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 *---------------------------------------------------------------*/
29struct hash_cell {
30	struct list_head name_list;
31	struct list_head uuid_list;
32
33	char *name;
34	char *uuid;
35	struct mapped_device *md;
36	struct dm_table *new_map;
37};
38
39/*
40 * A dummy definition to make RCU happy.
41 * struct dm_table should never be dereferenced in this file.
42 */
43struct dm_table {
44	int undefined__;
45};
46
47struct vers_iter {
48    size_t param_size;
49    struct dm_target_versions *vers, *old_vers;
50    char *end;
51    uint32_t flags;
52};
53
54
55#define NUM_BUCKETS 64
56#define MASK_BUCKETS (NUM_BUCKETS - 1)
57static struct list_head _name_buckets[NUM_BUCKETS];
58static struct list_head _uuid_buckets[NUM_BUCKETS];
59
60static void dm_hash_remove_all(bool keep_open_devices, bool mark_deferred, bool only_deferred);
61
62/*
63 * Guards access to both hash tables.
64 */
65static DECLARE_RWSEM(_hash_lock);
66
67/*
68 * Protects use of mdptr to obtain hash cell name and uuid from mapped device.
69 */
70static DEFINE_MUTEX(dm_hash_cells_mutex);
71
72static void init_buckets(struct list_head *buckets)
73{
74	unsigned int i;
75
76	for (i = 0; i < NUM_BUCKETS; i++)
77		INIT_LIST_HEAD(buckets + i);
78}
79
80static int dm_hash_init(void)
81{
82	init_buckets(_name_buckets);
83	init_buckets(_uuid_buckets);
84	return 0;
85}
86
87static void dm_hash_exit(void)
88{
89	dm_hash_remove_all(false, false, false);
90}
91
92/*-----------------------------------------------------------------
93 * Hash function:
94 * We're not really concerned with the str hash function being
95 * fast since it's only used by the ioctl interface.
96 *---------------------------------------------------------------*/
97static unsigned int hash_str(const char *str)
98{
99	const unsigned int hash_mult = 2654435387U;
100	unsigned int h = 0;
101
102	while (*str)
103		h = (h + (unsigned int) *str++) * hash_mult;
104
105	return h & MASK_BUCKETS;
106}
107
108/*-----------------------------------------------------------------
109 * Code for looking up a device by name
110 *---------------------------------------------------------------*/
111static struct hash_cell *__get_name_cell(const char *str)
112{
113	struct hash_cell *hc;
114	unsigned int h = hash_str(str);
115
116	list_for_each_entry (hc, _name_buckets + h, name_list)
117		if (!strcmp(hc->name, str)) {
118			dm_get(hc->md);
119			return hc;
120		}
121
122	return NULL;
123}
124
125static struct hash_cell *__get_uuid_cell(const char *str)
126{
127	struct hash_cell *hc;
128	unsigned int h = hash_str(str);
129
130	list_for_each_entry (hc, _uuid_buckets + h, uuid_list)
131		if (!strcmp(hc->uuid, str)) {
132			dm_get(hc->md);
133			return hc;
134		}
135
136	return NULL;
137}
138
139static struct hash_cell *__get_dev_cell(uint64_t dev)
140{
141	struct mapped_device *md;
142	struct hash_cell *hc;
143
144	md = dm_get_md(huge_decode_dev(dev));
145	if (!md)
146		return NULL;
147
148	hc = dm_get_mdptr(md);
149	if (!hc) {
150		dm_put(md);
151		return NULL;
152	}
153
154	return hc;
155}
156
157/*-----------------------------------------------------------------
158 * Inserting, removing and renaming a device.
159 *---------------------------------------------------------------*/
160static struct hash_cell *alloc_cell(const char *name, const char *uuid,
161				    struct mapped_device *md)
162{
163	struct hash_cell *hc;
164
165	hc = kmalloc(sizeof(*hc), GFP_KERNEL);
166	if (!hc)
167		return NULL;
168
169	hc->name = kstrdup(name, GFP_KERNEL);
170	if (!hc->name) {
171		kfree(hc);
172		return NULL;
173	}
174
175	if (!uuid)
176		hc->uuid = NULL;
177
178	else {
179		hc->uuid = kstrdup(uuid, GFP_KERNEL);
180		if (!hc->uuid) {
181			kfree(hc->name);
182			kfree(hc);
183			return NULL;
184		}
185	}
186
187	INIT_LIST_HEAD(&hc->name_list);
188	INIT_LIST_HEAD(&hc->uuid_list);
189	hc->md = md;
190	hc->new_map = NULL;
191	return hc;
192}
193
194static void free_cell(struct hash_cell *hc)
195{
196	if (hc) {
197		kfree(hc->name);
198		kfree(hc->uuid);
199		kfree(hc);
200	}
201}
202
203/*
204 * The kdev_t and uuid of a device can never change once it is
205 * initially inserted.
206 */
207static int dm_hash_insert(const char *name, const char *uuid, struct mapped_device *md)
208{
209	struct hash_cell *cell, *hc;
210
211	/*
212	 * Allocate the new cells.
213	 */
214	cell = alloc_cell(name, uuid, md);
215	if (!cell)
216		return -ENOMEM;
217
218	/*
219	 * Insert the cell into both hash tables.
220	 */
221	down_write(&_hash_lock);
222	hc = __get_name_cell(name);
223	if (hc) {
224		dm_put(hc->md);
225		goto bad;
226	}
227
228	list_add(&cell->name_list, _name_buckets + hash_str(name));
229
230	if (uuid) {
231		hc = __get_uuid_cell(uuid);
232		if (hc) {
233			list_del(&cell->name_list);
234			dm_put(hc->md);
235			goto bad;
236		}
237		list_add(&cell->uuid_list, _uuid_buckets + hash_str(uuid));
238	}
239	dm_get(md);
240	mutex_lock(&dm_hash_cells_mutex);
241	dm_set_mdptr(md, cell);
242	mutex_unlock(&dm_hash_cells_mutex);
243	up_write(&_hash_lock);
244
245	return 0;
246
247 bad:
248	up_write(&_hash_lock);
249	free_cell(cell);
250	return -EBUSY;
251}
252
253static struct dm_table *__hash_remove(struct hash_cell *hc)
254{
255	struct dm_table *table;
256	int srcu_idx;
257
258	/* remove from the dev hash */
259	list_del(&hc->uuid_list);
260	list_del(&hc->name_list);
261	mutex_lock(&dm_hash_cells_mutex);
262	dm_set_mdptr(hc->md, NULL);
263	mutex_unlock(&dm_hash_cells_mutex);
264
265	table = dm_get_live_table(hc->md, &srcu_idx);
266	if (table)
267		dm_table_event(table);
268	dm_put_live_table(hc->md, srcu_idx);
269
270	table = NULL;
271	if (hc->new_map)
272		table = hc->new_map;
273	dm_put(hc->md);
274	free_cell(hc);
275
276	return table;
277}
278
279static void dm_hash_remove_all(bool keep_open_devices, bool mark_deferred, bool only_deferred)
280{
281	int i, dev_skipped;
282	struct hash_cell *hc;
283	struct mapped_device *md;
284	struct dm_table *t;
285
286retry:
287	dev_skipped = 0;
288
289	down_write(&_hash_lock);
290
291	for (i = 0; i < NUM_BUCKETS; i++) {
292		list_for_each_entry(hc, _name_buckets + i, name_list) {
293			md = hc->md;
294			dm_get(md);
295
296			if (keep_open_devices &&
297			    dm_lock_for_deletion(md, mark_deferred, only_deferred)) {
298				dm_put(md);
299				dev_skipped++;
300				continue;
301			}
302
303			t = __hash_remove(hc);
304
305			up_write(&_hash_lock);
306
307			if (t) {
308				dm_sync_table(md);
309				dm_table_destroy(t);
310			}
311			dm_put(md);
312			if (likely(keep_open_devices))
313				dm_destroy(md);
314			else
315				dm_destroy_immediate(md);
316
317			/*
318			 * Some mapped devices may be using other mapped
319			 * devices, so repeat until we make no further
320			 * progress.  If a new mapped device is created
321			 * here it will also get removed.
322			 */
323			goto retry;
324		}
325	}
326
327	up_write(&_hash_lock);
328
329	if (dev_skipped)
330		DMWARN("remove_all left %d open device(s)", dev_skipped);
331}
332
333/*
334 * Set the uuid of a hash_cell that isn't already set.
335 */
336static void __set_cell_uuid(struct hash_cell *hc, char *new_uuid)
337{
338	mutex_lock(&dm_hash_cells_mutex);
339	hc->uuid = new_uuid;
340	mutex_unlock(&dm_hash_cells_mutex);
341
342	list_add(&hc->uuid_list, _uuid_buckets + hash_str(new_uuid));
343}
344
345/*
346 * Changes the name of a hash_cell and returns the old name for
347 * the caller to free.
348 */
349static char *__change_cell_name(struct hash_cell *hc, char *new_name)
350{
351	char *old_name;
352
353	/*
354	 * Rename and move the name cell.
355	 */
356	list_del(&hc->name_list);
357	old_name = hc->name;
358
359	mutex_lock(&dm_hash_cells_mutex);
360	hc->name = new_name;
361	mutex_unlock(&dm_hash_cells_mutex);
362
363	list_add(&hc->name_list, _name_buckets + hash_str(new_name));
364
365	return old_name;
366}
367
368static struct mapped_device *dm_hash_rename(struct dm_ioctl *param,
369					    const char *new)
370{
371	char *new_data, *old_name = NULL;
372	struct hash_cell *hc;
373	struct dm_table *table;
374	struct mapped_device *md;
375	unsigned change_uuid = (param->flags & DM_UUID_FLAG) ? 1 : 0;
376	int srcu_idx;
377
378	/*
379	 * duplicate new.
380	 */
381	new_data = kstrdup(new, GFP_KERNEL);
382	if (!new_data)
383		return ERR_PTR(-ENOMEM);
384
385	down_write(&_hash_lock);
386
387	/*
388	 * Is new free ?
389	 */
390	if (change_uuid)
391		hc = __get_uuid_cell(new);
392	else
393		hc = __get_name_cell(new);
394
395	if (hc) {
396		DMWARN("Unable to change %s on mapped device %s to one that "
397		       "already exists: %s",
398		       change_uuid ? "uuid" : "name",
399		       param->name, new);
400		dm_put(hc->md);
401		up_write(&_hash_lock);
402		kfree(new_data);
403		return ERR_PTR(-EBUSY);
404	}
405
406	/*
407	 * Is there such a device as 'old' ?
408	 */
409	hc = __get_name_cell(param->name);
410	if (!hc) {
411		DMWARN("Unable to rename non-existent device, %s to %s%s",
412		       param->name, change_uuid ? "uuid " : "", new);
413		up_write(&_hash_lock);
414		kfree(new_data);
415		return ERR_PTR(-ENXIO);
416	}
417
418	/*
419	 * Does this device already have a uuid?
420	 */
421	if (change_uuid && hc->uuid) {
422		DMWARN("Unable to change uuid of mapped device %s to %s "
423		       "because uuid is already set to %s",
424		       param->name, new, hc->uuid);
425		dm_put(hc->md);
426		up_write(&_hash_lock);
427		kfree(new_data);
428		return ERR_PTR(-EINVAL);
429	}
430
431	if (change_uuid)
432		__set_cell_uuid(hc, new_data);
433	else
434		old_name = __change_cell_name(hc, new_data);
435
436	/*
437	 * Wake up any dm event waiters.
438	 */
439	table = dm_get_live_table(hc->md, &srcu_idx);
440	if (table)
441		dm_table_event(table);
442	dm_put_live_table(hc->md, srcu_idx);
443
444	if (!dm_kobject_uevent(hc->md, KOBJ_CHANGE, param->event_nr))
445		param->flags |= DM_UEVENT_GENERATED_FLAG;
446
447	md = hc->md;
448	up_write(&_hash_lock);
449	kfree(old_name);
450
451	return md;
452}
453
454void dm_deferred_remove(void)
455{
456	dm_hash_remove_all(true, false, true);
457}
458
459/*-----------------------------------------------------------------
460 * Implementation of the ioctl commands
461 *---------------------------------------------------------------*/
462/*
463 * All the ioctl commands get dispatched to functions with this
464 * prototype.
465 */
466typedef int (*ioctl_fn)(struct dm_ioctl *param, size_t param_size);
467
468static int remove_all(struct dm_ioctl *param, size_t param_size)
469{
470	dm_hash_remove_all(true, !!(param->flags & DM_DEFERRED_REMOVE), false);
471	param->data_size = 0;
472	return 0;
473}
474
475/*
476 * Round up the ptr to an 8-byte boundary.
477 */
478#define ALIGN_MASK 7
479static inline void *align_ptr(void *ptr)
480{
481	return (void *) (((size_t) (ptr + ALIGN_MASK)) & ~ALIGN_MASK);
482}
483
484/*
485 * Retrieves the data payload buffer from an already allocated
486 * struct dm_ioctl.
487 */
488static void *get_result_buffer(struct dm_ioctl *param, size_t param_size,
489			       size_t *len)
490{
491	param->data_start = align_ptr(param + 1) - (void *) param;
492
493	if (param->data_start < param_size)
494		*len = param_size - param->data_start;
495	else
496		*len = 0;
497
498	return ((void *) param) + param->data_start;
499}
500
501static int list_devices(struct dm_ioctl *param, size_t param_size)
502{
503	unsigned int i;
504	struct hash_cell *hc;
505	size_t len, needed = 0;
506	struct gendisk *disk;
507	struct dm_name_list *nl, *old_nl = NULL;
508
509	down_write(&_hash_lock);
510
511	/*
512	 * Loop through all the devices working out how much
513	 * space we need.
514	 */
515	for (i = 0; i < NUM_BUCKETS; i++) {
516		list_for_each_entry (hc, _name_buckets + i, name_list) {
517			needed += sizeof(struct dm_name_list);
518			needed += strlen(hc->name) + 1;
519			needed += ALIGN_MASK;
520		}
521	}
522
523	/*
524	 * Grab our output buffer.
525	 */
526	nl = get_result_buffer(param, param_size, &len);
527	if (len < needed) {
528		param->flags |= DM_BUFFER_FULL_FLAG;
529		goto out;
530	}
531	param->data_size = param->data_start + needed;
532
533	nl->dev = 0;	/* Flags no data */
534
535	/*
536	 * Now loop through filling out the names.
537	 */
538	for (i = 0; i < NUM_BUCKETS; i++) {
539		list_for_each_entry (hc, _name_buckets + i, name_list) {
540			if (old_nl)
541				old_nl->next = (uint32_t) ((void *) nl -
542							   (void *) old_nl);
543			disk = dm_disk(hc->md);
544			nl->dev = huge_encode_dev(disk_devt(disk));
545			nl->next = 0;
546			strcpy(nl->name, hc->name);
547
548			old_nl = nl;
549			nl = align_ptr(((void *) ++nl) + strlen(hc->name) + 1);
550		}
551	}
552
553 out:
554	up_write(&_hash_lock);
555	return 0;
556}
557
558static void list_version_get_needed(struct target_type *tt, void *needed_param)
559{
560    size_t *needed = needed_param;
561
562    *needed += sizeof(struct dm_target_versions);
563    *needed += strlen(tt->name);
564    *needed += ALIGN_MASK;
565}
566
567static void list_version_get_info(struct target_type *tt, void *param)
568{
569    struct vers_iter *info = param;
570
571    /* Check space - it might have changed since the first iteration */
572    if ((char *)info->vers + sizeof(tt->version) + strlen(tt->name) + 1 >
573	info->end) {
574
575	info->flags = DM_BUFFER_FULL_FLAG;
576	return;
577    }
578
579    if (info->old_vers)
580	info->old_vers->next = (uint32_t) ((void *)info->vers -
581					   (void *)info->old_vers);
582    info->vers->version[0] = tt->version[0];
583    info->vers->version[1] = tt->version[1];
584    info->vers->version[2] = tt->version[2];
585    info->vers->next = 0;
586    strcpy(info->vers->name, tt->name);
587
588    info->old_vers = info->vers;
589    info->vers = align_ptr(((void *) ++info->vers) + strlen(tt->name) + 1);
590}
591
592static int list_versions(struct dm_ioctl *param, size_t param_size)
593{
594	size_t len, needed = 0;
595	struct dm_target_versions *vers;
596	struct vers_iter iter_info;
597
598	/*
599	 * Loop through all the devices working out how much
600	 * space we need.
601	 */
602	dm_target_iterate(list_version_get_needed, &needed);
603
604	/*
605	 * Grab our output buffer.
606	 */
607	vers = get_result_buffer(param, param_size, &len);
608	if (len < needed) {
609		param->flags |= DM_BUFFER_FULL_FLAG;
610		goto out;
611	}
612	param->data_size = param->data_start + needed;
613
614	iter_info.param_size = param_size;
615	iter_info.old_vers = NULL;
616	iter_info.vers = vers;
617	iter_info.flags = 0;
618	iter_info.end = (char *)vers+len;
619
620	/*
621	 * Now loop through filling out the names & versions.
622	 */
623	dm_target_iterate(list_version_get_info, &iter_info);
624	param->flags |= iter_info.flags;
625
626 out:
627	return 0;
628}
629
630static int check_name(const char *name)
631{
632	if (strchr(name, '/')) {
633		DMWARN("invalid device name");
634		return -EINVAL;
635	}
636
637	return 0;
638}
639
640/*
641 * On successful return, the caller must not attempt to acquire
642 * _hash_lock without first calling dm_put_live_table, because dm_table_destroy
643 * waits for this dm_put_live_table and could be called under this lock.
644 */
645static struct dm_table *dm_get_inactive_table(struct mapped_device *md, int *srcu_idx)
646{
647	struct hash_cell *hc;
648	struct dm_table *table = NULL;
649
650	/* increment rcu count, we don't care about the table pointer */
651	dm_get_live_table(md, srcu_idx);
652
653	down_read(&_hash_lock);
654	hc = dm_get_mdptr(md);
655	if (!hc || hc->md != md) {
656		DMWARN("device has been removed from the dev hash table.");
657		goto out;
658	}
659
660	table = hc->new_map;
661
662out:
663	up_read(&_hash_lock);
664
665	return table;
666}
667
668static struct dm_table *dm_get_live_or_inactive_table(struct mapped_device *md,
669						      struct dm_ioctl *param,
670						      int *srcu_idx)
671{
672	return (param->flags & DM_QUERY_INACTIVE_TABLE_FLAG) ?
673		dm_get_inactive_table(md, srcu_idx) : dm_get_live_table(md, srcu_idx);
674}
675
676/*
677 * Fills in a dm_ioctl structure, ready for sending back to
678 * userland.
679 */
680static void __dev_status(struct mapped_device *md, struct dm_ioctl *param)
681{
682	struct gendisk *disk = dm_disk(md);
683	struct dm_table *table;
684	int srcu_idx;
685
686	param->flags &= ~(DM_SUSPEND_FLAG | DM_READONLY_FLAG |
687			  DM_ACTIVE_PRESENT_FLAG | DM_INTERNAL_SUSPEND_FLAG);
688
689	if (dm_suspended_md(md))
690		param->flags |= DM_SUSPEND_FLAG;
691
692	if (dm_suspended_internally_md(md))
693		param->flags |= DM_INTERNAL_SUSPEND_FLAG;
694
695	if (dm_test_deferred_remove_flag(md))
696		param->flags |= DM_DEFERRED_REMOVE;
697
698	param->dev = huge_encode_dev(disk_devt(disk));
699
700	/*
701	 * Yes, this will be out of date by the time it gets back
702	 * to userland, but it is still very useful for
703	 * debugging.
704	 */
705	param->open_count = dm_open_count(md);
706
707	param->event_nr = dm_get_event_nr(md);
708	param->target_count = 0;
709
710	table = dm_get_live_table(md, &srcu_idx);
711	if (table) {
712		if (!(param->flags & DM_QUERY_INACTIVE_TABLE_FLAG)) {
713			if (get_disk_ro(disk))
714				param->flags |= DM_READONLY_FLAG;
715			param->target_count = dm_table_get_num_targets(table);
716		}
717
718		param->flags |= DM_ACTIVE_PRESENT_FLAG;
719	}
720	dm_put_live_table(md, srcu_idx);
721
722	if (param->flags & DM_QUERY_INACTIVE_TABLE_FLAG) {
723		int srcu_idx;
724		table = dm_get_inactive_table(md, &srcu_idx);
725		if (table) {
726			if (!(dm_table_get_mode(table) & FMODE_WRITE))
727				param->flags |= DM_READONLY_FLAG;
728			param->target_count = dm_table_get_num_targets(table);
729		}
730		dm_put_live_table(md, srcu_idx);
731	}
732}
733
734static int dev_create(struct dm_ioctl *param, size_t param_size)
735{
736	int r, m = DM_ANY_MINOR;
737	struct mapped_device *md;
738
739	r = check_name(param->name);
740	if (r)
741		return r;
742
743	if (param->flags & DM_PERSISTENT_DEV_FLAG)
744		m = MINOR(huge_decode_dev(param->dev));
745
746	r = dm_create(m, &md);
747	if (r)
748		return r;
749
750	r = dm_hash_insert(param->name, *param->uuid ? param->uuid : NULL, md);
751	if (r) {
752		dm_put(md);
753		dm_destroy(md);
754		return r;
755	}
756
757	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
758
759	__dev_status(md, param);
760
761	dm_put(md);
762
763	return 0;
764}
765
766/*
767 * Always use UUID for lookups if it's present, otherwise use name or dev.
768 */
769static struct hash_cell *__find_device_hash_cell(struct dm_ioctl *param)
770{
771	struct hash_cell *hc = NULL;
772
773	if (*param->uuid) {
774		if (*param->name || param->dev)
775			return NULL;
776
777		hc = __get_uuid_cell(param->uuid);
778		if (!hc)
779			return NULL;
780	} else if (*param->name) {
781		if (param->dev)
782			return NULL;
783
784		hc = __get_name_cell(param->name);
785		if (!hc)
786			return NULL;
787	} else if (param->dev) {
788		hc = __get_dev_cell(param->dev);
789		if (!hc)
790			return NULL;
791	} else
792		return NULL;
793
794	/*
795	 * Sneakily write in both the name and the uuid
796	 * while we have the cell.
797	 */
798	strlcpy(param->name, hc->name, sizeof(param->name));
799	if (hc->uuid)
800		strlcpy(param->uuid, hc->uuid, sizeof(param->uuid));
801	else
802		param->uuid[0] = '\0';
803
804	if (hc->new_map)
805		param->flags |= DM_INACTIVE_PRESENT_FLAG;
806	else
807		param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
808
809	return hc;
810}
811
812static struct mapped_device *find_device(struct dm_ioctl *param)
813{
814	struct hash_cell *hc;
815	struct mapped_device *md = NULL;
816
817	down_read(&_hash_lock);
818	hc = __find_device_hash_cell(param);
819	if (hc)
820		md = hc->md;
821	up_read(&_hash_lock);
822
823	return md;
824}
825
826static int dev_remove(struct dm_ioctl *param, size_t param_size)
827{
828	struct hash_cell *hc;
829	struct mapped_device *md;
830	int r;
831	struct dm_table *t;
832
833	down_write(&_hash_lock);
834	hc = __find_device_hash_cell(param);
835
836	if (!hc) {
837		DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
838		up_write(&_hash_lock);
839		return -ENXIO;
840	}
841
842	md = hc->md;
843
844	/*
845	 * Ensure the device is not open and nothing further can open it.
846	 */
847	r = dm_lock_for_deletion(md, !!(param->flags & DM_DEFERRED_REMOVE), false);
848	if (r) {
849		if (r == -EBUSY && param->flags & DM_DEFERRED_REMOVE) {
850			up_write(&_hash_lock);
851			dm_put(md);
852			return 0;
853		}
854		DMDEBUG_LIMIT("unable to remove open device %s", hc->name);
855		up_write(&_hash_lock);
856		dm_put(md);
857		return r;
858	}
859
860	t = __hash_remove(hc);
861	up_write(&_hash_lock);
862
863	if (t) {
864		dm_sync_table(md);
865		dm_table_destroy(t);
866	}
867
868	param->flags &= ~DM_DEFERRED_REMOVE;
869
870	if (!dm_kobject_uevent(md, KOBJ_REMOVE, param->event_nr))
871		param->flags |= DM_UEVENT_GENERATED_FLAG;
872
873	dm_put(md);
874	dm_destroy(md);
875	return 0;
876}
877
878/*
879 * Check a string doesn't overrun the chunk of
880 * memory we copied from userland.
881 */
882static int invalid_str(char *str, void *end)
883{
884	while ((void *) str < end)
885		if (!*str++)
886			return 0;
887
888	return -EINVAL;
889}
890
891static int dev_rename(struct dm_ioctl *param, size_t param_size)
892{
893	int r;
894	char *new_data = (char *) param + param->data_start;
895	struct mapped_device *md;
896	unsigned change_uuid = (param->flags & DM_UUID_FLAG) ? 1 : 0;
897
898	if (new_data < param->data ||
899	    invalid_str(new_data, (void *) param + param_size) || !*new_data ||
900	    strlen(new_data) > (change_uuid ? DM_UUID_LEN - 1 : DM_NAME_LEN - 1)) {
901		DMWARN("Invalid new mapped device name or uuid string supplied.");
902		return -EINVAL;
903	}
904
905	if (!change_uuid) {
906		r = check_name(new_data);
907		if (r)
908			return r;
909	}
910
911	md = dm_hash_rename(param, new_data);
912	if (IS_ERR(md))
913		return PTR_ERR(md);
914
915	__dev_status(md, param);
916	dm_put(md);
917
918	return 0;
919}
920
921static int dev_set_geometry(struct dm_ioctl *param, size_t param_size)
922{
923	int r = -EINVAL, x;
924	struct mapped_device *md;
925	struct hd_geometry geometry;
926	unsigned long indata[4];
927	char *geostr = (char *) param + param->data_start;
928	char dummy;
929
930	md = find_device(param);
931	if (!md)
932		return -ENXIO;
933
934	if (geostr < param->data ||
935	    invalid_str(geostr, (void *) param + param_size)) {
936		DMWARN("Invalid geometry supplied.");
937		goto out;
938	}
939
940	x = sscanf(geostr, "%lu %lu %lu %lu%c", indata,
941		   indata + 1, indata + 2, indata + 3, &dummy);
942
943	if (x != 4) {
944		DMWARN("Unable to interpret geometry settings.");
945		goto out;
946	}
947
948	if (indata[0] > 65535 || indata[1] > 255 ||
949	    indata[2] > 255 || indata[3] > ULONG_MAX) {
950		DMWARN("Geometry exceeds range limits.");
951		goto out;
952	}
953
954	geometry.cylinders = indata[0];
955	geometry.heads = indata[1];
956	geometry.sectors = indata[2];
957	geometry.start = indata[3];
958
959	r = dm_set_geometry(md, &geometry);
960
961	param->data_size = 0;
962
963out:
964	dm_put(md);
965	return r;
966}
967
968static int do_suspend(struct dm_ioctl *param)
969{
970	int r = 0;
971	unsigned suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
972	struct mapped_device *md;
973
974	md = find_device(param);
975	if (!md)
976		return -ENXIO;
977
978	if (param->flags & DM_SKIP_LOCKFS_FLAG)
979		suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
980	if (param->flags & DM_NOFLUSH_FLAG)
981		suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
982
983	if (!dm_suspended_md(md)) {
984		r = dm_suspend(md, suspend_flags);
985		if (r)
986			goto out;
987	}
988
989	__dev_status(md, param);
990
991out:
992	dm_put(md);
993
994	return r;
995}
996
997static int do_resume(struct dm_ioctl *param)
998{
999	int r = 0;
1000	unsigned suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
1001	struct hash_cell *hc;
1002	struct mapped_device *md;
1003	struct dm_table *new_map, *old_map = NULL;
1004
1005	down_write(&_hash_lock);
1006
1007	hc = __find_device_hash_cell(param);
1008	if (!hc) {
1009		DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
1010		up_write(&_hash_lock);
1011		return -ENXIO;
1012	}
1013
1014	md = hc->md;
1015
1016	new_map = hc->new_map;
1017	hc->new_map = NULL;
1018	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1019
1020	up_write(&_hash_lock);
1021
1022	/* Do we need to load a new map ? */
1023	if (new_map) {
1024		/* Suspend if it isn't already suspended */
1025		if (param->flags & DM_SKIP_LOCKFS_FLAG)
1026			suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
1027		if (param->flags & DM_NOFLUSH_FLAG)
1028			suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
1029		if (!dm_suspended_md(md))
1030			dm_suspend(md, suspend_flags);
1031
1032		old_map = dm_swap_table(md, new_map);
1033		if (IS_ERR(old_map)) {
1034			dm_sync_table(md);
1035			dm_table_destroy(new_map);
1036			dm_put(md);
1037			return PTR_ERR(old_map);
1038		}
1039
1040		if (dm_table_get_mode(new_map) & FMODE_WRITE)
1041			set_disk_ro(dm_disk(md), 0);
1042		else
1043			set_disk_ro(dm_disk(md), 1);
1044	}
1045
1046	if (dm_suspended_md(md)) {
1047		r = dm_resume(md);
1048		if (!r && !dm_kobject_uevent(md, KOBJ_CHANGE, param->event_nr))
1049			param->flags |= DM_UEVENT_GENERATED_FLAG;
1050	}
1051
1052	/*
1053	 * Since dm_swap_table synchronizes RCU, nobody should be in
1054	 * read-side critical section already.
1055	 */
1056	if (old_map)
1057		dm_table_destroy(old_map);
1058
1059	if (!r)
1060		__dev_status(md, param);
1061
1062	dm_put(md);
1063	return r;
1064}
1065
1066/*
1067 * Set or unset the suspension state of a device.
1068 * If the device already is in the requested state we just return its status.
1069 */
1070static int dev_suspend(struct dm_ioctl *param, size_t param_size)
1071{
1072	if (param->flags & DM_SUSPEND_FLAG)
1073		return do_suspend(param);
1074
1075	return do_resume(param);
1076}
1077
1078/*
1079 * Copies device info back to user space, used by
1080 * the create and info ioctls.
1081 */
1082static int dev_status(struct dm_ioctl *param, size_t param_size)
1083{
1084	struct mapped_device *md;
1085
1086	md = find_device(param);
1087	if (!md)
1088		return -ENXIO;
1089
1090	__dev_status(md, param);
1091	dm_put(md);
1092
1093	return 0;
1094}
1095
1096/*
1097 * Build up the status struct for each target
1098 */
1099static void retrieve_status(struct dm_table *table,
1100			    struct dm_ioctl *param, size_t param_size)
1101{
1102	unsigned int i, num_targets;
1103	struct dm_target_spec *spec;
1104	char *outbuf, *outptr;
1105	status_type_t type;
1106	size_t remaining, len, used = 0;
1107	unsigned status_flags = 0;
1108
1109	outptr = outbuf = get_result_buffer(param, param_size, &len);
1110
1111	if (param->flags & DM_STATUS_TABLE_FLAG)
1112		type = STATUSTYPE_TABLE;
1113	else
1114		type = STATUSTYPE_INFO;
1115
1116	/* Get all the target info */
1117	num_targets = dm_table_get_num_targets(table);
1118	for (i = 0; i < num_targets; i++) {
1119		struct dm_target *ti = dm_table_get_target(table, i);
1120		size_t l;
1121
1122		remaining = len - (outptr - outbuf);
1123		if (remaining <= sizeof(struct dm_target_spec)) {
1124			param->flags |= DM_BUFFER_FULL_FLAG;
1125			break;
1126		}
1127
1128		spec = (struct dm_target_spec *) outptr;
1129
1130		spec->status = 0;
1131		spec->sector_start = ti->begin;
1132		spec->length = ti->len;
1133		strncpy(spec->target_type, ti->type->name,
1134			sizeof(spec->target_type));
1135
1136		outptr += sizeof(struct dm_target_spec);
1137		remaining = len - (outptr - outbuf);
1138		if (remaining <= 0) {
1139			param->flags |= DM_BUFFER_FULL_FLAG;
1140			break;
1141		}
1142
1143		/* Get the status/table string from the target driver */
1144		if (ti->type->status) {
1145			if (param->flags & DM_NOFLUSH_FLAG)
1146				status_flags |= DM_STATUS_NOFLUSH_FLAG;
1147			ti->type->status(ti, type, status_flags, outptr, remaining);
1148		} else
1149			outptr[0] = '\0';
1150
1151		l = strlen(outptr) + 1;
1152		if (l == remaining) {
1153			param->flags |= DM_BUFFER_FULL_FLAG;
1154			break;
1155		}
1156
1157		outptr += l;
1158		used = param->data_start + (outptr - outbuf);
1159
1160		outptr = align_ptr(outptr);
1161		spec->next = outptr - outbuf;
1162	}
1163
1164	if (used)
1165		param->data_size = used;
1166
1167	param->target_count = num_targets;
1168}
1169
1170/*
1171 * Wait for a device to report an event
1172 */
1173static int dev_wait(struct dm_ioctl *param, size_t param_size)
1174{
1175	int r = 0;
1176	struct mapped_device *md;
1177	struct dm_table *table;
1178	int srcu_idx;
1179
1180	md = find_device(param);
1181	if (!md)
1182		return -ENXIO;
1183
1184	/*
1185	 * Wait for a notification event
1186	 */
1187	if (dm_wait_event(md, param->event_nr)) {
1188		r = -ERESTARTSYS;
1189		goto out;
1190	}
1191
1192	/*
1193	 * The userland program is going to want to know what
1194	 * changed to trigger the event, so we may as well tell
1195	 * him and save an ioctl.
1196	 */
1197	__dev_status(md, param);
1198
1199	table = dm_get_live_or_inactive_table(md, param, &srcu_idx);
1200	if (table)
1201		retrieve_status(table, param, param_size);
1202	dm_put_live_table(md, srcu_idx);
1203
1204out:
1205	dm_put(md);
1206
1207	return r;
1208}
1209
1210static inline fmode_t get_mode(struct dm_ioctl *param)
1211{
1212	fmode_t mode = FMODE_READ | FMODE_WRITE;
1213
1214	if (param->flags & DM_READONLY_FLAG)
1215		mode = FMODE_READ;
1216
1217	return mode;
1218}
1219
1220static int next_target(struct dm_target_spec *last, uint32_t next, void *end,
1221		       struct dm_target_spec **spec, char **target_params)
1222{
1223	*spec = (struct dm_target_spec *) ((unsigned char *) last + next);
1224	*target_params = (char *) (*spec + 1);
1225
1226	if (*spec < (last + 1))
1227		return -EINVAL;
1228
1229	return invalid_str(*target_params, end);
1230}
1231
1232static int populate_table(struct dm_table *table,
1233			  struct dm_ioctl *param, size_t param_size)
1234{
1235	int r;
1236	unsigned int i = 0;
1237	struct dm_target_spec *spec = (struct dm_target_spec *) param;
1238	uint32_t next = param->data_start;
1239	void *end = (void *) param + param_size;
1240	char *target_params;
1241
1242	if (!param->target_count) {
1243		DMWARN("populate_table: no targets specified");
1244		return -EINVAL;
1245	}
1246
1247	for (i = 0; i < param->target_count; i++) {
1248
1249		r = next_target(spec, next, end, &spec, &target_params);
1250		if (r) {
1251			DMWARN("unable to find target");
1252			return r;
1253		}
1254
1255		r = dm_table_add_target(table, spec->target_type,
1256					(sector_t) spec->sector_start,
1257					(sector_t) spec->length,
1258					target_params);
1259		if (r) {
1260			DMWARN("error adding target to table");
1261			return r;
1262		}
1263
1264		next = spec->next;
1265	}
1266
1267	return dm_table_complete(table);
1268}
1269
1270static int table_load(struct dm_ioctl *param, size_t param_size)
1271{
1272	int r;
1273	struct hash_cell *hc;
1274	struct dm_table *t, *old_map = NULL;
1275	struct mapped_device *md;
1276	struct target_type *immutable_target_type;
1277
1278	md = find_device(param);
1279	if (!md)
1280		return -ENXIO;
1281
1282	r = dm_table_create(&t, get_mode(param), param->target_count, md);
1283	if (r)
1284		goto err;
1285
1286	/* Protect md->type and md->queue against concurrent table loads. */
1287	dm_lock_md_type(md);
1288	r = populate_table(t, param, param_size);
1289	if (r)
1290		goto err_unlock_md_type;
1291
1292	immutable_target_type = dm_get_immutable_target_type(md);
1293	if (immutable_target_type &&
1294	    (immutable_target_type != dm_table_get_immutable_target_type(t))) {
1295		DMWARN("can't replace immutable target type %s",
1296		       immutable_target_type->name);
1297		r = -EINVAL;
1298		goto err_unlock_md_type;
1299	}
1300
1301	if (dm_get_md_type(md) == DM_TYPE_NONE) {
1302		/* Initial table load: acquire type of table. */
1303		dm_set_md_type(md, dm_table_get_type(t));
1304
1305		/* setup md->queue to reflect md's type (may block) */
1306		r = dm_setup_md_queue(md);
1307		if (r) {
1308			DMWARN("unable to set up device queue for new table.");
1309			goto err_unlock_md_type;
1310		}
1311	} else if (dm_get_md_type(md) != dm_table_get_type(t)) {
1312		DMWARN("can't change device type after initial table load.");
1313		r = -EINVAL;
1314		goto err_unlock_md_type;
1315	}
1316
1317	dm_unlock_md_type(md);
1318
1319	/* stage inactive table */
1320	down_write(&_hash_lock);
1321	hc = dm_get_mdptr(md);
1322	if (!hc || hc->md != md) {
1323		DMWARN("device has been removed from the dev hash table.");
1324		up_write(&_hash_lock);
1325		r = -ENXIO;
1326		goto err_destroy_table;
1327	}
1328
1329	if (hc->new_map)
1330		old_map = hc->new_map;
1331	hc->new_map = t;
1332	up_write(&_hash_lock);
1333
1334	param->flags |= DM_INACTIVE_PRESENT_FLAG;
1335	__dev_status(md, param);
1336
1337	if (old_map) {
1338		dm_sync_table(md);
1339		dm_table_destroy(old_map);
1340	}
1341
1342	dm_put(md);
1343
1344	return 0;
1345
1346err_unlock_md_type:
1347	dm_unlock_md_type(md);
1348err_destroy_table:
1349	dm_table_destroy(t);
1350err:
1351	dm_put(md);
1352
1353	return r;
1354}
1355
1356static int table_clear(struct dm_ioctl *param, size_t param_size)
1357{
1358	struct hash_cell *hc;
1359	struct mapped_device *md;
1360	struct dm_table *old_map = NULL;
1361
1362	down_write(&_hash_lock);
1363
1364	hc = __find_device_hash_cell(param);
1365	if (!hc) {
1366		DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
1367		up_write(&_hash_lock);
1368		return -ENXIO;
1369	}
1370
1371	if (hc->new_map) {
1372		old_map = hc->new_map;
1373		hc->new_map = NULL;
1374	}
1375
1376	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1377
1378	__dev_status(hc->md, param);
1379	md = hc->md;
1380	up_write(&_hash_lock);
1381	if (old_map) {
1382		dm_sync_table(md);
1383		dm_table_destroy(old_map);
1384	}
1385	dm_put(md);
1386
1387	return 0;
1388}
1389
1390/*
1391 * Retrieves a list of devices used by a particular dm device.
1392 */
1393static void retrieve_deps(struct dm_table *table,
1394			  struct dm_ioctl *param, size_t param_size)
1395{
1396	unsigned int count = 0;
1397	struct list_head *tmp;
1398	size_t len, needed;
1399	struct dm_dev_internal *dd;
1400	struct dm_target_deps *deps;
1401
1402	deps = get_result_buffer(param, param_size, &len);
1403
1404	/*
1405	 * Count the devices.
1406	 */
1407	list_for_each (tmp, dm_table_get_devices(table))
1408		count++;
1409
1410	/*
1411	 * Check we have enough space.
1412	 */
1413	needed = sizeof(*deps) + (sizeof(*deps->dev) * count);
1414	if (len < needed) {
1415		param->flags |= DM_BUFFER_FULL_FLAG;
1416		return;
1417	}
1418
1419	/*
1420	 * Fill in the devices.
1421	 */
1422	deps->count = count;
1423	count = 0;
1424	list_for_each_entry (dd, dm_table_get_devices(table), list)
1425		deps->dev[count++] = huge_encode_dev(dd->dm_dev->bdev->bd_dev);
1426
1427	param->data_size = param->data_start + needed;
1428}
1429
1430static int table_deps(struct dm_ioctl *param, size_t param_size)
1431{
1432	struct mapped_device *md;
1433	struct dm_table *table;
1434	int srcu_idx;
1435
1436	md = find_device(param);
1437	if (!md)
1438		return -ENXIO;
1439
1440	__dev_status(md, param);
1441
1442	table = dm_get_live_or_inactive_table(md, param, &srcu_idx);
1443	if (table)
1444		retrieve_deps(table, param, param_size);
1445	dm_put_live_table(md, srcu_idx);
1446
1447	dm_put(md);
1448
1449	return 0;
1450}
1451
1452/*
1453 * Return the status of a device as a text string for each
1454 * target.
1455 */
1456static int table_status(struct dm_ioctl *param, size_t param_size)
1457{
1458	struct mapped_device *md;
1459	struct dm_table *table;
1460	int srcu_idx;
1461
1462	md = find_device(param);
1463	if (!md)
1464		return -ENXIO;
1465
1466	__dev_status(md, param);
1467
1468	table = dm_get_live_or_inactive_table(md, param, &srcu_idx);
1469	if (table)
1470		retrieve_status(table, param, param_size);
1471	dm_put_live_table(md, srcu_idx);
1472
1473	dm_put(md);
1474
1475	return 0;
1476}
1477
1478/*
1479 * Process device-mapper dependent messages.  Messages prefixed with '@'
1480 * are processed by the DM core.  All others are delivered to the target.
1481 * Returns a number <= 1 if message was processed by device mapper.
1482 * Returns 2 if message should be delivered to the target.
1483 */
1484static int message_for_md(struct mapped_device *md, unsigned argc, char **argv,
1485			  char *result, unsigned maxlen)
1486{
1487	int r;
1488
1489	if (**argv != '@')
1490		return 2; /* no '@' prefix, deliver to target */
1491
1492	if (!strcasecmp(argv[0], "@cancel_deferred_remove")) {
1493		if (argc != 1) {
1494			DMERR("Invalid arguments for @cancel_deferred_remove");
1495			return -EINVAL;
1496		}
1497		return dm_cancel_deferred_remove(md);
1498	}
1499
1500	r = dm_stats_message(md, argc, argv, result, maxlen);
1501	if (r < 2)
1502		return r;
1503
1504	DMERR("Unsupported message sent to DM core: %s", argv[0]);
1505	return -EINVAL;
1506}
1507
1508/*
1509 * Pass a message to the target that's at the supplied device offset.
1510 */
1511static int target_message(struct dm_ioctl *param, size_t param_size)
1512{
1513	int r, argc;
1514	char **argv;
1515	struct mapped_device *md;
1516	struct dm_table *table;
1517	struct dm_target *ti;
1518	struct dm_target_msg *tmsg = (void *) param + param->data_start;
1519	size_t maxlen;
1520	char *result = get_result_buffer(param, param_size, &maxlen);
1521	int srcu_idx;
1522
1523	md = find_device(param);
1524	if (!md)
1525		return -ENXIO;
1526
1527	if (tmsg < (struct dm_target_msg *) param->data ||
1528	    invalid_str(tmsg->message, (void *) param + param_size)) {
1529		DMWARN("Invalid target message parameters.");
1530		r = -EINVAL;
1531		goto out;
1532	}
1533
1534	r = dm_split_args(&argc, &argv, tmsg->message);
1535	if (r) {
1536		DMWARN("Failed to split target message parameters");
1537		goto out;
1538	}
1539
1540	if (!argc) {
1541		DMWARN("Empty message received.");
1542		goto out_argv;
1543	}
1544
1545	r = message_for_md(md, argc, argv, result, maxlen);
1546	if (r <= 1)
1547		goto out_argv;
1548
1549	table = dm_get_live_table(md, &srcu_idx);
1550	if (!table)
1551		goto out_table;
1552
1553	if (dm_deleting_md(md)) {
1554		r = -ENXIO;
1555		goto out_table;
1556	}
1557
1558	ti = dm_table_find_target(table, tmsg->sector);
1559	if (!dm_target_is_valid(ti)) {
1560		DMWARN("Target message sector outside device.");
1561		r = -EINVAL;
1562	} else if (ti->type->message)
1563		r = ti->type->message(ti, argc, argv);
1564	else {
1565		DMWARN("Target type does not support messages");
1566		r = -EINVAL;
1567	}
1568
1569 out_table:
1570	dm_put_live_table(md, srcu_idx);
1571 out_argv:
1572	kfree(argv);
1573 out:
1574	if (r >= 0)
1575		__dev_status(md, param);
1576
1577	if (r == 1) {
1578		param->flags |= DM_DATA_OUT_FLAG;
1579		if (dm_message_test_buffer_overflow(result, maxlen))
1580			param->flags |= DM_BUFFER_FULL_FLAG;
1581		else
1582			param->data_size = param->data_start + strlen(result) + 1;
1583		r = 0;
1584	}
1585
1586	dm_put(md);
1587	return r;
1588}
1589
1590/*
1591 * The ioctl parameter block consists of two parts, a dm_ioctl struct
1592 * followed by a data buffer.  This flag is set if the second part,
1593 * which has a variable size, is not used by the function processing
1594 * the ioctl.
1595 */
1596#define IOCTL_FLAGS_NO_PARAMS	1
1597
1598/*-----------------------------------------------------------------
1599 * Implementation of open/close/ioctl on the special char
1600 * device.
1601 *---------------------------------------------------------------*/
1602static ioctl_fn lookup_ioctl(unsigned int cmd, int *ioctl_flags)
1603{
1604	static struct {
1605		int cmd;
1606		int flags;
1607		ioctl_fn fn;
1608	} _ioctls[] = {
1609		{DM_VERSION_CMD, 0, NULL}, /* version is dealt with elsewhere */
1610		{DM_REMOVE_ALL_CMD, IOCTL_FLAGS_NO_PARAMS, remove_all},
1611		{DM_LIST_DEVICES_CMD, 0, list_devices},
1612
1613		{DM_DEV_CREATE_CMD, IOCTL_FLAGS_NO_PARAMS, dev_create},
1614		{DM_DEV_REMOVE_CMD, IOCTL_FLAGS_NO_PARAMS, dev_remove},
1615		{DM_DEV_RENAME_CMD, 0, dev_rename},
1616		{DM_DEV_SUSPEND_CMD, IOCTL_FLAGS_NO_PARAMS, dev_suspend},
1617		{DM_DEV_STATUS_CMD, IOCTL_FLAGS_NO_PARAMS, dev_status},
1618		{DM_DEV_WAIT_CMD, 0, dev_wait},
1619
1620		{DM_TABLE_LOAD_CMD, 0, table_load},
1621		{DM_TABLE_CLEAR_CMD, IOCTL_FLAGS_NO_PARAMS, table_clear},
1622		{DM_TABLE_DEPS_CMD, 0, table_deps},
1623		{DM_TABLE_STATUS_CMD, 0, table_status},
1624
1625		{DM_LIST_VERSIONS_CMD, 0, list_versions},
1626
1627		{DM_TARGET_MSG_CMD, 0, target_message},
1628		{DM_DEV_SET_GEOMETRY_CMD, 0, dev_set_geometry}
1629	};
1630
1631	if (unlikely(cmd >= ARRAY_SIZE(_ioctls)))
1632		return NULL;
1633
1634	*ioctl_flags = _ioctls[cmd].flags;
1635	return _ioctls[cmd].fn;
1636}
1637
1638/*
1639 * As well as checking the version compatibility this always
1640 * copies the kernel interface version out.
1641 */
1642static int check_version(unsigned int cmd, struct dm_ioctl __user *user)
1643{
1644	uint32_t version[3];
1645	int r = 0;
1646
1647	if (copy_from_user(version, user->version, sizeof(version)))
1648		return -EFAULT;
1649
1650	if ((DM_VERSION_MAJOR != version[0]) ||
1651	    (DM_VERSION_MINOR < version[1])) {
1652		DMWARN("ioctl interface mismatch: "
1653		       "kernel(%u.%u.%u), user(%u.%u.%u), cmd(%d)",
1654		       DM_VERSION_MAJOR, DM_VERSION_MINOR,
1655		       DM_VERSION_PATCHLEVEL,
1656		       version[0], version[1], version[2], cmd);
1657		r = -EINVAL;
1658	}
1659
1660	/*
1661	 * Fill in the kernel version.
1662	 */
1663	version[0] = DM_VERSION_MAJOR;
1664	version[1] = DM_VERSION_MINOR;
1665	version[2] = DM_VERSION_PATCHLEVEL;
1666	if (copy_to_user(user->version, version, sizeof(version)))
1667		return -EFAULT;
1668
1669	return r;
1670}
1671
1672#define DM_PARAMS_KMALLOC	0x0001	/* Params alloced with kmalloc */
1673#define DM_PARAMS_VMALLOC	0x0002	/* Params alloced with vmalloc */
1674#define DM_WIPE_BUFFER		0x0010	/* Wipe input buffer before returning from ioctl */
1675
1676static void free_params(struct dm_ioctl *param, size_t param_size, int param_flags)
1677{
1678	if (param_flags & DM_WIPE_BUFFER)
1679		memset(param, 0, param_size);
1680
1681	if (param_flags & DM_PARAMS_KMALLOC)
1682		kfree(param);
1683	if (param_flags & DM_PARAMS_VMALLOC)
1684		vfree(param);
1685}
1686
1687static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl *param_kernel,
1688		       int ioctl_flags,
1689		       struct dm_ioctl **param, int *param_flags)
1690{
1691	struct dm_ioctl *dmi;
1692	int secure_data;
1693	const size_t minimum_data_size = sizeof(*param_kernel) - sizeof(param_kernel->data);
1694
1695	if (copy_from_user(param_kernel, user, minimum_data_size))
1696		return -EFAULT;
1697
1698	if (param_kernel->data_size < minimum_data_size)
1699		return -EINVAL;
1700
1701	secure_data = param_kernel->flags & DM_SECURE_DATA_FLAG;
1702
1703	*param_flags = secure_data ? DM_WIPE_BUFFER : 0;
1704
1705	if (ioctl_flags & IOCTL_FLAGS_NO_PARAMS) {
1706		dmi = param_kernel;
1707		dmi->data_size = minimum_data_size;
1708		goto data_copied;
1709	}
1710
1711	/*
1712	 * Try to avoid low memory issues when a device is suspended.
1713	 * Use kmalloc() rather than vmalloc() when we can.
1714	 */
1715	dmi = NULL;
1716	if (param_kernel->data_size <= KMALLOC_MAX_SIZE) {
1717		dmi = kmalloc(param_kernel->data_size, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
1718		if (dmi)
1719			*param_flags |= DM_PARAMS_KMALLOC;
1720	}
1721
1722	if (!dmi) {
1723		unsigned noio_flag;
1724		noio_flag = memalloc_noio_save();
1725		dmi = __vmalloc(param_kernel->data_size, GFP_NOIO | __GFP_REPEAT | __GFP_HIGH | __GFP_HIGHMEM, PAGE_KERNEL);
1726		memalloc_noio_restore(noio_flag);
1727		if (dmi)
1728			*param_flags |= DM_PARAMS_VMALLOC;
1729	}
1730
1731	if (!dmi) {
1732		if (secure_data && clear_user(user, param_kernel->data_size))
1733			return -EFAULT;
1734		return -ENOMEM;
1735	}
1736
1737	if (copy_from_user(dmi, user, param_kernel->data_size))
1738		goto bad;
1739
1740data_copied:
1741	/*
1742	 * Abort if something changed the ioctl data while it was being copied.
1743	 */
1744	if (dmi->data_size != param_kernel->data_size) {
1745		DMERR("rejecting ioctl: data size modified while processing parameters");
1746		goto bad;
1747	}
1748
1749	/* Wipe the user buffer so we do not return it to userspace */
1750	if (secure_data && clear_user(user, param_kernel->data_size))
1751		goto bad;
1752
1753	*param = dmi;
1754	return 0;
1755
1756bad:
1757	free_params(dmi, param_kernel->data_size, *param_flags);
1758
1759	return -EFAULT;
1760}
1761
1762static int validate_params(uint cmd, struct dm_ioctl *param)
1763{
1764	/* Always clear this flag */
1765	param->flags &= ~DM_BUFFER_FULL_FLAG;
1766	param->flags &= ~DM_UEVENT_GENERATED_FLAG;
1767	param->flags &= ~DM_SECURE_DATA_FLAG;
1768	param->flags &= ~DM_DATA_OUT_FLAG;
1769
1770	/* Ignores parameters */
1771	if (cmd == DM_REMOVE_ALL_CMD ||
1772	    cmd == DM_LIST_DEVICES_CMD ||
1773	    cmd == DM_LIST_VERSIONS_CMD)
1774		return 0;
1775
1776	if ((cmd == DM_DEV_CREATE_CMD)) {
1777		if (!*param->name) {
1778			DMWARN("name not supplied when creating device");
1779			return -EINVAL;
1780		}
1781	} else if ((*param->uuid && *param->name)) {
1782		DMWARN("only supply one of name or uuid, cmd(%u)", cmd);
1783		return -EINVAL;
1784	}
1785
1786	/* Ensure strings are terminated */
1787	param->name[DM_NAME_LEN - 1] = '\0';
1788	param->uuid[DM_UUID_LEN - 1] = '\0';
1789
1790	return 0;
1791}
1792
1793static int ctl_ioctl(uint command, struct dm_ioctl __user *user)
1794{
1795	int r = 0;
1796	int ioctl_flags;
1797	int param_flags;
1798	unsigned int cmd;
1799	struct dm_ioctl *uninitialized_var(param);
1800	ioctl_fn fn = NULL;
1801	size_t input_param_size;
1802	struct dm_ioctl param_kernel;
1803
1804	/* only root can play with this */
1805	if (!capable(CAP_SYS_ADMIN))
1806		return -EACCES;
1807
1808	if (_IOC_TYPE(command) != DM_IOCTL)
1809		return -ENOTTY;
1810
1811	cmd = _IOC_NR(command);
1812
1813	/*
1814	 * Check the interface version passed in.  This also
1815	 * writes out the kernel's interface version.
1816	 */
1817	r = check_version(cmd, user);
1818	if (r)
1819		return r;
1820
1821	/*
1822	 * Nothing more to do for the version command.
1823	 */
1824	if (cmd == DM_VERSION_CMD)
1825		return 0;
1826
1827	fn = lookup_ioctl(cmd, &ioctl_flags);
1828	if (!fn) {
1829		DMWARN("dm_ctl_ioctl: unknown command 0x%x", command);
1830		return -ENOTTY;
1831	}
1832
1833	/*
1834	 * Copy the parameters into kernel space.
1835	 */
1836	r = copy_params(user, &param_kernel, ioctl_flags, &param, &param_flags);
1837
1838	if (r)
1839		return r;
1840
1841	input_param_size = param->data_size;
1842	r = validate_params(cmd, param);
1843	if (r)
1844		goto out;
1845
1846	param->data_size = sizeof(*param);
1847	r = fn(param, input_param_size);
1848
1849	if (unlikely(param->flags & DM_BUFFER_FULL_FLAG) &&
1850	    unlikely(ioctl_flags & IOCTL_FLAGS_NO_PARAMS))
1851		DMERR("ioctl %d tried to output some data but has IOCTL_FLAGS_NO_PARAMS set", cmd);
1852
1853	/*
1854	 * Copy the results back to userland.
1855	 */
1856	if (!r && copy_to_user(user, param, param->data_size))
1857		r = -EFAULT;
1858
1859out:
1860	free_params(param, input_param_size, param_flags);
1861	return r;
1862}
1863
1864static long dm_ctl_ioctl(struct file *file, uint command, ulong u)
1865{
1866	return (long)ctl_ioctl(command, (struct dm_ioctl __user *)u);
1867}
1868
1869#ifdef CONFIG_COMPAT
1870static long dm_compat_ctl_ioctl(struct file *file, uint command, ulong u)
1871{
1872	return (long)dm_ctl_ioctl(file, command, (ulong) compat_ptr(u));
1873}
1874#else
1875#define dm_compat_ctl_ioctl NULL
1876#endif
1877
1878static const struct file_operations _ctl_fops = {
1879	.open = nonseekable_open,
1880	.unlocked_ioctl	 = dm_ctl_ioctl,
1881	.compat_ioctl = dm_compat_ctl_ioctl,
1882	.owner	 = THIS_MODULE,
1883	.llseek  = noop_llseek,
1884};
1885
1886static struct miscdevice _dm_misc = {
1887	.minor		= MAPPER_CTRL_MINOR,
1888	.name  		= DM_NAME,
1889	.nodename	= DM_DIR "/" DM_CONTROL_NODE,
1890	.fops  		= &_ctl_fops
1891};
1892
1893MODULE_ALIAS_MISCDEV(MAPPER_CTRL_MINOR);
1894MODULE_ALIAS("devname:" DM_DIR "/" DM_CONTROL_NODE);
1895
1896/*
1897 * Create misc character device and link to DM_DIR/control.
1898 */
1899int __init dm_interface_init(void)
1900{
1901	int r;
1902
1903	r = dm_hash_init();
1904	if (r)
1905		return r;
1906
1907	r = misc_register(&_dm_misc);
1908	if (r) {
1909		DMERR("misc_register failed for control device");
1910		dm_hash_exit();
1911		return r;
1912	}
1913
1914	DMINFO("%d.%d.%d%s initialised: %s", DM_VERSION_MAJOR,
1915	       DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL, DM_VERSION_EXTRA,
1916	       DM_DRIVER_EMAIL);
1917	return 0;
1918}
1919
1920void dm_interface_exit(void)
1921{
1922	if (misc_deregister(&_dm_misc) < 0)
1923		DMERR("misc_deregister failed for control device");
1924
1925	dm_hash_exit();
1926}
1927
1928/**
1929 * dm_copy_name_and_uuid - Copy mapped device name & uuid into supplied buffers
1930 * @md: Pointer to mapped_device
1931 * @name: Buffer (size DM_NAME_LEN) for name
1932 * @uuid: Buffer (size DM_UUID_LEN) for uuid or empty string if uuid not defined
1933 */
1934int dm_copy_name_and_uuid(struct mapped_device *md, char *name, char *uuid)
1935{
1936	int r = 0;
1937	struct hash_cell *hc;
1938
1939	if (!md)
1940		return -ENXIO;
1941
1942	mutex_lock(&dm_hash_cells_mutex);
1943	hc = dm_get_mdptr(md);
1944	if (!hc || hc->md != md) {
1945		r = -ENXIO;
1946		goto out;
1947	}
1948
1949	if (name)
1950		strcpy(name, hc->name);
1951	if (uuid)
1952		strcpy(uuid, hc->uuid ? : "");
1953
1954out:
1955	mutex_unlock(&dm_hash_cells_mutex);
1956
1957	return r;
1958}
1959