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
  19 #include <asm/uaccess.h>
  20
  21 #define DM_MSG_PREFIX "ioctl"
  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(int keep_open_devices);
  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         return 0;
  71 }
  72
  73 static void dm_hash_exit(void)
  74 {
  75         dm_hash_remove_all(0);
  76 }
  77
  78 /*-----------------------------------------------------------------
  79  * Hash function:
  80  * We're not really concerned with the str hash function being
  81  * fast since it's only used by the ioctl interface.
  82  *---------------------------------------------------------------*/
  83 static unsigned int hash_str(const char *str)
  84 {
  85         const unsigned int hash_mult = 2654435387U;
  86         unsigned int h = 0;
  87
  88         while (*str)
  89                 h = (h + (unsigned int) *str++) * hash_mult;
  90
  91         return h & MASK_BUCKETS;
  92 }
  93
  94 /*-----------------------------------------------------------------
  95  * Code for looking up a device by name
  96  *---------------------------------------------------------------*/
  97 static struct hash_cell *__get_name_cell(const char *str)
  98 {
  99         struct hash_cell *hc;
 100         unsigned int h = hash_str(str);
 101
 102         list_for_each_entry (hc, _name_buckets + h, name_list)
 103                 if (!strcmp(hc->name, str)) {
 104                         dm_get(hc->md);
 105                         return hc;
 106                 }
 107
 108         return NULL;
 109 }
 110
 111 static struct hash_cell *__get_uuid_cell(const char *str)
 112 {
 113         struct hash_cell *hc;
 114         unsigned int h = hash_str(str);
 115
 116         list_for_each_entry (hc, _uuid_buckets + h, uuid_list)
 117                 if (!strcmp(hc->uuid, str)) {
 118                         dm_get(hc->md);
 119                         return hc;
 120                 }
 121
 122         return NULL;
 123 }
 124
 125 /*-----------------------------------------------------------------
 126  * Inserting, removing and renaming a device.
 127  *---------------------------------------------------------------*/
 128 static struct hash_cell *alloc_cell(const char *name, const char *uuid,
 129                                     struct mapped_device *md)
 130 {
 131         struct hash_cell *hc;
 132
 133         hc = kmalloc(sizeof(*hc), GFP_KERNEL);
 134         if (!hc)
 135                 return NULL;
 136
 137         hc->name = kstrdup(name, GFP_KERNEL);
 138         if (!hc->name) {
 139                 kfree(hc);
 140                 return NULL;
 141         }
 142
 143         if (!uuid)
 144                 hc->uuid = NULL;
 145
 146         else {
 147                 hc->uuid = kstrdup(uuid, GFP_KERNEL);
 148                 if (!hc->uuid) {
 149                         kfree(hc->name);
 150                         kfree(hc);
 151                         return NULL;
 152                 }
 153         }
 154
 155         INIT_LIST_HEAD(&hc->name_list);
 156         INIT_LIST_HEAD(&hc->uuid_list);
 157         hc->md = md;
 158         hc->new_map = NULL;
 159         return hc;
 160 }
 161
 162 static void free_cell(struct hash_cell *hc)
 163 {
 164         if (hc) {
 165                 kfree(hc->name);
 166                 kfree(hc->uuid);
 167                 kfree(hc);
 168         }
 169 }
 170
 171 /*
 172  * The kdev_t and uuid of a device can never change once it is
 173  * initially inserted.
 174  */
 175 static int dm_hash_insert(const char *name, const char *uuid, struct mapped_device *md)
 176 {
 177         struct hash_cell *cell, *hc;
 178
 179         /*
 180          * Allocate the new cells.
 181          */
 182         cell = alloc_cell(name, uuid, md);
 183         if (!cell)
 184                 return -ENOMEM;
 185
 186         /*
 187          * Insert the cell into both hash tables.
 188          */
 189         down_write(&_hash_lock);
 190         hc = __get_name_cell(name);
 191         if (hc) {
 192                 dm_put(hc->md);
 193                 goto bad;
 194         }
 195
 196         list_add(&cell->name_list, _name_buckets + hash_str(name));
 197
 198         if (uuid) {
 199                 hc = __get_uuid_cell(uuid);
 200                 if (hc) {
 201                         list_del(&cell->name_list);
 202                         dm_put(hc->md);
 203                         goto bad;
 204                 }
 205                 list_add(&cell->uuid_list, _uuid_buckets + hash_str(uuid));
 206         }
 207         dm_get(md);
 208         dm_set_mdptr(md, cell);
 209         up_write(&_hash_lock);
 210
 211         return 0;
 212
 213  bad:
 214         up_write(&_hash_lock);
 215         free_cell(cell);
 216         return -EBUSY;
 217 }
 218
 219 static void __hash_remove(struct hash_cell *hc)
 220 {
 221         struct dm_table *table;
 222
 223         /* remove from the dev hash */
 224         list_del(&hc->uuid_list);
 225         list_del(&hc->name_list);
 226         dm_set_mdptr(hc->md, NULL);
 227
 228         table = dm_get_table(hc->md);
 229         if (table) {
 230                 dm_table_event(table);
 231                 dm_table_put(table);
 232         }
 233
 234         if (hc->new_map)
 235                 dm_table_put(hc->new_map);
 236         dm_put(hc->md);
 237         free_cell(hc);
 238 }
 239
 240 static void dm_hash_remove_all(int keep_open_devices)
 241 {
 242         int i, dev_skipped, dev_removed;
 243         struct hash_cell *hc;
 244         struct list_head *tmp, *n;
 245
 246         down_write(&_hash_lock);
 247
 248 retry:
 249         dev_skipped = dev_removed = 0;
 250         for (i = 0; i < NUM_BUCKETS; i++) {
 251                 list_for_each_safe (tmp, n, _name_buckets + i) {
 252                         hc = list_entry(tmp, struct hash_cell, name_list);
 253
 254                         if (keep_open_devices &&
 255                             dm_lock_for_deletion(hc->md)) {
 256                                 dev_skipped++;
 257                                 continue;
 258                         }
 259                         __hash_remove(hc);
 260                         dev_removed = 1;
 261                 }
 262         }
 263
 264         /*
 265          * Some mapped devices may be using other mapped devices, so if any
 266          * still exist, repeat until we make no further progress.
 267          */
 268         if (dev_skipped) {
 269                 if (dev_removed)
 270                         goto retry;
 271
 272                 DMWARN("remove_all left %d open device(s)", dev_skipped);
 273         }
 274
 275         up_write(&_hash_lock);
 276 }
 277
 278 static int dm_hash_rename(const char *old, const char *new)
 279 {
 280         char *new_name, *old_name;
 281         struct hash_cell *hc;
 282         struct dm_table *table;
 283
 284         /*
 285          * duplicate new.
 286          */
 287         new_name = kstrdup(new, GFP_KERNEL);
 288         if (!new_name)
 289                 return -ENOMEM;
 290
 291         down_write(&_hash_lock);
 292
 293         /*
 294          * Is new free ?
 295          */
 296         hc = __get_name_cell(new);
 297         if (hc) {
 298                 DMWARN("asked to rename to an already existing name %s -> %s",
 299                        old, new);
 300                 dm_put(hc->md);
 301                 up_write(&_hash_lock);
 302                 kfree(new_name);
 303                 return -EBUSY;
 304         }
 305
 306         /*
 307          * Is there such a device as 'old' ?
 308          */
 309         hc = __get_name_cell(old);
 310         if (!hc) {
 311                 DMWARN("asked to rename a non existent device %s -> %s",
 312                        old, new);
 313                 up_write(&_hash_lock);
 314                 kfree(new_name);
 315                 return -ENXIO;
 316         }
 317
 318         /*
 319          * rename and move the name cell.
 320          */
 321         list_del(&hc->name_list);
 322         old_name = hc->name;
 323         hc->name = new_name;
 324         list_add(&hc->name_list, _name_buckets + hash_str(new_name));
 325
 326         /*
 327          * Wake up any dm event waiters.
 328          */
 329         table = dm_get_table(hc->md);
 330         if (table) {
 331                 dm_table_event(table);
 332                 dm_table_put(table);
 333         }
 334
 335         dm_put(hc->md);
 336         up_write(&_hash_lock);
 337         kfree(old_name);
 338         return 0;
 339 }
 340
 341 /*-----------------------------------------------------------------
 342  * Implementation of the ioctl commands
 343  *---------------------------------------------------------------*/
 344 /*
 345  * All the ioctl commands get dispatched to functions with this
 346  * prototype.
 347  */
 348 typedef int (*ioctl_fn)(struct dm_ioctl *param, size_t param_size);
 349
 350 static int remove_all(struct dm_ioctl *param, size_t param_size)
 351 {
 352         dm_hash_remove_all(1);
 353         param->data_size = 0;
 354         return 0;
 355 }
 356
 357 /*
 358  * Round up the ptr to an 8-byte boundary.
 359  */
 360 #define ALIGN_MASK 7
 361 static inline void *align_ptr(void *ptr)
 362 {
 363         return (void *) (((size_t) (ptr + ALIGN_MASK)) & ~ALIGN_MASK);
 364 }
 365
 366 /*
 367  * Retrieves the data payload buffer from an already allocated
 368  * struct dm_ioctl.
 369  */
 370 static void *get_result_buffer(struct dm_ioctl *param, size_t param_size,
 371                                size_t *len)
 372 {
 373         param->data_start = align_ptr(param + 1) - (void *) param;
 374
 375         if (param->data_start < param_size)
 376                 *len = param_size - param->data_start;
 377         else
 378                 *len = 0;
 379
 380         return ((void *) param) + param->data_start;
 381 }
 382
 383 static int list_devices(struct dm_ioctl *param, size_t param_size)
 384 {
 385         unsigned int i;
 386         struct hash_cell *hc;
 387         size_t len, needed = 0;
 388         struct gendisk *disk;
 389         struct dm_name_list *nl, *old_nl = NULL;
 390
 391         down_write(&_hash_lock);
 392
 393         /*
 394          * Loop through all the devices working out how much
 395          * space we need.
 396          */
 397         for (i = 0; i < NUM_BUCKETS; i++) {
 398                 list_for_each_entry (hc, _name_buckets + i, name_list) {
 399                         needed += sizeof(struct dm_name_list);
 400                         needed += strlen(hc->name) + 1;
 401                         needed += ALIGN_MASK;
 402                 }
 403         }
 404
 405         /*
 406          * Grab our output buffer.
 407          */
 408         nl = get_result_buffer(param, param_size, &len);
 409         if (len < needed) {
 410                 param->flags |= DM_BUFFER_FULL_FLAG;
 411                 goto out;
 412         }
 413         param->data_size = param->data_start + needed;
 414
 415         nl->dev = 0;    /* Flags no data */
 416
 417         /*
 418          * Now loop through filling out the names.
 419          */
 420         for (i = 0; i < NUM_BUCKETS; i++) {
 421                 list_for_each_entry (hc, _name_buckets + i, name_list) {
 422                         if (old_nl)
 423                                 old_nl->next = (uint32_t) ((void *) nl -
 424                                                            (void *) old_nl);
 425                         disk = dm_disk(hc->md);
 426                         nl->dev = huge_encode_dev(MKDEV(disk->major, disk->first_minor));
 427                         nl->next = 0;
 428                         strcpy(nl->name, hc->name);
 429
 430                         old_nl = nl;
 431                         nl = align_ptr(((void *) ++nl) + strlen(hc->name) + 1);
 432                 }
 433         }
 434
 435  out:
 436         up_write(&_hash_lock);
 437         return 0;
 438 }
 439
 440 static void list_version_get_needed(struct target_type *tt, void *needed_param)
 441 {
 442     size_t *needed = needed_param;
 443
 444     *needed += sizeof(struct dm_target_versions);
 445     *needed += strlen(tt->name);
 446     *needed += ALIGN_MASK;
 447 }
 448
 449 static void list_version_get_info(struct target_type *tt, void *param)
 450 {
 451     struct vers_iter *info = param;
 452
 453     /* Check space - it might have changed since the first iteration */
 454     if ((char *)info->vers + sizeof(tt->version) + strlen(tt->name) + 1 >
 455         info->end) {
 456
 457         info->flags = DM_BUFFER_FULL_FLAG;
 458         return;
 459     }
 460
 461     if (info->old_vers)
 462         info->old_vers->next = (uint32_t) ((void *)info->vers -
 463                                            (void *)info->old_vers);
 464     info->vers->version[0] = tt->version[0];
 465     info->vers->version[1] = tt->version[1];
 466     info->vers->version[2] = tt->version[2];
 467     info->vers->next = 0;
 468     strcpy(info->vers->name, tt->name);
 469
 470     info->old_vers = info->vers;
 471     info->vers = align_ptr(((void *) ++info->vers) + strlen(tt->name) + 1);
 472 }
 473
 474 static int list_versions(struct dm_ioctl *param, size_t param_size)
 475 {
 476         size_t len, needed = 0;
 477         struct dm_target_versions *vers;
 478         struct vers_iter iter_info;
 479
 480         /*
 481          * Loop through all the devices working out how much
 482          * space we need.
 483          */
 484         dm_target_iterate(list_version_get_needed, &needed);
 485
 486         /*
 487          * Grab our output buffer.
 488          */
 489         vers = get_result_buffer(param, param_size, &len);
 490         if (len < needed) {
 491                 param->flags |= DM_BUFFER_FULL_FLAG;
 492                 goto out;
 493         }
 494         param->data_size = param->data_start + needed;
 495
 496         iter_info.param_size = param_size;
 497         iter_info.old_vers = NULL;
 498         iter_info.vers = vers;
 499         iter_info.flags = 0;
 500         iter_info.end = (char *)vers+len;
 501
 502         /*
 503          * Now loop through filling out the names & versions.
 504          */
 505         dm_target_iterate(list_version_get_info, &iter_info);
 506         param->flags |= iter_info.flags;
 507
 508  out:
 509         return 0;
 510 }
 511
 512
 513
 514 static int check_name(const char *name)
 515 {
 516         if (strchr(name, '/')) {
 517                 DMWARN("invalid device name");
 518                 return -EINVAL;
 519         }
 520
 521         return 0;
 522 }
 523
 524 /*
 525  * Fills in a dm_ioctl structure, ready for sending back to
 526  * userland.
 527  */
 528 static int __dev_status(struct mapped_device *md, struct dm_ioctl *param)
 529 {
 530         struct gendisk *disk = dm_disk(md);
 531         struct dm_table *table;
 532
 533         param->flags &= ~(DM_SUSPEND_FLAG | DM_READONLY_FLAG |
 534                           DM_ACTIVE_PRESENT_FLAG);
 535
 536         if (dm_suspended(md))
 537                 param->flags |= DM_SUSPEND_FLAG;
 538
 539         param->dev = huge_encode_dev(MKDEV(disk->major, disk->first_minor));
 540
 541         /*
 542          * Yes, this will be out of date by the time it gets back
 543          * to userland, but it is still very useful for
 544          * debugging.
 545          */
 546         param->open_count = dm_open_count(md);
 547
 548         if (disk->policy)
 549                 param->flags |= DM_READONLY_FLAG;
 550
 551         param->event_nr = dm_get_event_nr(md);
 552
 553         table = dm_get_table(md);
 554         if (table) {
 555                 param->flags |= DM_ACTIVE_PRESENT_FLAG;
 556                 param->target_count = dm_table_get_num_targets(table);
 557                 dm_table_put(table);
 558         } else
 559                 param->target_count = 0;
 560
 561         return 0;
 562 }
 563
 564 static int dev_create(struct dm_ioctl *param, size_t param_size)
 565 {
 566         int r, m = DM_ANY_MINOR;
 567         struct mapped_device *md;
 568
 569         r = check_name(param->name);
 570         if (r)
 571                 return r;
 572
 573         if (param->flags & DM_PERSISTENT_DEV_FLAG)
 574                 m = MINOR(huge_decode_dev(param->dev));
 575
 576         r = dm_create(m, &md);
 577         if (r)
 578                 return r;
 579
 580         r = dm_hash_insert(param->name, *param->uuid ? param->uuid : NULL, md);
 581         if (r) {
 582                 dm_put(md);
 583                 return r;
 584         }
 585
 586         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
 587
 588         r = __dev_status(md, param);
 589         dm_put(md);
 590
 591         return r;
 592 }
 593
 594 /*
 595  * Always use UUID for lookups if it's present, otherwise use name or dev.
 596  */
 597 static struct hash_cell *__find_device_hash_cell(struct dm_ioctl *param)
 598 {
 599         struct mapped_device *md;
 600         void *mdptr = NULL;
 601
 602         if (*param->uuid)
 603                 return __get_uuid_cell(param->uuid);
 604
 605         if (*param->name)
 606                 return __get_name_cell(param->name);
 607
 608         md = dm_get_md(huge_decode_dev(param->dev));
 609         if (!md)
 610                 goto out;
 611
 612         mdptr = dm_get_mdptr(md);
 613         if (!mdptr)
 614                 dm_put(md);
 615
 616 out:
 617         return mdptr;
 618 }
 619
 620 static struct mapped_device *find_device(struct dm_ioctl *param)
 621 {
 622         struct hash_cell *hc;
 623         struct mapped_device *md = NULL;
 624
 625         down_read(&_hash_lock);
 626         hc = __find_device_hash_cell(param);
 627         if (hc) {
 628                 md = hc->md;
 629
 630                 /*
 631                  * Sneakily write in both the name and the uuid
 632                  * while we have the cell.
 633                  */
 634                 strncpy(param->name, hc->name, sizeof(param->name));
 635                 if (hc->uuid)
 636                         strncpy(param->uuid, hc->uuid, sizeof(param->uuid)-1);
 637                 else
 638                         param->uuid[0] = '\0';
 639
 640                 if (hc->new_map)
 641                         param->flags |= DM_INACTIVE_PRESENT_FLAG;
 642                 else
 643                         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
 644         }
 645         up_read(&_hash_lock);
 646
 647         return md;
 648 }
 649
 650 static int dev_remove(struct dm_ioctl *param, size_t param_size)
 651 {
 652         struct hash_cell *hc;
 653         struct mapped_device *md;
 654         int r;
 655
 656         down_write(&_hash_lock);
 657         hc = __find_device_hash_cell(param);
 658
 659         if (!hc) {
 660                 DMWARN("device doesn't appear to be in the dev hash table.");
 661                 up_write(&_hash_lock);
 662                 return -ENXIO;
 663         }
 664
 665         md = hc->md;
 666
 667         /*
 668          * Ensure the device is not open and nothing further can open it.
 669          */
 670         r = dm_lock_for_deletion(md);
 671         if (r) {
 672                 DMWARN("unable to remove open device %s", hc->name);
 673                 up_write(&_hash_lock);
 674                 dm_put(md);
 675                 return r;
 676         }
 677
 678         __hash_remove(hc);
 679         up_write(&_hash_lock);
 680         dm_put(md);
 681         param->data_size = 0;
 682         return 0;
 683 }
 684
 685 /*
 686  * Check a string doesn't overrun the chunk of
 687  * memory we copied from userland.
 688  */
 689 static int invalid_str(char *str, void *end)
 690 {
 691         while ((void *) str < end)
 692                 if (!*str++)
 693                         return 0;
 694
 695         return -EINVAL;
 696 }
 697
 698 static int dev_rename(struct dm_ioctl *param, size_t param_size)
 699 {
 700         int r;
 701         char *new_name = (char *) param + param->data_start;
 702
 703         if (new_name < (char *) (param + 1) ||
 704             invalid_str(new_name, (void *) param + param_size)) {
 705                 DMWARN("Invalid new logical volume name supplied.");
 706                 return -EINVAL;
 707         }
 708
 709         r = check_name(new_name);
 710         if (r)
 711                 return r;
 712
 713         param->data_size = 0;
 714         return dm_hash_rename(param->name, new_name);
 715 }
 716
 717 static int dev_set_geometry(struct dm_ioctl *param, size_t param_size)
 718 {
 719         int r = -EINVAL, x;
 720         struct mapped_device *md;
 721         struct hd_geometry geometry;
 722         unsigned long indata[4];
 723         char *geostr = (char *) param + param->data_start;
 724
 725         md = find_device(param);
 726         if (!md)
 727                 return -ENXIO;
 728
 729         if (geostr < (char *) (param + 1) ||
 730             invalid_str(geostr, (void *) param + param_size)) {
 731                 DMWARN("Invalid geometry supplied.");
 732                 goto out;
 733         }
 734
 735         x = sscanf(geostr, "%lu %lu %lu %lu", indata,
 736                    indata + 1, indata + 2, indata + 3);
 737
 738         if (x != 4) {
 739                 DMWARN("Unable to interpret geometry settings.");
 740                 goto out;
 741         }
 742
 743         if (indata[0] > 65535 || indata[1] > 255 ||
 744             indata[2] > 255 || indata[3] > ULONG_MAX) {
 745                 DMWARN("Geometry exceeds range limits.");
 746                 goto out;
 747         }
 748
 749         geometry.cylinders = indata[0];
 750         geometry.heads = indata[1];
 751         geometry.sectors = indata[2];
 752         geometry.start = indata[3];
 753
 754         r = dm_set_geometry(md, &geometry);
 755         if (!r)
 756                 r = __dev_status(md, param);
 757
 758         param->data_size = 0;
 759
 760 out:
 761         dm_put(md);
 762         return r;
 763 }
 764
 765 static int do_suspend(struct dm_ioctl *param)
 766 {
 767         int r = 0;
 768         unsigned suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
 769         struct mapped_device *md;
 770
 771         md = find_device(param);
 772         if (!md)
 773                 return -ENXIO;
 774
 775         if (param->flags & DM_SKIP_LOCKFS_FLAG)
 776                 suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
 777         if (param->flags & DM_NOFLUSH_FLAG)
 778                 suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
 779
 780         if (!dm_suspended(md))
 781                 r = dm_suspend(md, suspend_flags);
 782
 783         if (!r)
 784                 r = __dev_status(md, param);
 785
 786         dm_put(md);
 787         return r;
 788 }
 789
 790 static int do_resume(struct dm_ioctl *param)
 791 {
 792         int r = 0;
 793         unsigned suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
 794         struct hash_cell *hc;
 795         struct mapped_device *md;
 796         struct dm_table *new_map;
 797
 798         down_write(&_hash_lock);
 799
 800         hc = __find_device_hash_cell(param);
 801         if (!hc) {
 802                 DMWARN("device doesn't appear to be in the dev hash table.");
 803                 up_write(&_hash_lock);
 804                 return -ENXIO;
 805         }
 806
 807         md = hc->md;
 808
 809         new_map = hc->new_map;
 810         hc->new_map = NULL;
 811         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
 812
 813         up_write(&_hash_lock);
 814
 815         /* Do we need to load a new map ? */
 816         if (new_map) {
 817                 /* Suspend if it isn't already suspended */
 818                 if (param->flags & DM_SKIP_LOCKFS_FLAG)
 819                         suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
 820                 if (param->flags & DM_NOFLUSH_FLAG)
 821                         suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
 822                 if (!dm_suspended(md))
 823                         dm_suspend(md, suspend_flags);
 824
 825                 r = dm_swap_table(md, new_map);
 826                 if (r) {
 827                         dm_put(md);
 828                         dm_table_put(new_map);
 829                         return r;
 830                 }
 831
 832                 if (dm_table_get_mode(new_map) & FMODE_WRITE)
 833                         set_disk_ro(dm_disk(md), 0);
 834                 else
 835                         set_disk_ro(dm_disk(md), 1);
 836
 837                 dm_table_put(new_map);
 838         }
 839
 840         if (dm_suspended(md))
 841                 r = dm_resume(md);
 842
 843         if (!r)
 844                 r = __dev_status(md, param);
 845
 846         dm_put(md);
 847         return r;
 848 }
 849
 850 /*
 851  * Set or unset the suspension state of a device.
 852  * If the device already is in the requested state we just return its status.
 853  */
 854 static int dev_suspend(struct dm_ioctl *param, size_t param_size)
 855 {
 856         if (param->flags & DM_SUSPEND_FLAG)
 857                 return do_suspend(param);
 858
 859         return do_resume(param);
 860 }
 861
 862 /*
 863  * Copies device info back to user space, used by
 864  * the create and info ioctls.
 865  */
 866 static int dev_status(struct dm_ioctl *param, size_t param_size)
 867 {
 868         int r;
 869         struct mapped_device *md;
 870
 871         md = find_device(param);
 872         if (!md)
 873                 return -ENXIO;
 874
 875         r = __dev_status(md, param);
 876         dm_put(md);
 877         return r;
 878 }
 879
 880 /*
 881  * Build up the status struct for each target
 882  */
 883 static void retrieve_status(struct dm_table *table,
 884                             struct dm_ioctl *param, size_t param_size)
 885 {
 886         unsigned int i, num_targets;
 887         struct dm_target_spec *spec;
 888         char *outbuf, *outptr;
 889         status_type_t type;
 890         size_t remaining, len, used = 0;
 891
 892         outptr = outbuf = get_result_buffer(param, param_size, &len);
 893
 894         if (param->flags & DM_STATUS_TABLE_FLAG)
 895                 type = STATUSTYPE_TABLE;
 896         else
 897                 type = STATUSTYPE_INFO;
 898
 899         /* Get all the target info */
 900         num_targets = dm_table_get_num_targets(table);
 901         for (i = 0; i < num_targets; i++) {
 902                 struct dm_target *ti = dm_table_get_target(table, i);
 903
 904                 remaining = len - (outptr - outbuf);
 905                 if (remaining <= sizeof(struct dm_target_spec)) {
 906                         param->flags |= DM_BUFFER_FULL_FLAG;
 907                         break;
 908                 }
 909
 910                 spec = (struct dm_target_spec *) outptr;
 911
 912                 spec->status = 0;
 913                 spec->sector_start = ti->begin;
 914                 spec->length = ti->len;
 915                 strncpy(spec->target_type, ti->type->name,
 916                         sizeof(spec->target_type));
 917
 918                 outptr += sizeof(struct dm_target_spec);
 919                 remaining = len - (outptr - outbuf);
 920                 if (remaining <= 0) {
 921                         param->flags |= DM_BUFFER_FULL_FLAG;
 922                         break;
 923                 }
 924
 925                 /* Get the status/table string from the target driver */
 926                 if (ti->type->status) {
 927                         if (ti->type->status(ti, type, outptr, remaining)) {
 928                                 param->flags |= DM_BUFFER_FULL_FLAG;
 929                                 break;
 930                         }
 931                 } else
 932                         outptr[0] = '\0';
 933
 934                 outptr += strlen(outptr) + 1;
 935                 used = param->data_start + (outptr - outbuf);
 936
 937                 outptr = align_ptr(outptr);
 938                 spec->next = outptr - outbuf;
 939         }
 940
 941         if (used)
 942                 param->data_size = used;
 943
 944         param->target_count = num_targets;
 945 }
 946
 947 /*
 948  * Wait for a device to report an event
 949  */
 950 static int dev_wait(struct dm_ioctl *param, size_t param_size)
 951 {
 952         int r;
 953         struct mapped_device *md;
 954         struct dm_table *table;
 955
 956         md = find_device(param);
 957         if (!md)
 958                 return -ENXIO;
 959
 960         /*
 961          * Wait for a notification event
 962          */
 963         if (dm_wait_event(md, param->event_nr)) {
 964                 r = -ERESTARTSYS;
 965                 goto out;
 966         }
 967
 968         /*
 969          * The userland program is going to want to know what
 970          * changed to trigger the event, so we may as well tell
 971          * him and save an ioctl.
 972          */
 973         r = __dev_status(md, param);
 974         if (r)
 975                 goto out;
 976
 977         table = dm_get_table(md);
 978         if (table) {
 979                 retrieve_status(table, param, param_size);
 980                 dm_table_put(table);
 981         }
 982
 983  out:
 984         dm_put(md);
 985         return r;
 986 }
 987
 988 static inline int get_mode(struct dm_ioctl *param)
 989 {
 990         int mode = FMODE_READ | FMODE_WRITE;
 991
 992         if (param->flags & DM_READONLY_FLAG)
 993                 mode = FMODE_READ;
 994
 995         return mode;
 996 }
 997
 998 static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
 999                        struct dm_target_spec **spec, char **target_params)
1000 {
1001         *spec = (struct dm_target_spec *) ((unsigned char *) last + next);
1002         *target_params = (char *) (*spec + 1);
1003
1004         if (*spec < (last + 1))
1005                 return -EINVAL;
1006
1007         return invalid_str(*target_params, end);
1008 }
1009
1010 static int populate_table(struct dm_table *table,
1011                           struct dm_ioctl *param, size_t param_size)
1012 {
1013         int r;
1014         unsigned int i = 0;
1015         struct dm_target_spec *spec = (struct dm_target_spec *) param;
1016         uint32_t next = param->data_start;
1017         void *end = (void *) param + param_size;
1018         char *target_params;
1019
1020         if (!param->target_count) {
1021                 DMWARN("populate_table: no targets specified");
1022                 return -EINVAL;
1023         }
1024
1025         for (i = 0; i < param->target_count; i++) {
1026
1027                 r = next_target(spec, next, end, &spec, &target_params);
1028                 if (r) {
1029                         DMWARN("unable to find target");
1030                         return r;
1031                 }
1032
1033                 r = dm_table_add_target(table, spec->target_type,
1034                                         (sector_t) spec->sector_start,
1035                                         (sector_t) spec->length,
1036                                         target_params);
1037                 if (r) {
1038                         DMWARN("error adding target to table");
1039                         return r;
1040                 }
1041
1042                 next = spec->next;
1043         }
1044
1045         return dm_table_complete(table);
1046 }
1047
1048 static int table_load(struct dm_ioctl *param, size_t param_size)
1049 {
1050         int r;
1051         struct hash_cell *hc;
1052         struct dm_table *t;
1053         struct mapped_device *md;
1054
1055         md = find_device(param);
1056         if (!md)
1057                 return -ENXIO;
1058
1059         r = dm_table_create(&t, get_mode(param), param->target_count, md);
1060         if (r)
1061                 goto out;
1062
1063         r = populate_table(t, param, param_size);
1064         if (r) {
1065                 dm_table_put(t);
1066                 goto out;
1067         }
1068
1069         down_write(&_hash_lock);
1070         hc = dm_get_mdptr(md);
1071         if (!hc || hc->md != md) {
1072                 DMWARN("device has been removed from the dev hash table.");
1073                 dm_table_put(t);
1074                 up_write(&_hash_lock);
1075                 r = -ENXIO;
1076                 goto out;
1077         }
1078
1079         if (hc->new_map)
1080                 dm_table_put(hc->new_map);
1081         hc->new_map = t;
1082         up_write(&_hash_lock);
1083
1084         param->flags |= DM_INACTIVE_PRESENT_FLAG;
1085         r = __dev_status(md, param);
1086
1087 out:
1088         dm_put(md);
1089
1090         return r;
1091 }
1092
1093 static int table_clear(struct dm_ioctl *param, size_t param_size)
1094 {
1095         int r;
1096         struct hash_cell *hc;
1097         struct mapped_device *md;
1098
1099         down_write(&_hash_lock);
1100
1101         hc = __find_device_hash_cell(param);
1102         if (!hc) {
1103                 DMWARN("device doesn't appear to be in the dev hash table.");
1104                 up_write(&_hash_lock);
1105                 return -ENXIO;
1106         }
1107
1108         if (hc->new_map) {
1109                 dm_table_put(hc->new_map);
1110                 hc->new_map = NULL;
1111         }
1112
1113         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1114
1115         r = __dev_status(hc->md, param);
1116         md = hc->md;
1117         up_write(&_hash_lock);
1118         dm_put(md);
1119         return r;
1120 }
1121
1122 /*
1123  * Retrieves a list of devices used by a particular dm device.
1124  */
1125 static void retrieve_deps(struct dm_table *table,
1126                           struct dm_ioctl *param, size_t param_size)
1127 {
1128         unsigned int count = 0;
1129         struct list_head *tmp;
1130         size_t len, needed;
1131         struct dm_dev *dd;
1132         struct dm_target_deps *deps;
1133
1134         deps = get_result_buffer(param, param_size, &len);
1135
1136         /*
1137          * Count the devices.
1138          */
1139         list_for_each (tmp, dm_table_get_devices(table))
1140                 count++;
1141
1142         /*
1143          * Check we have enough space.
1144          */
1145         needed = sizeof(*deps) + (sizeof(*deps->dev) * count);
1146         if (len < needed) {
1147                 param->flags |= DM_BUFFER_FULL_FLAG;
1148                 return;
1149         }
1150
1151         /*
1152          * Fill in the devices.
1153          */
1154         deps->count = count;
1155         count = 0;
1156         list_for_each_entry (dd, dm_table_get_devices(table), list)
1157                 deps->dev[count++] = huge_encode_dev(dd->bdev->bd_dev);
1158
1159         param->data_size = param->data_start + needed;
1160 }
1161
1162 static int table_deps(struct dm_ioctl *param, size_t param_size)
1163 {
1164         int r = 0;
1165         struct mapped_device *md;
1166         struct dm_table *table;
1167
1168         md = find_device(param);
1169         if (!md)
1170                 return -ENXIO;
1171
1172         r = __dev_status(md, param);
1173         if (r)
1174                 goto out;
1175
1176         table = dm_get_table(md);
1177         if (table) {
1178                 retrieve_deps(table, param, param_size);
1179                 dm_table_put(table);
1180         }
1181
1182  out:
1183         dm_put(md);
1184         return r;
1185 }
1186
1187 /*
1188  * Return the status of a device as a text string for each
1189  * target.
1190  */
1191 static int table_status(struct dm_ioctl *param, size_t param_size)
1192 {
1193         int r;
1194         struct mapped_device *md;
1195         struct dm_table *table;
1196
1197         md = find_device(param);
1198         if (!md)
1199                 return -ENXIO;
1200
1201         r = __dev_status(md, param);
1202         if (r)
1203                 goto out;
1204
1205         table = dm_get_table(md);
1206         if (table) {
1207                 retrieve_status(table, param, param_size);
1208                 dm_table_put(table);
1209         }
1210
1211  out:
1212         dm_put(md);
1213         return r;
1214 }
1215
1216 /*
1217  * Pass a message to the target that's at the supplied device offset.
1218  */
1219 static int target_message(struct dm_ioctl *param, size_t param_size)
1220 {
1221         int r, argc;
1222         char **argv;
1223         struct mapped_device *md;
1224         struct dm_table *table;
1225         struct dm_target *ti;
1226         struct dm_target_msg *tmsg = (void *) param + param->data_start;
1227
1228         md = find_device(param);
1229         if (!md)
1230                 return -ENXIO;
1231
1232         r = __dev_status(md, param);
1233         if (r)
1234                 goto out;
1235
1236         if (tmsg < (struct dm_target_msg *) (param + 1) ||
1237             invalid_str(tmsg->message, (void *) param + param_size)) {
1238                 DMWARN("Invalid target message parameters.");
1239                 r = -EINVAL;
1240                 goto out;
1241         }
1242
1243         r = dm_split_args(&argc, &argv, tmsg->message);
1244         if (r) {
1245                 DMWARN("Failed to split target message parameters");
1246                 goto out;
1247         }
1248
1249         table = dm_get_table(md);
1250         if (!table)
1251                 goto out_argv;
1252
1253         if (tmsg->sector >= dm_table_get_size(table)) {
1254                 DMWARN("Target message sector outside device.");
1255                 r = -EINVAL;
1256                 goto out_table;
1257         }
1258
1259         ti = dm_table_find_target(table, tmsg->sector);
1260         if (ti->type->message)
1261                 r = ti->type->message(ti, argc, argv);
1262         else {
1263                 DMWARN("Target type does not support messages");
1264                 r = -EINVAL;
1265         }
1266
1267  out_table:
1268         dm_table_put(table);
1269  out_argv:
1270         kfree(argv);
1271  out:
1272         param->data_size = 0;
1273         dm_put(md);
1274         return r;
1275 }
1276
1277 /*-----------------------------------------------------------------
1278  * Implementation of open/close/ioctl on the special char
1279  * device.
1280  *---------------------------------------------------------------*/
1281 static ioctl_fn lookup_ioctl(unsigned int cmd)
1282 {
1283         static struct {
1284                 int cmd;
1285                 ioctl_fn fn;
1286         } _ioctls[] = {
1287                 {DM_VERSION_CMD, NULL}, /* version is dealt with elsewhere */
1288                 {DM_REMOVE_ALL_CMD, remove_all},
1289                 {DM_LIST_DEVICES_CMD, list_devices},
1290
1291                 {DM_DEV_CREATE_CMD, dev_create},
1292                 {DM_DEV_REMOVE_CMD, dev_remove},
1293                 {DM_DEV_RENAME_CMD, dev_rename},
1294                 {DM_DEV_SUSPEND_CMD, dev_suspend},
1295                 {DM_DEV_STATUS_CMD, dev_status},
1296                 {DM_DEV_WAIT_CMD, dev_wait},
1297
1298                 {DM_TABLE_LOAD_CMD, table_load},
1299                 {DM_TABLE_CLEAR_CMD, table_clear},
1300                 {DM_TABLE_DEPS_CMD, table_deps},
1301                 {DM_TABLE_STATUS_CMD, table_status},
1302
1303                 {DM_LIST_VERSIONS_CMD, list_versions},
1304
1305                 {DM_TARGET_MSG_CMD, target_message},
1306                 {DM_DEV_SET_GEOMETRY_CMD, dev_set_geometry}
1307         };
1308
1309         return (cmd >= ARRAY_SIZE(_ioctls)) ? NULL : _ioctls[cmd].fn;
1310 }
1311
1312 /*
1313  * As well as checking the version compatibility this always
1314  * copies the kernel interface version out.
1315  */
1316 static int check_version(unsigned int cmd, struct dm_ioctl __user *user)
1317 {
1318         uint32_t version[3];
1319         int r = 0;
1320
1321         if (copy_from_user(version, user->version, sizeof(version)))
1322                 return -EFAULT;
1323
1324         if ((DM_VERSION_MAJOR != version[0]) ||
1325             (DM_VERSION_MINOR < version[1])) {
1326                 DMWARN("ioctl interface mismatch: "
1327                        "kernel(%u.%u.%u), user(%u.%u.%u), cmd(%d)",
1328                        DM_VERSION_MAJOR, DM_VERSION_MINOR,
1329                        DM_VERSION_PATCHLEVEL,
1330                        version[0], version[1], version[2], cmd);
1331                 r = -EINVAL;
1332         }
1333
1334         /*
1335          * Fill in the kernel version.
1336          */
1337         version[0] = DM_VERSION_MAJOR;
1338         version[1] = DM_VERSION_MINOR;
1339         version[2] = DM_VERSION_PATCHLEVEL;
1340         if (copy_to_user(user->version, version, sizeof(version)))
1341                 return -EFAULT;
1342
1343         return r;
1344 }
1345
1346 static void free_params(struct dm_ioctl *param)
1347 {
1348         vfree(param);
1349 }
1350
1351 static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl **param)
1352 {
1353         struct dm_ioctl tmp, *dmi;
1354
1355         if (copy_from_user(&tmp, user, sizeof(tmp)))
1356                 return -EFAULT;
1357
1358         if (tmp.data_size < sizeof(tmp))
1359                 return -EINVAL;
1360
1361         dmi = (struct dm_ioctl *) vmalloc(tmp.data_size);
1362         if (!dmi)
1363                 return -ENOMEM;
1364
1365         if (copy_from_user(dmi, user, tmp.data_size)) {
1366                 vfree(dmi);
1367                 return -EFAULT;
1368         }
1369
1370         *param = dmi;
1371         return 0;
1372 }
1373
1374 static int validate_params(uint cmd, struct dm_ioctl *param)
1375 {
1376         /* Always clear this flag */
1377         param->flags &= ~DM_BUFFER_FULL_FLAG;
1378
1379         /* Ignores parameters */
1380         if (cmd == DM_REMOVE_ALL_CMD ||
1381             cmd == DM_LIST_DEVICES_CMD ||
1382             cmd == DM_LIST_VERSIONS_CMD)
1383                 return 0;
1384
1385         if ((cmd == DM_DEV_CREATE_CMD)) {
1386                 if (!*param->name) {
1387                         DMWARN("name not supplied when creating device");
1388                         return -EINVAL;
1389                 }
1390         } else if ((*param->uuid && *param->name)) {
1391                 DMWARN("only supply one of name or uuid, cmd(%u)", cmd);
1392                 return -EINVAL;
1393         }
1394
1395         /* Ensure strings are terminated */
1396         param->name[DM_NAME_LEN - 1] = '\0';
1397         param->uuid[DM_UUID_LEN - 1] = '\0';
1398
1399         return 0;
1400 }
1401
1402 static int ctl_ioctl(struct inode *inode, struct file *file,
1403                      uint command, ulong u)
1404 {
1405         int r = 0;
1406         unsigned int cmd;
1407         struct dm_ioctl *param;
1408         struct dm_ioctl __user *user = (struct dm_ioctl __user *) u;
1409         ioctl_fn fn = NULL;
1410         size_t param_size;
1411
1412         /* only root can play with this */
1413         if (!capable(CAP_SYS_ADMIN))
1414                 return -EACCES;
1415
1416         if (_IOC_TYPE(command) != DM_IOCTL)
1417                 return -ENOTTY;
1418
1419         cmd = _IOC_NR(command);
1420
1421         /*
1422          * Check the interface version passed in.  This also
1423          * writes out the kernel's interface version.
1424          */
1425         r = check_version(cmd, user);
1426         if (r)
1427                 return r;
1428
1429         /*
1430          * Nothing more to do for the version command.
1431          */
1432         if (cmd == DM_VERSION_CMD)
1433                 return 0;
1434
1435         fn = lookup_ioctl(cmd);
1436         if (!fn) {
1437                 DMWARN("dm_ctl_ioctl: unknown command 0x%x", command);
1438                 return -ENOTTY;
1439         }
1440
1441         /*
1442          * Trying to avoid low memory issues when a device is
1443          * suspended.
1444          */
1445         current->flags |= PF_MEMALLOC;
1446
1447         /*
1448          * Copy the parameters into kernel space.
1449          */
1450         r = copy_params(user, &param);
1451
1452         current->flags &= ~PF_MEMALLOC;
1453
1454         if (r)
1455                 return r;
1456
1457         r = validate_params(cmd, param);
1458         if (r)
1459                 goto out;
1460
1461         param_size = param->data_size;
1462         param->data_size = sizeof(*param);
1463         r = fn(param, param_size);
1464
1465         /*
1466          * Copy the results back to userland.
1467          */
1468         if (!r && copy_to_user(user, param, param->data_size))
1469                 r = -EFAULT;
1470
1471  out:
1472         free_params(param);
1473         return r;
1474 }
1475
1476 static const struct file_operations _ctl_fops = {
1477         .ioctl   = ctl_ioctl,
1478         .owner   = THIS_MODULE,
1479 };
1480
1481 static struct miscdevice _dm_misc = {
1482         .minor          = MISC_DYNAMIC_MINOR,
1483         .name           = DM_NAME,
1484         .fops           = &_ctl_fops
1485 };
1486
1487 /*
1488  * Create misc character device and link to DM_DIR/control.
1489  */
1490 int __init dm_interface_init(void)
1491 {
1492         int r;
1493
1494         r = dm_hash_init();
1495         if (r)
1496                 return r;
1497
1498         r = misc_register(&_dm_misc);
1499         if (r) {
1500                 DMERR("misc_register failed for control device");
1501                 dm_hash_exit();
1502                 return r;
1503         }
1504
1505         DMINFO("%d.%d.%d%s initialised: %s", DM_VERSION_MAJOR,
1506                DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL, DM_VERSION_EXTRA,
1507                DM_DRIVER_EMAIL);
1508         return 0;
1509 }
1510
1511 void dm_interface_exit(void)
1512 {
1513         if (misc_deregister(&_dm_misc) < 0)
1514                 DMERR("misc_deregister failed for control device");
1515
1516         dm_hash_exit();
1517 }