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 NODE_SIZE L1_CACHE_BYTES
  28 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
  29 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
  30
  31 /*
  32  * Similar to ceiling(log_size(n))
  33  */
  34 static unsigned int int_log(unsigned int n, unsigned int base)
  35 {
  36         int result = 0;
  37
  38         while (n > 1) {
  39                 n = dm_div_up(n, base);
  40                 result++;
  41         }
  42
  43         return result;
  44 }
  45
  46 /*
  47  * Calculate the index of the child node of the n'th node k'th key.
  48  */
  49 static inline unsigned int get_child(unsigned int n, unsigned int k)
  50 {
  51         return (n * CHILDREN_PER_NODE) + k;
  52 }
  53
  54 /*
  55  * Return the n'th node of level l from table t.
  56  */
  57 static inline sector_t *get_node(struct dm_table *t,
  58                                  unsigned int l, unsigned int n)
  59 {
  60         return t->index[l] + (n * KEYS_PER_NODE);
  61 }
  62
  63 /*
  64  * Return the highest key that you could lookup from the n'th
  65  * node on level l of the btree.
  66  */
  67 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
  68 {
  69         for (; l < t->depth - 1; l++)
  70                 n = get_child(n, CHILDREN_PER_NODE - 1);
  71
  72         if (n >= t->counts[l])
  73                 return (sector_t) - 1;
  74
  75         return get_node(t, l, n)[KEYS_PER_NODE - 1];
  76 }
  77
  78 /*
  79  * Fills in a level of the btree based on the highs of the level
  80  * below it.
  81  */
  82 static int setup_btree_index(unsigned int l, struct dm_table *t)
  83 {
  84         unsigned int n, k;
  85         sector_t *node;
  86
  87         for (n = 0U; n < t->counts[l]; n++) {
  88                 node = get_node(t, l, n);
  89
  90                 for (k = 0U; k < KEYS_PER_NODE; k++)
  91                         node[k] = high(t, l + 1, get_child(n, k));
  92         }
  93
  94         return 0;
  95 }
  96
  97 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
  98 {
  99         unsigned long size;
 100         void *addr;
 101
 102         /*
 103          * Check that we're not going to overflow.
 104          */
 105         if (nmemb > (ULONG_MAX / elem_size))
 106                 return NULL;
 107
 108         size = nmemb * elem_size;
 109         addr = vzalloc(size);
 110
 111         return addr;
 112 }
 113 EXPORT_SYMBOL(dm_vcalloc);
 114
 115 /*
 116  * highs, and targets are managed as dynamic arrays during a
 117  * table load.
 118  */
 119 static int alloc_targets(struct dm_table *t, unsigned int num)
 120 {
 121         sector_t *n_highs;
 122         struct dm_target *n_targets;
 123
 124         /*
 125          * Allocate both the target array and offset array at once.
 126          */
 127         n_highs = (sector_t *) dm_vcalloc(num, sizeof(struct dm_target) +
 128                                           sizeof(sector_t));
 129         if (!n_highs)
 130                 return -ENOMEM;
 131
 132         n_targets = (struct dm_target *) (n_highs + num);
 133
 134         memset(n_highs, -1, sizeof(*n_highs) * num);
 135         vfree(t->highs);
 136
 137         t->num_allocated = num;
 138         t->highs = n_highs;
 139         t->targets = n_targets;
 140
 141         return 0;
 142 }
 143
 144 int dm_table_create(struct dm_table **result, fmode_t mode,
 145                     unsigned num_targets, struct mapped_device *md)
 146 {
 147         struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
 148
 149         if (!t)
 150                 return -ENOMEM;
 151
 152         INIT_LIST_HEAD(&t->devices);
 153
 154         if (!num_targets)
 155                 num_targets = KEYS_PER_NODE;
 156
 157         num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
 158
 159         if (!num_targets) {
 160                 kfree(t);
 161                 return -ENOMEM;
 162         }
 163
 164         if (alloc_targets(t, num_targets)) {
 165                 kfree(t);
 166                 return -ENOMEM;
 167         }
 168
 169         t->type = DM_TYPE_NONE;
 170         t->mode = mode;
 171         t->md = md;
 172         *result = t;
 173         return 0;
 174 }
 175
 176 static void free_devices(struct list_head *devices, struct mapped_device *md)
 177 {
 178         struct list_head *tmp, *next;
 179
 180         list_for_each_safe(tmp, next, devices) {
 181                 struct dm_dev_internal *dd =
 182                     list_entry(tmp, struct dm_dev_internal, list);
 183                 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
 184                        dm_device_name(md), dd->dm_dev->name);
 185                 dm_put_table_device(md, dd->dm_dev);
 186                 kfree(dd);
 187         }
 188 }
 189
 190 void dm_table_destroy(struct dm_table *t)
 191 {
 192         unsigned int i;
 193
 194         if (!t)
 195                 return;
 196
 197         /* free the indexes */
 198         if (t->depth >= 2)
 199                 vfree(t->index[t->depth - 2]);
 200
 201         /* free the targets */
 202         for (i = 0; i < t->num_targets; i++) {
 203                 struct dm_target *tgt = t->targets + i;
 204
 205                 if (tgt->type->dtr)
 206                         tgt->type->dtr(tgt);
 207
 208                 dm_put_target_type(tgt->type);
 209         }
 210
 211         vfree(t->highs);
 212
 213         /* free the device list */
 214         free_devices(&t->devices, t->md);
 215
 216         dm_free_md_mempools(t->mempools);
 217
 218         kfree(t);
 219 }
 220
 221 /*
 222  * See if we've already got a device in the list.
 223  */
 224 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
 225 {
 226         struct dm_dev_internal *dd;
 227
 228         list_for_each_entry (dd, l, list)
 229                 if (dd->dm_dev->bdev->bd_dev == dev)
 230                         return dd;
 231
 232         return NULL;
 233 }
 234
 235 /*
 236  * If possible, this checks an area of a destination device is invalid.
 237  */
 238 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
 239                                   sector_t start, sector_t len, void *data)
 240 {
 241         struct queue_limits *limits = data;
 242         struct block_device *bdev = dev->bdev;
 243         sector_t dev_size =
 244                 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
 245         unsigned short logical_block_size_sectors =
 246                 limits->logical_block_size >> SECTOR_SHIFT;
 247         char b[BDEVNAME_SIZE];
 248
 249         if (!dev_size)
 250                 return 0;
 251
 252         if ((start >= dev_size) || (start + len > dev_size)) {
 253                 DMWARN("%s: %s too small for target: "
 254                        "start=%llu, len=%llu, dev_size=%llu",
 255                        dm_device_name(ti->table->md), bdevname(bdev, b),
 256                        (unsigned long long)start,
 257                        (unsigned long long)len,
 258                        (unsigned long long)dev_size);
 259                 return 1;
 260         }
 261
 262         /*
 263          * If the target is mapped to zoned block device(s), check
 264          * that the zones are not partially mapped.
 265          */
 266         if (bdev_zoned_model(bdev) != BLK_ZONED_NONE) {
 267                 unsigned int zone_sectors = bdev_zone_sectors(bdev);
 268
 269                 if (start & (zone_sectors - 1)) {
 270                         DMWARN("%s: start=%llu not aligned to h/w zone size %u of %s",
 271                                dm_device_name(ti->table->md),
 272                                (unsigned long long)start,
 273                                zone_sectors, bdevname(bdev, b));
 274                         return 1;
 275                 }
 276
 277                 /*
 278                  * Note: The last zone of a zoned block device may be smaller
 279                  * than other zones. So for a target mapping the end of a
 280                  * zoned block device with such a zone, len would not be zone
 281                  * aligned. We do not allow such last smaller zone to be part
 282                  * of the mapping here to ensure that mappings with multiple
 283                  * devices do not end up with a smaller zone in the middle of
 284                  * the sector range.
 285                  */
 286                 if (len & (zone_sectors - 1)) {
 287                         DMWARN("%s: len=%llu not aligned to h/w zone size %u of %s",
 288                                dm_device_name(ti->table->md),
 289                                (unsigned long long)len,
 290                                zone_sectors, bdevname(bdev, b));
 291                         return 1;
 292                 }
 293         }
 294
 295         if (logical_block_size_sectors <= 1)
 296                 return 0;
 297
 298         if (start & (logical_block_size_sectors - 1)) {
 299                 DMWARN("%s: start=%llu not aligned to h/w "
 300                        "logical block size %u of %s",
 301                        dm_device_name(ti->table->md),
 302                        (unsigned long long)start,
 303                        limits->logical_block_size, bdevname(bdev, b));
 304                 return 1;
 305         }
 306
 307         if (len & (logical_block_size_sectors - 1)) {
 308                 DMWARN("%s: len=%llu not aligned to h/w "
 309                        "logical block size %u of %s",
 310                        dm_device_name(ti->table->md),
 311                        (unsigned long long)len,
 312                        limits->logical_block_size, bdevname(bdev, b));
 313                 return 1;
 314         }
 315
 316         return 0;
 317 }
 318
 319 /*
 320  * This upgrades the mode on an already open dm_dev, being
 321  * careful to leave things as they were if we fail to reopen the
 322  * device and not to touch the existing bdev field in case
 323  * it is accessed concurrently.
 324  */
 325 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
 326                         struct mapped_device *md)
 327 {
 328         int r;
 329         struct dm_dev *old_dev, *new_dev;
 330
 331         old_dev = dd->dm_dev;
 332
 333         r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
 334                                 dd->dm_dev->mode | new_mode, &new_dev);
 335         if (r)
 336                 return r;
 337
 338         dd->dm_dev = new_dev;
 339         dm_put_table_device(md, old_dev);
 340
 341         return 0;
 342 }
 343
 344 /*
 345  * Convert the path to a device
 346  */
 347 dev_t dm_get_dev_t(const char *path)
 348 {
 349         dev_t dev;
 350
 351         if (lookup_bdev(path, &dev))
 352                 dev = name_to_dev_t(path);
 353         return dev;
 354 }
 355 EXPORT_SYMBOL_GPL(dm_get_dev_t);
 356
 357 /*
 358  * Add a device to the list, or just increment the usage count if
 359  * it's already present.
 360  */
 361 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
 362                   struct dm_dev **result)
 363 {
 364         int r;
 365         dev_t dev;
 366         unsigned int major, minor;
 367         char dummy;
 368         struct dm_dev_internal *dd;
 369         struct dm_table *t = ti->table;
 370
 371         BUG_ON(!t);
 372
 373         if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
 374                 /* Extract the major/minor numbers */
 375                 dev = MKDEV(major, minor);
 376                 if (MAJOR(dev) != major || MINOR(dev) != minor)
 377                         return -EOVERFLOW;
 378         } else {
 379                 dev = dm_get_dev_t(path);
 380                 if (!dev)
 381                         return -ENODEV;
 382         }
 383
 384         dd = find_device(&t->devices, dev);
 385         if (!dd) {
 386                 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
 387                 if (!dd)
 388                         return -ENOMEM;
 389
 390                 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
 391                         kfree(dd);
 392                         return r;
 393                 }
 394
 395                 refcount_set(&dd->count, 1);
 396                 list_add(&dd->list, &t->devices);
 397                 goto out;
 398
 399         } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
 400                 r = upgrade_mode(dd, mode, t->md);
 401                 if (r)
 402                         return r;
 403         }
 404         refcount_inc(&dd->count);
 405 out:
 406         *result = dd->dm_dev;
 407         return 0;
 408 }
 409 EXPORT_SYMBOL(dm_get_device);
 410
 411 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
 412                                 sector_t start, sector_t len, void *data)
 413 {
 414         struct queue_limits *limits = data;
 415         struct block_device *bdev = dev->bdev;
 416         struct request_queue *q = bdev_get_queue(bdev);
 417         char b[BDEVNAME_SIZE];
 418
 419         if (unlikely(!q)) {
 420                 DMWARN("%s: Cannot set limits for nonexistent device %s",
 421                        dm_device_name(ti->table->md), bdevname(bdev, b));
 422                 return 0;
 423         }
 424
 425         if (blk_stack_limits(limits, &q->limits,
 426                         get_start_sect(bdev) + start) < 0)
 427                 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
 428                        "physical_block_size=%u, logical_block_size=%u, "
 429                        "alignment_offset=%u, start=%llu",
 430                        dm_device_name(ti->table->md), bdevname(bdev, b),
 431                        q->limits.physical_block_size,
 432                        q->limits.logical_block_size,
 433                        q->limits.alignment_offset,
 434                        (unsigned long long) start << SECTOR_SHIFT);
 435         return 0;
 436 }
 437
 438 /*
 439  * Decrement a device's use count and remove it if necessary.
 440  */
 441 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
 442 {
 443         int found = 0;
 444         struct list_head *devices = &ti->table->devices;
 445         struct dm_dev_internal *dd;
 446
 447         list_for_each_entry(dd, devices, list) {
 448                 if (dd->dm_dev == d) {
 449                         found = 1;
 450                         break;
 451                 }
 452         }
 453         if (!found) {
 454                 DMWARN("%s: device %s not in table devices list",
 455                        dm_device_name(ti->table->md), d->name);
 456                 return;
 457         }
 458         if (refcount_dec_and_test(&dd->count)) {
 459                 dm_put_table_device(ti->table->md, d);
 460                 list_del(&dd->list);
 461                 kfree(dd);
 462         }
 463 }
 464 EXPORT_SYMBOL(dm_put_device);
 465
 466 /*
 467  * Checks to see if the target joins onto the end of the table.
 468  */
 469 static int adjoin(struct dm_table *table, struct dm_target *ti)
 470 {
 471         struct dm_target *prev;
 472
 473         if (!table->num_targets)
 474                 return !ti->begin;
 475
 476         prev = &table->targets[table->num_targets - 1];
 477         return (ti->begin == (prev->begin + prev->len));
 478 }
 479
 480 /*
 481  * Used to dynamically allocate the arg array.
 482  *
 483  * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
 484  * process messages even if some device is suspended. These messages have a
 485  * small fixed number of arguments.
 486  *
 487  * On the other hand, dm-switch needs to process bulk data using messages and
 488  * excessive use of GFP_NOIO could cause trouble.
 489  */
 490 static char **realloc_argv(unsigned *size, char **old_argv)
 491 {
 492         char **argv;
 493         unsigned new_size;
 494         gfp_t gfp;
 495
 496         if (*size) {
 497                 new_size = *size * 2;
 498                 gfp = GFP_KERNEL;
 499         } else {
 500                 new_size = 8;
 501                 gfp = GFP_NOIO;
 502         }
 503         argv = kmalloc_array(new_size, sizeof(*argv), gfp);
 504         if (argv && old_argv) {
 505                 memcpy(argv, old_argv, *size * sizeof(*argv));
 506                 *size = new_size;
 507         }
 508
 509         kfree(old_argv);
 510         return argv;
 511 }
 512
 513 /*
 514  * Destructively splits up the argument list to pass to ctr.
 515  */
 516 int dm_split_args(int *argc, char ***argvp, char *input)
 517 {
 518         char *start, *end = input, *out, **argv = NULL;
 519         unsigned array_size = 0;
 520
 521         *argc = 0;
 522
 523         if (!input) {
 524                 *argvp = NULL;
 525                 return 0;
 526         }
 527
 528         argv = realloc_argv(&array_size, argv);
 529         if (!argv)
 530                 return -ENOMEM;
 531
 532         while (1) {
 533                 /* Skip whitespace */
 534                 start = skip_spaces(end);
 535
 536                 if (!*start)
 537                         break;  /* success, we hit the end */
 538
 539                 /* 'out' is used to remove any back-quotes */
 540                 end = out = start;
 541                 while (*end) {
 542                         /* Everything apart from '\0' can be quoted */
 543                         if (*end == '\\' && *(end + 1)) {
 544                                 *out++ = *(end + 1);
 545                                 end += 2;
 546                                 continue;
 547                         }
 548
 549                         if (isspace(*end))
 550                                 break;  /* end of token */
 551
 552                         *out++ = *end++;
 553                 }
 554
 555                 /* have we already filled the array ? */
 556                 if ((*argc + 1) > array_size) {
 557                         argv = realloc_argv(&array_size, argv);
 558                         if (!argv)
 559                                 return -ENOMEM;
 560                 }
 561
 562                 /* we know this is whitespace */
 563                 if (*end)
 564                         end++;
 565
 566                 /* terminate the string and put it in the array */
 567                 *out = '\0';
 568                 argv[*argc] = start;
 569                 (*argc)++;
 570         }
 571
 572         *argvp = argv;
 573         return 0;
 574 }
 575
 576 /*
 577  * Impose necessary and sufficient conditions on a devices's table such
 578  * that any incoming bio which respects its logical_block_size can be
 579  * processed successfully.  If it falls across the boundary between
 580  * two or more targets, the size of each piece it gets split into must
 581  * be compatible with the logical_block_size of the target processing it.
 582  */
 583 static int validate_hardware_logical_block_alignment(struct dm_table *table,
 584                                                  struct queue_limits *limits)
 585 {
 586         /*
 587          * This function uses arithmetic modulo the logical_block_size
 588          * (in units of 512-byte sectors).
 589          */
 590         unsigned short device_logical_block_size_sects =
 591                 limits->logical_block_size >> SECTOR_SHIFT;
 592
 593         /*
 594          * Offset of the start of the next table entry, mod logical_block_size.
 595          */
 596         unsigned short next_target_start = 0;
 597
 598         /*
 599          * Given an aligned bio that extends beyond the end of a
 600          * target, how many sectors must the next target handle?
 601          */
 602         unsigned short remaining = 0;
 603
 604         struct dm_target *ti;
 605         struct queue_limits ti_limits;
 606         unsigned i;
 607
 608         /*
 609          * Check each entry in the table in turn.
 610          */
 611         for (i = 0; i < dm_table_get_num_targets(table); i++) {
 612                 ti = dm_table_get_target(table, i);
 613
 614                 blk_set_stacking_limits(&ti_limits);
 615
 616                 /* combine all target devices' limits */
 617                 if (ti->type->iterate_devices)
 618                         ti->type->iterate_devices(ti, dm_set_device_limits,
 619                                                   &ti_limits);
 620
 621                 /*
 622                  * If the remaining sectors fall entirely within this
 623                  * table entry are they compatible with its logical_block_size?
 624                  */
 625                 if (remaining < ti->len &&
 626                     remaining & ((ti_limits.logical_block_size >>
 627                                   SECTOR_SHIFT) - 1))
 628                         break;  /* Error */
 629
 630                 next_target_start =
 631                     (unsigned short) ((next_target_start + ti->len) &
 632                                       (device_logical_block_size_sects - 1));
 633                 remaining = next_target_start ?
 634                     device_logical_block_size_sects - next_target_start : 0;
 635         }
 636
 637         if (remaining) {
 638                 DMWARN("%s: table line %u (start sect %llu len %llu) "
 639                        "not aligned to h/w logical block size %u",
 640                        dm_device_name(table->md), i,
 641                        (unsigned long long) ti->begin,
 642                        (unsigned long long) ti->len,
 643                        limits->logical_block_size);
 644                 return -EINVAL;
 645         }
 646
 647         return 0;
 648 }
 649
 650 int dm_table_add_target(struct dm_table *t, const char *type,
 651                         sector_t start, sector_t len, char *params)
 652 {
 653         int r = -EINVAL, argc;
 654         char **argv;
 655         struct dm_target *tgt;
 656
 657         if (t->singleton) {
 658                 DMERR("%s: target type %s must appear alone in table",
 659                       dm_device_name(t->md), t->targets->type->name);
 660                 return -EINVAL;
 661         }
 662
 663         BUG_ON(t->num_targets >= t->num_allocated);
 664
 665         tgt = t->targets + t->num_targets;
 666         memset(tgt, 0, sizeof(*tgt));
 667
 668         if (!len) {
 669                 DMERR("%s: zero-length target", dm_device_name(t->md));
 670                 return -EINVAL;
 671         }
 672
 673         tgt->type = dm_get_target_type(type);
 674         if (!tgt->type) {
 675                 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
 676                 return -EINVAL;
 677         }
 678
 679         if (dm_target_needs_singleton(tgt->type)) {
 680                 if (t->num_targets) {
 681                         tgt->error = "singleton target type must appear alone in table";
 682                         goto bad;
 683                 }
 684                 t->singleton = true;
 685         }
 686
 687         if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
 688                 tgt->error = "target type may not be included in a read-only table";
 689                 goto bad;
 690         }
 691
 692         if (t->immutable_target_type) {
 693                 if (t->immutable_target_type != tgt->type) {
 694                         tgt->error = "immutable target type cannot be mixed with other target types";
 695                         goto bad;
 696                 }
 697         } else if (dm_target_is_immutable(tgt->type)) {
 698                 if (t->num_targets) {
 699                         tgt->error = "immutable target type cannot be mixed with other target types";
 700                         goto bad;
 701                 }
 702                 t->immutable_target_type = tgt->type;
 703         }
 704
 705         if (dm_target_has_integrity(tgt->type))
 706                 t->integrity_added = 1;
 707
 708         tgt->table = t;
 709         tgt->begin = start;
 710         tgt->len = len;
 711         tgt->error = "Unknown error";
 712
 713         /*
 714          * Does this target adjoin the previous one ?
 715          */
 716         if (!adjoin(t, tgt)) {
 717                 tgt->error = "Gap in table";
 718                 goto bad;
 719         }
 720
 721         r = dm_split_args(&argc, &argv, params);
 722         if (r) {
 723                 tgt->error = "couldn't split parameters (insufficient memory)";
 724                 goto bad;
 725         }
 726
 727         r = tgt->type->ctr(tgt, argc, argv);
 728         kfree(argv);
 729         if (r)
 730                 goto bad;
 731
 732         t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
 733
 734         if (!tgt->num_discard_bios && tgt->discards_supported)
 735                 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
 736                        dm_device_name(t->md), type);
 737
 738         return 0;
 739
 740  bad:
 741         DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
 742         dm_put_target_type(tgt->type);
 743         return r;
 744 }
 745
 746 /*
 747  * Target argument parsing helpers.
 748  */
 749 static int validate_next_arg(const struct dm_arg *arg,
 750                              struct dm_arg_set *arg_set,
 751                              unsigned *value, char **error, unsigned grouped)
 752 {
 753         const char *arg_str = dm_shift_arg(arg_set);
 754         char dummy;
 755
 756         if (!arg_str ||
 757             (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
 758             (*value < arg->min) ||
 759             (*value > arg->max) ||
 760             (grouped && arg_set->argc < *value)) {
 761                 *error = arg->error;
 762                 return -EINVAL;
 763         }
 764
 765         return 0;
 766 }
 767
 768 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 769                 unsigned *value, char **error)
 770 {
 771         return validate_next_arg(arg, arg_set, value, error, 0);
 772 }
 773 EXPORT_SYMBOL(dm_read_arg);
 774
 775 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 776                       unsigned *value, char **error)
 777 {
 778         return validate_next_arg(arg, arg_set, value, error, 1);
 779 }
 780 EXPORT_SYMBOL(dm_read_arg_group);
 781
 782 const char *dm_shift_arg(struct dm_arg_set *as)
 783 {
 784         char *r;
 785
 786         if (as->argc) {
 787                 as->argc--;
 788                 r = *as->argv;
 789                 as->argv++;
 790                 return r;
 791         }
 792
 793         return NULL;
 794 }
 795 EXPORT_SYMBOL(dm_shift_arg);
 796
 797 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
 798 {
 799         BUG_ON(as->argc < num_args);
 800         as->argc -= num_args;
 801         as->argv += num_args;
 802 }
 803 EXPORT_SYMBOL(dm_consume_args);
 804
 805 static bool __table_type_bio_based(enum dm_queue_mode table_type)
 806 {
 807         return (table_type == DM_TYPE_BIO_BASED ||
 808                 table_type == DM_TYPE_DAX_BIO_BASED);
 809 }
 810
 811 static bool __table_type_request_based(enum dm_queue_mode table_type)
 812 {
 813         return table_type == DM_TYPE_REQUEST_BASED;
 814 }
 815
 816 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
 817 {
 818         t->type = type;
 819 }
 820 EXPORT_SYMBOL_GPL(dm_table_set_type);
 821
 822 /* validate the dax capability of the target device span */
 823 int device_supports_dax(struct dm_target *ti, struct dm_dev *dev,
 824                         sector_t start, sector_t len, void *data)
 825 {
 826         int blocksize = *(int *) data, id;
 827         bool rc;
 828
 829         id = dax_read_lock();
 830         rc = dax_supported(dev->dax_dev, dev->bdev, blocksize, start, len);
 831         dax_read_unlock(id);
 832
 833         return rc;
 834 }
 835
 836 /* Check devices support synchronous DAX */
 837 static int device_dax_synchronous(struct dm_target *ti, struct dm_dev *dev,
 838                                   sector_t start, sector_t len, void *data)
 839 {
 840         return dev->dax_dev && dax_synchronous(dev->dax_dev);
 841 }
 842
 843 bool dm_table_supports_dax(struct dm_table *t,
 844                            iterate_devices_callout_fn iterate_fn, int *blocksize)
 845 {
 846         struct dm_target *ti;
 847         unsigned i;
 848
 849         /* Ensure that all targets support DAX. */
 850         for (i = 0; i < dm_table_get_num_targets(t); i++) {
 851                 ti = dm_table_get_target(t, i);
 852
 853                 if (!ti->type->direct_access)
 854                         return false;
 855
 856                 if (!ti->type->iterate_devices ||
 857                     !ti->type->iterate_devices(ti, iterate_fn, blocksize))
 858                         return false;
 859         }
 860
 861         return true;
 862 }
 863
 864 static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev,
 865                                   sector_t start, sector_t len, void *data)
 866 {
 867         struct block_device *bdev = dev->bdev;
 868         struct request_queue *q = bdev_get_queue(bdev);
 869
 870         /* request-based cannot stack on partitions! */
 871         if (bdev_is_partition(bdev))
 872                 return false;
 873
 874         return queue_is_mq(q);
 875 }
 876
 877 static int dm_table_determine_type(struct dm_table *t)
 878 {
 879         unsigned i;
 880         unsigned bio_based = 0, request_based = 0, hybrid = 0;
 881         struct dm_target *tgt;
 882         struct list_head *devices = dm_table_get_devices(t);
 883         enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
 884         int page_size = PAGE_SIZE;
 885
 886         if (t->type != DM_TYPE_NONE) {
 887                 /* target already set the table's type */
 888                 if (t->type == DM_TYPE_BIO_BASED) {
 889                         /* possibly upgrade to a variant of bio-based */
 890                         goto verify_bio_based;
 891                 }
 892                 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
 893                 goto verify_rq_based;
 894         }
 895
 896         for (i = 0; i < t->num_targets; i++) {
 897                 tgt = t->targets + i;
 898                 if (dm_target_hybrid(tgt))
 899                         hybrid = 1;
 900                 else if (dm_target_request_based(tgt))
 901                         request_based = 1;
 902                 else
 903                         bio_based = 1;
 904
 905                 if (bio_based && request_based) {
 906                         DMERR("Inconsistent table: different target types"
 907                               " can't be mixed up");
 908                         return -EINVAL;
 909                 }
 910         }
 911
 912         if (hybrid && !bio_based && !request_based) {
 913                 /*
 914                  * The targets can work either way.
 915                  * Determine the type from the live device.
 916                  * Default to bio-based if device is new.
 917                  */
 918                 if (__table_type_request_based(live_md_type))
 919                         request_based = 1;
 920                 else
 921                         bio_based = 1;
 922         }
 923
 924         if (bio_based) {
 925 verify_bio_based:
 926                 /* We must use this table as bio-based */
 927                 t->type = DM_TYPE_BIO_BASED;
 928                 if (dm_table_supports_dax(t, device_supports_dax, &page_size) ||
 929                     (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
 930                         t->type = DM_TYPE_DAX_BIO_BASED;
 931                 }
 932                 return 0;
 933         }
 934
 935         BUG_ON(!request_based); /* No targets in this table */
 936
 937         t->type = DM_TYPE_REQUEST_BASED;
 938
 939 verify_rq_based:
 940         /*
 941          * Request-based dm supports only tables that have a single target now.
 942          * To support multiple targets, request splitting support is needed,
 943          * and that needs lots of changes in the block-layer.
 944          * (e.g. request completion process for partial completion.)
 945          */
 946         if (t->num_targets > 1) {
 947                 DMERR("request-based DM doesn't support multiple targets");
 948                 return -EINVAL;
 949         }
 950
 951         if (list_empty(devices)) {
 952                 int srcu_idx;
 953                 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
 954
 955                 /* inherit live table's type */
 956                 if (live_table)
 957                         t->type = live_table->type;
 958                 dm_put_live_table(t->md, srcu_idx);
 959                 return 0;
 960         }
 961
 962         tgt = dm_table_get_immutable_target(t);
 963         if (!tgt) {
 964                 DMERR("table load rejected: immutable target is required");
 965                 return -EINVAL;
 966         } else if (tgt->max_io_len) {
 967                 DMERR("table load rejected: immutable target that splits IO is not supported");
 968                 return -EINVAL;
 969         }
 970
 971         /* Non-request-stackable devices can't be used for request-based dm */
 972         if (!tgt->type->iterate_devices ||
 973             !tgt->type->iterate_devices(tgt, device_is_rq_stackable, NULL)) {
 974                 DMERR("table load rejected: including non-request-stackable devices");
 975                 return -EINVAL;
 976         }
 977
 978         return 0;
 979 }
 980
 981 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
 982 {
 983         return t->type;
 984 }
 985
 986 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
 987 {
 988         return t->immutable_target_type;
 989 }
 990
 991 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
 992 {
 993         /* Immutable target is implicitly a singleton */
 994         if (t->num_targets > 1 ||
 995             !dm_target_is_immutable(t->targets[0].type))
 996                 return NULL;
 997
 998         return t->targets;
 999 }
1000
1001 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1002 {
1003         struct dm_target *ti;
1004         unsigned i;
1005
1006         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1007                 ti = dm_table_get_target(t, i);
1008                 if (dm_target_is_wildcard(ti->type))
1009                         return ti;
1010         }
1011
1012         return NULL;
1013 }
1014
1015 bool dm_table_bio_based(struct dm_table *t)
1016 {
1017         return __table_type_bio_based(dm_table_get_type(t));
1018 }
1019
1020 bool dm_table_request_based(struct dm_table *t)
1021 {
1022         return __table_type_request_based(dm_table_get_type(t));
1023 }
1024
1025 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1026 {
1027         enum dm_queue_mode type = dm_table_get_type(t);
1028         unsigned per_io_data_size = 0;
1029         unsigned min_pool_size = 0;
1030         struct dm_target *ti;
1031         unsigned i;
1032
1033         if (unlikely(type == DM_TYPE_NONE)) {
1034                 DMWARN("no table type is set, can't allocate mempools");
1035                 return -EINVAL;
1036         }
1037
1038         if (__table_type_bio_based(type))
1039                 for (i = 0; i < t->num_targets; i++) {
1040                         ti = t->targets + i;
1041                         per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1042                         min_pool_size = max(min_pool_size, ti->num_flush_bios);
1043                 }
1044
1045         t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported,
1046                                            per_io_data_size, min_pool_size);
1047         if (!t->mempools)
1048                 return -ENOMEM;
1049
1050         return 0;
1051 }
1052
1053 void dm_table_free_md_mempools(struct dm_table *t)
1054 {
1055         dm_free_md_mempools(t->mempools);
1056         t->mempools = NULL;
1057 }
1058
1059 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1060 {
1061         return t->mempools;
1062 }
1063
1064 static int setup_indexes(struct dm_table *t)
1065 {
1066         int i;
1067         unsigned int total = 0;
1068         sector_t *indexes;
1069
1070         /* allocate the space for *all* the indexes */
1071         for (i = t->depth - 2; i >= 0; i--) {
1072                 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1073                 total += t->counts[i];
1074         }
1075
1076         indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1077         if (!indexes)
1078                 return -ENOMEM;
1079
1080         /* set up internal nodes, bottom-up */
1081         for (i = t->depth - 2; i >= 0; i--) {
1082                 t->index[i] = indexes;
1083                 indexes += (KEYS_PER_NODE * t->counts[i]);
1084                 setup_btree_index(i, t);
1085         }
1086
1087         return 0;
1088 }
1089
1090 /*
1091  * Builds the btree to index the map.
1092  */
1093 static int dm_table_build_index(struct dm_table *t)
1094 {
1095         int r = 0;
1096         unsigned int leaf_nodes;
1097
1098         /* how many indexes will the btree have ? */
1099         leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1100         t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1101
1102         /* leaf layer has already been set up */
1103         t->counts[t->depth - 1] = leaf_nodes;
1104         t->index[t->depth - 1] = t->highs;
1105
1106         if (t->depth >= 2)
1107                 r = setup_indexes(t);
1108
1109         return r;
1110 }
1111
1112 static bool integrity_profile_exists(struct gendisk *disk)
1113 {
1114         return !!blk_get_integrity(disk);
1115 }
1116
1117 /*
1118  * Get a disk whose integrity profile reflects the table's profile.
1119  * Returns NULL if integrity support was inconsistent or unavailable.
1120  */
1121 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1122 {
1123         struct list_head *devices = dm_table_get_devices(t);
1124         struct dm_dev_internal *dd = NULL;
1125         struct gendisk *prev_disk = NULL, *template_disk = NULL;
1126         unsigned i;
1127
1128         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1129                 struct dm_target *ti = dm_table_get_target(t, i);
1130                 if (!dm_target_passes_integrity(ti->type))
1131                         goto no_integrity;
1132         }
1133
1134         list_for_each_entry(dd, devices, list) {
1135                 template_disk = dd->dm_dev->bdev->bd_disk;
1136                 if (!integrity_profile_exists(template_disk))
1137                         goto no_integrity;
1138                 else if (prev_disk &&
1139                          blk_integrity_compare(prev_disk, template_disk) < 0)
1140                         goto no_integrity;
1141                 prev_disk = template_disk;
1142         }
1143
1144         return template_disk;
1145
1146 no_integrity:
1147         if (prev_disk)
1148                 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1149                        dm_device_name(t->md),
1150                        prev_disk->disk_name,
1151                        template_disk->disk_name);
1152         return NULL;
1153 }
1154
1155 /*
1156  * Register the mapped device for blk_integrity support if the
1157  * underlying devices have an integrity profile.  But all devices may
1158  * not have matching profiles (checking all devices isn't reliable
1159  * during table load because this table may use other DM device(s) which
1160  * must be resumed before they will have an initialized integity
1161  * profile).  Consequently, stacked DM devices force a 2 stage integrity
1162  * profile validation: First pass during table load, final pass during
1163  * resume.
1164  */
1165 static int dm_table_register_integrity(struct dm_table *t)
1166 {
1167         struct mapped_device *md = t->md;
1168         struct gendisk *template_disk = NULL;
1169
1170         /* If target handles integrity itself do not register it here. */
1171         if (t->integrity_added)
1172                 return 0;
1173
1174         template_disk = dm_table_get_integrity_disk(t);
1175         if (!template_disk)
1176                 return 0;
1177
1178         if (!integrity_profile_exists(dm_disk(md))) {
1179                 t->integrity_supported = true;
1180                 /*
1181                  * Register integrity profile during table load; we can do
1182                  * this because the final profile must match during resume.
1183                  */
1184                 blk_integrity_register(dm_disk(md),
1185                                        blk_get_integrity(template_disk));
1186                 return 0;
1187         }
1188
1189         /*
1190          * If DM device already has an initialized integrity
1191          * profile the new profile should not conflict.
1192          */
1193         if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1194                 DMWARN("%s: conflict with existing integrity profile: "
1195                        "%s profile mismatch",
1196                        dm_device_name(t->md),
1197                        template_disk->disk_name);
1198                 return 1;
1199         }
1200
1201         /* Preserve existing integrity profile */
1202         t->integrity_supported = true;
1203         return 0;
1204 }
1205
1206 /*
1207  * Prepares the table for use by building the indices,
1208  * setting the type, and allocating mempools.
1209  */
1210 int dm_table_complete(struct dm_table *t)
1211 {
1212         int r;
1213
1214         r = dm_table_determine_type(t);
1215         if (r) {
1216                 DMERR("unable to determine table type");
1217                 return r;
1218         }
1219
1220         r = dm_table_build_index(t);
1221         if (r) {
1222                 DMERR("unable to build btrees");
1223                 return r;
1224         }
1225
1226         r = dm_table_register_integrity(t);
1227         if (r) {
1228                 DMERR("could not register integrity profile.");
1229                 return r;
1230         }
1231
1232         r = dm_table_alloc_md_mempools(t, t->md);
1233         if (r)
1234                 DMERR("unable to allocate mempools");
1235
1236         return r;
1237 }
1238
1239 static DEFINE_MUTEX(_event_lock);
1240 void dm_table_event_callback(struct dm_table *t,
1241                              void (*fn)(void *), void *context)
1242 {
1243         mutex_lock(&_event_lock);
1244         t->event_fn = fn;
1245         t->event_context = context;
1246         mutex_unlock(&_event_lock);
1247 }
1248
1249 void dm_table_event(struct dm_table *t)
1250 {
1251         mutex_lock(&_event_lock);
1252         if (t->event_fn)
1253                 t->event_fn(t->event_context);
1254         mutex_unlock(&_event_lock);
1255 }
1256 EXPORT_SYMBOL(dm_table_event);
1257
1258 inline sector_t dm_table_get_size(struct dm_table *t)
1259 {
1260         return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1261 }
1262 EXPORT_SYMBOL(dm_table_get_size);
1263
1264 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1265 {
1266         if (index >= t->num_targets)
1267                 return NULL;
1268
1269         return t->targets + index;
1270 }
1271
1272 /*
1273  * Search the btree for the correct target.
1274  *
1275  * Caller should check returned pointer for NULL
1276  * to trap I/O beyond end of device.
1277  */
1278 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1279 {
1280         unsigned int l, n = 0, k = 0;
1281         sector_t *node;
1282
1283         if (unlikely(sector >= dm_table_get_size(t)))
1284                 return NULL;
1285
1286         for (l = 0; l < t->depth; l++) {
1287                 n = get_child(n, k);
1288                 node = get_node(t, l, n);
1289
1290                 for (k = 0; k < KEYS_PER_NODE; k++)
1291                         if (node[k] >= sector)
1292                                 break;
1293         }
1294
1295         return &t->targets[(KEYS_PER_NODE * n) + k];
1296 }
1297
1298 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1299                         sector_t start, sector_t len, void *data)
1300 {
1301         unsigned *num_devices = data;
1302
1303         (*num_devices)++;
1304
1305         return 0;
1306 }
1307
1308 /*
1309  * Check whether a table has no data devices attached using each
1310  * target's iterate_devices method.
1311  * Returns false if the result is unknown because a target doesn't
1312  * support iterate_devices.
1313  */
1314 bool dm_table_has_no_data_devices(struct dm_table *table)
1315 {
1316         struct dm_target *ti;
1317         unsigned i, num_devices;
1318
1319         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1320                 ti = dm_table_get_target(table, i);
1321
1322                 if (!ti->type->iterate_devices)
1323                         return false;
1324
1325                 num_devices = 0;
1326                 ti->type->iterate_devices(ti, count_device, &num_devices);
1327                 if (num_devices)
1328                         return false;
1329         }
1330
1331         return true;
1332 }
1333
1334 static int device_is_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1335                                  sector_t start, sector_t len, void *data)
1336 {
1337         struct request_queue *q = bdev_get_queue(dev->bdev);
1338         enum blk_zoned_model *zoned_model = data;
1339
1340         return q && blk_queue_zoned_model(q) == *zoned_model;
1341 }
1342
1343 static bool dm_table_supports_zoned_model(struct dm_table *t,
1344                                           enum blk_zoned_model zoned_model)
1345 {
1346         struct dm_target *ti;
1347         unsigned i;
1348
1349         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1350                 ti = dm_table_get_target(t, i);
1351
1352                 if (zoned_model == BLK_ZONED_HM &&
1353                     !dm_target_supports_zoned_hm(ti->type))
1354                         return false;
1355
1356                 if (!ti->type->iterate_devices ||
1357                     !ti->type->iterate_devices(ti, device_is_zoned_model, &zoned_model))
1358                         return false;
1359         }
1360
1361         return true;
1362 }
1363
1364 static int device_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1365                                        sector_t start, sector_t len, void *data)
1366 {
1367         struct request_queue *q = bdev_get_queue(dev->bdev);
1368         unsigned int *zone_sectors = data;
1369
1370         return q && blk_queue_zone_sectors(q) == *zone_sectors;
1371 }
1372
1373 static bool dm_table_matches_zone_sectors(struct dm_table *t,
1374                                           unsigned int zone_sectors)
1375 {
1376         struct dm_target *ti;
1377         unsigned i;
1378
1379         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1380                 ti = dm_table_get_target(t, i);
1381
1382                 if (!ti->type->iterate_devices ||
1383                     !ti->type->iterate_devices(ti, device_matches_zone_sectors, &zone_sectors))
1384                         return false;
1385         }
1386
1387         return true;
1388 }
1389
1390 static int validate_hardware_zoned_model(struct dm_table *table,
1391                                          enum blk_zoned_model zoned_model,
1392                                          unsigned int zone_sectors)
1393 {
1394         if (zoned_model == BLK_ZONED_NONE)
1395                 return 0;
1396
1397         if (!dm_table_supports_zoned_model(table, zoned_model)) {
1398                 DMERR("%s: zoned model is not consistent across all devices",
1399                       dm_device_name(table->md));
1400                 return -EINVAL;
1401         }
1402
1403         /* Check zone size validity and compatibility */
1404         if (!zone_sectors || !is_power_of_2(zone_sectors))
1405                 return -EINVAL;
1406
1407         if (!dm_table_matches_zone_sectors(table, zone_sectors)) {
1408                 DMERR("%s: zone sectors is not consistent across all devices",
1409                       dm_device_name(table->md));
1410                 return -EINVAL;
1411         }
1412
1413         return 0;
1414 }
1415
1416 /*
1417  * Establish the new table's queue_limits and validate them.
1418  */
1419 int dm_calculate_queue_limits(struct dm_table *table,
1420                               struct queue_limits *limits)
1421 {
1422         struct dm_target *ti;
1423         struct queue_limits ti_limits;
1424         unsigned i;
1425         enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1426         unsigned int zone_sectors = 0;
1427
1428         blk_set_stacking_limits(limits);
1429
1430         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1431                 blk_set_stacking_limits(&ti_limits);
1432
1433                 ti = dm_table_get_target(table, i);
1434
1435                 if (!ti->type->iterate_devices)
1436                         goto combine_limits;
1437
1438                 /*
1439                  * Combine queue limits of all the devices this target uses.
1440                  */
1441                 ti->type->iterate_devices(ti, dm_set_device_limits,
1442                                           &ti_limits);
1443
1444                 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1445                         /*
1446                          * After stacking all limits, validate all devices
1447                          * in table support this zoned model and zone sectors.
1448                          */
1449                         zoned_model = ti_limits.zoned;
1450                         zone_sectors = ti_limits.chunk_sectors;
1451                 }
1452
1453                 /* Set I/O hints portion of queue limits */
1454                 if (ti->type->io_hints)
1455                         ti->type->io_hints(ti, &ti_limits);
1456
1457                 /*
1458                  * Check each device area is consistent with the target's
1459                  * overall queue limits.
1460                  */
1461                 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1462                                               &ti_limits))
1463                         return -EINVAL;
1464
1465 combine_limits:
1466                 /*
1467                  * Merge this target's queue limits into the overall limits
1468                  * for the table.
1469                  */
1470                 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1471                         DMWARN("%s: adding target device "
1472                                "(start sect %llu len %llu) "
1473                                "caused an alignment inconsistency",
1474                                dm_device_name(table->md),
1475                                (unsigned long long) ti->begin,
1476                                (unsigned long long) ti->len);
1477         }
1478
1479         /*
1480          * Verify that the zoned model and zone sectors, as determined before
1481          * any .io_hints override, are the same across all devices in the table.
1482          * - this is especially relevant if .io_hints is emulating a disk-managed
1483          *   zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1484          * BUT...
1485          */
1486         if (limits->zoned != BLK_ZONED_NONE) {
1487                 /*
1488                  * ...IF the above limits stacking determined a zoned model
1489                  * validate that all of the table's devices conform to it.
1490                  */
1491                 zoned_model = limits->zoned;
1492                 zone_sectors = limits->chunk_sectors;
1493         }
1494         if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1495                 return -EINVAL;
1496
1497         return validate_hardware_logical_block_alignment(table, limits);
1498 }
1499
1500 /*
1501  * Verify that all devices have an integrity profile that matches the
1502  * DM device's registered integrity profile.  If the profiles don't
1503  * match then unregister the DM device's integrity profile.
1504  */
1505 static void dm_table_verify_integrity(struct dm_table *t)
1506 {
1507         struct gendisk *template_disk = NULL;
1508
1509         if (t->integrity_added)
1510                 return;
1511
1512         if (t->integrity_supported) {
1513                 /*
1514                  * Verify that the original integrity profile
1515                  * matches all the devices in this table.
1516                  */
1517                 template_disk = dm_table_get_integrity_disk(t);
1518                 if (template_disk &&
1519                     blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1520                         return;
1521         }
1522
1523         if (integrity_profile_exists(dm_disk(t->md))) {
1524                 DMWARN("%s: unable to establish an integrity profile",
1525                        dm_device_name(t->md));
1526                 blk_integrity_unregister(dm_disk(t->md));
1527         }
1528 }
1529
1530 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1531                                 sector_t start, sector_t len, void *data)
1532 {
1533         unsigned long flush = (unsigned long) data;
1534         struct request_queue *q = bdev_get_queue(dev->bdev);
1535
1536         return q && (q->queue_flags & flush);
1537 }
1538
1539 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1540 {
1541         struct dm_target *ti;
1542         unsigned i;
1543
1544         /*
1545          * Require at least one underlying device to support flushes.
1546          * t->devices includes internal dm devices such as mirror logs
1547          * so we need to use iterate_devices here, which targets
1548          * supporting flushes must provide.
1549          */
1550         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1551                 ti = dm_table_get_target(t, i);
1552
1553                 if (!ti->num_flush_bios)
1554                         continue;
1555
1556                 if (ti->flush_supported)
1557                         return true;
1558
1559                 if (ti->type->iterate_devices &&
1560                     ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1561                         return true;
1562         }
1563
1564         return false;
1565 }
1566
1567 static int device_dax_write_cache_enabled(struct dm_target *ti,
1568                                           struct dm_dev *dev, sector_t start,
1569                                           sector_t len, void *data)
1570 {
1571         struct dax_device *dax_dev = dev->dax_dev;
1572
1573         if (!dax_dev)
1574                 return false;
1575
1576         if (dax_write_cache_enabled(dax_dev))
1577                 return true;
1578         return false;
1579 }
1580
1581 static int dm_table_supports_dax_write_cache(struct dm_table *t)
1582 {
1583         struct dm_target *ti;
1584         unsigned i;
1585
1586         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1587                 ti = dm_table_get_target(t, i);
1588
1589                 if (ti->type->iterate_devices &&
1590                     ti->type->iterate_devices(ti,
1591                                 device_dax_write_cache_enabled, NULL))
1592                         return true;
1593         }
1594
1595         return false;
1596 }
1597
1598 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1599                             sector_t start, sector_t len, void *data)
1600 {
1601         struct request_queue *q = bdev_get_queue(dev->bdev);
1602
1603         return q && blk_queue_nonrot(q);
1604 }
1605
1606 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1607                              sector_t start, sector_t len, void *data)
1608 {
1609         struct request_queue *q = bdev_get_queue(dev->bdev);
1610
1611         return q && !blk_queue_add_random(q);
1612 }
1613
1614 static bool dm_table_all_devices_attribute(struct dm_table *t,
1615                                            iterate_devices_callout_fn func)
1616 {
1617         struct dm_target *ti;
1618         unsigned i;
1619
1620         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1621                 ti = dm_table_get_target(t, i);
1622
1623                 if (!ti->type->iterate_devices ||
1624                     !ti->type->iterate_devices(ti, func, NULL))
1625                         return false;
1626         }
1627
1628         return true;
1629 }
1630
1631 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1632                                          sector_t start, sector_t len, void *data)
1633 {
1634         struct request_queue *q = bdev_get_queue(dev->bdev);
1635
1636         return q && !q->limits.max_write_same_sectors;
1637 }
1638
1639 static bool dm_table_supports_write_same(struct dm_table *t)
1640 {
1641         struct dm_target *ti;
1642         unsigned i;
1643
1644         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1645                 ti = dm_table_get_target(t, i);
1646
1647                 if (!ti->num_write_same_bios)
1648                         return false;
1649
1650                 if (!ti->type->iterate_devices ||
1651                     ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1652                         return false;
1653         }
1654
1655         return true;
1656 }
1657
1658 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1659                                            sector_t start, sector_t len, void *data)
1660 {
1661         struct request_queue *q = bdev_get_queue(dev->bdev);
1662
1663         return q && !q->limits.max_write_zeroes_sectors;
1664 }
1665
1666 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1667 {
1668         struct dm_target *ti;
1669         unsigned i = 0;
1670
1671         while (i < dm_table_get_num_targets(t)) {
1672                 ti = dm_table_get_target(t, i++);
1673
1674                 if (!ti->num_write_zeroes_bios)
1675                         return false;
1676
1677                 if (!ti->type->iterate_devices ||
1678                     ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1679                         return false;
1680         }
1681
1682         return true;
1683 }
1684
1685 static int device_not_nowait_capable(struct dm_target *ti, struct dm_dev *dev,
1686                                      sector_t start, sector_t len, void *data)
1687 {
1688         struct request_queue *q = bdev_get_queue(dev->bdev);
1689
1690         return q && !blk_queue_nowait(q);
1691 }
1692
1693 static bool dm_table_supports_nowait(struct dm_table *t)
1694 {
1695         struct dm_target *ti;
1696         unsigned i = 0;
1697
1698         while (i < dm_table_get_num_targets(t)) {
1699                 ti = dm_table_get_target(t, i++);
1700
1701                 if (!dm_target_supports_nowait(ti->type))
1702                         return false;
1703
1704                 if (!ti->type->iterate_devices ||
1705                     ti->type->iterate_devices(ti, device_not_nowait_capable, NULL))
1706                         return false;
1707         }
1708
1709         return true;
1710 }
1711
1712 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1713                                       sector_t start, sector_t len, void *data)
1714 {
1715         struct request_queue *q = bdev_get_queue(dev->bdev);
1716
1717         return q && !blk_queue_discard(q);
1718 }
1719
1720 static bool dm_table_supports_discards(struct dm_table *t)
1721 {
1722         struct dm_target *ti;
1723         unsigned i;
1724
1725         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1726                 ti = dm_table_get_target(t, i);
1727
1728                 if (!ti->num_discard_bios)
1729                         return false;
1730
1731                 /*
1732                  * Either the target provides discard support (as implied by setting
1733                  * 'discards_supported') or it relies on _all_ data devices having
1734                  * discard support.
1735                  */
1736                 if (!ti->discards_supported &&
1737                     (!ti->type->iterate_devices ||
1738                      ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1739                         return false;
1740         }
1741
1742         return true;
1743 }
1744
1745 static int device_not_secure_erase_capable(struct dm_target *ti,
1746                                            struct dm_dev *dev, sector_t start,
1747                                            sector_t len, void *data)
1748 {
1749         struct request_queue *q = bdev_get_queue(dev->bdev);
1750
1751         return q && !blk_queue_secure_erase(q);
1752 }
1753
1754 static bool dm_table_supports_secure_erase(struct dm_table *t)
1755 {
1756         struct dm_target *ti;
1757         unsigned int i;
1758
1759         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1760                 ti = dm_table_get_target(t, i);
1761
1762                 if (!ti->num_secure_erase_bios)
1763                         return false;
1764
1765                 if (!ti->type->iterate_devices ||
1766                     ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1767                         return false;
1768         }
1769
1770         return true;
1771 }
1772
1773 static int device_requires_stable_pages(struct dm_target *ti,
1774                                         struct dm_dev *dev, sector_t start,
1775                                         sector_t len, void *data)
1776 {
1777         struct request_queue *q = bdev_get_queue(dev->bdev);
1778
1779         return q && blk_queue_stable_writes(q);
1780 }
1781
1782 /*
1783  * If any underlying device requires stable pages, a table must require
1784  * them as well.  Only targets that support iterate_devices are considered:
1785  * don't want error, zero, etc to require stable pages.
1786  */
1787 static bool dm_table_requires_stable_pages(struct dm_table *t)
1788 {
1789         struct dm_target *ti;
1790         unsigned i;
1791
1792         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1793                 ti = dm_table_get_target(t, i);
1794
1795                 if (ti->type->iterate_devices &&
1796                     ti->type->iterate_devices(ti, device_requires_stable_pages, NULL))
1797                         return true;
1798         }
1799
1800         return false;
1801 }
1802
1803 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1804                                struct queue_limits *limits)
1805 {
1806         bool wc = false, fua = false;
1807         int page_size = PAGE_SIZE;
1808
1809         /*
1810          * Copy table's limits to the DM device's request_queue
1811          */
1812         q->limits = *limits;
1813
1814         if (dm_table_supports_nowait(t))
1815                 blk_queue_flag_set(QUEUE_FLAG_NOWAIT, q);
1816         else
1817                 blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, q);
1818
1819         if (!dm_table_supports_discards(t)) {
1820                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
1821                 /* Must also clear discard limits... */
1822                 q->limits.max_discard_sectors = 0;
1823                 q->limits.max_hw_discard_sectors = 0;
1824                 q->limits.discard_granularity = 0;
1825                 q->limits.discard_alignment = 0;
1826                 q->limits.discard_misaligned = 0;
1827         } else
1828                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
1829
1830         if (dm_table_supports_secure_erase(t))
1831                 blk_queue_flag_set(QUEUE_FLAG_SECERASE, q);
1832
1833         if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1834                 wc = true;
1835                 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1836                         fua = true;
1837         }
1838         blk_queue_write_cache(q, wc, fua);
1839
1840         if (dm_table_supports_dax(t, device_supports_dax, &page_size)) {
1841                 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
1842                 if (dm_table_supports_dax(t, device_dax_synchronous, NULL))
1843                         set_dax_synchronous(t->md->dax_dev);
1844         }
1845         else
1846                 blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
1847
1848         if (dm_table_supports_dax_write_cache(t))
1849                 dax_write_cache(t->md->dax_dev, true);
1850
1851         /* Ensure that all underlying devices are non-rotational. */
1852         if (dm_table_all_devices_attribute(t, device_is_nonrot))
1853                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
1854         else
1855                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
1856
1857         if (!dm_table_supports_write_same(t))
1858                 q->limits.max_write_same_sectors = 0;
1859         if (!dm_table_supports_write_zeroes(t))
1860                 q->limits.max_write_zeroes_sectors = 0;
1861
1862         dm_table_verify_integrity(t);
1863
1864         /*
1865          * Some devices don't use blk_integrity but still want stable pages
1866          * because they do their own checksumming.
1867          */
1868         if (dm_table_requires_stable_pages(t))
1869                 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, q);
1870         else
1871                 blk_queue_flag_clear(QUEUE_FLAG_STABLE_WRITES, q);
1872
1873         /*
1874          * Determine whether or not this queue's I/O timings contribute
1875          * to the entropy pool, Only request-based targets use this.
1876          * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1877          * have it set.
1878          */
1879         if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1880                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
1881
1882         /*
1883          * For a zoned target, the number of zones should be updated for the
1884          * correct value to be exposed in sysfs queue/nr_zones. For a BIO based
1885          * target, this is all that is needed.
1886          */
1887 #ifdef CONFIG_BLK_DEV_ZONED
1888         if (blk_queue_is_zoned(q)) {
1889                 WARN_ON_ONCE(queue_is_mq(q));
1890                 q->nr_zones = blkdev_nr_zones(t->md->disk);
1891         }
1892 #endif
1893
1894         blk_queue_update_readahead(q);
1895 }
1896
1897 unsigned int dm_table_get_num_targets(struct dm_table *t)
1898 {
1899         return t->num_targets;
1900 }
1901
1902 struct list_head *dm_table_get_devices(struct dm_table *t)
1903 {
1904         return &t->devices;
1905 }
1906
1907 fmode_t dm_table_get_mode(struct dm_table *t)
1908 {
1909         return t->mode;
1910 }
1911 EXPORT_SYMBOL(dm_table_get_mode);
1912
1913 enum suspend_mode {
1914         PRESUSPEND,
1915         PRESUSPEND_UNDO,
1916         POSTSUSPEND,
1917 };
1918
1919 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1920 {
1921         int i = t->num_targets;
1922         struct dm_target *ti = t->targets;
1923
1924         lockdep_assert_held(&t->md->suspend_lock);
1925
1926         while (i--) {
1927                 switch (mode) {
1928                 case PRESUSPEND:
1929                         if (ti->type->presuspend)
1930                                 ti->type->presuspend(ti);
1931                         break;
1932                 case PRESUSPEND_UNDO:
1933                         if (ti->type->presuspend_undo)
1934                                 ti->type->presuspend_undo(ti);
1935                         break;
1936                 case POSTSUSPEND:
1937                         if (ti->type->postsuspend)
1938                                 ti->type->postsuspend(ti);
1939                         break;
1940                 }
1941                 ti++;
1942         }
1943 }
1944
1945 void dm_table_presuspend_targets(struct dm_table *t)
1946 {
1947         if (!t)
1948                 return;
1949
1950         suspend_targets(t, PRESUSPEND);
1951 }
1952
1953 void dm_table_presuspend_undo_targets(struct dm_table *t)
1954 {
1955         if (!t)
1956                 return;
1957
1958         suspend_targets(t, PRESUSPEND_UNDO);
1959 }
1960
1961 void dm_table_postsuspend_targets(struct dm_table *t)
1962 {
1963         if (!t)
1964                 return;
1965
1966         suspend_targets(t, POSTSUSPEND);
1967 }
1968
1969 int dm_table_resume_targets(struct dm_table *t)
1970 {
1971         int i, r = 0;
1972
1973         lockdep_assert_held(&t->md->suspend_lock);
1974
1975         for (i = 0; i < t->num_targets; i++) {
1976                 struct dm_target *ti = t->targets + i;
1977
1978                 if (!ti->type->preresume)
1979                         continue;
1980
1981                 r = ti->type->preresume(ti);
1982                 if (r) {
1983                         DMERR("%s: %s: preresume failed, error = %d",
1984                               dm_device_name(t->md), ti->type->name, r);
1985                         return r;
1986                 }
1987         }
1988
1989         for (i = 0; i < t->num_targets; i++) {
1990                 struct dm_target *ti = t->targets + i;
1991
1992                 if (ti->type->resume)
1993                         ti->type->resume(ti);
1994         }
1995
1996         return 0;
1997 }
1998
1999 struct mapped_device *dm_table_get_md(struct dm_table *t)
2000 {
2001         return t->md;
2002 }
2003 EXPORT_SYMBOL(dm_table_get_md);
2004
2005 const char *dm_table_device_name(struct dm_table *t)
2006 {
2007         return dm_device_name(t->md);
2008 }
2009 EXPORT_SYMBOL_GPL(dm_table_device_name);
2010
2011 void dm_table_run_md_queue_async(struct dm_table *t)
2012 {
2013         if (!dm_table_request_based(t))
2014                 return;
2015
2016         if (t->md->queue)
2017                 blk_mq_run_hw_queues(t->md->queue, true);
2018 }
2019 EXPORT_SYMBOL(dm_table_run_md_queue_async);
2020