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         struct dm_dev_internal *dd;
 435         struct dm_table *t = ti->table;
 436
 437         BUG_ON(!t);
 438
 439         dev = dm_get_dev_t(path);
 440         if (!dev)
 441                 return -ENODEV;
 442
 443         dd = find_device(&t->devices, dev);
 444         if (!dd) {
 445                 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
 446                 if (!dd)
 447                         return -ENOMEM;
 448
 449                 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
 450                         kfree(dd);
 451                         return r;
 452                 }
 453
 454                 refcount_set(&dd->count, 1);
 455                 list_add(&dd->list, &t->devices);
 456                 goto out;
 457
 458         } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
 459                 r = upgrade_mode(dd, mode, t->md);
 460                 if (r)
 461                         return r;
 462         }
 463         refcount_inc(&dd->count);
 464 out:
 465         *result = dd->dm_dev;
 466         return 0;
 467 }
 468 EXPORT_SYMBOL(dm_get_device);
 469
 470 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
 471                                 sector_t start, sector_t len, void *data)
 472 {
 473         struct queue_limits *limits = data;
 474         struct block_device *bdev = dev->bdev;
 475         struct request_queue *q = bdev_get_queue(bdev);
 476         char b[BDEVNAME_SIZE];
 477
 478         if (unlikely(!q)) {
 479                 DMWARN("%s: Cannot set limits for nonexistent device %s",
 480                        dm_device_name(ti->table->md), bdevname(bdev, b));
 481                 return 0;
 482         }
 483
 484         if (bdev_stack_limits(limits, bdev, start) < 0)
 485                 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
 486                        "physical_block_size=%u, logical_block_size=%u, "
 487                        "alignment_offset=%u, start=%llu",
 488                        dm_device_name(ti->table->md), bdevname(bdev, b),
 489                        q->limits.physical_block_size,
 490                        q->limits.logical_block_size,
 491                        q->limits.alignment_offset,
 492                        (unsigned long long) start << SECTOR_SHIFT);
 493
 494         limits->zoned = blk_queue_zoned_model(q);
 495
 496         return 0;
 497 }
 498
 499 /*
 500  * Decrement a device's use count and remove it if necessary.
 501  */
 502 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
 503 {
 504         int found = 0;
 505         struct list_head *devices = &ti->table->devices;
 506         struct dm_dev_internal *dd;
 507
 508         list_for_each_entry(dd, devices, list) {
 509                 if (dd->dm_dev == d) {
 510                         found = 1;
 511                         break;
 512                 }
 513         }
 514         if (!found) {
 515                 DMWARN("%s: device %s not in table devices list",
 516                        dm_device_name(ti->table->md), d->name);
 517                 return;
 518         }
 519         if (refcount_dec_and_test(&dd->count)) {
 520                 dm_put_table_device(ti->table->md, d);
 521                 list_del(&dd->list);
 522                 kfree(dd);
 523         }
 524 }
 525 EXPORT_SYMBOL(dm_put_device);
 526
 527 /*
 528  * Checks to see if the target joins onto the end of the table.
 529  */
 530 static int adjoin(struct dm_table *table, struct dm_target *ti)
 531 {
 532         struct dm_target *prev;
 533
 534         if (!table->num_targets)
 535                 return !ti->begin;
 536
 537         prev = &table->targets[table->num_targets - 1];
 538         return (ti->begin == (prev->begin + prev->len));
 539 }
 540
 541 /*
 542  * Used to dynamically allocate the arg array.
 543  *
 544  * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
 545  * process messages even if some device is suspended. These messages have a
 546  * small fixed number of arguments.
 547  *
 548  * On the other hand, dm-switch needs to process bulk data using messages and
 549  * excessive use of GFP_NOIO could cause trouble.
 550  */
 551 static char **realloc_argv(unsigned *array_size, char **old_argv)
 552 {
 553         char **argv;
 554         unsigned new_size;
 555         gfp_t gfp;
 556
 557         if (*array_size) {
 558                 new_size = *array_size * 2;
 559                 gfp = GFP_KERNEL;
 560         } else {
 561                 new_size = 8;
 562                 gfp = GFP_NOIO;
 563         }
 564         argv = kmalloc(new_size * sizeof(*argv), gfp);
 565         if (argv) {
 566                 memcpy(argv, old_argv, *array_size * sizeof(*argv));
 567                 *array_size = new_size;
 568         }
 569
 570         kfree(old_argv);
 571         return argv;
 572 }
 573
 574 /*
 575  * Destructively splits up the argument list to pass to ctr.
 576  */
 577 int dm_split_args(int *argc, char ***argvp, char *input)
 578 {
 579         char *start, *end = input, *out, **argv = NULL;
 580         unsigned array_size = 0;
 581
 582         *argc = 0;
 583
 584         if (!input) {
 585                 *argvp = NULL;
 586                 return 0;
 587         }
 588
 589         argv = realloc_argv(&array_size, argv);
 590         if (!argv)
 591                 return -ENOMEM;
 592
 593         while (1) {
 594                 /* Skip whitespace */
 595                 start = skip_spaces(end);
 596
 597                 if (!*start)
 598                         break;  /* success, we hit the end */
 599
 600                 /* 'out' is used to remove any back-quotes */
 601                 end = out = start;
 602                 while (*end) {
 603                         /* Everything apart from '\0' can be quoted */
 604                         if (*end == '\\' && *(end + 1)) {
 605                                 *out++ = *(end + 1);
 606                                 end += 2;
 607                                 continue;
 608                         }
 609
 610                         if (isspace(*end))
 611                                 break;  /* end of token */
 612
 613                         *out++ = *end++;
 614                 }
 615
 616                 /* have we already filled the array ? */
 617                 if ((*argc + 1) > array_size) {
 618                         argv = realloc_argv(&array_size, argv);
 619                         if (!argv)
 620                                 return -ENOMEM;
 621                 }
 622
 623                 /* we know this is whitespace */
 624                 if (*end)
 625                         end++;
 626
 627                 /* terminate the string and put it in the array */
 628                 *out = '\0';
 629                 argv[*argc] = start;
 630                 (*argc)++;
 631         }
 632
 633         *argvp = argv;
 634         return 0;
 635 }
 636
 637 /*
 638  * Impose necessary and sufficient conditions on a devices's table such
 639  * that any incoming bio which respects its logical_block_size can be
 640  * processed successfully.  If it falls across the boundary between
 641  * two or more targets, the size of each piece it gets split into must
 642  * be compatible with the logical_block_size of the target processing it.
 643  */
 644 static int validate_hardware_logical_block_alignment(struct dm_table *table,
 645                                                  struct queue_limits *limits)
 646 {
 647         /*
 648          * This function uses arithmetic modulo the logical_block_size
 649          * (in units of 512-byte sectors).
 650          */
 651         unsigned short device_logical_block_size_sects =
 652                 limits->logical_block_size >> SECTOR_SHIFT;
 653
 654         /*
 655          * Offset of the start of the next table entry, mod logical_block_size.
 656          */
 657         unsigned short next_target_start = 0;
 658
 659         /*
 660          * Given an aligned bio that extends beyond the end of a
 661          * target, how many sectors must the next target handle?
 662          */
 663         unsigned short remaining = 0;
 664
 665         struct dm_target *uninitialized_var(ti);
 666         struct queue_limits ti_limits;
 667         unsigned i;
 668
 669         /*
 670          * Check each entry in the table in turn.
 671          */
 672         for (i = 0; i < dm_table_get_num_targets(table); i++) {
 673                 ti = dm_table_get_target(table, i);
 674
 675                 blk_set_stacking_limits(&ti_limits);
 676
 677                 /* combine all target devices' limits */
 678                 if (ti->type->iterate_devices)
 679                         ti->type->iterate_devices(ti, dm_set_device_limits,
 680                                                   &ti_limits);
 681
 682                 /*
 683                  * If the remaining sectors fall entirely within this
 684                  * table entry are they compatible with its logical_block_size?
 685                  */
 686                 if (remaining < ti->len &&
 687                     remaining & ((ti_limits.logical_block_size >>
 688                                   SECTOR_SHIFT) - 1))
 689                         break;  /* Error */
 690
 691                 next_target_start =
 692                     (unsigned short) ((next_target_start + ti->len) &
 693                                       (device_logical_block_size_sects - 1));
 694                 remaining = next_target_start ?
 695                     device_logical_block_size_sects - next_target_start : 0;
 696         }
 697
 698         if (remaining) {
 699                 DMWARN("%s: table line %u (start sect %llu len %llu) "
 700                        "not aligned to h/w logical block size %u",
 701                        dm_device_name(table->md), i,
 702                        (unsigned long long) ti->begin,
 703                        (unsigned long long) ti->len,
 704                        limits->logical_block_size);
 705                 return -EINVAL;
 706         }
 707
 708         return 0;
 709 }
 710
 711 int dm_table_add_target(struct dm_table *t, const char *type,
 712                         sector_t start, sector_t len, char *params)
 713 {
 714         int r = -EINVAL, argc;
 715         char **argv;
 716         struct dm_target *tgt;
 717
 718         if (t->singleton) {
 719                 DMERR("%s: target type %s must appear alone in table",
 720                       dm_device_name(t->md), t->targets->type->name);
 721                 return -EINVAL;
 722         }
 723
 724         BUG_ON(t->num_targets >= t->num_allocated);
 725
 726         tgt = t->targets + t->num_targets;
 727         memset(tgt, 0, sizeof(*tgt));
 728
 729         if (!len) {
 730                 DMERR("%s: zero-length target", dm_device_name(t->md));
 731                 return -EINVAL;
 732         }
 733
 734         tgt->type = dm_get_target_type(type);
 735         if (!tgt->type) {
 736                 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
 737                 return -EINVAL;
 738         }
 739
 740         if (dm_target_needs_singleton(tgt->type)) {
 741                 if (t->num_targets) {
 742                         tgt->error = "singleton target type must appear alone in table";
 743                         goto bad;
 744                 }
 745                 t->singleton = true;
 746         }
 747
 748         if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
 749                 tgt->error = "target type may not be included in a read-only table";
 750                 goto bad;
 751         }
 752
 753         if (t->immutable_target_type) {
 754                 if (t->immutable_target_type != tgt->type) {
 755                         tgt->error = "immutable target type cannot be mixed with other target types";
 756                         goto bad;
 757                 }
 758         } else if (dm_target_is_immutable(tgt->type)) {
 759                 if (t->num_targets) {
 760                         tgt->error = "immutable target type cannot be mixed with other target types";
 761                         goto bad;
 762                 }
 763                 t->immutable_target_type = tgt->type;
 764         }
 765
 766         if (dm_target_has_integrity(tgt->type))
 767                 t->integrity_added = 1;
 768
 769         tgt->table = t;
 770         tgt->begin = start;
 771         tgt->len = len;
 772         tgt->error = "Unknown error";
 773
 774         /*
 775          * Does this target adjoin the previous one ?
 776          */
 777         if (!adjoin(t, tgt)) {
 778                 tgt->error = "Gap in table";
 779                 goto bad;
 780         }
 781
 782         r = dm_split_args(&argc, &argv, params);
 783         if (r) {
 784                 tgt->error = "couldn't split parameters (insufficient memory)";
 785                 goto bad;
 786         }
 787
 788         r = tgt->type->ctr(tgt, argc, argv);
 789         kfree(argv);
 790         if (r)
 791                 goto bad;
 792
 793         t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
 794
 795         if (!tgt->num_discard_bios && tgt->discards_supported)
 796                 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
 797                        dm_device_name(t->md), type);
 798
 799         return 0;
 800
 801  bad:
 802         DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
 803         dm_put_target_type(tgt->type);
 804         return r;
 805 }
 806
 807 /*
 808  * Target argument parsing helpers.
 809  */
 810 static int validate_next_arg(const struct dm_arg *arg,
 811                              struct dm_arg_set *arg_set,
 812                              unsigned *value, char **error, unsigned grouped)
 813 {
 814         const char *arg_str = dm_shift_arg(arg_set);
 815         char dummy;
 816
 817         if (!arg_str ||
 818             (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
 819             (*value < arg->min) ||
 820             (*value > arg->max) ||
 821             (grouped && arg_set->argc < *value)) {
 822                 *error = arg->error;
 823                 return -EINVAL;
 824         }
 825
 826         return 0;
 827 }
 828
 829 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 830                 unsigned *value, char **error)
 831 {
 832         return validate_next_arg(arg, arg_set, value, error, 0);
 833 }
 834 EXPORT_SYMBOL(dm_read_arg);
 835
 836 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 837                       unsigned *value, char **error)
 838 {
 839         return validate_next_arg(arg, arg_set, value, error, 1);
 840 }
 841 EXPORT_SYMBOL(dm_read_arg_group);
 842
 843 const char *dm_shift_arg(struct dm_arg_set *as)
 844 {
 845         char *r;
 846
 847         if (as->argc) {
 848                 as->argc--;
 849                 r = *as->argv;
 850                 as->argv++;
 851                 return r;
 852         }
 853
 854         return NULL;
 855 }
 856 EXPORT_SYMBOL(dm_shift_arg);
 857
 858 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
 859 {
 860         BUG_ON(as->argc < num_args);
 861         as->argc -= num_args;
 862         as->argv += num_args;
 863 }
 864 EXPORT_SYMBOL(dm_consume_args);
 865
 866 static bool __table_type_bio_based(enum dm_queue_mode table_type)
 867 {
 868         return (table_type == DM_TYPE_BIO_BASED ||
 869                 table_type == DM_TYPE_DAX_BIO_BASED ||
 870                 table_type == DM_TYPE_NVME_BIO_BASED);
 871 }
 872
 873 static bool __table_type_request_based(enum dm_queue_mode table_type)
 874 {
 875         return (table_type == DM_TYPE_REQUEST_BASED ||
 876                 table_type == DM_TYPE_MQ_REQUEST_BASED);
 877 }
 878
 879 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
 880 {
 881         t->type = type;
 882 }
 883 EXPORT_SYMBOL_GPL(dm_table_set_type);
 884
 885 static int device_supports_dax(struct dm_target *ti, struct dm_dev *dev,
 886                                sector_t start, sector_t len, void *data)
 887 {
 888         struct request_queue *q = bdev_get_queue(dev->bdev);
 889
 890         return q && blk_queue_dax(q);
 891 }
 892
 893 static bool dm_table_supports_dax(struct dm_table *t)
 894 {
 895         struct dm_target *ti;
 896         unsigned i;
 897
 898         /* Ensure that all targets support DAX. */
 899         for (i = 0; i < dm_table_get_num_targets(t); i++) {
 900                 ti = dm_table_get_target(t, i);
 901
 902                 if (!ti->type->direct_access)
 903                         return false;
 904
 905                 if (!ti->type->iterate_devices ||
 906                     !ti->type->iterate_devices(ti, device_supports_dax, NULL))
 907                         return false;
 908         }
 909
 910         return true;
 911 }
 912
 913 static bool dm_table_does_not_support_partial_completion(struct dm_table *t);
 914
 915 struct verify_rq_based_data {
 916         unsigned sq_count;
 917         unsigned mq_count;
 918 };
 919
 920 static int device_is_rq_based(struct dm_target *ti, struct dm_dev *dev,
 921                               sector_t start, sector_t len, void *data)
 922 {
 923         struct request_queue *q = bdev_get_queue(dev->bdev);
 924         struct verify_rq_based_data *v = data;
 925
 926         if (q->mq_ops)
 927                 v->mq_count++;
 928         else
 929                 v->sq_count++;
 930
 931         return queue_is_rq_based(q);
 932 }
 933
 934 static int dm_table_determine_type(struct dm_table *t)
 935 {
 936         unsigned i;
 937         unsigned bio_based = 0, request_based = 0, hybrid = 0;
 938         struct verify_rq_based_data v = {.sq_count = 0, .mq_count = 0};
 939         struct dm_target *tgt;
 940         struct list_head *devices = dm_table_get_devices(t);
 941         enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
 942
 943         if (t->type != DM_TYPE_NONE) {
 944                 /* target already set the table's type */
 945                 if (t->type == DM_TYPE_BIO_BASED) {
 946                         /* possibly upgrade to a variant of bio-based */
 947                         goto verify_bio_based;
 948                 }
 949                 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
 950                 BUG_ON(t->type == DM_TYPE_NVME_BIO_BASED);
 951                 goto verify_rq_based;
 952         }
 953
 954         for (i = 0; i < t->num_targets; i++) {
 955                 tgt = t->targets + i;
 956                 if (dm_target_hybrid(tgt))
 957                         hybrid = 1;
 958                 else if (dm_target_request_based(tgt))
 959                         request_based = 1;
 960                 else
 961                         bio_based = 1;
 962
 963                 if (bio_based && request_based) {
 964                         DMERR("Inconsistent table: different target types"
 965                               " can't be mixed up");
 966                         return -EINVAL;
 967                 }
 968         }
 969
 970         if (hybrid && !bio_based && !request_based) {
 971                 /*
 972                  * The targets can work either way.
 973                  * Determine the type from the live device.
 974                  * Default to bio-based if device is new.
 975                  */
 976                 if (__table_type_request_based(live_md_type))
 977                         request_based = 1;
 978                 else
 979                         bio_based = 1;
 980         }
 981
 982         if (bio_based) {
 983 verify_bio_based:
 984                 /* We must use this table as bio-based */
 985                 t->type = DM_TYPE_BIO_BASED;
 986                 if (dm_table_supports_dax(t) ||
 987                     (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
 988                         t->type = DM_TYPE_DAX_BIO_BASED;
 989                 } else {
 990                         /* Check if upgrading to NVMe bio-based is valid or required */
 991                         tgt = dm_table_get_immutable_target(t);
 992                         if (tgt && !tgt->max_io_len && dm_table_does_not_support_partial_completion(t)) {
 993                                 t->type = DM_TYPE_NVME_BIO_BASED;
 994                                 goto verify_rq_based; /* must be stacked directly on NVMe (blk-mq) */
 995                         } else if (list_empty(devices) && live_md_type == DM_TYPE_NVME_BIO_BASED) {
 996                                 t->type = DM_TYPE_NVME_BIO_BASED;
 997                         }
 998                 }
 999                 return 0;
1000         }
1001
1002         BUG_ON(!request_based); /* No targets in this table */
1003
1004         /*
1005          * The only way to establish DM_TYPE_MQ_REQUEST_BASED is by
1006          * having a compatible target use dm_table_set_type.
1007          */
1008         t->type = DM_TYPE_REQUEST_BASED;
1009
1010 verify_rq_based:
1011         /*
1012          * Request-based dm supports only tables that have a single target now.
1013          * To support multiple targets, request splitting support is needed,
1014          * and that needs lots of changes in the block-layer.
1015          * (e.g. request completion process for partial completion.)
1016          */
1017         if (t->num_targets > 1) {
1018                 DMERR("%s DM doesn't support multiple targets",
1019                       t->type == DM_TYPE_NVME_BIO_BASED ? "nvme bio-based" : "request-based");
1020                 return -EINVAL;
1021         }
1022
1023         if (list_empty(devices)) {
1024                 int srcu_idx;
1025                 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
1026
1027                 /* inherit live table's type and all_blk_mq */
1028                 if (live_table) {
1029                         t->type = live_table->type;
1030                         t->all_blk_mq = live_table->all_blk_mq;
1031                 }
1032                 dm_put_live_table(t->md, srcu_idx);
1033                 return 0;
1034         }
1035
1036         tgt = dm_table_get_immutable_target(t);
1037         if (!tgt) {
1038                 DMERR("table load rejected: immutable target is required");
1039                 return -EINVAL;
1040         } else if (tgt->max_io_len) {
1041                 DMERR("table load rejected: immutable target that splits IO is not supported");
1042                 return -EINVAL;
1043         }
1044
1045         /* Non-request-stackable devices can't be used for request-based dm */
1046         if (!tgt->type->iterate_devices ||
1047             !tgt->type->iterate_devices(tgt, device_is_rq_based, &v)) {
1048                 DMERR("table load rejected: including non-request-stackable devices");
1049                 return -EINVAL;
1050         }
1051         if (v.sq_count && v.mq_count) {
1052                 DMERR("table load rejected: not all devices are blk-mq request-stackable");
1053                 return -EINVAL;
1054         }
1055         t->all_blk_mq = v.mq_count > 0;
1056
1057         if (!t->all_blk_mq &&
1058             (t->type == DM_TYPE_MQ_REQUEST_BASED || t->type == DM_TYPE_NVME_BIO_BASED)) {
1059                 DMERR("table load rejected: all devices are not blk-mq request-stackable");
1060                 return -EINVAL;
1061         }
1062
1063         return 0;
1064 }
1065
1066 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
1067 {
1068         return t->type;
1069 }
1070
1071 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
1072 {
1073         return t->immutable_target_type;
1074 }
1075
1076 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
1077 {
1078         /* Immutable target is implicitly a singleton */
1079         if (t->num_targets > 1 ||
1080             !dm_target_is_immutable(t->targets[0].type))
1081                 return NULL;
1082
1083         return t->targets;
1084 }
1085
1086 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1087 {
1088         struct dm_target *ti;
1089         unsigned i;
1090
1091         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1092                 ti = dm_table_get_target(t, i);
1093                 if (dm_target_is_wildcard(ti->type))
1094                         return ti;
1095         }
1096
1097         return NULL;
1098 }
1099
1100 bool dm_table_bio_based(struct dm_table *t)
1101 {
1102         return __table_type_bio_based(dm_table_get_type(t));
1103 }
1104
1105 bool dm_table_request_based(struct dm_table *t)
1106 {
1107         return __table_type_request_based(dm_table_get_type(t));
1108 }
1109
1110 bool dm_table_all_blk_mq_devices(struct dm_table *t)
1111 {
1112         return t->all_blk_mq;
1113 }
1114
1115 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1116 {
1117         enum dm_queue_mode type = dm_table_get_type(t);
1118         unsigned per_io_data_size = 0;
1119         unsigned min_pool_size = 0;
1120         struct dm_target *ti;
1121         unsigned i;
1122
1123         if (unlikely(type == DM_TYPE_NONE)) {
1124                 DMWARN("no table type is set, can't allocate mempools");
1125                 return -EINVAL;
1126         }
1127
1128         if (__table_type_bio_based(type))
1129                 for (i = 0; i < t->num_targets; i++) {
1130                         ti = t->targets + i;
1131                         per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1132                         min_pool_size = max(min_pool_size, ti->num_flush_bios);
1133                 }
1134
1135         t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported,
1136                                            per_io_data_size, min_pool_size);
1137         if (!t->mempools)
1138                 return -ENOMEM;
1139
1140         return 0;
1141 }
1142
1143 void dm_table_free_md_mempools(struct dm_table *t)
1144 {
1145         dm_free_md_mempools(t->mempools);
1146         t->mempools = NULL;
1147 }
1148
1149 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1150 {
1151         return t->mempools;
1152 }
1153
1154 static int setup_indexes(struct dm_table *t)
1155 {
1156         int i;
1157         unsigned int total = 0;
1158         sector_t *indexes;
1159
1160         /* allocate the space for *all* the indexes */
1161         for (i = t->depth - 2; i >= 0; i--) {
1162                 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1163                 total += t->counts[i];
1164         }
1165
1166         indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1167         if (!indexes)
1168                 return -ENOMEM;
1169
1170         /* set up internal nodes, bottom-up */
1171         for (i = t->depth - 2; i >= 0; i--) {
1172                 t->index[i] = indexes;
1173                 indexes += (KEYS_PER_NODE * t->counts[i]);
1174                 setup_btree_index(i, t);
1175         }
1176
1177         return 0;
1178 }
1179
1180 /*
1181  * Builds the btree to index the map.
1182  */
1183 static int dm_table_build_index(struct dm_table *t)
1184 {
1185         int r = 0;
1186         unsigned int leaf_nodes;
1187
1188         /* how many indexes will the btree have ? */
1189         leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1190         t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1191
1192         /* leaf layer has already been set up */
1193         t->counts[t->depth - 1] = leaf_nodes;
1194         t->index[t->depth - 1] = t->highs;
1195
1196         if (t->depth >= 2)
1197                 r = setup_indexes(t);
1198
1199         return r;
1200 }
1201
1202 static bool integrity_profile_exists(struct gendisk *disk)
1203 {
1204         return !!blk_get_integrity(disk);
1205 }
1206
1207 /*
1208  * Get a disk whose integrity profile reflects the table's profile.
1209  * Returns NULL if integrity support was inconsistent or unavailable.
1210  */
1211 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1212 {
1213         struct list_head *devices = dm_table_get_devices(t);
1214         struct dm_dev_internal *dd = NULL;
1215         struct gendisk *prev_disk = NULL, *template_disk = NULL;
1216         unsigned i;
1217
1218         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1219                 struct dm_target *ti = dm_table_get_target(t, i);
1220                 if (!dm_target_passes_integrity(ti->type))
1221                         goto no_integrity;
1222         }
1223
1224         list_for_each_entry(dd, devices, list) {
1225                 template_disk = dd->dm_dev->bdev->bd_disk;
1226                 if (!integrity_profile_exists(template_disk))
1227                         goto no_integrity;
1228                 else if (prev_disk &&
1229                          blk_integrity_compare(prev_disk, template_disk) < 0)
1230                         goto no_integrity;
1231                 prev_disk = template_disk;
1232         }
1233
1234         return template_disk;
1235
1236 no_integrity:
1237         if (prev_disk)
1238                 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1239                        dm_device_name(t->md),
1240                        prev_disk->disk_name,
1241                        template_disk->disk_name);
1242         return NULL;
1243 }
1244
1245 /*
1246  * Register the mapped device for blk_integrity support if the
1247  * underlying devices have an integrity profile.  But all devices may
1248  * not have matching profiles (checking all devices isn't reliable
1249  * during table load because this table may use other DM device(s) which
1250  * must be resumed before they will have an initialized integity
1251  * profile).  Consequently, stacked DM devices force a 2 stage integrity
1252  * profile validation: First pass during table load, final pass during
1253  * resume.
1254  */
1255 static int dm_table_register_integrity(struct dm_table *t)
1256 {
1257         struct mapped_device *md = t->md;
1258         struct gendisk *template_disk = NULL;
1259
1260         /* If target handles integrity itself do not register it here. */
1261         if (t->integrity_added)
1262                 return 0;
1263
1264         template_disk = dm_table_get_integrity_disk(t);
1265         if (!template_disk)
1266                 return 0;
1267
1268         if (!integrity_profile_exists(dm_disk(md))) {
1269                 t->integrity_supported = true;
1270                 /*
1271                  * Register integrity profile during table load; we can do
1272                  * this because the final profile must match during resume.
1273                  */
1274                 blk_integrity_register(dm_disk(md),
1275                                        blk_get_integrity(template_disk));
1276                 return 0;
1277         }
1278
1279         /*
1280          * If DM device already has an initialized integrity
1281          * profile the new profile should not conflict.
1282          */
1283         if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1284                 DMWARN("%s: conflict with existing integrity profile: "
1285                        "%s profile mismatch",
1286                        dm_device_name(t->md),
1287                        template_disk->disk_name);
1288                 return 1;
1289         }
1290
1291         /* Preserve existing integrity profile */
1292         t->integrity_supported = true;
1293         return 0;
1294 }
1295
1296 /*
1297  * Prepares the table for use by building the indices,
1298  * setting the type, and allocating mempools.
1299  */
1300 int dm_table_complete(struct dm_table *t)
1301 {
1302         int r;
1303
1304         r = dm_table_determine_type(t);
1305         if (r) {
1306                 DMERR("unable to determine table type");
1307                 return r;
1308         }
1309
1310         r = dm_table_build_index(t);
1311         if (r) {
1312                 DMERR("unable to build btrees");
1313                 return r;
1314         }
1315
1316         r = dm_table_register_integrity(t);
1317         if (r) {
1318                 DMERR("could not register integrity profile.");
1319                 return r;
1320         }
1321
1322         r = dm_table_alloc_md_mempools(t, t->md);
1323         if (r)
1324                 DMERR("unable to allocate mempools");
1325
1326         return r;
1327 }
1328
1329 static DEFINE_MUTEX(_event_lock);
1330 void dm_table_event_callback(struct dm_table *t,
1331                              void (*fn)(void *), void *context)
1332 {
1333         mutex_lock(&_event_lock);
1334         t->event_fn = fn;
1335         t->event_context = context;
1336         mutex_unlock(&_event_lock);
1337 }
1338
1339 void dm_table_event(struct dm_table *t)
1340 {
1341         /*
1342          * You can no longer call dm_table_event() from interrupt
1343          * context, use a bottom half instead.
1344          */
1345         BUG_ON(in_interrupt());
1346
1347         mutex_lock(&_event_lock);
1348         if (t->event_fn)
1349                 t->event_fn(t->event_context);
1350         mutex_unlock(&_event_lock);
1351 }
1352 EXPORT_SYMBOL(dm_table_event);
1353
1354 sector_t dm_table_get_size(struct dm_table *t)
1355 {
1356         return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1357 }
1358 EXPORT_SYMBOL(dm_table_get_size);
1359
1360 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1361 {
1362         if (index >= t->num_targets)
1363                 return NULL;
1364
1365         return t->targets + index;
1366 }
1367
1368 /*
1369  * Search the btree for the correct target.
1370  *
1371  * Caller should check returned pointer with dm_target_is_valid()
1372  * to trap I/O beyond end of device.
1373  */
1374 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1375 {
1376         unsigned int l, n = 0, k = 0;
1377         sector_t *node;
1378
1379         for (l = 0; l < t->depth; l++) {
1380                 n = get_child(n, k);
1381                 node = get_node(t, l, n);
1382
1383                 for (k = 0; k < KEYS_PER_NODE; k++)
1384                         if (node[k] >= sector)
1385                                 break;
1386         }
1387
1388         return &t->targets[(KEYS_PER_NODE * n) + k];
1389 }
1390
1391 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1392                         sector_t start, sector_t len, void *data)
1393 {
1394         unsigned *num_devices = data;
1395
1396         (*num_devices)++;
1397
1398         return 0;
1399 }
1400
1401 /*
1402  * Check whether a table has no data devices attached using each
1403  * target's iterate_devices method.
1404  * Returns false if the result is unknown because a target doesn't
1405  * support iterate_devices.
1406  */
1407 bool dm_table_has_no_data_devices(struct dm_table *table)
1408 {
1409         struct dm_target *ti;
1410         unsigned i, num_devices;
1411
1412         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1413                 ti = dm_table_get_target(table, i);
1414
1415                 if (!ti->type->iterate_devices)
1416                         return false;
1417
1418                 num_devices = 0;
1419                 ti->type->iterate_devices(ti, count_device, &num_devices);
1420                 if (num_devices)
1421                         return false;
1422         }
1423
1424         return true;
1425 }
1426
1427 static int device_is_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1428                                  sector_t start, sector_t len, void *data)
1429 {
1430         struct request_queue *q = bdev_get_queue(dev->bdev);
1431         enum blk_zoned_model *zoned_model = data;
1432
1433         return q && blk_queue_zoned_model(q) == *zoned_model;
1434 }
1435
1436 static bool dm_table_supports_zoned_model(struct dm_table *t,
1437                                           enum blk_zoned_model zoned_model)
1438 {
1439         struct dm_target *ti;
1440         unsigned i;
1441
1442         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1443                 ti = dm_table_get_target(t, i);
1444
1445                 if (zoned_model == BLK_ZONED_HM &&
1446                     !dm_target_supports_zoned_hm(ti->type))
1447                         return false;
1448
1449                 if (!ti->type->iterate_devices ||
1450                     !ti->type->iterate_devices(ti, device_is_zoned_model, &zoned_model))
1451                         return false;
1452         }
1453
1454         return true;
1455 }
1456
1457 static int device_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1458                                        sector_t start, sector_t len, void *data)
1459 {
1460         struct request_queue *q = bdev_get_queue(dev->bdev);
1461         unsigned int *zone_sectors = data;
1462
1463         return q && blk_queue_zone_sectors(q) == *zone_sectors;
1464 }
1465
1466 static bool dm_table_matches_zone_sectors(struct dm_table *t,
1467                                           unsigned int zone_sectors)
1468 {
1469         struct dm_target *ti;
1470         unsigned i;
1471
1472         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1473                 ti = dm_table_get_target(t, i);
1474
1475                 if (!ti->type->iterate_devices ||
1476                     !ti->type->iterate_devices(ti, device_matches_zone_sectors, &zone_sectors))
1477                         return false;
1478         }
1479
1480         return true;
1481 }
1482
1483 static int validate_hardware_zoned_model(struct dm_table *table,
1484                                          enum blk_zoned_model zoned_model,
1485                                          unsigned int zone_sectors)
1486 {
1487         if (zoned_model == BLK_ZONED_NONE)
1488                 return 0;
1489
1490         if (!dm_table_supports_zoned_model(table, zoned_model)) {
1491                 DMERR("%s: zoned model is not consistent across all devices",
1492                       dm_device_name(table->md));
1493                 return -EINVAL;
1494         }
1495
1496         /* Check zone size validity and compatibility */
1497         if (!zone_sectors || !is_power_of_2(zone_sectors))
1498                 return -EINVAL;
1499
1500         if (!dm_table_matches_zone_sectors(table, zone_sectors)) {
1501                 DMERR("%s: zone sectors is not consistent across all devices",
1502                       dm_device_name(table->md));
1503                 return -EINVAL;
1504         }
1505
1506         return 0;
1507 }
1508
1509 /*
1510  * Establish the new table's queue_limits and validate them.
1511  */
1512 int dm_calculate_queue_limits(struct dm_table *table,
1513                               struct queue_limits *limits)
1514 {
1515         struct dm_target *ti;
1516         struct queue_limits ti_limits;
1517         unsigned i;
1518         enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1519         unsigned int zone_sectors = 0;
1520
1521         blk_set_stacking_limits(limits);
1522
1523         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1524                 blk_set_stacking_limits(&ti_limits);
1525
1526                 ti = dm_table_get_target(table, i);
1527
1528                 if (!ti->type->iterate_devices)
1529                         goto combine_limits;
1530
1531                 /*
1532                  * Combine queue limits of all the devices this target uses.
1533                  */
1534                 ti->type->iterate_devices(ti, dm_set_device_limits,
1535                                           &ti_limits);
1536
1537                 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1538                         /*
1539                          * After stacking all limits, validate all devices
1540                          * in table support this zoned model and zone sectors.
1541                          */
1542                         zoned_model = ti_limits.zoned;
1543                         zone_sectors = ti_limits.chunk_sectors;
1544                 }
1545
1546                 /* Set I/O hints portion of queue limits */
1547                 if (ti->type->io_hints)
1548                         ti->type->io_hints(ti, &ti_limits);
1549
1550                 /*
1551                  * Check each device area is consistent with the target's
1552                  * overall queue limits.
1553                  */
1554                 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1555                                               &ti_limits))
1556                         return -EINVAL;
1557
1558 combine_limits:
1559                 /*
1560                  * Merge this target's queue limits into the overall limits
1561                  * for the table.
1562                  */
1563                 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1564                         DMWARN("%s: adding target device "
1565                                "(start sect %llu len %llu) "
1566                                "caused an alignment inconsistency",
1567                                dm_device_name(table->md),
1568                                (unsigned long long) ti->begin,
1569                                (unsigned long long) ti->len);
1570
1571                 /*
1572                  * FIXME: this should likely be moved to blk_stack_limits(), would
1573                  * also eliminate limits->zoned stacking hack in dm_set_device_limits()
1574                  */
1575                 if (limits->zoned == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1576                         /*
1577                          * By default, the stacked limits zoned model is set to
1578                          * BLK_ZONED_NONE in blk_set_stacking_limits(). Update
1579                          * this model using the first target model reported
1580                          * that is not BLK_ZONED_NONE. This will be either the
1581                          * first target device zoned model or the model reported
1582                          * by the target .io_hints.
1583                          */
1584                         limits->zoned = ti_limits.zoned;
1585                 }
1586         }
1587
1588         /*
1589          * Verify that the zoned model and zone sectors, as determined before
1590          * any .io_hints override, are the same across all devices in the table.
1591          * - this is especially relevant if .io_hints is emulating a disk-managed
1592          *   zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1593          * BUT...
1594          */
1595         if (limits->zoned != BLK_ZONED_NONE) {
1596                 /*
1597                  * ...IF the above limits stacking determined a zoned model
1598                  * validate that all of the table's devices conform to it.
1599                  */
1600                 zoned_model = limits->zoned;
1601                 zone_sectors = limits->chunk_sectors;
1602         }
1603         if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1604                 return -EINVAL;
1605
1606         return validate_hardware_logical_block_alignment(table, limits);
1607 }
1608
1609 /*
1610  * Verify that all devices have an integrity profile that matches the
1611  * DM device's registered integrity profile.  If the profiles don't
1612  * match then unregister the DM device's integrity profile.
1613  */
1614 static void dm_table_verify_integrity(struct dm_table *t)
1615 {
1616         struct gendisk *template_disk = NULL;
1617
1618         if (t->integrity_added)
1619                 return;
1620
1621         if (t->integrity_supported) {
1622                 /*
1623                  * Verify that the original integrity profile
1624                  * matches all the devices in this table.
1625                  */
1626                 template_disk = dm_table_get_integrity_disk(t);
1627                 if (template_disk &&
1628                     blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1629                         return;
1630         }
1631
1632         if (integrity_profile_exists(dm_disk(t->md))) {
1633                 DMWARN("%s: unable to establish an integrity profile",
1634                        dm_device_name(t->md));
1635                 blk_integrity_unregister(dm_disk(t->md));
1636         }
1637 }
1638
1639 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1640                                 sector_t start, sector_t len, void *data)
1641 {
1642         unsigned long flush = (unsigned long) data;
1643         struct request_queue *q = bdev_get_queue(dev->bdev);
1644
1645         return q && (q->queue_flags & flush);
1646 }
1647
1648 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1649 {
1650         struct dm_target *ti;
1651         unsigned i;
1652
1653         /*
1654          * Require at least one underlying device to support flushes.
1655          * t->devices includes internal dm devices such as mirror logs
1656          * so we need to use iterate_devices here, which targets
1657          * supporting flushes must provide.
1658          */
1659         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1660                 ti = dm_table_get_target(t, i);
1661
1662                 if (!ti->num_flush_bios)
1663                         continue;
1664
1665                 if (ti->flush_supported)
1666                         return true;
1667
1668                 if (ti->type->iterate_devices &&
1669                     ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1670                         return true;
1671         }
1672
1673         return false;
1674 }
1675
1676 static int device_dax_write_cache_enabled(struct dm_target *ti,
1677                                           struct dm_dev *dev, sector_t start,
1678                                           sector_t len, void *data)
1679 {
1680         struct dax_device *dax_dev = dev->dax_dev;
1681
1682         if (!dax_dev)
1683                 return false;
1684
1685         if (dax_write_cache_enabled(dax_dev))
1686                 return true;
1687         return false;
1688 }
1689
1690 static int dm_table_supports_dax_write_cache(struct dm_table *t)
1691 {
1692         struct dm_target *ti;
1693         unsigned i;
1694
1695         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1696                 ti = dm_table_get_target(t, i);
1697
1698                 if (ti->type->iterate_devices &&
1699                     ti->type->iterate_devices(ti,
1700                                 device_dax_write_cache_enabled, NULL))
1701                         return true;
1702         }
1703
1704         return false;
1705 }
1706
1707 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1708                             sector_t start, sector_t len, void *data)
1709 {
1710         struct request_queue *q = bdev_get_queue(dev->bdev);
1711
1712         return q && blk_queue_nonrot(q);
1713 }
1714
1715 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1716                              sector_t start, sector_t len, void *data)
1717 {
1718         struct request_queue *q = bdev_get_queue(dev->bdev);
1719
1720         return q && !blk_queue_add_random(q);
1721 }
1722
1723 static int queue_supports_sg_merge(struct dm_target *ti, struct dm_dev *dev,
1724                                    sector_t start, sector_t len, void *data)
1725 {
1726         struct request_queue *q = bdev_get_queue(dev->bdev);
1727
1728         return q && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1729 }
1730
1731 static bool dm_table_all_devices_attribute(struct dm_table *t,
1732                                            iterate_devices_callout_fn func)
1733 {
1734         struct dm_target *ti;
1735         unsigned i;
1736
1737         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1738                 ti = dm_table_get_target(t, i);
1739
1740                 if (!ti->type->iterate_devices ||
1741                     !ti->type->iterate_devices(ti, func, NULL))
1742                         return false;
1743         }
1744
1745         return true;
1746 }
1747
1748 static int device_no_partial_completion(struct dm_target *ti, struct dm_dev *dev,
1749                                         sector_t start, sector_t len, void *data)
1750 {
1751         char b[BDEVNAME_SIZE];
1752
1753         /* For now, NVMe devices are the only devices of this class */
1754         return (strncmp(bdevname(dev->bdev, b), "nvme", 4) == 0);
1755 }
1756
1757 static bool dm_table_does_not_support_partial_completion(struct dm_table *t)
1758 {
1759         return dm_table_all_devices_attribute(t, device_no_partial_completion);
1760 }
1761
1762 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1763                                          sector_t start, sector_t len, void *data)
1764 {
1765         struct request_queue *q = bdev_get_queue(dev->bdev);
1766
1767         return q && !q->limits.max_write_same_sectors;
1768 }
1769
1770 static bool dm_table_supports_write_same(struct dm_table *t)
1771 {
1772         struct dm_target *ti;
1773         unsigned i;
1774
1775         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1776                 ti = dm_table_get_target(t, i);
1777
1778                 if (!ti->num_write_same_bios)
1779                         return false;
1780
1781                 if (!ti->type->iterate_devices ||
1782                     ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1783                         return false;
1784         }
1785
1786         return true;
1787 }
1788
1789 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1790                                            sector_t start, sector_t len, void *data)
1791 {
1792         struct request_queue *q = bdev_get_queue(dev->bdev);
1793
1794         return q && !q->limits.max_write_zeroes_sectors;
1795 }
1796
1797 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1798 {
1799         struct dm_target *ti;
1800         unsigned i = 0;
1801
1802         while (i < dm_table_get_num_targets(t)) {
1803                 ti = dm_table_get_target(t, i++);
1804
1805                 if (!ti->num_write_zeroes_bios)
1806                         return false;
1807
1808                 if (!ti->type->iterate_devices ||
1809                     ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1810                         return false;
1811         }
1812
1813         return true;
1814 }
1815
1816 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1817                                       sector_t start, sector_t len, void *data)
1818 {
1819         struct request_queue *q = bdev_get_queue(dev->bdev);
1820
1821         return q && !blk_queue_discard(q);
1822 }
1823
1824 static bool dm_table_supports_discards(struct dm_table *t)
1825 {
1826         struct dm_target *ti;
1827         unsigned i;
1828
1829         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1830                 ti = dm_table_get_target(t, i);
1831
1832                 if (!ti->num_discard_bios)
1833                         return false;
1834
1835                 /*
1836                  * Either the target provides discard support (as implied by setting
1837                  * 'discards_supported') or it relies on _all_ data devices having
1838                  * discard support.
1839                  */
1840                 if (!ti->discards_supported &&
1841                     (!ti->type->iterate_devices ||
1842                      ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1843                         return false;
1844         }
1845
1846         return true;
1847 }
1848
1849 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1850                                struct queue_limits *limits)
1851 {
1852         bool wc = false, fua = false;
1853
1854         /*
1855          * Copy table's limits to the DM device's request_queue
1856          */
1857         q->limits = *limits;
1858
1859         if (!dm_table_supports_discards(t)) {
1860                 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1861                 /* Must also clear discard limits... */
1862                 q->limits.max_discard_sectors = 0;
1863                 q->limits.max_hw_discard_sectors = 0;
1864                 q->limits.discard_granularity = 0;
1865                 q->limits.discard_alignment = 0;
1866                 q->limits.discard_misaligned = 0;
1867         } else
1868                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1869
1870         if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1871                 wc = true;
1872                 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1873                         fua = true;
1874         }
1875         blk_queue_write_cache(q, wc, fua);
1876
1877         if (dm_table_supports_dax(t))
1878                 queue_flag_set_unlocked(QUEUE_FLAG_DAX, q);
1879         if (dm_table_supports_dax_write_cache(t))
1880                 dax_write_cache(t->md->dax_dev, true);
1881
1882         /* Ensure that all underlying devices are non-rotational. */
1883         if (dm_table_all_devices_attribute(t, device_is_nonrot))
1884                 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1885         else
1886                 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1887
1888         if (!dm_table_supports_write_same(t))
1889                 q->limits.max_write_same_sectors = 0;
1890         if (!dm_table_supports_write_zeroes(t))
1891                 q->limits.max_write_zeroes_sectors = 0;
1892
1893         if (dm_table_all_devices_attribute(t, queue_supports_sg_merge))
1894                 queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1895         else
1896                 queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1897
1898         dm_table_verify_integrity(t);
1899
1900         /*
1901          * Determine whether or not this queue's I/O timings contribute
1902          * to the entropy pool, Only request-based targets use this.
1903          * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1904          * have it set.
1905          */
1906         if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1907                 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1908 }
1909
1910 unsigned int dm_table_get_num_targets(struct dm_table *t)
1911 {
1912         return t->num_targets;
1913 }
1914
1915 struct list_head *dm_table_get_devices(struct dm_table *t)
1916 {
1917         return &t->devices;
1918 }
1919
1920 fmode_t dm_table_get_mode(struct dm_table *t)
1921 {
1922         return t->mode;
1923 }
1924 EXPORT_SYMBOL(dm_table_get_mode);
1925
1926 enum suspend_mode {
1927         PRESUSPEND,
1928         PRESUSPEND_UNDO,
1929         POSTSUSPEND,
1930 };
1931
1932 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1933 {
1934         int i = t->num_targets;
1935         struct dm_target *ti = t->targets;
1936
1937         lockdep_assert_held(&t->md->suspend_lock);
1938
1939         while (i--) {
1940                 switch (mode) {
1941                 case PRESUSPEND:
1942                         if (ti->type->presuspend)
1943                                 ti->type->presuspend(ti);
1944                         break;
1945                 case PRESUSPEND_UNDO:
1946                         if (ti->type->presuspend_undo)
1947                                 ti->type->presuspend_undo(ti);
1948                         break;
1949                 case POSTSUSPEND:
1950                         if (ti->type->postsuspend)
1951                                 ti->type->postsuspend(ti);
1952                         break;
1953                 }
1954                 ti++;
1955         }
1956 }
1957
1958 void dm_table_presuspend_targets(struct dm_table *t)
1959 {
1960         if (!t)
1961                 return;
1962
1963         suspend_targets(t, PRESUSPEND);
1964 }
1965
1966 void dm_table_presuspend_undo_targets(struct dm_table *t)
1967 {
1968         if (!t)
1969                 return;
1970
1971         suspend_targets(t, PRESUSPEND_UNDO);
1972 }
1973
1974 void dm_table_postsuspend_targets(struct dm_table *t)
1975 {
1976         if (!t)
1977                 return;
1978
1979         suspend_targets(t, POSTSUSPEND);
1980 }
1981
1982 int dm_table_resume_targets(struct dm_table *t)
1983 {
1984         int i, r = 0;
1985
1986         lockdep_assert_held(&t->md->suspend_lock);
1987
1988         for (i = 0; i < t->num_targets; i++) {
1989                 struct dm_target *ti = t->targets + i;
1990
1991                 if (!ti->type->preresume)
1992                         continue;
1993
1994                 r = ti->type->preresume(ti);
1995                 if (r) {
1996                         DMERR("%s: %s: preresume failed, error = %d",
1997                               dm_device_name(t->md), ti->type->name, r);
1998                         return r;
1999                 }
2000         }
2001
2002         for (i = 0; i < t->num_targets; i++) {
2003                 struct dm_target *ti = t->targets + i;
2004
2005                 if (ti->type->resume)
2006                         ti->type->resume(ti);
2007         }
2008
2009         return 0;
2010 }
2011
2012 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
2013 {
2014         list_add(&cb->list, &t->target_callbacks);
2015 }
2016 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
2017
2018 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
2019 {
2020         struct dm_dev_internal *dd;
2021         struct list_head *devices = dm_table_get_devices(t);
2022         struct dm_target_callbacks *cb;
2023         int r = 0;
2024
2025         list_for_each_entry(dd, devices, list) {
2026                 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
2027                 char b[BDEVNAME_SIZE];
2028
2029                 if (likely(q))
2030                         r |= bdi_congested(q->backing_dev_info, bdi_bits);
2031                 else
2032                         DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
2033                                      dm_device_name(t->md),
2034                                      bdevname(dd->dm_dev->bdev, b));
2035         }
2036
2037         list_for_each_entry(cb, &t->target_callbacks, list)
2038                 if (cb->congested_fn)
2039                         r |= cb->congested_fn(cb, bdi_bits);
2040
2041         return r;
2042 }
2043
2044 struct mapped_device *dm_table_get_md(struct dm_table *t)
2045 {
2046         return t->md;
2047 }
2048 EXPORT_SYMBOL(dm_table_get_md);
2049
2050 void dm_table_run_md_queue_async(struct dm_table *t)
2051 {
2052         struct mapped_device *md;
2053         struct request_queue *queue;
2054         unsigned long flags;
2055
2056         if (!dm_table_request_based(t))
2057                 return;
2058
2059         md = dm_table_get_md(t);
2060         queue = dm_get_md_queue(md);
2061         if (queue) {
2062                 if (queue->mq_ops)
2063                         blk_mq_run_hw_queues(queue, true);
2064                 else {
2065                         spin_lock_irqsave(queue->queue_lock, flags);
2066                         blk_run_queue_async(queue);
2067                         spin_unlock_irqrestore(queue->queue_lock, flags);
2068                 }
2069         }
2070 }
2071 EXPORT_SYMBOL(dm_table_run_md_queue_async);
2072