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