drivers/md/dm-ioctl.c

   1 /*
   2  * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
   3  * Copyright (C) 2004 - 2005 Red Hat, Inc. All rights reserved.
   4  *
   5  * This file is released under the GPL.
   6  */
   7
   8 #include "dm.h"
   9
  10 #include <linux/module.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/miscdevice.h>
  13 #include <linux/init.h>
  14 #include <linux/wait.h>
  15 #include <linux/slab.h>
  16 #include <linux/devfs_fs_kernel.h>
  17 #include <linux/dm-ioctl.h>
  18 #include <linux/hdreg.h>
  19
  20 #include <asm/uaccess.h>
  21
  22 #define DM_DRIVER_EMAIL "dm-devel@redhat.com"
  23
  24 /*-----------------------------------------------------------------
  25  * The ioctl interface needs to be able to look up devices by
  26  * name or uuid.
  27  *---------------------------------------------------------------*/
  28 struct hash_cell {
  29         struct list_head name_list;
  30         struct list_head uuid_list;
  31
  32         char *name;
  33         char *uuid;
  34         struct mapped_device *md;
  35         struct dm_table *new_map;
  36 };
  37
  38 struct vers_iter {
  39     size_t param_size;
  40     struct dm_target_versions *vers, *old_vers;
  41     char *end;
  42     uint32_t flags;
  43 };
  44
  45
  46 #define NUM_BUCKETS 64
  47 #define MASK_BUCKETS (NUM_BUCKETS - 1)
  48 static struct list_head _name_buckets[NUM_BUCKETS];
  49 static struct list_head _uuid_buckets[NUM_BUCKETS];
  50
  51 static void dm_hash_remove_all(void);
  52
  53 /*
  54  * Guards access to both hash tables.
  55  */
  56 static DECLARE_RWSEM(_hash_lock);
  57
  58 static void init_buckets(struct list_head *buckets)
  59 {
  60         unsigned int i;
  61
  62         for (i = 0; i < NUM_BUCKETS; i++)
  63                 INIT_LIST_HEAD(buckets + i);
  64 }
  65
  66 static int dm_hash_init(void)
  67 {
  68         init_buckets(_name_buckets);
  69         init_buckets(_uuid_buckets);
  70         devfs_mk_dir(DM_DIR);
  71         return 0;
  72 }
  73
  74 static void dm_hash_exit(void)
  75 {
  76         dm_hash_remove_all();
  77         devfs_remove(DM_DIR);
  78 }
  79
  80 /*-----------------------------------------------------------------
  81  * Hash function:
  82  * We're not really concerned with the str hash function being
  83  * fast since it's only used by the ioctl interface.
  84  *---------------------------------------------------------------*/
  85 static unsigned int hash_str(const char *str)
  86 {
  87         const unsigned int hash_mult = 2654435387U;
  88         unsigned int h = 0;
  89
  90         while (*str)
  91                 h = (h + (unsigned int) *str++) * hash_mult;
  92
  93         return h & MASK_BUCKETS;
  94 }
  95
  96 /*-----------------------------------------------------------------
  97  * Code for looking up a device by name
  98  *---------------------------------------------------------------*/
  99 static struct hash_cell *__get_name_cell(const char *str)
 100 {
 101         struct hash_cell *hc;
 102         unsigned int h = hash_str(str);
 103
 104         list_for_each_entry (hc, _name_buckets + h, name_list)
 105                 if (!strcmp(hc->name, str))
 106                         return hc;
 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                         return hc;
 119
 120         return NULL;
 121 }
 122
 123 /*-----------------------------------------------------------------
 124  * Inserting, removing and renaming a device.
 125  *---------------------------------------------------------------*/
 126 static struct hash_cell *alloc_cell(const char *name, const char *uuid,
 127                                     struct mapped_device *md)
 128 {
 129         struct hash_cell *hc;
 130
 131         hc = kmalloc(sizeof(*hc), GFP_KERNEL);
 132         if (!hc)
 133                 return NULL;
 134
 135         hc->name = kstrdup(name, GFP_KERNEL);
 136         if (!hc->name) {
 137                 kfree(hc);
 138                 return NULL;
 139         }
 140
 141         if (!uuid)
 142                 hc->uuid = NULL;
 143
 144         else {
 145                 hc->uuid = kstrdup(uuid, GFP_KERNEL);
 146                 if (!hc->uuid) {
 147                         kfree(hc->name);
 148                         kfree(hc);
 149                         return NULL;
 150                 }
 151         }
 152
 153         INIT_LIST_HEAD(&hc->name_list);
 154         INIT_LIST_HEAD(&hc->uuid_list);
 155         hc->md = md;
 156         hc->new_map = NULL;
 157         return hc;
 158 }
 159
 160 static void free_cell(struct hash_cell *hc)
 161 {
 162         if (hc) {
 163                 kfree(hc->name);
 164                 kfree(hc->uuid);
 165                 kfree(hc);
 166         }
 167 }
 168
 169 /*
 170  * devfs stuff.
 171  */
 172 static int register_with_devfs(struct hash_cell *hc)
 173 {
 174         struct gendisk *disk = dm_disk(hc->md);
 175
 176         devfs_mk_bdev(MKDEV(disk->major, disk->first_minor),
 177                       S_IFBLK | S_IRUSR | S_IWUSR | S_IRGRP,
 178                       DM_DIR "/%s", hc->name);
 179         return 0;
 180 }
 181
 182 static int unregister_with_devfs(struct hash_cell *hc)
 183 {
 184         devfs_remove(DM_DIR"/%s", hc->name);
 185         return 0;
 186 }
 187
 188 /*
 189  * The kdev_t and uuid of a device can never change once it is
 190  * initially inserted.
 191  */
 192 static int dm_hash_insert(const char *name, const char *uuid, struct mapped_device *md)
 193 {
 194         struct hash_cell *cell;
 195
 196         /*
 197          * Allocate the new cells.
 198          */
 199         cell = alloc_cell(name, uuid, md);
 200         if (!cell)
 201                 return -ENOMEM;
 202
 203         /*
 204          * Insert the cell into both hash tables.
 205          */
 206         down_write(&_hash_lock);
 207         if (__get_name_cell(name))
 208                 goto bad;
 209
 210         list_add(&cell->name_list, _name_buckets + hash_str(name));
 211
 212         if (uuid) {
 213                 if (__get_uuid_cell(uuid)) {
 214                         list_del(&cell->name_list);
 215                         goto bad;
 216                 }
 217                 list_add(&cell->uuid_list, _uuid_buckets + hash_str(uuid));
 218         }
 219         register_with_devfs(cell);
 220         dm_get(md);
 221         dm_set_mdptr(md, cell);
 222         up_write(&_hash_lock);
 223
 224         return 0;
 225
 226  bad:
 227         up_write(&_hash_lock);
 228         free_cell(cell);
 229         return -EBUSY;
 230 }
 231
 232 static void __hash_remove(struct hash_cell *hc)
 233 {
 234         struct dm_table *table;
 235
 236         /* remove from the dev hash */
 237         list_del(&hc->uuid_list);
 238         list_del(&hc->name_list);
 239         unregister_with_devfs(hc);
 240         dm_set_mdptr(hc->md, NULL);
 241
 242         table = dm_get_table(hc->md);
 243         if (table) {
 244                 dm_table_event(table);
 245                 dm_table_put(table);
 246         }
 247
 248         if (hc->new_map)
 249                 dm_table_put(hc->new_map);
 250         dm_put(hc->md);
 251         free_cell(hc);
 252 }
 253
 254 static void dm_hash_remove_all(void)
 255 {
 256         int i;
 257         struct hash_cell *hc;
 258         struct list_head *tmp, *n;
 259
 260         down_write(&_hash_lock);
 261         for (i = 0; i < NUM_BUCKETS; i++) {
 262                 list_for_each_safe (tmp, n, _name_buckets + i) {
 263                         hc = list_entry(tmp, struct hash_cell, name_list);
 264                         __hash_remove(hc);
 265                 }
 266         }
 267         up_write(&_hash_lock);
 268 }
 269
 270 static int dm_hash_rename(const char *old, const char *new)
 271 {
 272         char *new_name, *old_name;
 273         struct hash_cell *hc;
 274         struct dm_table *table;
 275
 276         /*
 277          * duplicate new.
 278          */
 279         new_name = kstrdup(new, GFP_KERNEL);
 280         if (!new_name)
 281                 return -ENOMEM;
 282
 283         down_write(&_hash_lock);
 284
 285         /*
 286          * Is new free ?
 287          */
 288         hc = __get_name_cell(new);
 289         if (hc) {
 290                 DMWARN("asked to rename to an already existing name %s -> %s",
 291                        old, new);
 292                 up_write(&_hash_lock);
 293                 kfree(new_name);
 294                 return -EBUSY;
 295         }
 296
 297         /*
 298          * Is there such a device as 'old' ?
 299          */
 300         hc = __get_name_cell(old);
 301         if (!hc) {
 302                 DMWARN("asked to rename a non existent device %s -> %s",
 303                        old, new);
 304                 up_write(&_hash_lock);
 305                 kfree(new_name);
 306                 return -ENXIO;
 307         }
 308
 309         /*
 310          * rename and move the name cell.
 311          */
 312         unregister_with_devfs(hc);
 313
 314         list_del(&hc->name_list);
 315         old_name = hc->name;
 316         hc->name = new_name;
 317         list_add(&hc->name_list, _name_buckets + hash_str(new_name));
 318
 319         /* rename the device node in devfs */
 320         register_with_devfs(hc);
 321
 322         /*
 323          * Wake up any dm event waiters.
 324          */
 325         table = dm_get_table(hc->md);
 326         if (table) {
 327                 dm_table_event(table);
 328                 dm_table_put(table);
 329         }
 330
 331         up_write(&_hash_lock);
 332         kfree(old_name);
 333         return 0;
 334 }
 335
 336 /*-----------------------------------------------------------------
 337  * Implementation of the ioctl commands
 338  *---------------------------------------------------------------*/
 339 /*
 340  * All the ioctl commands get dispatched to functions with this
 341  * prototype.
 342  */
 343 typedef int (*ioctl_fn)(struct dm_ioctl *param, size_t param_size);
 344
 345 static int remove_all(struct dm_ioctl *param, size_t param_size)
 346 {
 347         dm_hash_remove_all();
 348         param->data_size = 0;
 349         return 0;
 350 }
 351
 352 /*
 353  * Round up the ptr to an 8-byte boundary.
 354  */
 355 #define ALIGN_MASK 7
 356 static inline void *align_ptr(void *ptr)
 357 {
 358         return (void *) (((size_t) (ptr + ALIGN_MASK)) & ~ALIGN_MASK);
 359 }
 360
 361 /*
 362  * Retrieves the data payload buffer from an already allocated
 363  * struct dm_ioctl.
 364  */
 365 static void *get_result_buffer(struct dm_ioctl *param, size_t param_size,
 366                                size_t *len)
 367 {
 368         param->data_start = align_ptr(param + 1) - (void *) param;
 369
 370         if (param->data_start < param_size)
 371                 *len = param_size - param->data_start;
 372         else
 373                 *len = 0;
 374
 375         return ((void *) param) + param->data_start;
 376 }
 377
 378 static int list_devices(struct dm_ioctl *param, size_t param_size)
 379 {
 380         unsigned int i;
 381         struct hash_cell *hc;
 382         size_t len, needed = 0;
 383         struct gendisk *disk;
 384         struct dm_name_list *nl, *old_nl = NULL;
 385
 386         down_write(&_hash_lock);
 387
 388         /*
 389          * Loop through all the devices working out how much
 390          * space we need.
 391          */
 392         for (i = 0; i < NUM_BUCKETS; i++) {
 393                 list_for_each_entry (hc, _name_buckets + i, name_list) {
 394                         needed += sizeof(struct dm_name_list);
 395                         needed += strlen(hc->name) + 1;
 396                         needed += ALIGN_MASK;
 397                 }
 398         }
 399
 400         /*
 401          * Grab our output buffer.
 402          */
 403         nl = get_result_buffer(param, param_size, &len);
 404         if (len < needed) {
 405                 param->flags |= DM_BUFFER_FULL_FLAG;
 406                 goto out;
 407         }
 408         param->data_size = param->data_start + needed;
 409
 410         nl->dev = 0;    /* Flags no data */
 411
 412         /*
 413          * Now loop through filling out the names.
 414          */
 415         for (i = 0; i < NUM_BUCKETS; i++) {
 416                 list_for_each_entry (hc, _name_buckets + i, name_list) {
 417                         if (old_nl)
 418                                 old_nl->next = (uint32_t) ((void *) nl -
 419                                                            (void *) old_nl);
 420                         disk = dm_disk(hc->md);
 421                         nl->dev = huge_encode_dev(MKDEV(disk->major, disk->first_minor));
 422                         nl->next = 0;
 423                         strcpy(nl->name, hc->name);
 424
 425                         old_nl = nl;
 426                         nl = align_ptr(((void *) ++nl) + strlen(hc->name) + 1);
 427                 }
 428         }
 429
 430  out:
 431         up_write(&_hash_lock);
 432         return 0;
 433 }
 434
 435 static void list_version_get_needed(struct target_type *tt, void *needed_param)
 436 {
 437     size_t *needed = needed_param;
 438
 439     *needed += sizeof(struct dm_target_versions);
 440     *needed += strlen(tt->name);
 441     *needed += ALIGN_MASK;
 442 }
 443
 444 static void list_version_get_info(struct target_type *tt, void *param)
 445 {
 446     struct vers_iter *info = param;
 447
 448     /* Check space - it might have changed since the first iteration */
 449     if ((char *)info->vers + sizeof(tt->version) + strlen(tt->name) + 1 >
 450         info->end) {
 451
 452         info->flags = DM_BUFFER_FULL_FLAG;
 453         return;
 454     }
 455
 456     if (info->old_vers)
 457         info->old_vers->next = (uint32_t) ((void *)info->vers -
 458                                            (void *)info->old_vers);
 459     info->vers->version[0] = tt->version[0];
 460     info->vers->version[1] = tt->version[1];
 461     info->vers->version[2] = tt->version[2];
 462     info->vers->next = 0;
 463     strcpy(info->vers->name, tt->name);
 464
 465     info->old_vers = info->vers;
 466     info->vers = align_ptr(((void *) ++info->vers) + strlen(tt->name) + 1);
 467 }
 468
 469 static int list_versions(struct dm_ioctl *param, size_t param_size)
 470 {
 471         size_t len, needed = 0;
 472         struct dm_target_versions *vers;
 473         struct vers_iter iter_info;
 474
 475         /*
 476          * Loop through all the devices working out how much
 477          * space we need.
 478          */
 479         dm_target_iterate(list_version_get_needed, &needed);
 480
 481         /*
 482          * Grab our output buffer.
 483          */
 484         vers = get_result_buffer(param, param_size, &len);
 485         if (len < needed) {
 486                 param->flags |= DM_BUFFER_FULL_FLAG;
 487                 goto out;
 488         }
 489         param->data_size = param->data_start + needed;
 490
 491         iter_info.param_size = param_size;
 492         iter_info.old_vers = NULL;
 493         iter_info.vers = vers;
 494         iter_info.flags = 0;
 495         iter_info.end = (char *)vers+len;
 496
 497         /*
 498          * Now loop through filling out the names & versions.
 499          */
 500         dm_target_iterate(list_version_get_info, &iter_info);
 501         param->flags |= iter_info.flags;
 502
 503  out:
 504         return 0;
 505 }
 506
 507
 508
 509 static int check_name(const char *name)
 510 {
 511         if (strchr(name, '/')) {
 512                 DMWARN("invalid device name");
 513                 return -EINVAL;
 514         }
 515
 516         return 0;
 517 }
 518
 519 /*
 520  * Fills in a dm_ioctl structure, ready for sending back to
 521  * userland.
 522  */
 523 static int __dev_status(struct mapped_device *md, struct dm_ioctl *param)
 524 {
 525         struct gendisk *disk = dm_disk(md);
 526         struct dm_table *table;
 527         struct block_device *bdev;
 528
 529         param->flags &= ~(DM_SUSPEND_FLAG | DM_READONLY_FLAG |
 530                           DM_ACTIVE_PRESENT_FLAG);
 531
 532         if (dm_suspended(md))
 533                 param->flags |= DM_SUSPEND_FLAG;
 534
 535         param->dev = huge_encode_dev(MKDEV(disk->major, disk->first_minor));
 536
 537         if (!(param->flags & DM_SKIP_BDGET_FLAG)) {
 538                 bdev = bdget_disk(disk, 0);
 539                 if (!bdev)
 540                         return -ENXIO;
 541
 542                 /*
 543                  * Yes, this will be out of date by the time it gets back
 544                  * to userland, but it is still very useful for
 545                  * debugging.
 546                  */
 547                 param->open_count = bdev->bd_openers;
 548                 bdput(bdev);
 549         } else
 550                 param->open_count = -1;
 551
 552         if (disk->policy)
 553                 param->flags |= DM_READONLY_FLAG;
 554
 555         param->event_nr = dm_get_event_nr(md);
 556
 557         table = dm_get_table(md);
 558         if (table) {
 559                 param->flags |= DM_ACTIVE_PRESENT_FLAG;
 560                 param->target_count = dm_table_get_num_targets(table);
 561                 dm_table_put(table);
 562         } else
 563                 param->target_count = 0;
 564
 565         return 0;
 566 }
 567
 568 static int dev_create(struct dm_ioctl *param, size_t param_size)
 569 {
 570         int r;
 571         struct mapped_device *md;
 572
 573         r = check_name(param->name);
 574         if (r)
 575                 return r;
 576
 577         if (param->flags & DM_PERSISTENT_DEV_FLAG)
 578                 r = dm_create_with_minor(MINOR(huge_decode_dev(param->dev)), &md);
 579         else
 580                 r = dm_create(&md);
 581
 582         if (r)
 583                 return r;
 584
 585         r = dm_hash_insert(param->name, *param->uuid ? param->uuid : NULL, md);
 586         if (r) {
 587                 dm_put(md);
 588                 return r;
 589         }
 590
 591         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
 592
 593         r = __dev_status(md, param);
 594         dm_put(md);
 595
 596         return r;
 597 }
 598
 599 /*
 600  * Always use UUID for lookups if it's present, otherwise use name or dev.
 601  */
 602 static struct hash_cell *__find_device_hash_cell(struct dm_ioctl *param)
 603 {
 604         struct mapped_device *md;
 605         void *mdptr = NULL;
 606
 607         if (*param->uuid)
 608                 return __get_uuid_cell(param->uuid);
 609
 610         if (*param->name)
 611                 return __get_name_cell(param->name);
 612
 613         md = dm_get_md(huge_decode_dev(param->dev));
 614         if (md) {
 615                 mdptr = dm_get_mdptr(md);
 616                 dm_put(md);
 617         }
 618
 619         return mdptr;
 620 }
 621
 622 static struct mapped_device *find_device(struct dm_ioctl *param)
 623 {
 624         struct hash_cell *hc;
 625         struct mapped_device *md = NULL;
 626
 627         down_read(&_hash_lock);
 628         hc = __find_device_hash_cell(param);
 629         if (hc) {
 630                 md = hc->md;
 631                 dm_get(md);
 632
 633                 /*
 634                  * Sneakily write in both the name and the uuid
 635                  * while we have the cell.
 636                  */
 637                 strncpy(param->name, hc->name, sizeof(param->name));
 638                 if (hc->uuid)
 639                         strncpy(param->uuid, hc->uuid, sizeof(param->uuid)-1);
 640                 else
 641                         param->uuid[0] = '\0';
 642
 643                 if (hc->new_map)
 644                         param->flags |= DM_INACTIVE_PRESENT_FLAG;
 645                 else
 646                         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
 647         }
 648         up_read(&_hash_lock);
 649
 650         return md;
 651 }
 652
 653 static int dev_remove(struct dm_ioctl *param, size_t param_size)
 654 {
 655         struct hash_cell *hc;
 656
 657         down_write(&_hash_lock);
 658         hc = __find_device_hash_cell(param);
 659
 660         if (!hc) {
 661                 DMWARN("device doesn't appear to be in the dev hash table.");
 662                 up_write(&_hash_lock);
 663                 return -ENXIO;
 664         }
 665
 666         __hash_remove(hc);
 667         up_write(&_hash_lock);
 668         param->data_size = 0;
 669         return 0;
 670 }
 671
 672 /*
 673  * Check a string doesn't overrun the chunk of
 674  * memory we copied from userland.
 675  */
 676 static int invalid_str(char *str, void *end)
 677 {
 678         while ((void *) str < end)
 679                 if (!*str++)
 680                         return 0;
 681
 682         return -EINVAL;
 683 }
 684
 685 static int dev_rename(struct dm_ioctl *param, size_t param_size)
 686 {
 687         int r;
 688         char *new_name = (char *) param + param->data_start;
 689
 690         if (new_name < (char *) (param + 1) ||
 691             invalid_str(new_name, (void *) param + param_size)) {
 692                 DMWARN("Invalid new logical volume name supplied.");
 693                 return -EINVAL;
 694         }
 695
 696         r = check_name(new_name);
 697         if (r)
 698                 return r;
 699
 700         param->data_size = 0;
 701         return dm_hash_rename(param->name, new_name);
 702 }
 703
 704 static int dev_set_geometry(struct dm_ioctl *param, size_t param_size)
 705 {
 706         int r = -EINVAL, x;
 707         struct mapped_device *md;
 708         struct hd_geometry geometry;
 709         unsigned long indata[4];
 710         char *geostr = (char *) param + param->data_start;
 711
 712         md = find_device(param);
 713         if (!md)
 714                 return -ENXIO;
 715
 716         if (geostr < (char *) (param + 1) ||
 717             invalid_str(geostr, (void *) param + param_size)) {
 718                 DMWARN("Invalid geometry supplied.");
 719                 goto out;
 720         }
 721
 722         x = sscanf(geostr, "%lu %lu %lu %lu", indata,
 723                    indata + 1, indata + 2, indata + 3);
 724
 725         if (x != 4) {
 726                 DMWARN("Unable to interpret geometry settings.");
 727                 goto out;
 728         }
 729
 730         if (indata[0] > 65535 || indata[1] > 255 ||
 731             indata[2] > 255 || indata[3] > ULONG_MAX) {
 732                 DMWARN("Geometry exceeds range limits.");
 733                 goto out;
 734         }
 735
 736         geometry.cylinders = indata[0];
 737         geometry.heads = indata[1];
 738         geometry.sectors = indata[2];
 739         geometry.start = indata[3];
 740
 741         r = dm_set_geometry(md, &geometry);
 742         if (!r)
 743                 r = __dev_status(md, param);
 744
 745         param->data_size = 0;
 746
 747 out:
 748         dm_put(md);
 749         return r;
 750 }
 751
 752 static int do_suspend(struct dm_ioctl *param)
 753 {
 754         int r = 0;
 755         int do_lockfs = 1;
 756         struct mapped_device *md;
 757
 758         md = find_device(param);
 759         if (!md)
 760                 return -ENXIO;
 761
 762         if (param->flags & DM_SKIP_LOCKFS_FLAG)
 763                 do_lockfs = 0;
 764
 765         if (!dm_suspended(md))
 766                 r = dm_suspend(md, do_lockfs);
 767
 768         if (!r)
 769                 r = __dev_status(md, param);
 770
 771         dm_put(md);
 772         return r;
 773 }
 774
 775 static int do_resume(struct dm_ioctl *param)
 776 {
 777         int r = 0;
 778         int do_lockfs = 1;
 779         struct hash_cell *hc;
 780         struct mapped_device *md;
 781         struct dm_table *new_map;
 782
 783         down_write(&_hash_lock);
 784
 785         hc = __find_device_hash_cell(param);
 786         if (!hc) {
 787                 DMWARN("device doesn't appear to be in the dev hash table.");
 788                 up_write(&_hash_lock);
 789                 return -ENXIO;
 790         }
 791
 792         md = hc->md;
 793         dm_get(md);
 794
 795         new_map = hc->new_map;
 796         hc->new_map = NULL;
 797         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
 798
 799         up_write(&_hash_lock);
 800
 801         /* Do we need to load a new map ? */
 802         if (new_map) {
 803                 /* Suspend if it isn't already suspended */
 804                 if (param->flags & DM_SKIP_LOCKFS_FLAG)
 805                         do_lockfs = 0;
 806                 if (!dm_suspended(md))
 807                         dm_suspend(md, do_lockfs);
 808
 809                 r = dm_swap_table(md, new_map);
 810                 if (r) {
 811                         dm_put(md);
 812                         dm_table_put(new_map);
 813                         return r;
 814                 }
 815
 816                 if (dm_table_get_mode(new_map) & FMODE_WRITE)
 817                         set_disk_ro(dm_disk(md), 0);
 818                 else
 819                         set_disk_ro(dm_disk(md), 1);
 820
 821                 dm_table_put(new_map);
 822         }
 823
 824         if (dm_suspended(md))
 825                 r = dm_resume(md);
 826
 827         if (!r)
 828                 r = __dev_status(md, param);
 829
 830         dm_put(md);
 831         return r;
 832 }
 833
 834 /*
 835  * Set or unset the suspension state of a device.
 836  * If the device already is in the requested state we just return its status.
 837  */
 838 static int dev_suspend(struct dm_ioctl *param, size_t param_size)
 839 {
 840         if (param->flags & DM_SUSPEND_FLAG)
 841                 return do_suspend(param);
 842
 843         return do_resume(param);
 844 }
 845
 846 /*
 847  * Copies device info back to user space, used by
 848  * the create and info ioctls.
 849  */
 850 static int dev_status(struct dm_ioctl *param, size_t param_size)
 851 {
 852         int r;
 853         struct mapped_device *md;
 854
 855         md = find_device(param);
 856         if (!md)
 857                 return -ENXIO;
 858
 859         r = __dev_status(md, param);
 860         dm_put(md);
 861         return r;
 862 }
 863
 864 /*
 865  * Build up the status struct for each target
 866  */
 867 static void retrieve_status(struct dm_table *table,
 868                             struct dm_ioctl *param, size_t param_size)
 869 {
 870         unsigned int i, num_targets;
 871         struct dm_target_spec *spec;
 872         char *outbuf, *outptr;
 873         status_type_t type;
 874         size_t remaining, len, used = 0;
 875
 876         outptr = outbuf = get_result_buffer(param, param_size, &len);
 877
 878         if (param->flags & DM_STATUS_TABLE_FLAG)
 879                 type = STATUSTYPE_TABLE;
 880         else
 881                 type = STATUSTYPE_INFO;
 882
 883         /* Get all the target info */
 884         num_targets = dm_table_get_num_targets(table);
 885         for (i = 0; i < num_targets; i++) {
 886                 struct dm_target *ti = dm_table_get_target(table, i);
 887
 888                 remaining = len - (outptr - outbuf);
 889                 if (remaining <= sizeof(struct dm_target_spec)) {
 890                         param->flags |= DM_BUFFER_FULL_FLAG;
 891                         break;
 892                 }
 893
 894                 spec = (struct dm_target_spec *) outptr;
 895
 896                 spec->status = 0;
 897                 spec->sector_start = ti->begin;
 898                 spec->length = ti->len;
 899                 strncpy(spec->target_type, ti->type->name,
 900                         sizeof(spec->target_type));
 901
 902                 outptr += sizeof(struct dm_target_spec);
 903                 remaining = len - (outptr - outbuf);
 904                 if (remaining <= 0) {
 905                         param->flags |= DM_BUFFER_FULL_FLAG;
 906                         break;
 907                 }
 908
 909                 /* Get the status/table string from the target driver */
 910                 if (ti->type->status) {
 911                         if (ti->type->status(ti, type, outptr, remaining)) {
 912                                 param->flags |= DM_BUFFER_FULL_FLAG;
 913                                 break;
 914                         }
 915                 } else
 916                         outptr[0] = '\0';
 917
 918                 outptr += strlen(outptr) + 1;
 919                 used = param->data_start + (outptr - outbuf);
 920
 921                 outptr = align_ptr(outptr);
 922                 spec->next = outptr - outbuf;
 923         }
 924
 925         if (used)
 926                 param->data_size = used;
 927
 928         param->target_count = num_targets;
 929 }
 930
 931 /*
 932  * Wait for a device to report an event
 933  */
 934 static int dev_wait(struct dm_ioctl *param, size_t param_size)
 935 {
 936         int r;
 937         struct mapped_device *md;
 938         struct dm_table *table;
 939
 940         md = find_device(param);
 941         if (!md)
 942                 return -ENXIO;
 943
 944         /*
 945          * Wait for a notification event
 946          */
 947         if (dm_wait_event(md, param->event_nr)) {
 948                 r = -ERESTARTSYS;
 949                 goto out;
 950         }
 951
 952         /*
 953          * The userland program is going to want to know what
 954          * changed to trigger the event, so we may as well tell
 955          * him and save an ioctl.
 956          */
 957         r = __dev_status(md, param);
 958         if (r)
 959                 goto out;
 960
 961         table = dm_get_table(md);
 962         if (table) {
 963                 retrieve_status(table, param, param_size);
 964                 dm_table_put(table);
 965         }
 966
 967  out:
 968         dm_put(md);
 969         return r;
 970 }
 971
 972 static inline int get_mode(struct dm_ioctl *param)
 973 {
 974         int mode = FMODE_READ | FMODE_WRITE;
 975
 976         if (param->flags & DM_READONLY_FLAG)
 977                 mode = FMODE_READ;
 978
 979         return mode;
 980 }
 981
 982 static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
 983                        struct dm_target_spec **spec, char **target_params)
 984 {
 985         *spec = (struct dm_target_spec *) ((unsigned char *) last + next);
 986         *target_params = (char *) (*spec + 1);
 987
 988         if (*spec < (last + 1))
 989                 return -EINVAL;
 990
 991         return invalid_str(*target_params, end);
 992 }
 993
 994 static int populate_table(struct dm_table *table,
 995                           struct dm_ioctl *param, size_t param_size)
 996 {
 997         int r;
 998         unsigned int i = 0;
 999         struct dm_target_spec *spec = (struct dm_target_spec *) param;
1000         uint32_t next = param->data_start;
1001         void *end = (void *) param + param_size;
1002         char *target_params;
1003
1004         if (!param->target_count) {
1005                 DMWARN("populate_table: no targets specified");
1006                 return -EINVAL;
1007         }
1008
1009         for (i = 0; i < param->target_count; i++) {
1010
1011                 r = next_target(spec, next, end, &spec, &target_params);
1012                 if (r) {
1013                         DMWARN("unable to find target");
1014                         return r;
1015                 }
1016
1017                 r = dm_table_add_target(table, spec->target_type,
1018                                         (sector_t) spec->sector_start,
1019                                         (sector_t) spec->length,
1020                                         target_params);
1021                 if (r) {
1022                         DMWARN("error adding target to table");
1023                         return r;
1024                 }
1025
1026                 next = spec->next;
1027         }
1028
1029         return dm_table_complete(table);
1030 }
1031
1032 static int table_load(struct dm_ioctl *param, size_t param_size)
1033 {
1034         int r;
1035         struct hash_cell *hc;
1036         struct dm_table *t;
1037         struct mapped_device *md;
1038
1039         md = find_device(param);
1040         if (!md)
1041                 return -ENXIO;
1042
1043         r = dm_table_create(&t, get_mode(param), param->target_count, md);
1044         if (r)
1045                 goto out;
1046
1047         r = populate_table(t, param, param_size);
1048         if (r) {
1049                 dm_table_put(t);
1050                 goto out;
1051         }
1052
1053         down_write(&_hash_lock);
1054         hc = dm_get_mdptr(md);
1055         if (!hc || hc->md != md) {
1056                 DMWARN("device has been removed from the dev hash table.");
1057                 dm_table_put(t);
1058                 up_write(&_hash_lock);
1059                 r = -ENXIO;
1060                 goto out;
1061         }
1062
1063         if (hc->new_map)
1064                 dm_table_put(hc->new_map);
1065         hc->new_map = t;
1066         up_write(&_hash_lock);
1067
1068         param->flags |= DM_INACTIVE_PRESENT_FLAG;
1069         r = __dev_status(md, param);
1070
1071 out:
1072         dm_put(md);
1073
1074         return r;
1075 }
1076
1077 static int table_clear(struct dm_ioctl *param, size_t param_size)
1078 {
1079         int r;
1080         struct hash_cell *hc;
1081
1082         down_write(&_hash_lock);
1083
1084         hc = __find_device_hash_cell(param);
1085         if (!hc) {
1086                 DMWARN("device doesn't appear to be in the dev hash table.");
1087                 up_write(&_hash_lock);
1088                 return -ENXIO;
1089         }
1090
1091         if (hc->new_map) {
1092                 dm_table_put(hc->new_map);
1093                 hc->new_map = NULL;
1094         }
1095
1096         param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1097
1098         r = __dev_status(hc->md, param);
1099         up_write(&_hash_lock);
1100         return r;
1101 }
1102
1103 /*
1104  * Retrieves a list of devices used by a particular dm device.
1105  */
1106 static void retrieve_deps(struct dm_table *table,
1107                           struct dm_ioctl *param, size_t param_size)
1108 {
1109         unsigned int count = 0;
1110         struct list_head *tmp;
1111         size_t len, needed;
1112         struct dm_dev *dd;
1113         struct dm_target_deps *deps;
1114
1115         deps = get_result_buffer(param, param_size, &len);
1116
1117         /*
1118          * Count the devices.
1119          */
1120         list_for_each (tmp, dm_table_get_devices(table))
1121                 count++;
1122
1123         /*
1124          * Check we have enough space.
1125          */
1126         needed = sizeof(*deps) + (sizeof(*deps->dev) * count);
1127         if (len < needed) {
1128                 param->flags |= DM_BUFFER_FULL_FLAG;
1129                 return;
1130         }
1131
1132         /*
1133          * Fill in the devices.
1134          */
1135         deps->count = count;
1136         count = 0;
1137         list_for_each_entry (dd, dm_table_get_devices(table), list)
1138                 deps->dev[count++] = huge_encode_dev(dd->bdev->bd_dev);
1139
1140         param->data_size = param->data_start + needed;
1141 }
1142
1143 static int table_deps(struct dm_ioctl *param, size_t param_size)
1144 {
1145         int r = 0;
1146         struct mapped_device *md;
1147         struct dm_table *table;
1148
1149         md = find_device(param);
1150         if (!md)
1151                 return -ENXIO;
1152
1153         r = __dev_status(md, param);
1154         if (r)
1155                 goto out;
1156
1157         table = dm_get_table(md);
1158         if (table) {
1159                 retrieve_deps(table, param, param_size);
1160                 dm_table_put(table);
1161         }
1162
1163  out:
1164         dm_put(md);
1165         return r;
1166 }
1167
1168 /*
1169  * Return the status of a device as a text string for each
1170  * target.
1171  */
1172 static int table_status(struct dm_ioctl *param, size_t param_size)
1173 {
1174         int r;
1175         struct mapped_device *md;
1176         struct dm_table *table;
1177
1178         md = find_device(param);
1179         if (!md)
1180                 return -ENXIO;
1181
1182         r = __dev_status(md, param);
1183         if (r)
1184                 goto out;
1185
1186         table = dm_get_table(md);
1187         if (table) {
1188                 retrieve_status(table, param, param_size);
1189                 dm_table_put(table);
1190         }
1191
1192  out:
1193         dm_put(md);
1194         return r;
1195 }
1196
1197 /*
1198  * Pass a message to the target that's at the supplied device offset.
1199  */
1200 static int target_message(struct dm_ioctl *param, size_t param_size)
1201 {
1202         int r, argc;
1203         char **argv;
1204         struct mapped_device *md;
1205         struct dm_table *table;
1206         struct dm_target *ti;
1207         struct dm_target_msg *tmsg = (void *) param + param->data_start;
1208
1209         md = find_device(param);
1210         if (!md)
1211                 return -ENXIO;
1212
1213         r = __dev_status(md, param);
1214         if (r)
1215                 goto out;
1216
1217         if (tmsg < (struct dm_target_msg *) (param + 1) ||
1218             invalid_str(tmsg->message, (void *) param + param_size)) {
1219                 DMWARN("Invalid target message parameters.");
1220                 r = -EINVAL;
1221                 goto out;
1222         }
1223
1224         r = dm_split_args(&argc, &argv, tmsg->message);
1225         if (r) {
1226                 DMWARN("Failed to split target message parameters");
1227                 goto out;
1228         }
1229
1230         table = dm_get_table(md);
1231         if (!table)
1232                 goto out_argv;
1233
1234         if (tmsg->sector >= dm_table_get_size(table)) {
1235                 DMWARN("Target message sector outside device.");
1236                 r = -EINVAL;
1237                 goto out_table;
1238         }
1239
1240         ti = dm_table_find_target(table, tmsg->sector);
1241         if (ti->type->message)
1242                 r = ti->type->message(ti, argc, argv);
1243         else {
1244                 DMWARN("Target type does not support messages");
1245                 r = -EINVAL;
1246         }
1247
1248  out_table:
1249         dm_table_put(table);
1250  out_argv:
1251         kfree(argv);
1252  out:
1253         param->data_size = 0;
1254         dm_put(md);
1255         return r;
1256 }
1257
1258 /*-----------------------------------------------------------------
1259  * Implementation of open/close/ioctl on the special char
1260  * device.
1261  *---------------------------------------------------------------*/
1262 static ioctl_fn lookup_ioctl(unsigned int cmd)
1263 {
1264         static struct {
1265                 int cmd;
1266                 ioctl_fn fn;
1267         } _ioctls[] = {
1268                 {DM_VERSION_CMD, NULL}, /* version is dealt with elsewhere */
1269                 {DM_REMOVE_ALL_CMD, remove_all},
1270                 {DM_LIST_DEVICES_CMD, list_devices},
1271
1272                 {DM_DEV_CREATE_CMD, dev_create},
1273                 {DM_DEV_REMOVE_CMD, dev_remove},
1274                 {DM_DEV_RENAME_CMD, dev_rename},
1275                 {DM_DEV_SUSPEND_CMD, dev_suspend},
1276                 {DM_DEV_STATUS_CMD, dev_status},
1277                 {DM_DEV_WAIT_CMD, dev_wait},
1278
1279                 {DM_TABLE_LOAD_CMD, table_load},
1280                 {DM_TABLE_CLEAR_CMD, table_clear},
1281                 {DM_TABLE_DEPS_CMD, table_deps},
1282                 {DM_TABLE_STATUS_CMD, table_status},
1283
1284                 {DM_LIST_VERSIONS_CMD, list_versions},
1285
1286                 {DM_TARGET_MSG_CMD, target_message},
1287                 {DM_DEV_SET_GEOMETRY_CMD, dev_set_geometry}
1288         };
1289
1290         return (cmd >= ARRAY_SIZE(_ioctls)) ? NULL : _ioctls[cmd].fn;
1291 }
1292
1293 /*
1294  * As well as checking the version compatibility this always
1295  * copies the kernel interface version out.
1296  */
1297 static int check_version(unsigned int cmd, struct dm_ioctl __user *user)
1298 {
1299         uint32_t version[3];
1300         int r = 0;
1301
1302         if (copy_from_user(version, user->version, sizeof(version)))
1303                 return -EFAULT;
1304
1305         if ((DM_VERSION_MAJOR != version[0]) ||
1306             (DM_VERSION_MINOR < version[1])) {
1307                 DMWARN("ioctl interface mismatch: "
1308                        "kernel(%u.%u.%u), user(%u.%u.%u), cmd(%d)",
1309                        DM_VERSION_MAJOR, DM_VERSION_MINOR,
1310                        DM_VERSION_PATCHLEVEL,
1311                        version[0], version[1], version[2], cmd);
1312                 r = -EINVAL;
1313         }
1314
1315         /*
1316          * Fill in the kernel version.
1317          */
1318         version[0] = DM_VERSION_MAJOR;
1319         version[1] = DM_VERSION_MINOR;
1320         version[2] = DM_VERSION_PATCHLEVEL;
1321         if (copy_to_user(user->version, version, sizeof(version)))
1322                 return -EFAULT;
1323
1324         return r;
1325 }
1326
1327 static void free_params(struct dm_ioctl *param)
1328 {
1329         vfree(param);
1330 }
1331
1332 static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl **param)
1333 {
1334         struct dm_ioctl tmp, *dmi;
1335
1336         if (copy_from_user(&tmp, user, sizeof(tmp)))
1337                 return -EFAULT;
1338
1339         if (tmp.data_size < sizeof(tmp))
1340                 return -EINVAL;
1341
1342         dmi = (struct dm_ioctl *) vmalloc(tmp.data_size);
1343         if (!dmi)
1344                 return -ENOMEM;
1345
1346         if (copy_from_user(dmi, user, tmp.data_size)) {
1347                 vfree(dmi);
1348                 return -EFAULT;
1349         }
1350
1351         *param = dmi;
1352         return 0;
1353 }
1354
1355 static int validate_params(uint cmd, struct dm_ioctl *param)
1356 {
1357         /* Always clear this flag */
1358         param->flags &= ~DM_BUFFER_FULL_FLAG;
1359
1360         /* Ignores parameters */
1361         if (cmd == DM_REMOVE_ALL_CMD ||
1362             cmd == DM_LIST_DEVICES_CMD ||
1363             cmd == DM_LIST_VERSIONS_CMD)
1364                 return 0;
1365
1366         if ((cmd == DM_DEV_CREATE_CMD)) {
1367                 if (!*param->name) {
1368                         DMWARN("name not supplied when creating device");
1369                         return -EINVAL;
1370                 }
1371         } else if ((*param->uuid && *param->name)) {
1372                 DMWARN("only supply one of name or uuid, cmd(%u)", cmd);
1373                 return -EINVAL;
1374         }
1375
1376         /* Ensure strings are terminated */
1377         param->name[DM_NAME_LEN - 1] = '\0';
1378         param->uuid[DM_UUID_LEN - 1] = '\0';
1379
1380         return 0;
1381 }
1382
1383 static int ctl_ioctl(struct inode *inode, struct file *file,
1384                      uint command, ulong u)
1385 {
1386         int r = 0;
1387         unsigned int cmd;
1388         struct dm_ioctl *param;
1389         struct dm_ioctl __user *user = (struct dm_ioctl __user *) u;
1390         ioctl_fn fn = NULL;
1391         size_t param_size;
1392
1393         /* only root can play with this */
1394         if (!capable(CAP_SYS_ADMIN))
1395                 return -EACCES;
1396
1397         if (_IOC_TYPE(command) != DM_IOCTL)
1398                 return -ENOTTY;
1399
1400         cmd = _IOC_NR(command);
1401
1402         /*
1403          * Check the interface version passed in.  This also
1404          * writes out the kernel's interface version.
1405          */
1406         r = check_version(cmd, user);
1407         if (r)
1408                 return r;
1409
1410         /*
1411          * Nothing more to do for the version command.
1412          */
1413         if (cmd == DM_VERSION_CMD)
1414                 return 0;
1415
1416         fn = lookup_ioctl(cmd);
1417         if (!fn) {
1418                 DMWARN("dm_ctl_ioctl: unknown command 0x%x", command);
1419                 return -ENOTTY;
1420         }
1421
1422         /*
1423          * Trying to avoid low memory issues when a device is
1424          * suspended.
1425          */
1426         current->flags |= PF_MEMALLOC;
1427
1428         /*
1429          * Copy the parameters into kernel space.
1430          */
1431         r = copy_params(user, &param);
1432
1433         current->flags &= ~PF_MEMALLOC;
1434
1435         if (r)
1436                 return r;
1437
1438         r = validate_params(cmd, param);
1439         if (r)
1440                 goto out;
1441
1442         param_size = param->data_size;
1443         param->data_size = sizeof(*param);
1444         r = fn(param, param_size);
1445
1446         /*
1447          * Copy the results back to userland.
1448          */
1449         if (!r && copy_to_user(user, param, param->data_size))
1450                 r = -EFAULT;
1451
1452  out:
1453         free_params(param);
1454         return r;
1455 }
1456
1457 static struct file_operations _ctl_fops = {
1458         .ioctl   = ctl_ioctl,
1459         .owner   = THIS_MODULE,
1460 };
1461
1462 static struct miscdevice _dm_misc = {
1463         .minor          = MISC_DYNAMIC_MINOR,
1464         .name           = DM_NAME,
1465         .devfs_name     = "mapper/control",
1466         .fops           = &_ctl_fops
1467 };
1468
1469 /*
1470  * Create misc character device and link to DM_DIR/control.
1471  */
1472 int __init dm_interface_init(void)
1473 {
1474         int r;
1475
1476         r = dm_hash_init();
1477         if (r)
1478                 return r;
1479
1480         r = misc_register(&_dm_misc);
1481         if (r) {
1482                 DMERR("misc_register failed for control device");
1483                 dm_hash_exit();
1484                 return r;
1485         }
1486
1487         DMINFO("%d.%d.%d%s initialised: %s", DM_VERSION_MAJOR,
1488                DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL, DM_VERSION_EXTRA,
1489                DM_DRIVER_EMAIL);
1490         return 0;
1491 }
1492
1493 void dm_interface_exit(void)
1494 {
1495         if (misc_deregister(&_dm_misc) < 0)
1496                 DMERR("misc_deregister failed for control device");
1497
1498         dm_hash_exit();
1499 }