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