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