linux/drivers/md/dm-ioctl.c
<<
>>
Prefs
   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-core.h"
   9
  10#include <linux/module.h>
  11#include <linux/vmalloc.h>
  12#include <linux/miscdevice.h>
  13#include <linux/sched/mm.h>
  14#include <linux/init.h>
  15#include <linux/wait.h>
  16#include <linux/slab.h>
  17#include <linux/rbtree.h>
  18#include <linux/dm-ioctl.h>
  19#include <linux/hdreg.h>
  20#include <linux/compat.h>
  21
  22#include <linux/uaccess.h>
  23
  24#define DM_MSG_PREFIX "ioctl"
  25#define DM_DRIVER_EMAIL "dm-devel@redhat.com"
  26
  27struct dm_file {
  28        /*
  29         * poll will wait until the global event number is greater than
  30         * this value.
  31         */
  32        volatile unsigned global_event_nr;
  33};
  34
  35/*-----------------------------------------------------------------
  36 * The ioctl interface needs to be able to look up devices by
  37 * name or uuid.
  38 *---------------------------------------------------------------*/
  39struct hash_cell {
  40        struct rb_node name_node;
  41        struct rb_node uuid_node;
  42        bool name_set;
  43        bool uuid_set;
  44
  45        char *name;
  46        char *uuid;
  47        struct mapped_device *md;
  48        struct dm_table *new_map;
  49};
  50
  51struct vers_iter {
  52    size_t param_size;
  53    struct dm_target_versions *vers, *old_vers;
  54    char *end;
  55    uint32_t flags;
  56};
  57
  58
  59static struct rb_root name_rb_tree = RB_ROOT;
  60static struct rb_root uuid_rb_tree = RB_ROOT;
  61
  62static void dm_hash_remove_all(bool keep_open_devices, bool mark_deferred, bool only_deferred);
  63
  64/*
  65 * Guards access to both hash tables.
  66 */
  67static DECLARE_RWSEM(_hash_lock);
  68
  69/*
  70 * Protects use of mdptr to obtain hash cell name and uuid from mapped device.
  71 */
  72static DEFINE_MUTEX(dm_hash_cells_mutex);
  73
  74static void dm_hash_exit(void)
  75{
  76        dm_hash_remove_all(false, false, false);
  77}
  78
  79/*-----------------------------------------------------------------
  80 * Code for looking up a device by name
  81 *---------------------------------------------------------------*/
  82static struct hash_cell *__get_name_cell(const char *str)
  83{
  84        struct rb_node *n = name_rb_tree.rb_node;
  85
  86        while (n) {
  87                struct hash_cell *hc = container_of(n, struct hash_cell, name_node);
  88                int c = strcmp(hc->name, str);
  89                if (!c) {
  90                        dm_get(hc->md);
  91                        return hc;
  92                }
  93                n = c >= 0 ? n->rb_left : n->rb_right;
  94        }
  95
  96        return NULL;
  97}
  98
  99static struct hash_cell *__get_uuid_cell(const char *str)
 100{
 101        struct rb_node *n = uuid_rb_tree.rb_node;
 102
 103        while (n) {
 104                struct hash_cell *hc = container_of(n, struct hash_cell, uuid_node);
 105                int c = strcmp(hc->uuid, str);
 106                if (!c) {
 107                        dm_get(hc->md);
 108                        return hc;
 109                }
 110                n = c >= 0 ? n->rb_left : n->rb_right;
 111        }
 112
 113        return NULL;
 114}
 115
 116static void __unlink_name(struct hash_cell *hc)
 117{
 118        if (hc->name_set) {
 119                hc->name_set = false;
 120                rb_erase(&hc->name_node, &name_rb_tree);
 121        }
 122}
 123
 124static void __unlink_uuid(struct hash_cell *hc)
 125{
 126        if (hc->uuid_set) {
 127                hc->uuid_set = false;
 128                rb_erase(&hc->uuid_node, &uuid_rb_tree);
 129        }
 130}
 131
 132static void __link_name(struct hash_cell *new_hc)
 133{
 134        struct rb_node **n, *parent;
 135
 136        __unlink_name(new_hc);
 137
 138        new_hc->name_set = true;
 139
 140        n = &name_rb_tree.rb_node;
 141        parent = NULL;
 142
 143        while (*n) {
 144                struct hash_cell *hc = container_of(*n, struct hash_cell, name_node);
 145                int c = strcmp(hc->name, new_hc->name);
 146                BUG_ON(!c);
 147                parent = *n;
 148                n = c >= 0 ? &hc->name_node.rb_left : &hc->name_node.rb_right;
 149        }
 150
 151        rb_link_node(&new_hc->name_node, parent, n);
 152        rb_insert_color(&new_hc->name_node, &name_rb_tree);
 153}
 154
 155static void __link_uuid(struct hash_cell *new_hc)
 156{
 157        struct rb_node **n, *parent;
 158
 159        __unlink_uuid(new_hc);
 160
 161        new_hc->uuid_set = true;
 162
 163        n = &uuid_rb_tree.rb_node;
 164        parent = NULL;
 165
 166        while (*n) {
 167                struct hash_cell *hc = container_of(*n, struct hash_cell, uuid_node);
 168                int c = strcmp(hc->uuid, new_hc->uuid);
 169                BUG_ON(!c);
 170                parent = *n;
 171                n = c > 0 ? &hc->uuid_node.rb_left : &hc->uuid_node.rb_right;
 172        }
 173
 174        rb_link_node(&new_hc->uuid_node, parent, n);
 175        rb_insert_color(&new_hc->uuid_node, &uuid_rb_tree);
 176}
 177
 178static struct hash_cell *__get_dev_cell(uint64_t dev)
 179{
 180        struct mapped_device *md;
 181        struct hash_cell *hc;
 182
 183        md = dm_get_md(huge_decode_dev(dev));
 184        if (!md)
 185                return NULL;
 186
 187        hc = dm_get_mdptr(md);
 188        if (!hc) {
 189                dm_put(md);
 190                return NULL;
 191        }
 192
 193        return hc;
 194}
 195
 196/*-----------------------------------------------------------------
 197 * Inserting, removing and renaming a device.
 198 *---------------------------------------------------------------*/
 199static struct hash_cell *alloc_cell(const char *name, const char *uuid,
 200                                    struct mapped_device *md)
 201{
 202        struct hash_cell *hc;
 203
 204        hc = kmalloc(sizeof(*hc), GFP_KERNEL);
 205        if (!hc)
 206                return NULL;
 207
 208        hc->name = kstrdup(name, GFP_KERNEL);
 209        if (!hc->name) {
 210                kfree(hc);
 211                return NULL;
 212        }
 213
 214        if (!uuid)
 215                hc->uuid = NULL;
 216
 217        else {
 218                hc->uuid = kstrdup(uuid, GFP_KERNEL);
 219                if (!hc->uuid) {
 220                        kfree(hc->name);
 221                        kfree(hc);
 222                        return NULL;
 223                }
 224        }
 225
 226        hc->name_set = hc->uuid_set = false;
 227        hc->md = md;
 228        hc->new_map = NULL;
 229        return hc;
 230}
 231
 232static void free_cell(struct hash_cell *hc)
 233{
 234        if (hc) {
 235                kfree(hc->name);
 236                kfree(hc->uuid);
 237                kfree(hc);
 238        }
 239}
 240
 241/*
 242 * The kdev_t and uuid of a device can never change once it is
 243 * initially inserted.
 244 */
 245static int dm_hash_insert(const char *name, const char *uuid, struct mapped_device *md)
 246{
 247        struct hash_cell *cell, *hc;
 248
 249        /*
 250         * Allocate the new cells.
 251         */
 252        cell = alloc_cell(name, uuid, md);
 253        if (!cell)
 254                return -ENOMEM;
 255
 256        /*
 257         * Insert the cell into both hash tables.
 258         */
 259        down_write(&_hash_lock);
 260        hc = __get_name_cell(name);
 261        if (hc) {
 262                dm_put(hc->md);
 263                goto bad;
 264        }
 265
 266        __link_name(cell);
 267
 268        if (uuid) {
 269                hc = __get_uuid_cell(uuid);
 270                if (hc) {
 271                        __unlink_name(cell);
 272                        dm_put(hc->md);
 273                        goto bad;
 274                }
 275                __link_uuid(cell);
 276        }
 277        dm_get(md);
 278        mutex_lock(&dm_hash_cells_mutex);
 279        dm_set_mdptr(md, cell);
 280        mutex_unlock(&dm_hash_cells_mutex);
 281        up_write(&_hash_lock);
 282
 283        return 0;
 284
 285 bad:
 286        up_write(&_hash_lock);
 287        free_cell(cell);
 288        return -EBUSY;
 289}
 290
 291static struct dm_table *__hash_remove(struct hash_cell *hc)
 292{
 293        struct dm_table *table;
 294        int srcu_idx;
 295
 296        /* remove from the dev trees */
 297        __unlink_name(hc);
 298        __unlink_uuid(hc);
 299        mutex_lock(&dm_hash_cells_mutex);
 300        dm_set_mdptr(hc->md, NULL);
 301        mutex_unlock(&dm_hash_cells_mutex);
 302
 303        table = dm_get_live_table(hc->md, &srcu_idx);
 304        if (table)
 305                dm_table_event(table);
 306        dm_put_live_table(hc->md, srcu_idx);
 307
 308        table = NULL;
 309        if (hc->new_map)
 310                table = hc->new_map;
 311        dm_put(hc->md);
 312        free_cell(hc);
 313
 314        return table;
 315}
 316
 317static void dm_hash_remove_all(bool keep_open_devices, bool mark_deferred, bool only_deferred)
 318{
 319        int dev_skipped;
 320        struct rb_node *n;
 321        struct hash_cell *hc;
 322        struct mapped_device *md;
 323        struct dm_table *t;
 324
 325retry:
 326        dev_skipped = 0;
 327
 328        down_write(&_hash_lock);
 329
 330        for (n = rb_first(&name_rb_tree); n; n = rb_next(n)) {
 331                hc = container_of(n, struct hash_cell, name_node);
 332                md = hc->md;
 333                dm_get(md);
 334
 335                if (keep_open_devices &&
 336                    dm_lock_for_deletion(md, mark_deferred, only_deferred)) {
 337                        dm_put(md);
 338                        dev_skipped++;
 339                        continue;
 340                }
 341
 342                t = __hash_remove(hc);
 343
 344                up_write(&_hash_lock);
 345
 346                if (t) {
 347                        dm_sync_table(md);
 348                        dm_table_destroy(t);
 349                }
 350                dm_put(md);
 351                if (likely(keep_open_devices))
 352                        dm_destroy(md);
 353                else
 354                        dm_destroy_immediate(md);
 355
 356                /*
 357                 * Some mapped devices may be using other mapped
 358                 * devices, so repeat until we make no further
 359                 * progress.  If a new mapped device is created
 360                 * here it will also get removed.
 361                 */
 362                goto retry;
 363        }
 364
 365        up_write(&_hash_lock);
 366
 367        if (dev_skipped)
 368                DMWARN("remove_all left %d open device(s)", dev_skipped);
 369}
 370
 371/*
 372 * Set the uuid of a hash_cell that isn't already set.
 373 */
 374static void __set_cell_uuid(struct hash_cell *hc, char *new_uuid)
 375{
 376        mutex_lock(&dm_hash_cells_mutex);
 377        hc->uuid = new_uuid;
 378        mutex_unlock(&dm_hash_cells_mutex);
 379
 380        __link_uuid(hc);
 381}
 382
 383/*
 384 * Changes the name of a hash_cell and returns the old name for
 385 * the caller to free.
 386 */
 387static char *__change_cell_name(struct hash_cell *hc, char *new_name)
 388{
 389        char *old_name;
 390
 391        /*
 392         * Rename and move the name cell.
 393         */
 394        __unlink_name(hc);
 395        old_name = hc->name;
 396
 397        mutex_lock(&dm_hash_cells_mutex);
 398        hc->name = new_name;
 399        mutex_unlock(&dm_hash_cells_mutex);
 400
 401        __link_name(hc);
 402
 403        return old_name;
 404}
 405
 406static struct mapped_device *dm_hash_rename(struct dm_ioctl *param,
 407                                            const char *new)
 408{
 409        char *new_data, *old_name = NULL;
 410        struct hash_cell *hc;
 411        struct dm_table *table;
 412        struct mapped_device *md;
 413        unsigned change_uuid = (param->flags & DM_UUID_FLAG) ? 1 : 0;
 414        int srcu_idx;
 415
 416        /*
 417         * duplicate new.
 418         */
 419        new_data = kstrdup(new, GFP_KERNEL);
 420        if (!new_data)
 421                return ERR_PTR(-ENOMEM);
 422
 423        down_write(&_hash_lock);
 424
 425        /*
 426         * Is new free ?
 427         */
 428        if (change_uuid)
 429                hc = __get_uuid_cell(new);
 430        else
 431                hc = __get_name_cell(new);
 432
 433        if (hc) {
 434                DMWARN("Unable to change %s on mapped device %s to one that "
 435                       "already exists: %s",
 436                       change_uuid ? "uuid" : "name",
 437                       param->name, new);
 438                dm_put(hc->md);
 439                up_write(&_hash_lock);
 440                kfree(new_data);
 441                return ERR_PTR(-EBUSY);
 442        }
 443
 444        /*
 445         * Is there such a device as 'old' ?
 446         */
 447        hc = __get_name_cell(param->name);
 448        if (!hc) {
 449                DMWARN("Unable to rename non-existent device, %s to %s%s",
 450                       param->name, change_uuid ? "uuid " : "", new);
 451                up_write(&_hash_lock);
 452                kfree(new_data);
 453                return ERR_PTR(-ENXIO);
 454        }
 455
 456        /*
 457         * Does this device already have a uuid?
 458         */
 459        if (change_uuid && hc->uuid) {
 460                DMWARN("Unable to change uuid of mapped device %s to %s "
 461                       "because uuid is already set to %s",
 462                       param->name, new, hc->uuid);
 463                dm_put(hc->md);
 464                up_write(&_hash_lock);
 465                kfree(new_data);
 466                return ERR_PTR(-EINVAL);
 467        }
 468
 469        if (change_uuid)
 470                __set_cell_uuid(hc, new_data);
 471        else
 472                old_name = __change_cell_name(hc, new_data);
 473
 474        /*
 475         * Wake up any dm event waiters.
 476         */
 477        table = dm_get_live_table(hc->md, &srcu_idx);
 478        if (table)
 479                dm_table_event(table);
 480        dm_put_live_table(hc->md, srcu_idx);
 481
 482        if (!dm_kobject_uevent(hc->md, KOBJ_CHANGE, param->event_nr))
 483                param->flags |= DM_UEVENT_GENERATED_FLAG;
 484
 485        md = hc->md;
 486        up_write(&_hash_lock);
 487        kfree(old_name);
 488
 489        return md;
 490}
 491
 492void dm_deferred_remove(void)
 493{
 494        dm_hash_remove_all(true, false, true);
 495}
 496
 497/*-----------------------------------------------------------------
 498 * Implementation of the ioctl commands
 499 *---------------------------------------------------------------*/
 500/*
 501 * All the ioctl commands get dispatched to functions with this
 502 * prototype.
 503 */
 504typedef int (*ioctl_fn)(struct file *filp, struct dm_ioctl *param, size_t param_size);
 505
 506static int remove_all(struct file *filp, struct dm_ioctl *param, size_t param_size)
 507{
 508        dm_hash_remove_all(true, !!(param->flags & DM_DEFERRED_REMOVE), false);
 509        param->data_size = 0;
 510        return 0;
 511}
 512
 513/*
 514 * Round up the ptr to an 8-byte boundary.
 515 */
 516#define ALIGN_MASK 7
 517static inline size_t align_val(size_t val)
 518{
 519        return (val + ALIGN_MASK) & ~ALIGN_MASK;
 520}
 521static inline void *align_ptr(void *ptr)
 522{
 523        return (void *)align_val((size_t)ptr);
 524}
 525
 526/*
 527 * Retrieves the data payload buffer from an already allocated
 528 * struct dm_ioctl.
 529 */
 530static void *get_result_buffer(struct dm_ioctl *param, size_t param_size,
 531                               size_t *len)
 532{
 533        param->data_start = align_ptr(param + 1) - (void *) param;
 534
 535        if (param->data_start < param_size)
 536                *len = param_size - param->data_start;
 537        else
 538                *len = 0;
 539
 540        return ((void *) param) + param->data_start;
 541}
 542
 543static bool filter_device(struct hash_cell *hc, const char *pfx_name, const char *pfx_uuid)
 544{
 545        const char *val;
 546        size_t val_len, pfx_len;
 547
 548        val = hc->name;
 549        val_len = strlen(val);
 550        pfx_len = strnlen(pfx_name, DM_NAME_LEN);
 551        if (pfx_len > val_len)
 552                return false;
 553        if (memcmp(val, pfx_name, pfx_len))
 554                return false;
 555
 556        val = hc->uuid ? hc->uuid : "";
 557        val_len = strlen(val);
 558        pfx_len = strnlen(pfx_uuid, DM_UUID_LEN);
 559        if (pfx_len > val_len)
 560                return false;
 561        if (memcmp(val, pfx_uuid, pfx_len))
 562                return false;
 563
 564        return true;
 565}
 566
 567static int list_devices(struct file *filp, struct dm_ioctl *param, size_t param_size)
 568{
 569        struct rb_node *n;
 570        struct hash_cell *hc;
 571        size_t len, needed = 0;
 572        struct gendisk *disk;
 573        struct dm_name_list *orig_nl, *nl, *old_nl = NULL;
 574        uint32_t *event_nr;
 575
 576        down_write(&_hash_lock);
 577
 578        /*
 579         * Loop through all the devices working out how much
 580         * space we need.
 581         */
 582        for (n = rb_first(&name_rb_tree); n; n = rb_next(n)) {
 583                hc = container_of(n, struct hash_cell, name_node);
 584                if (!filter_device(hc, param->name, param->uuid))
 585                        continue;
 586                needed += align_val(offsetof(struct dm_name_list, name) + strlen(hc->name) + 1);
 587                needed += align_val(sizeof(uint32_t) * 2);
 588                if (param->flags & DM_UUID_FLAG && hc->uuid)
 589                        needed += align_val(strlen(hc->uuid) + 1);
 590        }
 591
 592        /*
 593         * Grab our output buffer.
 594         */
 595        nl = orig_nl = get_result_buffer(param, param_size, &len);
 596        if (len < needed || len < sizeof(nl->dev)) {
 597                param->flags |= DM_BUFFER_FULL_FLAG;
 598                goto out;
 599        }
 600        param->data_size = param->data_start + needed;
 601
 602        nl->dev = 0;    /* Flags no data */
 603
 604        /*
 605         * Now loop through filling out the names.
 606         */
 607        for (n = rb_first(&name_rb_tree); n; n = rb_next(n)) {
 608                void *uuid_ptr;
 609                hc = container_of(n, struct hash_cell, name_node);
 610                if (!filter_device(hc, param->name, param->uuid))
 611                        continue;
 612                if (old_nl)
 613                        old_nl->next = (uint32_t) ((void *) nl -
 614                                                   (void *) old_nl);
 615                disk = dm_disk(hc->md);
 616                nl->dev = huge_encode_dev(disk_devt(disk));
 617                nl->next = 0;
 618                strcpy(nl->name, hc->name);
 619
 620                old_nl = nl;
 621                event_nr = align_ptr(nl->name + strlen(hc->name) + 1);
 622                event_nr[0] = dm_get_event_nr(hc->md);
 623                event_nr[1] = 0;
 624                uuid_ptr = align_ptr(event_nr + 2);
 625                if (param->flags & DM_UUID_FLAG) {
 626                        if (hc->uuid) {
 627                                event_nr[1] |= DM_NAME_LIST_FLAG_HAS_UUID;
 628                                strcpy(uuid_ptr, hc->uuid);
 629                                uuid_ptr = align_ptr(uuid_ptr + strlen(hc->uuid) + 1);
 630                        } else {
 631                                event_nr[1] |= DM_NAME_LIST_FLAG_DOESNT_HAVE_UUID;
 632                        }
 633                }
 634                nl = uuid_ptr;
 635        }
 636        /*
 637         * If mismatch happens, security may be compromised due to buffer
 638         * overflow, so it's better to crash.
 639         */
 640        BUG_ON((char *)nl - (char *)orig_nl != needed);
 641
 642 out:
 643        up_write(&_hash_lock);
 644        return 0;
 645}
 646
 647static void list_version_get_needed(struct target_type *tt, void *needed_param)
 648{
 649    size_t *needed = needed_param;
 650
 651    *needed += sizeof(struct dm_target_versions);
 652    *needed += strlen(tt->name);
 653    *needed += ALIGN_MASK;
 654}
 655
 656static void list_version_get_info(struct target_type *tt, void *param)
 657{
 658    struct vers_iter *info = param;
 659
 660    /* Check space - it might have changed since the first iteration */
 661    if ((char *)info->vers + sizeof(tt->version) + strlen(tt->name) + 1 >
 662        info->end) {
 663
 664        info->flags = DM_BUFFER_FULL_FLAG;
 665        return;
 666    }
 667
 668    if (info->old_vers)
 669        info->old_vers->next = (uint32_t) ((void *)info->vers -
 670                                           (void *)info->old_vers);
 671    info->vers->version[0] = tt->version[0];
 672    info->vers->version[1] = tt->version[1];
 673    info->vers->version[2] = tt->version[2];
 674    info->vers->next = 0;
 675    strcpy(info->vers->name, tt->name);
 676
 677    info->old_vers = info->vers;
 678    info->vers = align_ptr(((void *) ++info->vers) + strlen(tt->name) + 1);
 679}
 680
 681static int __list_versions(struct dm_ioctl *param, size_t param_size, const char *name)
 682{
 683        size_t len, needed = 0;
 684        struct dm_target_versions *vers;
 685        struct vers_iter iter_info;
 686        struct target_type *tt = NULL;
 687
 688        if (name) {
 689                tt = dm_get_target_type(name);
 690                if (!tt)
 691                        return -EINVAL;
 692        }
 693
 694        /*
 695         * Loop through all the devices working out how much
 696         * space we need.
 697         */
 698        if (!tt)
 699                dm_target_iterate(list_version_get_needed, &needed);
 700        else
 701                list_version_get_needed(tt, &needed);
 702
 703        /*
 704         * Grab our output buffer.
 705         */
 706        vers = get_result_buffer(param, param_size, &len);
 707        if (len < needed) {
 708                param->flags |= DM_BUFFER_FULL_FLAG;
 709                goto out;
 710        }
 711        param->data_size = param->data_start + needed;
 712
 713        iter_info.param_size = param_size;
 714        iter_info.old_vers = NULL;
 715        iter_info.vers = vers;
 716        iter_info.flags = 0;
 717        iter_info.end = (char *)vers+len;
 718
 719        /*
 720         * Now loop through filling out the names & versions.
 721         */
 722        if (!tt)
 723                dm_target_iterate(list_version_get_info, &iter_info);
 724        else
 725                list_version_get_info(tt, &iter_info);
 726        param->flags |= iter_info.flags;
 727
 728 out:
 729        if (tt)
 730                dm_put_target_type(tt);
 731        return 0;
 732}
 733
 734static int list_versions(struct file *filp, struct dm_ioctl *param, size_t param_size)
 735{
 736        return __list_versions(param, param_size, NULL);
 737}
 738
 739static int get_target_version(struct file *filp, struct dm_ioctl *param, size_t param_size)
 740{
 741        return __list_versions(param, param_size, param->name);
 742}
 743
 744static int check_name(const char *name)
 745{
 746        if (strchr(name, '/')) {
 747                DMWARN("invalid device name");
 748                return -EINVAL;
 749        }
 750
 751        return 0;
 752}
 753
 754/*
 755 * On successful return, the caller must not attempt to acquire
 756 * _hash_lock without first calling dm_put_live_table, because dm_table_destroy
 757 * waits for this dm_put_live_table and could be called under this lock.
 758 */
 759static struct dm_table *dm_get_inactive_table(struct mapped_device *md, int *srcu_idx)
 760{
 761        struct hash_cell *hc;
 762        struct dm_table *table = NULL;
 763
 764        /* increment rcu count, we don't care about the table pointer */
 765        dm_get_live_table(md, srcu_idx);
 766
 767        down_read(&_hash_lock);
 768        hc = dm_get_mdptr(md);
 769        if (!hc || hc->md != md) {
 770                DMWARN("device has been removed from the dev hash table.");
 771                goto out;
 772        }
 773
 774        table = hc->new_map;
 775
 776out:
 777        up_read(&_hash_lock);
 778
 779        return table;
 780}
 781
 782static struct dm_table *dm_get_live_or_inactive_table(struct mapped_device *md,
 783                                                      struct dm_ioctl *param,
 784                                                      int *srcu_idx)
 785{
 786        return (param->flags & DM_QUERY_INACTIVE_TABLE_FLAG) ?
 787                dm_get_inactive_table(md, srcu_idx) : dm_get_live_table(md, srcu_idx);
 788}
 789
 790/*
 791 * Fills in a dm_ioctl structure, ready for sending back to
 792 * userland.
 793 */
 794static void __dev_status(struct mapped_device *md, struct dm_ioctl *param)
 795{
 796        struct gendisk *disk = dm_disk(md);
 797        struct dm_table *table;
 798        int srcu_idx;
 799
 800        param->flags &= ~(DM_SUSPEND_FLAG | DM_READONLY_FLAG |
 801                          DM_ACTIVE_PRESENT_FLAG | DM_INTERNAL_SUSPEND_FLAG);
 802
 803        if (dm_suspended_md(md))
 804                param->flags |= DM_SUSPEND_FLAG;
 805
 806        if (dm_suspended_internally_md(md))
 807                param->flags |= DM_INTERNAL_SUSPEND_FLAG;
 808
 809        if (dm_test_deferred_remove_flag(md))
 810                param->flags |= DM_DEFERRED_REMOVE;
 811
 812        param->dev = huge_encode_dev(disk_devt(disk));
 813
 814        /*
 815         * Yes, this will be out of date by the time it gets back
 816         * to userland, but it is still very useful for
 817         * debugging.
 818         */
 819        param->open_count = dm_open_count(md);
 820
 821        param->event_nr = dm_get_event_nr(md);
 822        param->target_count = 0;
 823
 824        table = dm_get_live_table(md, &srcu_idx);
 825        if (table) {
 826                if (!(param->flags & DM_QUERY_INACTIVE_TABLE_FLAG)) {
 827                        if (get_disk_ro(disk))
 828                                param->flags |= DM_READONLY_FLAG;
 829                        param->target_count = dm_table_get_num_targets(table);
 830                }
 831
 832                param->flags |= DM_ACTIVE_PRESENT_FLAG;
 833        }
 834        dm_put_live_table(md, srcu_idx);
 835
 836        if (param->flags & DM_QUERY_INACTIVE_TABLE_FLAG) {
 837                int srcu_idx;
 838                table = dm_get_inactive_table(md, &srcu_idx);
 839                if (table) {
 840                        if (!(dm_table_get_mode(table) & FMODE_WRITE))
 841                                param->flags |= DM_READONLY_FLAG;
 842                        param->target_count = dm_table_get_num_targets(table);
 843                }
 844                dm_put_live_table(md, srcu_idx);
 845        }
 846}
 847
 848static int dev_create(struct file *filp, struct dm_ioctl *param, size_t param_size)
 849{
 850        int r, m = DM_ANY_MINOR;
 851        struct mapped_device *md;
 852
 853        r = check_name(param->name);
 854        if (r)
 855                return r;
 856
 857        if (param->flags & DM_PERSISTENT_DEV_FLAG)
 858                m = MINOR(huge_decode_dev(param->dev));
 859
 860        r = dm_create(m, &md);
 861        if (r)
 862                return r;
 863
 864        r = dm_hash_insert(param->name, *param->uuid ? param->uuid : NULL, md);
 865        if (r) {
 866                dm_put(md);
 867                dm_destroy(md);
 868                return r;
 869        }
 870
 871        param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
 872
 873        __dev_status(md, param);
 874
 875        dm_put(md);
 876
 877        return 0;
 878}
 879
 880/*
 881 * Always use UUID for lookups if it's present, otherwise use name or dev.
 882 */
 883static struct hash_cell *__find_device_hash_cell(struct dm_ioctl *param)
 884{
 885        struct hash_cell *hc = NULL;
 886
 887        if (*param->uuid) {
 888                if (*param->name || param->dev)
 889                        return NULL;
 890
 891                hc = __get_uuid_cell(param->uuid);
 892                if (!hc)
 893                        return NULL;
 894        } else if (*param->name) {
 895                if (param->dev)
 896                        return NULL;
 897
 898                hc = __get_name_cell(param->name);
 899                if (!hc)
 900                        return NULL;
 901        } else if (param->dev) {
 902                hc = __get_dev_cell(param->dev);
 903                if (!hc)
 904                        return NULL;
 905        } else
 906                return NULL;
 907
 908        /*
 909         * Sneakily write in both the name and the uuid
 910         * while we have the cell.
 911         */
 912        strlcpy(param->name, hc->name, sizeof(param->name));
 913        if (hc->uuid)
 914                strlcpy(param->uuid, hc->uuid, sizeof(param->uuid));
 915        else
 916                param->uuid[0] = '\0';
 917
 918        if (hc->new_map)
 919                param->flags |= DM_INACTIVE_PRESENT_FLAG;
 920        else
 921                param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
 922
 923        return hc;
 924}
 925
 926static struct mapped_device *find_device(struct dm_ioctl *param)
 927{
 928        struct hash_cell *hc;
 929        struct mapped_device *md = NULL;
 930
 931        down_read(&_hash_lock);
 932        hc = __find_device_hash_cell(param);
 933        if (hc)
 934                md = hc->md;
 935        up_read(&_hash_lock);
 936
 937        return md;
 938}
 939
 940static int dev_remove(struct file *filp, struct dm_ioctl *param, size_t param_size)
 941{
 942        struct hash_cell *hc;
 943        struct mapped_device *md;
 944        int r;
 945        struct dm_table *t;
 946
 947        down_write(&_hash_lock);
 948        hc = __find_device_hash_cell(param);
 949
 950        if (!hc) {
 951                DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
 952                up_write(&_hash_lock);
 953                return -ENXIO;
 954        }
 955
 956        md = hc->md;
 957
 958        /*
 959         * Ensure the device is not open and nothing further can open it.
 960         */
 961        r = dm_lock_for_deletion(md, !!(param->flags & DM_DEFERRED_REMOVE), false);
 962        if (r) {
 963                if (r == -EBUSY && param->flags & DM_DEFERRED_REMOVE) {
 964                        up_write(&_hash_lock);
 965                        dm_put(md);
 966                        return 0;
 967                }
 968                DMDEBUG_LIMIT("unable to remove open device %s", hc->name);
 969                up_write(&_hash_lock);
 970                dm_put(md);
 971                return r;
 972        }
 973
 974        t = __hash_remove(hc);
 975        up_write(&_hash_lock);
 976
 977        if (t) {
 978                dm_sync_table(md);
 979                dm_table_destroy(t);
 980        }
 981
 982        param->flags &= ~DM_DEFERRED_REMOVE;
 983
 984        if (!dm_kobject_uevent(md, KOBJ_REMOVE, param->event_nr))
 985                param->flags |= DM_UEVENT_GENERATED_FLAG;
 986
 987        dm_put(md);
 988        dm_destroy(md);
 989        return 0;
 990}
 991
 992/*
 993 * Check a string doesn't overrun the chunk of
 994 * memory we copied from userland.
 995 */
 996static int invalid_str(char *str, void *end)
 997{
 998        while ((void *) str < end)
 999                if (!*str++)
1000                        return 0;
1001
1002        return -EINVAL;
1003}
1004
1005static int dev_rename(struct file *filp, struct dm_ioctl *param, size_t param_size)
1006{
1007        int r;
1008        char *new_data = (char *) param + param->data_start;
1009        struct mapped_device *md;
1010        unsigned change_uuid = (param->flags & DM_UUID_FLAG) ? 1 : 0;
1011
1012        if (new_data < param->data ||
1013            invalid_str(new_data, (void *) param + param_size) || !*new_data ||
1014            strlen(new_data) > (change_uuid ? DM_UUID_LEN - 1 : DM_NAME_LEN - 1)) {
1015                DMWARN("Invalid new mapped device name or uuid string supplied.");
1016                return -EINVAL;
1017        }
1018
1019        if (!change_uuid) {
1020                r = check_name(new_data);
1021                if (r)
1022                        return r;
1023        }
1024
1025        md = dm_hash_rename(param, new_data);
1026        if (IS_ERR(md))
1027                return PTR_ERR(md);
1028
1029        __dev_status(md, param);
1030        dm_put(md);
1031
1032        return 0;
1033}
1034
1035static int dev_set_geometry(struct file *filp, struct dm_ioctl *param, size_t param_size)
1036{
1037        int r = -EINVAL, x;
1038        struct mapped_device *md;
1039        struct hd_geometry geometry;
1040        unsigned long indata[4];
1041        char *geostr = (char *) param + param->data_start;
1042        char dummy;
1043
1044        md = find_device(param);
1045        if (!md)
1046                return -ENXIO;
1047
1048        if (geostr < param->data ||
1049            invalid_str(geostr, (void *) param + param_size)) {
1050                DMWARN("Invalid geometry supplied.");
1051                goto out;
1052        }
1053
1054        x = sscanf(geostr, "%lu %lu %lu %lu%c", indata,
1055                   indata + 1, indata + 2, indata + 3, &dummy);
1056
1057        if (x != 4) {
1058                DMWARN("Unable to interpret geometry settings.");
1059                goto out;
1060        }
1061
1062        if (indata[0] > 65535 || indata[1] > 255 ||
1063            indata[2] > 255 || indata[3] > ULONG_MAX) {
1064                DMWARN("Geometry exceeds range limits.");
1065                goto out;
1066        }
1067
1068        geometry.cylinders = indata[0];
1069        geometry.heads = indata[1];
1070        geometry.sectors = indata[2];
1071        geometry.start = indata[3];
1072
1073        r = dm_set_geometry(md, &geometry);
1074
1075        param->data_size = 0;
1076
1077out:
1078        dm_put(md);
1079        return r;
1080}
1081
1082static int do_suspend(struct dm_ioctl *param)
1083{
1084        int r = 0;
1085        unsigned suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
1086        struct mapped_device *md;
1087
1088        md = find_device(param);
1089        if (!md)
1090                return -ENXIO;
1091
1092        if (param->flags & DM_SKIP_LOCKFS_FLAG)
1093                suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
1094        if (param->flags & DM_NOFLUSH_FLAG)
1095                suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
1096
1097        if (!dm_suspended_md(md)) {
1098                r = dm_suspend(md, suspend_flags);
1099                if (r)
1100                        goto out;
1101        }
1102
1103        __dev_status(md, param);
1104
1105out:
1106        dm_put(md);
1107
1108        return r;
1109}
1110
1111static int do_resume(struct dm_ioctl *param)
1112{
1113        int r = 0;
1114        unsigned suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
1115        struct hash_cell *hc;
1116        struct mapped_device *md;
1117        struct dm_table *new_map, *old_map = NULL;
1118
1119        down_write(&_hash_lock);
1120
1121        hc = __find_device_hash_cell(param);
1122        if (!hc) {
1123                DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
1124                up_write(&_hash_lock);
1125                return -ENXIO;
1126        }
1127
1128        md = hc->md;
1129
1130        new_map = hc->new_map;
1131        hc->new_map = NULL;
1132        param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1133
1134        up_write(&_hash_lock);
1135
1136        /* Do we need to load a new map ? */
1137        if (new_map) {
1138                /* Suspend if it isn't already suspended */
1139                if (param->flags & DM_SKIP_LOCKFS_FLAG)
1140                        suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
1141                if (param->flags & DM_NOFLUSH_FLAG)
1142                        suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
1143                if (!dm_suspended_md(md))
1144                        dm_suspend(md, suspend_flags);
1145
1146                old_map = dm_swap_table(md, new_map);
1147                if (IS_ERR(old_map)) {
1148                        dm_sync_table(md);
1149                        dm_table_destroy(new_map);
1150                        dm_put(md);
1151                        return PTR_ERR(old_map);
1152                }
1153
1154                if (dm_table_get_mode(new_map) & FMODE_WRITE)
1155                        set_disk_ro(dm_disk(md), 0);
1156                else
1157                        set_disk_ro(dm_disk(md), 1);
1158        }
1159
1160        if (dm_suspended_md(md)) {
1161                r = dm_resume(md);
1162                if (!r && !dm_kobject_uevent(md, KOBJ_CHANGE, param->event_nr))
1163                        param->flags |= DM_UEVENT_GENERATED_FLAG;
1164        }
1165
1166        /*
1167         * Since dm_swap_table synchronizes RCU, nobody should be in
1168         * read-side critical section already.
1169         */
1170        if (old_map)
1171                dm_table_destroy(old_map);
1172
1173        if (!r)
1174                __dev_status(md, param);
1175
1176        dm_put(md);
1177        return r;
1178}
1179
1180/*
1181 * Set or unset the suspension state of a device.
1182 * If the device already is in the requested state we just return its status.
1183 */
1184static int dev_suspend(struct file *filp, struct dm_ioctl *param, size_t param_size)
1185{
1186        if (param->flags & DM_SUSPEND_FLAG)
1187                return do_suspend(param);
1188
1189        return do_resume(param);
1190}
1191
1192/*
1193 * Copies device info back to user space, used by
1194 * the create and info ioctls.
1195 */
1196static int dev_status(struct file *filp, struct dm_ioctl *param, size_t param_size)
1197{
1198        struct mapped_device *md;
1199
1200        md = find_device(param);
1201        if (!md)
1202                return -ENXIO;
1203
1204        __dev_status(md, param);
1205        dm_put(md);
1206
1207        return 0;
1208}
1209
1210/*
1211 * Build up the status struct for each target
1212 */
1213static void retrieve_status(struct dm_table *table,
1214                            struct dm_ioctl *param, size_t param_size)
1215{
1216        unsigned int i, num_targets;
1217        struct dm_target_spec *spec;
1218        char *outbuf, *outptr;
1219        status_type_t type;
1220        size_t remaining, len, used = 0;
1221        unsigned status_flags = 0;
1222
1223        outptr = outbuf = get_result_buffer(param, param_size, &len);
1224
1225        if (param->flags & DM_STATUS_TABLE_FLAG)
1226                type = STATUSTYPE_TABLE;
1227        else
1228                type = STATUSTYPE_INFO;
1229
1230        /* Get all the target info */
1231        num_targets = dm_table_get_num_targets(table);
1232        for (i = 0; i < num_targets; i++) {
1233                struct dm_target *ti = dm_table_get_target(table, i);
1234                size_t l;
1235
1236                remaining = len - (outptr - outbuf);
1237                if (remaining <= sizeof(struct dm_target_spec)) {
1238                        param->flags |= DM_BUFFER_FULL_FLAG;
1239                        break;
1240                }
1241
1242                spec = (struct dm_target_spec *) outptr;
1243
1244                spec->status = 0;
1245                spec->sector_start = ti->begin;
1246                spec->length = ti->len;
1247                strncpy(spec->target_type, ti->type->name,
1248                        sizeof(spec->target_type) - 1);
1249
1250                outptr += sizeof(struct dm_target_spec);
1251                remaining = len - (outptr - outbuf);
1252                if (remaining <= 0) {
1253                        param->flags |= DM_BUFFER_FULL_FLAG;
1254                        break;
1255                }
1256
1257                /* Get the status/table string from the target driver */
1258                if (ti->type->status) {
1259                        if (param->flags & DM_NOFLUSH_FLAG)
1260                                status_flags |= DM_STATUS_NOFLUSH_FLAG;
1261                        ti->type->status(ti, type, status_flags, outptr, remaining);
1262                } else
1263                        outptr[0] = '\0';
1264
1265                l = strlen(outptr) + 1;
1266                if (l == remaining) {
1267                        param->flags |= DM_BUFFER_FULL_FLAG;
1268                        break;
1269                }
1270
1271                outptr += l;
1272                used = param->data_start + (outptr - outbuf);
1273
1274                outptr = align_ptr(outptr);
1275                spec->next = outptr - outbuf;
1276        }
1277
1278        if (used)
1279                param->data_size = used;
1280
1281        param->target_count = num_targets;
1282}
1283
1284/*
1285 * Wait for a device to report an event
1286 */
1287static int dev_wait(struct file *filp, struct dm_ioctl *param, size_t param_size)
1288{
1289        int r = 0;
1290        struct mapped_device *md;
1291        struct dm_table *table;
1292        int srcu_idx;
1293
1294        md = find_device(param);
1295        if (!md)
1296                return -ENXIO;
1297
1298        /*
1299         * Wait for a notification event
1300         */
1301        if (dm_wait_event(md, param->event_nr)) {
1302                r = -ERESTARTSYS;
1303                goto out;
1304        }
1305
1306        /*
1307         * The userland program is going to want to know what
1308         * changed to trigger the event, so we may as well tell
1309         * him and save an ioctl.
1310         */
1311        __dev_status(md, param);
1312
1313        table = dm_get_live_or_inactive_table(md, param, &srcu_idx);
1314        if (table)
1315                retrieve_status(table, param, param_size);
1316        dm_put_live_table(md, srcu_idx);
1317
1318out:
1319        dm_put(md);
1320
1321        return r;
1322}
1323
1324/*
1325 * Remember the global event number and make it possible to poll
1326 * for further events.
1327 */
1328static int dev_arm_poll(struct file *filp, struct dm_ioctl *param, size_t param_size)
1329{
1330        struct dm_file *priv = filp->private_data;
1331
1332        priv->global_event_nr = atomic_read(&dm_global_event_nr);
1333
1334        return 0;
1335}
1336
1337static inline fmode_t get_mode(struct dm_ioctl *param)
1338{
1339        fmode_t mode = FMODE_READ | FMODE_WRITE;
1340
1341        if (param->flags & DM_READONLY_FLAG)
1342                mode = FMODE_READ;
1343
1344        return mode;
1345}
1346
1347static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
1348                       struct dm_target_spec **spec, char **target_params)
1349{
1350        *spec = (struct dm_target_spec *) ((unsigned char *) last + next);
1351        *target_params = (char *) (*spec + 1);
1352
1353        if (*spec < (last + 1))
1354                return -EINVAL;
1355
1356        return invalid_str(*target_params, end);
1357}
1358
1359static int populate_table(struct dm_table *table,
1360                          struct dm_ioctl *param, size_t param_size)
1361{
1362        int r;
1363        unsigned int i = 0;
1364        struct dm_target_spec *spec = (struct dm_target_spec *) param;
1365        uint32_t next = param->data_start;
1366        void *end = (void *) param + param_size;
1367        char *target_params;
1368
1369        if (!param->target_count) {
1370                DMWARN("populate_table: no targets specified");
1371                return -EINVAL;
1372        }
1373
1374        for (i = 0; i < param->target_count; i++) {
1375
1376                r = next_target(spec, next, end, &spec, &target_params);
1377                if (r) {
1378                        DMWARN("unable to find target");
1379                        return r;
1380                }
1381
1382                r = dm_table_add_target(table, spec->target_type,
1383                                        (sector_t) spec->sector_start,
1384                                        (sector_t) spec->length,
1385                                        target_params);
1386                if (r) {
1387                        DMWARN("error adding target to table");
1388                        return r;
1389                }
1390
1391                next = spec->next;
1392        }
1393
1394        return dm_table_complete(table);
1395}
1396
1397static bool is_valid_type(enum dm_queue_mode cur, enum dm_queue_mode new)
1398{
1399        if (cur == new ||
1400            (cur == DM_TYPE_BIO_BASED && new == DM_TYPE_DAX_BIO_BASED))
1401                return true;
1402
1403        return false;
1404}
1405
1406static int table_load(struct file *filp, struct dm_ioctl *param, size_t param_size)
1407{
1408        int r;
1409        struct hash_cell *hc;
1410        struct dm_table *t, *old_map = NULL;
1411        struct mapped_device *md;
1412        struct target_type *immutable_target_type;
1413
1414        md = find_device(param);
1415        if (!md)
1416                return -ENXIO;
1417
1418        r = dm_table_create(&t, get_mode(param), param->target_count, md);
1419        if (r)
1420                goto err;
1421
1422        /* Protect md->type and md->queue against concurrent table loads. */
1423        dm_lock_md_type(md);
1424        r = populate_table(t, param, param_size);
1425        if (r)
1426                goto err_unlock_md_type;
1427
1428        immutable_target_type = dm_get_immutable_target_type(md);
1429        if (immutable_target_type &&
1430            (immutable_target_type != dm_table_get_immutable_target_type(t)) &&
1431            !dm_table_get_wildcard_target(t)) {
1432                DMWARN("can't replace immutable target type %s",
1433                       immutable_target_type->name);
1434                r = -EINVAL;
1435                goto err_unlock_md_type;
1436        }
1437
1438        if (dm_get_md_type(md) == DM_TYPE_NONE) {
1439                /* Initial table load: acquire type of table. */
1440                dm_set_md_type(md, dm_table_get_type(t));
1441
1442                /* setup md->queue to reflect md's type (may block) */
1443                r = dm_setup_md_queue(md, t);
1444                if (r) {
1445                        DMWARN("unable to set up device queue for new table.");
1446                        goto err_unlock_md_type;
1447                }
1448        } else if (!is_valid_type(dm_get_md_type(md), dm_table_get_type(t))) {
1449                DMWARN("can't change device type (old=%u vs new=%u) after initial table load.",
1450                       dm_get_md_type(md), dm_table_get_type(t));
1451                r = -EINVAL;
1452                goto err_unlock_md_type;
1453        }
1454
1455        dm_unlock_md_type(md);
1456
1457        /* stage inactive table */
1458        down_write(&_hash_lock);
1459        hc = dm_get_mdptr(md);
1460        if (!hc || hc->md != md) {
1461                DMWARN("device has been removed from the dev hash table.");
1462                up_write(&_hash_lock);
1463                r = -ENXIO;
1464                goto err_destroy_table;
1465        }
1466
1467        if (hc->new_map)
1468                old_map = hc->new_map;
1469        hc->new_map = t;
1470        up_write(&_hash_lock);
1471
1472        param->flags |= DM_INACTIVE_PRESENT_FLAG;
1473        __dev_status(md, param);
1474
1475        if (old_map) {
1476                dm_sync_table(md);
1477                dm_table_destroy(old_map);
1478        }
1479
1480        dm_put(md);
1481
1482        return 0;
1483
1484err_unlock_md_type:
1485        dm_unlock_md_type(md);
1486err_destroy_table:
1487        dm_table_destroy(t);
1488err:
1489        dm_put(md);
1490
1491        return r;
1492}
1493
1494static int table_clear(struct file *filp, struct dm_ioctl *param, size_t param_size)
1495{
1496        struct hash_cell *hc;
1497        struct mapped_device *md;
1498        struct dm_table *old_map = NULL;
1499
1500        down_write(&_hash_lock);
1501
1502        hc = __find_device_hash_cell(param);
1503        if (!hc) {
1504                DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
1505                up_write(&_hash_lock);
1506                return -ENXIO;
1507        }
1508
1509        if (hc->new_map) {
1510                old_map = hc->new_map;
1511                hc->new_map = NULL;
1512        }
1513
1514        param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1515
1516        __dev_status(hc->md, param);
1517        md = hc->md;
1518        up_write(&_hash_lock);
1519        if (old_map) {
1520                dm_sync_table(md);
1521                dm_table_destroy(old_map);
1522        }
1523        dm_put(md);
1524
1525        return 0;
1526}
1527
1528/*
1529 * Retrieves a list of devices used by a particular dm device.
1530 */
1531static void retrieve_deps(struct dm_table *table,
1532                          struct dm_ioctl *param, size_t param_size)
1533{
1534        unsigned int count = 0;
1535        struct list_head *tmp;
1536        size_t len, needed;
1537        struct dm_dev_internal *dd;
1538        struct dm_target_deps *deps;
1539
1540        deps = get_result_buffer(param, param_size, &len);
1541
1542        /*
1543         * Count the devices.
1544         */
1545        list_for_each (tmp, dm_table_get_devices(table))
1546                count++;
1547
1548        /*
1549         * Check we have enough space.
1550         */
1551        needed = struct_size(deps, dev, count);
1552        if (len < needed) {
1553                param->flags |= DM_BUFFER_FULL_FLAG;
1554                return;
1555        }
1556
1557        /*
1558         * Fill in the devices.
1559         */
1560        deps->count = count;
1561        count = 0;
1562        list_for_each_entry (dd, dm_table_get_devices(table), list)
1563                deps->dev[count++] = huge_encode_dev(dd->dm_dev->bdev->bd_dev);
1564
1565        param->data_size = param->data_start + needed;
1566}
1567
1568static int table_deps(struct file *filp, struct dm_ioctl *param, size_t param_size)
1569{
1570        struct mapped_device *md;
1571        struct dm_table *table;
1572        int srcu_idx;
1573
1574        md = find_device(param);
1575        if (!md)
1576                return -ENXIO;
1577
1578        __dev_status(md, param);
1579
1580        table = dm_get_live_or_inactive_table(md, param, &srcu_idx);
1581        if (table)
1582                retrieve_deps(table, param, param_size);
1583        dm_put_live_table(md, srcu_idx);
1584
1585        dm_put(md);
1586
1587        return 0;
1588}
1589
1590/*
1591 * Return the status of a device as a text string for each
1592 * target.
1593 */
1594static int table_status(struct file *filp, struct dm_ioctl *param, size_t param_size)
1595{
1596        struct mapped_device *md;
1597        struct dm_table *table;
1598        int srcu_idx;
1599
1600        md = find_device(param);
1601        if (!md)
1602                return -ENXIO;
1603
1604        __dev_status(md, param);
1605
1606        table = dm_get_live_or_inactive_table(md, param, &srcu_idx);
1607        if (table)
1608                retrieve_status(table, param, param_size);
1609        dm_put_live_table(md, srcu_idx);
1610
1611        dm_put(md);
1612
1613        return 0;
1614}
1615
1616/*
1617 * Process device-mapper dependent messages.  Messages prefixed with '@'
1618 * are processed by the DM core.  All others are delivered to the target.
1619 * Returns a number <= 1 if message was processed by device mapper.
1620 * Returns 2 if message should be delivered to the target.
1621 */
1622static int message_for_md(struct mapped_device *md, unsigned argc, char **argv,
1623                          char *result, unsigned maxlen)
1624{
1625        int r;
1626
1627        if (**argv != '@')
1628                return 2; /* no '@' prefix, deliver to target */
1629
1630        if (!strcasecmp(argv[0], "@cancel_deferred_remove")) {
1631                if (argc != 1) {
1632                        DMERR("Invalid arguments for @cancel_deferred_remove");
1633                        return -EINVAL;
1634                }
1635                return dm_cancel_deferred_remove(md);
1636        }
1637
1638        r = dm_stats_message(md, argc, argv, result, maxlen);
1639        if (r < 2)
1640                return r;
1641
1642        DMERR("Unsupported message sent to DM core: %s", argv[0]);
1643        return -EINVAL;
1644}
1645
1646/*
1647 * Pass a message to the target that's at the supplied device offset.
1648 */
1649static int target_message(struct file *filp, struct dm_ioctl *param, size_t param_size)
1650{
1651        int r, argc;
1652        char **argv;
1653        struct mapped_device *md;
1654        struct dm_table *table;
1655        struct dm_target *ti;
1656        struct dm_target_msg *tmsg = (void *) param + param->data_start;
1657        size_t maxlen;
1658        char *result = get_result_buffer(param, param_size, &maxlen);
1659        int srcu_idx;
1660
1661        md = find_device(param);
1662        if (!md)
1663                return -ENXIO;
1664
1665        if (tmsg < (struct dm_target_msg *) param->data ||
1666            invalid_str(tmsg->message, (void *) param + param_size)) {
1667                DMWARN("Invalid target message parameters.");
1668                r = -EINVAL;
1669                goto out;
1670        }
1671
1672        r = dm_split_args(&argc, &argv, tmsg->message);
1673        if (r) {
1674                DMWARN("Failed to split target message parameters");
1675                goto out;
1676        }
1677
1678        if (!argc) {
1679                DMWARN("Empty message received.");
1680                r = -EINVAL;
1681                goto out_argv;
1682        }
1683
1684        r = message_for_md(md, argc, argv, result, maxlen);
1685        if (r <= 1)
1686                goto out_argv;
1687
1688        table = dm_get_live_table(md, &srcu_idx);
1689        if (!table)
1690                goto out_table;
1691
1692        if (dm_deleting_md(md)) {
1693                r = -ENXIO;
1694                goto out_table;
1695        }
1696
1697        ti = dm_table_find_target(table, tmsg->sector);
1698        if (!ti) {
1699                DMWARN("Target message sector outside device.");
1700                r = -EINVAL;
1701        } else if (ti->type->message)
1702                r = ti->type->message(ti, argc, argv, result, maxlen);
1703        else {
1704                DMWARN("Target type does not support messages");
1705                r = -EINVAL;
1706        }
1707
1708 out_table:
1709        dm_put_live_table(md, srcu_idx);
1710 out_argv:
1711        kfree(argv);
1712 out:
1713        if (r >= 0)
1714                __dev_status(md, param);
1715
1716        if (r == 1) {
1717                param->flags |= DM_DATA_OUT_FLAG;
1718                if (dm_message_test_buffer_overflow(result, maxlen))
1719                        param->flags |= DM_BUFFER_FULL_FLAG;
1720                else
1721                        param->data_size = param->data_start + strlen(result) + 1;
1722                r = 0;
1723        }
1724
1725        dm_put(md);
1726        return r;
1727}
1728
1729/*
1730 * The ioctl parameter block consists of two parts, a dm_ioctl struct
1731 * followed by a data buffer.  This flag is set if the second part,
1732 * which has a variable size, is not used by the function processing
1733 * the ioctl.
1734 */
1735#define IOCTL_FLAGS_NO_PARAMS           1
1736#define IOCTL_FLAGS_ISSUE_GLOBAL_EVENT  2
1737
1738/*-----------------------------------------------------------------
1739 * Implementation of open/close/ioctl on the special char
1740 * device.
1741 *---------------------------------------------------------------*/
1742static ioctl_fn lookup_ioctl(unsigned int cmd, int *ioctl_flags)
1743{
1744        static const struct {
1745                int cmd;
1746                int flags;
1747                ioctl_fn fn;
1748        } _ioctls[] = {
1749                {DM_VERSION_CMD, 0, NULL}, /* version is dealt with elsewhere */
1750                {DM_REMOVE_ALL_CMD, IOCTL_FLAGS_NO_PARAMS | IOCTL_FLAGS_ISSUE_GLOBAL_EVENT, remove_all},
1751                {DM_LIST_DEVICES_CMD, 0, list_devices},
1752
1753                {DM_DEV_CREATE_CMD, IOCTL_FLAGS_NO_PARAMS | IOCTL_FLAGS_ISSUE_GLOBAL_EVENT, dev_create},
1754                {DM_DEV_REMOVE_CMD, IOCTL_FLAGS_NO_PARAMS | IOCTL_FLAGS_ISSUE_GLOBAL_EVENT, dev_remove},
1755                {DM_DEV_RENAME_CMD, IOCTL_FLAGS_ISSUE_GLOBAL_EVENT, dev_rename},
1756                {DM_DEV_SUSPEND_CMD, IOCTL_FLAGS_NO_PARAMS, dev_suspend},
1757                {DM_DEV_STATUS_CMD, IOCTL_FLAGS_NO_PARAMS, dev_status},
1758                {DM_DEV_WAIT_CMD, 0, dev_wait},
1759
1760                {DM_TABLE_LOAD_CMD, 0, table_load},
1761                {DM_TABLE_CLEAR_CMD, IOCTL_FLAGS_NO_PARAMS, table_clear},
1762                {DM_TABLE_DEPS_CMD, 0, table_deps},
1763                {DM_TABLE_STATUS_CMD, 0, table_status},
1764
1765                {DM_LIST_VERSIONS_CMD, 0, list_versions},
1766
1767                {DM_TARGET_MSG_CMD, 0, target_message},
1768                {DM_DEV_SET_GEOMETRY_CMD, 0, dev_set_geometry},
1769                {DM_DEV_ARM_POLL, IOCTL_FLAGS_NO_PARAMS, dev_arm_poll},
1770                {DM_GET_TARGET_VERSION, 0, get_target_version},
1771        };
1772
1773        if (unlikely(cmd >= ARRAY_SIZE(_ioctls)))
1774                return NULL;
1775
1776        *ioctl_flags = _ioctls[cmd].flags;
1777        return _ioctls[cmd].fn;
1778}
1779
1780/*
1781 * As well as checking the version compatibility this always
1782 * copies the kernel interface version out.
1783 */
1784static int check_version(unsigned int cmd, struct dm_ioctl __user *user)
1785{
1786        uint32_t version[3];
1787        int r = 0;
1788
1789        if (copy_from_user(version, user->version, sizeof(version)))
1790                return -EFAULT;
1791
1792        if ((DM_VERSION_MAJOR != version[0]) ||
1793            (DM_VERSION_MINOR < version[1])) {
1794                DMWARN("ioctl interface mismatch: "
1795                       "kernel(%u.%u.%u), user(%u.%u.%u), cmd(%d)",
1796                       DM_VERSION_MAJOR, DM_VERSION_MINOR,
1797                       DM_VERSION_PATCHLEVEL,
1798                       version[0], version[1], version[2], cmd);
1799                r = -EINVAL;
1800        }
1801
1802        /*
1803         * Fill in the kernel version.
1804         */
1805        version[0] = DM_VERSION_MAJOR;
1806        version[1] = DM_VERSION_MINOR;
1807        version[2] = DM_VERSION_PATCHLEVEL;
1808        if (copy_to_user(user->version, version, sizeof(version)))
1809                return -EFAULT;
1810
1811        return r;
1812}
1813
1814#define DM_PARAMS_MALLOC        0x0001  /* Params allocated with kvmalloc() */
1815#define DM_WIPE_BUFFER          0x0010  /* Wipe input buffer before returning from ioctl */
1816
1817static void free_params(struct dm_ioctl *param, size_t param_size, int param_flags)
1818{
1819        if (param_flags & DM_WIPE_BUFFER)
1820                memset(param, 0, param_size);
1821
1822        if (param_flags & DM_PARAMS_MALLOC)
1823                kvfree(param);
1824}
1825
1826static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl *param_kernel,
1827                       int ioctl_flags, struct dm_ioctl **param, int *param_flags)
1828{
1829        struct dm_ioctl *dmi;
1830        int secure_data;
1831        const size_t minimum_data_size = offsetof(struct dm_ioctl, data);
1832        unsigned noio_flag;
1833
1834        if (copy_from_user(param_kernel, user, minimum_data_size))
1835                return -EFAULT;
1836
1837        if (param_kernel->data_size < minimum_data_size)
1838                return -EINVAL;
1839
1840        secure_data = param_kernel->flags & DM_SECURE_DATA_FLAG;
1841
1842        *param_flags = secure_data ? DM_WIPE_BUFFER : 0;
1843
1844        if (ioctl_flags & IOCTL_FLAGS_NO_PARAMS) {
1845                dmi = param_kernel;
1846                dmi->data_size = minimum_data_size;
1847                goto data_copied;
1848        }
1849
1850        /*
1851         * Use __GFP_HIGH to avoid low memory issues when a device is
1852         * suspended and the ioctl is needed to resume it.
1853         * Use kmalloc() rather than vmalloc() when we can.
1854         */
1855        dmi = NULL;
1856        noio_flag = memalloc_noio_save();
1857        dmi = kvmalloc(param_kernel->data_size, GFP_KERNEL | __GFP_HIGH);
1858        memalloc_noio_restore(noio_flag);
1859
1860        if (!dmi) {
1861                if (secure_data && clear_user(user, param_kernel->data_size))
1862                        return -EFAULT;
1863                return -ENOMEM;
1864        }
1865
1866        *param_flags |= DM_PARAMS_MALLOC;
1867
1868        /* Copy from param_kernel (which was already copied from user) */
1869        memcpy(dmi, param_kernel, minimum_data_size);
1870
1871        if (copy_from_user(&dmi->data, (char __user *)user + minimum_data_size,
1872                           param_kernel->data_size - minimum_data_size))
1873                goto bad;
1874data_copied:
1875        /* Wipe the user buffer so we do not return it to userspace */
1876        if (secure_data && clear_user(user, param_kernel->data_size))
1877                goto bad;
1878
1879        *param = dmi;
1880        return 0;
1881
1882bad:
1883        free_params(dmi, param_kernel->data_size, *param_flags);
1884
1885        return -EFAULT;
1886}
1887
1888static int validate_params(uint cmd, struct dm_ioctl *param)
1889{
1890        /* Always clear this flag */
1891        param->flags &= ~DM_BUFFER_FULL_FLAG;
1892        param->flags &= ~DM_UEVENT_GENERATED_FLAG;
1893        param->flags &= ~DM_SECURE_DATA_FLAG;
1894        param->flags &= ~DM_DATA_OUT_FLAG;
1895
1896        /* Ignores parameters */
1897        if (cmd == DM_REMOVE_ALL_CMD ||
1898            cmd == DM_LIST_DEVICES_CMD ||
1899            cmd == DM_LIST_VERSIONS_CMD)
1900                return 0;
1901
1902        if (cmd == DM_DEV_CREATE_CMD) {
1903                if (!*param->name) {
1904                        DMWARN("name not supplied when creating device");
1905                        return -EINVAL;
1906                }
1907        } else if (*param->uuid && *param->name) {
1908                DMWARN("only supply one of name or uuid, cmd(%u)", cmd);
1909                return -EINVAL;
1910        }
1911
1912        /* Ensure strings are terminated */
1913        param->name[DM_NAME_LEN - 1] = '\0';
1914        param->uuid[DM_UUID_LEN - 1] = '\0';
1915
1916        return 0;
1917}
1918
1919static int ctl_ioctl(struct file *file, uint command, struct dm_ioctl __user *user)
1920{
1921        int r = 0;
1922        int ioctl_flags;
1923        int param_flags;
1924        unsigned int cmd;
1925        struct dm_ioctl *param;
1926        ioctl_fn fn = NULL;
1927        size_t input_param_size;
1928        struct dm_ioctl param_kernel;
1929
1930        /* only root can play with this */
1931        if (!capable(CAP_SYS_ADMIN))
1932                return -EACCES;
1933
1934        if (_IOC_TYPE(command) != DM_IOCTL)
1935                return -ENOTTY;
1936
1937        cmd = _IOC_NR(command);
1938
1939        /*
1940         * Check the interface version passed in.  This also
1941         * writes out the kernel's interface version.
1942         */
1943        r = check_version(cmd, user);
1944        if (r)
1945                return r;
1946
1947        /*
1948         * Nothing more to do for the version command.
1949         */
1950        if (cmd == DM_VERSION_CMD)
1951                return 0;
1952
1953        fn = lookup_ioctl(cmd, &ioctl_flags);
1954        if (!fn) {
1955                DMWARN("dm_ctl_ioctl: unknown command 0x%x", command);
1956                return -ENOTTY;
1957        }
1958
1959        /*
1960         * Copy the parameters into kernel space.
1961         */
1962        r = copy_params(user, &param_kernel, ioctl_flags, &param, &param_flags);
1963
1964        if (r)
1965                return r;
1966
1967        input_param_size = param->data_size;
1968        r = validate_params(cmd, param);
1969        if (r)
1970                goto out;
1971
1972        param->data_size = offsetof(struct dm_ioctl, data);
1973        r = fn(file, param, input_param_size);
1974
1975        if (unlikely(param->flags & DM_BUFFER_FULL_FLAG) &&
1976            unlikely(ioctl_flags & IOCTL_FLAGS_NO_PARAMS))
1977                DMERR("ioctl %d tried to output some data but has IOCTL_FLAGS_NO_PARAMS set", cmd);
1978
1979        if (!r && ioctl_flags & IOCTL_FLAGS_ISSUE_GLOBAL_EVENT)
1980                dm_issue_global_event();
1981
1982        /*
1983         * Copy the results back to userland.
1984         */
1985        if (!r && copy_to_user(user, param, param->data_size))
1986                r = -EFAULT;
1987
1988out:
1989        free_params(param, input_param_size, param_flags);
1990        return r;
1991}
1992
1993static long dm_ctl_ioctl(struct file *file, uint command, ulong u)
1994{
1995        return (long)ctl_ioctl(file, command, (struct dm_ioctl __user *)u);
1996}
1997
1998#ifdef CONFIG_COMPAT
1999static long dm_compat_ctl_ioctl(struct file *file, uint command, ulong u)
2000{
2001        return (long)dm_ctl_ioctl(file, command, (ulong) compat_ptr(u));
2002}
2003#else
2004#define dm_compat_ctl_ioctl NULL
2005#endif
2006
2007static int dm_open(struct inode *inode, struct file *filp)
2008{
2009        int r;
2010        struct dm_file *priv;
2011
2012        r = nonseekable_open(inode, filp);
2013        if (unlikely(r))
2014                return r;
2015
2016        priv = filp->private_data = kmalloc(sizeof(struct dm_file), GFP_KERNEL);
2017        if (!priv)
2018                return -ENOMEM;
2019
2020        priv->global_event_nr = atomic_read(&dm_global_event_nr);
2021
2022        return 0;
2023}
2024
2025static int dm_release(struct inode *inode, struct file *filp)
2026{
2027        kfree(filp->private_data);
2028        return 0;
2029}
2030
2031static __poll_t dm_poll(struct file *filp, poll_table *wait)
2032{
2033        struct dm_file *priv = filp->private_data;
2034        __poll_t mask = 0;
2035
2036        poll_wait(filp, &dm_global_eventq, wait);
2037
2038        if ((int)(atomic_read(&dm_global_event_nr) - priv->global_event_nr) > 0)
2039                mask |= EPOLLIN;
2040
2041        return mask;
2042}
2043
2044static const struct file_operations _ctl_fops = {
2045        .open    = dm_open,
2046        .release = dm_release,
2047        .poll    = dm_poll,
2048        .unlocked_ioctl  = dm_ctl_ioctl,
2049        .compat_ioctl = dm_compat_ctl_ioctl,
2050        .owner   = THIS_MODULE,
2051        .llseek  = noop_llseek,
2052};
2053
2054static struct miscdevice _dm_misc = {
2055        .minor          = MAPPER_CTRL_MINOR,
2056        .name           = DM_NAME,
2057        .nodename       = DM_DIR "/" DM_CONTROL_NODE,
2058        .fops           = &_ctl_fops
2059};
2060
2061MODULE_ALIAS_MISCDEV(MAPPER_CTRL_MINOR);
2062MODULE_ALIAS("devname:" DM_DIR "/" DM_CONTROL_NODE);
2063
2064/*
2065 * Create misc character device and link to DM_DIR/control.
2066 */
2067int __init dm_interface_init(void)
2068{
2069        int r;
2070
2071        r = misc_register(&_dm_misc);
2072        if (r) {
2073                DMERR("misc_register failed for control device");
2074                return r;
2075        }
2076
2077        DMINFO("%d.%d.%d%s initialised: %s", DM_VERSION_MAJOR,
2078               DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL, DM_VERSION_EXTRA,
2079               DM_DRIVER_EMAIL);
2080        return 0;
2081}
2082
2083void dm_interface_exit(void)
2084{
2085        misc_deregister(&_dm_misc);
2086        dm_hash_exit();
2087}
2088
2089/**
2090 * dm_copy_name_and_uuid - Copy mapped device name & uuid into supplied buffers
2091 * @md: Pointer to mapped_device
2092 * @name: Buffer (size DM_NAME_LEN) for name
2093 * @uuid: Buffer (size DM_UUID_LEN) for uuid or empty string if uuid not defined
2094 */
2095int dm_copy_name_and_uuid(struct mapped_device *md, char *name, char *uuid)
2096{
2097        int r = 0;
2098        struct hash_cell *hc;
2099
2100        if (!md)
2101                return -ENXIO;
2102
2103        mutex_lock(&dm_hash_cells_mutex);
2104        hc = dm_get_mdptr(md);
2105        if (!hc || hc->md != md) {
2106                r = -ENXIO;
2107                goto out;
2108        }
2109
2110        if (name)
2111                strcpy(name, hc->name);
2112        if (uuid)
2113                strcpy(uuid, hc->uuid ? : "");
2114
2115out:
2116        mutex_unlock(&dm_hash_cells_mutex);
2117
2118        return r;
2119}
2120EXPORT_SYMBOL_GPL(dm_copy_name_and_uuid);
2121
2122/**
2123 * dm_early_create - create a mapped device in early boot.
2124 *
2125 * @dmi: Contains main information of the device mapping to be created.
2126 * @spec_array: array of pointers to struct dm_target_spec. Describes the
2127 * mapping table of the device.
2128 * @target_params_array: array of strings with the parameters to a specific
2129 * target.
2130 *
2131 * Instead of having the struct dm_target_spec and the parameters for every
2132 * target embedded at the end of struct dm_ioctl (as performed in a normal
2133 * ioctl), pass them as arguments, so the caller doesn't need to serialize them.
2134 * The size of the spec_array and target_params_array is given by
2135 * @dmi->target_count.
2136 * This function is supposed to be called in early boot, so locking mechanisms
2137 * to protect against concurrent loads are not required.
2138 */
2139int __init dm_early_create(struct dm_ioctl *dmi,
2140                           struct dm_target_spec **spec_array,
2141                           char **target_params_array)
2142{
2143        int r, m = DM_ANY_MINOR;
2144        struct dm_table *t, *old_map;
2145        struct mapped_device *md;
2146        unsigned int i;
2147
2148        if (!dmi->target_count)
2149                return -EINVAL;
2150
2151        r = check_name(dmi->name);
2152        if (r)
2153                return r;
2154
2155        if (dmi->flags & DM_PERSISTENT_DEV_FLAG)
2156                m = MINOR(huge_decode_dev(dmi->dev));
2157
2158        /* alloc dm device */
2159        r = dm_create(m, &md);
2160        if (r)
2161                return r;
2162
2163        /* hash insert */
2164        r = dm_hash_insert(dmi->name, *dmi->uuid ? dmi->uuid : NULL, md);
2165        if (r)
2166                goto err_destroy_dm;
2167
2168        /* alloc table */
2169        r = dm_table_create(&t, get_mode(dmi), dmi->target_count, md);
2170        if (r)
2171                goto err_hash_remove;
2172
2173        /* add targets */
2174        for (i = 0; i < dmi->target_count; i++) {
2175                r = dm_table_add_target(t, spec_array[i]->target_type,
2176                                        (sector_t) spec_array[i]->sector_start,
2177                                        (sector_t) spec_array[i]->length,
2178                                        target_params_array[i]);
2179                if (r) {
2180                        DMWARN("error adding target to table");
2181                        goto err_destroy_table;
2182                }
2183        }
2184
2185        /* finish table */
2186        r = dm_table_complete(t);
2187        if (r)
2188                goto err_destroy_table;
2189
2190        md->type = dm_table_get_type(t);
2191        /* setup md->queue to reflect md's type (may block) */
2192        r = dm_setup_md_queue(md, t);
2193        if (r) {
2194                DMWARN("unable to set up device queue for new table.");
2195                goto err_destroy_table;
2196        }
2197
2198        /* Set new map */
2199        dm_suspend(md, 0);
2200        old_map = dm_swap_table(md, t);
2201        if (IS_ERR(old_map)) {
2202                r = PTR_ERR(old_map);
2203                goto err_destroy_table;
2204        }
2205        set_disk_ro(dm_disk(md), !!(dmi->flags & DM_READONLY_FLAG));
2206
2207        /* resume device */
2208        r = dm_resume(md);
2209        if (r)
2210                goto err_destroy_table;
2211
2212        DMINFO("%s (%s) is ready", md->disk->disk_name, dmi->name);
2213        dm_put(md);
2214        return 0;
2215
2216err_destroy_table:
2217        dm_table_destroy(t);
2218err_hash_remove:
2219        (void) __hash_remove(__get_name_cell(dmi->name));
2220        /* release reference from __get_name_cell */
2221        dm_put(md);
2222err_destroy_dm:
2223        dm_put(md);
2224        dm_destroy(md);
2225        return r;
2226}
2227