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