xref: /linux/drivers/md/dm-ioctl.c (revision 87c2ce3b9305b9b723faeedf6e32ef703ec9b33a)
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 	struct dm_table *table;
274 
275 	/*
276 	 * duplicate new.
277 	 */
278 	new_name = kstrdup(new, GFP_KERNEL);
279 	if (!new_name)
280 		return -ENOMEM;
281 
282 	down_write(&_hash_lock);
283 
284 	/*
285 	 * Is new free ?
286 	 */
287 	hc = __get_name_cell(new);
288 	if (hc) {
289 		DMWARN("asked to rename to an already existing name %s -> %s",
290 		       old, new);
291 		up_write(&_hash_lock);
292 		kfree(new_name);
293 		return -EBUSY;
294 	}
295 
296 	/*
297 	 * Is there such a device as 'old' ?
298 	 */
299 	hc = __get_name_cell(old);
300 	if (!hc) {
301 		DMWARN("asked to rename a non existent device %s -> %s",
302 		       old, new);
303 		up_write(&_hash_lock);
304 		kfree(new_name);
305 		return -ENXIO;
306 	}
307 
308 	/*
309 	 * rename and move the name cell.
310 	 */
311 	unregister_with_devfs(hc);
312 
313 	list_del(&hc->name_list);
314 	old_name = hc->name;
315 	hc->name = new_name;
316 	list_add(&hc->name_list, _name_buckets + hash_str(new_name));
317 
318 	/* rename the device node in devfs */
319 	register_with_devfs(hc);
320 
321 	/*
322 	 * Wake up any dm event waiters.
323 	 */
324 	table = dm_get_table(hc->md);
325 	if (table) {
326 		dm_table_event(table);
327 		dm_table_put(table);
328 	}
329 
330 	up_write(&_hash_lock);
331 	kfree(old_name);
332 	return 0;
333 }
334 
335 /*-----------------------------------------------------------------
336  * Implementation of the ioctl commands
337  *---------------------------------------------------------------*/
338 /*
339  * All the ioctl commands get dispatched to functions with this
340  * prototype.
341  */
342 typedef int (*ioctl_fn)(struct dm_ioctl *param, size_t param_size);
343 
344 static int remove_all(struct dm_ioctl *param, size_t param_size)
345 {
346 	dm_hash_remove_all();
347 	param->data_size = 0;
348 	return 0;
349 }
350 
351 /*
352  * Round up the ptr to an 8-byte boundary.
353  */
354 #define ALIGN_MASK 7
355 static inline void *align_ptr(void *ptr)
356 {
357 	return (void *) (((size_t) (ptr + ALIGN_MASK)) & ~ALIGN_MASK);
358 }
359 
360 /*
361  * Retrieves the data payload buffer from an already allocated
362  * struct dm_ioctl.
363  */
364 static void *get_result_buffer(struct dm_ioctl *param, size_t param_size,
365 			       size_t *len)
366 {
367 	param->data_start = align_ptr(param + 1) - (void *) param;
368 
369 	if (param->data_start < param_size)
370 		*len = param_size - param->data_start;
371 	else
372 		*len = 0;
373 
374 	return ((void *) param) + param->data_start;
375 }
376 
377 static int list_devices(struct dm_ioctl *param, size_t param_size)
378 {
379 	unsigned int i;
380 	struct hash_cell *hc;
381 	size_t len, needed = 0;
382 	struct gendisk *disk;
383 	struct dm_name_list *nl, *old_nl = NULL;
384 
385 	down_write(&_hash_lock);
386 
387 	/*
388 	 * Loop through all the devices working out how much
389 	 * space we need.
390 	 */
391 	for (i = 0; i < NUM_BUCKETS; i++) {
392 		list_for_each_entry (hc, _name_buckets + i, name_list) {
393 			needed += sizeof(struct dm_name_list);
394 			needed += strlen(hc->name) + 1;
395 			needed += ALIGN_MASK;
396 		}
397 	}
398 
399 	/*
400 	 * Grab our output buffer.
401 	 */
402 	nl = get_result_buffer(param, param_size, &len);
403 	if (len < needed) {
404 		param->flags |= DM_BUFFER_FULL_FLAG;
405 		goto out;
406 	}
407 	param->data_size = param->data_start + needed;
408 
409 	nl->dev = 0;	/* Flags no data */
410 
411 	/*
412 	 * Now loop through filling out the names.
413 	 */
414 	for (i = 0; i < NUM_BUCKETS; i++) {
415 		list_for_each_entry (hc, _name_buckets + i, name_list) {
416 			if (old_nl)
417 				old_nl->next = (uint32_t) ((void *) nl -
418 							   (void *) old_nl);
419 			disk = dm_disk(hc->md);
420 			nl->dev = huge_encode_dev(MKDEV(disk->major, disk->first_minor));
421 			nl->next = 0;
422 			strcpy(nl->name, hc->name);
423 
424 			old_nl = nl;
425 			nl = align_ptr(((void *) ++nl) + strlen(hc->name) + 1);
426 		}
427 	}
428 
429  out:
430 	up_write(&_hash_lock);
431 	return 0;
432 }
433 
434 static void list_version_get_needed(struct target_type *tt, void *needed_param)
435 {
436     size_t *needed = needed_param;
437 
438     *needed += sizeof(struct dm_target_versions);
439     *needed += strlen(tt->name);
440     *needed += ALIGN_MASK;
441 }
442 
443 static void list_version_get_info(struct target_type *tt, void *param)
444 {
445     struct vers_iter *info = param;
446 
447     /* Check space - it might have changed since the first iteration */
448     if ((char *)info->vers + sizeof(tt->version) + strlen(tt->name) + 1 >
449 	info->end) {
450 
451 	info->flags = DM_BUFFER_FULL_FLAG;
452 	return;
453     }
454 
455     if (info->old_vers)
456 	info->old_vers->next = (uint32_t) ((void *)info->vers -
457 					   (void *)info->old_vers);
458     info->vers->version[0] = tt->version[0];
459     info->vers->version[1] = tt->version[1];
460     info->vers->version[2] = tt->version[2];
461     info->vers->next = 0;
462     strcpy(info->vers->name, tt->name);
463 
464     info->old_vers = info->vers;
465     info->vers = align_ptr(((void *) ++info->vers) + strlen(tt->name) + 1);
466 }
467 
468 static int list_versions(struct dm_ioctl *param, size_t param_size)
469 {
470 	size_t len, needed = 0;
471 	struct dm_target_versions *vers;
472 	struct vers_iter iter_info;
473 
474 	/*
475 	 * Loop through all the devices working out how much
476 	 * space we need.
477 	 */
478 	dm_target_iterate(list_version_get_needed, &needed);
479 
480 	/*
481 	 * Grab our output buffer.
482 	 */
483 	vers = get_result_buffer(param, param_size, &len);
484 	if (len < needed) {
485 		param->flags |= DM_BUFFER_FULL_FLAG;
486 		goto out;
487 	}
488 	param->data_size = param->data_start + needed;
489 
490 	iter_info.param_size = param_size;
491 	iter_info.old_vers = NULL;
492 	iter_info.vers = vers;
493 	iter_info.flags = 0;
494 	iter_info.end = (char *)vers+len;
495 
496 	/*
497 	 * Now loop through filling out the names & versions.
498 	 */
499 	dm_target_iterate(list_version_get_info, &iter_info);
500 	param->flags |= iter_info.flags;
501 
502  out:
503 	return 0;
504 }
505 
506 
507 
508 static int check_name(const char *name)
509 {
510 	if (strchr(name, '/')) {
511 		DMWARN("invalid device name");
512 		return -EINVAL;
513 	}
514 
515 	return 0;
516 }
517 
518 /*
519  * Fills in a dm_ioctl structure, ready for sending back to
520  * userland.
521  */
522 static int __dev_status(struct mapped_device *md, struct dm_ioctl *param)
523 {
524 	struct gendisk *disk = dm_disk(md);
525 	struct dm_table *table;
526 	struct block_device *bdev;
527 
528 	param->flags &= ~(DM_SUSPEND_FLAG | DM_READONLY_FLAG |
529 			  DM_ACTIVE_PRESENT_FLAG);
530 
531 	if (dm_suspended(md))
532 		param->flags |= DM_SUSPEND_FLAG;
533 
534 	param->dev = huge_encode_dev(MKDEV(disk->major, disk->first_minor));
535 
536 	if (!(param->flags & DM_SKIP_BDGET_FLAG)) {
537 		bdev = bdget_disk(disk, 0);
538 		if (!bdev)
539 			return -ENXIO;
540 
541 		/*
542 		 * Yes, this will be out of date by the time it gets back
543 		 * to userland, but it is still very useful for
544 		 * debugging.
545 		 */
546 		param->open_count = bdev->bd_openers;
547 		bdput(bdev);
548 	} else
549 		param->open_count = -1;
550 
551 	if (disk->policy)
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;
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 		r = dm_create_with_minor(MINOR(huge_decode_dev(param->dev)), &md);
578 	else
579 		r = dm_create(&md);
580 
581 	if (r)
582 		return r;
583 
584 	r = dm_hash_insert(param->name, *param->uuid ? param->uuid : NULL, md);
585 	if (r) {
586 		dm_put(md);
587 		return r;
588 	}
589 
590 	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
591 
592 	r = __dev_status(md, param);
593 	dm_put(md);
594 
595 	return r;
596 }
597 
598 /*
599  * Always use UUID for lookups if it's present, otherwise use name or dev.
600  */
601 static inline struct hash_cell *__find_device_hash_cell(struct dm_ioctl *param)
602 {
603 	if (*param->uuid)
604 		return __get_uuid_cell(param->uuid);
605 	else if (*param->name)
606 		return __get_name_cell(param->name);
607 	else
608 		return dm_get_mdptr(huge_decode_dev(param->dev));
609 }
610 
611 static inline struct mapped_device *find_device(struct dm_ioctl *param)
612 {
613 	struct hash_cell *hc;
614 	struct mapped_device *md = NULL;
615 
616 	down_read(&_hash_lock);
617 	hc = __find_device_hash_cell(param);
618 	if (hc) {
619 		md = hc->md;
620 		dm_get(md);
621 
622 		/*
623 		 * Sneakily write in both the name and the uuid
624 		 * while we have the cell.
625 		 */
626 		strncpy(param->name, hc->name, sizeof(param->name));
627 		if (hc->uuid)
628 			strncpy(param->uuid, hc->uuid, sizeof(param->uuid)-1);
629 		else
630 			param->uuid[0] = '\0';
631 
632 		if (hc->new_map)
633 			param->flags |= DM_INACTIVE_PRESENT_FLAG;
634 		else
635 			param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
636 	}
637 	up_read(&_hash_lock);
638 
639 	return md;
640 }
641 
642 static int dev_remove(struct dm_ioctl *param, size_t param_size)
643 {
644 	struct hash_cell *hc;
645 
646 	down_write(&_hash_lock);
647 	hc = __find_device_hash_cell(param);
648 
649 	if (!hc) {
650 		DMWARN("device doesn't appear to be in the dev hash table.");
651 		up_write(&_hash_lock);
652 		return -ENXIO;
653 	}
654 
655 	__hash_remove(hc);
656 	up_write(&_hash_lock);
657 	param->data_size = 0;
658 	return 0;
659 }
660 
661 /*
662  * Check a string doesn't overrun the chunk of
663  * memory we copied from userland.
664  */
665 static int invalid_str(char *str, void *end)
666 {
667 	while ((void *) str < end)
668 		if (!*str++)
669 			return 0;
670 
671 	return -EINVAL;
672 }
673 
674 static int dev_rename(struct dm_ioctl *param, size_t param_size)
675 {
676 	int r;
677 	char *new_name = (char *) param + param->data_start;
678 
679 	if (new_name < (char *) (param + 1) ||
680 	    invalid_str(new_name, (void *) param + param_size)) {
681 		DMWARN("Invalid new logical volume name supplied.");
682 		return -EINVAL;
683 	}
684 
685 	r = check_name(new_name);
686 	if (r)
687 		return r;
688 
689 	param->data_size = 0;
690 	return dm_hash_rename(param->name, new_name);
691 }
692 
693 static int do_suspend(struct dm_ioctl *param)
694 {
695 	int r = 0;
696 	int do_lockfs = 1;
697 	struct mapped_device *md;
698 
699 	md = find_device(param);
700 	if (!md)
701 		return -ENXIO;
702 
703 	if (param->flags & DM_SKIP_LOCKFS_FLAG)
704 		do_lockfs = 0;
705 
706 	if (!dm_suspended(md))
707 		r = dm_suspend(md, do_lockfs);
708 
709 	if (!r)
710 		r = __dev_status(md, param);
711 
712 	dm_put(md);
713 	return r;
714 }
715 
716 static int do_resume(struct dm_ioctl *param)
717 {
718 	int r = 0;
719 	int do_lockfs = 1;
720 	struct hash_cell *hc;
721 	struct mapped_device *md;
722 	struct dm_table *new_map;
723 
724 	down_write(&_hash_lock);
725 
726 	hc = __find_device_hash_cell(param);
727 	if (!hc) {
728 		DMWARN("device doesn't appear to be in the dev hash table.");
729 		up_write(&_hash_lock);
730 		return -ENXIO;
731 	}
732 
733 	md = hc->md;
734 	dm_get(md);
735 
736 	new_map = hc->new_map;
737 	hc->new_map = NULL;
738 	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
739 
740 	up_write(&_hash_lock);
741 
742 	/* Do we need to load a new map ? */
743 	if (new_map) {
744 		/* Suspend if it isn't already suspended */
745 		if (param->flags & DM_SKIP_LOCKFS_FLAG)
746 			do_lockfs = 0;
747 		if (!dm_suspended(md))
748 			dm_suspend(md, do_lockfs);
749 
750 		r = dm_swap_table(md, new_map);
751 		if (r) {
752 			dm_put(md);
753 			dm_table_put(new_map);
754 			return r;
755 		}
756 
757 		if (dm_table_get_mode(new_map) & FMODE_WRITE)
758 			set_disk_ro(dm_disk(md), 0);
759 		else
760 			set_disk_ro(dm_disk(md), 1);
761 
762 		dm_table_put(new_map);
763 	}
764 
765 	if (dm_suspended(md))
766 		r = dm_resume(md);
767 
768 	if (!r)
769 		r = __dev_status(md, param);
770 
771 	dm_put(md);
772 	return r;
773 }
774 
775 /*
776  * Set or unset the suspension state of a device.
777  * If the device already is in the requested state we just return its status.
778  */
779 static int dev_suspend(struct dm_ioctl *param, size_t param_size)
780 {
781 	if (param->flags & DM_SUSPEND_FLAG)
782 		return do_suspend(param);
783 
784 	return do_resume(param);
785 }
786 
787 /*
788  * Copies device info back to user space, used by
789  * the create and info ioctls.
790  */
791 static int dev_status(struct dm_ioctl *param, size_t param_size)
792 {
793 	int r;
794 	struct mapped_device *md;
795 
796 	md = find_device(param);
797 	if (!md)
798 		return -ENXIO;
799 
800 	r = __dev_status(md, param);
801 	dm_put(md);
802 	return r;
803 }
804 
805 /*
806  * Build up the status struct for each target
807  */
808 static void retrieve_status(struct dm_table *table,
809 			    struct dm_ioctl *param, size_t param_size)
810 {
811 	unsigned int i, num_targets;
812 	struct dm_target_spec *spec;
813 	char *outbuf, *outptr;
814 	status_type_t type;
815 	size_t remaining, len, used = 0;
816 
817 	outptr = outbuf = get_result_buffer(param, param_size, &len);
818 
819 	if (param->flags & DM_STATUS_TABLE_FLAG)
820 		type = STATUSTYPE_TABLE;
821 	else
822 		type = STATUSTYPE_INFO;
823 
824 	/* Get all the target info */
825 	num_targets = dm_table_get_num_targets(table);
826 	for (i = 0; i < num_targets; i++) {
827 		struct dm_target *ti = dm_table_get_target(table, i);
828 
829 		remaining = len - (outptr - outbuf);
830 		if (remaining <= sizeof(struct dm_target_spec)) {
831 			param->flags |= DM_BUFFER_FULL_FLAG;
832 			break;
833 		}
834 
835 		spec = (struct dm_target_spec *) outptr;
836 
837 		spec->status = 0;
838 		spec->sector_start = ti->begin;
839 		spec->length = ti->len;
840 		strncpy(spec->target_type, ti->type->name,
841 			sizeof(spec->target_type));
842 
843 		outptr += sizeof(struct dm_target_spec);
844 		remaining = len - (outptr - outbuf);
845 		if (remaining <= 0) {
846 			param->flags |= DM_BUFFER_FULL_FLAG;
847 			break;
848 		}
849 
850 		/* Get the status/table string from the target driver */
851 		if (ti->type->status) {
852 			if (ti->type->status(ti, type, outptr, remaining)) {
853 				param->flags |= DM_BUFFER_FULL_FLAG;
854 				break;
855 			}
856 		} else
857 			outptr[0] = '\0';
858 
859 		outptr += strlen(outptr) + 1;
860 		used = param->data_start + (outptr - outbuf);
861 
862 		outptr = align_ptr(outptr);
863 		spec->next = outptr - outbuf;
864 	}
865 
866 	if (used)
867 		param->data_size = used;
868 
869 	param->target_count = num_targets;
870 }
871 
872 /*
873  * Wait for a device to report an event
874  */
875 static int dev_wait(struct dm_ioctl *param, size_t param_size)
876 {
877 	int r;
878 	struct mapped_device *md;
879 	struct dm_table *table;
880 
881 	md = find_device(param);
882 	if (!md)
883 		return -ENXIO;
884 
885 	/*
886 	 * Wait for a notification event
887 	 */
888 	if (dm_wait_event(md, param->event_nr)) {
889 		r = -ERESTARTSYS;
890 		goto out;
891 	}
892 
893 	/*
894 	 * The userland program is going to want to know what
895 	 * changed to trigger the event, so we may as well tell
896 	 * him and save an ioctl.
897 	 */
898 	r = __dev_status(md, param);
899 	if (r)
900 		goto out;
901 
902 	table = dm_get_table(md);
903 	if (table) {
904 		retrieve_status(table, param, param_size);
905 		dm_table_put(table);
906 	}
907 
908  out:
909 	dm_put(md);
910 	return r;
911 }
912 
913 static inline int get_mode(struct dm_ioctl *param)
914 {
915 	int mode = FMODE_READ | FMODE_WRITE;
916 
917 	if (param->flags & DM_READONLY_FLAG)
918 		mode = FMODE_READ;
919 
920 	return mode;
921 }
922 
923 static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
924 		       struct dm_target_spec **spec, char **target_params)
925 {
926 	*spec = (struct dm_target_spec *) ((unsigned char *) last + next);
927 	*target_params = (char *) (*spec + 1);
928 
929 	if (*spec < (last + 1))
930 		return -EINVAL;
931 
932 	return invalid_str(*target_params, end);
933 }
934 
935 static int populate_table(struct dm_table *table,
936 			  struct dm_ioctl *param, size_t param_size)
937 {
938 	int r;
939 	unsigned int i = 0;
940 	struct dm_target_spec *spec = (struct dm_target_spec *) param;
941 	uint32_t next = param->data_start;
942 	void *end = (void *) param + param_size;
943 	char *target_params;
944 
945 	if (!param->target_count) {
946 		DMWARN("populate_table: no targets specified");
947 		return -EINVAL;
948 	}
949 
950 	for (i = 0; i < param->target_count; i++) {
951 
952 		r = next_target(spec, next, end, &spec, &target_params);
953 		if (r) {
954 			DMWARN("unable to find target");
955 			return r;
956 		}
957 
958 		r = dm_table_add_target(table, spec->target_type,
959 					(sector_t) spec->sector_start,
960 					(sector_t) spec->length,
961 					target_params);
962 		if (r) {
963 			DMWARN("error adding target to table");
964 			return r;
965 		}
966 
967 		next = spec->next;
968 	}
969 
970 	return dm_table_complete(table);
971 }
972 
973 static int table_load(struct dm_ioctl *param, size_t param_size)
974 {
975 	int r;
976 	struct hash_cell *hc;
977 	struct dm_table *t;
978 
979 	r = dm_table_create(&t, get_mode(param), param->target_count);
980 	if (r)
981 		return r;
982 
983 	r = populate_table(t, param, param_size);
984 	if (r) {
985 		dm_table_put(t);
986 		return r;
987 	}
988 
989 	down_write(&_hash_lock);
990 	hc = __find_device_hash_cell(param);
991 	if (!hc) {
992 		DMWARN("device doesn't appear to be in the dev hash table.");
993 		up_write(&_hash_lock);
994 		dm_table_put(t);
995 		return -ENXIO;
996 	}
997 
998 	if (hc->new_map)
999 		dm_table_put(hc->new_map);
1000 	hc->new_map = t;
1001 	param->flags |= DM_INACTIVE_PRESENT_FLAG;
1002 
1003 	r = __dev_status(hc->md, param);
1004 	up_write(&_hash_lock);
1005 	return r;
1006 }
1007 
1008 static int table_clear(struct dm_ioctl *param, size_t param_size)
1009 {
1010 	int r;
1011 	struct hash_cell *hc;
1012 
1013 	down_write(&_hash_lock);
1014 
1015 	hc = __find_device_hash_cell(param);
1016 	if (!hc) {
1017 		DMWARN("device doesn't appear to be in the dev hash table.");
1018 		up_write(&_hash_lock);
1019 		return -ENXIO;
1020 	}
1021 
1022 	if (hc->new_map) {
1023 		dm_table_put(hc->new_map);
1024 		hc->new_map = NULL;
1025 	}
1026 
1027 	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1028 
1029 	r = __dev_status(hc->md, param);
1030 	up_write(&_hash_lock);
1031 	return r;
1032 }
1033 
1034 /*
1035  * Retrieves a list of devices used by a particular dm device.
1036  */
1037 static void retrieve_deps(struct dm_table *table,
1038 			  struct dm_ioctl *param, size_t param_size)
1039 {
1040 	unsigned int count = 0;
1041 	struct list_head *tmp;
1042 	size_t len, needed;
1043 	struct dm_dev *dd;
1044 	struct dm_target_deps *deps;
1045 
1046 	deps = get_result_buffer(param, param_size, &len);
1047 
1048 	/*
1049 	 * Count the devices.
1050 	 */
1051 	list_for_each (tmp, dm_table_get_devices(table))
1052 		count++;
1053 
1054 	/*
1055 	 * Check we have enough space.
1056 	 */
1057 	needed = sizeof(*deps) + (sizeof(*deps->dev) * count);
1058 	if (len < needed) {
1059 		param->flags |= DM_BUFFER_FULL_FLAG;
1060 		return;
1061 	}
1062 
1063 	/*
1064 	 * Fill in the devices.
1065 	 */
1066 	deps->count = count;
1067 	count = 0;
1068 	list_for_each_entry (dd, dm_table_get_devices(table), list)
1069 		deps->dev[count++] = huge_encode_dev(dd->bdev->bd_dev);
1070 
1071 	param->data_size = param->data_start + needed;
1072 }
1073 
1074 static int table_deps(struct dm_ioctl *param, size_t param_size)
1075 {
1076 	int r = 0;
1077 	struct mapped_device *md;
1078 	struct dm_table *table;
1079 
1080 	md = find_device(param);
1081 	if (!md)
1082 		return -ENXIO;
1083 
1084 	r = __dev_status(md, param);
1085 	if (r)
1086 		goto out;
1087 
1088 	table = dm_get_table(md);
1089 	if (table) {
1090 		retrieve_deps(table, param, param_size);
1091 		dm_table_put(table);
1092 	}
1093 
1094  out:
1095 	dm_put(md);
1096 	return r;
1097 }
1098 
1099 /*
1100  * Return the status of a device as a text string for each
1101  * target.
1102  */
1103 static int table_status(struct dm_ioctl *param, size_t param_size)
1104 {
1105 	int r;
1106 	struct mapped_device *md;
1107 	struct dm_table *table;
1108 
1109 	md = find_device(param);
1110 	if (!md)
1111 		return -ENXIO;
1112 
1113 	r = __dev_status(md, param);
1114 	if (r)
1115 		goto out;
1116 
1117 	table = dm_get_table(md);
1118 	if (table) {
1119 		retrieve_status(table, param, param_size);
1120 		dm_table_put(table);
1121 	}
1122 
1123  out:
1124 	dm_put(md);
1125 	return r;
1126 }
1127 
1128 /*
1129  * Pass a message to the target that's at the supplied device offset.
1130  */
1131 static int target_message(struct dm_ioctl *param, size_t param_size)
1132 {
1133 	int r, argc;
1134 	char **argv;
1135 	struct mapped_device *md;
1136 	struct dm_table *table;
1137 	struct dm_target *ti;
1138 	struct dm_target_msg *tmsg = (void *) param + param->data_start;
1139 
1140 	md = find_device(param);
1141 	if (!md)
1142 		return -ENXIO;
1143 
1144 	r = __dev_status(md, param);
1145 	if (r)
1146 		goto out;
1147 
1148 	if (tmsg < (struct dm_target_msg *) (param + 1) ||
1149 	    invalid_str(tmsg->message, (void *) param + param_size)) {
1150 		DMWARN("Invalid target message parameters.");
1151 		r = -EINVAL;
1152 		goto out;
1153 	}
1154 
1155 	r = dm_split_args(&argc, &argv, tmsg->message);
1156 	if (r) {
1157 		DMWARN("Failed to split target message parameters");
1158 		goto out;
1159 	}
1160 
1161 	table = dm_get_table(md);
1162 	if (!table)
1163 		goto out_argv;
1164 
1165 	if (tmsg->sector >= dm_table_get_size(table)) {
1166 		DMWARN("Target message sector outside device.");
1167 		r = -EINVAL;
1168 		goto out_table;
1169 	}
1170 
1171 	ti = dm_table_find_target(table, tmsg->sector);
1172 	if (ti->type->message)
1173 		r = ti->type->message(ti, argc, argv);
1174 	else {
1175 		DMWARN("Target type does not support messages");
1176 		r = -EINVAL;
1177 	}
1178 
1179  out_table:
1180 	dm_table_put(table);
1181  out_argv:
1182 	kfree(argv);
1183  out:
1184 	param->data_size = 0;
1185 	dm_put(md);
1186 	return r;
1187 }
1188 
1189 /*-----------------------------------------------------------------
1190  * Implementation of open/close/ioctl on the special char
1191  * device.
1192  *---------------------------------------------------------------*/
1193 static ioctl_fn lookup_ioctl(unsigned int cmd)
1194 {
1195 	static struct {
1196 		int cmd;
1197 		ioctl_fn fn;
1198 	} _ioctls[] = {
1199 		{DM_VERSION_CMD, NULL},	/* version is dealt with elsewhere */
1200 		{DM_REMOVE_ALL_CMD, remove_all},
1201 		{DM_LIST_DEVICES_CMD, list_devices},
1202 
1203 		{DM_DEV_CREATE_CMD, dev_create},
1204 		{DM_DEV_REMOVE_CMD, dev_remove},
1205 		{DM_DEV_RENAME_CMD, dev_rename},
1206 		{DM_DEV_SUSPEND_CMD, dev_suspend},
1207 		{DM_DEV_STATUS_CMD, dev_status},
1208 		{DM_DEV_WAIT_CMD, dev_wait},
1209 
1210 		{DM_TABLE_LOAD_CMD, table_load},
1211 		{DM_TABLE_CLEAR_CMD, table_clear},
1212 		{DM_TABLE_DEPS_CMD, table_deps},
1213 		{DM_TABLE_STATUS_CMD, table_status},
1214 
1215 		{DM_LIST_VERSIONS_CMD, list_versions},
1216 
1217 		{DM_TARGET_MSG_CMD, target_message}
1218 	};
1219 
1220 	return (cmd >= ARRAY_SIZE(_ioctls)) ? NULL : _ioctls[cmd].fn;
1221 }
1222 
1223 /*
1224  * As well as checking the version compatibility this always
1225  * copies the kernel interface version out.
1226  */
1227 static int check_version(unsigned int cmd, struct dm_ioctl __user *user)
1228 {
1229 	uint32_t version[3];
1230 	int r = 0;
1231 
1232 	if (copy_from_user(version, user->version, sizeof(version)))
1233 		return -EFAULT;
1234 
1235 	if ((DM_VERSION_MAJOR != version[0]) ||
1236 	    (DM_VERSION_MINOR < version[1])) {
1237 		DMWARN("ioctl interface mismatch: "
1238 		       "kernel(%u.%u.%u), user(%u.%u.%u), cmd(%d)",
1239 		       DM_VERSION_MAJOR, DM_VERSION_MINOR,
1240 		       DM_VERSION_PATCHLEVEL,
1241 		       version[0], version[1], version[2], cmd);
1242 		r = -EINVAL;
1243 	}
1244 
1245 	/*
1246 	 * Fill in the kernel version.
1247 	 */
1248 	version[0] = DM_VERSION_MAJOR;
1249 	version[1] = DM_VERSION_MINOR;
1250 	version[2] = DM_VERSION_PATCHLEVEL;
1251 	if (copy_to_user(user->version, version, sizeof(version)))
1252 		return -EFAULT;
1253 
1254 	return r;
1255 }
1256 
1257 static void free_params(struct dm_ioctl *param)
1258 {
1259 	vfree(param);
1260 }
1261 
1262 static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl **param)
1263 {
1264 	struct dm_ioctl tmp, *dmi;
1265 
1266 	if (copy_from_user(&tmp, user, sizeof(tmp)))
1267 		return -EFAULT;
1268 
1269 	if (tmp.data_size < sizeof(tmp))
1270 		return -EINVAL;
1271 
1272 	dmi = (struct dm_ioctl *) vmalloc(tmp.data_size);
1273 	if (!dmi)
1274 		return -ENOMEM;
1275 
1276 	if (copy_from_user(dmi, user, tmp.data_size)) {
1277 		vfree(dmi);
1278 		return -EFAULT;
1279 	}
1280 
1281 	*param = dmi;
1282 	return 0;
1283 }
1284 
1285 static int validate_params(uint cmd, struct dm_ioctl *param)
1286 {
1287 	/* Always clear this flag */
1288 	param->flags &= ~DM_BUFFER_FULL_FLAG;
1289 
1290 	/* Ignores parameters */
1291 	if (cmd == DM_REMOVE_ALL_CMD ||
1292 	    cmd == DM_LIST_DEVICES_CMD ||
1293 	    cmd == DM_LIST_VERSIONS_CMD)
1294 		return 0;
1295 
1296 	if ((cmd == DM_DEV_CREATE_CMD)) {
1297 		if (!*param->name) {
1298 			DMWARN("name not supplied when creating device");
1299 			return -EINVAL;
1300 		}
1301 	} else if ((*param->uuid && *param->name)) {
1302 		DMWARN("only supply one of name or uuid, cmd(%u)", cmd);
1303 		return -EINVAL;
1304 	}
1305 
1306 	/* Ensure strings are terminated */
1307 	param->name[DM_NAME_LEN - 1] = '\0';
1308 	param->uuid[DM_UUID_LEN - 1] = '\0';
1309 
1310 	return 0;
1311 }
1312 
1313 static int ctl_ioctl(struct inode *inode, struct file *file,
1314 		     uint command, ulong u)
1315 {
1316 	int r = 0;
1317 	unsigned int cmd;
1318 	struct dm_ioctl *param;
1319 	struct dm_ioctl __user *user = (struct dm_ioctl __user *) u;
1320 	ioctl_fn fn = NULL;
1321 	size_t param_size;
1322 
1323 	/* only root can play with this */
1324 	if (!capable(CAP_SYS_ADMIN))
1325 		return -EACCES;
1326 
1327 	if (_IOC_TYPE(command) != DM_IOCTL)
1328 		return -ENOTTY;
1329 
1330 	cmd = _IOC_NR(command);
1331 
1332 	/*
1333 	 * Check the interface version passed in.  This also
1334 	 * writes out the kernel's interface version.
1335 	 */
1336 	r = check_version(cmd, user);
1337 	if (r)
1338 		return r;
1339 
1340 	/*
1341 	 * Nothing more to do for the version command.
1342 	 */
1343 	if (cmd == DM_VERSION_CMD)
1344 		return 0;
1345 
1346 	fn = lookup_ioctl(cmd);
1347 	if (!fn) {
1348 		DMWARN("dm_ctl_ioctl: unknown command 0x%x", command);
1349 		return -ENOTTY;
1350 	}
1351 
1352 	/*
1353 	 * Trying to avoid low memory issues when a device is
1354 	 * suspended.
1355 	 */
1356 	current->flags |= PF_MEMALLOC;
1357 
1358 	/*
1359 	 * Copy the parameters into kernel space.
1360 	 */
1361 	r = copy_params(user, &param);
1362 	if (r) {
1363 		current->flags &= ~PF_MEMALLOC;
1364 		return r;
1365 	}
1366 
1367 	/*
1368 	 * FIXME: eventually we will remove the PF_MEMALLOC flag
1369 	 * here.  However the tools still do nasty things like
1370 	 * 'load' while a device is suspended.
1371 	 */
1372 
1373 	r = validate_params(cmd, param);
1374 	if (r)
1375 		goto out;
1376 
1377 	param_size = param->data_size;
1378 	param->data_size = sizeof(*param);
1379 	r = fn(param, param_size);
1380 
1381 	/*
1382 	 * Copy the results back to userland.
1383 	 */
1384 	if (!r && copy_to_user(user, param, param->data_size))
1385 		r = -EFAULT;
1386 
1387  out:
1388 	free_params(param);
1389 	current->flags &= ~PF_MEMALLOC;
1390 	return r;
1391 }
1392 
1393 static struct file_operations _ctl_fops = {
1394 	.ioctl	 = ctl_ioctl,
1395 	.owner	 = THIS_MODULE,
1396 };
1397 
1398 static struct miscdevice _dm_misc = {
1399 	.minor 		= MISC_DYNAMIC_MINOR,
1400 	.name  		= DM_NAME,
1401 	.devfs_name 	= "mapper/control",
1402 	.fops  		= &_ctl_fops
1403 };
1404 
1405 /*
1406  * Create misc character device and link to DM_DIR/control.
1407  */
1408 int __init dm_interface_init(void)
1409 {
1410 	int r;
1411 
1412 	r = dm_hash_init();
1413 	if (r)
1414 		return r;
1415 
1416 	r = misc_register(&_dm_misc);
1417 	if (r) {
1418 		DMERR("misc_register failed for control device");
1419 		dm_hash_exit();
1420 		return r;
1421 	}
1422 
1423 	DMINFO("%d.%d.%d%s initialised: %s", DM_VERSION_MAJOR,
1424 	       DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL, DM_VERSION_EXTRA,
1425 	       DM_DRIVER_EMAIL);
1426 	return 0;
1427 }
1428 
1429 void dm_interface_exit(void)
1430 {
1431 	if (misc_deregister(&_dm_misc) < 0)
1432 		DMERR("misc_deregister failed for control device");
1433 
1434 	dm_hash_exit();
1435 }
1436