drivers/md/dm-table.c

   1 /*
   2  * Copyright (C) 2001 Sistina Software (UK) Limited.
   3  * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
   4  *
   5  * This file is released under the GPL.
   6  */
   7
   8 #include "dm-core.h"
   9
  10 #include <linux/module.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/blkdev.h>
  13 #include <linux/namei.h>
  14 #include <linux/ctype.h>
  15 #include <linux/string.h>
  16 #include <linux/slab.h>
  17 #include <linux/interrupt.h>
  18 #include <linux/mutex.h>
  19 #include <linux/delay.h>
  20 #include <linux/atomic.h>
  21 #include <linux/blk-mq.h>
  22 #include <linux/mount.h>
  23 #include <linux/dax.h>
  24
  25 #define DM_MSG_PREFIX "table"
  26
  27 #define MAX_DEPTH 16
  28 #define NODE_SIZE L1_CACHE_BYTES
  29 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
  30 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
  31
  32 struct dm_table {
  33         struct mapped_device *md;
  34         enum dm_queue_mode type;
  35
  36         /* btree table */
  37         unsigned int depth;
  38         unsigned int counts[MAX_DEPTH]; /* in nodes */
  39         sector_t *index[MAX_DEPTH];
  40
  41         unsigned int num_targets;
  42         unsigned int num_allocated;
  43         sector_t *highs;
  44         struct dm_target *targets;
  45
  46         struct target_type *immutable_target_type;
  47
  48         bool integrity_supported:1;
  49         bool singleton: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          */
 167         n_highs = (sector_t *) dm_vcalloc(num, sizeof(struct dm_target) +
 168                                           sizeof(sector_t));
 169         if (!n_highs)
 170                 return -ENOMEM;
 171
 172         n_targets = (struct dm_target *) (n_highs + num);
 173
 174         memset(n_highs, -1, sizeof(*n_highs) * num);
 175         vfree(t->highs);
 176
 177         t->num_allocated = num;
 178         t->highs = n_highs;
 179         t->targets = n_targets;
 180
 181         return 0;
 182 }
 183
 184 int dm_table_create(struct dm_table **result, fmode_t mode,
 185                     unsigned num_targets, struct mapped_device *md)
 186 {
 187         struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
 188
 189         if (!t)
 190                 return -ENOMEM;
 191
 192         INIT_LIST_HEAD(&t->devices);
 193         INIT_LIST_HEAD(&t->target_callbacks);
 194
 195         if (!num_targets)
 196                 num_targets = KEYS_PER_NODE;
 197
 198         num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
 199
 200         if (!num_targets) {
 201                 kfree(t);
 202                 return -ENOMEM;
 203         }
 204
 205         if (alloc_targets(t, num_targets)) {
 206                 kfree(t);
 207                 return -ENOMEM;
 208         }
 209
 210         t->type = DM_TYPE_NONE;
 211         t->mode = mode;
 212         t->md = md;
 213         *result = t;
 214         return 0;
 215 }
 216
 217 static void free_devices(struct list_head *devices, struct mapped_device *md)
 218 {
 219         struct list_head *tmp, *next;
 220
 221         list_for_each_safe(tmp, next, devices) {
 222                 struct dm_dev_internal *dd =
 223                     list_entry(tmp, struct dm_dev_internal, list);
 224                 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
 225                        dm_device_name(md), dd->dm_dev->name);
 226                 dm_put_table_device(md, dd->dm_dev);
 227                 kfree(dd);
 228         }
 229 }
 230
 231 void dm_table_destroy(struct dm_table *t)
 232 {
 233         unsigned int i;
 234
 235         if (!t)
 236                 return;
 237
 238         /* free the indexes */
 239         if (t->depth >= 2)
 240                 vfree(t->index[t->depth - 2]);
 241
 242         /* free the targets */
 243         for (i = 0; i < t->num_targets; i++) {
 244                 struct dm_target *tgt = t->targets + i;
 245
 246                 if (tgt->type->dtr)
 247                         tgt->type->dtr(tgt);
 248
 249                 dm_put_target_type(tgt->type);
 250         }
 251
 252         vfree(t->highs);
 253
 254         /* free the device list */
 255         free_devices(&t->devices, t->md);
 256
 257         dm_free_md_mempools(t->mempools);
 258
 259         kfree(t);
 260 }
 261
 262 /*
 263  * See if we've already got a device in the list.
 264  */
 265 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
 266 {
 267         struct dm_dev_internal *dd;
 268
 269         list_for_each_entry (dd, l, list)
 270                 if (dd->dm_dev->bdev->bd_dev == dev)
 271                         return dd;
 272
 273         return NULL;
 274 }
 275
 276 /*
 277  * If possible, this checks an area of a destination device is invalid.
 278  */
 279 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
 280                                   sector_t start, sector_t len, void *data)
 281 {
 282         struct request_queue *q;
 283         struct queue_limits *limits = data;
 284         struct block_device *bdev = dev->bdev;
 285         sector_t dev_size =
 286                 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
 287         unsigned short logical_block_size_sectors =
 288                 limits->logical_block_size >> SECTOR_SHIFT;
 289         char b[BDEVNAME_SIZE];
 290
 291         /*
 292          * Some devices exist without request functions,
 293          * such as loop devices not yet bound to backing files.
 294          * Forbid the use of such devices.
 295          */
 296         q = bdev_get_queue(bdev);
 297         if (!q || !q->make_request_fn) {
 298                 DMWARN("%s: %s is not yet initialised: "
 299                        "start=%llu, len=%llu, dev_size=%llu",
 300                        dm_device_name(ti->table->md), bdevname(bdev, b),
 301                        (unsigned long long)start,
 302                        (unsigned long long)len,
 303                        (unsigned long long)dev_size);
 304                 return 1;
 305         }
 306
 307         if (!dev_size)
 308                 return 0;
 309
 310         if ((start >= dev_size) || (start + len > dev_size)) {
 311                 DMWARN("%s: %s too small for target: "
 312                        "start=%llu, len=%llu, dev_size=%llu",
 313                        dm_device_name(ti->table->md), bdevname(bdev, b),
 314                        (unsigned long long)start,
 315                        (unsigned long long)len,
 316                        (unsigned long long)dev_size);
 317                 return 1;
 318         }
 319
 320         /*
 321          * If the target is mapped to zoned block device(s), check
 322          * that the zones are not partially mapped.
 323          */
 324         if (bdev_zoned_model(bdev) != BLK_ZONED_NONE) {
 325                 unsigned int zone_sectors = bdev_zone_sectors(bdev);
 326
 327                 if (start & (zone_sectors - 1)) {
 328                         DMWARN("%s: start=%llu not aligned to h/w zone size %u of %s",
 329                                dm_device_name(ti->table->md),
 330                                (unsigned long long)start,
 331                                zone_sectors, bdevname(bdev, b));
 332                         return 1;
 333                 }
 334
 335                 /*
 336                  * Note: The last zone of a zoned block device may be smaller
 337                  * than other zones. So for a target mapping the end of a
 338                  * zoned block device with such a zone, len would not be zone
 339                  * aligned. We do not allow such last smaller zone to be part
 340                  * of the mapping here to ensure that mappings with multiple
 341                  * devices do not end up with a smaller zone in the middle of
 342                  * the sector range.
 343                  */
 344                 if (len & (zone_sectors - 1)) {
 345                         DMWARN("%s: len=%llu not aligned to h/w zone size %u of %s",
 346                                dm_device_name(ti->table->md),
 347                                (unsigned long long)len,
 348                                zone_sectors, bdevname(bdev, b));
 349                         return 1;
 350                 }
 351         }
 352
 353         if (logical_block_size_sectors <= 1)
 354                 return 0;
 355
 356         if (start & (logical_block_size_sectors - 1)) {
 357                 DMWARN("%s: start=%llu not aligned to h/w "
 358                        "logical block size %u of %s",
 359                        dm_device_name(ti->table->md),
 360                        (unsigned long long)start,
 361                        limits->logical_block_size, bdevname(bdev, b));
 362                 return 1;
 363         }
 364
 365         if (len & (logical_block_size_sectors - 1)) {
 366                 DMWARN("%s: len=%llu not aligned to h/w "
 367                        "logical block size %u of %s",
 368                        dm_device_name(ti->table->md),
 369                        (unsigned long long)len,
 370                        limits->logical_block_size, bdevname(bdev, b));
 371                 return 1;
 372         }
 373
 374         return 0;
 375 }
 376
 377 /*
 378  * This upgrades the mode on an already open dm_dev, being
 379  * careful to leave things as they were if we fail to reopen the
 380  * device and not to touch the existing bdev field in case
 381  * it is accessed concurrently inside dm_table_any_congested().
 382  */
 383 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
 384                         struct mapped_device *md)
 385 {
 386         int r;
 387         struct dm_dev *old_dev, *new_dev;
 388
 389         old_dev = dd->dm_dev;
 390
 391         r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
 392                                 dd->dm_dev->mode | new_mode, &new_dev);
 393         if (r)
 394                 return r;
 395
 396         dd->dm_dev = new_dev;
 397         dm_put_table_device(md, old_dev);
 398
 399         return 0;
 400 }
 401
 402 /*
 403  * Convert the path to a device
 404  */
 405 dev_t dm_get_dev_t(const char *path)
 406 {
 407         dev_t dev;
 408         struct block_device *bdev;
 409
 410         bdev = lookup_bdev(path);
 411         if (IS_ERR(bdev))
 412                 dev = name_to_dev_t(path);
 413         else {
 414                 dev = bdev->bd_dev;
 415                 bdput(bdev);
 416         }
 417
 418         return dev;
 419 }
 420 EXPORT_SYMBOL_GPL(dm_get_dev_t);
 421
 422 /*
 423  * Add a device to the list, or just increment the usage count if
 424  * it's already present.
 425  */
 426 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
 427                   struct dm_dev **result)
 428 {
 429         int r;
 430         dev_t dev;
 431         struct dm_dev_internal *dd;
 432         struct dm_table *t = ti->table;
 433
 434         BUG_ON(!t);
 435
 436         dev = dm_get_dev_t(path);
 437         if (!dev)
 438                 return -ENODEV;
 439
 440         dd = find_device(&t->devices, dev);
 441         if (!dd) {
 442                 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
 443                 if (!dd)
 444                         return -ENOMEM;
 445
 446                 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
 447                         kfree(dd);
 448                         return r;
 449                 }
 450
 451                 refcount_set(&dd->count, 1);
 452                 list_add(&dd->list, &t->devices);
 453                 goto out;
 454
 455         } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
 456                 r = upgrade_mode(dd, mode, t->md);
 457                 if (r)
 458                         return r;
 459         }
 460         refcount_inc(&dd->count);
 461 out:
 462         *result = dd->dm_dev;
 463         return 0;
 464 }
 465 EXPORT_SYMBOL(dm_get_device);
 466
 467 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
 468                                 sector_t start, sector_t len, void *data)
 469 {
 470         struct queue_limits *limits = data;
 471         struct block_device *bdev = dev->bdev;
 472         struct request_queue *q = bdev_get_queue(bdev);
 473         char b[BDEVNAME_SIZE];
 474
 475         if (unlikely(!q)) {
 476                 DMWARN("%s: Cannot set limits for nonexistent device %s",
 477                        dm_device_name(ti->table->md), bdevname(bdev, b));
 478                 return 0;
 479         }
 480
 481         if (bdev_stack_limits(limits, bdev, start) < 0)
 482                 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
 483                        "physical_block_size=%u, logical_block_size=%u, "
 484                        "alignment_offset=%u, start=%llu",
 485                        dm_device_name(ti->table->md), bdevname(bdev, b),
 486                        q->limits.physical_block_size,
 487                        q->limits.logical_block_size,
 488                        q->limits.alignment_offset,
 489                        (unsigned long long) start << SECTOR_SHIFT);
 490
 491         limits->zoned = blk_queue_zoned_model(q);
 492
 493         return 0;
 494 }
 495
 496 /*
 497  * Decrement a device's use count and remove it if necessary.
 498  */
 499 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
 500 {
 501         int found = 0;
 502         struct list_head *devices = &ti->table->devices;
 503         struct dm_dev_internal *dd;
 504
 505         list_for_each_entry(dd, devices, list) {
 506                 if (dd->dm_dev == d) {
 507                         found = 1;
 508                         break;
 509                 }
 510         }
 511         if (!found) {
 512                 DMWARN("%s: device %s not in table devices list",
 513                        dm_device_name(ti->table->md), d->name);
 514                 return;
 515         }
 516         if (refcount_dec_and_test(&dd->count)) {
 517                 dm_put_table_device(ti->table->md, d);
 518                 list_del(&dd->list);
 519                 kfree(dd);
 520         }
 521 }
 522 EXPORT_SYMBOL(dm_put_device);
 523
 524 /*
 525  * Checks to see if the target joins onto the end of the table.
 526  */
 527 static int adjoin(struct dm_table *table, struct dm_target *ti)
 528 {
 529         struct dm_target *prev;
 530
 531         if (!table->num_targets)
 532                 return !ti->begin;
 533
 534         prev = &table->targets[table->num_targets - 1];
 535         return (ti->begin == (prev->begin + prev->len));
 536 }
 537
 538 /*
 539  * Used to dynamically allocate the arg array.
 540  *
 541  * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
 542  * process messages even if some device is suspended. These messages have a
 543  * small fixed number of arguments.
 544  *
 545  * On the other hand, dm-switch needs to process bulk data using messages and
 546  * excessive use of GFP_NOIO could cause trouble.
 547  */
 548 static char **realloc_argv(unsigned *size, char **old_argv)
 549 {
 550         char **argv;
 551         unsigned new_size;
 552         gfp_t gfp;
 553
 554         if (*size) {
 555                 new_size = *size * 2;
 556                 gfp = GFP_KERNEL;
 557         } else {
 558                 new_size = 8;
 559                 gfp = GFP_NOIO;
 560         }
 561         argv = kmalloc_array(new_size, sizeof(*argv), gfp);
 562         if (argv && old_argv) {
 563                 memcpy(argv, old_argv, *size * sizeof(*argv));
 564                 *size = new_size;
 565         }
 566
 567         kfree(old_argv);
 568         return argv;
 569 }
 570
 571 /*
 572  * Destructively splits up the argument list to pass to ctr.
 573  */
 574 int dm_split_args(int *argc, char ***argvp, char *input)
 575 {
 576         char *start, *end = input, *out, **argv = NULL;
 577         unsigned array_size = 0;
 578
 579         *argc = 0;
 580
 581         if (!input) {
 582                 *argvp = NULL;
 583                 return 0;
 584         }
 585
 586         argv = realloc_argv(&array_size, argv);
 587         if (!argv)
 588                 return -ENOMEM;
 589
 590         while (1) {
 591                 /* Skip whitespace */
 592                 start = skip_spaces(end);
 593
 594                 if (!*start)
 595                         break;  /* success, we hit the end */
 596
 597                 /* 'out' is used to remove any back-quotes */
 598                 end = out = start;
 599                 while (*end) {
 600                         /* Everything apart from '\0' can be quoted */
 601                         if (*end == '\\' && *(end + 1)) {
 602                                 *out++ = *(end + 1);
 603                                 end += 2;
 604                                 continue;
 605                         }
 606
 607                         if (isspace(*end))
 608                                 break;  /* end of token */
 609
 610                         *out++ = *end++;
 611                 }
 612
 613                 /* have we already filled the array ? */
 614                 if ((*argc + 1) > array_size) {
 615                         argv = realloc_argv(&array_size, argv);
 616                         if (!argv)
 617                                 return -ENOMEM;
 618                 }
 619
 620                 /* we know this is whitespace */
 621                 if (*end)
 622                         end++;
 623
 624                 /* terminate the string and put it in the array */
 625                 *out = '\0';
 626                 argv[*argc] = start;
 627                 (*argc)++;
 628         }
 629
 630         *argvp = argv;
 631         return 0;
 632 }
 633
 634 /*
 635  * Impose necessary and sufficient conditions on a devices's table such
 636  * that any incoming bio which respects its logical_block_size can be
 637  * processed successfully.  If it falls across the boundary between
 638  * two or more targets, the size of each piece it gets split into must
 639  * be compatible with the logical_block_size of the target processing it.
 640  */
 641 static int validate_hardware_logical_block_alignment(struct dm_table *table,
 642                                                  struct queue_limits *limits)
 643 {
 644         /*
 645          * This function uses arithmetic modulo the logical_block_size
 646          * (in units of 512-byte sectors).
 647          */
 648         unsigned short device_logical_block_size_sects =
 649                 limits->logical_block_size >> SECTOR_SHIFT;
 650
 651         /*
 652          * Offset of the start of the next table entry, mod logical_block_size.
 653          */
 654         unsigned short next_target_start = 0;
 655
 656         /*
 657          * Given an aligned bio that extends beyond the end of a
 658          * target, how many sectors must the next target handle?
 659          */
 660         unsigned short remaining = 0;
 661
 662         struct dm_target *uninitialized_var(ti);
 663         struct queue_limits ti_limits;
 664         unsigned i;
 665
 666         /*
 667          * Check each entry in the table in turn.
 668          */
 669         for (i = 0; i < dm_table_get_num_targets(table); i++) {
 670                 ti = dm_table_get_target(table, i);
 671
 672                 blk_set_stacking_limits(&ti_limits);
 673
 674                 /* combine all target devices' limits */
 675                 if (ti->type->iterate_devices)
 676                         ti->type->iterate_devices(ti, dm_set_device_limits,
 677                                                   &ti_limits);
 678
 679                 /*
 680                  * If the remaining sectors fall entirely within this
 681                  * table entry are they compatible with its logical_block_size?
 682                  */
 683                 if (remaining < ti->len &&
 684                     remaining & ((ti_limits.logical_block_size >>
 685                                   SECTOR_SHIFT) - 1))
 686                         break;  /* Error */
 687
 688                 next_target_start =
 689                     (unsigned short) ((next_target_start + ti->len) &
 690                                       (device_logical_block_size_sects - 1));
 691                 remaining = next_target_start ?
 692                     device_logical_block_size_sects - next_target_start : 0;
 693         }
 694
 695         if (remaining) {
 696                 DMWARN("%s: table line %u (start sect %llu len %llu) "
 697                        "not aligned to h/w logical block size %u",
 698                        dm_device_name(table->md), i,
 699                        (unsigned long long) ti->begin,
 700                        (unsigned long long) ti->len,
 701                        limits->logical_block_size);
 702                 return -EINVAL;
 703         }
 704
 705         return 0;
 706 }
 707
 708 int dm_table_add_target(struct dm_table *t, const char *type,
 709                         sector_t start, sector_t len, char *params)
 710 {
 711         int r = -EINVAL, argc;
 712         char **argv;
 713         struct dm_target *tgt;
 714
 715         if (t->singleton) {
 716                 DMERR("%s: target type %s must appear alone in table",
 717                       dm_device_name(t->md), t->targets->type->name);
 718                 return -EINVAL;
 719         }
 720
 721         BUG_ON(t->num_targets >= t->num_allocated);
 722
 723         tgt = t->targets + t->num_targets;
 724         memset(tgt, 0, sizeof(*tgt));
 725
 726         if (!len) {
 727                 DMERR("%s: zero-length target", dm_device_name(t->md));
 728                 return -EINVAL;
 729         }
 730
 731         tgt->type = dm_get_target_type(type);
 732         if (!tgt->type) {
 733                 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
 734                 return -EINVAL;
 735         }
 736
 737         if (dm_target_needs_singleton(tgt->type)) {
 738                 if (t->num_targets) {
 739                         tgt->error = "singleton target type must appear alone in table";
 740                         goto bad;
 741                 }
 742                 t->singleton = true;
 743         }
 744
 745         if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
 746                 tgt->error = "target type may not be included in a read-only table";
 747                 goto bad;
 748         }
 749
 750         if (t->immutable_target_type) {
 751                 if (t->immutable_target_type != tgt->type) {
 752                         tgt->error = "immutable target type cannot be mixed with other target types";
 753                         goto bad;
 754                 }
 755         } else if (dm_target_is_immutable(tgt->type)) {
 756                 if (t->num_targets) {
 757                         tgt->error = "immutable target type cannot be mixed with other target types";
 758                         goto bad;
 759                 }
 760                 t->immutable_target_type = tgt->type;
 761         }
 762
 763         if (dm_target_has_integrity(tgt->type))
 764                 t->integrity_added = 1;
 765
 766         tgt->table = t;
 767         tgt->begin = start;
 768         tgt->len = len;
 769         tgt->error = "Unknown error";
 770
 771         /*
 772          * Does this target adjoin the previous one ?
 773          */
 774         if (!adjoin(t, tgt)) {
 775                 tgt->error = "Gap in table";
 776                 goto bad;
 777         }
 778
 779         r = dm_split_args(&argc, &argv, params);
 780         if (r) {
 781                 tgt->error = "couldn't split parameters (insufficient memory)";
 782                 goto bad;
 783         }
 784
 785         r = tgt->type->ctr(tgt, argc, argv);
 786         kfree(argv);
 787         if (r)
 788                 goto bad;
 789
 790         t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
 791
 792         if (!tgt->num_discard_bios && tgt->discards_supported)
 793                 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
 794                        dm_device_name(t->md), type);
 795
 796         return 0;
 797
 798  bad:
 799         DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
 800         dm_put_target_type(tgt->type);
 801         return r;
 802 }
 803
 804 /*
 805  * Target argument parsing helpers.
 806  */
 807 static int validate_next_arg(const struct dm_arg *arg,
 808                              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(const 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(const 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                 table_type == DM_TYPE_NVME_BIO_BASED);
 868 }
 869
 870 static bool __table_type_request_based(enum dm_queue_mode table_type)
 871 {
 872         return table_type == DM_TYPE_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 /* validate the dax capability of the target device span */
 882 int device_supports_dax(struct dm_target *ti, struct dm_dev *dev,
 883                         sector_t start, sector_t len, void *data)
 884 {
 885         int blocksize = *(int *) data;
 886
 887         return generic_fsdax_supported(dev->dax_dev, dev->bdev, blocksize,
 888                                        start, len);
 889 }
 890
 891 /* Check devices support synchronous DAX */
 892 static int device_dax_synchronous(struct dm_target *ti, struct dm_dev *dev,
 893                                   sector_t start, sector_t len, void *data)
 894 {
 895         return dev->dax_dev && dax_synchronous(dev->dax_dev);
 896 }
 897
 898 bool dm_table_supports_dax(struct dm_table *t,
 899                            iterate_devices_callout_fn iterate_fn, int *blocksize)
 900 {
 901         struct dm_target *ti;
 902         unsigned i;
 903
 904         /* Ensure that all targets support DAX. */
 905         for (i = 0; i < dm_table_get_num_targets(t); i++) {
 906                 ti = dm_table_get_target(t, i);
 907
 908                 if (!ti->type->direct_access)
 909                         return false;
 910
 911                 if (!ti->type->iterate_devices ||
 912                     !ti->type->iterate_devices(ti, iterate_fn, blocksize))
 913                         return false;
 914         }
 915
 916         return true;
 917 }
 918
 919 static bool dm_table_does_not_support_partial_completion(struct dm_table *t);
 920
 921 static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev,
 922                                   sector_t start, sector_t len, void *data)
 923 {
 924         struct block_device *bdev = dev->bdev;
 925         struct request_queue *q = bdev_get_queue(bdev);
 926
 927         /* request-based cannot stack on partitions! */
 928         if (bdev != bdev->bd_contains)
 929                 return false;
 930
 931         return queue_is_mq(q);
 932 }
 933
 934 static int dm_table_determine_type(struct dm_table *t)
 935 {
 936         unsigned i;
 937         unsigned bio_based = 0, request_based = 0, hybrid = 0;
 938         struct dm_target *tgt;
 939         struct list_head *devices = dm_table_get_devices(t);
 940         enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
 941         int page_size = PAGE_SIZE;
 942
 943         if (t->type != DM_TYPE_NONE) {
 944                 /* target already set the table's type */
 945                 if (t->type == DM_TYPE_BIO_BASED) {
 946                         /* possibly upgrade to a variant of bio-based */
 947                         goto verify_bio_based;
 948                 }
 949                 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
 950                 BUG_ON(t->type == DM_TYPE_NVME_BIO_BASED);
 951                 goto verify_rq_based;
 952         }
 953
 954         for (i = 0; i < t->num_targets; i++) {
 955                 tgt = t->targets + i;
 956                 if (dm_target_hybrid(tgt))
 957                         hybrid = 1;
 958                 else if (dm_target_request_based(tgt))
 959                         request_based = 1;
 960                 else
 961                         bio_based = 1;
 962
 963                 if (bio_based && request_based) {
 964                         DMERR("Inconsistent table: different target types"
 965                               " can't be mixed up");
 966                         return -EINVAL;
 967                 }
 968         }
 969
 970         if (hybrid && !bio_based && !request_based) {
 971                 /*
 972                  * The targets can work either way.
 973                  * Determine the type from the live device.
 974                  * Default to bio-based if device is new.
 975                  */
 976                 if (__table_type_request_based(live_md_type))
 977                         request_based = 1;
 978                 else
 979                         bio_based = 1;
 980         }
 981
 982         if (bio_based) {
 983 verify_bio_based:
 984                 /* We must use this table as bio-based */
 985                 t->type = DM_TYPE_BIO_BASED;
 986                 if (dm_table_supports_dax(t, device_supports_dax, &page_size) ||
 987                     (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
 988                         t->type = DM_TYPE_DAX_BIO_BASED;
 989                 } else {
 990                         /* Check if upgrading to NVMe bio-based is valid or required */
 991                         tgt = dm_table_get_immutable_target(t);
 992                         if (tgt && !tgt->max_io_len && dm_table_does_not_support_partial_completion(t)) {
 993                                 t->type = DM_TYPE_NVME_BIO_BASED;
 994                                 goto verify_rq_based; /* must be stacked directly on NVMe (blk-mq) */
 995                         } else if (list_empty(devices) && live_md_type == DM_TYPE_NVME_BIO_BASED) {
 996                                 t->type = DM_TYPE_NVME_BIO_BASED;
 997                         }
 998                 }
 999                 return 0;
1000         }
1001
1002         BUG_ON(!request_based); /* No targets in this table */
1003
1004         t->type = DM_TYPE_REQUEST_BASED;
1005
1006 verify_rq_based:
1007         /*
1008          * Request-based dm supports only tables that have a single target now.
1009          * To support multiple targets, request splitting support is needed,
1010          * and that needs lots of changes in the block-layer.
1011          * (e.g. request completion process for partial completion.)
1012          */
1013         if (t->num_targets > 1) {
1014                 DMERR("%s DM doesn't support multiple targets",
1015                       t->type == DM_TYPE_NVME_BIO_BASED ? "nvme bio-based" : "request-based");
1016                 return -EINVAL;
1017         }
1018
1019         if (list_empty(devices)) {
1020                 int srcu_idx;
1021                 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
1022
1023                 /* inherit live table's type */
1024                 if (live_table)
1025                         t->type = live_table->type;
1026                 dm_put_live_table(t->md, srcu_idx);
1027                 return 0;
1028         }
1029
1030         tgt = dm_table_get_immutable_target(t);
1031         if (!tgt) {
1032                 DMERR("table load rejected: immutable target is required");
1033                 return -EINVAL;
1034         } else if (tgt->max_io_len) {
1035                 DMERR("table load rejected: immutable target that splits IO is not supported");
1036                 return -EINVAL;
1037         }
1038
1039         /* Non-request-stackable devices can't be used for request-based dm */
1040         if (!tgt->type->iterate_devices ||
1041             !tgt->type->iterate_devices(tgt, device_is_rq_stackable, NULL)) {
1042                 DMERR("table load rejected: including non-request-stackable devices");
1043                 return -EINVAL;
1044         }
1045
1046         return 0;
1047 }
1048
1049 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
1050 {
1051         return t->type;
1052 }
1053
1054 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
1055 {
1056         return t->immutable_target_type;
1057 }
1058
1059 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
1060 {
1061         /* Immutable target is implicitly a singleton */
1062         if (t->num_targets > 1 ||
1063             !dm_target_is_immutable(t->targets[0].type))
1064                 return NULL;
1065
1066         return t->targets;
1067 }
1068
1069 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1070 {
1071         struct dm_target *ti;
1072         unsigned i;
1073
1074         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1075                 ti = dm_table_get_target(t, i);
1076                 if (dm_target_is_wildcard(ti->type))
1077                         return ti;
1078         }
1079
1080         return NULL;
1081 }
1082
1083 bool dm_table_bio_based(struct dm_table *t)
1084 {
1085         return __table_type_bio_based(dm_table_get_type(t));
1086 }
1087
1088 bool dm_table_request_based(struct dm_table *t)
1089 {
1090         return __table_type_request_based(dm_table_get_type(t));
1091 }
1092
1093 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1094 {
1095         enum dm_queue_mode type = dm_table_get_type(t);
1096         unsigned per_io_data_size = 0;
1097         unsigned min_pool_size = 0;
1098         struct dm_target *ti;
1099         unsigned i;
1100
1101         if (unlikely(type == DM_TYPE_NONE)) {
1102                 DMWARN("no table type is set, can't allocate mempools");
1103                 return -EINVAL;
1104         }
1105
1106         if (__table_type_bio_based(type))
1107                 for (i = 0; i < t->num_targets; i++) {
1108                         ti = t->targets + i;
1109                         per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1110                         min_pool_size = max(min_pool_size, ti->num_flush_bios);
1111                 }
1112
1113         t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported,
1114                                            per_io_data_size, min_pool_size);
1115         if (!t->mempools)
1116                 return -ENOMEM;
1117
1118         return 0;
1119 }
1120
1121 void dm_table_free_md_mempools(struct dm_table *t)
1122 {
1123         dm_free_md_mempools(t->mempools);
1124         t->mempools = NULL;
1125 }
1126
1127 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1128 {
1129         return t->mempools;
1130 }
1131
1132 static int setup_indexes(struct dm_table *t)
1133 {
1134         int i;
1135         unsigned int total = 0;
1136         sector_t *indexes;
1137
1138         /* allocate the space for *all* the indexes */
1139         for (i = t->depth - 2; i >= 0; i--) {
1140                 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1141                 total += t->counts[i];
1142         }
1143
1144         indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1145         if (!indexes)
1146                 return -ENOMEM;
1147
1148         /* set up internal nodes, bottom-up */
1149         for (i = t->depth - 2; i >= 0; i--) {
1150                 t->index[i] = indexes;
1151                 indexes += (KEYS_PER_NODE * t->counts[i]);
1152                 setup_btree_index(i, t);
1153         }
1154
1155         return 0;
1156 }
1157
1158 /*
1159  * Builds the btree to index the map.
1160  */
1161 static int dm_table_build_index(struct dm_table *t)
1162 {
1163         int r = 0;
1164         unsigned int leaf_nodes;
1165
1166         /* how many indexes will the btree have ? */
1167         leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1168         t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1169
1170         /* leaf layer has already been set up */
1171         t->counts[t->depth - 1] = leaf_nodes;
1172         t->index[t->depth - 1] = t->highs;
1173
1174         if (t->depth >= 2)
1175                 r = setup_indexes(t);
1176
1177         return r;
1178 }
1179
1180 static bool integrity_profile_exists(struct gendisk *disk)
1181 {
1182         return !!blk_get_integrity(disk);
1183 }
1184
1185 /*
1186  * Get a disk whose integrity profile reflects the table's profile.
1187  * Returns NULL if integrity support was inconsistent or unavailable.
1188  */
1189 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1190 {
1191         struct list_head *devices = dm_table_get_devices(t);
1192         struct dm_dev_internal *dd = NULL;
1193         struct gendisk *prev_disk = NULL, *template_disk = NULL;
1194         unsigned i;
1195
1196         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1197                 struct dm_target *ti = dm_table_get_target(t, i);
1198                 if (!dm_target_passes_integrity(ti->type))
1199                         goto no_integrity;
1200         }
1201
1202         list_for_each_entry(dd, devices, list) {
1203                 template_disk = dd->dm_dev->bdev->bd_disk;
1204                 if (!integrity_profile_exists(template_disk))
1205                         goto no_integrity;
1206                 else if (prev_disk &&
1207                          blk_integrity_compare(prev_disk, template_disk) < 0)
1208                         goto no_integrity;
1209                 prev_disk = template_disk;
1210         }
1211
1212         return template_disk;
1213
1214 no_integrity:
1215         if (prev_disk)
1216                 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1217                        dm_device_name(t->md),
1218                        prev_disk->disk_name,
1219                        template_disk->disk_name);
1220         return NULL;
1221 }
1222
1223 /*
1224  * Register the mapped device for blk_integrity support if the
1225  * underlying devices have an integrity profile.  But all devices may
1226  * not have matching profiles (checking all devices isn't reliable
1227  * during table load because this table may use other DM device(s) which
1228  * must be resumed before they will have an initialized integity
1229  * profile).  Consequently, stacked DM devices force a 2 stage integrity
1230  * profile validation: First pass during table load, final pass during
1231  * resume.
1232  */
1233 static int dm_table_register_integrity(struct dm_table *t)
1234 {
1235         struct mapped_device *md = t->md;
1236         struct gendisk *template_disk = NULL;
1237
1238         /* If target handles integrity itself do not register it here. */
1239         if (t->integrity_added)
1240                 return 0;
1241
1242         template_disk = dm_table_get_integrity_disk(t);
1243         if (!template_disk)
1244                 return 0;
1245
1246         if (!integrity_profile_exists(dm_disk(md))) {
1247                 t->integrity_supported = true;
1248                 /*
1249                  * Register integrity profile during table load; we can do
1250                  * this because the final profile must match during resume.
1251                  */
1252                 blk_integrity_register(dm_disk(md),
1253                                        blk_get_integrity(template_disk));
1254                 return 0;
1255         }
1256
1257         /*
1258          * If DM device already has an initialized integrity
1259          * profile the new profile should not conflict.
1260          */
1261         if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1262                 DMWARN("%s: conflict with existing integrity profile: "
1263                        "%s profile mismatch",
1264                        dm_device_name(t->md),
1265                        template_disk->disk_name);
1266                 return 1;
1267         }
1268
1269         /* Preserve existing integrity profile */
1270         t->integrity_supported = true;
1271         return 0;
1272 }
1273
1274 /*
1275  * Prepares the table for use by building the indices,
1276  * setting the type, and allocating mempools.
1277  */
1278 int dm_table_complete(struct dm_table *t)
1279 {
1280         int r;
1281
1282         r = dm_table_determine_type(t);
1283         if (r) {
1284                 DMERR("unable to determine table type");
1285                 return r;
1286         }
1287
1288         r = dm_table_build_index(t);
1289         if (r) {
1290                 DMERR("unable to build btrees");
1291                 return r;
1292         }
1293
1294         r = dm_table_register_integrity(t);
1295         if (r) {
1296                 DMERR("could not register integrity profile.");
1297                 return r;
1298         }
1299
1300         r = dm_table_alloc_md_mempools(t, t->md);
1301         if (r)
1302                 DMERR("unable to allocate mempools");
1303
1304         return r;
1305 }
1306
1307 static DEFINE_MUTEX(_event_lock);
1308 void dm_table_event_callback(struct dm_table *t,
1309                              void (*fn)(void *), void *context)
1310 {
1311         mutex_lock(&_event_lock);
1312         t->event_fn = fn;
1313         t->event_context = context;
1314         mutex_unlock(&_event_lock);
1315 }
1316
1317 void dm_table_event(struct dm_table *t)
1318 {
1319         /*
1320          * You can no longer call dm_table_event() from interrupt
1321          * context, use a bottom half instead.
1322          */
1323         BUG_ON(in_interrupt());
1324
1325         mutex_lock(&_event_lock);
1326         if (t->event_fn)
1327                 t->event_fn(t->event_context);
1328         mutex_unlock(&_event_lock);
1329 }
1330 EXPORT_SYMBOL(dm_table_event);
1331
1332 inline sector_t dm_table_get_size(struct dm_table *t)
1333 {
1334         return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1335 }
1336 EXPORT_SYMBOL(dm_table_get_size);
1337
1338 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1339 {
1340         if (index >= t->num_targets)
1341                 return NULL;
1342
1343         return t->targets + index;
1344 }
1345
1346 /*
1347  * Search the btree for the correct target.
1348  *
1349  * Caller should check returned pointer for NULL
1350  * to trap I/O beyond end of device.
1351  */
1352 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1353 {
1354         unsigned int l, n = 0, k = 0;
1355         sector_t *node;
1356
1357         if (unlikely(sector >= dm_table_get_size(t)))
1358                 return NULL;
1359
1360         for (l = 0; l < t->depth; l++) {
1361                 n = get_child(n, k);
1362                 node = get_node(t, l, n);
1363
1364                 for (k = 0; k < KEYS_PER_NODE; k++)
1365                         if (node[k] >= sector)
1366                                 break;
1367         }
1368
1369         return &t->targets[(KEYS_PER_NODE * n) + k];
1370 }
1371
1372 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1373                         sector_t start, sector_t len, void *data)
1374 {
1375         unsigned *num_devices = data;
1376
1377         (*num_devices)++;
1378
1379         return 0;
1380 }
1381
1382 /*
1383  * Check whether a table has no data devices attached using each
1384  * target's iterate_devices method.
1385  * Returns false if the result is unknown because a target doesn't
1386  * support iterate_devices.
1387  */
1388 bool dm_table_has_no_data_devices(struct dm_table *table)
1389 {
1390         struct dm_target *ti;
1391         unsigned i, num_devices;
1392
1393         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1394                 ti = dm_table_get_target(table, i);
1395
1396                 if (!ti->type->iterate_devices)
1397                         return false;
1398
1399                 num_devices = 0;
1400                 ti->type->iterate_devices(ti, count_device, &num_devices);
1401                 if (num_devices)
1402                         return false;
1403         }
1404
1405         return true;
1406 }
1407
1408 static int device_is_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1409                                  sector_t start, sector_t len, void *data)
1410 {
1411         struct request_queue *q = bdev_get_queue(dev->bdev);
1412         enum blk_zoned_model *zoned_model = data;
1413
1414         return q && blk_queue_zoned_model(q) == *zoned_model;
1415 }
1416
1417 static bool dm_table_supports_zoned_model(struct dm_table *t,
1418                                           enum blk_zoned_model zoned_model)
1419 {
1420         struct dm_target *ti;
1421         unsigned i;
1422
1423         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1424                 ti = dm_table_get_target(t, i);
1425
1426                 if (zoned_model == BLK_ZONED_HM &&
1427                     !dm_target_supports_zoned_hm(ti->type))
1428                         return false;
1429
1430                 if (!ti->type->iterate_devices ||
1431                     !ti->type->iterate_devices(ti, device_is_zoned_model, &zoned_model))
1432                         return false;
1433         }
1434
1435         return true;
1436 }
1437
1438 static int device_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1439                                        sector_t start, sector_t len, void *data)
1440 {
1441         struct request_queue *q = bdev_get_queue(dev->bdev);
1442         unsigned int *zone_sectors = data;
1443
1444         return q && blk_queue_zone_sectors(q) == *zone_sectors;
1445 }
1446
1447 static bool dm_table_matches_zone_sectors(struct dm_table *t,
1448                                           unsigned int zone_sectors)
1449 {
1450         struct dm_target *ti;
1451         unsigned i;
1452
1453         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1454                 ti = dm_table_get_target(t, i);
1455
1456                 if (!ti->type->iterate_devices ||
1457                     !ti->type->iterate_devices(ti, device_matches_zone_sectors, &zone_sectors))
1458                         return false;
1459         }
1460
1461         return true;
1462 }
1463
1464 static int validate_hardware_zoned_model(struct dm_table *table,
1465                                          enum blk_zoned_model zoned_model,
1466                                          unsigned int zone_sectors)
1467 {
1468         if (zoned_model == BLK_ZONED_NONE)
1469                 return 0;
1470
1471         if (!dm_table_supports_zoned_model(table, zoned_model)) {
1472                 DMERR("%s: zoned model is not consistent across all devices",
1473                       dm_device_name(table->md));
1474                 return -EINVAL;
1475         }
1476
1477         /* Check zone size validity and compatibility */
1478         if (!zone_sectors || !is_power_of_2(zone_sectors))
1479                 return -EINVAL;
1480
1481         if (!dm_table_matches_zone_sectors(table, zone_sectors)) {
1482                 DMERR("%s: zone sectors is not consistent across all devices",
1483                       dm_device_name(table->md));
1484                 return -EINVAL;
1485         }
1486
1487         return 0;
1488 }
1489
1490 /*
1491  * Establish the new table's queue_limits and validate them.
1492  */
1493 int dm_calculate_queue_limits(struct dm_table *table,
1494                               struct queue_limits *limits)
1495 {
1496         struct dm_target *ti;
1497         struct queue_limits ti_limits;
1498         unsigned i;
1499         enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1500         unsigned int zone_sectors = 0;
1501
1502         blk_set_stacking_limits(limits);
1503
1504         for (i = 0; i < dm_table_get_num_targets(table); i++) {
1505                 blk_set_stacking_limits(&ti_limits);
1506
1507                 ti = dm_table_get_target(table, i);
1508
1509                 if (!ti->type->iterate_devices)
1510                         goto combine_limits;
1511
1512                 /*
1513                  * Combine queue limits of all the devices this target uses.
1514                  */
1515                 ti->type->iterate_devices(ti, dm_set_device_limits,
1516                                           &ti_limits);
1517
1518                 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1519                         /*
1520                          * After stacking all limits, validate all devices
1521                          * in table support this zoned model and zone sectors.
1522                          */
1523                         zoned_model = ti_limits.zoned;
1524                         zone_sectors = ti_limits.chunk_sectors;
1525                 }
1526
1527                 /* Set I/O hints portion of queue limits */
1528                 if (ti->type->io_hints)
1529                         ti->type->io_hints(ti, &ti_limits);
1530
1531                 /*
1532                  * Check each device area is consistent with the target's
1533                  * overall queue limits.
1534                  */
1535                 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1536                                               &ti_limits))
1537                         return -EINVAL;
1538
1539 combine_limits:
1540                 /*
1541                  * Merge this target's queue limits into the overall limits
1542                  * for the table.
1543                  */
1544                 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1545                         DMWARN("%s: adding target device "
1546                                "(start sect %llu len %llu) "
1547                                "caused an alignment inconsistency",
1548                                dm_device_name(table->md),
1549                                (unsigned long long) ti->begin,
1550                                (unsigned long long) ti->len);
1551
1552                 /*
1553                  * FIXME: this should likely be moved to blk_stack_limits(), would
1554                  * also eliminate limits->zoned stacking hack in dm_set_device_limits()
1555                  */
1556                 if (limits->zoned == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1557                         /*
1558                          * By default, the stacked limits zoned model is set to
1559                          * BLK_ZONED_NONE in blk_set_stacking_limits(). Update
1560                          * this model using the first target model reported
1561                          * that is not BLK_ZONED_NONE. This will be either the
1562                          * first target device zoned model or the model reported
1563                          * by the target .io_hints.
1564                          */
1565                         limits->zoned = ti_limits.zoned;
1566                 }
1567         }
1568
1569         /*
1570          * Verify that the zoned model and zone sectors, as determined before
1571          * any .io_hints override, are the same across all devices in the table.
1572          * - this is especially relevant if .io_hints is emulating a disk-managed
1573          *   zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1574          * BUT...
1575          */
1576         if (limits->zoned != BLK_ZONED_NONE) {
1577                 /*
1578                  * ...IF the above limits stacking determined a zoned model
1579                  * validate that all of the table's devices conform to it.
1580                  */
1581                 zoned_model = limits->zoned;
1582                 zone_sectors = limits->chunk_sectors;
1583         }
1584         if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1585                 return -EINVAL;
1586
1587         return validate_hardware_logical_block_alignment(table, limits);
1588 }
1589
1590 /*
1591  * Verify that all devices have an integrity profile that matches the
1592  * DM device's registered integrity profile.  If the profiles don't
1593  * match then unregister the DM device's integrity profile.
1594  */
1595 static void dm_table_verify_integrity(struct dm_table *t)
1596 {
1597         struct gendisk *template_disk = NULL;
1598
1599         if (t->integrity_added)
1600                 return;
1601
1602         if (t->integrity_supported) {
1603                 /*
1604                  * Verify that the original integrity profile
1605                  * matches all the devices in this table.
1606                  */
1607                 template_disk = dm_table_get_integrity_disk(t);
1608                 if (template_disk &&
1609                     blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1610                         return;
1611         }
1612
1613         if (integrity_profile_exists(dm_disk(t->md))) {
1614                 DMWARN("%s: unable to establish an integrity profile",
1615                        dm_device_name(t->md));
1616                 blk_integrity_unregister(dm_disk(t->md));
1617         }
1618 }
1619
1620 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1621                                 sector_t start, sector_t len, void *data)
1622 {
1623         unsigned long flush = (unsigned long) data;
1624         struct request_queue *q = bdev_get_queue(dev->bdev);
1625
1626         return q && (q->queue_flags & flush);
1627 }
1628
1629 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1630 {
1631         struct dm_target *ti;
1632         unsigned i;
1633
1634         /*
1635          * Require at least one underlying device to support flushes.
1636          * t->devices includes internal dm devices such as mirror logs
1637          * so we need to use iterate_devices here, which targets
1638          * supporting flushes must provide.
1639          */
1640         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1641                 ti = dm_table_get_target(t, i);
1642
1643                 if (!ti->num_flush_bios)
1644                         continue;
1645
1646                 if (ti->flush_supported)
1647                         return true;
1648
1649                 if (ti->type->iterate_devices &&
1650                     ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1651                         return true;
1652         }
1653
1654         return false;
1655 }
1656
1657 static int device_dax_write_cache_enabled(struct dm_target *ti,
1658                                           struct dm_dev *dev, sector_t start,
1659                                           sector_t len, void *data)
1660 {
1661         struct dax_device *dax_dev = dev->dax_dev;
1662
1663         if (!dax_dev)
1664                 return false;
1665
1666         if (dax_write_cache_enabled(dax_dev))
1667                 return true;
1668         return false;
1669 }
1670
1671 static int dm_table_supports_dax_write_cache(struct dm_table *t)
1672 {
1673         struct dm_target *ti;
1674         unsigned i;
1675
1676         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1677                 ti = dm_table_get_target(t, i);
1678
1679                 if (ti->type->iterate_devices &&
1680                     ti->type->iterate_devices(ti,
1681                                 device_dax_write_cache_enabled, NULL))
1682                         return true;
1683         }
1684
1685         return false;
1686 }
1687
1688 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1689                             sector_t start, sector_t len, void *data)
1690 {
1691         struct request_queue *q = bdev_get_queue(dev->bdev);
1692
1693         return q && blk_queue_nonrot(q);
1694 }
1695
1696 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1697                              sector_t start, sector_t len, void *data)
1698 {
1699         struct request_queue *q = bdev_get_queue(dev->bdev);
1700
1701         return q && !blk_queue_add_random(q);
1702 }
1703
1704 static bool dm_table_all_devices_attribute(struct dm_table *t,
1705                                            iterate_devices_callout_fn func)
1706 {
1707         struct dm_target *ti;
1708         unsigned i;
1709
1710         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1711                 ti = dm_table_get_target(t, i);
1712
1713                 if (!ti->type->iterate_devices ||
1714                     !ti->type->iterate_devices(ti, func, NULL))
1715                         return false;
1716         }
1717
1718         return true;
1719 }
1720
1721 static int device_no_partial_completion(struct dm_target *ti, struct dm_dev *dev,
1722                                         sector_t start, sector_t len, void *data)
1723 {
1724         char b[BDEVNAME_SIZE];
1725
1726         /* For now, NVMe devices are the only devices of this class */
1727         return (strncmp(bdevname(dev->bdev, b), "nvme", 4) == 0);
1728 }
1729
1730 static bool dm_table_does_not_support_partial_completion(struct dm_table *t)
1731 {
1732         return dm_table_all_devices_attribute(t, device_no_partial_completion);
1733 }
1734
1735 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1736                                          sector_t start, sector_t len, void *data)
1737 {
1738         struct request_queue *q = bdev_get_queue(dev->bdev);
1739
1740         return q && !q->limits.max_write_same_sectors;
1741 }
1742
1743 static bool dm_table_supports_write_same(struct dm_table *t)
1744 {
1745         struct dm_target *ti;
1746         unsigned i;
1747
1748         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1749                 ti = dm_table_get_target(t, i);
1750
1751                 if (!ti->num_write_same_bios)
1752                         return false;
1753
1754                 if (!ti->type->iterate_devices ||
1755                     ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1756                         return false;
1757         }
1758
1759         return true;
1760 }
1761
1762 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1763                                            sector_t start, sector_t len, void *data)
1764 {
1765         struct request_queue *q = bdev_get_queue(dev->bdev);
1766
1767         return q && !q->limits.max_write_zeroes_sectors;
1768 }
1769
1770 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1771 {
1772         struct dm_target *ti;
1773         unsigned i = 0;
1774
1775         while (i < dm_table_get_num_targets(t)) {
1776                 ti = dm_table_get_target(t, i++);
1777
1778                 if (!ti->num_write_zeroes_bios)
1779                         return false;
1780
1781                 if (!ti->type->iterate_devices ||
1782                     ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1783                         return false;
1784         }
1785
1786         return true;
1787 }
1788
1789 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1790                                       sector_t start, sector_t len, void *data)
1791 {
1792         struct request_queue *q = bdev_get_queue(dev->bdev);
1793
1794         return q && !blk_queue_discard(q);
1795 }
1796
1797 static bool dm_table_supports_discards(struct dm_table *t)
1798 {
1799         struct dm_target *ti;
1800         unsigned i;
1801
1802         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1803                 ti = dm_table_get_target(t, i);
1804
1805                 if (!ti->num_discard_bios)
1806                         return false;
1807
1808                 /*
1809                  * Either the target provides discard support (as implied by setting
1810                  * 'discards_supported') or it relies on _all_ data devices having
1811                  * discard support.
1812                  */
1813                 if (!ti->discards_supported &&
1814                     (!ti->type->iterate_devices ||
1815                      ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1816                         return false;
1817         }
1818
1819         return true;
1820 }
1821
1822 static int device_not_secure_erase_capable(struct dm_target *ti,
1823                                            struct dm_dev *dev, sector_t start,
1824                                            sector_t len, void *data)
1825 {
1826         struct request_queue *q = bdev_get_queue(dev->bdev);
1827
1828         return q && !blk_queue_secure_erase(q);
1829 }
1830
1831 static bool dm_table_supports_secure_erase(struct dm_table *t)
1832 {
1833         struct dm_target *ti;
1834         unsigned int i;
1835
1836         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1837                 ti = dm_table_get_target(t, i);
1838
1839                 if (!ti->num_secure_erase_bios)
1840                         return false;
1841
1842                 if (!ti->type->iterate_devices ||
1843                     ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1844                         return false;
1845         }
1846
1847         return true;
1848 }
1849
1850 static int device_requires_stable_pages(struct dm_target *ti,
1851                                         struct dm_dev *dev, sector_t start,
1852                                         sector_t len, void *data)
1853 {
1854         struct request_queue *q = bdev_get_queue(dev->bdev);
1855
1856         return q && bdi_cap_stable_pages_required(q->backing_dev_info);
1857 }
1858
1859 /*
1860  * If any underlying device requires stable pages, a table must require
1861  * them as well.  Only targets that support iterate_devices are considered:
1862  * don't want error, zero, etc to require stable pages.
1863  */
1864 static bool dm_table_requires_stable_pages(struct dm_table *t)
1865 {
1866         struct dm_target *ti;
1867         unsigned i;
1868
1869         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1870                 ti = dm_table_get_target(t, i);
1871
1872                 if (ti->type->iterate_devices &&
1873                     ti->type->iterate_devices(ti, device_requires_stable_pages, NULL))
1874                         return true;
1875         }
1876
1877         return false;
1878 }
1879
1880 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1881                                struct queue_limits *limits)
1882 {
1883         bool wc = false, fua = false;
1884         int page_size = PAGE_SIZE;
1885
1886         /*
1887          * Copy table's limits to the DM device's request_queue
1888          */
1889         q->limits = *limits;
1890
1891         if (!dm_table_supports_discards(t)) {
1892                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
1893                 /* Must also clear discard limits... */
1894                 q->limits.max_discard_sectors = 0;
1895                 q->limits.max_hw_discard_sectors = 0;
1896                 q->limits.discard_granularity = 0;
1897                 q->limits.discard_alignment = 0;
1898                 q->limits.discard_misaligned = 0;
1899         } else
1900                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
1901
1902         if (dm_table_supports_secure_erase(t))
1903                 blk_queue_flag_set(QUEUE_FLAG_SECERASE, q);
1904
1905         if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1906                 wc = true;
1907                 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1908                         fua = true;
1909         }
1910         blk_queue_write_cache(q, wc, fua);
1911
1912         if (dm_table_supports_dax(t, device_supports_dax, &page_size)) {
1913                 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
1914                 if (dm_table_supports_dax(t, device_dax_synchronous, NULL))
1915                         set_dax_synchronous(t->md->dax_dev);
1916         }
1917         else
1918                 blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
1919
1920         if (dm_table_supports_dax_write_cache(t))
1921                 dax_write_cache(t->md->dax_dev, true);
1922
1923         /* Ensure that all underlying devices are non-rotational. */
1924         if (dm_table_all_devices_attribute(t, device_is_nonrot))
1925                 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
1926         else
1927                 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
1928
1929         if (!dm_table_supports_write_same(t))
1930                 q->limits.max_write_same_sectors = 0;
1931         if (!dm_table_supports_write_zeroes(t))
1932                 q->limits.max_write_zeroes_sectors = 0;
1933
1934         dm_table_verify_integrity(t);
1935
1936         /*
1937          * Some devices don't use blk_integrity but still want stable pages
1938          * because they do their own checksumming.
1939          */
1940         if (dm_table_requires_stable_pages(t))
1941                 q->backing_dev_info->capabilities |= BDI_CAP_STABLE_WRITES;
1942         else
1943                 q->backing_dev_info->capabilities &= ~BDI_CAP_STABLE_WRITES;
1944
1945         /*
1946          * Determine whether or not this queue's I/O timings contribute
1947          * to the entropy pool, Only request-based targets use this.
1948          * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1949          * have it set.
1950          */
1951         if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1952                 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
1953
1954         /*
1955          * For a zoned target, the number of zones should be updated for the
1956          * correct value to be exposed in sysfs queue/nr_zones. For a BIO based
1957          * target, this is all that is needed.
1958          */
1959 #ifdef CONFIG_BLK_DEV_ZONED
1960         if (blk_queue_is_zoned(q)) {
1961                 WARN_ON_ONCE(queue_is_mq(q));
1962                 q->nr_zones = blkdev_nr_zones(t->md->disk);
1963         }
1964 #endif
1965
1966         /* Allow reads to exceed readahead limits */
1967         q->backing_dev_info->io_pages = limits->max_sectors >> (PAGE_SHIFT - 9);
1968 }
1969
1970 unsigned int dm_table_get_num_targets(struct dm_table *t)
1971 {
1972         return t->num_targets;
1973 }
1974
1975 struct list_head *dm_table_get_devices(struct dm_table *t)
1976 {
1977         return &t->devices;
1978 }
1979
1980 fmode_t dm_table_get_mode(struct dm_table *t)
1981 {
1982         return t->mode;
1983 }
1984 EXPORT_SYMBOL(dm_table_get_mode);
1985
1986 enum suspend_mode {
1987         PRESUSPEND,
1988         PRESUSPEND_UNDO,
1989         POSTSUSPEND,
1990 };
1991
1992 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1993 {
1994         int i = t->num_targets;
1995         struct dm_target *ti = t->targets;
1996
1997         lockdep_assert_held(&t->md->suspend_lock);
1998
1999         while (i--) {
2000                 switch (mode) {
2001                 case PRESUSPEND:
2002                         if (ti->type->presuspend)
2003                                 ti->type->presuspend(ti);
2004                         break;
2005                 case PRESUSPEND_UNDO:
2006                         if (ti->type->presuspend_undo)
2007                                 ti->type->presuspend_undo(ti);
2008                         break;
2009                 case POSTSUSPEND:
2010                         if (ti->type->postsuspend)
2011                                 ti->type->postsuspend(ti);
2012                         break;
2013                 }
2014                 ti++;
2015         }
2016 }
2017
2018 void dm_table_presuspend_targets(struct dm_table *t)
2019 {
2020         if (!t)
2021                 return;
2022
2023         suspend_targets(t, PRESUSPEND);
2024 }
2025
2026 void dm_table_presuspend_undo_targets(struct dm_table *t)
2027 {
2028         if (!t)
2029                 return;
2030
2031         suspend_targets(t, PRESUSPEND_UNDO);
2032 }
2033
2034 void dm_table_postsuspend_targets(struct dm_table *t)
2035 {
2036         if (!t)
2037                 return;
2038
2039         suspend_targets(t, POSTSUSPEND);
2040 }
2041
2042 int dm_table_resume_targets(struct dm_table *t)
2043 {
2044         int i, r = 0;
2045
2046         lockdep_assert_held(&t->md->suspend_lock);
2047
2048         for (i = 0; i < t->num_targets; i++) {
2049                 struct dm_target *ti = t->targets + i;
2050
2051                 if (!ti->type->preresume)
2052                         continue;
2053
2054                 r = ti->type->preresume(ti);
2055                 if (r) {
2056                         DMERR("%s: %s: preresume failed, error = %d",
2057                               dm_device_name(t->md), ti->type->name, r);
2058                         return r;
2059                 }
2060         }
2061
2062         for (i = 0; i < t->num_targets; i++) {
2063                 struct dm_target *ti = t->targets + i;
2064
2065                 if (ti->type->resume)
2066                         ti->type->resume(ti);
2067         }
2068
2069         return 0;
2070 }
2071
2072 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
2073 {
2074         list_add(&cb->list, &t->target_callbacks);
2075 }
2076 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
2077
2078 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
2079 {
2080         struct dm_dev_internal *dd;
2081         struct list_head *devices = dm_table_get_devices(t);
2082         struct dm_target_callbacks *cb;
2083         int r = 0;
2084
2085         list_for_each_entry(dd, devices, list) {
2086                 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
2087                 char b[BDEVNAME_SIZE];
2088
2089                 if (likely(q))
2090                         r |= bdi_congested(q->backing_dev_info, bdi_bits);
2091                 else
2092                         DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
2093                                      dm_device_name(t->md),
2094                                      bdevname(dd->dm_dev->bdev, b));
2095         }
2096
2097         list_for_each_entry(cb, &t->target_callbacks, list)
2098                 if (cb->congested_fn)
2099                         r |= cb->congested_fn(cb, bdi_bits);
2100
2101         return r;
2102 }
2103
2104 struct mapped_device *dm_table_get_md(struct dm_table *t)
2105 {
2106         return t->md;
2107 }
2108 EXPORT_SYMBOL(dm_table_get_md);
2109
2110 const char *dm_table_device_name(struct dm_table *t)
2111 {
2112         return dm_device_name(t->md);
2113 }
2114 EXPORT_SYMBOL_GPL(dm_table_device_name);
2115
2116 void dm_table_run_md_queue_async(struct dm_table *t)
2117 {
2118         struct mapped_device *md;
2119         struct request_queue *queue;
2120
2121         if (!dm_table_request_based(t))
2122                 return;
2123
2124         md = dm_table_get_md(t);
2125         queue = dm_get_md_queue(md);
2126         if (queue)
2127                 blk_mq_run_hw_queues(queue, true);
2128 }
2129 EXPORT_SYMBOL(dm_table_run_md_queue_async);
2130