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