; i < t->num_targets; i++) {
 244                 struct dm_target *tgt = t->targets + i;
 245
 246                 if (tgt->type->dtr)
 247                         tgt->type->dtr(tgt);
 248
 249                 dm_put_target_type(tgt->type);
 250         }
 251
 252         vfree(t->highs);
 253
 254         /* free the device list */
 255         free_devices(&t->devices, t->md);
 256
 257         dm_free_md_mempools(t->mempools);
 258
 259         kfree(t);
 260 }
 261
 262 /*
 263  * See if we've already got a device in the list.
 264  */
 265 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
 266 {
 267         struct dm_dev_internal *dd;
 268
 269         list_for_each_entry (dd, l, list)
 270                 if (dd->dm_dev->bdev->bd_dev == dev)
 271                         return dd;
 272
 273         return NULL;
 274 }
 275
 276 /*
 277  * If possible, this checks an area of a destination device is invalid.
 278  */
 279 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
 280                                   sector_t start, sector_t len, void *data)
 281 {
 282         struct request_queue *q;
 283         struct queue_limits *limits = data;
 284         struct block_device *bdev = dev->bdev;
 285         sector_t dev_size =
 286                 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
 287         unsigned short logical_block_size_sectors =
 288                 limits->logical_block_size >> SECTOR_SHIFT;
 289         char b[BDEVNAME_SIZE];
 290
 291         /*
 292          * Some devices exist without request functions,
 293          * such as loop devices not yet bound to backing files.
 294          * Forbid the use of such devices.
 295          */
 296         q = bdev_get_queue(bdev);
 297         if (!q || !q->make_request_fn) {
 298                 DMWARN("%s: %s is not yet initialised: "
 299                        "start=%llu, len=%llu, dev_size=%llu",
 300                        dm_device_name(ti->table->md), bdevname(bdev, b),
 301                        (unsigned long long)start,
 302                        (unsigned long long)len,
 303                        (unsigned long long)dev_size);
 304                 return 1;
 305         }
 306
 307         if (!dev_size)
 308                 return 0;
 309
 310         if ((start >= dev_size) || (start + len > dev_size)) {
 311                 DMWARN("%s: %s too small for target: "
 312                        "start=%llu, len=%llu, dev_size=%llu",
 313                        dm_device_name(ti->table->md), bdevname(bdev, b),
 314                        (unsigned long long)start,
 315                        (unsigned long long)len,
 316                        (unsigned long long)dev_size);
 317                 return 1;
 318         }
 319
 320         /*
 321          * If the target is mapped to zoned block device(s), check
 322          * that the zones are not partially mapped.
 323          */
 324         if (bdev_zoned_model(bdev) != BLK_ZONED_NONE) {
 325                 unsigned int zone_sectors = bdev_zone_sectors(bdev);
 326
 327                 if (start & (zone_sectors - 1)) {
 328                         DMWARN("%s: start=%llu not aligned to h/w zone size %u of %s",
 329                                dm_device_name(ti->table->md),
 330                                (unsigned long long)start,
 331                                zone_sectors, bdevname(bdev, b));
 332                         return 1;
 333                 }
 334
 335                 /*
 336                  * Note: The last zone of a zoned block device may be smaller
 337                  * than other zones. So for a target mapping the end of a
 338                  * zoned block device with such a zone, len would not be zone
 339                  * aligned. We do not allow such last smaller zone to be part
 340                  * of the mapping here to ensure that mappings with multiple
 341                  * devices do not end up with a smaller zone in the middle of
 342                  * the sector range.
 343                  */
 344                 if (len & (zone_sectors - 1)) {
 345                         DMWARN("%s: len=%llu not aligned to h/w zone size %u of %s",
 346                                dm_device_name(ti->table->md),
 347                                (unsigned long long)len,
 348                                zone_sectors, bdevname(bdev, b));
 349                         return 1;
 350                 }
 351         }
 352
 353         if (logical_block_size_sectors <= 1)
 354                 return 0;
 355
 356         if (start & (logical_block_size_sectors - 1)) {
 357                 DMWARN("%s: start=%llu not aligned to h/w "
 358                        "logical block size %u of %s",
 359                        dm_device_name(ti->table->md),
 360                        (unsigned long long)start,
 361                        limits->logical_block_size, bdevname(bdev, b));
 362                 return 1;
 363         }
 364
 365         if (len & (logical_block_size_sectors - 1)) {
 366                 DMWARN("%s: len=%llu not aligned to h/w "
 367                        "logical block size %u of %s",
 368                        dm_device_name(ti->table->md),
 369                        (unsigned long long)len,
 370                        limits->logical_block_size, bdevname(bdev, b));
 371                 return 1;
 372         }
 373
 374         return 0;
 375 }
 376
 377 /*
 378  * This upgrades the mode on an already open dm_dev, being
 379  * careful to leave things as they were if we fail to reopen the
 380  * device and not to touch the existing bdev field in case
 381  * it is accessed concurrently inside dm_table_any_congested().
 382  */
 383 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
 384                         struct mapped_device *md)
 385 {
 386         int r;
 387         struct dm_dev *old_dev, *new_dev;
 388
 389         old_dev = dd->dm_dev;
 390
 391         r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
 392                                 dd->dm_dev->mode | new_mode, &new_dev);
 393         if (r)
 394                 return r;
 395
 396         dd->dm_dev = new_dev;
 397         dm_put_table_device(md, old_dev);
 398
 399         return 0;
 400 }
 401
 402 /*
 403  * Convert the path to a device
 404  */
 405 dev_t dm_get_dev_t(const char *path)
 406 {
 407         dev_t dev;
 408         struct block_device *bdev;
 409
 410         bdev = lookup_bdev(path);
 411         if (IS_ERR(bdev))
 412                 dev = name_to_dev_t(path);
 413         else {
 414                 dev = bdev->bd_dev;
 415                 bdput(bdev);
 416         }
 417
 418         return dev;
 419 }
 420 EXPORT_SYMBOL_GPL(dm_get_dev_t);
 421
 422 /*
 423  * Add a device to the list, or just increment the usage count if
 424  * it's already present.
 425  */
 426 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
 427                   struct dm_dev **result)
 428 {
 429         int r;
 430         dev_t dev;
 431         unsigned int major, minor;
 432         char dummy;
 433         struct dm_dev_internal *dd;
 434         struct dm_table *t = ti->table;
 435
 436         BUG_ON(!t);
 437
 438         if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
 439                 /* Extract the major/minor numbers */
 440                 dev = MKDEV(major, minor);
 441                 if (MAJOR(dev) != major || MINOR(dev) != minor)
 442                         return -EOVERFLOW;
 443         } else {
 444                 dev = dm_get_dev_t(path);
 445                 if (!dev)
 446                         return -ENODEV;
 447         }
 448
 449         dd = find_device(&t->devices, dev);
 450         if (!dd) {
 451                 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
 452                 if (!dd)
 453                         return -ENOMEM;
 454
 455                 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
 456                         kfree(dd);
 457                         return r;
 458                 }
 459
 460                 refcount_set(&dd->count, 1);
 461                 list_add(&dd->list, &t->devices);
 462                 goto out;
 463
 464         } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
 465                 r = upgrade_mode(dd, mode, t->md);
 466                 if (r)
 467                         return r;
 468         }
 469         refcount_inc(&dd->count);
 470 out:
 471         *result = dd->dm_dev;
 472         return 0;
 473 }
 474 EXPORT_SYMBOL(dm_get_device);
 475
 476 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
 477                                 sector_t start, sector_t len, void *data)
 478 {
 479         struct queue_limits *limits = data;
 480         struct block_device *bdev = dev->bdev;
 481         struct request_queue *q = bdev_get_queue(bdev);
 482         char b[BDEVNAME_SIZE];
 483
 484         if (unlikely(!q)) {
 485                 DMWARN("%s: Cannot set limits for nonexistent device %s",
 486                        dm_device_name(ti->table->md), bdevname(bdev, b));
 487                 return 0;
 488         }
 489
 490         if (bdev_stack_limits(limits, bdev, start) < 0)
 491                 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
 492                        "physical_block_size=%u, logical_block_size=%u, "
 493                        "alignment_offset=%u, start=%llu",
 494                        dm_device_name(ti->table->md), bdevname(bdev, b),
 495                        q->limits.physical_block_size,
 496                        q->limits.logical_block_size,
 497                        q->limits.alignment_offset,
 498                        (unsigned long long) start << SECTOR_SHIFT);
 499
 500         limits->zoned = blk_queue_zoned_model(q);
 501
 502         return 0;
 503 }
 504
 505 /*
 506  * Decrement a device's use count and remove it if necessary.
 507  */
 508 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
 509 {
 510         int found = 0;
 511         struct list_head *devices = &ti->table->devices;
 512         struct dm_dev_internal *dd;
 513
 514         list_for_each_entry(dd, devices, list) {
 515                 if (dd->dm_dev == d) {
 516                         found = 1;
 517                         break;
 518                 }
 519         }
 520         if (!found) {
 521                 DMWARN("%s: device %s not in table devices list",
 522                        dm_device_name(ti->table->md), d->name);
 523                 return;
 524         }
 525         if (refcount_dec_and_test(&dd->count)) {
 526                 dm_put_table_device(ti->table->md, d);
 527                 list_del(&dd->list);
 528                 kfree(dd);
 529         }
 530 }
 531 EXPORT_SYMBOL(dm_put_device);
 532
 533 /*
 534  * Checks to see if the target joins onto the end of the table.
 535  */
 536 static int adjoin(struct dm_table *table, struct dm_target *ti)
 537 {
 538         struct dm_target *prev;
 539
 540         if (!table->num_targets)
 541                 return !ti->begin;
 542
 543         prev = &table->targets[table->num_targets - 1];
 544         return (ti->begin == (prev->begin + prev->len));
 545 }
 546
 547 /*
 548  * Used to dynamically allocate the arg array.
 549  *
 550  * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
 551  * process messages even if some device is suspended. These messages have a
 552  * small fixed number of arguments.
 553  *
 554  * On the other hand, dm-switch needs to process bulk data using messages and
 555  * excessive use of GFP_NOIO could cause trouble.
 556  */
 557 static char **realloc_argv(unsigned *size, char **old_argv)
 558 {
 559         char **argv;
 560         unsigned new_size;
 561         gfp_t gfp;
 562
 563         if (*size) {
 564                 new_size = *size * 2;
 565                 gfp = GFP_KERNEL;
 566         } else {
 567                 new_size = 8;
 568                 gfp = GFP_NOIO;
 569         }
 570         argv = kmalloc_array(new_size, sizeof(*argv), gfp);
 571         if (argv && old_argv) {
 572                 memcpy(argv, old_argv, *size * sizeof(*argv));
 573                 *size = new_size;
 574         }
 575
 576         kfree(old_argv);
 577         return argv;
 578 }
 579
 580 /*
 581  * Destructively splits up the argument list to pass to ctr.
 582  */
 583 int dm_split_args(int *argc, char ***argvp, char *input)
 584 {
 585         char *start, *end = input, *out, **argv = NULL;
 586         unsigned array_size = 0;
 587
 588         *argc = 0;
 589
 590         if (!input) {
 591                 *argvp = NULL;
 592                 return 0;
 593         }
 594
 595         argv = realloc_argv(&array_size, argv);
 596         if (!argv)
 597                 return -ENOMEM;
 598
 599         while (1) {
 600                 /* Skip whitespace */
 601                 start = skip_spaces(end);
 602
 603                 if (!*start)
 604                         break;  /* success, we hit the end */
 605
 606                 /* 'out' is used to remove any back-quotes */
 607                 end = out = start;
 608                 while (*end) {
 609                         /* Everything apart from '\0' can be quoted */
 610                         if (*end == '\\' && *(end + 1)) {
 611                                 *out++ = *(end + 1);
 612                                 end += 2;
 613                                 continue;
 614                         }
 615
 616                         if (isspace(*end))
 617                                 break;  /* end of token */
 618
 619                         *out++ = *end++;
 620                 }
 621
 622                 /* have we already filled the array ? */
 623                 if ((*argc + 1) > array_size) {
 624                         argv = realloc_argv(&array_size, argv);
 625                         if (!argv)
 626                                 return -ENOMEM;
 627                 }
 628
 629                 /* we know this is whitespace */
 630                 if (*end)
 631                         end++;
 632
 633                 /* terminate the string and put it in the array */
 634                 *out = '\0';
 635                 argv[*argc] = start;
 636                 (*argc)++;
 637         }
 638
 639         *argvp = argv;
 640         return 0;
 641 }
 642
 643 /*
 644  * Impose necessary and sufficient conditions on a devices's table such
 645  * that any incoming bio which respects its logical_block_size can be
 646  * processed successfully.  If it falls across the boundary between
 647  * two or more targets, the size of each piece it gets split into must
 648  * be compatible with the logical_block_size of the target processing it.
 649  */
 650 static int validate_hardware_logical_block_alignment(struct dm_table *table,
 651                                                  struct queue_limits *limits)
 652 {
 653         /*
 654          * This function uses arithmetic modulo the logical_block_size
 655          * (in units of 512-byte sectors).
 656          */
 657         unsigned short device_logical_block_size_sects =
 658                 limits->logical_block_size >> SECTOR_SHIFT;
 659
 660         /*
 661          * Offset of the start of the next table entry, mod logical_block_size.
 662          */
 663         unsigned short next_target_start = 0;
 664
 665         /*
 666          * Given an aligned bio that extends beyond the end of a
 667          * target, how many sectors must the next target handle?
 668          */
 669         unsigned short remaining = 0;
 670
 671         struct dm_target *uninitialized_var(ti);
 672         struct queue_limits ti_limits;
 673         unsigned i;
 674
 675         /*
 676          * Check each entry in the table in turn.
 677          */
 678         for (i = 0; i < dm_table_get_num_targets(table); i++) {
 679                 ti = dm_table_get_target(table, i);
 680
 681                 blk_set_stacking_limits(&ti_limits);
 682
 683                 /* combine all target devices' limits */
 684                 if (ti->type->iterate_devices)
 685                         ti->type->iterate_devices(ti, dm_set_device_limits,
 686                                                   &ti_limits);
 687
 688                 /*
 689                  * If the remaining sectors fall entirely within this
 690                  * table entry are they compatible with its logical_block_size?
 691                  */
 692                 if (remaining < ti->len &&
 693                     remaining & ((ti_limits.logical_block_size >>
 694                                   SECTOR_SHIFT) - 1))
 695                         break;  /* Error */
 696
 697                 next_target_start =
 698                     (unsigned short) ((next_target_start + ti->len) &
 699                                       (device_logical_block_size_sects - 1));
 700                 remaining = next_target_start ?
 701                     device_logical_block_size_sects - next_target_start : 0;
 702         }
 703
 704         if (remaining) {
 705                 DMWARN("%s: table line %u (start sect %llu len %llu) "
 706                        "not aligned to h/w logical block size %u",
 707                        dm_device_name(table->md), i,
 708                        (unsigned long long) ti->begin,
 709                        (unsigned long long) ti->len,
 710                        limits->logical_block_size);
 711                 return -EINVAL;
 712         }
 713
 714         return 0;
 715 }
 716
 717 int dm_table_add_target(struct dm_table *t, const char *type,
 718                         sector_t start, sector_t len, char *params)
 719 {
 720         int r = -EINVAL, argc;
 721         char **argv;
 722         struct dm_target *tgt;
 723
 724         if (t->singleton) {
 725                 DMERR("%s: target type %s must appear alone in table",
 726                       dm_device_name(t->md), t->targets->type->name);
 727                 return -EINVAL;
 728         }
 729
 730         BUG_ON(t->num_targets >= t->num_allocated);
 731
 732         tgt = t->targets + t->num_targets;
 733         memset(tgt, 0, sizeof(*tgt));
 734
 735         if (!len) {
 736                 DMERR("%s: zero-length target", dm_device_name(t->md));
 737                 return -EINVAL;
 738         }
 739
 740         tgt->type = dm_get_target_type(type);
 741         if (!tgt->type) {
 742                 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
 743                 return -EINVAL;
 744         }
 745
 746         if (dm_target_needs_singleton(tgt->type)) {
 747                 if (t->num_targets) {
 748                         tgt->error = "singleton target type must appear alone in table";
 749                         goto bad;
 750                 }
 751                 t->singleton = true;
 752         }
 753
 754         if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
 755                 tgt->error = "target type may not be included in a read-only table";
 756                 goto bad;
 757         }
 758
 759         if (t->immutable_target_type) {
 760                 if (t->immutable_target_type != tgt->type) {
 761                         tgt->error = "immutable target type cannot be mixed with other target types";
 762                         goto bad;
 763                 }
 764         } else if (dm_target_is_immutable(tgt->type)) {
 765                 if (t->num_targets) {
 766                         tgt->error = "immutable target type cannot be mixed with other target types";
 767                         goto bad;
 768                 }
 769                 t->immutable_target_type = tgt->type;
 770         }
 771
 772         if (dm_target_has_integrity(tgt->type))
 773                 t->integrity_added = 1;
 774
 775         tgt->table = t;
 776         tgt->begin = start;
 777         tgt->len = len;
 778         tgt->error = "Unknown error";
 779
 780         /*
 781          * Does this target adjoin the previous one ?
 782          */
 783         if (!adjoin(t, tgt)) {
 784                 tgt->error = "Gap in table";
 785                 goto bad;
 786         }
 787
 788         r = dm_split_args(&argc, &argv, params);
 789         if (r) {
 790                 tgt->error = "couldn't split parameters (insufficient memory)";
 791                 goto bad;
 792         }
 793
 794         r = tgt->type->ctr(tgt, argc, argv);
 795         kfree(argv);
 796         if (r)
 797                 goto bad;
 798
 799         t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
 800
 801         if (!tgt->num_discard_bios && tgt->discards_supported)
 802                 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
 803                        dm_device_name(t->md), type);
 804
 805         return 0;
 806
 807  bad:
 808         DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
 809         dm_put_target_type(tgt->type);
 810         return r;
 811 }
 812
 813 /*
 814  * Target argument parsing helpers.
 815  */
 816 static int validate_next_arg(const struct dm_arg *arg,
 817                              struct dm_arg_set *arg_set,
 818                              unsigned *value, char **error, unsigned grouped)
 819 {
 820         const char *arg_str = dm_shift_arg(arg_set);
 821         char dummy;
 822
 823         if (!arg_str ||
 824             (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
 825             (*value < arg->min) ||
 826             (*value > arg->max) ||
 827             (grouped && arg_set->argc < *value)) {
 828                 *error = arg->error;
 829                 return -EINVAL;
 830         }
 831
 832         return 0;
 833 }
 834
 835 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 836                 unsigned *value, char **error)
 837 {
 838         return validate_next_arg(arg, arg_set, value, error, 0);
 839 }
 840 EXPORT_SYMBOL(dm_read_arg);
 841
 842 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 843                       unsigned *value, char **error)
 844 {
 845         return validate_next_arg(arg, arg_set, value, error, 1);
 846 }
 847 EXPORT_SYMBOL(dm_read_arg_group);
 848
 849 const char *dm_shift_arg(struct dm_arg_set *as)
 850 {
 851         char *r;
 852
 853         if (as->argc) {
 854                 as->argc--;
 855                 r = *as->argv;
 856                 as->argv++;
 857                 return r;
 858         }
 859
 860         return NULL;
 861 }
 862 EXPORT_SYMBOL(dm_shift_arg);
 863
 864 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
 865 {
 866         BUG_ON(as->argc < num_args);
 867         as->argc -= num_args;
 868         as->argv += num_args;
 869 }
 870 EXPORT_SYMBOL(dm_consume_args);
 871
 872 static bool __table_type_bio_based(enum dm_queue_mode table_type)
 873 {
 874         return (table_type == DM_TYPE_BIO_BASED ||
 875                 table_type == DM_TYPE_DAX_BIO_BASED);
 876 }
 877
 878 static bool __table_type_request_based(enum dm_queue_mode table_type)
 879 {
 880         return table_type == DM_TYPE_REQUEST_BASED;
 881 }
 882
 883 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
 884 {
 885         t->type = type;
 886 }
 887 EXPORT_SYMBOL_GPL(dm_table_set_type);
 888
 889 /* validate the dax capability of the target device span */
 890 int device_not_dax_capable(struct dm_target *ti, struct dm_dev *dev,
 891                         sector_t start, sector_t len, void *data)
 892 {
 893         int blocksize = *(int *) data, id;
 894         bool rc;
 895
 896         id = dax_read_lock();
 897         rc = !dax_supported(dev->dax_dev, dev->bdev, blocksize, start, len);
 898         dax_read_unlock(id);
 899
 900         return rc;
 901 }
 902
 903 /* Check devices support synchronous DAX */
 904 static int device_not_dax_synchronous_capable(struct dm_target *ti, struct dm_dev *dev,
 905                                               sector_t start, sector_t len, void *data)
 906 {
 907         return !dev->dax_dev || !dax_synchronous(dev->dax_dev);
 908 }
 909
 910 bool dm_table_supports_dax(struct dm_table *t,
 911                            iterate_devices_callout_fn iterate_fn, int *blocksize)
 912 {
 913         struct dm_target *ti;
 914         unsigned i;
 915
 916         /* Ensure that all targets support DAX. */
 917         for (i = 0; i < dm_table_get_num_targets(t); i++) {
 918                 ti = dm_table_get_target(t, i);
 919
 920                 if (!ti->type->direct_access)
 921                         return false;
 922
 923                 if (!ti->type->iterate_devices ||
 924                     ti->type->iterate_devices(ti, iterate_fn, blocksize))
 925                         return false;
 926         }
 927
 928         return true;
 929 }
 930
 931 static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev,
 932                                   sector_t start, sector_t len, void *data)
 933 {
 934         struct block_device *bdev = dev->bdev;
 935         struct request_queue *q = bdev_get_queue(bdev);
 936
 937         /* request-based cannot stack on partitions! */
 938         if (bdev != bdev->bd_contains)
 939                 return false;
 940
 941         return queue_is_mq(q);
 942 }
 943
 944 static int dm_table_determine_type(struct dm_table *t)
 945 {
 946         unsigned i;
 947         unsigned bio_based = 0, request_based = 0, hybrid = 0;
 948         struct dm_target *tgt;
 949         struct list_head *devices = dm_table_get_devices(t);
 950         enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
 951         int page_size = PAGE_SIZE;
 952
 953         if (t->type != DM_TYPE_NONE) {
 954                 /* target already set the table's type */
 955                 if (t->type == DM_TYPE_BIO_BASED) {
 956                         /* possibly upgrade to a variant of bio-based */
 957                         goto verify_bio_based;
 958                 }
 959                 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
 960                 goto verify_rq_based;
 961         }
 962
 963         for (i = 0; i < t->num_targets; i++) {
 964                 tgt = t->targets + i;
 965                 if (dm_target_hybrid(tgt))
 966                         hybrid = 1;
 967                 else if (dm_target_request_based(tgt))
 968                         request_based = 1;
 969                 else
 970                         bio_based = 1;
 971
 972                 if (bio_based && request_based) {
 973                         DMERR("Inconsistent table: different target types"
 974                               " can't be mixed up");
 975                         return -EINVAL;
 976                 }
 977         }
 978
 979         if (hybrid && !bio_based && !request_based) {
 980                 /*
 981                  * The targets can work either way.
 982                  * Determine the type from the live device.
 983                  * Default to bio-based if device is new.
 984                  */
 985                 if (__table_type_request_based(live_md_type))
 986                         request_based = 1;
 987                 else
 988                         bio_based = 1;
 989         }
 990
 991         if (bio_based) {
 992 verify_bio_based:
 993                 /* We must use this table as bio-based */
 994                 t->type = DM_TYPE_BIO_BASED;
 995                 if (dm_table_supports_dax(t, device_not_dax_capable, &page_size) ||
 996                     (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
 997                         t->type = DM_TYPE_DAX_BIO_BASED;
 998                 }
 999                 return 0;
1000         }
1001
1002         BUG_ON(!request_based); /* No targets in this table */
1003
1004         t->type = DM_TYPE_REQUEST_BASED;
1005
1006 verify_rq_based:
1007         /*
1008          * Request-based dm supports only tables that have a single target now.
1009          * To support multiple targets, request splitting support is needed,
1010          * and that needs lots of changes in the block-layer.
1011          * (e.g. request completion process for partial completion.)
1012          */
1013         if (t->num_targets > 1) {
1014                 DMERR("request-based DM doesn't support multiple targets");
1015                 return -EINVAL;
1016         }
1017
1018         if (list_empty(devices)) {
1019                 int srcu_idx;
1020                 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
1021
1022                 /* inherit live table's type */
1023                 if (live_table)
1024                         t->type = live_table->type;
1025                 dm_put_live_table(t->md, srcu_idx);
1026                 return 0;
1027         }
1028
1029         tgt = dm_table_get_immutable_target(t);
1030         if (!tgt) {
1031                 DMERR("table load rejected: immutable target is required");
1032                 return -EINVAL;
1033         } else if (tgt->max_io_len) {
1034                 DMERR("table load rejected: immutable target that splits IO is not supported");
1035                 return -EINVAL;
1036         }
1037
1038         /* Non-request-stackable devices can't be used for request-based dm */
1039         if (!tgt->type->iterate_devices ||
1040             !tgt->type->iterate_devices(tgt, device_is_rq_stackable, NULL)) {
1041                 DMERR("table load rejected: including non-request-stackable devices");
1042                 return -EINVAL;
1043         }
1044
1045         return 0;
1046 }
1047
1048 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
1049 {
1050         return t->type;
1051 }
1052
1053 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
1054 {
1055         return t->immutable_target_type;
1056 }
1057
1058 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
1059 {
1060         /* Immutable target is implicitly a singleton */
1061         if (t->num_targets > 1 ||
1062             !dm_target_is_immutable(t->targets[0].type))
1063                 return NULL;
1064
1065         return t->targets;
1066 }
1067
1068 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1069 {
1070         struct dm_target *ti;
1071         unsigned i;
1072
1073         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1074                 ti = dm_table_get_target(t, i);
1075                 if (dm_target_is_wildcard(ti->type))
1076                         return ti;
1077         }
1078
1079         return NULL;
1080 }
1081
1082 bool dm_table_bio_based(struct dm_table *t)
1083 {
1084         return __table_type_bio_based(dm_table_get_type(t));
1085 }
1086
1087 bool dm_table_request_based(struct dm_table *t)
1088 {
1089         return __table_type_request_based(dm_table_get_type(t));
1090 }
1091
1092 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1093 {
1094         enum dm_queue_mode type = dm_table_get_type(t);
1095         unsigned per_io_data_size = 0;
1096         unsigned min_pool_size = 0;
1097         struct dm_target *ti;
1098         unsigned i;
1099
1100         if (unlikely(type == DM_TYPE_NONE)) {
1101                 DMWARN("no table type is set, can't allocate mempools");
1102                 return -EINVAL;
1103         }
1104
1105         if (__table_type_bio_based(type))
1106                 for (i = 0; i < t->num_targets; i++) {
1107                         ti = t->targets + i;
1108                         per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1109                         min_pool_size = max(min_pool_size, ti->num_flush_bios);
1110                 }
1111
1112         t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported,
1113                                            per_io_data_size, min_pool_size);
1114         if (!t->mempools)
1115                 return -ENOMEM;
1116
1117         return 0;
1118 }
1119
1120 void dm_table_free_md_mempools(struct dm_table *t)
1121 {
1122         dm_free_md_mempools(t->mempools);
1123         t->mempools = NULL;
1124 }
1125
1126 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1127 {
1128         return t->mempools;
1129 }
1130
1131 static int setup_indexes(struct dm_table *t)
1132 {
1133         int i;
1134         unsigned int total = 0;
1135         sector_t *indexes;
1136
1137         /* allocate the space for *all* the indexes */
1138         for (i = t->depth - 2; i >= 0; i--) {
1139                 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1140                 total += t->counts[i];
1141         }
1142
1143         indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1144         if (!indexes)
1145                 return -ENOMEM;
1146
1147         /* set up internal nodes, bottom-up */
1148         for (i = t->depth - 2; i >= 0; i--) {
1149                 t->index[i] = indexes;
1150                 indexes += (KEYS_PER_NODE * t->counts[i]);
1151                 setup_btree_index(i, t);
1152         }
1153
1154         return 0;
1155 }
1156
1157 /*
1158  * Builds the btree to index the map.
1159  */
1160 static int dm_table_build_index(struct dm_table *t)
1161 {
1162         int r = 0;
1163         unsigned int leaf_nodes;
1164
1165         /* how many indexes will the btree have ? */
1166         leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1167         t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1168
1169         /* leaf layer has already been set up */
1170         t->counts[t->depth - 1] = leaf_nodes;
1171         t->index[t->depth - 1] = t->highs;
1172
1173         if (t->depth >= 2)
1174                 r = setup_indexes(t);
1175
1176         return r;
1177 }
1178
1179 static bool integrity_profile_exists(struct gendisk *disk)
1180 {
1181         return !!blk_get_integrity(disk);
1182 }
1183
1184 /*
1185  * Get a disk whose integrity profile reflects the table's profile.
1186  * Returns NULL if integrity support was inconsistent or unavailable.
1187  */
1188 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1189 {
1190         struct list_head *devices = dm_table_get_devices(t);
1191         struct dm_dev_internal *dd = NULL;
1192         struct gendisk *prev_disk = NULL, *template_disk = NULL;
1193         unsigned i;
1194
1195         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1196                 struct dm_target *ti = dm_table_get_target(t, i);
1197                 if (!dm_target_passes_integrity(ti->type))
1198                         goto no_integrity;
1199         }
1200
1201         list_for_each_entry(dd, devices, list) {
1202                 template_disk = dd->dm_dev->bdev->bd_disk;
1203                 if (!integrity_profile_exists(template_disk))
1204                         goto no_integrity;
1205                 else if (prev_disk &&
1206                          blk_integrity_compare(prev_disk, template_disk) < 0)
1207                         goto no_integrity;
1208                 prev_disk = template_disk;
1209         }
1210
1211         return template_disk;
1212
1213 no_integrity:
1214         if (prev_disk)
1215                 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1216                        dm_device_name(t->md),
1217                        prev_disk->disk_name,
1218                        template_disk->disk_name);
1219         return NULL;
1220 }
1221
1222 /*
1223  * Register the mapped device for blk_integrity support if the
1224  * underlying devices have an integrity profile.  But all devices may
1225  * not have matching profiles (checking all devices isn't reliable
1226  * during table load because this table may use other DM device(s) which
1227  * must be resumed before they will have an initialized integity
1228  * profile).  Consequently, stacked DM devices force a 2 stage integrity
1229  * profile validation: First pass during table load, final pass during
1230  * resume.
1231  */
1232 static int dm_table_register_integrity(struct dm_table *t)
1233 {
1234         struct mapped_device *md = t->md;
1235         struct gendisk *template_disk = NULL;
1236
1237         /* If target handles integrity itself do not register it here. */
1238         if (t->integrity_added)
1239                 return 0;
1240
1241         template_disk = dm_table_get_integrity_disk(t);
1242         if (!template_disk)
1243                 return 0;
1244
1245         if (!integrity_profile_exists(dm_disk(md))) {
1246                 t->integrity_supported = true;
1247                 /*
1248                  * Register integrity profile during table load; we can do
1249                  * this because the final profile must match during resume.
1250                  */
1251                 blk_integrity_register(dm_disk(md),
1252                                        blk_get_integrity(template_disk));
1253                 return 0;
1254         }
1255
1256         /*
1257          * If DM device already has an initialized integrity
1258          * profile the new profile should not conflict.
1259          */
1260         if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1261                 DMWARN("%s: conflict with existing integrity profile: "
1262                        "%s profile mismatch",
1263                        dm_device_name(t->md),
1264                        template_disk->disk_name);
1265                 return 1;
1266         }
1267
1268         /* Preserve existing integrity profile */
1269         t->integrity_supported = true;
1270         return 0;
1271 }
1272
1273 /*
1274  * Prepares the table for use by building the indices,
1275  * setting the type, and allocating mempools.
1276  */
1277 int dm_table_complete(struct dm_table *t)
1278 {
1279         int r;
1280
1281         r = dm_table_determine_type(t);
1282         if (r) {
1283                 DMERR("unable to determine table type");
1284                 return r;
1285         }
1286
1287         r = dm_table_build_index(t);
1288         if (r) {
1289                 DMERR("unable to build btrees");
1290                 return r;
1291         }
1292
1293         r = dm_table_register_integrity(t);
1294         if (r) {
1295                 DMERR("could not register integrity profile.");
1296                 return r;
1297         }
1298
1299         r = dm_table_alloc_md_mempools(t, t->md);
1300         if (r)
1301                 DMERR("unable to allocate mempools");
1302
1303         return r;
1304 }
1305
1306 static DEFINE_MUTEX(_event_lock);
1307 void dm_table_event_callback(struct dm_table *t,
1308                              void (*fn)(void *), void *context)
1309 {
1310         mutex_lock(&_event_lock);
1311         t->event_fn = fn;
1312         t->event_context = context;
1313         mutex_unlock(&_event_lock);
1314 }
1315
1316 void dm_table_event(struct dm_table *t)
1317 {
1318         mutex_lock(&_event_lock);
1319         if (t->event_fn)
1320                 t->event_fn(t->event_context);
1321         mutex_unlock(&_event_lock);
1322 }
1323 EXPORT_SYMBOL(dm_table_event);
1324
1325 inline sector_t dm_table_get_size(struct dm_table *t)
1326 {
1327         return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1328 }
1329 EXPORT_SYMBOL(dm_table_get_size);
1330
1331 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1332 {
1333         if (index >= t->num_targets)
1334                 return NULL;
1335
1336         return t->targets + index;
1337 }
1338
1339 /*
1340  * Search the btree for the correct target.
1341  *
1342  * Caller should check returned pointer for NULL
1343  * to trap I/O beyond end of device.
1344  */
1345 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1346 {
1347         unsigned int l, n = 0, k = 0;
1348         sector_t *node;
1349
1350         if (unlikely(sector >= dm_table_get_size(t)))
1351                 return NULL;
1352
1353         for (l = 0; l < t->depth; l++) {
1354                 n = get_child(n, k);
1355                 node = get_node(t, l, n);
1356
1357                 for (k = 0; k < KEYS_PER_NODE; k++)
1358                         if (node[k] >= sector)
1359                                 break;
1360         }
1361
1362         return &t->targets[(KEYS_PER_NODE * n) + k];
1363 }
1364
1365 /*
1366  * type->iterate_devices() should be called when the sanity check needs to
1367  * iterate and check all underlying data devices. iterate_devices() will
1368  * iterate all underlying data devices until it encounters a non-zero return
1369  * code, returned by whether the input iterate_devices_callout_fn, or
1370  * iterate_devices() itself internally.
1371  *
1372  * For some target type (e.g. dm-stripe), one call of iterate_devices() may
1373  * iterate multiple underlying devices internally, in which case a non-zero
1374  * return code returned by iterate_devices_callout_fn will stop the iteration
1375  * in advance.
1376  *
1377  * Cases requiring _any_ underlying device supporting some kind of attribute,
1378  * should use the iteration structure like dm_table_any_dev_attr(), or call
1379  * it directly. @func should handle semantics of positive examples, e.g.
1380  * capable of something.
1381  *
1382  * Cases requiring _all_ underlying devices supporting some kind of attribute,
1383  * should use the iteration structure like dm_table_supports_nowait() or
1384  * dm_table_supports_discards(). Or introduce dm_table_all_devs_attr() that
1385  * uses an @anti_func that handle semantics of counter examples, e.g. not
1386  * capable of something. So: return !dm_table_any_dev_attr(t, anti_func, data);
1387  */
1388 static bool dm_table_any_dev_attr(struct dm_table *t,
1389                                   iterate_devices_callout_fn func, void *data)
1390 {
1391         struct dm_target *ti;
1392         unsigned int i;
1393
1394         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1395                 ti = dm_table_get_target(t, i);
1396
1397                 if (ti->type->iterate_devices &&
1398                     ti->type->iterate_devices(ti, func, data))
1399                         return true;
1400         }
1401
1402         return false;
1403 }
1404
1405 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1406                         sector_t start, sector_t len, void *data)
1407 {
1408         unsigned *num_devices = data;
1409
1410         (*num_devices)++;
1411
1412         return 0;
1413 }
1414
1415 /*
1416  * Check whether a table has no data devices attached using each
1417  * target's iterate_devices method.
1418  * Returns false if the result is unknown because a target doesn't
1419  * support iterate_devices.
1420  */
1421 bool dm_table_has_no_data_devices(struct dm_table *table)
1422 {
1423         struct dm_target *ti;
1424         unsigned i, num_devices;
1425
1426         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1427                 ti = dm_table_get_target(table, i);
1428
1429                 if (!ti->type->iterate_devices)
1430                         return false;
1431
1432                 num_devices = 0;
1433                 ti->type->iterate_devices(ti, count_device, &num_devices);
1434                 if (num_devices)
1435                         return false;
1436         }
1437
1438         return true;
1439 }
1440
1441 static int device_not_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1442                                   sector_t start, sector_t len, void *data)
1443 {
1444         struct request_queue *q = bdev_get_queue(dev->bdev);
1445         enum blk_zoned_model *zoned_model = data;
1446
1447         return !q || blk_queue_zoned_model(q) != *zoned_model;
1448 }
1449
1450 static bool dm_table_supports_zoned_model(struct dm_table *t,
1451                                           enum blk_zoned_model zoned_model)
1452 {
1453         struct dm_target *ti;
1454         unsigned i;
1455
1456         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1457                 ti = dm_table_get_target(t, i);
1458
1459                 if (zoned_model == BLK_ZONED_HM &&
1460                     !dm_target_supports_zoned_hm(ti->type))
1461                         return false;
1462
1463                 if (!ti->type->iterate_devices ||
1464                     ti->type->iterate_devices(ti, device_not_zoned_model, &zoned_model))
1465                         return false;
1466         }
1467
1468         return true;
1469 }
1470
1471 static int device_not_matches_zone_sectors(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         unsigned int *zone_sectors = data;
1476
1477         return !q || blk_queue_zone_sectors(q) != *zone_sectors;
1478 }
1479
1480 static int validate_hardware_zoned_model(struct dm_table *table,
1481                                          enum blk_zoned_model zoned_model,
1482                                          unsigned int zone_sectors)
1483 {
1484         if (zoned_model == BLK_ZONED_NONE)
1485                 return 0;
1486
1487         if (!dm_table_supports_zoned_model(table, zoned_model)) {
1488                 DMERR("%s: zoned model is not consistent across all devices",
1489                       dm_device_name(table->md));
1490                 return -EINVAL;
1491         }
1492
1493         /* Check zone size validity and compatibility */
1494         if (!zone_sectors || !is_power_of_2(zone_sectors))
1495                 return -EINVAL;
1496
1497         if (dm_table_any_dev_attr(table, device_not_matches_zone_sectors, &zone_sectors)) {
1498                 DMERR("%s: zone sectors is not consistent across all devices",
1499                       dm_device_name(table->md));
1500                 return -EINVAL;
1501         }
1502
1503         return 0;
1504 }
1505
1506 /*
1507  * Establish the new table's queue_limits and validate them.
1508  */
1509 int dm_calculate_queue_limits(struct dm_table *table,
1510                               struct queue_limits *limits)
1511 {
1512         struct dm_target *ti;
1513         struct queue_limits ti_limits;
1514         unsigned i;
1515         enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1516         unsigned int zone_sectors = 0;
1517
1518         blk_set_stacking_limits(limits);
1519
1520         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1521                 blk_set_stacking_limits(&ti_limits);
1522
1523                 ti = dm_table_get_target(table, i);
1524
1525                 if (!ti->type->iterate_devices)
1526                         goto combine_limits;
1527
1528                 /*
1529                  * Combine queue limits of all the devices this target uses.
1530                  */
1531                 ti->type->iterate_devices(ti, dm_set_device_limits,
1532                                           &ti_limits);
1533
1534                 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1535                         /*
1536                          * After stacking all limits, validate all devices
1537                          * in table support this zoned model and zone sectors.
1538                          */
1539                         zoned_model = ti_limits.zoned;
1540                         zone_sectors = ti_limits.chunk_sectors;
1541                 }
1542
1543                 /* Set I/O hints portion of queue limits */
1544                 if (ti->type->io_hints)
1545                         ti->type->io_hints(ti, &ti_limits);
1546
1547                 /*
1548                  * Check each device area is consistent with the target's
1549                  * overall queue limits.
1550                  */
1551                 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1552                                               &ti_limits))
1553                         return -EINVAL;
1554
1555 combine_limits:
1556                 /*
1557                  * Merge this target's queue limits into the overall limits
1558                  * for the table.
1559                  */
1560                 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1561                         DMWARN("%s: adding target device "
1562                                "(start sect %llu len %llu) "
1563                                "caused an alignment inconsistency",
1564                                dm_device_name(table->md),
1565                                (unsigned long long) ti->begin,
1566                                (unsigned long long) ti->len);
1567
1568                 /*
1569                  * FIXME: this should likely be moved to blk_stack_limits(), would
1570                  * also eliminate limits->zoned stacking hack in dm_set_device_limits()
1571                  */
1572                 if (limits->zoned == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1573                         /*
1574                          * By default, the stacked limits zoned model is set to
1575                          * BLK_ZONED_NONE in blk_set_stacking_limits(). Update
1576                          * this model using the first target model reported
1577                          * that is not BLK_ZONED_NONE. This will be either the
1578                          * first target device zoned model or the model reported
1579                          * by the target .io_hints.
1580                          */
1581                         limits->zoned = ti_limits.zoned;
1582                 }
1583         }
1584
1585         /*
1586          * Verify that the zoned model and zone sectors, as determined before
1587          * any .io_hints override, are the same across all devices in the table.
1588          * - this is especially relevant if .io_hints is emulating a disk-managed
1589          *   zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1590          * BUT...
1591          */
1592         if (limits->zoned != BLK_ZONED_NONE) {
1593                 /*
1594                  * ...IF the above limits stacking determined a zoned model
1595                  * validate that all of the table's devices conform to it.
1596                  */
1597                 zoned_model = limits->zoned;
1598                 zone_sectors = limits->chunk_sectors;
1599         }
1600         if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1601                 return -EINVAL;
1602
1603         return validate_hardware_logical_block_alignment(table, limits);
1604 }
1605
1606 /*
1607  * Verify that all devices have an integrity profile that matches the
1608  * DM device's registered integrity profile.  If the profiles don't
1609  * match then unregister the DM device's integrity profile.
1610  */
1611 static void dm_table_verify_integrity(struct dm_table *t)
1612 {
1613         struct gendisk *template_disk = NULL;
1614
1615         if (t->integrity_added)
1616                 return;
1617
1618         if (t->integrity_supported) {
1619                 /*
1620                  * Verify that the original integrity profile
1621                  * matches all the devices in this table.
1622                  */
1623                 template_disk = dm_table_get_integrity_disk(t);
1624                 if (template_disk &&
1625                     blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1626                         return;
1627         }
1628
1629         if (integrity_profile_exists(dm_disk(t->md))) {
1630                 DMWARN("%s: unable to establish an integrity profile",
1631                        dm_device_name(t->md));
1632                 blk_integrity_unregister(dm_disk(t->md));
1633         }
1634 }
1635
1636 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1637                                 sector_t start, sector_t len, void *data)
1638 {
1639         unsigned long flush = (unsigned long) data;
1640         struct request_queue *q = bdev_get_queue(dev->bdev);
1641
1642         return q && (q->queue_flags & flush);
1643 }
1644
1645 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1646 {
1647         struct dm_target *ti;
1648         unsigned i;
1649
1650         /*
1651          * Require at least one underlying device to support flushes.
1652          * t->devices includes internal dm devices such as mirror logs
1653          * so we need to use iterate_devices here, which targets
1654          * supporting flushes must provide.
1655          */
1656         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1657                 ti = dm_table_get_target(t, i);
1658
1659                 if (!ti->num_flush_bios)
1660                         continue;
1661
1662                 if (ti->flush_supported)
1663                         return true;
1664
1665                 if (ti->type->iterate_devices &&
1666                     ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1667                         return true;
1668         }
1669
1670         return false;
1671 }
1672
1673 static int device_dax_write_cache_enabled(struct dm_target *ti,
1674                                           struct dm_dev *dev, sector_t start,
1675                                           sector_t len, void *data)
1676 {
1677         struct dax_device *dax_dev = dev->dax_dev;
1678
1679         if (!dax_dev)
1680                 return false;
1681
1682         if (dax_write_cache_enabled(dax_dev))
1683                 return true;
1684         return false;
1685 }
1686
1687 static int device_is_rotational(struct dm_target *ti, struct dm_dev *dev,
1688                                 sector_t start, sector_t len, void *data)
1689 {
1690         struct request_queue *q = bdev_get_queue(dev->bdev);
1691
1692         return q && !blk_queue_nonrot(q);
1693 }
1694
1695 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1696                              sector_t start, sector_t len, void *data)
1697 {
1698         struct request_queue *q = bdev_get_queue(dev->bdev);
1699
1700         return q && !blk_queue_add_random(q);
1701 }
1702
1703 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1704                                          sector_t start, sector_t len, void *data)
1705 {
1706         struct request_queue *q = bdev_get_queue(dev->bdev);
1707
1708         return q && !q->limits.max_write_same_sectors;
1709 }
1710
1711 static bool dm_table_supports_write_same(struct dm_table *t)
1712 {
1713         struct dm_target *ti;
1714         unsigned i;
1715
1716         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1717                 ti = dm_table_get_target(t, i);
1718
1719                 if (!ti->num_write_same_bios)
1720                         return false;
1721
1722                 if (!ti->type->iterate_devices ||
1723                     ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1724                         return false;
1725         }
1726
1727         return true;
1728 }
1729
1730 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1731                                            sector_t start, sector_t len, void *data)
1732 {
1733         struct request_queue *q = bdev_get_queue(dev->bdev);
1734
1735         return q && !q->limits.max_write_zeroes_sectors;
1736 }
1737
1738 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1739 {
1740         struct dm_target *ti;
1741         unsigned i = 0;
1742
1743         while (i < dm_table_get_num_targets(t)) {
1744                 ti = dm_table_get_target(t, i++);
1745
1746                 if (!ti->num_write_zeroes_bios)
1747                         return false;
1748
1749                 if (!ti->type->iterate_devices ||
1750                     ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1751                         return false;
1752         }
1753
1754         return true;
1755 }
1756
1757 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1758                                       sector_t start, sector_t len, void *data)
1759 {
1760         struct request_queue *q = bdev_get_queue(dev->bdev);
1761
1762         return q && !blk_queue_discard(q);
1763 }
1764
1765 static bool dm_table_supports_discards(struct dm_table *t)
1766 {
1767         struct dm_target *ti;
1768         unsigned i;
1769
1770         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1771                 ti = dm_table_get_target(t, i);
1772
1773                 if (!ti->num_discard_bios)
1774                         return false;
1775
1776                 /*
1777                  * Either the target provides discard support (as implied by setting
1778                  * 'discards_supported') or it relies on _all_ data devices having
1779                  * discard support.
1780                  */
1781                 if (!ti->discards_supported &&
1782                     (!ti->type->iterate_devices ||
1783                      ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1784                         return false;
1785         }
1786
1787         return true;
1788 }
1789
1790 static int device_not_secure_erase_capable(struct dm_target *ti,
1791                                            struct dm_dev *dev, sector_t start,
1792                                            sector_t len, void *data)
1793 {
1794         struct request_queue *q = bdev_get_queue(dev->bdev);
1795
1796         return q && !blk_queue_secure_erase(q);
1797 }
1798
1799 static bool dm_table_supports_secure_erase(struct dm_table *t)
1800 {
1801         struct dm_target *ti;
1802         unsigned int i;
1803
1804         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1805                 ti = dm_table_get_target(t, i);
1806
1807                 if (!ti->num_secure_erase_bios)
1808                         return false;
1809
1810                 if (!ti->type->iterate_devices ||
1811                     ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1812                         return false;
1813         }
1814
1815         return true;
1816 }
1817
1818 static int device_requires_stable_pages(struct dm_target *ti,
1819                                         struct dm_dev *dev, sector_t start,
1820                                         sector_t len, void *data)
1821 {
1822         struct request_queue *q = bdev_get_queue(dev->bdev);
1823
1824         return q && bdi_cap_stable_pages_required(q->backing_dev_info);
1825 }
1826
1827 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1828                                struct queue_limits *limits)
1829 {
1830         bool wc = false, fua = false;
1831         int page_size = PAGE_SIZE;
1832
1833         /*
1834          * Copy table's limits to the DM device's request_queue
1835          */
1836         q->limits = *limits;
1837
1838         if (!dm_table_supports_discards(t)) {
1839                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
1840                 /* Must also clear discard limits... */
1841                 q->limits.max_discard_sectors = 0;
1842                 q->limits.max_hw_discard_sectors = 0;
1843                 q->limits.discard_granularity = 0;
1844                 q->limits.discard_alignment = 0;
1845                 q->limits.discard_misaligned = 0;
1846         } else
1847                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
1848
1849         if (dm_table_supports_secure_erase(t))
1850                 blk_queue_flag_set(QUEUE_FLAG_SECERASE, q);
1851
1852         if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1853                 wc = true;
1854                 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1855                         fua = true;
1856         }
1857         blk_queue_write_cache(q, wc, fua);
1858
1859         if (dm_table_supports_dax(t, device_not_dax_capable, &page_size)) {
1860                 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
1861                 if (dm_table_supports_dax(t, device_not_dax_synchronous_capable, NULL))
1862                         set_dax_synchronous(t->md->dax_dev);
1863         }
1864         else
1865                 blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
1866
1867         if (dm_table_any_dev_attr(t, device_dax_write_cache_enabled, NULL))
1868                 dax_write_cache(t->md->dax_dev, true);
1869
1870         /* Ensure that all underlying devices are non-rotational. */
1871         if (dm_table_any_dev_attr(t, device_is_rotational, NULL))
1872                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
1873         else
1874                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
1875
1876         if (!dm_table_supports_write_same(t))
1877                 q->limits.max_write_same_sectors = 0;
1878         if (!dm_table_supports_write_zeroes(t))
1879                 q->limits.max_write_zeroes_sectors = 0;
1880
1881         dm_table_verify_integrity(t);
1882
1883         /*
1884          * Some devices don't use blk_integrity but still want stable pages
1885          * because they do their own checksumming.
1886          * If any underlying device requires stable pages, a table must require
1887          * them as well.  Only targets that support iterate_devices are considered:
1888          * don't want error, zero, etc to require stable pages.
1889          */
1890         if (dm_table_any_dev_attr(t, device_requires_stable_pages, NULL))
1891                 q->backing_dev_info->capabilities |= BDI_CAP_STABLE_WRITES;
1892         else
1893                 q->backing_dev_info->capabilities &= ~BDI_CAP_STABLE_WRITES;
1894
1895         /*
1896          * Determine whether or not this queue's I/O timings contribute
1897          * to the entropy pool, Only request-based targets use this.
1898          * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1899          * have it set.
1900          */
1901         if (blk_queue_add_random(q) &&
1902             dm_table_any_dev_attr(t, device_is_not_random, NULL))
1903                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
1904
1905         /*
1906          * For a zoned target, the number of zones should be updated for the
1907          * correct value to be exposed in sysfs queue/nr_zones. For a BIO based
1908          * target, this is all that is needed. For a request based target, the
1909          * queue zone bitmaps must also be updated.
1910          * Use blk_revalidate_disk_zones() to handle this.
1911          */
1912         if (blk_queue_is_zoned(q))
1913                 blk_revalidate_disk_zones(t->md->disk);
1914
1915         /* Allow reads to exceed readahead limits */
1916         q->backing_dev_info->io_pages = limits->max_sectors >> (PAGE_SHIFT - 9);
1917 }
1918
1919 unsigned int dm_table_get_num_targets(struct dm_table *t)
1920 {
1921         return t->num_targets;
1922 }
1923
1924 struct list_head *dm_table_get_devices(struct dm_table *t)
1925 {
1926         return &t->devices;
1927 }
1928
1929 fmode_t dm_table_get_mode(struct dm_table *t)
1930 {
1931         return t->mode;
1932 }
1933 EXPORT_SYMBOL(dm_table_get_mode);
1934
1935 enum suspend_mode {
1936         PRESUSPEND,
1937         PRESUSPEND_UNDO,
1938         POSTSUSPEND,
1939 };
1940
1941 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1942 {
1943         int i = t->num_targets;
1944         struct dm_target *ti = t->targets;
1945
1946         lockdep_assert_held(&t->md->suspend_lock);
1947
1948         while (i--) {
1949                 switch (mode) {
1950                 case PRESUSPEND:
1951                         if (ti->type->presuspend)
1952                                 ti->type->presuspend(ti);
1953                         break;
1954                 case PRESUSPEND_UNDO:
1955                         if (ti->type->presuspend_undo)
1956                                 ti->type->presuspend_undo(ti);
1957                         break;
1958                 case POSTSUSPEND:
1959                         if (ti->type->postsuspend)
1960                                 ti->type->postsuspend(ti);
1961                         break;
1962                 }
1963                 ti++;
1964         }
1965 }
1966
1967 void dm_table_presuspend_targets(struct dm_table *t)
1968 {
1969         if (!t)
1970                 return;
1971
1972         suspend_targets(t, PRESUSPEND);
1973 }
1974
1975 void dm_table_presuspend_undo_targets(struct dm_table *t)
1976 {
1977         if (!t)
1978                 return;
1979
1980         suspend_targets(t, PRESUSPEND_UNDO);
1981 }
1982
1983 void dm_table_postsuspend_targets(struct dm_table *t)
1984 {
1985         if (!t)
1986                 return;
1987
1988         suspend_targets(t, POSTSUSPEND);
1989 }
1990
1991 int dm_table_resume_targets(struct dm_table *t)
1992 {
1993         int i, r = 0;
1994
1995         lockdep_assert_held(&t->md->suspend_lock);
1996
1997         for (i = 0; i < t->num_targets; i++) {
1998                 struct dm_target *ti = t->targets + i;
1999
2000                 if (!ti->type->preresume)
2001                         continue;
2002
2003                 r = ti->type->preresume(ti);
2004                 if (r) {
2005                         DMERR("%s: %s: preresume failed, error = %d",
2006                               dm_device_name(t->md), ti->type->name, r);
2007                         return r;
2008                 }
2009         }
2010
2011         for (i = 0; i < t->num_targets; i++) {
2012                 struct dm_target *ti = t->targets + i;
2013
2014                 if (ti->type->resume)
2015                         ti->type->resume(ti);
2016         }
2017
2018         return 0;
2019 }
2020
2021 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
2022 {
2023         list_add(&cb->list, &t->target_callbacks);
2024 }
2025 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
2026
2027 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
2028 {
2029         struct dm_dev_internal *dd;
2030         struct list_head *devices = dm_table_get_devices(t);
2031         struct dm_target_callbacks *cb;
2032         int r = 0;
2033
2034         list_for_each_entry(dd, devices, list) {
2035                 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
2036                 char b[BDEVNAME_SIZE];
2037
2038                 if (likely(q))
2039                         r |= bdi_congested(q->backing_dev_info, bdi_bits);
2040                 else
2041                         DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
2042                                      dm_device_name(t->md),
2043                                      bdevname(dd->dm_dev->bdev, b));
2044         }
2045
2046         list_for_each_entry(cb, &t->target_callbacks, list)
2047                 if (cb->congested_fn)
2048                         r |= cb->congested_fn(cb, bdi_bits);
2049
2050         return r;
2051 }
2052
2053 struct mapped_device *dm_table_get_md(struct dm_table *t)
2054 {
2055         return t->md;
2056 }
2057 EXPORT_SYMBOL(dm_table_get_md);
2058
2059 const char *dm_table_device_name(struct dm_table *t)
2060 {
2061         return dm_device_name(t->md);
2062 }
2063 EXPORT_SYMBOL_GPL(dm_table_device_name);
2064
2065 void dm_table_run_md_queue_async(struct dm_table *t)
2066 {
2067         struct mapped_device *md;
2068         struct request_queue *queue;
2069
2070         if (!dm_table_request_based(t))
2071                 return;
2072
2073         md = dm_table_get_md(t);
2074         queue = dm_get_md_queue(md);
2075         if (queue)
2076                 blk_mq_run_hw_queues(queue, true);
2077 }
2078 EXPORT_SYMBOL(dm_table_run_md_queue_async);
2079