drivers/md/dm-table.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) 2001 Sistina Software (UK) Limited.
   4  * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
   5  *
   6  * This file is released under the GPL.
   7  */
   8
   9 #include "dm-core.h"
  10 #include "dm-rq.h"
  11
  12 #include <linux/module.h>
  13 #include <linux/vmalloc.h>
  14 #include <linux/blkdev.h>
  15 #include <linux/blk-integrity.h>
  16 #include <linux/namei.h>
  17 #include <linux/ctype.h>
  18 #include <linux/string.h>
  19 #include <linux/slab.h>
  20 #include <linux/interrupt.h>
  21 #include <linux/mutex.h>
  22 #include <linux/delay.h>
  23 #include <linux/atomic.h>
  24 #include <linux/blk-mq.h>
  25 #include <linux/mount.h>
  26 #include <linux/dax.h>
  27
  28 #define DM_MSG_PREFIX "table"
  29
  30 #define NODE_SIZE L1_CACHE_BYTES
  31 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
  32 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
  33
  34 /*
  35  * Similar to ceiling(log_size(n))
  36  */
  37 static unsigned int int_log(unsigned int n, unsigned int base)
  38 {
  39         int result = 0;
  40
  41         while (n > 1) {
  42                 n = dm_div_up(n, base);
  43                 result++;
  44         }
  45
  46         return result;
  47 }
  48
  49 /*
  50  * Calculate the index of the child node of the n'th node k'th key.
  51  */
  52 static inline unsigned int get_child(unsigned int n, unsigned int k)
  53 {
  54         return (n * CHILDREN_PER_NODE) + k;
  55 }
  56
  57 /*
  58  * Return the n'th node of level l from table t.
  59  */
  60 static inline sector_t *get_node(struct dm_table *t,
  61                                  unsigned int l, unsigned int n)
  62 {
  63         return t->index[l] + (n * KEYS_PER_NODE);
  64 }
  65
  66 /*
  67  * Return the highest key that you could lookup from the n'th
  68  * node on level l of the btree.
  69  */
  70 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
  71 {
  72         for (; l < t->depth - 1; l++)
  73                 n = get_child(n, CHILDREN_PER_NODE - 1);
  74
  75         if (n >= t->counts[l])
  76                 return (sector_t) -1;
  77
  78         return get_node(t, l, n)[KEYS_PER_NODE - 1];
  79 }
  80
  81 /*
  82  * Fills in a level of the btree based on the highs of the level
  83  * below it.
  84  */
  85 static int setup_btree_index(unsigned int l, struct dm_table *t)
  86 {
  87         unsigned int n, k;
  88         sector_t *node;
  89
  90         for (n = 0U; n < t->counts[l]; n++) {
  91                 node = get_node(t, l, n);
  92
  93                 for (k = 0U; k < KEYS_PER_NODE; k++)
  94                         node[k] = high(t, l + 1, get_child(n, k));
  95         }
  96
  97         return 0;
  98 }
  99
 100 /*
 101  * highs, and targets are managed as dynamic arrays during a
 102  * table load.
 103  */
 104 static int alloc_targets(struct dm_table *t, unsigned int num)
 105 {
 106         sector_t *n_highs;
 107         struct dm_target *n_targets;
 108
 109         /*
 110          * Allocate both the target array and offset array at once.
 111          */
 112         n_highs = kvcalloc(num, sizeof(struct dm_target) + sizeof(sector_t),
 113                            GFP_KERNEL);
 114         if (!n_highs)
 115                 return -ENOMEM;
 116
 117         n_targets = (struct dm_target *) (n_highs + num);
 118
 119         memset(n_highs, -1, sizeof(*n_highs) * num);
 120         kvfree(t->highs);
 121
 122         t->num_allocated = num;
 123         t->highs = n_highs;
 124         t->targets = n_targets;
 125
 126         return 0;
 127 }
 128
 129 int dm_table_create(struct dm_table **result, blk_mode_t mode,
 130                     unsigned int num_targets, struct mapped_device *md)
 131 {
 132         struct dm_table *t;
 133
 134         if (num_targets > DM_MAX_TARGETS)
 135                 return -EOVERFLOW;
 136
 137         t = kzalloc(sizeof(*t), GFP_KERNEL);
 138
 139         if (!t)
 140                 return -ENOMEM;
 141
 142         INIT_LIST_HEAD(&t->devices);
 143         init_rwsem(&t->devices_lock);
 144
 145         if (!num_targets)
 146                 num_targets = KEYS_PER_NODE;
 147
 148         num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
 149
 150         if (!num_targets) {
 151                 kfree(t);
 152                 return -EOVERFLOW;
 153         }
 154
 155         if (alloc_targets(t, num_targets)) {
 156                 kfree(t);
 157                 return -ENOMEM;
 158         }
 159
 160         t->type = DM_TYPE_NONE;
 161         t->mode = mode;
 162         t->md = md;
 163         t->flush_bypasses_map = true;
 164         *result = t;
 165         return 0;
 166 }
 167
 168 static void free_devices(struct list_head *devices, struct mapped_device *md)
 169 {
 170         struct list_head *tmp, *next;
 171
 172         list_for_each_safe(tmp, next, devices) {
 173                 struct dm_dev_internal *dd =
 174                     list_entry(tmp, struct dm_dev_internal, list);
 175                 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
 176                        dm_device_name(md), dd->dm_dev->name);
 177                 dm_put_table_device(md, dd->dm_dev);
 178                 kfree(dd);
 179         }
 180 }
 181
 182 static void dm_table_destroy_crypto_profile(struct dm_table *t);
 183
 184 void dm_table_destroy(struct dm_table *t)
 185 {
 186         if (!t)
 187                 return;
 188
 189         /* free the indexes */
 190         if (t->depth >= 2)
 191                 kvfree(t->index[t->depth - 2]);
 192
 193         /* free the targets */
 194         for (unsigned int i = 0; i < t->num_targets; i++) {
 195                 struct dm_target *ti = dm_table_get_target(t, i);
 196
 197                 if (ti->type->dtr)
 198                         ti->type->dtr(ti);
 199
 200                 dm_put_target_type(ti->type);
 201         }
 202
 203         kvfree(t->highs);
 204
 205         /* free the device list */
 206         free_devices(&t->devices, t->md);
 207
 208         dm_free_md_mempools(t->mempools);
 209
 210         dm_table_destroy_crypto_profile(t);
 211
 212         kfree(t);
 213 }
 214
 215 /*
 216  * See if we've already got a device in the list.
 217  */
 218 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
 219 {
 220         struct dm_dev_internal *dd;
 221
 222         list_for_each_entry(dd, l, list)
 223                 if (dd->dm_dev->bdev->bd_dev == dev)
 224                         return dd;
 225
 226         return NULL;
 227 }
 228
 229 /*
 230  * If possible, this checks an area of a destination device is invalid.
 231  */
 232 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
 233                                   sector_t start, sector_t len, void *data)
 234 {
 235         struct queue_limits *limits = data;
 236         struct block_device *bdev = dev->bdev;
 237         sector_t dev_size = bdev_nr_sectors(bdev);
 238         unsigned short logical_block_size_sectors =
 239                 limits->logical_block_size >> SECTOR_SHIFT;
 240
 241         if (!dev_size)
 242                 return 0;
 243
 244         if ((start >= dev_size) || (start + len > dev_size)) {
 245                 DMERR("%s: %pg too small for target: start=%llu, len=%llu, dev_size=%llu",
 246                       dm_device_name(ti->table->md), bdev,
 247                       (unsigned long long)start,
 248                       (unsigned long long)len,
 249                       (unsigned long long)dev_size);
 250                 return 1;
 251         }
 252
 253         /*
 254          * If the target is mapped to zoned block device(s), check
 255          * that the zones are not partially mapped.
 256          */
 257         if (bdev_is_zoned(bdev)) {
 258                 unsigned int zone_sectors = bdev_zone_sectors(bdev);
 259
 260                 if (start & (zone_sectors - 1)) {
 261                         DMERR("%s: start=%llu not aligned to h/w zone size %u of %pg",
 262                               dm_device_name(ti->table->md),
 263                               (unsigned long long)start,
 264                               zone_sectors, bdev);
 265                         return 1;
 266                 }
 267
 268                 /*
 269                  * Note: The last zone of a zoned block device may be smaller
 270                  * than other zones. So for a target mapping the end of a
 271                  * zoned block device with such a zone, len would not be zone
 272                  * aligned. We do not allow such last smaller zone to be part
 273                  * of the mapping here to ensure that mappings with multiple
 274                  * devices do not end up with a smaller zone in the middle of
 275                  * the sector range.
 276                  */
 277                 if (len & (zone_sectors - 1)) {
 278                         DMERR("%s: len=%llu not aligned to h/w zone size %u of %pg",
 279                               dm_device_name(ti->table->md),
 280                               (unsigned long long)len,
 281                               zone_sectors, bdev);
 282                         return 1;
 283                 }
 284         }
 285
 286         if (logical_block_size_sectors <= 1)
 287                 return 0;
 288
 289         if (start & (logical_block_size_sectors - 1)) {
 290                 DMERR("%s: start=%llu not aligned to h/w logical block size %u of %pg",
 291                       dm_device_name(ti->table->md),
 292                       (unsigned long long)start,
 293                       limits->logical_block_size, bdev);
 294                 return 1;
 295         }
 296
 297         if (len & (logical_block_size_sectors - 1)) {
 298                 DMERR("%s: len=%llu not aligned to h/w logical block size %u of %pg",
 299                       dm_device_name(ti->table->md),
 300                       (unsigned long long)len,
 301                       limits->logical_block_size, bdev);
 302                 return 1;
 303         }
 304
 305         return 0;
 306 }
 307
 308 /*
 309  * This upgrades the mode on an already open dm_dev, being
 310  * careful to leave things as they were if we fail to reopen the
 311  * device and not to touch the existing bdev field in case
 312  * it is accessed concurrently.
 313  */
 314 static int upgrade_mode(struct dm_dev_internal *dd, blk_mode_t new_mode,
 315                         struct mapped_device *md)
 316 {
 317         int r;
 318         struct dm_dev *old_dev, *new_dev;
 319
 320         old_dev = dd->dm_dev;
 321
 322         r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
 323                                 dd->dm_dev->mode | new_mode, &new_dev);
 324         if (r)
 325                 return r;
 326
 327         dd->dm_dev = new_dev;
 328         dm_put_table_device(md, old_dev);
 329
 330         return 0;
 331 }
 332
 333 /*
 334  * Note: the __ref annotation is because this function can call the __init
 335  * marked early_lookup_bdev when called during early boot code from dm-init.c.
 336  */
 337 int __ref dm_devt_from_path(const char *path, dev_t *dev_p)
 338 {
 339         int r;
 340         dev_t dev;
 341         unsigned int major, minor;
 342         char dummy;
 343
 344         if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
 345                 /* Extract the major/minor numbers */
 346                 dev = MKDEV(major, minor);
 347                 if (MAJOR(dev) != major || MINOR(dev) != minor)
 348                         return -EOVERFLOW;
 349         } else {
 350                 r = lookup_bdev(path, &dev);
 351 #ifndef MODULE
 352                 if (r && system_state < SYSTEM_RUNNING)
 353                         r = early_lookup_bdev(path, &dev);
 354 #endif
 355                 if (r)
 356                         return r;
 357         }
 358         *dev_p = dev;
 359         return 0;
 360 }
 361 EXPORT_SYMBOL(dm_devt_from_path);
 362
 363 /*
 364  * Add a device to the list, or just increment the usage count if
 365  * it's already present.
 366  */
 367 int dm_get_device(struct dm_target *ti, const char *path, blk_mode_t mode,
 368                   struct dm_dev **result)
 369 {
 370         int r;
 371         dev_t dev;
 372         struct dm_dev_internal *dd;
 373         struct dm_table *t = ti->table;
 374
 375         BUG_ON(!t);
 376
 377         r = dm_devt_from_path(path, &dev);
 378         if (r)
 379                 return r;
 380
 381         if (dev == disk_devt(t->md->disk))
 382                 return -EINVAL;
 383
 384         down_write(&t->devices_lock);
 385
 386         dd = find_device(&t->devices, dev);
 387         if (!dd) {
 388                 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
 389                 if (!dd) {
 390                         r = -ENOMEM;
 391                         goto unlock_ret_r;
 392                 }
 393
 394                 r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev);
 395                 if (r) {
 396                         kfree(dd);
 397                         goto unlock_ret_r;
 398                 }
 399
 400                 refcount_set(&dd->count, 1);
 401                 list_add(&dd->list, &t->devices);
 402                 goto out;
 403
 404         } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
 405                 r = upgrade_mode(dd, mode, t->md);
 406                 if (r)
 407                         goto unlock_ret_r;
 408         }
 409         refcount_inc(&dd->count);
 410 out:
 411         up_write(&t->devices_lock);
 412         *result = dd->dm_dev;
 413         return 0;
 414
 415 unlock_ret_r:
 416         up_write(&t->devices_lock);
 417         return r;
 418 }
 419 EXPORT_SYMBOL(dm_get_device);
 420
 421 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
 422                                 sector_t start, sector_t len, void *data)
 423 {
 424         struct queue_limits *limits = data;
 425         struct block_device *bdev = dev->bdev;
 426         struct request_queue *q = bdev_get_queue(bdev);
 427
 428         if (unlikely(!q)) {
 429                 DMWARN("%s: Cannot set limits for nonexistent device %pg",
 430                        dm_device_name(ti->table->md), bdev);
 431                 return 0;
 432         }
 433
 434         if (blk_stack_limits(limits, &q->limits,
 435                         get_start_sect(bdev) + start) < 0)
 436                 DMWARN("%s: adding target device %pg caused an alignment inconsistency: "
 437                        "physical_block_size=%u, logical_block_size=%u, "
 438                        "alignment_offset=%u, start=%llu",
 439                        dm_device_name(ti->table->md), bdev,
 440                        q->limits.physical_block_size,
 441                        q->limits.logical_block_size,
 442                        q->limits.alignment_offset,
 443                        (unsigned long long) start << SECTOR_SHIFT);
 444
 445         /*
 446          * Only stack the integrity profile if the target doesn't have native
 447          * integrity support.
 448          */
 449         if (!dm_target_has_integrity(ti->type))
 450                 queue_limits_stack_integrity_bdev(limits, bdev);
 451         return 0;
 452 }
 453
 454 /*
 455  * Decrement a device's use count and remove it if necessary.
 456  */
 457 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
 458 {
 459         int found = 0;
 460         struct dm_table *t = ti->table;
 461         struct list_head *devices = &t->devices;
 462         struct dm_dev_internal *dd;
 463
 464         down_write(&t->devices_lock);
 465
 466         list_for_each_entry(dd, devices, list) {
 467                 if (dd->dm_dev == d) {
 468                         found = 1;
 469                         break;
 470                 }
 471         }
 472         if (!found) {
 473                 DMERR("%s: device %s not in table devices list",
 474                       dm_device_name(t->md), d->name);
 475                 goto unlock_ret;
 476         }
 477         if (refcount_dec_and_test(&dd->count)) {
 478                 dm_put_table_device(t->md, d);
 479                 list_del(&dd->list);
 480                 kfree(dd);
 481         }
 482
 483 unlock_ret:
 484         up_write(&t->devices_lock);
 485 }
 486 EXPORT_SYMBOL(dm_put_device);
 487
 488 /*
 489  * Checks to see if the target joins onto the end of the table.
 490  */
 491 static int adjoin(struct dm_table *t, struct dm_target *ti)
 492 {
 493         struct dm_target *prev;
 494
 495         if (!t->num_targets)
 496                 return !ti->begin;
 497
 498         prev = &t->targets[t->num_targets - 1];
 499         return (ti->begin == (prev->begin + prev->len));
 500 }
 501
 502 /*
 503  * Used to dynamically allocate the arg array.
 504  *
 505  * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
 506  * process messages even if some device is suspended. These messages have a
 507  * small fixed number of arguments.
 508  *
 509  * On the other hand, dm-switch needs to process bulk data using messages and
 510  * excessive use of GFP_NOIO could cause trouble.
 511  */
 512 static char **realloc_argv(unsigned int *size, char **old_argv)
 513 {
 514         char **argv;
 515         unsigned int new_size;
 516         gfp_t gfp;
 517
 518         if (*size) {
 519                 new_size = *size * 2;
 520                 gfp = GFP_KERNEL;
 521         } else {
 522                 new_size = 8;
 523                 gfp = GFP_NOIO;
 524         }
 525         argv = kmalloc_array(new_size, sizeof(*argv), gfp);
 526         if (argv && old_argv) {
 527                 memcpy(argv, old_argv, *size * sizeof(*argv));
 528                 *size = new_size;
 529         }
 530
 531         kfree(old_argv);
 532         return argv;
 533 }
 534
 535 /*
 536  * Destructively splits up the argument list to pass to ctr.
 537  */
 538 int dm_split_args(int *argc, char ***argvp, char *input)
 539 {
 540         char *start, *end = input, *out, **argv = NULL;
 541         unsigned int array_size = 0;
 542
 543         *argc = 0;
 544
 545         if (!input) {
 546                 *argvp = NULL;
 547                 return 0;
 548         }
 549
 550         argv = realloc_argv(&array_size, argv);
 551         if (!argv)
 552                 return -ENOMEM;
 553
 554         while (1) {
 555                 /* Skip whitespace */
 556                 start = skip_spaces(end);
 557
 558                 if (!*start)
 559                         break;  /* success, we hit the end */
 560
 561                 /* 'out' is used to remove any back-quotes */
 562                 end = out = start;
 563                 while (*end) {
 564                         /* Everything apart from '\0' can be quoted */
 565                         if (*end == '\\' && *(end + 1)) {
 566                                 *out++ = *(end + 1);
 567                                 end += 2;
 568                                 continue;
 569                         }
 570
 571                         if (isspace(*end))
 572                                 break;  /* end of token */
 573
 574                         *out++ = *end++;
 575                 }
 576
 577                 /* have we already filled the array ? */
 578                 if ((*argc + 1) > array_size) {
 579                         argv = realloc_argv(&array_size, argv);
 580                         if (!argv)
 581                                 return -ENOMEM;
 582                 }
 583
 584                 /* we know this is whitespace */
 585                 if (*end)
 586                         end++;
 587
 588                 /* terminate the string and put it in the array */
 589                 *out = '\0';
 590                 argv[*argc] = start;
 591                 (*argc)++;
 592         }
 593
 594         *argvp = argv;
 595         return 0;
 596 }
 597
 598 static void dm_set_stacking_limits(struct queue_limits *limits)
 599 {
 600         blk_set_stacking_limits(limits);
 601         limits->features |= BLK_FEAT_IO_STAT | BLK_FEAT_NOWAIT | BLK_FEAT_POLL;
 602 }
 603
 604 /*
 605  * Impose necessary and sufficient conditions on a devices's table such
 606  * that any incoming bio which respects its logical_block_size can be
 607  * processed successfully.  If it falls across the boundary between
 608  * two or more targets, the size of each piece it gets split into must
 609  * be compatible with the logical_block_size of the target processing it.
 610  */
 611 static int validate_hardware_logical_block_alignment(struct dm_table *t,
 612                                                      struct queue_limits *limits)
 613 {
 614         /*
 615          * This function uses arithmetic modulo the logical_block_size
 616          * (in units of 512-byte sectors).
 617          */
 618         unsigned short device_logical_block_size_sects =
 619                 limits->logical_block_size >> SECTOR_SHIFT;
 620
 621         /*
 622          * Offset of the start of the next table entry, mod logical_block_size.
 623          */
 624         unsigned short next_target_start = 0;
 625
 626         /*
 627          * Given an aligned bio that extends beyond the end of a
 628          * target, how many sectors must the next target handle?
 629          */
 630         unsigned short remaining = 0;
 631
 632         struct dm_target *ti;
 633         struct queue_limits ti_limits;
 634         unsigned int i;
 635
 636         /*
 637          * Check each entry in the table in turn.
 638          */
 639         for (i = 0; i < t->num_targets; i++) {
 640                 ti = dm_table_get_target(t, i);
 641
 642                 dm_set_stacking_limits(&ti_limits);
 643
 644                 /* combine all target devices' limits */
 645                 if (ti->type->iterate_devices)
 646                         ti->type->iterate_devices(ti, dm_set_device_limits,
 647                                                   &ti_limits);
 648
 649                 /*
 650                  * If the remaining sectors fall entirely within this
 651                  * table entry are they compatible with its logical_block_size?
 652                  */
 653                 if (remaining < ti->len &&
 654                     remaining & ((ti_limits.logical_block_size >>
 655                                   SECTOR_SHIFT) - 1))
 656                         break;  /* Error */
 657
 658                 next_target_start =
 659                     (unsigned short) ((next_target_start + ti->len) &
 660                                       (device_logical_block_size_sects - 1));
 661                 remaining = next_target_start ?
 662                     device_logical_block_size_sects - next_target_start : 0;
 663         }
 664
 665         if (remaining) {
 666                 DMERR("%s: table line %u (start sect %llu len %llu) "
 667                       "not aligned to h/w logical block size %u",
 668                       dm_device_name(t->md), i,
 669                       (unsigned long long) ti->begin,
 670                       (unsigned long long) ti->len,
 671                       limits->logical_block_size);
 672                 return -EINVAL;
 673         }
 674
 675         return 0;
 676 }
 677
 678 int dm_table_add_target(struct dm_table *t, const char *type,
 679                         sector_t start, sector_t len, char *params)
 680 {
 681         int r = -EINVAL, argc;
 682         char **argv;
 683         struct dm_target *ti;
 684
 685         if (t->singleton) {
 686                 DMERR("%s: target type %s must appear alone in table",
 687                       dm_device_name(t->md), t->targets->type->name);
 688                 return -EINVAL;
 689         }
 690
 691         BUG_ON(t->num_targets >= t->num_allocated);
 692
 693         ti = t->targets + t->num_targets;
 694         memset(ti, 0, sizeof(*ti));
 695
 696         if (!len) {
 697                 DMERR("%s: zero-length target", dm_device_name(t->md));
 698                 return -EINVAL;
 699         }
 700
 701         ti->type = dm_get_target_type(type);
 702         if (!ti->type) {
 703                 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
 704                 return -EINVAL;
 705         }
 706
 707         if (dm_target_needs_singleton(ti->type)) {
 708                 if (t->num_targets) {
 709                         ti->error = "singleton target type must appear alone in table";
 710                         goto bad;
 711                 }
 712                 t->singleton = true;
 713         }
 714
 715         if (dm_target_always_writeable(ti->type) &&
 716             !(t->mode & BLK_OPEN_WRITE)) {
 717                 ti->error = "target type may not be included in a read-only table";
 718                 goto bad;
 719         }
 720
 721         if (t->immutable_target_type) {
 722                 if (t->immutable_target_type != ti->type) {
 723                         ti->error = "immutable target type cannot be mixed with other target types";
 724                         goto bad;
 725                 }
 726         } else if (dm_target_is_immutable(ti->type)) {
 727                 if (t->num_targets) {
 728                         ti->error = "immutable target type cannot be mixed with other target types";
 729                         goto bad;
 730                 }
 731                 t->immutable_target_type = ti->type;
 732         }
 733
 734         ti->table = t;
 735         ti->begin = start;
 736         ti->len = len;
 737         ti->error = "Unknown error";
 738
 739         /*
 740          * Does this target adjoin the previous one ?
 741          */
 742         if (!adjoin(t, ti)) {
 743                 ti->error = "Gap in table";
 744                 goto bad;
 745         }
 746
 747         r = dm_split_args(&argc, &argv, params);
 748         if (r) {
 749                 ti->error = "couldn't split parameters";
 750                 goto bad;
 751         }
 752
 753         r = ti->type->ctr(ti, argc, argv);
 754         kfree(argv);
 755         if (r)
 756                 goto bad;
 757
 758         t->highs[t->num_targets++] = ti->begin + ti->len - 1;
 759
 760         if (!ti->num_discard_bios && ti->discards_supported)
 761                 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
 762                        dm_device_name(t->md), type);
 763
 764         if (ti->limit_swap_bios && !static_key_enabled(&swap_bios_enabled.key))
 765                 static_branch_enable(&swap_bios_enabled);
 766
 767         if (!ti->flush_bypasses_map)
 768                 t->flush_bypasses_map = false;
 769
 770         return 0;
 771
 772  bad:
 773         DMERR("%s: %s: %s (%pe)", dm_device_name(t->md), type, ti->error, ERR_PTR(r));
 774         dm_put_target_type(ti->type);
 775         return r;
 776 }
 777
 778 /*
 779  * Target argument parsing helpers.
 780  */
 781 static int validate_next_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 782                              unsigned int *value, char **error, unsigned int grouped)
 783 {
 784         const char *arg_str = dm_shift_arg(arg_set);
 785         char dummy;
 786
 787         if (!arg_str ||
 788             (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
 789             (*value < arg->min) ||
 790             (*value > arg->max) ||
 791             (grouped && arg_set->argc < *value)) {
 792                 *error = arg->error;
 793                 return -EINVAL;
 794         }
 795
 796         return 0;
 797 }
 798
 799 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 800                 unsigned int *value, char **error)
 801 {
 802         return validate_next_arg(arg, arg_set, value, error, 0);
 803 }
 804 EXPORT_SYMBOL(dm_read_arg);
 805
 806 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 807                       unsigned int *value, char **error)
 808 {
 809         return validate_next_arg(arg, arg_set, value, error, 1);
 810 }
 811 EXPORT_SYMBOL(dm_read_arg_group);
 812
 813 const char *dm_shift_arg(struct dm_arg_set *as)
 814 {
 815         char *r;
 816
 817         if (as->argc) {
 818                 as->argc--;
 819                 r = *as->argv;
 820                 as->argv++;
 821                 return r;
 822         }
 823
 824         return NULL;
 825 }
 826 EXPORT_SYMBOL(dm_shift_arg);
 827
 828 void dm_consume_args(struct dm_arg_set *as, unsigned int num_args)
 829 {
 830         BUG_ON(as->argc < num_args);
 831         as->argc -= num_args;
 832         as->argv += num_args;
 833 }
 834 EXPORT_SYMBOL(dm_consume_args);
 835
 836 static bool __table_type_bio_based(enum dm_queue_mode table_type)
 837 {
 838         return (table_type == DM_TYPE_BIO_BASED ||
 839                 table_type == DM_TYPE_DAX_BIO_BASED);
 840 }
 841
 842 static bool __table_type_request_based(enum dm_queue_mode table_type)
 843 {
 844         return table_type == DM_TYPE_REQUEST_BASED;
 845 }
 846
 847 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
 848 {
 849         t->type = type;
 850 }
 851 EXPORT_SYMBOL_GPL(dm_table_set_type);
 852
 853 /* validate the dax capability of the target device span */
 854 static int device_not_dax_capable(struct dm_target *ti, struct dm_dev *dev,
 855                         sector_t start, sector_t len, void *data)
 856 {
 857         if (dev->dax_dev)
 858                 return false;
 859
 860         DMDEBUG("%pg: error: dax unsupported by block device", dev->bdev);
 861         return true;
 862 }
 863
 864 /* Check devices support synchronous DAX */
 865 static int device_not_dax_synchronous_capable(struct dm_target *ti, struct dm_dev *dev,
 866                                               sector_t start, sector_t len, void *data)
 867 {
 868         return !dev->dax_dev || !dax_synchronous(dev->dax_dev);
 869 }
 870
 871 static bool dm_table_supports_dax(struct dm_table *t,
 872                                   iterate_devices_callout_fn iterate_fn)
 873 {
 874         /* Ensure that all targets support DAX. */
 875         for (unsigned int i = 0; i < t->num_targets; i++) {
 876                 struct dm_target *ti = dm_table_get_target(t, i);
 877
 878                 if (!ti->type->direct_access)
 879                         return false;
 880
 881                 if (dm_target_is_wildcard(ti->type) ||
 882                     !ti->type->iterate_devices ||
 883                     ti->type->iterate_devices(ti, iterate_fn, NULL))
 884                         return false;
 885         }
 886
 887         return true;
 888 }
 889
 890 static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev,
 891                                   sector_t start, sector_t len, void *data)
 892 {
 893         struct block_device *bdev = dev->bdev;
 894         struct request_queue *q = bdev_get_queue(bdev);
 895
 896         /* request-based cannot stack on partitions! */
 897         if (bdev_is_partition(bdev))
 898                 return false;
 899
 900         return queue_is_mq(q);
 901 }
 902
 903 static int dm_table_determine_type(struct dm_table *t)
 904 {
 905         unsigned int bio_based = 0, request_based = 0, hybrid = 0;
 906         struct dm_target *ti;
 907         struct list_head *devices = dm_table_get_devices(t);
 908         enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
 909
 910         if (t->type != DM_TYPE_NONE) {
 911                 /* target already set the table's type */
 912                 if (t->type == DM_TYPE_BIO_BASED) {
 913                         /* possibly upgrade to a variant of bio-based */
 914                         goto verify_bio_based;
 915                 }
 916                 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
 917                 goto verify_rq_based;
 918         }
 919
 920         for (unsigned int i = 0; i < t->num_targets; i++) {
 921                 ti = dm_table_get_target(t, i);
 922                 if (dm_target_hybrid(ti))
 923                         hybrid = 1;
 924                 else if (dm_target_request_based(ti))
 925                         request_based = 1;
 926                 else
 927                         bio_based = 1;
 928
 929                 if (bio_based && request_based) {
 930                         DMERR("Inconsistent table: different target types can't be mixed up");
 931                         return -EINVAL;
 932                 }
 933         }
 934
 935         if (hybrid && !bio_based && !request_based) {
 936                 /*
 937                  * The targets can work either way.
 938                  * Determine the type from the live device.
 939                  * Default to bio-based if device is new.
 940                  */
 941                 if (__table_type_request_based(live_md_type))
 942                         request_based = 1;
 943                 else
 944                         bio_based = 1;
 945         }
 946
 947         if (bio_based) {
 948 verify_bio_based:
 949                 /* We must use this table as bio-based */
 950                 t->type = DM_TYPE_BIO_BASED;
 951                 if (dm_table_supports_dax(t, device_not_dax_capable) ||
 952                     (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
 953                         t->type = DM_TYPE_DAX_BIO_BASED;
 954                 }
 955                 return 0;
 956         }
 957
 958         BUG_ON(!request_based); /* No targets in this table */
 959
 960         t->type = DM_TYPE_REQUEST_BASED;
 961
 962 verify_rq_based:
 963         /*
 964          * Request-based dm supports only tables that have a single target now.
 965          * To support multiple targets, request splitting support is needed,
 966          * and that needs lots of changes in the block-layer.
 967          * (e.g. request completion process for partial completion.)
 968          */
 969         if (t->num_targets > 1) {
 970                 DMERR("request-based DM doesn't support multiple targets");
 971                 return -EINVAL;
 972         }
 973
 974         if (list_empty(devices)) {
 975                 int srcu_idx;
 976                 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
 977
 978                 /* inherit live table's type */
 979                 if (live_table)
 980                         t->type = live_table->type;
 981                 dm_put_live_table(t->md, srcu_idx);
 982                 return 0;
 983         }
 984
 985         ti = dm_table_get_immutable_target(t);
 986         if (!ti) {
 987                 DMERR("table load rejected: immutable target is required");
 988                 return -EINVAL;
 989         } else if (ti->max_io_len) {
 990                 DMERR("table load rejected: immutable target that splits IO is not supported");
 991                 return -EINVAL;
 992         }
 993
 994         /* Non-request-stackable devices can't be used for request-based dm */
 995         if (!ti->type->iterate_devices ||
 996             !ti->type->iterate_devices(ti, device_is_rq_stackable, NULL)) {
 997                 DMERR("table load rejected: including non-request-stackable devices");
 998                 return -EINVAL;
 999         }
1000
1001         return 0;
1002 }
1003
1004 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
1005 {
1006         return t->type;
1007 }
1008
1009 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
1010 {
1011         return t->immutable_target_type;
1012 }
1013
1014 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
1015 {
1016         /* Immutable target is implicitly a singleton */
1017         if (t->num_targets > 1 ||
1018             !dm_target_is_immutable(t->targets[0].type))
1019                 return NULL;
1020
1021         return t->targets;
1022 }
1023
1024 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1025 {
1026         for (unsigned int i = 0; i < t->num_targets; i++) {
1027                 struct dm_target *ti = dm_table_get_target(t, i);
1028
1029                 if (dm_target_is_wildcard(ti->type))
1030                         return ti;
1031         }
1032
1033         return NULL;
1034 }
1035
1036 bool dm_table_bio_based(struct dm_table *t)
1037 {
1038         return __table_type_bio_based(dm_table_get_type(t));
1039 }
1040
1041 bool dm_table_request_based(struct dm_table *t)
1042 {
1043         return __table_type_request_based(dm_table_get_type(t));
1044 }
1045
1046 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1047 {
1048         enum dm_queue_mode type = dm_table_get_type(t);
1049         unsigned int per_io_data_size = 0, front_pad, io_front_pad;
1050         unsigned int min_pool_size = 0, pool_size;
1051         struct dm_md_mempools *pools;
1052         unsigned int bioset_flags = 0;
1053         bool mempool_needs_integrity = t->integrity_supported;
1054
1055         if (unlikely(type == DM_TYPE_NONE)) {
1056                 DMERR("no table type is set, can't allocate mempools");
1057                 return -EINVAL;
1058         }
1059
1060         pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
1061         if (!pools)
1062                 return -ENOMEM;
1063
1064         if (type == DM_TYPE_REQUEST_BASED) {
1065                 pool_size = dm_get_reserved_rq_based_ios();
1066                 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
1067                 goto init_bs;
1068         }
1069
1070         if (md->queue->limits.features & BLK_FEAT_POLL)
1071                 bioset_flags |= BIOSET_PERCPU_CACHE;
1072
1073         for (unsigned int i = 0; i < t->num_targets; i++) {
1074                 struct dm_target *ti = dm_table_get_target(t, i);
1075
1076                 per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1077                 min_pool_size = max(min_pool_size, ti->num_flush_bios);
1078
1079                 mempool_needs_integrity |= ti->mempool_needs_integrity;
1080         }
1081         pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
1082         front_pad = roundup(per_io_data_size,
1083                 __alignof__(struct dm_target_io)) + DM_TARGET_IO_BIO_OFFSET;
1084
1085         io_front_pad = roundup(per_io_data_size,
1086                 __alignof__(struct dm_io)) + DM_IO_BIO_OFFSET;
1087         if (bioset_init(&pools->io_bs, pool_size, io_front_pad, bioset_flags))
1088                 goto out_free_pools;
1089         if (mempool_needs_integrity &&
1090             bioset_integrity_create(&pools->io_bs, pool_size))
1091                 goto out_free_pools;
1092 init_bs:
1093         if (bioset_init(&pools->bs, pool_size, front_pad, 0))
1094                 goto out_free_pools;
1095         if (mempool_needs_integrity &&
1096             bioset_integrity_create(&pools->bs, pool_size))
1097                 goto out_free_pools;
1098
1099         t->mempools = pools;
1100         return 0;
1101
1102 out_free_pools:
1103         dm_free_md_mempools(pools);
1104         return -ENOMEM;
1105 }
1106
1107 static int setup_indexes(struct dm_table *t)
1108 {
1109         int i;
1110         unsigned int total = 0;
1111         sector_t *indexes;
1112
1113         /* allocate the space for *all* the indexes */
1114         for (i = t->depth - 2; i >= 0; i--) {
1115                 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1116                 total += t->counts[i];
1117         }
1118
1119         indexes = kvcalloc(total, NODE_SIZE, GFP_KERNEL);
1120         if (!indexes)
1121                 return -ENOMEM;
1122
1123         /* set up internal nodes, bottom-up */
1124         for (i = t->depth - 2; i >= 0; i--) {
1125                 t->index[i] = indexes;
1126                 indexes += (KEYS_PER_NODE * t->counts[i]);
1127                 setup_btree_index(i, t);
1128         }
1129
1130         return 0;
1131 }
1132
1133 /*
1134  * Builds the btree to index the map.
1135  */
1136 static int dm_table_build_index(struct dm_table *t)
1137 {
1138         int r = 0;
1139         unsigned int leaf_nodes;
1140
1141         /* how many indexes will the btree have ? */
1142         leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1143         t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1144
1145         /* leaf layer has already been set up */
1146         t->counts[t->depth - 1] = leaf_nodes;
1147         t->index[t->depth - 1] = t->highs;
1148
1149         if (t->depth >= 2)
1150                 r = setup_indexes(t);
1151
1152         return r;
1153 }
1154
1155 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
1156
1157 struct dm_crypto_profile {
1158         struct blk_crypto_profile profile;
1159         struct mapped_device *md;
1160 };
1161
1162 static int dm_keyslot_evict_callback(struct dm_target *ti, struct dm_dev *dev,
1163                                      sector_t start, sector_t len, void *data)
1164 {
1165         const struct blk_crypto_key *key = data;
1166
1167         blk_crypto_evict_key(dev->bdev, key);
1168         return 0;
1169 }
1170
1171 /*
1172  * When an inline encryption key is evicted from a device-mapper device, evict
1173  * it from all the underlying devices.
1174  */
1175 static int dm_keyslot_evict(struct blk_crypto_profile *profile,
1176                             const struct blk_crypto_key *key, unsigned int slot)
1177 {
1178         struct mapped_device *md =
1179                 container_of(profile, struct dm_crypto_profile, profile)->md;
1180         struct dm_table *t;
1181         int srcu_idx;
1182
1183         t = dm_get_live_table(md, &srcu_idx);
1184         if (!t)
1185                 return 0;
1186
1187         for (unsigned int i = 0; i < t->num_targets; i++) {
1188                 struct dm_target *ti = dm_table_get_target(t, i);
1189
1190                 if (!ti->type->iterate_devices)
1191                         continue;
1192                 ti->type->iterate_devices(ti, dm_keyslot_evict_callback,
1193                                           (void *)key);
1194         }
1195
1196         dm_put_live_table(md, srcu_idx);
1197         return 0;
1198 }
1199
1200 static int
1201 device_intersect_crypto_capabilities(struct dm_target *ti, struct dm_dev *dev,
1202                                      sector_t start, sector_t len, void *data)
1203 {
1204         struct blk_crypto_profile *parent = data;
1205         struct blk_crypto_profile *child =
1206                 bdev_get_queue(dev->bdev)->crypto_profile;
1207
1208         blk_crypto_intersect_capabilities(parent, child);
1209         return 0;
1210 }
1211
1212 void dm_destroy_crypto_profile(struct blk_crypto_profile *profile)
1213 {
1214         struct dm_crypto_profile *dmcp = container_of(profile,
1215                                                       struct dm_crypto_profile,
1216                                                       profile);
1217
1218         if (!profile)
1219                 return;
1220
1221         blk_crypto_profile_destroy(profile);
1222         kfree(dmcp);
1223 }
1224
1225 static void dm_table_destroy_crypto_profile(struct dm_table *t)
1226 {
1227         dm_destroy_crypto_profile(t->crypto_profile);
1228         t->crypto_profile = NULL;
1229 }
1230
1231 /*
1232  * Constructs and initializes t->crypto_profile with a crypto profile that
1233  * represents the common set of crypto capabilities of the devices described by
1234  * the dm_table.  However, if the constructed crypto profile doesn't support all
1235  * crypto capabilities that are supported by the current mapped_device, it
1236  * returns an error instead, since we don't support removing crypto capabilities
1237  * on table changes.  Finally, if the constructed crypto profile is "empty" (has
1238  * no crypto capabilities at all), it just sets t->crypto_profile to NULL.
1239  */
1240 static int dm_table_construct_crypto_profile(struct dm_table *t)
1241 {
1242         struct dm_crypto_profile *dmcp;
1243         struct blk_crypto_profile *profile;
1244         unsigned int i;
1245         bool empty_profile = true;
1246
1247         dmcp = kmalloc(sizeof(*dmcp), GFP_KERNEL);
1248         if (!dmcp)
1249                 return -ENOMEM;
1250         dmcp->md = t->md;
1251
1252         profile = &dmcp->profile;
1253         blk_crypto_profile_init(profile, 0);
1254         profile->ll_ops.keyslot_evict = dm_keyslot_evict;
1255         profile->max_dun_bytes_supported = UINT_MAX;
1256         memset(profile->modes_supported, 0xFF,
1257                sizeof(profile->modes_supported));
1258
1259         for (i = 0; i < t->num_targets; i++) {
1260                 struct dm_target *ti = dm_table_get_target(t, i);
1261
1262                 if (!dm_target_passes_crypto(ti->type)) {
1263                         blk_crypto_intersect_capabilities(profile, NULL);
1264                         break;
1265                 }
1266                 if (!ti->type->iterate_devices)
1267                         continue;
1268                 ti->type->iterate_devices(ti,
1269                                           device_intersect_crypto_capabilities,
1270                                           profile);
1271         }
1272
1273         if (t->md->queue &&
1274             !blk_crypto_has_capabilities(profile,
1275                                          t->md->queue->crypto_profile)) {
1276                 DMERR("Inline encryption capabilities of new DM table were more restrictive than the old table's. This is not supported!");
1277                 dm_destroy_crypto_profile(profile);
1278                 return -EINVAL;
1279         }
1280
1281         /*
1282          * If the new profile doesn't actually support any crypto capabilities,
1283          * we may as well represent it with a NULL profile.
1284          */
1285         for (i = 0; i < ARRAY_SIZE(profile->modes_supported); i++) {
1286                 if (profile->modes_supported[i]) {
1287                         empty_profile = false;
1288                         break;
1289                 }
1290         }
1291
1292         if (empty_profile) {
1293                 dm_destroy_crypto_profile(profile);
1294                 profile = NULL;
1295         }
1296
1297         /*
1298          * t->crypto_profile is only set temporarily while the table is being
1299          * set up, and it gets set to NULL after the profile has been
1300          * transferred to the request_queue.
1301          */
1302         t->crypto_profile = profile;
1303
1304         return 0;
1305 }
1306
1307 static void dm_update_crypto_profile(struct request_queue *q,
1308                                      struct dm_table *t)
1309 {
1310         if (!t->crypto_profile)
1311                 return;
1312
1313         /* Make the crypto profile less restrictive. */
1314         if (!q->crypto_profile) {
1315                 blk_crypto_register(t->crypto_profile, q);
1316         } else {
1317                 blk_crypto_update_capabilities(q->crypto_profile,
1318                                                t->crypto_profile);
1319                 dm_destroy_crypto_profile(t->crypto_profile);
1320         }
1321         t->crypto_profile = NULL;
1322 }
1323
1324 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1325
1326 static int dm_table_construct_crypto_profile(struct dm_table *t)
1327 {
1328         return 0;
1329 }
1330
1331 void dm_destroy_crypto_profile(struct blk_crypto_profile *profile)
1332 {
1333 }
1334
1335 static void dm_table_destroy_crypto_profile(struct dm_table *t)
1336 {
1337 }
1338
1339 static void dm_update_crypto_profile(struct request_queue *q,
1340                                      struct dm_table *t)
1341 {
1342 }
1343
1344 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1345
1346 /*
1347  * Prepares the table for use by building the indices,
1348  * setting the type, and allocating mempools.
1349  */
1350 int dm_table_complete(struct dm_table *t)
1351 {
1352         int r;
1353
1354         r = dm_table_determine_type(t);
1355         if (r) {
1356                 DMERR("unable to determine table type");
1357                 return r;
1358         }
1359
1360         r = dm_table_build_index(t);
1361         if (r) {
1362                 DMERR("unable to build btrees");
1363                 return r;
1364         }
1365
1366         r = dm_table_construct_crypto_profile(t);
1367         if (r) {
1368                 DMERR("could not construct crypto profile.");
1369                 return r;
1370         }
1371
1372         r = dm_table_alloc_md_mempools(t, t->md);
1373         if (r)
1374                 DMERR("unable to allocate mempools");
1375
1376         return r;
1377 }
1378
1379 static DEFINE_MUTEX(_event_lock);
1380 void dm_table_event_callback(struct dm_table *t,
1381                              void (*fn)(void *), void *context)
1382 {
1383         mutex_lock(&_event_lock);
1384         t->event_fn = fn;
1385         t->event_context = context;
1386         mutex_unlock(&_event_lock);
1387 }
1388
1389 void dm_table_event(struct dm_table *t)
1390 {
1391         mutex_lock(&_event_lock);
1392         if (t->event_fn)
1393                 t->event_fn(t->event_context);
1394         mutex_unlock(&_event_lock);
1395 }
1396 EXPORT_SYMBOL(dm_table_event);
1397
1398 inline sector_t dm_table_get_size(struct dm_table *t)
1399 {
1400         return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1401 }
1402 EXPORT_SYMBOL(dm_table_get_size);
1403
1404 /*
1405  * Search the btree for the correct target.
1406  *
1407  * Caller should check returned pointer for NULL
1408  * to trap I/O beyond end of device.
1409  */
1410 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1411 {
1412         unsigned int l, n = 0, k = 0;
1413         sector_t *node;
1414
1415         if (unlikely(sector >= dm_table_get_size(t)))
1416                 return NULL;
1417
1418         for (l = 0; l < t->depth; l++) {
1419                 n = get_child(n, k);
1420                 node = get_node(t, l, n);
1421
1422                 for (k = 0; k < KEYS_PER_NODE; k++)
1423                         if (node[k] >= sector)
1424                                 break;
1425         }
1426
1427         return &t->targets[(KEYS_PER_NODE * n) + k];
1428 }
1429
1430 /*
1431  * type->iterate_devices() should be called when the sanity check needs to
1432  * iterate and check all underlying data devices. iterate_devices() will
1433  * iterate all underlying data devices until it encounters a non-zero return
1434  * code, returned by whether the input iterate_devices_callout_fn, or
1435  * iterate_devices() itself internally.
1436  *
1437  * For some target type (e.g. dm-stripe), one call of iterate_devices() may
1438  * iterate multiple underlying devices internally, in which case a non-zero
1439  * return code returned by iterate_devices_callout_fn will stop the iteration
1440  * in advance.
1441  *
1442  * Cases requiring _any_ underlying device supporting some kind of attribute,
1443  * should use the iteration structure like dm_table_any_dev_attr(), or call
1444  * it directly. @func should handle semantics of positive examples, e.g.
1445  * capable of something.
1446  *
1447  * Cases requiring _all_ underlying devices supporting some kind of attribute,
1448  * should use the iteration structure like dm_table_supports_nowait() or
1449  * dm_table_supports_discards(). Or introduce dm_table_all_devs_attr() that
1450  * uses an @anti_func that handle semantics of counter examples, e.g. not
1451  * capable of something. So: return !dm_table_any_dev_attr(t, anti_func, data);
1452  */
1453 static bool dm_table_any_dev_attr(struct dm_table *t,
1454                                   iterate_devices_callout_fn func, void *data)
1455 {
1456         for (unsigned int i = 0; i < t->num_targets; i++) {
1457                 struct dm_target *ti = dm_table_get_target(t, i);
1458
1459                 if (ti->type->iterate_devices &&
1460                     ti->type->iterate_devices(ti, func, data))
1461                         return true;
1462         }
1463
1464         return false;
1465 }
1466
1467 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1468                         sector_t start, sector_t len, void *data)
1469 {
1470         unsigned int *num_devices = data;
1471
1472         (*num_devices)++;
1473
1474         return 0;
1475 }
1476
1477 /*
1478  * Check whether a table has no data devices attached using each
1479  * target's iterate_devices method.
1480  * Returns false if the result is unknown because a target doesn't
1481  * support iterate_devices.
1482  */
1483 bool dm_table_has_no_data_devices(struct dm_table *t)
1484 {
1485         for (unsigned int i = 0; i < t->num_targets; i++) {
1486                 struct dm_target *ti = dm_table_get_target(t, i);
1487                 unsigned int num_devices = 0;
1488
1489                 if (!ti->type->iterate_devices)
1490                         return false;
1491
1492                 ti->type->iterate_devices(ti, count_device, &num_devices);
1493                 if (num_devices)
1494                         return false;
1495         }
1496
1497         return true;
1498 }
1499
1500 static int device_not_zoned(struct dm_target *ti, struct dm_dev *dev,
1501                             sector_t start, sector_t len, void *data)
1502 {
1503         bool *zoned = data;
1504
1505         return bdev_is_zoned(dev->bdev) != *zoned;
1506 }
1507
1508 static int device_is_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1509                                  sector_t start, sector_t len, void *data)
1510 {
1511         return bdev_is_zoned(dev->bdev);
1512 }
1513
1514 /*
1515  * Check the device zoned model based on the target feature flag. If the target
1516  * has the DM_TARGET_ZONED_HM feature flag set, host-managed zoned devices are
1517  * also accepted but all devices must have the same zoned model. If the target
1518  * has the DM_TARGET_MIXED_ZONED_MODEL feature set, the devices can have any
1519  * zoned model with all zoned devices having the same zone size.
1520  */
1521 static bool dm_table_supports_zoned(struct dm_table *t, bool zoned)
1522 {
1523         for (unsigned int i = 0; i < t->num_targets; i++) {
1524                 struct dm_target *ti = dm_table_get_target(t, i);
1525
1526                 /*
1527                  * For the wildcard target (dm-error), if we do not have a
1528                  * backing device, we must always return false. If we have a
1529                  * backing device, the result must depend on checking zoned
1530                  * model, like for any other target. So for this, check directly
1531                  * if the target backing device is zoned as we get "false" when
1532                  * dm-error was set without a backing device.
1533                  */
1534                 if (dm_target_is_wildcard(ti->type) &&
1535                     !ti->type->iterate_devices(ti, device_is_zoned_model, NULL))
1536                         return false;
1537
1538                 if (dm_target_supports_zoned_hm(ti->type)) {
1539                         if (!ti->type->iterate_devices ||
1540                             ti->type->iterate_devices(ti, device_not_zoned,
1541                                                       &zoned))
1542                                 return false;
1543                 } else if (!dm_target_supports_mixed_zoned_model(ti->type)) {
1544                         if (zoned)
1545                                 return false;
1546                 }
1547         }
1548
1549         return true;
1550 }
1551
1552 static int device_not_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1553                                            sector_t start, sector_t len, void *data)
1554 {
1555         unsigned int *zone_sectors = data;
1556
1557         if (!bdev_is_zoned(dev->bdev))
1558                 return 0;
1559         return bdev_zone_sectors(dev->bdev) != *zone_sectors;
1560 }
1561
1562 /*
1563  * Check consistency of zoned model and zone sectors across all targets. For
1564  * zone sectors, if the destination device is a zoned block device, it shall
1565  * have the specified zone_sectors.
1566  */
1567 static int validate_hardware_zoned(struct dm_table *t, bool zoned,
1568                                    unsigned int zone_sectors)
1569 {
1570         if (!zoned)
1571                 return 0;
1572
1573         if (!dm_table_supports_zoned(t, zoned)) {
1574                 DMERR("%s: zoned model is not consistent across all devices",
1575                       dm_device_name(t->md));
1576                 return -EINVAL;
1577         }
1578
1579         /* Check zone size validity and compatibility */
1580         if (!zone_sectors || !is_power_of_2(zone_sectors))
1581                 return -EINVAL;
1582
1583         if (dm_table_any_dev_attr(t, device_not_matches_zone_sectors, &zone_sectors)) {
1584                 DMERR("%s: zone sectors is not consistent across all zoned devices",
1585                       dm_device_name(t->md));
1586                 return -EINVAL;
1587         }
1588
1589         return 0;
1590 }
1591
1592 /*
1593  * Establish the new table's queue_limits and validate them.
1594  */
1595 int dm_calculate_queue_limits(struct dm_table *t,
1596                               struct queue_limits *limits)
1597 {
1598         struct queue_limits ti_limits;
1599         unsigned int zone_sectors = 0;
1600         bool zoned = false;
1601
1602         dm_set_stacking_limits(limits);
1603
1604         t->integrity_supported = true;
1605         for (unsigned int i = 0; i < t->num_targets; i++) {
1606                 struct dm_target *ti = dm_table_get_target(t, i);
1607
1608                 if (!dm_target_passes_integrity(ti->type))
1609                         t->integrity_supported = false;
1610         }
1611
1612         for (unsigned int i = 0; i < t->num_targets; i++) {
1613                 struct dm_target *ti = dm_table_get_target(t, i);
1614
1615                 dm_set_stacking_limits(&ti_limits);
1616
1617                 if (!ti->type->iterate_devices) {
1618                         /* Set I/O hints portion of queue limits */
1619                         if (ti->type->io_hints)
1620                                 ti->type->io_hints(ti, &ti_limits);
1621                         goto combine_limits;
1622                 }
1623
1624                 /*
1625                  * Combine queue limits of all the devices this target uses.
1626                  */
1627                 ti->type->iterate_devices(ti, dm_set_device_limits,
1628                                           &ti_limits);
1629
1630                 if (!zoned && (ti_limits.features & BLK_FEAT_ZONED)) {
1631                         /*
1632                          * After stacking all limits, validate all devices
1633                          * in table support this zoned model and zone sectors.
1634                          */
1635                         zoned = (ti_limits.features & BLK_FEAT_ZONED);
1636                         zone_sectors = ti_limits.chunk_sectors;
1637                 }
1638
1639                 /* Set I/O hints portion of queue limits */
1640                 if (ti->type->io_hints)
1641                         ti->type->io_hints(ti, &ti_limits);
1642
1643                 /*
1644                  * Check each device area is consistent with the target's
1645                  * overall queue limits.
1646                  */
1647                 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1648                                               &ti_limits))
1649                         return -EINVAL;
1650
1651 combine_limits:
1652                 /*
1653                  * Merge this target's queue limits into the overall limits
1654                  * for the table.
1655                  */
1656                 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1657                         DMWARN("%s: adding target device (start sect %llu len %llu) "
1658                                "caused an alignment inconsistency",
1659                                dm_device_name(t->md),
1660                                (unsigned long long) ti->begin,
1661                                (unsigned long long) ti->len);
1662
1663                 if (t->integrity_supported ||
1664                     dm_target_has_integrity(ti->type)) {
1665                         if (!queue_limits_stack_integrity(limits, &ti_limits)) {
1666                                 DMWARN("%s: adding target device (start sect %llu len %llu) "
1667                                        "disabled integrity support due to incompatibility",
1668                                        dm_device_name(t->md),
1669                                        (unsigned long long) ti->begin,
1670                                        (unsigned long long) ti->len);
1671                                 t->integrity_supported = false;
1672                         }
1673                 }
1674         }
1675
1676         /*
1677          * Verify that the zoned model and zone sectors, as determined before
1678          * any .io_hints override, are the same across all devices in the table.
1679          * - this is especially relevant if .io_hints is emulating a disk-managed
1680          *   zoned model on host-managed zoned block devices.
1681          * BUT...
1682          */
1683         if (limits->features & BLK_FEAT_ZONED) {
1684                 /*
1685                  * ...IF the above limits stacking determined a zoned model
1686                  * validate that all of the table's devices conform to it.
1687                  */
1688                 zoned = limits->features & BLK_FEAT_ZONED;
1689                 zone_sectors = limits->chunk_sectors;
1690         }
1691         if (validate_hardware_zoned(t, zoned, zone_sectors))
1692                 return -EINVAL;
1693
1694         return validate_hardware_logical_block_alignment(t, limits);
1695 }
1696
1697 /*
1698  * Check if a target requires flush support even if none of the underlying
1699  * devices need it (e.g. to persist target-specific metadata).
1700  */
1701 static bool dm_table_supports_flush(struct dm_table *t)
1702 {
1703         for (unsigned int i = 0; i < t->num_targets; i++) {
1704                 struct dm_target *ti = dm_table_get_target(t, i);
1705
1706                 if (ti->num_flush_bios && ti->flush_supported)
1707                         return true;
1708         }
1709
1710         return false;
1711 }
1712
1713 static int device_dax_write_cache_enabled(struct dm_target *ti,
1714                                           struct dm_dev *dev, sector_t start,
1715                                           sector_t len, void *data)
1716 {
1717         struct dax_device *dax_dev = dev->dax_dev;
1718
1719         if (!dax_dev)
1720                 return false;
1721
1722         if (dax_write_cache_enabled(dax_dev))
1723                 return true;
1724         return false;
1725 }
1726
1727 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1728                                            sector_t start, sector_t len, void *data)
1729 {
1730         struct request_queue *q = bdev_get_queue(dev->bdev);
1731
1732         return !q->limits.max_write_zeroes_sectors;
1733 }
1734
1735 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1736 {
1737         for (unsigned int i = 0; i < t->num_targets; i++) {
1738                 struct dm_target *ti = dm_table_get_target(t, i);
1739
1740                 if (!ti->num_write_zeroes_bios)
1741                         return false;
1742
1743                 if (!ti->type->iterate_devices ||
1744                     ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1745                         return false;
1746         }
1747
1748         return true;
1749 }
1750
1751 static bool dm_table_supports_nowait(struct dm_table *t)
1752 {
1753         for (unsigned int i = 0; i < t->num_targets; i++) {
1754                 struct dm_target *ti = dm_table_get_target(t, i);
1755
1756                 if (!dm_target_supports_nowait(ti->type))
1757                         return false;
1758         }
1759
1760         return true;
1761 }
1762
1763 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1764                                       sector_t start, sector_t len, void *data)
1765 {
1766         return !bdev_max_discard_sectors(dev->bdev);
1767 }
1768
1769 static bool dm_table_supports_discards(struct dm_table *t)
1770 {
1771         for (unsigned int i = 0; i < t->num_targets; i++) {
1772                 struct dm_target *ti = dm_table_get_target(t, i);
1773
1774                 if (!ti->num_discard_bios)
1775                         return false;
1776
1777                 /*
1778                  * Either the target provides discard support (as implied by setting
1779                  * 'discards_supported') or it relies on _all_ data devices having
1780                  * discard support.
1781                  */
1782                 if (!ti->discards_supported &&
1783                     (!ti->type->iterate_devices ||
1784                      ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1785                         return false;
1786         }
1787
1788         return true;
1789 }
1790
1791 static int device_not_secure_erase_capable(struct dm_target *ti,
1792                                            struct dm_dev *dev, sector_t start,
1793                                            sector_t len, void *data)
1794 {
1795         return !bdev_max_secure_erase_sectors(dev->bdev);
1796 }
1797
1798 static bool dm_table_supports_secure_erase(struct dm_table *t)
1799 {
1800         for (unsigned int i = 0; i < t->num_targets; i++) {
1801                 struct dm_target *ti = dm_table_get_target(t, i);
1802
1803                 if (!ti->num_secure_erase_bios)
1804                         return false;
1805
1806                 if (!ti->type->iterate_devices ||
1807                     ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1808                         return false;
1809         }
1810
1811         return true;
1812 }
1813
1814 int dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1815                               struct queue_limits *limits)
1816 {
1817         int r;
1818
1819         if (!dm_table_supports_nowait(t))
1820                 limits->features &= ~BLK_FEAT_NOWAIT;
1821
1822         /*
1823          * The current polling impementation does not support request based
1824          * stacking.
1825          */
1826         if (!__table_type_bio_based(t->type))
1827                 limits->features &= ~BLK_FEAT_POLL;
1828
1829         if (!dm_table_supports_discards(t)) {
1830                 limits->max_hw_discard_sectors = 0;
1831                 limits->discard_granularity = 0;
1832                 limits->discard_alignment = 0;
1833         }
1834
1835         if (!dm_table_supports_write_zeroes(t))
1836                 limits->max_write_zeroes_sectors = 0;
1837
1838         if (!dm_table_supports_secure_erase(t))
1839                 limits->max_secure_erase_sectors = 0;
1840
1841         if (dm_table_supports_flush(t))
1842                 limits->features |= BLK_FEAT_WRITE_CACHE | BLK_FEAT_FUA;
1843
1844         if (dm_table_supports_dax(t, device_not_dax_capable)) {
1845                 limits->features |= BLK_FEAT_DAX;
1846                 if (dm_table_supports_dax(t, device_not_dax_synchronous_capable))
1847                         set_dax_synchronous(t->md->dax_dev);
1848         } else
1849                 limits->features &= ~BLK_FEAT_DAX;
1850
1851         if (dm_table_any_dev_attr(t, device_dax_write_cache_enabled, NULL))
1852                 dax_write_cache(t->md->dax_dev, true);
1853
1854         /* For a zoned table, setup the zone related queue attributes. */
1855         if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
1856             (limits->features & BLK_FEAT_ZONED)) {
1857                 r = dm_set_zones_restrictions(t, q, limits);
1858                 if (r)
1859                         return r;
1860         }
1861
1862         r = queue_limits_set(q, limits);
1863         if (r)
1864                 return r;
1865
1866         /*
1867          * Now that the limits are set, check the zones mapped by the table
1868          * and setup the resources for zone append emulation if necessary.
1869          */
1870         if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
1871             (limits->features & BLK_FEAT_ZONED)) {
1872                 r = dm_revalidate_zones(t, q);
1873                 if (r)
1874                         return r;
1875         }
1876
1877         dm_update_crypto_profile(q, t);
1878         return 0;
1879 }
1880
1881 struct list_head *dm_table_get_devices(struct dm_table *t)
1882 {
1883         return &t->devices;
1884 }
1885
1886 blk_mode_t dm_table_get_mode(struct dm_table *t)
1887 {
1888         return t->mode;
1889 }
1890 EXPORT_SYMBOL(dm_table_get_mode);
1891
1892 enum suspend_mode {
1893         PRESUSPEND,
1894         PRESUSPEND_UNDO,
1895         POSTSUSPEND,
1896 };
1897
1898 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1899 {
1900         lockdep_assert_held(&t->md->suspend_lock);
1901
1902         for (unsigned int i = 0; i < t->num_targets; i++) {
1903                 struct dm_target *ti = dm_table_get_target(t, i);
1904
1905                 switch (mode) {
1906                 case PRESUSPEND:
1907                         if (ti->type->presuspend)
1908                                 ti->type->presuspend(ti);
1909                         break;
1910                 case PRESUSPEND_UNDO:
1911                         if (ti->type->presuspend_undo)
1912                                 ti->type->presuspend_undo(ti);
1913                         break;
1914                 case POSTSUSPEND:
1915                         if (ti->type->postsuspend)
1916                                 ti->type->postsuspend(ti);
1917                         break;
1918                 }
1919         }
1920 }
1921
1922 void dm_table_presuspend_targets(struct dm_table *t)
1923 {
1924         if (!t)
1925                 return;
1926
1927         suspend_targets(t, PRESUSPEND);
1928 }
1929
1930 void dm_table_presuspend_undo_targets(struct dm_table *t)
1931 {
1932         if (!t)
1933                 return;
1934
1935         suspend_targets(t, PRESUSPEND_UNDO);
1936 }
1937
1938 void dm_table_postsuspend_targets(struct dm_table *t)
1939 {
1940         if (!t)
1941                 return;
1942
1943         suspend_targets(t, POSTSUSPEND);
1944 }
1945
1946 int dm_table_resume_targets(struct dm_table *t)
1947 {
1948         unsigned int i;
1949         int r = 0;
1950
1951         lockdep_assert_held(&t->md->suspend_lock);
1952
1953         for (i = 0; i < t->num_targets; i++) {
1954                 struct dm_target *ti = dm_table_get_target(t, i);
1955
1956                 if (!ti->type->preresume)
1957                         continue;
1958
1959                 r = ti->type->preresume(ti);
1960                 if (r) {
1961                         DMERR("%s: %s: preresume failed, error = %d",
1962                               dm_device_name(t->md), ti->type->name, r);
1963                         return r;
1964                 }
1965         }
1966
1967         for (i = 0; i < t->num_targets; i++) {
1968                 struct dm_target *ti = dm_table_get_target(t, i);
1969
1970                 if (ti->type->resume)
1971                         ti->type->resume(ti);
1972         }
1973
1974         return 0;
1975 }
1976
1977 struct mapped_device *dm_table_get_md(struct dm_table *t)
1978 {
1979         return t->md;
1980 }
1981 EXPORT_SYMBOL(dm_table_get_md);
1982
1983 const char *dm_table_device_name(struct dm_table *t)
1984 {
1985         return dm_device_name(t->md);
1986 }
1987 EXPORT_SYMBOL_GPL(dm_table_device_name);
1988
1989 void dm_table_run_md_queue_async(struct dm_table *t)
1990 {
1991         if (!dm_table_request_based(t))
1992                 return;
1993
1994         if (t->md->queue)
1995                 blk_mq_run_hw_queues(t->md->queue, true);
1996 }
1997 EXPORT_SYMBOL(dm_table_run_md_queue_async);
1998