xref: /openbmc/linux/drivers/md/dm-ioctl.c (revision f7777dcc)
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 /*
40  * A dummy definition to make RCU happy.
41  * struct dm_table should never be dereferenced in this file.
42  */
43 struct dm_table {
44 	int undefined__;
45 };
46 
47 struct vers_iter {
48     size_t param_size;
49     struct dm_target_versions *vers, *old_vers;
50     char *end;
51     uint32_t flags;
52 };
53 
54 
55 #define NUM_BUCKETS 64
56 #define MASK_BUCKETS (NUM_BUCKETS - 1)
57 static struct list_head _name_buckets[NUM_BUCKETS];
58 static struct list_head _uuid_buckets[NUM_BUCKETS];
59 
60 static void dm_hash_remove_all(int keep_open_devices);
61 
62 /*
63  * Guards access to both hash tables.
64  */
65 static DECLARE_RWSEM(_hash_lock);
66 
67 /*
68  * Protects use of mdptr to obtain hash cell name and uuid from mapped device.
69  */
70 static DEFINE_MUTEX(dm_hash_cells_mutex);
71 
72 static void init_buckets(struct list_head *buckets)
73 {
74 	unsigned int i;
75 
76 	for (i = 0; i < NUM_BUCKETS; i++)
77 		INIT_LIST_HEAD(buckets + i);
78 }
79 
80 static int dm_hash_init(void)
81 {
82 	init_buckets(_name_buckets);
83 	init_buckets(_uuid_buckets);
84 	return 0;
85 }
86 
87 static void dm_hash_exit(void)
88 {
89 	dm_hash_remove_all(0);
90 }
91 
92 /*-----------------------------------------------------------------
93  * Hash function:
94  * We're not really concerned with the str hash function being
95  * fast since it's only used by the ioctl interface.
96  *---------------------------------------------------------------*/
97 static unsigned int hash_str(const char *str)
98 {
99 	const unsigned int hash_mult = 2654435387U;
100 	unsigned int h = 0;
101 
102 	while (*str)
103 		h = (h + (unsigned int) *str++) * hash_mult;
104 
105 	return h & MASK_BUCKETS;
106 }
107 
108 /*-----------------------------------------------------------------
109  * Code for looking up a device by name
110  *---------------------------------------------------------------*/
111 static struct hash_cell *__get_name_cell(const char *str)
112 {
113 	struct hash_cell *hc;
114 	unsigned int h = hash_str(str);
115 
116 	list_for_each_entry (hc, _name_buckets + h, name_list)
117 		if (!strcmp(hc->name, str)) {
118 			dm_get(hc->md);
119 			return hc;
120 		}
121 
122 	return NULL;
123 }
124 
125 static struct hash_cell *__get_uuid_cell(const char *str)
126 {
127 	struct hash_cell *hc;
128 	unsigned int h = hash_str(str);
129 
130 	list_for_each_entry (hc, _uuid_buckets + h, uuid_list)
131 		if (!strcmp(hc->uuid, str)) {
132 			dm_get(hc->md);
133 			return hc;
134 		}
135 
136 	return NULL;
137 }
138 
139 static struct hash_cell *__get_dev_cell(uint64_t dev)
140 {
141 	struct mapped_device *md;
142 	struct hash_cell *hc;
143 
144 	md = dm_get_md(huge_decode_dev(dev));
145 	if (!md)
146 		return NULL;
147 
148 	hc = dm_get_mdptr(md);
149 	if (!hc) {
150 		dm_put(md);
151 		return NULL;
152 	}
153 
154 	return hc;
155 }
156 
157 /*-----------------------------------------------------------------
158  * Inserting, removing and renaming a device.
159  *---------------------------------------------------------------*/
160 static struct hash_cell *alloc_cell(const char *name, const char *uuid,
161 				    struct mapped_device *md)
162 {
163 	struct hash_cell *hc;
164 
165 	hc = kmalloc(sizeof(*hc), GFP_KERNEL);
166 	if (!hc)
167 		return NULL;
168 
169 	hc->name = kstrdup(name, GFP_KERNEL);
170 	if (!hc->name) {
171 		kfree(hc);
172 		return NULL;
173 	}
174 
175 	if (!uuid)
176 		hc->uuid = NULL;
177 
178 	else {
179 		hc->uuid = kstrdup(uuid, GFP_KERNEL);
180 		if (!hc->uuid) {
181 			kfree(hc->name);
182 			kfree(hc);
183 			return NULL;
184 		}
185 	}
186 
187 	INIT_LIST_HEAD(&hc->name_list);
188 	INIT_LIST_HEAD(&hc->uuid_list);
189 	hc->md = md;
190 	hc->new_map = NULL;
191 	return hc;
192 }
193 
194 static void free_cell(struct hash_cell *hc)
195 {
196 	if (hc) {
197 		kfree(hc->name);
198 		kfree(hc->uuid);
199 		kfree(hc);
200 	}
201 }
202 
203 /*
204  * The kdev_t and uuid of a device can never change once it is
205  * initially inserted.
206  */
207 static int dm_hash_insert(const char *name, const char *uuid, struct mapped_device *md)
208 {
209 	struct hash_cell *cell, *hc;
210 
211 	/*
212 	 * Allocate the new cells.
213 	 */
214 	cell = alloc_cell(name, uuid, md);
215 	if (!cell)
216 		return -ENOMEM;
217 
218 	/*
219 	 * Insert the cell into both hash tables.
220 	 */
221 	down_write(&_hash_lock);
222 	hc = __get_name_cell(name);
223 	if (hc) {
224 		dm_put(hc->md);
225 		goto bad;
226 	}
227 
228 	list_add(&cell->name_list, _name_buckets + hash_str(name));
229 
230 	if (uuid) {
231 		hc = __get_uuid_cell(uuid);
232 		if (hc) {
233 			list_del(&cell->name_list);
234 			dm_put(hc->md);
235 			goto bad;
236 		}
237 		list_add(&cell->uuid_list, _uuid_buckets + hash_str(uuid));
238 	}
239 	dm_get(md);
240 	mutex_lock(&dm_hash_cells_mutex);
241 	dm_set_mdptr(md, cell);
242 	mutex_unlock(&dm_hash_cells_mutex);
243 	up_write(&_hash_lock);
244 
245 	return 0;
246 
247  bad:
248 	up_write(&_hash_lock);
249 	free_cell(cell);
250 	return -EBUSY;
251 }
252 
253 static struct dm_table *__hash_remove(struct hash_cell *hc)
254 {
255 	struct dm_table *table;
256 	int srcu_idx;
257 
258 	/* remove from the dev hash */
259 	list_del(&hc->uuid_list);
260 	list_del(&hc->name_list);
261 	mutex_lock(&dm_hash_cells_mutex);
262 	dm_set_mdptr(hc->md, NULL);
263 	mutex_unlock(&dm_hash_cells_mutex);
264 
265 	table = dm_get_live_table(hc->md, &srcu_idx);
266 	if (table)
267 		dm_table_event(table);
268 	dm_put_live_table(hc->md, srcu_idx);
269 
270 	table = NULL;
271 	if (hc->new_map)
272 		table = hc->new_map;
273 	dm_put(hc->md);
274 	free_cell(hc);
275 
276 	return table;
277 }
278 
279 static void dm_hash_remove_all(int keep_open_devices)
280 {
281 	int i, dev_skipped;
282 	struct hash_cell *hc;
283 	struct mapped_device *md;
284 	struct dm_table *t;
285 
286 retry:
287 	dev_skipped = 0;
288 
289 	down_write(&_hash_lock);
290 
291 	for (i = 0; i < NUM_BUCKETS; i++) {
292 		list_for_each_entry(hc, _name_buckets + i, name_list) {
293 			md = hc->md;
294 			dm_get(md);
295 
296 			if (keep_open_devices && dm_lock_for_deletion(md)) {
297 				dm_put(md);
298 				dev_skipped++;
299 				continue;
300 			}
301 
302 			t = __hash_remove(hc);
303 
304 			up_write(&_hash_lock);
305 
306 			if (t) {
307 				dm_sync_table(md);
308 				dm_table_destroy(t);
309 			}
310 			dm_put(md);
311 			if (likely(keep_open_devices))
312 				dm_destroy(md);
313 			else
314 				dm_destroy_immediate(md);
315 
316 			/*
317 			 * Some mapped devices may be using other mapped
318 			 * devices, so repeat until we make no further
319 			 * progress.  If a new mapped device is created
320 			 * here it will also get removed.
321 			 */
322 			goto retry;
323 		}
324 	}
325 
326 	up_write(&_hash_lock);
327 
328 	if (dev_skipped)
329 		DMWARN("remove_all left %d open device(s)", dev_skipped);
330 }
331 
332 /*
333  * Set the uuid of a hash_cell that isn't already set.
334  */
335 static void __set_cell_uuid(struct hash_cell *hc, char *new_uuid)
336 {
337 	mutex_lock(&dm_hash_cells_mutex);
338 	hc->uuid = new_uuid;
339 	mutex_unlock(&dm_hash_cells_mutex);
340 
341 	list_add(&hc->uuid_list, _uuid_buckets + hash_str(new_uuid));
342 }
343 
344 /*
345  * Changes the name of a hash_cell and returns the old name for
346  * the caller to free.
347  */
348 static char *__change_cell_name(struct hash_cell *hc, char *new_name)
349 {
350 	char *old_name;
351 
352 	/*
353 	 * Rename and move the name cell.
354 	 */
355 	list_del(&hc->name_list);
356 	old_name = hc->name;
357 
358 	mutex_lock(&dm_hash_cells_mutex);
359 	hc->name = new_name;
360 	mutex_unlock(&dm_hash_cells_mutex);
361 
362 	list_add(&hc->name_list, _name_buckets + hash_str(new_name));
363 
364 	return old_name;
365 }
366 
367 static struct mapped_device *dm_hash_rename(struct dm_ioctl *param,
368 					    const char *new)
369 {
370 	char *new_data, *old_name = NULL;
371 	struct hash_cell *hc;
372 	struct dm_table *table;
373 	struct mapped_device *md;
374 	unsigned change_uuid = (param->flags & DM_UUID_FLAG) ? 1 : 0;
375 	int srcu_idx;
376 
377 	/*
378 	 * duplicate new.
379 	 */
380 	new_data = kstrdup(new, GFP_KERNEL);
381 	if (!new_data)
382 		return ERR_PTR(-ENOMEM);
383 
384 	down_write(&_hash_lock);
385 
386 	/*
387 	 * Is new free ?
388 	 */
389 	if (change_uuid)
390 		hc = __get_uuid_cell(new);
391 	else
392 		hc = __get_name_cell(new);
393 
394 	if (hc) {
395 		DMWARN("Unable to change %s on mapped device %s to one that "
396 		       "already exists: %s",
397 		       change_uuid ? "uuid" : "name",
398 		       param->name, new);
399 		dm_put(hc->md);
400 		up_write(&_hash_lock);
401 		kfree(new_data);
402 		return ERR_PTR(-EBUSY);
403 	}
404 
405 	/*
406 	 * Is there such a device as 'old' ?
407 	 */
408 	hc = __get_name_cell(param->name);
409 	if (!hc) {
410 		DMWARN("Unable to rename non-existent device, %s to %s%s",
411 		       param->name, change_uuid ? "uuid " : "", new);
412 		up_write(&_hash_lock);
413 		kfree(new_data);
414 		return ERR_PTR(-ENXIO);
415 	}
416 
417 	/*
418 	 * Does this device already have a uuid?
419 	 */
420 	if (change_uuid && hc->uuid) {
421 		DMWARN("Unable to change uuid of mapped device %s to %s "
422 		       "because uuid is already set to %s",
423 		       param->name, new, hc->uuid);
424 		dm_put(hc->md);
425 		up_write(&_hash_lock);
426 		kfree(new_data);
427 		return ERR_PTR(-EINVAL);
428 	}
429 
430 	if (change_uuid)
431 		__set_cell_uuid(hc, new_data);
432 	else
433 		old_name = __change_cell_name(hc, new_data);
434 
435 	/*
436 	 * Wake up any dm event waiters.
437 	 */
438 	table = dm_get_live_table(hc->md, &srcu_idx);
439 	if (table)
440 		dm_table_event(table);
441 	dm_put_live_table(hc->md, srcu_idx);
442 
443 	if (!dm_kobject_uevent(hc->md, KOBJ_CHANGE, param->event_nr))
444 		param->flags |= DM_UEVENT_GENERATED_FLAG;
445 
446 	md = hc->md;
447 	up_write(&_hash_lock);
448 	kfree(old_name);
449 
450 	return md;
451 }
452 
453 /*-----------------------------------------------------------------
454  * Implementation of the ioctl commands
455  *---------------------------------------------------------------*/
456 /*
457  * All the ioctl commands get dispatched to functions with this
458  * prototype.
459  */
460 typedef int (*ioctl_fn)(struct dm_ioctl *param, size_t param_size);
461 
462 static int remove_all(struct dm_ioctl *param, size_t param_size)
463 {
464 	dm_hash_remove_all(1);
465 	param->data_size = 0;
466 	return 0;
467 }
468 
469 /*
470  * Round up the ptr to an 8-byte boundary.
471  */
472 #define ALIGN_MASK 7
473 static inline void *align_ptr(void *ptr)
474 {
475 	return (void *) (((size_t) (ptr + ALIGN_MASK)) & ~ALIGN_MASK);
476 }
477 
478 /*
479  * Retrieves the data payload buffer from an already allocated
480  * struct dm_ioctl.
481  */
482 static void *get_result_buffer(struct dm_ioctl *param, size_t param_size,
483 			       size_t *len)
484 {
485 	param->data_start = align_ptr(param + 1) - (void *) param;
486 
487 	if (param->data_start < param_size)
488 		*len = param_size - param->data_start;
489 	else
490 		*len = 0;
491 
492 	return ((void *) param) + param->data_start;
493 }
494 
495 static int list_devices(struct dm_ioctl *param, size_t param_size)
496 {
497 	unsigned int i;
498 	struct hash_cell *hc;
499 	size_t len, needed = 0;
500 	struct gendisk *disk;
501 	struct dm_name_list *nl, *old_nl = NULL;
502 
503 	down_write(&_hash_lock);
504 
505 	/*
506 	 * Loop through all the devices working out how much
507 	 * space we need.
508 	 */
509 	for (i = 0; i < NUM_BUCKETS; i++) {
510 		list_for_each_entry (hc, _name_buckets + i, name_list) {
511 			needed += sizeof(struct dm_name_list);
512 			needed += strlen(hc->name) + 1;
513 			needed += ALIGN_MASK;
514 		}
515 	}
516 
517 	/*
518 	 * Grab our output buffer.
519 	 */
520 	nl = get_result_buffer(param, param_size, &len);
521 	if (len < needed) {
522 		param->flags |= DM_BUFFER_FULL_FLAG;
523 		goto out;
524 	}
525 	param->data_size = param->data_start + needed;
526 
527 	nl->dev = 0;	/* Flags no data */
528 
529 	/*
530 	 * Now loop through filling out the names.
531 	 */
532 	for (i = 0; i < NUM_BUCKETS; i++) {
533 		list_for_each_entry (hc, _name_buckets + i, name_list) {
534 			if (old_nl)
535 				old_nl->next = (uint32_t) ((void *) nl -
536 							   (void *) old_nl);
537 			disk = dm_disk(hc->md);
538 			nl->dev = huge_encode_dev(disk_devt(disk));
539 			nl->next = 0;
540 			strcpy(nl->name, hc->name);
541 
542 			old_nl = nl;
543 			nl = align_ptr(((void *) ++nl) + strlen(hc->name) + 1);
544 		}
545 	}
546 
547  out:
548 	up_write(&_hash_lock);
549 	return 0;
550 }
551 
552 static void list_version_get_needed(struct target_type *tt, void *needed_param)
553 {
554     size_t *needed = needed_param;
555 
556     *needed += sizeof(struct dm_target_versions);
557     *needed += strlen(tt->name);
558     *needed += ALIGN_MASK;
559 }
560 
561 static void list_version_get_info(struct target_type *tt, void *param)
562 {
563     struct vers_iter *info = param;
564 
565     /* Check space - it might have changed since the first iteration */
566     if ((char *)info->vers + sizeof(tt->version) + strlen(tt->name) + 1 >
567 	info->end) {
568 
569 	info->flags = DM_BUFFER_FULL_FLAG;
570 	return;
571     }
572 
573     if (info->old_vers)
574 	info->old_vers->next = (uint32_t) ((void *)info->vers -
575 					   (void *)info->old_vers);
576     info->vers->version[0] = tt->version[0];
577     info->vers->version[1] = tt->version[1];
578     info->vers->version[2] = tt->version[2];
579     info->vers->next = 0;
580     strcpy(info->vers->name, tt->name);
581 
582     info->old_vers = info->vers;
583     info->vers = align_ptr(((void *) ++info->vers) + strlen(tt->name) + 1);
584 }
585 
586 static int list_versions(struct dm_ioctl *param, size_t param_size)
587 {
588 	size_t len, needed = 0;
589 	struct dm_target_versions *vers;
590 	struct vers_iter iter_info;
591 
592 	/*
593 	 * Loop through all the devices working out how much
594 	 * space we need.
595 	 */
596 	dm_target_iterate(list_version_get_needed, &needed);
597 
598 	/*
599 	 * Grab our output buffer.
600 	 */
601 	vers = get_result_buffer(param, param_size, &len);
602 	if (len < needed) {
603 		param->flags |= DM_BUFFER_FULL_FLAG;
604 		goto out;
605 	}
606 	param->data_size = param->data_start + needed;
607 
608 	iter_info.param_size = param_size;
609 	iter_info.old_vers = NULL;
610 	iter_info.vers = vers;
611 	iter_info.flags = 0;
612 	iter_info.end = (char *)vers+len;
613 
614 	/*
615 	 * Now loop through filling out the names & versions.
616 	 */
617 	dm_target_iterate(list_version_get_info, &iter_info);
618 	param->flags |= iter_info.flags;
619 
620  out:
621 	return 0;
622 }
623 
624 static int check_name(const char *name)
625 {
626 	if (strchr(name, '/')) {
627 		DMWARN("invalid device name");
628 		return -EINVAL;
629 	}
630 
631 	return 0;
632 }
633 
634 /*
635  * On successful return, the caller must not attempt to acquire
636  * _hash_lock without first calling dm_table_put, because dm_table_destroy
637  * waits for this dm_table_put and could be called under this lock.
638  */
639 static struct dm_table *dm_get_inactive_table(struct mapped_device *md, int *srcu_idx)
640 {
641 	struct hash_cell *hc;
642 	struct dm_table *table = NULL;
643 
644 	/* increment rcu count, we don't care about the table pointer */
645 	dm_get_live_table(md, srcu_idx);
646 
647 	down_read(&_hash_lock);
648 	hc = dm_get_mdptr(md);
649 	if (!hc || hc->md != md) {
650 		DMWARN("device has been removed from the dev hash table.");
651 		goto out;
652 	}
653 
654 	table = hc->new_map;
655 
656 out:
657 	up_read(&_hash_lock);
658 
659 	return table;
660 }
661 
662 static struct dm_table *dm_get_live_or_inactive_table(struct mapped_device *md,
663 						      struct dm_ioctl *param,
664 						      int *srcu_idx)
665 {
666 	return (param->flags & DM_QUERY_INACTIVE_TABLE_FLAG) ?
667 		dm_get_inactive_table(md, srcu_idx) : dm_get_live_table(md, srcu_idx);
668 }
669 
670 /*
671  * Fills in a dm_ioctl structure, ready for sending back to
672  * userland.
673  */
674 static void __dev_status(struct mapped_device *md, struct dm_ioctl *param)
675 {
676 	struct gendisk *disk = dm_disk(md);
677 	struct dm_table *table;
678 	int srcu_idx;
679 
680 	param->flags &= ~(DM_SUSPEND_FLAG | DM_READONLY_FLAG |
681 			  DM_ACTIVE_PRESENT_FLAG);
682 
683 	if (dm_suspended_md(md))
684 		param->flags |= DM_SUSPEND_FLAG;
685 
686 	param->dev = huge_encode_dev(disk_devt(disk));
687 
688 	/*
689 	 * Yes, this will be out of date by the time it gets back
690 	 * to userland, but it is still very useful for
691 	 * debugging.
692 	 */
693 	param->open_count = dm_open_count(md);
694 
695 	param->event_nr = dm_get_event_nr(md);
696 	param->target_count = 0;
697 
698 	table = dm_get_live_table(md, &srcu_idx);
699 	if (table) {
700 		if (!(param->flags & DM_QUERY_INACTIVE_TABLE_FLAG)) {
701 			if (get_disk_ro(disk))
702 				param->flags |= DM_READONLY_FLAG;
703 			param->target_count = dm_table_get_num_targets(table);
704 		}
705 
706 		param->flags |= DM_ACTIVE_PRESENT_FLAG;
707 	}
708 	dm_put_live_table(md, srcu_idx);
709 
710 	if (param->flags & DM_QUERY_INACTIVE_TABLE_FLAG) {
711 		int srcu_idx;
712 		table = dm_get_inactive_table(md, &srcu_idx);
713 		if (table) {
714 			if (!(dm_table_get_mode(table) & FMODE_WRITE))
715 				param->flags |= DM_READONLY_FLAG;
716 			param->target_count = dm_table_get_num_targets(table);
717 		}
718 		dm_put_live_table(md, srcu_idx);
719 	}
720 }
721 
722 static int dev_create(struct dm_ioctl *param, size_t param_size)
723 {
724 	int r, m = DM_ANY_MINOR;
725 	struct mapped_device *md;
726 
727 	r = check_name(param->name);
728 	if (r)
729 		return r;
730 
731 	if (param->flags & DM_PERSISTENT_DEV_FLAG)
732 		m = MINOR(huge_decode_dev(param->dev));
733 
734 	r = dm_create(m, &md);
735 	if (r)
736 		return r;
737 
738 	r = dm_hash_insert(param->name, *param->uuid ? param->uuid : NULL, md);
739 	if (r) {
740 		dm_put(md);
741 		dm_destroy(md);
742 		return r;
743 	}
744 
745 	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
746 
747 	__dev_status(md, param);
748 
749 	dm_put(md);
750 
751 	return 0;
752 }
753 
754 /*
755  * Always use UUID for lookups if it's present, otherwise use name or dev.
756  */
757 static struct hash_cell *__find_device_hash_cell(struct dm_ioctl *param)
758 {
759 	struct hash_cell *hc = NULL;
760 
761 	if (*param->uuid) {
762 		if (*param->name || param->dev)
763 			return NULL;
764 
765 		hc = __get_uuid_cell(param->uuid);
766 		if (!hc)
767 			return NULL;
768 	} else if (*param->name) {
769 		if (param->dev)
770 			return NULL;
771 
772 		hc = __get_name_cell(param->name);
773 		if (!hc)
774 			return NULL;
775 	} else if (param->dev) {
776 		hc = __get_dev_cell(param->dev);
777 		if (!hc)
778 			return NULL;
779 	} else
780 		return NULL;
781 
782 	/*
783 	 * Sneakily write in both the name and the uuid
784 	 * while we have the cell.
785 	 */
786 	strlcpy(param->name, hc->name, sizeof(param->name));
787 	if (hc->uuid)
788 		strlcpy(param->uuid, hc->uuid, sizeof(param->uuid));
789 	else
790 		param->uuid[0] = '\0';
791 
792 	if (hc->new_map)
793 		param->flags |= DM_INACTIVE_PRESENT_FLAG;
794 	else
795 		param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
796 
797 	return hc;
798 }
799 
800 static struct mapped_device *find_device(struct dm_ioctl *param)
801 {
802 	struct hash_cell *hc;
803 	struct mapped_device *md = NULL;
804 
805 	down_read(&_hash_lock);
806 	hc = __find_device_hash_cell(param);
807 	if (hc)
808 		md = hc->md;
809 	up_read(&_hash_lock);
810 
811 	return md;
812 }
813 
814 static int dev_remove(struct dm_ioctl *param, size_t param_size)
815 {
816 	struct hash_cell *hc;
817 	struct mapped_device *md;
818 	int r;
819 	struct dm_table *t;
820 
821 	down_write(&_hash_lock);
822 	hc = __find_device_hash_cell(param);
823 
824 	if (!hc) {
825 		DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
826 		up_write(&_hash_lock);
827 		return -ENXIO;
828 	}
829 
830 	md = hc->md;
831 
832 	/*
833 	 * Ensure the device is not open and nothing further can open it.
834 	 */
835 	r = dm_lock_for_deletion(md);
836 	if (r) {
837 		DMDEBUG_LIMIT("unable to remove open device %s", hc->name);
838 		up_write(&_hash_lock);
839 		dm_put(md);
840 		return r;
841 	}
842 
843 	t = __hash_remove(hc);
844 	up_write(&_hash_lock);
845 
846 	if (t) {
847 		dm_sync_table(md);
848 		dm_table_destroy(t);
849 	}
850 
851 	if (!dm_kobject_uevent(md, KOBJ_REMOVE, param->event_nr))
852 		param->flags |= DM_UEVENT_GENERATED_FLAG;
853 
854 	dm_put(md);
855 	dm_destroy(md);
856 	return 0;
857 }
858 
859 /*
860  * Check a string doesn't overrun the chunk of
861  * memory we copied from userland.
862  */
863 static int invalid_str(char *str, void *end)
864 {
865 	while ((void *) str < end)
866 		if (!*str++)
867 			return 0;
868 
869 	return -EINVAL;
870 }
871 
872 static int dev_rename(struct dm_ioctl *param, size_t param_size)
873 {
874 	int r;
875 	char *new_data = (char *) param + param->data_start;
876 	struct mapped_device *md;
877 	unsigned change_uuid = (param->flags & DM_UUID_FLAG) ? 1 : 0;
878 
879 	if (new_data < param->data ||
880 	    invalid_str(new_data, (void *) param + param_size) || !*new_data ||
881 	    strlen(new_data) > (change_uuid ? DM_UUID_LEN - 1 : DM_NAME_LEN - 1)) {
882 		DMWARN("Invalid new mapped device name or uuid string supplied.");
883 		return -EINVAL;
884 	}
885 
886 	if (!change_uuid) {
887 		r = check_name(new_data);
888 		if (r)
889 			return r;
890 	}
891 
892 	md = dm_hash_rename(param, new_data);
893 	if (IS_ERR(md))
894 		return PTR_ERR(md);
895 
896 	__dev_status(md, param);
897 	dm_put(md);
898 
899 	return 0;
900 }
901 
902 static int dev_set_geometry(struct dm_ioctl *param, size_t param_size)
903 {
904 	int r = -EINVAL, x;
905 	struct mapped_device *md;
906 	struct hd_geometry geometry;
907 	unsigned long indata[4];
908 	char *geostr = (char *) param + param->data_start;
909 	char dummy;
910 
911 	md = find_device(param);
912 	if (!md)
913 		return -ENXIO;
914 
915 	if (geostr < param->data ||
916 	    invalid_str(geostr, (void *) param + param_size)) {
917 		DMWARN("Invalid geometry supplied.");
918 		goto out;
919 	}
920 
921 	x = sscanf(geostr, "%lu %lu %lu %lu%c", indata,
922 		   indata + 1, indata + 2, indata + 3, &dummy);
923 
924 	if (x != 4) {
925 		DMWARN("Unable to interpret geometry settings.");
926 		goto out;
927 	}
928 
929 	if (indata[0] > 65535 || indata[1] > 255 ||
930 	    indata[2] > 255 || indata[3] > ULONG_MAX) {
931 		DMWARN("Geometry exceeds range limits.");
932 		goto out;
933 	}
934 
935 	geometry.cylinders = indata[0];
936 	geometry.heads = indata[1];
937 	geometry.sectors = indata[2];
938 	geometry.start = indata[3];
939 
940 	r = dm_set_geometry(md, &geometry);
941 
942 	param->data_size = 0;
943 
944 out:
945 	dm_put(md);
946 	return r;
947 }
948 
949 static int do_suspend(struct dm_ioctl *param)
950 {
951 	int r = 0;
952 	unsigned suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
953 	struct mapped_device *md;
954 
955 	md = find_device(param);
956 	if (!md)
957 		return -ENXIO;
958 
959 	if (param->flags & DM_SKIP_LOCKFS_FLAG)
960 		suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
961 	if (param->flags & DM_NOFLUSH_FLAG)
962 		suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
963 
964 	if (!dm_suspended_md(md)) {
965 		r = dm_suspend(md, suspend_flags);
966 		if (r)
967 			goto out;
968 	}
969 
970 	__dev_status(md, param);
971 
972 out:
973 	dm_put(md);
974 
975 	return r;
976 }
977 
978 static int do_resume(struct dm_ioctl *param)
979 {
980 	int r = 0;
981 	unsigned suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
982 	struct hash_cell *hc;
983 	struct mapped_device *md;
984 	struct dm_table *new_map, *old_map = NULL;
985 
986 	down_write(&_hash_lock);
987 
988 	hc = __find_device_hash_cell(param);
989 	if (!hc) {
990 		DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
991 		up_write(&_hash_lock);
992 		return -ENXIO;
993 	}
994 
995 	md = hc->md;
996 
997 	new_map = hc->new_map;
998 	hc->new_map = NULL;
999 	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1000 
1001 	up_write(&_hash_lock);
1002 
1003 	/* Do we need to load a new map ? */
1004 	if (new_map) {
1005 		/* Suspend if it isn't already suspended */
1006 		if (param->flags & DM_SKIP_LOCKFS_FLAG)
1007 			suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
1008 		if (param->flags & DM_NOFLUSH_FLAG)
1009 			suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
1010 		if (!dm_suspended_md(md))
1011 			dm_suspend(md, suspend_flags);
1012 
1013 		old_map = dm_swap_table(md, new_map);
1014 		if (IS_ERR(old_map)) {
1015 			dm_sync_table(md);
1016 			dm_table_destroy(new_map);
1017 			dm_put(md);
1018 			return PTR_ERR(old_map);
1019 		}
1020 
1021 		if (dm_table_get_mode(new_map) & FMODE_WRITE)
1022 			set_disk_ro(dm_disk(md), 0);
1023 		else
1024 			set_disk_ro(dm_disk(md), 1);
1025 	}
1026 
1027 	if (dm_suspended_md(md)) {
1028 		r = dm_resume(md);
1029 		if (!r && !dm_kobject_uevent(md, KOBJ_CHANGE, param->event_nr))
1030 			param->flags |= DM_UEVENT_GENERATED_FLAG;
1031 	}
1032 
1033 	/*
1034 	 * Since dm_swap_table synchronizes RCU, nobody should be in
1035 	 * read-side critical section already.
1036 	 */
1037 	if (old_map)
1038 		dm_table_destroy(old_map);
1039 
1040 	if (!r)
1041 		__dev_status(md, param);
1042 
1043 	dm_put(md);
1044 	return r;
1045 }
1046 
1047 /*
1048  * Set or unset the suspension state of a device.
1049  * If the device already is in the requested state we just return its status.
1050  */
1051 static int dev_suspend(struct dm_ioctl *param, size_t param_size)
1052 {
1053 	if (param->flags & DM_SUSPEND_FLAG)
1054 		return do_suspend(param);
1055 
1056 	return do_resume(param);
1057 }
1058 
1059 /*
1060  * Copies device info back to user space, used by
1061  * the create and info ioctls.
1062  */
1063 static int dev_status(struct dm_ioctl *param, size_t param_size)
1064 {
1065 	struct mapped_device *md;
1066 
1067 	md = find_device(param);
1068 	if (!md)
1069 		return -ENXIO;
1070 
1071 	__dev_status(md, param);
1072 	dm_put(md);
1073 
1074 	return 0;
1075 }
1076 
1077 /*
1078  * Build up the status struct for each target
1079  */
1080 static void retrieve_status(struct dm_table *table,
1081 			    struct dm_ioctl *param, size_t param_size)
1082 {
1083 	unsigned int i, num_targets;
1084 	struct dm_target_spec *spec;
1085 	char *outbuf, *outptr;
1086 	status_type_t type;
1087 	size_t remaining, len, used = 0;
1088 	unsigned status_flags = 0;
1089 
1090 	outptr = outbuf = get_result_buffer(param, param_size, &len);
1091 
1092 	if (param->flags & DM_STATUS_TABLE_FLAG)
1093 		type = STATUSTYPE_TABLE;
1094 	else
1095 		type = STATUSTYPE_INFO;
1096 
1097 	/* Get all the target info */
1098 	num_targets = dm_table_get_num_targets(table);
1099 	for (i = 0; i < num_targets; i++) {
1100 		struct dm_target *ti = dm_table_get_target(table, i);
1101 		size_t l;
1102 
1103 		remaining = len - (outptr - outbuf);
1104 		if (remaining <= sizeof(struct dm_target_spec)) {
1105 			param->flags |= DM_BUFFER_FULL_FLAG;
1106 			break;
1107 		}
1108 
1109 		spec = (struct dm_target_spec *) outptr;
1110 
1111 		spec->status = 0;
1112 		spec->sector_start = ti->begin;
1113 		spec->length = ti->len;
1114 		strncpy(spec->target_type, ti->type->name,
1115 			sizeof(spec->target_type));
1116 
1117 		outptr += sizeof(struct dm_target_spec);
1118 		remaining = len - (outptr - outbuf);
1119 		if (remaining <= 0) {
1120 			param->flags |= DM_BUFFER_FULL_FLAG;
1121 			break;
1122 		}
1123 
1124 		/* Get the status/table string from the target driver */
1125 		if (ti->type->status) {
1126 			if (param->flags & DM_NOFLUSH_FLAG)
1127 				status_flags |= DM_STATUS_NOFLUSH_FLAG;
1128 			ti->type->status(ti, type, status_flags, outptr, remaining);
1129 		} else
1130 			outptr[0] = '\0';
1131 
1132 		l = strlen(outptr) + 1;
1133 		if (l == remaining) {
1134 			param->flags |= DM_BUFFER_FULL_FLAG;
1135 			break;
1136 		}
1137 
1138 		outptr += l;
1139 		used = param->data_start + (outptr - outbuf);
1140 
1141 		outptr = align_ptr(outptr);
1142 		spec->next = outptr - outbuf;
1143 	}
1144 
1145 	if (used)
1146 		param->data_size = used;
1147 
1148 	param->target_count = num_targets;
1149 }
1150 
1151 /*
1152  * Wait for a device to report an event
1153  */
1154 static int dev_wait(struct dm_ioctl *param, size_t param_size)
1155 {
1156 	int r = 0;
1157 	struct mapped_device *md;
1158 	struct dm_table *table;
1159 	int srcu_idx;
1160 
1161 	md = find_device(param);
1162 	if (!md)
1163 		return -ENXIO;
1164 
1165 	/*
1166 	 * Wait for a notification event
1167 	 */
1168 	if (dm_wait_event(md, param->event_nr)) {
1169 		r = -ERESTARTSYS;
1170 		goto out;
1171 	}
1172 
1173 	/*
1174 	 * The userland program is going to want to know what
1175 	 * changed to trigger the event, so we may as well tell
1176 	 * him and save an ioctl.
1177 	 */
1178 	__dev_status(md, param);
1179 
1180 	table = dm_get_live_or_inactive_table(md, param, &srcu_idx);
1181 	if (table)
1182 		retrieve_status(table, param, param_size);
1183 	dm_put_live_table(md, srcu_idx);
1184 
1185 out:
1186 	dm_put(md);
1187 
1188 	return r;
1189 }
1190 
1191 static inline fmode_t get_mode(struct dm_ioctl *param)
1192 {
1193 	fmode_t mode = FMODE_READ | FMODE_WRITE;
1194 
1195 	if (param->flags & DM_READONLY_FLAG)
1196 		mode = FMODE_READ;
1197 
1198 	return mode;
1199 }
1200 
1201 static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
1202 		       struct dm_target_spec **spec, char **target_params)
1203 {
1204 	*spec = (struct dm_target_spec *) ((unsigned char *) last + next);
1205 	*target_params = (char *) (*spec + 1);
1206 
1207 	if (*spec < (last + 1))
1208 		return -EINVAL;
1209 
1210 	return invalid_str(*target_params, end);
1211 }
1212 
1213 static int populate_table(struct dm_table *table,
1214 			  struct dm_ioctl *param, size_t param_size)
1215 {
1216 	int r;
1217 	unsigned int i = 0;
1218 	struct dm_target_spec *spec = (struct dm_target_spec *) param;
1219 	uint32_t next = param->data_start;
1220 	void *end = (void *) param + param_size;
1221 	char *target_params;
1222 
1223 	if (!param->target_count) {
1224 		DMWARN("populate_table: no targets specified");
1225 		return -EINVAL;
1226 	}
1227 
1228 	for (i = 0; i < param->target_count; i++) {
1229 
1230 		r = next_target(spec, next, end, &spec, &target_params);
1231 		if (r) {
1232 			DMWARN("unable to find target");
1233 			return r;
1234 		}
1235 
1236 		r = dm_table_add_target(table, spec->target_type,
1237 					(sector_t) spec->sector_start,
1238 					(sector_t) spec->length,
1239 					target_params);
1240 		if (r) {
1241 			DMWARN("error adding target to table");
1242 			return r;
1243 		}
1244 
1245 		next = spec->next;
1246 	}
1247 
1248 	return dm_table_complete(table);
1249 }
1250 
1251 static int table_load(struct dm_ioctl *param, size_t param_size)
1252 {
1253 	int r;
1254 	struct hash_cell *hc;
1255 	struct dm_table *t, *old_map = NULL;
1256 	struct mapped_device *md;
1257 	struct target_type *immutable_target_type;
1258 
1259 	md = find_device(param);
1260 	if (!md)
1261 		return -ENXIO;
1262 
1263 	r = dm_table_create(&t, get_mode(param), param->target_count, md);
1264 	if (r)
1265 		goto err;
1266 
1267 	/* Protect md->type and md->queue against concurrent table loads. */
1268 	dm_lock_md_type(md);
1269 	r = populate_table(t, param, param_size);
1270 	if (r)
1271 		goto err_unlock_md_type;
1272 
1273 	immutable_target_type = dm_get_immutable_target_type(md);
1274 	if (immutable_target_type &&
1275 	    (immutable_target_type != dm_table_get_immutable_target_type(t))) {
1276 		DMWARN("can't replace immutable target type %s",
1277 		       immutable_target_type->name);
1278 		r = -EINVAL;
1279 		goto err_unlock_md_type;
1280 	}
1281 
1282 	if (dm_get_md_type(md) == DM_TYPE_NONE)
1283 		/* Initial table load: acquire type of table. */
1284 		dm_set_md_type(md, dm_table_get_type(t));
1285 	else if (dm_get_md_type(md) != dm_table_get_type(t)) {
1286 		DMWARN("can't change device type after initial table load.");
1287 		r = -EINVAL;
1288 		goto err_unlock_md_type;
1289 	}
1290 
1291 	/* setup md->queue to reflect md's type (may block) */
1292 	r = dm_setup_md_queue(md);
1293 	if (r) {
1294 		DMWARN("unable to set up device queue for new table.");
1295 		goto err_unlock_md_type;
1296 	}
1297 	dm_unlock_md_type(md);
1298 
1299 	/* stage inactive table */
1300 	down_write(&_hash_lock);
1301 	hc = dm_get_mdptr(md);
1302 	if (!hc || hc->md != md) {
1303 		DMWARN("device has been removed from the dev hash table.");
1304 		up_write(&_hash_lock);
1305 		r = -ENXIO;
1306 		goto err_destroy_table;
1307 	}
1308 
1309 	if (hc->new_map)
1310 		old_map = hc->new_map;
1311 	hc->new_map = t;
1312 	up_write(&_hash_lock);
1313 
1314 	param->flags |= DM_INACTIVE_PRESENT_FLAG;
1315 	__dev_status(md, param);
1316 
1317 	if (old_map) {
1318 		dm_sync_table(md);
1319 		dm_table_destroy(old_map);
1320 	}
1321 
1322 	dm_put(md);
1323 
1324 	return 0;
1325 
1326 err_unlock_md_type:
1327 	dm_unlock_md_type(md);
1328 err_destroy_table:
1329 	dm_table_destroy(t);
1330 err:
1331 	dm_put(md);
1332 
1333 	return r;
1334 }
1335 
1336 static int table_clear(struct dm_ioctl *param, size_t param_size)
1337 {
1338 	struct hash_cell *hc;
1339 	struct mapped_device *md;
1340 	struct dm_table *old_map = NULL;
1341 
1342 	down_write(&_hash_lock);
1343 
1344 	hc = __find_device_hash_cell(param);
1345 	if (!hc) {
1346 		DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
1347 		up_write(&_hash_lock);
1348 		return -ENXIO;
1349 	}
1350 
1351 	if (hc->new_map) {
1352 		old_map = hc->new_map;
1353 		hc->new_map = NULL;
1354 	}
1355 
1356 	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1357 
1358 	__dev_status(hc->md, param);
1359 	md = hc->md;
1360 	up_write(&_hash_lock);
1361 	if (old_map) {
1362 		dm_sync_table(md);
1363 		dm_table_destroy(old_map);
1364 	}
1365 	dm_put(md);
1366 
1367 	return 0;
1368 }
1369 
1370 /*
1371  * Retrieves a list of devices used by a particular dm device.
1372  */
1373 static void retrieve_deps(struct dm_table *table,
1374 			  struct dm_ioctl *param, size_t param_size)
1375 {
1376 	unsigned int count = 0;
1377 	struct list_head *tmp;
1378 	size_t len, needed;
1379 	struct dm_dev_internal *dd;
1380 	struct dm_target_deps *deps;
1381 
1382 	deps = get_result_buffer(param, param_size, &len);
1383 
1384 	/*
1385 	 * Count the devices.
1386 	 */
1387 	list_for_each (tmp, dm_table_get_devices(table))
1388 		count++;
1389 
1390 	/*
1391 	 * Check we have enough space.
1392 	 */
1393 	needed = sizeof(*deps) + (sizeof(*deps->dev) * count);
1394 	if (len < needed) {
1395 		param->flags |= DM_BUFFER_FULL_FLAG;
1396 		return;
1397 	}
1398 
1399 	/*
1400 	 * Fill in the devices.
1401 	 */
1402 	deps->count = count;
1403 	count = 0;
1404 	list_for_each_entry (dd, dm_table_get_devices(table), list)
1405 		deps->dev[count++] = huge_encode_dev(dd->dm_dev.bdev->bd_dev);
1406 
1407 	param->data_size = param->data_start + needed;
1408 }
1409 
1410 static int table_deps(struct dm_ioctl *param, size_t param_size)
1411 {
1412 	struct mapped_device *md;
1413 	struct dm_table *table;
1414 	int srcu_idx;
1415 
1416 	md = find_device(param);
1417 	if (!md)
1418 		return -ENXIO;
1419 
1420 	__dev_status(md, param);
1421 
1422 	table = dm_get_live_or_inactive_table(md, param, &srcu_idx);
1423 	if (table)
1424 		retrieve_deps(table, param, param_size);
1425 	dm_put_live_table(md, srcu_idx);
1426 
1427 	dm_put(md);
1428 
1429 	return 0;
1430 }
1431 
1432 /*
1433  * Return the status of a device as a text string for each
1434  * target.
1435  */
1436 static int table_status(struct dm_ioctl *param, size_t param_size)
1437 {
1438 	struct mapped_device *md;
1439 	struct dm_table *table;
1440 	int srcu_idx;
1441 
1442 	md = find_device(param);
1443 	if (!md)
1444 		return -ENXIO;
1445 
1446 	__dev_status(md, param);
1447 
1448 	table = dm_get_live_or_inactive_table(md, param, &srcu_idx);
1449 	if (table)
1450 		retrieve_status(table, param, param_size);
1451 	dm_put_live_table(md, srcu_idx);
1452 
1453 	dm_put(md);
1454 
1455 	return 0;
1456 }
1457 
1458 /*
1459  * Process device-mapper dependent messages.  Messages prefixed with '@'
1460  * are processed by the DM core.  All others are delivered to the target.
1461  * Returns a number <= 1 if message was processed by device mapper.
1462  * Returns 2 if message should be delivered to the target.
1463  */
1464 static int message_for_md(struct mapped_device *md, unsigned argc, char **argv,
1465 			  char *result, unsigned maxlen)
1466 {
1467 	int r;
1468 
1469 	if (**argv != '@')
1470 		return 2; /* no '@' prefix, deliver to target */
1471 
1472 	r = dm_stats_message(md, argc, argv, result, maxlen);
1473 	if (r < 2)
1474 		return r;
1475 
1476 	DMERR("Unsupported message sent to DM core: %s", argv[0]);
1477 	return -EINVAL;
1478 }
1479 
1480 /*
1481  * Pass a message to the target that's at the supplied device offset.
1482  */
1483 static int target_message(struct dm_ioctl *param, size_t param_size)
1484 {
1485 	int r, argc;
1486 	char **argv;
1487 	struct mapped_device *md;
1488 	struct dm_table *table;
1489 	struct dm_target *ti;
1490 	struct dm_target_msg *tmsg = (void *) param + param->data_start;
1491 	size_t maxlen;
1492 	char *result = get_result_buffer(param, param_size, &maxlen);
1493 	int srcu_idx;
1494 
1495 	md = find_device(param);
1496 	if (!md)
1497 		return -ENXIO;
1498 
1499 	if (tmsg < (struct dm_target_msg *) param->data ||
1500 	    invalid_str(tmsg->message, (void *) param + param_size)) {
1501 		DMWARN("Invalid target message parameters.");
1502 		r = -EINVAL;
1503 		goto out;
1504 	}
1505 
1506 	r = dm_split_args(&argc, &argv, tmsg->message);
1507 	if (r) {
1508 		DMWARN("Failed to split target message parameters");
1509 		goto out;
1510 	}
1511 
1512 	if (!argc) {
1513 		DMWARN("Empty message received.");
1514 		goto out_argv;
1515 	}
1516 
1517 	r = message_for_md(md, argc, argv, result, maxlen);
1518 	if (r <= 1)
1519 		goto out_argv;
1520 
1521 	table = dm_get_live_table(md, &srcu_idx);
1522 	if (!table)
1523 		goto out_table;
1524 
1525 	if (dm_deleting_md(md)) {
1526 		r = -ENXIO;
1527 		goto out_table;
1528 	}
1529 
1530 	ti = dm_table_find_target(table, tmsg->sector);
1531 	if (!dm_target_is_valid(ti)) {
1532 		DMWARN("Target message sector outside device.");
1533 		r = -EINVAL;
1534 	} else if (ti->type->message)
1535 		r = ti->type->message(ti, argc, argv);
1536 	else {
1537 		DMWARN("Target type does not support messages");
1538 		r = -EINVAL;
1539 	}
1540 
1541  out_table:
1542 	dm_put_live_table(md, srcu_idx);
1543  out_argv:
1544 	kfree(argv);
1545  out:
1546 	if (r >= 0)
1547 		__dev_status(md, param);
1548 
1549 	if (r == 1) {
1550 		param->flags |= DM_DATA_OUT_FLAG;
1551 		if (dm_message_test_buffer_overflow(result, maxlen))
1552 			param->flags |= DM_BUFFER_FULL_FLAG;
1553 		else
1554 			param->data_size = param->data_start + strlen(result) + 1;
1555 		r = 0;
1556 	}
1557 
1558 	dm_put(md);
1559 	return r;
1560 }
1561 
1562 /*
1563  * The ioctl parameter block consists of two parts, a dm_ioctl struct
1564  * followed by a data buffer.  This flag is set if the second part,
1565  * which has a variable size, is not used by the function processing
1566  * the ioctl.
1567  */
1568 #define IOCTL_FLAGS_NO_PARAMS	1
1569 
1570 /*-----------------------------------------------------------------
1571  * Implementation of open/close/ioctl on the special char
1572  * device.
1573  *---------------------------------------------------------------*/
1574 static ioctl_fn lookup_ioctl(unsigned int cmd, int *ioctl_flags)
1575 {
1576 	static struct {
1577 		int cmd;
1578 		int flags;
1579 		ioctl_fn fn;
1580 	} _ioctls[] = {
1581 		{DM_VERSION_CMD, 0, NULL}, /* version is dealt with elsewhere */
1582 		{DM_REMOVE_ALL_CMD, IOCTL_FLAGS_NO_PARAMS, remove_all},
1583 		{DM_LIST_DEVICES_CMD, 0, list_devices},
1584 
1585 		{DM_DEV_CREATE_CMD, IOCTL_FLAGS_NO_PARAMS, dev_create},
1586 		{DM_DEV_REMOVE_CMD, IOCTL_FLAGS_NO_PARAMS, dev_remove},
1587 		{DM_DEV_RENAME_CMD, 0, dev_rename},
1588 		{DM_DEV_SUSPEND_CMD, IOCTL_FLAGS_NO_PARAMS, dev_suspend},
1589 		{DM_DEV_STATUS_CMD, IOCTL_FLAGS_NO_PARAMS, dev_status},
1590 		{DM_DEV_WAIT_CMD, 0, dev_wait},
1591 
1592 		{DM_TABLE_LOAD_CMD, 0, table_load},
1593 		{DM_TABLE_CLEAR_CMD, IOCTL_FLAGS_NO_PARAMS, table_clear},
1594 		{DM_TABLE_DEPS_CMD, 0, table_deps},
1595 		{DM_TABLE_STATUS_CMD, 0, table_status},
1596 
1597 		{DM_LIST_VERSIONS_CMD, 0, list_versions},
1598 
1599 		{DM_TARGET_MSG_CMD, 0, target_message},
1600 		{DM_DEV_SET_GEOMETRY_CMD, 0, dev_set_geometry}
1601 	};
1602 
1603 	if (unlikely(cmd >= ARRAY_SIZE(_ioctls)))
1604 		return NULL;
1605 
1606 	*ioctl_flags = _ioctls[cmd].flags;
1607 	return _ioctls[cmd].fn;
1608 }
1609 
1610 /*
1611  * As well as checking the version compatibility this always
1612  * copies the kernel interface version out.
1613  */
1614 static int check_version(unsigned int cmd, struct dm_ioctl __user *user)
1615 {
1616 	uint32_t version[3];
1617 	int r = 0;
1618 
1619 	if (copy_from_user(version, user->version, sizeof(version)))
1620 		return -EFAULT;
1621 
1622 	if ((DM_VERSION_MAJOR != version[0]) ||
1623 	    (DM_VERSION_MINOR < version[1])) {
1624 		DMWARN("ioctl interface mismatch: "
1625 		       "kernel(%u.%u.%u), user(%u.%u.%u), cmd(%d)",
1626 		       DM_VERSION_MAJOR, DM_VERSION_MINOR,
1627 		       DM_VERSION_PATCHLEVEL,
1628 		       version[0], version[1], version[2], cmd);
1629 		r = -EINVAL;
1630 	}
1631 
1632 	/*
1633 	 * Fill in the kernel version.
1634 	 */
1635 	version[0] = DM_VERSION_MAJOR;
1636 	version[1] = DM_VERSION_MINOR;
1637 	version[2] = DM_VERSION_PATCHLEVEL;
1638 	if (copy_to_user(user->version, version, sizeof(version)))
1639 		return -EFAULT;
1640 
1641 	return r;
1642 }
1643 
1644 #define DM_PARAMS_KMALLOC	0x0001	/* Params alloced with kmalloc */
1645 #define DM_PARAMS_VMALLOC	0x0002	/* Params alloced with vmalloc */
1646 #define DM_WIPE_BUFFER		0x0010	/* Wipe input buffer before returning from ioctl */
1647 
1648 static void free_params(struct dm_ioctl *param, size_t param_size, int param_flags)
1649 {
1650 	if (param_flags & DM_WIPE_BUFFER)
1651 		memset(param, 0, param_size);
1652 
1653 	if (param_flags & DM_PARAMS_KMALLOC)
1654 		kfree(param);
1655 	if (param_flags & DM_PARAMS_VMALLOC)
1656 		vfree(param);
1657 }
1658 
1659 static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl *param_kernel,
1660 		       int ioctl_flags,
1661 		       struct dm_ioctl **param, int *param_flags)
1662 {
1663 	struct dm_ioctl *dmi;
1664 	int secure_data;
1665 	const size_t minimum_data_size = sizeof(*param_kernel) - sizeof(param_kernel->data);
1666 
1667 	if (copy_from_user(param_kernel, user, minimum_data_size))
1668 		return -EFAULT;
1669 
1670 	if (param_kernel->data_size < minimum_data_size)
1671 		return -EINVAL;
1672 
1673 	secure_data = param_kernel->flags & DM_SECURE_DATA_FLAG;
1674 
1675 	*param_flags = secure_data ? DM_WIPE_BUFFER : 0;
1676 
1677 	if (ioctl_flags & IOCTL_FLAGS_NO_PARAMS) {
1678 		dmi = param_kernel;
1679 		dmi->data_size = minimum_data_size;
1680 		goto data_copied;
1681 	}
1682 
1683 	/*
1684 	 * Try to avoid low memory issues when a device is suspended.
1685 	 * Use kmalloc() rather than vmalloc() when we can.
1686 	 */
1687 	dmi = NULL;
1688 	if (param_kernel->data_size <= KMALLOC_MAX_SIZE) {
1689 		dmi = kmalloc(param_kernel->data_size, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
1690 		if (dmi)
1691 			*param_flags |= DM_PARAMS_KMALLOC;
1692 	}
1693 
1694 	if (!dmi) {
1695 		unsigned noio_flag;
1696 		noio_flag = memalloc_noio_save();
1697 		dmi = __vmalloc(param_kernel->data_size, GFP_NOIO | __GFP_REPEAT | __GFP_HIGH | __GFP_HIGHMEM, PAGE_KERNEL);
1698 		memalloc_noio_restore(noio_flag);
1699 		if (dmi)
1700 			*param_flags |= DM_PARAMS_VMALLOC;
1701 	}
1702 
1703 	if (!dmi) {
1704 		if (secure_data && clear_user(user, param_kernel->data_size))
1705 			return -EFAULT;
1706 		return -ENOMEM;
1707 	}
1708 
1709 	if (copy_from_user(dmi, user, param_kernel->data_size))
1710 		goto bad;
1711 
1712 data_copied:
1713 	/*
1714 	 * Abort if something changed the ioctl data while it was being copied.
1715 	 */
1716 	if (dmi->data_size != param_kernel->data_size) {
1717 		DMERR("rejecting ioctl: data size modified while processing parameters");
1718 		goto bad;
1719 	}
1720 
1721 	/* Wipe the user buffer so we do not return it to userspace */
1722 	if (secure_data && clear_user(user, param_kernel->data_size))
1723 		goto bad;
1724 
1725 	*param = dmi;
1726 	return 0;
1727 
1728 bad:
1729 	free_params(dmi, param_kernel->data_size, *param_flags);
1730 
1731 	return -EFAULT;
1732 }
1733 
1734 static int validate_params(uint cmd, struct dm_ioctl *param)
1735 {
1736 	/* Always clear this flag */
1737 	param->flags &= ~DM_BUFFER_FULL_FLAG;
1738 	param->flags &= ~DM_UEVENT_GENERATED_FLAG;
1739 	param->flags &= ~DM_SECURE_DATA_FLAG;
1740 	param->flags &= ~DM_DATA_OUT_FLAG;
1741 
1742 	/* Ignores parameters */
1743 	if (cmd == DM_REMOVE_ALL_CMD ||
1744 	    cmd == DM_LIST_DEVICES_CMD ||
1745 	    cmd == DM_LIST_VERSIONS_CMD)
1746 		return 0;
1747 
1748 	if ((cmd == DM_DEV_CREATE_CMD)) {
1749 		if (!*param->name) {
1750 			DMWARN("name not supplied when creating device");
1751 			return -EINVAL;
1752 		}
1753 	} else if ((*param->uuid && *param->name)) {
1754 		DMWARN("only supply one of name or uuid, cmd(%u)", cmd);
1755 		return -EINVAL;
1756 	}
1757 
1758 	/* Ensure strings are terminated */
1759 	param->name[DM_NAME_LEN - 1] = '\0';
1760 	param->uuid[DM_UUID_LEN - 1] = '\0';
1761 
1762 	return 0;
1763 }
1764 
1765 static int ctl_ioctl(uint command, struct dm_ioctl __user *user)
1766 {
1767 	int r = 0;
1768 	int ioctl_flags;
1769 	int param_flags;
1770 	unsigned int cmd;
1771 	struct dm_ioctl *uninitialized_var(param);
1772 	ioctl_fn fn = NULL;
1773 	size_t input_param_size;
1774 	struct dm_ioctl param_kernel;
1775 
1776 	/* only root can play with this */
1777 	if (!capable(CAP_SYS_ADMIN))
1778 		return -EACCES;
1779 
1780 	if (_IOC_TYPE(command) != DM_IOCTL)
1781 		return -ENOTTY;
1782 
1783 	cmd = _IOC_NR(command);
1784 
1785 	/*
1786 	 * Check the interface version passed in.  This also
1787 	 * writes out the kernel's interface version.
1788 	 */
1789 	r = check_version(cmd, user);
1790 	if (r)
1791 		return r;
1792 
1793 	/*
1794 	 * Nothing more to do for the version command.
1795 	 */
1796 	if (cmd == DM_VERSION_CMD)
1797 		return 0;
1798 
1799 	fn = lookup_ioctl(cmd, &ioctl_flags);
1800 	if (!fn) {
1801 		DMWARN("dm_ctl_ioctl: unknown command 0x%x", command);
1802 		return -ENOTTY;
1803 	}
1804 
1805 	/*
1806 	 * Copy the parameters into kernel space.
1807 	 */
1808 	r = copy_params(user, &param_kernel, ioctl_flags, &param, &param_flags);
1809 
1810 	if (r)
1811 		return r;
1812 
1813 	input_param_size = param->data_size;
1814 	r = validate_params(cmd, param);
1815 	if (r)
1816 		goto out;
1817 
1818 	param->data_size = sizeof(*param);
1819 	r = fn(param, input_param_size);
1820 
1821 	if (unlikely(param->flags & DM_BUFFER_FULL_FLAG) &&
1822 	    unlikely(ioctl_flags & IOCTL_FLAGS_NO_PARAMS))
1823 		DMERR("ioctl %d tried to output some data but has IOCTL_FLAGS_NO_PARAMS set", cmd);
1824 
1825 	/*
1826 	 * Copy the results back to userland.
1827 	 */
1828 	if (!r && copy_to_user(user, param, param->data_size))
1829 		r = -EFAULT;
1830 
1831 out:
1832 	free_params(param, input_param_size, param_flags);
1833 	return r;
1834 }
1835 
1836 static long dm_ctl_ioctl(struct file *file, uint command, ulong u)
1837 {
1838 	return (long)ctl_ioctl(command, (struct dm_ioctl __user *)u);
1839 }
1840 
1841 #ifdef CONFIG_COMPAT
1842 static long dm_compat_ctl_ioctl(struct file *file, uint command, ulong u)
1843 {
1844 	return (long)dm_ctl_ioctl(file, command, (ulong) compat_ptr(u));
1845 }
1846 #else
1847 #define dm_compat_ctl_ioctl NULL
1848 #endif
1849 
1850 static const struct file_operations _ctl_fops = {
1851 	.open = nonseekable_open,
1852 	.unlocked_ioctl	 = dm_ctl_ioctl,
1853 	.compat_ioctl = dm_compat_ctl_ioctl,
1854 	.owner	 = THIS_MODULE,
1855 	.llseek  = noop_llseek,
1856 };
1857 
1858 static struct miscdevice _dm_misc = {
1859 	.minor		= MAPPER_CTRL_MINOR,
1860 	.name  		= DM_NAME,
1861 	.nodename	= DM_DIR "/" DM_CONTROL_NODE,
1862 	.fops  		= &_ctl_fops
1863 };
1864 
1865 MODULE_ALIAS_MISCDEV(MAPPER_CTRL_MINOR);
1866 MODULE_ALIAS("devname:" DM_DIR "/" DM_CONTROL_NODE);
1867 
1868 /*
1869  * Create misc character device and link to DM_DIR/control.
1870  */
1871 int __init dm_interface_init(void)
1872 {
1873 	int r;
1874 
1875 	r = dm_hash_init();
1876 	if (r)
1877 		return r;
1878 
1879 	r = misc_register(&_dm_misc);
1880 	if (r) {
1881 		DMERR("misc_register failed for control device");
1882 		dm_hash_exit();
1883 		return r;
1884 	}
1885 
1886 	DMINFO("%d.%d.%d%s initialised: %s", DM_VERSION_MAJOR,
1887 	       DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL, DM_VERSION_EXTRA,
1888 	       DM_DRIVER_EMAIL);
1889 	return 0;
1890 }
1891 
1892 void dm_interface_exit(void)
1893 {
1894 	if (misc_deregister(&_dm_misc) < 0)
1895 		DMERR("misc_deregister failed for control device");
1896 
1897 	dm_hash_exit();
1898 }
1899 
1900 /**
1901  * dm_copy_name_and_uuid - Copy mapped device name & uuid into supplied buffers
1902  * @md: Pointer to mapped_device
1903  * @name: Buffer (size DM_NAME_LEN) for name
1904  * @uuid: Buffer (size DM_UUID_LEN) for uuid or empty string if uuid not defined
1905  */
1906 int dm_copy_name_and_uuid(struct mapped_device *md, char *name, char *uuid)
1907 {
1908 	int r = 0;
1909 	struct hash_cell *hc;
1910 
1911 	if (!md)
1912 		return -ENXIO;
1913 
1914 	mutex_lock(&dm_hash_cells_mutex);
1915 	hc = dm_get_mdptr(md);
1916 	if (!hc || hc->md != md) {
1917 		r = -ENXIO;
1918 		goto out;
1919 	}
1920 
1921 	if (name)
1922 		strcpy(name, hc->name);
1923 	if (uuid)
1924 		strcpy(uuid, hc->uuid ? : "");
1925 
1926 out:
1927 	mutex_unlock(&dm_hash_cells_mutex);
1928 
1929 	return r;
1930 }
1931