drivers/md/dm-ioctl.c

   1 /*
   2  * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
   3  * Copyright (C) 2004 - 2006 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/dm-ioctl.h>
  17 #include <linux/hdreg.h>
  18 #include <linux/compat.h>
  19
  20 #include <asm/uaccess.h>
  21
  22 #define DM_MSG_PREFIX "ioctl"
  23 #define DM_DRIVER_EMAIL "dm-devel@redhat.com"
  24
  25 /*-----------------------------------------------------------------
  26  * The ioctl interface needs to be able to look up devices by
  27  * name or uuid.
  28  *---------------------------------------------------------------*/
  29 struct hash_cell {
  30         struct list_head name_list;
  31         struct list_head uuid_list;
  32
  33         char *name;
  34         char *uuid;
  35         struct mapped_device *md;
  36         struct dm_table *new_map;
  37 };
  38
  39 struct vers_iter {
  40     size_t param_size;
  41     struct dm_target_versions *vers, *old_vers;
  42     char *end;
  43     uint32_t flags;
  44 };
  45
  46
  47 #define NUM_BUCKETS 64
  48 #define MASK_BUCKETS (NUM_BUCKETS - 1)
  49 static struct list_head _name_buckets[NUM_BUCKETS];
  50 static struct list_head _uuid_buckets[NUM_BUCKETS];
  51
  52 static void dm_hash_remove_all(int keep_open_devices);
  53
  54 /*
  55  * Guards access to both hash tables.
  56  */
  57 static DECLARE_RWSEM(_hash_lock);
  58
  59 static void init_buckets(struct list_head *buckets)
  60 {
  61         unsigned int i;
  62
  63         for (i = 0; i < NUM_BUCKETS; i++)
  64                 INIT_LIST_HEAD(buckets + i);
  65 }
  66
  67 static int dm_hash_init(void)
  68 {
  69         init_buckets(_name_buckets);
  70         init_buckets(_uuid_buckets);
  71         return 0;
  72 }
  73
  74 static void dm_hash_exit(void)
  75 {
  76         dm_hash_remove_all(0);
  77 }
  78
  79 /*-----------------------------------------------------------------
  80  * Hash function:
  81  * We're not really concerned with the str hash function being
  82  * fast since it's only used by the ioctl interface.
  83  *---------------------------------------------------------------*/
  84 static unsigned int hash_str(const char *str)
  85 {
  86         const unsigned int hash_mult = 2654435387U;
  87         unsigned int h = 0;
  88
  89         while (*str)
  90                 h = (h + (unsigned int) *str++) * hash_mult;
  91
  92         return h & MASK_BUCKETS;
  93 }
  94
  95 /*-----------------------------------------------------------------
  96  * Code for looking up a device by name
  97  *---------------------------------------------------------------*/
  98 static struct hash_cell *__get_name_cell(const char *str)
  99 {
 100         struct hash_cell *hc;
 101         unsigned int h = hash_str(str);
 102
 103         list_for_each_entry (hc, _name_buckets + h, name_list)
 104                 if (!strcmp(hc->name, str)) {
 105                         dm_get(hc->md);
 106                         return hc;
 107                 }
 108
 109         return NULL;
 110 }
 111
 112 static struct hash_cell *__get_uuid_cell(const char *str)
 113 {
 114         struct hash_cell *hc;
 115         unsigned int h = hash_str(str);
 116
 117         list_for_each_entry (hc, _uuid_buckets + h, uuid_list)
 118                 if (!strcmp(hc->uuid, str)) {
 119                         dm_get(hc->md);
 120                         return hc;
 121                 }
 122
 123         return NULL;
 124 }
 125
 126 /*-----------------------------------------------------------------
 127  * Inserting, removing and renaming a device.
 128  *---------------------------------------------------------------*/
 129 static struct hash_cell *alloc_cell(const char *name, const char *uuid,
 130                                     struct mapped_device *md)
 131 {
 132         struct hash_cell *hc;
 133
 134         hc = kmalloc(sizeof(*hc), GFP_KERNEL);
 135         if (!hc)
 136                 return NULL;
 137
 138         hc->name = kstrdup(name, GFP_KERNEL);
 139         if (!hc->name) {
 140                 kfree(hc);
 141                 return NULL;
 142         }
 143
 144         if (!uuid)
 145                 hc->uuid = NULL;
 146
 147         else {
 148                 hc->uuid = kstrdup(uuid, GFP_KERNEL);
 149                 if (!hc->uuid) {
 150                         kfree(hc->name);
 151                         kfree(hc);
 152                         return NULL;
 153                 }
 154         }
 155
 156         INIT_LIST_HEAD(&hc->name_list);
 157         INIT_LIST_HEAD(&hc->uuid_list);
 158         hc->md = md;
 159         hc->new_map = NULL;
 160         return hc;
 161 }
 162
 163 static void free_cell(struct hash_cell *hc)
 164 {
 165         if (hc) {
 166                 kfree(hc->name);
 167                 kfree(hc->uuid);
 168                 kfree(hc);
 169         }
 170 }
 171
 172 /*
 173  * The kdev_t and uuid of a device can never change once it is
 174  * initially inserted.
 175  */
 176 static int dm_hash_insert(const char *name, const char *uuid, struct mapped_device *md)
 177 {
 178         struct hash_cell *cell, *hc;
 179
 180         /*
 181          * Allocate the new cells.
 182          */
 183         cell = alloc_cell(name, uuid, md);
 184         if (!cell)
 185                 return -ENOMEM;
 186
 187         /*
 188          * Insert the cell into both hash tables.
 189          */
 190         down_write(&_hash_lock);
 191         hc = __get_name_cell(name);
 192         if (hc) {
 193                 dm_put(hc->md);
 194                 goto bad;
 195         }
 196
 197         list_add(&cell->name_list, _name_buckets + hash_str(name));
 198
 199         if (uuid) {
 200                 hc = __get_uuid_cell(uuid);
 201                 if (hc) {
 202                         list_del(&cell->name_list);
 203                         dm_put(hc->md);
 204                         goto bad;
 205                 }
 206                 list_add(&cell->uuid_list, _uuid_buckets + hash_str(uuid));
 207         }
 208         dm_get(md);
 209         dm_set_mdptr(md, cell);
 210         up_write(&_hash_lock);
 211
 212         return 0;
 213
 214  bad:
 215         up_write(&_hash_lock);
 216         free_cell(cell);
 217         return -EBUSY;
 218 }
 219
 220 static void __hash_remove(struct hash_cell *hc)
 221 {
 222         struct dm_table *table;
 223
 224         /* remove from the dev hash */
 225         list_del(&hc->uuid_list);
 226         list_del(&hc->name_list);
 227         dm_set_mdptr(hc->md, NULL);
 228
 229         table = dm_get_table(hc->md);
 230         if (table) {
 231                 dm_table_event(table);
 232                 dm_table_put(table);
 233         }
 234
 235         if (hc->new_map)
 236                 dm_table_destroy(hc->new_map);
 237         dm_put(hc->md);
 238         free_cell(hc);
 239 }
 240
 241 static void dm_hash_remove_all(int keep_open_devices)
 242 {
 243         int i, dev_skipped, dev_removed;
 244         struct hash_cell *hc;
 245         struct list_head *tmp, *n;
 246
 247         down_write(&_hash_lock);
 248
 249 retry:
 250         dev_skipped = dev_removed = 0;
 251         for (i = 0; i < NUM_BUCKETS; i++) {
 252                 list_for_each_safe (tmp, n, _name_buckets + i) {
 253                         hc = list_entry(tmp, struct hash_cell, name_list);
 254
 255                         if (keep_open_devices &&
 256                             dm_lock_for_deletion(hc->md)) {
 257                                 dev_skipped++;
 258                                 continue;
 259                         }
 260                         __hash_remove(hc);
 261                         dev_removed = 1;
 262                 }
 263         }
 264
 265         /*
 266          * Some mapped devices may be using other mapped devices, so if any
 267          * still exist, repeat until we make no further progress.
 268          */
 269         if (dev_skipped) {
 270                 if (dev_removed)
 271                         goto retry;
 272
 273                 DMWARN("remove_all left %d open device(s)", dev_skipped);
 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         list_del(&hc->name_list);
 323         old_name = hc->name;
 324         hc->name = new_name;
 325         list_add(&hc->name_list, _name_buckets + hash_str(new_name));
 326
 327         /*
 328          * Wake up any dm event waiters.
 329          */
 330         table = dm_get_table(hc->md);
 331         if (table) {
 332                 dm_table_event(table);
 333                 dm_table_put(table);
 334         }
 335
 336         dm_kobject_uevent(hc->md);
 337
 338         dm_put(hc->md);
 339         up_write(&_hash_lock);
 340         kfree(old_name);
 341         return 0;
 342 }
 343
 344 /*-----------------------------------------------------------------
 345  * Implementation of the ioctl commands
 346  *---------------------------------------------------------------*/
 347 /*
 348  * All the ioctl commands get dispatched to functions with this
 349  * prototype.
 350  */
 351 typedef int (*ioctl_fn)(struct dm_ioctl *param, size_t param_size);
 352
 353 static int remove_all(struct dm_ioctl *param, size_t param_size)
 354 {
 355         dm_hash_remove_all(1);
 356         param->data_size = 0;
 357         return 0;
 358 }
 359
 360 /*
 361  * Round up the ptr to an 8-byte boundary.
 362  */
 363 #define ALIGN_MASK 7
 364 static inline void *align_ptr(void *ptr)
 365 {
 366         return (void *) (((size_t) (ptr + ALIGN_MASK)) & ~ALIGN_MASK);
 367 }
 368
 369 /*
 370  * Retrieves the data payload buffer from an already allocated
 371  * struct dm_ioctl.
 372  */
 373 static void *get_result_buffer(struct dm_ioctl *param, size_t param_size,
 374                                size_t *len)
 375 {
 376         param->data_start = align_ptr(param + 1) - (void *) param;
 377
 378         if (param->data_start < param_size)
 379                 *len = param_size - param->data_start;
 380         else
 381                 *len = 0;
 382
 383         return ((void *) param) + param->data_start;
 384 }
 385
 386 static int list_devices(struct dm_ioctl *param, size_t param_size)
 387 {
 388         unsigned int i;
 389         struct hash_cell *hc;
 390         size_t len, needed = 0;
 391         struct gendisk *disk;
 392         struct dm_name_list *nl, *old_nl = NULL;
 393
 394         down_write(&_hash_lock);
 395
 396         /*
 397          * Loop through all the devices working out how much
 398          * space we need.
 399          */
 400         for (i = 0; i < NUM_BUCKETS; i++) {
 401                 list_for_each_entry (hc, _name_buckets + i, name_list) {
 402                         needed += sizeof(struct dm_name_list);
 403                         needed += strlen(hc->name) + 1;
 404                         needed += ALIGN_MASK;
 405                 }
 406         }
 407
 408         /*
 409          * Grab our output buffer.
 410          */
 411         nl = get_result_buffer(param, param_size, &len);
 412         if (len < needed) {
 413                 param->flags |= DM_BUFFER_FULL_FLAG;
 414                 goto out;
 415         }
 416         param->data_size = param->data_start + needed;
 417
 418         nl->dev = 0;    /* Flags no data */
 419
 420         /*
 421          * Now loop through filling out the names.
 422          */
 423         for (i = 0; i < NUM_BUCKETS; i++) {
 424                 list_for_each_entry (hc, _name_buckets + i, name_list) {
 425                         if (old_nl)
 426                                 old_nl->next = (uint32_t) ((void *) nl -
 427                                                            (void *) old_nl);
 428                         disk = dm_disk(hc->md);
 429                         nl->dev = huge_encode_dev(disk_devt(disk));
 430                         nl->next = 0;
 431                         strcpy(nl->name, hc->name);
 432
 433                         old_nl = nl;
 434                         nl = align_ptr(((void *) ++nl) + strlen(hc->name) + 1);
 435                 }
 436         }
 437
 438  out:
 439         up_write(&_hash_lock);
 440         return 0;
 441 }
 442
 443 static void list_version_get_needed(struct target_type *tt, void *needed_param)
 444 {
 445     size_t *needed = needed_param;
 446
 447     *needed += sizeof(struct dm_target_versions);
 448     *needed += strlen(tt->name);
 449     *needed += ALIGN_MASK;
 450 }
 451
 452 static void list_version_get_info(struct target_type *tt, void *param)
 453 {
 454     struct vers_iter *info = param;
 455
 456     /* Check space - it might have changed since the first iteration */
 457     if ((char *)info->vers + sizeof(tt->version) + strlen(tt->name) + 1 >
 458         info->end) {
 459
 460         info->flags = DM_BUFFER_FULL_FLAG;
 461         return;
 462     }
 463
 464     if (info->old_vers)
 465         info->old_vers->next = (uint32_t) ((void *)info->vers -
 466                                            (void *)info->old_vers);
 467     info->vers->version[0] = tt->version[0];
 468     info->vers->version[1] = tt->version[1];
 469     info->vers->version[2] = tt->version[2];
 470     info->vers->next = 0;
 471     strcpy(info->vers->name, tt->name);
 472
 473     info->old_vers = info->vers;
 474     info->vers = align_ptr(((void *) ++info->vers) + strlen(tt->name) + 1);
 475 }
 476
 477 static int list_versions(struct dm_ioctl *param, size_t param_size)
 478 {
 479         size_t len, needed = 0;
 480         struct dm_target_versions *vers;
 481         struct vers_iter iter_info;
 482
 483         /*
 484          * Loop through all the devices working out how much
 485          * space we need.
 486          */
 487         dm_target_iterate(list_version_get_needed, &needed);
 488
 489         /*
 490          * Grab our output buffer.
 491          */
 492         vers = get_result_buffer(param, param_size, &len);
 493         if (len < needed) {
 494                 param->flags |= DM_BUFFER_FULL_FLAG;
 495                 goto out;
 496         }
 497         param->data_size = param->data_start + needed;
 498
 499         iter_info.param_size = param_size;
 500         iter_info.old_vers = NULL;
 501         iter_info.vers = vers;
 502         iter_info.flags = 0;
 503         iter_info.end = (char *)vers+len;
 504
 505         /*
 506          * Now loop through filling out the names & versions.
 507          */
 508         dm_target_iterate(list_version_get_info, &iter_info);
 509         param->flags |= iter_info.flags;
 510
 511  out:
 512         return 0;
 513 }
 514
 515
 516
 517 static int check_name(const char *name)
 518 {
 519         if (strchr(name, '/')) {
 520                 DMWARN("invalid device name");
 521                 return -EINVAL;
 522         }
 523
 524         return 0;
 525 }
 526
 527 /*
 528  * Fills in a dm_ioctl structure, ready for sending back to
 529  * userland.
 530  */
 531 static int __dev_status(struct mapped_device *md, struct dm_ioctl *param)
 532 {
 533         struct gendisk *disk = dm_disk(md);
 534         struct dm_table *table;
 535
 536         param->flags &= ~(DM_SUSPEND_FLAG | DM_READONLY_FLAG |
 537                           DM_ACTIVE_PRESENT_FLAG);
 538
 539         if (dm_suspended(md))
 540                 param->flags |= DM_SUSPEND_FLAG;
 541
 542         param->dev = huge_encode_dev(disk_devt(disk));
 543
 544         /*
 545          * Yes, this will be out of date by the time it gets back
 546          * to userland, but it is still very useful for
 547          * debugging.
 548          */
 549         param->open_count = dm_open_count(md);
 550
 551         if (get_disk_ro(disk))
 552                 param->flags |= DM_READONLY_FLAG;
 553
 554         param->event_nr = dm_get_event_nr(md);
 555
 556         table = dm_get_table(md);
 557         if (table) {
 558                 param->flags |= DM_ACTIVE_PRESENT_FLAG;
 559                 param->target_count = dm_table_get_num_targets(table);
 560                 dm_table_put(table);
 561         } else
 562                 param->target_count = 0;
 563
 564         return 0;
 565 }
 566
 567 static int dev_create(struct dm_ioctl *param, size_t param_size)
 568 {
 569         int r, m = DM_ANY_MINOR;
 570         struct mapped_device *md;
 571
 572         r = check_name(param->name);
 573         if (r)
 574                 return r;
 575
 576         if (param->flags & DM_PERSISTENT_DEV_FLAG)
 577                 m = MINOR(huge_decode_dev(param->dev));
 578
 579         r = dm_create(m, &md);
 580         if (r)
 581                 return r;
 582
 583         r = dm_hash_insert(param->name, *param->uuid ? param->uuid : NULL, md);
 584         if (r) {
 585                 dm_put(md);
 586                 return r;
 587         }
 588
 589         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
 590
 591         r = __dev_status(md, param);
 592         dm_put(md);
 593
 594         return r;
 595 }
 596
 597 /*
 598  * Always use UUID for lookups if it's present, otherwise use name or dev.
 599  */
 600 static struct hash_cell *__find_device_hash_cell(struct dm_ioctl *param)
 601 {
 602         struct mapped_device *md;
 603         void *mdptr = NULL;
 604
 605         if (*param->uuid)
 606                 return __get_uuid_cell(param->uuid);
 607
 608         if (*param->name)
 609                 return __get_name_cell(param->name);
 610
 611         md = dm_get_md(huge_decode_dev(param->dev));
 612         if (!md)
 613                 goto out;
 614
 615         mdptr = dm_get_mdptr(md);
 616         if (!mdptr)
 617                 dm_put(md);
 618
 619 out:
 620         return mdptr;
 621 }
 622
 623 static struct mapped_device *find_device(struct dm_ioctl *param)
 624 {
 625         struct hash_cell *hc;
 626         struct mapped_device *md = NULL;
 627
 628         down_read(&_hash_lock);
 629         hc = __find_device_hash_cell(param);
 630         if (hc) {
 631                 md = hc->md;
 632
 633                 /*
 634                  * Sneakily write in both the name and the uuid
 635                  * while we have the cell.
 636                  */
 637                 strncpy(param->name, hc->name, sizeof(param->name));
 638                 if (hc->uuid)
 639                         strncpy(param->uuid, hc->uuid, sizeof(param->uuid)-1);
 640                 else
 641                         param->uuid[0] = '\0';
 642
 643                 if (hc->new_map)
 644                         param->flags |= DM_INACTIVE_PRESENT_FLAG;
 645                 else
 646                         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
 647         }
 648         up_read(&_hash_lock);
 649
 650         return md;
 651 }
 652
 653 static int dev_remove(struct dm_ioctl *param, size_t param_size)
 654 {
 655         struct hash_cell *hc;
 656         struct mapped_device *md;
 657         int r;
 658
 659         down_write(&_hash_lock);
 660         hc = __find_device_hash_cell(param);
 661
 662         if (!hc) {
 663                 DMWARN("device doesn't appear to be in the dev hash table.");
 664                 up_write(&_hash_lock);
 665                 return -ENXIO;
 666         }
 667
 668         md = hc->md;
 669
 670         /*
 671          * Ensure the device is not open and nothing further can open it.
 672          */
 673         r = dm_lock_for_deletion(md);
 674         if (r) {
 675                 DMWARN("unable to remove open device %s", hc->name);
 676                 up_write(&_hash_lock);
 677                 dm_put(md);
 678                 return r;
 679         }
 680
 681         __hash_remove(hc);
 682         up_write(&_hash_lock);
 683         dm_put(md);
 684         param->data_size = 0;
 685         return 0;
 686 }
 687
 688 /*
 689  * Check a string doesn't overrun the chunk of
 690  * memory we copied from userland.
 691  */
 692 static int invalid_str(char *str, void *end)
 693 {
 694         while ((void *) str < end)
 695                 if (!*str++)
 696                         return 0;
 697
 698         return -EINVAL;
 699 }
 700
 701 static int dev_rename(struct dm_ioctl *param, size_t param_size)
 702 {
 703         int r;
 704         char *new_name = (char *) param + param->data_start;
 705
 706         if (new_name < param->data ||
 707             invalid_str(new_name, (void *) param + param_size) ||
 708             strlen(new_name) > DM_NAME_LEN - 1) {
 709                 DMWARN("Invalid new logical volume name supplied.");
 710                 return -EINVAL;
 711         }
 712
 713         r = check_name(new_name);
 714         if (r)
 715                 return r;
 716
 717         param->data_size = 0;
 718         return dm_hash_rename(param->name, new_name);
 719 }
 720
 721 static int dev_set_geometry(struct dm_ioctl *param, size_t param_size)
 722 {
 723         int r = -EINVAL, x;
 724         struct mapped_device *md;
 725         struct hd_geometry geometry;
 726         unsigned long indata[4];
 727         char *geostr = (char *) param + param->data_start;
 728
 729         md = find_device(param);
 730         if (!md)
 731                 return -ENXIO;
 732
 733         if (geostr < param->data ||
 734             invalid_str(geostr, (void *) param + param_size)) {
 735                 DMWARN("Invalid geometry supplied.");
 736                 goto out;
 737         }
 738
 739         x = sscanf(geostr, "%lu %lu %lu %lu", indata,
 740                    indata + 1, indata + 2, indata + 3);
 741
 742         if (x != 4) {
 743                 DMWARN("Unable to interpret geometry settings.");
 744                 goto out;
 745         }
 746
 747         if (indata[0] > 65535 || indata[1] > 255 ||
 748             indata[2] > 255 || indata[3] > ULONG_MAX) {
 749                 DMWARN("Geometry exceeds range limits.");
 750                 goto out;
 751         }
 752
 753         geometry.cylinders = indata[0];
 754         geometry.heads = indata[1];
 755         geometry.sectors = indata[2];
 756         geometry.start = indata[3];
 757
 758         r = dm_set_geometry(md, &geometry);
 759         if (!r)
 760                 r = __dev_status(md, param);
 761
 762         param->data_size = 0;
 763
 764 out:
 765         dm_put(md);
 766         return r;
 767 }
 768
 769 static int do_suspend(struct dm_ioctl *param)
 770 {
 771         int r = 0;
 772         unsigned suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
 773         struct mapped_device *md;
 774
 775         md = find_device(param);
 776         if (!md)
 777                 return -ENXIO;
 778
 779         if (param->flags & DM_SKIP_LOCKFS_FLAG)
 780                 suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
 781         if (param->flags & DM_NOFLUSH_FLAG)
 782                 suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
 783
 784         if (!dm_suspended(md))
 785                 r = dm_suspend(md, suspend_flags);
 786
 787         if (!r)
 788                 r = __dev_status(md, param);
 789
 790         dm_put(md);
 791         return r;
 792 }
 793
 794 static int do_resume(struct dm_ioctl *param)
 795 {
 796         int r = 0;
 797         unsigned suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
 798         struct hash_cell *hc;
 799         struct mapped_device *md;
 800         struct dm_table *new_map;
 801
 802         down_write(&_hash_lock);
 803
 804         hc = __find_device_hash_cell(param);
 805         if (!hc) {
 806                 DMWARN("device doesn't appear to be in the dev hash table.");
 807                 up_write(&_hash_lock);
 808                 return -ENXIO;
 809         }
 810
 811         md = hc->md;
 812
 813         new_map = hc->new_map;
 814         hc->new_map = NULL;
 815         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
 816
 817         up_write(&_hash_lock);
 818
 819         /* Do we need to load a new map ? */
 820         if (new_map) {
 821                 /* Suspend if it isn't already suspended */
 822                 if (param->flags & DM_SKIP_LOCKFS_FLAG)
 823                         suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
 824                 if (param->flags & DM_NOFLUSH_FLAG)
 825                         suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
 826                 if (!dm_suspended(md))
 827                         dm_suspend(md, suspend_flags);
 828
 829                 r = dm_swap_table(md, new_map);
 830                 if (r) {
 831                         dm_table_destroy(new_map);
 832                         dm_put(md);
 833                         return r;
 834                 }
 835
 836                 if (dm_table_get_mode(new_map) & FMODE_WRITE)
 837                         set_disk_ro(dm_disk(md), 0);
 838                 else
 839                         set_disk_ro(dm_disk(md), 1);
 840         }
 841
 842         if (dm_suspended(md))
 843                 r = dm_resume(md);
 844
 845         if (!r)
 846                 r = __dev_status(md, param);
 847
 848         dm_put(md);
 849         return r;
 850 }
 851
 852 /*
 853  * Set or unset the suspension state of a device.
 854  * If the device already is in the requested state we just return its status.
 855  */
 856 static int dev_suspend(struct dm_ioctl *param, size_t param_size)
 857 {
 858         if (param->flags & DM_SUSPEND_FLAG)
 859                 return do_suspend(param);
 860
 861         return do_resume(param);
 862 }
 863
 864 /*
 865  * Copies device info back to user space, used by
 866  * the create and info ioctls.
 867  */
 868 static int dev_status(struct dm_ioctl *param, size_t param_size)
 869 {
 870         int r;
 871         struct mapped_device *md;
 872
 873         md = find_device(param);
 874         if (!md)
 875                 return -ENXIO;
 876
 877         r = __dev_status(md, param);
 878         dm_put(md);
 879         return r;
 880 }
 881
 882 /*
 883  * Build up the status struct for each target
 884  */
 885 static void retrieve_status(struct dm_table *table,
 886                             struct dm_ioctl *param, size_t param_size)
 887 {
 888         unsigned int i, num_targets;
 889         struct dm_target_spec *spec;
 890         char *outbuf, *outptr;
 891         status_type_t type;
 892         size_t remaining, len, used = 0;
 893
 894         outptr = outbuf = get_result_buffer(param, param_size, &len);
 895
 896         if (param->flags & DM_STATUS_TABLE_FLAG)
 897                 type = STATUSTYPE_TABLE;
 898         else
 899                 type = STATUSTYPE_INFO;
 900
 901         /* Get all the target info */
 902         num_targets = dm_table_get_num_targets(table);
 903         for (i = 0; i < num_targets; i++) {
 904                 struct dm_target *ti = dm_table_get_target(table, i);
 905
 906                 remaining = len - (outptr - outbuf);
 907                 if (remaining <= sizeof(struct dm_target_spec)) {
 908                         param->flags |= DM_BUFFER_FULL_FLAG;
 909                         break;
 910                 }
 911
 912                 spec = (struct dm_target_spec *) outptr;
 913
 914                 spec->status = 0;
 915                 spec->sector_start = ti->begin;
 916                 spec->length = ti->len;
 917                 strncpy(spec->target_type, ti->type->name,
 918                         sizeof(spec->target_type));
 919
 920                 outptr += sizeof(struct dm_target_spec);
 921                 remaining = len - (outptr - outbuf);
 922                 if (remaining <= 0) {
 923                         param->flags |= DM_BUFFER_FULL_FLAG;
 924                         break;
 925                 }
 926
 927                 /* Get the status/table string from the target driver */
 928                 if (ti->type->status) {
 929                         if (ti->type->status(ti, type, outptr, remaining)) {
 930                                 param->flags |= DM_BUFFER_FULL_FLAG;
 931                                 break;
 932                         }
 933                 } else
 934                         outptr[0] = '\0';
 935
 936                 outptr += strlen(outptr) + 1;
 937                 used = param->data_start + (outptr - outbuf);
 938
 939                 outptr = align_ptr(outptr);
 940                 spec->next = outptr - outbuf;
 941         }
 942
 943         if (used)
 944                 param->data_size = used;
 945
 946         param->target_count = num_targets;
 947 }
 948
 949 /*
 950  * Wait for a device to report an event
 951  */
 952 static int dev_wait(struct dm_ioctl *param, size_t param_size)
 953 {
 954         int r;
 955         struct mapped_device *md;
 956         struct dm_table *table;
 957
 958         md = find_device(param);
 959         if (!md)
 960                 return -ENXIO;
 961
 962         /*
 963          * Wait for a notification event
 964          */
 965         if (dm_wait_event(md, param->event_nr)) {
 966                 r = -ERESTARTSYS;
 967                 goto out;
 968         }
 969
 970         /*
 971          * The userland program is going to want to know what
 972          * changed to trigger the event, so we may as well tell
 973          * him and save an ioctl.
 974          */
 975         r = __dev_status(md, param);
 976         if (r)
 977                 goto out;
 978
 979         table = dm_get_table(md);
 980         if (table) {
 981                 retrieve_status(table, param, param_size);
 982                 dm_table_put(table);
 983         }
 984
 985  out:
 986         dm_put(md);
 987         return r;
 988 }
 989
 990 static inline fmode_t get_mode(struct dm_ioctl *param)
 991 {
 992         fmode_t mode = FMODE_READ | FMODE_WRITE;
 993
 994         if (param->flags & DM_READONLY_FLAG)
 995                 mode = FMODE_READ;
 996
 997         return mode;
 998 }
 999
1000 static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
1001                        struct dm_target_spec **spec, char **target_params)
1002 {
1003         *spec = (struct dm_target_spec *) ((unsigned char *) last + next);
1004         *target_params = (char *) (*spec + 1);
1005
1006         if (*spec < (last + 1))
1007                 return -EINVAL;
1008
1009         return invalid_str(*target_params, end);
1010 }
1011
1012 static int populate_table(struct dm_table *table,
1013                           struct dm_ioctl *param, size_t param_size)
1014 {
1015         int r;
1016         unsigned int i = 0;
1017         struct dm_target_spec *spec = (struct dm_target_spec *) param;
1018         uint32_t next = param->data_start;
1019         void *end = (void *) param + param_size;
1020         char *target_params;
1021
1022         if (!param->target_count) {
1023                 DMWARN("populate_table: no targets specified");
1024                 return -EINVAL;
1025         }
1026
1027         for (i = 0; i < param->target_count; i++) {
1028
1029                 r = next_target(spec, next, end, &spec, &target_params);
1030                 if (r) {
1031                         DMWARN("unable to find target");
1032                         return r;
1033                 }
1034
1035                 r = dm_table_add_target(table, spec->target_type,
1036                                         (sector_t) spec->sector_start,
1037                                         (sector_t) spec->length,
1038                                         target_params);
1039                 if (r) {
1040                         DMWARN("error adding target to table");
1041                         return r;
1042                 }
1043
1044                 next = spec->next;
1045         }
1046
1047         return dm_table_complete(table);
1048 }
1049
1050 static int table_prealloc_integrity(struct dm_table *t,
1051                                     struct mapped_device *md)
1052 {
1053         struct list_head *devices = dm_table_get_devices(t);
1054         struct dm_dev_internal *dd;
1055
1056         list_for_each_entry(dd, devices, list)
1057                 if (bdev_get_integrity(dd->dm_dev.bdev))
1058                         return blk_integrity_register(dm_disk(md), NULL);
1059
1060         return 0;
1061 }
1062
1063 static int table_load(struct dm_ioctl *param, size_t param_size)
1064 {
1065         int r;
1066         struct hash_cell *hc;
1067         struct dm_table *t;
1068         struct mapped_device *md;
1069
1070         md = find_device(param);
1071         if (!md)
1072                 return -ENXIO;
1073
1074         r = dm_table_create(&t, get_mode(param), param->target_count, md);
1075         if (r)
1076                 goto out;
1077
1078         r = populate_table(t, param, param_size);
1079         if (r) {
1080                 dm_table_destroy(t);
1081                 goto out;
1082         }
1083
1084         r = table_prealloc_integrity(t, md);
1085         if (r) {
1086                 DMERR("%s: could not register integrity profile.",
1087                       dm_device_name(md));
1088                 dm_table_destroy(t);
1089                 goto out;
1090         }
1091
1092         down_write(&_hash_lock);
1093         hc = dm_get_mdptr(md);
1094         if (!hc || hc->md != md) {
1095                 DMWARN("device has been removed from the dev hash table.");
1096                 dm_table_destroy(t);
1097                 up_write(&_hash_lock);
1098                 r = -ENXIO;
1099                 goto out;
1100         }
1101
1102         if (hc->new_map)
1103                 dm_table_destroy(hc->new_map);
1104         hc->new_map = t;
1105         up_write(&_hash_lock);
1106
1107         param->flags |= DM_INACTIVE_PRESENT_FLAG;
1108         r = __dev_status(md, param);
1109
1110 out:
1111         dm_put(md);
1112
1113         return r;
1114 }
1115
1116 static int table_clear(struct dm_ioctl *param, size_t param_size)
1117 {
1118         int r;
1119         struct hash_cell *hc;
1120         struct mapped_device *md;
1121
1122         down_write(&_hash_lock);
1123
1124         hc = __find_device_hash_cell(param);
1125         if (!hc) {
1126                 DMWARN("device doesn't appear to be in the dev hash table.");
1127                 up_write(&_hash_lock);
1128                 return -ENXIO;
1129         }
1130
1131         if (hc->new_map) {
1132                 dm_table_destroy(hc->new_map);
1133                 hc->new_map = NULL;
1134         }
1135
1136         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1137
1138         r = __dev_status(hc->md, param);
1139         md = hc->md;
1140         up_write(&_hash_lock);
1141         dm_put(md);
1142         return r;
1143 }
1144
1145 /*
1146  * Retrieves a list of devices used by a particular dm device.
1147  */
1148 static void retrieve_deps(struct dm_table *table,
1149                           struct dm_ioctl *param, size_t param_size)
1150 {
1151         unsigned int count = 0;
1152         struct list_head *tmp;
1153         size_t len, needed;
1154         struct dm_dev_internal *dd;
1155         struct dm_target_deps *deps;
1156
1157         deps = get_result_buffer(param, param_size, &len);
1158
1159         /*
1160          * Count the devices.
1161          */
1162         list_for_each (tmp, dm_table_get_devices(table))
1163                 count++;
1164
1165         /*
1166          * Check we have enough space.
1167          */
1168         needed = sizeof(*deps) + (sizeof(*deps->dev) * count);
1169         if (len < needed) {
1170                 param->flags |= DM_BUFFER_FULL_FLAG;
1171                 return;
1172         }
1173
1174         /*
1175          * Fill in the devices.
1176          */
1177         deps->count = count;
1178         count = 0;
1179         list_for_each_entry (dd, dm_table_get_devices(table), list)
1180                 deps->dev[count++] = huge_encode_dev(dd->dm_dev.bdev->bd_dev);
1181
1182         param->data_size = param->data_start + needed;
1183 }
1184
1185 static int table_deps(struct dm_ioctl *param, size_t param_size)
1186 {
1187         int r = 0;
1188         struct mapped_device *md;
1189         struct dm_table *table;
1190
1191         md = find_device(param);
1192         if (!md)
1193                 return -ENXIO;
1194
1195         r = __dev_status(md, param);
1196         if (r)
1197                 goto out;
1198
1199         table = dm_get_table(md);
1200         if (table) {
1201                 retrieve_deps(table, param, param_size);
1202                 dm_table_put(table);
1203         }
1204
1205  out:
1206         dm_put(md);
1207         return r;
1208 }
1209
1210 /*
1211  * Return the status of a device as a text string for each
1212  * target.
1213  */
1214 static int table_status(struct dm_ioctl *param, size_t param_size)
1215 {
1216         int r;
1217         struct mapped_device *md;
1218         struct dm_table *table;
1219
1220         md = find_device(param);
1221         if (!md)
1222                 return -ENXIO;
1223
1224         r = __dev_status(md, param);
1225         if (r)
1226                 goto out;
1227
1228         table = dm_get_table(md);
1229         if (table) {
1230                 retrieve_status(table, param, param_size);
1231                 dm_table_put(table);
1232         }
1233
1234  out:
1235         dm_put(md);
1236         return r;
1237 }
1238
1239 /*
1240  * Pass a message to the target that's at the supplied device offset.
1241  */
1242 static int target_message(struct dm_ioctl *param, size_t param_size)
1243 {
1244         int r, argc;
1245         char **argv;
1246         struct mapped_device *md;
1247         struct dm_table *table;
1248         struct dm_target *ti;
1249         struct dm_target_msg *tmsg = (void *) param + param->data_start;
1250
1251         md = find_device(param);
1252         if (!md)
1253                 return -ENXIO;
1254
1255         r = __dev_status(md, param);
1256         if (r)
1257                 goto out;
1258
1259         if (tmsg < (struct dm_target_msg *) param->data ||
1260             invalid_str(tmsg->message, (void *) param + param_size)) {
1261                 DMWARN("Invalid target message parameters.");
1262                 r = -EINVAL;
1263                 goto out;
1264         }
1265
1266         r = dm_split_args(&argc, &argv, tmsg->message);
1267         if (r) {
1268                 DMWARN("Failed to split target message parameters");
1269                 goto out;
1270         }
1271
1272         table = dm_get_table(md);
1273         if (!table)
1274                 goto out_argv;
1275
1276         ti = dm_table_find_target(table, tmsg->sector);
1277         if (!dm_target_is_valid(ti)) {
1278                 DMWARN("Target message sector outside device.");
1279                 r = -EINVAL;
1280         } else if (ti->type->message)
1281                 r = ti->type->message(ti, argc, argv);
1282         else {
1283                 DMWARN("Target type does not support messages");
1284                 r = -EINVAL;
1285         }
1286
1287         dm_table_put(table);
1288  out_argv:
1289         kfree(argv);
1290  out:
1291         param->data_size = 0;
1292         dm_put(md);
1293         return r;
1294 }
1295
1296 /*-----------------------------------------------------------------
1297  * Implementation of open/close/ioctl on the special char
1298  * device.
1299  *---------------------------------------------------------------*/
1300 static ioctl_fn lookup_ioctl(unsigned int cmd)
1301 {
1302         static struct {
1303                 int cmd;
1304                 ioctl_fn fn;
1305         } _ioctls[] = {
1306                 {DM_VERSION_CMD, NULL}, /* version is dealt with elsewhere */
1307                 {DM_REMOVE_ALL_CMD, remove_all},
1308                 {DM_LIST_DEVICES_CMD, list_devices},
1309
1310                 {DM_DEV_CREATE_CMD, dev_create},
1311                 {DM_DEV_REMOVE_CMD, dev_remove},
1312                 {DM_DEV_RENAME_CMD, dev_rename},
1313                 {DM_DEV_SUSPEND_CMD, dev_suspend},
1314                 {DM_DEV_STATUS_CMD, dev_status},
1315                 {DM_DEV_WAIT_CMD, dev_wait},
1316
1317                 {DM_TABLE_LOAD_CMD, table_load},
1318                 {DM_TABLE_CLEAR_CMD, table_clear},
1319                 {DM_TABLE_DEPS_CMD, table_deps},
1320                 {DM_TABLE_STATUS_CMD, table_status},
1321
1322                 {DM_LIST_VERSIONS_CMD, list_versions},
1323
1324                 {DM_TARGET_MSG_CMD, target_message},
1325                 {DM_DEV_SET_GEOMETRY_CMD, dev_set_geometry}
1326         };
1327
1328         return (cmd >= ARRAY_SIZE(_ioctls)) ? NULL : _ioctls[cmd].fn;
1329 }
1330
1331 /*
1332  * As well as checking the version compatibility this always
1333  * copies the kernel interface version out.
1334  */
1335 static int check_version(unsigned int cmd, struct dm_ioctl __user *user)
1336 {
1337         uint32_t version[3];
1338         int r = 0;
1339
1340         if (copy_from_user(version, user->version, sizeof(version)))
1341                 return -EFAULT;
1342
1343         if ((DM_VERSION_MAJOR != version[0]) ||
1344             (DM_VERSION_MINOR < version[1])) {
1345                 DMWARN("ioctl interface mismatch: "
1346                        "kernel(%u.%u.%u), user(%u.%u.%u), cmd(%d)",
1347                        DM_VERSION_MAJOR, DM_VERSION_MINOR,
1348                        DM_VERSION_PATCHLEVEL,
1349                        version[0], version[1], version[2], cmd);
1350                 r = -EINVAL;
1351         }
1352
1353         /*
1354          * Fill in the kernel version.
1355          */
1356         version[0] = DM_VERSION_MAJOR;
1357         version[1] = DM_VERSION_MINOR;
1358         version[2] = DM_VERSION_PATCHLEVEL;
1359         if (copy_to_user(user->version, version, sizeof(version)))
1360                 return -EFAULT;
1361
1362         return r;
1363 }
1364
1365 static void free_params(struct dm_ioctl *param)
1366 {
1367         vfree(param);
1368 }
1369
1370 static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl **param)
1371 {
1372         struct dm_ioctl tmp, *dmi;
1373
1374         if (copy_from_user(&tmp, user, sizeof(tmp) - sizeof(tmp.data)))
1375                 return -EFAULT;
1376
1377         if (tmp.data_size < (sizeof(tmp) - sizeof(tmp.data)))
1378                 return -EINVAL;
1379
1380         dmi = vmalloc(tmp.data_size);
1381         if (!dmi)
1382                 return -ENOMEM;
1383
1384         if (copy_from_user(dmi, user, tmp.data_size)) {
1385                 vfree(dmi);
1386                 return -EFAULT;
1387         }
1388
1389         *param = dmi;
1390         return 0;
1391 }
1392
1393 static int validate_params(uint cmd, struct dm_ioctl *param)
1394 {
1395         /* Always clear this flag */
1396         param->flags &= ~DM_BUFFER_FULL_FLAG;
1397
1398         /* Ignores parameters */
1399         if (cmd == DM_REMOVE_ALL_CMD ||
1400             cmd == DM_LIST_DEVICES_CMD ||
1401             cmd == DM_LIST_VERSIONS_CMD)
1402                 return 0;
1403
1404         if ((cmd == DM_DEV_CREATE_CMD)) {
1405                 if (!*param->name) {
1406                         DMWARN("name not supplied when creating device");
1407                         return -EINVAL;
1408                 }
1409         } else if ((*param->uuid && *param->name)) {
1410                 DMWARN("only supply one of name or uuid, cmd(%u)", cmd);
1411                 return -EINVAL;
1412         }
1413
1414         /* Ensure strings are terminated */
1415         param->name[DM_NAME_LEN - 1] = '\0';
1416         param->uuid[DM_UUID_LEN - 1] = '\0';
1417
1418         return 0;
1419 }
1420
1421 static int ctl_ioctl(uint command, struct dm_ioctl __user *user)
1422 {
1423         int r = 0;
1424         unsigned int cmd;
1425         struct dm_ioctl *uninitialized_var(param);
1426         ioctl_fn fn = NULL;
1427         size_t param_size;
1428
1429         /* only root can play with this */
1430         if (!capable(CAP_SYS_ADMIN))
1431                 return -EACCES;
1432
1433         if (_IOC_TYPE(command) != DM_IOCTL)
1434                 return -ENOTTY;
1435
1436         cmd = _IOC_NR(command);
1437
1438         /*
1439          * Check the interface version passed in.  This also
1440          * writes out the kernel's interface version.
1441          */
1442         r = check_version(cmd, user);
1443         if (r)
1444                 return r;
1445
1446         /*
1447          * Nothing more to do for the version command.
1448          */
1449         if (cmd == DM_VERSION_CMD)
1450                 return 0;
1451
1452         fn = lookup_ioctl(cmd);
1453         if (!fn) {
1454                 DMWARN("dm_ctl_ioctl: unknown command 0x%x", command);
1455                 return -ENOTTY;
1456         }
1457
1458         /*
1459          * Trying to avoid low memory issues when a device is
1460          * suspended.
1461          */
1462         current->flags |= PF_MEMALLOC;
1463
1464         /*
1465          * Copy the parameters into kernel space.
1466          */
1467         r = copy_params(user, &param);
1468
1469         current->flags &= ~PF_MEMALLOC;
1470
1471         if (r)
1472                 return r;
1473
1474         r = validate_params(cmd, param);
1475         if (r)
1476                 goto out;
1477
1478         param_size = param->data_size;
1479         param->data_size = sizeof(*param);
1480         r = fn(param, param_size);
1481
1482         /*
1483          * Copy the results back to userland.
1484          */
1485         if (!r && copy_to_user(user, param, param->data_size))
1486                 r = -EFAULT;
1487
1488  out:
1489         free_params(param);
1490         return r;
1491 }
1492
1493 static long dm_ctl_ioctl(struct file *file, uint command, ulong u)
1494 {
1495         return (long)ctl_ioctl(command, (struct dm_ioctl __user *)u);
1496 }
1497
1498 #ifdef CONFIG_COMPAT
1499 static long dm_compat_ctl_ioctl(struct file *file, uint command, ulong u)
1500 {
1501         return (long)dm_ctl_ioctl(file, command, (ulong) compat_ptr(u));
1502 }
1503 #else
1504 #define dm_compat_ctl_ioctl NULL
1505 #endif
1506
1507 static const struct file_operations _ctl_fops = {
1508         .unlocked_ioctl  = dm_ctl_ioctl,
1509         .compat_ioctl = dm_compat_ctl_ioctl,
1510         .owner   = THIS_MODULE,
1511 };
1512
1513 static struct miscdevice _dm_misc = {
1514         .minor          = MISC_DYNAMIC_MINOR,
1515         .name           = DM_NAME,
1516         .devnode        = "mapper/control",
1517         .fops           = &_ctl_fops
1518 };
1519
1520 /*
1521  * Create misc character device and link to DM_DIR/control.
1522  */
1523 int __init dm_interface_init(void)
1524 {
1525         int r;
1526
1527         r = dm_hash_init();
1528         if (r)
1529                 return r;
1530
1531         r = misc_register(&_dm_misc);
1532         if (r) {
1533                 DMERR("misc_register failed for control device");
1534                 dm_hash_exit();
1535                 return r;
1536         }
1537
1538         DMINFO("%d.%d.%d%s initialised: %s", DM_VERSION_MAJOR,
1539                DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL, DM_VERSION_EXTRA,
1540                DM_DRIVER_EMAIL);
1541         return 0;
1542 }
1543
1544 void dm_interface_exit(void)
1545 {
1546         if (misc_deregister(&_dm_misc) < 0)
1547                 DMERR("misc_deregister failed for control device");
1548
1549         dm_hash_exit();
1550 }
1551
1552 /**
1553  * dm_copy_name_and_uuid - Copy mapped device name & uuid into supplied buffers
1554  * @md: Pointer to mapped_device
1555  * @name: Buffer (size DM_NAME_LEN) for name
1556  * @uuid: Buffer (size DM_UUID_LEN) for uuid or empty string if uuid not defined
1557  */
1558 int dm_copy_name_and_uuid(struct mapped_device *md, char *name, char *uuid)
1559 {
1560         int r = 0;
1561         struct hash_cell *hc;
1562
1563         if (!md)
1564                 return -ENXIO;
1565
1566         dm_get(md);
1567         down_read(&_hash_lock);
1568         hc = dm_get_mdptr(md);
1569         if (!hc || hc->md != md) {
1570                 r = -ENXIO;
1571                 goto out;
1572         }
1573
1574         if (name)
1575                 strcpy(name, hc->name);
1576         if (uuid)
1577                 strcpy(uuid, hc->uuid ? : "");
1578
1579 out:
1580         up_read(&_hash_lock);
1581         dm_put(md);
1582
1583         return r;
1584 }