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/devfs_fs_kernel.h>
  17 #include <linux/dm-ioctl.h>
  18 #include <linux/hdreg.h>
  19
  20 #include <asm/uaccess.h>
  21
  22 #define DM_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         devfs_mk_dir(DM_DIR);
  72         return 0;
  73 }
  74
  75 static void dm_hash_exit(void)
  76 {
  77         dm_hash_remove_all(0);
  78         devfs_remove(DM_DIR);
  79 }
  80
  81 /*-----------------------------------------------------------------
  82  * Hash function:
  83  * We're not really concerned with the str hash function being
  84  * fast since it's only used by the ioctl interface.
  85  *---------------------------------------------------------------*/
  86 static unsigned int hash_str(const char *str)
  87 {
  88         const unsigned int hash_mult = 2654435387U;
  89         unsigned int h = 0;
  90
  91         while (*str)
  92                 h = (h + (unsigned int) *str++) * hash_mult;
  93
  94         return h & MASK_BUCKETS;
  95 }
  96
  97 /*-----------------------------------------------------------------
  98  * Code for looking up a device by name
  99  *---------------------------------------------------------------*/
 100 static struct hash_cell *__get_name_cell(const char *str)
 101 {
 102         struct hash_cell *hc;
 103         unsigned int h = hash_str(str);
 104
 105         list_for_each_entry (hc, _name_buckets + h, name_list)
 106                 if (!strcmp(hc->name, str)) {
 107                         dm_get(hc->md);
 108                         return hc;
 109                 }
 110
 111         return NULL;
 112 }
 113
 114 static struct hash_cell *__get_uuid_cell(const char *str)
 115 {
 116         struct hash_cell *hc;
 117         unsigned int h = hash_str(str);
 118
 119         list_for_each_entry (hc, _uuid_buckets + h, uuid_list)
 120                 if (!strcmp(hc->uuid, str)) {
 121                         dm_get(hc->md);
 122                         return hc;
 123                 }
 124
 125         return NULL;
 126 }
 127
 128 /*-----------------------------------------------------------------
 129  * Inserting, removing and renaming a device.
 130  *---------------------------------------------------------------*/
 131 static struct hash_cell *alloc_cell(const char *name, const char *uuid,
 132                                     struct mapped_device *md)
 133 {
 134         struct hash_cell *hc;
 135
 136         hc = kmalloc(sizeof(*hc), GFP_KERNEL);
 137         if (!hc)
 138                 return NULL;
 139
 140         hc->name = kstrdup(name, GFP_KERNEL);
 141         if (!hc->name) {
 142                 kfree(hc);
 143                 return NULL;
 144         }
 145
 146         if (!uuid)
 147                 hc->uuid = NULL;
 148
 149         else {
 150                 hc->uuid = kstrdup(uuid, GFP_KERNEL);
 151                 if (!hc->uuid) {
 152                         kfree(hc->name);
 153                         kfree(hc);
 154                         return NULL;
 155                 }
 156         }
 157
 158         INIT_LIST_HEAD(&hc->name_list);
 159         INIT_LIST_HEAD(&hc->uuid_list);
 160         hc->md = md;
 161         hc->new_map = NULL;
 162         return hc;
 163 }
 164
 165 static void free_cell(struct hash_cell *hc)
 166 {
 167         if (hc) {
 168                 kfree(hc->name);
 169                 kfree(hc->uuid);
 170                 kfree(hc);
 171         }
 172 }
 173
 174 /*
 175  * devfs stuff.
 176  */
 177 static int register_with_devfs(struct hash_cell *hc)
 178 {
 179         struct gendisk *disk = dm_disk(hc->md);
 180
 181         devfs_mk_bdev(MKDEV(disk->major, disk->first_minor),
 182                       S_IFBLK | S_IRUSR | S_IWUSR | S_IRGRP,
 183                       DM_DIR "/%s", hc->name);
 184         return 0;
 185 }
 186
 187 static int unregister_with_devfs(struct hash_cell *hc)
 188 {
 189         devfs_remove(DM_DIR"/%s", hc->name);
 190         return 0;
 191 }
 192
 193 /*
 194  * The kdev_t and uuid of a device can never change once it is
 195  * initially inserted.
 196  */
 197 static int dm_hash_insert(const char *name, const char *uuid, struct mapped_device *md)
 198 {
 199         struct hash_cell *cell, *hc;
 200
 201         /*
 202          * Allocate the new cells.
 203          */
 204         cell = alloc_cell(name, uuid, md);
 205         if (!cell)
 206                 return -ENOMEM;
 207
 208         /*
 209          * Insert the cell into both hash tables.
 210          */
 211         down_write(&_hash_lock);
 212         hc = __get_name_cell(name);
 213         if (hc) {
 214                 dm_put(hc->md);
 215                 goto bad;
 216         }
 217
 218         list_add(&cell->name_list, _name_buckets + hash_str(name));
 219
 220         if (uuid) {
 221                 hc = __get_uuid_cell(uuid);
 222                 if (hc) {
 223                         list_del(&cell->name_list);
 224                         dm_put(hc->md);
 225                         goto bad;
 226                 }
 227                 list_add(&cell->uuid_list, _uuid_buckets + hash_str(uuid));
 228         }
 229         register_with_devfs(cell);
 230         dm_get(md);
 231         dm_set_mdptr(md, cell);
 232         up_write(&_hash_lock);
 233
 234         return 0;
 235
 236  bad:
 237         up_write(&_hash_lock);
 238         free_cell(cell);
 239         return -EBUSY;
 240 }
 241
 242 static void __hash_remove(struct hash_cell *hc)
 243 {
 244         struct dm_table *table;
 245
 246         /* remove from the dev hash */
 247         list_del(&hc->uuid_list);
 248         list_del(&hc->name_list);
 249         unregister_with_devfs(hc);
 250         dm_set_mdptr(hc->md, NULL);
 251
 252         table = dm_get_table(hc->md);
 253         if (table) {
 254                 dm_table_event(table);
 255                 dm_table_put(table);
 256         }
 257
 258         if (hc->new_map)
 259                 dm_table_put(hc->new_map);
 260         dm_put(hc->md);
 261         free_cell(hc);
 262 }
 263
 264 static void dm_hash_remove_all(int keep_open_devices)
 265 {
 266         int i, dev_skipped, dev_removed;
 267         struct hash_cell *hc;
 268         struct list_head *tmp, *n;
 269
 270         down_write(&_hash_lock);
 271
 272 retry:
 273         dev_skipped = dev_removed = 0;
 274         for (i = 0; i < NUM_BUCKETS; i++) {
 275                 list_for_each_safe (tmp, n, _name_buckets + i) {
 276                         hc = list_entry(tmp, struct hash_cell, name_list);
 277
 278                         if (keep_open_devices &&
 279                             dm_lock_for_deletion(hc->md)) {
 280                                 dev_skipped++;
 281                                 continue;
 282                         }
 283                         __hash_remove(hc);
 284                         dev_removed = 1;
 285                 }
 286         }
 287
 288         /*
 289          * Some mapped devices may be using other mapped devices, so if any
 290          * still exist, repeat until we make no further progress.
 291          */
 292         if (dev_skipped) {
 293                 if (dev_removed)
 294                         goto retry;
 295
 296                 DMWARN("remove_all left %d open device(s)", dev_skipped);
 297         }
 298
 299         up_write(&_hash_lock);
 300 }
 301
 302 static int dm_hash_rename(const char *old, const char *new)
 303 {
 304         char *new_name, *old_name;
 305         struct hash_cell *hc;
 306         struct dm_table *table;
 307
 308         /*
 309          * duplicate new.
 310          */
 311         new_name = kstrdup(new, GFP_KERNEL);
 312         if (!new_name)
 313                 return -ENOMEM;
 314
 315         down_write(&_hash_lock);
 316
 317         /*
 318          * Is new free ?
 319          */
 320         hc = __get_name_cell(new);
 321         if (hc) {
 322                 DMWARN("asked to rename to an already existing name %s -> %s",
 323                        old, new);
 324                 dm_put(hc->md);
 325                 up_write(&_hash_lock);
 326                 kfree(new_name);
 327                 return -EBUSY;
 328         }
 329
 330         /*
 331          * Is there such a device as 'old' ?
 332          */
 333         hc = __get_name_cell(old);
 334         if (!hc) {
 335                 DMWARN("asked to rename a non existent device %s -> %s",
 336                        old, new);
 337                 up_write(&_hash_lock);
 338                 kfree(new_name);
 339                 return -ENXIO;
 340         }
 341
 342         /*
 343          * rename and move the name cell.
 344          */
 345         unregister_with_devfs(hc);
 346
 347         list_del(&hc->name_list);
 348         old_name = hc->name;
 349         hc->name = new_name;
 350         list_add(&hc->name_list, _name_buckets + hash_str(new_name));
 351
 352         /* rename the device node in devfs */
 353         register_with_devfs(hc);
 354
 355         /*
 356          * Wake up any dm event waiters.
 357          */
 358         table = dm_get_table(hc->md);
 359         if (table) {
 360                 dm_table_event(table);
 361                 dm_table_put(table);
 362         }
 363
 364         dm_put(hc->md);
 365         up_write(&_hash_lock);
 366         kfree(old_name);
 367         return 0;
 368 }
 369
 370 /*-----------------------------------------------------------------
 371  * Implementation of the ioctl commands
 372  *---------------------------------------------------------------*/
 373 /*
 374  * All the ioctl commands get dispatched to functions with this
 375  * prototype.
 376  */
 377 typedef int (*ioctl_fn)(struct dm_ioctl *param, size_t param_size);
 378
 379 static int remove_all(struct dm_ioctl *param, size_t param_size)
 380 {
 381         dm_hash_remove_all(1);
 382         param->data_size = 0;
 383         return 0;
 384 }
 385
 386 /*
 387  * Round up the ptr to an 8-byte boundary.
 388  */
 389 #define ALIGN_MASK 7
 390 static inline void *align_ptr(void *ptr)
 391 {
 392         return (void *) (((size_t) (ptr + ALIGN_MASK)) & ~ALIGN_MASK);
 393 }
 394
 395 /*
 396  * Retrieves the data payload buffer from an already allocated
 397  * struct dm_ioctl.
 398  */
 399 static void *get_result_buffer(struct dm_ioctl *param, size_t param_size,
 400                                size_t *len)
 401 {
 402         param->data_start = align_ptr(param + 1) - (void *) param;
 403
 404         if (param->data_start < param_size)
 405                 *len = param_size - param->data_start;
 406         else
 407                 *len = 0;
 408
 409         return ((void *) param) + param->data_start;
 410 }
 411
 412 static int list_devices(struct dm_ioctl *param, size_t param_size)
 413 {
 414         unsigned int i;
 415         struct hash_cell *hc;
 416         size_t len, needed = 0;
 417         struct gendisk *disk;
 418         struct dm_name_list *nl, *old_nl = NULL;
 419
 420         down_write(&_hash_lock);
 421
 422         /*
 423          * Loop through all the devices working out how much
 424          * space we need.
 425          */
 426         for (i = 0; i < NUM_BUCKETS; i++) {
 427                 list_for_each_entry (hc, _name_buckets + i, name_list) {
 428                         needed += sizeof(struct dm_name_list);
 429                         needed += strlen(hc->name) + 1;
 430                         needed += ALIGN_MASK;
 431                 }
 432         }
 433
 434         /*
 435          * Grab our output buffer.
 436          */
 437         nl = get_result_buffer(param, param_size, &len);
 438         if (len < needed) {
 439                 param->flags |= DM_BUFFER_FULL_FLAG;
 440                 goto out;
 441         }
 442         param->data_size = param->data_start + needed;
 443
 444         nl->dev = 0;    /* Flags no data */
 445
 446         /*
 447          * Now loop through filling out the names.
 448          */
 449         for (i = 0; i < NUM_BUCKETS; i++) {
 450                 list_for_each_entry (hc, _name_buckets + i, name_list) {
 451                         if (old_nl)
 452                                 old_nl->next = (uint32_t) ((void *) nl -
 453                                                            (void *) old_nl);
 454                         disk = dm_disk(hc->md);
 455                         nl->dev = huge_encode_dev(MKDEV(disk->major, disk->first_minor));
 456                         nl->next = 0;
 457                         strcpy(nl->name, hc->name);
 458
 459                         old_nl = nl;
 460                         nl = align_ptr(((void *) ++nl) + strlen(hc->name) + 1);
 461                 }
 462         }
 463
 464  out:
 465         up_write(&_hash_lock);
 466         return 0;
 467 }
 468
 469 static void list_version_get_needed(struct target_type *tt, void *needed_param)
 470 {
 471     size_t *needed = needed_param;
 472
 473     *needed += sizeof(struct dm_target_versions);
 474     *needed += strlen(tt->name);
 475     *needed += ALIGN_MASK;
 476 }
 477
 478 static void list_version_get_info(struct target_type *tt, void *param)
 479 {
 480     struct vers_iter *info = param;
 481
 482     /* Check space - it might have changed since the first iteration */
 483     if ((char *)info->vers + sizeof(tt->version) + strlen(tt->name) + 1 >
 484         info->end) {
 485
 486         info->flags = DM_BUFFER_FULL_FLAG;
 487         return;
 488     }
 489
 490     if (info->old_vers)
 491         info->old_vers->next = (uint32_t) ((void *)info->vers -
 492                                            (void *)info->old_vers);
 493     info->vers->version[0] = tt->version[0];
 494     info->vers->version[1] = tt->version[1];
 495     info->vers->version[2] = tt->version[2];
 496     info->vers->next = 0;
 497     strcpy(info->vers->name, tt->name);
 498
 499     info->old_vers = info->vers;
 500     info->vers = align_ptr(((void *) ++info->vers) + strlen(tt->name) + 1);
 501 }
 502
 503 static int list_versions(struct dm_ioctl *param, size_t param_size)
 504 {
 505         size_t len, needed = 0;
 506         struct dm_target_versions *vers;
 507         struct vers_iter iter_info;
 508
 509         /*
 510          * Loop through all the devices working out how much
 511          * space we need.
 512          */
 513         dm_target_iterate(list_version_get_needed, &needed);
 514
 515         /*
 516          * Grab our output buffer.
 517          */
 518         vers = get_result_buffer(param, param_size, &len);
 519         if (len < needed) {
 520                 param->flags |= DM_BUFFER_FULL_FLAG;
 521                 goto out;
 522         }
 523         param->data_size = param->data_start + needed;
 524
 525         iter_info.param_size = param_size;
 526         iter_info.old_vers = NULL;
 527         iter_info.vers = vers;
 528         iter_info.flags = 0;
 529         iter_info.end = (char *)vers+len;
 530
 531         /*
 532          * Now loop through filling out the names & versions.
 533          */
 534         dm_target_iterate(list_version_get_info, &iter_info);
 535         param->flags |= iter_info.flags;
 536
 537  out:
 538         return 0;
 539 }
 540
 541
 542
 543 static int check_name(const char *name)
 544 {
 545         if (strchr(name, '/')) {
 546                 DMWARN("invalid device name");
 547                 return -EINVAL;
 548         }
 549
 550         return 0;
 551 }
 552
 553 /*
 554  * Fills in a dm_ioctl structure, ready for sending back to
 555  * userland.
 556  */
 557 static int __dev_status(struct mapped_device *md, struct dm_ioctl *param)
 558 {
 559         struct gendisk *disk = dm_disk(md);
 560         struct dm_table *table;
 561
 562         param->flags &= ~(DM_SUSPEND_FLAG | DM_READONLY_FLAG |
 563                           DM_ACTIVE_PRESENT_FLAG);
 564
 565         if (dm_suspended(md))
 566                 param->flags |= DM_SUSPEND_FLAG;
 567
 568         param->dev = huge_encode_dev(MKDEV(disk->major, disk->first_minor));
 569
 570         /*
 571          * Yes, this will be out of date by the time it gets back
 572          * to userland, but it is still very useful for
 573          * debugging.
 574          */
 575         param->open_count = dm_open_count(md);
 576
 577         if (disk->policy)
 578                 param->flags |= DM_READONLY_FLAG;
 579
 580         param->event_nr = dm_get_event_nr(md);
 581
 582         table = dm_get_table(md);
 583         if (table) {
 584                 param->flags |= DM_ACTIVE_PRESENT_FLAG;
 585                 param->target_count = dm_table_get_num_targets(table);
 586                 dm_table_put(table);
 587         } else
 588                 param->target_count = 0;
 589
 590         return 0;
 591 }
 592
 593 static int dev_create(struct dm_ioctl *param, size_t param_size)
 594 {
 595         int r, m = DM_ANY_MINOR;
 596         struct mapped_device *md;
 597
 598         r = check_name(param->name);
 599         if (r)
 600                 return r;
 601
 602         if (param->flags & DM_PERSISTENT_DEV_FLAG)
 603                 m = MINOR(huge_decode_dev(param->dev));
 604
 605         r = dm_create(m, &md);
 606         if (r)
 607                 return r;
 608
 609         r = dm_hash_insert(param->name, *param->uuid ? param->uuid : NULL, md);
 610         if (r) {
 611                 dm_put(md);
 612                 return r;
 613         }
 614
 615         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
 616
 617         r = __dev_status(md, param);
 618         dm_put(md);
 619
 620         return r;
 621 }
 622
 623 /*
 624  * Always use UUID for lookups if it's present, otherwise use name or dev.
 625  */
 626 static struct hash_cell *__find_device_hash_cell(struct dm_ioctl *param)
 627 {
 628         struct mapped_device *md;
 629         void *mdptr = NULL;
 630
 631         if (*param->uuid)
 632                 return __get_uuid_cell(param->uuid);
 633
 634         if (*param->name)
 635                 return __get_name_cell(param->name);
 636
 637         md = dm_get_md(huge_decode_dev(param->dev));
 638         if (md)
 639                 mdptr = dm_get_mdptr(md);
 640
 641         return mdptr;
 642 }
 643
 644 static struct mapped_device *find_device(struct dm_ioctl *param)
 645 {
 646         struct hash_cell *hc;
 647         struct mapped_device *md = NULL;
 648
 649         down_read(&_hash_lock);
 650         hc = __find_device_hash_cell(param);
 651         if (hc) {
 652                 md = hc->md;
 653
 654                 /*
 655                  * Sneakily write in both the name and the uuid
 656                  * while we have the cell.
 657                  */
 658                 strncpy(param->name, hc->name, sizeof(param->name));
 659                 if (hc->uuid)
 660                         strncpy(param->uuid, hc->uuid, sizeof(param->uuid)-1);
 661                 else
 662                         param->uuid[0] = '\0';
 663
 664                 if (hc->new_map)
 665                         param->flags |= DM_INACTIVE_PRESENT_FLAG;
 666                 else
 667                         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
 668         }
 669         up_read(&_hash_lock);
 670
 671         return md;
 672 }
 673
 674 static int dev_remove(struct dm_ioctl *param, size_t param_size)
 675 {
 676         struct hash_cell *hc;
 677         struct mapped_device *md;
 678         int r;
 679
 680         down_write(&_hash_lock);
 681         hc = __find_device_hash_cell(param);
 682
 683         if (!hc) {
 684                 DMWARN("device doesn't appear to be in the dev hash table.");
 685                 up_write(&_hash_lock);
 686                 return -ENXIO;
 687         }
 688
 689         md = hc->md;
 690
 691         /*
 692          * Ensure the device is not open and nothing further can open it.
 693          */
 694         r = dm_lock_for_deletion(md);
 695         if (r) {
 696                 DMWARN("unable to remove open device %s", hc->name);
 697                 up_write(&_hash_lock);
 698                 dm_put(md);
 699                 return r;
 700         }
 701
 702         __hash_remove(hc);
 703         up_write(&_hash_lock);
 704         dm_put(md);
 705         param->data_size = 0;
 706         return 0;
 707 }
 708
 709 /*
 710  * Check a string doesn't overrun the chunk of
 711  * memory we copied from userland.
 712  */
 713 static int invalid_str(char *str, void *end)
 714 {
 715         while ((void *) str < end)
 716                 if (!*str++)
 717                         return 0;
 718
 719         return -EINVAL;
 720 }
 721
 722 static int dev_rename(struct dm_ioctl *param, size_t param_size)
 723 {
 724         int r;
 725         char *new_name = (char *) param + param->data_start;
 726
 727         if (new_name < (char *) (param + 1) ||
 728             invalid_str(new_name, (void *) param + param_size)) {
 729                 DMWARN("Invalid new logical volume name supplied.");
 730                 return -EINVAL;
 731         }
 732
 733         r = check_name(new_name);
 734         if (r)
 735                 return r;
 736
 737         param->data_size = 0;
 738         return dm_hash_rename(param->name, new_name);
 739 }
 740
 741 static int dev_set_geometry(struct dm_ioctl *param, size_t param_size)
 742 {
 743         int r = -EINVAL, x;
 744         struct mapped_device *md;
 745         struct hd_geometry geometry;
 746         unsigned long indata[4];
 747         char *geostr = (char *) param + param->data_start;
 748
 749         md = find_device(param);
 750         if (!md)
 751                 return -ENXIO;
 752
 753         if (geostr < (char *) (param + 1) ||
 754             invalid_str(geostr, (void *) param + param_size)) {
 755                 DMWARN("Invalid geometry supplied.");
 756                 goto out;
 757         }
 758
 759         x = sscanf(geostr, "%lu %lu %lu %lu", indata,
 760                    indata + 1, indata + 2, indata + 3);
 761
 762         if (x != 4) {
 763                 DMWARN("Unable to interpret geometry settings.");
 764                 goto out;
 765         }
 766
 767         if (indata[0] > 65535 || indata[1] > 255 ||
 768             indata[2] > 255 || indata[3] > ULONG_MAX) {
 769                 DMWARN("Geometry exceeds range limits.");
 770                 goto out;
 771         }
 772
 773         geometry.cylinders = indata[0];
 774         geometry.heads = indata[1];
 775         geometry.sectors = indata[2];
 776         geometry.start = indata[3];
 777
 778         r = dm_set_geometry(md, &geometry);
 779         if (!r)
 780                 r = __dev_status(md, param);
 781
 782         param->data_size = 0;
 783
 784 out:
 785         dm_put(md);
 786         return r;
 787 }
 788
 789 static int do_suspend(struct dm_ioctl *param)
 790 {
 791         int r = 0;
 792         int do_lockfs = 1;
 793         struct mapped_device *md;
 794
 795         md = find_device(param);
 796         if (!md)
 797                 return -ENXIO;
 798
 799         if (param->flags & DM_SKIP_LOCKFS_FLAG)
 800                 do_lockfs = 0;
 801
 802         if (!dm_suspended(md))
 803                 r = dm_suspend(md, do_lockfs);
 804
 805         if (!r)
 806                 r = __dev_status(md, param);
 807
 808         dm_put(md);
 809         return r;
 810 }
 811
 812 static int do_resume(struct dm_ioctl *param)
 813 {
 814         int r = 0;
 815         int do_lockfs = 1;
 816         struct hash_cell *hc;
 817         struct mapped_device *md;
 818         struct dm_table *new_map;
 819
 820         down_write(&_hash_lock);
 821
 822         hc = __find_device_hash_cell(param);
 823         if (!hc) {
 824                 DMWARN("device doesn't appear to be in the dev hash table.");
 825                 up_write(&_hash_lock);
 826                 return -ENXIO;
 827         }
 828
 829         md = hc->md;
 830
 831         new_map = hc->new_map;
 832         hc->new_map = NULL;
 833         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
 834
 835         up_write(&_hash_lock);
 836
 837         /* Do we need to load a new map ? */
 838         if (new_map) {
 839                 /* Suspend if it isn't already suspended */
 840                 if (param->flags & DM_SKIP_LOCKFS_FLAG)
 841                         do_lockfs = 0;
 842                 if (!dm_suspended(md))
 843                         dm_suspend(md, do_lockfs);
 844
 845                 r = dm_swap_table(md, new_map);
 846                 if (r) {
 847                         dm_put(md);
 848                         dm_table_put(new_map);
 849                         return r;
 850                 }
 851
 852                 if (dm_table_get_mode(new_map) & FMODE_WRITE)
 853                         set_disk_ro(dm_disk(md), 0);
 854                 else
 855                         set_disk_ro(dm_disk(md), 1);
 856
 857                 dm_table_put(new_map);
 858         }
 859
 860         if (dm_suspended(md))
 861                 r = dm_resume(md);
 862
 863         if (!r)
 864                 r = __dev_status(md, param);
 865
 866         dm_put(md);
 867         return r;
 868 }
 869
 870 /*
 871  * Set or unset the suspension state of a device.
 872  * If the device already is in the requested state we just return its status.
 873  */
 874 static int dev_suspend(struct dm_ioctl *param, size_t param_size)
 875 {
 876         if (param->flags & DM_SUSPEND_FLAG)
 877                 return do_suspend(param);
 878
 879         return do_resume(param);
 880 }
 881
 882 /*
 883  * Copies device info back to user space, used by
 884  * the create and info ioctls.
 885  */
 886 static int dev_status(struct dm_ioctl *param, size_t param_size)
 887 {
 888         int r;
 889         struct mapped_device *md;
 890
 891         md = find_device(param);
 892         if (!md)
 893                 return -ENXIO;
 894
 895         r = __dev_status(md, param);
 896         dm_put(md);
 897         return r;
 898 }
 899
 900 /*
 901  * Build up the status struct for each target
 902  */
 903 static void retrieve_status(struct dm_table *table,
 904                             struct dm_ioctl *param, size_t param_size)
 905 {
 906         unsigned int i, num_targets;
 907         struct dm_target_spec *spec;
 908         char *outbuf, *outptr;
 909         status_type_t type;
 910         size_t remaining, len, used = 0;
 911
 912         outptr = outbuf = get_result_buffer(param, param_size, &len);
 913
 914         if (param->flags & DM_STATUS_TABLE_FLAG)
 915                 type = STATUSTYPE_TABLE;
 916         else
 917                 type = STATUSTYPE_INFO;
 918
 919         /* Get all the target info */
 920         num_targets = dm_table_get_num_targets(table);
 921         for (i = 0; i < num_targets; i++) {
 922                 struct dm_target *ti = dm_table_get_target(table, i);
 923
 924                 remaining = len - (outptr - outbuf);
 925                 if (remaining <= sizeof(struct dm_target_spec)) {
 926                         param->flags |= DM_BUFFER_FULL_FLAG;
 927                         break;
 928                 }
 929
 930                 spec = (struct dm_target_spec *) outptr;
 931
 932                 spec->status = 0;
 933                 spec->sector_start = ti->begin;
 934                 spec->length = ti->len;
 935                 strncpy(spec->target_type, ti->type->name,
 936                         sizeof(spec->target_type));
 937
 938                 outptr += sizeof(struct dm_target_spec);
 939                 remaining = len - (outptr - outbuf);
 940                 if (remaining <= 0) {
 941                         param->flags |= DM_BUFFER_FULL_FLAG;
 942                         break;
 943                 }
 944
 945                 /* Get the status/table string from the target driver */
 946                 if (ti->type->status) {
 947                         if (ti->type->status(ti, type, outptr, remaining)) {
 948                                 param->flags |= DM_BUFFER_FULL_FLAG;
 949                                 break;
 950                         }
 951                 } else
 952                         outptr[0] = '\0';
 953
 954                 outptr += strlen(outptr) + 1;
 955                 used = param->data_start + (outptr - outbuf);
 956
 957                 outptr = align_ptr(outptr);
 958                 spec->next = outptr - outbuf;
 959         }
 960
 961         if (used)
 962                 param->data_size = used;
 963
 964         param->target_count = num_targets;
 965 }
 966
 967 /*
 968  * Wait for a device to report an event
 969  */
 970 static int dev_wait(struct dm_ioctl *param, size_t param_size)
 971 {
 972         int r;
 973         struct mapped_device *md;
 974         struct dm_table *table;
 975
 976         md = find_device(param);
 977         if (!md)
 978                 return -ENXIO;
 979
 980         /*
 981          * Wait for a notification event
 982          */
 983         if (dm_wait_event(md, param->event_nr)) {
 984                 r = -ERESTARTSYS;
 985                 goto out;
 986         }
 987
 988         /*
 989          * The userland program is going to want to know what
 990          * changed to trigger the event, so we may as well tell
 991          * him and save an ioctl.
 992          */
 993         r = __dev_status(md, param);
 994         if (r)
 995                 goto out;
 996
 997         table = dm_get_table(md);
 998         if (table) {
 999                 retrieve_status(table, param, param_size);
1000                 dm_table_put(table);
1001         }
1002
1003  out:
1004         dm_put(md);
1005         return r;
1006 }
1007
1008 static inline int get_mode(struct dm_ioctl *param)
1009 {
1010         int mode = FMODE_READ | FMODE_WRITE;
1011
1012         if (param->flags & DM_READONLY_FLAG)
1013                 mode = FMODE_READ;
1014
1015         return mode;
1016 }
1017
1018 static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
1019                        struct dm_target_spec **spec, char **target_params)
1020 {
1021         *spec = (struct dm_target_spec *) ((unsigned char *) last + next);
1022         *target_params = (char *) (*spec + 1);
1023
1024         if (*spec < (last + 1))
1025                 return -EINVAL;
1026
1027         return invalid_str(*target_params, end);
1028 }
1029
1030 static int populate_table(struct dm_table *table,
1031                           struct dm_ioctl *param, size_t param_size)
1032 {
1033         int r;
1034         unsigned int i = 0;
1035         struct dm_target_spec *spec = (struct dm_target_spec *) param;
1036         uint32_t next = param->data_start;
1037         void *end = (void *) param + param_size;
1038         char *target_params;
1039
1040         if (!param->target_count) {
1041                 DMWARN("populate_table: no targets specified");
1042                 return -EINVAL;
1043         }
1044
1045         for (i = 0; i < param->target_count; i++) {
1046
1047                 r = next_target(spec, next, end, &spec, &target_params);
1048                 if (r) {
1049                         DMWARN("unable to find target");
1050                         return r;
1051                 }
1052
1053                 r = dm_table_add_target(table, spec->target_type,
1054                                         (sector_t) spec->sector_start,
1055                                         (sector_t) spec->length,
1056                                         target_params);
1057                 if (r) {
1058                         DMWARN("error adding target to table");
1059                         return r;
1060                 }
1061
1062                 next = spec->next;
1063         }
1064
1065         return dm_table_complete(table);
1066 }
1067
1068 static int table_load(struct dm_ioctl *param, size_t param_size)
1069 {
1070         int r;
1071         struct hash_cell *hc;
1072         struct dm_table *t;
1073         struct mapped_device *md;
1074
1075         md = find_device(param);
1076         if (!md)
1077                 return -ENXIO;
1078
1079         r = dm_table_create(&t, get_mode(param), param->target_count, md);
1080         if (r)
1081                 goto out;
1082
1083         r = populate_table(t, param, param_size);
1084         if (r) {
1085                 dm_table_put(t);
1086                 goto out;
1087         }
1088
1089         down_write(&_hash_lock);
1090         hc = dm_get_mdptr(md);
1091         if (!hc || hc->md != md) {
1092                 DMWARN("device has been removed from the dev hash table.");
1093                 dm_table_put(t);
1094                 up_write(&_hash_lock);
1095                 r = -ENXIO;
1096                 goto out;
1097         }
1098
1099         if (hc->new_map)
1100                 dm_table_put(hc->new_map);
1101         hc->new_map = t;
1102         up_write(&_hash_lock);
1103
1104         param->flags |= DM_INACTIVE_PRESENT_FLAG;
1105         r = __dev_status(md, param);
1106
1107 out:
1108         dm_put(md);
1109
1110         return r;
1111 }
1112
1113 static int table_clear(struct dm_ioctl *param, size_t param_size)
1114 {
1115         int r;
1116         struct hash_cell *hc;
1117         struct mapped_device *md;
1118
1119         down_write(&_hash_lock);
1120
1121         hc = __find_device_hash_cell(param);
1122         if (!hc) {
1123                 DMWARN("device doesn't appear to be in the dev hash table.");
1124                 up_write(&_hash_lock);
1125                 return -ENXIO;
1126         }
1127
1128         if (hc->new_map) {
1129                 dm_table_put(hc->new_map);
1130                 hc->new_map = NULL;
1131         }
1132
1133         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1134
1135         r = __dev_status(hc->md, param);
1136         md = hc->md;
1137         up_write(&_hash_lock);
1138         dm_put(md);
1139         return r;
1140 }
1141
1142 /*
1143  * Retrieves a list of devices used by a particular dm device.
1144  */
1145 static void retrieve_deps(struct dm_table *table,
1146                           struct dm_ioctl *param, size_t param_size)
1147 {
1148         unsigned int count = 0;
1149         struct list_head *tmp;
1150         size_t len, needed;
1151         struct dm_dev *dd;
1152         struct dm_target_deps *deps;
1153
1154         deps = get_result_buffer(param, param_size, &len);
1155
1156         /*
1157          * Count the devices.
1158          */
1159         list_for_each (tmp, dm_table_get_devices(table))
1160                 count++;
1161
1162         /*
1163          * Check we have enough space.
1164          */
1165         needed = sizeof(*deps) + (sizeof(*deps->dev) * count);
1166         if (len < needed) {
1167                 param->flags |= DM_BUFFER_FULL_FLAG;
1168                 return;
1169         }
1170
1171         /*
1172          * Fill in the devices.
1173          */
1174         deps->count = count;
1175         count = 0;
1176         list_for_each_entry (dd, dm_table_get_devices(table), list)
1177                 deps->dev[count++] = huge_encode_dev(dd->bdev->bd_dev);
1178
1179         param->data_size = param->data_start + needed;
1180 }
1181
1182 static int table_deps(struct dm_ioctl *param, size_t param_size)
1183 {
1184         int r = 0;
1185         struct mapped_device *md;
1186         struct dm_table *table;
1187
1188         md = find_device(param);
1189         if (!md)
1190                 return -ENXIO;
1191
1192         r = __dev_status(md, param);
1193         if (r)
1194                 goto out;
1195
1196         table = dm_get_table(md);
1197         if (table) {
1198                 retrieve_deps(table, param, param_size);
1199                 dm_table_put(table);
1200         }
1201
1202  out:
1203         dm_put(md);
1204         return r;
1205 }
1206
1207 /*
1208  * Return the status of a device as a text string for each
1209  * target.
1210  */
1211 static int table_status(struct dm_ioctl *param, size_t param_size)
1212 {
1213         int r;
1214         struct mapped_device *md;
1215         struct dm_table *table;
1216
1217         md = find_device(param);
1218         if (!md)
1219                 return -ENXIO;
1220
1221         r = __dev_status(md, param);
1222         if (r)
1223                 goto out;
1224
1225         table = dm_get_table(md);
1226         if (table) {
1227                 retrieve_status(table, param, param_size);
1228                 dm_table_put(table);
1229         }
1230
1231  out:
1232         dm_put(md);
1233         return r;
1234 }
1235
1236 /*
1237  * Pass a message to the target that's at the supplied device offset.
1238  */
1239 static int target_message(struct dm_ioctl *param, size_t param_size)
1240 {
1241         int r, argc;
1242         char **argv;
1243         struct mapped_device *md;
1244         struct dm_table *table;
1245         struct dm_target *ti;
1246         struct dm_target_msg *tmsg = (void *) param + param->data_start;
1247
1248         md = find_device(param);
1249         if (!md)
1250                 return -ENXIO;
1251
1252         r = __dev_status(md, param);
1253         if (r)
1254                 goto out;
1255
1256         if (tmsg < (struct dm_target_msg *) (param + 1) ||
1257             invalid_str(tmsg->message, (void *) param + param_size)) {
1258                 DMWARN("Invalid target message parameters.");
1259                 r = -EINVAL;
1260                 goto out;
1261         }
1262
1263         r = dm_split_args(&argc, &argv, tmsg->message);
1264         if (r) {
1265                 DMWARN("Failed to split target message parameters");
1266                 goto out;
1267         }
1268
1269         table = dm_get_table(md);
1270         if (!table)
1271                 goto out_argv;
1272
1273         if (tmsg->sector >= dm_table_get_size(table)) {
1274                 DMWARN("Target message sector outside device.");
1275                 r = -EINVAL;
1276                 goto out_table;
1277         }
1278
1279         ti = dm_table_find_target(table, tmsg->sector);
1280         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  out_table:
1288         dm_table_put(table);
1289  out_argv:
1290         kfree(argv);
1291  out:
1292         param->data_size = 0;
1293         dm_put(md);
1294         return r;
1295 }
1296
1297 /*-----------------------------------------------------------------
1298  * Implementation of open/close/ioctl on the special char
1299  * device.
1300  *---------------------------------------------------------------*/
1301 static ioctl_fn lookup_ioctl(unsigned int cmd)
1302 {
1303         static struct {
1304                 int cmd;
1305                 ioctl_fn fn;
1306         } _ioctls[] = {
1307                 {DM_VERSION_CMD, NULL}, /* version is dealt with elsewhere */
1308                 {DM_REMOVE_ALL_CMD, remove_all},
1309                 {DM_LIST_DEVICES_CMD, list_devices},
1310
1311                 {DM_DEV_CREATE_CMD, dev_create},
1312                 {DM_DEV_REMOVE_CMD, dev_remove},
1313                 {DM_DEV_RENAME_CMD, dev_rename},
1314                 {DM_DEV_SUSPEND_CMD, dev_suspend},
1315                 {DM_DEV_STATUS_CMD, dev_status},
1316                 {DM_DEV_WAIT_CMD, dev_wait},
1317
1318                 {DM_TABLE_LOAD_CMD, table_load},
1319                 {DM_TABLE_CLEAR_CMD, table_clear},
1320                 {DM_TABLE_DEPS_CMD, table_deps},
1321                 {DM_TABLE_STATUS_CMD, table_status},
1322
1323                 {DM_LIST_VERSIONS_CMD, list_versions},
1324
1325                 {DM_TARGET_MSG_CMD, target_message},
1326                 {DM_DEV_SET_GEOMETRY_CMD, dev_set_geometry}
1327         };
1328
1329         return (cmd >= ARRAY_SIZE(_ioctls)) ? NULL : _ioctls[cmd].fn;
1330 }
1331
1332 /*
1333  * As well as checking the version compatibility this always
1334  * copies the kernel interface version out.
1335  */
1336 static int check_version(unsigned int cmd, struct dm_ioctl __user *user)
1337 {
1338         uint32_t version[3];
1339         int r = 0;
1340
1341         if (copy_from_user(version, user->version, sizeof(version)))
1342                 return -EFAULT;
1343
1344         if ((DM_VERSION_MAJOR != version[0]) ||
1345             (DM_VERSION_MINOR < version[1])) {
1346                 DMWARN("ioctl interface mismatch: "
1347                        "kernel(%u.%u.%u), user(%u.%u.%u), cmd(%d)",
1348                        DM_VERSION_MAJOR, DM_VERSION_MINOR,
1349                        DM_VERSION_PATCHLEVEL,
1350                        version[0], version[1], version[2], cmd);
1351                 r = -EINVAL;
1352         }
1353
1354         /*
1355          * Fill in the kernel version.
1356          */
1357         version[0] = DM_VERSION_MAJOR;
1358         version[1] = DM_VERSION_MINOR;
1359         version[2] = DM_VERSION_PATCHLEVEL;
1360         if (copy_to_user(user->version, version, sizeof(version)))
1361                 return -EFAULT;
1362
1363         return r;
1364 }
1365
1366 static void free_params(struct dm_ioctl *param)
1367 {
1368         vfree(param);
1369 }
1370
1371 static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl **param)
1372 {
1373         struct dm_ioctl tmp, *dmi;
1374
1375         if (copy_from_user(&tmp, user, sizeof(tmp)))
1376                 return -EFAULT;
1377
1378         if (tmp.data_size < sizeof(tmp))
1379                 return -EINVAL;
1380
1381         dmi = (struct dm_ioctl *) vmalloc(tmp.data_size);
1382         if (!dmi)
1383                 return -ENOMEM;
1384
1385         if (copy_from_user(dmi, user, tmp.data_size)) {
1386                 vfree(dmi);
1387                 return -EFAULT;
1388         }
1389
1390         *param = dmi;
1391         return 0;
1392 }
1393
1394 static int validate_params(uint cmd, struct dm_ioctl *param)
1395 {
1396         /* Always clear this flag */
1397         param->flags &= ~DM_BUFFER_FULL_FLAG;
1398
1399         /* Ignores parameters */
1400         if (cmd == DM_REMOVE_ALL_CMD ||
1401             cmd == DM_LIST_DEVICES_CMD ||
1402             cmd == DM_LIST_VERSIONS_CMD)
1403                 return 0;
1404
1405         if ((cmd == DM_DEV_CREATE_CMD)) {
1406                 if (!*param->name) {
1407                         DMWARN("name not supplied when creating device");
1408                         return -EINVAL;
1409                 }
1410         } else if ((*param->uuid && *param->name)) {
1411                 DMWARN("only supply one of name or uuid, cmd(%u)", cmd);
1412                 return -EINVAL;
1413         }
1414
1415         /* Ensure strings are terminated */
1416         param->name[DM_NAME_LEN - 1] = '\0';
1417         param->uuid[DM_UUID_LEN - 1] = '\0';
1418
1419         return 0;
1420 }
1421
1422 static int ctl_ioctl(struct inode *inode, struct file *file,
1423                      uint command, ulong u)
1424 {
1425         int r = 0;
1426         unsigned int cmd;
1427         struct dm_ioctl *param;
1428         struct dm_ioctl __user *user = (struct dm_ioctl __user *) u;
1429         ioctl_fn fn = NULL;
1430         size_t param_size;
1431
1432         /* only root can play with this */
1433         if (!capable(CAP_SYS_ADMIN))
1434                 return -EACCES;
1435
1436         if (_IOC_TYPE(command) != DM_IOCTL)
1437                 return -ENOTTY;
1438
1439         cmd = _IOC_NR(command);
1440
1441         /*
1442          * Check the interface version passed in.  This also
1443          * writes out the kernel's interface version.
1444          */
1445         r = check_version(cmd, user);
1446         if (r)
1447                 return r;
1448
1449         /*
1450          * Nothing more to do for the version command.
1451          */
1452         if (cmd == DM_VERSION_CMD)
1453                 return 0;
1454
1455         fn = lookup_ioctl(cmd);
1456         if (!fn) {
1457                 DMWARN("dm_ctl_ioctl: unknown command 0x%x", command);
1458                 return -ENOTTY;
1459         }
1460
1461         /*
1462          * Trying to avoid low memory issues when a device is
1463          * suspended.
1464          */
1465         current->flags |= PF_MEMALLOC;
1466
1467         /*
1468          * Copy the parameters into kernel space.
1469          */
1470         r = copy_params(user, &param);
1471
1472         current->flags &= ~PF_MEMALLOC;
1473
1474         if (r)
1475                 return r;
1476
1477         r = validate_params(cmd, param);
1478         if (r)
1479                 goto out;
1480
1481         param_size = param->data_size;
1482         param->data_size = sizeof(*param);
1483         r = fn(param, param_size);
1484
1485         /*
1486          * Copy the results back to userland.
1487          */
1488         if (!r && copy_to_user(user, param, param->data_size))
1489                 r = -EFAULT;
1490
1491  out:
1492         free_params(param);
1493         return r;
1494 }
1495
1496 static struct file_operations _ctl_fops = {
1497         .ioctl   = ctl_ioctl,
1498         .owner   = THIS_MODULE,
1499 };
1500
1501 static struct miscdevice _dm_misc = {
1502         .minor          = MISC_DYNAMIC_MINOR,
1503         .name           = DM_NAME,
1504         .devfs_name     = "mapper/control",
1505         .fops           = &_ctl_fops
1506 };
1507
1508 /*
1509  * Create misc character device and link to DM_DIR/control.
1510  */
1511 int __init dm_interface_init(void)
1512 {
1513         int r;
1514
1515         r = dm_hash_init();
1516         if (r)
1517                 return r;
1518
1519         r = misc_register(&_dm_misc);
1520         if (r) {
1521                 DMERR("misc_register failed for control device");
1522                 dm_hash_exit();
1523                 return r;
1524         }
1525
1526         DMINFO("%d.%d.%d%s initialised: %s", DM_VERSION_MAJOR,
1527                DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL, DM_VERSION_EXTRA,
1528                DM_DRIVER_EMAIL);
1529         return 0;
1530 }
1531
1532 void dm_interface_exit(void)
1533 {
1534         if (misc_deregister(&_dm_misc) < 0)
1535                 DMERR("misc_deregister failed for control device");
1536
1537         dm_hash_exit();
1538 }