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         unsigned 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
  51         /*
  52          * Indicates the rw permissions for the new logical
  53          * device.  This should be a combination of FMODE_READ
  54          * and FMODE_WRITE.
  55          */
  56         fmode_t mode;
  57
  58         /* a list of devices used by this table */
  59         struct list_head devices;
  60
  61         /* events get handed up using this callback */
  62         void (*event_fn)(void *);
  63         void *event_context;
  64
  65         struct dm_md_mempools *mempools;
  66
  67         struct list_head target_callbacks;
  68 };
  69
  70 /*
  71  * Similar to ceiling(log_size(n))
  72  */
  73 static unsigned int int_log(unsigned int n, unsigned int base)
  74 {
  75         int result = 0;
  76
  77         while (n > 1) {
  78                 n = dm_div_up(n, base);
  79                 result++;
  80         }
  81
  82         return result;
  83 }
  84
  85 /*
  86  * Calculate the index of the child node of the n'th node k'th key.
  87  */
  88 static inline unsigned int get_child(unsigned int n, unsigned int k)
  89 {
  90         return (n * CHILDREN_PER_NODE) + k;
  91 }
  92
  93 /*
  94  * Return the n'th node of level l from table t.
  95  */
  96 static inline sector_t *get_node(struct dm_table *t,
  97                                  unsigned int l, unsigned int n)
  98 {
  99         return t->index[l] + (n * KEYS_PER_NODE);
 100 }
 101
 102 /*
 103  * Return the highest key that you could lookup from the n'th
 104  * node on level l of the btree.
 105  */
 106 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
 107 {
 108         for (; l < t->depth - 1; l++)
 109                 n = get_child(n, CHILDREN_PER_NODE - 1);
 110
 111         if (n >= t->counts[l])
 112                 return (sector_t) - 1;
 113
 114         return get_node(t, l, n)[KEYS_PER_NODE - 1];
 115 }
 116
 117 /*
 118  * Fills in a level of the btree based on the highs of the level
 119  * below it.
 120  */
 121 static int setup_btree_index(unsigned int l, struct dm_table *t)
 122 {
 123         unsigned int n, k;
 124         sector_t *node;
 125
 126         for (n = 0U; n < t->counts[l]; n++) {
 127                 node = get_node(t, l, n);
 128
 129                 for (k = 0U; k < KEYS_PER_NODE; k++)
 130                         node[k] = high(t, l + 1, get_child(n, k));
 131         }
 132
 133         return 0;
 134 }
 135
 136 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
 137 {
 138         unsigned long size;
 139         void *addr;
 140
 141         /*
 142          * Check that we're not going to overflow.
 143          */
 144         if (nmemb > (ULONG_MAX / elem_size))
 145                 return NULL;
 146
 147         size = nmemb * elem_size;
 148         addr = vzalloc(size);
 149
 150         return addr;
 151 }
 152 EXPORT_SYMBOL(dm_vcalloc);
 153
 154 /*
 155  * highs, and targets are managed as dynamic arrays during a
 156  * table load.
 157  */
 158 static int alloc_targets(struct dm_table *t, unsigned int num)
 159 {
 160         sector_t *n_highs;
 161         struct dm_target *n_targets;
 162
 163         /*
 164          * Allocate both the target array and offset array at once.
 165          * Append an empty entry to catch sectors beyond the end of
 166          * the device.
 167          */
 168         n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
 169                                           sizeof(sector_t));
 170         if (!n_highs)
 171                 return -ENOMEM;
 172
 173         n_targets = (struct dm_target *) (n_highs + num);
 174
 175         memset(n_highs, -1, sizeof(*n_highs) * num);
 176         vfree(t->highs);
 177
 178         t->num_allocated = num;
 179         t->highs = n_highs;
 180         t->targets = n_targets;
 181
 182         return 0;
 183 }
 184
 185 int dm_table_create(struct dm_table **result, fmode_t mode,
 186                     unsigned num_targets, struct mapped_device *md)
 187 {
 188         struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
 189
 190         if (!t)
 191                 return -ENOMEM;
 192
 193         INIT_LIST_HEAD(&t->devices);
 194         INIT_LIST_HEAD(&t->target_callbacks);
 195
 196         if (!num_targets)
 197                 num_targets = KEYS_PER_NODE;
 198
 199         num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
 200
 201         if (!num_targets) {
 202                 kfree(t);
 203                 return -ENOMEM;
 204         }
 205
 206         if (alloc_targets(t, num_targets)) {
 207                 kfree(t);
 208                 return -ENOMEM;
 209         }
 210
 211         t->type = DM_TYPE_NONE;
 212         t->mode = mode;
 213         t->md = md;
 214         *result = t;
 215         return 0;
 216 }
 217
 218 static void free_devices(struct list_head *devices, struct mapped_device *md)
 219 {
 220         struct list_head *tmp, *next;
 221
 222         list_for_each_safe(tmp, next, devices) {
 223                 struct dm_dev_internal *dd =
 224                     list_entry(tmp, struct dm_dev_internal, list);
 225                 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
 226                        dm_device_name(md), dd->dm_dev->name);
 227                 dm_put_table_device(md, dd->dm_dev);
 228                 kfree(dd);
 229         }
 230 }
 231
 232 void dm_table_destroy(struct dm_table *t)
 233 {
 234         unsigned int i;
 235
 236         if (!t)
 237                 return;
 238
 239         /* free the indexes */
 240         if (t->depth >= 2)
 241                 vfree(t->index[t->depth - 2]);
 242
 243         /* free the targets */
 244         for (i = 0; i < t->num_targets; i++) {
 245                 struct dm_target *tgt = t->targets + i;
 246
 247                 if (tgt->type->dtr)
 248                         tgt->type->dtr(tgt);
 249
 250                 dm_put_target_type(tgt->type);
 251         }
 252
 253         vfree(t->highs);
 254
 255         /* free the device list */
 256         free_devices(&t->devices, t->md);
 257
 258         dm_free_md_mempools(t->mempools);
 259
 260         kfree(t);
 261 }
 262
 263 /*
 264  * See if we've already got a device in the list.
 265  */
 266 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
 267 {
 268         struct dm_dev_internal *dd;
 269
 270         list_for_each_entry (dd, l, list)
 271                 if (dd->dm_dev->bdev->bd_dev == dev)
 272                         return dd;
 273
 274         return NULL;
 275 }
 276
 277 /*
 278  * If possible, this checks an area of a destination device is invalid.
 279  */
 280 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
 281                                   sector_t start, sector_t len, void *data)
 282 {
 283         struct request_queue *q;
 284         struct queue_limits *limits = data;
 285         struct block_device *bdev = dev->bdev;
 286         sector_t dev_size =
 287                 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
 288         unsigned short logical_block_size_sectors =
 289                 limits->logical_block_size >> SECTOR_SHIFT;
 290         char b[BDEVNAME_SIZE];
 291
 292         /*
 293          * Some devices exist without request functions,
 294          * such as loop devices not yet bound to backing files.
 295          * Forbid the use of such devices.
 296          */
 297         q = bdev_get_queue(bdev);
 298         if (!q || !q->make_request_fn) {
 299                 DMWARN("%s: %s is not yet initialised: "
 300                        "start=%llu, len=%llu, dev_size=%llu",
 301                        dm_device_name(ti->table->md), bdevname(bdev, b),
 302                        (unsigned long long)start,
 303                        (unsigned long long)len,
 304                        (unsigned long long)dev_size);
 305                 return 1;
 306         }
 307
 308         if (!dev_size)
 309                 return 0;
 310
 311         if ((start >= dev_size) || (start + len > dev_size)) {
 312                 DMWARN("%s: %s too small for target: "
 313                        "start=%llu, len=%llu, dev_size=%llu",
 314                        dm_device_name(ti->table->md), bdevname(bdev, b),
 315                        (unsigned long long)start,
 316                        (unsigned long long)len,
 317                        (unsigned long long)dev_size);
 318                 return 1;
 319         }
 320
 321         if (logical_block_size_sectors <= 1)
 322                 return 0;
 323
 324         if (start & (logical_block_size_sectors - 1)) {
 325                 DMWARN("%s: start=%llu not aligned to h/w "
 326                        "logical block size %u of %s",
 327                        dm_device_name(ti->table->md),
 328                        (unsigned long long)start,
 329                        limits->logical_block_size, bdevname(bdev, b));
 330                 return 1;
 331         }
 332
 333         if (len & (logical_block_size_sectors - 1)) {
 334                 DMWARN("%s: len=%llu not aligned to h/w "
 335                        "logical block size %u of %s",
 336                        dm_device_name(ti->table->md),
 337                        (unsigned long long)len,
 338                        limits->logical_block_size, bdevname(bdev, b));
 339                 return 1;
 340         }
 341
 342         return 0;
 343 }
 344
 345 /*
 346  * This upgrades the mode on an already open dm_dev, being
 347  * careful to leave things as they were if we fail to reopen the
 348  * device and not to touch the existing bdev field in case
 349  * it is accessed concurrently inside dm_table_any_congested().
 350  */
 351 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
 352                         struct mapped_device *md)
 353 {
 354         int r;
 355         struct dm_dev *old_dev, *new_dev;
 356
 357         old_dev = dd->dm_dev;
 358
 359         r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
 360                                 dd->dm_dev->mode | new_mode, &new_dev);
 361         if (r)
 362                 return r;
 363
 364         dd->dm_dev = new_dev;
 365         dm_put_table_device(md, old_dev);
 366
 367         return 0;
 368 }
 369
 370 /*
 371  * Convert the path to a device
 372  */
 373 dev_t dm_get_dev_t(const char *path)
 374 {
 375         dev_t uninitialized_var(dev);
 376         struct block_device *bdev;
 377
 378         bdev = lookup_bdev(path);
 379         if (IS_ERR(bdev))
 380                 dev = name_to_dev_t(path);
 381         else {
 382                 dev = bdev->bd_dev;
 383                 bdput(bdev);
 384         }
 385
 386         return dev;
 387 }
 388 EXPORT_SYMBOL_GPL(dm_get_dev_t);
 389
 390 /*
 391  * Add a device to the list, or just increment the usage count if
 392  * it's already present.
 393  */
 394 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
 395                   struct dm_dev **result)
 396 {
 397         int r;
 398         dev_t dev;
 399         struct dm_dev_internal *dd;
 400         struct dm_table *t = ti->table;
 401
 402         BUG_ON(!t);
 403
 404         dev = dm_get_dev_t(path);
 405         if (!dev)
 406                 return -ENODEV;
 407
 408         dd = find_device(&t->devices, dev);
 409         if (!dd) {
 410                 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
 411                 if (!dd)
 412                         return -ENOMEM;
 413
 414                 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
 415                         kfree(dd);
 416                         return r;
 417                 }
 418
 419                 atomic_set(&dd->count, 0);
 420                 list_add(&dd->list, &t->devices);
 421
 422         } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
 423                 r = upgrade_mode(dd, mode, t->md);
 424                 if (r)
 425                         return r;
 426         }
 427         atomic_inc(&dd->count);
 428
 429         *result = dd->dm_dev;
 430         return 0;
 431 }
 432 EXPORT_SYMBOL(dm_get_device);
 433
 434 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
 435                                 sector_t start, sector_t len, void *data)
 436 {
 437         struct queue_limits *limits = data;
 438         struct block_device *bdev = dev->bdev;
 439         struct request_queue *q = bdev_get_queue(bdev);
 440         char b[BDEVNAME_SIZE];
 441
 442         if (unlikely(!q)) {
 443                 DMWARN("%s: Cannot set limits for nonexistent device %s",
 444                        dm_device_name(ti->table->md), bdevname(bdev, b));
 445                 return 0;
 446         }
 447
 448         if (bdev_stack_limits(limits, bdev, start) < 0)
 449                 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
 450                        "physical_block_size=%u, logical_block_size=%u, "
 451                        "alignment_offset=%u, start=%llu",
 452                        dm_device_name(ti->table->md), bdevname(bdev, b),
 453                        q->limits.physical_block_size,
 454                        q->limits.logical_block_size,
 455                        q->limits.alignment_offset,
 456                        (unsigned long long) start << SECTOR_SHIFT);
 457
 458         return 0;
 459 }
 460
 461 /*
 462  * Decrement a device's use count and remove it if necessary.
 463  */
 464 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
 465 {
 466         int found = 0;
 467         struct list_head *devices = &ti->table->devices;
 468         struct dm_dev_internal *dd;
 469
 470         list_for_each_entry(dd, devices, list) {
 471                 if (dd->dm_dev == d) {
 472                         found = 1;
 473                         break;
 474                 }
 475         }
 476         if (!found) {
 477                 DMWARN("%s: device %s not in table devices list",
 478                        dm_device_name(ti->table->md), d->name);
 479                 return;
 480         }
 481         if (atomic_dec_and_test(&dd->count)) {
 482                 dm_put_table_device(ti->table->md, d);
 483                 list_del(&dd->list);
 484                 kfree(dd);
 485         }
 486 }
 487 EXPORT_SYMBOL(dm_put_device);
 488
 489 /*
 490  * Checks to see if the target joins onto the end of the table.
 491  */
 492 static int adjoin(struct dm_table *table, struct dm_target *ti)
 493 {
 494         struct dm_target *prev;
 495
 496         if (!table->num_targets)
 497                 return !ti->begin;
 498
 499         prev = &table->targets[table->num_targets - 1];
 500         return (ti->begin == (prev->begin + prev->len));
 501 }
 502
 503 /*
 504  * Used to dynamically allocate the arg array.
 505  *
 506  * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
 507  * process messages even if some device is suspended. These messages have a
 508  * small fixed number of arguments.
 509  *
 510  * On the other hand, dm-switch needs to process bulk data using messages and
 511  * excessive use of GFP_NOIO could cause trouble.
 512  */
 513 static char **realloc_argv(unsigned *array_size, char **old_argv)
 514 {
 515         char **argv;
 516         unsigned new_size;
 517         gfp_t gfp;
 518
 519         if (*array_size) {
 520                 new_size = *array_size * 2;
 521                 gfp = GFP_KERNEL;
 522         } else {
 523                 new_size = 8;
 524                 gfp = GFP_NOIO;
 525         }
 526         argv = kmalloc(new_size * sizeof(*argv), gfp);
 527         if (argv) {
 528                 memcpy(argv, old_argv, *array_size * sizeof(*argv));
 529                 *array_size = new_size;
 530         }
 531
 532         kfree(old_argv);
 533         return argv;
 534 }
 535
 536 /*
 537  * Destructively splits up the argument list to pass to ctr.
 538  */
 539 int dm_split_args(int *argc, char ***argvp, char *input)
 540 {
 541         char *start, *end = input, *out, **argv = NULL;
 542         unsigned array_size = 0;
 543
 544         *argc = 0;
 545
 546         if (!input) {
 547                 *argvp = NULL;
 548                 return 0;
 549         }
 550
 551         argv = realloc_argv(&array_size, argv);
 552         if (!argv)
 553                 return -ENOMEM;
 554
 555         while (1) {
 556                 /* Skip whitespace */
 557                 start = skip_spaces(end);
 558
 559                 if (!*start)
 560                         break;  /* success, we hit the end */
 561
 562                 /* 'out' is used to remove any back-quotes */
 563                 end = out = start;
 564                 while (*end) {
 565                         /* Everything apart from '\0' can be quoted */
 566                         if (*end == '\\' && *(end + 1)) {
 567                                 *out++ = *(end + 1);
 568                                 end += 2;
 569                                 continue;
 570                         }
 571
 572                         if (isspace(*end))
 573                                 break;  /* end of token */
 574
 575                         *out++ = *end++;
 576                 }
 577
 578                 /* have we already filled the array ? */
 579                 if ((*argc + 1) > array_size) {
 580                         argv = realloc_argv(&array_size, argv);
 581                         if (!argv)
 582                                 return -ENOMEM;
 583                 }
 584
 585                 /* we know this is whitespace */
 586                 if (*end)
 587                         end++;
 588
 589                 /* terminate the string and put it in the array */
 590                 *out = '\0';
 591                 argv[*argc] = start;
 592                 (*argc)++;
 593         }
 594
 595         *argvp = argv;
 596         return 0;
 597 }
 598
 599 /*
 600  * Impose necessary and sufficient conditions on a devices's table such
 601  * that any incoming bio which respects its logical_block_size can be
 602  * processed successfully.  If it falls across the boundary between
 603  * two or more targets, the size of each piece it gets split into must
 604  * be compatible with the logical_block_size of the target processing it.
 605  */
 606 static int validate_hardware_logical_block_alignment(struct dm_table *table,
 607                                                  struct queue_limits *limits)
 608 {
 609         /*
 610          * This function uses arithmetic modulo the logical_block_size
 611          * (in units of 512-byte sectors).
 612          */
 613         unsigned short device_logical_block_size_sects =
 614                 limits->logical_block_size >> SECTOR_SHIFT;
 615
 616         /*
 617          * Offset of the start of the next table entry, mod logical_block_size.
 618          */
 619         unsigned short next_target_start = 0;
 620
 621         /*
 622          * Given an aligned bio that extends beyond the end of a
 623          * target, how many sectors must the next target handle?
 624          */
 625         unsigned short remaining = 0;
 626
 627         struct dm_target *uninitialized_var(ti);
 628         struct queue_limits ti_limits;
 629         unsigned i = 0;
 630
 631         /*
 632          * Check each entry in the table in turn.
 633          */
 634         while (i < dm_table_get_num_targets(table)) {
 635                 ti = dm_table_get_target(table, i++);
 636
 637                 blk_set_stacking_limits(&ti_limits);
 638
 639                 /* combine all target devices' limits */
 640                 if (ti->type->iterate_devices)
 641                         ti->type->iterate_devices(ti, dm_set_device_limits,
 642                                                   &ti_limits);
 643
 644                 /*
 645                  * If the remaining sectors fall entirely within this
 646                  * table entry are they compatible with its logical_block_size?
 647                  */
 648                 if (remaining < ti->len &&
 649                     remaining & ((ti_limits.logical_block_size >>
 650                                   SECTOR_SHIFT) - 1))
 651                         break;  /* Error */
 652
 653                 next_target_start =
 654                     (unsigned short) ((next_target_start + ti->len) &
 655                                       (device_logical_block_size_sects - 1));
 656                 remaining = next_target_start ?
 657                     device_logical_block_size_sects - next_target_start : 0;
 658         }
 659
 660         if (remaining) {
 661                 DMWARN("%s: table line %u (start sect %llu len %llu) "
 662                        "not aligned to h/w logical block size %u",
 663                        dm_device_name(table->md), i,
 664                        (unsigned long long) ti->begin,
 665                        (unsigned long long) ti->len,
 666                        limits->logical_block_size);
 667                 return -EINVAL;
 668         }
 669
 670         return 0;
 671 }
 672
 673 int dm_table_add_target(struct dm_table *t, const char *type,
 674                         sector_t start, sector_t len, char *params)
 675 {
 676         int r = -EINVAL, argc;
 677         char **argv;
 678         struct dm_target *tgt;
 679
 680         if (t->singleton) {
 681                 DMERR("%s: target type %s must appear alone in table",
 682                       dm_device_name(t->md), t->targets->type->name);
 683                 return -EINVAL;
 684         }
 685
 686         BUG_ON(t->num_targets >= t->num_allocated);
 687
 688         tgt = t->targets + t->num_targets;
 689         memset(tgt, 0, sizeof(*tgt));
 690
 691         if (!len) {
 692                 DMERR("%s: zero-length target", dm_device_name(t->md));
 693                 return -EINVAL;
 694         }
 695
 696         tgt->type = dm_get_target_type(type);
 697         if (!tgt->type) {
 698                 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
 699                       type);
 700                 return -EINVAL;
 701         }
 702
 703         if (dm_target_needs_singleton(tgt->type)) {
 704                 if (t->num_targets) {
 705                         DMERR("%s: target type %s must appear alone in table",
 706                               dm_device_name(t->md), type);
 707                         return -EINVAL;
 708                 }
 709                 t->singleton = true;
 710         }
 711
 712         if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
 713                 DMERR("%s: target type %s may not be included in read-only tables",
 714                       dm_device_name(t->md), type);
 715                 return -EINVAL;
 716         }
 717
 718         if (t->immutable_target_type) {
 719                 if (t->immutable_target_type != tgt->type) {
 720                         DMERR("%s: immutable target type %s cannot be mixed with other target types",
 721                               dm_device_name(t->md), t->immutable_target_type->name);
 722                         return -EINVAL;
 723                 }
 724         } else if (dm_target_is_immutable(tgt->type)) {
 725                 if (t->num_targets) {
 726                         DMERR("%s: immutable target type %s cannot be mixed with other target types",
 727                               dm_device_name(t->md), tgt->type->name);
 728                         return -EINVAL;
 729                 }
 730                 t->immutable_target_type = tgt->type;
 731         }
 732
 733         tgt->table = t;
 734         tgt->begin = start;
 735         tgt->len = len;
 736         tgt->error = "Unknown error";
 737
 738         /*
 739          * Does this target adjoin the previous one ?
 740          */
 741         if (!adjoin(t, tgt)) {
 742                 tgt->error = "Gap in table";
 743                 r = -EINVAL;
 744                 goto bad;
 745         }
 746
 747         r = dm_split_args(&argc, &argv, params);
 748         if (r) {
 749                 tgt->error = "couldn't split parameters (insufficient memory)";
 750                 goto bad;
 751         }
 752
 753         r = tgt->type->ctr(tgt, argc, argv);
 754         kfree(argv);
 755         if (r)
 756                 goto bad;
 757
 758         t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
 759
 760         if (!tgt->num_discard_bios && tgt->discards_supported)
 761                 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
 762                        dm_device_name(t->md), type);
 763
 764         return 0;
 765
 766  bad:
 767         DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
 768         dm_put_target_type(tgt->type);
 769         return r;
 770 }
 771
 772 /*
 773  * Target argument parsing helpers.
 774  */
 775 static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
 776                              unsigned *value, char **error, unsigned grouped)
 777 {
 778         const char *arg_str = dm_shift_arg(arg_set);
 779         char dummy;
 780
 781         if (!arg_str ||
 782             (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
 783             (*value < arg->min) ||
 784             (*value > arg->max) ||
 785             (grouped && arg_set->argc < *value)) {
 786                 *error = arg->error;
 787                 return -EINVAL;
 788         }
 789
 790         return 0;
 791 }
 792
 793 int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
 794                 unsigned *value, char **error)
 795 {
 796         return validate_next_arg(arg, arg_set, value, error, 0);
 797 }
 798 EXPORT_SYMBOL(dm_read_arg);
 799
 800 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
 801                       unsigned *value, char **error)
 802 {
 803         return validate_next_arg(arg, arg_set, value, error, 1);
 804 }
 805 EXPORT_SYMBOL(dm_read_arg_group);
 806
 807 const char *dm_shift_arg(struct dm_arg_set *as)
 808 {
 809         char *r;
 810
 811         if (as->argc) {
 812                 as->argc--;
 813                 r = *as->argv;
 814                 as->argv++;
 815                 return r;
 816         }
 817
 818         return NULL;
 819 }
 820 EXPORT_SYMBOL(dm_shift_arg);
 821
 822 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
 823 {
 824         BUG_ON(as->argc < num_args);
 825         as->argc -= num_args;
 826         as->argv += num_args;
 827 }
 828 EXPORT_SYMBOL(dm_consume_args);
 829
 830 static bool __table_type_bio_based(unsigned table_type)
 831 {
 832         return (table_type == DM_TYPE_BIO_BASED ||
 833                 table_type == DM_TYPE_DAX_BIO_BASED);
 834 }
 835
 836 static bool __table_type_request_based(unsigned table_type)
 837 {
 838         return (table_type == DM_TYPE_REQUEST_BASED ||
 839                 table_type == DM_TYPE_MQ_REQUEST_BASED);
 840 }
 841
 842 void dm_table_set_type(struct dm_table *t, unsigned type)
 843 {
 844         t->type = type;
 845 }
 846 EXPORT_SYMBOL_GPL(dm_table_set_type);
 847
 848 static int device_supports_dax(struct dm_target *ti, struct dm_dev *dev,
 849                                sector_t start, sector_t len, void *data)
 850 {
 851         struct request_queue *q = bdev_get_queue(dev->bdev);
 852
 853         return q && blk_queue_dax(q);
 854 }
 855
 856 static bool dm_table_supports_dax(struct dm_table *t)
 857 {
 858         struct dm_target *ti;
 859         unsigned i = 0;
 860
 861         /* Ensure that all targets support DAX. */
 862         while (i < dm_table_get_num_targets(t)) {
 863                 ti = dm_table_get_target(t, i++);
 864
 865                 if (!ti->type->direct_access)
 866                         return false;
 867
 868                 if (!ti->type->iterate_devices ||
 869                     !ti->type->iterate_devices(ti, device_supports_dax, NULL))
 870                         return false;
 871         }
 872
 873         return true;
 874 }
 875
 876 static int dm_table_determine_type(struct dm_table *t)
 877 {
 878         unsigned i;
 879         unsigned bio_based = 0, request_based = 0, hybrid = 0;
 880         bool verify_blk_mq = false;
 881         struct dm_target *tgt;
 882         struct dm_dev_internal *dd;
 883         struct list_head *devices = dm_table_get_devices(t);
 884         unsigned live_md_type = dm_get_md_type(t->md);
 885
 886         if (t->type != DM_TYPE_NONE) {
 887                 /* target already set the table's type */
 888                 if (t->type == DM_TYPE_BIO_BASED)
 889                         return 0;
 890                 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
 891                 goto verify_rq_based;
 892         }
 893
 894         for (i = 0; i < t->num_targets; i++) {
 895                 tgt = t->targets + i;
 896                 if (dm_target_hybrid(tgt))
 897                         hybrid = 1;
 898                 else if (dm_target_request_based(tgt))
 899                         request_based = 1;
 900                 else
 901                         bio_based = 1;
 902
 903                 if (bio_based && request_based) {
 904                         DMWARN("Inconsistent table: different target types"
 905                                " can't be mixed up");
 906                         return -EINVAL;
 907                 }
 908         }
 909
 910         if (hybrid && !bio_based && !request_based) {
 911                 /*
 912                  * The targets can work either way.
 913                  * Determine the type from the live device.
 914                  * Default to bio-based if device is new.
 915                  */
 916                 if (__table_type_request_based(live_md_type))
 917                         request_based = 1;
 918                 else
 919                         bio_based = 1;
 920         }
 921
 922         if (bio_based) {
 923                 /* We must use this table as bio-based */
 924                 t->type = DM_TYPE_BIO_BASED;
 925                 if (dm_table_supports_dax(t) ||
 926                     (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED))
 927                         t->type = DM_TYPE_DAX_BIO_BASED;
 928                 return 0;
 929         }
 930
 931         BUG_ON(!request_based); /* No targets in this table */
 932
 933         if (list_empty(devices) && __table_type_request_based(live_md_type)) {
 934                 /* inherit live MD type */
 935                 t->type = live_md_type;
 936                 return 0;
 937         }
 938
 939         /*
 940          * The only way to establish DM_TYPE_MQ_REQUEST_BASED is by
 941          * having a compatible target use dm_table_set_type.
 942          */
 943         t->type = DM_TYPE_REQUEST_BASED;
 944
 945 verify_rq_based:
 946         /*
 947          * Request-based dm supports only tables that have a single target now.
 948          * To support multiple targets, request splitting support is needed,
 949          * and that needs lots of changes in the block-layer.
 950          * (e.g. request completion process for partial completion.)
 951          */
 952         if (t->num_targets > 1) {
 953                 DMWARN("Request-based dm doesn't support multiple targets yet");
 954                 return -EINVAL;
 955         }
 956
 957         /* Non-request-stackable devices can't be used for request-based dm */
 958         list_for_each_entry(dd, devices, list) {
 959                 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
 960
 961                 if (!blk_queue_stackable(q)) {
 962                         DMERR("table load rejected: including"
 963                               " non-request-stackable devices");
 964                         return -EINVAL;
 965                 }
 966
 967                 if (q->mq_ops)
 968                         verify_blk_mq = true;
 969         }
 970
 971         if (verify_blk_mq) {
 972                 /* verify _all_ devices in the table are blk-mq devices */
 973                 list_for_each_entry(dd, devices, list)
 974                         if (!bdev_get_queue(dd->dm_dev->bdev)->mq_ops) {
 975                                 DMERR("table load rejected: not all devices"
 976                                       " are blk-mq request-stackable");
 977                                 return -EINVAL;
 978                         }
 979
 980                 t->all_blk_mq = true;
 981         }
 982
 983         return 0;
 984 }
 985
 986 unsigned dm_table_get_type(struct dm_table *t)
 987 {
 988         return t->type;
 989 }
 990
 991 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
 992 {
 993         return t->immutable_target_type;
 994 }
 995
 996 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
 997 {
 998         /* Immutable target is implicitly a singleton */
 999         if (t->num_targets > 1 ||
1000             !dm_target_is_immutable(t->targets[0].type))
1001                 return NULL;
1002
1003         return t->targets;
1004 }
1005
1006 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1007 {
1008         struct dm_target *uninitialized_var(ti);
1009         unsigned i = 0;
1010
1011         while (i < dm_table_get_num_targets(t)) {
1012                 ti = dm_table_get_target(t, i++);
1013                 if (dm_target_is_wildcard(ti->type))
1014                         return ti;
1015         }
1016
1017         return NULL;
1018 }
1019
1020 bool dm_table_bio_based(struct dm_table *t)
1021 {
1022         return __table_type_bio_based(dm_table_get_type(t));
1023 }
1024
1025 bool dm_table_request_based(struct dm_table *t)
1026 {
1027         return __table_type_request_based(dm_table_get_type(t));
1028 }
1029
1030 bool dm_table_all_blk_mq_devices(struct dm_table *t)
1031 {
1032         return t->all_blk_mq;
1033 }
1034
1035 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1036 {
1037         unsigned type = dm_table_get_type(t);
1038         unsigned per_io_data_size = 0;
1039         struct dm_target *tgt;
1040         unsigned i;
1041
1042         if (unlikely(type == DM_TYPE_NONE)) {
1043                 DMWARN("no table type is set, can't allocate mempools");
1044                 return -EINVAL;
1045         }
1046
1047         if (__table_type_bio_based(type))
1048                 for (i = 0; i < t->num_targets; i++) {
1049                         tgt = t->targets + i;
1050                         per_io_data_size = max(per_io_data_size, tgt->per_io_data_size);
1051                 }
1052
1053         t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported, per_io_data_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
1134         list_for_each_entry(dd, devices, list) {
1135                 template_disk = dd->dm_dev->bdev->bd_disk;
1136                 if (!integrity_profile_exists(template_disk))
1137                         goto no_integrity;
1138                 else if (prev_disk &&
1139                          blk_integrity_compare(prev_disk, template_disk) < 0)
1140                         goto no_integrity;
1141                 prev_disk = template_disk;
1142         }
1143
1144         return template_disk;
1145
1146 no_integrity:
1147         if (prev_disk)
1148                 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1149                        dm_device_name(t->md),
1150                        prev_disk->disk_name,
1151                        template_disk->disk_name);
1152         return NULL;
1153 }
1154
1155 /*
1156  * Register the mapped device for blk_integrity support if the
1157  * underlying devices have an integrity profile.  But all devices may
1158  * not have matching profiles (checking all devices isn't reliable
1159  * during table load because this table may use other DM device(s) which
1160  * must be resumed before they will have an initialized integity
1161  * profile).  Consequently, stacked DM devices force a 2 stage integrity
1162  * profile validation: First pass during table load, final pass during
1163  * resume.
1164  */
1165 static int dm_table_register_integrity(struct dm_table *t)
1166 {
1167         struct mapped_device *md = t->md;
1168         struct gendisk *template_disk = NULL;
1169
1170         template_disk = dm_table_get_integrity_disk(t);
1171         if (!template_disk)
1172                 return 0;
1173
1174         if (!integrity_profile_exists(dm_disk(md))) {
1175                 t->integrity_supported = true;
1176                 /*
1177                  * Register integrity profile during table load; we can do
1178                  * this because the final profile must match during resume.
1179                  */
1180                 blk_integrity_register(dm_disk(md),
1181                                        blk_get_integrity(template_disk));
1182                 return 0;
1183         }
1184
1185         /*
1186          * If DM device already has an initialized integrity
1187          * profile the new profile should not conflict.
1188          */
1189         if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1190                 DMWARN("%s: conflict with existing integrity profile: "
1191                        "%s profile mismatch",
1192                        dm_device_name(t->md),
1193                        template_disk->disk_name);
1194                 return 1;
1195         }
1196
1197         /* Preserve existing integrity profile */
1198         t->integrity_supported = true;
1199         return 0;
1200 }
1201
1202 /*
1203  * Prepares the table for use by building the indices,
1204  * setting the type, and allocating mempools.
1205  */
1206 int dm_table_complete(struct dm_table *t)
1207 {
1208         int r;
1209
1210         r = dm_table_determine_type(t);
1211         if (r) {
1212                 DMERR("unable to determine table type");
1213                 return r;
1214         }
1215
1216         r = dm_table_build_index(t);
1217         if (r) {
1218                 DMERR("unable to build btrees");
1219                 return r;
1220         }
1221
1222         r = dm_table_register_integrity(t);
1223         if (r) {
1224                 DMERR("could not register integrity profile.");
1225                 return r;
1226         }
1227
1228         r = dm_table_alloc_md_mempools(t, t->md);
1229         if (r)
1230                 DMERR("unable to allocate mempools");
1231
1232         return r;
1233 }
1234
1235 static DEFINE_MUTEX(_event_lock);
1236 void dm_table_event_callback(struct dm_table *t,
1237                              void (*fn)(void *), void *context)
1238 {
1239         mutex_lock(&_event_lock);
1240         t->event_fn = fn;
1241         t->event_context = context;
1242         mutex_unlock(&_event_lock);
1243 }
1244
1245 void dm_table_event(struct dm_table *t)
1246 {
1247         /*
1248          * You can no longer call dm_table_event() from interrupt
1249          * context, use a bottom half instead.
1250          */
1251         BUG_ON(in_interrupt());
1252
1253         mutex_lock(&_event_lock);
1254         if (t->event_fn)
1255                 t->event_fn(t->event_context);
1256         mutex_unlock(&_event_lock);
1257 }
1258 EXPORT_SYMBOL(dm_table_event);
1259
1260 sector_t dm_table_get_size(struct dm_table *t)
1261 {
1262         return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1263 }
1264 EXPORT_SYMBOL(dm_table_get_size);
1265
1266 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1267 {
1268         if (index >= t->num_targets)
1269                 return NULL;
1270
1271         return t->targets + index;
1272 }
1273
1274 /*
1275  * Search the btree for the correct target.
1276  *
1277  * Caller should check returned pointer with dm_target_is_valid()
1278  * to trap I/O beyond end of device.
1279  */
1280 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1281 {
1282         unsigned int l, n = 0, k = 0;
1283         sector_t *node;
1284
1285         for (l = 0; l < t->depth; l++) {
1286                 n = get_child(n, k);
1287                 node = get_node(t, l, n);
1288
1289                 for (k = 0; k < KEYS_PER_NODE; k++)
1290                         if (node[k] >= sector)
1291                                 break;
1292         }
1293
1294         return &t->targets[(KEYS_PER_NODE * n) + k];
1295 }
1296
1297 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1298                         sector_t start, sector_t len, void *data)
1299 {
1300         unsigned *num_devices = data;
1301
1302         (*num_devices)++;
1303
1304         return 0;
1305 }
1306
1307 /*
1308  * Check whether a table has no data devices attached using each
1309  * target's iterate_devices method.
1310  * Returns false if the result is unknown because a target doesn't
1311  * support iterate_devices.
1312  */
1313 bool dm_table_has_no_data_devices(struct dm_table *table)
1314 {
1315         struct dm_target *uninitialized_var(ti);
1316         unsigned i = 0, num_devices = 0;
1317
1318         while (i < dm_table_get_num_targets(table)) {
1319                 ti = dm_table_get_target(table, i++);
1320
1321                 if (!ti->type->iterate_devices)
1322                         return false;
1323
1324                 ti->type->iterate_devices(ti, count_device, &num_devices);
1325                 if (num_devices)
1326                         return false;
1327         }
1328
1329         return true;
1330 }
1331
1332 /*
1333  * Establish the new table's queue_limits and validate them.
1334  */
1335 int dm_calculate_queue_limits(struct dm_table *table,
1336                               struct queue_limits *limits)
1337 {
1338         struct dm_target *uninitialized_var(ti);
1339         struct queue_limits ti_limits;
1340         unsigned i = 0;
1341
1342         blk_set_stacking_limits(limits);
1343
1344         while (i < dm_table_get_num_targets(table)) {
1345                 blk_set_stacking_limits(&ti_limits);
1346
1347                 ti = dm_table_get_target(table, i++);
1348
1349                 if (!ti->type->iterate_devices)
1350                         goto combine_limits;
1351
1352                 /*
1353                  * Combine queue limits of all the devices this target uses.
1354                  */
1355                 ti->type->iterate_devices(ti, dm_set_device_limits,
1356                                           &ti_limits);
1357
1358                 /* Set I/O hints portion of queue limits */
1359                 if (ti->type->io_hints)
1360                         ti->type->io_hints(ti, &ti_limits);
1361
1362                 /*
1363                  * Check each device area is consistent with the target's
1364                  * overall queue limits.
1365                  */
1366                 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1367                                               &ti_limits))
1368                         return -EINVAL;
1369
1370 combine_limits:
1371                 /*
1372                  * Merge this target's queue limits into the overall limits
1373                  * for the table.
1374                  */
1375                 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1376                         DMWARN("%s: adding target device "
1377                                "(start sect %llu len %llu) "
1378                                "caused an alignment inconsistency",
1379                                dm_device_name(table->md),
1380                                (unsigned long long) ti->begin,
1381                                (unsigned long long) ti->len);
1382         }
1383
1384         return validate_hardware_logical_block_alignment(table, limits);
1385 }
1386
1387 /*
1388  * Verify that all devices have an integrity profile that matches the
1389  * DM device's registered integrity profile.  If the profiles don't
1390  * match then unregister the DM device's integrity profile.
1391  */
1392 static void dm_table_verify_integrity(struct dm_table *t)
1393 {
1394         struct gendisk *template_disk = NULL;
1395
1396         if (t->integrity_supported) {
1397                 /*
1398                  * Verify that the original integrity profile
1399                  * matches all the devices in this table.
1400                  */
1401                 template_disk = dm_table_get_integrity_disk(t);
1402                 if (template_disk &&
1403                     blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1404                         return;
1405         }
1406
1407         if (integrity_profile_exists(dm_disk(t->md))) {
1408                 DMWARN("%s: unable to establish an integrity profile",
1409                        dm_device_name(t->md));
1410                 blk_integrity_unregister(dm_disk(t->md));
1411         }
1412 }
1413
1414 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1415                                 sector_t start, sector_t len, void *data)
1416 {
1417         unsigned long flush = (unsigned long) data;
1418         struct request_queue *q = bdev_get_queue(dev->bdev);
1419
1420         return q && (q->queue_flags & flush);
1421 }
1422
1423 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1424 {
1425         struct dm_target *ti;
1426         unsigned i = 0;
1427
1428         /*
1429          * Require at least one underlying device to support flushes.
1430          * t->devices includes internal dm devices such as mirror logs
1431          * so we need to use iterate_devices here, which targets
1432          * supporting flushes must provide.
1433          */
1434         while (i < dm_table_get_num_targets(t)) {
1435                 ti = dm_table_get_target(t, i++);
1436
1437                 if (!ti->num_flush_bios)
1438                         continue;
1439
1440                 if (ti->flush_supported)
1441                         return true;
1442
1443                 if (ti->type->iterate_devices &&
1444                     ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1445                         return true;
1446         }
1447
1448         return false;
1449 }
1450
1451 static bool dm_table_discard_zeroes_data(struct dm_table *t)
1452 {
1453         struct dm_target *ti;
1454         unsigned i = 0;
1455
1456         /* Ensure that all targets supports discard_zeroes_data. */
1457         while (i < dm_table_get_num_targets(t)) {
1458                 ti = dm_table_get_target(t, i++);
1459
1460                 if (ti->discard_zeroes_data_unsupported)
1461                         return false;
1462         }
1463
1464         return true;
1465 }
1466
1467 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1468                             sector_t start, sector_t len, void *data)
1469 {
1470         struct request_queue *q = bdev_get_queue(dev->bdev);
1471
1472         return q && blk_queue_nonrot(q);
1473 }
1474
1475 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1476                              sector_t start, sector_t len, void *data)
1477 {
1478         struct request_queue *q = bdev_get_queue(dev->bdev);
1479
1480         return q && !blk_queue_add_random(q);
1481 }
1482
1483 static int queue_supports_sg_merge(struct dm_target *ti, struct dm_dev *dev,
1484                                    sector_t start, sector_t len, void *data)
1485 {
1486         struct request_queue *q = bdev_get_queue(dev->bdev);
1487
1488         return q && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1489 }
1490
1491 static bool dm_table_all_devices_attribute(struct dm_table *t,
1492                                            iterate_devices_callout_fn func)
1493 {
1494         struct dm_target *ti;
1495         unsigned i = 0;
1496
1497         while (i < dm_table_get_num_targets(t)) {
1498                 ti = dm_table_get_target(t, i++);
1499
1500                 if (!ti->type->iterate_devices ||
1501                     !ti->type->iterate_devices(ti, func, NULL))
1502                         return false;
1503         }
1504
1505         return true;
1506 }
1507
1508 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1509                                          sector_t start, sector_t len, void *data)
1510 {
1511         struct request_queue *q = bdev_get_queue(dev->bdev);
1512
1513         return q && !q->limits.max_write_same_sectors;
1514 }
1515
1516 static bool dm_table_supports_write_same(struct dm_table *t)
1517 {
1518         struct dm_target *ti;
1519         unsigned i = 0;
1520
1521         while (i < dm_table_get_num_targets(t)) {
1522                 ti = dm_table_get_target(t, i++);
1523
1524                 if (!ti->num_write_same_bios)
1525                         return false;
1526
1527                 if (!ti->type->iterate_devices ||
1528                     ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1529                         return false;
1530         }
1531
1532         return true;
1533 }
1534
1535 static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1536                                   sector_t start, sector_t len, void *data)
1537 {
1538         struct request_queue *q = bdev_get_queue(dev->bdev);
1539
1540         return q && blk_queue_discard(q);
1541 }
1542
1543 static bool dm_table_supports_discards(struct dm_table *t)
1544 {
1545         struct dm_target *ti;
1546         unsigned i = 0;
1547
1548         /*
1549          * Unless any target used by the table set discards_supported,
1550          * require at least one underlying device to support discards.
1551          * t->devices includes internal dm devices such as mirror logs
1552          * so we need to use iterate_devices here, which targets
1553          * supporting discard selectively must provide.
1554          */
1555         while (i < dm_table_get_num_targets(t)) {
1556                 ti = dm_table_get_target(t, i++);
1557
1558                 if (!ti->num_discard_bios)
1559                         continue;
1560
1561                 if (ti->discards_supported)
1562                         return true;
1563
1564                 if (ti->type->iterate_devices &&
1565                     ti->type->iterate_devices(ti, device_discard_capable, NULL))
1566                         return true;
1567         }
1568
1569         return false;
1570 }
1571
1572 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1573                                struct queue_limits *limits)
1574 {
1575         bool wc = false, fua = false;
1576
1577         /*
1578          * Copy table's limits to the DM device's request_queue
1579          */
1580         q->limits = *limits;
1581
1582         if (!dm_table_supports_discards(t))
1583                 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1584         else
1585                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1586
1587         if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1588                 wc = true;
1589                 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1590                         fua = true;
1591         }
1592         blk_queue_write_cache(q, wc, fua);
1593
1594         if (!dm_table_discard_zeroes_data(t))
1595                 q->limits.discard_zeroes_data = 0;
1596
1597         /* Ensure that all underlying devices are non-rotational. */
1598         if (dm_table_all_devices_attribute(t, device_is_nonrot))
1599                 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1600         else
1601                 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1602
1603         if (!dm_table_supports_write_same(t))
1604                 q->limits.max_write_same_sectors = 0;
1605
1606         if (dm_table_all_devices_attribute(t, queue_supports_sg_merge))
1607                 queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1608         else
1609                 queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1610
1611         dm_table_verify_integrity(t);
1612
1613         /*
1614          * Determine whether or not this queue's I/O timings contribute
1615          * to the entropy pool, Only request-based targets use this.
1616          * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1617          * have it set.
1618          */
1619         if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1620                 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1621
1622         /*
1623          * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1624          * visible to other CPUs because, once the flag is set, incoming bios
1625          * are processed by request-based dm, which refers to the queue
1626          * settings.
1627          * Until the flag set, bios are passed to bio-based dm and queued to
1628          * md->deferred where queue settings are not needed yet.
1629          * Those bios are passed to request-based dm at the resume time.
1630          */
1631         smp_mb();
1632         if (dm_table_request_based(t))
1633                 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1634 }
1635
1636 unsigned int dm_table_get_num_targets(struct dm_table *t)
1637 {
1638         return t->num_targets;
1639 }
1640
1641 struct list_head *dm_table_get_devices(struct dm_table *t)
1642 {
1643         return &t->devices;
1644 }
1645
1646 fmode_t dm_table_get_mode(struct dm_table *t)
1647 {
1648         return t->mode;
1649 }
1650 EXPORT_SYMBOL(dm_table_get_mode);
1651
1652 enum suspend_mode {
1653         PRESUSPEND,
1654         PRESUSPEND_UNDO,
1655         POSTSUSPEND,
1656 };
1657
1658 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1659 {
1660         int i = t->num_targets;
1661         struct dm_target *ti = t->targets;
1662
1663         while (i--) {
1664                 switch (mode) {
1665                 case PRESUSPEND:
1666                         if (ti->type->presuspend)
1667                                 ti->type->presuspend(ti);
1668                         break;
1669                 case PRESUSPEND_UNDO:
1670                         if (ti->type->presuspend_undo)
1671                                 ti->type->presuspend_undo(ti);
1672                         break;
1673                 case POSTSUSPEND:
1674                         if (ti->type->postsuspend)
1675                                 ti->type->postsuspend(ti);
1676                         break;
1677                 }
1678                 ti++;
1679         }
1680 }
1681
1682 void dm_table_presuspend_targets(struct dm_table *t)
1683 {
1684         if (!t)
1685                 return;
1686
1687         suspend_targets(t, PRESUSPEND);
1688 }
1689
1690 void dm_table_presuspend_undo_targets(struct dm_table *t)
1691 {
1692         if (!t)
1693                 return;
1694
1695         suspend_targets(t, PRESUSPEND_UNDO);
1696 }
1697
1698 void dm_table_postsuspend_targets(struct dm_table *t)
1699 {
1700         if (!t)
1701                 return;
1702
1703         suspend_targets(t, POSTSUSPEND);
1704 }
1705
1706 int dm_table_resume_targets(struct dm_table *t)
1707 {
1708         int i, r = 0;
1709
1710         for (i = 0; i < t->num_targets; i++) {
1711                 struct dm_target *ti = t->targets + i;
1712
1713                 if (!ti->type->preresume)
1714                         continue;
1715
1716                 r = ti->type->preresume(ti);
1717                 if (r) {
1718                         DMERR("%s: %s: preresume failed, error = %d",
1719                               dm_device_name(t->md), ti->type->name, r);
1720                         return r;
1721                 }
1722         }
1723
1724         for (i = 0; i < t->num_targets; i++) {
1725                 struct dm_target *ti = t->targets + i;
1726
1727                 if (ti->type->resume)
1728                         ti->type->resume(ti);
1729         }
1730
1731         return 0;
1732 }
1733
1734 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1735 {
1736         list_add(&cb->list, &t->target_callbacks);
1737 }
1738 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1739
1740 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1741 {
1742         struct dm_dev_internal *dd;
1743         struct list_head *devices = dm_table_get_devices(t);
1744         struct dm_target_callbacks *cb;
1745         int r = 0;
1746
1747         list_for_each_entry(dd, devices, list) {
1748                 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1749                 char b[BDEVNAME_SIZE];
1750
1751                 if (likely(q))
1752                         r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1753                 else
1754                         DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1755                                      dm_device_name(t->md),
1756                                      bdevname(dd->dm_dev->bdev, b));
1757         }
1758
1759         list_for_each_entry(cb, &t->target_callbacks, list)
1760                 if (cb->congested_fn)
1761                         r |= cb->congested_fn(cb, bdi_bits);
1762
1763         return r;
1764 }
1765
1766 struct mapped_device *dm_table_get_md(struct dm_table *t)
1767 {
1768         return t->md;
1769 }
1770 EXPORT_SYMBOL(dm_table_get_md);
1771
1772 void dm_table_run_md_queue_async(struct dm_table *t)
1773 {
1774         struct mapped_device *md;
1775         struct request_queue *queue;
1776         unsigned long flags;
1777
1778         if (!dm_table_request_based(t))
1779                 return;
1780
1781         md = dm_table_get_md(t);
1782         queue = dm_get_md_queue(md);
1783         if (queue) {
1784                 if (queue->mq_ops)
1785                         blk_mq_run_hw_queues(queue, true);
1786                 else {
1787                         spin_lock_irqsave(queue->queue_lock, flags);
1788                         blk_run_queue_async(queue);
1789                         spin_unlock_irqrestore(queue->queue_lock, flags);
1790                 }
1791         }
1792 }
1793 EXPORT_SYMBOL(dm_table_run_md_queue_async);
1794