linux/drivers/md/dm-table.c
<<
>>
Prefs
   1/*
   2 * Copyright (C) 2001 Sistina Software (UK) Limited.
   3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
   4 *
   5 * This file is released under the GPL.
   6 */
   7
   8#include "dm.h"
   9
  10#include <linux/module.h>
  11#include <linux/vmalloc.h>
  12#include <linux/blkdev.h>
  13#include <linux/namei.h>
  14#include <linux/ctype.h>
  15#include <linux/string.h>
  16#include <linux/slab.h>
  17#include <linux/interrupt.h>
  18#include <linux/mutex.h>
  19#include <linux/delay.h>
  20#include <linux/atomic.h>
  21
  22#define DM_MSG_PREFIX "table"
  23
  24#define MAX_DEPTH 16
  25#define NODE_SIZE L1_CACHE_BYTES
  26#define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
  27#define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
  28
  29struct dm_table {
  30        struct mapped_device *md;
  31        unsigned type;
  32
  33        /* btree table */
  34        unsigned int depth;
  35        unsigned int counts[MAX_DEPTH]; /* in nodes */
  36        sector_t *index[MAX_DEPTH];
  37
  38        unsigned int num_targets;
  39        unsigned int num_allocated;
  40        sector_t *highs;
  41        struct dm_target *targets;
  42
  43        struct target_type *immutable_target_type;
  44        unsigned integrity_supported:1;
  45        unsigned singleton:1;
  46
  47        /*
  48         * Indicates the rw permissions for the new logical
  49         * device.  This should be a combination of FMODE_READ
  50         * and FMODE_WRITE.
  51         */
  52        fmode_t mode;
  53
  54        /* a list of devices used by this table */
  55        struct list_head devices;
  56
  57        /* events get handed up using this callback */
  58        void (*event_fn)(void *);
  59        void *event_context;
  60
  61        struct dm_md_mempools *mempools;
  62
  63        struct list_head target_callbacks;
  64};
  65
  66/*
  67 * Similar to ceiling(log_size(n))
  68 */
  69static unsigned int int_log(unsigned int n, unsigned int base)
  70{
  71        int result = 0;
  72
  73        while (n > 1) {
  74                n = dm_div_up(n, base);
  75                result++;
  76        }
  77
  78        return result;
  79}
  80
  81/*
  82 * Calculate the index of the child node of the n'th node k'th key.
  83 */
  84static inline unsigned int get_child(unsigned int n, unsigned int k)
  85{
  86        return (n * CHILDREN_PER_NODE) + k;
  87}
  88
  89/*
  90 * Return the n'th node of level l from table t.
  91 */
  92static inline sector_t *get_node(struct dm_table *t,
  93                                 unsigned int l, unsigned int n)
  94{
  95        return t->index[l] + (n * KEYS_PER_NODE);
  96}
  97
  98/*
  99 * Return the highest key that you could lookup from the n'th
 100 * node on level l of the btree.
 101 */
 102static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
 103{
 104        for (; l < t->depth - 1; l++)
 105                n = get_child(n, CHILDREN_PER_NODE - 1);
 106
 107        if (n >= t->counts[l])
 108                return (sector_t) - 1;
 109
 110        return get_node(t, l, n)[KEYS_PER_NODE - 1];
 111}
 112
 113/*
 114 * Fills in a level of the btree based on the highs of the level
 115 * below it.
 116 */
 117static int setup_btree_index(unsigned int l, struct dm_table *t)
 118{
 119        unsigned int n, k;
 120        sector_t *node;
 121
 122        for (n = 0U; n < t->counts[l]; n++) {
 123                node = get_node(t, l, n);
 124
 125                for (k = 0U; k < KEYS_PER_NODE; k++)
 126                        node[k] = high(t, l + 1, get_child(n, k));
 127        }
 128
 129        return 0;
 130}
 131
 132void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
 133{
 134        unsigned long size;
 135        void *addr;
 136
 137        /*
 138         * Check that we're not going to overflow.
 139         */
 140        if (nmemb > (ULONG_MAX / elem_size))
 141                return NULL;
 142
 143        size = nmemb * elem_size;
 144        addr = vzalloc(size);
 145
 146        return addr;
 147}
 148EXPORT_SYMBOL(dm_vcalloc);
 149
 150/*
 151 * highs, and targets are managed as dynamic arrays during a
 152 * table load.
 153 */
 154static int alloc_targets(struct dm_table *t, unsigned int num)
 155{
 156        sector_t *n_highs;
 157        struct dm_target *n_targets;
 158        int n = t->num_targets;
 159
 160        /*
 161         * Allocate both the target array and offset array at once.
 162         * Append an empty entry to catch sectors beyond the end of
 163         * the device.
 164         */
 165        n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
 166                                          sizeof(sector_t));
 167        if (!n_highs)
 168                return -ENOMEM;
 169
 170        n_targets = (struct dm_target *) (n_highs + num);
 171
 172        if (n) {
 173                memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
 174                memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
 175        }
 176
 177        memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
 178        vfree(t->highs);
 179
 180        t->num_allocated = num;
 181        t->highs = n_highs;
 182        t->targets = n_targets;
 183
 184        return 0;
 185}
 186
 187int dm_table_create(struct dm_table **result, fmode_t mode,
 188                    unsigned num_targets, struct mapped_device *md)
 189{
 190        struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
 191
 192        if (!t)
 193                return -ENOMEM;
 194
 195        INIT_LIST_HEAD(&t->devices);
 196        INIT_LIST_HEAD(&t->target_callbacks);
 197
 198        if (!num_targets)
 199                num_targets = KEYS_PER_NODE;
 200
 201        num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
 202
 203        if (alloc_targets(t, num_targets)) {
 204                kfree(t);
 205                return -ENOMEM;
 206        }
 207
 208        t->mode = mode;
 209        t->md = md;
 210        *result = t;
 211        return 0;
 212}
 213
 214static void free_devices(struct list_head *devices)
 215{
 216        struct list_head *tmp, *next;
 217
 218        list_for_each_safe(tmp, next, devices) {
 219                struct dm_dev_internal *dd =
 220                    list_entry(tmp, struct dm_dev_internal, list);
 221                DMWARN("dm_table_destroy: dm_put_device call missing for %s",
 222                       dd->dm_dev.name);
 223                kfree(dd);
 224        }
 225}
 226
 227void dm_table_destroy(struct dm_table *t)
 228{
 229        unsigned int i;
 230
 231        if (!t)
 232                return;
 233
 234        /* free the indexes */
 235        if (t->depth >= 2)
 236                vfree(t->index[t->depth - 2]);
 237
 238        /* free the targets */
 239        for (i = 0; i < t->num_targets; i++) {
 240                struct dm_target *tgt = t->targets + i;
 241
 242                if (tgt->type->dtr)
 243                        tgt->type->dtr(tgt);
 244
 245                dm_put_target_type(tgt->type);
 246        }
 247
 248        vfree(t->highs);
 249
 250        /* free the device list */
 251        free_devices(&t->devices);
 252
 253        dm_free_md_mempools(t->mempools);
 254
 255        kfree(t);
 256}
 257
 258/*
 259 * Checks to see if we need to extend highs or targets.
 260 */
 261static inline int check_space(struct dm_table *t)
 262{
 263        if (t->num_targets >= t->num_allocated)
 264                return alloc_targets(t, t->num_allocated * 2);
 265
 266        return 0;
 267}
 268
 269/*
 270 * See if we've already got a device in the list.
 271 */
 272static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
 273{
 274        struct dm_dev_internal *dd;
 275
 276        list_for_each_entry (dd, l, list)
 277                if (dd->dm_dev.bdev->bd_dev == dev)
 278                        return dd;
 279
 280        return NULL;
 281}
 282
 283/*
 284 * Open a device so we can use it as a map destination.
 285 */
 286static int open_dev(struct dm_dev_internal *d, dev_t dev,
 287                    struct mapped_device *md)
 288{
 289        static char *_claim_ptr = "I belong to device-mapper";
 290        struct block_device *bdev;
 291
 292        int r;
 293
 294        BUG_ON(d->dm_dev.bdev);
 295
 296        bdev = blkdev_get_by_dev(dev, d->dm_dev.mode | FMODE_EXCL, _claim_ptr);
 297        if (IS_ERR(bdev))
 298                return PTR_ERR(bdev);
 299
 300        r = bd_link_disk_holder(bdev, dm_disk(md));
 301        if (r) {
 302                blkdev_put(bdev, d->dm_dev.mode | FMODE_EXCL);
 303                return r;
 304        }
 305
 306        d->dm_dev.bdev = bdev;
 307        return 0;
 308}
 309
 310/*
 311 * Close a device that we've been using.
 312 */
 313static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
 314{
 315        if (!d->dm_dev.bdev)
 316                return;
 317
 318        bd_unlink_disk_holder(d->dm_dev.bdev, dm_disk(md));
 319        blkdev_put(d->dm_dev.bdev, d->dm_dev.mode | FMODE_EXCL);
 320        d->dm_dev.bdev = NULL;
 321}
 322
 323/*
 324 * If possible, this checks an area of a destination device is invalid.
 325 */
 326static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
 327                                  sector_t start, sector_t len, void *data)
 328{
 329        struct request_queue *q;
 330        struct queue_limits *limits = data;
 331        struct block_device *bdev = dev->bdev;
 332        sector_t dev_size =
 333                i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
 334        unsigned short logical_block_size_sectors =
 335                limits->logical_block_size >> SECTOR_SHIFT;
 336        char b[BDEVNAME_SIZE];
 337
 338        /*
 339         * Some devices exist without request functions,
 340         * such as loop devices not yet bound to backing files.
 341         * Forbid the use of such devices.
 342         */
 343        q = bdev_get_queue(bdev);
 344        if (!q || !q->make_request_fn) {
 345                DMWARN("%s: %s is not yet initialised: "
 346                       "start=%llu, len=%llu, dev_size=%llu",
 347                       dm_device_name(ti->table->md), bdevname(bdev, b),
 348                       (unsigned long long)start,
 349                       (unsigned long long)len,
 350                       (unsigned long long)dev_size);
 351                return 1;
 352        }
 353
 354        if (!dev_size)
 355                return 0;
 356
 357        if ((start >= dev_size) || (start + len > dev_size)) {
 358                DMWARN("%s: %s too small for target: "
 359                       "start=%llu, len=%llu, dev_size=%llu",
 360                       dm_device_name(ti->table->md), bdevname(bdev, b),
 361                       (unsigned long long)start,
 362                       (unsigned long long)len,
 363                       (unsigned long long)dev_size);
 364                return 1;
 365        }
 366
 367        if (logical_block_size_sectors <= 1)
 368                return 0;
 369
 370        if (start & (logical_block_size_sectors - 1)) {
 371                DMWARN("%s: start=%llu not aligned to h/w "
 372                       "logical block size %u of %s",
 373                       dm_device_name(ti->table->md),
 374                       (unsigned long long)start,
 375                       limits->logical_block_size, bdevname(bdev, b));
 376                return 1;
 377        }
 378
 379        if (len & (logical_block_size_sectors - 1)) {
 380                DMWARN("%s: len=%llu not aligned to h/w "
 381                       "logical block size %u of %s",
 382                       dm_device_name(ti->table->md),
 383                       (unsigned long long)len,
 384                       limits->logical_block_size, bdevname(bdev, b));
 385                return 1;
 386        }
 387
 388        return 0;
 389}
 390
 391/*
 392 * This upgrades the mode on an already open dm_dev, being
 393 * careful to leave things as they were if we fail to reopen the
 394 * device and not to touch the existing bdev field in case
 395 * it is accessed concurrently inside dm_table_any_congested().
 396 */
 397static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
 398                        struct mapped_device *md)
 399{
 400        int r;
 401        struct dm_dev_internal dd_new, dd_old;
 402
 403        dd_new = dd_old = *dd;
 404
 405        dd_new.dm_dev.mode |= new_mode;
 406        dd_new.dm_dev.bdev = NULL;
 407
 408        r = open_dev(&dd_new, dd->dm_dev.bdev->bd_dev, md);
 409        if (r)
 410                return r;
 411
 412        dd->dm_dev.mode |= new_mode;
 413        close_dev(&dd_old, md);
 414
 415        return 0;
 416}
 417
 418/*
 419 * Add a device to the list, or just increment the usage count if
 420 * it's already present.
 421 */
 422int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
 423                  struct dm_dev **result)
 424{
 425        int r;
 426        dev_t uninitialized_var(dev);
 427        struct dm_dev_internal *dd;
 428        unsigned int major, minor;
 429        struct dm_table *t = ti->table;
 430        char dummy;
 431
 432        BUG_ON(!t);
 433
 434        if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
 435                /* Extract the major/minor numbers */
 436                dev = MKDEV(major, minor);
 437                if (MAJOR(dev) != major || MINOR(dev) != minor)
 438                        return -EOVERFLOW;
 439        } else {
 440                /* convert the path to a device */
 441                struct block_device *bdev = lookup_bdev(path);
 442
 443                if (IS_ERR(bdev))
 444                        return PTR_ERR(bdev);
 445                dev = bdev->bd_dev;
 446                bdput(bdev);
 447        }
 448
 449        dd = find_device(&t->devices, dev);
 450        if (!dd) {
 451                dd = kmalloc(sizeof(*dd), GFP_KERNEL);
 452                if (!dd)
 453                        return -ENOMEM;
 454
 455                dd->dm_dev.mode = mode;
 456                dd->dm_dev.bdev = NULL;
 457
 458                if ((r = open_dev(dd, dev, t->md))) {
 459                        kfree(dd);
 460                        return r;
 461                }
 462
 463                format_dev_t(dd->dm_dev.name, dev);
 464
 465                atomic_set(&dd->count, 0);
 466                list_add(&dd->list, &t->devices);
 467
 468        } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) {
 469                r = upgrade_mode(dd, mode, t->md);
 470                if (r)
 471                        return r;
 472        }
 473        atomic_inc(&dd->count);
 474
 475        *result = &dd->dm_dev;
 476        return 0;
 477}
 478EXPORT_SYMBOL(dm_get_device);
 479
 480int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
 481                         sector_t start, sector_t len, void *data)
 482{
 483        struct queue_limits *limits = data;
 484        struct block_device *bdev = dev->bdev;
 485        struct request_queue *q = bdev_get_queue(bdev);
 486        char b[BDEVNAME_SIZE];
 487
 488        if (unlikely(!q)) {
 489                DMWARN("%s: Cannot set limits for nonexistent device %s",
 490                       dm_device_name(ti->table->md), bdevname(bdev, b));
 491                return 0;
 492        }
 493
 494        if (bdev_stack_limits(limits, bdev, start) < 0)
 495                DMWARN("%s: adding target device %s caused an alignment inconsistency: "
 496                       "physical_block_size=%u, logical_block_size=%u, "
 497                       "alignment_offset=%u, start=%llu",
 498                       dm_device_name(ti->table->md), bdevname(bdev, b),
 499                       q->limits.physical_block_size,
 500                       q->limits.logical_block_size,
 501                       q->limits.alignment_offset,
 502                       (unsigned long long) start << SECTOR_SHIFT);
 503
 504        /*
 505         * Check if merge fn is supported.
 506         * If not we'll force DM to use PAGE_SIZE or
 507         * smaller I/O, just to be safe.
 508         */
 509        if (dm_queue_merge_is_compulsory(q) && !ti->type->merge)
 510                blk_limits_max_hw_sectors(limits,
 511                                          (unsigned int) (PAGE_SIZE >> 9));
 512        return 0;
 513}
 514EXPORT_SYMBOL_GPL(dm_set_device_limits);
 515
 516/*
 517 * Decrement a device's use count and remove it if necessary.
 518 */
 519void dm_put_device(struct dm_target *ti, struct dm_dev *d)
 520{
 521        struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal,
 522                                                  dm_dev);
 523
 524        if (atomic_dec_and_test(&dd->count)) {
 525                close_dev(dd, ti->table->md);
 526                list_del(&dd->list);
 527                kfree(dd);
 528        }
 529}
 530EXPORT_SYMBOL(dm_put_device);
 531
 532/*
 533 * Checks to see if the target joins onto the end of the table.
 534 */
 535static int adjoin(struct dm_table *table, struct dm_target *ti)
 536{
 537        struct dm_target *prev;
 538
 539        if (!table->num_targets)
 540                return !ti->begin;
 541
 542        prev = &table->targets[table->num_targets - 1];
 543        return (ti->begin == (prev->begin + prev->len));
 544}
 545
 546/*
 547 * Used to dynamically allocate the arg array.
 548 */
 549static char **realloc_argv(unsigned *array_size, char **old_argv)
 550{
 551        char **argv;
 552        unsigned new_size;
 553
 554        new_size = *array_size ? *array_size * 2 : 64;
 555        argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
 556        if (argv) {
 557                memcpy(argv, old_argv, *array_size * sizeof(*argv));
 558                *array_size = new_size;
 559        }
 560
 561        kfree(old_argv);
 562        return argv;
 563}
 564
 565/*
 566 * Destructively splits up the argument list to pass to ctr.
 567 */
 568int dm_split_args(int *argc, char ***argvp, char *input)
 569{
 570        char *start, *end = input, *out, **argv = NULL;
 571        unsigned array_size = 0;
 572
 573        *argc = 0;
 574
 575        if (!input) {
 576                *argvp = NULL;
 577                return 0;
 578        }
 579
 580        argv = realloc_argv(&array_size, argv);
 581        if (!argv)
 582                return -ENOMEM;
 583
 584        while (1) {
 585                /* Skip whitespace */
 586                start = skip_spaces(end);
 587
 588                if (!*start)
 589                        break;  /* success, we hit the end */
 590
 591                /* 'out' is used to remove any back-quotes */
 592                end = out = start;
 593                while (*end) {
 594                        /* Everything apart from '\0' can be quoted */
 595                        if (*end == '\\' && *(end + 1)) {
 596                                *out++ = *(end + 1);
 597                                end += 2;
 598                                continue;
 599                        }
 600
 601                        if (isspace(*end))
 602                                break;  /* end of token */
 603
 604                        *out++ = *end++;
 605                }
 606
 607                /* have we already filled the array ? */
 608                if ((*argc + 1) > array_size) {
 609                        argv = realloc_argv(&array_size, argv);
 610                        if (!argv)
 611                                return -ENOMEM;
 612                }
 613
 614                /* we know this is whitespace */
 615                if (*end)
 616                        end++;
 617
 618                /* terminate the string and put it in the array */
 619                *out = '\0';
 620                argv[*argc] = start;
 621                (*argc)++;
 622        }
 623
 624        *argvp = argv;
 625        return 0;
 626}
 627
 628/*
 629 * Impose necessary and sufficient conditions on a devices's table such
 630 * that any incoming bio which respects its logical_block_size can be
 631 * processed successfully.  If it falls across the boundary between
 632 * two or more targets, the size of each piece it gets split into must
 633 * be compatible with the logical_block_size of the target processing it.
 634 */
 635static int validate_hardware_logical_block_alignment(struct dm_table *table,
 636                                                 struct queue_limits *limits)
 637{
 638        /*
 639         * This function uses arithmetic modulo the logical_block_size
 640         * (in units of 512-byte sectors).
 641         */
 642        unsigned short device_logical_block_size_sects =
 643                limits->logical_block_size >> SECTOR_SHIFT;
 644
 645        /*
 646         * Offset of the start of the next table entry, mod logical_block_size.
 647         */
 648        unsigned short next_target_start = 0;
 649
 650        /*
 651         * Given an aligned bio that extends beyond the end of a
 652         * target, how many sectors must the next target handle?
 653         */
 654        unsigned short remaining = 0;
 655
 656        struct dm_target *uninitialized_var(ti);
 657        struct queue_limits ti_limits;
 658        unsigned i = 0;
 659
 660        /*
 661         * Check each entry in the table in turn.
 662         */
 663        while (i < dm_table_get_num_targets(table)) {
 664                ti = dm_table_get_target(table, i++);
 665
 666                blk_set_stacking_limits(&ti_limits);
 667
 668                /* combine all target devices' limits */
 669                if (ti->type->iterate_devices)
 670                        ti->type->iterate_devices(ti, dm_set_device_limits,
 671                                                  &ti_limits);
 672
 673                /*
 674                 * If the remaining sectors fall entirely within this
 675                 * table entry are they compatible with its logical_block_size?
 676                 */
 677                if (remaining < ti->len &&
 678                    remaining & ((ti_limits.logical_block_size >>
 679                                  SECTOR_SHIFT) - 1))
 680                        break;  /* Error */
 681
 682                next_target_start =
 683                    (unsigned short) ((next_target_start + ti->len) &
 684                                      (device_logical_block_size_sects - 1));
 685                remaining = next_target_start ?
 686                    device_logical_block_size_sects - next_target_start : 0;
 687        }
 688
 689        if (remaining) {
 690                DMWARN("%s: table line %u (start sect %llu len %llu) "
 691                       "not aligned to h/w logical block size %u",
 692                       dm_device_name(table->md), i,
 693                       (unsigned long long) ti->begin,
 694                       (unsigned long long) ti->len,
 695                       limits->logical_block_size);
 696                return -EINVAL;
 697        }
 698
 699        return 0;
 700}
 701
 702int dm_table_add_target(struct dm_table *t, const char *type,
 703                        sector_t start, sector_t len, char *params)
 704{
 705        int r = -EINVAL, argc;
 706        char **argv;
 707        struct dm_target *tgt;
 708
 709        if (t->singleton) {
 710                DMERR("%s: target type %s must appear alone in table",
 711                      dm_device_name(t->md), t->targets->type->name);
 712                return -EINVAL;
 713        }
 714
 715        if ((r = check_space(t)))
 716                return r;
 717
 718        tgt = t->targets + t->num_targets;
 719        memset(tgt, 0, sizeof(*tgt));
 720
 721        if (!len) {
 722                DMERR("%s: zero-length target", dm_device_name(t->md));
 723                return -EINVAL;
 724        }
 725
 726        tgt->type = dm_get_target_type(type);
 727        if (!tgt->type) {
 728                DMERR("%s: %s: unknown target type", dm_device_name(t->md),
 729                      type);
 730                return -EINVAL;
 731        }
 732
 733        if (dm_target_needs_singleton(tgt->type)) {
 734                if (t->num_targets) {
 735                        DMERR("%s: target type %s must appear alone in table",
 736                              dm_device_name(t->md), type);
 737                        return -EINVAL;
 738                }
 739                t->singleton = 1;
 740        }
 741
 742        if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
 743                DMERR("%s: target type %s may not be included in read-only tables",
 744                      dm_device_name(t->md), type);
 745                return -EINVAL;
 746        }
 747
 748        if (t->immutable_target_type) {
 749                if (t->immutable_target_type != tgt->type) {
 750                        DMERR("%s: immutable target type %s cannot be mixed with other target types",
 751                              dm_device_name(t->md), t->immutable_target_type->name);
 752                        return -EINVAL;
 753                }
 754        } else if (dm_target_is_immutable(tgt->type)) {
 755                if (t->num_targets) {
 756                        DMERR("%s: immutable target type %s cannot be mixed with other target types",
 757                              dm_device_name(t->md), tgt->type->name);
 758                        return -EINVAL;
 759                }
 760                t->immutable_target_type = tgt->type;
 761        }
 762
 763        tgt->table = t;
 764        tgt->begin = start;
 765        tgt->len = len;
 766        tgt->error = "Unknown error";
 767
 768        /*
 769         * Does this target adjoin the previous one ?
 770         */
 771        if (!adjoin(t, tgt)) {
 772                tgt->error = "Gap in table";
 773                r = -EINVAL;
 774                goto bad;
 775        }
 776
 777        r = dm_split_args(&argc, &argv, params);
 778        if (r) {
 779                tgt->error = "couldn't split parameters (insufficient memory)";
 780                goto bad;
 781        }
 782
 783        r = tgt->type->ctr(tgt, argc, argv);
 784        kfree(argv);
 785        if (r)
 786                goto bad;
 787
 788        t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
 789
 790        if (!tgt->num_discard_bios && tgt->discards_supported)
 791                DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
 792                       dm_device_name(t->md), type);
 793
 794        return 0;
 795
 796 bad:
 797        DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
 798        dm_put_target_type(tgt->type);
 799        return r;
 800}
 801
 802/*
 803 * Target argument parsing helpers.
 804 */
 805static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
 806                             unsigned *value, char **error, unsigned grouped)
 807{
 808        const char *arg_str = dm_shift_arg(arg_set);
 809        char dummy;
 810
 811        if (!arg_str ||
 812            (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
 813            (*value < arg->min) ||
 814            (*value > arg->max) ||
 815            (grouped && arg_set->argc < *value)) {
 816                *error = arg->error;
 817                return -EINVAL;
 818        }
 819
 820        return 0;
 821}
 822
 823int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
 824                unsigned *value, char **error)
 825{
 826        return validate_next_arg(arg, arg_set, value, error, 0);
 827}
 828EXPORT_SYMBOL(dm_read_arg);
 829
 830int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
 831                      unsigned *value, char **error)
 832{
 833        return validate_next_arg(arg, arg_set, value, error, 1);
 834}
 835EXPORT_SYMBOL(dm_read_arg_group);
 836
 837const char *dm_shift_arg(struct dm_arg_set *as)
 838{
 839        char *r;
 840
 841        if (as->argc) {
 842                as->argc--;
 843                r = *as->argv;
 844                as->argv++;
 845                return r;
 846        }
 847
 848        return NULL;
 849}
 850EXPORT_SYMBOL(dm_shift_arg);
 851
 852void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
 853{
 854        BUG_ON(as->argc < num_args);
 855        as->argc -= num_args;
 856        as->argv += num_args;
 857}
 858EXPORT_SYMBOL(dm_consume_args);
 859
 860static int dm_table_set_type(struct dm_table *t)
 861{
 862        unsigned i;
 863        unsigned bio_based = 0, request_based = 0;
 864        struct dm_target *tgt;
 865        struct dm_dev_internal *dd;
 866        struct list_head *devices;
 867
 868        for (i = 0; i < t->num_targets; i++) {
 869                tgt = t->targets + i;
 870                if (dm_target_request_based(tgt))
 871                        request_based = 1;
 872                else
 873                        bio_based = 1;
 874
 875                if (bio_based && request_based) {
 876                        DMWARN("Inconsistent table: different target types"
 877                               " can't be mixed up");
 878                        return -EINVAL;
 879                }
 880        }
 881
 882        if (bio_based) {
 883                /* We must use this table as bio-based */
 884                t->type = DM_TYPE_BIO_BASED;
 885                return 0;
 886        }
 887
 888        BUG_ON(!request_based); /* No targets in this table */
 889
 890        /* Non-request-stackable devices can't be used for request-based dm */
 891        devices = dm_table_get_devices(t);
 892        list_for_each_entry(dd, devices, list) {
 893                if (!blk_queue_stackable(bdev_get_queue(dd->dm_dev.bdev))) {
 894                        DMWARN("table load rejected: including"
 895                               " non-request-stackable devices");
 896                        return -EINVAL;
 897                }
 898        }
 899
 900        /*
 901         * Request-based dm supports only tables that have a single target now.
 902         * To support multiple targets, request splitting support is needed,
 903         * and that needs lots of changes in the block-layer.
 904         * (e.g. request completion process for partial completion.)
 905         */
 906        if (t->num_targets > 1) {
 907                DMWARN("Request-based dm doesn't support multiple targets yet");
 908                return -EINVAL;
 909        }
 910
 911        t->type = DM_TYPE_REQUEST_BASED;
 912
 913        return 0;
 914}
 915
 916unsigned dm_table_get_type(struct dm_table *t)
 917{
 918        return t->type;
 919}
 920
 921struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
 922{
 923        return t->immutable_target_type;
 924}
 925
 926bool dm_table_request_based(struct dm_table *t)
 927{
 928        return dm_table_get_type(t) == DM_TYPE_REQUEST_BASED;
 929}
 930
 931int dm_table_alloc_md_mempools(struct dm_table *t)
 932{
 933        unsigned type = dm_table_get_type(t);
 934        unsigned per_bio_data_size = 0;
 935        struct dm_target *tgt;
 936        unsigned i;
 937
 938        if (unlikely(type == DM_TYPE_NONE)) {
 939                DMWARN("no table type is set, can't allocate mempools");
 940                return -EINVAL;
 941        }
 942
 943        if (type == DM_TYPE_BIO_BASED)
 944                for (i = 0; i < t->num_targets; i++) {
 945                        tgt = t->targets + i;
 946                        per_bio_data_size = max(per_bio_data_size, tgt->per_bio_data_size);
 947                }
 948
 949        t->mempools = dm_alloc_md_mempools(type, t->integrity_supported, per_bio_data_size);
 950        if (!t->mempools)
 951                return -ENOMEM;
 952
 953        return 0;
 954}
 955
 956void dm_table_free_md_mempools(struct dm_table *t)
 957{
 958        dm_free_md_mempools(t->mempools);
 959        t->mempools = NULL;
 960}
 961
 962struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
 963{
 964        return t->mempools;
 965}
 966
 967static int setup_indexes(struct dm_table *t)
 968{
 969        int i;
 970        unsigned int total = 0;
 971        sector_t *indexes;
 972
 973        /* allocate the space for *all* the indexes */
 974        for (i = t->depth - 2; i >= 0; i--) {
 975                t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
 976                total += t->counts[i];
 977        }
 978
 979        indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
 980        if (!indexes)
 981                return -ENOMEM;
 982
 983        /* set up internal nodes, bottom-up */
 984        for (i = t->depth - 2; i >= 0; i--) {
 985                t->index[i] = indexes;
 986                indexes += (KEYS_PER_NODE * t->counts[i]);
 987                setup_btree_index(i, t);
 988        }
 989
 990        return 0;
 991}
 992
 993/*
 994 * Builds the btree to index the map.
 995 */
 996static int dm_table_build_index(struct dm_table *t)
 997{
 998        int r = 0;
 999        unsigned int leaf_nodes;
1000
1001        /* how many indexes will the btree have ? */
1002        leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1003        t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1004
1005        /* leaf layer has already been set up */
1006        t->counts[t->depth - 1] = leaf_nodes;
1007        t->index[t->depth - 1] = t->highs;
1008
1009        if (t->depth >= 2)
1010                r = setup_indexes(t);
1011
1012        return r;
1013}
1014
1015/*
1016 * Get a disk whose integrity profile reflects the table's profile.
1017 * If %match_all is true, all devices' profiles must match.
1018 * If %match_all is false, all devices must at least have an
1019 * allocated integrity profile; but uninitialized is ok.
1020 * Returns NULL if integrity support was inconsistent or unavailable.
1021 */
1022static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t,
1023                                                    bool match_all)
1024{
1025        struct list_head *devices = dm_table_get_devices(t);
1026        struct dm_dev_internal *dd = NULL;
1027        struct gendisk *prev_disk = NULL, *template_disk = NULL;
1028
1029        list_for_each_entry(dd, devices, list) {
1030                template_disk = dd->dm_dev.bdev->bd_disk;
1031                if (!blk_get_integrity(template_disk))
1032                        goto no_integrity;
1033                if (!match_all && !blk_integrity_is_initialized(template_disk))
1034                        continue; /* skip uninitialized profiles */
1035                else if (prev_disk &&
1036                         blk_integrity_compare(prev_disk, template_disk) < 0)
1037                        goto no_integrity;
1038                prev_disk = template_disk;
1039        }
1040
1041        return template_disk;
1042
1043no_integrity:
1044        if (prev_disk)
1045                DMWARN("%s: integrity not set: %s and %s profile mismatch",
1046                       dm_device_name(t->md),
1047                       prev_disk->disk_name,
1048                       template_disk->disk_name);
1049        return NULL;
1050}
1051
1052/*
1053 * Register the mapped device for blk_integrity support if
1054 * the underlying devices have an integrity profile.  But all devices
1055 * may not have matching profiles (checking all devices isn't reliable
1056 * during table load because this table may use other DM device(s) which
1057 * must be resumed before they will have an initialized integity profile).
1058 * Stacked DM devices force a 2 stage integrity profile validation:
1059 * 1 - during load, validate all initialized integrity profiles match
1060 * 2 - during resume, validate all integrity profiles match
1061 */
1062static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md)
1063{
1064        struct gendisk *template_disk = NULL;
1065
1066        template_disk = dm_table_get_integrity_disk(t, false);
1067        if (!template_disk)
1068                return 0;
1069
1070        if (!blk_integrity_is_initialized(dm_disk(md))) {
1071                t->integrity_supported = 1;
1072                return blk_integrity_register(dm_disk(md), NULL);
1073        }
1074
1075        /*
1076         * If DM device already has an initalized integrity
1077         * profile the new profile should not conflict.
1078         */
1079        if (blk_integrity_is_initialized(template_disk) &&
1080            blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1081                DMWARN("%s: conflict with existing integrity profile: "
1082                       "%s profile mismatch",
1083                       dm_device_name(t->md),
1084                       template_disk->disk_name);
1085                return 1;
1086        }
1087
1088        /* Preserve existing initialized integrity profile */
1089        t->integrity_supported = 1;
1090        return 0;
1091}
1092
1093/*
1094 * Prepares the table for use by building the indices,
1095 * setting the type, and allocating mempools.
1096 */
1097int dm_table_complete(struct dm_table *t)
1098{
1099        int r;
1100
1101        r = dm_table_set_type(t);
1102        if (r) {
1103                DMERR("unable to set table type");
1104                return r;
1105        }
1106
1107        r = dm_table_build_index(t);
1108        if (r) {
1109                DMERR("unable to build btrees");
1110                return r;
1111        }
1112
1113        r = dm_table_prealloc_integrity(t, t->md);
1114        if (r) {
1115                DMERR("could not register integrity profile.");
1116                return r;
1117        }
1118
1119        r = dm_table_alloc_md_mempools(t);
1120        if (r)
1121                DMERR("unable to allocate mempools");
1122
1123        return r;
1124}
1125
1126static DEFINE_MUTEX(_event_lock);
1127void dm_table_event_callback(struct dm_table *t,
1128                             void (*fn)(void *), void *context)
1129{
1130        mutex_lock(&_event_lock);
1131        t->event_fn = fn;
1132        t->event_context = context;
1133        mutex_unlock(&_event_lock);
1134}
1135
1136void dm_table_event(struct dm_table *t)
1137{
1138        /*
1139         * You can no longer call dm_table_event() from interrupt
1140         * context, use a bottom half instead.
1141         */
1142        BUG_ON(in_interrupt());
1143
1144        mutex_lock(&_event_lock);
1145        if (t->event_fn)
1146                t->event_fn(t->event_context);
1147        mutex_unlock(&_event_lock);
1148}
1149EXPORT_SYMBOL(dm_table_event);
1150
1151sector_t dm_table_get_size(struct dm_table *t)
1152{
1153        return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1154}
1155EXPORT_SYMBOL(dm_table_get_size);
1156
1157struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1158{
1159        if (index >= t->num_targets)
1160                return NULL;
1161
1162        return t->targets + index;
1163}
1164
1165/*
1166 * Search the btree for the correct target.
1167 *
1168 * Caller should check returned pointer with dm_target_is_valid()
1169 * to trap I/O beyond end of device.
1170 */
1171struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1172{
1173        unsigned int l, n = 0, k = 0;
1174        sector_t *node;
1175
1176        for (l = 0; l < t->depth; l++) {
1177                n = get_child(n, k);
1178                node = get_node(t, l, n);
1179
1180                for (k = 0; k < KEYS_PER_NODE; k++)
1181                        if (node[k] >= sector)
1182                                break;
1183        }
1184
1185        return &t->targets[(KEYS_PER_NODE * n) + k];
1186}
1187
1188static int count_device(struct dm_target *ti, struct dm_dev *dev,
1189                        sector_t start, sector_t len, void *data)
1190{
1191        unsigned *num_devices = data;
1192
1193        (*num_devices)++;
1194
1195        return 0;
1196}
1197
1198/*
1199 * Check whether a table has no data devices attached using each
1200 * target's iterate_devices method.
1201 * Returns false if the result is unknown because a target doesn't
1202 * support iterate_devices.
1203 */
1204bool dm_table_has_no_data_devices(struct dm_table *table)
1205{
1206        struct dm_target *uninitialized_var(ti);
1207        unsigned i = 0, num_devices = 0;
1208
1209        while (i < dm_table_get_num_targets(table)) {
1210                ti = dm_table_get_target(table, i++);
1211
1212                if (!ti->type->iterate_devices)
1213                        return false;
1214
1215                ti->type->iterate_devices(ti, count_device, &num_devices);
1216                if (num_devices)
1217                        return false;
1218        }
1219
1220        return true;
1221}
1222
1223/*
1224 * Establish the new table's queue_limits and validate them.
1225 */
1226int dm_calculate_queue_limits(struct dm_table *table,
1227                              struct queue_limits *limits)
1228{
1229        struct dm_target *uninitialized_var(ti);
1230        struct queue_limits ti_limits;
1231        unsigned i = 0;
1232
1233        blk_set_stacking_limits(limits);
1234
1235        while (i < dm_table_get_num_targets(table)) {
1236                blk_set_stacking_limits(&ti_limits);
1237
1238                ti = dm_table_get_target(table, i++);
1239
1240                if (!ti->type->iterate_devices)
1241                        goto combine_limits;
1242
1243                /*
1244                 * Combine queue limits of all the devices this target uses.
1245                 */
1246                ti->type->iterate_devices(ti, dm_set_device_limits,
1247                                          &ti_limits);
1248
1249                /* Set I/O hints portion of queue limits */
1250                if (ti->type->io_hints)
1251                        ti->type->io_hints(ti, &ti_limits);
1252
1253                /*
1254                 * Check each device area is consistent with the target's
1255                 * overall queue limits.
1256                 */
1257                if (ti->type->iterate_devices(ti, device_area_is_invalid,
1258                                              &ti_limits))
1259                        return -EINVAL;
1260
1261combine_limits:
1262                /*
1263                 * Merge this target's queue limits into the overall limits
1264                 * for the table.
1265                 */
1266                if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1267                        DMWARN("%s: adding target device "
1268                               "(start sect %llu len %llu) "
1269                               "caused an alignment inconsistency",
1270                               dm_device_name(table->md),
1271                               (unsigned long long) ti->begin,
1272                               (unsigned long long) ti->len);
1273        }
1274
1275        return validate_hardware_logical_block_alignment(table, limits);
1276}
1277
1278/*
1279 * Set the integrity profile for this device if all devices used have
1280 * matching profiles.  We're quite deep in the resume path but still
1281 * don't know if all devices (particularly DM devices this device
1282 * may be stacked on) have matching profiles.  Even if the profiles
1283 * don't match we have no way to fail (to resume) at this point.
1284 */
1285static void dm_table_set_integrity(struct dm_table *t)
1286{
1287        struct gendisk *template_disk = NULL;
1288
1289        if (!blk_get_integrity(dm_disk(t->md)))
1290                return;
1291
1292        template_disk = dm_table_get_integrity_disk(t, true);
1293        if (template_disk)
1294                blk_integrity_register(dm_disk(t->md),
1295                                       blk_get_integrity(template_disk));
1296        else if (blk_integrity_is_initialized(dm_disk(t->md)))
1297                DMWARN("%s: device no longer has a valid integrity profile",
1298                       dm_device_name(t->md));
1299        else
1300                DMWARN("%s: unable to establish an integrity profile",
1301                       dm_device_name(t->md));
1302}
1303
1304static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1305                                sector_t start, sector_t len, void *data)
1306{
1307        unsigned flush = (*(unsigned *)data);
1308        struct request_queue *q = bdev_get_queue(dev->bdev);
1309
1310        return q && (q->flush_flags & flush);
1311}
1312
1313static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1314{
1315        struct dm_target *ti;
1316        unsigned i = 0;
1317
1318        /*
1319         * Require at least one underlying device to support flushes.
1320         * t->devices includes internal dm devices such as mirror logs
1321         * so we need to use iterate_devices here, which targets
1322         * supporting flushes must provide.
1323         */
1324        while (i < dm_table_get_num_targets(t)) {
1325                ti = dm_table_get_target(t, i++);
1326
1327                if (!ti->num_flush_bios)
1328                        continue;
1329
1330                if (ti->flush_supported)
1331                        return 1;
1332
1333                if (ti->type->iterate_devices &&
1334                    ti->type->iterate_devices(ti, device_flush_capable, &flush))
1335                        return 1;
1336        }
1337
1338        return 0;
1339}
1340
1341static bool dm_table_discard_zeroes_data(struct dm_table *t)
1342{
1343        struct dm_target *ti;
1344        unsigned i = 0;
1345
1346        /* Ensure that all targets supports discard_zeroes_data. */
1347        while (i < dm_table_get_num_targets(t)) {
1348                ti = dm_table_get_target(t, i++);
1349
1350                if (ti->discard_zeroes_data_unsupported)
1351                        return 0;
1352        }
1353
1354        return 1;
1355}
1356
1357static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1358                            sector_t start, sector_t len, void *data)
1359{
1360        struct request_queue *q = bdev_get_queue(dev->bdev);
1361
1362        return q && blk_queue_nonrot(q);
1363}
1364
1365static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1366                             sector_t start, sector_t len, void *data)
1367{
1368        struct request_queue *q = bdev_get_queue(dev->bdev);
1369
1370        return q && !blk_queue_add_random(q);
1371}
1372
1373static bool dm_table_all_devices_attribute(struct dm_table *t,
1374                                           iterate_devices_callout_fn func)
1375{
1376        struct dm_target *ti;
1377        unsigned i = 0;
1378
1379        while (i < dm_table_get_num_targets(t)) {
1380                ti = dm_table_get_target(t, i++);
1381
1382                if (!ti->type->iterate_devices ||
1383                    !ti->type->iterate_devices(ti, func, NULL))
1384                        return 0;
1385        }
1386
1387        return 1;
1388}
1389
1390static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1391                                         sector_t start, sector_t len, void *data)
1392{
1393        struct request_queue *q = bdev_get_queue(dev->bdev);
1394
1395        return q && !q->limits.max_write_same_sectors;
1396}
1397
1398static bool dm_table_supports_write_same(struct dm_table *t)
1399{
1400        struct dm_target *ti;
1401        unsigned i = 0;
1402
1403        while (i < dm_table_get_num_targets(t)) {
1404                ti = dm_table_get_target(t, i++);
1405
1406                if (!ti->num_write_same_bios)
1407                        return false;
1408
1409                if (!ti->type->iterate_devices ||
1410                    ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1411                        return false;
1412        }
1413
1414        return true;
1415}
1416
1417void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1418                               struct queue_limits *limits)
1419{
1420        unsigned flush = 0;
1421
1422        /*
1423         * Copy table's limits to the DM device's request_queue
1424         */
1425        q->limits = *limits;
1426
1427        if (!dm_table_supports_discards(t))
1428                queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1429        else
1430                queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1431
1432        if (dm_table_supports_flush(t, REQ_FLUSH)) {
1433                flush |= REQ_FLUSH;
1434                if (dm_table_supports_flush(t, REQ_FUA))
1435                        flush |= REQ_FUA;
1436        }
1437        blk_queue_flush(q, flush);
1438
1439        if (!dm_table_discard_zeroes_data(t))
1440                q->limits.discard_zeroes_data = 0;
1441
1442        /* Ensure that all underlying devices are non-rotational. */
1443        if (dm_table_all_devices_attribute(t, device_is_nonrot))
1444                queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1445        else
1446                queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1447
1448        if (!dm_table_supports_write_same(t))
1449                q->limits.max_write_same_sectors = 0;
1450
1451        dm_table_set_integrity(t);
1452
1453        /*
1454         * Determine whether or not this queue's I/O timings contribute
1455         * to the entropy pool, Only request-based targets use this.
1456         * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1457         * have it set.
1458         */
1459        if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1460                queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1461
1462        /*
1463         * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1464         * visible to other CPUs because, once the flag is set, incoming bios
1465         * are processed by request-based dm, which refers to the queue
1466         * settings.
1467         * Until the flag set, bios are passed to bio-based dm and queued to
1468         * md->deferred where queue settings are not needed yet.
1469         * Those bios are passed to request-based dm at the resume time.
1470         */
1471        smp_mb();
1472        if (dm_table_request_based(t))
1473                queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1474}
1475
1476unsigned int dm_table_get_num_targets(struct dm_table *t)
1477{
1478        return t->num_targets;
1479}
1480
1481struct list_head *dm_table_get_devices(struct dm_table *t)
1482{
1483        return &t->devices;
1484}
1485
1486fmode_t dm_table_get_mode(struct dm_table *t)
1487{
1488        return t->mode;
1489}
1490EXPORT_SYMBOL(dm_table_get_mode);
1491
1492static void suspend_targets(struct dm_table *t, unsigned postsuspend)
1493{
1494        int i = t->num_targets;
1495        struct dm_target *ti = t->targets;
1496
1497        while (i--) {
1498                if (postsuspend) {
1499                        if (ti->type->postsuspend)
1500                                ti->type->postsuspend(ti);
1501                } else if (ti->type->presuspend)
1502                        ti->type->presuspend(ti);
1503
1504                ti++;
1505        }
1506}
1507
1508void dm_table_presuspend_targets(struct dm_table *t)
1509{
1510        if (!t)
1511                return;
1512
1513        suspend_targets(t, 0);
1514}
1515
1516void dm_table_postsuspend_targets(struct dm_table *t)
1517{
1518        if (!t)
1519                return;
1520
1521        suspend_targets(t, 1);
1522}
1523
1524int dm_table_resume_targets(struct dm_table *t)
1525{
1526        int i, r = 0;
1527
1528        for (i = 0; i < t->num_targets; i++) {
1529                struct dm_target *ti = t->targets + i;
1530
1531                if (!ti->type->preresume)
1532                        continue;
1533
1534                r = ti->type->preresume(ti);
1535                if (r)
1536                        return r;
1537        }
1538
1539        for (i = 0; i < t->num_targets; i++) {
1540                struct dm_target *ti = t->targets + i;
1541
1542                if (ti->type->resume)
1543                        ti->type->resume(ti);
1544        }
1545
1546        return 0;
1547}
1548
1549void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1550{
1551        list_add(&cb->list, &t->target_callbacks);
1552}
1553EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1554
1555int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1556{
1557        struct dm_dev_internal *dd;
1558        struct list_head *devices = dm_table_get_devices(t);
1559        struct dm_target_callbacks *cb;
1560        int r = 0;
1561
1562        list_for_each_entry(dd, devices, list) {
1563                struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
1564                char b[BDEVNAME_SIZE];
1565
1566                if (likely(q))
1567                        r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1568                else
1569                        DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1570                                     dm_device_name(t->md),
1571                                     bdevname(dd->dm_dev.bdev, b));
1572        }
1573
1574        list_for_each_entry(cb, &t->target_callbacks, list)
1575                if (cb->congested_fn)
1576                        r |= cb->congested_fn(cb, bdi_bits);
1577
1578        return r;
1579}
1580
1581int dm_table_any_busy_target(struct dm_table *t)
1582{
1583        unsigned i;
1584        struct dm_target *ti;
1585
1586        for (i = 0; i < t->num_targets; i++) {
1587                ti = t->targets + i;
1588                if (ti->type->busy && ti->type->busy(ti))
1589                        return 1;
1590        }
1591
1592        return 0;
1593}
1594
1595struct mapped_device *dm_table_get_md(struct dm_table *t)
1596{
1597        return t->md;
1598}
1599EXPORT_SYMBOL(dm_table_get_md);
1600
1601static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1602                                  sector_t start, sector_t len, void *data)
1603{
1604        struct request_queue *q = bdev_get_queue(dev->bdev);
1605
1606        return q && blk_queue_discard(q);
1607}
1608
1609bool dm_table_supports_discards(struct dm_table *t)
1610{
1611        struct dm_target *ti;
1612        unsigned i = 0;
1613
1614        /*
1615         * Unless any target used by the table set discards_supported,
1616         * require at least one underlying device to support discards.
1617         * t->devices includes internal dm devices such as mirror logs
1618         * so we need to use iterate_devices here, which targets
1619         * supporting discard selectively must provide.
1620         */
1621        while (i < dm_table_get_num_targets(t)) {
1622                ti = dm_table_get_target(t, i++);
1623
1624                if (!ti->num_discard_bios)
1625                        continue;
1626
1627                if (ti->discards_supported)
1628                        return 1;
1629
1630                if (ti->type->iterate_devices &&
1631                    ti->type->iterate_devices(ti, device_discard_capable, NULL))
1632                        return 1;
1633        }
1634
1635        return 0;
1636}
1637
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.