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, fmode_t mode, 130 unsigned int num_targets, struct mapped_device *md) 131 { 132 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL); 133 134 if (!t) 135 return -ENOMEM; 136 137 INIT_LIST_HEAD(&t->devices); 138 139 if (!num_targets) 140 num_targets = KEYS_PER_NODE; 141 142 num_targets = dm_round_up(num_targets, KEYS_PER_NODE); 143 144 if (!num_targets) { 145 kfree(t); 146 return -ENOMEM; 147 } 148 149 if (alloc_targets(t, num_targets)) { 150 kfree(t); 151 return -ENOMEM; 152 } 153 154 t->type = DM_TYPE_NONE; 155 t->mode = mode; 156 t->md = md; 157 *result = t; 158 return 0; 159 } 160 161 static void free_devices(struct list_head *devices, struct mapped_device *md) 162 { 163 struct list_head *tmp, *next; 164 165 list_for_each_safe(tmp, next, devices) { 166 struct dm_dev_internal *dd = 167 list_entry(tmp, struct dm_dev_internal, list); 168 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s", 169 dm_device_name(md), dd->dm_dev->name); 170 dm_put_table_device(md, dd->dm_dev); 171 kfree(dd); 172 } 173 } 174 175 static void dm_table_destroy_crypto_profile(struct dm_table *t); 176 177 void dm_table_destroy(struct dm_table *t) 178 { 179 if (!t) 180 return; 181 182 /* free the indexes */ 183 if (t->depth >= 2) 184 kvfree(t->index[t->depth - 2]); 185 186 /* free the targets */ 187 for (unsigned int i = 0; i < t->num_targets; i++) { 188 struct dm_target *ti = dm_table_get_target(t, i); 189 190 if (ti->type->dtr) 191 ti->type->dtr(ti); 192 193 dm_put_target_type(ti->type); 194 } 195 196 kvfree(t->highs); 197 198 /* free the device list */ 199 free_devices(&t->devices, t->md); 200 201 dm_free_md_mempools(t->mempools); 202 203 dm_table_destroy_crypto_profile(t); 204 205 kfree(t); 206 } 207 208 /* 209 * See if we've already got a device in the list. 210 */ 211 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev) 212 { 213 struct dm_dev_internal *dd; 214 215 list_for_each_entry(dd, l, list) 216 if (dd->dm_dev->bdev->bd_dev == dev) 217 return dd; 218 219 return NULL; 220 } 221 222 /* 223 * If possible, this checks an area of a destination device is invalid. 224 */ 225 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev, 226 sector_t start, sector_t len, void *data) 227 { 228 struct queue_limits *limits = data; 229 struct block_device *bdev = dev->bdev; 230 sector_t dev_size = bdev_nr_sectors(bdev); 231 unsigned short logical_block_size_sectors = 232 limits->logical_block_size >> SECTOR_SHIFT; 233 234 if (!dev_size) 235 return 0; 236 237 if ((start >= dev_size) || (start + len > dev_size)) { 238 DMERR("%s: %pg too small for target: start=%llu, len=%llu, dev_size=%llu", 239 dm_device_name(ti->table->md), bdev, 240 (unsigned long long)start, 241 (unsigned long long)len, 242 (unsigned long long)dev_size); 243 return 1; 244 } 245 246 /* 247 * If the target is mapped to zoned block device(s), check 248 * that the zones are not partially mapped. 249 */ 250 if (bdev_is_zoned(bdev)) { 251 unsigned int zone_sectors = bdev_zone_sectors(bdev); 252 253 if (start & (zone_sectors - 1)) { 254 DMERR("%s: start=%llu not aligned to h/w zone size %u of %pg", 255 dm_device_name(ti->table->md), 256 (unsigned long long)start, 257 zone_sectors, bdev); 258 return 1; 259 } 260 261 /* 262 * Note: The last zone of a zoned block device may be smaller 263 * than other zones. So for a target mapping the end of a 264 * zoned block device with such a zone, len would not be zone 265 * aligned. We do not allow such last smaller zone to be part 266 * of the mapping here to ensure that mappings with multiple 267 * devices do not end up with a smaller zone in the middle of 268 * the sector range. 269 */ 270 if (len & (zone_sectors - 1)) { 271 DMERR("%s: len=%llu not aligned to h/w zone size %u of %pg", 272 dm_device_name(ti->table->md), 273 (unsigned long long)len, 274 zone_sectors, bdev); 275 return 1; 276 } 277 } 278 279 if (logical_block_size_sectors <= 1) 280 return 0; 281 282 if (start & (logical_block_size_sectors - 1)) { 283 DMERR("%s: start=%llu not aligned to h/w logical block size %u of %pg", 284 dm_device_name(ti->table->md), 285 (unsigned long long)start, 286 limits->logical_block_size, bdev); 287 return 1; 288 } 289 290 if (len & (logical_block_size_sectors - 1)) { 291 DMERR("%s: len=%llu not aligned to h/w logical block size %u of %pg", 292 dm_device_name(ti->table->md), 293 (unsigned long long)len, 294 limits->logical_block_size, bdev); 295 return 1; 296 } 297 298 return 0; 299 } 300 301 /* 302 * This upgrades the mode on an already open dm_dev, being 303 * careful to leave things as they were if we fail to reopen the 304 * device and not to touch the existing bdev field in case 305 * it is accessed concurrently. 306 */ 307 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode, 308 struct mapped_device *md) 309 { 310 int r; 311 struct dm_dev *old_dev, *new_dev; 312 313 old_dev = dd->dm_dev; 314 315 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev, 316 dd->dm_dev->mode | new_mode, &new_dev); 317 if (r) 318 return r; 319 320 dd->dm_dev = new_dev; 321 dm_put_table_device(md, old_dev); 322 323 return 0; 324 } 325 326 /* 327 * Add a device to the list, or just increment the usage count if 328 * it's already present. 329 * 330 * Note: the __ref annotation is because this function can call the __init 331 * marked early_lookup_bdev when called during early boot code from dm-init.c. 332 */ 333 int __ref dm_get_device(struct dm_target *ti, const char *path, fmode_t mode, 334 struct dm_dev **result) 335 { 336 int r; 337 dev_t dev; 338 unsigned int major, minor; 339 char dummy; 340 struct dm_dev_internal *dd; 341 struct dm_table *t = ti->table; 342 343 BUG_ON(!t); 344 345 if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) { 346 /* Extract the major/minor numbers */ 347 dev = MKDEV(major, minor); 348 if (MAJOR(dev) != major || MINOR(dev) != minor) 349 return -EOVERFLOW; 350 } else { 351 r = lookup_bdev(path, &dev); 352 #ifndef MODULE 353 if (r && system_state < SYSTEM_RUNNING) 354 r = early_lookup_bdev(path, &dev); 355 #endif 356 if (r) 357 return r; 358 } 359 if (dev == disk_devt(t->md->disk)) 360 return -EINVAL; 361 362 dd = find_device(&t->devices, dev); 363 if (!dd) { 364 dd = kmalloc(sizeof(*dd), GFP_KERNEL); 365 if (!dd) 366 return -ENOMEM; 367 368 r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev); 369 if (r) { 370 kfree(dd); 371 return r; 372 } 373 374 refcount_set(&dd->count, 1); 375 list_add(&dd->list, &t->devices); 376 goto out; 377 378 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) { 379 r = upgrade_mode(dd, mode, t->md); 380 if (r) 381 return r; 382 } 383 refcount_inc(&dd->count); 384 out: 385 *result = dd->dm_dev; 386 return 0; 387 } 388 EXPORT_SYMBOL(dm_get_device); 389 390 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev, 391 sector_t start, sector_t len, void *data) 392 { 393 struct queue_limits *limits = data; 394 struct block_device *bdev = dev->bdev; 395 struct request_queue *q = bdev_get_queue(bdev); 396 397 if (unlikely(!q)) { 398 DMWARN("%s: Cannot set limits for nonexistent device %pg", 399 dm_device_name(ti->table->md), bdev); 400 return 0; 401 } 402 403 if (blk_stack_limits(limits, &q->limits, 404 get_start_sect(bdev) + start) < 0) 405 DMWARN("%s: adding target device %pg caused an alignment inconsistency: " 406 "physical_block_size=%u, logical_block_size=%u, " 407 "alignment_offset=%u, start=%llu", 408 dm_device_name(ti->table->md), bdev, 409 q->limits.physical_block_size, 410 q->limits.logical_block_size, 411 q->limits.alignment_offset, 412 (unsigned long long) start << SECTOR_SHIFT); 413 return 0; 414 } 415 416 /* 417 * Decrement a device's use count and remove it if necessary. 418 */ 419 void dm_put_device(struct dm_target *ti, struct dm_dev *d) 420 { 421 int found = 0; 422 struct list_head *devices = &ti->table->devices; 423 struct dm_dev_internal *dd; 424 425 list_for_each_entry(dd, devices, list) { 426 if (dd->dm_dev == d) { 427 found = 1; 428 break; 429 } 430 } 431 if (!found) { 432 DMERR("%s: device %s not in table devices list", 433 dm_device_name(ti->table->md), d->name); 434 return; 435 } 436 if (refcount_dec_and_test(&dd->count)) { 437 dm_put_table_device(ti->table->md, d); 438 list_del(&dd->list); 439 kfree(dd); 440 } 441 } 442 EXPORT_SYMBOL(dm_put_device); 443 444 /* 445 * Checks to see if the target joins onto the end of the table. 446 */ 447 static int adjoin(struct dm_table *t, struct dm_target *ti) 448 { 449 struct dm_target *prev; 450 451 if (!t->num_targets) 452 return !ti->begin; 453 454 prev = &t->targets[t->num_targets - 1]; 455 return (ti->begin == (prev->begin + prev->len)); 456 } 457 458 /* 459 * Used to dynamically allocate the arg array. 460 * 461 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must 462 * process messages even if some device is suspended. These messages have a 463 * small fixed number of arguments. 464 * 465 * On the other hand, dm-switch needs to process bulk data using messages and 466 * excessive use of GFP_NOIO could cause trouble. 467 */ 468 static char **realloc_argv(unsigned int *size, char **old_argv) 469 { 470 char **argv; 471 unsigned int new_size; 472 gfp_t gfp; 473 474 if (*size) { 475 new_size = *size * 2; 476 gfp = GFP_KERNEL; 477 } else { 478 new_size = 8; 479 gfp = GFP_NOIO; 480 } 481 argv = kmalloc_array(new_size, sizeof(*argv), gfp); 482 if (argv && old_argv) { 483 memcpy(argv, old_argv, *size * sizeof(*argv)); 484 *size = new_size; 485 } 486 487 kfree(old_argv); 488 return argv; 489 } 490 491 /* 492 * Destructively splits up the argument list to pass to ctr. 493 */ 494 int dm_split_args(int *argc, char ***argvp, char *input) 495 { 496 char *start, *end = input, *out, **argv = NULL; 497 unsigned int array_size = 0; 498 499 *argc = 0; 500 501 if (!input) { 502 *argvp = NULL; 503 return 0; 504 } 505 506 argv = realloc_argv(&array_size, argv); 507 if (!argv) 508 return -ENOMEM; 509 510 while (1) { 511 /* Skip whitespace */ 512 start = skip_spaces(end); 513 514 if (!*start) 515 break; /* success, we hit the end */ 516 517 /* 'out' is used to remove any back-quotes */ 518 end = out = start; 519 while (*end) { 520 /* Everything apart from '\0' can be quoted */ 521 if (*end == '\\' && *(end + 1)) { 522 *out++ = *(end + 1); 523 end += 2; 524 continue; 525 } 526 527 if (isspace(*end)) 528 break; /* end of token */ 529 530 *out++ = *end++; 531 } 532 533 /* have we already filled the array ? */ 534 if ((*argc + 1) > array_size) { 535 argv = realloc_argv(&array_size, argv); 536 if (!argv) 537 return -ENOMEM; 538 } 539 540 /* we know this is whitespace */ 541 if (*end) 542 end++; 543 544 /* terminate the string and put it in the array */ 545 *out = '\0'; 546 argv[*argc] = start; 547 (*argc)++; 548 } 549 550 *argvp = argv; 551 return 0; 552 } 553 554 /* 555 * Impose necessary and sufficient conditions on a devices's table such 556 * that any incoming bio which respects its logical_block_size can be 557 * processed successfully. If it falls across the boundary between 558 * two or more targets, the size of each piece it gets split into must 559 * be compatible with the logical_block_size of the target processing it. 560 */ 561 static int validate_hardware_logical_block_alignment(struct dm_table *t, 562 struct queue_limits *limits) 563 { 564 /* 565 * This function uses arithmetic modulo the logical_block_size 566 * (in units of 512-byte sectors). 567 */ 568 unsigned short device_logical_block_size_sects = 569 limits->logical_block_size >> SECTOR_SHIFT; 570 571 /* 572 * Offset of the start of the next table entry, mod logical_block_size. 573 */ 574 unsigned short next_target_start = 0; 575 576 /* 577 * Given an aligned bio that extends beyond the end of a 578 * target, how many sectors must the next target handle? 579 */ 580 unsigned short remaining = 0; 581 582 struct dm_target *ti; 583 struct queue_limits ti_limits; 584 unsigned int i; 585 586 /* 587 * Check each entry in the table in turn. 588 */ 589 for (i = 0; i < t->num_targets; i++) { 590 ti = dm_table_get_target(t, i); 591 592 blk_set_stacking_limits(&ti_limits); 593 594 /* combine all target devices' limits */ 595 if (ti->type->iterate_devices) 596 ti->type->iterate_devices(ti, dm_set_device_limits, 597 &ti_limits); 598 599 /* 600 * If the remaining sectors fall entirely within this 601 * table entry are they compatible with its logical_block_size? 602 */ 603 if (remaining < ti->len && 604 remaining & ((ti_limits.logical_block_size >> 605 SECTOR_SHIFT) - 1)) 606 break; /* Error */ 607 608 next_target_start = 609 (unsigned short) ((next_target_start + ti->len) & 610 (device_logical_block_size_sects - 1)); 611 remaining = next_target_start ? 612 device_logical_block_size_sects - next_target_start : 0; 613 } 614 615 if (remaining) { 616 DMERR("%s: table line %u (start sect %llu len %llu) " 617 "not aligned to h/w logical block size %u", 618 dm_device_name(t->md), i, 619 (unsigned long long) ti->begin, 620 (unsigned long long) ti->len, 621 limits->logical_block_size); 622 return -EINVAL; 623 } 624 625 return 0; 626 } 627 628 int dm_table_add_target(struct dm_table *t, const char *type, 629 sector_t start, sector_t len, char *params) 630 { 631 int r = -EINVAL, argc; 632 char **argv; 633 struct dm_target *ti; 634 635 if (t->singleton) { 636 DMERR("%s: target type %s must appear alone in table", 637 dm_device_name(t->md), t->targets->type->name); 638 return -EINVAL; 639 } 640 641 BUG_ON(t->num_targets >= t->num_allocated); 642 643 ti = t->targets + t->num_targets; 644 memset(ti, 0, sizeof(*ti)); 645 646 if (!len) { 647 DMERR("%s: zero-length target", dm_device_name(t->md)); 648 return -EINVAL; 649 } 650 651 ti->type = dm_get_target_type(type); 652 if (!ti->type) { 653 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type); 654 return -EINVAL; 655 } 656 657 if (dm_target_needs_singleton(ti->type)) { 658 if (t->num_targets) { 659 ti->error = "singleton target type must appear alone in table"; 660 goto bad; 661 } 662 t->singleton = true; 663 } 664 665 if (dm_target_always_writeable(ti->type) && !(t->mode & FMODE_WRITE)) { 666 ti->error = "target type may not be included in a read-only table"; 667 goto bad; 668 } 669 670 if (t->immutable_target_type) { 671 if (t->immutable_target_type != ti->type) { 672 ti->error = "immutable target type cannot be mixed with other target types"; 673 goto bad; 674 } 675 } else if (dm_target_is_immutable(ti->type)) { 676 if (t->num_targets) { 677 ti->error = "immutable target type cannot be mixed with other target types"; 678 goto bad; 679 } 680 t->immutable_target_type = ti->type; 681 } 682 683 if (dm_target_has_integrity(ti->type)) 684 t->integrity_added = 1; 685 686 ti->table = t; 687 ti->begin = start; 688 ti->len = len; 689 ti->error = "Unknown error"; 690 691 /* 692 * Does this target adjoin the previous one ? 693 */ 694 if (!adjoin(t, ti)) { 695 ti->error = "Gap in table"; 696 goto bad; 697 } 698 699 r = dm_split_args(&argc, &argv, params); 700 if (r) { 701 ti->error = "couldn't split parameters"; 702 goto bad; 703 } 704 705 r = ti->type->ctr(ti, argc, argv); 706 kfree(argv); 707 if (r) 708 goto bad; 709 710 t->highs[t->num_targets++] = ti->begin + ti->len - 1; 711 712 if (!ti->num_discard_bios && ti->discards_supported) 713 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.", 714 dm_device_name(t->md), type); 715 716 if (ti->limit_swap_bios && !static_key_enabled(&swap_bios_enabled.key)) 717 static_branch_enable(&swap_bios_enabled); 718 719 return 0; 720 721 bad: 722 DMERR("%s: %s: %s (%pe)", dm_device_name(t->md), type, ti->error, ERR_PTR(r)); 723 dm_put_target_type(ti->type); 724 return r; 725 } 726 727 /* 728 * Target argument parsing helpers. 729 */ 730 static int validate_next_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set, 731 unsigned int *value, char **error, unsigned int grouped) 732 { 733 const char *arg_str = dm_shift_arg(arg_set); 734 char dummy; 735 736 if (!arg_str || 737 (sscanf(arg_str, "%u%c", value, &dummy) != 1) || 738 (*value < arg->min) || 739 (*value > arg->max) || 740 (grouped && arg_set->argc < *value)) { 741 *error = arg->error; 742 return -EINVAL; 743 } 744 745 return 0; 746 } 747 748 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set, 749 unsigned int *value, char **error) 750 { 751 return validate_next_arg(arg, arg_set, value, error, 0); 752 } 753 EXPORT_SYMBOL(dm_read_arg); 754 755 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set, 756 unsigned int *value, char **error) 757 { 758 return validate_next_arg(arg, arg_set, value, error, 1); 759 } 760 EXPORT_SYMBOL(dm_read_arg_group); 761 762 const char *dm_shift_arg(struct dm_arg_set *as) 763 { 764 char *r; 765 766 if (as->argc) { 767 as->argc--; 768 r = *as->argv; 769 as->argv++; 770 return r; 771 } 772 773 return NULL; 774 } 775 EXPORT_SYMBOL(dm_shift_arg); 776 777 void dm_consume_args(struct dm_arg_set *as, unsigned int num_args) 778 { 779 BUG_ON(as->argc < num_args); 780 as->argc -= num_args; 781 as->argv += num_args; 782 } 783 EXPORT_SYMBOL(dm_consume_args); 784 785 static bool __table_type_bio_based(enum dm_queue_mode table_type) 786 { 787 return (table_type == DM_TYPE_BIO_BASED || 788 table_type == DM_TYPE_DAX_BIO_BASED); 789 } 790 791 static bool __table_type_request_based(enum dm_queue_mode table_type) 792 { 793 return table_type == DM_TYPE_REQUEST_BASED; 794 } 795 796 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type) 797 { 798 t->type = type; 799 } 800 EXPORT_SYMBOL_GPL(dm_table_set_type); 801 802 /* validate the dax capability of the target device span */ 803 static int device_not_dax_capable(struct dm_target *ti, struct dm_dev *dev, 804 sector_t start, sector_t len, void *data) 805 { 806 if (dev->dax_dev) 807 return false; 808 809 DMDEBUG("%pg: error: dax unsupported by block device", dev->bdev); 810 return true; 811 } 812 813 /* Check devices support synchronous DAX */ 814 static int device_not_dax_synchronous_capable(struct dm_target *ti, struct dm_dev *dev, 815 sector_t start, sector_t len, void *data) 816 { 817 return !dev->dax_dev || !dax_synchronous(dev->dax_dev); 818 } 819 820 static bool dm_table_supports_dax(struct dm_table *t, 821 iterate_devices_callout_fn iterate_fn) 822 { 823 /* Ensure that all targets support DAX. */ 824 for (unsigned int i = 0; i < t->num_targets; i++) { 825 struct dm_target *ti = dm_table_get_target(t, i); 826 827 if (!ti->type->direct_access) 828 return false; 829 830 if (!ti->type->iterate_devices || 831 ti->type->iterate_devices(ti, iterate_fn, NULL)) 832 return false; 833 } 834 835 return true; 836 } 837 838 static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev, 839 sector_t start, sector_t len, void *data) 840 { 841 struct block_device *bdev = dev->bdev; 842 struct request_queue *q = bdev_get_queue(bdev); 843 844 /* request-based cannot stack on partitions! */ 845 if (bdev_is_partition(bdev)) 846 return false; 847 848 return queue_is_mq(q); 849 } 850 851 static int dm_table_determine_type(struct dm_table *t) 852 { 853 unsigned int bio_based = 0, request_based = 0, hybrid = 0; 854 struct dm_target *ti; 855 struct list_head *devices = dm_table_get_devices(t); 856 enum dm_queue_mode live_md_type = dm_get_md_type(t->md); 857 858 if (t->type != DM_TYPE_NONE) { 859 /* target already set the table's type */ 860 if (t->type == DM_TYPE_BIO_BASED) { 861 /* possibly upgrade to a variant of bio-based */ 862 goto verify_bio_based; 863 } 864 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED); 865 goto verify_rq_based; 866 } 867 868 for (unsigned int i = 0; i < t->num_targets; i++) { 869 ti = dm_table_get_target(t, i); 870 if (dm_target_hybrid(ti)) 871 hybrid = 1; 872 else if (dm_target_request_based(ti)) 873 request_based = 1; 874 else 875 bio_based = 1; 876 877 if (bio_based && request_based) { 878 DMERR("Inconsistent table: different target types can't be mixed up"); 879 return -EINVAL; 880 } 881 } 882 883 if (hybrid && !bio_based && !request_based) { 884 /* 885 * The targets can work either way. 886 * Determine the type from the live device. 887 * Default to bio-based if device is new. 888 */ 889 if (__table_type_request_based(live_md_type)) 890 request_based = 1; 891 else 892 bio_based = 1; 893 } 894 895 if (bio_based) { 896 verify_bio_based: 897 /* We must use this table as bio-based */ 898 t->type = DM_TYPE_BIO_BASED; 899 if (dm_table_supports_dax(t, device_not_dax_capable) || 900 (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) { 901 t->type = DM_TYPE_DAX_BIO_BASED; 902 } 903 return 0; 904 } 905 906 BUG_ON(!request_based); /* No targets in this table */ 907 908 t->type = DM_TYPE_REQUEST_BASED; 909 910 verify_rq_based: 911 /* 912 * Request-based dm supports only tables that have a single target now. 913 * To support multiple targets, request splitting support is needed, 914 * and that needs lots of changes in the block-layer. 915 * (e.g. request completion process for partial completion.) 916 */ 917 if (t->num_targets > 1) { 918 DMERR("request-based DM doesn't support multiple targets"); 919 return -EINVAL; 920 } 921 922 if (list_empty(devices)) { 923 int srcu_idx; 924 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx); 925 926 /* inherit live table's type */ 927 if (live_table) 928 t->type = live_table->type; 929 dm_put_live_table(t->md, srcu_idx); 930 return 0; 931 } 932 933 ti = dm_table_get_immutable_target(t); 934 if (!ti) { 935 DMERR("table load rejected: immutable target is required"); 936 return -EINVAL; 937 } else if (ti->max_io_len) { 938 DMERR("table load rejected: immutable target that splits IO is not supported"); 939 return -EINVAL; 940 } 941 942 /* Non-request-stackable devices can't be used for request-based dm */ 943 if (!ti->type->iterate_devices || 944 !ti->type->iterate_devices(ti, device_is_rq_stackable, NULL)) { 945 DMERR("table load rejected: including non-request-stackable devices"); 946 return -EINVAL; 947 } 948 949 return 0; 950 } 951 952 enum dm_queue_mode dm_table_get_type(struct dm_table *t) 953 { 954 return t->type; 955 } 956 957 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t) 958 { 959 return t->immutable_target_type; 960 } 961 962 struct dm_target *dm_table_get_immutable_target(struct dm_table *t) 963 { 964 /* Immutable target is implicitly a singleton */ 965 if (t->num_targets > 1 || 966 !dm_target_is_immutable(t->targets[0].type)) 967 return NULL; 968 969 return t->targets; 970 } 971 972 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t) 973 { 974 for (unsigned int i = 0; i < t->num_targets; i++) { 975 struct dm_target *ti = dm_table_get_target(t, i); 976 977 if (dm_target_is_wildcard(ti->type)) 978 return ti; 979 } 980 981 return NULL; 982 } 983 984 bool dm_table_bio_based(struct dm_table *t) 985 { 986 return __table_type_bio_based(dm_table_get_type(t)); 987 } 988 989 bool dm_table_request_based(struct dm_table *t) 990 { 991 return __table_type_request_based(dm_table_get_type(t)); 992 } 993 994 static bool dm_table_supports_poll(struct dm_table *t); 995 996 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md) 997 { 998 enum dm_queue_mode type = dm_table_get_type(t); 999 unsigned int per_io_data_size = 0, front_pad, io_front_pad; 1000 unsigned int min_pool_size = 0, pool_size; 1001 struct dm_md_mempools *pools; 1002 1003 if (unlikely(type == DM_TYPE_NONE)) { 1004 DMERR("no table type is set, can't allocate mempools"); 1005 return -EINVAL; 1006 } 1007 1008 pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id); 1009 if (!pools) 1010 return -ENOMEM; 1011 1012 if (type == DM_TYPE_REQUEST_BASED) { 1013 pool_size = dm_get_reserved_rq_based_ios(); 1014 front_pad = offsetof(struct dm_rq_clone_bio_info, clone); 1015 goto init_bs; 1016 } 1017 1018 for (unsigned int i = 0; i < t->num_targets; i++) { 1019 struct dm_target *ti = dm_table_get_target(t, i); 1020 1021 per_io_data_size = max(per_io_data_size, ti->per_io_data_size); 1022 min_pool_size = max(min_pool_size, ti->num_flush_bios); 1023 } 1024 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size); 1025 front_pad = roundup(per_io_data_size, 1026 __alignof__(struct dm_target_io)) + DM_TARGET_IO_BIO_OFFSET; 1027 1028 io_front_pad = roundup(per_io_data_size, 1029 __alignof__(struct dm_io)) + DM_IO_BIO_OFFSET; 1030 if (bioset_init(&pools->io_bs, pool_size, io_front_pad, 1031 dm_table_supports_poll(t) ? BIOSET_PERCPU_CACHE : 0)) 1032 goto out_free_pools; 1033 if (t->integrity_supported && 1034 bioset_integrity_create(&pools->io_bs, pool_size)) 1035 goto out_free_pools; 1036 init_bs: 1037 if (bioset_init(&pools->bs, pool_size, front_pad, 0)) 1038 goto out_free_pools; 1039 if (t->integrity_supported && 1040 bioset_integrity_create(&pools->bs, pool_size)) 1041 goto out_free_pools; 1042 1043 t->mempools = pools; 1044 return 0; 1045 1046 out_free_pools: 1047 dm_free_md_mempools(pools); 1048 return -ENOMEM; 1049 } 1050 1051 static int setup_indexes(struct dm_table *t) 1052 { 1053 int i; 1054 unsigned int total = 0; 1055 sector_t *indexes; 1056 1057 /* allocate the space for *all* the indexes */ 1058 for (i = t->depth - 2; i >= 0; i--) { 1059 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE); 1060 total += t->counts[i]; 1061 } 1062 1063 indexes = kvcalloc(total, NODE_SIZE, GFP_KERNEL); 1064 if (!indexes) 1065 return -ENOMEM; 1066 1067 /* set up internal nodes, bottom-up */ 1068 for (i = t->depth - 2; i >= 0; i--) { 1069 t->index[i] = indexes; 1070 indexes += (KEYS_PER_NODE * t->counts[i]); 1071 setup_btree_index(i, t); 1072 } 1073 1074 return 0; 1075 } 1076 1077 /* 1078 * Builds the btree to index the map. 1079 */ 1080 static int dm_table_build_index(struct dm_table *t) 1081 { 1082 int r = 0; 1083 unsigned int leaf_nodes; 1084 1085 /* how many indexes will the btree have ? */ 1086 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE); 1087 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE); 1088 1089 /* leaf layer has already been set up */ 1090 t->counts[t->depth - 1] = leaf_nodes; 1091 t->index[t->depth - 1] = t->highs; 1092 1093 if (t->depth >= 2) 1094 r = setup_indexes(t); 1095 1096 return r; 1097 } 1098 1099 static bool integrity_profile_exists(struct gendisk *disk) 1100 { 1101 return !!blk_get_integrity(disk); 1102 } 1103 1104 /* 1105 * Get a disk whose integrity profile reflects the table's profile. 1106 * Returns NULL if integrity support was inconsistent or unavailable. 1107 */ 1108 static struct gendisk *dm_table_get_integrity_disk(struct dm_table *t) 1109 { 1110 struct list_head *devices = dm_table_get_devices(t); 1111 struct dm_dev_internal *dd = NULL; 1112 struct gendisk *prev_disk = NULL, *template_disk = NULL; 1113 1114 for (unsigned int i = 0; i < t->num_targets; i++) { 1115 struct dm_target *ti = dm_table_get_target(t, i); 1116 1117 if (!dm_target_passes_integrity(ti->type)) 1118 goto no_integrity; 1119 } 1120 1121 list_for_each_entry(dd, devices, list) { 1122 template_disk = dd->dm_dev->bdev->bd_disk; 1123 if (!integrity_profile_exists(template_disk)) 1124 goto no_integrity; 1125 else if (prev_disk && 1126 blk_integrity_compare(prev_disk, template_disk) < 0) 1127 goto no_integrity; 1128 prev_disk = template_disk; 1129 } 1130 1131 return template_disk; 1132 1133 no_integrity: 1134 if (prev_disk) 1135 DMWARN("%s: integrity not set: %s and %s profile mismatch", 1136 dm_device_name(t->md), 1137 prev_disk->disk_name, 1138 template_disk->disk_name); 1139 return NULL; 1140 } 1141 1142 /* 1143 * Register the mapped device for blk_integrity support if the 1144 * underlying devices have an integrity profile. But all devices may 1145 * not have matching profiles (checking all devices isn't reliable 1146 * during table load because this table may use other DM device(s) which 1147 * must be resumed before they will have an initialized integity 1148 * profile). Consequently, stacked DM devices force a 2 stage integrity 1149 * profile validation: First pass during table load, final pass during 1150 * resume. 1151 */ 1152 static int dm_table_register_integrity(struct dm_table *t) 1153 { 1154 struct mapped_device *md = t->md; 1155 struct gendisk *template_disk = NULL; 1156 1157 /* If target handles integrity itself do not register it here. */ 1158 if (t->integrity_added) 1159 return 0; 1160 1161 template_disk = dm_table_get_integrity_disk(t); 1162 if (!template_disk) 1163 return 0; 1164 1165 if (!integrity_profile_exists(dm_disk(md))) { 1166 t->integrity_supported = true; 1167 /* 1168 * Register integrity profile during table load; we can do 1169 * this because the final profile must match during resume. 1170 */ 1171 blk_integrity_register(dm_disk(md), 1172 blk_get_integrity(template_disk)); 1173 return 0; 1174 } 1175 1176 /* 1177 * If DM device already has an initialized integrity 1178 * profile the new profile should not conflict. 1179 */ 1180 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) { 1181 DMERR("%s: conflict with existing integrity profile: %s profile mismatch", 1182 dm_device_name(t->md), 1183 template_disk->disk_name); 1184 return 1; 1185 } 1186 1187 /* Preserve existing integrity profile */ 1188 t->integrity_supported = true; 1189 return 0; 1190 } 1191 1192 #ifdef CONFIG_BLK_INLINE_ENCRYPTION 1193 1194 struct dm_crypto_profile { 1195 struct blk_crypto_profile profile; 1196 struct mapped_device *md; 1197 }; 1198 1199 static int dm_keyslot_evict_callback(struct dm_target *ti, struct dm_dev *dev, 1200 sector_t start, sector_t len, void *data) 1201 { 1202 const struct blk_crypto_key *key = data; 1203 1204 blk_crypto_evict_key(dev->bdev, key); 1205 return 0; 1206 } 1207 1208 /* 1209 * When an inline encryption key is evicted from a device-mapper device, evict 1210 * it from all the underlying devices. 1211 */ 1212 static int dm_keyslot_evict(struct blk_crypto_profile *profile, 1213 const struct blk_crypto_key *key, unsigned int slot) 1214 { 1215 struct mapped_device *md = 1216 container_of(profile, struct dm_crypto_profile, profile)->md; 1217 struct dm_table *t; 1218 int srcu_idx; 1219 1220 t = dm_get_live_table(md, &srcu_idx); 1221 if (!t) 1222 return 0; 1223 1224 for (unsigned int i = 0; i < t->num_targets; i++) { 1225 struct dm_target *ti = dm_table_get_target(t, i); 1226 1227 if (!ti->type->iterate_devices) 1228 continue; 1229 ti->type->iterate_devices(ti, dm_keyslot_evict_callback, 1230 (void *)key); 1231 } 1232 1233 dm_put_live_table(md, srcu_idx); 1234 return 0; 1235 } 1236 1237 static int 1238 device_intersect_crypto_capabilities(struct dm_target *ti, struct dm_dev *dev, 1239 sector_t start, sector_t len, void *data) 1240 { 1241 struct blk_crypto_profile *parent = data; 1242 struct blk_crypto_profile *child = 1243 bdev_get_queue(dev->bdev)->crypto_profile; 1244 1245 blk_crypto_intersect_capabilities(parent, child); 1246 return 0; 1247 } 1248 1249 void dm_destroy_crypto_profile(struct blk_crypto_profile *profile) 1250 { 1251 struct dm_crypto_profile *dmcp = container_of(profile, 1252 struct dm_crypto_profile, 1253 profile); 1254 1255 if (!profile) 1256 return; 1257 1258 blk_crypto_profile_destroy(profile); 1259 kfree(dmcp); 1260 } 1261 1262 static void dm_table_destroy_crypto_profile(struct dm_table *t) 1263 { 1264 dm_destroy_crypto_profile(t->crypto_profile); 1265 t->crypto_profile = NULL; 1266 } 1267 1268 /* 1269 * Constructs and initializes t->crypto_profile with a crypto profile that 1270 * represents the common set of crypto capabilities of the devices described by 1271 * the dm_table. However, if the constructed crypto profile doesn't support all 1272 * crypto capabilities that are supported by the current mapped_device, it 1273 * returns an error instead, since we don't support removing crypto capabilities 1274 * on table changes. Finally, if the constructed crypto profile is "empty" (has 1275 * no crypto capabilities at all), it just sets t->crypto_profile to NULL. 1276 */ 1277 static int dm_table_construct_crypto_profile(struct dm_table *t) 1278 { 1279 struct dm_crypto_profile *dmcp; 1280 struct blk_crypto_profile *profile; 1281 unsigned int i; 1282 bool empty_profile = true; 1283 1284 dmcp = kmalloc(sizeof(*dmcp), GFP_KERNEL); 1285 if (!dmcp) 1286 return -ENOMEM; 1287 dmcp->md = t->md; 1288 1289 profile = &dmcp->profile; 1290 blk_crypto_profile_init(profile, 0); 1291 profile->ll_ops.keyslot_evict = dm_keyslot_evict; 1292 profile->max_dun_bytes_supported = UINT_MAX; 1293 memset(profile->modes_supported, 0xFF, 1294 sizeof(profile->modes_supported)); 1295 1296 for (i = 0; i < t->num_targets; i++) { 1297 struct dm_target *ti = dm_table_get_target(t, i); 1298 1299 if (!dm_target_passes_crypto(ti->type)) { 1300 blk_crypto_intersect_capabilities(profile, NULL); 1301 break; 1302 } 1303 if (!ti->type->iterate_devices) 1304 continue; 1305 ti->type->iterate_devices(ti, 1306 device_intersect_crypto_capabilities, 1307 profile); 1308 } 1309 1310 if (t->md->queue && 1311 !blk_crypto_has_capabilities(profile, 1312 t->md->queue->crypto_profile)) { 1313 DMERR("Inline encryption capabilities of new DM table were more restrictive than the old table's. This is not supported!"); 1314 dm_destroy_crypto_profile(profile); 1315 return -EINVAL; 1316 } 1317 1318 /* 1319 * If the new profile doesn't actually support any crypto capabilities, 1320 * we may as well represent it with a NULL profile. 1321 */ 1322 for (i = 0; i < ARRAY_SIZE(profile->modes_supported); i++) { 1323 if (profile->modes_supported[i]) { 1324 empty_profile = false; 1325 break; 1326 } 1327 } 1328 1329 if (empty_profile) { 1330 dm_destroy_crypto_profile(profile); 1331 profile = NULL; 1332 } 1333 1334 /* 1335 * t->crypto_profile is only set temporarily while the table is being 1336 * set up, and it gets set to NULL after the profile has been 1337 * transferred to the request_queue. 1338 */ 1339 t->crypto_profile = profile; 1340 1341 return 0; 1342 } 1343 1344 static void dm_update_crypto_profile(struct request_queue *q, 1345 struct dm_table *t) 1346 { 1347 if (!t->crypto_profile) 1348 return; 1349 1350 /* Make the crypto profile less restrictive. */ 1351 if (!q->crypto_profile) { 1352 blk_crypto_register(t->crypto_profile, q); 1353 } else { 1354 blk_crypto_update_capabilities(q->crypto_profile, 1355 t->crypto_profile); 1356 dm_destroy_crypto_profile(t->crypto_profile); 1357 } 1358 t->crypto_profile = NULL; 1359 } 1360 1361 #else /* CONFIG_BLK_INLINE_ENCRYPTION */ 1362 1363 static int dm_table_construct_crypto_profile(struct dm_table *t) 1364 { 1365 return 0; 1366 } 1367 1368 void dm_destroy_crypto_profile(struct blk_crypto_profile *profile) 1369 { 1370 } 1371 1372 static void dm_table_destroy_crypto_profile(struct dm_table *t) 1373 { 1374 } 1375 1376 static void dm_update_crypto_profile(struct request_queue *q, 1377 struct dm_table *t) 1378 { 1379 } 1380 1381 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */ 1382 1383 /* 1384 * Prepares the table for use by building the indices, 1385 * setting the type, and allocating mempools. 1386 */ 1387 int dm_table_complete(struct dm_table *t) 1388 { 1389 int r; 1390 1391 r = dm_table_determine_type(t); 1392 if (r) { 1393 DMERR("unable to determine table type"); 1394 return r; 1395 } 1396 1397 r = dm_table_build_index(t); 1398 if (r) { 1399 DMERR("unable to build btrees"); 1400 return r; 1401 } 1402 1403 r = dm_table_register_integrity(t); 1404 if (r) { 1405 DMERR("could not register integrity profile."); 1406 return r; 1407 } 1408 1409 r = dm_table_construct_crypto_profile(t); 1410 if (r) { 1411 DMERR("could not construct crypto profile."); 1412 return r; 1413 } 1414 1415 r = dm_table_alloc_md_mempools(t, t->md); 1416 if (r) 1417 DMERR("unable to allocate mempools"); 1418 1419 return r; 1420 } 1421 1422 static DEFINE_MUTEX(_event_lock); 1423 void dm_table_event_callback(struct dm_table *t, 1424 void (*fn)(void *), void *context) 1425 { 1426 mutex_lock(&_event_lock); 1427 t->event_fn = fn; 1428 t->event_context = context; 1429 mutex_unlock(&_event_lock); 1430 } 1431 1432 void dm_table_event(struct dm_table *t) 1433 { 1434 mutex_lock(&_event_lock); 1435 if (t->event_fn) 1436 t->event_fn(t->event_context); 1437 mutex_unlock(&_event_lock); 1438 } 1439 EXPORT_SYMBOL(dm_table_event); 1440 1441 inline sector_t dm_table_get_size(struct dm_table *t) 1442 { 1443 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0; 1444 } 1445 EXPORT_SYMBOL(dm_table_get_size); 1446 1447 /* 1448 * Search the btree for the correct target. 1449 * 1450 * Caller should check returned pointer for NULL 1451 * to trap I/O beyond end of device. 1452 */ 1453 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector) 1454 { 1455 unsigned int l, n = 0, k = 0; 1456 sector_t *node; 1457 1458 if (unlikely(sector >= dm_table_get_size(t))) 1459 return NULL; 1460 1461 for (l = 0; l < t->depth; l++) { 1462 n = get_child(n, k); 1463 node = get_node(t, l, n); 1464 1465 for (k = 0; k < KEYS_PER_NODE; k++) 1466 if (node[k] >= sector) 1467 break; 1468 } 1469 1470 return &t->targets[(KEYS_PER_NODE * n) + k]; 1471 } 1472 1473 static int device_not_poll_capable(struct dm_target *ti, struct dm_dev *dev, 1474 sector_t start, sector_t len, void *data) 1475 { 1476 struct request_queue *q = bdev_get_queue(dev->bdev); 1477 1478 return !test_bit(QUEUE_FLAG_POLL, &q->queue_flags); 1479 } 1480 1481 /* 1482 * type->iterate_devices() should be called when the sanity check needs to 1483 * iterate and check all underlying data devices. iterate_devices() will 1484 * iterate all underlying data devices until it encounters a non-zero return 1485 * code, returned by whether the input iterate_devices_callout_fn, or 1486 * iterate_devices() itself internally. 1487 * 1488 * For some target type (e.g. dm-stripe), one call of iterate_devices() may 1489 * iterate multiple underlying devices internally, in which case a non-zero 1490 * return code returned by iterate_devices_callout_fn will stop the iteration 1491 * in advance. 1492 * 1493 * Cases requiring _any_ underlying device supporting some kind of attribute, 1494 * should use the iteration structure like dm_table_any_dev_attr(), or call 1495 * it directly. @func should handle semantics of positive examples, e.g. 1496 * capable of something. 1497 * 1498 * Cases requiring _all_ underlying devices supporting some kind of attribute, 1499 * should use the iteration structure like dm_table_supports_nowait() or 1500 * dm_table_supports_discards(). Or introduce dm_table_all_devs_attr() that 1501 * uses an @anti_func that handle semantics of counter examples, e.g. not 1502 * capable of something. So: return !dm_table_any_dev_attr(t, anti_func, data); 1503 */ 1504 static bool dm_table_any_dev_attr(struct dm_table *t, 1505 iterate_devices_callout_fn func, void *data) 1506 { 1507 for (unsigned int i = 0; i < t->num_targets; i++) { 1508 struct dm_target *ti = dm_table_get_target(t, i); 1509 1510 if (ti->type->iterate_devices && 1511 ti->type->iterate_devices(ti, func, data)) 1512 return true; 1513 } 1514 1515 return false; 1516 } 1517 1518 static int count_device(struct dm_target *ti, struct dm_dev *dev, 1519 sector_t start, sector_t len, void *data) 1520 { 1521 unsigned int *num_devices = data; 1522 1523 (*num_devices)++; 1524 1525 return 0; 1526 } 1527 1528 static bool dm_table_supports_poll(struct dm_table *t) 1529 { 1530 for (unsigned int i = 0; i < t->num_targets; i++) { 1531 struct dm_target *ti = dm_table_get_target(t, i); 1532 1533 if (!ti->type->iterate_devices || 1534 ti->type->iterate_devices(ti, device_not_poll_capable, NULL)) 1535 return false; 1536 } 1537 1538 return true; 1539 } 1540 1541 /* 1542 * Check whether a table has no data devices attached using each 1543 * target's iterate_devices method. 1544 * Returns false if the result is unknown because a target doesn't 1545 * support iterate_devices. 1546 */ 1547 bool dm_table_has_no_data_devices(struct dm_table *t) 1548 { 1549 for (unsigned int i = 0; i < t->num_targets; i++) { 1550 struct dm_target *ti = dm_table_get_target(t, i); 1551 unsigned int num_devices = 0; 1552 1553 if (!ti->type->iterate_devices) 1554 return false; 1555 1556 ti->type->iterate_devices(ti, count_device, &num_devices); 1557 if (num_devices) 1558 return false; 1559 } 1560 1561 return true; 1562 } 1563 1564 static int device_not_zoned_model(struct dm_target *ti, struct dm_dev *dev, 1565 sector_t start, sector_t len, void *data) 1566 { 1567 struct request_queue *q = bdev_get_queue(dev->bdev); 1568 enum blk_zoned_model *zoned_model = data; 1569 1570 return blk_queue_zoned_model(q) != *zoned_model; 1571 } 1572 1573 /* 1574 * Check the device zoned model based on the target feature flag. If the target 1575 * has the DM_TARGET_ZONED_HM feature flag set, host-managed zoned devices are 1576 * also accepted but all devices must have the same zoned model. If the target 1577 * has the DM_TARGET_MIXED_ZONED_MODEL feature set, the devices can have any 1578 * zoned model with all zoned devices having the same zone size. 1579 */ 1580 static bool dm_table_supports_zoned_model(struct dm_table *t, 1581 enum blk_zoned_model zoned_model) 1582 { 1583 for (unsigned int i = 0; i < t->num_targets; i++) { 1584 struct dm_target *ti = dm_table_get_target(t, i); 1585 1586 if (dm_target_supports_zoned_hm(ti->type)) { 1587 if (!ti->type->iterate_devices || 1588 ti->type->iterate_devices(ti, device_not_zoned_model, 1589 &zoned_model)) 1590 return false; 1591 } else if (!dm_target_supports_mixed_zoned_model(ti->type)) { 1592 if (zoned_model == BLK_ZONED_HM) 1593 return false; 1594 } 1595 } 1596 1597 return true; 1598 } 1599 1600 static int device_not_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev, 1601 sector_t start, sector_t len, void *data) 1602 { 1603 unsigned int *zone_sectors = data; 1604 1605 if (!bdev_is_zoned(dev->bdev)) 1606 return 0; 1607 return bdev_zone_sectors(dev->bdev) != *zone_sectors; 1608 } 1609 1610 /* 1611 * Check consistency of zoned model and zone sectors across all targets. For 1612 * zone sectors, if the destination device is a zoned block device, it shall 1613 * have the specified zone_sectors. 1614 */ 1615 static int validate_hardware_zoned_model(struct dm_table *t, 1616 enum blk_zoned_model zoned_model, 1617 unsigned int zone_sectors) 1618 { 1619 if (zoned_model == BLK_ZONED_NONE) 1620 return 0; 1621 1622 if (!dm_table_supports_zoned_model(t, zoned_model)) { 1623 DMERR("%s: zoned model is not consistent across all devices", 1624 dm_device_name(t->md)); 1625 return -EINVAL; 1626 } 1627 1628 /* Check zone size validity and compatibility */ 1629 if (!zone_sectors || !is_power_of_2(zone_sectors)) 1630 return -EINVAL; 1631 1632 if (dm_table_any_dev_attr(t, device_not_matches_zone_sectors, &zone_sectors)) { 1633 DMERR("%s: zone sectors is not consistent across all zoned devices", 1634 dm_device_name(t->md)); 1635 return -EINVAL; 1636 } 1637 1638 return 0; 1639 } 1640 1641 /* 1642 * Establish the new table's queue_limits and validate them. 1643 */ 1644 int dm_calculate_queue_limits(struct dm_table *t, 1645 struct queue_limits *limits) 1646 { 1647 struct queue_limits ti_limits; 1648 enum blk_zoned_model zoned_model = BLK_ZONED_NONE; 1649 unsigned int zone_sectors = 0; 1650 1651 blk_set_stacking_limits(limits); 1652 1653 for (unsigned int i = 0; i < t->num_targets; i++) { 1654 struct dm_target *ti = dm_table_get_target(t, i); 1655 1656 blk_set_stacking_limits(&ti_limits); 1657 1658 if (!ti->type->iterate_devices) { 1659 /* Set I/O hints portion of queue limits */ 1660 if (ti->type->io_hints) 1661 ti->type->io_hints(ti, &ti_limits); 1662 goto combine_limits; 1663 } 1664 1665 /* 1666 * Combine queue limits of all the devices this target uses. 1667 */ 1668 ti->type->iterate_devices(ti, dm_set_device_limits, 1669 &ti_limits); 1670 1671 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) { 1672 /* 1673 * After stacking all limits, validate all devices 1674 * in table support this zoned model and zone sectors. 1675 */ 1676 zoned_model = ti_limits.zoned; 1677 zone_sectors = ti_limits.chunk_sectors; 1678 } 1679 1680 /* Set I/O hints portion of queue limits */ 1681 if (ti->type->io_hints) 1682 ti->type->io_hints(ti, &ti_limits); 1683 1684 /* 1685 * Check each device area is consistent with the target's 1686 * overall queue limits. 1687 */ 1688 if (ti->type->iterate_devices(ti, device_area_is_invalid, 1689 &ti_limits)) 1690 return -EINVAL; 1691 1692 combine_limits: 1693 /* 1694 * Merge this target's queue limits into the overall limits 1695 * for the table. 1696 */ 1697 if (blk_stack_limits(limits, &ti_limits, 0) < 0) 1698 DMWARN("%s: adding target device (start sect %llu len %llu) " 1699 "caused an alignment inconsistency", 1700 dm_device_name(t->md), 1701 (unsigned long long) ti->begin, 1702 (unsigned long long) ti->len); 1703 } 1704 1705 /* 1706 * Verify that the zoned model and zone sectors, as determined before 1707 * any .io_hints override, are the same across all devices in the table. 1708 * - this is especially relevant if .io_hints is emulating a disk-managed 1709 * zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices. 1710 * BUT... 1711 */ 1712 if (limits->zoned != BLK_ZONED_NONE) { 1713 /* 1714 * ...IF the above limits stacking determined a zoned model 1715 * validate that all of the table's devices conform to it. 1716 */ 1717 zoned_model = limits->zoned; 1718 zone_sectors = limits->chunk_sectors; 1719 } 1720 if (validate_hardware_zoned_model(t, zoned_model, zone_sectors)) 1721 return -EINVAL; 1722 1723 return validate_hardware_logical_block_alignment(t, limits); 1724 } 1725 1726 /* 1727 * Verify that all devices have an integrity profile that matches the 1728 * DM device's registered integrity profile. If the profiles don't 1729 * match then unregister the DM device's integrity profile. 1730 */ 1731 static void dm_table_verify_integrity(struct dm_table *t) 1732 { 1733 struct gendisk *template_disk = NULL; 1734 1735 if (t->integrity_added) 1736 return; 1737 1738 if (t->integrity_supported) { 1739 /* 1740 * Verify that the original integrity profile 1741 * matches all the devices in this table. 1742 */ 1743 template_disk = dm_table_get_integrity_disk(t); 1744 if (template_disk && 1745 blk_integrity_compare(dm_disk(t->md), template_disk) >= 0) 1746 return; 1747 } 1748 1749 if (integrity_profile_exists(dm_disk(t->md))) { 1750 DMWARN("%s: unable to establish an integrity profile", 1751 dm_device_name(t->md)); 1752 blk_integrity_unregister(dm_disk(t->md)); 1753 } 1754 } 1755 1756 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev, 1757 sector_t start, sector_t len, void *data) 1758 { 1759 unsigned long flush = (unsigned long) data; 1760 struct request_queue *q = bdev_get_queue(dev->bdev); 1761 1762 return (q->queue_flags & flush); 1763 } 1764 1765 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush) 1766 { 1767 /* 1768 * Require at least one underlying device to support flushes. 1769 * t->devices includes internal dm devices such as mirror logs 1770 * so we need to use iterate_devices here, which targets 1771 * supporting flushes must provide. 1772 */ 1773 for (unsigned int i = 0; i < t->num_targets; i++) { 1774 struct dm_target *ti = dm_table_get_target(t, i); 1775 1776 if (!ti->num_flush_bios) 1777 continue; 1778 1779 if (ti->flush_supported) 1780 return true; 1781 1782 if (ti->type->iterate_devices && 1783 ti->type->iterate_devices(ti, device_flush_capable, (void *) flush)) 1784 return true; 1785 } 1786 1787 return false; 1788 } 1789 1790 static int device_dax_write_cache_enabled(struct dm_target *ti, 1791 struct dm_dev *dev, sector_t start, 1792 sector_t len, void *data) 1793 { 1794 struct dax_device *dax_dev = dev->dax_dev; 1795 1796 if (!dax_dev) 1797 return false; 1798 1799 if (dax_write_cache_enabled(dax_dev)) 1800 return true; 1801 return false; 1802 } 1803 1804 static int device_is_rotational(struct dm_target *ti, struct dm_dev *dev, 1805 sector_t start, sector_t len, void *data) 1806 { 1807 return !bdev_nonrot(dev->bdev); 1808 } 1809 1810 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev, 1811 sector_t start, sector_t len, void *data) 1812 { 1813 struct request_queue *q = bdev_get_queue(dev->bdev); 1814 1815 return !blk_queue_add_random(q); 1816 } 1817 1818 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev, 1819 sector_t start, sector_t len, void *data) 1820 { 1821 struct request_queue *q = bdev_get_queue(dev->bdev); 1822 1823 return !q->limits.max_write_zeroes_sectors; 1824 } 1825 1826 static bool dm_table_supports_write_zeroes(struct dm_table *t) 1827 { 1828 for (unsigned int i = 0; i < t->num_targets; i++) { 1829 struct dm_target *ti = dm_table_get_target(t, i); 1830 1831 if (!ti->num_write_zeroes_bios) 1832 return false; 1833 1834 if (!ti->type->iterate_devices || 1835 ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL)) 1836 return false; 1837 } 1838 1839 return true; 1840 } 1841 1842 static int device_not_nowait_capable(struct dm_target *ti, struct dm_dev *dev, 1843 sector_t start, sector_t len, void *data) 1844 { 1845 return !bdev_nowait(dev->bdev); 1846 } 1847 1848 static bool dm_table_supports_nowait(struct dm_table *t) 1849 { 1850 for (unsigned int i = 0; i < t->num_targets; i++) { 1851 struct dm_target *ti = dm_table_get_target(t, i); 1852 1853 if (!dm_target_supports_nowait(ti->type)) 1854 return false; 1855 1856 if (!ti->type->iterate_devices || 1857 ti->type->iterate_devices(ti, device_not_nowait_capable, NULL)) 1858 return false; 1859 } 1860 1861 return true; 1862 } 1863 1864 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev, 1865 sector_t start, sector_t len, void *data) 1866 { 1867 return !bdev_max_discard_sectors(dev->bdev); 1868 } 1869 1870 static bool dm_table_supports_discards(struct dm_table *t) 1871 { 1872 for (unsigned int i = 0; i < t->num_targets; i++) { 1873 struct dm_target *ti = dm_table_get_target(t, i); 1874 1875 if (!ti->num_discard_bios) 1876 return false; 1877 1878 /* 1879 * Either the target provides discard support (as implied by setting 1880 * 'discards_supported') or it relies on _all_ data devices having 1881 * discard support. 1882 */ 1883 if (!ti->discards_supported && 1884 (!ti->type->iterate_devices || 1885 ti->type->iterate_devices(ti, device_not_discard_capable, NULL))) 1886 return false; 1887 } 1888 1889 return true; 1890 } 1891 1892 static int device_not_secure_erase_capable(struct dm_target *ti, 1893 struct dm_dev *dev, sector_t start, 1894 sector_t len, void *data) 1895 { 1896 return !bdev_max_secure_erase_sectors(dev->bdev); 1897 } 1898 1899 static bool dm_table_supports_secure_erase(struct dm_table *t) 1900 { 1901 for (unsigned int i = 0; i < t->num_targets; i++) { 1902 struct dm_target *ti = dm_table_get_target(t, i); 1903 1904 if (!ti->num_secure_erase_bios) 1905 return false; 1906 1907 if (!ti->type->iterate_devices || 1908 ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL)) 1909 return false; 1910 } 1911 1912 return true; 1913 } 1914 1915 static int device_requires_stable_pages(struct dm_target *ti, 1916 struct dm_dev *dev, sector_t start, 1917 sector_t len, void *data) 1918 { 1919 return bdev_stable_writes(dev->bdev); 1920 } 1921 1922 int dm_table_set_restrictions(struct dm_table *t, struct request_queue *q, 1923 struct queue_limits *limits) 1924 { 1925 bool wc = false, fua = false; 1926 int r; 1927 1928 /* 1929 * Copy table's limits to the DM device's request_queue 1930 */ 1931 q->limits = *limits; 1932 1933 if (dm_table_supports_nowait(t)) 1934 blk_queue_flag_set(QUEUE_FLAG_NOWAIT, q); 1935 else 1936 blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, q); 1937 1938 if (!dm_table_supports_discards(t)) { 1939 q->limits.max_discard_sectors = 0; 1940 q->limits.max_hw_discard_sectors = 0; 1941 q->limits.discard_granularity = 0; 1942 q->limits.discard_alignment = 0; 1943 q->limits.discard_misaligned = 0; 1944 } 1945 1946 if (!dm_table_supports_secure_erase(t)) 1947 q->limits.max_secure_erase_sectors = 0; 1948 1949 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) { 1950 wc = true; 1951 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA))) 1952 fua = true; 1953 } 1954 blk_queue_write_cache(q, wc, fua); 1955 1956 if (dm_table_supports_dax(t, device_not_dax_capable)) { 1957 blk_queue_flag_set(QUEUE_FLAG_DAX, q); 1958 if (dm_table_supports_dax(t, device_not_dax_synchronous_capable)) 1959 set_dax_synchronous(t->md->dax_dev); 1960 } else 1961 blk_queue_flag_clear(QUEUE_FLAG_DAX, q); 1962 1963 if (dm_table_any_dev_attr(t, device_dax_write_cache_enabled, NULL)) 1964 dax_write_cache(t->md->dax_dev, true); 1965 1966 /* Ensure that all underlying devices are non-rotational. */ 1967 if (dm_table_any_dev_attr(t, device_is_rotational, NULL)) 1968 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); 1969 else 1970 blk_queue_flag_set(QUEUE_FLAG_NONROT, q); 1971 1972 if (!dm_table_supports_write_zeroes(t)) 1973 q->limits.max_write_zeroes_sectors = 0; 1974 1975 dm_table_verify_integrity(t); 1976 1977 /* 1978 * Some devices don't use blk_integrity but still want stable pages 1979 * because they do their own checksumming. 1980 * If any underlying device requires stable pages, a table must require 1981 * them as well. Only targets that support iterate_devices are considered: 1982 * don't want error, zero, etc to require stable pages. 1983 */ 1984 if (dm_table_any_dev_attr(t, device_requires_stable_pages, NULL)) 1985 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, q); 1986 else 1987 blk_queue_flag_clear(QUEUE_FLAG_STABLE_WRITES, q); 1988 1989 /* 1990 * Determine whether or not this queue's I/O timings contribute 1991 * to the entropy pool, Only request-based targets use this. 1992 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not 1993 * have it set. 1994 */ 1995 if (blk_queue_add_random(q) && 1996 dm_table_any_dev_attr(t, device_is_not_random, NULL)) 1997 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q); 1998 1999 /* 2000 * For a zoned target, setup the zones related queue attributes 2001 * and resources necessary for zone append emulation if necessary. 2002 */ 2003 if (blk_queue_is_zoned(q)) { 2004 r = dm_set_zones_restrictions(t, q); 2005 if (r) 2006 return r; 2007 if (!static_key_enabled(&zoned_enabled.key)) 2008 static_branch_enable(&zoned_enabled); 2009 } 2010 2011 dm_update_crypto_profile(q, t); 2012 disk_update_readahead(t->md->disk); 2013 2014 /* 2015 * Check for request-based device is left to 2016 * dm_mq_init_request_queue()->blk_mq_init_allocated_queue(). 2017 * 2018 * For bio-based device, only set QUEUE_FLAG_POLL when all 2019 * underlying devices supporting polling. 2020 */ 2021 if (__table_type_bio_based(t->type)) { 2022 if (dm_table_supports_poll(t)) 2023 blk_queue_flag_set(QUEUE_FLAG_POLL, q); 2024 else 2025 blk_queue_flag_clear(QUEUE_FLAG_POLL, q); 2026 } 2027 2028 return 0; 2029 } 2030 2031 struct list_head *dm_table_get_devices(struct dm_table *t) 2032 { 2033 return &t->devices; 2034 } 2035 2036 fmode_t dm_table_get_mode(struct dm_table *t) 2037 { 2038 return t->mode; 2039 } 2040 EXPORT_SYMBOL(dm_table_get_mode); 2041 2042 enum suspend_mode { 2043 PRESUSPEND, 2044 PRESUSPEND_UNDO, 2045 POSTSUSPEND, 2046 }; 2047 2048 static void suspend_targets(struct dm_table *t, enum suspend_mode mode) 2049 { 2050 lockdep_assert_held(&t->md->suspend_lock); 2051 2052 for (unsigned int i = 0; i < t->num_targets; i++) { 2053 struct dm_target *ti = dm_table_get_target(t, i); 2054 2055 switch (mode) { 2056 case PRESUSPEND: 2057 if (ti->type->presuspend) 2058 ti->type->presuspend(ti); 2059 break; 2060 case PRESUSPEND_UNDO: 2061 if (ti->type->presuspend_undo) 2062 ti->type->presuspend_undo(ti); 2063 break; 2064 case POSTSUSPEND: 2065 if (ti->type->postsuspend) 2066 ti->type->postsuspend(ti); 2067 break; 2068 } 2069 } 2070 } 2071 2072 void dm_table_presuspend_targets(struct dm_table *t) 2073 { 2074 if (!t) 2075 return; 2076 2077 suspend_targets(t, PRESUSPEND); 2078 } 2079 2080 void dm_table_presuspend_undo_targets(struct dm_table *t) 2081 { 2082 if (!t) 2083 return; 2084 2085 suspend_targets(t, PRESUSPEND_UNDO); 2086 } 2087 2088 void dm_table_postsuspend_targets(struct dm_table *t) 2089 { 2090 if (!t) 2091 return; 2092 2093 suspend_targets(t, POSTSUSPEND); 2094 } 2095 2096 int dm_table_resume_targets(struct dm_table *t) 2097 { 2098 unsigned int i; 2099 int r = 0; 2100 2101 lockdep_assert_held(&t->md->suspend_lock); 2102 2103 for (i = 0; i < t->num_targets; i++) { 2104 struct dm_target *ti = dm_table_get_target(t, i); 2105 2106 if (!ti->type->preresume) 2107 continue; 2108 2109 r = ti->type->preresume(ti); 2110 if (r) { 2111 DMERR("%s: %s: preresume failed, error = %d", 2112 dm_device_name(t->md), ti->type->name, r); 2113 return r; 2114 } 2115 } 2116 2117 for (i = 0; i < t->num_targets; i++) { 2118 struct dm_target *ti = dm_table_get_target(t, i); 2119 2120 if (ti->type->resume) 2121 ti->type->resume(ti); 2122 } 2123 2124 return 0; 2125 } 2126 2127 struct mapped_device *dm_table_get_md(struct dm_table *t) 2128 { 2129 return t->md; 2130 } 2131 EXPORT_SYMBOL(dm_table_get_md); 2132 2133 const char *dm_table_device_name(struct dm_table *t) 2134 { 2135 return dm_device_name(t->md); 2136 } 2137 EXPORT_SYMBOL_GPL(dm_table_device_name); 2138 2139 void dm_table_run_md_queue_async(struct dm_table *t) 2140 { 2141 if (!dm_table_request_based(t)) 2142 return; 2143 2144 if (t->md->queue) 2145 blk_mq_run_hw_queues(t->md->queue, true); 2146 } 2147 EXPORT_SYMBOL(dm_table_run_md_queue_async); 2148 2149