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
  24 #define DM_MSG_PREFIX "table"
  25
  26 #define MAX_DEPTH 16
  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 struct dm_table {
  32         struct mapped_device *md;
  33         enum dm_queue_mode type;
  34
  35         /* btree table */
  36         unsigned int depth;
  37         unsigned int counts[MAX_DEPTH]; /* in nodes */
  38         sector_t *index[MAX_DEPTH];
  39
  40         unsigned int num_targets;
  41         unsigned int num_allocated;
  42         sector_t *highs;
  43         struct dm_target *targets;
  44
  45         struct target_type *immutable_target_type;
  46
  47         bool integrity_supported:1;
  48         bool singleton:1;
  49         bool all_blk_mq:1;
  50         unsigned integrity_added:1;
  51
  52         /*
  53          * Indicates the rw permissions for the new logical
  54          * device.  This should be a combination of FMODE_READ
  55          * and FMODE_WRITE.
  56          */
  57         fmode_t mode;
  58
  59         /* a list of devices used by this table */
  60         struct list_head devices;
  61
  62         /* events get handed up using this callback */
  63         void (*event_fn)(void *);
  64         void *event_context;
  65
  66         struct dm_md_mempools *mempools;
  67
  68         struct list_head target_callbacks;
  69 };
  70
  71 /*
  72  * Similar to ceiling(log_size(n))
  73  */
  74 static unsigned int int_log(unsigned int n, unsigned int base)
  75 {
  76         int result = 0;
  77
  78         while (n > 1) {
  79                 n = dm_div_up(n, base);
  80                 result++;
  81         }
  82
  83         return result;
  84 }
  85
  86 /*
  87  * Calculate the index of the child node of the n'th node k'th key.
  88  */
  89 static inline unsigned int get_child(unsigned int n, unsigned int k)
  90 {
  91         return (n * CHILDREN_PER_NODE) + k;
  92 }
  93
  94 /*
  95  * Return the n'th node of level l from table t.
  96  */
  97 static inline sector_t *get_node(struct dm_table *t,
  98                                  unsigned int l, unsigned int n)
  99 {
 100         return t->index[l] + (n * KEYS_PER_NODE);
 101 }
 102
 103 /*
 104  * Return the highest key that you could lookup from the n'th
 105  * node on level l of the btree.
 106  */
 107 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
 108 {
 109         for (; l < t->depth - 1; l++)
 110                 n = get_child(n, CHILDREN_PER_NODE - 1);
 111
 112         if (n >= t->counts[l])
 113                 return (sector_t) - 1;
 114
 115         return get_node(t, l, n)[KEYS_PER_NODE - 1];
 116 }
 117
 118 /*
 119  * Fills in a level of the btree based on the highs of the level
 120  * below it.
 121  */
 122 static int setup_btree_index(unsigned int l, struct dm_table *t)
 123 {
 124         unsigned int n, k;
 125         sector_t *node;
 126
 127         for (n = 0U; n < t->counts[l]; n++) {
 128                 node = get_node(t, l, n);
 129
 130                 for (k = 0U; k < KEYS_PER_NODE; k++)
 131                         node[k] = high(t, l + 1, get_child(n, k));
 132         }
 133
 134         return 0;
 135 }
 136
 137 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
 138 {
 139         unsigned long size;
 140         void *addr;
 141
 142         /*
 143          * Check that we're not going to overflow.
 144          */
 145         if (nmemb > (ULONG_MAX / elem_size))
 146                 return NULL;
 147
 148         size = nmemb * elem_size;
 149         addr = vzalloc(size);
 150
 151         return addr;
 152 }
 153 EXPORT_SYMBOL(dm_vcalloc);
 154
 155 /*
 156  * highs, and targets are managed as dynamic arrays during a
 157  * table load.
 158  */
 159 static int alloc_targets(struct dm_table *t, unsigned int num)
 160 {
 161         sector_t *n_highs;
 162         struct dm_target *n_targets;
 163
 164         /*
 165          * Allocate both the target array and offset array at once.
 166          * Append an empty entry to catch sectors beyond the end of
 167          * the device.
 168          */
 169         n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
 170                                           sizeof(sector_t));
 171         if (!n_highs)
 172                 return -ENOMEM;
 173
 174         n_targets = (struct dm_target *) (n_highs + num);
 175
 176         memset(n_highs, -1, sizeof(*n_highs) * num);
 177         vfree(t->highs);
 178
 179         t->num_allocated = num;
 180         t->highs = n_highs;
 181         t->targets = n_targets;
 182
 183         return 0;
 184 }
 185
 186 int dm_table_create(struct dm_table **result, fmode_t mode,
 187                     unsigned num_targets, struct mapped_device *md)
 188 {
 189         struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
 190
 191         if (!t)
 192                 return -ENOMEM;
 193
 194         INIT_LIST_HEAD(&t->devices);
 195         INIT_LIST_HEAD(&t->target_callbacks);
 196
 197         if (!num_targets)
 198                 num_targets = KEYS_PER_NODE;
 199
 200         num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
 201
 202         if (!num_targets) {
 203                 kfree(t);
 204                 return -ENOMEM;
 205         }
 206
 207         if (alloc_targets(t, num_targets)) {
 208                 kfree(t);
 209                 return -ENOMEM;
 210         }
 211
 212         t->type = DM_TYPE_NONE;
 213         t->mode = mode;
 214         t->md = md;
 215         *result = t;
 216         return 0;
 217 }
 218
 219 static void free_devices(struct list_head *devices, struct mapped_device *md)
 220 {
 221         struct list_head *tmp, *next;
 222
 223         list_for_each_safe(tmp, next, devices) {
 224                 struct dm_dev_internal *dd =
 225                     list_entry(tmp, struct dm_dev_internal, list);
 226                 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
 227                        dm_device_name(md), dd->dm_dev->name);
 228                 dm_put_table_device(md, dd->dm_dev);
 229                 kfree(dd);
 230         }
 231 }
 232
 233 void dm_table_destroy(struct dm_table *t)
 234 {
 235         unsigned int i;
 236
 237         if (!t)
 238                 return;
 239
 240         /* free the indexes */
 241         if (t->depth >= 2)
 242                 vfree(t->index[t->depth - 2]);
 243
 244         /* free the targets */
 245         for (i = 0; i < t->num_targets; i++) {
 246                 struct dm_target *tgt = t->targets + i;
 247
 248                 if (tgt->type->dtr)
 249                         tgt->type->dtr(tgt);
 250
 251                 dm_put_target_type(tgt->type);
 252         }
 253
 254         vfree(t->highs);
 255
 256         /* free the device list */
 257         free_devices(&t->devices, t->md);
 258
 259         dm_free_md_mempools(t->mempools);
 260
 261         kfree(t);
 262 }
 263
 264 /*
 265  * See if we've already got a device in the list.
 266  */
 267 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
 268 {
 269         struct dm_dev_internal *dd;
 270
 271         list_for_each_entry (dd, l, list)
 272                 if (dd->dm_dev->bdev->bd_dev == dev)
 273                         return dd;
 274
 275         return NULL;
 276 }
 277
 278 /*
 279  * If possible, this checks an area of a destination device is invalid.
 280  */
 281 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
 282                                   sector_t start, sector_t len, void *data)
 283 {
 284         struct request_queue *q;
 285         struct queue_limits *limits = data;
 286         struct block_device *bdev = dev->bdev;
 287         sector_t dev_size =
 288                 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
 289         unsigned short logical_block_size_sectors =
 290                 limits->logical_block_size >> SECTOR_SHIFT;
 291         char b[BDEVNAME_SIZE];
 292
 293         /*
 294          * Some devices exist without request functions,
 295          * such as loop devices not yet bound to backing files.
 296          * Forbid the use of such devices.
 297          */
 298         q = bdev_get_queue(bdev);
 299         if (!q || !q->make_request_fn) {
 300                 DMWARN("%s: %s is not yet initialised: "
 301                        "start=%llu, len=%llu, dev_size=%llu",
 302                        dm_device_name(ti->table->md), bdevname(bdev, b),
 303                        (unsigned long long)start,
 304                        (unsigned long long)len,
 305                        (unsigned long long)dev_size);
 306                 return 1;
 307         }
 308
 309         if (!dev_size)
 310                 return 0;
 311
 312         if ((start >= dev_size) || (start + len > dev_size)) {
 313                 DMWARN("%s: %s too small for target: "
 314                        "start=%llu, len=%llu, dev_size=%llu",
 315                        dm_device_name(ti->table->md), bdevname(bdev, b),
 316                        (unsigned long long)start,
 317                        (unsigned long long)len,
 318                        (unsigned long long)dev_size);
 319                 return 1;
 320         }
 321
 322         /*
 323          * If the target is mapped to zoned block device(s), check
 324          * that the zones are not partially mapped.
 325          */
 326         if (bdev_zoned_model(bdev) != BLK_ZONED_NONE) {
 327                 unsigned int zone_sectors = bdev_zone_sectors(bdev);
 328
 329                 if (start & (zone_sectors - 1)) {
 330                         DMWARN("%s: start=%llu not aligned to h/w zone size %u of %s",
 331                                dm_device_name(ti->table->md),
 332                                (unsigned long long)start,
 333                                zone_sectors, bdevname(bdev, b));
 334                         return 1;
 335                 }
 336
 337                 /*
 338                  * Note: The last zone of a zoned block device may be smaller
 339                  * than other zones. So for a target mapping the end of a
 340                  * zoned block device with such a zone, len would not be zone
 341                  * aligned. We do not allow such last smaller zone to be part
 342                  * of the mapping here to ensure that mappings with multiple
 343                  * devices do not end up with a smaller zone in the middle of
 344                  * the sector range.
 345                  */
 346                 if (len & (zone_sectors - 1)) {
 347                         DMWARN("%s: len=%llu not aligned to h/w zone size %u of %s",
 348                                dm_device_name(ti->table->md),
 349                                (unsigned long long)len,
 350                                zone_sectors, bdevname(bdev, b));
 351                         return 1;
 352                 }
 353         }
 354
 355         if (logical_block_size_sectors <= 1)
 356                 return 0;
 357
 358         if (start & (logical_block_size_sectors - 1)) {
 359                 DMWARN("%s: start=%llu not aligned to h/w "
 360                        "logical block size %u of %s",
 361                        dm_device_name(ti->table->md),
 362                        (unsigned long long)start,
 363                        limits->logical_block_size, bdevname(bdev, b));
 364                 return 1;
 365         }
 366
 367         if (len & (logical_block_size_sectors - 1)) {
 368                 DMWARN("%s: len=%llu not aligned to h/w "
 369                        "logical block size %u of %s",
 370                        dm_device_name(ti->table->md),
 371                        (unsigned long long)len,
 372                        limits->logical_block_size, bdevname(bdev, b));
 373                 return 1;
 374         }
 375
 376         return 0;
 377 }
 378
 379 /*
 380  * This upgrades the mode on an already open dm_dev, being
 381  * careful to leave things as they were if we fail to reopen the
 382  * device and not to touch the existing bdev field in case
 383  * it is accessed concurrently inside dm_table_any_congested().
 384  */
 385 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
 386                         struct mapped_device *md)
 387 {
 388         int r;
 389         struct dm_dev *old_dev, *new_dev;
 390
 391         old_dev = dd->dm_dev;
 392
 393         r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
 394                                 dd->dm_dev->mode | new_mode, &new_dev);
 395         if (r)
 396                 return r;
 397
 398         dd->dm_dev = new_dev;
 399         dm_put_table_device(md, old_dev);
 400
 401         return 0;
 402 }
 403
 404 /*
 405  * Convert the path to a device
 406  */
 407 dev_t dm_get_dev_t(const char *path)
 408 {
 409         dev_t dev;
 410         struct block_device *bdev;
 411
 412         bdev = lookup_bdev(path);
 413         if (IS_ERR(bdev))
 414                 dev = name_to_dev_t(path);
 415         else {
 416                 dev = bdev->bd_dev;
 417                 bdput(bdev);
 418         }
 419
 420         return dev;
 421 }
 422 EXPORT_SYMBOL_GPL(dm_get_dev_t);
 423
 424 /*
 425  * Add a device to the list, or just increment the usage count if
 426  * it's already present.
 427  */
 428 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
 429                   struct dm_dev **result)
 430 {
 431         int r;
 432         dev_t dev;
 433         struct dm_dev_internal *dd;
 434         struct dm_table *t = ti->table;
 435
 436         BUG_ON(!t);
 437
 438         dev = dm_get_dev_t(path);
 439         if (!dev)
 440                 return -ENODEV;
 441
 442         dd = find_device(&t->devices, dev);
 443         if (!dd) {
 444                 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
 445                 if (!dd)
 446                         return -ENOMEM;
 447
 448                 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
 449                         kfree(dd);
 450                         return r;
 451                 }
 452
 453                 atomic_set(&dd->count, 0);
 454                 list_add(&dd->list, &t->devices);
 455
 456         } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
 457                 r = upgrade_mode(dd, mode, t->md);
 458                 if (r)
 459                         return r;
 460         }
 461         atomic_inc(&dd->count);
 462
 463         *result = dd->dm_dev;
 464         return 0;
 465 }
 466 EXPORT_SYMBOL(dm_get_device);
 467
 468 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
 469                                 sector_t start, sector_t len, void *data)
 470 {
 471         struct queue_limits *limits = data;
 472         struct block_device *bdev = dev->bdev;
 473         struct request_queue *q = bdev_get_queue(bdev);
 474         char b[BDEVNAME_SIZE];
 475
 476         if (unlikely(!q)) {
 477                 DMWARN("%s: Cannot set limits for nonexistent device %s",
 478                        dm_device_name(ti->table->md), bdevname(bdev, b));
 479                 return 0;
 480         }
 481
 482         if (bdev_stack_limits(limits, bdev, start) < 0)
 483                 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
 484                        "physical_block_size=%u, logical_block_size=%u, "
 485                        "alignment_offset=%u, start=%llu",
 486                        dm_device_name(ti->table->md), bdevname(bdev, b),
 487                        q->limits.physical_block_size,
 488                        q->limits.logical_block_size,
 489                        q->limits.alignment_offset,
 490                        (unsigned long long) start << SECTOR_SHIFT);
 491
 492         limits->zoned = blk_queue_zoned_model(q);
 493
 494         return 0;
 495 }
 496
 497 /*
 498  * Decrement a device's use count and remove it if necessary.
 499  */
 500 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
 501 {
 502         int found = 0;
 503         struct list_head *devices = &ti->table->devices;
 504         struct dm_dev_internal *dd;
 505
 506         list_for_each_entry(dd, devices, list) {
 507                 if (dd->dm_dev == d) {
 508                         found = 1;
 509                         break;
 510                 }
 511         }
 512         if (!found) {
 513                 DMWARN("%s: device %s not in table devices list",
 514                        dm_device_name(ti->table->md), d->name);
 515                 return;
 516         }
 517         if (atomic_dec_and_test(&dd->count)) {
 518                 dm_put_table_device(ti->table->md, d);
 519                 list_del(&dd->list);
 520                 kfree(dd);
 521         }
 522 }
 523 EXPORT_SYMBOL(dm_put_device);
 524
 525 /*
 526  * Checks to see if the target joins onto the end of the table.
 527  */
 528 static int adjoin(struct dm_table *table, struct dm_target *ti)
 529 {
 530         struct dm_target *prev;
 531
 532         if (!table->num_targets)
 533                 return !ti->begin;
 534
 535         prev = &table->targets[table->num_targets - 1];
 536         return (ti->begin == (prev->begin + prev->len));
 537 }
 538
 539 /*
 540  * Used to dynamically allocate the arg array.
 541  *
 542  * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
 543  * process messages even if some device is suspended. These messages have a
 544  * small fixed number of arguments.
 545  *
 546  * On the other hand, dm-switch needs to process bulk data using messages and
 547  * excessive use of GFP_NOIO could cause trouble.
 548  */
 549 static char **realloc_argv(unsigned *array_size, char **old_argv)
 550 {
 551         char **argv;
 552         unsigned new_size;
 553         gfp_t gfp;
 554
 555         if (*array_size) {
 556                 new_size = *array_size * 2;
 557                 gfp = GFP_KERNEL;
 558         } else {
 559                 new_size = 8;
 560                 gfp = GFP_NOIO;
 561         }
 562         argv = kmalloc(new_size * sizeof(*argv), gfp);
 563         if (argv) {
 564                 memcpy(argv, old_argv, *array_size * sizeof(*argv));
 565                 *array_size = new_size;
 566         }
 567
 568         kfree(old_argv);
 569         return argv;
 570 }
 571
 572 /*
 573  * Destructively splits up the argument list to pass to ctr.
 574  */
 575 int dm_split_args(int *argc, char ***argvp, char *input)
 576 {
 577         char *start, *end = input, *out, **argv = NULL;
 578         unsigned array_size = 0;
 579
 580         *argc = 0;
 581
 582         if (!input) {
 583                 *argvp = NULL;
 584                 return 0;
 585         }
 586
 587         argv = realloc_argv(&array_size, argv);
 588         if (!argv)
 589                 return -ENOMEM;
 590
 591         while (1) {
 592                 /* Skip whitespace */
 593                 start = skip_spaces(end);
 594
 595                 if (!*start)
 596                         break;  /* success, we hit the end */
 597
 598                 /* 'out' is used to remove any back-quotes */
 599                 end = out = start;
 600                 while (*end) {
 601                         /* Everything apart from '\0' can be quoted */
 602                         if (*end == '\\' && *(end + 1)) {
 603                                 *out++ = *(end + 1);
 604                                 end += 2;
 605                                 continue;
 606                         }
 607
 608                         if (isspace(*end))
 609                                 break;  /* end of token */
 610
 611                         *out++ = *end++;
 612                 }
 613
 614                 /* have we already filled the array ? */
 615                 if ((*argc + 1) > array_size) {
 616                         argv = realloc_argv(&array_size, argv);
 617                         if (!argv)
 618                                 return -ENOMEM;
 619                 }
 620
 621                 /* we know this is whitespace */
 622                 if (*end)
 623                         end++;
 624
 625                 /* terminate the string and put it in the array */
 626                 *out = '\0';
 627                 argv[*argc] = start;
 628                 (*argc)++;
 629         }
 630
 631         *argvp = argv;
 632         return 0;
 633 }
 634
 635 /*
 636  * Impose necessary and sufficient conditions on a devices's table such
 637  * that any incoming bio which respects its logical_block_size can be
 638  * processed successfully.  If it falls across the boundary between
 639  * two or more targets, the size of each piece it gets split into must
 640  * be compatible with the logical_block_size of the target processing it.
 641  */
 642 static int validate_hardware_logical_block_alignment(struct dm_table *table,
 643                                                  struct queue_limits *limits)
 644 {
 645         /*
 646          * This function uses arithmetic modulo the logical_block_size
 647          * (in units of 512-byte sectors).
 648          */
 649         unsigned short device_logical_block_size_sects =
 650                 limits->logical_block_size >> SECTOR_SHIFT;
 651
 652         /*
 653          * Offset of the start of the next table entry, mod logical_block_size.
 654          */
 655         unsigned short next_target_start = 0;
 656
 657         /*
 658          * Given an aligned bio that extends beyond the end of a
 659          * target, how many sectors must the next target handle?
 660          */
 661         unsigned short remaining = 0;
 662
 663         struct dm_target *uninitialized_var(ti);
 664         struct queue_limits ti_limits;
 665         unsigned i;
 666
 667         /*
 668          * Check each entry in the table in turn.
 669          */
 670         for (i = 0; i < dm_table_get_num_targets(table); i++) {
 671                 ti = dm_table_get_target(table, i);
 672
 673                 blk_set_stacking_limits(&ti_limits);
 674
 675                 /* combine all target devices' limits */
 676                 if (ti->type->iterate_devices)
 677                         ti->type->iterate_devices(ti, dm_set_device_limits,
 678                                                   &ti_limits);
 679
 680                 /*
 681                  * If the remaining sectors fall entirely within this
 682                  * table entry are they compatible with its logical_block_size?
 683                  */
 684                 if (remaining < ti->len &&
 685                     remaining & ((ti_limits.logical_block_size >>
 686                                   SECTOR_SHIFT) - 1))
 687                         break;  /* Error */
 688
 689                 next_target_start =
 690                     (unsigned short) ((next_target_start + ti->len) &
 691                                       (device_logical_block_size_sects - 1));
 692                 remaining = next_target_start ?
 693                     device_logical_block_size_sects - next_target_start : 0;
 694         }
 695
 696         if (remaining) {
 697                 DMWARN("%s: table line %u (start sect %llu len %llu) "
 698                        "not aligned to h/w logical block size %u",
 699                        dm_device_name(table->md), i,
 700                        (unsigned long long) ti->begin,
 701                        (unsigned long long) ti->len,
 702                        limits->logical_block_size);
 703                 return -EINVAL;
 704         }
 705
 706         return 0;
 707 }
 708
 709 int dm_table_add_target(struct dm_table *t, const char *type,
 710                         sector_t start, sector_t len, char *params)
 711 {
 712         int r = -EINVAL, argc;
 713         char **argv;
 714         struct dm_target *tgt;
 715
 716         if (t->singleton) {
 717                 DMERR("%s: target type %s must appear alone in table",
 718                       dm_device_name(t->md), t->targets->type->name);
 719                 return -EINVAL;
 720         }
 721
 722         BUG_ON(t->num_targets >= t->num_allocated);
 723
 724         tgt = t->targets + t->num_targets;
 725         memset(tgt, 0, sizeof(*tgt));
 726
 727         if (!len) {
 728                 DMERR("%s: zero-length target", dm_device_name(t->md));
 729                 return -EINVAL;
 730         }
 731
 732         tgt->type = dm_get_target_type(type);
 733         if (!tgt->type) {
 734                 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
 735                 return -EINVAL;
 736         }
 737
 738         if (dm_target_needs_singleton(tgt->type)) {
 739                 if (t->num_targets) {
 740                         tgt->error = "singleton target type must appear alone in table";
 741                         goto bad;
 742                 }
 743                 t->singleton = true;
 744         }
 745
 746         if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
 747                 tgt->error = "target type may not be included in a read-only table";
 748                 goto bad;
 749         }
 750
 751         if (t->immutable_target_type) {
 752                 if (t->immutable_target_type != tgt->type) {
 753                         tgt->error = "immutable target type cannot be mixed with other target types";
 754                         goto bad;
 755                 }
 756         } else if (dm_target_is_immutable(tgt->type)) {
 757                 if (t->num_targets) {
 758                         tgt->error = "immutable target type cannot be mixed with other target types";
 759                         goto bad;
 760                 }
 761                 t->immutable_target_type = tgt->type;
 762         }
 763
 764         if (dm_target_has_integrity(tgt->type))
 765                 t->integrity_added = 1;
 766
 767         tgt->table = t;
 768         tgt->begin = start;
 769         tgt->len = len;
 770         tgt->error = "Unknown error";
 771
 772         /*
 773          * Does this target adjoin the previous one ?
 774          */
 775         if (!adjoin(t, tgt)) {
 776                 tgt->error = "Gap in table";
 777                 goto bad;
 778         }
 779
 780         r = dm_split_args(&argc, &argv, params);
 781         if (r) {
 782                 tgt->error = "couldn't split parameters (insufficient memory)";
 783                 goto bad;
 784         }
 785
 786         r = tgt->type->ctr(tgt, argc, argv);
 787         kfree(argv);
 788         if (r)
 789                 goto bad;
 790
 791         t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
 792
 793         if (!tgt->num_discard_bios && tgt->discards_supported)
 794                 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
 795                        dm_device_name(t->md), type);
 796
 797         return 0;
 798
 799  bad:
 800         DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
 801         dm_put_target_type(tgt->type);
 802         return r;
 803 }
 804
 805 /*
 806  * Target argument parsing helpers.
 807  */
 808 static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
 809                              unsigned *value, char **error, unsigned grouped)
 810 {
 811         const char *arg_str = dm_shift_arg(arg_set);
 812         char dummy;
 813
 814         if (!arg_str ||
 815             (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
 816             (*value < arg->min) ||
 817             (*value > arg->max) ||
 818             (grouped && arg_set->argc < *value)) {
 819                 *error = arg->error;
 820                 return -EINVAL;
 821         }
 822
 823         return 0;
 824 }
 825
 826 int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
 827                 unsigned *value, char **error)
 828 {
 829         return validate_next_arg(arg, arg_set, value, error, 0);
 830 }
 831 EXPORT_SYMBOL(dm_read_arg);
 832
 833 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
 834                       unsigned *value, char **error)
 835 {
 836         return validate_next_arg(arg, arg_set, value, error, 1);
 837 }
 838 EXPORT_SYMBOL(dm_read_arg_group);
 839
 840 const char *dm_shift_arg(struct dm_arg_set *as)
 841 {
 842         char *r;
 843
 844         if (as->argc) {
 845                 as->argc--;
 846                 r = *as->argv;
 847                 as->argv++;
 848                 return r;
 849         }
 850
 851         return NULL;
 852 }
 853 EXPORT_SYMBOL(dm_shift_arg);
 854
 855 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
 856 {
 857         BUG_ON(as->argc < num_args);
 858         as->argc -= num_args;
 859         as->argv += num_args;
 860 }
 861 EXPORT_SYMBOL(dm_consume_args);
 862
 863 static bool __table_type_bio_based(enum dm_queue_mode table_type)
 864 {
 865         return (table_type == DM_TYPE_BIO_BASED ||
 866                 table_type == DM_TYPE_DAX_BIO_BASED);
 867 }
 868
 869 static bool __table_type_request_based(enum dm_queue_mode table_type)
 870 {
 871         return (table_type == DM_TYPE_REQUEST_BASED ||
 872                 table_type == DM_TYPE_MQ_REQUEST_BASED);
 873 }
 874
 875 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
 876 {
 877         t->type = type;
 878 }
 879 EXPORT_SYMBOL_GPL(dm_table_set_type);
 880
 881 static int device_supports_dax(struct dm_target *ti, struct dm_dev *dev,
 882                                sector_t start, sector_t len, void *data)
 883 {
 884         struct request_queue *q = bdev_get_queue(dev->bdev);
 885
 886         return q && blk_queue_dax(q);
 887 }
 888
 889 static bool dm_table_supports_dax(struct dm_table *t)
 890 {
 891         struct dm_target *ti;
 892         unsigned i;
 893
 894         /* Ensure that all targets support DAX. */
 895         for (i = 0; i < dm_table_get_num_targets(t); i++) {
 896                 ti = dm_table_get_target(t, i);
 897
 898                 if (!ti->type->direct_access)
 899                         return false;
 900
 901                 if (!ti->type->iterate_devices ||
 902                     !ti->type->iterate_devices(ti, device_supports_dax, NULL))
 903                         return false;
 904         }
 905
 906         return true;
 907 }
 908
 909 static int dm_table_determine_type(struct dm_table *t)
 910 {
 911         unsigned i;
 912         unsigned bio_based = 0, request_based = 0, hybrid = 0;
 913         unsigned sq_count = 0, mq_count = 0;
 914         struct dm_target *tgt;
 915         struct dm_dev_internal *dd;
 916         struct list_head *devices = dm_table_get_devices(t);
 917         enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
 918
 919         if (t->type != DM_TYPE_NONE) {
 920                 /* target already set the table's type */
 921                 if (t->type == DM_TYPE_BIO_BASED)
 922                         return 0;
 923                 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
 924                 goto verify_rq_based;
 925         }
 926
 927         for (i = 0; i < t->num_targets; i++) {
 928                 tgt = t->targets + i;
 929                 if (dm_target_hybrid(tgt))
 930                         hybrid = 1;
 931                 else if (dm_target_request_based(tgt))
 932                         request_based = 1;
 933                 else
 934                         bio_based = 1;
 935
 936                 if (bio_based && request_based) {
 937                         DMWARN("Inconsistent table: different target types"
 938                                " can't be mixed up");
 939                         return -EINVAL;
 940                 }
 941         }
 942
 943         if (hybrid && !bio_based && !request_based) {
 944                 /*
 945                  * The targets can work either way.
 946                  * Determine the type from the live device.
 947                  * Default to bio-based if device is new.
 948                  */
 949                 if (__table_type_request_based(live_md_type))
 950                         request_based = 1;
 951                 else
 952                         bio_based = 1;
 953         }
 954
 955         if (bio_based) {
 956                 /* We must use this table as bio-based */
 957                 t->type = DM_TYPE_BIO_BASED;
 958                 if (dm_table_supports_dax(t) ||
 959                     (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED))
 960                         t->type = DM_TYPE_DAX_BIO_BASED;
 961                 return 0;
 962         }
 963
 964         BUG_ON(!request_based); /* No targets in this table */
 965
 966         /*
 967          * The only way to establish DM_TYPE_MQ_REQUEST_BASED is by
 968          * having a compatible target use dm_table_set_type.
 969          */
 970         t->type = DM_TYPE_REQUEST_BASED;
 971
 972 verify_rq_based:
 973         /*
 974          * Request-based dm supports only tables that have a single target now.
 975          * To support multiple targets, request splitting support is needed,
 976          * and that needs lots of changes in the block-layer.
 977          * (e.g. request completion process for partial completion.)
 978          */
 979         if (t->num_targets > 1) {
 980                 DMWARN("Request-based dm doesn't support multiple targets yet");
 981                 return -EINVAL;
 982         }
 983
 984         if (list_empty(devices)) {
 985                 int srcu_idx;
 986                 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
 987
 988                 /* inherit live table's type and all_blk_mq */
 989                 if (live_table) {
 990                         t->type = live_table->type;
 991                         t->all_blk_mq = live_table->all_blk_mq;
 992                 }
 993                 dm_put_live_table(t->md, srcu_idx);
 994                 return 0;
 995         }
 996
 997         /* Non-request-stackable devices can't be used for request-based dm */
 998         list_for_each_entry(dd, devices, list) {
 999                 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1000
1001                 if (!blk_queue_stackable(q)) {
1002                         DMERR("table load rejected: including"
1003                               " non-request-stackable devices");
1004                         return -EINVAL;
1005                 }
1006
1007                 if (q->mq_ops)
1008                         mq_count++;
1009                 else
1010                         sq_count++;
1011         }
1012         if (sq_count && mq_count) {
1013                 DMERR("table load rejected: not all devices are blk-mq request-stackable");
1014                 return -EINVAL;
1015         }
1016         t->all_blk_mq = mq_count > 0;
1017
1018         if (t->type == DM_TYPE_MQ_REQUEST_BASED && !t->all_blk_mq) {
1019                 DMERR("table load rejected: all devices are not blk-mq request-stackable");
1020                 return -EINVAL;
1021         }
1022
1023         return 0;
1024 }
1025
1026 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
1027 {
1028         return t->type;
1029 }
1030
1031 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
1032 {
1033         return t->immutable_target_type;
1034 }
1035
1036 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
1037 {
1038         /* Immutable target is implicitly a singleton */
1039         if (t->num_targets > 1 ||
1040             !dm_target_is_immutable(t->targets[0].type))
1041                 return NULL;
1042
1043         return t->targets;
1044 }
1045
1046 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1047 {
1048         struct dm_target *ti;
1049         unsigned i;
1050
1051         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1052                 ti = dm_table_get_target(t, i);
1053                 if (dm_target_is_wildcard(ti->type))
1054                         return ti;
1055         }
1056
1057         return NULL;
1058 }
1059
1060 bool dm_table_bio_based(struct dm_table *t)
1061 {
1062         return __table_type_bio_based(dm_table_get_type(t));
1063 }
1064
1065 bool dm_table_request_based(struct dm_table *t)
1066 {
1067         return __table_type_request_based(dm_table_get_type(t));
1068 }
1069
1070 bool dm_table_all_blk_mq_devices(struct dm_table *t)
1071 {
1072         return t->all_blk_mq;
1073 }
1074
1075 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1076 {
1077         enum dm_queue_mode type = dm_table_get_type(t);
1078         unsigned per_io_data_size = 0;
1079         struct dm_target *tgt;
1080         unsigned i;
1081
1082         if (unlikely(type == DM_TYPE_NONE)) {
1083                 DMWARN("no table type is set, can't allocate mempools");
1084                 return -EINVAL;
1085         }
1086
1087         if (__table_type_bio_based(type))
1088                 for (i = 0; i < t->num_targets; i++) {
1089                         tgt = t->targets + i;
1090                         per_io_data_size = max(per_io_data_size, tgt->per_io_data_size);
1091                 }
1092
1093         t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported, per_io_data_size);
1094         if (!t->mempools)
1095                 return -ENOMEM;
1096
1097         return 0;
1098 }
1099
1100 void dm_table_free_md_mempools(struct dm_table *t)
1101 {
1102         dm_free_md_mempools(t->mempools);
1103         t->mempools = NULL;
1104 }
1105
1106 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1107 {
1108         return t->mempools;
1109 }
1110
1111 static int setup_indexes(struct dm_table *t)
1112 {
1113         int i;
1114         unsigned int total = 0;
1115         sector_t *indexes;
1116
1117         /* allocate the space for *all* the indexes */
1118         for (i = t->depth - 2; i >= 0; i--) {
1119                 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1120                 total += t->counts[i];
1121         }
1122
1123         indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1124         if (!indexes)
1125                 return -ENOMEM;
1126
1127         /* set up internal nodes, bottom-up */
1128         for (i = t->depth - 2; i >= 0; i--) {
1129                 t->index[i] = indexes;
1130                 indexes += (KEYS_PER_NODE * t->counts[i]);
1131                 setup_btree_index(i, t);
1132         }
1133
1134         return 0;
1135 }
1136
1137 /*
1138  * Builds the btree to index the map.
1139  */
1140 static int dm_table_build_index(struct dm_table *t)
1141 {
1142         int r = 0;
1143         unsigned int leaf_nodes;
1144
1145         /* how many indexes will the btree have ? */
1146         leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1147         t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1148
1149         /* leaf layer has already been set up */
1150         t->counts[t->depth - 1] = leaf_nodes;
1151         t->index[t->depth - 1] = t->highs;
1152
1153         if (t->depth >= 2)
1154                 r = setup_indexes(t);
1155
1156         return r;
1157 }
1158
1159 static bool integrity_profile_exists(struct gendisk *disk)
1160 {
1161         return !!blk_get_integrity(disk);
1162 }
1163
1164 /*
1165  * Get a disk whose integrity profile reflects the table's profile.
1166  * Returns NULL if integrity support was inconsistent or unavailable.
1167  */
1168 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1169 {
1170         struct list_head *devices = dm_table_get_devices(t);
1171         struct dm_dev_internal *dd = NULL;
1172         struct gendisk *prev_disk = NULL, *template_disk = NULL;
1173         unsigned i;
1174
1175         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1176                 struct dm_target *ti = dm_table_get_target(t, i);
1177                 if (!dm_target_passes_integrity(ti->type))
1178                         goto no_integrity;
1179         }
1180
1181         list_for_each_entry(dd, devices, list) {
1182                 template_disk = dd->dm_dev->bdev->bd_disk;
1183                 if (!integrity_profile_exists(template_disk))
1184                         goto no_integrity;
1185                 else if (prev_disk &&
1186                          blk_integrity_compare(prev_disk, template_disk) < 0)
1187                         goto no_integrity;
1188                 prev_disk = template_disk;
1189         }
1190
1191         return template_disk;
1192
1193 no_integrity:
1194         if (prev_disk)
1195                 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1196                        dm_device_name(t->md),
1197                        prev_disk->disk_name,
1198                        template_disk->disk_name);
1199         return NULL;
1200 }
1201
1202 /*
1203  * Register the mapped device for blk_integrity support if the
1204  * underlying devices have an integrity profile.  But all devices may
1205  * not have matching profiles (checking all devices isn't reliable
1206  * during table load because this table may use other DM device(s) which
1207  * must be resumed before they will have an initialized integity
1208  * profile).  Consequently, stacked DM devices force a 2 stage integrity
1209  * profile validation: First pass during table load, final pass during
1210  * resume.
1211  */
1212 static int dm_table_register_integrity(struct dm_table *t)
1213 {
1214         struct mapped_device *md = t->md;
1215         struct gendisk *template_disk = NULL;
1216
1217         /* If target handles integrity itself do not register it here. */
1218         if (t->integrity_added)
1219                 return 0;
1220
1221         template_disk = dm_table_get_integrity_disk(t);
1222         if (!template_disk)
1223                 return 0;
1224
1225         if (!integrity_profile_exists(dm_disk(md))) {
1226                 t->integrity_supported = true;
1227                 /*
1228                  * Register integrity profile during table load; we can do
1229                  * this because the final profile must match during resume.
1230                  */
1231                 blk_integrity_register(dm_disk(md),
1232                                        blk_get_integrity(template_disk));
1233                 return 0;
1234         }
1235
1236         /*
1237          * If DM device already has an initialized integrity
1238          * profile the new profile should not conflict.
1239          */
1240         if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1241                 DMWARN("%s: conflict with existing integrity profile: "
1242                        "%s profile mismatch",
1243                        dm_device_name(t->md),
1244                        template_disk->disk_name);
1245                 return 1;
1246         }
1247
1248         /* Preserve existing integrity profile */
1249         t->integrity_supported = true;
1250         return 0;
1251 }
1252
1253 /*
1254  * Prepares the table for use by building the indices,
1255  * setting the type, and allocating mempools.
1256  */
1257 int dm_table_complete(struct dm_table *t)
1258 {
1259         int r;
1260
1261         r = dm_table_determine_type(t);
1262         if (r) {
1263                 DMERR("unable to determine table type");
1264                 return r;
1265         }
1266
1267         r = dm_table_build_index(t);
1268         if (r) {
1269                 DMERR("unable to build btrees");
1270                 return r;
1271         }
1272
1273         r = dm_table_register_integrity(t);
1274         if (r) {
1275                 DMERR("could not register integrity profile.");
1276                 return r;
1277         }
1278
1279         r = dm_table_alloc_md_mempools(t, t->md);
1280         if (r)
1281                 DMERR("unable to allocate mempools");
1282
1283         return r;
1284 }
1285
1286 static DEFINE_MUTEX(_event_lock);
1287 void dm_table_event_callback(struct dm_table *t,
1288                              void (*fn)(void *), void *context)
1289 {
1290         mutex_lock(&_event_lock);
1291         t->event_fn = fn;
1292         t->event_context = context;
1293         mutex_unlock(&_event_lock);
1294 }
1295
1296 void dm_table_event(struct dm_table *t)
1297 {
1298         /*
1299          * You can no longer call dm_table_event() from interrupt
1300          * context, use a bottom half instead.
1301          */
1302         BUG_ON(in_interrupt());
1303
1304         mutex_lock(&_event_lock);
1305         if (t->event_fn)
1306                 t->event_fn(t->event_context);
1307         mutex_unlock(&_event_lock);
1308 }
1309 EXPORT_SYMBOL(dm_table_event);
1310
1311 sector_t dm_table_get_size(struct dm_table *t)
1312 {
1313         return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1314 }
1315 EXPORT_SYMBOL(dm_table_get_size);
1316
1317 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1318 {
1319         if (index >= t->num_targets)
1320                 return NULL;
1321
1322         return t->targets + index;
1323 }
1324
1325 /*
1326  * Search the btree for the correct target.
1327  *
1328  * Caller should check returned pointer with dm_target_is_valid()
1329  * to trap I/O beyond end of device.
1330  */
1331 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1332 {
1333         unsigned int l, n = 0, k = 0;
1334         sector_t *node;
1335
1336         for (l = 0; l < t->depth; l++) {
1337                 n = get_child(n, k);
1338                 node = get_node(t, l, n);
1339
1340                 for (k = 0; k < KEYS_PER_NODE; k++)
1341                         if (node[k] >= sector)
1342                                 break;
1343         }
1344
1345         return &t->targets[(KEYS_PER_NODE * n) + k];
1346 }
1347
1348 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1349                         sector_t start, sector_t len, void *data)
1350 {
1351         unsigned *num_devices = data;
1352
1353         (*num_devices)++;
1354
1355         return 0;
1356 }
1357
1358 /*
1359  * Check whether a table has no data devices attached using each
1360  * target's iterate_devices method.
1361  * Returns false if the result is unknown because a target doesn't
1362  * support iterate_devices.
1363  */
1364 bool dm_table_has_no_data_devices(struct dm_table *table)
1365 {
1366         struct dm_target *ti;
1367         unsigned i, num_devices;
1368
1369         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1370                 ti = dm_table_get_target(table, i);
1371
1372                 if (!ti->type->iterate_devices)
1373                         return false;
1374
1375                 num_devices = 0;
1376                 ti->type->iterate_devices(ti, count_device, &num_devices);
1377                 if (num_devices)
1378                         return false;
1379         }
1380
1381         return true;
1382 }
1383
1384 static int device_is_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1385                                  sector_t start, sector_t len, void *data)
1386 {
1387         struct request_queue *q = bdev_get_queue(dev->bdev);
1388         enum blk_zoned_model *zoned_model = data;
1389
1390         return q && blk_queue_zoned_model(q) == *zoned_model;
1391 }
1392
1393 static bool dm_table_supports_zoned_model(struct dm_table *t,
1394                                           enum blk_zoned_model zoned_model)
1395 {
1396         struct dm_target *ti;
1397         unsigned i;
1398
1399         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1400                 ti = dm_table_get_target(t, i);
1401
1402                 if (zoned_model == BLK_ZONED_HM &&
1403                     !dm_target_supports_zoned_hm(ti->type))
1404                         return false;
1405
1406                 if (!ti->type->iterate_devices ||
1407                     !ti->type->iterate_devices(ti, device_is_zoned_model, &zoned_model))
1408                         return false;
1409         }
1410
1411         return true;
1412 }
1413
1414 static int device_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1415                                        sector_t start, sector_t len, void *data)
1416 {
1417         struct request_queue *q = bdev_get_queue(dev->bdev);
1418         unsigned int *zone_sectors = data;
1419
1420         return q && blk_queue_zone_sectors(q) == *zone_sectors;
1421 }
1422
1423 static bool dm_table_matches_zone_sectors(struct dm_table *t,
1424                                           unsigned int zone_sectors)
1425 {
1426         struct dm_target *ti;
1427         unsigned i;
1428
1429         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1430                 ti = dm_table_get_target(t, i);
1431
1432                 if (!ti->type->iterate_devices ||
1433                     !ti->type->iterate_devices(ti, device_matches_zone_sectors, &zone_sectors))
1434                         return false;
1435         }
1436
1437         return true;
1438 }
1439
1440 static int validate_hardware_zoned_model(struct dm_table *table,
1441                                          enum blk_zoned_model zoned_model,
1442                                          unsigned int zone_sectors)
1443 {
1444         if (zoned_model == BLK_ZONED_NONE)
1445                 return 0;
1446
1447         if (!dm_table_supports_zoned_model(table, zoned_model)) {
1448                 DMERR("%s: zoned model is not consistent across all devices",
1449                       dm_device_name(table->md));
1450                 return -EINVAL;
1451         }
1452
1453         /* Check zone size validity and compatibility */
1454         if (!zone_sectors || !is_power_of_2(zone_sectors))
1455                 return -EINVAL;
1456
1457         if (!dm_table_matches_zone_sectors(table, zone_sectors)) {
1458                 DMERR("%s: zone sectors is not consistent across all devices",
1459                       dm_device_name(table->md));
1460                 return -EINVAL;
1461         }
1462
1463         return 0;
1464 }
1465
1466 /*
1467  * Establish the new table's queue_limits and validate them.
1468  */
1469 int dm_calculate_queue_limits(struct dm_table *table,
1470                               struct queue_limits *limits)
1471 {
1472         struct dm_target *ti;
1473         struct queue_limits ti_limits;
1474         unsigned i;
1475         enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1476         unsigned int zone_sectors = 0;
1477
1478         blk_set_stacking_limits(limits);
1479
1480         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1481                 blk_set_stacking_limits(&ti_limits);
1482
1483                 ti = dm_table_get_target(table, i);
1484
1485                 if (!ti->type->iterate_devices)
1486                         goto combine_limits;
1487
1488                 /*
1489                  * Combine queue limits of all the devices this target uses.
1490                  */
1491                 ti->type->iterate_devices(ti, dm_set_device_limits,
1492                                           &ti_limits);
1493
1494                 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1495                         /*
1496                          * After stacking all limits, validate all devices
1497                          * in table support this zoned model and zone sectors.
1498                          */
1499                         zoned_model = ti_limits.zoned;
1500                         zone_sectors = ti_limits.chunk_sectors;
1501                 }
1502
1503                 /* Set I/O hints portion of queue limits */
1504                 if (ti->type->io_hints)
1505                         ti->type->io_hints(ti, &ti_limits);
1506
1507                 /*
1508                  * Check each device area is consistent with the target's
1509                  * overall queue limits.
1510                  */
1511                 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1512                                               &ti_limits))
1513                         return -EINVAL;
1514
1515 combine_limits:
1516                 /*
1517                  * Merge this target's queue limits into the overall limits
1518                  * for the table.
1519                  */
1520                 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1521                         DMWARN("%s: adding target device "
1522                                "(start sect %llu len %llu) "
1523                                "caused an alignment inconsistency",
1524                                dm_device_name(table->md),
1525                                (unsigned long long) ti->begin,
1526                                (unsigned long long) ti->len);
1527
1528                 /*
1529                  * FIXME: this should likely be moved to blk_stack_limits(), would
1530                  * also eliminate limits->zoned stacking hack in dm_set_device_limits()
1531                  */
1532                 if (limits->zoned == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1533                         /*
1534                          * By default, the stacked limits zoned model is set to
1535                          * BLK_ZONED_NONE in blk_set_stacking_limits(). Update
1536                          * this model using the first target model reported
1537                          * that is not BLK_ZONED_NONE. This will be either the
1538                          * first target device zoned model or the model reported
1539                          * by the target .io_hints.
1540                          */
1541                         limits->zoned = ti_limits.zoned;
1542                 }
1543         }
1544
1545         /*
1546          * Verify that the zoned model and zone sectors, as determined before
1547          * any .io_hints override, are the same across all devices in the table.
1548          * - this is especially relevant if .io_hints is emulating a disk-managed
1549          *   zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1550          * BUT...
1551          */
1552         if (limits->zoned != BLK_ZONED_NONE) {
1553                 /*
1554                  * ...IF the above limits stacking determined a zoned model
1555                  * validate that all of the table's devices conform to it.
1556                  */
1557                 zoned_model = limits->zoned;
1558                 zone_sectors = limits->chunk_sectors;
1559         }
1560         if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1561                 return -EINVAL;
1562
1563         return validate_hardware_logical_block_alignment(table, limits);
1564 }
1565
1566 /*
1567  * Verify that all devices have an integrity profile that matches the
1568  * DM device's registered integrity profile.  If the profiles don't
1569  * match then unregister the DM device's integrity profile.
1570  */
1571 static void dm_table_verify_integrity(struct dm_table *t)
1572 {
1573         struct gendisk *template_disk = NULL;
1574
1575         if (t->integrity_added)
1576                 return;
1577
1578         if (t->integrity_supported) {
1579                 /*
1580                  * Verify that the original integrity profile
1581                  * matches all the devices in this table.
1582                  */
1583                 template_disk = dm_table_get_integrity_disk(t);
1584                 if (template_disk &&
1585                     blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1586                         return;
1587         }
1588
1589         if (integrity_profile_exists(dm_disk(t->md))) {
1590                 DMWARN("%s: unable to establish an integrity profile",
1591                        dm_device_name(t->md));
1592                 blk_integrity_unregister(dm_disk(t->md));
1593         }
1594 }
1595
1596 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1597                                 sector_t start, sector_t len, void *data)
1598 {
1599         unsigned long flush = (unsigned long) data;
1600         struct request_queue *q = bdev_get_queue(dev->bdev);
1601
1602         return q && (q->queue_flags & flush);
1603 }
1604
1605 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1606 {
1607         struct dm_target *ti;
1608         unsigned i;
1609
1610         /*
1611          * Require at least one underlying device to support flushes.
1612          * t->devices includes internal dm devices such as mirror logs
1613          * so we need to use iterate_devices here, which targets
1614          * supporting flushes must provide.
1615          */
1616         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1617                 ti = dm_table_get_target(t, i);
1618
1619                 if (!ti->num_flush_bios)
1620                         continue;
1621
1622                 if (ti->flush_supported)
1623                         return true;
1624
1625                 if (ti->type->iterate_devices &&
1626                     ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1627                         return true;
1628         }
1629
1630         return false;
1631 }
1632
1633 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1634                             sector_t start, sector_t len, void *data)
1635 {
1636         struct request_queue *q = bdev_get_queue(dev->bdev);
1637
1638         return q && blk_queue_nonrot(q);
1639 }
1640
1641 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1642                              sector_t start, sector_t len, void *data)
1643 {
1644         struct request_queue *q = bdev_get_queue(dev->bdev);
1645
1646         return q && !blk_queue_add_random(q);
1647 }
1648
1649 static int queue_supports_sg_merge(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 && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1655 }
1656
1657 static bool dm_table_all_devices_attribute(struct dm_table *t,
1658                                            iterate_devices_callout_fn func)
1659 {
1660         struct dm_target *ti;
1661         unsigned i;
1662
1663         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1664                 ti = dm_table_get_target(t, i);
1665
1666                 if (!ti->type->iterate_devices ||
1667                     !ti->type->iterate_devices(ti, func, NULL))
1668                         return false;
1669         }
1670
1671         return true;
1672 }
1673
1674 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1675                                          sector_t start, sector_t len, void *data)
1676 {
1677         struct request_queue *q = bdev_get_queue(dev->bdev);
1678
1679         return q && !q->limits.max_write_same_sectors;
1680 }
1681
1682 static bool dm_table_supports_write_same(struct dm_table *t)
1683 {
1684         struct dm_target *ti;
1685         unsigned i;
1686
1687         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1688                 ti = dm_table_get_target(t, i);
1689
1690                 if (!ti->num_write_same_bios)
1691                         return false;
1692
1693                 if (!ti->type->iterate_devices ||
1694                     ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1695                         return false;
1696         }
1697
1698         return true;
1699 }
1700
1701 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1702                                            sector_t start, sector_t len, void *data)
1703 {
1704         struct request_queue *q = bdev_get_queue(dev->bdev);
1705
1706         return q && !q->limits.max_write_zeroes_sectors;
1707 }
1708
1709 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1710 {
1711         struct dm_target *ti;
1712         unsigned i = 0;
1713
1714         while (i < dm_table_get_num_targets(t)) {
1715                 ti = dm_table_get_target(t, i++);
1716
1717                 if (!ti->num_write_zeroes_bios)
1718                         return false;
1719
1720                 if (!ti->type->iterate_devices ||
1721                     ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1722                         return false;
1723         }
1724
1725         return true;
1726 }
1727
1728
1729 static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1730                                   sector_t start, sector_t len, void *data)
1731 {
1732         struct request_queue *q = bdev_get_queue(dev->bdev);
1733
1734         return q && blk_queue_discard(q);
1735 }
1736
1737 static bool dm_table_supports_discards(struct dm_table *t)
1738 {
1739         struct dm_target *ti;
1740         unsigned i;
1741
1742         /*
1743          * Unless any target used by the table set discards_supported,
1744          * require at least one underlying device to support discards.
1745          * t->devices includes internal dm devices such as mirror logs
1746          * so we need to use iterate_devices here, which targets
1747          * supporting discard selectively must provide.
1748          */
1749         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1750                 ti = dm_table_get_target(t, i);
1751
1752                 if (!ti->num_discard_bios)
1753                         continue;
1754
1755                 if (ti->discards_supported)
1756                         return true;
1757
1758                 if (ti->type->iterate_devices &&
1759                     ti->type->iterate_devices(ti, device_discard_capable, NULL))
1760                         return true;
1761         }
1762
1763         return false;
1764 }
1765
1766 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1767                                struct queue_limits *limits)
1768 {
1769         bool wc = false, fua = false;
1770
1771         /*
1772          * Copy table's limits to the DM device's request_queue
1773          */
1774         q->limits = *limits;
1775
1776         if (!dm_table_supports_discards(t))
1777                 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1778         else
1779                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1780
1781         if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1782                 wc = true;
1783                 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1784                         fua = true;
1785         }
1786         blk_queue_write_cache(q, wc, fua);
1787
1788         /* Ensure that all underlying devices are non-rotational. */
1789         if (dm_table_all_devices_attribute(t, device_is_nonrot))
1790                 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1791         else
1792                 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1793
1794         if (!dm_table_supports_write_same(t))
1795                 q->limits.max_write_same_sectors = 0;
1796         if (!dm_table_supports_write_zeroes(t))
1797                 q->limits.max_write_zeroes_sectors = 0;
1798
1799         if (dm_table_all_devices_attribute(t, queue_supports_sg_merge))
1800                 queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1801         else
1802                 queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1803
1804         dm_table_verify_integrity(t);
1805
1806         /*
1807          * Determine whether or not this queue's I/O timings contribute
1808          * to the entropy pool, Only request-based targets use this.
1809          * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1810          * have it set.
1811          */
1812         if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1813                 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1814
1815         /*
1816          * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1817          * visible to other CPUs because, once the flag is set, incoming bios
1818          * are processed by request-based dm, which refers to the queue
1819          * settings.
1820          * Until the flag set, bios are passed to bio-based dm and queued to
1821          * md->deferred where queue settings are not needed yet.
1822          * Those bios are passed to request-based dm at the resume time.
1823          */
1824         smp_mb();
1825         if (dm_table_request_based(t))
1826                 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1827 }
1828
1829 unsigned int dm_table_get_num_targets(struct dm_table *t)
1830 {
1831         return t->num_targets;
1832 }
1833
1834 struct list_head *dm_table_get_devices(struct dm_table *t)
1835 {
1836         return &t->devices;
1837 }
1838
1839 fmode_t dm_table_get_mode(struct dm_table *t)
1840 {
1841         return t->mode;
1842 }
1843 EXPORT_SYMBOL(dm_table_get_mode);
1844
1845 enum suspend_mode {
1846         PRESUSPEND,
1847         PRESUSPEND_UNDO,
1848         POSTSUSPEND,
1849 };
1850
1851 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1852 {
1853         int i = t->num_targets;
1854         struct dm_target *ti = t->targets;
1855
1856         lockdep_assert_held(&t->md->suspend_lock);
1857
1858         while (i--) {
1859                 switch (mode) {
1860                 case PRESUSPEND:
1861                         if (ti->type->presuspend)
1862                                 ti->type->presuspend(ti);
1863                         break;
1864                 case PRESUSPEND_UNDO:
1865                         if (ti->type->presuspend_undo)
1866                                 ti->type->presuspend_undo(ti);
1867                         break;
1868                 case POSTSUSPEND:
1869                         if (ti->type->postsuspend)
1870                                 ti->type->postsuspend(ti);
1871                         break;
1872                 }
1873                 ti++;
1874         }
1875 }
1876
1877 void dm_table_presuspend_targets(struct dm_table *t)
1878 {
1879         if (!t)
1880                 return;
1881
1882         suspend_targets(t, PRESUSPEND);
1883 }
1884
1885 void dm_table_presuspend_undo_targets(struct dm_table *t)
1886 {
1887         if (!t)
1888                 return;
1889
1890         suspend_targets(t, PRESUSPEND_UNDO);
1891 }
1892
1893 void dm_table_postsuspend_targets(struct dm_table *t)
1894 {
1895         if (!t)
1896                 return;
1897
1898         suspend_targets(t, POSTSUSPEND);
1899 }
1900
1901 int dm_table_resume_targets(struct dm_table *t)
1902 {
1903         int i, r = 0;
1904
1905         lockdep_assert_held(&t->md->suspend_lock);
1906
1907         for (i = 0; i < t->num_targets; i++) {
1908                 struct dm_target *ti = t->targets + i;
1909
1910                 if (!ti->type->preresume)
1911                         continue;
1912
1913                 r = ti->type->preresume(ti);
1914                 if (r) {
1915                         DMERR("%s: %s: preresume failed, error = %d",
1916                               dm_device_name(t->md), ti->type->name, r);
1917                         return r;
1918                 }
1919         }
1920
1921         for (i = 0; i < t->num_targets; i++) {
1922                 struct dm_target *ti = t->targets + i;
1923
1924                 if (ti->type->resume)
1925                         ti->type->resume(ti);
1926         }
1927
1928         return 0;
1929 }
1930
1931 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1932 {
1933         list_add(&cb->list, &t->target_callbacks);
1934 }
1935 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1936
1937 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1938 {
1939         struct dm_dev_internal *dd;
1940         struct list_head *devices = dm_table_get_devices(t);
1941         struct dm_target_callbacks *cb;
1942         int r = 0;
1943
1944         list_for_each_entry(dd, devices, list) {
1945                 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1946                 char b[BDEVNAME_SIZE];
1947
1948                 if (likely(q))
1949                         r |= bdi_congested(q->backing_dev_info, bdi_bits);
1950                 else
1951                         DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1952                                      dm_device_name(t->md),
1953                                      bdevname(dd->dm_dev->bdev, b));
1954         }
1955
1956         list_for_each_entry(cb, &t->target_callbacks, list)
1957                 if (cb->congested_fn)
1958                         r |= cb->congested_fn(cb, bdi_bits);
1959
1960         return r;
1961 }
1962
1963 struct mapped_device *dm_table_get_md(struct dm_table *t)
1964 {
1965         return t->md;
1966 }
1967 EXPORT_SYMBOL(dm_table_get_md);
1968
1969 void dm_table_run_md_queue_async(struct dm_table *t)
1970 {
1971         struct mapped_device *md;
1972         struct request_queue *queue;
1973         unsigned long flags;
1974
1975         if (!dm_table_request_based(t))
1976                 return;
1977
1978         md = dm_table_get_md(t);
1979         queue = dm_get_md_queue(md);
1980         if (queue) {
1981                 if (queue->mq_ops)
1982                         blk_mq_run_hw_queues(queue, true);
1983                 else {
1984                         spin_lock_irqsave(queue->queue_lock, flags);
1985                         blk_run_queue_async(queue);
1986                         spin_unlock_irqrestore(queue->queue_lock, flags);
1987                 }
1988         }
1989 }
1990 EXPORT_SYMBOL(dm_table_run_md_queue_async);
1991