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-core.h"
   9
  10 #include <linux/module.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/blkdev.h>
  13 #include <linux/namei.h>
  14 #include <linux/ctype.h>
  15 #include <linux/string.h>
  16 #include <linux/slab.h>
  17 #include <linux/interrupt.h>
  18 #include <linux/mutex.h>
  19 #include <linux/delay.h>
  20 #include <linux/atomic.h>
  21 #include <linux/blk-mq.h>
  22 #include <linux/mount.h>
  23
  24 #define DM_MSG_PREFIX "table"
  25
  26 #define MAX_DEPTH 16
  27 #define NODE_SIZE L1_CACHE_BYTES
  28 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
  29 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
  30
  31 struct dm_table {
  32         struct mapped_device *md;
  33         unsigned type;
  34
  35         /* btree table */
  36         unsigned int depth;
  37         unsigned int counts[MAX_DEPTH]; /* in nodes */
  38         sector_t *index[MAX_DEPTH];
  39
  40         unsigned int num_targets;
  41         unsigned int num_allocated;
  42         sector_t *highs;
  43         struct dm_target *targets;
  44
  45         struct target_type *immutable_target_type;
  46
  47         bool integrity_supported:1;
  48         bool singleton:1;
  49         bool all_blk_mq:1;
  50
  51         /*
  52          * Indicates the rw permissions for the new logical
  53          * device.  This should be a combination of FMODE_READ
  54          * and FMODE_WRITE.
  55          */
  56         fmode_t mode;
  57
  58         /* a list of devices used by this table */
  59         struct list_head devices;
  60
  61         /* events get handed up using this callback */
  62         void (*event_fn)(void *);
  63         void *event_context;
  64
  65         struct dm_md_mempools *mempools;
  66
  67         struct list_head target_callbacks;
  68 };
  69
  70 /*
  71  * Similar to ceiling(log_size(n))
  72  */
  73 static unsigned int int_log(unsigned int n, unsigned int base)
  74 {
  75         int result = 0;
  76
  77         while (n > 1) {
  78                 n = dm_div_up(n, base);
  79                 result++;
  80         }
  81
  82         return result;
  83 }
  84
  85 /*
  86  * Calculate the index of the child node of the n'th node k'th key.
  87  */
  88 static inline unsigned int get_child(unsigned int n, unsigned int k)
  89 {
  90         return (n * CHILDREN_PER_NODE) + k;
  91 }
  92
  93 /*
  94  * Return the n'th node of level l from table t.
  95  */
  96 static inline sector_t *get_node(struct dm_table *t,
  97                                  unsigned int l, unsigned int n)
  98 {
  99         return t->index[l] + (n * KEYS_PER_NODE);
 100 }
 101
 102 /*
 103  * Return the highest key that you could lookup from the n'th
 104  * node on level l of the btree.
 105  */
 106 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
 107 {
 108         for (; l < t->depth - 1; l++)
 109                 n = get_child(n, CHILDREN_PER_NODE - 1);
 110
 111         if (n >= t->counts[l])
 112                 return (sector_t) - 1;
 113
 114         return get_node(t, l, n)[KEYS_PER_NODE - 1];
 115 }
 116
 117 /*
 118  * Fills in a level of the btree based on the highs of the level
 119  * below it.
 120  */
 121 static int setup_btree_index(unsigned int l, struct dm_table *t)
 122 {
 123         unsigned int n, k;
 124         sector_t *node;
 125
 126         for (n = 0U; n < t->counts[l]; n++) {
 127                 node = get_node(t, l, n);
 128
 129                 for (k = 0U; k < KEYS_PER_NODE; k++)
 130                         node[k] = high(t, l + 1, get_child(n, k));
 131         }
 132
 133         return 0;
 134 }
 135
 136 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
 137 {
 138         unsigned long size;
 139         void *addr;
 140
 141         /*
 142          * Check that we're not going to overflow.
 143          */
 144         if (nmemb > (ULONG_MAX / elem_size))
 145                 return NULL;
 146
 147         size = nmemb * elem_size;
 148         addr = vzalloc(size);
 149
 150         return addr;
 151 }
 152 EXPORT_SYMBOL(dm_vcalloc);
 153
 154 /*
 155  * highs, and targets are managed as dynamic arrays during a
 156  * table load.
 157  */
 158 static int alloc_targets(struct dm_table *t, unsigned int num)
 159 {
 160         sector_t *n_highs;
 161         struct dm_target *n_targets;
 162
 163         /*
 164          * Allocate both the target array and offset array at once.
 165          * Append an empty entry to catch sectors beyond the end of
 166          * the device.
 167          */
 168         n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
 169                                           sizeof(sector_t));
 170         if (!n_highs)
 171                 return -ENOMEM;
 172
 173         n_targets = (struct dm_target *) (n_highs + num);
 174
 175         memset(n_highs, -1, sizeof(*n_highs) * num);
 176         vfree(t->highs);
 177
 178         t->num_allocated = num;
 179         t->highs = n_highs;
 180         t->targets = n_targets;
 181
 182         return 0;
 183 }
 184
 185 int dm_table_create(struct dm_table **result, fmode_t mode,
 186                     unsigned num_targets, struct mapped_device *md)
 187 {
 188         struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
 189
 190         if (!t)
 191                 return -ENOMEM;
 192
 193         INIT_LIST_HEAD(&t->devices);
 194         INIT_LIST_HEAD(&t->target_callbacks);
 195
 196         if (!num_targets)
 197                 num_targets = KEYS_PER_NODE;
 198
 199         num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
 200
 201         if (!num_targets) {
 202                 kfree(t);
 203                 return -ENOMEM;
 204         }
 205
 206         if (alloc_targets(t, num_targets)) {
 207                 kfree(t);
 208                 return -ENOMEM;
 209         }
 210
 211         t->type = DM_TYPE_NONE;
 212         t->mode = mode;
 213         t->md = md;
 214         *result = t;
 215         return 0;
 216 }
 217
 218 static void free_devices(struct list_head *devices, struct mapped_device *md)
 219 {
 220         struct list_head *tmp, *next;
 221
 222         list_for_each_safe(tmp, next, devices) {
 223                 struct dm_dev_internal *dd =
 224                     list_entry(tmp, struct dm_dev_internal, list);
 225                 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
 226                        dm_device_name(md), dd->dm_dev->name);
 227                 dm_put_table_device(md, dd->dm_dev);
 228                 kfree(dd);
 229         }
 230 }
 231
 232 void dm_table_destroy(struct dm_table *t)
 233 {
 234         unsigned int i;
 235
 236         if (!t)
 237                 return;
 238
 239         /* free the indexes */
 240         if (t->depth >= 2)
 241                 vfree(t->index[t->depth - 2]);
 242
 243         /* free the targets */
 244         for (i = 0; i < t->num_targets; i++) {
 245                 struct dm_target *tgt = t->targets + i;
 246
 247                 if (tgt->type->dtr)
 248                         tgt->type->dtr(tgt);
 249
 250                 dm_put_target_type(tgt->type);
 251         }
 252
 253         vfree(t->highs);
 254
 255         /* free the device list */
 256         free_devices(&t->devices, t->md);
 257
 258         dm_free_md_mempools(t->mempools);
 259
 260         kfree(t);
 261 }
 262
 263 /*
 264  * See if we've already got a device in the list.
 265  */
 266 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
 267 {
 268         struct dm_dev_internal *dd;
 269
 270         list_for_each_entry (dd, l, list)
 271                 if (dd->dm_dev->bdev->bd_dev == dev)
 272                         return dd;
 273
 274         return NULL;
 275 }
 276
 277 /*
 278  * If possible, this checks an area of a destination device is invalid.
 279  */
 280 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
 281                                   sector_t start, sector_t len, void *data)
 282 {
 283         struct request_queue *q;
 284         struct queue_limits *limits = data;
 285         struct block_device *bdev = dev->bdev;
 286         sector_t dev_size =
 287                 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
 288         unsigned short logical_block_size_sectors =
 289                 limits->logical_block_size >> SECTOR_SHIFT;
 290         char b[BDEVNAME_SIZE];
 291
 292         /*
 293          * Some devices exist without request functions,
 294          * such as loop devices not yet bound to backing files.
 295          * Forbid the use of such devices.
 296          */
 297         q = bdev_get_queue(bdev);
 298         if (!q || !q->make_request_fn) {
 299                 DMWARN("%s: %s is not yet initialised: "
 300                        "start=%llu, len=%llu, dev_size=%llu",
 301                        dm_device_name(ti->table->md), bdevname(bdev, b),
 302                        (unsigned long long)start,
 303                        (unsigned long long)len,
 304                        (unsigned long long)dev_size);
 305                 return 1;
 306         }
 307
 308         if (!dev_size)
 309                 return 0;
 310
 311         if ((start >= dev_size) || (start + len > dev_size)) {
 312                 DMWARN("%s: %s too small for target: "
 313                        "start=%llu, len=%llu, dev_size=%llu",
 314                        dm_device_name(ti->table->md), bdevname(bdev, b),
 315                        (unsigned long long)start,
 316                        (unsigned long long)len,
 317                        (unsigned long long)dev_size);
 318                 return 1;
 319         }
 320
 321         if (logical_block_size_sectors <= 1)
 322                 return 0;
 323
 324         if (start & (logical_block_size_sectors - 1)) {
 325                 DMWARN("%s: start=%llu not aligned to h/w "
 326                        "logical block size %u of %s",
 327                        dm_device_name(ti->table->md),
 328                        (unsigned long long)start,
 329                        limits->logical_block_size, bdevname(bdev, b));
 330                 return 1;
 331         }
 332
 333         if (len & (logical_block_size_sectors - 1)) {
 334                 DMWARN("%s: len=%llu not aligned to h/w "
 335                        "logical block size %u of %s",
 336                        dm_device_name(ti->table->md),
 337                        (unsigned long long)len,
 338                        limits->logical_block_size, bdevname(bdev, b));
 339                 return 1;
 340         }
 341
 342         return 0;
 343 }
 344
 345 /*
 346  * This upgrades the mode on an already open dm_dev, being
 347  * careful to leave things as they were if we fail to reopen the
 348  * device and not to touch the existing bdev field in case
 349  * it is accessed concurrently inside dm_table_any_congested().
 350  */
 351 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
 352                         struct mapped_device *md)
 353 {
 354         int r;
 355         struct dm_dev *old_dev, *new_dev;
 356
 357         old_dev = dd->dm_dev;
 358
 359         r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
 360                                 dd->dm_dev->mode | new_mode, &new_dev);
 361         if (r)
 362                 return r;
 363
 364         dd->dm_dev = new_dev;
 365         dm_put_table_device(md, old_dev);
 366
 367         return 0;
 368 }
 369
 370 /*
 371  * Convert the path to a device
 372  */
 373 dev_t dm_get_dev_t(const char *path)
 374 {
 375         dev_t uninitialized_var(dev);
 376         struct block_device *bdev;
 377
 378         bdev = lookup_bdev(path);
 379         if (IS_ERR(bdev))
 380                 dev = name_to_dev_t(path);
 381         else {
 382                 dev = bdev->bd_dev;
 383                 bdput(bdev);
 384         }
 385
 386         return dev;
 387 }
 388 EXPORT_SYMBOL_GPL(dm_get_dev_t);
 389
 390 /*
 391  * Add a device to the list, or just increment the usage count if
 392  * it's already present.
 393  */
 394 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
 395                   struct dm_dev **result)
 396 {
 397         int r;
 398         dev_t dev;
 399         unsigned int major, minor;
 400         char dummy;
 401         struct dm_dev_internal *dd;
 402         struct dm_table *t = ti->table;
 403
 404         BUG_ON(!t);
 405
 406         if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
 407                 /* Extract the major/minor numbers */
 408                 dev = MKDEV(major, minor);
 409                 if (MAJOR(dev) != major || MINOR(dev) != minor)
 410                         return -EOVERFLOW;
 411         } else {
 412                 dev = dm_get_dev_t(path);
 413                 if (!dev)
 414                         return -ENODEV;
 415         }
 416
 417         dd = find_device(&t->devices, dev);
 418         if (!dd) {
 419                 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
 420                 if (!dd)
 421                         return -ENOMEM;
 422
 423                 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
 424                         kfree(dd);
 425                         return r;
 426                 }
 427
 428                 atomic_set(&dd->count, 0);
 429                 list_add(&dd->list, &t->devices);
 430
 431         } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
 432                 r = upgrade_mode(dd, mode, t->md);
 433                 if (r)
 434                         return r;
 435         }
 436         atomic_inc(&dd->count);
 437
 438         *result = dd->dm_dev;
 439         return 0;
 440 }
 441 EXPORT_SYMBOL(dm_get_device);
 442
 443 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
 444                                 sector_t start, sector_t len, void *data)
 445 {
 446         struct queue_limits *limits = data;
 447         struct block_device *bdev = dev->bdev;
 448         struct request_queue *q = bdev_get_queue(bdev);
 449         char b[BDEVNAME_SIZE];
 450
 451         if (unlikely(!q)) {
 452                 DMWARN("%s: Cannot set limits for nonexistent device %s",
 453                        dm_device_name(ti->table->md), bdevname(bdev, b));
 454                 return 0;
 455         }
 456
 457         if (bdev_stack_limits(limits, bdev, start) < 0)
 458                 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
 459                        "physical_block_size=%u, logical_block_size=%u, "
 460                        "alignment_offset=%u, start=%llu",
 461                        dm_device_name(ti->table->md), bdevname(bdev, b),
 462                        q->limits.physical_block_size,
 463                        q->limits.logical_block_size,
 464                        q->limits.alignment_offset,
 465                        (unsigned long long) start << SECTOR_SHIFT);
 466
 467         return 0;
 468 }
 469
 470 /*
 471  * Decrement a device's use count and remove it if necessary.
 472  */
 473 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
 474 {
 475         int found = 0;
 476         struct list_head *devices = &ti->table->devices;
 477         struct dm_dev_internal *dd;
 478
 479         list_for_each_entry(dd, devices, list) {
 480                 if (dd->dm_dev == d) {
 481                         found = 1;
 482                         break;
 483                 }
 484         }
 485         if (!found) {
 486                 DMWARN("%s: device %s not in table devices list",
 487                        dm_device_name(ti->table->md), d->name);
 488                 return;
 489         }
 490         if (atomic_dec_and_test(&dd->count)) {
 491                 dm_put_table_device(ti->table->md, d);
 492                 list_del(&dd->list);
 493                 kfree(dd);
 494         }
 495 }
 496 EXPORT_SYMBOL(dm_put_device);
 497
 498 /*
 499  * Checks to see if the target joins onto the end of the table.
 500  */
 501 static int adjoin(struct dm_table *table, struct dm_target *ti)
 502 {
 503         struct dm_target *prev;
 504
 505         if (!table->num_targets)
 506                 return !ti->begin;
 507
 508         prev = &table->targets[table->num_targets - 1];
 509         return (ti->begin == (prev->begin + prev->len));
 510 }
 511
 512 /*
 513  * Used to dynamically allocate the arg array.
 514  *
 515  * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
 516  * process messages even if some device is suspended. These messages have a
 517  * small fixed number of arguments.
 518  *
 519  * On the other hand, dm-switch needs to process bulk data using messages and
 520  * excessive use of GFP_NOIO could cause trouble.
 521  */
 522 static char **realloc_argv(unsigned *size, char **old_argv)
 523 {
 524         char **argv;
 525         unsigned new_size;
 526         gfp_t gfp;
 527
 528         if (*size) {
 529                 new_size = *size * 2;
 530                 gfp = GFP_KERNEL;
 531         } else {
 532                 new_size = 8;
 533                 gfp = GFP_NOIO;
 534         }
 535         argv = kmalloc(new_size * sizeof(*argv), gfp);
 536         if (argv) {
 537                 memcpy(argv, old_argv, *size * sizeof(*argv));
 538                 *size = new_size;
 539         }
 540
 541         kfree(old_argv);
 542         return argv;
 543 }
 544
 545 /*
 546  * Destructively splits up the argument list to pass to ctr.
 547  */
 548 int dm_split_args(int *argc, char ***argvp, char *input)
 549 {
 550         char *start, *end = input, *out, **argv = NULL;
 551         unsigned array_size = 0;
 552
 553         *argc = 0;
 554
 555         if (!input) {
 556                 *argvp = NULL;
 557                 return 0;
 558         }
 559
 560         argv = realloc_argv(&array_size, argv);
 561         if (!argv)
 562                 return -ENOMEM;
 563
 564         while (1) {
 565                 /* Skip whitespace */
 566                 start = skip_spaces(end);
 567
 568                 if (!*start)
 569                         break;  /* success, we hit the end */
 570
 571                 /* 'out' is used to remove any back-quotes */
 572                 end = out = start;
 573                 while (*end) {
 574                         /* Everything apart from '\0' can be quoted */
 575                         if (*end == '\\' && *(end + 1)) {
 576                                 *out++ = *(end + 1);
 577                                 end += 2;
 578                                 continue;
 579                         }
 580
 581                         if (isspace(*end))
 582                                 break;  /* end of token */
 583
 584                         *out++ = *end++;
 585                 }
 586
 587                 /* have we already filled the array ? */
 588                 if ((*argc + 1) > array_size) {
 589                         argv = realloc_argv(&array_size, argv);
 590                         if (!argv)
 591                                 return -ENOMEM;
 592                 }
 593
 594                 /* we know this is whitespace */
 595                 if (*end)
 596                         end++;
 597
 598                 /* terminate the string and put it in the array */
 599                 *out = '\0';
 600                 argv[*argc] = start;
 601                 (*argc)++;
 602         }
 603
 604         *argvp = argv;
 605         return 0;
 606 }
 607
 608 /*
 609  * Impose necessary and sufficient conditions on a devices's table such
 610  * that any incoming bio which respects its logical_block_size can be
 611  * processed successfully.  If it falls across the boundary between
 612  * two or more targets, the size of each piece it gets split into must
 613  * be compatible with the logical_block_size of the target processing it.
 614  */
 615 static int validate_hardware_logical_block_alignment(struct dm_table *table,
 616                                                  struct queue_limits *limits)
 617 {
 618         /*
 619          * This function uses arithmetic modulo the logical_block_size
 620          * (in units of 512-byte sectors).
 621          */
 622         unsigned short device_logical_block_size_sects =
 623                 limits->logical_block_size >> SECTOR_SHIFT;
 624
 625         /*
 626          * Offset of the start of the next table entry, mod logical_block_size.
 627          */
 628         unsigned short next_target_start = 0;
 629
 630         /*
 631          * Given an aligned bio that extends beyond the end of a
 632          * target, how many sectors must the next target handle?
 633          */
 634         unsigned short remaining = 0;
 635
 636         struct dm_target *uninitialized_var(ti);
 637         struct queue_limits ti_limits;
 638         unsigned i = 0;
 639
 640         /*
 641          * Check each entry in the table in turn.
 642          */
 643         while (i < dm_table_get_num_targets(table)) {
 644                 ti = dm_table_get_target(table, i++);
 645
 646                 blk_set_stacking_limits(&ti_limits);
 647
 648                 /* combine all target devices' limits */
 649                 if (ti->type->iterate_devices)
 650                         ti->type->iterate_devices(ti, dm_set_device_limits,
 651                                                   &ti_limits);
 652
 653                 /*
 654                  * If the remaining sectors fall entirely within this
 655                  * table entry are they compatible with its logical_block_size?
 656                  */
 657                 if (remaining < ti->len &&
 658                     remaining & ((ti_limits.logical_block_size >>
 659                                   SECTOR_SHIFT) - 1))
 660                         break;  /* Error */
 661
 662                 next_target_start =
 663                     (unsigned short) ((next_target_start + ti->len) &
 664                                       (device_logical_block_size_sects - 1));
 665                 remaining = next_target_start ?
 666                     device_logical_block_size_sects - next_target_start : 0;
 667         }
 668
 669         if (remaining) {
 670                 DMWARN("%s: table line %u (start sect %llu len %llu) "
 671                        "not aligned to h/w logical block size %u",
 672                        dm_device_name(table->md), i,
 673                        (unsigned long long) ti->begin,
 674                        (unsigned long long) ti->len,
 675                        limits->logical_block_size);
 676                 return -EINVAL;
 677         }
 678
 679         return 0;
 680 }
 681
 682 int dm_table_add_target(struct dm_table *t, const char *type,
 683                         sector_t start, sector_t len, char *params)
 684 {
 685         int r = -EINVAL, argc;
 686         char **argv;
 687         struct dm_target *tgt;
 688
 689         if (t->singleton) {
 690                 DMERR("%s: target type %s must appear alone in table",
 691                       dm_device_name(t->md), t->targets->type->name);
 692                 return -EINVAL;
 693         }
 694
 695         BUG_ON(t->num_targets >= t->num_allocated);
 696
 697         tgt = t->targets + t->num_targets;
 698         memset(tgt, 0, sizeof(*tgt));
 699
 700         if (!len) {
 701                 DMERR("%s: zero-length target", dm_device_name(t->md));
 702                 return -EINVAL;
 703         }
 704
 705         tgt->type = dm_get_target_type(type);
 706         if (!tgt->type) {
 707                 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
 708                 return -EINVAL;
 709         }
 710
 711         if (dm_target_needs_singleton(tgt->type)) {
 712                 if (t->num_targets) {
 713                         tgt->error = "singleton target type must appear alone in table";
 714                         goto bad;
 715                 }
 716                 t->singleton = true;
 717         }
 718
 719         if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
 720                 tgt->error = "target type may not be included in a read-only table";
 721                 goto bad;
 722         }
 723
 724         if (t->immutable_target_type) {
 725                 if (t->immutable_target_type != tgt->type) {
 726                         tgt->error = "immutable target type cannot be mixed with other target types";
 727                         goto bad;
 728                 }
 729         } else if (dm_target_is_immutable(tgt->type)) {
 730                 if (t->num_targets) {
 731                         tgt->error = "immutable target type cannot be mixed with other target types";
 732                         goto bad;
 733                 }
 734                 t->immutable_target_type = tgt->type;
 735         }
 736
 737         tgt->table = t;
 738         tgt->begin = start;
 739         tgt->len = len;
 740         tgt->error = "Unknown error";
 741
 742         /*
 743          * Does this target adjoin the previous one ?
 744          */
 745         if (!adjoin(t, tgt)) {
 746                 tgt->error = "Gap in table";
 747                 goto bad;
 748         }
 749
 750         r = dm_split_args(&argc, &argv, params);
 751         if (r) {
 752                 tgt->error = "couldn't split parameters (insufficient memory)";
 753                 goto bad;
 754         }
 755
 756         r = tgt->type->ctr(tgt, argc, argv);
 757         kfree(argv);
 758         if (r)
 759                 goto bad;
 760
 761         t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
 762
 763         if (!tgt->num_discard_bios && tgt->discards_supported)
 764                 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
 765                        dm_device_name(t->md), type);
 766
 767         return 0;
 768
 769  bad:
 770         DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
 771         dm_put_target_type(tgt->type);
 772         return r;
 773 }
 774
 775 /*
 776  * Target argument parsing helpers.
 777  */
 778 static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
 779                              unsigned *value, char **error, unsigned grouped)
 780 {
 781         const char *arg_str = dm_shift_arg(arg_set);
 782         char dummy;
 783
 784         if (!arg_str ||
 785             (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
 786             (*value < arg->min) ||
 787             (*value > arg->max) ||
 788             (grouped && arg_set->argc < *value)) {
 789                 *error = arg->error;
 790                 return -EINVAL;
 791         }
 792
 793         return 0;
 794 }
 795
 796 int dm_read_arg(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, 0);
 800 }
 801 EXPORT_SYMBOL(dm_read_arg);
 802
 803 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
 804                       unsigned *value, char **error)
 805 {
 806         return validate_next_arg(arg, arg_set, value, error, 1);
 807 }
 808 EXPORT_SYMBOL(dm_read_arg_group);
 809
 810 const char *dm_shift_arg(struct dm_arg_set *as)
 811 {
 812         char *r;
 813
 814         if (as->argc) {
 815                 as->argc--;
 816                 r = *as->argv;
 817                 as->argv++;
 818                 return r;
 819         }
 820
 821         return NULL;
 822 }
 823 EXPORT_SYMBOL(dm_shift_arg);
 824
 825 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
 826 {
 827         BUG_ON(as->argc < num_args);
 828         as->argc -= num_args;
 829         as->argv += num_args;
 830 }
 831 EXPORT_SYMBOL(dm_consume_args);
 832
 833 static bool __table_type_bio_based(unsigned table_type)
 834 {
 835         return (table_type == DM_TYPE_BIO_BASED ||
 836                 table_type == DM_TYPE_DAX_BIO_BASED);
 837 }
 838
 839 static bool __table_type_request_based(unsigned table_type)
 840 {
 841         return (table_type == DM_TYPE_REQUEST_BASED ||
 842                 table_type == DM_TYPE_MQ_REQUEST_BASED);
 843 }
 844
 845 void dm_table_set_type(struct dm_table *t, unsigned type)
 846 {
 847         t->type = type;
 848 }
 849 EXPORT_SYMBOL_GPL(dm_table_set_type);
 850
 851 static int device_not_dax_capable(struct dm_target *ti, struct dm_dev *dev,
 852                                   sector_t start, sector_t len, void *data)
 853 {
 854         struct request_queue *q = bdev_get_queue(dev->bdev);
 855
 856         return q && !blk_queue_dax(q);
 857 }
 858
 859 static bool dm_table_supports_dax(struct dm_table *t)
 860 {
 861         struct dm_target *ti;
 862         unsigned i = 0;
 863
 864         /* Ensure that all targets support DAX. */
 865         while (i < dm_table_get_num_targets(t)) {
 866                 ti = dm_table_get_target(t, i++);
 867
 868                 if (!ti->type->direct_access)
 869                         return false;
 870
 871                 if (!ti->type->iterate_devices ||
 872                     ti->type->iterate_devices(ti, device_not_dax_capable, NULL))
 873                         return false;
 874         }
 875
 876         return true;
 877 }
 878
 879 static int dm_table_determine_type(struct dm_table *t)
 880 {
 881         unsigned i;
 882         unsigned bio_based = 0, request_based = 0, hybrid = 0;
 883         bool verify_blk_mq = false;
 884         struct dm_target *tgt;
 885         struct dm_dev_internal *dd;
 886         struct list_head *devices = dm_table_get_devices(t);
 887         unsigned live_md_type = dm_get_md_type(t->md);
 888
 889         if (t->type != DM_TYPE_NONE) {
 890                 /* target already set the table's type */
 891                 if (t->type == DM_TYPE_BIO_BASED)
 892                         return 0;
 893                 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
 894                 goto verify_rq_based;
 895         }
 896
 897         for (i = 0; i < t->num_targets; i++) {
 898                 tgt = t->targets + i;
 899                 if (dm_target_hybrid(tgt))
 900                         hybrid = 1;
 901                 else if (dm_target_request_based(tgt))
 902                         request_based = 1;
 903                 else
 904                         bio_based = 1;
 905
 906                 if (bio_based && request_based) {
 907                         DMWARN("Inconsistent table: different target types"
 908                                " can't be mixed up");
 909                         return -EINVAL;
 910                 }
 911         }
 912
 913         if (hybrid && !bio_based && !request_based) {
 914                 /*
 915                  * The targets can work either way.
 916                  * Determine the type from the live device.
 917                  * Default to bio-based if device is new.
 918                  */
 919                 if (__table_type_request_based(live_md_type))
 920                         request_based = 1;
 921                 else
 922                         bio_based = 1;
 923         }
 924
 925         if (bio_based) {
 926                 /* We must use this table as bio-based */
 927                 t->type = DM_TYPE_BIO_BASED;
 928                 if (dm_table_supports_dax(t) ||
 929                     (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED))
 930                         t->type = DM_TYPE_DAX_BIO_BASED;
 931                 return 0;
 932         }
 933
 934         BUG_ON(!request_based); /* No targets in this table */
 935
 936         /*
 937          * The only way to establish DM_TYPE_MQ_REQUEST_BASED is by
 938          * having a compatible target use dm_table_set_type.
 939          */
 940         t->type = DM_TYPE_REQUEST_BASED;
 941
 942 verify_rq_based:
 943         /*
 944          * Request-based dm supports only tables that have a single target now.
 945          * To support multiple targets, request splitting support is needed,
 946          * and that needs lots of changes in the block-layer.
 947          * (e.g. request completion process for partial completion.)
 948          */
 949         if (t->num_targets > 1) {
 950                 DMWARN("Request-based dm doesn't support multiple targets yet");
 951                 return -EINVAL;
 952         }
 953
 954         if (list_empty(devices)) {
 955                 int srcu_idx;
 956                 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
 957
 958                 /* inherit live table's type and all_blk_mq */
 959                 if (live_table) {
 960                         t->type = live_table->type;
 961                         t->all_blk_mq = live_table->all_blk_mq;
 962                 }
 963                 dm_put_live_table(t->md, srcu_idx);
 964                 return 0;
 965         }
 966
 967         /* Non-request-stackable devices can't be used for request-based dm */
 968         list_for_each_entry(dd, devices, list) {
 969                 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
 970
 971                 if (!blk_queue_stackable(q)) {
 972                         DMERR("table load rejected: including"
 973                               " non-request-stackable devices");
 974                         return -EINVAL;
 975                 }
 976
 977                 if (q->mq_ops)
 978                         verify_blk_mq = true;
 979         }
 980
 981         if (verify_blk_mq) {
 982                 /* verify _all_ devices in the table are blk-mq devices */
 983                 list_for_each_entry(dd, devices, list)
 984                         if (!bdev_get_queue(dd->dm_dev->bdev)->mq_ops) {
 985                                 DMERR("table load rejected: not all devices"
 986                                       " are blk-mq request-stackable");
 987                                 return -EINVAL;
 988                         }
 989
 990                 t->all_blk_mq = true;
 991         }
 992
 993         if (t->type == DM_TYPE_MQ_REQUEST_BASED && !t->all_blk_mq) {
 994                 DMERR("table load rejected: all devices are not blk-mq request-stackable");
 995                 return -EINVAL;
 996         }
 997
 998         return 0;
 999 }
1000
1001 unsigned dm_table_get_type(struct dm_table *t)
1002 {
1003         return t->type;
1004 }
1005
1006 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
1007 {
1008         return t->immutable_target_type;
1009 }
1010
1011 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
1012 {
1013         /* Immutable target is implicitly a singleton */
1014         if (t->num_targets > 1 ||
1015             !dm_target_is_immutable(t->targets[0].type))
1016                 return NULL;
1017
1018         return t->targets;
1019 }
1020
1021 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1022 {
1023         struct dm_target *uninitialized_var(ti);
1024         unsigned i = 0;
1025
1026         while (i < dm_table_get_num_targets(t)) {
1027                 ti = dm_table_get_target(t, i++);
1028                 if (dm_target_is_wildcard(ti->type))
1029                         return ti;
1030         }
1031
1032         return NULL;
1033 }
1034
1035 bool dm_table_bio_based(struct dm_table *t)
1036 {
1037         return __table_type_bio_based(dm_table_get_type(t));
1038 }
1039
1040 bool dm_table_request_based(struct dm_table *t)
1041 {
1042         return __table_type_request_based(dm_table_get_type(t));
1043 }
1044
1045 bool dm_table_all_blk_mq_devices(struct dm_table *t)
1046 {
1047         return t->all_blk_mq;
1048 }
1049
1050 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1051 {
1052         unsigned type = dm_table_get_type(t);
1053         unsigned per_io_data_size = 0;
1054         struct dm_target *tgt;
1055         unsigned i;
1056
1057         if (unlikely(type == DM_TYPE_NONE)) {
1058                 DMWARN("no table type is set, can't allocate mempools");
1059                 return -EINVAL;
1060         }
1061
1062         if (__table_type_bio_based(type))
1063                 for (i = 0; i < t->num_targets; i++) {
1064                         tgt = t->targets + i;
1065                         per_io_data_size = max(per_io_data_size, tgt->per_io_data_size);
1066                 }
1067
1068         t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported, per_io_data_size);
1069         if (!t->mempools)
1070                 return -ENOMEM;
1071
1072         return 0;
1073 }
1074
1075 void dm_table_free_md_mempools(struct dm_table *t)
1076 {
1077         dm_free_md_mempools(t->mempools);
1078         t->mempools = NULL;
1079 }
1080
1081 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1082 {
1083         return t->mempools;
1084 }
1085
1086 static int setup_indexes(struct dm_table *t)
1087 {
1088         int i;
1089         unsigned int total = 0;
1090         sector_t *indexes;
1091
1092         /* allocate the space for *all* the indexes */
1093         for (i = t->depth - 2; i >= 0; i--) {
1094                 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1095                 total += t->counts[i];
1096         }
1097
1098         indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1099         if (!indexes)
1100                 return -ENOMEM;
1101
1102         /* set up internal nodes, bottom-up */
1103         for (i = t->depth - 2; i >= 0; i--) {
1104                 t->index[i] = indexes;
1105                 indexes += (KEYS_PER_NODE * t->counts[i]);
1106                 setup_btree_index(i, t);
1107         }
1108
1109         return 0;
1110 }
1111
1112 /*
1113  * Builds the btree to index the map.
1114  */
1115 static int dm_table_build_index(struct dm_table *t)
1116 {
1117         int r = 0;
1118         unsigned int leaf_nodes;
1119
1120         /* how many indexes will the btree have ? */
1121         leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1122         t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1123
1124         /* leaf layer has already been set up */
1125         t->counts[t->depth - 1] = leaf_nodes;
1126         t->index[t->depth - 1] = t->highs;
1127
1128         if (t->depth >= 2)
1129                 r = setup_indexes(t);
1130
1131         return r;
1132 }
1133
1134 static bool integrity_profile_exists(struct gendisk *disk)
1135 {
1136         return !!blk_get_integrity(disk);
1137 }
1138
1139 /*
1140  * Get a disk whose integrity profile reflects the table's profile.
1141  * Returns NULL if integrity support was inconsistent or unavailable.
1142  */
1143 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1144 {
1145         struct list_head *devices = dm_table_get_devices(t);
1146         struct dm_dev_internal *dd = NULL;
1147         struct gendisk *prev_disk = NULL, *template_disk = NULL;
1148
1149         list_for_each_entry(dd, devices, list) {
1150                 template_disk = dd->dm_dev->bdev->bd_disk;
1151                 if (!integrity_profile_exists(template_disk))
1152                         goto no_integrity;
1153                 else if (prev_disk &&
1154                          blk_integrity_compare(prev_disk, template_disk) < 0)
1155                         goto no_integrity;
1156                 prev_disk = template_disk;
1157         }
1158
1159         return template_disk;
1160
1161 no_integrity:
1162         if (prev_disk)
1163                 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1164                        dm_device_name(t->md),
1165                        prev_disk->disk_name,
1166                        template_disk->disk_name);
1167         return NULL;
1168 }
1169
1170 /*
1171  * Register the mapped device for blk_integrity support if the
1172  * underlying devices have an integrity profile.  But all devices may
1173  * not have matching profiles (checking all devices isn't reliable
1174  * during table load because this table may use other DM device(s) which
1175  * must be resumed before they will have an initialized integity
1176  * profile).  Consequently, stacked DM devices force a 2 stage integrity
1177  * profile validation: First pass during table load, final pass during
1178  * resume.
1179  */
1180 static int dm_table_register_integrity(struct dm_table *t)
1181 {
1182         struct mapped_device *md = t->md;
1183         struct gendisk *template_disk = NULL;
1184
1185         template_disk = dm_table_get_integrity_disk(t);
1186         if (!template_disk)
1187                 return 0;
1188
1189         if (!integrity_profile_exists(dm_disk(md))) {
1190                 t->integrity_supported = true;
1191                 /*
1192                  * Register integrity profile during table load; we can do
1193                  * this because the final profile must match during resume.
1194                  */
1195                 blk_integrity_register(dm_disk(md),
1196                                        blk_get_integrity(template_disk));
1197                 return 0;
1198         }
1199
1200         /*
1201          * If DM device already has an initialized integrity
1202          * profile the new profile should not conflict.
1203          */
1204         if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1205                 DMWARN("%s: conflict with existing integrity profile: "
1206                        "%s profile mismatch",
1207                        dm_device_name(t->md),
1208                        template_disk->disk_name);
1209                 return 1;
1210         }
1211
1212         /* Preserve existing integrity profile */
1213         t->integrity_supported = true;
1214         return 0;
1215 }
1216
1217 /*
1218  * Prepares the table for use by building the indices,
1219  * setting the type, and allocating mempools.
1220  */
1221 int dm_table_complete(struct dm_table *t)
1222 {
1223         int r;
1224
1225         r = dm_table_determine_type(t);
1226         if (r) {
1227                 DMERR("unable to determine table type");
1228                 return r;
1229         }
1230
1231         r = dm_table_build_index(t);
1232         if (r) {
1233                 DMERR("unable to build btrees");
1234                 return r;
1235         }
1236
1237         r = dm_table_register_integrity(t);
1238         if (r) {
1239                 DMERR("could not register integrity profile.");
1240                 return r;
1241         }
1242
1243         r = dm_table_alloc_md_mempools(t, t->md);
1244         if (r)
1245                 DMERR("unable to allocate mempools");
1246
1247         return r;
1248 }
1249
1250 static DEFINE_MUTEX(_event_lock);
1251 void dm_table_event_callback(struct dm_table *t,
1252                              void (*fn)(void *), void *context)
1253 {
1254         mutex_lock(&_event_lock);
1255         t->event_fn = fn;
1256         t->event_context = context;
1257         mutex_unlock(&_event_lock);
1258 }
1259
1260 void dm_table_event(struct dm_table *t)
1261 {
1262         mutex_lock(&_event_lock);
1263         if (t->event_fn)
1264                 t->event_fn(t->event_context);
1265         mutex_unlock(&_event_lock);
1266 }
1267 EXPORT_SYMBOL(dm_table_event);
1268
1269 inline sector_t dm_table_get_size(struct dm_table *t)
1270 {
1271         return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1272 }
1273 EXPORT_SYMBOL(dm_table_get_size);
1274
1275 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1276 {
1277         if (index >= t->num_targets)
1278                 return NULL;
1279
1280         return t->targets + index;
1281 }
1282
1283 /*
1284  * Search the btree for the correct target.
1285  *
1286  * Caller should check returned pointer with dm_target_is_valid()
1287  * to trap I/O beyond end of device.
1288  */
1289 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1290 {
1291         unsigned int l, n = 0, k = 0;
1292         sector_t *node;
1293
1294         if (unlikely(sector >= dm_table_get_size(t)))
1295                 return &t->targets[t->num_targets];
1296
1297         for (l = 0; l < t->depth; l++) {
1298                 n = get_child(n, k);
1299                 node = get_node(t, l, n);
1300
1301                 for (k = 0; k < KEYS_PER_NODE; k++)
1302                         if (node[k] >= sector)
1303                                 break;
1304         }
1305
1306         return &t->targets[(KEYS_PER_NODE * n) + k];
1307 }
1308
1309 /*
1310  * type->iterate_devices() should be called when the sanity check needs to
1311  * iterate and check all underlying data devices. iterate_devices() will
1312  * iterate all underlying data devices until it encounters a non-zero return
1313  * code, returned by whether the input iterate_devices_callout_fn, or
1314  * iterate_devices() itself internally.
1315  *
1316  * For some target type (e.g. dm-stripe), one call of iterate_devices() may
1317  * iterate multiple underlying devices internally, in which case a non-zero
1318  * return code returned by iterate_devices_callout_fn will stop the iteration
1319  * in advance.
1320  *
1321  * Cases requiring _any_ underlying device supporting some kind of attribute,
1322  * should use the iteration structure like dm_table_any_dev_attr(), or call
1323  * it directly. @func should handle semantics of positive examples, e.g.
1324  * capable of something.
1325  *
1326  * Cases requiring _all_ underlying devices supporting some kind of attribute,
1327  * should use the iteration structure like dm_table_supports_nowait() or
1328  * dm_table_supports_discards(). Or introduce dm_table_all_devs_attr() that
1329  * uses an @anti_func that handle semantics of counter examples, e.g. not
1330  * capable of something. So: return !dm_table_any_dev_attr(t, anti_func);
1331  */
1332 static bool dm_table_any_dev_attr(struct dm_table *t,
1333                                   iterate_devices_callout_fn func)
1334 {
1335         struct dm_target *ti;
1336         unsigned int i;
1337
1338         for (i = 0; i < dm_table_get_num_targets(t); i++) {
1339                 ti = dm_table_get_target(t, i);
1340
1341                 if (ti->type->iterate_devices &&
1342                     ti->type->iterate_devices(ti, func, NULL))
1343                         return true;
1344         }
1345
1346         return false;
1347 }
1348
1349 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1350                         sector_t start, sector_t len, void *data)
1351 {
1352         unsigned *num_devices = data;
1353
1354         (*num_devices)++;
1355
1356         return 0;
1357 }
1358
1359 /*
1360  * Check whether a table has no data devices attached using each
1361  * target's iterate_devices method.
1362  * Returns false if the result is unknown because a target doesn't
1363  * support iterate_devices.
1364  */
1365 bool dm_table_has_no_data_devices(struct dm_table *table)
1366 {
1367         struct dm_target *uninitialized_var(ti);
1368         unsigned i = 0, num_devices = 0;
1369
1370         while (i < dm_table_get_num_targets(table)) {
1371                 ti = dm_table_get_target(table, i++);
1372
1373                 if (!ti->type->iterate_devices)
1374                         return false;
1375
1376                 ti->type->iterate_devices(ti, count_device, &num_devices);
1377                 if (num_devices)
1378                         return false;
1379         }
1380
1381         return true;
1382 }
1383
1384 /*
1385  * Establish the new table's queue_limits and validate them.
1386  */
1387 int dm_calculate_queue_limits(struct dm_table *table,
1388                               struct queue_limits *limits)
1389 {
1390         struct dm_target *uninitialized_var(ti);
1391         struct queue_limits ti_limits;
1392         unsigned i = 0;
1393
1394         blk_set_stacking_limits(limits);
1395
1396         while (i < dm_table_get_num_targets(table)) {
1397                 blk_set_stacking_limits(&ti_limits);
1398
1399                 ti = dm_table_get_target(table, i++);
1400
1401                 if (!ti->type->iterate_devices)
1402                         goto combine_limits;
1403
1404                 /*
1405                  * Combine queue limits of all the devices this target uses.
1406                  */
1407                 ti->type->iterate_devices(ti, dm_set_device_limits,
1408                                           &ti_limits);
1409
1410                 /* Set I/O hints portion of queue limits */
1411                 if (ti->type->io_hints)
1412                         ti->type->io_hints(ti, &ti_limits);
1413
1414                 /*
1415                  * Check each device area is consistent with the target's
1416                  * overall queue limits.
1417                  */
1418                 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1419                                               &ti_limits))
1420                         return -EINVAL;
1421
1422 combine_limits:
1423                 /*
1424                  * Merge this target's queue limits into the overall limits
1425                  * for the table.
1426                  */
1427                 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1428                         DMWARN("%s: adding target device "
1429                                "(start sect %llu len %llu) "
1430                                "caused an alignment inconsistency",
1431                                dm_device_name(table->md),
1432                                (unsigned long long) ti->begin,
1433                                (unsigned long long) ti->len);
1434         }
1435
1436         return validate_hardware_logical_block_alignment(table, limits);
1437 }
1438
1439 /*
1440  * Verify that all devices have an integrity profile that matches the
1441  * DM device's registered integrity profile.  If the profiles don't
1442  * match then unregister the DM device's integrity profile.
1443  */
1444 static void dm_table_verify_integrity(struct dm_table *t)
1445 {
1446         struct gendisk *template_disk = NULL;
1447
1448         if (t->integrity_supported) {
1449                 /*
1450                  * Verify that the original integrity profile
1451                  * matches all the devices in this table.
1452                  */
1453                 template_disk = dm_table_get_integrity_disk(t);
1454                 if (template_disk &&
1455                     blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1456                         return;
1457         }
1458
1459         if (integrity_profile_exists(dm_disk(t->md))) {
1460                 DMWARN("%s: unable to establish an integrity profile",
1461                        dm_device_name(t->md));
1462                 blk_integrity_unregister(dm_disk(t->md));
1463         }
1464 }
1465
1466 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1467                                 sector_t start, sector_t len, void *data)
1468 {
1469         unsigned long flush = (unsigned long) data;
1470         struct request_queue *q = bdev_get_queue(dev->bdev);
1471
1472         return q && (q->queue_flags & flush);
1473 }
1474
1475 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1476 {
1477         struct dm_target *ti;
1478         unsigned i = 0;
1479
1480         /*
1481          * Require at least one underlying device to support flushes.
1482          * t->devices includes internal dm devices such as mirror logs
1483          * so we need to use iterate_devices here, which targets
1484          * supporting flushes must provide.
1485          */
1486         while (i < dm_table_get_num_targets(t)) {
1487                 ti = dm_table_get_target(t, i++);
1488
1489                 if (!ti->num_flush_bios)
1490                         continue;
1491
1492                 if (ti->flush_supported)
1493                         return true;
1494
1495                 if (ti->type->iterate_devices &&
1496                     ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1497                         return true;
1498         }
1499
1500         return false;
1501 }
1502
1503 static bool dm_table_discard_zeroes_data(struct dm_table *t)
1504 {
1505         struct dm_target *ti;
1506         unsigned i = 0;
1507
1508         /* Ensure that all targets supports discard_zeroes_data. */
1509         while (i < dm_table_get_num_targets(t)) {
1510                 ti = dm_table_get_target(t, i++);
1511
1512                 if (ti->discard_zeroes_data_unsupported)
1513                         return false;
1514         }
1515
1516         return true;
1517 }
1518
1519 static int device_is_rotational(struct dm_target *ti, struct dm_dev *dev,
1520                                 sector_t start, sector_t len, void *data)
1521 {
1522         struct request_queue *q = bdev_get_queue(dev->bdev);
1523
1524         return q && !blk_queue_nonrot(q);
1525 }
1526
1527 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1528                              sector_t start, sector_t len, void *data)
1529 {
1530         struct request_queue *q = bdev_get_queue(dev->bdev);
1531
1532         return q && !blk_queue_add_random(q);
1533 }
1534
1535 static int queue_no_sg_merge(struct dm_target *ti, struct dm_dev *dev,
1536                              sector_t start, sector_t len, void *data)
1537 {
1538         struct request_queue *q = bdev_get_queue(dev->bdev);
1539
1540         return q && test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1541 }
1542
1543 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1544                                          sector_t start, sector_t len, void *data)
1545 {
1546         struct request_queue *q = bdev_get_queue(dev->bdev);
1547
1548         return q && !q->limits.max_write_same_sectors;
1549 }
1550
1551 static bool dm_table_supports_write_same(struct dm_table *t)
1552 {
1553         struct dm_target *ti;
1554         unsigned i = 0;
1555
1556         while (i < dm_table_get_num_targets(t)) {
1557                 ti = dm_table_get_target(t, i++);
1558
1559                 if (!ti->num_write_same_bios)
1560                         return false;
1561
1562                 if (!ti->type->iterate_devices ||
1563                     ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1564                         return false;
1565         }
1566
1567         return true;
1568 }
1569
1570 static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1571                                   sector_t start, sector_t len, void *data)
1572 {
1573         struct request_queue *q = bdev_get_queue(dev->bdev);
1574
1575         return q && blk_queue_discard(q);
1576 }
1577
1578 static bool dm_table_supports_discards(struct dm_table *t)
1579 {
1580         struct dm_target *ti;
1581         unsigned i = 0;
1582
1583         /*
1584          * Unless any target used by the table set discards_supported,
1585          * require at least one underlying device to support discards.
1586          * t->devices includes internal dm devices such as mirror logs
1587          * so we need to use iterate_devices here, which targets
1588          * supporting discard selectively must provide.
1589          */
1590         while (i < dm_table_get_num_targets(t)) {
1591                 ti = dm_table_get_target(t, i++);
1592
1593                 if (!ti->num_discard_bios)
1594                         continue;
1595
1596                 if (ti->discards_supported)
1597                         return true;
1598
1599                 if (ti->type->iterate_devices &&
1600                     ti->type->iterate_devices(ti, device_discard_capable, NULL))
1601                         return true;
1602         }
1603
1604         return false;
1605 }
1606
1607 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1608                                struct queue_limits *limits)
1609 {
1610         bool wc = false, fua = false;
1611
1612         /*
1613          * Copy table's limits to the DM device's request_queue
1614          */
1615         q->limits = *limits;
1616
1617         if (!dm_table_supports_discards(t))
1618                 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1619         else
1620                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1621
1622         if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1623                 wc = true;
1624                 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1625                         fua = true;
1626         }
1627         blk_queue_write_cache(q, wc, fua);
1628
1629         if (!dm_table_discard_zeroes_data(t))
1630                 q->limits.discard_zeroes_data = 0;
1631
1632         /* Ensure that all underlying devices are non-rotational. */
1633         if (dm_table_any_dev_attr(t, device_is_rotational))
1634                 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1635         else
1636                 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1637
1638         if (!dm_table_supports_write_same(t))
1639                 q->limits.max_write_same_sectors = 0;
1640
1641         if (dm_table_any_dev_attr(t, queue_no_sg_merge))
1642                 queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1643         else
1644                 queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1645
1646         dm_table_verify_integrity(t);
1647
1648         /*
1649          * Determine whether or not this queue's I/O timings contribute
1650          * to the entropy pool, Only request-based targets use this.
1651          * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1652          * have it set.
1653          */
1654         if (blk_queue_add_random(q) && dm_table_any_dev_attr(t, device_is_not_random))
1655                 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1656
1657         /*
1658          * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1659          * visible to other CPUs because, once the flag is set, incoming bios
1660          * are processed by request-based dm, which refers to the queue
1661          * settings.
1662          * Until the flag set, bios are passed to bio-based dm and queued to
1663          * md->deferred where queue settings are not needed yet.
1664          * Those bios are passed to request-based dm at the resume time.
1665          */
1666         smp_mb();
1667         if (dm_table_request_based(t))
1668                 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1669 }
1670
1671 unsigned int dm_table_get_num_targets(struct dm_table *t)
1672 {
1673         return t->num_targets;
1674 }
1675
1676 struct list_head *dm_table_get_devices(struct dm_table *t)
1677 {
1678         return &t->devices;
1679 }
1680
1681 fmode_t dm_table_get_mode(struct dm_table *t)
1682 {
1683         return t->mode;
1684 }
1685 EXPORT_SYMBOL(dm_table_get_mode);
1686
1687 enum suspend_mode {
1688         PRESUSPEND,
1689         PRESUSPEND_UNDO,
1690         POSTSUSPEND,
1691 };
1692
1693 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1694 {
1695         int i = t->num_targets;
1696         struct dm_target *ti = t->targets;
1697
1698         while (i--) {
1699                 switch (mode) {
1700                 case PRESUSPEND:
1701                         if (ti->type->presuspend)
1702                                 ti->type->presuspend(ti);
1703                         break;
1704                 case PRESUSPEND_UNDO:
1705                         if (ti->type->presuspend_undo)
1706                                 ti->type->presuspend_undo(ti);
1707                         break;
1708                 case POSTSUSPEND:
1709                         if (ti->type->postsuspend)
1710                                 ti->type->postsuspend(ti);
1711                         break;
1712                 }
1713                 ti++;
1714         }
1715 }
1716
1717 void dm_table_presuspend_targets(struct dm_table *t)
1718 {
1719         if (!t)
1720                 return;
1721
1722         suspend_targets(t, PRESUSPEND);
1723 }
1724
1725 void dm_table_presuspend_undo_targets(struct dm_table *t)
1726 {
1727         if (!t)
1728                 return;
1729
1730         suspend_targets(t, PRESUSPEND_UNDO);
1731 }
1732
1733 void dm_table_postsuspend_targets(struct dm_table *t)
1734 {
1735         if (!t)
1736                 return;
1737
1738         suspend_targets(t, POSTSUSPEND);
1739 }
1740
1741 int dm_table_resume_targets(struct dm_table *t)
1742 {
1743         int i, r = 0;
1744
1745         for (i = 0; i < t->num_targets; i++) {
1746                 struct dm_target *ti = t->targets + i;
1747
1748                 if (!ti->type->preresume)
1749                         continue;
1750
1751                 r = ti->type->preresume(ti);
1752                 if (r) {
1753                         DMERR("%s: %s: preresume failed, error = %d",
1754                               dm_device_name(t->md), ti->type->name, r);
1755                         return r;
1756                 }
1757         }
1758
1759         for (i = 0; i < t->num_targets; i++) {
1760                 struct dm_target *ti = t->targets + i;
1761
1762                 if (ti->type->resume)
1763                         ti->type->resume(ti);
1764         }
1765
1766         return 0;
1767 }
1768
1769 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1770 {
1771         list_add(&cb->list, &t->target_callbacks);
1772 }
1773 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1774
1775 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1776 {
1777         struct dm_dev_internal *dd;
1778         struct list_head *devices = dm_table_get_devices(t);
1779         struct dm_target_callbacks *cb;
1780         int r = 0;
1781
1782         list_for_each_entry(dd, devices, list) {
1783                 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1784                 char b[BDEVNAME_SIZE];
1785
1786                 if (likely(q))
1787                         r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1788                 else
1789                         DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1790                                      dm_device_name(t->md),
1791                                      bdevname(dd->dm_dev->bdev, b));
1792         }
1793
1794         list_for_each_entry(cb, &t->target_callbacks, list)
1795                 if (cb->congested_fn)
1796                         r |= cb->congested_fn(cb, bdi_bits);
1797
1798         return r;
1799 }
1800
1801 struct mapped_device *dm_table_get_md(struct dm_table *t)
1802 {
1803         return t->md;
1804 }
1805 EXPORT_SYMBOL(dm_table_get_md);
1806
1807 void dm_table_run_md_queue_async(struct dm_table *t)
1808 {
1809         struct mapped_device *md;
1810         struct request_queue *queue;
1811         unsigned long flags;
1812
1813         if (!dm_table_request_based(t))
1814                 return;
1815
1816         md = dm_table_get_md(t);
1817         queue = dm_get_md_queue(md);
1818         if (queue) {
1819                 if (queue->mq_ops)
1820                         blk_mq_run_hw_queues(queue, true);
1821                 else {
1822                         spin_lock_irqsave(queue->queue_lock, flags);
1823                         blk_run_queue_async(queue);
1824                         spin_unlock_irqrestore(queue->queue_lock, flags);
1825                 }
1826         }
1827 }
1828 EXPORT_SYMBOL(dm_table_run_md_queue_async);
1829