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/dm-ioctl.h>
  18#include <linux/hdreg.h>
  19#include <linux/compat.h>
  20
  21#include <linux/uaccess.h>
  22
  23#define DM_MSG_PREFIX "ioctl"
  24#define DM_DRIVER_EMAIL "dm-devel@redhat.com"
  25
  26struct dm_file {
  27	/*
  28	 * poll will wait until the global event number is greater than
  29	 * this value.
  30	 */
  31	volatile unsigned global_event_nr;
  32};
  33
  34/*-----------------------------------------------------------------
  35 * The ioctl interface needs to be able to look up devices by
  36 * name or uuid.
  37 *---------------------------------------------------------------*/
  38struct hash_cell {
  39	struct list_head name_list;
  40	struct list_head uuid_list;
  41
  42	char *name;
  43	char *uuid;
  44	struct mapped_device *md;
  45	struct dm_table *new_map;
  46};
  47
 
 
 
 
 
 
 
 
  48struct vers_iter {
  49    size_t param_size;
  50    struct dm_target_versions *vers, *old_vers;
  51    char *end;
  52    uint32_t flags;
  53};
  54
  55
  56#define NUM_BUCKETS 64
  57#define MASK_BUCKETS (NUM_BUCKETS - 1)
  58static struct list_head _name_buckets[NUM_BUCKETS];
  59static struct list_head _uuid_buckets[NUM_BUCKETS];
  60
  61static void dm_hash_remove_all(bool keep_open_devices, bool mark_deferred, bool only_deferred);
  62
  63/*
  64 * Guards access to both hash tables.
  65 */
  66static DECLARE_RWSEM(_hash_lock);
  67
  68/*
  69 * Protects use of mdptr to obtain hash cell name and uuid from mapped device.
  70 */
  71static DEFINE_MUTEX(dm_hash_cells_mutex);
  72
  73static void init_buckets(struct list_head *buckets)
  74{
  75	unsigned int i;
  76
  77	for (i = 0; i < NUM_BUCKETS; i++)
  78		INIT_LIST_HEAD(buckets + i);
  79}
  80
  81static int dm_hash_init(void)
  82{
  83	init_buckets(_name_buckets);
  84	init_buckets(_uuid_buckets);
  85	return 0;
  86}
  87
  88static void dm_hash_exit(void)
  89{
  90	dm_hash_remove_all(false, false, false);
  91}
  92
  93/*-----------------------------------------------------------------
  94 * Hash function:
  95 * We're not really concerned with the str hash function being
  96 * fast since it's only used by the ioctl interface.
  97 *---------------------------------------------------------------*/
  98static unsigned int hash_str(const char *str)
  99{
 100	const unsigned int hash_mult = 2654435387U;
 101	unsigned int h = 0;
 102
 103	while (*str)
 104		h = (h + (unsigned int) *str++) * hash_mult;
 105
 106	return h & MASK_BUCKETS;
 107}
 108
 109/*-----------------------------------------------------------------
 110 * Code for looking up a device by name
 111 *---------------------------------------------------------------*/
 112static struct hash_cell *__get_name_cell(const char *str)
 113{
 114	struct hash_cell *hc;
 115	unsigned int h = hash_str(str);
 116
 117	list_for_each_entry (hc, _name_buckets + h, name_list)
 118		if (!strcmp(hc->name, str)) {
 119			dm_get(hc->md);
 120			return hc;
 121		}
 122
 123	return NULL;
 124}
 125
 126static struct hash_cell *__get_uuid_cell(const char *str)
 127{
 128	struct hash_cell *hc;
 129	unsigned int h = hash_str(str);
 130
 131	list_for_each_entry (hc, _uuid_buckets + h, uuid_list)
 132		if (!strcmp(hc->uuid, str)) {
 133			dm_get(hc->md);
 134			return hc;
 135		}
 136
 137	return NULL;
 138}
 139
 140static struct hash_cell *__get_dev_cell(uint64_t dev)
 141{
 142	struct mapped_device *md;
 143	struct hash_cell *hc;
 144
 145	md = dm_get_md(huge_decode_dev(dev));
 146	if (!md)
 147		return NULL;
 148
 149	hc = dm_get_mdptr(md);
 150	if (!hc) {
 151		dm_put(md);
 152		return NULL;
 153	}
 154
 155	return hc;
 156}
 157
 158/*-----------------------------------------------------------------
 159 * Inserting, removing and renaming a device.
 160 *---------------------------------------------------------------*/
 161static struct hash_cell *alloc_cell(const char *name, const char *uuid,
 162				    struct mapped_device *md)
 163{
 164	struct hash_cell *hc;
 165
 166	hc = kmalloc(sizeof(*hc), GFP_KERNEL);
 167	if (!hc)
 168		return NULL;
 169
 170	hc->name = kstrdup(name, GFP_KERNEL);
 171	if (!hc->name) {
 172		kfree(hc);
 173		return NULL;
 174	}
 175
 176	if (!uuid)
 177		hc->uuid = NULL;
 178
 179	else {
 180		hc->uuid = kstrdup(uuid, GFP_KERNEL);
 181		if (!hc->uuid) {
 182			kfree(hc->name);
 183			kfree(hc);
 184			return NULL;
 185		}
 186	}
 187
 188	INIT_LIST_HEAD(&hc->name_list);
 189	INIT_LIST_HEAD(&hc->uuid_list);
 190	hc->md = md;
 191	hc->new_map = NULL;
 192	return hc;
 193}
 194
 195static void free_cell(struct hash_cell *hc)
 196{
 197	if (hc) {
 198		kfree(hc->name);
 199		kfree(hc->uuid);
 200		kfree(hc);
 201	}
 202}
 203
 204/*
 205 * The kdev_t and uuid of a device can never change once it is
 206 * initially inserted.
 207 */
 208static int dm_hash_insert(const char *name, const char *uuid, struct mapped_device *md)
 209{
 210	struct hash_cell *cell, *hc;
 211
 212	/*
 213	 * Allocate the new cells.
 214	 */
 215	cell = alloc_cell(name, uuid, md);
 216	if (!cell)
 217		return -ENOMEM;
 218
 219	/*
 220	 * Insert the cell into both hash tables.
 221	 */
 222	down_write(&_hash_lock);
 223	hc = __get_name_cell(name);
 224	if (hc) {
 225		dm_put(hc->md);
 226		goto bad;
 227	}
 228
 229	list_add(&cell->name_list, _name_buckets + hash_str(name));
 230
 231	if (uuid) {
 232		hc = __get_uuid_cell(uuid);
 233		if (hc) {
 234			list_del(&cell->name_list);
 235			dm_put(hc->md);
 236			goto bad;
 237		}
 238		list_add(&cell->uuid_list, _uuid_buckets + hash_str(uuid));
 239	}
 240	dm_get(md);
 241	mutex_lock(&dm_hash_cells_mutex);
 242	dm_set_mdptr(md, cell);
 243	mutex_unlock(&dm_hash_cells_mutex);
 244	up_write(&_hash_lock);
 245
 246	return 0;
 247
 248 bad:
 249	up_write(&_hash_lock);
 250	free_cell(cell);
 251	return -EBUSY;
 252}
 253
 254static struct dm_table *__hash_remove(struct hash_cell *hc)
 255{
 256	struct dm_table *table;
 257	int srcu_idx;
 258
 259	/* remove from the dev hash */
 260	list_del(&hc->uuid_list);
 261	list_del(&hc->name_list);
 262	mutex_lock(&dm_hash_cells_mutex);
 263	dm_set_mdptr(hc->md, NULL);
 264	mutex_unlock(&dm_hash_cells_mutex);
 265
 266	table = dm_get_live_table(hc->md, &srcu_idx);
 267	if (table)
 268		dm_table_event(table);
 269	dm_put_live_table(hc->md, srcu_idx);
 270
 271	table = NULL;
 272	if (hc->new_map)
 273		table = hc->new_map;
 274	dm_put(hc->md);
 275	free_cell(hc);
 276
 277	return table;
 278}
 279
 280static void dm_hash_remove_all(bool keep_open_devices, bool mark_deferred, bool only_deferred)
 281{
 282	int i, dev_skipped;
 283	struct hash_cell *hc;
 284	struct mapped_device *md;
 285	struct dm_table *t;
 286
 287retry:
 288	dev_skipped = 0;
 289
 290	down_write(&_hash_lock);
 291
 292	for (i = 0; i < NUM_BUCKETS; i++) {
 293		list_for_each_entry(hc, _name_buckets + i, name_list) {
 294			md = hc->md;
 295			dm_get(md);
 296
 297			if (keep_open_devices &&
 298			    dm_lock_for_deletion(md, mark_deferred, only_deferred)) {
 299				dm_put(md);
 300				dev_skipped++;
 301				continue;
 302			}
 303
 304			t = __hash_remove(hc);
 305
 306			up_write(&_hash_lock);
 307
 308			if (t) {
 309				dm_sync_table(md);
 310				dm_table_destroy(t);
 311			}
 312			dm_put(md);
 313			if (likely(keep_open_devices))
 314				dm_destroy(md);
 315			else
 316				dm_destroy_immediate(md);
 317
 318			/*
 319			 * Some mapped devices may be using other mapped
 320			 * devices, so repeat until we make no further
 321			 * progress.  If a new mapped device is created
 322			 * here it will also get removed.
 323			 */
 324			goto retry;
 325		}
 326	}
 327
 328	up_write(&_hash_lock);
 329
 330	if (dev_skipped)
 331		DMWARN("remove_all left %d open device(s)", dev_skipped);
 332}
 333
 334/*
 335 * Set the uuid of a hash_cell that isn't already set.
 336 */
 337static void __set_cell_uuid(struct hash_cell *hc, char *new_uuid)
 338{
 339	mutex_lock(&dm_hash_cells_mutex);
 340	hc->uuid = new_uuid;
 341	mutex_unlock(&dm_hash_cells_mutex);
 342
 343	list_add(&hc->uuid_list, _uuid_buckets + hash_str(new_uuid));
 344}
 345
 346/*
 347 * Changes the name of a hash_cell and returns the old name for
 348 * the caller to free.
 349 */
 350static char *__change_cell_name(struct hash_cell *hc, char *new_name)
 351{
 352	char *old_name;
 353
 354	/*
 355	 * Rename and move the name cell.
 356	 */
 357	list_del(&hc->name_list);
 358	old_name = hc->name;
 359
 360	mutex_lock(&dm_hash_cells_mutex);
 361	hc->name = new_name;
 362	mutex_unlock(&dm_hash_cells_mutex);
 363
 364	list_add(&hc->name_list, _name_buckets + hash_str(new_name));
 365
 366	return old_name;
 367}
 368
 369static struct mapped_device *dm_hash_rename(struct dm_ioctl *param,
 370					    const char *new)
 371{
 372	char *new_data, *old_name = NULL;
 373	struct hash_cell *hc;
 374	struct dm_table *table;
 375	struct mapped_device *md;
 376	unsigned change_uuid = (param->flags & DM_UUID_FLAG) ? 1 : 0;
 377	int srcu_idx;
 378
 379	/*
 380	 * duplicate new.
 381	 */
 382	new_data = kstrdup(new, GFP_KERNEL);
 383	if (!new_data)
 384		return ERR_PTR(-ENOMEM);
 385
 386	down_write(&_hash_lock);
 387
 388	/*
 389	 * Is new free ?
 390	 */
 391	if (change_uuid)
 392		hc = __get_uuid_cell(new);
 393	else
 394		hc = __get_name_cell(new);
 395
 396	if (hc) {
 397		DMWARN("Unable to change %s on mapped device %s to one that "
 398		       "already exists: %s",
 399		       change_uuid ? "uuid" : "name",
 400		       param->name, new);
 401		dm_put(hc->md);
 402		up_write(&_hash_lock);
 403		kfree(new_data);
 404		return ERR_PTR(-EBUSY);
 405	}
 406
 407	/*
 408	 * Is there such a device as 'old' ?
 409	 */
 410	hc = __get_name_cell(param->name);
 411	if (!hc) {
 412		DMWARN("Unable to rename non-existent device, %s to %s%s",
 413		       param->name, change_uuid ? "uuid " : "", new);
 414		up_write(&_hash_lock);
 415		kfree(new_data);
 416		return ERR_PTR(-ENXIO);
 417	}
 418
 419	/*
 420	 * Does this device already have a uuid?
 421	 */
 422	if (change_uuid && hc->uuid) {
 423		DMWARN("Unable to change uuid of mapped device %s to %s "
 424		       "because uuid is already set to %s",
 425		       param->name, new, hc->uuid);
 426		dm_put(hc->md);
 427		up_write(&_hash_lock);
 428		kfree(new_data);
 429		return ERR_PTR(-EINVAL);
 430	}
 431
 432	if (change_uuid)
 433		__set_cell_uuid(hc, new_data);
 434	else
 435		old_name = __change_cell_name(hc, new_data);
 436
 437	/*
 438	 * Wake up any dm event waiters.
 439	 */
 440	table = dm_get_live_table(hc->md, &srcu_idx);
 441	if (table)
 442		dm_table_event(table);
 443	dm_put_live_table(hc->md, srcu_idx);
 444
 445	if (!dm_kobject_uevent(hc->md, KOBJ_CHANGE, param->event_nr))
 446		param->flags |= DM_UEVENT_GENERATED_FLAG;
 447
 448	md = hc->md;
 449	up_write(&_hash_lock);
 450	kfree(old_name);
 451
 452	return md;
 453}
 454
 455void dm_deferred_remove(void)
 456{
 457	dm_hash_remove_all(true, false, true);
 458}
 459
 460/*-----------------------------------------------------------------
 461 * Implementation of the ioctl commands
 462 *---------------------------------------------------------------*/
 463/*
 464 * All the ioctl commands get dispatched to functions with this
 465 * prototype.
 466 */
 467typedef int (*ioctl_fn)(struct file *filp, struct dm_ioctl *param, size_t param_size);
 468
 469static int remove_all(struct file *filp, struct dm_ioctl *param, size_t param_size)
 470{
 471	dm_hash_remove_all(true, !!(param->flags & DM_DEFERRED_REMOVE), false);
 472	param->data_size = 0;
 473	return 0;
 474}
 475
 476/*
 477 * Round up the ptr to an 8-byte boundary.
 478 */
 479#define ALIGN_MASK 7
 480static inline size_t align_val(size_t val)
 481{
 482	return (val + ALIGN_MASK) & ~ALIGN_MASK;
 483}
 484static inline void *align_ptr(void *ptr)
 485{
 486	return (void *)align_val((size_t)ptr);
 487}
 488
 489/*
 490 * Retrieves the data payload buffer from an already allocated
 491 * struct dm_ioctl.
 492 */
 493static void *get_result_buffer(struct dm_ioctl *param, size_t param_size,
 494			       size_t *len)
 495{
 496	param->data_start = align_ptr(param + 1) - (void *) param;
 497
 498	if (param->data_start < param_size)
 499		*len = param_size - param->data_start;
 500	else
 501		*len = 0;
 502
 503	return ((void *) param) + param->data_start;
 504}
 505
 506static int list_devices(struct file *filp, struct dm_ioctl *param, size_t param_size)
 507{
 508	unsigned int i;
 509	struct hash_cell *hc;
 510	size_t len, needed = 0;
 511	struct gendisk *disk;
 512	struct dm_name_list *orig_nl, *nl, *old_nl = NULL;
 513	uint32_t *event_nr;
 514
 515	down_write(&_hash_lock);
 516
 517	/*
 518	 * Loop through all the devices working out how much
 519	 * space we need.
 520	 */
 521	for (i = 0; i < NUM_BUCKETS; i++) {
 522		list_for_each_entry (hc, _name_buckets + i, name_list) {
 523			needed += align_val(offsetof(struct dm_name_list, name) + strlen(hc->name) + 1);
 524			needed += align_val(sizeof(uint32_t));
 
 525		}
 526	}
 527
 528	/*
 529	 * Grab our output buffer.
 530	 */
 531	nl = orig_nl = get_result_buffer(param, param_size, &len);
 532	if (len < needed) {
 533		param->flags |= DM_BUFFER_FULL_FLAG;
 534		goto out;
 535	}
 536	param->data_size = param->data_start + needed;
 537
 538	nl->dev = 0;	/* Flags no data */
 539
 540	/*
 541	 * Now loop through filling out the names.
 542	 */
 543	for (i = 0; i < NUM_BUCKETS; i++) {
 544		list_for_each_entry (hc, _name_buckets + i, name_list) {
 545			if (old_nl)
 546				old_nl->next = (uint32_t) ((void *) nl -
 547							   (void *) old_nl);
 548			disk = dm_disk(hc->md);
 549			nl->dev = huge_encode_dev(disk_devt(disk));
 550			nl->next = 0;
 551			strcpy(nl->name, hc->name);
 552
 553			old_nl = nl;
 554			event_nr = align_ptr(nl->name + strlen(hc->name) + 1);
 555			*event_nr = dm_get_event_nr(hc->md);
 556			nl = align_ptr(event_nr + 1);
 557		}
 558	}
 559	/*
 560	 * If mismatch happens, security may be compromised due to buffer
 561	 * overflow, so it's better to crash.
 562	 */
 563	BUG_ON((char *)nl - (char *)orig_nl != needed);
 564
 565 out:
 566	up_write(&_hash_lock);
 567	return 0;
 568}
 569
 570static void list_version_get_needed(struct target_type *tt, void *needed_param)
 571{
 572    size_t *needed = needed_param;
 573
 574    *needed += sizeof(struct dm_target_versions);
 575    *needed += strlen(tt->name);
 576    *needed += ALIGN_MASK;
 577}
 578
 579static void list_version_get_info(struct target_type *tt, void *param)
 580{
 581    struct vers_iter *info = param;
 582
 583    /* Check space - it might have changed since the first iteration */
 584    if ((char *)info->vers + sizeof(tt->version) + strlen(tt->name) + 1 >
 585	info->end) {
 586
 587	info->flags = DM_BUFFER_FULL_FLAG;
 588	return;
 589    }
 590
 591    if (info->old_vers)
 592	info->old_vers->next = (uint32_t) ((void *)info->vers -
 593					   (void *)info->old_vers);
 594    info->vers->version[0] = tt->version[0];
 595    info->vers->version[1] = tt->version[1];
 596    info->vers->version[2] = tt->version[2];
 597    info->vers->next = 0;
 598    strcpy(info->vers->name, tt->name);
 599
 600    info->old_vers = info->vers;
 601    info->vers = align_ptr(((void *) ++info->vers) + strlen(tt->name) + 1);
 602}
 603
 604static int __list_versions(struct dm_ioctl *param, size_t param_size, const char *name)
 605{
 606	size_t len, needed = 0;
 607	struct dm_target_versions *vers;
 608	struct vers_iter iter_info;
 609	struct target_type *tt = NULL;
 610
 611	if (name) {
 612		tt = dm_get_target_type(name);
 613		if (!tt)
 614			return -EINVAL;
 615	}
 616
 617	/*
 618	 * Loop through all the devices working out how much
 619	 * space we need.
 620	 */
 621	if (!tt)
 622		dm_target_iterate(list_version_get_needed, &needed);
 623	else
 624		list_version_get_needed(tt, &needed);
 625
 626	/*
 627	 * Grab our output buffer.
 628	 */
 629	vers = get_result_buffer(param, param_size, &len);
 630	if (len < needed) {
 631		param->flags |= DM_BUFFER_FULL_FLAG;
 632		goto out;
 633	}
 634	param->data_size = param->data_start + needed;
 635
 636	iter_info.param_size = param_size;
 637	iter_info.old_vers = NULL;
 638	iter_info.vers = vers;
 639	iter_info.flags = 0;
 640	iter_info.end = (char *)vers+len;
 641
 642	/*
 643	 * Now loop through filling out the names & versions.
 644	 */
 645	if (!tt)
 646		dm_target_iterate(list_version_get_info, &iter_info);
 647	else
 648		list_version_get_info(tt, &iter_info);
 649	param->flags |= iter_info.flags;
 650
 651 out:
 652	if (tt)
 653		dm_put_target_type(tt);
 654	return 0;
 655}
 656
 657static int list_versions(struct file *filp, struct dm_ioctl *param, size_t param_size)
 658{
 659	return __list_versions(param, param_size, NULL);
 660}
 661
 662static int get_target_version(struct file *filp, struct dm_ioctl *param, size_t param_size)
 663{
 664	return __list_versions(param, param_size, param->name);
 665}
 666
 667static int check_name(const char *name)
 668{
 669	if (strchr(name, '/')) {
 670		DMWARN("invalid device name");
 671		return -EINVAL;
 672	}
 673
 674	return 0;
 675}
 676
 677/*
 678 * On successful return, the caller must not attempt to acquire
 679 * _hash_lock without first calling dm_put_live_table, because dm_table_destroy
 680 * waits for this dm_put_live_table and could be called under this lock.
 681 */
 682static struct dm_table *dm_get_inactive_table(struct mapped_device *md, int *srcu_idx)
 683{
 684	struct hash_cell *hc;
 685	struct dm_table *table = NULL;
 686
 687	/* increment rcu count, we don't care about the table pointer */
 688	dm_get_live_table(md, srcu_idx);
 689
 690	down_read(&_hash_lock);
 691	hc = dm_get_mdptr(md);
 692	if (!hc || hc->md != md) {
 693		DMWARN("device has been removed from the dev hash table.");
 694		goto out;
 695	}
 696
 697	table = hc->new_map;
 698
 699out:
 700	up_read(&_hash_lock);
 701
 702	return table;
 703}
 704
 705static struct dm_table *dm_get_live_or_inactive_table(struct mapped_device *md,
 706						      struct dm_ioctl *param,
 707						      int *srcu_idx)
 708{
 709	return (param->flags & DM_QUERY_INACTIVE_TABLE_FLAG) ?
 710		dm_get_inactive_table(md, srcu_idx) : dm_get_live_table(md, srcu_idx);
 711}
 712
 713/*
 714 * Fills in a dm_ioctl structure, ready for sending back to
 715 * userland.
 716 */
 717static void __dev_status(struct mapped_device *md, struct dm_ioctl *param)
 718{
 719	struct gendisk *disk = dm_disk(md);
 720	struct dm_table *table;
 721	int srcu_idx;
 722
 723	param->flags &= ~(DM_SUSPEND_FLAG | DM_READONLY_FLAG |
 724			  DM_ACTIVE_PRESENT_FLAG | DM_INTERNAL_SUSPEND_FLAG);
 725
 726	if (dm_suspended_md(md))
 727		param->flags |= DM_SUSPEND_FLAG;
 728
 729	if (dm_suspended_internally_md(md))
 730		param->flags |= DM_INTERNAL_SUSPEND_FLAG;
 731
 732	if (dm_test_deferred_remove_flag(md))
 733		param->flags |= DM_DEFERRED_REMOVE;
 734
 735	param->dev = huge_encode_dev(disk_devt(disk));
 736
 737	/*
 738	 * Yes, this will be out of date by the time it gets back
 739	 * to userland, but it is still very useful for
 740	 * debugging.
 741	 */
 742	param->open_count = dm_open_count(md);
 743
 744	param->event_nr = dm_get_event_nr(md);
 745	param->target_count = 0;
 746
 747	table = dm_get_live_table(md, &srcu_idx);
 748	if (table) {
 749		if (!(param->flags & DM_QUERY_INACTIVE_TABLE_FLAG)) {
 750			if (get_disk_ro(disk))
 751				param->flags |= DM_READONLY_FLAG;
 752			param->target_count = dm_table_get_num_targets(table);
 753		}
 754
 755		param->flags |= DM_ACTIVE_PRESENT_FLAG;
 756	}
 757	dm_put_live_table(md, srcu_idx);
 758
 759	if (param->flags & DM_QUERY_INACTIVE_TABLE_FLAG) {
 760		int srcu_idx;
 761		table = dm_get_inactive_table(md, &srcu_idx);
 762		if (table) {
 763			if (!(dm_table_get_mode(table) & FMODE_WRITE))
 764				param->flags |= DM_READONLY_FLAG;
 765			param->target_count = dm_table_get_num_targets(table);
 766		}
 767		dm_put_live_table(md, srcu_idx);
 768	}
 769}
 770
 771static int dev_create(struct file *filp, struct dm_ioctl *param, size_t param_size)
 772{
 773	int r, m = DM_ANY_MINOR;
 774	struct mapped_device *md;
 775
 776	r = check_name(param->name);
 777	if (r)
 778		return r;
 779
 780	if (param->flags & DM_PERSISTENT_DEV_FLAG)
 781		m = MINOR(huge_decode_dev(param->dev));
 782
 783	r = dm_create(m, &md);
 784	if (r)
 785		return r;
 786
 787	r = dm_hash_insert(param->name, *param->uuid ? param->uuid : NULL, md);
 788	if (r) {
 789		dm_put(md);
 790		dm_destroy(md);
 791		return r;
 792	}
 793
 794	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
 795
 796	__dev_status(md, param);
 797
 798	dm_put(md);
 799
 800	return 0;
 801}
 802
 803/*
 804 * Always use UUID for lookups if it's present, otherwise use name or dev.
 805 */
 806static struct hash_cell *__find_device_hash_cell(struct dm_ioctl *param)
 807{
 808	struct hash_cell *hc = NULL;
 809
 810	if (*param->uuid) {
 811		if (*param->name || param->dev)
 812			return NULL;
 813
 814		hc = __get_uuid_cell(param->uuid);
 815		if (!hc)
 816			return NULL;
 817	} else if (*param->name) {
 818		if (param->dev)
 819			return NULL;
 820
 821		hc = __get_name_cell(param->name);
 822		if (!hc)
 823			return NULL;
 824	} else if (param->dev) {
 825		hc = __get_dev_cell(param->dev);
 826		if (!hc)
 827			return NULL;
 828	} else
 829		return NULL;
 830
 831	/*
 832	 * Sneakily write in both the name and the uuid
 833	 * while we have the cell.
 834	 */
 835	strlcpy(param->name, hc->name, sizeof(param->name));
 836	if (hc->uuid)
 837		strlcpy(param->uuid, hc->uuid, sizeof(param->uuid));
 838	else
 839		param->uuid[0] = '\0';
 840
 841	if (hc->new_map)
 842		param->flags |= DM_INACTIVE_PRESENT_FLAG;
 843	else
 844		param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
 845
 846	return hc;
 847}
 848
 849static struct mapped_device *find_device(struct dm_ioctl *param)
 850{
 851	struct hash_cell *hc;
 852	struct mapped_device *md = NULL;
 853
 854	down_read(&_hash_lock);
 855	hc = __find_device_hash_cell(param);
 856	if (hc)
 857		md = hc->md;
 858	up_read(&_hash_lock);
 859
 860	return md;
 861}
 862
 863static int dev_remove(struct file *filp, struct dm_ioctl *param, size_t param_size)
 864{
 865	struct hash_cell *hc;
 866	struct mapped_device *md;
 867	int r;
 868	struct dm_table *t;
 869
 870	down_write(&_hash_lock);
 871	hc = __find_device_hash_cell(param);
 872
 873	if (!hc) {
 874		DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
 875		up_write(&_hash_lock);
 876		return -ENXIO;
 877	}
 878
 879	md = hc->md;
 880
 881	/*
 882	 * Ensure the device is not open and nothing further can open it.
 883	 */
 884	r = dm_lock_for_deletion(md, !!(param->flags & DM_DEFERRED_REMOVE), false);
 885	if (r) {
 886		if (r == -EBUSY && param->flags & DM_DEFERRED_REMOVE) {
 887			up_write(&_hash_lock);
 888			dm_put(md);
 889			return 0;
 890		}
 891		DMDEBUG_LIMIT("unable to remove open device %s", hc->name);
 892		up_write(&_hash_lock);
 893		dm_put(md);
 894		return r;
 895	}
 896
 897	t = __hash_remove(hc);
 898	up_write(&_hash_lock);
 899
 900	if (t) {
 901		dm_sync_table(md);
 902		dm_table_destroy(t);
 903	}
 904
 905	param->flags &= ~DM_DEFERRED_REMOVE;
 906
 907	if (!dm_kobject_uevent(md, KOBJ_REMOVE, param->event_nr))
 908		param->flags |= DM_UEVENT_GENERATED_FLAG;
 909
 910	dm_put(md);
 911	dm_destroy(md);
 912	return 0;
 913}
 914
 915/*
 916 * Check a string doesn't overrun the chunk of
 917 * memory we copied from userland.
 918 */
 919static int invalid_str(char *str, void *end)
 920{
 921	while ((void *) str < end)
 922		if (!*str++)
 923			return 0;
 924
 925	return -EINVAL;
 926}
 927
 928static int dev_rename(struct file *filp, struct dm_ioctl *param, size_t param_size)
 929{
 930	int r;
 931	char *new_data = (char *) param + param->data_start;
 932	struct mapped_device *md;
 933	unsigned change_uuid = (param->flags & DM_UUID_FLAG) ? 1 : 0;
 934
 935	if (new_data < param->data ||
 936	    invalid_str(new_data, (void *) param + param_size) || !*new_data ||
 937	    strlen(new_data) > (change_uuid ? DM_UUID_LEN - 1 : DM_NAME_LEN - 1)) {
 938		DMWARN("Invalid new mapped device name or uuid string supplied.");
 939		return -EINVAL;
 940	}
 941
 942	if (!change_uuid) {
 943		r = check_name(new_data);
 944		if (r)
 945			return r;
 946	}
 947
 948	md = dm_hash_rename(param, new_data);
 949	if (IS_ERR(md))
 950		return PTR_ERR(md);
 951
 952	__dev_status(md, param);
 953	dm_put(md);
 954
 955	return 0;
 956}
 957
 958static int dev_set_geometry(struct file *filp, struct dm_ioctl *param, size_t param_size)
 959{
 960	int r = -EINVAL, x;
 961	struct mapped_device *md;
 962	struct hd_geometry geometry;
 963	unsigned long indata[4];
 964	char *geostr = (char *) param + param->data_start;
 965	char dummy;
 966
 967	md = find_device(param);
 968	if (!md)
 969		return -ENXIO;
 970
 971	if (geostr < param->data ||
 972	    invalid_str(geostr, (void *) param + param_size)) {
 973		DMWARN("Invalid geometry supplied.");
 974		goto out;
 975	}
 976
 977	x = sscanf(geostr, "%lu %lu %lu %lu%c", indata,
 978		   indata + 1, indata + 2, indata + 3, &dummy);
 979
 980	if (x != 4) {
 981		DMWARN("Unable to interpret geometry settings.");
 982		goto out;
 983	}
 984
 985	if (indata[0] > 65535 || indata[1] > 255 ||
 986	    indata[2] > 255 || indata[3] > ULONG_MAX) {
 987		DMWARN("Geometry exceeds range limits.");
 988		goto out;
 989	}
 990
 991	geometry.cylinders = indata[0];
 992	geometry.heads = indata[1];
 993	geometry.sectors = indata[2];
 994	geometry.start = indata[3];
 995
 996	r = dm_set_geometry(md, &geometry);
 997
 998	param->data_size = 0;
 999
1000out:
1001	dm_put(md);
1002	return r;
1003}
1004
1005static int do_suspend(struct dm_ioctl *param)
1006{
1007	int r = 0;
1008	unsigned suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
1009	struct mapped_device *md;
1010
1011	md = find_device(param);
1012	if (!md)
1013		return -ENXIO;
1014
1015	if (param->flags & DM_SKIP_LOCKFS_FLAG)
1016		suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
1017	if (param->flags & DM_NOFLUSH_FLAG)
1018		suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
1019
1020	if (!dm_suspended_md(md)) {
1021		r = dm_suspend(md, suspend_flags);
1022		if (r)
1023			goto out;
1024	}
1025
1026	__dev_status(md, param);
1027
1028out:
1029	dm_put(md);
1030
1031	return r;
1032}
1033
1034static int do_resume(struct dm_ioctl *param)
1035{
1036	int r = 0;
1037	unsigned suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
1038	struct hash_cell *hc;
1039	struct mapped_device *md;
1040	struct dm_table *new_map, *old_map = NULL;
1041
1042	down_write(&_hash_lock);
1043
1044	hc = __find_device_hash_cell(param);
1045	if (!hc) {
1046		DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
1047		up_write(&_hash_lock);
1048		return -ENXIO;
1049	}
1050
1051	md = hc->md;
1052
1053	new_map = hc->new_map;
1054	hc->new_map = NULL;
1055	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1056
1057	up_write(&_hash_lock);
1058
1059	/* Do we need to load a new map ? */
1060	if (new_map) {
1061		/* Suspend if it isn't already suspended */
1062		if (param->flags & DM_SKIP_LOCKFS_FLAG)
1063			suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
1064		if (param->flags & DM_NOFLUSH_FLAG)
1065			suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
1066		if (!dm_suspended_md(md))
1067			dm_suspend(md, suspend_flags);
1068
1069		old_map = dm_swap_table(md, new_map);
1070		if (IS_ERR(old_map)) {
1071			dm_sync_table(md);
1072			dm_table_destroy(new_map);
1073			dm_put(md);
1074			return PTR_ERR(old_map);
1075		}
1076
1077		if (dm_table_get_mode(new_map) & FMODE_WRITE)
1078			set_disk_ro(dm_disk(md), 0);
1079		else
1080			set_disk_ro(dm_disk(md), 1);
1081	}
1082
1083	if (dm_suspended_md(md)) {
1084		r = dm_resume(md);
1085		if (!r && !dm_kobject_uevent(md, KOBJ_CHANGE, param->event_nr))
1086			param->flags |= DM_UEVENT_GENERATED_FLAG;
1087	}
1088
1089	/*
1090	 * Since dm_swap_table synchronizes RCU, nobody should be in
1091	 * read-side critical section already.
1092	 */
1093	if (old_map)
1094		dm_table_destroy(old_map);
1095
1096	if (!r)
1097		__dev_status(md, param);
1098
1099	dm_put(md);
1100	return r;
1101}
1102
1103/*
1104 * Set or unset the suspension state of a device.
1105 * If the device already is in the requested state we just return its status.
1106 */
1107static int dev_suspend(struct file *filp, struct dm_ioctl *param, size_t param_size)
1108{
1109	if (param->flags & DM_SUSPEND_FLAG)
1110		return do_suspend(param);
1111
1112	return do_resume(param);
1113}
1114
1115/*
1116 * Copies device info back to user space, used by
1117 * the create and info ioctls.
1118 */
1119static int dev_status(struct file *filp, struct dm_ioctl *param, size_t param_size)
1120{
1121	struct mapped_device *md;
1122
1123	md = find_device(param);
1124	if (!md)
1125		return -ENXIO;
1126
1127	__dev_status(md, param);
1128	dm_put(md);
1129
1130	return 0;
1131}
1132
1133/*
1134 * Build up the status struct for each target
1135 */
1136static void retrieve_status(struct dm_table *table,
1137			    struct dm_ioctl *param, size_t param_size)
1138{
1139	unsigned int i, num_targets;
1140	struct dm_target_spec *spec;
1141	char *outbuf, *outptr;
1142	status_type_t type;
1143	size_t remaining, len, used = 0;
1144	unsigned status_flags = 0;
1145
1146	outptr = outbuf = get_result_buffer(param, param_size, &len);
1147
1148	if (param->flags & DM_STATUS_TABLE_FLAG)
1149		type = STATUSTYPE_TABLE;
1150	else
1151		type = STATUSTYPE_INFO;
1152
1153	/* Get all the target info */
1154	num_targets = dm_table_get_num_targets(table);
1155	for (i = 0; i < num_targets; i++) {
1156		struct dm_target *ti = dm_table_get_target(table, i);
1157		size_t l;
1158
1159		remaining = len - (outptr - outbuf);
1160		if (remaining <= sizeof(struct dm_target_spec)) {
1161			param->flags |= DM_BUFFER_FULL_FLAG;
1162			break;
1163		}
1164
1165		spec = (struct dm_target_spec *) outptr;
1166
1167		spec->status = 0;
1168		spec->sector_start = ti->begin;
1169		spec->length = ti->len;
1170		strncpy(spec->target_type, ti->type->name,
1171			sizeof(spec->target_type) - 1);
1172
1173		outptr += sizeof(struct dm_target_spec);
1174		remaining = len - (outptr - outbuf);
1175		if (remaining <= 0) {
1176			param->flags |= DM_BUFFER_FULL_FLAG;
1177			break;
1178		}
1179
1180		/* Get the status/table string from the target driver */
1181		if (ti->type->status) {
1182			if (param->flags & DM_NOFLUSH_FLAG)
1183				status_flags |= DM_STATUS_NOFLUSH_FLAG;
1184			ti->type->status(ti, type, status_flags, outptr, remaining);
1185		} else
1186			outptr[0] = '\0';
1187
1188		l = strlen(outptr) + 1;
1189		if (l == remaining) {
1190			param->flags |= DM_BUFFER_FULL_FLAG;
1191			break;
1192		}
1193
1194		outptr += l;
1195		used = param->data_start + (outptr - outbuf);
1196
1197		outptr = align_ptr(outptr);
1198		spec->next = outptr - outbuf;
1199	}
1200
1201	if (used)
1202		param->data_size = used;
1203
1204	param->target_count = num_targets;
1205}
1206
1207/*
1208 * Wait for a device to report an event
1209 */
1210static int dev_wait(struct file *filp, struct dm_ioctl *param, size_t param_size)
1211{
1212	int r = 0;
1213	struct mapped_device *md;
1214	struct dm_table *table;
1215	int srcu_idx;
1216
1217	md = find_device(param);
1218	if (!md)
1219		return -ENXIO;
1220
1221	/*
1222	 * Wait for a notification event
1223	 */
1224	if (dm_wait_event(md, param->event_nr)) {
1225		r = -ERESTARTSYS;
1226		goto out;
1227	}
1228
1229	/*
1230	 * The userland program is going to want to know what
1231	 * changed to trigger the event, so we may as well tell
1232	 * him and save an ioctl.
1233	 */
1234	__dev_status(md, param);
1235
1236	table = dm_get_live_or_inactive_table(md, param, &srcu_idx);
1237	if (table)
1238		retrieve_status(table, param, param_size);
1239	dm_put_live_table(md, srcu_idx);
1240
1241out:
1242	dm_put(md);
1243
1244	return r;
1245}
1246
1247/*
1248 * Remember the global event number and make it possible to poll
1249 * for further events.
1250 */
1251static int dev_arm_poll(struct file *filp, struct dm_ioctl *param, size_t param_size)
1252{
1253	struct dm_file *priv = filp->private_data;
1254
1255	priv->global_event_nr = atomic_read(&dm_global_event_nr);
1256
1257	return 0;
1258}
1259
1260static inline fmode_t get_mode(struct dm_ioctl *param)
1261{
1262	fmode_t mode = FMODE_READ | FMODE_WRITE;
1263
1264	if (param->flags & DM_READONLY_FLAG)
1265		mode = FMODE_READ;
1266
1267	return mode;
1268}
1269
1270static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
1271		       struct dm_target_spec **spec, char **target_params)
1272{
1273	*spec = (struct dm_target_spec *) ((unsigned char *) last + next);
1274	*target_params = (char *) (*spec + 1);
1275
1276	if (*spec < (last + 1))
1277		return -EINVAL;
1278
1279	return invalid_str(*target_params, end);
1280}
1281
1282static int populate_table(struct dm_table *table,
1283			  struct dm_ioctl *param, size_t param_size)
1284{
1285	int r;
1286	unsigned int i = 0;
1287	struct dm_target_spec *spec = (struct dm_target_spec *) param;
1288	uint32_t next = param->data_start;
1289	void *end = (void *) param + param_size;
1290	char *target_params;
1291
1292	if (!param->target_count) {
1293		DMWARN("populate_table: no targets specified");
1294		return -EINVAL;
1295	}
1296
1297	for (i = 0; i < param->target_count; i++) {
1298
1299		r = next_target(spec, next, end, &spec, &target_params);
1300		if (r) {
1301			DMWARN("unable to find target");
1302			return r;
1303		}
1304
1305		r = dm_table_add_target(table, spec->target_type,
1306					(sector_t) spec->sector_start,
1307					(sector_t) spec->length,
1308					target_params);
1309		if (r) {
1310			DMWARN("error adding target to table");
1311			return r;
1312		}
1313
1314		next = spec->next;
1315	}
1316
1317	return dm_table_complete(table);
1318}
1319
1320static bool is_valid_type(enum dm_queue_mode cur, enum dm_queue_mode new)
1321{
1322	if (cur == new ||
1323	    (cur == DM_TYPE_BIO_BASED && new == DM_TYPE_DAX_BIO_BASED))
1324		return true;
1325
1326	return false;
1327}
1328
1329static int table_load(struct file *filp, struct dm_ioctl *param, size_t param_size)
1330{
1331	int r;
1332	struct hash_cell *hc;
1333	struct dm_table *t, *old_map = NULL;
1334	struct mapped_device *md;
1335	struct target_type *immutable_target_type;
1336
1337	md = find_device(param);
1338	if (!md)
1339		return -ENXIO;
1340
1341	r = dm_table_create(&t, get_mode(param), param->target_count, md);
1342	if (r)
1343		goto err;
1344
1345	/* Protect md->type and md->queue against concurrent table loads. */
1346	dm_lock_md_type(md);
1347	r = populate_table(t, param, param_size);
1348	if (r)
1349		goto err_unlock_md_type;
1350
1351	immutable_target_type = dm_get_immutable_target_type(md);
1352	if (immutable_target_type &&
1353	    (immutable_target_type != dm_table_get_immutable_target_type(t)) &&
1354	    !dm_table_get_wildcard_target(t)) {
1355		DMWARN("can't replace immutable target type %s",
1356		       immutable_target_type->name);
1357		r = -EINVAL;
1358		goto err_unlock_md_type;
1359	}
1360
1361	if (dm_get_md_type(md) == DM_TYPE_NONE) {
1362		/* Initial table load: acquire type of table. */
1363		dm_set_md_type(md, dm_table_get_type(t));
1364
1365		/* setup md->queue to reflect md's type (may block) */
1366		r = dm_setup_md_queue(md, t);
1367		if (r) {
1368			DMWARN("unable to set up device queue for new table.");
1369			goto err_unlock_md_type;
1370		}
1371	} else if (!is_valid_type(dm_get_md_type(md), dm_table_get_type(t))) {
1372		DMWARN("can't change device type (old=%u vs new=%u) after initial table load.",
1373		       dm_get_md_type(md), dm_table_get_type(t));
1374		r = -EINVAL;
1375		goto err_unlock_md_type;
1376	}
1377
1378	dm_unlock_md_type(md);
1379
1380	/* stage inactive table */
1381	down_write(&_hash_lock);
1382	hc = dm_get_mdptr(md);
1383	if (!hc || hc->md != md) {
1384		DMWARN("device has been removed from the dev hash table.");
1385		up_write(&_hash_lock);
1386		r = -ENXIO;
1387		goto err_destroy_table;
1388	}
1389
1390	if (hc->new_map)
1391		old_map = hc->new_map;
1392	hc->new_map = t;
1393	up_write(&_hash_lock);
1394
1395	param->flags |= DM_INACTIVE_PRESENT_FLAG;
1396	__dev_status(md, param);
1397
1398	if (old_map) {
1399		dm_sync_table(md);
1400		dm_table_destroy(old_map);
1401	}
1402
1403	dm_put(md);
1404
1405	return 0;
1406
1407err_unlock_md_type:
1408	dm_unlock_md_type(md);
1409err_destroy_table:
1410	dm_table_destroy(t);
1411err:
1412	dm_put(md);
1413
1414	return r;
1415}
1416
1417static int table_clear(struct file *filp, struct dm_ioctl *param, size_t param_size)
1418{
1419	struct hash_cell *hc;
1420	struct mapped_device *md;
1421	struct dm_table *old_map = NULL;
1422
1423	down_write(&_hash_lock);
1424
1425	hc = __find_device_hash_cell(param);
1426	if (!hc) {
1427		DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
1428		up_write(&_hash_lock);
1429		return -ENXIO;
1430	}
1431
1432	if (hc->new_map) {
1433		old_map = hc->new_map;
1434		hc->new_map = NULL;
1435	}
1436
1437	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1438
1439	__dev_status(hc->md, param);
1440	md = hc->md;
1441	up_write(&_hash_lock);
1442	if (old_map) {
1443		dm_sync_table(md);
1444		dm_table_destroy(old_map);
1445	}
1446	dm_put(md);
1447
1448	return 0;
1449}
1450
1451/*
1452 * Retrieves a list of devices used by a particular dm device.
1453 */
1454static void retrieve_deps(struct dm_table *table,
1455			  struct dm_ioctl *param, size_t param_size)
1456{
1457	unsigned int count = 0;
1458	struct list_head *tmp;
1459	size_t len, needed;
1460	struct dm_dev_internal *dd;
1461	struct dm_target_deps *deps;
1462
1463	deps = get_result_buffer(param, param_size, &len);
1464
1465	/*
1466	 * Count the devices.
1467	 */
1468	list_for_each (tmp, dm_table_get_devices(table))
1469		count++;
1470
1471	/*
1472	 * Check we have enough space.
1473	 */
1474	needed = struct_size(deps, dev, count);
1475	if (len < needed) {
1476		param->flags |= DM_BUFFER_FULL_FLAG;
1477		return;
1478	}
1479
1480	/*
1481	 * Fill in the devices.
1482	 */
1483	deps->count = count;
1484	count = 0;
1485	list_for_each_entry (dd, dm_table_get_devices(table), list)
1486		deps->dev[count++] = huge_encode_dev(dd->dm_dev->bdev->bd_dev);
1487
1488	param->data_size = param->data_start + needed;
1489}
1490
1491static int table_deps(struct file *filp, struct dm_ioctl *param, size_t param_size)
1492{
1493	struct mapped_device *md;
1494	struct dm_table *table;
1495	int srcu_idx;
1496
1497	md = find_device(param);
1498	if (!md)
1499		return -ENXIO;
1500
1501	__dev_status(md, param);
1502
1503	table = dm_get_live_or_inactive_table(md, param, &srcu_idx);
1504	if (table)
1505		retrieve_deps(table, param, param_size);
1506	dm_put_live_table(md, srcu_idx);
1507
1508	dm_put(md);
1509
1510	return 0;
1511}
1512
1513/*
1514 * Return the status of a device as a text string for each
1515 * target.
1516 */
1517static int table_status(struct file *filp, struct dm_ioctl *param, size_t param_size)
1518{
1519	struct mapped_device *md;
1520	struct dm_table *table;
1521	int srcu_idx;
1522
1523	md = find_device(param);
1524	if (!md)
1525		return -ENXIO;
1526
1527	__dev_status(md, param);
1528
1529	table = dm_get_live_or_inactive_table(md, param, &srcu_idx);
1530	if (table)
1531		retrieve_status(table, param, param_size);
1532	dm_put_live_table(md, srcu_idx);
1533
1534	dm_put(md);
1535
1536	return 0;
1537}
1538
1539/*
1540 * Process device-mapper dependent messages.  Messages prefixed with '@'
1541 * are processed by the DM core.  All others are delivered to the target.
1542 * Returns a number <= 1 if message was processed by device mapper.
1543 * Returns 2 if message should be delivered to the target.
1544 */
1545static int message_for_md(struct mapped_device *md, unsigned argc, char **argv,
1546			  char *result, unsigned maxlen)
1547{
1548	int r;
1549
1550	if (**argv != '@')
1551		return 2; /* no '@' prefix, deliver to target */
1552
1553	if (!strcasecmp(argv[0], "@cancel_deferred_remove")) {
1554		if (argc != 1) {
1555			DMERR("Invalid arguments for @cancel_deferred_remove");
1556			return -EINVAL;
1557		}
1558		return dm_cancel_deferred_remove(md);
1559	}
1560
1561	r = dm_stats_message(md, argc, argv, result, maxlen);
1562	if (r < 2)
1563		return r;
1564
1565	DMERR("Unsupported message sent to DM core: %s", argv[0]);
1566	return -EINVAL;
1567}
1568
1569/*
1570 * Pass a message to the target that's at the supplied device offset.
1571 */
1572static int target_message(struct file *filp, struct dm_ioctl *param, size_t param_size)
1573{
1574	int r, argc;
1575	char **argv;
1576	struct mapped_device *md;
1577	struct dm_table *table;
1578	struct dm_target *ti;
1579	struct dm_target_msg *tmsg = (void *) param + param->data_start;
1580	size_t maxlen;
1581	char *result = get_result_buffer(param, param_size, &maxlen);
1582	int srcu_idx;
1583
1584	md = find_device(param);
1585	if (!md)
1586		return -ENXIO;
1587
1588	if (tmsg < (struct dm_target_msg *) param->data ||
1589	    invalid_str(tmsg->message, (void *) param + param_size)) {
1590		DMWARN("Invalid target message parameters.");
1591		r = -EINVAL;
1592		goto out;
1593	}
1594
1595	r = dm_split_args(&argc, &argv, tmsg->message);
1596	if (r) {
1597		DMWARN("Failed to split target message parameters");
1598		goto out;
1599	}
1600
1601	if (!argc) {
1602		DMWARN("Empty message received.");
1603		goto out_argv;
1604	}
1605
1606	r = message_for_md(md, argc, argv, result, maxlen);
1607	if (r <= 1)
1608		goto out_argv;
1609
1610	table = dm_get_live_table(md, &srcu_idx);
1611	if (!table)
1612		goto out_table;
1613
1614	if (dm_deleting_md(md)) {
1615		r = -ENXIO;
1616		goto out_table;
1617	}
1618
1619	ti = dm_table_find_target(table, tmsg->sector);
1620	if (!ti) {
1621		DMWARN("Target message sector outside device.");
1622		r = -EINVAL;
1623	} else if (ti->type->message)
1624		r = ti->type->message(ti, argc, argv, result, maxlen);
1625	else {
1626		DMWARN("Target type does not support messages");
1627		r = -EINVAL;
1628	}
1629
1630 out_table:
1631	dm_put_live_table(md, srcu_idx);
1632 out_argv:
1633	kfree(argv);
1634 out:
1635	if (r >= 0)
1636		__dev_status(md, param);
1637
1638	if (r == 1) {
1639		param->flags |= DM_DATA_OUT_FLAG;
1640		if (dm_message_test_buffer_overflow(result, maxlen))
1641			param->flags |= DM_BUFFER_FULL_FLAG;
1642		else
1643			param->data_size = param->data_start + strlen(result) + 1;
1644		r = 0;
1645	}
1646
1647	dm_put(md);
1648	return r;
1649}
1650
1651/*
1652 * The ioctl parameter block consists of two parts, a dm_ioctl struct
1653 * followed by a data buffer.  This flag is set if the second part,
1654 * which has a variable size, is not used by the function processing
1655 * the ioctl.
1656 */
1657#define IOCTL_FLAGS_NO_PARAMS		1
1658#define IOCTL_FLAGS_ISSUE_GLOBAL_EVENT	2
1659
1660/*-----------------------------------------------------------------
1661 * Implementation of open/close/ioctl on the special char
1662 * device.
1663 *---------------------------------------------------------------*/
1664static ioctl_fn lookup_ioctl(unsigned int cmd, int *ioctl_flags)
1665{
1666	static const struct {
1667		int cmd;
1668		int flags;
1669		ioctl_fn fn;
1670	} _ioctls[] = {
1671		{DM_VERSION_CMD, 0, NULL}, /* version is dealt with elsewhere */
1672		{DM_REMOVE_ALL_CMD, IOCTL_FLAGS_NO_PARAMS | IOCTL_FLAGS_ISSUE_GLOBAL_EVENT, remove_all},
1673		{DM_LIST_DEVICES_CMD, 0, list_devices},
1674
1675		{DM_DEV_CREATE_CMD, IOCTL_FLAGS_NO_PARAMS | IOCTL_FLAGS_ISSUE_GLOBAL_EVENT, dev_create},
1676		{DM_DEV_REMOVE_CMD, IOCTL_FLAGS_NO_PARAMS | IOCTL_FLAGS_ISSUE_GLOBAL_EVENT, dev_remove},
1677		{DM_DEV_RENAME_CMD, IOCTL_FLAGS_ISSUE_GLOBAL_EVENT, dev_rename},
1678		{DM_DEV_SUSPEND_CMD, IOCTL_FLAGS_NO_PARAMS, dev_suspend},
1679		{DM_DEV_STATUS_CMD, IOCTL_FLAGS_NO_PARAMS, dev_status},
1680		{DM_DEV_WAIT_CMD, 0, dev_wait},
1681
1682		{DM_TABLE_LOAD_CMD, 0, table_load},
1683		{DM_TABLE_CLEAR_CMD, IOCTL_FLAGS_NO_PARAMS, table_clear},
1684		{DM_TABLE_DEPS_CMD, 0, table_deps},
1685		{DM_TABLE_STATUS_CMD, 0, table_status},
1686
1687		{DM_LIST_VERSIONS_CMD, 0, list_versions},
1688
1689		{DM_TARGET_MSG_CMD, 0, target_message},
1690		{DM_DEV_SET_GEOMETRY_CMD, 0, dev_set_geometry},
1691		{DM_DEV_ARM_POLL, IOCTL_FLAGS_NO_PARAMS, dev_arm_poll},
1692		{DM_GET_TARGET_VERSION, 0, get_target_version},
1693	};
1694
1695	if (unlikely(cmd >= ARRAY_SIZE(_ioctls)))
1696		return NULL;
1697
1698	*ioctl_flags = _ioctls[cmd].flags;
1699	return _ioctls[cmd].fn;
1700}
1701
1702/*
1703 * As well as checking the version compatibility this always
1704 * copies the kernel interface version out.
1705 */
1706static int check_version(unsigned int cmd, struct dm_ioctl __user *user)
1707{
1708	uint32_t version[3];
1709	int r = 0;
1710
1711	if (copy_from_user(version, user->version, sizeof(version)))
1712		return -EFAULT;
1713
1714	if ((DM_VERSION_MAJOR != version[0]) ||
1715	    (DM_VERSION_MINOR < version[1])) {
1716		DMWARN("ioctl interface mismatch: "
1717		       "kernel(%u.%u.%u), user(%u.%u.%u), cmd(%d)",
1718		       DM_VERSION_MAJOR, DM_VERSION_MINOR,
1719		       DM_VERSION_PATCHLEVEL,
1720		       version[0], version[1], version[2], cmd);
1721		r = -EINVAL;
1722	}
1723
1724	/*
1725	 * Fill in the kernel version.
1726	 */
1727	version[0] = DM_VERSION_MAJOR;
1728	version[1] = DM_VERSION_MINOR;
1729	version[2] = DM_VERSION_PATCHLEVEL;
1730	if (copy_to_user(user->version, version, sizeof(version)))
1731		return -EFAULT;
1732
1733	return r;
1734}
1735
1736#define DM_PARAMS_MALLOC	0x0001	/* Params allocated with kvmalloc() */
1737#define DM_WIPE_BUFFER		0x0010	/* Wipe input buffer before returning from ioctl */
1738
1739static void free_params(struct dm_ioctl *param, size_t param_size, int param_flags)
1740{
1741	if (param_flags & DM_WIPE_BUFFER)
1742		memset(param, 0, param_size);
1743
1744	if (param_flags & DM_PARAMS_MALLOC)
1745		kvfree(param);
1746}
1747
1748static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl *param_kernel,
1749		       int ioctl_flags, struct dm_ioctl **param, int *param_flags)
 
1750{
1751	struct dm_ioctl *dmi;
1752	int secure_data;
1753	const size_t minimum_data_size = offsetof(struct dm_ioctl, data);
1754	unsigned noio_flag;
1755
1756	if (copy_from_user(param_kernel, user, minimum_data_size))
1757		return -EFAULT;
1758
1759	if (param_kernel->data_size < minimum_data_size)
1760		return -EINVAL;
1761
1762	secure_data = param_kernel->flags & DM_SECURE_DATA_FLAG;
1763
1764	*param_flags = secure_data ? DM_WIPE_BUFFER : 0;
1765
1766	if (ioctl_flags & IOCTL_FLAGS_NO_PARAMS) {
1767		dmi = param_kernel;
1768		dmi->data_size = minimum_data_size;
1769		goto data_copied;
1770	}
1771
1772	/*
1773	 * Use __GFP_HIGH to avoid low memory issues when a device is
1774	 * suspended and the ioctl is needed to resume it.
1775	 * Use kmalloc() rather than vmalloc() when we can.
1776	 */
1777	dmi = NULL;
1778	noio_flag = memalloc_noio_save();
1779	dmi = kvmalloc(param_kernel->data_size, GFP_KERNEL | __GFP_HIGH);
1780	memalloc_noio_restore(noio_flag);
 
 
 
 
 
 
1781
1782	if (!dmi) {
1783		if (secure_data && clear_user(user, param_kernel->data_size))
1784			return -EFAULT;
1785		return -ENOMEM;
1786	}
1787
1788	*param_flags |= DM_PARAMS_MALLOC;
1789
1790	/* Copy from param_kernel (which was already copied from user) */
1791	memcpy(dmi, param_kernel, minimum_data_size);
1792
1793	if (copy_from_user(&dmi->data, (char __user *)user + minimum_data_size,
1794			   param_kernel->data_size - minimum_data_size))
1795		goto bad;
 
1796data_copied:
 
 
 
 
 
 
 
 
1797	/* Wipe the user buffer so we do not return it to userspace */
1798	if (secure_data && clear_user(user, param_kernel->data_size))
1799		goto bad;
1800
1801	*param = dmi;
1802	return 0;
1803
1804bad:
1805	free_params(dmi, param_kernel->data_size, *param_flags);
1806
1807	return -EFAULT;
1808}
1809
1810static int validate_params(uint cmd, struct dm_ioctl *param)
1811{
1812	/* Always clear this flag */
1813	param->flags &= ~DM_BUFFER_FULL_FLAG;
1814	param->flags &= ~DM_UEVENT_GENERATED_FLAG;
1815	param->flags &= ~DM_SECURE_DATA_FLAG;
1816	param->flags &= ~DM_DATA_OUT_FLAG;
1817
1818	/* Ignores parameters */
1819	if (cmd == DM_REMOVE_ALL_CMD ||
1820	    cmd == DM_LIST_DEVICES_CMD ||
1821	    cmd == DM_LIST_VERSIONS_CMD)
1822		return 0;
1823
1824	if (cmd == DM_DEV_CREATE_CMD) {
1825		if (!*param->name) {
1826			DMWARN("name not supplied when creating device");
1827			return -EINVAL;
1828		}
1829	} else if (*param->uuid && *param->name) {
1830		DMWARN("only supply one of name or uuid, cmd(%u)", cmd);
1831		return -EINVAL;
1832	}
1833
1834	/* Ensure strings are terminated */
1835	param->name[DM_NAME_LEN - 1] = '\0';
1836	param->uuid[DM_UUID_LEN - 1] = '\0';
1837
1838	return 0;
1839}
1840
1841static int ctl_ioctl(struct file *file, uint command, struct dm_ioctl __user *user)
1842{
1843	int r = 0;
1844	int ioctl_flags;
1845	int param_flags;
1846	unsigned int cmd;
1847	struct dm_ioctl *param;
1848	ioctl_fn fn = NULL;
1849	size_t input_param_size;
1850	struct dm_ioctl param_kernel;
1851
1852	/* only root can play with this */
1853	if (!capable(CAP_SYS_ADMIN))
1854		return -EACCES;
1855
1856	if (_IOC_TYPE(command) != DM_IOCTL)
1857		return -ENOTTY;
1858
1859	cmd = _IOC_NR(command);
1860
1861	/*
1862	 * Check the interface version passed in.  This also
1863	 * writes out the kernel's interface version.
1864	 */
1865	r = check_version(cmd, user);
1866	if (r)
1867		return r;
1868
1869	/*
1870	 * Nothing more to do for the version command.
1871	 */
1872	if (cmd == DM_VERSION_CMD)
1873		return 0;
1874
1875	fn = lookup_ioctl(cmd, &ioctl_flags);
1876	if (!fn) {
1877		DMWARN("dm_ctl_ioctl: unknown command 0x%x", command);
1878		return -ENOTTY;
1879	}
1880
1881	/*
1882	 * Copy the parameters into kernel space.
1883	 */
1884	r = copy_params(user, &param_kernel, ioctl_flags, &param, &param_flags);
1885
1886	if (r)
1887		return r;
1888
1889	input_param_size = param->data_size;
1890	r = validate_params(cmd, param);
1891	if (r)
1892		goto out;
1893
1894	param->data_size = offsetof(struct dm_ioctl, data);
1895	r = fn(file, param, input_param_size);
1896
1897	if (unlikely(param->flags & DM_BUFFER_FULL_FLAG) &&
1898	    unlikely(ioctl_flags & IOCTL_FLAGS_NO_PARAMS))
1899		DMERR("ioctl %d tried to output some data but has IOCTL_FLAGS_NO_PARAMS set", cmd);
1900
1901	if (!r && ioctl_flags & IOCTL_FLAGS_ISSUE_GLOBAL_EVENT)
1902		dm_issue_global_event();
1903
1904	/*
1905	 * Copy the results back to userland.
1906	 */
1907	if (!r && copy_to_user(user, param, param->data_size))
1908		r = -EFAULT;
1909
1910out:
1911	free_params(param, input_param_size, param_flags);
1912	return r;
1913}
1914
1915static long dm_ctl_ioctl(struct file *file, uint command, ulong u)
1916{
1917	return (long)ctl_ioctl(file, command, (struct dm_ioctl __user *)u);
1918}
1919
1920#ifdef CONFIG_COMPAT
1921static long dm_compat_ctl_ioctl(struct file *file, uint command, ulong u)
1922{
1923	return (long)dm_ctl_ioctl(file, command, (ulong) compat_ptr(u));
1924}
1925#else
1926#define dm_compat_ctl_ioctl NULL
1927#endif
1928
1929static int dm_open(struct inode *inode, struct file *filp)
1930{
1931	int r;
1932	struct dm_file *priv;
1933
1934	r = nonseekable_open(inode, filp);
1935	if (unlikely(r))
1936		return r;
1937
1938	priv = filp->private_data = kmalloc(sizeof(struct dm_file), GFP_KERNEL);
1939	if (!priv)
1940		return -ENOMEM;
1941
1942	priv->global_event_nr = atomic_read(&dm_global_event_nr);
1943
1944	return 0;
1945}
1946
1947static int dm_release(struct inode *inode, struct file *filp)
1948{
1949	kfree(filp->private_data);
1950	return 0;
1951}
1952
1953static __poll_t dm_poll(struct file *filp, poll_table *wait)
1954{
1955	struct dm_file *priv = filp->private_data;
1956	__poll_t mask = 0;
1957
1958	poll_wait(filp, &dm_global_eventq, wait);
1959
1960	if ((int)(atomic_read(&dm_global_event_nr) - priv->global_event_nr) > 0)
1961		mask |= EPOLLIN;
1962
1963	return mask;
1964}
1965
1966static const struct file_operations _ctl_fops = {
1967	.open    = dm_open,
1968	.release = dm_release,
1969	.poll    = dm_poll,
1970	.unlocked_ioctl	 = dm_ctl_ioctl,
1971	.compat_ioctl = dm_compat_ctl_ioctl,
1972	.owner	 = THIS_MODULE,
1973	.llseek  = noop_llseek,
1974};
1975
1976static struct miscdevice _dm_misc = {
1977	.minor		= MAPPER_CTRL_MINOR,
1978	.name  		= DM_NAME,
1979	.nodename	= DM_DIR "/" DM_CONTROL_NODE,
1980	.fops  		= &_ctl_fops
1981};
1982
1983MODULE_ALIAS_MISCDEV(MAPPER_CTRL_MINOR);
1984MODULE_ALIAS("devname:" DM_DIR "/" DM_CONTROL_NODE);
1985
1986/*
1987 * Create misc character device and link to DM_DIR/control.
1988 */
1989int __init dm_interface_init(void)
1990{
1991	int r;
1992
1993	r = dm_hash_init();
1994	if (r)
1995		return r;
1996
1997	r = misc_register(&_dm_misc);
1998	if (r) {
1999		DMERR("misc_register failed for control device");
2000		dm_hash_exit();
2001		return r;
2002	}
2003
2004	DMINFO("%d.%d.%d%s initialised: %s", DM_VERSION_MAJOR,
2005	       DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL, DM_VERSION_EXTRA,
2006	       DM_DRIVER_EMAIL);
2007	return 0;
2008}
2009
2010void dm_interface_exit(void)
2011{
2012	misc_deregister(&_dm_misc);
2013	dm_hash_exit();
2014}
2015
2016/**
2017 * dm_copy_name_and_uuid - Copy mapped device name & uuid into supplied buffers
2018 * @md: Pointer to mapped_device
2019 * @name: Buffer (size DM_NAME_LEN) for name
2020 * @uuid: Buffer (size DM_UUID_LEN) for uuid or empty string if uuid not defined
2021 */
2022int dm_copy_name_and_uuid(struct mapped_device *md, char *name, char *uuid)
2023{
2024	int r = 0;
2025	struct hash_cell *hc;
2026
2027	if (!md)
2028		return -ENXIO;
2029
2030	mutex_lock(&dm_hash_cells_mutex);
2031	hc = dm_get_mdptr(md);
2032	if (!hc || hc->md != md) {
2033		r = -ENXIO;
2034		goto out;
2035	}
2036
2037	if (name)
2038		strcpy(name, hc->name);
2039	if (uuid)
2040		strcpy(uuid, hc->uuid ? : "");
2041
2042out:
2043	mutex_unlock(&dm_hash_cells_mutex);
2044
2045	return r;
2046}
2047
2048
2049/**
2050 * dm_early_create - create a mapped device in early boot.
2051 *
2052 * @dmi: Contains main information of the device mapping to be created.
2053 * @spec_array: array of pointers to struct dm_target_spec. Describes the
2054 * mapping table of the device.
2055 * @target_params_array: array of strings with the parameters to a specific
2056 * target.
2057 *
2058 * Instead of having the struct dm_target_spec and the parameters for every
2059 * target embedded at the end of struct dm_ioctl (as performed in a normal
2060 * ioctl), pass them as arguments, so the caller doesn't need to serialize them.
2061 * The size of the spec_array and target_params_array is given by
2062 * @dmi->target_count.
2063 * This function is supposed to be called in early boot, so locking mechanisms
2064 * to protect against concurrent loads are not required.
2065 */
2066int __init dm_early_create(struct dm_ioctl *dmi,
2067			   struct dm_target_spec **spec_array,
2068			   char **target_params_array)
2069{
2070	int r, m = DM_ANY_MINOR;
2071	struct dm_table *t, *old_map;
2072	struct mapped_device *md;
2073	unsigned int i;
2074
2075	if (!dmi->target_count)
2076		return -EINVAL;
2077
2078	r = check_name(dmi->name);
2079	if (r)
2080		return r;
2081
2082	if (dmi->flags & DM_PERSISTENT_DEV_FLAG)
2083		m = MINOR(huge_decode_dev(dmi->dev));
2084
2085	/* alloc dm device */
2086	r = dm_create(m, &md);
2087	if (r)
2088		return r;
2089
2090	/* hash insert */
2091	r = dm_hash_insert(dmi->name, *dmi->uuid ? dmi->uuid : NULL, md);
2092	if (r)
2093		goto err_destroy_dm;
2094
2095	/* alloc table */
2096	r = dm_table_create(&t, get_mode(dmi), dmi->target_count, md);
2097	if (r)
2098		goto err_hash_remove;
2099
2100	/* add targets */
2101	for (i = 0; i < dmi->target_count; i++) {
2102		r = dm_table_add_target(t, spec_array[i]->target_type,
2103					(sector_t) spec_array[i]->sector_start,
2104					(sector_t) spec_array[i]->length,
2105					target_params_array[i]);
2106		if (r) {
2107			DMWARN("error adding target to table");
2108			goto err_destroy_table;
2109		}
2110	}
2111
2112	/* finish table */
2113	r = dm_table_complete(t);
2114	if (r)
2115		goto err_destroy_table;
2116
2117	md->type = dm_table_get_type(t);
2118	/* setup md->queue to reflect md's type (may block) */
2119	r = dm_setup_md_queue(md, t);
2120	if (r) {
2121		DMWARN("unable to set up device queue for new table.");
2122		goto err_destroy_table;
2123	}
2124
2125	/* Set new map */
2126	dm_suspend(md, 0);
2127	old_map = dm_swap_table(md, t);
2128	if (IS_ERR(old_map)) {
2129		r = PTR_ERR(old_map);
2130		goto err_destroy_table;
2131	}
2132	set_disk_ro(dm_disk(md), !!(dmi->flags & DM_READONLY_FLAG));
2133
2134	/* resume device */
2135	r = dm_resume(md);
2136	if (r)
2137		goto err_destroy_table;
2138
2139	DMINFO("%s (%s) is ready", md->disk->disk_name, dmi->name);
2140	dm_put(md);
2141	return 0;
2142
2143err_destroy_table:
2144	dm_table_destroy(t);
2145err_hash_remove:
2146	(void) __hash_remove(__get_name_cell(dmi->name));
2147	/* release reference from __get_name_cell */
2148	dm_put(md);
2149err_destroy_dm:
2150	dm_put(md);
2151	dm_destroy(md);
2152	return r;
2153}