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 *size, char **old_argv)
 552 {
 553         char **argv;
 554         unsigned new_size;
 555         gfp_t gfp;
 556
 557         if (*size) {
 558                 new_size = *size * 2;
 559                 gfp = GFP_KERNEL;
 560         } else {
 561                 new_size = 8;
 562                 gfp = GFP_NOIO;
 563         }
 564         argv = kmalloc_array(new_size, sizeof(*argv), gfp);
 565         if (argv) {
 566                 memcpy(argv, old_argv, *size * sizeof(*argv));
 567                 *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         return bdev_dax_supported(dev->bdev, PAGE_SIZE);
 889 }
 890
 891 static bool dm_table_supports_dax(struct dm_table *t)
 892 {
 893         struct dm_target *ti;
 894         unsigned i;
 895
 896         /* Ensure that all targets support DAX. */
 897         for (i = 0; i < dm_table_get_num_targets(t); i++) {
 898                 ti = dm_table_get_target(t, i);
 899
 900                 if (!ti->type->direct_access)
 901                         return false;
 902
 903                 if (!ti->type->iterate_devices ||
 904                     !ti->type->iterate_devices(ti, device_supports_dax, NULL))
 905                         return false;
 906         }
 907
 908         return true;
 909 }
 910
 911 static bool dm_table_does_not_support_partial_completion(struct dm_table *t);
 912
 913 struct verify_rq_based_data {
 914         unsigned sq_count;
 915         unsigned mq_count;
 916 };
 917
 918 static int device_is_rq_based(struct dm_target *ti, struct dm_dev *dev,
 919                               sector_t start, sector_t len, void *data)
 920 {
 921         struct request_queue *q = bdev_get_queue(dev->bdev);
 922         struct verify_rq_based_data *v = data;
 923
 924         if (q->mq_ops)
 925                 v->mq_count++;
 926         else
 927                 v->sq_count++;
 928
 929         return queue_is_rq_based(q);
 930 }
 931
 932 static int dm_table_determine_type(struct dm_table *t)
 933 {
 934         unsigned i;
 935         unsigned bio_based = 0, request_based = 0, hybrid = 0;
 936         struct verify_rq_based_data v = {.sq_count = 0, .mq_count = 0};
 937         struct dm_target *tgt;
 938         struct list_head *devices = dm_table_get_devices(t);
 939         enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
 940
 941         if (t->type != DM_TYPE_NONE) {
 942                 /* target already set the table's type */
 943                 if (t->type == DM_TYPE_BIO_BASED) {
 944                         /* possibly upgrade to a variant of bio-based */
 945                         goto verify_bio_based;
 946                 }
 947                 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
 948                 BUG_ON(t->type == DM_TYPE_NVME_BIO_BASED);
 949                 goto verify_rq_based;
 950         }
 951
 952         for (i = 0; i < t->num_targets; i++) {
 953                 tgt = t->targets + i;
 954                 if (dm_target_hybrid(tgt))
 955                         hybrid = 1;
 956                 else if (dm_target_request_based(tgt))
 957                         request_based = 1;
 958                 else
 959                         bio_based = 1;
 960
 961                 if (bio_based && request_based) {
 962                         DMERR("Inconsistent table: different target types"
 963                               " can't be mixed up");
 964                         return -EINVAL;
 965                 }
 966         }
 967
 968         if (hybrid && !bio_based && !request_based) {
 969                 /*
 970                  * The targets can work either way.
 971                  * Determine the type from the live device.
 972                  * Default to bio-based if device is new.
 973                  */
 974                 if (__table_type_request_based(live_md_type))
 975                         request_based = 1;
 976                 else
 977                         bio_based = 1;
 978         }
 979
 980         if (bio_based) {
 981 verify_bio_based:
 982                 /* We must use this table as bio-based */
 983                 t->type = DM_TYPE_BIO_BASED;
 984                 if (dm_table_supports_dax(t) ||
 985                     (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
 986                         t->type = DM_TYPE_DAX_BIO_BASED;
 987                 } else {
 988                         /* Check if upgrading to NVMe bio-based is valid or required */
 989                         tgt = dm_table_get_immutable_target(t);
 990                         if (tgt && !tgt->max_io_len && dm_table_does_not_support_partial_completion(t)) {
 991                                 t->type = DM_TYPE_NVME_BIO_BASED;
 992                                 goto verify_rq_based; /* must be stacked directly on NVMe (blk-mq) */
 993                         } else if (list_empty(devices) && live_md_type == DM_TYPE_NVME_BIO_BASED) {
 994                                 t->type = DM_TYPE_NVME_BIO_BASED;
 995                         }
 996                 }
 997                 return 0;
 998         }
 999
1000         BUG_ON(!request_based); /* No targets in this table */
1001
1002         /*
1003          * The only way to establish DM_TYPE_MQ_REQUEST_BASED is by
1004          * having a compatible target use dm_table_set_type.
1005          */
1006         t->type = DM_TYPE_REQUEST_BASED;
1007
1008 verify_rq_based:
1009         /*
1010          * Request-based dm supports only tables that have a single target now.
1011          * To support multiple targets, request splitting support is needed,
1012          * and that needs lots of changes in the block-layer.
1013          * (e.g. request completion process for partial completion.)
1014          */
1015         if (t->num_targets > 1) {
1016                 DMERR("%s DM doesn't support multiple targets",
1017                       t->type == DM_TYPE_NVME_BIO_BASED ? "nvme bio-based" : "request-based");
1018                 return -EINVAL;
1019         }
1020
1021         if (list_empty(devices)) {
1022                 int srcu_idx;
1023                 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
1024
1025                 /* inherit live table's type and all_blk_mq */
1026                 if (live_table) {
1027                         t->type = live_table->type;
1028                         t->all_blk_mq = live_table->all_blk_mq;
1029                 }
1030                 dm_put_live_table(t->md, srcu_idx);
1031                 return 0;
1032         }
1033
1034         tgt = dm_table_get_immutable_target(t);
1035         if (!tgt) {
1036                 DMERR("table load rejected: immutable target is required");
1037                 return -EINVAL;
1038         } else if (tgt->max_io_len) {
1039                 DMERR("table load rejected: immutable target that splits IO is not supported");
1040                 return -EINVAL;
1041         }
1042
1043         /* Non-request-stackable devices can't be used for request-based dm */
1044         if (!tgt->type->iterate_devices ||
1045             !tgt->type->iterate_devices(tgt, device_is_rq_based, &v)) {
1046                 DMERR("table load rejected: including non-request-stackable devices");
1047                 return -EINVAL;
1048         }
1049         if (v.sq_count && v.mq_count) {
1050                 DMERR("table load rejected: not all devices are blk-mq request-stackable");
1051                 return -EINVAL;
1052         }
1053         t->all_blk_mq = v.mq_count > 0;
1054
1055         if (!t->all_blk_mq &&
1056             (t->type == DM_TYPE_MQ_REQUEST_BASED || t->type == DM_TYPE_NVME_BIO_BASED)) {
1057                 DMERR("table load rejected: all devices are not blk-mq request-stackable");
1058                 return -EINVAL;
1059         }
1060
1061         return 0;
1062 }
1063
1064 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
1065 {
1066         return t->type;
1067 }
1068
1069 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
1070 {
1071         return t->immutable_target_type;
1072 }
1073
1074 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
1075 {
1076         /* Immutable target is implicitly a singleton */
1077         if (t->num_targets > 1 ||
1078             !dm_target_is_immutable(t->targets[0].type))
1079                 return NULL;
1080
1081         return t->targets;
1082 }
1083
1084 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1085 {
1086         struct dm_target *ti;
1087         unsigned i;
1088
1089         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1090                 ti = dm_table_get_target(t, i);
1091                 if (dm_target_is_wildcard(ti->type))
1092                         return ti;
1093         }
1094
1095         return NULL;
1096 }
1097
1098 bool dm_table_bio_based(struct dm_table *t)
1099 {
1100         return __table_type_bio_based(dm_table_get_type(t));
1101 }
1102
1103 bool dm_table_request_based(struct dm_table *t)
1104 {
1105         return __table_type_request_based(dm_table_get_type(t));
1106 }
1107
1108 bool dm_table_all_blk_mq_devices(struct dm_table *t)
1109 {
1110         return t->all_blk_mq;
1111 }
1112
1113 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1114 {
1115         enum dm_queue_mode type = dm_table_get_type(t);
1116         unsigned per_io_data_size = 0;
1117         unsigned min_pool_size = 0;
1118         struct dm_target *ti;
1119         unsigned i;
1120
1121         if (unlikely(type == DM_TYPE_NONE)) {
1122                 DMWARN("no table type is set, can't allocate mempools");
1123                 return -EINVAL;
1124         }
1125
1126         if (__table_type_bio_based(type))
1127                 for (i = 0; i < t->num_targets; i++) {
1128                         ti = t->targets + i;
1129                         per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1130                         min_pool_size = max(min_pool_size, ti->num_flush_bios);
1131                 }
1132
1133         t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported,
1134                                            per_io_data_size, min_pool_size);
1135         if (!t->mempools)
1136                 return -ENOMEM;
1137
1138         return 0;
1139 }
1140
1141 void dm_table_free_md_mempools(struct dm_table *t)
1142 {
1143         dm_free_md_mempools(t->mempools);
1144         t->mempools = NULL;
1145 }
1146
1147 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1148 {
1149         return t->mempools;
1150 }
1151
1152 static int setup_indexes(struct dm_table *t)
1153 {
1154         int i;
1155         unsigned int total = 0;
1156         sector_t *indexes;
1157
1158         /* allocate the space for *all* the indexes */
1159         for (i = t->depth - 2; i >= 0; i--) {
1160                 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1161                 total += t->counts[i];
1162         }
1163
1164         indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1165         if (!indexes)
1166                 return -ENOMEM;
1167
1168         /* set up internal nodes, bottom-up */
1169         for (i = t->depth - 2; i >= 0; i--) {
1170                 t->index[i] = indexes;
1171                 indexes += (KEYS_PER_NODE * t->counts[i]);
1172                 setup_btree_index(i, t);
1173         }
1174
1175         return 0;
1176 }
1177
1178 /*
1179  * Builds the btree to index the map.
1180  */
1181 static int dm_table_build_index(struct dm_table *t)
1182 {
1183         int r = 0;
1184         unsigned int leaf_nodes;
1185
1186         /* how many indexes will the btree have ? */
1187         leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1188         t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1189
1190         /* leaf layer has already been set up */
1191         t->counts[t->depth - 1] = leaf_nodes;
1192         t->index[t->depth - 1] = t->highs;
1193
1194         if (t->depth >= 2)
1195                 r = setup_indexes(t);
1196
1197         return r;
1198 }
1199
1200 static bool integrity_profile_exists(struct gendisk *disk)
1201 {
1202         return !!blk_get_integrity(disk);
1203 }
1204
1205 /*
1206  * Get a disk whose integrity profile reflects the table's profile.
1207  * Returns NULL if integrity support was inconsistent or unavailable.
1208  */
1209 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1210 {
1211         struct list_head *devices = dm_table_get_devices(t);
1212         struct dm_dev_internal *dd = NULL;
1213         struct gendisk *prev_disk = NULL, *template_disk = NULL;
1214         unsigned i;
1215
1216         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1217                 struct dm_target *ti = dm_table_get_target(t, i);
1218                 if (!dm_target_passes_integrity(ti->type))
1219                         goto no_integrity;
1220         }
1221
1222         list_for_each_entry(dd, devices, list) {
1223                 template_disk = dd->dm_dev->bdev->bd_disk;
1224                 if (!integrity_profile_exists(template_disk))
1225                         goto no_integrity;
1226                 else if (prev_disk &&
1227                          blk_integrity_compare(prev_disk, template_disk) < 0)
1228                         goto no_integrity;
1229                 prev_disk = template_disk;
1230         }
1231
1232         return template_disk;
1233
1234 no_integrity:
1235         if (prev_disk)
1236                 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1237                        dm_device_name(t->md),
1238                        prev_disk->disk_name,
1239                        template_disk->disk_name);
1240         return NULL;
1241 }
1242
1243 /*
1244  * Register the mapped device for blk_integrity support if the
1245  * underlying devices have an integrity profile.  But all devices may
1246  * not have matching profiles (checking all devices isn't reliable
1247  * during table load because this table may use other DM device(s) which
1248  * must be resumed before they will have an initialized integity
1249  * profile).  Consequently, stacked DM devices force a 2 stage integrity
1250  * profile validation: First pass during table load, final pass during
1251  * resume.
1252  */
1253 static int dm_table_register_integrity(struct dm_table *t)
1254 {
1255         struct mapped_device *md = t->md;
1256         struct gendisk *template_disk = NULL;
1257
1258         /* If target handles integrity itself do not register it here. */
1259         if (t->integrity_added)
1260                 return 0;
1261
1262         template_disk = dm_table_get_integrity_disk(t);
1263         if (!template_disk)
1264                 return 0;
1265
1266         if (!integrity_profile_exists(dm_disk(md))) {
1267                 t->integrity_supported = true;
1268                 /*
1269                  * Register integrity profile during table load; we can do
1270                  * this because the final profile must match during resume.
1271                  */
1272                 blk_integrity_register(dm_disk(md),
1273                                        blk_get_integrity(template_disk));
1274                 return 0;
1275         }
1276
1277         /*
1278          * If DM device already has an initialized integrity
1279          * profile the new profile should not conflict.
1280          */
1281         if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1282                 DMWARN("%s: conflict with existing integrity profile: "
1283                        "%s profile mismatch",
1284                        dm_device_name(t->md),
1285                        template_disk->disk_name);
1286                 return 1;
1287         }
1288
1289         /* Preserve existing integrity profile */
1290         t->integrity_supported = true;
1291         return 0;
1292 }
1293
1294 /*
1295  * Prepares the table for use by building the indices,
1296  * setting the type, and allocating mempools.
1297  */
1298 int dm_table_complete(struct dm_table *t)
1299 {
1300         int r;
1301
1302         r = dm_table_determine_type(t);
1303         if (r) {
1304                 DMERR("unable to determine table type");
1305                 return r;
1306         }
1307
1308         r = dm_table_build_index(t);
1309         if (r) {
1310                 DMERR("unable to build btrees");
1311                 return r;
1312         }
1313
1314         r = dm_table_register_integrity(t);
1315         if (r) {
1316                 DMERR("could not register integrity profile.");
1317                 return r;
1318         }
1319
1320         r = dm_table_alloc_md_mempools(t, t->md);
1321         if (r)
1322                 DMERR("unable to allocate mempools");
1323
1324         return r;
1325 }
1326
1327 static DEFINE_MUTEX(_event_lock);
1328 void dm_table_event_callback(struct dm_table *t,
1329                              void (*fn)(void *), void *context)
1330 {
1331         mutex_lock(&_event_lock);
1332         t->event_fn = fn;
1333         t->event_context = context;
1334         mutex_unlock(&_event_lock);
1335 }
1336
1337 void dm_table_event(struct dm_table *t)
1338 {
1339         /*
1340          * You can no longer call dm_table_event() from interrupt
1341          * context, use a bottom half instead.
1342          */
1343         BUG_ON(in_interrupt());
1344
1345         mutex_lock(&_event_lock);
1346         if (t->event_fn)
1347                 t->event_fn(t->event_context);
1348         mutex_unlock(&_event_lock);
1349 }
1350 EXPORT_SYMBOL(dm_table_event);
1351
1352 sector_t dm_table_get_size(struct dm_table *t)
1353 {
1354         return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1355 }
1356 EXPORT_SYMBOL(dm_table_get_size);
1357
1358 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1359 {
1360         if (index >= t->num_targets)
1361                 return NULL;
1362
1363         return t->targets + index;
1364 }
1365
1366 /*
1367  * Search the btree for the correct target.
1368  *
1369  * Caller should check returned pointer with dm_target_is_valid()
1370  * to trap I/O beyond end of device.
1371  */
1372 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1373 {
1374         unsigned int l, n = 0, k = 0;
1375         sector_t *node;
1376
1377         for (l = 0; l < t->depth; l++) {
1378                 n = get_child(n, k);
1379                 node = get_node(t, l, n);
1380
1381                 for (k = 0; k < KEYS_PER_NODE; k++)
1382                         if (node[k] >= sector)
1383                                 break;
1384         }
1385
1386         return &t->targets[(KEYS_PER_NODE * n) + k];
1387 }
1388
1389 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1390                         sector_t start, sector_t len, void *data)
1391 {
1392         unsigned *num_devices = data;
1393
1394         (*num_devices)++;
1395
1396         return 0;
1397 }
1398
1399 /*
1400  * Check whether a table has no data devices attached using each
1401  * target's iterate_devices method.
1402  * Returns false if the result is unknown because a target doesn't
1403  * support iterate_devices.
1404  */
1405 bool dm_table_has_no_data_devices(struct dm_table *table)
1406 {
1407         struct dm_target *ti;
1408         unsigned i, num_devices;
1409
1410         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1411                 ti = dm_table_get_target(table, i);
1412
1413                 if (!ti->type->iterate_devices)
1414                         return false;
1415
1416                 num_devices = 0;
1417                 ti->type->iterate_devices(ti, count_device, &num_devices);
1418                 if (num_devices)
1419                         return false;
1420         }
1421
1422         return true;
1423 }
1424
1425 static int device_is_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1426                                  sector_t start, sector_t len, void *data)
1427 {
1428         struct request_queue *q = bdev_get_queue(dev->bdev);
1429         enum blk_zoned_model *zoned_model = data;
1430
1431         return q && blk_queue_zoned_model(q) == *zoned_model;
1432 }
1433
1434 static bool dm_table_supports_zoned_model(struct dm_table *t,
1435                                           enum blk_zoned_model zoned_model)
1436 {
1437         struct dm_target *ti;
1438         unsigned i;
1439
1440         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1441                 ti = dm_table_get_target(t, i);
1442
1443                 if (zoned_model == BLK_ZONED_HM &&
1444                     !dm_target_supports_zoned_hm(ti->type))
1445                         return false;
1446
1447                 if (!ti->type->iterate_devices ||
1448                     !ti->type->iterate_devices(ti, device_is_zoned_model, &zoned_model))
1449                         return false;
1450         }
1451
1452         return true;
1453 }
1454
1455 static int device_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1456                                        sector_t start, sector_t len, void *data)
1457 {
1458         struct request_queue *q = bdev_get_queue(dev->bdev);
1459         unsigned int *zone_sectors = data;
1460
1461         return q && blk_queue_zone_sectors(q) == *zone_sectors;
1462 }
1463
1464 static bool dm_table_matches_zone_sectors(struct dm_table *t,
1465                                           unsigned int zone_sectors)
1466 {
1467         struct dm_target *ti;
1468         unsigned i;
1469
1470         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1471                 ti = dm_table_get_target(t, i);
1472
1473                 if (!ti->type->iterate_devices ||
1474                     !ti->type->iterate_devices(ti, device_matches_zone_sectors, &zone_sectors))
1475                         return false;
1476         }
1477
1478         return true;
1479 }
1480
1481 static int validate_hardware_zoned_model(struct dm_table *table,
1482                                          enum blk_zoned_model zoned_model,
1483                                          unsigned int zone_sectors)
1484 {
1485         if (zoned_model == BLK_ZONED_NONE)
1486                 return 0;
1487
1488         if (!dm_table_supports_zoned_model(table, zoned_model)) {
1489                 DMERR("%s: zoned model is not consistent across all devices",
1490                       dm_device_name(table->md));
1491                 return -EINVAL;
1492         }
1493
1494         /* Check zone size validity and compatibility */
1495         if (!zone_sectors || !is_power_of_2(zone_sectors))
1496                 return -EINVAL;
1497
1498         if (!dm_table_matches_zone_sectors(table, zone_sectors)) {
1499                 DMERR("%s: zone sectors is not consistent across all devices",
1500                       dm_device_name(table->md));
1501                 return -EINVAL;
1502         }
1503
1504         return 0;
1505 }
1506
1507 /*
1508  * Establish the new table's queue_limits and validate them.
1509  */
1510 int dm_calculate_queue_limits(struct dm_table *table,
1511                               struct queue_limits *limits)
1512 {
1513         struct dm_target *ti;
1514         struct queue_limits ti_limits;
1515         unsigned i;
1516         enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1517         unsigned int zone_sectors = 0;
1518
1519         blk_set_stacking_limits(limits);
1520
1521         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1522                 blk_set_stacking_limits(&ti_limits);
1523
1524                 ti = dm_table_get_target(table, i);
1525
1526                 if (!ti->type->iterate_devices)
1527                         goto combine_limits;
1528
1529                 /*
1530                  * Combine queue limits of all the devices this target uses.
1531                  */
1532                 ti->type->iterate_devices(ti, dm_set_device_limits,
1533                                           &ti_limits);
1534
1535                 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1536                         /*
1537                          * After stacking all limits, validate all devices
1538                          * in table support this zoned model and zone sectors.
1539                          */
1540                         zoned_model = ti_limits.zoned;
1541                         zone_sectors = ti_limits.chunk_sectors;
1542                 }
1543
1544                 /* Set I/O hints portion of queue limits */
1545                 if (ti->type->io_hints)
1546                         ti->type->io_hints(ti, &ti_limits);
1547
1548                 /*
1549                  * Check each device area is consistent with the target's
1550                  * overall queue limits.
1551                  */
1552                 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1553                                               &ti_limits))
1554                         return -EINVAL;
1555
1556 combine_limits:
1557                 /*
1558                  * Merge this target's queue limits into the overall limits
1559                  * for the table.
1560                  */
1561                 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1562                         DMWARN("%s: adding target device "
1563                                "(start sect %llu len %llu) "
1564                                "caused an alignment inconsistency",
1565                                dm_device_name(table->md),
1566                                (unsigned long long) ti->begin,
1567                                (unsigned long long) ti->len);
1568
1569                 /*
1570                  * FIXME: this should likely be moved to blk_stack_limits(), would
1571                  * also eliminate limits->zoned stacking hack in dm_set_device_limits()
1572                  */
1573                 if (limits->zoned == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1574                         /*
1575                          * By default, the stacked limits zoned model is set to
1576                          * BLK_ZONED_NONE in blk_set_stacking_limits(). Update
1577                          * this model using the first target model reported
1578                          * that is not BLK_ZONED_NONE. This will be either the
1579                          * first target device zoned model or the model reported
1580                          * by the target .io_hints.
1581                          */
1582                         limits->zoned = ti_limits.zoned;
1583                 }
1584         }
1585
1586         /*
1587          * Verify that the zoned model and zone sectors, as determined before
1588          * any .io_hints override, are the same across all devices in the table.
1589          * - this is especially relevant if .io_hints is emulating a disk-managed
1590          *   zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1591          * BUT...
1592          */
1593         if (limits->zoned != BLK_ZONED_NONE) {
1594                 /*
1595                  * ...IF the above limits stacking determined a zoned model
1596                  * validate that all of the table's devices conform to it.
1597                  */
1598                 zoned_model = limits->zoned;
1599                 zone_sectors = limits->chunk_sectors;
1600         }
1601         if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1602                 return -EINVAL;
1603
1604         return validate_hardware_logical_block_alignment(table, limits);
1605 }
1606
1607 /*
1608  * Verify that all devices have an integrity profile that matches the
1609  * DM device's registered integrity profile.  If the profiles don't
1610  * match then unregister the DM device's integrity profile.
1611  */
1612 static void dm_table_verify_integrity(struct dm_table *t)
1613 {
1614         struct gendisk *template_disk = NULL;
1615
1616         if (t->integrity_added)
1617                 return;
1618
1619         if (t->integrity_supported) {
1620                 /*
1621                  * Verify that the original integrity profile
1622                  * matches all the devices in this table.
1623                  */
1624                 template_disk = dm_table_get_integrity_disk(t);
1625                 if (template_disk &&
1626                     blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1627                         return;
1628         }
1629
1630         if (integrity_profile_exists(dm_disk(t->md))) {
1631                 DMWARN("%s: unable to establish an integrity profile",
1632                        dm_device_name(t->md));
1633                 blk_integrity_unregister(dm_disk(t->md));
1634         }
1635 }
1636
1637 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1638                                 sector_t start, sector_t len, void *data)
1639 {
1640         unsigned long flush = (unsigned long) data;
1641         struct request_queue *q = bdev_get_queue(dev->bdev);
1642
1643         return q && (q->queue_flags & flush);
1644 }
1645
1646 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1647 {
1648         struct dm_target *ti;
1649         unsigned i;
1650
1651         /*
1652          * Require at least one underlying device to support flushes.
1653          * t->devices includes internal dm devices such as mirror logs
1654          * so we need to use iterate_devices here, which targets
1655          * supporting flushes must provide.
1656          */
1657         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1658                 ti = dm_table_get_target(t, i);
1659
1660                 if (!ti->num_flush_bios)
1661                         continue;
1662
1663                 if (ti->flush_supported)
1664                         return true;
1665
1666                 if (ti->type->iterate_devices &&
1667                     ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1668                         return true;
1669         }
1670
1671         return false;
1672 }
1673
1674 static int device_dax_write_cache_enabled(struct dm_target *ti,
1675                                           struct dm_dev *dev, sector_t start,
1676                                           sector_t len, void *data)
1677 {
1678         struct dax_device *dax_dev = dev->dax_dev;
1679
1680         if (!dax_dev)
1681                 return false;
1682
1683         if (dax_write_cache_enabled(dax_dev))
1684                 return true;
1685         return false;
1686 }
1687
1688 static int dm_table_supports_dax_write_cache(struct dm_table *t)
1689 {
1690         struct dm_target *ti;
1691         unsigned i;
1692
1693         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1694                 ti = dm_table_get_target(t, i);
1695
1696                 if (ti->type->iterate_devices &&
1697                     ti->type->iterate_devices(ti,
1698                                 device_dax_write_cache_enabled, NULL))
1699                         return true;
1700         }
1701
1702         return false;
1703 }
1704
1705 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1706                             sector_t start, sector_t len, void *data)
1707 {
1708         struct request_queue *q = bdev_get_queue(dev->bdev);
1709
1710         return q && blk_queue_nonrot(q);
1711 }
1712
1713 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1714                              sector_t start, sector_t len, void *data)
1715 {
1716         struct request_queue *q = bdev_get_queue(dev->bdev);
1717
1718         return q && !blk_queue_add_random(q);
1719 }
1720
1721 static int queue_supports_sg_merge(struct dm_target *ti, struct dm_dev *dev,
1722                                    sector_t start, sector_t len, void *data)
1723 {
1724         struct request_queue *q = bdev_get_queue(dev->bdev);
1725
1726         return q && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1727 }
1728
1729 static bool dm_table_all_devices_attribute(struct dm_table *t,
1730                                            iterate_devices_callout_fn func)
1731 {
1732         struct dm_target *ti;
1733         unsigned i;
1734
1735         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1736                 ti = dm_table_get_target(t, i);
1737
1738                 if (!ti->type->iterate_devices ||
1739                     !ti->type->iterate_devices(ti, func, NULL))
1740                         return false;
1741         }
1742
1743         return true;
1744 }
1745
1746 static int device_no_partial_completion(struct dm_target *ti, struct dm_dev *dev,
1747                                         sector_t start, sector_t len, void *data)
1748 {
1749         char b[BDEVNAME_SIZE];
1750
1751         /* For now, NVMe devices are the only devices of this class */
1752         return (strncmp(bdevname(dev->bdev, b), "nvme", 4) == 0);
1753 }
1754
1755 static bool dm_table_does_not_support_partial_completion(struct dm_table *t)
1756 {
1757         return dm_table_all_devices_attribute(t, device_no_partial_completion);
1758 }
1759
1760 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1761                                          sector_t start, sector_t len, void *data)
1762 {
1763         struct request_queue *q = bdev_get_queue(dev->bdev);
1764
1765         return q && !q->limits.max_write_same_sectors;
1766 }
1767
1768 static bool dm_table_supports_write_same(struct dm_table *t)
1769 {
1770         struct dm_target *ti;
1771         unsigned i;
1772
1773         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1774                 ti = dm_table_get_target(t, i);
1775
1776                 if (!ti->num_write_same_bios)
1777                         return false;
1778
1779                 if (!ti->type->iterate_devices ||
1780                     ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1781                         return false;
1782         }
1783
1784         return true;
1785 }
1786
1787 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1788                                            sector_t start, sector_t len, void *data)
1789 {
1790         struct request_queue *q = bdev_get_queue(dev->bdev);
1791
1792         return q && !q->limits.max_write_zeroes_sectors;
1793 }
1794
1795 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1796 {
1797         struct dm_target *ti;
1798         unsigned i = 0;
1799
1800         while (i < dm_table_get_num_targets(t)) {
1801                 ti = dm_table_get_target(t, i++);
1802
1803                 if (!ti->num_write_zeroes_bios)
1804                         return false;
1805
1806                 if (!ti->type->iterate_devices ||
1807                     ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1808                         return false;
1809         }
1810
1811         return true;
1812 }
1813
1814 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1815                                       sector_t start, sector_t len, void *data)
1816 {
1817         struct request_queue *q = bdev_get_queue(dev->bdev);
1818
1819         return q && !blk_queue_discard(q);
1820 }
1821
1822 static bool dm_table_supports_discards(struct dm_table *t)
1823 {
1824         struct dm_target *ti;
1825         unsigned i;
1826
1827         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1828                 ti = dm_table_get_target(t, i);
1829
1830                 if (!ti->num_discard_bios)
1831                         return false;
1832
1833                 /*
1834                  * Either the target provides discard support (as implied by setting
1835                  * 'discards_supported') or it relies on _all_ data devices having
1836                  * discard support.
1837                  */
1838                 if (!ti->discards_supported &&
1839                     (!ti->type->iterate_devices ||
1840                      ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1841                         return false;
1842         }
1843
1844         return true;
1845 }
1846
1847 static int device_not_secure_erase_capable(struct dm_target *ti,
1848                                            struct dm_dev *dev, sector_t start,
1849                                            sector_t len, void *data)
1850 {
1851         struct request_queue *q = bdev_get_queue(dev->bdev);
1852
1853         return q && !blk_queue_secure_erase(q);
1854 }
1855
1856 static bool dm_table_supports_secure_erase(struct dm_table *t)
1857 {
1858         struct dm_target *ti;
1859         unsigned int i;
1860
1861         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1862                 ti = dm_table_get_target(t, i);
1863
1864                 if (!ti->num_secure_erase_bios)
1865                         return false;
1866
1867                 if (!ti->type->iterate_devices ||
1868                     ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1869                         return false;
1870         }
1871
1872         return true;
1873 }
1874
1875 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1876                                struct queue_limits *limits)
1877 {
1878         bool wc = false, fua = false;
1879
1880         /*
1881          * Copy table's limits to the DM device's request_queue
1882          */
1883         q->limits = *limits;
1884
1885         if (!dm_table_supports_discards(t)) {
1886                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
1887                 /* Must also clear discard limits... */
1888                 q->limits.max_discard_sectors = 0;
1889                 q->limits.max_hw_discard_sectors = 0;
1890                 q->limits.discard_granularity = 0;
1891                 q->limits.discard_alignment = 0;
1892                 q->limits.discard_misaligned = 0;
1893         } else
1894                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
1895
1896         if (dm_table_supports_secure_erase(t))
1897                 blk_queue_flag_set(QUEUE_FLAG_SECERASE, q);
1898
1899         if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1900                 wc = true;
1901                 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1902                         fua = true;
1903         }
1904         blk_queue_write_cache(q, wc, fua);
1905
1906         if (dm_table_supports_dax(t))
1907                 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
1908         else
1909                 blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
1910
1911         if (dm_table_supports_dax_write_cache(t))
1912                 dax_write_cache(t->md->dax_dev, true);
1913
1914         /* Ensure that all underlying devices are non-rotational. */
1915         if (dm_table_all_devices_attribute(t, device_is_nonrot))
1916                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
1917         else
1918                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
1919
1920         if (!dm_table_supports_write_same(t))
1921                 q->limits.max_write_same_sectors = 0;
1922         if (!dm_table_supports_write_zeroes(t))
1923                 q->limits.max_write_zeroes_sectors = 0;
1924
1925         if (dm_table_all_devices_attribute(t, queue_supports_sg_merge))
1926                 blk_queue_flag_clear(QUEUE_FLAG_NO_SG_MERGE, q);
1927         else
1928                 blk_queue_flag_set(QUEUE_FLAG_NO_SG_MERGE, q);
1929
1930         dm_table_verify_integrity(t);
1931
1932         /*
1933          * Determine whether or not this queue's I/O timings contribute
1934          * to the entropy pool, Only request-based targets use this.
1935          * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1936          * have it set.
1937          */
1938         if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1939                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
1940 }
1941
1942 unsigned int dm_table_get_num_targets(struct dm_table *t)
1943 {
1944         return t->num_targets;
1945 }
1946
1947 struct list_head *dm_table_get_devices(struct dm_table *t)
1948 {
1949         return &t->devices;
1950 }
1951
1952 fmode_t dm_table_get_mode(struct dm_table *t)
1953 {
1954         return t->mode;
1955 }
1956 EXPORT_SYMBOL(dm_table_get_mode);
1957
1958 enum suspend_mode {
1959         PRESUSPEND,
1960         PRESUSPEND_UNDO,
1961         POSTSUSPEND,
1962 };
1963
1964 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1965 {
1966         int i = t->num_targets;
1967         struct dm_target *ti = t->targets;
1968
1969         lockdep_assert_held(&t->md->suspend_lock);
1970
1971         while (i--) {
1972                 switch (mode) {
1973                 case PRESUSPEND:
1974                         if (ti->type->presuspend)
1975                                 ti->type->presuspend(ti);
1976                         break;
1977                 case PRESUSPEND_UNDO:
1978                         if (ti->type->presuspend_undo)
1979                                 ti->type->presuspend_undo(ti);
1980                         break;
1981                 case POSTSUSPEND:
1982                         if (ti->type->postsuspend)
1983                                 ti->type->postsuspend(ti);
1984                         break;
1985                 }
1986                 ti++;
1987         }
1988 }
1989
1990 void dm_table_presuspend_targets(struct dm_table *t)
1991 {
1992         if (!t)
1993                 return;
1994
1995         suspend_targets(t, PRESUSPEND);
1996 }
1997
1998 void dm_table_presuspend_undo_targets(struct dm_table *t)
1999 {
2000         if (!t)
2001                 return;
2002
2003         suspend_targets(t, PRESUSPEND_UNDO);
2004 }
2005
2006 void dm_table_postsuspend_targets(struct dm_table *t)
2007 {
2008         if (!t)
2009                 return;
2010
2011         suspend_targets(t, POSTSUSPEND);
2012 }
2013
2014 int dm_table_resume_targets(struct dm_table *t)
2015 {
2016         int i, r = 0;
2017
2018         lockdep_assert_held(&t->md->suspend_lock);
2019
2020         for (i = 0; i < t->num_targets; i++) {
2021                 struct dm_target *ti = t->targets + i;
2022
2023                 if (!ti->type->preresume)
2024                         continue;
2025
2026                 r = ti->type->preresume(ti);
2027                 if (r) {
2028                         DMERR("%s: %s: preresume failed, error = %d",
2029                               dm_device_name(t->md), ti->type->name, r);
2030                         return r;
2031                 }
2032         }
2033
2034         for (i = 0; i < t->num_targets; i++) {
2035                 struct dm_target *ti = t->targets + i;
2036
2037                 if (ti->type->resume)
2038                         ti->type->resume(ti);
2039         }
2040
2041         return 0;
2042 }
2043
2044 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
2045 {
2046         list_add(&cb->list, &t->target_callbacks);
2047 }
2048 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
2049
2050 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
2051 {
2052         struct dm_dev_internal *dd;
2053         struct list_head *devices = dm_table_get_devices(t);
2054         struct dm_target_callbacks *cb;
2055         int r = 0;
2056
2057         list_for_each_entry(dd, devices, list) {
2058                 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
2059                 char b[BDEVNAME_SIZE];
2060
2061                 if (likely(q))
2062                         r |= bdi_congested(q->backing_dev_info, bdi_bits);
2063                 else
2064                         DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
2065                                      dm_device_name(t->md),
2066                                      bdevname(dd->dm_dev->bdev, b));
2067         }
2068
2069         list_for_each_entry(cb, &t->target_callbacks, list)
2070                 if (cb->congested_fn)
2071                         r |= cb->congested_fn(cb, bdi_bits);
2072
2073         return r;
2074 }
2075
2076 struct mapped_device *dm_table_get_md(struct dm_table *t)
2077 {
2078         return t->md;
2079 }
2080 EXPORT_SYMBOL(dm_table_get_md);
2081
2082 void dm_table_run_md_queue_async(struct dm_table *t)
2083 {
2084         struct mapped_device *md;
2085         struct request_queue *queue;
2086         unsigned long flags;
2087
2088         if (!dm_table_request_based(t))
2089                 return;
2090
2091         md = dm_table_get_md(t);
2092         queue = dm_get_md_queue(md);
2093         if (queue) {
2094                 if (queue->mq_ops)
2095                         blk_mq_run_hw_queues(queue, true);
2096                 else {
2097                         spin_lock_irqsave(queue->queue_lock, flags);
2098                         blk_run_queue_async(queue);
2099                         spin_unlock_irqrestore(queue->queue_lock, flags);
2100                 }
2101         }
2102 }
2103 EXPORT_SYMBOL(dm_table_run_md_queue_async);
2104