xref: /openbmc/linux/drivers/md/dm-ioctl.c (revision 39b6f3aa)
1 /*
2  * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3  * Copyright (C) 2004 - 2006 Red Hat, Inc. All rights reserved.
4  *
5  * This file is released under the GPL.
6  */
7 
8 #include "dm.h"
9 
10 #include <linux/module.h>
11 #include <linux/vmalloc.h>
12 #include <linux/miscdevice.h>
13 #include <linux/init.h>
14 #include <linux/wait.h>
15 #include <linux/slab.h>
16 #include <linux/dm-ioctl.h>
17 #include <linux/hdreg.h>
18 #include <linux/compat.h>
19 
20 #include <asm/uaccess.h>
21 
22 #define DM_MSG_PREFIX "ioctl"
23 #define DM_DRIVER_EMAIL "dm-devel@redhat.com"
24 
25 /*-----------------------------------------------------------------
26  * The ioctl interface needs to be able to look up devices by
27  * name or uuid.
28  *---------------------------------------------------------------*/
29 struct hash_cell {
30 	struct list_head name_list;
31 	struct list_head uuid_list;
32 
33 	char *name;
34 	char *uuid;
35 	struct mapped_device *md;
36 	struct dm_table *new_map;
37 };
38 
39 struct vers_iter {
40     size_t param_size;
41     struct dm_target_versions *vers, *old_vers;
42     char *end;
43     uint32_t flags;
44 };
45 
46 
47 #define NUM_BUCKETS 64
48 #define MASK_BUCKETS (NUM_BUCKETS - 1)
49 static struct list_head _name_buckets[NUM_BUCKETS];
50 static struct list_head _uuid_buckets[NUM_BUCKETS];
51 
52 static void dm_hash_remove_all(int keep_open_devices);
53 
54 /*
55  * Guards access to both hash tables.
56  */
57 static DECLARE_RWSEM(_hash_lock);
58 
59 /*
60  * Protects use of mdptr to obtain hash cell name and uuid from mapped device.
61  */
62 static DEFINE_MUTEX(dm_hash_cells_mutex);
63 
64 static void init_buckets(struct list_head *buckets)
65 {
66 	unsigned int i;
67 
68 	for (i = 0; i < NUM_BUCKETS; i++)
69 		INIT_LIST_HEAD(buckets + i);
70 }
71 
72 static int dm_hash_init(void)
73 {
74 	init_buckets(_name_buckets);
75 	init_buckets(_uuid_buckets);
76 	return 0;
77 }
78 
79 static void dm_hash_exit(void)
80 {
81 	dm_hash_remove_all(0);
82 }
83 
84 /*-----------------------------------------------------------------
85  * Hash function:
86  * We're not really concerned with the str hash function being
87  * fast since it's only used by the ioctl interface.
88  *---------------------------------------------------------------*/
89 static unsigned int hash_str(const char *str)
90 {
91 	const unsigned int hash_mult = 2654435387U;
92 	unsigned int h = 0;
93 
94 	while (*str)
95 		h = (h + (unsigned int) *str++) * hash_mult;
96 
97 	return h & MASK_BUCKETS;
98 }
99 
100 /*-----------------------------------------------------------------
101  * Code for looking up a device by name
102  *---------------------------------------------------------------*/
103 static struct hash_cell *__get_name_cell(const char *str)
104 {
105 	struct hash_cell *hc;
106 	unsigned int h = hash_str(str);
107 
108 	list_for_each_entry (hc, _name_buckets + h, name_list)
109 		if (!strcmp(hc->name, str)) {
110 			dm_get(hc->md);
111 			return hc;
112 		}
113 
114 	return NULL;
115 }
116 
117 static struct hash_cell *__get_uuid_cell(const char *str)
118 {
119 	struct hash_cell *hc;
120 	unsigned int h = hash_str(str);
121 
122 	list_for_each_entry (hc, _uuid_buckets + h, uuid_list)
123 		if (!strcmp(hc->uuid, str)) {
124 			dm_get(hc->md);
125 			return hc;
126 		}
127 
128 	return NULL;
129 }
130 
131 static struct hash_cell *__get_dev_cell(uint64_t dev)
132 {
133 	struct mapped_device *md;
134 	struct hash_cell *hc;
135 
136 	md = dm_get_md(huge_decode_dev(dev));
137 	if (!md)
138 		return NULL;
139 
140 	hc = dm_get_mdptr(md);
141 	if (!hc) {
142 		dm_put(md);
143 		return NULL;
144 	}
145 
146 	return hc;
147 }
148 
149 /*-----------------------------------------------------------------
150  * Inserting, removing and renaming a device.
151  *---------------------------------------------------------------*/
152 static struct hash_cell *alloc_cell(const char *name, const char *uuid,
153 				    struct mapped_device *md)
154 {
155 	struct hash_cell *hc;
156 
157 	hc = kmalloc(sizeof(*hc), GFP_KERNEL);
158 	if (!hc)
159 		return NULL;
160 
161 	hc->name = kstrdup(name, GFP_KERNEL);
162 	if (!hc->name) {
163 		kfree(hc);
164 		return NULL;
165 	}
166 
167 	if (!uuid)
168 		hc->uuid = NULL;
169 
170 	else {
171 		hc->uuid = kstrdup(uuid, GFP_KERNEL);
172 		if (!hc->uuid) {
173 			kfree(hc->name);
174 			kfree(hc);
175 			return NULL;
176 		}
177 	}
178 
179 	INIT_LIST_HEAD(&hc->name_list);
180 	INIT_LIST_HEAD(&hc->uuid_list);
181 	hc->md = md;
182 	hc->new_map = NULL;
183 	return hc;
184 }
185 
186 static void free_cell(struct hash_cell *hc)
187 {
188 	if (hc) {
189 		kfree(hc->name);
190 		kfree(hc->uuid);
191 		kfree(hc);
192 	}
193 }
194 
195 /*
196  * The kdev_t and uuid of a device can never change once it is
197  * initially inserted.
198  */
199 static int dm_hash_insert(const char *name, const char *uuid, struct mapped_device *md)
200 {
201 	struct hash_cell *cell, *hc;
202 
203 	/*
204 	 * Allocate the new cells.
205 	 */
206 	cell = alloc_cell(name, uuid, md);
207 	if (!cell)
208 		return -ENOMEM;
209 
210 	/*
211 	 * Insert the cell into both hash tables.
212 	 */
213 	down_write(&_hash_lock);
214 	hc = __get_name_cell(name);
215 	if (hc) {
216 		dm_put(hc->md);
217 		goto bad;
218 	}
219 
220 	list_add(&cell->name_list, _name_buckets + hash_str(name));
221 
222 	if (uuid) {
223 		hc = __get_uuid_cell(uuid);
224 		if (hc) {
225 			list_del(&cell->name_list);
226 			dm_put(hc->md);
227 			goto bad;
228 		}
229 		list_add(&cell->uuid_list, _uuid_buckets + hash_str(uuid));
230 	}
231 	dm_get(md);
232 	mutex_lock(&dm_hash_cells_mutex);
233 	dm_set_mdptr(md, cell);
234 	mutex_unlock(&dm_hash_cells_mutex);
235 	up_write(&_hash_lock);
236 
237 	return 0;
238 
239  bad:
240 	up_write(&_hash_lock);
241 	free_cell(cell);
242 	return -EBUSY;
243 }
244 
245 static void __hash_remove(struct hash_cell *hc)
246 {
247 	struct dm_table *table;
248 
249 	/* remove from the dev hash */
250 	list_del(&hc->uuid_list);
251 	list_del(&hc->name_list);
252 	mutex_lock(&dm_hash_cells_mutex);
253 	dm_set_mdptr(hc->md, NULL);
254 	mutex_unlock(&dm_hash_cells_mutex);
255 
256 	table = dm_get_live_table(hc->md);
257 	if (table) {
258 		dm_table_event(table);
259 		dm_table_put(table);
260 	}
261 
262 	if (hc->new_map)
263 		dm_table_destroy(hc->new_map);
264 	dm_put(hc->md);
265 	free_cell(hc);
266 }
267 
268 static void dm_hash_remove_all(int keep_open_devices)
269 {
270 	int i, dev_skipped;
271 	struct hash_cell *hc;
272 	struct mapped_device *md;
273 
274 retry:
275 	dev_skipped = 0;
276 
277 	down_write(&_hash_lock);
278 
279 	for (i = 0; i < NUM_BUCKETS; i++) {
280 		list_for_each_entry(hc, _name_buckets + i, name_list) {
281 			md = hc->md;
282 			dm_get(md);
283 
284 			if (keep_open_devices && dm_lock_for_deletion(md)) {
285 				dm_put(md);
286 				dev_skipped++;
287 				continue;
288 			}
289 
290 			__hash_remove(hc);
291 
292 			up_write(&_hash_lock);
293 
294 			dm_put(md);
295 			if (likely(keep_open_devices))
296 				dm_destroy(md);
297 			else
298 				dm_destroy_immediate(md);
299 
300 			/*
301 			 * Some mapped devices may be using other mapped
302 			 * devices, so repeat until we make no further
303 			 * progress.  If a new mapped device is created
304 			 * here it will also get removed.
305 			 */
306 			goto retry;
307 		}
308 	}
309 
310 	up_write(&_hash_lock);
311 
312 	if (dev_skipped)
313 		DMWARN("remove_all left %d open device(s)", dev_skipped);
314 }
315 
316 /*
317  * Set the uuid of a hash_cell that isn't already set.
318  */
319 static void __set_cell_uuid(struct hash_cell *hc, char *new_uuid)
320 {
321 	mutex_lock(&dm_hash_cells_mutex);
322 	hc->uuid = new_uuid;
323 	mutex_unlock(&dm_hash_cells_mutex);
324 
325 	list_add(&hc->uuid_list, _uuid_buckets + hash_str(new_uuid));
326 }
327 
328 /*
329  * Changes the name of a hash_cell and returns the old name for
330  * the caller to free.
331  */
332 static char *__change_cell_name(struct hash_cell *hc, char *new_name)
333 {
334 	char *old_name;
335 
336 	/*
337 	 * Rename and move the name cell.
338 	 */
339 	list_del(&hc->name_list);
340 	old_name = hc->name;
341 
342 	mutex_lock(&dm_hash_cells_mutex);
343 	hc->name = new_name;
344 	mutex_unlock(&dm_hash_cells_mutex);
345 
346 	list_add(&hc->name_list, _name_buckets + hash_str(new_name));
347 
348 	return old_name;
349 }
350 
351 static struct mapped_device *dm_hash_rename(struct dm_ioctl *param,
352 					    const char *new)
353 {
354 	char *new_data, *old_name = NULL;
355 	struct hash_cell *hc;
356 	struct dm_table *table;
357 	struct mapped_device *md;
358 	unsigned change_uuid = (param->flags & DM_UUID_FLAG) ? 1 : 0;
359 
360 	/*
361 	 * duplicate new.
362 	 */
363 	new_data = kstrdup(new, GFP_KERNEL);
364 	if (!new_data)
365 		return ERR_PTR(-ENOMEM);
366 
367 	down_write(&_hash_lock);
368 
369 	/*
370 	 * Is new free ?
371 	 */
372 	if (change_uuid)
373 		hc = __get_uuid_cell(new);
374 	else
375 		hc = __get_name_cell(new);
376 
377 	if (hc) {
378 		DMWARN("Unable to change %s on mapped device %s to one that "
379 		       "already exists: %s",
380 		       change_uuid ? "uuid" : "name",
381 		       param->name, new);
382 		dm_put(hc->md);
383 		up_write(&_hash_lock);
384 		kfree(new_data);
385 		return ERR_PTR(-EBUSY);
386 	}
387 
388 	/*
389 	 * Is there such a device as 'old' ?
390 	 */
391 	hc = __get_name_cell(param->name);
392 	if (!hc) {
393 		DMWARN("Unable to rename non-existent device, %s to %s%s",
394 		       param->name, change_uuid ? "uuid " : "", new);
395 		up_write(&_hash_lock);
396 		kfree(new_data);
397 		return ERR_PTR(-ENXIO);
398 	}
399 
400 	/*
401 	 * Does this device already have a uuid?
402 	 */
403 	if (change_uuid && hc->uuid) {
404 		DMWARN("Unable to change uuid of mapped device %s to %s "
405 		       "because uuid is already set to %s",
406 		       param->name, new, hc->uuid);
407 		dm_put(hc->md);
408 		up_write(&_hash_lock);
409 		kfree(new_data);
410 		return ERR_PTR(-EINVAL);
411 	}
412 
413 	if (change_uuid)
414 		__set_cell_uuid(hc, new_data);
415 	else
416 		old_name = __change_cell_name(hc, new_data);
417 
418 	/*
419 	 * Wake up any dm event waiters.
420 	 */
421 	table = dm_get_live_table(hc->md);
422 	if (table) {
423 		dm_table_event(table);
424 		dm_table_put(table);
425 	}
426 
427 	if (!dm_kobject_uevent(hc->md, KOBJ_CHANGE, param->event_nr))
428 		param->flags |= DM_UEVENT_GENERATED_FLAG;
429 
430 	md = hc->md;
431 	up_write(&_hash_lock);
432 	kfree(old_name);
433 
434 	return md;
435 }
436 
437 /*-----------------------------------------------------------------
438  * Implementation of the ioctl commands
439  *---------------------------------------------------------------*/
440 /*
441  * All the ioctl commands get dispatched to functions with this
442  * prototype.
443  */
444 typedef int (*ioctl_fn)(struct dm_ioctl *param, size_t param_size);
445 
446 static int remove_all(struct dm_ioctl *param, size_t param_size)
447 {
448 	dm_hash_remove_all(1);
449 	param->data_size = 0;
450 	return 0;
451 }
452 
453 /*
454  * Round up the ptr to an 8-byte boundary.
455  */
456 #define ALIGN_MASK 7
457 static inline void *align_ptr(void *ptr)
458 {
459 	return (void *) (((size_t) (ptr + ALIGN_MASK)) & ~ALIGN_MASK);
460 }
461 
462 /*
463  * Retrieves the data payload buffer from an already allocated
464  * struct dm_ioctl.
465  */
466 static void *get_result_buffer(struct dm_ioctl *param, size_t param_size,
467 			       size_t *len)
468 {
469 	param->data_start = align_ptr(param + 1) - (void *) param;
470 
471 	if (param->data_start < param_size)
472 		*len = param_size - param->data_start;
473 	else
474 		*len = 0;
475 
476 	return ((void *) param) + param->data_start;
477 }
478 
479 static int list_devices(struct dm_ioctl *param, size_t param_size)
480 {
481 	unsigned int i;
482 	struct hash_cell *hc;
483 	size_t len, needed = 0;
484 	struct gendisk *disk;
485 	struct dm_name_list *nl, *old_nl = NULL;
486 
487 	down_write(&_hash_lock);
488 
489 	/*
490 	 * Loop through all the devices working out how much
491 	 * space we need.
492 	 */
493 	for (i = 0; i < NUM_BUCKETS; i++) {
494 		list_for_each_entry (hc, _name_buckets + i, name_list) {
495 			needed += sizeof(struct dm_name_list);
496 			needed += strlen(hc->name) + 1;
497 			needed += ALIGN_MASK;
498 		}
499 	}
500 
501 	/*
502 	 * Grab our output buffer.
503 	 */
504 	nl = get_result_buffer(param, param_size, &len);
505 	if (len < needed) {
506 		param->flags |= DM_BUFFER_FULL_FLAG;
507 		goto out;
508 	}
509 	param->data_size = param->data_start + needed;
510 
511 	nl->dev = 0;	/* Flags no data */
512 
513 	/*
514 	 * Now loop through filling out the names.
515 	 */
516 	for (i = 0; i < NUM_BUCKETS; i++) {
517 		list_for_each_entry (hc, _name_buckets + i, name_list) {
518 			if (old_nl)
519 				old_nl->next = (uint32_t) ((void *) nl -
520 							   (void *) old_nl);
521 			disk = dm_disk(hc->md);
522 			nl->dev = huge_encode_dev(disk_devt(disk));
523 			nl->next = 0;
524 			strcpy(nl->name, hc->name);
525 
526 			old_nl = nl;
527 			nl = align_ptr(((void *) ++nl) + strlen(hc->name) + 1);
528 		}
529 	}
530 
531  out:
532 	up_write(&_hash_lock);
533 	return 0;
534 }
535 
536 static void list_version_get_needed(struct target_type *tt, void *needed_param)
537 {
538     size_t *needed = needed_param;
539 
540     *needed += sizeof(struct dm_target_versions);
541     *needed += strlen(tt->name);
542     *needed += ALIGN_MASK;
543 }
544 
545 static void list_version_get_info(struct target_type *tt, void *param)
546 {
547     struct vers_iter *info = param;
548 
549     /* Check space - it might have changed since the first iteration */
550     if ((char *)info->vers + sizeof(tt->version) + strlen(tt->name) + 1 >
551 	info->end) {
552 
553 	info->flags = DM_BUFFER_FULL_FLAG;
554 	return;
555     }
556 
557     if (info->old_vers)
558 	info->old_vers->next = (uint32_t) ((void *)info->vers -
559 					   (void *)info->old_vers);
560     info->vers->version[0] = tt->version[0];
561     info->vers->version[1] = tt->version[1];
562     info->vers->version[2] = tt->version[2];
563     info->vers->next = 0;
564     strcpy(info->vers->name, tt->name);
565 
566     info->old_vers = info->vers;
567     info->vers = align_ptr(((void *) ++info->vers) + strlen(tt->name) + 1);
568 }
569 
570 static int list_versions(struct dm_ioctl *param, size_t param_size)
571 {
572 	size_t len, needed = 0;
573 	struct dm_target_versions *vers;
574 	struct vers_iter iter_info;
575 
576 	/*
577 	 * Loop through all the devices working out how much
578 	 * space we need.
579 	 */
580 	dm_target_iterate(list_version_get_needed, &needed);
581 
582 	/*
583 	 * Grab our output buffer.
584 	 */
585 	vers = get_result_buffer(param, param_size, &len);
586 	if (len < needed) {
587 		param->flags |= DM_BUFFER_FULL_FLAG;
588 		goto out;
589 	}
590 	param->data_size = param->data_start + needed;
591 
592 	iter_info.param_size = param_size;
593 	iter_info.old_vers = NULL;
594 	iter_info.vers = vers;
595 	iter_info.flags = 0;
596 	iter_info.end = (char *)vers+len;
597 
598 	/*
599 	 * Now loop through filling out the names & versions.
600 	 */
601 	dm_target_iterate(list_version_get_info, &iter_info);
602 	param->flags |= iter_info.flags;
603 
604  out:
605 	return 0;
606 }
607 
608 static int check_name(const char *name)
609 {
610 	if (strchr(name, '/')) {
611 		DMWARN("invalid device name");
612 		return -EINVAL;
613 	}
614 
615 	return 0;
616 }
617 
618 /*
619  * On successful return, the caller must not attempt to acquire
620  * _hash_lock without first calling dm_table_put, because dm_table_destroy
621  * waits for this dm_table_put and could be called under this lock.
622  */
623 static struct dm_table *dm_get_inactive_table(struct mapped_device *md)
624 {
625 	struct hash_cell *hc;
626 	struct dm_table *table = NULL;
627 
628 	down_read(&_hash_lock);
629 	hc = dm_get_mdptr(md);
630 	if (!hc || hc->md != md) {
631 		DMWARN("device has been removed from the dev hash table.");
632 		goto out;
633 	}
634 
635 	table = hc->new_map;
636 	if (table)
637 		dm_table_get(table);
638 
639 out:
640 	up_read(&_hash_lock);
641 
642 	return table;
643 }
644 
645 static struct dm_table *dm_get_live_or_inactive_table(struct mapped_device *md,
646 						      struct dm_ioctl *param)
647 {
648 	return (param->flags & DM_QUERY_INACTIVE_TABLE_FLAG) ?
649 		dm_get_inactive_table(md) : dm_get_live_table(md);
650 }
651 
652 /*
653  * Fills in a dm_ioctl structure, ready for sending back to
654  * userland.
655  */
656 static void __dev_status(struct mapped_device *md, struct dm_ioctl *param)
657 {
658 	struct gendisk *disk = dm_disk(md);
659 	struct dm_table *table;
660 
661 	param->flags &= ~(DM_SUSPEND_FLAG | DM_READONLY_FLAG |
662 			  DM_ACTIVE_PRESENT_FLAG);
663 
664 	if (dm_suspended_md(md))
665 		param->flags |= DM_SUSPEND_FLAG;
666 
667 	param->dev = huge_encode_dev(disk_devt(disk));
668 
669 	/*
670 	 * Yes, this will be out of date by the time it gets back
671 	 * to userland, but it is still very useful for
672 	 * debugging.
673 	 */
674 	param->open_count = dm_open_count(md);
675 
676 	param->event_nr = dm_get_event_nr(md);
677 	param->target_count = 0;
678 
679 	table = dm_get_live_table(md);
680 	if (table) {
681 		if (!(param->flags & DM_QUERY_INACTIVE_TABLE_FLAG)) {
682 			if (get_disk_ro(disk))
683 				param->flags |= DM_READONLY_FLAG;
684 			param->target_count = dm_table_get_num_targets(table);
685 		}
686 		dm_table_put(table);
687 
688 		param->flags |= DM_ACTIVE_PRESENT_FLAG;
689 	}
690 
691 	if (param->flags & DM_QUERY_INACTIVE_TABLE_FLAG) {
692 		table = dm_get_inactive_table(md);
693 		if (table) {
694 			if (!(dm_table_get_mode(table) & FMODE_WRITE))
695 				param->flags |= DM_READONLY_FLAG;
696 			param->target_count = dm_table_get_num_targets(table);
697 			dm_table_put(table);
698 		}
699 	}
700 }
701 
702 static int dev_create(struct dm_ioctl *param, size_t param_size)
703 {
704 	int r, m = DM_ANY_MINOR;
705 	struct mapped_device *md;
706 
707 	r = check_name(param->name);
708 	if (r)
709 		return r;
710 
711 	if (param->flags & DM_PERSISTENT_DEV_FLAG)
712 		m = MINOR(huge_decode_dev(param->dev));
713 
714 	r = dm_create(m, &md);
715 	if (r)
716 		return r;
717 
718 	r = dm_hash_insert(param->name, *param->uuid ? param->uuid : NULL, md);
719 	if (r) {
720 		dm_put(md);
721 		dm_destroy(md);
722 		return r;
723 	}
724 
725 	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
726 
727 	__dev_status(md, param);
728 
729 	dm_put(md);
730 
731 	return 0;
732 }
733 
734 /*
735  * Always use UUID for lookups if it's present, otherwise use name or dev.
736  */
737 static struct hash_cell *__find_device_hash_cell(struct dm_ioctl *param)
738 {
739 	struct hash_cell *hc = NULL;
740 
741 	if (*param->uuid) {
742 		if (*param->name || param->dev)
743 			return NULL;
744 
745 		hc = __get_uuid_cell(param->uuid);
746 		if (!hc)
747 			return NULL;
748 	} else if (*param->name) {
749 		if (param->dev)
750 			return NULL;
751 
752 		hc = __get_name_cell(param->name);
753 		if (!hc)
754 			return NULL;
755 	} else if (param->dev) {
756 		hc = __get_dev_cell(param->dev);
757 		if (!hc)
758 			return NULL;
759 	} else
760 		return NULL;
761 
762 	/*
763 	 * Sneakily write in both the name and the uuid
764 	 * while we have the cell.
765 	 */
766 	strlcpy(param->name, hc->name, sizeof(param->name));
767 	if (hc->uuid)
768 		strlcpy(param->uuid, hc->uuid, sizeof(param->uuid));
769 	else
770 		param->uuid[0] = '\0';
771 
772 	if (hc->new_map)
773 		param->flags |= DM_INACTIVE_PRESENT_FLAG;
774 	else
775 		param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
776 
777 	return hc;
778 }
779 
780 static struct mapped_device *find_device(struct dm_ioctl *param)
781 {
782 	struct hash_cell *hc;
783 	struct mapped_device *md = NULL;
784 
785 	down_read(&_hash_lock);
786 	hc = __find_device_hash_cell(param);
787 	if (hc)
788 		md = hc->md;
789 	up_read(&_hash_lock);
790 
791 	return md;
792 }
793 
794 static int dev_remove(struct dm_ioctl *param, size_t param_size)
795 {
796 	struct hash_cell *hc;
797 	struct mapped_device *md;
798 	int r;
799 
800 	down_write(&_hash_lock);
801 	hc = __find_device_hash_cell(param);
802 
803 	if (!hc) {
804 		DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
805 		up_write(&_hash_lock);
806 		return -ENXIO;
807 	}
808 
809 	md = hc->md;
810 
811 	/*
812 	 * Ensure the device is not open and nothing further can open it.
813 	 */
814 	r = dm_lock_for_deletion(md);
815 	if (r) {
816 		DMDEBUG_LIMIT("unable to remove open device %s", hc->name);
817 		up_write(&_hash_lock);
818 		dm_put(md);
819 		return r;
820 	}
821 
822 	__hash_remove(hc);
823 	up_write(&_hash_lock);
824 
825 	if (!dm_kobject_uevent(md, KOBJ_REMOVE, param->event_nr))
826 		param->flags |= DM_UEVENT_GENERATED_FLAG;
827 
828 	dm_put(md);
829 	dm_destroy(md);
830 	return 0;
831 }
832 
833 /*
834  * Check a string doesn't overrun the chunk of
835  * memory we copied from userland.
836  */
837 static int invalid_str(char *str, void *end)
838 {
839 	while ((void *) str < end)
840 		if (!*str++)
841 			return 0;
842 
843 	return -EINVAL;
844 }
845 
846 static int dev_rename(struct dm_ioctl *param, size_t param_size)
847 {
848 	int r;
849 	char *new_data = (char *) param + param->data_start;
850 	struct mapped_device *md;
851 	unsigned change_uuid = (param->flags & DM_UUID_FLAG) ? 1 : 0;
852 
853 	if (new_data < param->data ||
854 	    invalid_str(new_data, (void *) param + param_size) ||
855 	    strlen(new_data) > (change_uuid ? DM_UUID_LEN - 1 : DM_NAME_LEN - 1)) {
856 		DMWARN("Invalid new mapped device name or uuid string supplied.");
857 		return -EINVAL;
858 	}
859 
860 	if (!change_uuid) {
861 		r = check_name(new_data);
862 		if (r)
863 			return r;
864 	}
865 
866 	md = dm_hash_rename(param, new_data);
867 	if (IS_ERR(md))
868 		return PTR_ERR(md);
869 
870 	__dev_status(md, param);
871 	dm_put(md);
872 
873 	return 0;
874 }
875 
876 static int dev_set_geometry(struct dm_ioctl *param, size_t param_size)
877 {
878 	int r = -EINVAL, x;
879 	struct mapped_device *md;
880 	struct hd_geometry geometry;
881 	unsigned long indata[4];
882 	char *geostr = (char *) param + param->data_start;
883 	char dummy;
884 
885 	md = find_device(param);
886 	if (!md)
887 		return -ENXIO;
888 
889 	if (geostr < param->data ||
890 	    invalid_str(geostr, (void *) param + param_size)) {
891 		DMWARN("Invalid geometry supplied.");
892 		goto out;
893 	}
894 
895 	x = sscanf(geostr, "%lu %lu %lu %lu%c", indata,
896 		   indata + 1, indata + 2, indata + 3, &dummy);
897 
898 	if (x != 4) {
899 		DMWARN("Unable to interpret geometry settings.");
900 		goto out;
901 	}
902 
903 	if (indata[0] > 65535 || indata[1] > 255 ||
904 	    indata[2] > 255 || indata[3] > ULONG_MAX) {
905 		DMWARN("Geometry exceeds range limits.");
906 		goto out;
907 	}
908 
909 	geometry.cylinders = indata[0];
910 	geometry.heads = indata[1];
911 	geometry.sectors = indata[2];
912 	geometry.start = indata[3];
913 
914 	r = dm_set_geometry(md, &geometry);
915 
916 	param->data_size = 0;
917 
918 out:
919 	dm_put(md);
920 	return r;
921 }
922 
923 static int do_suspend(struct dm_ioctl *param)
924 {
925 	int r = 0;
926 	unsigned suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
927 	struct mapped_device *md;
928 
929 	md = find_device(param);
930 	if (!md)
931 		return -ENXIO;
932 
933 	if (param->flags & DM_SKIP_LOCKFS_FLAG)
934 		suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
935 	if (param->flags & DM_NOFLUSH_FLAG)
936 		suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
937 
938 	if (!dm_suspended_md(md)) {
939 		r = dm_suspend(md, suspend_flags);
940 		if (r)
941 			goto out;
942 	}
943 
944 	__dev_status(md, param);
945 
946 out:
947 	dm_put(md);
948 
949 	return r;
950 }
951 
952 static int do_resume(struct dm_ioctl *param)
953 {
954 	int r = 0;
955 	unsigned suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
956 	struct hash_cell *hc;
957 	struct mapped_device *md;
958 	struct dm_table *new_map, *old_map = NULL;
959 
960 	down_write(&_hash_lock);
961 
962 	hc = __find_device_hash_cell(param);
963 	if (!hc) {
964 		DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
965 		up_write(&_hash_lock);
966 		return -ENXIO;
967 	}
968 
969 	md = hc->md;
970 
971 	new_map = hc->new_map;
972 	hc->new_map = NULL;
973 	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
974 
975 	up_write(&_hash_lock);
976 
977 	/* Do we need to load a new map ? */
978 	if (new_map) {
979 		/* Suspend if it isn't already suspended */
980 		if (param->flags & DM_SKIP_LOCKFS_FLAG)
981 			suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
982 		if (param->flags & DM_NOFLUSH_FLAG)
983 			suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
984 		if (!dm_suspended_md(md))
985 			dm_suspend(md, suspend_flags);
986 
987 		old_map = dm_swap_table(md, new_map);
988 		if (IS_ERR(old_map)) {
989 			dm_table_destroy(new_map);
990 			dm_put(md);
991 			return PTR_ERR(old_map);
992 		}
993 
994 		if (dm_table_get_mode(new_map) & FMODE_WRITE)
995 			set_disk_ro(dm_disk(md), 0);
996 		else
997 			set_disk_ro(dm_disk(md), 1);
998 	}
999 
1000 	if (dm_suspended_md(md)) {
1001 		r = dm_resume(md);
1002 		if (!r && !dm_kobject_uevent(md, KOBJ_CHANGE, param->event_nr))
1003 			param->flags |= DM_UEVENT_GENERATED_FLAG;
1004 	}
1005 
1006 	if (old_map)
1007 		dm_table_destroy(old_map);
1008 
1009 	if (!r)
1010 		__dev_status(md, param);
1011 
1012 	dm_put(md);
1013 	return r;
1014 }
1015 
1016 /*
1017  * Set or unset the suspension state of a device.
1018  * If the device already is in the requested state we just return its status.
1019  */
1020 static int dev_suspend(struct dm_ioctl *param, size_t param_size)
1021 {
1022 	if (param->flags & DM_SUSPEND_FLAG)
1023 		return do_suspend(param);
1024 
1025 	return do_resume(param);
1026 }
1027 
1028 /*
1029  * Copies device info back to user space, used by
1030  * the create and info ioctls.
1031  */
1032 static int dev_status(struct dm_ioctl *param, size_t param_size)
1033 {
1034 	struct mapped_device *md;
1035 
1036 	md = find_device(param);
1037 	if (!md)
1038 		return -ENXIO;
1039 
1040 	__dev_status(md, param);
1041 	dm_put(md);
1042 
1043 	return 0;
1044 }
1045 
1046 /*
1047  * Build up the status struct for each target
1048  */
1049 static void retrieve_status(struct dm_table *table,
1050 			    struct dm_ioctl *param, size_t param_size)
1051 {
1052 	unsigned int i, num_targets;
1053 	struct dm_target_spec *spec;
1054 	char *outbuf, *outptr;
1055 	status_type_t type;
1056 	size_t remaining, len, used = 0;
1057 	unsigned status_flags = 0;
1058 
1059 	outptr = outbuf = get_result_buffer(param, param_size, &len);
1060 
1061 	if (param->flags & DM_STATUS_TABLE_FLAG)
1062 		type = STATUSTYPE_TABLE;
1063 	else
1064 		type = STATUSTYPE_INFO;
1065 
1066 	/* Get all the target info */
1067 	num_targets = dm_table_get_num_targets(table);
1068 	for (i = 0; i < num_targets; i++) {
1069 		struct dm_target *ti = dm_table_get_target(table, i);
1070 		size_t l;
1071 
1072 		remaining = len - (outptr - outbuf);
1073 		if (remaining <= sizeof(struct dm_target_spec)) {
1074 			param->flags |= DM_BUFFER_FULL_FLAG;
1075 			break;
1076 		}
1077 
1078 		spec = (struct dm_target_spec *) outptr;
1079 
1080 		spec->status = 0;
1081 		spec->sector_start = ti->begin;
1082 		spec->length = ti->len;
1083 		strncpy(spec->target_type, ti->type->name,
1084 			sizeof(spec->target_type));
1085 
1086 		outptr += sizeof(struct dm_target_spec);
1087 		remaining = len - (outptr - outbuf);
1088 		if (remaining <= 0) {
1089 			param->flags |= DM_BUFFER_FULL_FLAG;
1090 			break;
1091 		}
1092 
1093 		/* Get the status/table string from the target driver */
1094 		if (ti->type->status) {
1095 			if (param->flags & DM_NOFLUSH_FLAG)
1096 				status_flags |= DM_STATUS_NOFLUSH_FLAG;
1097 			ti->type->status(ti, type, status_flags, outptr, remaining);
1098 		} else
1099 			outptr[0] = '\0';
1100 
1101 		l = strlen(outptr) + 1;
1102 		if (l == remaining) {
1103 			param->flags |= DM_BUFFER_FULL_FLAG;
1104 			break;
1105 		}
1106 
1107 		outptr += l;
1108 		used = param->data_start + (outptr - outbuf);
1109 
1110 		outptr = align_ptr(outptr);
1111 		spec->next = outptr - outbuf;
1112 	}
1113 
1114 	if (used)
1115 		param->data_size = used;
1116 
1117 	param->target_count = num_targets;
1118 }
1119 
1120 /*
1121  * Wait for a device to report an event
1122  */
1123 static int dev_wait(struct dm_ioctl *param, size_t param_size)
1124 {
1125 	int r = 0;
1126 	struct mapped_device *md;
1127 	struct dm_table *table;
1128 
1129 	md = find_device(param);
1130 	if (!md)
1131 		return -ENXIO;
1132 
1133 	/*
1134 	 * Wait for a notification event
1135 	 */
1136 	if (dm_wait_event(md, param->event_nr)) {
1137 		r = -ERESTARTSYS;
1138 		goto out;
1139 	}
1140 
1141 	/*
1142 	 * The userland program is going to want to know what
1143 	 * changed to trigger the event, so we may as well tell
1144 	 * him and save an ioctl.
1145 	 */
1146 	__dev_status(md, param);
1147 
1148 	table = dm_get_live_or_inactive_table(md, param);
1149 	if (table) {
1150 		retrieve_status(table, param, param_size);
1151 		dm_table_put(table);
1152 	}
1153 
1154 out:
1155 	dm_put(md);
1156 
1157 	return r;
1158 }
1159 
1160 static inline fmode_t get_mode(struct dm_ioctl *param)
1161 {
1162 	fmode_t mode = FMODE_READ | FMODE_WRITE;
1163 
1164 	if (param->flags & DM_READONLY_FLAG)
1165 		mode = FMODE_READ;
1166 
1167 	return mode;
1168 }
1169 
1170 static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
1171 		       struct dm_target_spec **spec, char **target_params)
1172 {
1173 	*spec = (struct dm_target_spec *) ((unsigned char *) last + next);
1174 	*target_params = (char *) (*spec + 1);
1175 
1176 	if (*spec < (last + 1))
1177 		return -EINVAL;
1178 
1179 	return invalid_str(*target_params, end);
1180 }
1181 
1182 static int populate_table(struct dm_table *table,
1183 			  struct dm_ioctl *param, size_t param_size)
1184 {
1185 	int r;
1186 	unsigned int i = 0;
1187 	struct dm_target_spec *spec = (struct dm_target_spec *) param;
1188 	uint32_t next = param->data_start;
1189 	void *end = (void *) param + param_size;
1190 	char *target_params;
1191 
1192 	if (!param->target_count) {
1193 		DMWARN("populate_table: no targets specified");
1194 		return -EINVAL;
1195 	}
1196 
1197 	for (i = 0; i < param->target_count; i++) {
1198 
1199 		r = next_target(spec, next, end, &spec, &target_params);
1200 		if (r) {
1201 			DMWARN("unable to find target");
1202 			return r;
1203 		}
1204 
1205 		r = dm_table_add_target(table, spec->target_type,
1206 					(sector_t) spec->sector_start,
1207 					(sector_t) spec->length,
1208 					target_params);
1209 		if (r) {
1210 			DMWARN("error adding target to table");
1211 			return r;
1212 		}
1213 
1214 		next = spec->next;
1215 	}
1216 
1217 	return dm_table_complete(table);
1218 }
1219 
1220 static int table_load(struct dm_ioctl *param, size_t param_size)
1221 {
1222 	int r;
1223 	struct hash_cell *hc;
1224 	struct dm_table *t;
1225 	struct mapped_device *md;
1226 	struct target_type *immutable_target_type;
1227 
1228 	md = find_device(param);
1229 	if (!md)
1230 		return -ENXIO;
1231 
1232 	r = dm_table_create(&t, get_mode(param), param->target_count, md);
1233 	if (r)
1234 		goto out;
1235 
1236 	r = populate_table(t, param, param_size);
1237 	if (r) {
1238 		dm_table_destroy(t);
1239 		goto out;
1240 	}
1241 
1242 	immutable_target_type = dm_get_immutable_target_type(md);
1243 	if (immutable_target_type &&
1244 	    (immutable_target_type != dm_table_get_immutable_target_type(t))) {
1245 		DMWARN("can't replace immutable target type %s",
1246 		       immutable_target_type->name);
1247 		dm_table_destroy(t);
1248 		r = -EINVAL;
1249 		goto out;
1250 	}
1251 
1252 	/* Protect md->type and md->queue against concurrent table loads. */
1253 	dm_lock_md_type(md);
1254 	if (dm_get_md_type(md) == DM_TYPE_NONE)
1255 		/* Initial table load: acquire type of table. */
1256 		dm_set_md_type(md, dm_table_get_type(t));
1257 	else if (dm_get_md_type(md) != dm_table_get_type(t)) {
1258 		DMWARN("can't change device type after initial table load.");
1259 		dm_table_destroy(t);
1260 		dm_unlock_md_type(md);
1261 		r = -EINVAL;
1262 		goto out;
1263 	}
1264 
1265 	/* setup md->queue to reflect md's type (may block) */
1266 	r = dm_setup_md_queue(md);
1267 	if (r) {
1268 		DMWARN("unable to set up device queue for new table.");
1269 		dm_table_destroy(t);
1270 		dm_unlock_md_type(md);
1271 		goto out;
1272 	}
1273 	dm_unlock_md_type(md);
1274 
1275 	/* stage inactive table */
1276 	down_write(&_hash_lock);
1277 	hc = dm_get_mdptr(md);
1278 	if (!hc || hc->md != md) {
1279 		DMWARN("device has been removed from the dev hash table.");
1280 		dm_table_destroy(t);
1281 		up_write(&_hash_lock);
1282 		r = -ENXIO;
1283 		goto out;
1284 	}
1285 
1286 	if (hc->new_map)
1287 		dm_table_destroy(hc->new_map);
1288 	hc->new_map = t;
1289 	up_write(&_hash_lock);
1290 
1291 	param->flags |= DM_INACTIVE_PRESENT_FLAG;
1292 	__dev_status(md, param);
1293 
1294 out:
1295 	dm_put(md);
1296 
1297 	return r;
1298 }
1299 
1300 static int table_clear(struct dm_ioctl *param, size_t param_size)
1301 {
1302 	struct hash_cell *hc;
1303 	struct mapped_device *md;
1304 
1305 	down_write(&_hash_lock);
1306 
1307 	hc = __find_device_hash_cell(param);
1308 	if (!hc) {
1309 		DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
1310 		up_write(&_hash_lock);
1311 		return -ENXIO;
1312 	}
1313 
1314 	if (hc->new_map) {
1315 		dm_table_destroy(hc->new_map);
1316 		hc->new_map = NULL;
1317 	}
1318 
1319 	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1320 
1321 	__dev_status(hc->md, param);
1322 	md = hc->md;
1323 	up_write(&_hash_lock);
1324 	dm_put(md);
1325 
1326 	return 0;
1327 }
1328 
1329 /*
1330  * Retrieves a list of devices used by a particular dm device.
1331  */
1332 static void retrieve_deps(struct dm_table *table,
1333 			  struct dm_ioctl *param, size_t param_size)
1334 {
1335 	unsigned int count = 0;
1336 	struct list_head *tmp;
1337 	size_t len, needed;
1338 	struct dm_dev_internal *dd;
1339 	struct dm_target_deps *deps;
1340 
1341 	deps = get_result_buffer(param, param_size, &len);
1342 
1343 	/*
1344 	 * Count the devices.
1345 	 */
1346 	list_for_each (tmp, dm_table_get_devices(table))
1347 		count++;
1348 
1349 	/*
1350 	 * Check we have enough space.
1351 	 */
1352 	needed = sizeof(*deps) + (sizeof(*deps->dev) * count);
1353 	if (len < needed) {
1354 		param->flags |= DM_BUFFER_FULL_FLAG;
1355 		return;
1356 	}
1357 
1358 	/*
1359 	 * Fill in the devices.
1360 	 */
1361 	deps->count = count;
1362 	count = 0;
1363 	list_for_each_entry (dd, dm_table_get_devices(table), list)
1364 		deps->dev[count++] = huge_encode_dev(dd->dm_dev.bdev->bd_dev);
1365 
1366 	param->data_size = param->data_start + needed;
1367 }
1368 
1369 static int table_deps(struct dm_ioctl *param, size_t param_size)
1370 {
1371 	struct mapped_device *md;
1372 	struct dm_table *table;
1373 
1374 	md = find_device(param);
1375 	if (!md)
1376 		return -ENXIO;
1377 
1378 	__dev_status(md, param);
1379 
1380 	table = dm_get_live_or_inactive_table(md, param);
1381 	if (table) {
1382 		retrieve_deps(table, param, param_size);
1383 		dm_table_put(table);
1384 	}
1385 
1386 	dm_put(md);
1387 
1388 	return 0;
1389 }
1390 
1391 /*
1392  * Return the status of a device as a text string for each
1393  * target.
1394  */
1395 static int table_status(struct dm_ioctl *param, size_t param_size)
1396 {
1397 	struct mapped_device *md;
1398 	struct dm_table *table;
1399 
1400 	md = find_device(param);
1401 	if (!md)
1402 		return -ENXIO;
1403 
1404 	__dev_status(md, param);
1405 
1406 	table = dm_get_live_or_inactive_table(md, param);
1407 	if (table) {
1408 		retrieve_status(table, param, param_size);
1409 		dm_table_put(table);
1410 	}
1411 
1412 	dm_put(md);
1413 
1414 	return 0;
1415 }
1416 
1417 static bool buffer_test_overflow(char *result, unsigned maxlen)
1418 {
1419 	return !maxlen || strlen(result) + 1 >= maxlen;
1420 }
1421 
1422 /*
1423  * Process device-mapper dependent messages.
1424  * Returns a number <= 1 if message was processed by device mapper.
1425  * Returns 2 if message should be delivered to the target.
1426  */
1427 static int message_for_md(struct mapped_device *md, unsigned argc, char **argv,
1428 			  char *result, unsigned maxlen)
1429 {
1430 	return 2;
1431 }
1432 
1433 /*
1434  * Pass a message to the target that's at the supplied device offset.
1435  */
1436 static int target_message(struct dm_ioctl *param, size_t param_size)
1437 {
1438 	int r, argc;
1439 	char **argv;
1440 	struct mapped_device *md;
1441 	struct dm_table *table;
1442 	struct dm_target *ti;
1443 	struct dm_target_msg *tmsg = (void *) param + param->data_start;
1444 	size_t maxlen;
1445 	char *result = get_result_buffer(param, param_size, &maxlen);
1446 
1447 	md = find_device(param);
1448 	if (!md)
1449 		return -ENXIO;
1450 
1451 	if (tmsg < (struct dm_target_msg *) param->data ||
1452 	    invalid_str(tmsg->message, (void *) param + param_size)) {
1453 		DMWARN("Invalid target message parameters.");
1454 		r = -EINVAL;
1455 		goto out;
1456 	}
1457 
1458 	r = dm_split_args(&argc, &argv, tmsg->message);
1459 	if (r) {
1460 		DMWARN("Failed to split target message parameters");
1461 		goto out;
1462 	}
1463 
1464 	if (!argc) {
1465 		DMWARN("Empty message received.");
1466 		goto out_argv;
1467 	}
1468 
1469 	r = message_for_md(md, argc, argv, result, maxlen);
1470 	if (r <= 1)
1471 		goto out_argv;
1472 
1473 	table = dm_get_live_table(md);
1474 	if (!table)
1475 		goto out_argv;
1476 
1477 	if (dm_deleting_md(md)) {
1478 		r = -ENXIO;
1479 		goto out_table;
1480 	}
1481 
1482 	ti = dm_table_find_target(table, tmsg->sector);
1483 	if (!dm_target_is_valid(ti)) {
1484 		DMWARN("Target message sector outside device.");
1485 		r = -EINVAL;
1486 	} else if (ti->type->message)
1487 		r = ti->type->message(ti, argc, argv);
1488 	else {
1489 		DMWARN("Target type does not support messages");
1490 		r = -EINVAL;
1491 	}
1492 
1493  out_table:
1494 	dm_table_put(table);
1495  out_argv:
1496 	kfree(argv);
1497  out:
1498 	if (r >= 0)
1499 		__dev_status(md, param);
1500 
1501 	if (r == 1) {
1502 		param->flags |= DM_DATA_OUT_FLAG;
1503 		if (buffer_test_overflow(result, maxlen))
1504 			param->flags |= DM_BUFFER_FULL_FLAG;
1505 		else
1506 			param->data_size = param->data_start + strlen(result) + 1;
1507 		r = 0;
1508 	}
1509 
1510 	dm_put(md);
1511 	return r;
1512 }
1513 
1514 /*
1515  * The ioctl parameter block consists of two parts, a dm_ioctl struct
1516  * followed by a data buffer.  This flag is set if the second part,
1517  * which has a variable size, is not used by the function processing
1518  * the ioctl.
1519  */
1520 #define IOCTL_FLAGS_NO_PARAMS	1
1521 
1522 /*-----------------------------------------------------------------
1523  * Implementation of open/close/ioctl on the special char
1524  * device.
1525  *---------------------------------------------------------------*/
1526 static ioctl_fn lookup_ioctl(unsigned int cmd, int *ioctl_flags)
1527 {
1528 	static struct {
1529 		int cmd;
1530 		int flags;
1531 		ioctl_fn fn;
1532 	} _ioctls[] = {
1533 		{DM_VERSION_CMD, 0, NULL}, /* version is dealt with elsewhere */
1534 		{DM_REMOVE_ALL_CMD, IOCTL_FLAGS_NO_PARAMS, remove_all},
1535 		{DM_LIST_DEVICES_CMD, 0, list_devices},
1536 
1537 		{DM_DEV_CREATE_CMD, IOCTL_FLAGS_NO_PARAMS, dev_create},
1538 		{DM_DEV_REMOVE_CMD, IOCTL_FLAGS_NO_PARAMS, dev_remove},
1539 		{DM_DEV_RENAME_CMD, 0, dev_rename},
1540 		{DM_DEV_SUSPEND_CMD, IOCTL_FLAGS_NO_PARAMS, dev_suspend},
1541 		{DM_DEV_STATUS_CMD, IOCTL_FLAGS_NO_PARAMS, dev_status},
1542 		{DM_DEV_WAIT_CMD, 0, dev_wait},
1543 
1544 		{DM_TABLE_LOAD_CMD, 0, table_load},
1545 		{DM_TABLE_CLEAR_CMD, IOCTL_FLAGS_NO_PARAMS, table_clear},
1546 		{DM_TABLE_DEPS_CMD, 0, table_deps},
1547 		{DM_TABLE_STATUS_CMD, 0, table_status},
1548 
1549 		{DM_LIST_VERSIONS_CMD, 0, list_versions},
1550 
1551 		{DM_TARGET_MSG_CMD, 0, target_message},
1552 		{DM_DEV_SET_GEOMETRY_CMD, 0, dev_set_geometry}
1553 	};
1554 
1555 	if (unlikely(cmd >= ARRAY_SIZE(_ioctls)))
1556 		return NULL;
1557 
1558 	*ioctl_flags = _ioctls[cmd].flags;
1559 	return _ioctls[cmd].fn;
1560 }
1561 
1562 /*
1563  * As well as checking the version compatibility this always
1564  * copies the kernel interface version out.
1565  */
1566 static int check_version(unsigned int cmd, struct dm_ioctl __user *user)
1567 {
1568 	uint32_t version[3];
1569 	int r = 0;
1570 
1571 	if (copy_from_user(version, user->version, sizeof(version)))
1572 		return -EFAULT;
1573 
1574 	if ((DM_VERSION_MAJOR != version[0]) ||
1575 	    (DM_VERSION_MINOR < version[1])) {
1576 		DMWARN("ioctl interface mismatch: "
1577 		       "kernel(%u.%u.%u), user(%u.%u.%u), cmd(%d)",
1578 		       DM_VERSION_MAJOR, DM_VERSION_MINOR,
1579 		       DM_VERSION_PATCHLEVEL,
1580 		       version[0], version[1], version[2], cmd);
1581 		r = -EINVAL;
1582 	}
1583 
1584 	/*
1585 	 * Fill in the kernel version.
1586 	 */
1587 	version[0] = DM_VERSION_MAJOR;
1588 	version[1] = DM_VERSION_MINOR;
1589 	version[2] = DM_VERSION_PATCHLEVEL;
1590 	if (copy_to_user(user->version, version, sizeof(version)))
1591 		return -EFAULT;
1592 
1593 	return r;
1594 }
1595 
1596 #define DM_PARAMS_KMALLOC	0x0001	/* Params alloced with kmalloc */
1597 #define DM_PARAMS_VMALLOC	0x0002	/* Params alloced with vmalloc */
1598 #define DM_WIPE_BUFFER		0x0010	/* Wipe input buffer before returning from ioctl */
1599 
1600 static void free_params(struct dm_ioctl *param, size_t param_size, int param_flags)
1601 {
1602 	if (param_flags & DM_WIPE_BUFFER)
1603 		memset(param, 0, param_size);
1604 
1605 	if (param_flags & DM_PARAMS_KMALLOC)
1606 		kfree(param);
1607 	if (param_flags & DM_PARAMS_VMALLOC)
1608 		vfree(param);
1609 }
1610 
1611 static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl *param_kernel,
1612 		       int ioctl_flags,
1613 		       struct dm_ioctl **param, int *param_flags)
1614 {
1615 	struct dm_ioctl *dmi;
1616 	int secure_data;
1617 	const size_t minimum_data_size = sizeof(*param_kernel) - sizeof(param_kernel->data);
1618 
1619 	if (copy_from_user(param_kernel, user, minimum_data_size))
1620 		return -EFAULT;
1621 
1622 	if (param_kernel->data_size < minimum_data_size)
1623 		return -EINVAL;
1624 
1625 	secure_data = param_kernel->flags & DM_SECURE_DATA_FLAG;
1626 
1627 	*param_flags = secure_data ? DM_WIPE_BUFFER : 0;
1628 
1629 	if (ioctl_flags & IOCTL_FLAGS_NO_PARAMS) {
1630 		dmi = param_kernel;
1631 		dmi->data_size = minimum_data_size;
1632 		goto data_copied;
1633 	}
1634 
1635 	/*
1636 	 * Try to avoid low memory issues when a device is suspended.
1637 	 * Use kmalloc() rather than vmalloc() when we can.
1638 	 */
1639 	dmi = NULL;
1640 	if (param_kernel->data_size <= KMALLOC_MAX_SIZE) {
1641 		dmi = kmalloc(param_kernel->data_size, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
1642 		if (dmi)
1643 			*param_flags |= DM_PARAMS_KMALLOC;
1644 	}
1645 
1646 	if (!dmi) {
1647 		dmi = __vmalloc(param_kernel->data_size, GFP_NOIO | __GFP_REPEAT | __GFP_HIGH, PAGE_KERNEL);
1648 		if (dmi)
1649 			*param_flags |= DM_PARAMS_VMALLOC;
1650 	}
1651 
1652 	if (!dmi) {
1653 		if (secure_data && clear_user(user, param_kernel->data_size))
1654 			return -EFAULT;
1655 		return -ENOMEM;
1656 	}
1657 
1658 	if (copy_from_user(dmi, user, param_kernel->data_size))
1659 		goto bad;
1660 
1661 data_copied:
1662 	/*
1663 	 * Abort if something changed the ioctl data while it was being copied.
1664 	 */
1665 	if (dmi->data_size != param_kernel->data_size) {
1666 		DMERR("rejecting ioctl: data size modified while processing parameters");
1667 		goto bad;
1668 	}
1669 
1670 	/* Wipe the user buffer so we do not return it to userspace */
1671 	if (secure_data && clear_user(user, param_kernel->data_size))
1672 		goto bad;
1673 
1674 	*param = dmi;
1675 	return 0;
1676 
1677 bad:
1678 	free_params(dmi, param_kernel->data_size, *param_flags);
1679 
1680 	return -EFAULT;
1681 }
1682 
1683 static int validate_params(uint cmd, struct dm_ioctl *param)
1684 {
1685 	/* Always clear this flag */
1686 	param->flags &= ~DM_BUFFER_FULL_FLAG;
1687 	param->flags &= ~DM_UEVENT_GENERATED_FLAG;
1688 	param->flags &= ~DM_SECURE_DATA_FLAG;
1689 	param->flags &= ~DM_DATA_OUT_FLAG;
1690 
1691 	/* Ignores parameters */
1692 	if (cmd == DM_REMOVE_ALL_CMD ||
1693 	    cmd == DM_LIST_DEVICES_CMD ||
1694 	    cmd == DM_LIST_VERSIONS_CMD)
1695 		return 0;
1696 
1697 	if ((cmd == DM_DEV_CREATE_CMD)) {
1698 		if (!*param->name) {
1699 			DMWARN("name not supplied when creating device");
1700 			return -EINVAL;
1701 		}
1702 	} else if ((*param->uuid && *param->name)) {
1703 		DMWARN("only supply one of name or uuid, cmd(%u)", cmd);
1704 		return -EINVAL;
1705 	}
1706 
1707 	/* Ensure strings are terminated */
1708 	param->name[DM_NAME_LEN - 1] = '\0';
1709 	param->uuid[DM_UUID_LEN - 1] = '\0';
1710 
1711 	return 0;
1712 }
1713 
1714 static int ctl_ioctl(uint command, struct dm_ioctl __user *user)
1715 {
1716 	int r = 0;
1717 	int ioctl_flags;
1718 	int param_flags;
1719 	unsigned int cmd;
1720 	struct dm_ioctl *uninitialized_var(param);
1721 	ioctl_fn fn = NULL;
1722 	size_t input_param_size;
1723 	struct dm_ioctl param_kernel;
1724 
1725 	/* only root can play with this */
1726 	if (!capable(CAP_SYS_ADMIN))
1727 		return -EACCES;
1728 
1729 	if (_IOC_TYPE(command) != DM_IOCTL)
1730 		return -ENOTTY;
1731 
1732 	cmd = _IOC_NR(command);
1733 
1734 	/*
1735 	 * Check the interface version passed in.  This also
1736 	 * writes out the kernel's interface version.
1737 	 */
1738 	r = check_version(cmd, user);
1739 	if (r)
1740 		return r;
1741 
1742 	/*
1743 	 * Nothing more to do for the version command.
1744 	 */
1745 	if (cmd == DM_VERSION_CMD)
1746 		return 0;
1747 
1748 	fn = lookup_ioctl(cmd, &ioctl_flags);
1749 	if (!fn) {
1750 		DMWARN("dm_ctl_ioctl: unknown command 0x%x", command);
1751 		return -ENOTTY;
1752 	}
1753 
1754 	/*
1755 	 * Copy the parameters into kernel space.
1756 	 */
1757 	r = copy_params(user, &param_kernel, ioctl_flags, &param, &param_flags);
1758 
1759 	if (r)
1760 		return r;
1761 
1762 	input_param_size = param->data_size;
1763 	r = validate_params(cmd, param);
1764 	if (r)
1765 		goto out;
1766 
1767 	param->data_size = sizeof(*param);
1768 	r = fn(param, input_param_size);
1769 
1770 	if (unlikely(param->flags & DM_BUFFER_FULL_FLAG) &&
1771 	    unlikely(ioctl_flags & IOCTL_FLAGS_NO_PARAMS))
1772 		DMERR("ioctl %d tried to output some data but has IOCTL_FLAGS_NO_PARAMS set", cmd);
1773 
1774 	/*
1775 	 * Copy the results back to userland.
1776 	 */
1777 	if (!r && copy_to_user(user, param, param->data_size))
1778 		r = -EFAULT;
1779 
1780 out:
1781 	free_params(param, input_param_size, param_flags);
1782 	return r;
1783 }
1784 
1785 static long dm_ctl_ioctl(struct file *file, uint command, ulong u)
1786 {
1787 	return (long)ctl_ioctl(command, (struct dm_ioctl __user *)u);
1788 }
1789 
1790 #ifdef CONFIG_COMPAT
1791 static long dm_compat_ctl_ioctl(struct file *file, uint command, ulong u)
1792 {
1793 	return (long)dm_ctl_ioctl(file, command, (ulong) compat_ptr(u));
1794 }
1795 #else
1796 #define dm_compat_ctl_ioctl NULL
1797 #endif
1798 
1799 static const struct file_operations _ctl_fops = {
1800 	.open = nonseekable_open,
1801 	.unlocked_ioctl	 = dm_ctl_ioctl,
1802 	.compat_ioctl = dm_compat_ctl_ioctl,
1803 	.owner	 = THIS_MODULE,
1804 	.llseek  = noop_llseek,
1805 };
1806 
1807 static struct miscdevice _dm_misc = {
1808 	.minor		= MAPPER_CTRL_MINOR,
1809 	.name  		= DM_NAME,
1810 	.nodename	= DM_DIR "/" DM_CONTROL_NODE,
1811 	.fops  		= &_ctl_fops
1812 };
1813 
1814 MODULE_ALIAS_MISCDEV(MAPPER_CTRL_MINOR);
1815 MODULE_ALIAS("devname:" DM_DIR "/" DM_CONTROL_NODE);
1816 
1817 /*
1818  * Create misc character device and link to DM_DIR/control.
1819  */
1820 int __init dm_interface_init(void)
1821 {
1822 	int r;
1823 
1824 	r = dm_hash_init();
1825 	if (r)
1826 		return r;
1827 
1828 	r = misc_register(&_dm_misc);
1829 	if (r) {
1830 		DMERR("misc_register failed for control device");
1831 		dm_hash_exit();
1832 		return r;
1833 	}
1834 
1835 	DMINFO("%d.%d.%d%s initialised: %s", DM_VERSION_MAJOR,
1836 	       DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL, DM_VERSION_EXTRA,
1837 	       DM_DRIVER_EMAIL);
1838 	return 0;
1839 }
1840 
1841 void dm_interface_exit(void)
1842 {
1843 	if (misc_deregister(&_dm_misc) < 0)
1844 		DMERR("misc_deregister failed for control device");
1845 
1846 	dm_hash_exit();
1847 }
1848 
1849 /**
1850  * dm_copy_name_and_uuid - Copy mapped device name & uuid into supplied buffers
1851  * @md: Pointer to mapped_device
1852  * @name: Buffer (size DM_NAME_LEN) for name
1853  * @uuid: Buffer (size DM_UUID_LEN) for uuid or empty string if uuid not defined
1854  */
1855 int dm_copy_name_and_uuid(struct mapped_device *md, char *name, char *uuid)
1856 {
1857 	int r = 0;
1858 	struct hash_cell *hc;
1859 
1860 	if (!md)
1861 		return -ENXIO;
1862 
1863 	mutex_lock(&dm_hash_cells_mutex);
1864 	hc = dm_get_mdptr(md);
1865 	if (!hc || hc->md != md) {
1866 		r = -ENXIO;
1867 		goto out;
1868 	}
1869 
1870 	if (name)
1871 		strcpy(name, hc->name);
1872 	if (uuid)
1873 		strcpy(uuid, hc->uuid ? : "");
1874 
1875 out:
1876 	mutex_unlock(&dm_hash_cells_mutex);
1877 
1878 	return r;
1879 }
1880