xref: /openbmc/linux/drivers/md/dm-ioctl.c (revision 384740dc)
1 /*
2  * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3  * Copyright (C) 2004 - 2006 Red Hat, Inc. All rights reserved.
4  *
5  * This file is released under the GPL.
6  */
7 
8 #include "dm.h"
9 
10 #include <linux/module.h>
11 #include <linux/vmalloc.h>
12 #include <linux/miscdevice.h>
13 #include <linux/init.h>
14 #include <linux/wait.h>
15 #include <linux/slab.h>
16 #include <linux/dm-ioctl.h>
17 #include <linux/hdreg.h>
18 #include <linux/compat.h>
19 
20 #include <asm/uaccess.h>
21 
22 #define DM_MSG_PREFIX "ioctl"
23 #define DM_DRIVER_EMAIL "dm-devel@redhat.com"
24 
25 /*-----------------------------------------------------------------
26  * The ioctl interface needs to be able to look up devices by
27  * name or uuid.
28  *---------------------------------------------------------------*/
29 struct hash_cell {
30 	struct list_head name_list;
31 	struct list_head uuid_list;
32 
33 	char *name;
34 	char *uuid;
35 	struct mapped_device *md;
36 	struct dm_table *new_map;
37 };
38 
39 struct vers_iter {
40     size_t param_size;
41     struct dm_target_versions *vers, *old_vers;
42     char *end;
43     uint32_t flags;
44 };
45 
46 
47 #define NUM_BUCKETS 64
48 #define MASK_BUCKETS (NUM_BUCKETS - 1)
49 static struct list_head _name_buckets[NUM_BUCKETS];
50 static struct list_head _uuid_buckets[NUM_BUCKETS];
51 
52 static void dm_hash_remove_all(int keep_open_devices);
53 
54 /*
55  * Guards access to both hash tables.
56  */
57 static DECLARE_RWSEM(_hash_lock);
58 
59 static void init_buckets(struct list_head *buckets)
60 {
61 	unsigned int i;
62 
63 	for (i = 0; i < NUM_BUCKETS; i++)
64 		INIT_LIST_HEAD(buckets + i);
65 }
66 
67 static int dm_hash_init(void)
68 {
69 	init_buckets(_name_buckets);
70 	init_buckets(_uuid_buckets);
71 	return 0;
72 }
73 
74 static void dm_hash_exit(void)
75 {
76 	dm_hash_remove_all(0);
77 }
78 
79 /*-----------------------------------------------------------------
80  * Hash function:
81  * We're not really concerned with the str hash function being
82  * fast since it's only used by the ioctl interface.
83  *---------------------------------------------------------------*/
84 static unsigned int hash_str(const char *str)
85 {
86 	const unsigned int hash_mult = 2654435387U;
87 	unsigned int h = 0;
88 
89 	while (*str)
90 		h = (h + (unsigned int) *str++) * hash_mult;
91 
92 	return h & MASK_BUCKETS;
93 }
94 
95 /*-----------------------------------------------------------------
96  * Code for looking up a device by name
97  *---------------------------------------------------------------*/
98 static struct hash_cell *__get_name_cell(const char *str)
99 {
100 	struct hash_cell *hc;
101 	unsigned int h = hash_str(str);
102 
103 	list_for_each_entry (hc, _name_buckets + h, name_list)
104 		if (!strcmp(hc->name, str)) {
105 			dm_get(hc->md);
106 			return hc;
107 		}
108 
109 	return NULL;
110 }
111 
112 static struct hash_cell *__get_uuid_cell(const char *str)
113 {
114 	struct hash_cell *hc;
115 	unsigned int h = hash_str(str);
116 
117 	list_for_each_entry (hc, _uuid_buckets + h, uuid_list)
118 		if (!strcmp(hc->uuid, str)) {
119 			dm_get(hc->md);
120 			return hc;
121 		}
122 
123 	return NULL;
124 }
125 
126 /*-----------------------------------------------------------------
127  * Inserting, removing and renaming a device.
128  *---------------------------------------------------------------*/
129 static struct hash_cell *alloc_cell(const char *name, const char *uuid,
130 				    struct mapped_device *md)
131 {
132 	struct hash_cell *hc;
133 
134 	hc = kmalloc(sizeof(*hc), GFP_KERNEL);
135 	if (!hc)
136 		return NULL;
137 
138 	hc->name = kstrdup(name, GFP_KERNEL);
139 	if (!hc->name) {
140 		kfree(hc);
141 		return NULL;
142 	}
143 
144 	if (!uuid)
145 		hc->uuid = NULL;
146 
147 	else {
148 		hc->uuid = kstrdup(uuid, GFP_KERNEL);
149 		if (!hc->uuid) {
150 			kfree(hc->name);
151 			kfree(hc);
152 			return NULL;
153 		}
154 	}
155 
156 	INIT_LIST_HEAD(&hc->name_list);
157 	INIT_LIST_HEAD(&hc->uuid_list);
158 	hc->md = md;
159 	hc->new_map = NULL;
160 	return hc;
161 }
162 
163 static void free_cell(struct hash_cell *hc)
164 {
165 	if (hc) {
166 		kfree(hc->name);
167 		kfree(hc->uuid);
168 		kfree(hc);
169 	}
170 }
171 
172 /*
173  * The kdev_t and uuid of a device can never change once it is
174  * initially inserted.
175  */
176 static int dm_hash_insert(const char *name, const char *uuid, struct mapped_device *md)
177 {
178 	struct hash_cell *cell, *hc;
179 
180 	/*
181 	 * Allocate the new cells.
182 	 */
183 	cell = alloc_cell(name, uuid, md);
184 	if (!cell)
185 		return -ENOMEM;
186 
187 	/*
188 	 * Insert the cell into both hash tables.
189 	 */
190 	down_write(&_hash_lock);
191 	hc = __get_name_cell(name);
192 	if (hc) {
193 		dm_put(hc->md);
194 		goto bad;
195 	}
196 
197 	list_add(&cell->name_list, _name_buckets + hash_str(name));
198 
199 	if (uuid) {
200 		hc = __get_uuid_cell(uuid);
201 		if (hc) {
202 			list_del(&cell->name_list);
203 			dm_put(hc->md);
204 			goto bad;
205 		}
206 		list_add(&cell->uuid_list, _uuid_buckets + hash_str(uuid));
207 	}
208 	dm_get(md);
209 	dm_set_mdptr(md, cell);
210 	up_write(&_hash_lock);
211 
212 	return 0;
213 
214  bad:
215 	up_write(&_hash_lock);
216 	free_cell(cell);
217 	return -EBUSY;
218 }
219 
220 static void __hash_remove(struct hash_cell *hc)
221 {
222 	struct dm_table *table;
223 
224 	/* remove from the dev hash */
225 	list_del(&hc->uuid_list);
226 	list_del(&hc->name_list);
227 	dm_set_mdptr(hc->md, NULL);
228 
229 	table = dm_get_table(hc->md);
230 	if (table) {
231 		dm_table_event(table);
232 		dm_table_put(table);
233 	}
234 
235 	if (hc->new_map)
236 		dm_table_put(hc->new_map);
237 	dm_put(hc->md);
238 	free_cell(hc);
239 }
240 
241 static void dm_hash_remove_all(int keep_open_devices)
242 {
243 	int i, dev_skipped, dev_removed;
244 	struct hash_cell *hc;
245 	struct list_head *tmp, *n;
246 
247 	down_write(&_hash_lock);
248 
249 retry:
250 	dev_skipped = dev_removed = 0;
251 	for (i = 0; i < NUM_BUCKETS; i++) {
252 		list_for_each_safe (tmp, n, _name_buckets + i) {
253 			hc = list_entry(tmp, struct hash_cell, name_list);
254 
255 			if (keep_open_devices &&
256 			    dm_lock_for_deletion(hc->md)) {
257 				dev_skipped++;
258 				continue;
259 			}
260 			__hash_remove(hc);
261 			dev_removed = 1;
262 		}
263 	}
264 
265 	/*
266 	 * Some mapped devices may be using other mapped devices, so if any
267 	 * still exist, repeat until we make no further progress.
268 	 */
269 	if (dev_skipped) {
270 		if (dev_removed)
271 			goto retry;
272 
273 		DMWARN("remove_all left %d open device(s)", dev_skipped);
274 	}
275 
276 	up_write(&_hash_lock);
277 }
278 
279 static int dm_hash_rename(const char *old, const char *new)
280 {
281 	char *new_name, *old_name;
282 	struct hash_cell *hc;
283 	struct dm_table *table;
284 
285 	/*
286 	 * duplicate new.
287 	 */
288 	new_name = kstrdup(new, GFP_KERNEL);
289 	if (!new_name)
290 		return -ENOMEM;
291 
292 	down_write(&_hash_lock);
293 
294 	/*
295 	 * Is new free ?
296 	 */
297 	hc = __get_name_cell(new);
298 	if (hc) {
299 		DMWARN("asked to rename to an already existing name %s -> %s",
300 		       old, new);
301 		dm_put(hc->md);
302 		up_write(&_hash_lock);
303 		kfree(new_name);
304 		return -EBUSY;
305 	}
306 
307 	/*
308 	 * Is there such a device as 'old' ?
309 	 */
310 	hc = __get_name_cell(old);
311 	if (!hc) {
312 		DMWARN("asked to rename a non existent device %s -> %s",
313 		       old, new);
314 		up_write(&_hash_lock);
315 		kfree(new_name);
316 		return -ENXIO;
317 	}
318 
319 	/*
320 	 * rename and move the name cell.
321 	 */
322 	list_del(&hc->name_list);
323 	old_name = hc->name;
324 	hc->name = new_name;
325 	list_add(&hc->name_list, _name_buckets + hash_str(new_name));
326 
327 	/*
328 	 * Wake up any dm event waiters.
329 	 */
330 	table = dm_get_table(hc->md);
331 	if (table) {
332 		dm_table_event(table);
333 		dm_table_put(table);
334 	}
335 
336 	dm_kobject_uevent(hc->md);
337 
338 	dm_put(hc->md);
339 	up_write(&_hash_lock);
340 	kfree(old_name);
341 	return 0;
342 }
343 
344 /*-----------------------------------------------------------------
345  * Implementation of the ioctl commands
346  *---------------------------------------------------------------*/
347 /*
348  * All the ioctl commands get dispatched to functions with this
349  * prototype.
350  */
351 typedef int (*ioctl_fn)(struct dm_ioctl *param, size_t param_size);
352 
353 static int remove_all(struct dm_ioctl *param, size_t param_size)
354 {
355 	dm_hash_remove_all(1);
356 	param->data_size = 0;
357 	return 0;
358 }
359 
360 /*
361  * Round up the ptr to an 8-byte boundary.
362  */
363 #define ALIGN_MASK 7
364 static inline void *align_ptr(void *ptr)
365 {
366 	return (void *) (((size_t) (ptr + ALIGN_MASK)) & ~ALIGN_MASK);
367 }
368 
369 /*
370  * Retrieves the data payload buffer from an already allocated
371  * struct dm_ioctl.
372  */
373 static void *get_result_buffer(struct dm_ioctl *param, size_t param_size,
374 			       size_t *len)
375 {
376 	param->data_start = align_ptr(param + 1) - (void *) param;
377 
378 	if (param->data_start < param_size)
379 		*len = param_size - param->data_start;
380 	else
381 		*len = 0;
382 
383 	return ((void *) param) + param->data_start;
384 }
385 
386 static int list_devices(struct dm_ioctl *param, size_t param_size)
387 {
388 	unsigned int i;
389 	struct hash_cell *hc;
390 	size_t len, needed = 0;
391 	struct gendisk *disk;
392 	struct dm_name_list *nl, *old_nl = NULL;
393 
394 	down_write(&_hash_lock);
395 
396 	/*
397 	 * Loop through all the devices working out how much
398 	 * space we need.
399 	 */
400 	for (i = 0; i < NUM_BUCKETS; i++) {
401 		list_for_each_entry (hc, _name_buckets + i, name_list) {
402 			needed += sizeof(struct dm_name_list);
403 			needed += strlen(hc->name) + 1;
404 			needed += ALIGN_MASK;
405 		}
406 	}
407 
408 	/*
409 	 * Grab our output buffer.
410 	 */
411 	nl = get_result_buffer(param, param_size, &len);
412 	if (len < needed) {
413 		param->flags |= DM_BUFFER_FULL_FLAG;
414 		goto out;
415 	}
416 	param->data_size = param->data_start + needed;
417 
418 	nl->dev = 0;	/* Flags no data */
419 
420 	/*
421 	 * Now loop through filling out the names.
422 	 */
423 	for (i = 0; i < NUM_BUCKETS; i++) {
424 		list_for_each_entry (hc, _name_buckets + i, name_list) {
425 			if (old_nl)
426 				old_nl->next = (uint32_t) ((void *) nl -
427 							   (void *) old_nl);
428 			disk = dm_disk(hc->md);
429 			nl->dev = huge_encode_dev(disk_devt(disk));
430 			nl->next = 0;
431 			strcpy(nl->name, hc->name);
432 
433 			old_nl = nl;
434 			nl = align_ptr(((void *) ++nl) + strlen(hc->name) + 1);
435 		}
436 	}
437 
438  out:
439 	up_write(&_hash_lock);
440 	return 0;
441 }
442 
443 static void list_version_get_needed(struct target_type *tt, void *needed_param)
444 {
445     size_t *needed = needed_param;
446 
447     *needed += sizeof(struct dm_target_versions);
448     *needed += strlen(tt->name);
449     *needed += ALIGN_MASK;
450 }
451 
452 static void list_version_get_info(struct target_type *tt, void *param)
453 {
454     struct vers_iter *info = param;
455 
456     /* Check space - it might have changed since the first iteration */
457     if ((char *)info->vers + sizeof(tt->version) + strlen(tt->name) + 1 >
458 	info->end) {
459 
460 	info->flags = DM_BUFFER_FULL_FLAG;
461 	return;
462     }
463 
464     if (info->old_vers)
465 	info->old_vers->next = (uint32_t) ((void *)info->vers -
466 					   (void *)info->old_vers);
467     info->vers->version[0] = tt->version[0];
468     info->vers->version[1] = tt->version[1];
469     info->vers->version[2] = tt->version[2];
470     info->vers->next = 0;
471     strcpy(info->vers->name, tt->name);
472 
473     info->old_vers = info->vers;
474     info->vers = align_ptr(((void *) ++info->vers) + strlen(tt->name) + 1);
475 }
476 
477 static int list_versions(struct dm_ioctl *param, size_t param_size)
478 {
479 	size_t len, needed = 0;
480 	struct dm_target_versions *vers;
481 	struct vers_iter iter_info;
482 
483 	/*
484 	 * Loop through all the devices working out how much
485 	 * space we need.
486 	 */
487 	dm_target_iterate(list_version_get_needed, &needed);
488 
489 	/*
490 	 * Grab our output buffer.
491 	 */
492 	vers = get_result_buffer(param, param_size, &len);
493 	if (len < needed) {
494 		param->flags |= DM_BUFFER_FULL_FLAG;
495 		goto out;
496 	}
497 	param->data_size = param->data_start + needed;
498 
499 	iter_info.param_size = param_size;
500 	iter_info.old_vers = NULL;
501 	iter_info.vers = vers;
502 	iter_info.flags = 0;
503 	iter_info.end = (char *)vers+len;
504 
505 	/*
506 	 * Now loop through filling out the names & versions.
507 	 */
508 	dm_target_iterate(list_version_get_info, &iter_info);
509 	param->flags |= iter_info.flags;
510 
511  out:
512 	return 0;
513 }
514 
515 
516 
517 static int check_name(const char *name)
518 {
519 	if (strchr(name, '/')) {
520 		DMWARN("invalid device name");
521 		return -EINVAL;
522 	}
523 
524 	return 0;
525 }
526 
527 /*
528  * Fills in a dm_ioctl structure, ready for sending back to
529  * userland.
530  */
531 static int __dev_status(struct mapped_device *md, struct dm_ioctl *param)
532 {
533 	struct gendisk *disk = dm_disk(md);
534 	struct dm_table *table;
535 
536 	param->flags &= ~(DM_SUSPEND_FLAG | DM_READONLY_FLAG |
537 			  DM_ACTIVE_PRESENT_FLAG);
538 
539 	if (dm_suspended(md))
540 		param->flags |= DM_SUSPEND_FLAG;
541 
542 	param->dev = huge_encode_dev(disk_devt(disk));
543 
544 	/*
545 	 * Yes, this will be out of date by the time it gets back
546 	 * to userland, but it is still very useful for
547 	 * debugging.
548 	 */
549 	param->open_count = dm_open_count(md);
550 
551 	if (get_disk_ro(disk))
552 		param->flags |= DM_READONLY_FLAG;
553 
554 	param->event_nr = dm_get_event_nr(md);
555 
556 	table = dm_get_table(md);
557 	if (table) {
558 		param->flags |= DM_ACTIVE_PRESENT_FLAG;
559 		param->target_count = dm_table_get_num_targets(table);
560 		dm_table_put(table);
561 	} else
562 		param->target_count = 0;
563 
564 	return 0;
565 }
566 
567 static int dev_create(struct dm_ioctl *param, size_t param_size)
568 {
569 	int r, m = DM_ANY_MINOR;
570 	struct mapped_device *md;
571 
572 	r = check_name(param->name);
573 	if (r)
574 		return r;
575 
576 	if (param->flags & DM_PERSISTENT_DEV_FLAG)
577 		m = MINOR(huge_decode_dev(param->dev));
578 
579 	r = dm_create(m, &md);
580 	if (r)
581 		return r;
582 
583 	r = dm_hash_insert(param->name, *param->uuid ? param->uuid : NULL, md);
584 	if (r) {
585 		dm_put(md);
586 		return r;
587 	}
588 
589 	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
590 
591 	r = __dev_status(md, param);
592 	dm_put(md);
593 
594 	return r;
595 }
596 
597 /*
598  * Always use UUID for lookups if it's present, otherwise use name or dev.
599  */
600 static struct hash_cell *__find_device_hash_cell(struct dm_ioctl *param)
601 {
602 	struct mapped_device *md;
603 	void *mdptr = NULL;
604 
605 	if (*param->uuid)
606 		return __get_uuid_cell(param->uuid);
607 
608 	if (*param->name)
609 		return __get_name_cell(param->name);
610 
611 	md = dm_get_md(huge_decode_dev(param->dev));
612 	if (!md)
613 		goto out;
614 
615 	mdptr = dm_get_mdptr(md);
616 	if (!mdptr)
617 		dm_put(md);
618 
619 out:
620 	return mdptr;
621 }
622 
623 static struct mapped_device *find_device(struct dm_ioctl *param)
624 {
625 	struct hash_cell *hc;
626 	struct mapped_device *md = NULL;
627 
628 	down_read(&_hash_lock);
629 	hc = __find_device_hash_cell(param);
630 	if (hc) {
631 		md = hc->md;
632 
633 		/*
634 		 * Sneakily write in both the name and the uuid
635 		 * while we have the cell.
636 		 */
637 		strncpy(param->name, hc->name, sizeof(param->name));
638 		if (hc->uuid)
639 			strncpy(param->uuid, hc->uuid, sizeof(param->uuid)-1);
640 		else
641 			param->uuid[0] = '\0';
642 
643 		if (hc->new_map)
644 			param->flags |= DM_INACTIVE_PRESENT_FLAG;
645 		else
646 			param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
647 	}
648 	up_read(&_hash_lock);
649 
650 	return md;
651 }
652 
653 static int dev_remove(struct dm_ioctl *param, size_t param_size)
654 {
655 	struct hash_cell *hc;
656 	struct mapped_device *md;
657 	int r;
658 
659 	down_write(&_hash_lock);
660 	hc = __find_device_hash_cell(param);
661 
662 	if (!hc) {
663 		DMWARN("device doesn't appear to be in the dev hash table.");
664 		up_write(&_hash_lock);
665 		return -ENXIO;
666 	}
667 
668 	md = hc->md;
669 
670 	/*
671 	 * Ensure the device is not open and nothing further can open it.
672 	 */
673 	r = dm_lock_for_deletion(md);
674 	if (r) {
675 		DMWARN("unable to remove open device %s", hc->name);
676 		up_write(&_hash_lock);
677 		dm_put(md);
678 		return r;
679 	}
680 
681 	__hash_remove(hc);
682 	up_write(&_hash_lock);
683 	dm_put(md);
684 	param->data_size = 0;
685 	return 0;
686 }
687 
688 /*
689  * Check a string doesn't overrun the chunk of
690  * memory we copied from userland.
691  */
692 static int invalid_str(char *str, void *end)
693 {
694 	while ((void *) str < end)
695 		if (!*str++)
696 			return 0;
697 
698 	return -EINVAL;
699 }
700 
701 static int dev_rename(struct dm_ioctl *param, size_t param_size)
702 {
703 	int r;
704 	char *new_name = (char *) param + param->data_start;
705 
706 	if (new_name < param->data ||
707 	    invalid_str(new_name, (void *) param + param_size)) {
708 		DMWARN("Invalid new logical volume name supplied.");
709 		return -EINVAL;
710 	}
711 
712 	r = check_name(new_name);
713 	if (r)
714 		return r;
715 
716 	param->data_size = 0;
717 	return dm_hash_rename(param->name, new_name);
718 }
719 
720 static int dev_set_geometry(struct dm_ioctl *param, size_t param_size)
721 {
722 	int r = -EINVAL, x;
723 	struct mapped_device *md;
724 	struct hd_geometry geometry;
725 	unsigned long indata[4];
726 	char *geostr = (char *) param + param->data_start;
727 
728 	md = find_device(param);
729 	if (!md)
730 		return -ENXIO;
731 
732 	if (geostr < param->data ||
733 	    invalid_str(geostr, (void *) param + param_size)) {
734 		DMWARN("Invalid geometry supplied.");
735 		goto out;
736 	}
737 
738 	x = sscanf(geostr, "%lu %lu %lu %lu", indata,
739 		   indata + 1, indata + 2, indata + 3);
740 
741 	if (x != 4) {
742 		DMWARN("Unable to interpret geometry settings.");
743 		goto out;
744 	}
745 
746 	if (indata[0] > 65535 || indata[1] > 255 ||
747 	    indata[2] > 255 || indata[3] > ULONG_MAX) {
748 		DMWARN("Geometry exceeds range limits.");
749 		goto out;
750 	}
751 
752 	geometry.cylinders = indata[0];
753 	geometry.heads = indata[1];
754 	geometry.sectors = indata[2];
755 	geometry.start = indata[3];
756 
757 	r = dm_set_geometry(md, &geometry);
758 	if (!r)
759 		r = __dev_status(md, param);
760 
761 	param->data_size = 0;
762 
763 out:
764 	dm_put(md);
765 	return r;
766 }
767 
768 static int do_suspend(struct dm_ioctl *param)
769 {
770 	int r = 0;
771 	unsigned suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
772 	struct mapped_device *md;
773 
774 	md = find_device(param);
775 	if (!md)
776 		return -ENXIO;
777 
778 	if (param->flags & DM_SKIP_LOCKFS_FLAG)
779 		suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
780 	if (param->flags & DM_NOFLUSH_FLAG)
781 		suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
782 
783 	if (!dm_suspended(md))
784 		r = dm_suspend(md, suspend_flags);
785 
786 	if (!r)
787 		r = __dev_status(md, param);
788 
789 	dm_put(md);
790 	return r;
791 }
792 
793 static int do_resume(struct dm_ioctl *param)
794 {
795 	int r = 0;
796 	unsigned suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
797 	struct hash_cell *hc;
798 	struct mapped_device *md;
799 	struct dm_table *new_map;
800 
801 	down_write(&_hash_lock);
802 
803 	hc = __find_device_hash_cell(param);
804 	if (!hc) {
805 		DMWARN("device doesn't appear to be in the dev hash table.");
806 		up_write(&_hash_lock);
807 		return -ENXIO;
808 	}
809 
810 	md = hc->md;
811 
812 	new_map = hc->new_map;
813 	hc->new_map = NULL;
814 	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
815 
816 	up_write(&_hash_lock);
817 
818 	/* Do we need to load a new map ? */
819 	if (new_map) {
820 		/* Suspend if it isn't already suspended */
821 		if (param->flags & DM_SKIP_LOCKFS_FLAG)
822 			suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
823 		if (param->flags & DM_NOFLUSH_FLAG)
824 			suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
825 		if (!dm_suspended(md))
826 			dm_suspend(md, suspend_flags);
827 
828 		r = dm_swap_table(md, new_map);
829 		if (r) {
830 			dm_put(md);
831 			dm_table_put(new_map);
832 			return r;
833 		}
834 
835 		if (dm_table_get_mode(new_map) & FMODE_WRITE)
836 			set_disk_ro(dm_disk(md), 0);
837 		else
838 			set_disk_ro(dm_disk(md), 1);
839 
840 		dm_table_put(new_map);
841 	}
842 
843 	if (dm_suspended(md))
844 		r = dm_resume(md);
845 
846 	if (!r)
847 		r = __dev_status(md, param);
848 
849 	dm_put(md);
850 	return r;
851 }
852 
853 /*
854  * Set or unset the suspension state of a device.
855  * If the device already is in the requested state we just return its status.
856  */
857 static int dev_suspend(struct dm_ioctl *param, size_t param_size)
858 {
859 	if (param->flags & DM_SUSPEND_FLAG)
860 		return do_suspend(param);
861 
862 	return do_resume(param);
863 }
864 
865 /*
866  * Copies device info back to user space, used by
867  * the create and info ioctls.
868  */
869 static int dev_status(struct dm_ioctl *param, size_t param_size)
870 {
871 	int r;
872 	struct mapped_device *md;
873 
874 	md = find_device(param);
875 	if (!md)
876 		return -ENXIO;
877 
878 	r = __dev_status(md, param);
879 	dm_put(md);
880 	return r;
881 }
882 
883 /*
884  * Build up the status struct for each target
885  */
886 static void retrieve_status(struct dm_table *table,
887 			    struct dm_ioctl *param, size_t param_size)
888 {
889 	unsigned int i, num_targets;
890 	struct dm_target_spec *spec;
891 	char *outbuf, *outptr;
892 	status_type_t type;
893 	size_t remaining, len, used = 0;
894 
895 	outptr = outbuf = get_result_buffer(param, param_size, &len);
896 
897 	if (param->flags & DM_STATUS_TABLE_FLAG)
898 		type = STATUSTYPE_TABLE;
899 	else
900 		type = STATUSTYPE_INFO;
901 
902 	/* Get all the target info */
903 	num_targets = dm_table_get_num_targets(table);
904 	for (i = 0; i < num_targets; i++) {
905 		struct dm_target *ti = dm_table_get_target(table, i);
906 
907 		remaining = len - (outptr - outbuf);
908 		if (remaining <= sizeof(struct dm_target_spec)) {
909 			param->flags |= DM_BUFFER_FULL_FLAG;
910 			break;
911 		}
912 
913 		spec = (struct dm_target_spec *) outptr;
914 
915 		spec->status = 0;
916 		spec->sector_start = ti->begin;
917 		spec->length = ti->len;
918 		strncpy(spec->target_type, ti->type->name,
919 			sizeof(spec->target_type));
920 
921 		outptr += sizeof(struct dm_target_spec);
922 		remaining = len - (outptr - outbuf);
923 		if (remaining <= 0) {
924 			param->flags |= DM_BUFFER_FULL_FLAG;
925 			break;
926 		}
927 
928 		/* Get the status/table string from the target driver */
929 		if (ti->type->status) {
930 			if (ti->type->status(ti, type, outptr, remaining)) {
931 				param->flags |= DM_BUFFER_FULL_FLAG;
932 				break;
933 			}
934 		} else
935 			outptr[0] = '\0';
936 
937 		outptr += strlen(outptr) + 1;
938 		used = param->data_start + (outptr - outbuf);
939 
940 		outptr = align_ptr(outptr);
941 		spec->next = outptr - outbuf;
942 	}
943 
944 	if (used)
945 		param->data_size = used;
946 
947 	param->target_count = num_targets;
948 }
949 
950 /*
951  * Wait for a device to report an event
952  */
953 static int dev_wait(struct dm_ioctl *param, size_t param_size)
954 {
955 	int r;
956 	struct mapped_device *md;
957 	struct dm_table *table;
958 
959 	md = find_device(param);
960 	if (!md)
961 		return -ENXIO;
962 
963 	/*
964 	 * Wait for a notification event
965 	 */
966 	if (dm_wait_event(md, param->event_nr)) {
967 		r = -ERESTARTSYS;
968 		goto out;
969 	}
970 
971 	/*
972 	 * The userland program is going to want to know what
973 	 * changed to trigger the event, so we may as well tell
974 	 * him and save an ioctl.
975 	 */
976 	r = __dev_status(md, param);
977 	if (r)
978 		goto out;
979 
980 	table = dm_get_table(md);
981 	if (table) {
982 		retrieve_status(table, param, param_size);
983 		dm_table_put(table);
984 	}
985 
986  out:
987 	dm_put(md);
988 	return r;
989 }
990 
991 static inline int get_mode(struct dm_ioctl *param)
992 {
993 	int mode = FMODE_READ | FMODE_WRITE;
994 
995 	if (param->flags & DM_READONLY_FLAG)
996 		mode = FMODE_READ;
997 
998 	return mode;
999 }
1000 
1001 static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
1002 		       struct dm_target_spec **spec, char **target_params)
1003 {
1004 	*spec = (struct dm_target_spec *) ((unsigned char *) last + next);
1005 	*target_params = (char *) (*spec + 1);
1006 
1007 	if (*spec < (last + 1))
1008 		return -EINVAL;
1009 
1010 	return invalid_str(*target_params, end);
1011 }
1012 
1013 static int populate_table(struct dm_table *table,
1014 			  struct dm_ioctl *param, size_t param_size)
1015 {
1016 	int r;
1017 	unsigned int i = 0;
1018 	struct dm_target_spec *spec = (struct dm_target_spec *) param;
1019 	uint32_t next = param->data_start;
1020 	void *end = (void *) param + param_size;
1021 	char *target_params;
1022 
1023 	if (!param->target_count) {
1024 		DMWARN("populate_table: no targets specified");
1025 		return -EINVAL;
1026 	}
1027 
1028 	for (i = 0; i < param->target_count; i++) {
1029 
1030 		r = next_target(spec, next, end, &spec, &target_params);
1031 		if (r) {
1032 			DMWARN("unable to find target");
1033 			return r;
1034 		}
1035 
1036 		r = dm_table_add_target(table, spec->target_type,
1037 					(sector_t) spec->sector_start,
1038 					(sector_t) spec->length,
1039 					target_params);
1040 		if (r) {
1041 			DMWARN("error adding target to table");
1042 			return r;
1043 		}
1044 
1045 		next = spec->next;
1046 	}
1047 
1048 	return dm_table_complete(table);
1049 }
1050 
1051 static int table_load(struct dm_ioctl *param, size_t param_size)
1052 {
1053 	int r;
1054 	struct hash_cell *hc;
1055 	struct dm_table *t;
1056 	struct mapped_device *md;
1057 
1058 	md = find_device(param);
1059 	if (!md)
1060 		return -ENXIO;
1061 
1062 	r = dm_table_create(&t, get_mode(param), param->target_count, md);
1063 	if (r)
1064 		goto out;
1065 
1066 	r = populate_table(t, param, param_size);
1067 	if (r) {
1068 		dm_table_put(t);
1069 		goto out;
1070 	}
1071 
1072 	down_write(&_hash_lock);
1073 	hc = dm_get_mdptr(md);
1074 	if (!hc || hc->md != md) {
1075 		DMWARN("device has been removed from the dev hash table.");
1076 		dm_table_put(t);
1077 		up_write(&_hash_lock);
1078 		r = -ENXIO;
1079 		goto out;
1080 	}
1081 
1082 	if (hc->new_map)
1083 		dm_table_put(hc->new_map);
1084 	hc->new_map = t;
1085 	up_write(&_hash_lock);
1086 
1087 	param->flags |= DM_INACTIVE_PRESENT_FLAG;
1088 	r = __dev_status(md, param);
1089 
1090 out:
1091 	dm_put(md);
1092 
1093 	return r;
1094 }
1095 
1096 static int table_clear(struct dm_ioctl *param, size_t param_size)
1097 {
1098 	int r;
1099 	struct hash_cell *hc;
1100 	struct mapped_device *md;
1101 
1102 	down_write(&_hash_lock);
1103 
1104 	hc = __find_device_hash_cell(param);
1105 	if (!hc) {
1106 		DMWARN("device doesn't appear to be in the dev hash table.");
1107 		up_write(&_hash_lock);
1108 		return -ENXIO;
1109 	}
1110 
1111 	if (hc->new_map) {
1112 		dm_table_put(hc->new_map);
1113 		hc->new_map = NULL;
1114 	}
1115 
1116 	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1117 
1118 	r = __dev_status(hc->md, param);
1119 	md = hc->md;
1120 	up_write(&_hash_lock);
1121 	dm_put(md);
1122 	return r;
1123 }
1124 
1125 /*
1126  * Retrieves a list of devices used by a particular dm device.
1127  */
1128 static void retrieve_deps(struct dm_table *table,
1129 			  struct dm_ioctl *param, size_t param_size)
1130 {
1131 	unsigned int count = 0;
1132 	struct list_head *tmp;
1133 	size_t len, needed;
1134 	struct dm_dev_internal *dd;
1135 	struct dm_target_deps *deps;
1136 
1137 	deps = get_result_buffer(param, param_size, &len);
1138 
1139 	/*
1140 	 * Count the devices.
1141 	 */
1142 	list_for_each (tmp, dm_table_get_devices(table))
1143 		count++;
1144 
1145 	/*
1146 	 * Check we have enough space.
1147 	 */
1148 	needed = sizeof(*deps) + (sizeof(*deps->dev) * count);
1149 	if (len < needed) {
1150 		param->flags |= DM_BUFFER_FULL_FLAG;
1151 		return;
1152 	}
1153 
1154 	/*
1155 	 * Fill in the devices.
1156 	 */
1157 	deps->count = count;
1158 	count = 0;
1159 	list_for_each_entry (dd, dm_table_get_devices(table), list)
1160 		deps->dev[count++] = huge_encode_dev(dd->dm_dev.bdev->bd_dev);
1161 
1162 	param->data_size = param->data_start + needed;
1163 }
1164 
1165 static int table_deps(struct dm_ioctl *param, size_t param_size)
1166 {
1167 	int r = 0;
1168 	struct mapped_device *md;
1169 	struct dm_table *table;
1170 
1171 	md = find_device(param);
1172 	if (!md)
1173 		return -ENXIO;
1174 
1175 	r = __dev_status(md, param);
1176 	if (r)
1177 		goto out;
1178 
1179 	table = dm_get_table(md);
1180 	if (table) {
1181 		retrieve_deps(table, param, param_size);
1182 		dm_table_put(table);
1183 	}
1184 
1185  out:
1186 	dm_put(md);
1187 	return r;
1188 }
1189 
1190 /*
1191  * Return the status of a device as a text string for each
1192  * target.
1193  */
1194 static int table_status(struct dm_ioctl *param, size_t param_size)
1195 {
1196 	int r;
1197 	struct mapped_device *md;
1198 	struct dm_table *table;
1199 
1200 	md = find_device(param);
1201 	if (!md)
1202 		return -ENXIO;
1203 
1204 	r = __dev_status(md, param);
1205 	if (r)
1206 		goto out;
1207 
1208 	table = dm_get_table(md);
1209 	if (table) {
1210 		retrieve_status(table, param, param_size);
1211 		dm_table_put(table);
1212 	}
1213 
1214  out:
1215 	dm_put(md);
1216 	return r;
1217 }
1218 
1219 /*
1220  * Pass a message to the target that's at the supplied device offset.
1221  */
1222 static int target_message(struct dm_ioctl *param, size_t param_size)
1223 {
1224 	int r, argc;
1225 	char **argv;
1226 	struct mapped_device *md;
1227 	struct dm_table *table;
1228 	struct dm_target *ti;
1229 	struct dm_target_msg *tmsg = (void *) param + param->data_start;
1230 
1231 	md = find_device(param);
1232 	if (!md)
1233 		return -ENXIO;
1234 
1235 	r = __dev_status(md, param);
1236 	if (r)
1237 		goto out;
1238 
1239 	if (tmsg < (struct dm_target_msg *) param->data ||
1240 	    invalid_str(tmsg->message, (void *) param + param_size)) {
1241 		DMWARN("Invalid target message parameters.");
1242 		r = -EINVAL;
1243 		goto out;
1244 	}
1245 
1246 	r = dm_split_args(&argc, &argv, tmsg->message);
1247 	if (r) {
1248 		DMWARN("Failed to split target message parameters");
1249 		goto out;
1250 	}
1251 
1252 	table = dm_get_table(md);
1253 	if (!table)
1254 		goto out_argv;
1255 
1256 	ti = dm_table_find_target(table, tmsg->sector);
1257 	if (!dm_target_is_valid(ti)) {
1258 		DMWARN("Target message sector outside device.");
1259 		r = -EINVAL;
1260 	} else if (ti->type->message)
1261 		r = ti->type->message(ti, argc, argv);
1262 	else {
1263 		DMWARN("Target type does not support messages");
1264 		r = -EINVAL;
1265 	}
1266 
1267 	dm_table_put(table);
1268  out_argv:
1269 	kfree(argv);
1270  out:
1271 	param->data_size = 0;
1272 	dm_put(md);
1273 	return r;
1274 }
1275 
1276 /*-----------------------------------------------------------------
1277  * Implementation of open/close/ioctl on the special char
1278  * device.
1279  *---------------------------------------------------------------*/
1280 static ioctl_fn lookup_ioctl(unsigned int cmd)
1281 {
1282 	static struct {
1283 		int cmd;
1284 		ioctl_fn fn;
1285 	} _ioctls[] = {
1286 		{DM_VERSION_CMD, NULL},	/* version is dealt with elsewhere */
1287 		{DM_REMOVE_ALL_CMD, remove_all},
1288 		{DM_LIST_DEVICES_CMD, list_devices},
1289 
1290 		{DM_DEV_CREATE_CMD, dev_create},
1291 		{DM_DEV_REMOVE_CMD, dev_remove},
1292 		{DM_DEV_RENAME_CMD, dev_rename},
1293 		{DM_DEV_SUSPEND_CMD, dev_suspend},
1294 		{DM_DEV_STATUS_CMD, dev_status},
1295 		{DM_DEV_WAIT_CMD, dev_wait},
1296 
1297 		{DM_TABLE_LOAD_CMD, table_load},
1298 		{DM_TABLE_CLEAR_CMD, table_clear},
1299 		{DM_TABLE_DEPS_CMD, table_deps},
1300 		{DM_TABLE_STATUS_CMD, table_status},
1301 
1302 		{DM_LIST_VERSIONS_CMD, list_versions},
1303 
1304 		{DM_TARGET_MSG_CMD, target_message},
1305 		{DM_DEV_SET_GEOMETRY_CMD, dev_set_geometry}
1306 	};
1307 
1308 	return (cmd >= ARRAY_SIZE(_ioctls)) ? NULL : _ioctls[cmd].fn;
1309 }
1310 
1311 /*
1312  * As well as checking the version compatibility this always
1313  * copies the kernel interface version out.
1314  */
1315 static int check_version(unsigned int cmd, struct dm_ioctl __user *user)
1316 {
1317 	uint32_t version[3];
1318 	int r = 0;
1319 
1320 	if (copy_from_user(version, user->version, sizeof(version)))
1321 		return -EFAULT;
1322 
1323 	if ((DM_VERSION_MAJOR != version[0]) ||
1324 	    (DM_VERSION_MINOR < version[1])) {
1325 		DMWARN("ioctl interface mismatch: "
1326 		       "kernel(%u.%u.%u), user(%u.%u.%u), cmd(%d)",
1327 		       DM_VERSION_MAJOR, DM_VERSION_MINOR,
1328 		       DM_VERSION_PATCHLEVEL,
1329 		       version[0], version[1], version[2], cmd);
1330 		r = -EINVAL;
1331 	}
1332 
1333 	/*
1334 	 * Fill in the kernel version.
1335 	 */
1336 	version[0] = DM_VERSION_MAJOR;
1337 	version[1] = DM_VERSION_MINOR;
1338 	version[2] = DM_VERSION_PATCHLEVEL;
1339 	if (copy_to_user(user->version, version, sizeof(version)))
1340 		return -EFAULT;
1341 
1342 	return r;
1343 }
1344 
1345 static void free_params(struct dm_ioctl *param)
1346 {
1347 	vfree(param);
1348 }
1349 
1350 static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl **param)
1351 {
1352 	struct dm_ioctl tmp, *dmi;
1353 
1354 	if (copy_from_user(&tmp, user, sizeof(tmp) - sizeof(tmp.data)))
1355 		return -EFAULT;
1356 
1357 	if (tmp.data_size < (sizeof(tmp) - sizeof(tmp.data)))
1358 		return -EINVAL;
1359 
1360 	dmi = vmalloc(tmp.data_size);
1361 	if (!dmi)
1362 		return -ENOMEM;
1363 
1364 	if (copy_from_user(dmi, user, tmp.data_size)) {
1365 		vfree(dmi);
1366 		return -EFAULT;
1367 	}
1368 
1369 	*param = dmi;
1370 	return 0;
1371 }
1372 
1373 static int validate_params(uint cmd, struct dm_ioctl *param)
1374 {
1375 	/* Always clear this flag */
1376 	param->flags &= ~DM_BUFFER_FULL_FLAG;
1377 
1378 	/* Ignores parameters */
1379 	if (cmd == DM_REMOVE_ALL_CMD ||
1380 	    cmd == DM_LIST_DEVICES_CMD ||
1381 	    cmd == DM_LIST_VERSIONS_CMD)
1382 		return 0;
1383 
1384 	if ((cmd == DM_DEV_CREATE_CMD)) {
1385 		if (!*param->name) {
1386 			DMWARN("name not supplied when creating device");
1387 			return -EINVAL;
1388 		}
1389 	} else if ((*param->uuid && *param->name)) {
1390 		DMWARN("only supply one of name or uuid, cmd(%u)", cmd);
1391 		return -EINVAL;
1392 	}
1393 
1394 	/* Ensure strings are terminated */
1395 	param->name[DM_NAME_LEN - 1] = '\0';
1396 	param->uuid[DM_UUID_LEN - 1] = '\0';
1397 
1398 	return 0;
1399 }
1400 
1401 static int ctl_ioctl(uint command, struct dm_ioctl __user *user)
1402 {
1403 	int r = 0;
1404 	unsigned int cmd;
1405 	struct dm_ioctl *uninitialized_var(param);
1406 	ioctl_fn fn = NULL;
1407 	size_t param_size;
1408 
1409 	/* only root can play with this */
1410 	if (!capable(CAP_SYS_ADMIN))
1411 		return -EACCES;
1412 
1413 	if (_IOC_TYPE(command) != DM_IOCTL)
1414 		return -ENOTTY;
1415 
1416 	cmd = _IOC_NR(command);
1417 
1418 	/*
1419 	 * Check the interface version passed in.  This also
1420 	 * writes out the kernel's interface version.
1421 	 */
1422 	r = check_version(cmd, user);
1423 	if (r)
1424 		return r;
1425 
1426 	/*
1427 	 * Nothing more to do for the version command.
1428 	 */
1429 	if (cmd == DM_VERSION_CMD)
1430 		return 0;
1431 
1432 	fn = lookup_ioctl(cmd);
1433 	if (!fn) {
1434 		DMWARN("dm_ctl_ioctl: unknown command 0x%x", command);
1435 		return -ENOTTY;
1436 	}
1437 
1438 	/*
1439 	 * Trying to avoid low memory issues when a device is
1440 	 * suspended.
1441 	 */
1442 	current->flags |= PF_MEMALLOC;
1443 
1444 	/*
1445 	 * Copy the parameters into kernel space.
1446 	 */
1447 	r = copy_params(user, &param);
1448 
1449 	current->flags &= ~PF_MEMALLOC;
1450 
1451 	if (r)
1452 		return r;
1453 
1454 	r = validate_params(cmd, param);
1455 	if (r)
1456 		goto out;
1457 
1458 	param_size = param->data_size;
1459 	param->data_size = sizeof(*param);
1460 	r = fn(param, param_size);
1461 
1462 	/*
1463 	 * Copy the results back to userland.
1464 	 */
1465 	if (!r && copy_to_user(user, param, param->data_size))
1466 		r = -EFAULT;
1467 
1468  out:
1469 	free_params(param);
1470 	return r;
1471 }
1472 
1473 static long dm_ctl_ioctl(struct file *file, uint command, ulong u)
1474 {
1475 	return (long)ctl_ioctl(command, (struct dm_ioctl __user *)u);
1476 }
1477 
1478 #ifdef CONFIG_COMPAT
1479 static long dm_compat_ctl_ioctl(struct file *file, uint command, ulong u)
1480 {
1481 	return (long)dm_ctl_ioctl(file, command, (ulong) compat_ptr(u));
1482 }
1483 #else
1484 #define dm_compat_ctl_ioctl NULL
1485 #endif
1486 
1487 static const struct file_operations _ctl_fops = {
1488 	.unlocked_ioctl	 = dm_ctl_ioctl,
1489 	.compat_ioctl = dm_compat_ctl_ioctl,
1490 	.owner	 = THIS_MODULE,
1491 };
1492 
1493 static struct miscdevice _dm_misc = {
1494 	.minor 		= MISC_DYNAMIC_MINOR,
1495 	.name  		= DM_NAME,
1496 	.fops  		= &_ctl_fops
1497 };
1498 
1499 /*
1500  * Create misc character device and link to DM_DIR/control.
1501  */
1502 int __init dm_interface_init(void)
1503 {
1504 	int r;
1505 
1506 	r = dm_hash_init();
1507 	if (r)
1508 		return r;
1509 
1510 	r = misc_register(&_dm_misc);
1511 	if (r) {
1512 		DMERR("misc_register failed for control device");
1513 		dm_hash_exit();
1514 		return r;
1515 	}
1516 
1517 	DMINFO("%d.%d.%d%s initialised: %s", DM_VERSION_MAJOR,
1518 	       DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL, DM_VERSION_EXTRA,
1519 	       DM_DRIVER_EMAIL);
1520 	return 0;
1521 }
1522 
1523 void dm_interface_exit(void)
1524 {
1525 	if (misc_deregister(&_dm_misc) < 0)
1526 		DMERR("misc_deregister failed for control device");
1527 
1528 	dm_hash_exit();
1529 }
1530 
1531 /**
1532  * dm_copy_name_and_uuid - Copy mapped device name & uuid into supplied buffers
1533  * @md: Pointer to mapped_device
1534  * @name: Buffer (size DM_NAME_LEN) for name
1535  * @uuid: Buffer (size DM_UUID_LEN) for uuid or empty string if uuid not defined
1536  */
1537 int dm_copy_name_and_uuid(struct mapped_device *md, char *name, char *uuid)
1538 {
1539 	int r = 0;
1540 	struct hash_cell *hc;
1541 
1542 	if (!md)
1543 		return -ENXIO;
1544 
1545 	dm_get(md);
1546 	down_read(&_hash_lock);
1547 	hc = dm_get_mdptr(md);
1548 	if (!hc || hc->md != md) {
1549 		r = -ENXIO;
1550 		goto out;
1551 	}
1552 
1553 	strcpy(name, hc->name);
1554 	strcpy(uuid, hc->uuid ? : "");
1555 
1556 out:
1557 	up_read(&_hash_lock);
1558 	dm_put(md);
1559 
1560 	return r;
1561 }
1562