xref: /linux/drivers/md/dm-ioctl.c (revision 630ae80ea1dd253609cb50cff87f3248f901aca3)
1 /*
2  * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3  * Copyright (C) 2004 - 2006 Red Hat, Inc. All rights reserved.
4  *
5  * This file is released under the GPL.
6  */
7 
8 #include "dm-core.h"
9 #include "dm-ima.h"
10 #include <linux/module.h>
11 #include <linux/vmalloc.h>
12 #include <linux/miscdevice.h>
13 #include <linux/sched/mm.h>
14 #include <linux/init.h>
15 #include <linux/wait.h>
16 #include <linux/slab.h>
17 #include <linux/rbtree.h>
18 #include <linux/dm-ioctl.h>
19 #include <linux/hdreg.h>
20 #include <linux/compat.h>
21 #include <linux/nospec.h>
22 
23 #include <linux/uaccess.h>
24 #include <linux/ima.h>
25 
26 #define DM_MSG_PREFIX "ioctl"
27 #define DM_DRIVER_EMAIL "dm-devel@redhat.com"
28 
29 struct dm_file {
30 	/*
31 	 * poll will wait until the global event number is greater than
32 	 * this value.
33 	 */
34 	volatile unsigned global_event_nr;
35 };
36 
37 /*-----------------------------------------------------------------
38  * The ioctl interface needs to be able to look up devices by
39  * name or uuid.
40  *---------------------------------------------------------------*/
41 struct hash_cell {
42 	struct rb_node name_node;
43 	struct rb_node uuid_node;
44 	bool name_set;
45 	bool uuid_set;
46 
47 	char *name;
48 	char *uuid;
49 	struct mapped_device *md;
50 	struct dm_table *new_map;
51 };
52 
53 struct vers_iter {
54     size_t param_size;
55     struct dm_target_versions *vers, *old_vers;
56     char *end;
57     uint32_t flags;
58 };
59 
60 
61 static struct rb_root name_rb_tree = RB_ROOT;
62 static struct rb_root uuid_rb_tree = RB_ROOT;
63 
64 static void dm_hash_remove_all(bool keep_open_devices, bool mark_deferred, bool only_deferred);
65 
66 /*
67  * Guards access to both hash tables.
68  */
69 static DECLARE_RWSEM(_hash_lock);
70 
71 /*
72  * Protects use of mdptr to obtain hash cell name and uuid from mapped device.
73  */
74 static DEFINE_MUTEX(dm_hash_cells_mutex);
75 
76 static void dm_hash_exit(void)
77 {
78 	dm_hash_remove_all(false, false, false);
79 }
80 
81 /*-----------------------------------------------------------------
82  * Code for looking up a device by name
83  *---------------------------------------------------------------*/
84 static struct hash_cell *__get_name_cell(const char *str)
85 {
86 	struct rb_node *n = name_rb_tree.rb_node;
87 
88 	while (n) {
89 		struct hash_cell *hc = container_of(n, struct hash_cell, name_node);
90 		int c = strcmp(hc->name, str);
91 		if (!c) {
92 			dm_get(hc->md);
93 			return hc;
94 		}
95 		n = c >= 0 ? n->rb_left : n->rb_right;
96 	}
97 
98 	return NULL;
99 }
100 
101 static struct hash_cell *__get_uuid_cell(const char *str)
102 {
103 	struct rb_node *n = uuid_rb_tree.rb_node;
104 
105 	while (n) {
106 		struct hash_cell *hc = container_of(n, struct hash_cell, uuid_node);
107 		int c = strcmp(hc->uuid, str);
108 		if (!c) {
109 			dm_get(hc->md);
110 			return hc;
111 		}
112 		n = c >= 0 ? n->rb_left : n->rb_right;
113 	}
114 
115 	return NULL;
116 }
117 
118 static void __unlink_name(struct hash_cell *hc)
119 {
120 	if (hc->name_set) {
121 		hc->name_set = false;
122 		rb_erase(&hc->name_node, &name_rb_tree);
123 	}
124 }
125 
126 static void __unlink_uuid(struct hash_cell *hc)
127 {
128 	if (hc->uuid_set) {
129 		hc->uuid_set = false;
130 		rb_erase(&hc->uuid_node, &uuid_rb_tree);
131 	}
132 }
133 
134 static void __link_name(struct hash_cell *new_hc)
135 {
136 	struct rb_node **n, *parent;
137 
138 	__unlink_name(new_hc);
139 
140 	new_hc->name_set = true;
141 
142 	n = &name_rb_tree.rb_node;
143 	parent = NULL;
144 
145 	while (*n) {
146 		struct hash_cell *hc = container_of(*n, struct hash_cell, name_node);
147 		int c = strcmp(hc->name, new_hc->name);
148 		BUG_ON(!c);
149 		parent = *n;
150 		n = c >= 0 ? &hc->name_node.rb_left : &hc->name_node.rb_right;
151 	}
152 
153 	rb_link_node(&new_hc->name_node, parent, n);
154 	rb_insert_color(&new_hc->name_node, &name_rb_tree);
155 }
156 
157 static void __link_uuid(struct hash_cell *new_hc)
158 {
159 	struct rb_node **n, *parent;
160 
161 	__unlink_uuid(new_hc);
162 
163 	new_hc->uuid_set = true;
164 
165 	n = &uuid_rb_tree.rb_node;
166 	parent = NULL;
167 
168 	while (*n) {
169 		struct hash_cell *hc = container_of(*n, struct hash_cell, uuid_node);
170 		int c = strcmp(hc->uuid, new_hc->uuid);
171 		BUG_ON(!c);
172 		parent = *n;
173 		n = c > 0 ? &hc->uuid_node.rb_left : &hc->uuid_node.rb_right;
174 	}
175 
176 	rb_link_node(&new_hc->uuid_node, parent, n);
177 	rb_insert_color(&new_hc->uuid_node, &uuid_rb_tree);
178 }
179 
180 static struct hash_cell *__get_dev_cell(uint64_t dev)
181 {
182 	struct mapped_device *md;
183 	struct hash_cell *hc;
184 
185 	md = dm_get_md(huge_decode_dev(dev));
186 	if (!md)
187 		return NULL;
188 
189 	hc = dm_get_mdptr(md);
190 	if (!hc) {
191 		dm_put(md);
192 		return NULL;
193 	}
194 
195 	return hc;
196 }
197 
198 /*-----------------------------------------------------------------
199  * Inserting, removing and renaming a device.
200  *---------------------------------------------------------------*/
201 static struct hash_cell *alloc_cell(const char *name, const char *uuid,
202 				    struct mapped_device *md)
203 {
204 	struct hash_cell *hc;
205 
206 	hc = kmalloc(sizeof(*hc), GFP_KERNEL);
207 	if (!hc)
208 		return NULL;
209 
210 	hc->name = kstrdup(name, GFP_KERNEL);
211 	if (!hc->name) {
212 		kfree(hc);
213 		return NULL;
214 	}
215 
216 	if (!uuid)
217 		hc->uuid = NULL;
218 
219 	else {
220 		hc->uuid = kstrdup(uuid, GFP_KERNEL);
221 		if (!hc->uuid) {
222 			kfree(hc->name);
223 			kfree(hc);
224 			return NULL;
225 		}
226 	}
227 
228 	hc->name_set = hc->uuid_set = false;
229 	hc->md = md;
230 	hc->new_map = NULL;
231 	return hc;
232 }
233 
234 static void free_cell(struct hash_cell *hc)
235 {
236 	if (hc) {
237 		kfree(hc->name);
238 		kfree(hc->uuid);
239 		kfree(hc);
240 	}
241 }
242 
243 /*
244  * The kdev_t and uuid of a device can never change once it is
245  * initially inserted.
246  */
247 static int dm_hash_insert(const char *name, const char *uuid, struct mapped_device *md)
248 {
249 	struct hash_cell *cell, *hc;
250 
251 	/*
252 	 * Allocate the new cells.
253 	 */
254 	cell = alloc_cell(name, uuid, md);
255 	if (!cell)
256 		return -ENOMEM;
257 
258 	/*
259 	 * Insert the cell into both hash tables.
260 	 */
261 	down_write(&_hash_lock);
262 	hc = __get_name_cell(name);
263 	if (hc) {
264 		dm_put(hc->md);
265 		goto bad;
266 	}
267 
268 	__link_name(cell);
269 
270 	if (uuid) {
271 		hc = __get_uuid_cell(uuid);
272 		if (hc) {
273 			__unlink_name(cell);
274 			dm_put(hc->md);
275 			goto bad;
276 		}
277 		__link_uuid(cell);
278 	}
279 	dm_get(md);
280 	mutex_lock(&dm_hash_cells_mutex);
281 	dm_set_mdptr(md, cell);
282 	mutex_unlock(&dm_hash_cells_mutex);
283 	up_write(&_hash_lock);
284 
285 	return 0;
286 
287  bad:
288 	up_write(&_hash_lock);
289 	free_cell(cell);
290 	return -EBUSY;
291 }
292 
293 static struct dm_table *__hash_remove(struct hash_cell *hc)
294 {
295 	struct dm_table *table;
296 	int srcu_idx;
297 
298 	/* remove from the dev trees */
299 	__unlink_name(hc);
300 	__unlink_uuid(hc);
301 	mutex_lock(&dm_hash_cells_mutex);
302 	dm_set_mdptr(hc->md, NULL);
303 	mutex_unlock(&dm_hash_cells_mutex);
304 
305 	table = dm_get_live_table(hc->md, &srcu_idx);
306 	if (table)
307 		dm_table_event(table);
308 	dm_put_live_table(hc->md, srcu_idx);
309 
310 	table = NULL;
311 	if (hc->new_map)
312 		table = hc->new_map;
313 	dm_put(hc->md);
314 	free_cell(hc);
315 
316 	return table;
317 }
318 
319 static void dm_hash_remove_all(bool keep_open_devices, bool mark_deferred, bool only_deferred)
320 {
321 	int dev_skipped;
322 	struct rb_node *n;
323 	struct hash_cell *hc;
324 	struct mapped_device *md;
325 	struct dm_table *t;
326 
327 retry:
328 	dev_skipped = 0;
329 
330 	down_write(&_hash_lock);
331 
332 	for (n = rb_first(&name_rb_tree); n; n = rb_next(n)) {
333 		hc = container_of(n, struct hash_cell, name_node);
334 		md = hc->md;
335 		dm_get(md);
336 
337 		if (keep_open_devices &&
338 		    dm_lock_for_deletion(md, mark_deferred, only_deferred)) {
339 			dm_put(md);
340 			dev_skipped++;
341 			continue;
342 		}
343 
344 		t = __hash_remove(hc);
345 
346 		up_write(&_hash_lock);
347 
348 		if (t) {
349 			dm_sync_table(md);
350 			dm_table_destroy(t);
351 		}
352 		dm_ima_measure_on_device_remove(md, true);
353 		dm_put(md);
354 		if (likely(keep_open_devices))
355 			dm_destroy(md);
356 		else
357 			dm_destroy_immediate(md);
358 
359 		/*
360 		 * Some mapped devices may be using other mapped
361 		 * devices, so repeat until we make no further
362 		 * progress.  If a new mapped device is created
363 		 * here it will also get removed.
364 		 */
365 		goto retry;
366 	}
367 
368 	up_write(&_hash_lock);
369 
370 	if (dev_skipped)
371 		DMWARN("remove_all left %d open device(s)", dev_skipped);
372 }
373 
374 /*
375  * Set the uuid of a hash_cell that isn't already set.
376  */
377 static void __set_cell_uuid(struct hash_cell *hc, char *new_uuid)
378 {
379 	mutex_lock(&dm_hash_cells_mutex);
380 	hc->uuid = new_uuid;
381 	mutex_unlock(&dm_hash_cells_mutex);
382 
383 	__link_uuid(hc);
384 }
385 
386 /*
387  * Changes the name of a hash_cell and returns the old name for
388  * the caller to free.
389  */
390 static char *__change_cell_name(struct hash_cell *hc, char *new_name)
391 {
392 	char *old_name;
393 
394 	/*
395 	 * Rename and move the name cell.
396 	 */
397 	__unlink_name(hc);
398 	old_name = hc->name;
399 
400 	mutex_lock(&dm_hash_cells_mutex);
401 	hc->name = new_name;
402 	mutex_unlock(&dm_hash_cells_mutex);
403 
404 	__link_name(hc);
405 
406 	return old_name;
407 }
408 
409 static struct mapped_device *dm_hash_rename(struct dm_ioctl *param,
410 					    const char *new)
411 {
412 	char *new_data, *old_name = NULL;
413 	struct hash_cell *hc;
414 	struct dm_table *table;
415 	struct mapped_device *md;
416 	unsigned change_uuid = (param->flags & DM_UUID_FLAG) ? 1 : 0;
417 	int srcu_idx;
418 
419 	/*
420 	 * duplicate new.
421 	 */
422 	new_data = kstrdup(new, GFP_KERNEL);
423 	if (!new_data)
424 		return ERR_PTR(-ENOMEM);
425 
426 	down_write(&_hash_lock);
427 
428 	/*
429 	 * Is new free ?
430 	 */
431 	if (change_uuid)
432 		hc = __get_uuid_cell(new);
433 	else
434 		hc = __get_name_cell(new);
435 
436 	if (hc) {
437 		DMERR("Unable to change %s on mapped device %s to one that "
438 		      "already exists: %s",
439 		      change_uuid ? "uuid" : "name",
440 		      param->name, new);
441 		dm_put(hc->md);
442 		up_write(&_hash_lock);
443 		kfree(new_data);
444 		return ERR_PTR(-EBUSY);
445 	}
446 
447 	/*
448 	 * Is there such a device as 'old' ?
449 	 */
450 	hc = __get_name_cell(param->name);
451 	if (!hc) {
452 		DMERR("Unable to rename non-existent device, %s to %s%s",
453 		      param->name, change_uuid ? "uuid " : "", new);
454 		up_write(&_hash_lock);
455 		kfree(new_data);
456 		return ERR_PTR(-ENXIO);
457 	}
458 
459 	/*
460 	 * Does this device already have a uuid?
461 	 */
462 	if (change_uuid && hc->uuid) {
463 		DMERR("Unable to change uuid of mapped device %s to %s "
464 		      "because uuid is already set to %s",
465 		      param->name, new, hc->uuid);
466 		dm_put(hc->md);
467 		up_write(&_hash_lock);
468 		kfree(new_data);
469 		return ERR_PTR(-EINVAL);
470 	}
471 
472 	if (change_uuid)
473 		__set_cell_uuid(hc, new_data);
474 	else
475 		old_name = __change_cell_name(hc, new_data);
476 
477 	/*
478 	 * Wake up any dm event waiters.
479 	 */
480 	table = dm_get_live_table(hc->md, &srcu_idx);
481 	if (table)
482 		dm_table_event(table);
483 	dm_put_live_table(hc->md, srcu_idx);
484 
485 	if (!dm_kobject_uevent(hc->md, KOBJ_CHANGE, param->event_nr))
486 		param->flags |= DM_UEVENT_GENERATED_FLAG;
487 
488 	md = hc->md;
489 
490 	dm_ima_measure_on_device_rename(md);
491 
492 	up_write(&_hash_lock);
493 	kfree(old_name);
494 
495 	return md;
496 }
497 
498 void dm_deferred_remove(void)
499 {
500 	dm_hash_remove_all(true, false, true);
501 }
502 
503 /*-----------------------------------------------------------------
504  * Implementation of the ioctl commands
505  *---------------------------------------------------------------*/
506 /*
507  * All the ioctl commands get dispatched to functions with this
508  * prototype.
509  */
510 typedef int (*ioctl_fn)(struct file *filp, struct dm_ioctl *param, size_t param_size);
511 
512 static int remove_all(struct file *filp, struct dm_ioctl *param, size_t param_size)
513 {
514 	dm_hash_remove_all(true, !!(param->flags & DM_DEFERRED_REMOVE), false);
515 	param->data_size = 0;
516 	return 0;
517 }
518 
519 /*
520  * Round up the ptr to an 8-byte boundary.
521  */
522 #define ALIGN_MASK 7
523 static inline size_t align_val(size_t val)
524 {
525 	return (val + ALIGN_MASK) & ~ALIGN_MASK;
526 }
527 static inline void *align_ptr(void *ptr)
528 {
529 	return (void *)align_val((size_t)ptr);
530 }
531 
532 /*
533  * Retrieves the data payload buffer from an already allocated
534  * struct dm_ioctl.
535  */
536 static void *get_result_buffer(struct dm_ioctl *param, size_t param_size,
537 			       size_t *len)
538 {
539 	param->data_start = align_ptr(param + 1) - (void *) param;
540 
541 	if (param->data_start < param_size)
542 		*len = param_size - param->data_start;
543 	else
544 		*len = 0;
545 
546 	return ((void *) param) + param->data_start;
547 }
548 
549 static bool filter_device(struct hash_cell *hc, const char *pfx_name, const char *pfx_uuid)
550 {
551 	const char *val;
552 	size_t val_len, pfx_len;
553 
554 	val = hc->name;
555 	val_len = strlen(val);
556 	pfx_len = strnlen(pfx_name, DM_NAME_LEN);
557 	if (pfx_len > val_len)
558 		return false;
559 	if (memcmp(val, pfx_name, pfx_len))
560 		return false;
561 
562 	val = hc->uuid ? hc->uuid : "";
563 	val_len = strlen(val);
564 	pfx_len = strnlen(pfx_uuid, DM_UUID_LEN);
565 	if (pfx_len > val_len)
566 		return false;
567 	if (memcmp(val, pfx_uuid, pfx_len))
568 		return false;
569 
570 	return true;
571 }
572 
573 static int list_devices(struct file *filp, struct dm_ioctl *param, size_t param_size)
574 {
575 	struct rb_node *n;
576 	struct hash_cell *hc;
577 	size_t len, needed = 0;
578 	struct gendisk *disk;
579 	struct dm_name_list *orig_nl, *nl, *old_nl = NULL;
580 	uint32_t *event_nr;
581 
582 	down_write(&_hash_lock);
583 
584 	/*
585 	 * Loop through all the devices working out how much
586 	 * space we need.
587 	 */
588 	for (n = rb_first(&name_rb_tree); n; n = rb_next(n)) {
589 		hc = container_of(n, struct hash_cell, name_node);
590 		if (!filter_device(hc, param->name, param->uuid))
591 			continue;
592 		needed += align_val(offsetof(struct dm_name_list, name) + strlen(hc->name) + 1);
593 		needed += align_val(sizeof(uint32_t) * 2);
594 		if (param->flags & DM_UUID_FLAG && hc->uuid)
595 			needed += align_val(strlen(hc->uuid) + 1);
596 	}
597 
598 	/*
599 	 * Grab our output buffer.
600 	 */
601 	nl = orig_nl = get_result_buffer(param, param_size, &len);
602 	if (len < needed || len < sizeof(nl->dev)) {
603 		param->flags |= DM_BUFFER_FULL_FLAG;
604 		goto out;
605 	}
606 	param->data_size = param->data_start + needed;
607 
608 	nl->dev = 0;	/* Flags no data */
609 
610 	/*
611 	 * Now loop through filling out the names.
612 	 */
613 	for (n = rb_first(&name_rb_tree); n; n = rb_next(n)) {
614 		void *uuid_ptr;
615 		hc = container_of(n, struct hash_cell, name_node);
616 		if (!filter_device(hc, param->name, param->uuid))
617 			continue;
618 		if (old_nl)
619 			old_nl->next = (uint32_t) ((void *) nl -
620 						   (void *) old_nl);
621 		disk = dm_disk(hc->md);
622 		nl->dev = huge_encode_dev(disk_devt(disk));
623 		nl->next = 0;
624 		strcpy(nl->name, hc->name);
625 
626 		old_nl = nl;
627 		event_nr = align_ptr(nl->name + strlen(hc->name) + 1);
628 		event_nr[0] = dm_get_event_nr(hc->md);
629 		event_nr[1] = 0;
630 		uuid_ptr = align_ptr(event_nr + 2);
631 		if (param->flags & DM_UUID_FLAG) {
632 			if (hc->uuid) {
633 				event_nr[1] |= DM_NAME_LIST_FLAG_HAS_UUID;
634 				strcpy(uuid_ptr, hc->uuid);
635 				uuid_ptr = align_ptr(uuid_ptr + strlen(hc->uuid) + 1);
636 			} else {
637 				event_nr[1] |= DM_NAME_LIST_FLAG_DOESNT_HAVE_UUID;
638 			}
639 		}
640 		nl = uuid_ptr;
641 	}
642 	/*
643 	 * If mismatch happens, security may be compromised due to buffer
644 	 * overflow, so it's better to crash.
645 	 */
646 	BUG_ON((char *)nl - (char *)orig_nl != needed);
647 
648  out:
649 	up_write(&_hash_lock);
650 	return 0;
651 }
652 
653 static void list_version_get_needed(struct target_type *tt, void *needed_param)
654 {
655     size_t *needed = needed_param;
656 
657     *needed += sizeof(struct dm_target_versions);
658     *needed += strlen(tt->name);
659     *needed += ALIGN_MASK;
660 }
661 
662 static void list_version_get_info(struct target_type *tt, void *param)
663 {
664     struct vers_iter *info = param;
665 
666     /* Check space - it might have changed since the first iteration */
667     if ((char *)info->vers + sizeof(tt->version) + strlen(tt->name) + 1 >
668 	info->end) {
669 
670 	info->flags = DM_BUFFER_FULL_FLAG;
671 	return;
672     }
673 
674     if (info->old_vers)
675 	info->old_vers->next = (uint32_t) ((void *)info->vers -
676 					   (void *)info->old_vers);
677     info->vers->version[0] = tt->version[0];
678     info->vers->version[1] = tt->version[1];
679     info->vers->version[2] = tt->version[2];
680     info->vers->next = 0;
681     strcpy(info->vers->name, tt->name);
682 
683     info->old_vers = info->vers;
684     info->vers = align_ptr(((void *) ++info->vers) + strlen(tt->name) + 1);
685 }
686 
687 static int __list_versions(struct dm_ioctl *param, size_t param_size, const char *name)
688 {
689 	size_t len, needed = 0;
690 	struct dm_target_versions *vers;
691 	struct vers_iter iter_info;
692 	struct target_type *tt = NULL;
693 
694 	if (name) {
695 		tt = dm_get_target_type(name);
696 		if (!tt)
697 			return -EINVAL;
698 	}
699 
700 	/*
701 	 * Loop through all the devices working out how much
702 	 * space we need.
703 	 */
704 	if (!tt)
705 		dm_target_iterate(list_version_get_needed, &needed);
706 	else
707 		list_version_get_needed(tt, &needed);
708 
709 	/*
710 	 * Grab our output buffer.
711 	 */
712 	vers = get_result_buffer(param, param_size, &len);
713 	if (len < needed) {
714 		param->flags |= DM_BUFFER_FULL_FLAG;
715 		goto out;
716 	}
717 	param->data_size = param->data_start + needed;
718 
719 	iter_info.param_size = param_size;
720 	iter_info.old_vers = NULL;
721 	iter_info.vers = vers;
722 	iter_info.flags = 0;
723 	iter_info.end = (char *)vers+len;
724 
725 	/*
726 	 * Now loop through filling out the names & versions.
727 	 */
728 	if (!tt)
729 		dm_target_iterate(list_version_get_info, &iter_info);
730 	else
731 		list_version_get_info(tt, &iter_info);
732 	param->flags |= iter_info.flags;
733 
734  out:
735 	if (tt)
736 		dm_put_target_type(tt);
737 	return 0;
738 }
739 
740 static int list_versions(struct file *filp, struct dm_ioctl *param, size_t param_size)
741 {
742 	return __list_versions(param, param_size, NULL);
743 }
744 
745 static int get_target_version(struct file *filp, struct dm_ioctl *param, size_t param_size)
746 {
747 	return __list_versions(param, param_size, param->name);
748 }
749 
750 static int check_name(const char *name)
751 {
752 	if (strchr(name, '/')) {
753 		DMERR("invalid device name");
754 		return -EINVAL;
755 	}
756 
757 	return 0;
758 }
759 
760 /*
761  * On successful return, the caller must not attempt to acquire
762  * _hash_lock without first calling dm_put_live_table, because dm_table_destroy
763  * waits for this dm_put_live_table and could be called under this lock.
764  */
765 static struct dm_table *dm_get_inactive_table(struct mapped_device *md, int *srcu_idx)
766 {
767 	struct hash_cell *hc;
768 	struct dm_table *table = NULL;
769 
770 	/* increment rcu count, we don't care about the table pointer */
771 	dm_get_live_table(md, srcu_idx);
772 
773 	down_read(&_hash_lock);
774 	hc = dm_get_mdptr(md);
775 	if (!hc || hc->md != md) {
776 		DMERR("device has been removed from the dev hash table.");
777 		goto out;
778 	}
779 
780 	table = hc->new_map;
781 
782 out:
783 	up_read(&_hash_lock);
784 
785 	return table;
786 }
787 
788 static struct dm_table *dm_get_live_or_inactive_table(struct mapped_device *md,
789 						      struct dm_ioctl *param,
790 						      int *srcu_idx)
791 {
792 	return (param->flags & DM_QUERY_INACTIVE_TABLE_FLAG) ?
793 		dm_get_inactive_table(md, srcu_idx) : dm_get_live_table(md, srcu_idx);
794 }
795 
796 /*
797  * Fills in a dm_ioctl structure, ready for sending back to
798  * userland.
799  */
800 static void __dev_status(struct mapped_device *md, struct dm_ioctl *param)
801 {
802 	struct gendisk *disk = dm_disk(md);
803 	struct dm_table *table;
804 	int srcu_idx;
805 
806 	param->flags &= ~(DM_SUSPEND_FLAG | DM_READONLY_FLAG |
807 			  DM_ACTIVE_PRESENT_FLAG | DM_INTERNAL_SUSPEND_FLAG);
808 
809 	if (dm_suspended_md(md))
810 		param->flags |= DM_SUSPEND_FLAG;
811 
812 	if (dm_suspended_internally_md(md))
813 		param->flags |= DM_INTERNAL_SUSPEND_FLAG;
814 
815 	if (dm_test_deferred_remove_flag(md))
816 		param->flags |= DM_DEFERRED_REMOVE;
817 
818 	param->dev = huge_encode_dev(disk_devt(disk));
819 
820 	/*
821 	 * Yes, this will be out of date by the time it gets back
822 	 * to userland, but it is still very useful for
823 	 * debugging.
824 	 */
825 	param->open_count = dm_open_count(md);
826 
827 	param->event_nr = dm_get_event_nr(md);
828 	param->target_count = 0;
829 
830 	table = dm_get_live_table(md, &srcu_idx);
831 	if (table) {
832 		if (!(param->flags & DM_QUERY_INACTIVE_TABLE_FLAG)) {
833 			if (get_disk_ro(disk))
834 				param->flags |= DM_READONLY_FLAG;
835 			param->target_count = table->num_targets;
836 		}
837 
838 		param->flags |= DM_ACTIVE_PRESENT_FLAG;
839 	}
840 	dm_put_live_table(md, srcu_idx);
841 
842 	if (param->flags & DM_QUERY_INACTIVE_TABLE_FLAG) {
843 		int srcu_idx;
844 		table = dm_get_inactive_table(md, &srcu_idx);
845 		if (table) {
846 			if (!(dm_table_get_mode(table) & FMODE_WRITE))
847 				param->flags |= DM_READONLY_FLAG;
848 			param->target_count = table->num_targets;
849 		}
850 		dm_put_live_table(md, srcu_idx);
851 	}
852 }
853 
854 static int dev_create(struct file *filp, struct dm_ioctl *param, size_t param_size)
855 {
856 	int r, m = DM_ANY_MINOR;
857 	struct mapped_device *md;
858 
859 	r = check_name(param->name);
860 	if (r)
861 		return r;
862 
863 	if (param->flags & DM_PERSISTENT_DEV_FLAG)
864 		m = MINOR(huge_decode_dev(param->dev));
865 
866 	r = dm_create(m, &md);
867 	if (r)
868 		return r;
869 
870 	r = dm_hash_insert(param->name, *param->uuid ? param->uuid : NULL, md);
871 	if (r) {
872 		dm_put(md);
873 		dm_destroy(md);
874 		return r;
875 	}
876 
877 	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
878 
879 	__dev_status(md, param);
880 
881 	dm_put(md);
882 
883 	return 0;
884 }
885 
886 /*
887  * Always use UUID for lookups if it's present, otherwise use name or dev.
888  */
889 static struct hash_cell *__find_device_hash_cell(struct dm_ioctl *param)
890 {
891 	struct hash_cell *hc = NULL;
892 
893 	if (*param->uuid) {
894 		if (*param->name || param->dev) {
895 			DMERR("Invalid ioctl structure: uuid %s, name %s, dev %llx",
896 			      param->uuid, param->name, (unsigned long long)param->dev);
897 			return NULL;
898 		}
899 
900 		hc = __get_uuid_cell(param->uuid);
901 		if (!hc)
902 			return NULL;
903 	} else if (*param->name) {
904 		if (param->dev) {
905 			DMERR("Invalid ioctl structure: name %s, dev %llx",
906 			      param->name, (unsigned long long)param->dev);
907 			return NULL;
908 		}
909 
910 		hc = __get_name_cell(param->name);
911 		if (!hc)
912 			return NULL;
913 	} else if (param->dev) {
914 		hc = __get_dev_cell(param->dev);
915 		if (!hc)
916 			return NULL;
917 	} else
918 		return NULL;
919 
920 	/*
921 	 * Sneakily write in both the name and the uuid
922 	 * while we have the cell.
923 	 */
924 	strlcpy(param->name, hc->name, sizeof(param->name));
925 	if (hc->uuid)
926 		strlcpy(param->uuid, hc->uuid, sizeof(param->uuid));
927 	else
928 		param->uuid[0] = '\0';
929 
930 	if (hc->new_map)
931 		param->flags |= DM_INACTIVE_PRESENT_FLAG;
932 	else
933 		param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
934 
935 	return hc;
936 }
937 
938 static struct mapped_device *find_device(struct dm_ioctl *param)
939 {
940 	struct hash_cell *hc;
941 	struct mapped_device *md = NULL;
942 
943 	down_read(&_hash_lock);
944 	hc = __find_device_hash_cell(param);
945 	if (hc)
946 		md = hc->md;
947 	up_read(&_hash_lock);
948 
949 	return md;
950 }
951 
952 static int dev_remove(struct file *filp, struct dm_ioctl *param, size_t param_size)
953 {
954 	struct hash_cell *hc;
955 	struct mapped_device *md;
956 	int r;
957 	struct dm_table *t;
958 
959 	down_write(&_hash_lock);
960 	hc = __find_device_hash_cell(param);
961 
962 	if (!hc) {
963 		DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
964 		up_write(&_hash_lock);
965 		return -ENXIO;
966 	}
967 
968 	md = hc->md;
969 
970 	/*
971 	 * Ensure the device is not open and nothing further can open it.
972 	 */
973 	r = dm_lock_for_deletion(md, !!(param->flags & DM_DEFERRED_REMOVE), false);
974 	if (r) {
975 		if (r == -EBUSY && param->flags & DM_DEFERRED_REMOVE) {
976 			up_write(&_hash_lock);
977 			dm_put(md);
978 			return 0;
979 		}
980 		DMDEBUG_LIMIT("unable to remove open device %s", hc->name);
981 		up_write(&_hash_lock);
982 		dm_put(md);
983 		return r;
984 	}
985 
986 	t = __hash_remove(hc);
987 	up_write(&_hash_lock);
988 
989 	if (t) {
990 		dm_sync_table(md);
991 		dm_table_destroy(t);
992 	}
993 
994 	param->flags &= ~DM_DEFERRED_REMOVE;
995 
996 	dm_ima_measure_on_device_remove(md, false);
997 
998 	if (!dm_kobject_uevent(md, KOBJ_REMOVE, param->event_nr))
999 		param->flags |= DM_UEVENT_GENERATED_FLAG;
1000 
1001 	dm_put(md);
1002 	dm_destroy(md);
1003 	return 0;
1004 }
1005 
1006 /*
1007  * Check a string doesn't overrun the chunk of
1008  * memory we copied from userland.
1009  */
1010 static int invalid_str(char *str, void *end)
1011 {
1012 	while ((void *) str < end)
1013 		if (!*str++)
1014 			return 0;
1015 
1016 	return -EINVAL;
1017 }
1018 
1019 static int dev_rename(struct file *filp, struct dm_ioctl *param, size_t param_size)
1020 {
1021 	int r;
1022 	char *new_data = (char *) param + param->data_start;
1023 	struct mapped_device *md;
1024 	unsigned change_uuid = (param->flags & DM_UUID_FLAG) ? 1 : 0;
1025 
1026 	if (new_data < param->data ||
1027 	    invalid_str(new_data, (void *) param + param_size) || !*new_data ||
1028 	    strlen(new_data) > (change_uuid ? DM_UUID_LEN - 1 : DM_NAME_LEN - 1)) {
1029 		DMERR("Invalid new mapped device name or uuid string supplied.");
1030 		return -EINVAL;
1031 	}
1032 
1033 	if (!change_uuid) {
1034 		r = check_name(new_data);
1035 		if (r)
1036 			return r;
1037 	}
1038 
1039 	md = dm_hash_rename(param, new_data);
1040 	if (IS_ERR(md))
1041 		return PTR_ERR(md);
1042 
1043 	__dev_status(md, param);
1044 	dm_put(md);
1045 
1046 	return 0;
1047 }
1048 
1049 static int dev_set_geometry(struct file *filp, struct dm_ioctl *param, size_t param_size)
1050 {
1051 	int r = -EINVAL, x;
1052 	struct mapped_device *md;
1053 	struct hd_geometry geometry;
1054 	unsigned long indata[4];
1055 	char *geostr = (char *) param + param->data_start;
1056 	char dummy;
1057 
1058 	md = find_device(param);
1059 	if (!md)
1060 		return -ENXIO;
1061 
1062 	if (geostr < param->data ||
1063 	    invalid_str(geostr, (void *) param + param_size)) {
1064 		DMERR("Invalid geometry supplied.");
1065 		goto out;
1066 	}
1067 
1068 	x = sscanf(geostr, "%lu %lu %lu %lu%c", indata,
1069 		   indata + 1, indata + 2, indata + 3, &dummy);
1070 
1071 	if (x != 4) {
1072 		DMERR("Unable to interpret geometry settings.");
1073 		goto out;
1074 	}
1075 
1076 	if (indata[0] > 65535 || indata[1] > 255 ||
1077 	    indata[2] > 255 || indata[3] > ULONG_MAX) {
1078 		DMERR("Geometry exceeds range limits.");
1079 		goto out;
1080 	}
1081 
1082 	geometry.cylinders = indata[0];
1083 	geometry.heads = indata[1];
1084 	geometry.sectors = indata[2];
1085 	geometry.start = indata[3];
1086 
1087 	r = dm_set_geometry(md, &geometry);
1088 
1089 	param->data_size = 0;
1090 
1091 out:
1092 	dm_put(md);
1093 	return r;
1094 }
1095 
1096 static int do_suspend(struct dm_ioctl *param)
1097 {
1098 	int r = 0;
1099 	unsigned suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
1100 	struct mapped_device *md;
1101 
1102 	md = find_device(param);
1103 	if (!md)
1104 		return -ENXIO;
1105 
1106 	if (param->flags & DM_SKIP_LOCKFS_FLAG)
1107 		suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
1108 	if (param->flags & DM_NOFLUSH_FLAG)
1109 		suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
1110 
1111 	if (!dm_suspended_md(md)) {
1112 		r = dm_suspend(md, suspend_flags);
1113 		if (r)
1114 			goto out;
1115 	}
1116 
1117 	__dev_status(md, param);
1118 
1119 out:
1120 	dm_put(md);
1121 
1122 	return r;
1123 }
1124 
1125 static int do_resume(struct dm_ioctl *param)
1126 {
1127 	int r = 0;
1128 	unsigned suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
1129 	struct hash_cell *hc;
1130 	struct mapped_device *md;
1131 	struct dm_table *new_map, *old_map = NULL;
1132 
1133 	down_write(&_hash_lock);
1134 
1135 	hc = __find_device_hash_cell(param);
1136 	if (!hc) {
1137 		DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
1138 		up_write(&_hash_lock);
1139 		return -ENXIO;
1140 	}
1141 
1142 	md = hc->md;
1143 
1144 	new_map = hc->new_map;
1145 	hc->new_map = NULL;
1146 	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1147 
1148 	up_write(&_hash_lock);
1149 
1150 	/* Do we need to load a new map ? */
1151 	if (new_map) {
1152 		/* Suspend if it isn't already suspended */
1153 		if (param->flags & DM_SKIP_LOCKFS_FLAG)
1154 			suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
1155 		if (param->flags & DM_NOFLUSH_FLAG)
1156 			suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
1157 		if (!dm_suspended_md(md))
1158 			dm_suspend(md, suspend_flags);
1159 
1160 		old_map = dm_swap_table(md, new_map);
1161 		if (IS_ERR(old_map)) {
1162 			dm_sync_table(md);
1163 			dm_table_destroy(new_map);
1164 			dm_put(md);
1165 			return PTR_ERR(old_map);
1166 		}
1167 
1168 		if (dm_table_get_mode(new_map) & FMODE_WRITE)
1169 			set_disk_ro(dm_disk(md), 0);
1170 		else
1171 			set_disk_ro(dm_disk(md), 1);
1172 	}
1173 
1174 	if (dm_suspended_md(md)) {
1175 		r = dm_resume(md);
1176 		if (!r) {
1177 			dm_ima_measure_on_device_resume(md, new_map ? true : false);
1178 
1179 			if (!dm_kobject_uevent(md, KOBJ_CHANGE, param->event_nr))
1180 				param->flags |= DM_UEVENT_GENERATED_FLAG;
1181 		}
1182 	}
1183 
1184 	/*
1185 	 * Since dm_swap_table synchronizes RCU, nobody should be in
1186 	 * read-side critical section already.
1187 	 */
1188 	if (old_map)
1189 		dm_table_destroy(old_map);
1190 
1191 	if (!r)
1192 		__dev_status(md, param);
1193 
1194 	dm_put(md);
1195 	return r;
1196 }
1197 
1198 /*
1199  * Set or unset the suspension state of a device.
1200  * If the device already is in the requested state we just return its status.
1201  */
1202 static int dev_suspend(struct file *filp, struct dm_ioctl *param, size_t param_size)
1203 {
1204 	if (param->flags & DM_SUSPEND_FLAG)
1205 		return do_suspend(param);
1206 
1207 	return do_resume(param);
1208 }
1209 
1210 /*
1211  * Copies device info back to user space, used by
1212  * the create and info ioctls.
1213  */
1214 static int dev_status(struct file *filp, struct dm_ioctl *param, size_t param_size)
1215 {
1216 	struct mapped_device *md;
1217 
1218 	md = find_device(param);
1219 	if (!md)
1220 		return -ENXIO;
1221 
1222 	__dev_status(md, param);
1223 	dm_put(md);
1224 
1225 	return 0;
1226 }
1227 
1228 /*
1229  * Build up the status struct for each target
1230  */
1231 static void retrieve_status(struct dm_table *table,
1232 			    struct dm_ioctl *param, size_t param_size)
1233 {
1234 	unsigned int i, num_targets;
1235 	struct dm_target_spec *spec;
1236 	char *outbuf, *outptr;
1237 	status_type_t type;
1238 	size_t remaining, len, used = 0;
1239 	unsigned status_flags = 0;
1240 
1241 	outptr = outbuf = get_result_buffer(param, param_size, &len);
1242 
1243 	if (param->flags & DM_STATUS_TABLE_FLAG)
1244 		type = STATUSTYPE_TABLE;
1245 	else if (param->flags & DM_IMA_MEASUREMENT_FLAG)
1246 		type = STATUSTYPE_IMA;
1247 	else
1248 		type = STATUSTYPE_INFO;
1249 
1250 	/* Get all the target info */
1251 	num_targets = table->num_targets;
1252 	for (i = 0; i < num_targets; i++) {
1253 		struct dm_target *ti = dm_table_get_target(table, i);
1254 		size_t l;
1255 
1256 		remaining = len - (outptr - outbuf);
1257 		if (remaining <= sizeof(struct dm_target_spec)) {
1258 			param->flags |= DM_BUFFER_FULL_FLAG;
1259 			break;
1260 		}
1261 
1262 		spec = (struct dm_target_spec *) outptr;
1263 
1264 		spec->status = 0;
1265 		spec->sector_start = ti->begin;
1266 		spec->length = ti->len;
1267 		strncpy(spec->target_type, ti->type->name,
1268 			sizeof(spec->target_type) - 1);
1269 
1270 		outptr += sizeof(struct dm_target_spec);
1271 		remaining = len - (outptr - outbuf);
1272 		if (remaining <= 0) {
1273 			param->flags |= DM_BUFFER_FULL_FLAG;
1274 			break;
1275 		}
1276 
1277 		/* Get the status/table string from the target driver */
1278 		if (ti->type->status) {
1279 			if (param->flags & DM_NOFLUSH_FLAG)
1280 				status_flags |= DM_STATUS_NOFLUSH_FLAG;
1281 			ti->type->status(ti, type, status_flags, outptr, remaining);
1282 		} else
1283 			outptr[0] = '\0';
1284 
1285 		l = strlen(outptr) + 1;
1286 		if (l == remaining) {
1287 			param->flags |= DM_BUFFER_FULL_FLAG;
1288 			break;
1289 		}
1290 
1291 		outptr += l;
1292 		used = param->data_start + (outptr - outbuf);
1293 
1294 		outptr = align_ptr(outptr);
1295 		spec->next = outptr - outbuf;
1296 	}
1297 
1298 	if (used)
1299 		param->data_size = used;
1300 
1301 	param->target_count = num_targets;
1302 }
1303 
1304 /*
1305  * Wait for a device to report an event
1306  */
1307 static int dev_wait(struct file *filp, struct dm_ioctl *param, size_t param_size)
1308 {
1309 	int r = 0;
1310 	struct mapped_device *md;
1311 	struct dm_table *table;
1312 	int srcu_idx;
1313 
1314 	md = find_device(param);
1315 	if (!md)
1316 		return -ENXIO;
1317 
1318 	/*
1319 	 * Wait for a notification event
1320 	 */
1321 	if (dm_wait_event(md, param->event_nr)) {
1322 		r = -ERESTARTSYS;
1323 		goto out;
1324 	}
1325 
1326 	/*
1327 	 * The userland program is going to want to know what
1328 	 * changed to trigger the event, so we may as well tell
1329 	 * him and save an ioctl.
1330 	 */
1331 	__dev_status(md, param);
1332 
1333 	table = dm_get_live_or_inactive_table(md, param, &srcu_idx);
1334 	if (table)
1335 		retrieve_status(table, param, param_size);
1336 	dm_put_live_table(md, srcu_idx);
1337 
1338 out:
1339 	dm_put(md);
1340 
1341 	return r;
1342 }
1343 
1344 /*
1345  * Remember the global event number and make it possible to poll
1346  * for further events.
1347  */
1348 static int dev_arm_poll(struct file *filp, struct dm_ioctl *param, size_t param_size)
1349 {
1350 	struct dm_file *priv = filp->private_data;
1351 
1352 	priv->global_event_nr = atomic_read(&dm_global_event_nr);
1353 
1354 	return 0;
1355 }
1356 
1357 static inline fmode_t get_mode(struct dm_ioctl *param)
1358 {
1359 	fmode_t mode = FMODE_READ | FMODE_WRITE;
1360 
1361 	if (param->flags & DM_READONLY_FLAG)
1362 		mode = FMODE_READ;
1363 
1364 	return mode;
1365 }
1366 
1367 static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
1368 		       struct dm_target_spec **spec, char **target_params)
1369 {
1370 	*spec = (struct dm_target_spec *) ((unsigned char *) last + next);
1371 	*target_params = (char *) (*spec + 1);
1372 
1373 	if (*spec < (last + 1))
1374 		return -EINVAL;
1375 
1376 	return invalid_str(*target_params, end);
1377 }
1378 
1379 static int populate_table(struct dm_table *table,
1380 			  struct dm_ioctl *param, size_t param_size)
1381 {
1382 	int r;
1383 	unsigned int i = 0;
1384 	struct dm_target_spec *spec = (struct dm_target_spec *) param;
1385 	uint32_t next = param->data_start;
1386 	void *end = (void *) param + param_size;
1387 	char *target_params;
1388 
1389 	if (!param->target_count) {
1390 		DMERR("populate_table: no targets specified");
1391 		return -EINVAL;
1392 	}
1393 
1394 	for (i = 0; i < param->target_count; i++) {
1395 
1396 		r = next_target(spec, next, end, &spec, &target_params);
1397 		if (r) {
1398 			DMERR("unable to find target");
1399 			return r;
1400 		}
1401 
1402 		r = dm_table_add_target(table, spec->target_type,
1403 					(sector_t) spec->sector_start,
1404 					(sector_t) spec->length,
1405 					target_params);
1406 		if (r) {
1407 			DMERR("error adding target to table");
1408 			return r;
1409 		}
1410 
1411 		next = spec->next;
1412 	}
1413 
1414 	return dm_table_complete(table);
1415 }
1416 
1417 static bool is_valid_type(enum dm_queue_mode cur, enum dm_queue_mode new)
1418 {
1419 	if (cur == new ||
1420 	    (cur == DM_TYPE_BIO_BASED && new == DM_TYPE_DAX_BIO_BASED))
1421 		return true;
1422 
1423 	return false;
1424 }
1425 
1426 static int table_load(struct file *filp, struct dm_ioctl *param, size_t param_size)
1427 {
1428 	int r;
1429 	struct hash_cell *hc;
1430 	struct dm_table *t, *old_map = NULL;
1431 	struct mapped_device *md;
1432 	struct target_type *immutable_target_type;
1433 
1434 	md = find_device(param);
1435 	if (!md)
1436 		return -ENXIO;
1437 
1438 	r = dm_table_create(&t, get_mode(param), param->target_count, md);
1439 	if (r)
1440 		goto err;
1441 
1442 	/* Protect md->type and md->queue against concurrent table loads. */
1443 	dm_lock_md_type(md);
1444 	r = populate_table(t, param, param_size);
1445 	if (r)
1446 		goto err_unlock_md_type;
1447 
1448 	dm_ima_measure_on_table_load(t, STATUSTYPE_IMA);
1449 
1450 	immutable_target_type = dm_get_immutable_target_type(md);
1451 	if (immutable_target_type &&
1452 	    (immutable_target_type != dm_table_get_immutable_target_type(t)) &&
1453 	    !dm_table_get_wildcard_target(t)) {
1454 		DMERR("can't replace immutable target type %s",
1455 		      immutable_target_type->name);
1456 		r = -EINVAL;
1457 		goto err_unlock_md_type;
1458 	}
1459 
1460 	if (dm_get_md_type(md) == DM_TYPE_NONE) {
1461 		/* setup md->queue to reflect md's type (may block) */
1462 		r = dm_setup_md_queue(md, t);
1463 		if (r) {
1464 			DMERR("unable to set up device queue for new table.");
1465 			goto err_unlock_md_type;
1466 		}
1467 	} else if (!is_valid_type(dm_get_md_type(md), dm_table_get_type(t))) {
1468 		DMERR("can't change device type (old=%u vs new=%u) after initial table load.",
1469 		      dm_get_md_type(md), dm_table_get_type(t));
1470 		r = -EINVAL;
1471 		goto err_unlock_md_type;
1472 	}
1473 
1474 	dm_unlock_md_type(md);
1475 
1476 	/* stage inactive table */
1477 	down_write(&_hash_lock);
1478 	hc = dm_get_mdptr(md);
1479 	if (!hc || hc->md != md) {
1480 		DMERR("device has been removed from the dev hash table.");
1481 		up_write(&_hash_lock);
1482 		r = -ENXIO;
1483 		goto err_destroy_table;
1484 	}
1485 
1486 	if (hc->new_map)
1487 		old_map = hc->new_map;
1488 	hc->new_map = t;
1489 	up_write(&_hash_lock);
1490 
1491 	param->flags |= DM_INACTIVE_PRESENT_FLAG;
1492 	__dev_status(md, param);
1493 
1494 	if (old_map) {
1495 		dm_sync_table(md);
1496 		dm_table_destroy(old_map);
1497 	}
1498 
1499 	dm_put(md);
1500 
1501 	return 0;
1502 
1503 err_unlock_md_type:
1504 	dm_unlock_md_type(md);
1505 err_destroy_table:
1506 	dm_table_destroy(t);
1507 err:
1508 	dm_put(md);
1509 
1510 	return r;
1511 }
1512 
1513 static int table_clear(struct file *filp, struct dm_ioctl *param, size_t param_size)
1514 {
1515 	struct hash_cell *hc;
1516 	struct mapped_device *md;
1517 	struct dm_table *old_map = NULL;
1518 	bool has_new_map = false;
1519 
1520 	down_write(&_hash_lock);
1521 
1522 	hc = __find_device_hash_cell(param);
1523 	if (!hc) {
1524 		DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
1525 		up_write(&_hash_lock);
1526 		return -ENXIO;
1527 	}
1528 
1529 	if (hc->new_map) {
1530 		old_map = hc->new_map;
1531 		hc->new_map = NULL;
1532 		has_new_map = true;
1533 	}
1534 
1535 	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1536 
1537 	__dev_status(hc->md, param);
1538 	md = hc->md;
1539 	up_write(&_hash_lock);
1540 	if (old_map) {
1541 		dm_sync_table(md);
1542 		dm_table_destroy(old_map);
1543 	}
1544 	dm_ima_measure_on_table_clear(md, has_new_map);
1545 	dm_put(md);
1546 
1547 	return 0;
1548 }
1549 
1550 /*
1551  * Retrieves a list of devices used by a particular dm device.
1552  */
1553 static void retrieve_deps(struct dm_table *table,
1554 			  struct dm_ioctl *param, size_t param_size)
1555 {
1556 	unsigned int count = 0;
1557 	struct list_head *tmp;
1558 	size_t len, needed;
1559 	struct dm_dev_internal *dd;
1560 	struct dm_target_deps *deps;
1561 
1562 	deps = get_result_buffer(param, param_size, &len);
1563 
1564 	/*
1565 	 * Count the devices.
1566 	 */
1567 	list_for_each (tmp, dm_table_get_devices(table))
1568 		count++;
1569 
1570 	/*
1571 	 * Check we have enough space.
1572 	 */
1573 	needed = struct_size(deps, dev, count);
1574 	if (len < needed) {
1575 		param->flags |= DM_BUFFER_FULL_FLAG;
1576 		return;
1577 	}
1578 
1579 	/*
1580 	 * Fill in the devices.
1581 	 */
1582 	deps->count = count;
1583 	count = 0;
1584 	list_for_each_entry (dd, dm_table_get_devices(table), list)
1585 		deps->dev[count++] = huge_encode_dev(dd->dm_dev->bdev->bd_dev);
1586 
1587 	param->data_size = param->data_start + needed;
1588 }
1589 
1590 static int table_deps(struct file *filp, struct dm_ioctl *param, size_t param_size)
1591 {
1592 	struct mapped_device *md;
1593 	struct dm_table *table;
1594 	int srcu_idx;
1595 
1596 	md = find_device(param);
1597 	if (!md)
1598 		return -ENXIO;
1599 
1600 	__dev_status(md, param);
1601 
1602 	table = dm_get_live_or_inactive_table(md, param, &srcu_idx);
1603 	if (table)
1604 		retrieve_deps(table, param, param_size);
1605 	dm_put_live_table(md, srcu_idx);
1606 
1607 	dm_put(md);
1608 
1609 	return 0;
1610 }
1611 
1612 /*
1613  * Return the status of a device as a text string for each
1614  * target.
1615  */
1616 static int table_status(struct file *filp, struct dm_ioctl *param, size_t param_size)
1617 {
1618 	struct mapped_device *md;
1619 	struct dm_table *table;
1620 	int srcu_idx;
1621 
1622 	md = find_device(param);
1623 	if (!md)
1624 		return -ENXIO;
1625 
1626 	__dev_status(md, param);
1627 
1628 	table = dm_get_live_or_inactive_table(md, param, &srcu_idx);
1629 	if (table)
1630 		retrieve_status(table, param, param_size);
1631 	dm_put_live_table(md, srcu_idx);
1632 
1633 	dm_put(md);
1634 
1635 	return 0;
1636 }
1637 
1638 /*
1639  * Process device-mapper dependent messages.  Messages prefixed with '@'
1640  * are processed by the DM core.  All others are delivered to the target.
1641  * Returns a number <= 1 if message was processed by device mapper.
1642  * Returns 2 if message should be delivered to the target.
1643  */
1644 static int message_for_md(struct mapped_device *md, unsigned argc, char **argv,
1645 			  char *result, unsigned maxlen)
1646 {
1647 	int r;
1648 
1649 	if (**argv != '@')
1650 		return 2; /* no '@' prefix, deliver to target */
1651 
1652 	if (!strcasecmp(argv[0], "@cancel_deferred_remove")) {
1653 		if (argc != 1) {
1654 			DMERR("Invalid arguments for @cancel_deferred_remove");
1655 			return -EINVAL;
1656 		}
1657 		return dm_cancel_deferred_remove(md);
1658 	}
1659 
1660 	r = dm_stats_message(md, argc, argv, result, maxlen);
1661 	if (r < 2)
1662 		return r;
1663 
1664 	DMERR("Unsupported message sent to DM core: %s", argv[0]);
1665 	return -EINVAL;
1666 }
1667 
1668 /*
1669  * Pass a message to the target that's at the supplied device offset.
1670  */
1671 static int target_message(struct file *filp, struct dm_ioctl *param, size_t param_size)
1672 {
1673 	int r, argc;
1674 	char **argv;
1675 	struct mapped_device *md;
1676 	struct dm_table *table;
1677 	struct dm_target *ti;
1678 	struct dm_target_msg *tmsg = (void *) param + param->data_start;
1679 	size_t maxlen;
1680 	char *result = get_result_buffer(param, param_size, &maxlen);
1681 	int srcu_idx;
1682 
1683 	md = find_device(param);
1684 	if (!md)
1685 		return -ENXIO;
1686 
1687 	if (tmsg < (struct dm_target_msg *) param->data ||
1688 	    invalid_str(tmsg->message, (void *) param + param_size)) {
1689 		DMERR("Invalid target message parameters.");
1690 		r = -EINVAL;
1691 		goto out;
1692 	}
1693 
1694 	r = dm_split_args(&argc, &argv, tmsg->message);
1695 	if (r) {
1696 		DMERR("Failed to split target message parameters");
1697 		goto out;
1698 	}
1699 
1700 	if (!argc) {
1701 		DMERR("Empty message received.");
1702 		r = -EINVAL;
1703 		goto out_argv;
1704 	}
1705 
1706 	r = message_for_md(md, argc, argv, result, maxlen);
1707 	if (r <= 1)
1708 		goto out_argv;
1709 
1710 	table = dm_get_live_table(md, &srcu_idx);
1711 	if (!table)
1712 		goto out_table;
1713 
1714 	if (dm_deleting_md(md)) {
1715 		r = -ENXIO;
1716 		goto out_table;
1717 	}
1718 
1719 	ti = dm_table_find_target(table, tmsg->sector);
1720 	if (!ti) {
1721 		DMERR("Target message sector outside device.");
1722 		r = -EINVAL;
1723 	} else if (ti->type->message)
1724 		r = ti->type->message(ti, argc, argv, result, maxlen);
1725 	else {
1726 		DMERR("Target type does not support messages");
1727 		r = -EINVAL;
1728 	}
1729 
1730  out_table:
1731 	dm_put_live_table(md, srcu_idx);
1732  out_argv:
1733 	kfree(argv);
1734  out:
1735 	if (r >= 0)
1736 		__dev_status(md, param);
1737 
1738 	if (r == 1) {
1739 		param->flags |= DM_DATA_OUT_FLAG;
1740 		if (dm_message_test_buffer_overflow(result, maxlen))
1741 			param->flags |= DM_BUFFER_FULL_FLAG;
1742 		else
1743 			param->data_size = param->data_start + strlen(result) + 1;
1744 		r = 0;
1745 	}
1746 
1747 	dm_put(md);
1748 	return r;
1749 }
1750 
1751 /*
1752  * The ioctl parameter block consists of two parts, a dm_ioctl struct
1753  * followed by a data buffer.  This flag is set if the second part,
1754  * which has a variable size, is not used by the function processing
1755  * the ioctl.
1756  */
1757 #define IOCTL_FLAGS_NO_PARAMS		1
1758 #define IOCTL_FLAGS_ISSUE_GLOBAL_EVENT	2
1759 
1760 /*-----------------------------------------------------------------
1761  * Implementation of open/close/ioctl on the special char
1762  * device.
1763  *---------------------------------------------------------------*/
1764 static ioctl_fn lookup_ioctl(unsigned int cmd, int *ioctl_flags)
1765 {
1766 	static const struct {
1767 		int cmd;
1768 		int flags;
1769 		ioctl_fn fn;
1770 	} _ioctls[] = {
1771 		{DM_VERSION_CMD, 0, NULL}, /* version is dealt with elsewhere */
1772 		{DM_REMOVE_ALL_CMD, IOCTL_FLAGS_NO_PARAMS | IOCTL_FLAGS_ISSUE_GLOBAL_EVENT, remove_all},
1773 		{DM_LIST_DEVICES_CMD, 0, list_devices},
1774 
1775 		{DM_DEV_CREATE_CMD, IOCTL_FLAGS_NO_PARAMS | IOCTL_FLAGS_ISSUE_GLOBAL_EVENT, dev_create},
1776 		{DM_DEV_REMOVE_CMD, IOCTL_FLAGS_NO_PARAMS | IOCTL_FLAGS_ISSUE_GLOBAL_EVENT, dev_remove},
1777 		{DM_DEV_RENAME_CMD, IOCTL_FLAGS_ISSUE_GLOBAL_EVENT, dev_rename},
1778 		{DM_DEV_SUSPEND_CMD, IOCTL_FLAGS_NO_PARAMS, dev_suspend},
1779 		{DM_DEV_STATUS_CMD, IOCTL_FLAGS_NO_PARAMS, dev_status},
1780 		{DM_DEV_WAIT_CMD, 0, dev_wait},
1781 
1782 		{DM_TABLE_LOAD_CMD, 0, table_load},
1783 		{DM_TABLE_CLEAR_CMD, IOCTL_FLAGS_NO_PARAMS, table_clear},
1784 		{DM_TABLE_DEPS_CMD, 0, table_deps},
1785 		{DM_TABLE_STATUS_CMD, 0, table_status},
1786 
1787 		{DM_LIST_VERSIONS_CMD, 0, list_versions},
1788 
1789 		{DM_TARGET_MSG_CMD, 0, target_message},
1790 		{DM_DEV_SET_GEOMETRY_CMD, 0, dev_set_geometry},
1791 		{DM_DEV_ARM_POLL, IOCTL_FLAGS_NO_PARAMS, dev_arm_poll},
1792 		{DM_GET_TARGET_VERSION, 0, get_target_version},
1793 	};
1794 
1795 	if (unlikely(cmd >= ARRAY_SIZE(_ioctls)))
1796 		return NULL;
1797 
1798 	cmd = array_index_nospec(cmd, ARRAY_SIZE(_ioctls));
1799 	*ioctl_flags = _ioctls[cmd].flags;
1800 	return _ioctls[cmd].fn;
1801 }
1802 
1803 /*
1804  * As well as checking the version compatibility this always
1805  * copies the kernel interface version out.
1806  */
1807 static int check_version(unsigned int cmd, struct dm_ioctl __user *user)
1808 {
1809 	uint32_t version[3];
1810 	int r = 0;
1811 
1812 	if (copy_from_user(version, user->version, sizeof(version)))
1813 		return -EFAULT;
1814 
1815 	if ((DM_VERSION_MAJOR != version[0]) ||
1816 	    (DM_VERSION_MINOR < version[1])) {
1817 		DMERR("ioctl interface mismatch: "
1818 		      "kernel(%u.%u.%u), user(%u.%u.%u), cmd(%d)",
1819 		      DM_VERSION_MAJOR, DM_VERSION_MINOR,
1820 		      DM_VERSION_PATCHLEVEL,
1821 		      version[0], version[1], version[2], cmd);
1822 		r = -EINVAL;
1823 	}
1824 
1825 	/*
1826 	 * Fill in the kernel version.
1827 	 */
1828 	version[0] = DM_VERSION_MAJOR;
1829 	version[1] = DM_VERSION_MINOR;
1830 	version[2] = DM_VERSION_PATCHLEVEL;
1831 	if (copy_to_user(user->version, version, sizeof(version)))
1832 		return -EFAULT;
1833 
1834 	return r;
1835 }
1836 
1837 #define DM_PARAMS_MALLOC	0x0001	/* Params allocated with kvmalloc() */
1838 #define DM_WIPE_BUFFER		0x0010	/* Wipe input buffer before returning from ioctl */
1839 
1840 static void free_params(struct dm_ioctl *param, size_t param_size, int param_flags)
1841 {
1842 	if (param_flags & DM_WIPE_BUFFER)
1843 		memset(param, 0, param_size);
1844 
1845 	if (param_flags & DM_PARAMS_MALLOC)
1846 		kvfree(param);
1847 }
1848 
1849 static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl *param_kernel,
1850 		       int ioctl_flags, struct dm_ioctl **param, int *param_flags)
1851 {
1852 	struct dm_ioctl *dmi;
1853 	int secure_data;
1854 	const size_t minimum_data_size = offsetof(struct dm_ioctl, data);
1855 	unsigned noio_flag;
1856 
1857 	if (copy_from_user(param_kernel, user, minimum_data_size))
1858 		return -EFAULT;
1859 
1860 	if (param_kernel->data_size < minimum_data_size) {
1861 		DMERR("Invalid data size in the ioctl structure: %u",
1862 		      param_kernel->data_size);
1863 		return -EINVAL;
1864 	}
1865 
1866 	secure_data = param_kernel->flags & DM_SECURE_DATA_FLAG;
1867 
1868 	*param_flags = secure_data ? DM_WIPE_BUFFER : 0;
1869 
1870 	if (ioctl_flags & IOCTL_FLAGS_NO_PARAMS) {
1871 		dmi = param_kernel;
1872 		dmi->data_size = minimum_data_size;
1873 		goto data_copied;
1874 	}
1875 
1876 	/*
1877 	 * Use __GFP_HIGH to avoid low memory issues when a device is
1878 	 * suspended and the ioctl is needed to resume it.
1879 	 * Use kmalloc() rather than vmalloc() when we can.
1880 	 */
1881 	dmi = NULL;
1882 	noio_flag = memalloc_noio_save();
1883 	dmi = kvmalloc(param_kernel->data_size, GFP_KERNEL | __GFP_HIGH);
1884 	memalloc_noio_restore(noio_flag);
1885 
1886 	if (!dmi) {
1887 		if (secure_data && clear_user(user, param_kernel->data_size))
1888 			return -EFAULT;
1889 		return -ENOMEM;
1890 	}
1891 
1892 	*param_flags |= DM_PARAMS_MALLOC;
1893 
1894 	/* Copy from param_kernel (which was already copied from user) */
1895 	memcpy(dmi, param_kernel, minimum_data_size);
1896 
1897 	if (copy_from_user(&dmi->data, (char __user *)user + minimum_data_size,
1898 			   param_kernel->data_size - minimum_data_size))
1899 		goto bad;
1900 data_copied:
1901 	/* Wipe the user buffer so we do not return it to userspace */
1902 	if (secure_data && clear_user(user, param_kernel->data_size))
1903 		goto bad;
1904 
1905 	*param = dmi;
1906 	return 0;
1907 
1908 bad:
1909 	free_params(dmi, param_kernel->data_size, *param_flags);
1910 
1911 	return -EFAULT;
1912 }
1913 
1914 static int validate_params(uint cmd, struct dm_ioctl *param)
1915 {
1916 	/* Always clear this flag */
1917 	param->flags &= ~DM_BUFFER_FULL_FLAG;
1918 	param->flags &= ~DM_UEVENT_GENERATED_FLAG;
1919 	param->flags &= ~DM_SECURE_DATA_FLAG;
1920 	param->flags &= ~DM_DATA_OUT_FLAG;
1921 
1922 	/* Ignores parameters */
1923 	if (cmd == DM_REMOVE_ALL_CMD ||
1924 	    cmd == DM_LIST_DEVICES_CMD ||
1925 	    cmd == DM_LIST_VERSIONS_CMD)
1926 		return 0;
1927 
1928 	if (cmd == DM_DEV_CREATE_CMD) {
1929 		if (!*param->name) {
1930 			DMERR("name not supplied when creating device");
1931 			return -EINVAL;
1932 		}
1933 	} else if (*param->uuid && *param->name) {
1934 		DMERR("only supply one of name or uuid, cmd(%u)", cmd);
1935 		return -EINVAL;
1936 	}
1937 
1938 	/* Ensure strings are terminated */
1939 	param->name[DM_NAME_LEN - 1] = '\0';
1940 	param->uuid[DM_UUID_LEN - 1] = '\0';
1941 
1942 	return 0;
1943 }
1944 
1945 static int ctl_ioctl(struct file *file, uint command, struct dm_ioctl __user *user)
1946 {
1947 	int r = 0;
1948 	int ioctl_flags;
1949 	int param_flags;
1950 	unsigned int cmd;
1951 	struct dm_ioctl *param;
1952 	ioctl_fn fn = NULL;
1953 	size_t input_param_size;
1954 	struct dm_ioctl param_kernel;
1955 
1956 	/* only root can play with this */
1957 	if (!capable(CAP_SYS_ADMIN))
1958 		return -EACCES;
1959 
1960 	if (_IOC_TYPE(command) != DM_IOCTL)
1961 		return -ENOTTY;
1962 
1963 	cmd = _IOC_NR(command);
1964 
1965 	/*
1966 	 * Check the interface version passed in.  This also
1967 	 * writes out the kernel's interface version.
1968 	 */
1969 	r = check_version(cmd, user);
1970 	if (r)
1971 		return r;
1972 
1973 	/*
1974 	 * Nothing more to do for the version command.
1975 	 */
1976 	if (cmd == DM_VERSION_CMD)
1977 		return 0;
1978 
1979 	fn = lookup_ioctl(cmd, &ioctl_flags);
1980 	if (!fn) {
1981 		DMERR("dm_ctl_ioctl: unknown command 0x%x", command);
1982 		return -ENOTTY;
1983 	}
1984 
1985 	/*
1986 	 * Copy the parameters into kernel space.
1987 	 */
1988 	r = copy_params(user, &param_kernel, ioctl_flags, &param, &param_flags);
1989 
1990 	if (r)
1991 		return r;
1992 
1993 	input_param_size = param->data_size;
1994 	r = validate_params(cmd, param);
1995 	if (r)
1996 		goto out;
1997 
1998 	param->data_size = offsetof(struct dm_ioctl, data);
1999 	r = fn(file, param, input_param_size);
2000 
2001 	if (unlikely(param->flags & DM_BUFFER_FULL_FLAG) &&
2002 	    unlikely(ioctl_flags & IOCTL_FLAGS_NO_PARAMS))
2003 		DMERR("ioctl %d tried to output some data but has IOCTL_FLAGS_NO_PARAMS set", cmd);
2004 
2005 	if (!r && ioctl_flags & IOCTL_FLAGS_ISSUE_GLOBAL_EVENT)
2006 		dm_issue_global_event();
2007 
2008 	/*
2009 	 * Copy the results back to userland.
2010 	 */
2011 	if (!r && copy_to_user(user, param, param->data_size))
2012 		r = -EFAULT;
2013 
2014 out:
2015 	free_params(param, input_param_size, param_flags);
2016 	return r;
2017 }
2018 
2019 static long dm_ctl_ioctl(struct file *file, uint command, ulong u)
2020 {
2021 	return (long)ctl_ioctl(file, command, (struct dm_ioctl __user *)u);
2022 }
2023 
2024 #ifdef CONFIG_COMPAT
2025 static long dm_compat_ctl_ioctl(struct file *file, uint command, ulong u)
2026 {
2027 	return (long)dm_ctl_ioctl(file, command, (ulong) compat_ptr(u));
2028 }
2029 #else
2030 #define dm_compat_ctl_ioctl NULL
2031 #endif
2032 
2033 static int dm_open(struct inode *inode, struct file *filp)
2034 {
2035 	int r;
2036 	struct dm_file *priv;
2037 
2038 	r = nonseekable_open(inode, filp);
2039 	if (unlikely(r))
2040 		return r;
2041 
2042 	priv = filp->private_data = kmalloc(sizeof(struct dm_file), GFP_KERNEL);
2043 	if (!priv)
2044 		return -ENOMEM;
2045 
2046 	priv->global_event_nr = atomic_read(&dm_global_event_nr);
2047 
2048 	return 0;
2049 }
2050 
2051 static int dm_release(struct inode *inode, struct file *filp)
2052 {
2053 	kfree(filp->private_data);
2054 	return 0;
2055 }
2056 
2057 static __poll_t dm_poll(struct file *filp, poll_table *wait)
2058 {
2059 	struct dm_file *priv = filp->private_data;
2060 	__poll_t mask = 0;
2061 
2062 	poll_wait(filp, &dm_global_eventq, wait);
2063 
2064 	if ((int)(atomic_read(&dm_global_event_nr) - priv->global_event_nr) > 0)
2065 		mask |= EPOLLIN;
2066 
2067 	return mask;
2068 }
2069 
2070 static const struct file_operations _ctl_fops = {
2071 	.open    = dm_open,
2072 	.release = dm_release,
2073 	.poll    = dm_poll,
2074 	.unlocked_ioctl	 = dm_ctl_ioctl,
2075 	.compat_ioctl = dm_compat_ctl_ioctl,
2076 	.owner	 = THIS_MODULE,
2077 	.llseek  = noop_llseek,
2078 };
2079 
2080 static struct miscdevice _dm_misc = {
2081 	.minor		= MAPPER_CTRL_MINOR,
2082 	.name  		= DM_NAME,
2083 	.nodename	= DM_DIR "/" DM_CONTROL_NODE,
2084 	.fops  		= &_ctl_fops
2085 };
2086 
2087 MODULE_ALIAS_MISCDEV(MAPPER_CTRL_MINOR);
2088 MODULE_ALIAS("devname:" DM_DIR "/" DM_CONTROL_NODE);
2089 
2090 /*
2091  * Create misc character device and link to DM_DIR/control.
2092  */
2093 int __init dm_interface_init(void)
2094 {
2095 	int r;
2096 
2097 	r = misc_register(&_dm_misc);
2098 	if (r) {
2099 		DMERR("misc_register failed for control device");
2100 		return r;
2101 	}
2102 
2103 	DMINFO("%d.%d.%d%s initialised: %s", DM_VERSION_MAJOR,
2104 	       DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL, DM_VERSION_EXTRA,
2105 	       DM_DRIVER_EMAIL);
2106 	return 0;
2107 }
2108 
2109 void dm_interface_exit(void)
2110 {
2111 	misc_deregister(&_dm_misc);
2112 	dm_hash_exit();
2113 }
2114 
2115 /**
2116  * dm_copy_name_and_uuid - Copy mapped device name & uuid into supplied buffers
2117  * @md: Pointer to mapped_device
2118  * @name: Buffer (size DM_NAME_LEN) for name
2119  * @uuid: Buffer (size DM_UUID_LEN) for uuid or empty string if uuid not defined
2120  */
2121 int dm_copy_name_and_uuid(struct mapped_device *md, char *name, char *uuid)
2122 {
2123 	int r = 0;
2124 	struct hash_cell *hc;
2125 
2126 	if (!md)
2127 		return -ENXIO;
2128 
2129 	mutex_lock(&dm_hash_cells_mutex);
2130 	hc = dm_get_mdptr(md);
2131 	if (!hc || hc->md != md) {
2132 		r = -ENXIO;
2133 		goto out;
2134 	}
2135 
2136 	if (name)
2137 		strcpy(name, hc->name);
2138 	if (uuid)
2139 		strcpy(uuid, hc->uuid ? : "");
2140 
2141 out:
2142 	mutex_unlock(&dm_hash_cells_mutex);
2143 
2144 	return r;
2145 }
2146 EXPORT_SYMBOL_GPL(dm_copy_name_and_uuid);
2147 
2148 /**
2149  * dm_early_create - create a mapped device in early boot.
2150  *
2151  * @dmi: Contains main information of the device mapping to be created.
2152  * @spec_array: array of pointers to struct dm_target_spec. Describes the
2153  * mapping table of the device.
2154  * @target_params_array: array of strings with the parameters to a specific
2155  * target.
2156  *
2157  * Instead of having the struct dm_target_spec and the parameters for every
2158  * target embedded at the end of struct dm_ioctl (as performed in a normal
2159  * ioctl), pass them as arguments, so the caller doesn't need to serialize them.
2160  * The size of the spec_array and target_params_array is given by
2161  * @dmi->target_count.
2162  * This function is supposed to be called in early boot, so locking mechanisms
2163  * to protect against concurrent loads are not required.
2164  */
2165 int __init dm_early_create(struct dm_ioctl *dmi,
2166 			   struct dm_target_spec **spec_array,
2167 			   char **target_params_array)
2168 {
2169 	int r, m = DM_ANY_MINOR;
2170 	struct dm_table *t, *old_map;
2171 	struct mapped_device *md;
2172 	unsigned int i;
2173 
2174 	if (!dmi->target_count)
2175 		return -EINVAL;
2176 
2177 	r = check_name(dmi->name);
2178 	if (r)
2179 		return r;
2180 
2181 	if (dmi->flags & DM_PERSISTENT_DEV_FLAG)
2182 		m = MINOR(huge_decode_dev(dmi->dev));
2183 
2184 	/* alloc dm device */
2185 	r = dm_create(m, &md);
2186 	if (r)
2187 		return r;
2188 
2189 	/* hash insert */
2190 	r = dm_hash_insert(dmi->name, *dmi->uuid ? dmi->uuid : NULL, md);
2191 	if (r)
2192 		goto err_destroy_dm;
2193 
2194 	/* alloc table */
2195 	r = dm_table_create(&t, get_mode(dmi), dmi->target_count, md);
2196 	if (r)
2197 		goto err_hash_remove;
2198 
2199 	/* add targets */
2200 	for (i = 0; i < dmi->target_count; i++) {
2201 		r = dm_table_add_target(t, spec_array[i]->target_type,
2202 					(sector_t) spec_array[i]->sector_start,
2203 					(sector_t) spec_array[i]->length,
2204 					target_params_array[i]);
2205 		if (r) {
2206 			DMERR("error adding target to table");
2207 			goto err_destroy_table;
2208 		}
2209 	}
2210 
2211 	/* finish table */
2212 	r = dm_table_complete(t);
2213 	if (r)
2214 		goto err_destroy_table;
2215 
2216 	/* setup md->queue to reflect md's type (may block) */
2217 	r = dm_setup_md_queue(md, t);
2218 	if (r) {
2219 		DMERR("unable to set up device queue for new table.");
2220 		goto err_destroy_table;
2221 	}
2222 
2223 	/* Set new map */
2224 	dm_suspend(md, 0);
2225 	old_map = dm_swap_table(md, t);
2226 	if (IS_ERR(old_map)) {
2227 		r = PTR_ERR(old_map);
2228 		goto err_destroy_table;
2229 	}
2230 	set_disk_ro(dm_disk(md), !!(dmi->flags & DM_READONLY_FLAG));
2231 
2232 	/* resume device */
2233 	r = dm_resume(md);
2234 	if (r)
2235 		goto err_destroy_table;
2236 
2237 	DMINFO("%s (%s) is ready", md->disk->disk_name, dmi->name);
2238 	dm_put(md);
2239 	return 0;
2240 
2241 err_destroy_table:
2242 	dm_table_destroy(t);
2243 err_hash_remove:
2244 	(void) __hash_remove(__get_name_cell(dmi->name));
2245 	/* release reference from __get_name_cell */
2246 	dm_put(md);
2247 err_destroy_dm:
2248 	dm_put(md);
2249 	dm_destroy(md);
2250 	return r;
2251 }
2252