drivers/md/dm-ioctl.c

   1 /*
   2  * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
   3  * Copyright (C) 2004 - 2005 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/miscdevice.h>
  13 #include <linux/init.h>
  14 #include <linux/wait.h>
  15 #include <linux/slab.h>
  16 #include <linux/devfs_fs_kernel.h>
  17 #include <linux/dm-ioctl.h>
  18 #include <linux/hdreg.h>
  19
  20 #include <asm/uaccess.h>
  21
  22 #define DM_DRIVER_EMAIL "dm-devel@redhat.com"
  23
  24 /*-----------------------------------------------------------------
  25  * The ioctl interface needs to be able to look up devices by
  26  * name or uuid.
  27  *---------------------------------------------------------------*/
  28 struct hash_cell {
  29         struct list_head name_list;
  30         struct list_head uuid_list;
  31
  32         char *name;
  33         char *uuid;
  34         struct mapped_device *md;
  35         struct dm_table *new_map;
  36 };
  37
  38 struct vers_iter {
  39     size_t param_size;
  40     struct dm_target_versions *vers, *old_vers;
  41     char *end;
  42     uint32_t flags;
  43 };
  44
  45
  46 #define NUM_BUCKETS 64
  47 #define MASK_BUCKETS (NUM_BUCKETS - 1)
  48 static struct list_head _name_buckets[NUM_BUCKETS];
  49 static struct list_head _uuid_buckets[NUM_BUCKETS];
  50
  51 static void dm_hash_remove_all(void);
  52
  53 /*
  54  * Guards access to both hash tables.
  55  */
  56 static DECLARE_RWSEM(_hash_lock);
  57
  58 static void init_buckets(struct list_head *buckets)
  59 {
  60         unsigned int i;
  61
  62         for (i = 0; i < NUM_BUCKETS; i++)
  63                 INIT_LIST_HEAD(buckets + i);
  64 }
  65
  66 static int dm_hash_init(void)
  67 {
  68         init_buckets(_name_buckets);
  69         init_buckets(_uuid_buckets);
  70         devfs_mk_dir(DM_DIR);
  71         return 0;
  72 }
  73
  74 static void dm_hash_exit(void)
  75 {
  76         dm_hash_remove_all();
  77         devfs_remove(DM_DIR);
  78 }
  79
  80 /*-----------------------------------------------------------------
  81  * Hash function:
  82  * We're not really concerned with the str hash function being
  83  * fast since it's only used by the ioctl interface.
  84  *---------------------------------------------------------------*/
  85 static unsigned int hash_str(const char *str)
  86 {
  87         const unsigned int hash_mult = 2654435387U;
  88         unsigned int h = 0;
  89
  90         while (*str)
  91                 h = (h + (unsigned int) *str++) * hash_mult;
  92
  93         return h & MASK_BUCKETS;
  94 }
  95
  96 /*-----------------------------------------------------------------
  97  * Code for looking up a device by name
  98  *---------------------------------------------------------------*/
  99 static struct hash_cell *__get_name_cell(const char *str)
 100 {
 101         struct hash_cell *hc;
 102         unsigned int h = hash_str(str);
 103
 104         list_for_each_entry (hc, _name_buckets + h, name_list)
 105                 if (!strcmp(hc->name, str)) {
 106                         dm_get(hc->md);
 107                         return hc;
 108                 }
 109
 110         return NULL;
 111 }
 112
 113 static struct hash_cell *__get_uuid_cell(const char *str)
 114 {
 115         struct hash_cell *hc;
 116         unsigned int h = hash_str(str);
 117
 118         list_for_each_entry (hc, _uuid_buckets + h, uuid_list)
 119                 if (!strcmp(hc->uuid, str)) {
 120                         dm_get(hc->md);
 121                         return hc;
 122                 }
 123
 124         return NULL;
 125 }
 126
 127 /*-----------------------------------------------------------------
 128  * Inserting, removing and renaming a device.
 129  *---------------------------------------------------------------*/
 130 static struct hash_cell *alloc_cell(const char *name, const char *uuid,
 131                                     struct mapped_device *md)
 132 {
 133         struct hash_cell *hc;
 134
 135         hc = kmalloc(sizeof(*hc), GFP_KERNEL);
 136         if (!hc)
 137                 return NULL;
 138
 139         hc->name = kstrdup(name, GFP_KERNEL);
 140         if (!hc->name) {
 141                 kfree(hc);
 142                 return NULL;
 143         }
 144
 145         if (!uuid)
 146                 hc->uuid = NULL;
 147
 148         else {
 149                 hc->uuid = kstrdup(uuid, GFP_KERNEL);
 150                 if (!hc->uuid) {
 151                         kfree(hc->name);
 152                         kfree(hc);
 153                         return NULL;
 154                 }
 155         }
 156
 157         INIT_LIST_HEAD(&hc->name_list);
 158         INIT_LIST_HEAD(&hc->uuid_list);
 159         hc->md = md;
 160         hc->new_map = NULL;
 161         return hc;
 162 }
 163
 164 static void free_cell(struct hash_cell *hc)
 165 {
 166         if (hc) {
 167                 kfree(hc->name);
 168                 kfree(hc->uuid);
 169                 kfree(hc);
 170         }
 171 }
 172
 173 /*
 174  * devfs stuff.
 175  */
 176 static int register_with_devfs(struct hash_cell *hc)
 177 {
 178         struct gendisk *disk = dm_disk(hc->md);
 179
 180         devfs_mk_bdev(MKDEV(disk->major, disk->first_minor),
 181                       S_IFBLK | S_IRUSR | S_IWUSR | S_IRGRP,
 182                       DM_DIR "/%s", hc->name);
 183         return 0;
 184 }
 185
 186 static int unregister_with_devfs(struct hash_cell *hc)
 187 {
 188         devfs_remove(DM_DIR"/%s", hc->name);
 189         return 0;
 190 }
 191
 192 /*
 193  * The kdev_t and uuid of a device can never change once it is
 194  * initially inserted.
 195  */
 196 static int dm_hash_insert(const char *name, const char *uuid, struct mapped_device *md)
 197 {
 198         struct hash_cell *cell, *hc;
 199
 200         /*
 201          * Allocate the new cells.
 202          */
 203         cell = alloc_cell(name, uuid, md);
 204         if (!cell)
 205                 return -ENOMEM;
 206
 207         /*
 208          * Insert the cell into both hash tables.
 209          */
 210         down_write(&_hash_lock);
 211         hc = __get_name_cell(name);
 212         if (hc) {
 213                 dm_put(hc->md);
 214                 goto bad;
 215         }
 216
 217         list_add(&cell->name_list, _name_buckets + hash_str(name));
 218
 219         if (uuid) {
 220                 hc = __get_uuid_cell(uuid);
 221                 if (hc) {
 222                         list_del(&cell->name_list);
 223                         dm_put(hc->md);
 224                         goto bad;
 225                 }
 226                 list_add(&cell->uuid_list, _uuid_buckets + hash_str(uuid));
 227         }
 228         register_with_devfs(cell);
 229         dm_get(md);
 230         dm_set_mdptr(md, cell);
 231         up_write(&_hash_lock);
 232
 233         return 0;
 234
 235  bad:
 236         up_write(&_hash_lock);
 237         free_cell(cell);
 238         return -EBUSY;
 239 }
 240
 241 static void __hash_remove(struct hash_cell *hc)
 242 {
 243         struct dm_table *table;
 244
 245         /* remove from the dev hash */
 246         list_del(&hc->uuid_list);
 247         list_del(&hc->name_list);
 248         unregister_with_devfs(hc);
 249         dm_set_mdptr(hc->md, NULL);
 250
 251         table = dm_get_table(hc->md);
 252         if (table) {
 253                 dm_table_event(table);
 254                 dm_table_put(table);
 255         }
 256
 257         if (hc->new_map)
 258                 dm_table_put(hc->new_map);
 259         dm_put(hc->md);
 260         free_cell(hc);
 261 }
 262
 263 static void dm_hash_remove_all(void)
 264 {
 265         int i;
 266         struct hash_cell *hc;
 267         struct list_head *tmp, *n;
 268
 269         down_write(&_hash_lock);
 270         for (i = 0; i < NUM_BUCKETS; i++) {
 271                 list_for_each_safe (tmp, n, _name_buckets + i) {
 272                         hc = list_entry(tmp, struct hash_cell, name_list);
 273                         __hash_remove(hc);
 274                 }
 275         }
 276         up_write(&_hash_lock);
 277 }
 278
 279 static int dm_hash_rename(const char *old, const char *new)
 280 {
 281         char *new_name, *old_name;
 282         struct hash_cell *hc;
 283         struct dm_table *table;
 284
 285         /*
 286          * duplicate new.
 287          */
 288         new_name = kstrdup(new, GFP_KERNEL);
 289         if (!new_name)
 290                 return -ENOMEM;
 291
 292         down_write(&_hash_lock);
 293
 294         /*
 295          * Is new free ?
 296          */
 297         hc = __get_name_cell(new);
 298         if (hc) {
 299                 DMWARN("asked to rename to an already existing name %s -> %s",
 300                        old, new);
 301                 dm_put(hc->md);
 302                 up_write(&_hash_lock);
 303                 kfree(new_name);
 304                 return -EBUSY;
 305         }
 306
 307         /*
 308          * Is there such a device as 'old' ?
 309          */
 310         hc = __get_name_cell(old);
 311         if (!hc) {
 312                 DMWARN("asked to rename a non existent device %s -> %s",
 313                        old, new);
 314                 up_write(&_hash_lock);
 315                 kfree(new_name);
 316                 return -ENXIO;
 317         }
 318
 319         /*
 320          * rename and move the name cell.
 321          */
 322         unregister_with_devfs(hc);
 323
 324         list_del(&hc->name_list);
 325         old_name = hc->name;
 326         hc->name = new_name;
 327         list_add(&hc->name_list, _name_buckets + hash_str(new_name));
 328
 329         /* rename the device node in devfs */
 330         register_with_devfs(hc);
 331
 332         /*
 333          * Wake up any dm event waiters.
 334          */
 335         table = dm_get_table(hc->md);
 336         if (table) {
 337                 dm_table_event(table);
 338                 dm_table_put(table);
 339         }
 340
 341         dm_put(hc->md);
 342         up_write(&_hash_lock);
 343         kfree(old_name);
 344         return 0;
 345 }
 346
 347 /*-----------------------------------------------------------------
 348  * Implementation of the ioctl commands
 349  *---------------------------------------------------------------*/
 350 /*
 351  * All the ioctl commands get dispatched to functions with this
 352  * prototype.
 353  */
 354 typedef int (*ioctl_fn)(struct dm_ioctl *param, size_t param_size);
 355
 356 static int remove_all(struct dm_ioctl *param, size_t param_size)
 357 {
 358         dm_hash_remove_all();
 359         param->data_size = 0;
 360         return 0;
 361 }
 362
 363 /*
 364  * Round up the ptr to an 8-byte boundary.
 365  */
 366 #define ALIGN_MASK 7
 367 static inline void *align_ptr(void *ptr)
 368 {
 369         return (void *) (((size_t) (ptr + ALIGN_MASK)) & ~ALIGN_MASK);
 370 }
 371
 372 /*
 373  * Retrieves the data payload buffer from an already allocated
 374  * struct dm_ioctl.
 375  */
 376 static void *get_result_buffer(struct dm_ioctl *param, size_t param_size,
 377                                size_t *len)
 378 {
 379         param->data_start = align_ptr(param + 1) - (void *) param;
 380
 381         if (param->data_start < param_size)
 382                 *len = param_size - param->data_start;
 383         else
 384                 *len = 0;
 385
 386         return ((void *) param) + param->data_start;
 387 }
 388
 389 static int list_devices(struct dm_ioctl *param, size_t param_size)
 390 {
 391         unsigned int i;
 392         struct hash_cell *hc;
 393         size_t len, needed = 0;
 394         struct gendisk *disk;
 395         struct dm_name_list *nl, *old_nl = NULL;
 396
 397         down_write(&_hash_lock);
 398
 399         /*
 400          * Loop through all the devices working out how much
 401          * space we need.
 402          */
 403         for (i = 0; i < NUM_BUCKETS; i++) {
 404                 list_for_each_entry (hc, _name_buckets + i, name_list) {
 405                         needed += sizeof(struct dm_name_list);
 406                         needed += strlen(hc->name) + 1;
 407                         needed += ALIGN_MASK;
 408                 }
 409         }
 410
 411         /*
 412          * Grab our output buffer.
 413          */
 414         nl = get_result_buffer(param, param_size, &len);
 415         if (len < needed) {
 416                 param->flags |= DM_BUFFER_FULL_FLAG;
 417                 goto out;
 418         }
 419         param->data_size = param->data_start + needed;
 420
 421         nl->dev = 0;    /* Flags no data */
 422
 423         /*
 424          * Now loop through filling out the names.
 425          */
 426         for (i = 0; i < NUM_BUCKETS; i++) {
 427                 list_for_each_entry (hc, _name_buckets + i, name_list) {
 428                         if (old_nl)
 429                                 old_nl->next = (uint32_t) ((void *) nl -
 430                                                            (void *) old_nl);
 431                         disk = dm_disk(hc->md);
 432                         nl->dev = huge_encode_dev(MKDEV(disk->major, disk->first_minor));
 433                         nl->next = 0;
 434                         strcpy(nl->name, hc->name);
 435
 436                         old_nl = nl;
 437                         nl = align_ptr(((void *) ++nl) + strlen(hc->name) + 1);
 438                 }
 439         }
 440
 441  out:
 442         up_write(&_hash_lock);
 443         return 0;
 444 }
 445
 446 static void list_version_get_needed(struct target_type *tt, void *needed_param)
 447 {
 448     size_t *needed = needed_param;
 449
 450     *needed += sizeof(struct dm_target_versions);
 451     *needed += strlen(tt->name);
 452     *needed += ALIGN_MASK;
 453 }
 454
 455 static void list_version_get_info(struct target_type *tt, void *param)
 456 {
 457     struct vers_iter *info = param;
 458
 459     /* Check space - it might have changed since the first iteration */
 460     if ((char *)info->vers + sizeof(tt->version) + strlen(tt->name) + 1 >
 461         info->end) {
 462
 463         info->flags = DM_BUFFER_FULL_FLAG;
 464         return;
 465     }
 466
 467     if (info->old_vers)
 468         info->old_vers->next = (uint32_t) ((void *)info->vers -
 469                                            (void *)info->old_vers);
 470     info->vers->version[0] = tt->version[0];
 471     info->vers->version[1] = tt->version[1];
 472     info->vers->version[2] = tt->version[2];
 473     info->vers->next = 0;
 474     strcpy(info->vers->name, tt->name);
 475
 476     info->old_vers = info->vers;
 477     info->vers = align_ptr(((void *) ++info->vers) + strlen(tt->name) + 1);
 478 }
 479
 480 static int list_versions(struct dm_ioctl *param, size_t param_size)
 481 {
 482         size_t len, needed = 0;
 483         struct dm_target_versions *vers;
 484         struct vers_iter iter_info;
 485
 486         /*
 487          * Loop through all the devices working out how much
 488          * space we need.
 489          */
 490         dm_target_iterate(list_version_get_needed, &needed);
 491
 492         /*
 493          * Grab our output buffer.
 494          */
 495         vers = get_result_buffer(param, param_size, &len);
 496         if (len < needed) {
 497                 param->flags |= DM_BUFFER_FULL_FLAG;
 498                 goto out;
 499         }
 500         param->data_size = param->data_start + needed;
 501
 502         iter_info.param_size = param_size;
 503         iter_info.old_vers = NULL;
 504         iter_info.vers = vers;
 505         iter_info.flags = 0;
 506         iter_info.end = (char *)vers+len;
 507
 508         /*
 509          * Now loop through filling out the names & versions.
 510          */
 511         dm_target_iterate(list_version_get_info, &iter_info);
 512         param->flags |= iter_info.flags;
 513
 514  out:
 515         return 0;
 516 }
 517
 518
 519
 520 static int check_name(const char *name)
 521 {
 522         if (strchr(name, '/')) {
 523                 DMWARN("invalid device name");
 524                 return -EINVAL;
 525         }
 526
 527         return 0;
 528 }
 529
 530 /*
 531  * Fills in a dm_ioctl structure, ready for sending back to
 532  * userland.
 533  */
 534 static int __dev_status(struct mapped_device *md, struct dm_ioctl *param)
 535 {
 536         struct gendisk *disk = dm_disk(md);
 537         struct dm_table *table;
 538         struct block_device *bdev;
 539
 540         param->flags &= ~(DM_SUSPEND_FLAG | DM_READONLY_FLAG |
 541                           DM_ACTIVE_PRESENT_FLAG);
 542
 543         if (dm_suspended(md))
 544                 param->flags |= DM_SUSPEND_FLAG;
 545
 546         param->dev = huge_encode_dev(MKDEV(disk->major, disk->first_minor));
 547
 548         if (!(param->flags & DM_SKIP_BDGET_FLAG)) {
 549                 bdev = bdget_disk(disk, 0);
 550                 if (!bdev)
 551                         return -ENXIO;
 552
 553                 /*
 554                  * Yes, this will be out of date by the time it gets back
 555                  * to userland, but it is still very useful for
 556                  * debugging.
 557                  */
 558                 param->open_count = bdev->bd_openers;
 559                 bdput(bdev);
 560         } else
 561                 param->open_count = -1;
 562
 563         if (disk->policy)
 564                 param->flags |= DM_READONLY_FLAG;
 565
 566         param->event_nr = dm_get_event_nr(md);
 567
 568         table = dm_get_table(md);
 569         if (table) {
 570                 param->flags |= DM_ACTIVE_PRESENT_FLAG;
 571                 param->target_count = dm_table_get_num_targets(table);
 572                 dm_table_put(table);
 573         } else
 574                 param->target_count = 0;
 575
 576         return 0;
 577 }
 578
 579 static int dev_create(struct dm_ioctl *param, size_t param_size)
 580 {
 581         int r;
 582         struct mapped_device *md;
 583
 584         r = check_name(param->name);
 585         if (r)
 586                 return r;
 587
 588         if (param->flags & DM_PERSISTENT_DEV_FLAG)
 589                 r = dm_create_with_minor(MINOR(huge_decode_dev(param->dev)), &md);
 590         else
 591                 r = dm_create(&md);
 592
 593         if (r)
 594                 return r;
 595
 596         r = dm_hash_insert(param->name, *param->uuid ? param->uuid : NULL, md);
 597         if (r) {
 598                 dm_put(md);
 599                 return r;
 600         }
 601
 602         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
 603
 604         r = __dev_status(md, param);
 605         dm_put(md);
 606
 607         return r;
 608 }
 609
 610 /*
 611  * Always use UUID for lookups if it's present, otherwise use name or dev.
 612  */
 613 static struct hash_cell *__find_device_hash_cell(struct dm_ioctl *param)
 614 {
 615         struct mapped_device *md;
 616         void *mdptr = NULL;
 617
 618         if (*param->uuid)
 619                 return __get_uuid_cell(param->uuid);
 620
 621         if (*param->name)
 622                 return __get_name_cell(param->name);
 623
 624         md = dm_get_md(huge_decode_dev(param->dev));
 625         if (md)
 626                 mdptr = dm_get_mdptr(md);
 627
 628         return mdptr;
 629 }
 630
 631 static struct mapped_device *find_device(struct dm_ioctl *param)
 632 {
 633         struct hash_cell *hc;
 634         struct mapped_device *md = NULL;
 635
 636         down_read(&_hash_lock);
 637         hc = __find_device_hash_cell(param);
 638         if (hc) {
 639                 md = hc->md;
 640
 641                 /*
 642                  * Sneakily write in both the name and the uuid
 643                  * while we have the cell.
 644                  */
 645                 strncpy(param->name, hc->name, sizeof(param->name));
 646                 if (hc->uuid)
 647                         strncpy(param->uuid, hc->uuid, sizeof(param->uuid)-1);
 648                 else
 649                         param->uuid[0] = '\0';
 650
 651                 if (hc->new_map)
 652                         param->flags |= DM_INACTIVE_PRESENT_FLAG;
 653                 else
 654                         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
 655         }
 656         up_read(&_hash_lock);
 657
 658         return md;
 659 }
 660
 661 static int dev_remove(struct dm_ioctl *param, size_t param_size)
 662 {
 663         struct hash_cell *hc;
 664         struct mapped_device *md;
 665
 666         down_write(&_hash_lock);
 667         hc = __find_device_hash_cell(param);
 668
 669         if (!hc) {
 670                 DMWARN("device doesn't appear to be in the dev hash table.");
 671                 up_write(&_hash_lock);
 672                 return -ENXIO;
 673         }
 674
 675         md = hc->md;
 676
 677         __hash_remove(hc);
 678         up_write(&_hash_lock);
 679         dm_put(md);
 680         param->data_size = 0;
 681         return 0;
 682 }
 683
 684 /*
 685  * Check a string doesn't overrun the chunk of
 686  * memory we copied from userland.
 687  */
 688 static int invalid_str(char *str, void *end)
 689 {
 690         while ((void *) str < end)
 691                 if (!*str++)
 692                         return 0;
 693
 694         return -EINVAL;
 695 }
 696
 697 static int dev_rename(struct dm_ioctl *param, size_t param_size)
 698 {
 699         int r;
 700         char *new_name = (char *) param + param->data_start;
 701
 702         if (new_name < (char *) (param + 1) ||
 703             invalid_str(new_name, (void *) param + param_size)) {
 704                 DMWARN("Invalid new logical volume name supplied.");
 705                 return -EINVAL;
 706         }
 707
 708         r = check_name(new_name);
 709         if (r)
 710                 return r;
 711
 712         param->data_size = 0;
 713         return dm_hash_rename(param->name, new_name);
 714 }
 715
 716 static int dev_set_geometry(struct dm_ioctl *param, size_t param_size)
 717 {
 718         int r = -EINVAL, x;
 719         struct mapped_device *md;
 720         struct hd_geometry geometry;
 721         unsigned long indata[4];
 722         char *geostr = (char *) param + param->data_start;
 723
 724         md = find_device(param);
 725         if (!md)
 726                 return -ENXIO;
 727
 728         if (geostr < (char *) (param + 1) ||
 729             invalid_str(geostr, (void *) param + param_size)) {
 730                 DMWARN("Invalid geometry supplied.");
 731                 goto out;
 732         }
 733
 734         x = sscanf(geostr, "%lu %lu %lu %lu", indata,
 735                    indata + 1, indata + 2, indata + 3);
 736
 737         if (x != 4) {
 738                 DMWARN("Unable to interpret geometry settings.");
 739                 goto out;
 740         }
 741
 742         if (indata[0] > 65535 || indata[1] > 255 ||
 743             indata[2] > 255 || indata[3] > ULONG_MAX) {
 744                 DMWARN("Geometry exceeds range limits.");
 745                 goto out;
 746         }
 747
 748         geometry.cylinders = indata[0];
 749         geometry.heads = indata[1];
 750         geometry.sectors = indata[2];
 751         geometry.start = indata[3];
 752
 753         r = dm_set_geometry(md, &geometry);
 754         if (!r)
 755                 r = __dev_status(md, param);
 756
 757         param->data_size = 0;
 758
 759 out:
 760         dm_put(md);
 761         return r;
 762 }
 763
 764 static int do_suspend(struct dm_ioctl *param)
 765 {
 766         int r = 0;
 767         int do_lockfs = 1;
 768         struct mapped_device *md;
 769
 770         md = find_device(param);
 771         if (!md)
 772                 return -ENXIO;
 773
 774         if (param->flags & DM_SKIP_LOCKFS_FLAG)
 775                 do_lockfs = 0;
 776
 777         if (!dm_suspended(md))
 778                 r = dm_suspend(md, do_lockfs);
 779
 780         if (!r)
 781                 r = __dev_status(md, param);
 782
 783         dm_put(md);
 784         return r;
 785 }
 786
 787 static int do_resume(struct dm_ioctl *param)
 788 {
 789         int r = 0;
 790         int do_lockfs = 1;
 791         struct hash_cell *hc;
 792         struct mapped_device *md;
 793         struct dm_table *new_map;
 794
 795         down_write(&_hash_lock);
 796
 797         hc = __find_device_hash_cell(param);
 798         if (!hc) {
 799                 DMWARN("device doesn't appear to be in the dev hash table.");
 800                 up_write(&_hash_lock);
 801                 return -ENXIO;
 802         }
 803
 804         md = hc->md;
 805
 806         new_map = hc->new_map;
 807         hc->new_map = NULL;
 808         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
 809
 810         up_write(&_hash_lock);
 811
 812         /* Do we need to load a new map ? */
 813         if (new_map) {
 814                 /* Suspend if it isn't already suspended */
 815                 if (param->flags & DM_SKIP_LOCKFS_FLAG)
 816                         do_lockfs = 0;
 817                 if (!dm_suspended(md))
 818                         dm_suspend(md, do_lockfs);
 819
 820                 r = dm_swap_table(md, new_map);
 821                 if (r) {
 822                         dm_put(md);
 823                         dm_table_put(new_map);
 824                         return r;
 825                 }
 826
 827                 if (dm_table_get_mode(new_map) & FMODE_WRITE)
 828                         set_disk_ro(dm_disk(md), 0);
 829                 else
 830                         set_disk_ro(dm_disk(md), 1);
 831
 832                 dm_table_put(new_map);
 833         }
 834
 835         if (dm_suspended(md))
 836                 r = dm_resume(md);
 837
 838         if (!r)
 839                 r = __dev_status(md, param);
 840
 841         dm_put(md);
 842         return r;
 843 }
 844
 845 /*
 846  * Set or unset the suspension state of a device.
 847  * If the device already is in the requested state we just return its status.
 848  */
 849 static int dev_suspend(struct dm_ioctl *param, size_t param_size)
 850 {
 851         if (param->flags & DM_SUSPEND_FLAG)
 852                 return do_suspend(param);
 853
 854         return do_resume(param);
 855 }
 856
 857 /*
 858  * Copies device info back to user space, used by
 859  * the create and info ioctls.
 860  */
 861 static int dev_status(struct dm_ioctl *param, size_t param_size)
 862 {
 863         int r;
 864         struct mapped_device *md;
 865
 866         md = find_device(param);
 867         if (!md)
 868                 return -ENXIO;
 869
 870         r = __dev_status(md, param);
 871         dm_put(md);
 872         return r;
 873 }
 874
 875 /*
 876  * Build up the status struct for each target
 877  */
 878 static void retrieve_status(struct dm_table *table,
 879                             struct dm_ioctl *param, size_t param_size)
 880 {
 881         unsigned int i, num_targets;
 882         struct dm_target_spec *spec;
 883         char *outbuf, *outptr;
 884         status_type_t type;
 885         size_t remaining, len, used = 0;
 886
 887         outptr = outbuf = get_result_buffer(param, param_size, &len);
 888
 889         if (param->flags & DM_STATUS_TABLE_FLAG)
 890                 type = STATUSTYPE_TABLE;
 891         else
 892                 type = STATUSTYPE_INFO;
 893
 894         /* Get all the target info */
 895         num_targets = dm_table_get_num_targets(table);
 896         for (i = 0; i < num_targets; i++) {
 897                 struct dm_target *ti = dm_table_get_target(table, i);
 898
 899                 remaining = len - (outptr - outbuf);
 900                 if (remaining <= sizeof(struct dm_target_spec)) {
 901                         param->flags |= DM_BUFFER_FULL_FLAG;
 902                         break;
 903                 }
 904
 905                 spec = (struct dm_target_spec *) outptr;
 906
 907                 spec->status = 0;
 908                 spec->sector_start = ti->begin;
 909                 spec->length = ti->len;
 910                 strncpy(spec->target_type, ti->type->name,
 911                         sizeof(spec->target_type));
 912
 913                 outptr += sizeof(struct dm_target_spec);
 914                 remaining = len - (outptr - outbuf);
 915                 if (remaining <= 0) {
 916                         param->flags |= DM_BUFFER_FULL_FLAG;
 917                         break;
 918                 }
 919
 920                 /* Get the status/table string from the target driver */
 921                 if (ti->type->status) {
 922                         if (ti->type->status(ti, type, outptr, remaining)) {
 923                                 param->flags |= DM_BUFFER_FULL_FLAG;
 924                                 break;
 925                         }
 926                 } else
 927                         outptr[0] = '\0';
 928
 929                 outptr += strlen(outptr) + 1;
 930                 used = param->data_start + (outptr - outbuf);
 931
 932                 outptr = align_ptr(outptr);
 933                 spec->next = outptr - outbuf;
 934         }
 935
 936         if (used)
 937                 param->data_size = used;
 938
 939         param->target_count = num_targets;
 940 }
 941
 942 /*
 943  * Wait for a device to report an event
 944  */
 945 static int dev_wait(struct dm_ioctl *param, size_t param_size)
 946 {
 947         int r;
 948         struct mapped_device *md;
 949         struct dm_table *table;
 950
 951         md = find_device(param);
 952         if (!md)
 953                 return -ENXIO;
 954
 955         /*
 956          * Wait for a notification event
 957          */
 958         if (dm_wait_event(md, param->event_nr)) {
 959                 r = -ERESTARTSYS;
 960                 goto out;
 961         }
 962
 963         /*
 964          * The userland program is going to want to know what
 965          * changed to trigger the event, so we may as well tell
 966          * him and save an ioctl.
 967          */
 968         r = __dev_status(md, param);
 969         if (r)
 970                 goto out;
 971
 972         table = dm_get_table(md);
 973         if (table) {
 974                 retrieve_status(table, param, param_size);
 975                 dm_table_put(table);
 976         }
 977
 978  out:
 979         dm_put(md);
 980         return r;
 981 }
 982
 983 static inline int get_mode(struct dm_ioctl *param)
 984 {
 985         int mode = FMODE_READ | FMODE_WRITE;
 986
 987         if (param->flags & DM_READONLY_FLAG)
 988                 mode = FMODE_READ;
 989
 990         return mode;
 991 }
 992
 993 static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
 994                        struct dm_target_spec **spec, char **target_params)
 995 {
 996         *spec = (struct dm_target_spec *) ((unsigned char *) last + next);
 997         *target_params = (char *) (*spec + 1);
 998
 999         if (*spec < (last + 1))
1000                 return -EINVAL;
1001
1002         return invalid_str(*target_params, end);
1003 }
1004
1005 static int populate_table(struct dm_table *table,
1006                           struct dm_ioctl *param, size_t param_size)
1007 {
1008         int r;
1009         unsigned int i = 0;
1010         struct dm_target_spec *spec = (struct dm_target_spec *) param;
1011         uint32_t next = param->data_start;
1012         void *end = (void *) param + param_size;
1013         char *target_params;
1014
1015         if (!param->target_count) {
1016                 DMWARN("populate_table: no targets specified");
1017                 return -EINVAL;
1018         }
1019
1020         for (i = 0; i < param->target_count; i++) {
1021
1022                 r = next_target(spec, next, end, &spec, &target_params);
1023                 if (r) {
1024                         DMWARN("unable to find target");
1025                         return r;
1026                 }
1027
1028                 r = dm_table_add_target(table, spec->target_type,
1029                                         (sector_t) spec->sector_start,
1030                                         (sector_t) spec->length,
1031                                         target_params);
1032                 if (r) {
1033                         DMWARN("error adding target to table");
1034                         return r;
1035                 }
1036
1037                 next = spec->next;
1038         }
1039
1040         return dm_table_complete(table);
1041 }
1042
1043 static int table_load(struct dm_ioctl *param, size_t param_size)
1044 {
1045         int r;
1046         struct hash_cell *hc;
1047         struct dm_table *t;
1048         struct mapped_device *md;
1049
1050         md = find_device(param);
1051         if (!md)
1052                 return -ENXIO;
1053
1054         r = dm_table_create(&t, get_mode(param), param->target_count, md);
1055         if (r)
1056                 goto out;
1057
1058         r = populate_table(t, param, param_size);
1059         if (r) {
1060                 dm_table_put(t);
1061                 goto out;
1062         }
1063
1064         down_write(&_hash_lock);
1065         hc = dm_get_mdptr(md);
1066         if (!hc || hc->md != md) {
1067                 DMWARN("device has been removed from the dev hash table.");
1068                 dm_table_put(t);
1069                 up_write(&_hash_lock);
1070                 r = -ENXIO;
1071                 goto out;
1072         }
1073
1074         if (hc->new_map)
1075                 dm_table_put(hc->new_map);
1076         hc->new_map = t;
1077         up_write(&_hash_lock);
1078
1079         param->flags |= DM_INACTIVE_PRESENT_FLAG;
1080         r = __dev_status(md, param);
1081
1082 out:
1083         dm_put(md);
1084
1085         return r;
1086 }
1087
1088 static int table_clear(struct dm_ioctl *param, size_t param_size)
1089 {
1090         int r;
1091         struct hash_cell *hc;
1092         struct mapped_device *md;
1093
1094         down_write(&_hash_lock);
1095
1096         hc = __find_device_hash_cell(param);
1097         if (!hc) {
1098                 DMWARN("device doesn't appear to be in the dev hash table.");
1099                 up_write(&_hash_lock);
1100                 return -ENXIO;
1101         }
1102
1103         if (hc->new_map) {
1104                 dm_table_put(hc->new_map);
1105                 hc->new_map = NULL;
1106         }
1107
1108         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1109
1110         r = __dev_status(hc->md, param);
1111         md = hc->md;
1112         up_write(&_hash_lock);
1113         dm_put(md);
1114         return r;
1115 }
1116
1117 /*
1118  * Retrieves a list of devices used by a particular dm device.
1119  */
1120 static void retrieve_deps(struct dm_table *table,
1121                           struct dm_ioctl *param, size_t param_size)
1122 {
1123         unsigned int count = 0;
1124         struct list_head *tmp;
1125         size_t len, needed;
1126         struct dm_dev *dd;
1127         struct dm_target_deps *deps;
1128
1129         deps = get_result_buffer(param, param_size, &len);
1130
1131         /*
1132          * Count the devices.
1133          */
1134         list_for_each (tmp, dm_table_get_devices(table))
1135                 count++;
1136
1137         /*
1138          * Check we have enough space.
1139          */
1140         needed = sizeof(*deps) + (sizeof(*deps->dev) * count);
1141         if (len < needed) {
1142                 param->flags |= DM_BUFFER_FULL_FLAG;
1143                 return;
1144         }
1145
1146         /*
1147          * Fill in the devices.
1148          */
1149         deps->count = count;
1150         count = 0;
1151         list_for_each_entry (dd, dm_table_get_devices(table), list)
1152                 deps->dev[count++] = huge_encode_dev(dd->bdev->bd_dev);
1153
1154         param->data_size = param->data_start + needed;
1155 }
1156
1157 static int table_deps(struct dm_ioctl *param, size_t param_size)
1158 {
1159         int r = 0;
1160         struct mapped_device *md;
1161         struct dm_table *table;
1162
1163         md = find_device(param);
1164         if (!md)
1165                 return -ENXIO;
1166
1167         r = __dev_status(md, param);
1168         if (r)
1169                 goto out;
1170
1171         table = dm_get_table(md);
1172         if (table) {
1173                 retrieve_deps(table, param, param_size);
1174                 dm_table_put(table);
1175         }
1176
1177  out:
1178         dm_put(md);
1179         return r;
1180 }
1181
1182 /*
1183  * Return the status of a device as a text string for each
1184  * target.
1185  */
1186 static int table_status(struct dm_ioctl *param, size_t param_size)
1187 {
1188         int r;
1189         struct mapped_device *md;
1190         struct dm_table *table;
1191
1192         md = find_device(param);
1193         if (!md)
1194                 return -ENXIO;
1195
1196         r = __dev_status(md, param);
1197         if (r)
1198                 goto out;
1199
1200         table = dm_get_table(md);
1201         if (table) {
1202                 retrieve_status(table, param, param_size);
1203                 dm_table_put(table);
1204         }
1205
1206  out:
1207         dm_put(md);
1208         return r;
1209 }
1210
1211 /*
1212  * Pass a message to the target that's at the supplied device offset.
1213  */
1214 static int target_message(struct dm_ioctl *param, size_t param_size)
1215 {
1216         int r, argc;
1217         char **argv;
1218         struct mapped_device *md;
1219         struct dm_table *table;
1220         struct dm_target *ti;
1221         struct dm_target_msg *tmsg = (void *) param + param->data_start;
1222
1223         md = find_device(param);
1224         if (!md)
1225                 return -ENXIO;
1226
1227         r = __dev_status(md, param);
1228         if (r)
1229                 goto out;
1230
1231         if (tmsg < (struct dm_target_msg *) (param + 1) ||
1232             invalid_str(tmsg->message, (void *) param + param_size)) {
1233                 DMWARN("Invalid target message parameters.");
1234                 r = -EINVAL;
1235                 goto out;
1236         }
1237
1238         r = dm_split_args(&argc, &argv, tmsg->message);
1239         if (r) {
1240                 DMWARN("Failed to split target message parameters");
1241                 goto out;
1242         }
1243
1244         table = dm_get_table(md);
1245         if (!table)
1246                 goto out_argv;
1247
1248         if (tmsg->sector >= dm_table_get_size(table)) {
1249                 DMWARN("Target message sector outside device.");
1250                 r = -EINVAL;
1251                 goto out_table;
1252         }
1253
1254         ti = dm_table_find_target(table, tmsg->sector);
1255         if (ti->type->message)
1256                 r = ti->type->message(ti, argc, argv);
1257         else {
1258                 DMWARN("Target type does not support messages");
1259                 r = -EINVAL;
1260         }
1261
1262  out_table:
1263         dm_table_put(table);
1264  out_argv:
1265         kfree(argv);
1266  out:
1267         param->data_size = 0;
1268         dm_put(md);
1269         return r;
1270 }
1271
1272 /*-----------------------------------------------------------------
1273  * Implementation of open/close/ioctl on the special char
1274  * device.
1275  *---------------------------------------------------------------*/
1276 static ioctl_fn lookup_ioctl(unsigned int cmd)
1277 {
1278         static struct {
1279                 int cmd;
1280                 ioctl_fn fn;
1281         } _ioctls[] = {
1282                 {DM_VERSION_CMD, NULL}, /* version is dealt with elsewhere */
1283                 {DM_REMOVE_ALL_CMD, remove_all},
1284                 {DM_LIST_DEVICES_CMD, list_devices},
1285
1286                 {DM_DEV_CREATE_CMD, dev_create},
1287                 {DM_DEV_REMOVE_CMD, dev_remove},
1288                 {DM_DEV_RENAME_CMD, dev_rename},
1289                 {DM_DEV_SUSPEND_CMD, dev_suspend},
1290                 {DM_DEV_STATUS_CMD, dev_status},
1291                 {DM_DEV_WAIT_CMD, dev_wait},
1292
1293                 {DM_TABLE_LOAD_CMD, table_load},
1294                 {DM_TABLE_CLEAR_CMD, table_clear},
1295                 {DM_TABLE_DEPS_CMD, table_deps},
1296                 {DM_TABLE_STATUS_CMD, table_status},
1297
1298                 {DM_LIST_VERSIONS_CMD, list_versions},
1299
1300                 {DM_TARGET_MSG_CMD, target_message},
1301                 {DM_DEV_SET_GEOMETRY_CMD, dev_set_geometry}
1302         };
1303
1304         return (cmd >= ARRAY_SIZE(_ioctls)) ? NULL : _ioctls[cmd].fn;
1305 }
1306
1307 /*
1308  * As well as checking the version compatibility this always
1309  * copies the kernel interface version out.
1310  */
1311 static int check_version(unsigned int cmd, struct dm_ioctl __user *user)
1312 {
1313         uint32_t version[3];
1314         int r = 0;
1315
1316         if (copy_from_user(version, user->version, sizeof(version)))
1317                 return -EFAULT;
1318
1319         if ((DM_VERSION_MAJOR != version[0]) ||
1320             (DM_VERSION_MINOR < version[1])) {
1321                 DMWARN("ioctl interface mismatch: "
1322                        "kernel(%u.%u.%u), user(%u.%u.%u), cmd(%d)",
1323                        DM_VERSION_MAJOR, DM_VERSION_MINOR,
1324                        DM_VERSION_PATCHLEVEL,
1325                        version[0], version[1], version[2], cmd);
1326                 r = -EINVAL;
1327         }
1328
1329         /*
1330          * Fill in the kernel version.
1331          */
1332         version[0] = DM_VERSION_MAJOR;
1333         version[1] = DM_VERSION_MINOR;
1334         version[2] = DM_VERSION_PATCHLEVEL;
1335         if (copy_to_user(user->version, version, sizeof(version)))
1336                 return -EFAULT;
1337
1338         return r;
1339 }
1340
1341 static void free_params(struct dm_ioctl *param)
1342 {
1343         vfree(param);
1344 }
1345
1346 static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl **param)
1347 {
1348         struct dm_ioctl tmp, *dmi;
1349
1350         if (copy_from_user(&tmp, user, sizeof(tmp)))
1351                 return -EFAULT;
1352
1353         if (tmp.data_size < sizeof(tmp))
1354                 return -EINVAL;
1355
1356         dmi = (struct dm_ioctl *) vmalloc(tmp.data_size);
1357         if (!dmi)
1358                 return -ENOMEM;
1359
1360         if (copy_from_user(dmi, user, tmp.data_size)) {
1361                 vfree(dmi);
1362                 return -EFAULT;
1363         }
1364
1365         *param = dmi;
1366         return 0;
1367 }
1368
1369 static int validate_params(uint cmd, struct dm_ioctl *param)
1370 {
1371         /* Always clear this flag */
1372         param->flags &= ~DM_BUFFER_FULL_FLAG;
1373
1374         /* Ignores parameters */
1375         if (cmd == DM_REMOVE_ALL_CMD ||
1376             cmd == DM_LIST_DEVICES_CMD ||
1377             cmd == DM_LIST_VERSIONS_CMD)
1378                 return 0;
1379
1380         if ((cmd == DM_DEV_CREATE_CMD)) {
1381                 if (!*param->name) {
1382                         DMWARN("name not supplied when creating device");
1383                         return -EINVAL;
1384                 }
1385         } else if ((*param->uuid && *param->name)) {
1386                 DMWARN("only supply one of name or uuid, cmd(%u)", cmd);
1387                 return -EINVAL;
1388         }
1389
1390         /* Ensure strings are terminated */
1391         param->name[DM_NAME_LEN - 1] = '\0';
1392         param->uuid[DM_UUID_LEN - 1] = '\0';
1393
1394         return 0;
1395 }
1396
1397 static int ctl_ioctl(struct inode *inode, struct file *file,
1398                      uint command, ulong u)
1399 {
1400         int r = 0;
1401         unsigned int cmd;
1402         struct dm_ioctl *param;
1403         struct dm_ioctl __user *user = (struct dm_ioctl __user *) u;
1404         ioctl_fn fn = NULL;
1405         size_t param_size;
1406
1407         /* only root can play with this */
1408         if (!capable(CAP_SYS_ADMIN))
1409                 return -EACCES;
1410
1411         if (_IOC_TYPE(command) != DM_IOCTL)
1412                 return -ENOTTY;
1413
1414         cmd = _IOC_NR(command);
1415
1416         /*
1417          * Check the interface version passed in.  This also
1418          * writes out the kernel's interface version.
1419          */
1420         r = check_version(cmd, user);
1421         if (r)
1422                 return r;
1423
1424         /*
1425          * Nothing more to do for the version command.
1426          */
1427         if (cmd == DM_VERSION_CMD)
1428                 return 0;
1429
1430         fn = lookup_ioctl(cmd);
1431         if (!fn) {
1432                 DMWARN("dm_ctl_ioctl: unknown command 0x%x", command);
1433                 return -ENOTTY;
1434         }
1435
1436         /*
1437          * Trying to avoid low memory issues when a device is
1438          * suspended.
1439          */
1440         current->flags |= PF_MEMALLOC;
1441
1442         /*
1443          * Copy the parameters into kernel space.
1444          */
1445         r = copy_params(user, &param);
1446
1447         current->flags &= ~PF_MEMALLOC;
1448
1449         if (r)
1450                 return r;
1451
1452         r = validate_params(cmd, param);
1453         if (r)
1454                 goto out;
1455
1456         param_size = param->data_size;
1457         param->data_size = sizeof(*param);
1458         r = fn(param, param_size);
1459
1460         /*
1461          * Copy the results back to userland.
1462          */
1463         if (!r && copy_to_user(user, param, param->data_size))
1464                 r = -EFAULT;
1465
1466  out:
1467         free_params(param);
1468         return r;
1469 }
1470
1471 static struct file_operations _ctl_fops = {
1472         .ioctl   = ctl_ioctl,
1473         .owner   = THIS_MODULE,
1474 };
1475
1476 static struct miscdevice _dm_misc = {
1477         .minor          = MISC_DYNAMIC_MINOR,
1478         .name           = DM_NAME,
1479         .devfs_name     = "mapper/control",
1480         .fops           = &_ctl_fops
1481 };
1482
1483 /*
1484  * Create misc character device and link to DM_DIR/control.
1485  */
1486 int __init dm_interface_init(void)
1487 {
1488         int r;
1489
1490         r = dm_hash_init();
1491         if (r)
1492                 return r;
1493
1494         r = misc_register(&_dm_misc);
1495         if (r) {
1496                 DMERR("misc_register failed for control device");
1497                 dm_hash_exit();
1498                 return r;
1499         }
1500
1501         DMINFO("%d.%d.%d%s initialised: %s", DM_VERSION_MAJOR,
1502                DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL, DM_VERSION_EXTRA,
1503                DM_DRIVER_EMAIL);
1504         return 0;
1505 }
1506
1507 void dm_interface_exit(void)
1508 {
1509         if (misc_deregister(&_dm_misc) < 0)
1510                 DMERR("misc_deregister failed for control device");
1511
1512         dm_hash_exit();
1513 }