xref: /linux/drivers/md/dm-ioctl.c (revision 9ce7677cfd7cd871adb457c80bea3b581b839641)
1 /*
2  * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3  * Copyright (C) 2004 - 2005 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/devfs_fs_kernel.h>
17 #include <linux/dm-ioctl.h>
18 
19 #include <asm/uaccess.h>
20 
21 #define DM_DRIVER_EMAIL "dm-devel@redhat.com"
22 
23 /*-----------------------------------------------------------------
24  * The ioctl interface needs to be able to look up devices by
25  * name or uuid.
26  *---------------------------------------------------------------*/
27 struct hash_cell {
28 	struct list_head name_list;
29 	struct list_head uuid_list;
30 
31 	char *name;
32 	char *uuid;
33 	struct mapped_device *md;
34 	struct dm_table *new_map;
35 };
36 
37 struct vers_iter {
38     size_t param_size;
39     struct dm_target_versions *vers, *old_vers;
40     char *end;
41     uint32_t flags;
42 };
43 
44 
45 #define NUM_BUCKETS 64
46 #define MASK_BUCKETS (NUM_BUCKETS - 1)
47 static struct list_head _name_buckets[NUM_BUCKETS];
48 static struct list_head _uuid_buckets[NUM_BUCKETS];
49 
50 static void dm_hash_remove_all(void);
51 
52 /*
53  * Guards access to both hash tables.
54  */
55 static DECLARE_RWSEM(_hash_lock);
56 
57 static void init_buckets(struct list_head *buckets)
58 {
59 	unsigned int i;
60 
61 	for (i = 0; i < NUM_BUCKETS; i++)
62 		INIT_LIST_HEAD(buckets + i);
63 }
64 
65 static int dm_hash_init(void)
66 {
67 	init_buckets(_name_buckets);
68 	init_buckets(_uuid_buckets);
69 	devfs_mk_dir(DM_DIR);
70 	return 0;
71 }
72 
73 static void dm_hash_exit(void)
74 {
75 	dm_hash_remove_all();
76 	devfs_remove(DM_DIR);
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 			return hc;
106 
107 	return NULL;
108 }
109 
110 static struct hash_cell *__get_uuid_cell(const char *str)
111 {
112 	struct hash_cell *hc;
113 	unsigned int h = hash_str(str);
114 
115 	list_for_each_entry (hc, _uuid_buckets + h, uuid_list)
116 		if (!strcmp(hc->uuid, str))
117 			return hc;
118 
119 	return NULL;
120 }
121 
122 /*-----------------------------------------------------------------
123  * Inserting, removing and renaming a device.
124  *---------------------------------------------------------------*/
125 static struct hash_cell *alloc_cell(const char *name, const char *uuid,
126 				    struct mapped_device *md)
127 {
128 	struct hash_cell *hc;
129 
130 	hc = kmalloc(sizeof(*hc), GFP_KERNEL);
131 	if (!hc)
132 		return NULL;
133 
134 	hc->name = kstrdup(name, GFP_KERNEL);
135 	if (!hc->name) {
136 		kfree(hc);
137 		return NULL;
138 	}
139 
140 	if (!uuid)
141 		hc->uuid = NULL;
142 
143 	else {
144 		hc->uuid = kstrdup(uuid, GFP_KERNEL);
145 		if (!hc->uuid) {
146 			kfree(hc->name);
147 			kfree(hc);
148 			return NULL;
149 		}
150 	}
151 
152 	INIT_LIST_HEAD(&hc->name_list);
153 	INIT_LIST_HEAD(&hc->uuid_list);
154 	hc->md = md;
155 	hc->new_map = NULL;
156 	return hc;
157 }
158 
159 static void free_cell(struct hash_cell *hc)
160 {
161 	if (hc) {
162 		kfree(hc->name);
163 		kfree(hc->uuid);
164 		kfree(hc);
165 	}
166 }
167 
168 /*
169  * devfs stuff.
170  */
171 static int register_with_devfs(struct hash_cell *hc)
172 {
173 	struct gendisk *disk = dm_disk(hc->md);
174 
175 	devfs_mk_bdev(MKDEV(disk->major, disk->first_minor),
176 		      S_IFBLK | S_IRUSR | S_IWUSR | S_IRGRP,
177 		      DM_DIR "/%s", hc->name);
178 	return 0;
179 }
180 
181 static int unregister_with_devfs(struct hash_cell *hc)
182 {
183 	devfs_remove(DM_DIR"/%s", hc->name);
184 	return 0;
185 }
186 
187 /*
188  * The kdev_t and uuid of a device can never change once it is
189  * initially inserted.
190  */
191 static int dm_hash_insert(const char *name, const char *uuid, struct mapped_device *md)
192 {
193 	struct hash_cell *cell;
194 
195 	/*
196 	 * Allocate the new cells.
197 	 */
198 	cell = alloc_cell(name, uuid, md);
199 	if (!cell)
200 		return -ENOMEM;
201 
202 	/*
203 	 * Insert the cell into both hash tables.
204 	 */
205 	down_write(&_hash_lock);
206 	if (__get_name_cell(name))
207 		goto bad;
208 
209 	list_add(&cell->name_list, _name_buckets + hash_str(name));
210 
211 	if (uuid) {
212 		if (__get_uuid_cell(uuid)) {
213 			list_del(&cell->name_list);
214 			goto bad;
215 		}
216 		list_add(&cell->uuid_list, _uuid_buckets + hash_str(uuid));
217 	}
218 	register_with_devfs(cell);
219 	dm_get(md);
220 	dm_set_mdptr(md, cell);
221 	up_write(&_hash_lock);
222 
223 	return 0;
224 
225  bad:
226 	up_write(&_hash_lock);
227 	free_cell(cell);
228 	return -EBUSY;
229 }
230 
231 static void __hash_remove(struct hash_cell *hc)
232 {
233 	struct dm_table *table;
234 
235 	/* remove from the dev hash */
236 	list_del(&hc->uuid_list);
237 	list_del(&hc->name_list);
238 	unregister_with_devfs(hc);
239 	dm_set_mdptr(hc->md, NULL);
240 
241 	table = dm_get_table(hc->md);
242 	if (table) {
243 		dm_table_event(table);
244 		dm_table_put(table);
245 	}
246 
247 	dm_put(hc->md);
248 	if (hc->new_map)
249 		dm_table_put(hc->new_map);
250 	free_cell(hc);
251 }
252 
253 static void dm_hash_remove_all(void)
254 {
255 	int i;
256 	struct hash_cell *hc;
257 	struct list_head *tmp, *n;
258 
259 	down_write(&_hash_lock);
260 	for (i = 0; i < NUM_BUCKETS; i++) {
261 		list_for_each_safe (tmp, n, _name_buckets + i) {
262 			hc = list_entry(tmp, struct hash_cell, name_list);
263 			__hash_remove(hc);
264 		}
265 	}
266 	up_write(&_hash_lock);
267 }
268 
269 static int dm_hash_rename(const char *old, const char *new)
270 {
271 	char *new_name, *old_name;
272 	struct hash_cell *hc;
273 
274 	/*
275 	 * duplicate new.
276 	 */
277 	new_name = kstrdup(new, GFP_KERNEL);
278 	if (!new_name)
279 		return -ENOMEM;
280 
281 	down_write(&_hash_lock);
282 
283 	/*
284 	 * Is new free ?
285 	 */
286 	hc = __get_name_cell(new);
287 	if (hc) {
288 		DMWARN("asked to rename to an already existing name %s -> %s",
289 		       old, new);
290 		up_write(&_hash_lock);
291 		kfree(new_name);
292 		return -EBUSY;
293 	}
294 
295 	/*
296 	 * Is there such a device as 'old' ?
297 	 */
298 	hc = __get_name_cell(old);
299 	if (!hc) {
300 		DMWARN("asked to rename a non existent device %s -> %s",
301 		       old, new);
302 		up_write(&_hash_lock);
303 		kfree(new_name);
304 		return -ENXIO;
305 	}
306 
307 	/*
308 	 * rename and move the name cell.
309 	 */
310 	unregister_with_devfs(hc);
311 
312 	list_del(&hc->name_list);
313 	old_name = hc->name;
314 	hc->name = new_name;
315 	list_add(&hc->name_list, _name_buckets + hash_str(new_name));
316 
317 	/* rename the device node in devfs */
318 	register_with_devfs(hc);
319 
320 	up_write(&_hash_lock);
321 	kfree(old_name);
322 	return 0;
323 }
324 
325 /*-----------------------------------------------------------------
326  * Implementation of the ioctl commands
327  *---------------------------------------------------------------*/
328 /*
329  * All the ioctl commands get dispatched to functions with this
330  * prototype.
331  */
332 typedef int (*ioctl_fn)(struct dm_ioctl *param, size_t param_size);
333 
334 static int remove_all(struct dm_ioctl *param, size_t param_size)
335 {
336 	dm_hash_remove_all();
337 	param->data_size = 0;
338 	return 0;
339 }
340 
341 /*
342  * Round up the ptr to an 8-byte boundary.
343  */
344 #define ALIGN_MASK 7
345 static inline void *align_ptr(void *ptr)
346 {
347 	return (void *) (((size_t) (ptr + ALIGN_MASK)) & ~ALIGN_MASK);
348 }
349 
350 /*
351  * Retrieves the data payload buffer from an already allocated
352  * struct dm_ioctl.
353  */
354 static void *get_result_buffer(struct dm_ioctl *param, size_t param_size,
355 			       size_t *len)
356 {
357 	param->data_start = align_ptr(param + 1) - (void *) param;
358 
359 	if (param->data_start < param_size)
360 		*len = param_size - param->data_start;
361 	else
362 		*len = 0;
363 
364 	return ((void *) param) + param->data_start;
365 }
366 
367 static int list_devices(struct dm_ioctl *param, size_t param_size)
368 {
369 	unsigned int i;
370 	struct hash_cell *hc;
371 	size_t len, needed = 0;
372 	struct gendisk *disk;
373 	struct dm_name_list *nl, *old_nl = NULL;
374 
375 	down_write(&_hash_lock);
376 
377 	/*
378 	 * Loop through all the devices working out how much
379 	 * space we need.
380 	 */
381 	for (i = 0; i < NUM_BUCKETS; i++) {
382 		list_for_each_entry (hc, _name_buckets + i, name_list) {
383 			needed += sizeof(struct dm_name_list);
384 			needed += strlen(hc->name) + 1;
385 			needed += ALIGN_MASK;
386 		}
387 	}
388 
389 	/*
390 	 * Grab our output buffer.
391 	 */
392 	nl = get_result_buffer(param, param_size, &len);
393 	if (len < needed) {
394 		param->flags |= DM_BUFFER_FULL_FLAG;
395 		goto out;
396 	}
397 	param->data_size = param->data_start + needed;
398 
399 	nl->dev = 0;	/* Flags no data */
400 
401 	/*
402 	 * Now loop through filling out the names.
403 	 */
404 	for (i = 0; i < NUM_BUCKETS; i++) {
405 		list_for_each_entry (hc, _name_buckets + i, name_list) {
406 			if (old_nl)
407 				old_nl->next = (uint32_t) ((void *) nl -
408 							   (void *) old_nl);
409 			disk = dm_disk(hc->md);
410 			nl->dev = huge_encode_dev(MKDEV(disk->major, disk->first_minor));
411 			nl->next = 0;
412 			strcpy(nl->name, hc->name);
413 
414 			old_nl = nl;
415 			nl = align_ptr(((void *) ++nl) + strlen(hc->name) + 1);
416 		}
417 	}
418 
419  out:
420 	up_write(&_hash_lock);
421 	return 0;
422 }
423 
424 static void list_version_get_needed(struct target_type *tt, void *needed_param)
425 {
426     size_t *needed = needed_param;
427 
428     *needed += sizeof(struct dm_target_versions);
429     *needed += strlen(tt->name);
430     *needed += ALIGN_MASK;
431 }
432 
433 static void list_version_get_info(struct target_type *tt, void *param)
434 {
435     struct vers_iter *info = param;
436 
437     /* Check space - it might have changed since the first iteration */
438     if ((char *)info->vers + sizeof(tt->version) + strlen(tt->name) + 1 >
439 	info->end) {
440 
441 	info->flags = DM_BUFFER_FULL_FLAG;
442 	return;
443     }
444 
445     if (info->old_vers)
446 	info->old_vers->next = (uint32_t) ((void *)info->vers -
447 					   (void *)info->old_vers);
448     info->vers->version[0] = tt->version[0];
449     info->vers->version[1] = tt->version[1];
450     info->vers->version[2] = tt->version[2];
451     info->vers->next = 0;
452     strcpy(info->vers->name, tt->name);
453 
454     info->old_vers = info->vers;
455     info->vers = align_ptr(((void *) ++info->vers) + strlen(tt->name) + 1);
456 }
457 
458 static int list_versions(struct dm_ioctl *param, size_t param_size)
459 {
460 	size_t len, needed = 0;
461 	struct dm_target_versions *vers;
462 	struct vers_iter iter_info;
463 
464 	/*
465 	 * Loop through all the devices working out how much
466 	 * space we need.
467 	 */
468 	dm_target_iterate(list_version_get_needed, &needed);
469 
470 	/*
471 	 * Grab our output buffer.
472 	 */
473 	vers = get_result_buffer(param, param_size, &len);
474 	if (len < needed) {
475 		param->flags |= DM_BUFFER_FULL_FLAG;
476 		goto out;
477 	}
478 	param->data_size = param->data_start + needed;
479 
480 	iter_info.param_size = param_size;
481 	iter_info.old_vers = NULL;
482 	iter_info.vers = vers;
483 	iter_info.flags = 0;
484 	iter_info.end = (char *)vers+len;
485 
486 	/*
487 	 * Now loop through filling out the names & versions.
488 	 */
489 	dm_target_iterate(list_version_get_info, &iter_info);
490 	param->flags |= iter_info.flags;
491 
492  out:
493 	return 0;
494 }
495 
496 
497 
498 static int check_name(const char *name)
499 {
500 	if (strchr(name, '/')) {
501 		DMWARN("invalid device name");
502 		return -EINVAL;
503 	}
504 
505 	return 0;
506 }
507 
508 /*
509  * Fills in a dm_ioctl structure, ready for sending back to
510  * userland.
511  */
512 static int __dev_status(struct mapped_device *md, struct dm_ioctl *param)
513 {
514 	struct gendisk *disk = dm_disk(md);
515 	struct dm_table *table;
516 	struct block_device *bdev;
517 
518 	param->flags &= ~(DM_SUSPEND_FLAG | DM_READONLY_FLAG |
519 			  DM_ACTIVE_PRESENT_FLAG);
520 
521 	if (dm_suspended(md))
522 		param->flags |= DM_SUSPEND_FLAG;
523 
524 	param->dev = huge_encode_dev(MKDEV(disk->major, disk->first_minor));
525 
526 	if (!(param->flags & DM_SKIP_BDGET_FLAG)) {
527 		bdev = bdget_disk(disk, 0);
528 		if (!bdev)
529 			return -ENXIO;
530 
531 		/*
532 		 * Yes, this will be out of date by the time it gets back
533 		 * to userland, but it is still very useful for
534 		 * debugging.
535 		 */
536 		param->open_count = bdev->bd_openers;
537 		bdput(bdev);
538 	} else
539 		param->open_count = -1;
540 
541 	if (disk->policy)
542 		param->flags |= DM_READONLY_FLAG;
543 
544 	param->event_nr = dm_get_event_nr(md);
545 
546 	table = dm_get_table(md);
547 	if (table) {
548 		param->flags |= DM_ACTIVE_PRESENT_FLAG;
549 		param->target_count = dm_table_get_num_targets(table);
550 		dm_table_put(table);
551 	} else
552 		param->target_count = 0;
553 
554 	return 0;
555 }
556 
557 static int dev_create(struct dm_ioctl *param, size_t param_size)
558 {
559 	int r;
560 	struct mapped_device *md;
561 
562 	r = check_name(param->name);
563 	if (r)
564 		return r;
565 
566 	if (param->flags & DM_PERSISTENT_DEV_FLAG)
567 		r = dm_create_with_minor(MINOR(huge_decode_dev(param->dev)), &md);
568 	else
569 		r = dm_create(&md);
570 
571 	if (r)
572 		return r;
573 
574 	r = dm_hash_insert(param->name, *param->uuid ? param->uuid : NULL, md);
575 	if (r) {
576 		dm_put(md);
577 		return r;
578 	}
579 
580 	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
581 
582 	r = __dev_status(md, param);
583 	dm_put(md);
584 
585 	return r;
586 }
587 
588 /*
589  * Always use UUID for lookups if it's present, otherwise use name or dev.
590  */
591 static inline struct hash_cell *__find_device_hash_cell(struct dm_ioctl *param)
592 {
593 	if (*param->uuid)
594 		return __get_uuid_cell(param->uuid);
595 	else if (*param->name)
596 		return __get_name_cell(param->name);
597 	else
598 		return dm_get_mdptr(huge_decode_dev(param->dev));
599 }
600 
601 static inline struct mapped_device *find_device(struct dm_ioctl *param)
602 {
603 	struct hash_cell *hc;
604 	struct mapped_device *md = NULL;
605 
606 	down_read(&_hash_lock);
607 	hc = __find_device_hash_cell(param);
608 	if (hc) {
609 		md = hc->md;
610 		dm_get(md);
611 
612 		/*
613 		 * Sneakily write in both the name and the uuid
614 		 * while we have the cell.
615 		 */
616 		strncpy(param->name, hc->name, sizeof(param->name));
617 		if (hc->uuid)
618 			strncpy(param->uuid, hc->uuid, sizeof(param->uuid)-1);
619 		else
620 			param->uuid[0] = '\0';
621 
622 		if (hc->new_map)
623 			param->flags |= DM_INACTIVE_PRESENT_FLAG;
624 		else
625 			param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
626 	}
627 	up_read(&_hash_lock);
628 
629 	return md;
630 }
631 
632 static int dev_remove(struct dm_ioctl *param, size_t param_size)
633 {
634 	struct hash_cell *hc;
635 
636 	down_write(&_hash_lock);
637 	hc = __find_device_hash_cell(param);
638 
639 	if (!hc) {
640 		DMWARN("device doesn't appear to be in the dev hash table.");
641 		up_write(&_hash_lock);
642 		return -ENXIO;
643 	}
644 
645 	__hash_remove(hc);
646 	up_write(&_hash_lock);
647 	param->data_size = 0;
648 	return 0;
649 }
650 
651 /*
652  * Check a string doesn't overrun the chunk of
653  * memory we copied from userland.
654  */
655 static int invalid_str(char *str, void *end)
656 {
657 	while ((void *) str < end)
658 		if (!*str++)
659 			return 0;
660 
661 	return -EINVAL;
662 }
663 
664 static int dev_rename(struct dm_ioctl *param, size_t param_size)
665 {
666 	int r;
667 	char *new_name = (char *) param + param->data_start;
668 
669 	if (new_name < (char *) (param + 1) ||
670 	    invalid_str(new_name, (void *) param + param_size)) {
671 		DMWARN("Invalid new logical volume name supplied.");
672 		return -EINVAL;
673 	}
674 
675 	r = check_name(new_name);
676 	if (r)
677 		return r;
678 
679 	param->data_size = 0;
680 	return dm_hash_rename(param->name, new_name);
681 }
682 
683 static int do_suspend(struct dm_ioctl *param)
684 {
685 	int r = 0;
686 	struct mapped_device *md;
687 
688 	md = find_device(param);
689 	if (!md)
690 		return -ENXIO;
691 
692 	if (!dm_suspended(md))
693 		r = dm_suspend(md);
694 
695 	if (!r)
696 		r = __dev_status(md, param);
697 
698 	dm_put(md);
699 	return r;
700 }
701 
702 static int do_resume(struct dm_ioctl *param)
703 {
704 	int r = 0;
705 	struct hash_cell *hc;
706 	struct mapped_device *md;
707 	struct dm_table *new_map;
708 
709 	down_write(&_hash_lock);
710 
711 	hc = __find_device_hash_cell(param);
712 	if (!hc) {
713 		DMWARN("device doesn't appear to be in the dev hash table.");
714 		up_write(&_hash_lock);
715 		return -ENXIO;
716 	}
717 
718 	md = hc->md;
719 	dm_get(md);
720 
721 	new_map = hc->new_map;
722 	hc->new_map = NULL;
723 	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
724 
725 	up_write(&_hash_lock);
726 
727 	/* Do we need to load a new map ? */
728 	if (new_map) {
729 		/* Suspend if it isn't already suspended */
730 		if (!dm_suspended(md))
731 			dm_suspend(md);
732 
733 		r = dm_swap_table(md, new_map);
734 		if (r) {
735 			dm_put(md);
736 			dm_table_put(new_map);
737 			return r;
738 		}
739 
740 		if (dm_table_get_mode(new_map) & FMODE_WRITE)
741 			set_disk_ro(dm_disk(md), 0);
742 		else
743 			set_disk_ro(dm_disk(md), 1);
744 
745 		dm_table_put(new_map);
746 	}
747 
748 	if (dm_suspended(md))
749 		r = dm_resume(md);
750 
751 	if (!r)
752 		r = __dev_status(md, param);
753 
754 	dm_put(md);
755 	return r;
756 }
757 
758 /*
759  * Set or unset the suspension state of a device.
760  * If the device already is in the requested state we just return its status.
761  */
762 static int dev_suspend(struct dm_ioctl *param, size_t param_size)
763 {
764 	if (param->flags & DM_SUSPEND_FLAG)
765 		return do_suspend(param);
766 
767 	return do_resume(param);
768 }
769 
770 /*
771  * Copies device info back to user space, used by
772  * the create and info ioctls.
773  */
774 static int dev_status(struct dm_ioctl *param, size_t param_size)
775 {
776 	int r;
777 	struct mapped_device *md;
778 
779 	md = find_device(param);
780 	if (!md)
781 		return -ENXIO;
782 
783 	r = __dev_status(md, param);
784 	dm_put(md);
785 	return r;
786 }
787 
788 /*
789  * Build up the status struct for each target
790  */
791 static void retrieve_status(struct dm_table *table,
792 			    struct dm_ioctl *param, size_t param_size)
793 {
794 	unsigned int i, num_targets;
795 	struct dm_target_spec *spec;
796 	char *outbuf, *outptr;
797 	status_type_t type;
798 	size_t remaining, len, used = 0;
799 
800 	outptr = outbuf = get_result_buffer(param, param_size, &len);
801 
802 	if (param->flags & DM_STATUS_TABLE_FLAG)
803 		type = STATUSTYPE_TABLE;
804 	else
805 		type = STATUSTYPE_INFO;
806 
807 	/* Get all the target info */
808 	num_targets = dm_table_get_num_targets(table);
809 	for (i = 0; i < num_targets; i++) {
810 		struct dm_target *ti = dm_table_get_target(table, i);
811 
812 		remaining = len - (outptr - outbuf);
813 		if (remaining <= sizeof(struct dm_target_spec)) {
814 			param->flags |= DM_BUFFER_FULL_FLAG;
815 			break;
816 		}
817 
818 		spec = (struct dm_target_spec *) outptr;
819 
820 		spec->status = 0;
821 		spec->sector_start = ti->begin;
822 		spec->length = ti->len;
823 		strncpy(spec->target_type, ti->type->name,
824 			sizeof(spec->target_type));
825 
826 		outptr += sizeof(struct dm_target_spec);
827 		remaining = len - (outptr - outbuf);
828 		if (remaining <= 0) {
829 			param->flags |= DM_BUFFER_FULL_FLAG;
830 			break;
831 		}
832 
833 		/* Get the status/table string from the target driver */
834 		if (ti->type->status) {
835 			if (ti->type->status(ti, type, outptr, remaining)) {
836 				param->flags |= DM_BUFFER_FULL_FLAG;
837 				break;
838 			}
839 		} else
840 			outptr[0] = '\0';
841 
842 		outptr += strlen(outptr) + 1;
843 		used = param->data_start + (outptr - outbuf);
844 
845 		outptr = align_ptr(outptr);
846 		spec->next = outptr - outbuf;
847 	}
848 
849 	if (used)
850 		param->data_size = used;
851 
852 	param->target_count = num_targets;
853 }
854 
855 /*
856  * Wait for a device to report an event
857  */
858 static int dev_wait(struct dm_ioctl *param, size_t param_size)
859 {
860 	int r;
861 	struct mapped_device *md;
862 	struct dm_table *table;
863 
864 	md = find_device(param);
865 	if (!md)
866 		return -ENXIO;
867 
868 	/*
869 	 * Wait for a notification event
870 	 */
871 	if (dm_wait_event(md, param->event_nr)) {
872 		r = -ERESTARTSYS;
873 		goto out;
874 	}
875 
876 	/*
877 	 * The userland program is going to want to know what
878 	 * changed to trigger the event, so we may as well tell
879 	 * him and save an ioctl.
880 	 */
881 	r = __dev_status(md, param);
882 	if (r)
883 		goto out;
884 
885 	table = dm_get_table(md);
886 	if (table) {
887 		retrieve_status(table, param, param_size);
888 		dm_table_put(table);
889 	}
890 
891  out:
892 	dm_put(md);
893 	return r;
894 }
895 
896 static inline int get_mode(struct dm_ioctl *param)
897 {
898 	int mode = FMODE_READ | FMODE_WRITE;
899 
900 	if (param->flags & DM_READONLY_FLAG)
901 		mode = FMODE_READ;
902 
903 	return mode;
904 }
905 
906 static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
907 		       struct dm_target_spec **spec, char **target_params)
908 {
909 	*spec = (struct dm_target_spec *) ((unsigned char *) last + next);
910 	*target_params = (char *) (*spec + 1);
911 
912 	if (*spec < (last + 1))
913 		return -EINVAL;
914 
915 	return invalid_str(*target_params, end);
916 }
917 
918 static int populate_table(struct dm_table *table,
919 			  struct dm_ioctl *param, size_t param_size)
920 {
921 	int r;
922 	unsigned int i = 0;
923 	struct dm_target_spec *spec = (struct dm_target_spec *) param;
924 	uint32_t next = param->data_start;
925 	void *end = (void *) param + param_size;
926 	char *target_params;
927 
928 	if (!param->target_count) {
929 		DMWARN("populate_table: no targets specified");
930 		return -EINVAL;
931 	}
932 
933 	for (i = 0; i < param->target_count; i++) {
934 
935 		r = next_target(spec, next, end, &spec, &target_params);
936 		if (r) {
937 			DMWARN("unable to find target");
938 			return r;
939 		}
940 
941 		r = dm_table_add_target(table, spec->target_type,
942 					(sector_t) spec->sector_start,
943 					(sector_t) spec->length,
944 					target_params);
945 		if (r) {
946 			DMWARN("error adding target to table");
947 			return r;
948 		}
949 
950 		next = spec->next;
951 	}
952 
953 	return dm_table_complete(table);
954 }
955 
956 static int table_load(struct dm_ioctl *param, size_t param_size)
957 {
958 	int r;
959 	struct hash_cell *hc;
960 	struct dm_table *t;
961 
962 	r = dm_table_create(&t, get_mode(param), param->target_count);
963 	if (r)
964 		return r;
965 
966 	r = populate_table(t, param, param_size);
967 	if (r) {
968 		dm_table_put(t);
969 		return r;
970 	}
971 
972 	down_write(&_hash_lock);
973 	hc = __find_device_hash_cell(param);
974 	if (!hc) {
975 		DMWARN("device doesn't appear to be in the dev hash table.");
976 		up_write(&_hash_lock);
977 		dm_table_put(t);
978 		return -ENXIO;
979 	}
980 
981 	if (hc->new_map)
982 		dm_table_put(hc->new_map);
983 	hc->new_map = t;
984 	param->flags |= DM_INACTIVE_PRESENT_FLAG;
985 
986 	r = __dev_status(hc->md, param);
987 	up_write(&_hash_lock);
988 	return r;
989 }
990 
991 static int table_clear(struct dm_ioctl *param, size_t param_size)
992 {
993 	int r;
994 	struct hash_cell *hc;
995 
996 	down_write(&_hash_lock);
997 
998 	hc = __find_device_hash_cell(param);
999 	if (!hc) {
1000 		DMWARN("device doesn't appear to be in the dev hash table.");
1001 		up_write(&_hash_lock);
1002 		return -ENXIO;
1003 	}
1004 
1005 	if (hc->new_map) {
1006 		dm_table_put(hc->new_map);
1007 		hc->new_map = NULL;
1008 	}
1009 
1010 	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1011 
1012 	r = __dev_status(hc->md, param);
1013 	up_write(&_hash_lock);
1014 	return r;
1015 }
1016 
1017 /*
1018  * Retrieves a list of devices used by a particular dm device.
1019  */
1020 static void retrieve_deps(struct dm_table *table,
1021 			  struct dm_ioctl *param, size_t param_size)
1022 {
1023 	unsigned int count = 0;
1024 	struct list_head *tmp;
1025 	size_t len, needed;
1026 	struct dm_dev *dd;
1027 	struct dm_target_deps *deps;
1028 
1029 	deps = get_result_buffer(param, param_size, &len);
1030 
1031 	/*
1032 	 * Count the devices.
1033 	 */
1034 	list_for_each (tmp, dm_table_get_devices(table))
1035 		count++;
1036 
1037 	/*
1038 	 * Check we have enough space.
1039 	 */
1040 	needed = sizeof(*deps) + (sizeof(*deps->dev) * count);
1041 	if (len < needed) {
1042 		param->flags |= DM_BUFFER_FULL_FLAG;
1043 		return;
1044 	}
1045 
1046 	/*
1047 	 * Fill in the devices.
1048 	 */
1049 	deps->count = count;
1050 	count = 0;
1051 	list_for_each_entry (dd, dm_table_get_devices(table), list)
1052 		deps->dev[count++] = huge_encode_dev(dd->bdev->bd_dev);
1053 
1054 	param->data_size = param->data_start + needed;
1055 }
1056 
1057 static int table_deps(struct dm_ioctl *param, size_t param_size)
1058 {
1059 	int r = 0;
1060 	struct mapped_device *md;
1061 	struct dm_table *table;
1062 
1063 	md = find_device(param);
1064 	if (!md)
1065 		return -ENXIO;
1066 
1067 	r = __dev_status(md, param);
1068 	if (r)
1069 		goto out;
1070 
1071 	table = dm_get_table(md);
1072 	if (table) {
1073 		retrieve_deps(table, param, param_size);
1074 		dm_table_put(table);
1075 	}
1076 
1077  out:
1078 	dm_put(md);
1079 	return r;
1080 }
1081 
1082 /*
1083  * Return the status of a device as a text string for each
1084  * target.
1085  */
1086 static int table_status(struct dm_ioctl *param, size_t param_size)
1087 {
1088 	int r;
1089 	struct mapped_device *md;
1090 	struct dm_table *table;
1091 
1092 	md = find_device(param);
1093 	if (!md)
1094 		return -ENXIO;
1095 
1096 	r = __dev_status(md, param);
1097 	if (r)
1098 		goto out;
1099 
1100 	table = dm_get_table(md);
1101 	if (table) {
1102 		retrieve_status(table, param, param_size);
1103 		dm_table_put(table);
1104 	}
1105 
1106  out:
1107 	dm_put(md);
1108 	return r;
1109 }
1110 
1111 /*
1112  * Pass a message to the target that's at the supplied device offset.
1113  */
1114 static int target_message(struct dm_ioctl *param, size_t param_size)
1115 {
1116 	int r, argc;
1117 	char **argv;
1118 	struct mapped_device *md;
1119 	struct dm_table *table;
1120 	struct dm_target *ti;
1121 	struct dm_target_msg *tmsg = (void *) param + param->data_start;
1122 
1123 	md = find_device(param);
1124 	if (!md)
1125 		return -ENXIO;
1126 
1127 	r = __dev_status(md, param);
1128 	if (r)
1129 		goto out;
1130 
1131 	if (tmsg < (struct dm_target_msg *) (param + 1) ||
1132 	    invalid_str(tmsg->message, (void *) param + param_size)) {
1133 		DMWARN("Invalid target message parameters.");
1134 		r = -EINVAL;
1135 		goto out;
1136 	}
1137 
1138 	r = dm_split_args(&argc, &argv, tmsg->message);
1139 	if (r) {
1140 		DMWARN("Failed to split target message parameters");
1141 		goto out;
1142 	}
1143 
1144 	table = dm_get_table(md);
1145 	if (!table)
1146 		goto out_argv;
1147 
1148 	if (tmsg->sector >= dm_table_get_size(table)) {
1149 		DMWARN("Target message sector outside device.");
1150 		r = -EINVAL;
1151 		goto out_table;
1152 	}
1153 
1154 	ti = dm_table_find_target(table, tmsg->sector);
1155 	if (ti->type->message)
1156 		r = ti->type->message(ti, argc, argv);
1157 	else {
1158 		DMWARN("Target type does not support messages");
1159 		r = -EINVAL;
1160 	}
1161 
1162  out_table:
1163 	dm_table_put(table);
1164  out_argv:
1165 	kfree(argv);
1166  out:
1167 	param->data_size = 0;
1168 	dm_put(md);
1169 	return r;
1170 }
1171 
1172 /*-----------------------------------------------------------------
1173  * Implementation of open/close/ioctl on the special char
1174  * device.
1175  *---------------------------------------------------------------*/
1176 static ioctl_fn lookup_ioctl(unsigned int cmd)
1177 {
1178 	static struct {
1179 		int cmd;
1180 		ioctl_fn fn;
1181 	} _ioctls[] = {
1182 		{DM_VERSION_CMD, NULL},	/* version is dealt with elsewhere */
1183 		{DM_REMOVE_ALL_CMD, remove_all},
1184 		{DM_LIST_DEVICES_CMD, list_devices},
1185 
1186 		{DM_DEV_CREATE_CMD, dev_create},
1187 		{DM_DEV_REMOVE_CMD, dev_remove},
1188 		{DM_DEV_RENAME_CMD, dev_rename},
1189 		{DM_DEV_SUSPEND_CMD, dev_suspend},
1190 		{DM_DEV_STATUS_CMD, dev_status},
1191 		{DM_DEV_WAIT_CMD, dev_wait},
1192 
1193 		{DM_TABLE_LOAD_CMD, table_load},
1194 		{DM_TABLE_CLEAR_CMD, table_clear},
1195 		{DM_TABLE_DEPS_CMD, table_deps},
1196 		{DM_TABLE_STATUS_CMD, table_status},
1197 
1198 		{DM_LIST_VERSIONS_CMD, list_versions},
1199 
1200 		{DM_TARGET_MSG_CMD, target_message}
1201 	};
1202 
1203 	return (cmd >= ARRAY_SIZE(_ioctls)) ? NULL : _ioctls[cmd].fn;
1204 }
1205 
1206 /*
1207  * As well as checking the version compatibility this always
1208  * copies the kernel interface version out.
1209  */
1210 static int check_version(unsigned int cmd, struct dm_ioctl __user *user)
1211 {
1212 	uint32_t version[3];
1213 	int r = 0;
1214 
1215 	if (copy_from_user(version, user->version, sizeof(version)))
1216 		return -EFAULT;
1217 
1218 	if ((DM_VERSION_MAJOR != version[0]) ||
1219 	    (DM_VERSION_MINOR < version[1])) {
1220 		DMWARN("ioctl interface mismatch: "
1221 		       "kernel(%u.%u.%u), user(%u.%u.%u), cmd(%d)",
1222 		       DM_VERSION_MAJOR, DM_VERSION_MINOR,
1223 		       DM_VERSION_PATCHLEVEL,
1224 		       version[0], version[1], version[2], cmd);
1225 		r = -EINVAL;
1226 	}
1227 
1228 	/*
1229 	 * Fill in the kernel version.
1230 	 */
1231 	version[0] = DM_VERSION_MAJOR;
1232 	version[1] = DM_VERSION_MINOR;
1233 	version[2] = DM_VERSION_PATCHLEVEL;
1234 	if (copy_to_user(user->version, version, sizeof(version)))
1235 		return -EFAULT;
1236 
1237 	return r;
1238 }
1239 
1240 static void free_params(struct dm_ioctl *param)
1241 {
1242 	vfree(param);
1243 }
1244 
1245 static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl **param)
1246 {
1247 	struct dm_ioctl tmp, *dmi;
1248 
1249 	if (copy_from_user(&tmp, user, sizeof(tmp)))
1250 		return -EFAULT;
1251 
1252 	if (tmp.data_size < sizeof(tmp))
1253 		return -EINVAL;
1254 
1255 	dmi = (struct dm_ioctl *) vmalloc(tmp.data_size);
1256 	if (!dmi)
1257 		return -ENOMEM;
1258 
1259 	if (copy_from_user(dmi, user, tmp.data_size)) {
1260 		vfree(dmi);
1261 		return -EFAULT;
1262 	}
1263 
1264 	*param = dmi;
1265 	return 0;
1266 }
1267 
1268 static int validate_params(uint cmd, struct dm_ioctl *param)
1269 {
1270 	/* Always clear this flag */
1271 	param->flags &= ~DM_BUFFER_FULL_FLAG;
1272 
1273 	/* Ignores parameters */
1274 	if (cmd == DM_REMOVE_ALL_CMD ||
1275 	    cmd == DM_LIST_DEVICES_CMD ||
1276 	    cmd == DM_LIST_VERSIONS_CMD)
1277 		return 0;
1278 
1279 	if ((cmd == DM_DEV_CREATE_CMD)) {
1280 		if (!*param->name) {
1281 			DMWARN("name not supplied when creating device");
1282 			return -EINVAL;
1283 		}
1284 	} else if ((*param->uuid && *param->name)) {
1285 		DMWARN("only supply one of name or uuid, cmd(%u)", cmd);
1286 		return -EINVAL;
1287 	}
1288 
1289 	/* Ensure strings are terminated */
1290 	param->name[DM_NAME_LEN - 1] = '\0';
1291 	param->uuid[DM_UUID_LEN - 1] = '\0';
1292 
1293 	return 0;
1294 }
1295 
1296 static int ctl_ioctl(struct inode *inode, struct file *file,
1297 		     uint command, ulong u)
1298 {
1299 	int r = 0;
1300 	unsigned int cmd;
1301 	struct dm_ioctl *param;
1302 	struct dm_ioctl __user *user = (struct dm_ioctl __user *) u;
1303 	ioctl_fn fn = NULL;
1304 	size_t param_size;
1305 
1306 	/* only root can play with this */
1307 	if (!capable(CAP_SYS_ADMIN))
1308 		return -EACCES;
1309 
1310 	if (_IOC_TYPE(command) != DM_IOCTL)
1311 		return -ENOTTY;
1312 
1313 	cmd = _IOC_NR(command);
1314 
1315 	/*
1316 	 * Check the interface version passed in.  This also
1317 	 * writes out the kernel's interface version.
1318 	 */
1319 	r = check_version(cmd, user);
1320 	if (r)
1321 		return r;
1322 
1323 	/*
1324 	 * Nothing more to do for the version command.
1325 	 */
1326 	if (cmd == DM_VERSION_CMD)
1327 		return 0;
1328 
1329 	fn = lookup_ioctl(cmd);
1330 	if (!fn) {
1331 		DMWARN("dm_ctl_ioctl: unknown command 0x%x", command);
1332 		return -ENOTTY;
1333 	}
1334 
1335 	/*
1336 	 * Trying to avoid low memory issues when a device is
1337 	 * suspended.
1338 	 */
1339 	current->flags |= PF_MEMALLOC;
1340 
1341 	/*
1342 	 * Copy the parameters into kernel space.
1343 	 */
1344 	r = copy_params(user, &param);
1345 	if (r) {
1346 		current->flags &= ~PF_MEMALLOC;
1347 		return r;
1348 	}
1349 
1350 	/*
1351 	 * FIXME: eventually we will remove the PF_MEMALLOC flag
1352 	 * here.  However the tools still do nasty things like
1353 	 * 'load' while a device is suspended.
1354 	 */
1355 
1356 	r = validate_params(cmd, param);
1357 	if (r)
1358 		goto out;
1359 
1360 	param_size = param->data_size;
1361 	param->data_size = sizeof(*param);
1362 	r = fn(param, param_size);
1363 
1364 	/*
1365 	 * Copy the results back to userland.
1366 	 */
1367 	if (!r && copy_to_user(user, param, param->data_size))
1368 		r = -EFAULT;
1369 
1370  out:
1371 	free_params(param);
1372 	current->flags &= ~PF_MEMALLOC;
1373 	return r;
1374 }
1375 
1376 static struct file_operations _ctl_fops = {
1377 	.ioctl	 = ctl_ioctl,
1378 	.owner	 = THIS_MODULE,
1379 };
1380 
1381 static struct miscdevice _dm_misc = {
1382 	.minor 		= MISC_DYNAMIC_MINOR,
1383 	.name  		= DM_NAME,
1384 	.devfs_name 	= "mapper/control",
1385 	.fops  		= &_ctl_fops
1386 };
1387 
1388 /*
1389  * Create misc character device and link to DM_DIR/control.
1390  */
1391 int __init dm_interface_init(void)
1392 {
1393 	int r;
1394 
1395 	r = dm_hash_init();
1396 	if (r)
1397 		return r;
1398 
1399 	r = misc_register(&_dm_misc);
1400 	if (r) {
1401 		DMERR("misc_register failed for control device");
1402 		dm_hash_exit();
1403 		return r;
1404 	}
1405 
1406 	DMINFO("%d.%d.%d%s initialised: %s", DM_VERSION_MAJOR,
1407 	       DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL, DM_VERSION_EXTRA,
1408 	       DM_DRIVER_EMAIL);
1409 	return 0;
1410 }
1411 
1412 void dm_interface_exit(void)
1413 {
1414 	if (misc_deregister(&_dm_misc) < 0)
1415 		DMERR("misc_deregister failed for control device");
1416 
1417 	dm_hash_exit();
1418 }
1419