drivers/md/dm-table.c

   1 /*
   2  * Copyright (C) 2001 Sistina Software (UK) Limited.
   3  * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
   4  *
   5  * This file is released under the GPL.
   6  */
   7
   8 #include "dm.h"
   9
  10 #include <linux/module.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/blkdev.h>
  13 #include <linux/namei.h>
  14 #include <linux/ctype.h>
  15 #include <linux/string.h>
  16 #include <linux/slab.h>
  17 #include <linux/interrupt.h>
  18 #include <linux/mutex.h>
  19 #include <linux/delay.h>
  20 #include <linux/atomic.h>
  21 #include <linux/blk-mq.h>
  22 #include <linux/mount.h>
  23
  24 #define DM_MSG_PREFIX "table"
  25
  26 #define MAX_DEPTH 16
  27 #define NODE_SIZE L1_CACHE_BYTES
  28 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
  29 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
  30
  31 struct dm_table {
  32         struct mapped_device *md;
  33         unsigned type;
  34
  35         /* btree table */
  36         unsigned int depth;
  37         unsigned int counts[MAX_DEPTH]; /* in nodes */
  38         sector_t *index[MAX_DEPTH];
  39
  40         unsigned int num_targets;
  41         unsigned int num_allocated;
  42         sector_t *highs;
  43         struct dm_target *targets;
  44
  45         struct target_type *immutable_target_type;
  46         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  */
  71 static 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  */
  86 static 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  */
  94 static 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  */
 104 static 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  */
 119 static 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
 134 void *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 }
 150 EXPORT_SYMBOL(dm_vcalloc);
 151
 152 /*
 153  * highs, and targets are managed as dynamic arrays during a
 154  * table load.
 155  */
 156 static 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
 183 int 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
 215 static 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
 229 void 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  */
 263 static 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  */
 277 static 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  */
 348 static 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  * Add a device to the list, or just increment the usage count if
 369  * it's already present.
 370  */
 371 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
 372                   struct dm_dev **result)
 373 {
 374         int r;
 375         dev_t uninitialized_var(dev);
 376         struct dm_dev_internal *dd;
 377         struct dm_table *t = ti->table;
 378         struct block_device *bdev;
 379
 380         BUG_ON(!t);
 381
 382         /* convert the path to a device */
 383         bdev = lookup_bdev(path);
 384         if (IS_ERR(bdev)) {
 385                 dev = name_to_dev_t(path);
 386                 if (!dev)
 387                         return -ENODEV;
 388         } else {
 389                 dev = bdev->bd_dev;
 390                 bdput(bdev);
 391         }
 392
 393         dd = find_device(&t->devices, dev);
 394         if (!dd) {
 395                 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
 396                 if (!dd)
 397                         return -ENOMEM;
 398
 399                 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
 400                         kfree(dd);
 401                         return r;
 402                 }
 403
 404                 atomic_set(&dd->count, 0);
 405                 list_add(&dd->list, &t->devices);
 406
 407         } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
 408                 r = upgrade_mode(dd, mode, t->md);
 409                 if (r)
 410                         return r;
 411         }
 412         atomic_inc(&dd->count);
 413
 414         *result = dd->dm_dev;
 415         return 0;
 416 }
 417 EXPORT_SYMBOL(dm_get_device);
 418
 419 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
 420                                 sector_t start, sector_t len, void *data)
 421 {
 422         struct queue_limits *limits = data;
 423         struct block_device *bdev = dev->bdev;
 424         struct request_queue *q = bdev_get_queue(bdev);
 425         char b[BDEVNAME_SIZE];
 426
 427         if (unlikely(!q)) {
 428                 DMWARN("%s: Cannot set limits for nonexistent device %s",
 429                        dm_device_name(ti->table->md), bdevname(bdev, b));
 430                 return 0;
 431         }
 432
 433         if (bdev_stack_limits(limits, bdev, start) < 0)
 434                 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
 435                        "physical_block_size=%u, logical_block_size=%u, "
 436                        "alignment_offset=%u, start=%llu",
 437                        dm_device_name(ti->table->md), bdevname(bdev, b),
 438                        q->limits.physical_block_size,
 439                        q->limits.logical_block_size,
 440                        q->limits.alignment_offset,
 441                        (unsigned long long) start << SECTOR_SHIFT);
 442
 443         return 0;
 444 }
 445
 446 /*
 447  * Decrement a device's use count and remove it if necessary.
 448  */
 449 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
 450 {
 451         int found = 0;
 452         struct list_head *devices = &ti->table->devices;
 453         struct dm_dev_internal *dd;
 454
 455         list_for_each_entry(dd, devices, list) {
 456                 if (dd->dm_dev == d) {
 457                         found = 1;
 458                         break;
 459                 }
 460         }
 461         if (!found) {
 462                 DMWARN("%s: device %s not in table devices list",
 463                        dm_device_name(ti->table->md), d->name);
 464                 return;
 465         }
 466         if (atomic_dec_and_test(&dd->count)) {
 467                 dm_put_table_device(ti->table->md, d);
 468                 list_del(&dd->list);
 469                 kfree(dd);
 470         }
 471 }
 472 EXPORT_SYMBOL(dm_put_device);
 473
 474 /*
 475  * Checks to see if the target joins onto the end of the table.
 476  */
 477 static int adjoin(struct dm_table *table, struct dm_target *ti)
 478 {
 479         struct dm_target *prev;
 480
 481         if (!table->num_targets)
 482                 return !ti->begin;
 483
 484         prev = &table->targets[table->num_targets - 1];
 485         return (ti->begin == (prev->begin + prev->len));
 486 }
 487
 488 /*
 489  * Used to dynamically allocate the arg array.
 490  *
 491  * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
 492  * process messages even if some device is suspended. These messages have a
 493  * small fixed number of arguments.
 494  *
 495  * On the other hand, dm-switch needs to process bulk data using messages and
 496  * excessive use of GFP_NOIO could cause trouble.
 497  */
 498 static char **realloc_argv(unsigned *array_size, char **old_argv)
 499 {
 500         char **argv;
 501         unsigned new_size;
 502         gfp_t gfp;
 503
 504         if (*array_size) {
 505                 new_size = *array_size * 2;
 506                 gfp = GFP_KERNEL;
 507         } else {
 508                 new_size = 8;
 509                 gfp = GFP_NOIO;
 510         }
 511         argv = kmalloc(new_size * sizeof(*argv), gfp);
 512         if (argv) {
 513                 memcpy(argv, old_argv, *array_size * sizeof(*argv));
 514                 *array_size = new_size;
 515         }
 516
 517         kfree(old_argv);
 518         return argv;
 519 }
 520
 521 /*
 522  * Destructively splits up the argument list to pass to ctr.
 523  */
 524 int dm_split_args(int *argc, char ***argvp, char *input)
 525 {
 526         char *start, *end = input, *out, **argv = NULL;
 527         unsigned array_size = 0;
 528
 529         *argc = 0;
 530
 531         if (!input) {
 532                 *argvp = NULL;
 533                 return 0;
 534         }
 535
 536         argv = realloc_argv(&array_size, argv);
 537         if (!argv)
 538                 return -ENOMEM;
 539
 540         while (1) {
 541                 /* Skip whitespace */
 542                 start = skip_spaces(end);
 543
 544                 if (!*start)
 545                         break;  /* success, we hit the end */
 546
 547                 /* 'out' is used to remove any back-quotes */
 548                 end = out = start;
 549                 while (*end) {
 550                         /* Everything apart from '\0' can be quoted */
 551                         if (*end == '\\' && *(end + 1)) {
 552                                 *out++ = *(end + 1);
 553                                 end += 2;
 554                                 continue;
 555                         }
 556
 557                         if (isspace(*end))
 558                                 break;  /* end of token */
 559
 560                         *out++ = *end++;
 561                 }
 562
 563                 /* have we already filled the array ? */
 564                 if ((*argc + 1) > array_size) {
 565                         argv = realloc_argv(&array_size, argv);
 566                         if (!argv)
 567                                 return -ENOMEM;
 568                 }
 569
 570                 /* we know this is whitespace */
 571                 if (*end)
 572                         end++;
 573
 574                 /* terminate the string and put it in the array */
 575                 *out = '\0';
 576                 argv[*argc] = start;
 577                 (*argc)++;
 578         }
 579
 580         *argvp = argv;
 581         return 0;
 582 }
 583
 584 /*
 585  * Impose necessary and sufficient conditions on a devices's table such
 586  * that any incoming bio which respects its logical_block_size can be
 587  * processed successfully.  If it falls across the boundary between
 588  * two or more targets, the size of each piece it gets split into must
 589  * be compatible with the logical_block_size of the target processing it.
 590  */
 591 static int validate_hardware_logical_block_alignment(struct dm_table *table,
 592                                                  struct queue_limits *limits)
 593 {
 594         /*
 595          * This function uses arithmetic modulo the logical_block_size
 596          * (in units of 512-byte sectors).
 597          */
 598         unsigned short device_logical_block_size_sects =
 599                 limits->logical_block_size >> SECTOR_SHIFT;
 600
 601         /*
 602          * Offset of the start of the next table entry, mod logical_block_size.
 603          */
 604         unsigned short next_target_start = 0;
 605
 606         /*
 607          * Given an aligned bio that extends beyond the end of a
 608          * target, how many sectors must the next target handle?
 609          */
 610         unsigned short remaining = 0;
 611
 612         struct dm_target *uninitialized_var(ti);
 613         struct queue_limits ti_limits;
 614         unsigned i = 0;
 615
 616         /*
 617          * Check each entry in the table in turn.
 618          */
 619         while (i < dm_table_get_num_targets(table)) {
 620                 ti = dm_table_get_target(table, i++);
 621
 622                 blk_set_stacking_limits(&ti_limits);
 623
 624                 /* combine all target devices' limits */
 625                 if (ti->type->iterate_devices)
 626                         ti->type->iterate_devices(ti, dm_set_device_limits,
 627                                                   &ti_limits);
 628
 629                 /*
 630                  * If the remaining sectors fall entirely within this
 631                  * table entry are they compatible with its logical_block_size?
 632                  */
 633                 if (remaining < ti->len &&
 634                     remaining & ((ti_limits.logical_block_size >>
 635                                   SECTOR_SHIFT) - 1))
 636                         break;  /* Error */
 637
 638                 next_target_start =
 639                     (unsigned short) ((next_target_start + ti->len) &
 640                                       (device_logical_block_size_sects - 1));
 641                 remaining = next_target_start ?
 642                     device_logical_block_size_sects - next_target_start : 0;
 643         }
 644
 645         if (remaining) {
 646                 DMWARN("%s: table line %u (start sect %llu len %llu) "
 647                        "not aligned to h/w logical block size %u",
 648                        dm_device_name(table->md), i,
 649                        (unsigned long long) ti->begin,
 650                        (unsigned long long) ti->len,
 651                        limits->logical_block_size);
 652                 return -EINVAL;
 653         }
 654
 655         return 0;
 656 }
 657
 658 int dm_table_add_target(struct dm_table *t, const char *type,
 659                         sector_t start, sector_t len, char *params)
 660 {
 661         int r = -EINVAL, argc;
 662         char **argv;
 663         struct dm_target *tgt;
 664
 665         if (t->singleton) {
 666                 DMERR("%s: target type %s must appear alone in table",
 667                       dm_device_name(t->md), t->targets->type->name);
 668                 return -EINVAL;
 669         }
 670
 671         BUG_ON(t->num_targets >= t->num_allocated);
 672
 673         tgt = t->targets + t->num_targets;
 674         memset(tgt, 0, sizeof(*tgt));
 675
 676         if (!len) {
 677                 DMERR("%s: zero-length target", dm_device_name(t->md));
 678                 return -EINVAL;
 679         }
 680
 681         tgt->type = dm_get_target_type(type);
 682         if (!tgt->type) {
 683                 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
 684                       type);
 685                 return -EINVAL;
 686         }
 687
 688         if (dm_target_needs_singleton(tgt->type)) {
 689                 if (t->num_targets) {
 690                         DMERR("%s: target type %s must appear alone in table",
 691                               dm_device_name(t->md), type);
 692                         return -EINVAL;
 693                 }
 694                 t->singleton = 1;
 695         }
 696
 697         if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
 698                 DMERR("%s: target type %s may not be included in read-only tables",
 699                       dm_device_name(t->md), type);
 700                 return -EINVAL;
 701         }
 702
 703         if (t->immutable_target_type) {
 704                 if (t->immutable_target_type != tgt->type) {
 705                         DMERR("%s: immutable target type %s cannot be mixed with other target types",
 706                               dm_device_name(t->md), t->immutable_target_type->name);
 707                         return -EINVAL;
 708                 }
 709         } else if (dm_target_is_immutable(tgt->type)) {
 710                 if (t->num_targets) {
 711                         DMERR("%s: immutable target type %s cannot be mixed with other target types",
 712                               dm_device_name(t->md), tgt->type->name);
 713                         return -EINVAL;
 714                 }
 715                 t->immutable_target_type = tgt->type;
 716         }
 717
 718         tgt->table = t;
 719         tgt->begin = start;
 720         tgt->len = len;
 721         tgt->error = "Unknown error";
 722
 723         /*
 724          * Does this target adjoin the previous one ?
 725          */
 726         if (!adjoin(t, tgt)) {
 727                 tgt->error = "Gap in table";
 728                 r = -EINVAL;
 729                 goto bad;
 730         }
 731
 732         r = dm_split_args(&argc, &argv, params);
 733         if (r) {
 734                 tgt->error = "couldn't split parameters (insufficient memory)";
 735                 goto bad;
 736         }
 737
 738         r = tgt->type->ctr(tgt, argc, argv);
 739         kfree(argv);
 740         if (r)
 741                 goto bad;
 742
 743         t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
 744
 745         if (!tgt->num_discard_bios && tgt->discards_supported)
 746                 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
 747                        dm_device_name(t->md), type);
 748
 749         return 0;
 750
 751  bad:
 752         DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
 753         dm_put_target_type(tgt->type);
 754         return r;
 755 }
 756
 757 /*
 758  * Target argument parsing helpers.
 759  */
 760 static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
 761                              unsigned *value, char **error, unsigned grouped)
 762 {
 763         const char *arg_str = dm_shift_arg(arg_set);
 764         char dummy;
 765
 766         if (!arg_str ||
 767             (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
 768             (*value < arg->min) ||
 769             (*value > arg->max) ||
 770             (grouped && arg_set->argc < *value)) {
 771                 *error = arg->error;
 772                 return -EINVAL;
 773         }
 774
 775         return 0;
 776 }
 777
 778 int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
 779                 unsigned *value, char **error)
 780 {
 781         return validate_next_arg(arg, arg_set, value, error, 0);
 782 }
 783 EXPORT_SYMBOL(dm_read_arg);
 784
 785 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
 786                       unsigned *value, char **error)
 787 {
 788         return validate_next_arg(arg, arg_set, value, error, 1);
 789 }
 790 EXPORT_SYMBOL(dm_read_arg_group);
 791
 792 const char *dm_shift_arg(struct dm_arg_set *as)
 793 {
 794         char *r;
 795
 796         if (as->argc) {
 797                 as->argc--;
 798                 r = *as->argv;
 799                 as->argv++;
 800                 return r;
 801         }
 802
 803         return NULL;
 804 }
 805 EXPORT_SYMBOL(dm_shift_arg);
 806
 807 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
 808 {
 809         BUG_ON(as->argc < num_args);
 810         as->argc -= num_args;
 811         as->argv += num_args;
 812 }
 813 EXPORT_SYMBOL(dm_consume_args);
 814
 815 static bool __table_type_request_based(unsigned table_type)
 816 {
 817         return (table_type == DM_TYPE_REQUEST_BASED ||
 818                 table_type == DM_TYPE_MQ_REQUEST_BASED);
 819 }
 820
 821 static int dm_table_set_type(struct dm_table *t)
 822 {
 823         unsigned i;
 824         unsigned bio_based = 0, request_based = 0, hybrid = 0;
 825         bool use_blk_mq = false;
 826         struct dm_target *tgt;
 827         struct dm_dev_internal *dd;
 828         struct list_head *devices;
 829         unsigned live_md_type = dm_get_md_type(t->md);
 830
 831         for (i = 0; i < t->num_targets; i++) {
 832                 tgt = t->targets + i;
 833                 if (dm_target_hybrid(tgt))
 834                         hybrid = 1;
 835                 else if (dm_target_request_based(tgt))
 836                         request_based = 1;
 837                 else
 838                         bio_based = 1;
 839
 840                 if (bio_based && request_based) {
 841                         DMWARN("Inconsistent table: different target types"
 842                                " can't be mixed up");
 843                         return -EINVAL;
 844                 }
 845         }
 846
 847         if (hybrid && !bio_based && !request_based) {
 848                 /*
 849                  * The targets can work either way.
 850                  * Determine the type from the live device.
 851                  * Default to bio-based if device is new.
 852                  */
 853                 if (__table_type_request_based(live_md_type))
 854                         request_based = 1;
 855                 else
 856                         bio_based = 1;
 857         }
 858
 859         if (bio_based) {
 860                 /* We must use this table as bio-based */
 861                 t->type = DM_TYPE_BIO_BASED;
 862                 return 0;
 863         }
 864
 865         BUG_ON(!request_based); /* No targets in this table */
 866
 867         /*
 868          * Request-based dm supports only tables that have a single target now.
 869          * To support multiple targets, request splitting support is needed,
 870          * and that needs lots of changes in the block-layer.
 871          * (e.g. request completion process for partial completion.)
 872          */
 873         if (t->num_targets > 1) {
 874                 DMWARN("Request-based dm doesn't support multiple targets yet");
 875                 return -EINVAL;
 876         }
 877
 878         /* Non-request-stackable devices can't be used for request-based dm */
 879         devices = dm_table_get_devices(t);
 880         list_for_each_entry(dd, devices, list) {
 881                 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
 882
 883                 if (!blk_queue_stackable(q)) {
 884                         DMERR("table load rejected: including"
 885                               " non-request-stackable devices");
 886                         return -EINVAL;
 887                 }
 888
 889                 if (q->mq_ops)
 890                         use_blk_mq = true;
 891         }
 892
 893         if (use_blk_mq) {
 894                 /* verify _all_ devices in the table are blk-mq devices */
 895                 list_for_each_entry(dd, devices, list)
 896                         if (!bdev_get_queue(dd->dm_dev->bdev)->mq_ops) {
 897                                 DMERR("table load rejected: not all devices"
 898                                       " are blk-mq request-stackable");
 899                                 return -EINVAL;
 900                         }
 901                 t->type = DM_TYPE_MQ_REQUEST_BASED;
 902
 903         } else if (list_empty(devices) && __table_type_request_based(live_md_type)) {
 904                 /* inherit live MD type */
 905                 t->type = live_md_type;
 906
 907         } else
 908                 t->type = DM_TYPE_REQUEST_BASED;
 909
 910         return 0;
 911 }
 912
 913 unsigned dm_table_get_type(struct dm_table *t)
 914 {
 915         return t->type;
 916 }
 917
 918 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
 919 {
 920         return t->immutable_target_type;
 921 }
 922
 923 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
 924 {
 925         /* Immutable target is implicitly a singleton */
 926         if (t->num_targets > 1 ||
 927             !dm_target_is_immutable(t->targets[0].type))
 928                 return NULL;
 929
 930         return t->targets;
 931 }
 932
 933 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
 934 {
 935         struct dm_target *uninitialized_var(ti);
 936         unsigned i = 0;
 937
 938         while (i < dm_table_get_num_targets(t)) {
 939                 ti = dm_table_get_target(t, i++);
 940                 if (dm_target_is_wildcard(ti->type))
 941                         return ti;
 942         }
 943
 944         return NULL;
 945 }
 946
 947 bool dm_table_request_based(struct dm_table *t)
 948 {
 949         return __table_type_request_based(dm_table_get_type(t));
 950 }
 951
 952 bool dm_table_mq_request_based(struct dm_table *t)
 953 {
 954         return dm_table_get_type(t) == DM_TYPE_MQ_REQUEST_BASED;
 955 }
 956
 957 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
 958 {
 959         unsigned type = dm_table_get_type(t);
 960         unsigned per_bio_data_size = 0;
 961         struct dm_target *tgt;
 962         unsigned i;
 963
 964         if (unlikely(type == DM_TYPE_NONE)) {
 965                 DMWARN("no table type is set, can't allocate mempools");
 966                 return -EINVAL;
 967         }
 968
 969         if (type == DM_TYPE_BIO_BASED)
 970                 for (i = 0; i < t->num_targets; i++) {
 971                         tgt = t->targets + i;
 972                         per_bio_data_size = max(per_bio_data_size, tgt->per_bio_data_size);
 973                 }
 974
 975         t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported, per_bio_data_size);
 976         if (!t->mempools)
 977                 return -ENOMEM;
 978
 979         return 0;
 980 }
 981
 982 void dm_table_free_md_mempools(struct dm_table *t)
 983 {
 984         dm_free_md_mempools(t->mempools);
 985         t->mempools = NULL;
 986 }
 987
 988 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
 989 {
 990         return t->mempools;
 991 }
 992
 993 static int setup_indexes(struct dm_table *t)
 994 {
 995         int i;
 996         unsigned int total = 0;
 997         sector_t *indexes;
 998
 999         /* allocate the space for *all* the indexes */
1000         for (i = t->depth - 2; i >= 0; i--) {
1001                 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1002                 total += t->counts[i];
1003         }
1004
1005         indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1006         if (!indexes)
1007                 return -ENOMEM;
1008
1009         /* set up internal nodes, bottom-up */
1010         for (i = t->depth - 2; i >= 0; i--) {
1011                 t->index[i] = indexes;
1012                 indexes += (KEYS_PER_NODE * t->counts[i]);
1013                 setup_btree_index(i, t);
1014         }
1015
1016         return 0;
1017 }
1018
1019 /*
1020  * Builds the btree to index the map.
1021  */
1022 static int dm_table_build_index(struct dm_table *t)
1023 {
1024         int r = 0;
1025         unsigned int leaf_nodes;
1026
1027         /* how many indexes will the btree have ? */
1028         leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1029         t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1030
1031         /* leaf layer has already been set up */
1032         t->counts[t->depth - 1] = leaf_nodes;
1033         t->index[t->depth - 1] = t->highs;
1034
1035         if (t->depth >= 2)
1036                 r = setup_indexes(t);
1037
1038         return r;
1039 }
1040
1041 static bool integrity_profile_exists(struct gendisk *disk)
1042 {
1043         return !!blk_get_integrity(disk);
1044 }
1045
1046 /*
1047  * Get a disk whose integrity profile reflects the table's profile.
1048  * Returns NULL if integrity support was inconsistent or unavailable.
1049  */
1050 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1051 {
1052         struct list_head *devices = dm_table_get_devices(t);
1053         struct dm_dev_internal *dd = NULL;
1054         struct gendisk *prev_disk = NULL, *template_disk = NULL;
1055
1056         list_for_each_entry(dd, devices, list) {
1057                 template_disk = dd->dm_dev->bdev->bd_disk;
1058                 if (!integrity_profile_exists(template_disk))
1059                         goto no_integrity;
1060                 else if (prev_disk &&
1061                          blk_integrity_compare(prev_disk, template_disk) < 0)
1062                         goto no_integrity;
1063                 prev_disk = template_disk;
1064         }
1065
1066         return template_disk;
1067
1068 no_integrity:
1069         if (prev_disk)
1070                 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1071                        dm_device_name(t->md),
1072                        prev_disk->disk_name,
1073                        template_disk->disk_name);
1074         return NULL;
1075 }
1076
1077 /*
1078  * Register the mapped device for blk_integrity support if the
1079  * underlying devices have an integrity profile.  But all devices may
1080  * not have matching profiles (checking all devices isn't reliable
1081  * during table load because this table may use other DM device(s) which
1082  * must be resumed before they will have an initialized integity
1083  * profile).  Consequently, stacked DM devices force a 2 stage integrity
1084  * profile validation: First pass during table load, final pass during
1085  * resume.
1086  */
1087 static int dm_table_register_integrity(struct dm_table *t)
1088 {
1089         struct mapped_device *md = t->md;
1090         struct gendisk *template_disk = NULL;
1091
1092         template_disk = dm_table_get_integrity_disk(t);
1093         if (!template_disk)
1094                 return 0;
1095
1096         if (!integrity_profile_exists(dm_disk(md))) {
1097                 t->integrity_supported = 1;
1098                 /*
1099                  * Register integrity profile during table load; we can do
1100                  * this because the final profile must match during resume.
1101                  */
1102                 blk_integrity_register(dm_disk(md),
1103                                        blk_get_integrity(template_disk));
1104                 return 0;
1105         }
1106
1107         /*
1108          * If DM device already has an initialized integrity
1109          * profile the new profile should not conflict.
1110          */
1111         if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1112                 DMWARN("%s: conflict with existing integrity profile: "
1113                        "%s profile mismatch",
1114                        dm_device_name(t->md),
1115                        template_disk->disk_name);
1116                 return 1;
1117         }
1118
1119         /* Preserve existing integrity profile */
1120         t->integrity_supported = 1;
1121         return 0;
1122 }
1123
1124 /*
1125  * Prepares the table for use by building the indices,
1126  * setting the type, and allocating mempools.
1127  */
1128 int dm_table_complete(struct dm_table *t)
1129 {
1130         int r;
1131
1132         r = dm_table_set_type(t);
1133         if (r) {
1134                 DMERR("unable to set table type");
1135                 return r;
1136         }
1137
1138         r = dm_table_build_index(t);
1139         if (r) {
1140                 DMERR("unable to build btrees");
1141                 return r;
1142         }
1143
1144         r = dm_table_register_integrity(t);
1145         if (r) {
1146                 DMERR("could not register integrity profile.");
1147                 return r;
1148         }
1149
1150         r = dm_table_alloc_md_mempools(t, t->md);
1151         if (r)
1152                 DMERR("unable to allocate mempools");
1153
1154         return r;
1155 }
1156
1157 static DEFINE_MUTEX(_event_lock);
1158 void dm_table_event_callback(struct dm_table *t,
1159                              void (*fn)(void *), void *context)
1160 {
1161         mutex_lock(&_event_lock);
1162         t->event_fn = fn;
1163         t->event_context = context;
1164         mutex_unlock(&_event_lock);
1165 }
1166
1167 void dm_table_event(struct dm_table *t)
1168 {
1169         /*
1170          * You can no longer call dm_table_event() from interrupt
1171          * context, use a bottom half instead.
1172          */
1173         BUG_ON(in_interrupt());
1174
1175         mutex_lock(&_event_lock);
1176         if (t->event_fn)
1177                 t->event_fn(t->event_context);
1178         mutex_unlock(&_event_lock);
1179 }
1180 EXPORT_SYMBOL(dm_table_event);
1181
1182 sector_t dm_table_get_size(struct dm_table *t)
1183 {
1184         return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1185 }
1186 EXPORT_SYMBOL(dm_table_get_size);
1187
1188 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1189 {
1190         if (index >= t->num_targets)
1191                 return NULL;
1192
1193         return t->targets + index;
1194 }
1195
1196 /*
1197  * Search the btree for the correct target.
1198  *
1199  * Caller should check returned pointer with dm_target_is_valid()
1200  * to trap I/O beyond end of device.
1201  */
1202 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1203 {
1204         unsigned int l, n = 0, k = 0;
1205         sector_t *node;
1206
1207         for (l = 0; l < t->depth; l++) {
1208                 n = get_child(n, k);
1209                 node = get_node(t, l, n);
1210
1211                 for (k = 0; k < KEYS_PER_NODE; k++)
1212                         if (node[k] >= sector)
1213                                 break;
1214         }
1215
1216         return &t->targets[(KEYS_PER_NODE * n) + k];
1217 }
1218
1219 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1220                         sector_t start, sector_t len, void *data)
1221 {
1222         unsigned *num_devices = data;
1223
1224         (*num_devices)++;
1225
1226         return 0;
1227 }
1228
1229 /*
1230  * Check whether a table has no data devices attached using each
1231  * target's iterate_devices method.
1232  * Returns false if the result is unknown because a target doesn't
1233  * support iterate_devices.
1234  */
1235 bool dm_table_has_no_data_devices(struct dm_table *table)
1236 {
1237         struct dm_target *uninitialized_var(ti);
1238         unsigned i = 0, num_devices = 0;
1239
1240         while (i < dm_table_get_num_targets(table)) {
1241                 ti = dm_table_get_target(table, i++);
1242
1243                 if (!ti->type->iterate_devices)
1244                         return false;
1245
1246                 ti->type->iterate_devices(ti, count_device, &num_devices);
1247                 if (num_devices)
1248                         return false;
1249         }
1250
1251         return true;
1252 }
1253
1254 /*
1255  * Establish the new table's queue_limits and validate them.
1256  */
1257 int dm_calculate_queue_limits(struct dm_table *table,
1258                               struct queue_limits *limits)
1259 {
1260         struct dm_target *uninitialized_var(ti);
1261         struct queue_limits ti_limits;
1262         unsigned i = 0;
1263
1264         blk_set_stacking_limits(limits);
1265
1266         while (i < dm_table_get_num_targets(table)) {
1267                 blk_set_stacking_limits(&ti_limits);
1268
1269                 ti = dm_table_get_target(table, i++);
1270
1271                 if (!ti->type->iterate_devices)
1272                         goto combine_limits;
1273
1274                 /*
1275                  * Combine queue limits of all the devices this target uses.
1276                  */
1277                 ti->type->iterate_devices(ti, dm_set_device_limits,
1278                                           &ti_limits);
1279
1280                 /* Set I/O hints portion of queue limits */
1281                 if (ti->type->io_hints)
1282                         ti->type->io_hints(ti, &ti_limits);
1283
1284                 /*
1285                  * Check each device area is consistent with the target's
1286                  * overall queue limits.
1287                  */
1288                 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1289                                               &ti_limits))
1290                         return -EINVAL;
1291
1292 combine_limits:
1293                 /*
1294                  * Merge this target's queue limits into the overall limits
1295                  * for the table.
1296                  */
1297                 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1298                         DMWARN("%s: adding target device "
1299                                "(start sect %llu len %llu) "
1300                                "caused an alignment inconsistency",
1301                                dm_device_name(table->md),
1302                                (unsigned long long) ti->begin,
1303                                (unsigned long long) ti->len);
1304         }
1305
1306         return validate_hardware_logical_block_alignment(table, limits);
1307 }
1308
1309 /*
1310  * Verify that all devices have an integrity profile that matches the
1311  * DM device's registered integrity profile.  If the profiles don't
1312  * match then unregister the DM device's integrity profile.
1313  */
1314 static void dm_table_verify_integrity(struct dm_table *t)
1315 {
1316         struct gendisk *template_disk = NULL;
1317
1318         if (t->integrity_supported) {
1319                 /*
1320                  * Verify that the original integrity profile
1321                  * matches all the devices in this table.
1322                  */
1323                 template_disk = dm_table_get_integrity_disk(t);
1324                 if (template_disk &&
1325                     blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1326                         return;
1327         }
1328
1329         if (integrity_profile_exists(dm_disk(t->md))) {
1330                 DMWARN("%s: unable to establish an integrity profile",
1331                        dm_device_name(t->md));
1332                 blk_integrity_unregister(dm_disk(t->md));
1333         }
1334 }
1335
1336 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1337                                 sector_t start, sector_t len, void *data)
1338 {
1339         unsigned flush = (*(unsigned *)data);
1340         struct request_queue *q = bdev_get_queue(dev->bdev);
1341
1342         return q && (q->flush_flags & flush);
1343 }
1344
1345 static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1346 {
1347         struct dm_target *ti;
1348         unsigned i = 0;
1349
1350         /*
1351          * Require at least one underlying device to support flushes.
1352          * t->devices includes internal dm devices such as mirror logs
1353          * so we need to use iterate_devices here, which targets
1354          * supporting flushes must provide.
1355          */
1356         while (i < dm_table_get_num_targets(t)) {
1357                 ti = dm_table_get_target(t, i++);
1358
1359                 if (!ti->num_flush_bios)
1360                         continue;
1361
1362                 if (ti->flush_supported)
1363                         return true;
1364
1365                 if (ti->type->iterate_devices &&
1366                     ti->type->iterate_devices(ti, device_flush_capable, &flush))
1367                         return true;
1368         }
1369
1370         return false;
1371 }
1372
1373 static bool dm_table_discard_zeroes_data(struct dm_table *t)
1374 {
1375         struct dm_target *ti;
1376         unsigned i = 0;
1377
1378         /* Ensure that all targets supports discard_zeroes_data. */
1379         while (i < dm_table_get_num_targets(t)) {
1380                 ti = dm_table_get_target(t, i++);
1381
1382                 if (ti->discard_zeroes_data_unsupported)
1383                         return false;
1384         }
1385
1386         return true;
1387 }
1388
1389 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1390                             sector_t start, sector_t len, void *data)
1391 {
1392         struct request_queue *q = bdev_get_queue(dev->bdev);
1393
1394         return q && blk_queue_nonrot(q);
1395 }
1396
1397 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1398                              sector_t start, sector_t len, void *data)
1399 {
1400         struct request_queue *q = bdev_get_queue(dev->bdev);
1401
1402         return q && !blk_queue_add_random(q);
1403 }
1404
1405 static int queue_supports_sg_merge(struct dm_target *ti, struct dm_dev *dev,
1406                                    sector_t start, sector_t len, void *data)
1407 {
1408         struct request_queue *q = bdev_get_queue(dev->bdev);
1409
1410         return q && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1411 }
1412
1413 static bool dm_table_all_devices_attribute(struct dm_table *t,
1414                                            iterate_devices_callout_fn func)
1415 {
1416         struct dm_target *ti;
1417         unsigned i = 0;
1418
1419         while (i < dm_table_get_num_targets(t)) {
1420                 ti = dm_table_get_target(t, i++);
1421
1422                 if (!ti->type->iterate_devices ||
1423                     !ti->type->iterate_devices(ti, func, NULL))
1424                         return false;
1425         }
1426
1427         return true;
1428 }
1429
1430 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1431                                          sector_t start, sector_t len, void *data)
1432 {
1433         struct request_queue *q = bdev_get_queue(dev->bdev);
1434
1435         return q && !q->limits.max_write_same_sectors;
1436 }
1437
1438 static bool dm_table_supports_write_same(struct dm_table *t)
1439 {
1440         struct dm_target *ti;
1441         unsigned i = 0;
1442
1443         while (i < dm_table_get_num_targets(t)) {
1444                 ti = dm_table_get_target(t, i++);
1445
1446                 if (!ti->num_write_same_bios)
1447                         return false;
1448
1449                 if (!ti->type->iterate_devices ||
1450                     ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1451                         return false;
1452         }
1453
1454         return true;
1455 }
1456
1457 static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1458                                   sector_t start, sector_t len, void *data)
1459 {
1460         struct request_queue *q = bdev_get_queue(dev->bdev);
1461
1462         return q && blk_queue_discard(q);
1463 }
1464
1465 static bool dm_table_supports_discards(struct dm_table *t)
1466 {
1467         struct dm_target *ti;
1468         unsigned i = 0;
1469
1470         /*
1471          * Unless any target used by the table set discards_supported,
1472          * require at least one underlying device to support discards.
1473          * t->devices includes internal dm devices such as mirror logs
1474          * so we need to use iterate_devices here, which targets
1475          * supporting discard selectively must provide.
1476          */
1477         while (i < dm_table_get_num_targets(t)) {
1478                 ti = dm_table_get_target(t, i++);
1479
1480                 if (!ti->num_discard_bios)
1481                         continue;
1482
1483                 if (ti->discards_supported)
1484                         return true;
1485
1486                 if (ti->type->iterate_devices &&
1487                     ti->type->iterate_devices(ti, device_discard_capable, NULL))
1488                         return true;
1489         }
1490
1491         return false;
1492 }
1493
1494 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1495                                struct queue_limits *limits)
1496 {
1497         unsigned flush = 0;
1498
1499         /*
1500          * Copy table's limits to the DM device's request_queue
1501          */
1502         q->limits = *limits;
1503
1504         if (!dm_table_supports_discards(t))
1505                 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1506         else
1507                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1508
1509         if (dm_table_supports_flush(t, REQ_FLUSH)) {
1510                 flush |= REQ_FLUSH;
1511                 if (dm_table_supports_flush(t, REQ_FUA))
1512                         flush |= REQ_FUA;
1513         }
1514         blk_queue_flush(q, flush);
1515
1516         if (!dm_table_discard_zeroes_data(t))
1517                 q->limits.discard_zeroes_data = 0;
1518
1519         /* Ensure that all underlying devices are non-rotational. */
1520         if (dm_table_all_devices_attribute(t, device_is_nonrot))
1521                 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1522         else
1523                 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1524
1525         if (!dm_table_supports_write_same(t))
1526                 q->limits.max_write_same_sectors = 0;
1527
1528         if (dm_table_all_devices_attribute(t, queue_supports_sg_merge))
1529                 queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1530         else
1531                 queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1532
1533         dm_table_verify_integrity(t);
1534
1535         /*
1536          * Determine whether or not this queue's I/O timings contribute
1537          * to the entropy pool, Only request-based targets use this.
1538          * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1539          * have it set.
1540          */
1541         if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1542                 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1543
1544         /*
1545          * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1546          * visible to other CPUs because, once the flag is set, incoming bios
1547          * are processed by request-based dm, which refers to the queue
1548          * settings.
1549          * Until the flag set, bios are passed to bio-based dm and queued to
1550          * md->deferred where queue settings are not needed yet.
1551          * Those bios are passed to request-based dm at the resume time.
1552          */
1553         smp_mb();
1554         if (dm_table_request_based(t))
1555                 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1556 }
1557
1558 unsigned int dm_table_get_num_targets(struct dm_table *t)
1559 {
1560         return t->num_targets;
1561 }
1562
1563 struct list_head *dm_table_get_devices(struct dm_table *t)
1564 {
1565         return &t->devices;
1566 }
1567
1568 fmode_t dm_table_get_mode(struct dm_table *t)
1569 {
1570         return t->mode;
1571 }
1572 EXPORT_SYMBOL(dm_table_get_mode);
1573
1574 enum suspend_mode {
1575         PRESUSPEND,
1576         PRESUSPEND_UNDO,
1577         POSTSUSPEND,
1578 };
1579
1580 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1581 {
1582         int i = t->num_targets;
1583         struct dm_target *ti = t->targets;
1584
1585         while (i--) {
1586                 switch (mode) {
1587                 case PRESUSPEND:
1588                         if (ti->type->presuspend)
1589                                 ti->type->presuspend(ti);
1590                         break;
1591                 case PRESUSPEND_UNDO:
1592                         if (ti->type->presuspend_undo)
1593                                 ti->type->presuspend_undo(ti);
1594                         break;
1595                 case POSTSUSPEND:
1596                         if (ti->type->postsuspend)
1597                                 ti->type->postsuspend(ti);
1598                         break;
1599                 }
1600                 ti++;
1601         }
1602 }
1603
1604 void dm_table_presuspend_targets(struct dm_table *t)
1605 {
1606         if (!t)
1607                 return;
1608
1609         suspend_targets(t, PRESUSPEND);
1610 }
1611
1612 void dm_table_presuspend_undo_targets(struct dm_table *t)
1613 {
1614         if (!t)
1615                 return;
1616
1617         suspend_targets(t, PRESUSPEND_UNDO);
1618 }
1619
1620 void dm_table_postsuspend_targets(struct dm_table *t)
1621 {
1622         if (!t)
1623                 return;
1624
1625         suspend_targets(t, POSTSUSPEND);
1626 }
1627
1628 int dm_table_resume_targets(struct dm_table *t)
1629 {
1630         int i, r = 0;
1631
1632         for (i = 0; i < t->num_targets; i++) {
1633                 struct dm_target *ti = t->targets + i;
1634
1635                 if (!ti->type->preresume)
1636                         continue;
1637
1638                 r = ti->type->preresume(ti);
1639                 if (r) {
1640                         DMERR("%s: %s: preresume failed, error = %d",
1641                               dm_device_name(t->md), ti->type->name, r);
1642                         return r;
1643                 }
1644         }
1645
1646         for (i = 0; i < t->num_targets; i++) {
1647                 struct dm_target *ti = t->targets + i;
1648
1649                 if (ti->type->resume)
1650                         ti->type->resume(ti);
1651         }
1652
1653         return 0;
1654 }
1655
1656 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1657 {
1658         list_add(&cb->list, &t->target_callbacks);
1659 }
1660 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1661
1662 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1663 {
1664         struct dm_dev_internal *dd;
1665         struct list_head *devices = dm_table_get_devices(t);
1666         struct dm_target_callbacks *cb;
1667         int r = 0;
1668
1669         list_for_each_entry(dd, devices, list) {
1670                 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1671                 char b[BDEVNAME_SIZE];
1672
1673                 if (likely(q))
1674                         r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1675                 else
1676                         DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1677                                      dm_device_name(t->md),
1678                                      bdevname(dd->dm_dev->bdev, b));
1679         }
1680
1681         list_for_each_entry(cb, &t->target_callbacks, list)
1682                 if (cb->congested_fn)
1683                         r |= cb->congested_fn(cb, bdi_bits);
1684
1685         return r;
1686 }
1687
1688 struct mapped_device *dm_table_get_md(struct dm_table *t)
1689 {
1690         return t->md;
1691 }
1692 EXPORT_SYMBOL(dm_table_get_md);
1693
1694 void dm_table_run_md_queue_async(struct dm_table *t)
1695 {
1696         struct mapped_device *md;
1697         struct request_queue *queue;
1698         unsigned long flags;
1699
1700         if (!dm_table_request_based(t))
1701                 return;
1702
1703         md = dm_table_get_md(t);
1704         queue = dm_get_md_queue(md);
1705         if (queue) {
1706                 if (queue->mq_ops)
1707                         blk_mq_run_hw_queues(queue, true);
1708                 else {
1709                         spin_lock_irqsave(queue->queue_lock, flags);
1710                         blk_run_queue_async(queue);
1711                         spin_unlock_irqrestore(queue->queue_lock, flags);
1712                 }
1713         }
1714 }
1715 EXPORT_SYMBOL(dm_table_run_md_queue_async);
1716