1/* 2 * Copyright (C) 2001 Sistina Software (UK) Limited. 3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. 4 * 5 * This file is released under the GPL. 6 */ 7 8#include "dm.h" 9 10#include <linux/module.h> 11#include <linux/vmalloc.h> 12#include <linux/blkdev.h> 13#include <linux/namei.h> 14#include <linux/ctype.h> 15#include <linux/string.h> 16#include <linux/slab.h> 17#include <linux/interrupt.h> 18#include <linux/mutex.h> 19#include <linux/delay.h> 20#include <linux/atomic.h> 21#include <linux/blk-mq.h> 22#include <linux/mount.h> 23 24#define DM_MSG_PREFIX "table" 25 26#define MAX_DEPTH 16 27#define NODE_SIZE L1_CACHE_BYTES 28#define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t)) 29#define CHILDREN_PER_NODE (KEYS_PER_NODE + 1) 30 31struct dm_table { 32 struct mapped_device *md; 33 unsigned type; 34 35 /* btree table */ 36 unsigned int depth; 37 unsigned int counts[MAX_DEPTH]; /* in nodes */ 38 sector_t *index[MAX_DEPTH]; 39 40 unsigned int num_targets; 41 unsigned int num_allocated; 42 sector_t *highs; 43 struct dm_target *targets; 44 45 struct target_type *immutable_target_type; 46 unsigned integrity_supported:1; 47 unsigned singleton:1; 48 49 /* 50 * Indicates the rw permissions for the new logical 51 * device. This should be a combination of FMODE_READ 52 * and FMODE_WRITE. 53 */ 54 fmode_t mode; 55 56 /* a list of devices used by this table */ 57 struct list_head devices; 58 59 /* events get handed up using this callback */ 60 void (*event_fn)(void *); 61 void *event_context; 62 63 struct dm_md_mempools *mempools; 64 65 struct list_head target_callbacks; 66}; 67 68/* 69 * Similar to ceiling(log_size(n)) 70 */ 71static unsigned int int_log(unsigned int n, unsigned int base) 72{ 73 int result = 0; 74 75 while (n > 1) { 76 n = dm_div_up(n, base); 77 result++; 78 } 79 80 return result; 81} 82 83/* 84 * Calculate the index of the child node of the n'th node k'th key. 85 */ 86static inline unsigned int get_child(unsigned int n, unsigned int k) 87{ 88 return (n * CHILDREN_PER_NODE) + k; 89} 90 91/* 92 * Return the n'th node of level l from table t. 93 */ 94static inline sector_t *get_node(struct dm_table *t, 95 unsigned int l, unsigned int n) 96{ 97 return t->index[l] + (n * KEYS_PER_NODE); 98} 99 100/* 101 * Return the highest key that you could lookup from the n'th 102 * node on level l of the btree. 103 */ 104static sector_t high(struct dm_table *t, unsigned int l, unsigned int n) 105{ 106 for (; l < t->depth - 1; l++) 107 n = get_child(n, CHILDREN_PER_NODE - 1); 108 109 if (n >= t->counts[l]) 110 return (sector_t) - 1; 111 112 return get_node(t, l, n)[KEYS_PER_NODE - 1]; 113} 114 115/* 116 * Fills in a level of the btree based on the highs of the level 117 * below it. 118 */ 119static int setup_btree_index(unsigned int l, struct dm_table *t) 120{ 121 unsigned int n, k; 122 sector_t *node; 123 124 for (n = 0U; n < t->counts[l]; n++) { 125 node = get_node(t, l, n); 126 127 for (k = 0U; k < KEYS_PER_NODE; k++) 128 node[k] = high(t, l + 1, get_child(n, k)); 129 } 130 131 return 0; 132} 133 134void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size) 135{ 136 unsigned long size; 137 void *addr; 138 139 /* 140 * Check that we're not going to overflow. 141 */ 142 if (nmemb > (ULONG_MAX / elem_size)) 143 return NULL; 144 145 size = nmemb * elem_size; 146 addr = vzalloc(size); 147 148 return addr; 149} 150EXPORT_SYMBOL(dm_vcalloc); 151 152/* 153 * highs, and targets are managed as dynamic arrays during a 154 * table load. 155 */ 156static int alloc_targets(struct dm_table *t, unsigned int num) 157{ 158 sector_t *n_highs; 159 struct dm_target *n_targets; 160 161 /* 162 * Allocate both the target array and offset array at once. 163 * Append an empty entry to catch sectors beyond the end of 164 * the device. 165 */ 166 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) + 167 sizeof(sector_t)); 168 if (!n_highs) 169 return -ENOMEM; 170 171 n_targets = (struct dm_target *) (n_highs + num); 172 173 memset(n_highs, -1, sizeof(*n_highs) * num); 174 vfree(t->highs); 175 176 t->num_allocated = num; 177 t->highs = n_highs; 178 t->targets = n_targets; 179 180 return 0; 181} 182 183int dm_table_create(struct dm_table **result, fmode_t mode, 184 unsigned num_targets, struct mapped_device *md) 185{ 186 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL); 187 188 if (!t) 189 return -ENOMEM; 190 191 INIT_LIST_HEAD(&t->devices); 192 INIT_LIST_HEAD(&t->target_callbacks); 193 194 if (!num_targets) 195 num_targets = KEYS_PER_NODE; 196 197 num_targets = dm_round_up(num_targets, KEYS_PER_NODE); 198 199 if (!num_targets) { 200 kfree(t); 201 return -ENOMEM; 202 } 203 204 if (alloc_targets(t, num_targets)) { 205 kfree(t); 206 return -ENOMEM; 207 } 208 209 t->mode = mode; 210 t->md = md; 211 *result = t; 212 return 0; 213} 214 215static void free_devices(struct list_head *devices, struct mapped_device *md) 216{ 217 struct list_head *tmp, *next; 218 219 list_for_each_safe(tmp, next, devices) { 220 struct dm_dev_internal *dd = 221 list_entry(tmp, struct dm_dev_internal, list); 222 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s", 223 dm_device_name(md), dd->dm_dev->name); 224 dm_put_table_device(md, dd->dm_dev); 225 kfree(dd); 226 } 227} 228 229void dm_table_destroy(struct dm_table *t) 230{ 231 unsigned int i; 232 233 if (!t) 234 return; 235 236 /* free the indexes */ 237 if (t->depth >= 2) 238 vfree(t->index[t->depth - 2]); 239 240 /* free the targets */ 241 for (i = 0; i < t->num_targets; i++) { 242 struct dm_target *tgt = t->targets + i; 243 244 if (tgt->type->dtr) 245 tgt->type->dtr(tgt); 246 247 dm_put_target_type(tgt->type); 248 } 249 250 vfree(t->highs); 251 252 /* free the device list */ 253 free_devices(&t->devices, t->md); 254 255 dm_free_md_mempools(t->mempools); 256 257 kfree(t); 258} 259 260/* 261 * See if we've already got a device in the list. 262 */ 263static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev) 264{ 265 struct dm_dev_internal *dd; 266 267 list_for_each_entry (dd, l, list) 268 if (dd->dm_dev->bdev->bd_dev == dev) 269 return dd; 270 271 return NULL; 272} 273 274/* 275 * If possible, this checks an area of a destination device is invalid. 276 */ 277static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev, 278 sector_t start, sector_t len, void *data) 279{ 280 struct request_queue *q; 281 struct queue_limits *limits = data; 282 struct block_device *bdev = dev->bdev; 283 sector_t dev_size = 284 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT; 285 unsigned short logical_block_size_sectors = 286 limits->logical_block_size >> SECTOR_SHIFT; 287 char b[BDEVNAME_SIZE]; 288 289 /* 290 * Some devices exist without request functions, 291 * such as loop devices not yet bound to backing files. 292 * Forbid the use of such devices. 293 */ 294 q = bdev_get_queue(bdev); 295 if (!q || !q->make_request_fn) { 296 DMWARN("%s: %s is not yet initialised: " 297 "start=%llu, len=%llu, dev_size=%llu", 298 dm_device_name(ti->table->md), bdevname(bdev, b), 299 (unsigned long long)start, 300 (unsigned long long)len, 301 (unsigned long long)dev_size); 302 return 1; 303 } 304 305 if (!dev_size) 306 return 0; 307 308 if ((start >= dev_size) || (start + len > dev_size)) { 309 DMWARN("%s: %s too small for target: " 310 "start=%llu, len=%llu, dev_size=%llu", 311 dm_device_name(ti->table->md), bdevname(bdev, b), 312 (unsigned long long)start, 313 (unsigned long long)len, 314 (unsigned long long)dev_size); 315 return 1; 316 } 317 318 if (logical_block_size_sectors <= 1) 319 return 0; 320 321 if (start & (logical_block_size_sectors - 1)) { 322 DMWARN("%s: start=%llu not aligned to h/w " 323 "logical block size %u of %s", 324 dm_device_name(ti->table->md), 325 (unsigned long long)start, 326 limits->logical_block_size, bdevname(bdev, b)); 327 return 1; 328 } 329 330 if (len & (logical_block_size_sectors - 1)) { 331 DMWARN("%s: len=%llu not aligned to h/w " 332 "logical block size %u of %s", 333 dm_device_name(ti->table->md), 334 (unsigned long long)len, 335 limits->logical_block_size, bdevname(bdev, b)); 336 return 1; 337 } 338 339 return 0; 340} 341 342/* 343 * This upgrades the mode on an already open dm_dev, being 344 * careful to leave things as they were if we fail to reopen the 345 * device and not to touch the existing bdev field in case 346 * it is accessed concurrently inside dm_table_any_congested(). 347 */ 348static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode, 349 struct mapped_device *md) 350{ 351 int r; 352 struct dm_dev *old_dev, *new_dev; 353 354 old_dev = dd->dm_dev; 355 356 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev, 357 dd->dm_dev->mode | new_mode, &new_dev); 358 if (r) 359 return r; 360 361 dd->dm_dev = new_dev; 362 dm_put_table_device(md, old_dev); 363 364 return 0; 365} 366 367/* 368 * Convert the path to a device 369 */ 370dev_t dm_get_dev_t(const char *path) 371{ 372 dev_t uninitialized_var(dev); 373 struct block_device *bdev; 374 375 bdev = lookup_bdev(path); 376 if (IS_ERR(bdev)) 377 dev = name_to_dev_t(path); 378 else { 379 dev = bdev->bd_dev; 380 bdput(bdev); 381 } 382 383 return dev; 384} 385EXPORT_SYMBOL_GPL(dm_get_dev_t); 386 387/* 388 * Add a device to the list, or just increment the usage count if 389 * it's already present. 390 */ 391int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode, 392 struct dm_dev **result) 393{ 394 int r; 395 dev_t dev; 396 struct dm_dev_internal *dd; 397 struct dm_table *t = ti->table; 398 399 BUG_ON(!t); 400 401 dev = dm_get_dev_t(path); 402 if (!dev) 403 return -ENODEV; 404 405 dd = find_device(&t->devices, dev); 406 if (!dd) { 407 dd = kmalloc(sizeof(*dd), GFP_KERNEL); 408 if (!dd) 409 return -ENOMEM; 410 411 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) { 412 kfree(dd); 413 return r; 414 } 415 416 atomic_set(&dd->count, 0); 417 list_add(&dd->list, &t->devices); 418 419 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) { 420 r = upgrade_mode(dd, mode, t->md); 421 if (r) 422 return r; 423 } 424 atomic_inc(&dd->count); 425 426 *result = dd->dm_dev; 427 return 0; 428} 429EXPORT_SYMBOL(dm_get_device); 430 431static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev, 432 sector_t start, sector_t len, void *data) 433{ 434 struct queue_limits *limits = data; 435 struct block_device *bdev = dev->bdev; 436 struct request_queue *q = bdev_get_queue(bdev); 437 char b[BDEVNAME_SIZE]; 438 439 if (unlikely(!q)) { 440 DMWARN("%s: Cannot set limits for nonexistent device %s", 441 dm_device_name(ti->table->md), bdevname(bdev, b)); 442 return 0; 443 } 444 445 if (bdev_stack_limits(limits, bdev, start) < 0) 446 DMWARN("%s: adding target device %s caused an alignment inconsistency: " 447 "physical_block_size=%u, logical_block_size=%u, " 448 "alignment_offset=%u, start=%llu", 449 dm_device_name(ti->table->md), bdevname(bdev, b), 450 q->limits.physical_block_size, 451 q->limits.logical_block_size, 452 q->limits.alignment_offset, 453 (unsigned long long) start << SECTOR_SHIFT); 454 455 return 0; 456} 457 458/* 459 * Decrement a device's use count and remove it if necessary. 460 */ 461void dm_put_device(struct dm_target *ti, struct dm_dev *d) 462{ 463 int found = 0; 464 struct list_head *devices = &ti->table->devices; 465 struct dm_dev_internal *dd; 466 467 list_for_each_entry(dd, devices, list) { 468 if (dd->dm_dev == d) { 469 found = 1; 470 break; 471 } 472 } 473 if (!found) { 474 DMWARN("%s: device %s not in table devices list", 475 dm_device_name(ti->table->md), d->name); 476 return; 477 } 478 if (atomic_dec_and_test(&dd->count)) { 479 dm_put_table_device(ti->table->md, d); 480 list_del(&dd->list); 481 kfree(dd); 482 } 483} 484EXPORT_SYMBOL(dm_put_device); 485 486/* 487 * Checks to see if the target joins onto the end of the table. 488 */ 489static int adjoin(struct dm_table *table, struct dm_target *ti) 490{ 491 struct dm_target *prev; 492 493 if (!table->num_targets) 494 return !ti->begin; 495 496 prev = &table->targets[table->num_targets - 1]; 497 return (ti->begin == (prev->begin + prev->len)); 498} 499 500/* 501 * Used to dynamically allocate the arg array. 502 * 503 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must 504 * process messages even if some device is suspended. These messages have a 505 * small fixed number of arguments. 506 * 507 * On the other hand, dm-switch needs to process bulk data using messages and 508 * excessive use of GFP_NOIO could cause trouble. 509 */ 510static char **realloc_argv(unsigned *array_size, char **old_argv) 511{ 512 char **argv; 513 unsigned new_size; 514 gfp_t gfp; 515 516 if (*array_size) { 517 new_size = *array_size * 2; 518 gfp = GFP_KERNEL; 519 } else { 520 new_size = 8; 521 gfp = GFP_NOIO; 522 } 523 argv = kmalloc(new_size * sizeof(*argv), gfp); 524 if (argv) { 525 memcpy(argv, old_argv, *array_size * sizeof(*argv)); 526 *array_size = new_size; 527 } 528 529 kfree(old_argv); 530 return argv; 531} 532 533/* 534 * Destructively splits up the argument list to pass to ctr. 535 */ 536int dm_split_args(int *argc, char ***argvp, char *input) 537{ 538 char *start, *end = input, *out, **argv = NULL; 539 unsigned array_size = 0; 540 541 *argc = 0; 542 543 if (!input) { 544 *argvp = NULL; 545 return 0; 546 } 547 548 argv = realloc_argv(&array_size, argv); 549 if (!argv) 550 return -ENOMEM; 551 552 while (1) { 553 /* Skip whitespace */ 554 start = skip_spaces(end); 555 556 if (!*start) 557 break; /* success, we hit the end */ 558 559 /* 'out' is used to remove any back-quotes */ 560 end = out = start; 561 while (*end) { 562 /* Everything apart from '\0' can be quoted */ 563 if (*end == '\\' && *(end + 1)) { 564 *out++ = *(end + 1); 565 end += 2; 566 continue; 567 } 568 569 if (isspace(*end)) 570 break; /* end of token */ 571 572 *out++ = *end++; 573 } 574 575 /* have we already filled the array ? */ 576 if ((*argc + 1) > array_size) { 577 argv = realloc_argv(&array_size, argv); 578 if (!argv) 579 return -ENOMEM; 580 } 581 582 /* we know this is whitespace */ 583 if (*end) 584 end++; 585 586 /* terminate the string and put it in the array */ 587 *out = '\0'; 588 argv[*argc] = start; 589 (*argc)++; 590 } 591 592 *argvp = argv; 593 return 0; 594} 595 596/* 597 * Impose necessary and sufficient conditions on a devices's table such 598 * that any incoming bio which respects its logical_block_size can be 599 * processed successfully. If it falls across the boundary between 600 * two or more targets, the size of each piece it gets split into must 601 * be compatible with the logical_block_size of the target processing it. 602 */ 603static int validate_hardware_logical_block_alignment(struct dm_table *table, 604 struct queue_limits *limits) 605{ 606 /* 607 * This function uses arithmetic modulo the logical_block_size 608 * (in units of 512-byte sectors). 609 */ 610 unsigned short device_logical_block_size_sects = 611 limits->logical_block_size >> SECTOR_SHIFT; 612 613 /* 614 * Offset of the start of the next table entry, mod logical_block_size. 615 */ 616 unsigned short next_target_start = 0; 617 618 /* 619 * Given an aligned bio that extends beyond the end of a 620 * target, how many sectors must the next target handle? 621 */ 622 unsigned short remaining = 0; 623 624 struct dm_target *uninitialized_var(ti); 625 struct queue_limits ti_limits; 626 unsigned i = 0; 627 628 /* 629 * Check each entry in the table in turn. 630 */ 631 while (i < dm_table_get_num_targets(table)) { 632 ti = dm_table_get_target(table, i++); 633 634 blk_set_stacking_limits(&ti_limits); 635 636 /* combine all target devices' limits */ 637 if (ti->type->iterate_devices) 638 ti->type->iterate_devices(ti, dm_set_device_limits, 639 &ti_limits); 640 641 /* 642 * If the remaining sectors fall entirely within this 643 * table entry are they compatible with its logical_block_size? 644 */ 645 if (remaining < ti->len && 646 remaining & ((ti_limits.logical_block_size >> 647 SECTOR_SHIFT) - 1)) 648 break; /* Error */ 649 650 next_target_start = 651 (unsigned short) ((next_target_start + ti->len) & 652 (device_logical_block_size_sects - 1)); 653 remaining = next_target_start ? 654 device_logical_block_size_sects - next_target_start : 0; 655 } 656 657 if (remaining) { 658 DMWARN("%s: table line %u (start sect %llu len %llu) " 659 "not aligned to h/w logical block size %u", 660 dm_device_name(table->md), i, 661 (unsigned long long) ti->begin, 662 (unsigned long long) ti->len, 663 limits->logical_block_size); 664 return -EINVAL; 665 } 666 667 return 0; 668} 669 670int dm_table_add_target(struct dm_table *t, const char *type, 671 sector_t start, sector_t len, char *params) 672{ 673 int r = -EINVAL, argc; 674 char **argv; 675 struct dm_target *tgt; 676 677 if (t->singleton) { 678 DMERR("%s: target type %s must appear alone in table", 679 dm_device_name(t->md), t->targets->type->name); 680 return -EINVAL; 681 } 682 683 BUG_ON(t->num_targets >= t->num_allocated); 684 685 tgt = t->targets + t->num_targets; 686 memset(tgt, 0, sizeof(*tgt)); 687 688 if (!len) { 689 DMERR("%s: zero-length target", dm_device_name(t->md)); 690 return -EINVAL; 691 } 692 693 tgt->type = dm_get_target_type(type); 694 if (!tgt->type) { 695 DMERR("%s: %s: unknown target type", dm_device_name(t->md), 696 type); 697 return -EINVAL; 698 } 699 700 if (dm_target_needs_singleton(tgt->type)) { 701 if (t->num_targets) { 702 DMERR("%s: target type %s must appear alone in table", 703 dm_device_name(t->md), type); 704 return -EINVAL; 705 } 706 t->singleton = 1; 707 } 708 709 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) { 710 DMERR("%s: target type %s may not be included in read-only tables", 711 dm_device_name(t->md), type); 712 return -EINVAL; 713 } 714 715 if (t->immutable_target_type) { 716 if (t->immutable_target_type != tgt->type) { 717 DMERR("%s: immutable target type %s cannot be mixed with other target types", 718 dm_device_name(t->md), t->immutable_target_type->name); 719 return -EINVAL; 720 } 721 } else if (dm_target_is_immutable(tgt->type)) { 722 if (t->num_targets) { 723 DMERR("%s: immutable target type %s cannot be mixed with other target types", 724 dm_device_name(t->md), tgt->type->name); 725 return -EINVAL; 726 } 727 t->immutable_target_type = tgt->type; 728 } 729 730 tgt->table = t; 731 tgt->begin = start; 732 tgt->len = len; 733 tgt->error = "Unknown error"; 734 735 /* 736 * Does this target adjoin the previous one ? 737 */ 738 if (!adjoin(t, tgt)) { 739 tgt->error = "Gap in table"; 740 r = -EINVAL; 741 goto bad; 742 } 743 744 r = dm_split_args(&argc, &argv, params); 745 if (r) { 746 tgt->error = "couldn't split parameters (insufficient memory)"; 747 goto bad; 748 } 749 750 r = tgt->type->ctr(tgt, argc, argv); 751 kfree(argv); 752 if (r) 753 goto bad; 754 755 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1; 756 757 if (!tgt->num_discard_bios && tgt->discards_supported) 758 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.", 759 dm_device_name(t->md), type); 760 761 return 0; 762 763 bad: 764 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error); 765 dm_put_target_type(tgt->type); 766 return r; 767} 768 769/* 770 * Target argument parsing helpers. 771 */ 772static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set, 773 unsigned *value, char **error, unsigned grouped) 774{ 775 const char *arg_str = dm_shift_arg(arg_set); 776 char dummy; 777 778 if (!arg_str || 779 (sscanf(arg_str, "%u%c", value, &dummy) != 1) || 780 (*value < arg->min) || 781 (*value > arg->max) || 782 (grouped && arg_set->argc < *value)) { 783 *error = arg->error; 784 return -EINVAL; 785 } 786 787 return 0; 788} 789 790int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set, 791 unsigned *value, char **error) 792{ 793 return validate_next_arg(arg, arg_set, value, error, 0); 794} 795EXPORT_SYMBOL(dm_read_arg); 796 797int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set, 798 unsigned *value, char **error) 799{ 800 return validate_next_arg(arg, arg_set, value, error, 1); 801} 802EXPORT_SYMBOL(dm_read_arg_group); 803 804const char *dm_shift_arg(struct dm_arg_set *as) 805{ 806 char *r; 807 808 if (as->argc) { 809 as->argc--; 810 r = *as->argv; 811 as->argv++; 812 return r; 813 } 814 815 return NULL; 816} 817EXPORT_SYMBOL(dm_shift_arg); 818 819void dm_consume_args(struct dm_arg_set *as, unsigned num_args) 820{ 821 BUG_ON(as->argc < num_args); 822 as->argc -= num_args; 823 as->argv += num_args; 824} 825EXPORT_SYMBOL(dm_consume_args); 826 827static bool __table_type_request_based(unsigned table_type) 828{ 829 return (table_type == DM_TYPE_REQUEST_BASED || 830 table_type == DM_TYPE_MQ_REQUEST_BASED); 831} 832 833static int dm_table_set_type(struct dm_table *t) 834{ 835 unsigned i; 836 unsigned bio_based = 0, request_based = 0, hybrid = 0; 837 bool use_blk_mq = false; 838 struct dm_target *tgt; 839 struct dm_dev_internal *dd; 840 struct list_head *devices; 841 unsigned live_md_type = dm_get_md_type(t->md); 842 843 for (i = 0; i < t->num_targets; i++) { 844 tgt = t->targets + i; 845 if (dm_target_hybrid(tgt)) 846 hybrid = 1; 847 else if (dm_target_request_based(tgt)) 848 request_based = 1; 849 else 850 bio_based = 1; 851 852 if (bio_based && request_based) { 853 DMWARN("Inconsistent table: different target types" 854 " can't be mixed up"); 855 return -EINVAL; 856 } 857 } 858 859 if (hybrid && !bio_based && !request_based) { 860 /* 861 * The targets can work either way. 862 * Determine the type from the live device. 863 * Default to bio-based if device is new. 864 */ 865 if (__table_type_request_based(live_md_type)) 866 request_based = 1; 867 else 868 bio_based = 1; 869 } 870 871 if (bio_based) { 872 /* We must use this table as bio-based */ 873 t->type = DM_TYPE_BIO_BASED; 874 return 0; 875 } 876 877 BUG_ON(!request_based); /* No targets in this table */ 878 879 /* 880 * Request-based dm supports only tables that have a single target now. 881 * To support multiple targets, request splitting support is needed, 882 * and that needs lots of changes in the block-layer. 883 * (e.g. request completion process for partial completion.) 884 */ 885 if (t->num_targets > 1) { 886 DMWARN("Request-based dm doesn't support multiple targets yet"); 887 return -EINVAL; 888 } 889 890 /* Non-request-stackable devices can't be used for request-based dm */ 891 devices = dm_table_get_devices(t); 892 list_for_each_entry(dd, devices, list) { 893 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev); 894 895 if (!blk_queue_stackable(q)) { 896 DMERR("table load rejected: including" 897 " non-request-stackable devices"); 898 return -EINVAL; 899 } 900 901 if (q->mq_ops) 902 use_blk_mq = true; 903 } 904 905 if (use_blk_mq) { 906 /* verify _all_ devices in the table are blk-mq devices */ 907 list_for_each_entry(dd, devices, list) 908 if (!bdev_get_queue(dd->dm_dev->bdev)->mq_ops) { 909 DMERR("table load rejected: not all devices" 910 " are blk-mq request-stackable"); 911 return -EINVAL; 912 } 913 t->type = DM_TYPE_MQ_REQUEST_BASED; 914 915 } else if (list_empty(devices) && __table_type_request_based(live_md_type)) { 916 /* inherit live MD type */ 917 t->type = live_md_type; 918 919 } else 920 t->type = DM_TYPE_REQUEST_BASED; 921 922 return 0; 923} 924 925unsigned dm_table_get_type(struct dm_table *t) 926{ 927 return t->type; 928} 929 930struct target_type *dm_table_get_immutable_target_type(struct dm_table *t) 931{ 932 return t->immutable_target_type; 933} 934 935bool dm_table_request_based(struct dm_table *t) 936{ 937 return __table_type_request_based(dm_table_get_type(t)); 938} 939 940bool dm_table_mq_request_based(struct dm_table *t) 941{ 942 return dm_table_get_type(t) == DM_TYPE_MQ_REQUEST_BASED; 943} 944 945static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md) 946{ 947 unsigned type = dm_table_get_type(t); 948 unsigned per_bio_data_size = 0; 949 struct dm_target *tgt; 950 unsigned i; 951 952 if (unlikely(type == DM_TYPE_NONE)) { 953 DMWARN("no table type is set, can't allocate mempools"); 954 return -EINVAL; 955 } 956 957 if (type == DM_TYPE_BIO_BASED) 958 for (i = 0; i < t->num_targets; i++) { 959 tgt = t->targets + i; 960 per_bio_data_size = max(per_bio_data_size, tgt->per_bio_data_size); 961 } 962 963 t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported, per_bio_data_size); 964 if (!t->mempools) 965 return -ENOMEM; 966 967 return 0; 968} 969 970void dm_table_free_md_mempools(struct dm_table *t) 971{ 972 dm_free_md_mempools(t->mempools); 973 t->mempools = NULL; 974} 975 976struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t) 977{ 978 return t->mempools; 979} 980 981static int setup_indexes(struct dm_table *t) 982{ 983 int i; 984 unsigned int total = 0; 985 sector_t *indexes; 986 987 /* allocate the space for *all* the indexes */ 988 for (i = t->depth - 2; i >= 0; i--) { 989 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE); 990 total += t->counts[i]; 991 } 992 993 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE); 994 if (!indexes) 995 return -ENOMEM; 996 997 /* set up internal nodes, bottom-up */ 998 for (i = t->depth - 2; i >= 0; i--) { 999 t->index[i] = indexes; 1000 indexes += (KEYS_PER_NODE * t->counts[i]); 1001 setup_btree_index(i, t); 1002 } 1003 1004 return 0; 1005} 1006 1007/* 1008 * Builds the btree to index the map. 1009 */ 1010static int dm_table_build_index(struct dm_table *t) 1011{ 1012 int r = 0; 1013 unsigned int leaf_nodes; 1014 1015 /* how many indexes will the btree have ? */ 1016 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE); 1017 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE); 1018 1019 /* leaf layer has already been set up */ 1020 t->counts[t->depth - 1] = leaf_nodes; 1021 t->index[t->depth - 1] = t->highs; 1022 1023 if (t->depth >= 2) 1024 r = setup_indexes(t); 1025 1026 return r; 1027} 1028 1029static bool integrity_profile_exists(struct gendisk *disk) 1030{ 1031 return !!blk_get_integrity(disk); 1032} 1033 1034/* 1035 * Get a disk whose integrity profile reflects the table's profile. 1036 * Returns NULL if integrity support was inconsistent or unavailable. 1037 */ 1038static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t) 1039{ 1040 struct list_head *devices = dm_table_get_devices(t); 1041 struct dm_dev_internal *dd = NULL; 1042 struct gendisk *prev_disk = NULL, *template_disk = NULL; 1043 1044 list_for_each_entry(dd, devices, list) { 1045 template_disk = dd->dm_dev->bdev->bd_disk; 1046 if (!integrity_profile_exists(template_disk)) 1047 goto no_integrity; 1048 else if (prev_disk && 1049 blk_integrity_compare(prev_disk, template_disk) < 0) 1050 goto no_integrity; 1051 prev_disk = template_disk; 1052 } 1053 1054 return template_disk; 1055 1056no_integrity: 1057 if (prev_disk) 1058 DMWARN("%s: integrity not set: %s and %s profile mismatch", 1059 dm_device_name(t->md), 1060 prev_disk->disk_name, 1061 template_disk->disk_name); 1062 return NULL; 1063} 1064 1065/* 1066 * Register the mapped device for blk_integrity support if the 1067 * underlying devices have an integrity profile. But all devices may 1068 * not have matching profiles (checking all devices isn't reliable 1069 * during table load because this table may use other DM device(s) which 1070 * must be resumed before they will have an initialized integity 1071 * profile). Consequently, stacked DM devices force a 2 stage integrity 1072 * profile validation: First pass during table load, final pass during 1073 * resume. 1074 */ 1075static int dm_table_register_integrity(struct dm_table *t) 1076{ 1077 struct mapped_device *md = t->md; 1078 struct gendisk *template_disk = NULL; 1079 1080 template_disk = dm_table_get_integrity_disk(t); 1081 if (!template_disk) 1082 return 0; 1083 1084 if (!integrity_profile_exists(dm_disk(md))) { 1085 t->integrity_supported = 1; 1086 /* 1087 * Register integrity profile during table load; we can do 1088 * this because the final profile must match during resume. 1089 */ 1090 blk_integrity_register(dm_disk(md), 1091 blk_get_integrity(template_disk)); 1092 return 0; 1093 } 1094 1095 /* 1096 * If DM device already has an initialized integrity 1097 * profile the new profile should not conflict. 1098 */ 1099 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) { 1100 DMWARN("%s: conflict with existing integrity profile: " 1101 "%s profile mismatch", 1102 dm_device_name(t->md), 1103 template_disk->disk_name); 1104 return 1; 1105 } 1106 1107 /* Preserve existing integrity profile */ 1108 t->integrity_supported = 1; 1109 return 0; 1110} 1111 1112/* 1113 * Prepares the table for use by building the indices, 1114 * setting the type, and allocating mempools. 1115 */ 1116int dm_table_complete(struct dm_table *t) 1117{ 1118 int r; 1119 1120 r = dm_table_set_type(t); 1121 if (r) { 1122 DMERR("unable to set table type"); 1123 return r; 1124 } 1125 1126 r = dm_table_build_index(t); 1127 if (r) { 1128 DMERR("unable to build btrees"); 1129 return r; 1130 } 1131 1132 r = dm_table_register_integrity(t); 1133 if (r) { 1134 DMERR("could not register integrity profile."); 1135 return r; 1136 } 1137 1138 r = dm_table_alloc_md_mempools(t, t->md); 1139 if (r) 1140 DMERR("unable to allocate mempools"); 1141 1142 return r; 1143} 1144 1145static DEFINE_MUTEX(_event_lock); 1146void dm_table_event_callback(struct dm_table *t, 1147 void (*fn)(void *), void *context) 1148{ 1149 mutex_lock(&_event_lock); 1150 t->event_fn = fn; 1151 t->event_context = context; 1152 mutex_unlock(&_event_lock); 1153} 1154 1155void dm_table_event(struct dm_table *t) 1156{ 1157 /* 1158 * You can no longer call dm_table_event() from interrupt 1159 * context, use a bottom half instead. 1160 */ 1161 BUG_ON(in_interrupt()); 1162 1163 mutex_lock(&_event_lock); 1164 if (t->event_fn) 1165 t->event_fn(t->event_context); 1166 mutex_unlock(&_event_lock); 1167} 1168EXPORT_SYMBOL(dm_table_event); 1169 1170sector_t dm_table_get_size(struct dm_table *t) 1171{ 1172 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0; 1173} 1174EXPORT_SYMBOL(dm_table_get_size); 1175 1176struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index) 1177{ 1178 if (index >= t->num_targets) 1179 return NULL; 1180 1181 return t->targets + index; 1182} 1183 1184/* 1185 * Search the btree for the correct target. 1186 * 1187 * Caller should check returned pointer with dm_target_is_valid() 1188 * to trap I/O beyond end of device. 1189 */ 1190struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector) 1191{ 1192 unsigned int l, n = 0, k = 0; 1193 sector_t *node; 1194 1195 for (l = 0; l < t->depth; l++) { 1196 n = get_child(n, k); 1197 node = get_node(t, l, n); 1198 1199 for (k = 0; k < KEYS_PER_NODE; k++) 1200 if (node[k] >= sector) 1201 break; 1202 } 1203 1204 return &t->targets[(KEYS_PER_NODE * n) + k]; 1205} 1206 1207static int count_device(struct dm_target *ti, struct dm_dev *dev, 1208 sector_t start, sector_t len, void *data) 1209{ 1210 unsigned *num_devices = data; 1211 1212 (*num_devices)++; 1213 1214 return 0; 1215} 1216 1217/* 1218 * Check whether a table has no data devices attached using each 1219 * target's iterate_devices method. 1220 * Returns false if the result is unknown because a target doesn't 1221 * support iterate_devices. 1222 */ 1223bool dm_table_has_no_data_devices(struct dm_table *table) 1224{ 1225 struct dm_target *uninitialized_var(ti); 1226 unsigned i = 0, num_devices = 0; 1227 1228 while (i < dm_table_get_num_targets(table)) { 1229 ti = dm_table_get_target(table, i++); 1230 1231 if (!ti->type->iterate_devices) 1232 return false; 1233 1234 ti->type->iterate_devices(ti, count_device, &num_devices); 1235 if (num_devices) 1236 return false; 1237 } 1238 1239 return true; 1240} 1241 1242/* 1243 * Establish the new table's queue_limits and validate them. 1244 */ 1245int dm_calculate_queue_limits(struct dm_table *table, 1246 struct queue_limits *limits) 1247{ 1248 struct dm_target *uninitialized_var(ti); 1249 struct queue_limits ti_limits; 1250 unsigned i = 0; 1251 1252 blk_set_stacking_limits(limits); 1253 1254 while (i < dm_table_get_num_targets(table)) { 1255 blk_set_stacking_limits(&ti_limits); 1256 1257 ti = dm_table_get_target(table, i++); 1258 1259 if (!ti->type->iterate_devices) 1260 goto combine_limits; 1261 1262 /* 1263 * Combine queue limits of all the devices this target uses. 1264 */ 1265 ti->type->iterate_devices(ti, dm_set_device_limits, 1266 &ti_limits); 1267 1268 /* Set I/O hints portion of queue limits */ 1269 if (ti->type->io_hints) 1270 ti->type->io_hints(ti, &ti_limits); 1271 1272 /* 1273 * Check each device area is consistent with the target's 1274 * overall queue limits. 1275 */ 1276 if (ti->type->iterate_devices(ti, device_area_is_invalid, 1277 &ti_limits)) 1278 return -EINVAL; 1279 1280combine_limits: 1281 /* 1282 * Merge this target's queue limits into the overall limits 1283 * for the table. 1284 */ 1285 if (blk_stack_limits(limits, &ti_limits, 0) < 0) 1286 DMWARN("%s: adding target device " 1287 "(start sect %llu len %llu) " 1288 "caused an alignment inconsistency", 1289 dm_device_name(table->md), 1290 (unsigned long long) ti->begin, 1291 (unsigned long long) ti->len); 1292 } 1293 1294 return validate_hardware_logical_block_alignment(table, limits); 1295} 1296 1297/* 1298 * Verify that all devices have an integrity profile that matches the 1299 * DM device's registered integrity profile. If the profiles don't 1300 * match then unregister the DM device's integrity profile. 1301 */ 1302static void dm_table_verify_integrity(struct dm_table *t) 1303{ 1304 struct gendisk *template_disk = NULL; 1305 1306 if (t->integrity_supported) { 1307 /* 1308 * Verify that the original integrity profile 1309 * matches all the devices in this table. 1310 */ 1311 template_disk = dm_table_get_integrity_disk(t); 1312 if (template_disk && 1313 blk_integrity_compare(dm_disk(t->md), template_disk) >= 0) 1314 return; 1315 } 1316 1317 if (integrity_profile_exists(dm_disk(t->md))) { 1318 DMWARN("%s: unable to establish an integrity profile", 1319 dm_device_name(t->md)); 1320 blk_integrity_unregister(dm_disk(t->md)); 1321 } 1322} 1323 1324static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev, 1325 sector_t start, sector_t len, void *data) 1326{ 1327 unsigned flush = (*(unsigned *)data); 1328 struct request_queue *q = bdev_get_queue(dev->bdev); 1329 1330 return q && (q->flush_flags & flush); 1331} 1332 1333static bool dm_table_supports_flush(struct dm_table *t, unsigned flush) 1334{ 1335 struct dm_target *ti; 1336 unsigned i = 0; 1337 1338 /* 1339 * Require at least one underlying device to support flushes. 1340 * t->devices includes internal dm devices such as mirror logs 1341 * so we need to use iterate_devices here, which targets 1342 * supporting flushes must provide. 1343 */ 1344 while (i < dm_table_get_num_targets(t)) { 1345 ti = dm_table_get_target(t, i++); 1346 1347 if (!ti->num_flush_bios) 1348 continue; 1349 1350 if (ti->flush_supported) 1351 return true; 1352 1353 if (ti->type->iterate_devices && 1354 ti->type->iterate_devices(ti, device_flush_capable, &flush)) 1355 return true; 1356 } 1357 1358 return false; 1359} 1360 1361static bool dm_table_discard_zeroes_data(struct dm_table *t) 1362{ 1363 struct dm_target *ti; 1364 unsigned i = 0; 1365 1366 /* Ensure that all targets supports discard_zeroes_data. */ 1367 while (i < dm_table_get_num_targets(t)) { 1368 ti = dm_table_get_target(t, i++); 1369 1370 if (ti->discard_zeroes_data_unsupported) 1371 return false; 1372 } 1373 1374 return true; 1375} 1376 1377static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev, 1378 sector_t start, sector_t len, void *data) 1379{ 1380 struct request_queue *q = bdev_get_queue(dev->bdev); 1381 1382 return q && blk_queue_nonrot(q); 1383} 1384 1385static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev, 1386 sector_t start, sector_t len, void *data) 1387{ 1388 struct request_queue *q = bdev_get_queue(dev->bdev); 1389 1390 return q && !blk_queue_add_random(q); 1391} 1392 1393static int queue_supports_sg_merge(struct dm_target *ti, struct dm_dev *dev, 1394 sector_t start, sector_t len, void *data) 1395{ 1396 struct request_queue *q = bdev_get_queue(dev->bdev); 1397 1398 return q && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags); 1399} 1400 1401static bool dm_table_all_devices_attribute(struct dm_table *t, 1402 iterate_devices_callout_fn func) 1403{ 1404 struct dm_target *ti; 1405 unsigned i = 0; 1406 1407 while (i < dm_table_get_num_targets(t)) { 1408 ti = dm_table_get_target(t, i++); 1409 1410 if (!ti->type->iterate_devices || 1411 !ti->type->iterate_devices(ti, func, NULL)) 1412 return false; 1413 } 1414 1415 return true; 1416} 1417 1418static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev, 1419 sector_t start, sector_t len, void *data) 1420{ 1421 struct request_queue *q = bdev_get_queue(dev->bdev); 1422 1423 return q && !q->limits.max_write_same_sectors; 1424} 1425 1426static bool dm_table_supports_write_same(struct dm_table *t) 1427{ 1428 struct dm_target *ti; 1429 unsigned i = 0; 1430 1431 while (i < dm_table_get_num_targets(t)) { 1432 ti = dm_table_get_target(t, i++); 1433 1434 if (!ti->num_write_same_bios) 1435 return false; 1436 1437 if (!ti->type->iterate_devices || 1438 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL)) 1439 return false; 1440 } 1441 1442 return true; 1443} 1444 1445static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev, 1446 sector_t start, sector_t len, void *data) 1447{ 1448 struct request_queue *q = bdev_get_queue(dev->bdev); 1449 1450 return q && blk_queue_discard(q); 1451} 1452 1453static bool dm_table_supports_discards(struct dm_table *t) 1454{ 1455 struct dm_target *ti; 1456 unsigned i = 0; 1457 1458 /* 1459 * Unless any target used by the table set discards_supported, 1460 * require at least one underlying device to support discards. 1461 * t->devices includes internal dm devices such as mirror logs 1462 * so we need to use iterate_devices here, which targets 1463 * supporting discard selectively must provide. 1464 */ 1465 while (i < dm_table_get_num_targets(t)) { 1466 ti = dm_table_get_target(t, i++); 1467 1468 if (!ti->num_discard_bios) 1469 continue; 1470 1471 if (ti->discards_supported) 1472 return true; 1473 1474 if (ti->type->iterate_devices && 1475 ti->type->iterate_devices(ti, device_discard_capable, NULL)) 1476 return true; 1477 } 1478 1479 return false; 1480} 1481 1482void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q, 1483 struct queue_limits *limits) 1484{ 1485 unsigned flush = 0; 1486 1487 /* 1488 * Copy table's limits to the DM device's request_queue 1489 */ 1490 q->limits = *limits; 1491 1492 if (!dm_table_supports_discards(t)) 1493 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); 1494 else 1495 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q); 1496 1497 if (dm_table_supports_flush(t, REQ_FLUSH)) { 1498 flush |= REQ_FLUSH; 1499 if (dm_table_supports_flush(t, REQ_FUA)) 1500 flush |= REQ_FUA; 1501 } 1502 blk_queue_flush(q, flush); 1503 1504 if (!dm_table_discard_zeroes_data(t)) 1505 q->limits.discard_zeroes_data = 0; 1506 1507 /* Ensure that all underlying devices are non-rotational. */ 1508 if (dm_table_all_devices_attribute(t, device_is_nonrot)) 1509 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q); 1510 else 1511 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q); 1512 1513 if (!dm_table_supports_write_same(t)) 1514 q->limits.max_write_same_sectors = 0; 1515 1516 if (dm_table_all_devices_attribute(t, queue_supports_sg_merge)) 1517 queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q); 1518 else 1519 queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q); 1520 1521 dm_table_verify_integrity(t); 1522 1523 /* 1524 * Determine whether or not this queue's I/O timings contribute 1525 * to the entropy pool, Only request-based targets use this. 1526 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not 1527 * have it set. 1528 */ 1529 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random)) 1530 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q); 1531 1532 /* 1533 * QUEUE_FLAG_STACKABLE must be set after all queue settings are 1534 * visible to other CPUs because, once the flag is set, incoming bios 1535 * are processed by request-based dm, which refers to the queue 1536 * settings. 1537 * Until the flag set, bios are passed to bio-based dm and queued to 1538 * md->deferred where queue settings are not needed yet. 1539 * Those bios are passed to request-based dm at the resume time. 1540 */ 1541 smp_mb(); 1542 if (dm_table_request_based(t)) 1543 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q); 1544} 1545 1546unsigned int dm_table_get_num_targets(struct dm_table *t) 1547{ 1548 return t->num_targets; 1549} 1550 1551struct list_head *dm_table_get_devices(struct dm_table *t) 1552{ 1553 return &t->devices; 1554} 1555 1556fmode_t dm_table_get_mode(struct dm_table *t) 1557{ 1558 return t->mode; 1559} 1560EXPORT_SYMBOL(dm_table_get_mode); 1561 1562enum suspend_mode { 1563 PRESUSPEND, 1564 PRESUSPEND_UNDO, 1565 POSTSUSPEND, 1566}; 1567 1568static void suspend_targets(struct dm_table *t, enum suspend_mode mode) 1569{ 1570 int i = t->num_targets; 1571 struct dm_target *ti = t->targets; 1572 1573 while (i--) { 1574 switch (mode) { 1575 case PRESUSPEND: 1576 if (ti->type->presuspend) 1577 ti->type->presuspend(ti); 1578 break; 1579 case PRESUSPEND_UNDO: 1580 if (ti->type->presuspend_undo) 1581 ti->type->presuspend_undo(ti); 1582 break; 1583 case POSTSUSPEND: 1584 if (ti->type->postsuspend) 1585 ti->type->postsuspend(ti); 1586 break; 1587 } 1588 ti++; 1589 } 1590} 1591 1592void dm_table_presuspend_targets(struct dm_table *t) 1593{ 1594 if (!t) 1595 return; 1596 1597 suspend_targets(t, PRESUSPEND); 1598} 1599 1600void dm_table_presuspend_undo_targets(struct dm_table *t) 1601{ 1602 if (!t) 1603 return; 1604 1605 suspend_targets(t, PRESUSPEND_UNDO); 1606} 1607 1608void dm_table_postsuspend_targets(struct dm_table *t) 1609{ 1610 if (!t) 1611 return; 1612 1613 suspend_targets(t, POSTSUSPEND); 1614} 1615 1616int dm_table_resume_targets(struct dm_table *t) 1617{ 1618 int i, r = 0; 1619 1620 for (i = 0; i < t->num_targets; i++) { 1621 struct dm_target *ti = t->targets + i; 1622 1623 if (!ti->type->preresume) 1624 continue; 1625 1626 r = ti->type->preresume(ti); 1627 if (r) { 1628 DMERR("%s: %s: preresume failed, error = %d", 1629 dm_device_name(t->md), ti->type->name, r); 1630 return r; 1631 } 1632 } 1633 1634 for (i = 0; i < t->num_targets; i++) { 1635 struct dm_target *ti = t->targets + i; 1636 1637 if (ti->type->resume) 1638 ti->type->resume(ti); 1639 } 1640 1641 return 0; 1642} 1643 1644void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb) 1645{ 1646 list_add(&cb->list, &t->target_callbacks); 1647} 1648EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks); 1649 1650int dm_table_any_congested(struct dm_table *t, int bdi_bits) 1651{ 1652 struct dm_dev_internal *dd; 1653 struct list_head *devices = dm_table_get_devices(t); 1654 struct dm_target_callbacks *cb; 1655 int r = 0; 1656 1657 list_for_each_entry(dd, devices, list) { 1658 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev); 1659 char b[BDEVNAME_SIZE]; 1660 1661 if (likely(q)) 1662 r |= bdi_congested(&q->backing_dev_info, bdi_bits); 1663 else 1664 DMWARN_LIMIT("%s: any_congested: nonexistent device %s", 1665 dm_device_name(t->md), 1666 bdevname(dd->dm_dev->bdev, b)); 1667 } 1668 1669 list_for_each_entry(cb, &t->target_callbacks, list) 1670 if (cb->congested_fn) 1671 r |= cb->congested_fn(cb, bdi_bits); 1672 1673 return r; 1674} 1675 1676struct mapped_device *dm_table_get_md(struct dm_table *t) 1677{ 1678 return t->md; 1679} 1680EXPORT_SYMBOL(dm_table_get_md); 1681 1682void dm_table_run_md_queue_async(struct dm_table *t) 1683{ 1684 struct mapped_device *md; 1685 struct request_queue *queue; 1686 unsigned long flags; 1687 1688 if (!dm_table_request_based(t)) 1689 return; 1690 1691 md = dm_table_get_md(t); 1692 queue = dm_get_md_queue(md); 1693 if (queue) { 1694 if (queue->mq_ops) 1695 blk_mq_run_hw_queues(queue, true); 1696 else { 1697 spin_lock_irqsave(queue->queue_lock, flags); 1698 blk_run_queue_async(queue); 1699 spin_unlock_irqrestore(queue->queue_lock, flags); 1700 } 1701 } 1702} 1703EXPORT_SYMBOL(dm_table_run_md_queue_async); 1704 1705