linux/drivers/md/dm-table.c
<<
>>
Prefs
   1/*
   2 * Copyright (C) 2001 Sistina Software (UK) Limited.
   3 * Copyright (C) 2004 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/slab.h>
  16#include <linux/interrupt.h>
  17#include <linux/mutex.h>
  18#include <asm/atomic.h>
  19
  20#define DM_MSG_PREFIX "table"
  21
  22#define MAX_DEPTH 16
  23#define NODE_SIZE L1_CACHE_BYTES
  24#define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
  25#define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
  26
  27struct dm_table {
  28        struct mapped_device *md;
  29        atomic_t holders;
  30
  31        /* btree table */
  32        unsigned int depth;
  33        unsigned int counts[MAX_DEPTH]; /* in nodes */
  34        sector_t *index[MAX_DEPTH];
  35
  36        unsigned int num_targets;
  37        unsigned int num_allocated;
  38        sector_t *highs;
  39        struct dm_target *targets;
  40
  41        /*
  42         * Indicates the rw permissions for the new logical
  43         * device.  This should be a combination of FMODE_READ
  44         * and FMODE_WRITE.
  45         */
  46        int mode;
  47
  48        /* a list of devices used by this table */
  49        struct list_head devices;
  50
  51        /*
  52         * These are optimistic limits taken from all the
  53         * targets, some targets will need smaller limits.
  54         */
  55        struct io_restrictions limits;
  56
  57        /* events get handed up using this callback */
  58        void (*event_fn)(void *);
  59        void *event_context;
  60};
  61
  62/*
  63 * Similar to ceiling(log_size(n))
  64 */
  65static unsigned int int_log(unsigned int n, unsigned int base)
  66{
  67        int result = 0;
  68
  69        while (n > 1) {
  70                n = dm_div_up(n, base);
  71                result++;
  72        }
  73
  74        return result;
  75}
  76
  77/*
  78 * Returns the minimum that is _not_ zero, unless both are zero.
  79 */
  80#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
  81
  82/*
  83 * Combine two io_restrictions, always taking the lower value.
  84 */
  85static void combine_restrictions_low(struct io_restrictions *lhs,
  86                                     struct io_restrictions *rhs)
  87{
  88        lhs->max_sectors =
  89                min_not_zero(lhs->max_sectors, rhs->max_sectors);
  90
  91        lhs->max_phys_segments =
  92                min_not_zero(lhs->max_phys_segments, rhs->max_phys_segments);
  93
  94        lhs->max_hw_segments =
  95                min_not_zero(lhs->max_hw_segments, rhs->max_hw_segments);
  96
  97        lhs->hardsect_size = max(lhs->hardsect_size, rhs->hardsect_size);
  98
  99        lhs->max_segment_size =
 100                min_not_zero(lhs->max_segment_size, rhs->max_segment_size);
 101
 102        lhs->max_hw_sectors =
 103                min_not_zero(lhs->max_hw_sectors, rhs->max_hw_sectors);
 104
 105        lhs->seg_boundary_mask =
 106                min_not_zero(lhs->seg_boundary_mask, rhs->seg_boundary_mask);
 107
 108        lhs->bounce_pfn = min_not_zero(lhs->bounce_pfn, rhs->bounce_pfn);
 109
 110        lhs->no_cluster |= rhs->no_cluster;
 111}
 112
 113/*
 114 * Calculate the index of the child node of the n'th node k'th key.
 115 */
 116static inline unsigned int get_child(unsigned int n, unsigned int k)
 117{
 118        return (n * CHILDREN_PER_NODE) + k;
 119}
 120
 121/*
 122 * Return the n'th node of level l from table t.
 123 */
 124static inline sector_t *get_node(struct dm_table *t,
 125                                 unsigned int l, unsigned int n)
 126{
 127        return t->index[l] + (n * KEYS_PER_NODE);
 128}
 129
 130/*
 131 * Return the highest key that you could lookup from the n'th
 132 * node on level l of the btree.
 133 */
 134static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
 135{
 136        for (; l < t->depth - 1; l++)
 137                n = get_child(n, CHILDREN_PER_NODE - 1);
 138
 139        if (n >= t->counts[l])
 140                return (sector_t) - 1;
 141
 142        return get_node(t, l, n)[KEYS_PER_NODE - 1];
 143}
 144
 145/*
 146 * Fills in a level of the btree based on the highs of the level
 147 * below it.
 148 */
 149static int setup_btree_index(unsigned int l, struct dm_table *t)
 150{
 151        unsigned int n, k;
 152        sector_t *node;
 153
 154        for (n = 0U; n < t->counts[l]; n++) {
 155                node = get_node(t, l, n);
 156
 157                for (k = 0U; k < KEYS_PER_NODE; k++)
 158                        node[k] = high(t, l + 1, get_child(n, k));
 159        }
 160
 161        return 0;
 162}
 163
 164void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
 165{
 166        unsigned long size;
 167        void *addr;
 168
 169        /*
 170         * Check that we're not going to overflow.
 171         */
 172        if (nmemb > (ULONG_MAX / elem_size))
 173                return NULL;
 174
 175        size = nmemb * elem_size;
 176        addr = vmalloc(size);
 177        if (addr)
 178                memset(addr, 0, size);
 179
 180        return addr;
 181}
 182
 183/*
 184 * highs, and targets are managed as dynamic arrays during a
 185 * table load.
 186 */
 187static int alloc_targets(struct dm_table *t, unsigned int num)
 188{
 189        sector_t *n_highs;
 190        struct dm_target *n_targets;
 191        int n = t->num_targets;
 192
 193        /*
 194         * Allocate both the target array and offset array at once.
 195         * Append an empty entry to catch sectors beyond the end of
 196         * the device.
 197         */
 198        n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
 199                                          sizeof(sector_t));
 200        if (!n_highs)
 201                return -ENOMEM;
 202
 203        n_targets = (struct dm_target *) (n_highs + num);
 204
 205        if (n) {
 206                memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
 207                memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
 208        }
 209
 210        memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
 211        vfree(t->highs);
 212
 213        t->num_allocated = num;
 214        t->highs = n_highs;
 215        t->targets = n_targets;
 216
 217        return 0;
 218}
 219
 220int dm_table_create(struct dm_table **result, int mode,
 221                    unsigned num_targets, struct mapped_device *md)
 222{
 223        struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
 224
 225        if (!t)
 226                return -ENOMEM;
 227
 228        INIT_LIST_HEAD(&t->devices);
 229        atomic_set(&t->holders, 1);
 230
 231        if (!num_targets)
 232                num_targets = KEYS_PER_NODE;
 233
 234        num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
 235
 236        if (alloc_targets(t, num_targets)) {
 237                kfree(t);
 238                t = NULL;
 239                return -ENOMEM;
 240        }
 241
 242        t->mode = mode;
 243        t->md = md;
 244        *result = t;
 245        return 0;
 246}
 247
 248int dm_create_error_table(struct dm_table **result, struct mapped_device *md)
 249{
 250        struct dm_table *t;
 251        sector_t dev_size = 1;
 252        int r;
 253
 254        /*
 255         * Find current size of device.
 256         * Default to 1 sector if inactive.
 257         */
 258        t = dm_get_table(md);
 259        if (t) {
 260                dev_size = dm_table_get_size(t);
 261                dm_table_put(t);
 262        }
 263
 264        r = dm_table_create(&t, FMODE_READ, 1, md);
 265        if (r)
 266                return r;
 267
 268        r = dm_table_add_target(t, "error", 0, dev_size, NULL);
 269        if (r)
 270                goto out;
 271
 272        r = dm_table_complete(t);
 273        if (r)
 274                goto out;
 275
 276        *result = t;
 277
 278out:
 279        if (r)
 280                dm_table_put(t);
 281
 282        return r;
 283}
 284EXPORT_SYMBOL_GPL(dm_create_error_table);
 285
 286static void free_devices(struct list_head *devices)
 287{
 288        struct list_head *tmp, *next;
 289
 290        for (tmp = devices->next; tmp != devices; tmp = next) {
 291                struct dm_dev *dd = list_entry(tmp, struct dm_dev, list);
 292                next = tmp->next;
 293                kfree(dd);
 294        }
 295}
 296
 297static void table_destroy(struct dm_table *t)
 298{
 299        unsigned int i;
 300
 301        /* free the indexes (see dm_table_complete) */
 302        if (t->depth >= 2)
 303                vfree(t->index[t->depth - 2]);
 304
 305        /* free the targets */
 306        for (i = 0; i < t->num_targets; i++) {
 307                struct dm_target *tgt = t->targets + i;
 308
 309                if (tgt->type->dtr)
 310                        tgt->type->dtr(tgt);
 311
 312                dm_put_target_type(tgt->type);
 313        }
 314
 315        vfree(t->highs);
 316
 317        /* free the device list */
 318        if (t->devices.next != &t->devices) {
 319                DMWARN("devices still present during destroy: "
 320                       "dm_table_remove_device calls missing");
 321
 322                free_devices(&t->devices);
 323        }
 324
 325        kfree(t);
 326}
 327
 328void dm_table_get(struct dm_table *t)
 329{
 330        atomic_inc(&t->holders);
 331}
 332
 333void dm_table_put(struct dm_table *t)
 334{
 335        if (!t)
 336                return;
 337
 338        if (atomic_dec_and_test(&t->holders))
 339                table_destroy(t);
 340}
 341
 342/*
 343 * Checks to see if we need to extend highs or targets.
 344 */
 345static inline int check_space(struct dm_table *t)
 346{
 347        if (t->num_targets >= t->num_allocated)
 348                return alloc_targets(t, t->num_allocated * 2);
 349
 350        return 0;
 351}
 352
 353/*
 354 * Convert a device path to a dev_t.
 355 */
 356static int lookup_device(const char *path, dev_t *dev)
 357{
 358        int r;
 359        struct nameidata nd;
 360        struct inode *inode;
 361
 362        if ((r = path_lookup(path, LOOKUP_FOLLOW, &nd)))
 363                return r;
 364
 365        inode = nd.dentry->d_inode;
 366        if (!inode) {
 367                r = -ENOENT;
 368                goto out;
 369        }
 370
 371        if (!S_ISBLK(inode->i_mode)) {
 372                r = -ENOTBLK;
 373                goto out;
 374        }
 375
 376        *dev = inode->i_rdev;
 377
 378 out:
 379        path_release(&nd);
 380        return r;
 381}
 382
 383/*
 384 * See if we've already got a device in the list.
 385 */
 386static struct dm_dev *find_device(struct list_head *l, dev_t dev)
 387{
 388        struct dm_dev *dd;
 389
 390        list_for_each_entry (dd, l, list)
 391                if (dd->bdev->bd_dev == dev)
 392                        return dd;
 393
 394        return NULL;
 395}
 396
 397/*
 398 * Open a device so we can use it as a map destination.
 399 */
 400static int open_dev(struct dm_dev *d, dev_t dev, struct mapped_device *md)
 401{
 402        static char *_claim_ptr = "I belong to device-mapper";
 403        struct block_device *bdev;
 404
 405        int r;
 406
 407        BUG_ON(d->bdev);
 408
 409        bdev = open_by_devnum(dev, d->mode);
 410        if (IS_ERR(bdev))
 411                return PTR_ERR(bdev);
 412        r = bd_claim_by_disk(bdev, _claim_ptr, dm_disk(md));
 413        if (r)
 414                blkdev_put(bdev);
 415        else
 416                d->bdev = bdev;
 417        return r;
 418}
 419
 420/*
 421 * Close a device that we've been using.
 422 */
 423static void close_dev(struct dm_dev *d, struct mapped_device *md)
 424{
 425        if (!d->bdev)
 426                return;
 427
 428        bd_release_from_disk(d->bdev, dm_disk(md));
 429        blkdev_put(d->bdev);
 430        d->bdev = NULL;
 431}
 432
 433/*
 434 * If possible, this checks an area of a destination device is valid.
 435 */
 436static int check_device_area(struct dm_dev *dd, sector_t start, sector_t len)
 437{
 438        sector_t dev_size = dd->bdev->bd_inode->i_size >> SECTOR_SHIFT;
 439
 440        if (!dev_size)
 441                return 1;
 442
 443        return ((start < dev_size) && (len <= (dev_size - start)));
 444}
 445
 446/*
 447 * This upgrades the mode on an already open dm_dev.  Being
 448 * careful to leave things as they were if we fail to reopen the
 449 * device.
 450 */
 451static int upgrade_mode(struct dm_dev *dd, int new_mode, struct mapped_device *md)
 452{
 453        int r;
 454        struct dm_dev dd_copy;
 455        dev_t dev = dd->bdev->bd_dev;
 456
 457        dd_copy = *dd;
 458
 459        dd->mode |= new_mode;
 460        dd->bdev = NULL;
 461        r = open_dev(dd, dev, md);
 462        if (!r)
 463                close_dev(&dd_copy, md);
 464        else
 465                *dd = dd_copy;
 466
 467        return r;
 468}
 469
 470/*
 471 * Add a device to the list, or just increment the usage count if
 472 * it's already present.
 473 */
 474static int __table_get_device(struct dm_table *t, struct dm_target *ti,
 475                              const char *path, sector_t start, sector_t len,
 476                              int mode, struct dm_dev **result)
 477{
 478        int r;
 479        dev_t dev;
 480        struct dm_dev *dd;
 481        unsigned int major, minor;
 482
 483        BUG_ON(!t);
 484
 485        if (sscanf(path, "%u:%u", &major, &minor) == 2) {
 486                /* Extract the major/minor numbers */
 487                dev = MKDEV(major, minor);
 488                if (MAJOR(dev) != major || MINOR(dev) != minor)
 489                        return -EOVERFLOW;
 490        } else {
 491                /* convert the path to a device */
 492                if ((r = lookup_device(path, &dev)))
 493                        return r;
 494        }
 495
 496        dd = find_device(&t->devices, dev);
 497        if (!dd) {
 498                dd = kmalloc(sizeof(*dd), GFP_KERNEL);
 499                if (!dd)
 500                        return -ENOMEM;
 501
 502                dd->mode = mode;
 503                dd->bdev = NULL;
 504
 505                if ((r = open_dev(dd, dev, t->md))) {
 506                        kfree(dd);
 507                        return r;
 508                }
 509
 510                format_dev_t(dd->name, dev);
 511
 512                atomic_set(&dd->count, 0);
 513                list_add(&dd->list, &t->devices);
 514
 515        } else if (dd->mode != (mode | dd->mode)) {
 516                r = upgrade_mode(dd, mode, t->md);
 517                if (r)
 518                        return r;
 519        }
 520        atomic_inc(&dd->count);
 521
 522        if (!check_device_area(dd, start, len)) {
 523                DMWARN("device %s too small for target", path);
 524                dm_put_device(ti, dd);
 525                return -EINVAL;
 526        }
 527
 528        *result = dd;
 529
 530        return 0;
 531}
 532
 533void dm_set_device_limits(struct dm_target *ti, struct block_device *bdev)
 534{
 535        struct request_queue *q = bdev_get_queue(bdev);
 536        struct io_restrictions *rs = &ti->limits;
 537
 538        /*
 539         * Combine the device limits low.
 540         *
 541         * FIXME: if we move an io_restriction struct
 542         *        into q this would just be a call to
 543         *        combine_restrictions_low()
 544         */
 545        rs->max_sectors =
 546                min_not_zero(rs->max_sectors, q->max_sectors);
 547
 548        /* FIXME: Device-Mapper on top of RAID-0 breaks because DM
 549         *        currently doesn't honor MD's merge_bvec_fn routine.
 550         *        In this case, we'll force DM to use PAGE_SIZE or
 551         *        smaller I/O, just to be safe. A better fix is in the
 552         *        works, but add this for the time being so it will at
 553         *        least operate correctly.
 554         */
 555        if (q->merge_bvec_fn)
 556                rs->max_sectors =
 557                        min_not_zero(rs->max_sectors,
 558                                     (unsigned int) (PAGE_SIZE >> 9));
 559
 560        rs->max_phys_segments =
 561                min_not_zero(rs->max_phys_segments,
 562                             q->max_phys_segments);
 563
 564        rs->max_hw_segments =
 565                min_not_zero(rs->max_hw_segments, q->max_hw_segments);
 566
 567        rs->hardsect_size = max(rs->hardsect_size, q->hardsect_size);
 568
 569        rs->max_segment_size =
 570                min_not_zero(rs->max_segment_size, q->max_segment_size);
 571
 572        rs->max_hw_sectors =
 573                min_not_zero(rs->max_hw_sectors, q->max_hw_sectors);
 574
 575        rs->seg_boundary_mask =
 576                min_not_zero(rs->seg_boundary_mask,
 577                             q->seg_boundary_mask);
 578
 579        rs->bounce_pfn = min_not_zero(rs->bounce_pfn, q->bounce_pfn);
 580
 581        rs->no_cluster |= !test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
 582}
 583EXPORT_SYMBOL_GPL(dm_set_device_limits);
 584
 585int dm_get_device(struct dm_target *ti, const char *path, sector_t start,
 586                  sector_t len, int mode, struct dm_dev **result)
 587{
 588        int r = __table_get_device(ti->table, ti, path,
 589                                   start, len, mode, result);
 590
 591        if (!r)
 592                dm_set_device_limits(ti, (*result)->bdev);
 593
 594        return r;
 595}
 596
 597/*
 598 * Decrement a devices use count and remove it if necessary.
 599 */
 600void dm_put_device(struct dm_target *ti, struct dm_dev *dd)
 601{
 602        if (atomic_dec_and_test(&dd->count)) {
 603                close_dev(dd, ti->table->md);
 604                list_del(&dd->list);
 605                kfree(dd);
 606        }
 607}
 608
 609/*
 610 * Checks to see if the target joins onto the end of the table.
 611 */
 612static int adjoin(struct dm_table *table, struct dm_target *ti)
 613{
 614        struct dm_target *prev;
 615
 616        if (!table->num_targets)
 617                return !ti->begin;
 618
 619        prev = &table->targets[table->num_targets - 1];
 620        return (ti->begin == (prev->begin + prev->len));
 621}
 622
 623/*
 624 * Used to dynamically allocate the arg array.
 625 */
 626static char **realloc_argv(unsigned *array_size, char **old_argv)
 627{
 628        char **argv;
 629        unsigned new_size;
 630
 631        new_size = *array_size ? *array_size * 2 : 64;
 632        argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
 633        if (argv) {
 634                memcpy(argv, old_argv, *array_size * sizeof(*argv));
 635                *array_size = new_size;
 636        }
 637
 638        kfree(old_argv);
 639        return argv;
 640}
 641
 642/*
 643 * Destructively splits up the argument list to pass to ctr.
 644 */
 645int dm_split_args(int *argc, char ***argvp, char *input)
 646{
 647        char *start, *end = input, *out, **argv = NULL;
 648        unsigned array_size = 0;
 649
 650        *argc = 0;
 651
 652        if (!input) {
 653                *argvp = NULL;
 654                return 0;
 655        }
 656
 657        argv = realloc_argv(&array_size, argv);
 658        if (!argv)
 659                return -ENOMEM;
 660
 661        while (1) {
 662                start = end;
 663
 664                /* Skip whitespace */
 665                while (*start && isspace(*start))
 666                        start++;
 667
 668                if (!*start)
 669                        break;  /* success, we hit the end */
 670
 671                /* 'out' is used to remove any back-quotes */
 672                end = out = start;
 673                while (*end) {
 674                        /* Everything apart from '\0' can be quoted */
 675                        if (*end == '\\' && *(end + 1)) {
 676                                *out++ = *(end + 1);
 677                                end += 2;
 678                                continue;
 679                        }
 680
 681                        if (isspace(*end))
 682                                break;  /* end of token */
 683
 684                        *out++ = *end++;
 685                }
 686
 687                /* have we already filled the array ? */
 688                if ((*argc + 1) > array_size) {
 689                        argv = realloc_argv(&array_size, argv);
 690                        if (!argv)
 691                                return -ENOMEM;
 692                }
 693
 694                /* we know this is whitespace */
 695                if (*end)
 696                        end++;
 697
 698                /* terminate the string and put it in the array */
 699                *out = '\0';
 700                argv[*argc] = start;
 701                (*argc)++;
 702        }
 703
 704        *argvp = argv;
 705        return 0;
 706}
 707
 708static void check_for_valid_limits(struct io_restrictions *rs)
 709{
 710        if (!rs->max_sectors)
 711                rs->max_sectors = SAFE_MAX_SECTORS;
 712        if (!rs->max_hw_sectors)
 713                rs->max_hw_sectors = SAFE_MAX_SECTORS;
 714        if (!rs->max_phys_segments)
 715                rs->max_phys_segments = MAX_PHYS_SEGMENTS;
 716        if (!rs->max_hw_segments)
 717                rs->max_hw_segments = MAX_HW_SEGMENTS;
 718        if (!rs->hardsect_size)
 719                rs->hardsect_size = 1 << SECTOR_SHIFT;
 720        if (!rs->max_segment_size)
 721                rs->max_segment_size = MAX_SEGMENT_SIZE;
 722        if (!rs->seg_boundary_mask)
 723                rs->seg_boundary_mask = -1;
 724        if (!rs->bounce_pfn)
 725                rs->bounce_pfn = -1;
 726}
 727
 728int dm_table_add_target(struct dm_table *t, const char *type,
 729                        sector_t start, sector_t len, char *params)
 730{
 731        int r = -EINVAL, argc;
 732        char **argv;
 733        struct dm_target *tgt;
 734
 735        if ((r = check_space(t)))
 736                return r;
 737
 738        tgt = t->targets + t->num_targets;
 739        memset(tgt, 0, sizeof(*tgt));
 740
 741        if (!len) {
 742                DMERR("%s: zero-length target", dm_device_name(t->md));
 743                return -EINVAL;
 744        }
 745
 746        tgt->type = dm_get_target_type(type);
 747        if (!tgt->type) {
 748                DMERR("%s: %s: unknown target type", dm_device_name(t->md),
 749                      type);
 750                return -EINVAL;
 751        }
 752
 753        tgt->table = t;
 754        tgt->begin = start;
 755        tgt->len = len;
 756        tgt->error = "Unknown error";
 757
 758        /*
 759         * Does this target adjoin the previous one ?
 760         */
 761        if (!adjoin(t, tgt)) {
 762                tgt->error = "Gap in table";
 763                r = -EINVAL;
 764                goto bad;
 765        }
 766
 767        r = dm_split_args(&argc, &argv, params);
 768        if (r) {
 769                tgt->error = "couldn't split parameters (insufficient memory)";
 770                goto bad;
 771        }
 772
 773        r = tgt->type->ctr(tgt, argc, argv);
 774        kfree(argv);
 775        if (r)
 776                goto bad;
 777
 778        t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
 779
 780        /* FIXME: the plan is to combine high here and then have
 781         * the merge fn apply the target level restrictions. */
 782        combine_restrictions_low(&t->limits, &tgt->limits);
 783        return 0;
 784
 785 bad:
 786        DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
 787        dm_put_target_type(tgt->type);
 788        return r;
 789}
 790
 791static int setup_indexes(struct dm_table *t)
 792{
 793        int i;
 794        unsigned int total = 0;
 795        sector_t *indexes;
 796
 797        /* allocate the space for *all* the indexes */
 798        for (i = t->depth - 2; i >= 0; i--) {
 799                t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
 800                total += t->counts[i];
 801        }
 802
 803        indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
 804        if (!indexes)
 805                return -ENOMEM;
 806
 807        /* set up internal nodes, bottom-up */
 808        for (i = t->depth - 2, total = 0; i >= 0; i--) {
 809                t->index[i] = indexes;
 810                indexes += (KEYS_PER_NODE * t->counts[i]);
 811                setup_btree_index(i, t);
 812        }
 813
 814        return 0;
 815}
 816
 817/*
 818 * Builds the btree to index the map.
 819 */
 820int dm_table_complete(struct dm_table *t)
 821{
 822        int r = 0;
 823        unsigned int leaf_nodes;
 824
 825        check_for_valid_limits(&t->limits);
 826
 827        /* how many indexes will the btree have ? */
 828        leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
 829        t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
 830
 831        /* leaf layer has already been set up */
 832        t->counts[t->depth - 1] = leaf_nodes;
 833        t->index[t->depth - 1] = t->highs;
 834
 835        if (t->depth >= 2)
 836                r = setup_indexes(t);
 837
 838        return r;
 839}
 840
 841static DEFINE_MUTEX(_event_lock);
 842void dm_table_event_callback(struct dm_table *t,
 843                             void (*fn)(void *), void *context)
 844{
 845        mutex_lock(&_event_lock);
 846        t->event_fn = fn;
 847        t->event_context = context;
 848        mutex_unlock(&_event_lock);
 849}
 850
 851void dm_table_event(struct dm_table *t)
 852{
 853        /*
 854         * You can no longer call dm_table_event() from interrupt
 855         * context, use a bottom half instead.
 856         */
 857        BUG_ON(in_interrupt());
 858
 859        mutex_lock(&_event_lock);
 860        if (t->event_fn)
 861                t->event_fn(t->event_context);
 862        mutex_unlock(&_event_lock);
 863}
 864
 865sector_t dm_table_get_size(struct dm_table *t)
 866{
 867        return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
 868}
 869
 870struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
 871{
 872        if (index >= t->num_targets)
 873                return NULL;
 874
 875        return t->targets + index;
 876}
 877
 878/*
 879 * Search the btree for the correct target.
 880 *
 881 * Caller should check returned pointer with dm_target_is_valid()
 882 * to trap I/O beyond end of device.
 883 */
 884struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
 885{
 886        unsigned int l, n = 0, k = 0;
 887        sector_t *node;
 888
 889        for (l = 0; l < t->depth; l++) {
 890                n = get_child(n, k);
 891                node = get_node(t, l, n);
 892
 893                for (k = 0; k < KEYS_PER_NODE; k++)
 894                        if (node[k] >= sector)
 895                                break;
 896        }
 897
 898        return &t->targets[(KEYS_PER_NODE * n) + k];
 899}
 900
 901void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q)
 902{
 903        /*
 904         * Make sure we obey the optimistic sub devices
 905         * restrictions.
 906         */
 907        blk_queue_max_sectors(q, t->limits.max_sectors);
 908        q->max_phys_segments = t->limits.max_phys_segments;
 909        q->max_hw_segments = t->limits.max_hw_segments;
 910        q->hardsect_size = t->limits.hardsect_size;
 911        q->max_segment_size = t->limits.max_segment_size;
 912        q->max_hw_sectors = t->limits.max_hw_sectors;
 913        q->seg_boundary_mask = t->limits.seg_boundary_mask;
 914        q->bounce_pfn = t->limits.bounce_pfn;
 915        if (t->limits.no_cluster)
 916                q->queue_flags &= ~(1 << QUEUE_FLAG_CLUSTER);
 917        else
 918                q->queue_flags |= (1 << QUEUE_FLAG_CLUSTER);
 919
 920}
 921
 922unsigned int dm_table_get_num_targets(struct dm_table *t)
 923{
 924        return t->num_targets;
 925}
 926
 927struct list_head *dm_table_get_devices(struct dm_table *t)
 928{
 929        return &t->devices;
 930}
 931
 932int dm_table_get_mode(struct dm_table *t)
 933{
 934        return t->mode;
 935}
 936
 937static void suspend_targets(struct dm_table *t, unsigned postsuspend)
 938{
 939        int i = t->num_targets;
 940        struct dm_target *ti = t->targets;
 941
 942        while (i--) {
 943                if (postsuspend) {
 944                        if (ti->type->postsuspend)
 945                                ti->type->postsuspend(ti);
 946                } else if (ti->type->presuspend)
 947                        ti->type->presuspend(ti);
 948
 949                ti++;
 950        }
 951}
 952
 953void dm_table_presuspend_targets(struct dm_table *t)
 954{
 955        if (!t)
 956                return;
 957
 958        return suspend_targets(t, 0);
 959}
 960
 961void dm_table_postsuspend_targets(struct dm_table *t)
 962{
 963        if (!t)
 964                return;
 965
 966        return suspend_targets(t, 1);
 967}
 968
 969int dm_table_resume_targets(struct dm_table *t)
 970{
 971        int i, r = 0;
 972
 973        for (i = 0; i < t->num_targets; i++) {
 974                struct dm_target *ti = t->targets + i;
 975
 976                if (!ti->type->preresume)
 977                        continue;
 978
 979                r = ti->type->preresume(ti);
 980                if (r)
 981                        return r;
 982        }
 983
 984        for (i = 0; i < t->num_targets; i++) {
 985                struct dm_target *ti = t->targets + i;
 986
 987                if (ti->type->resume)
 988                        ti->type->resume(ti);
 989        }
 990
 991        return 0;
 992}
 993
 994int dm_table_any_congested(struct dm_table *t, int bdi_bits)
 995{
 996        struct list_head *d, *devices;
 997        int r = 0;
 998
 999        devices = dm_table_get_devices(t);
1000        for (d = devices->next; d != devices; d = d->next) {
1001                struct dm_dev *dd = list_entry(d, struct dm_dev, list);
1002                struct request_queue *q = bdev_get_queue(dd->bdev);
1003                r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1004        }
1005
1006        return r;
1007}
1008
1009void dm_table_unplug_all(struct dm_table *t)
1010{
1011        struct list_head *d, *devices = dm_table_get_devices(t);
1012
1013        for (d = devices->next; d != devices; d = d->next) {
1014                struct dm_dev *dd = list_entry(d, struct dm_dev, list);
1015                struct request_queue *q = bdev_get_queue(dd->bdev);
1016
1017                blk_unplug(q);
1018        }
1019}
1020
1021struct mapped_device *dm_table_get_md(struct dm_table *t)
1022{
1023        dm_get(t->md);
1024
1025        return t->md;
1026}
1027
1028EXPORT_SYMBOL(dm_vcalloc);
1029EXPORT_SYMBOL(dm_get_device);
1030EXPORT_SYMBOL(dm_put_device);
1031EXPORT_SYMBOL(dm_table_event);
1032EXPORT_SYMBOL(dm_table_get_size);
1033EXPORT_SYMBOL(dm_table_get_mode);
1034EXPORT_SYMBOL(dm_table_get_md);
1035EXPORT_SYMBOL(dm_table_put);
1036EXPORT_SYMBOL(dm_table_get);
1037EXPORT_SYMBOL(dm_table_unplug_all);
1038