1	// SPDX-License-Identifier: GPL-2.0-only
2	/*
3	* Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
4	* Copyright (C) 2004 - 2006 Red Hat, Inc. All rights reserved.
5	*
6	* This file is released under the GPL.
7	*/
8
9	#include "dm-core.h"
10	#include "dm-ima.h"
11	#include <linux/module.h>
12	#include <linux/vmalloc.h>
13	#include <linux/miscdevice.h>
14	#include <linux/sched/mm.h>
15	#include <linux/init.h>
16	#include <linux/wait.h>
17	#include <linux/slab.h>
18	#include <linux/rbtree.h>
19	#include <linux/dm-ioctl.h>
20	#include <linux/hdreg.h>
21	#include <linux/compat.h>
22	#include <linux/nospec.h>
23
24	#include <linux/uaccess.h>
25	#include <linux/ima.h>
26
27	#define DM_MSG_PREFIX "ioctl"
28	#define DM_DRIVER_EMAIL "dm-devel@redhat.com"
29
30	struct dm_file {
31	/*
32	* poll will wait until the global event number is greater than
33	* this value.
34	*/
35	volatile unsigned int global_event_nr;
36	};
37
38	/*
39	*---------------------------------------------------------------
40	* The ioctl interface needs to be able to look up devices by
41	* name or uuid.
42	*---------------------------------------------------------------
43	*/
44	struct hash_cell {
45	struct rb_node name_node;
46	struct rb_node uuid_node;
47	bool name_set;
48	bool uuid_set;
49
50	char *name;
51	char *uuid;
52	struct mapped_device *md;
53	struct dm_table *new_map;
54	};
55
56	struct vers_iter {
57	size_t param_size;
58	struct dm_target_versions *vers, *old_vers;
59	char *end;
60	uint32_t flags;
61	};
62
63
64	static struct rb_root name_rb_tree = RB_ROOT;
65	static struct rb_root uuid_rb_tree = RB_ROOT;
66
67	static void dm_hash_remove_all(bool keep_open_devices, bool mark_deferred, bool only_deferred);
68
69	/*
70	* Guards access to both hash tables.
71	*/
72	static DECLARE_RWSEM(_hash_lock);
73
74	/*
75	* Protects use of mdptr to obtain hash cell name and uuid from mapped device.
76	*/
77	static DEFINE_MUTEX(dm_hash_cells_mutex);
78
79	static void dm_hash_exit(void)
80	{
81	dm_hash_remove_all(keep_open_devices: false, mark_deferred: false, only_deferred: false);
82	}
83
84	/*
85	*---------------------------------------------------------------
86	* Code for looking up a device by name
87	*---------------------------------------------------------------
88	*/
89	static struct hash_cell *__get_name_cell(const char *str)
90	{
91	struct rb_node *n = name_rb_tree.rb_node;
92
93	while (n) {
94	struct hash_cell *hc = container_of(n, struct hash_cell, name_node);
95	int c;
96
97	c = strcmp(hc->name, str);
98	if (!c) {
99	dm_get(md: hc->md);
100	return hc;
101	}
102	n = c >= 0 ? n->rb_left : n->rb_right;
103	}
104
105	return NULL;
106	}
107
108	static struct hash_cell *__get_uuid_cell(const char *str)
109	{
110	struct rb_node *n = uuid_rb_tree.rb_node;
111
112	while (n) {
113	struct hash_cell *hc = container_of(n, struct hash_cell, uuid_node);
114	int c;
115
116	c = strcmp(hc->uuid, str);
117	if (!c) {
118	dm_get(md: hc->md);
119	return hc;
120	}
121	n = c >= 0 ? n->rb_left : n->rb_right;
122	}
123
124	return NULL;
125	}
126
127	static void __unlink_name(struct hash_cell *hc)
128	{
129	if (hc->name_set) {
130	hc->name_set = false;
131	rb_erase(&hc->name_node, &name_rb_tree);
132	}
133	}
134
135	static void __unlink_uuid(struct hash_cell *hc)
136	{
137	if (hc->uuid_set) {
138	hc->uuid_set = false;
139	rb_erase(&hc->uuid_node, &uuid_rb_tree);
140	}
141	}
142
143	static void __link_name(struct hash_cell *new_hc)
144	{
145	struct rb_node **n, *parent;
146
147	__unlink_name(hc: new_hc);
148
149	new_hc->name_set = true;
150
151	n = &name_rb_tree.rb_node;
152	parent = NULL;
153
154	while (*n) {
155	struct hash_cell *hc = container_of(*n, struct hash_cell, name_node);
156	int c;
157
158	c = strcmp(hc->name, new_hc->name);
159	BUG_ON(!c);
160	parent = *n;
161	n = c >= 0 ? &hc->name_node.rb_left : &hc->name_node.rb_right;
162	}
163
164	rb_link_node(node: &new_hc->name_node, parent, rb_link: n);
165	rb_insert_color(&new_hc->name_node, &name_rb_tree);
166	}
167
168	static void __link_uuid(struct hash_cell *new_hc)
169	{
170	struct rb_node **n, *parent;
171
172	__unlink_uuid(hc: new_hc);
173
174	new_hc->uuid_set = true;
175
176	n = &uuid_rb_tree.rb_node;
177	parent = NULL;
178
179	while (*n) {
180	struct hash_cell *hc = container_of(*n, struct hash_cell, uuid_node);
181	int c;
182
183	c = strcmp(hc->uuid, new_hc->uuid);
184	BUG_ON(!c);
185	parent = *n;
186	n = c > 0 ? &hc->uuid_node.rb_left : &hc->uuid_node.rb_right;
187	}
188
189	rb_link_node(node: &new_hc->uuid_node, parent, rb_link: n);
190	rb_insert_color(&new_hc->uuid_node, &uuid_rb_tree);
191	}
192
193	static struct hash_cell *__get_dev_cell(uint64_t dev)
194	{
195	struct mapped_device *md;
196	struct hash_cell *hc;
197
198	md = dm_get_md(dev: huge_decode_dev(dev));
199	if (!md)
200	return NULL;
201
202	hc = dm_get_mdptr(md);
203	if (!hc) {
204	dm_put(md);
205	return NULL;
206	}
207
208	return hc;
209	}
210
211	/*
212	*---------------------------------------------------------------
213	* Inserting, removing and renaming a device.
214	*---------------------------------------------------------------
215	*/
216	static struct hash_cell *alloc_cell(const char *name, const char *uuid,
217	struct mapped_device *md)
218	{
219	struct hash_cell *hc;
220
221	hc = kmalloc(size: sizeof(*hc), GFP_KERNEL);
222	if (!hc)
223	return NULL;
224
225	hc->name = kstrdup(s: name, GFP_KERNEL);
226	if (!hc->name) {
227	kfree(objp: hc);
228	return NULL;
229	}
230
231	if (!uuid)
232	hc->uuid = NULL;
233
234	else {
235	hc->uuid = kstrdup(s: uuid, GFP_KERNEL);
236	if (!hc->uuid) {
237	kfree(objp: hc->name);
238	kfree(objp: hc);
239	return NULL;
240	}
241	}
242
243	hc->name_set = hc->uuid_set = false;
244	hc->md = md;
245	hc->new_map = NULL;
246	return hc;
247	}
248
249	static void free_cell(struct hash_cell *hc)
250	{
251	if (hc) {
252	kfree(objp: hc->name);
253	kfree(objp: hc->uuid);
254	kfree(objp: hc);
255	}
256	}
257
258	/*
259	* The kdev_t and uuid of a device can never change once it is
260	* initially inserted.
261	*/
262	static int dm_hash_insert(const char *name, const char *uuid, struct mapped_device *md)
263	{
264	struct hash_cell *cell, *hc;
265
266	/*
267	* Allocate the new cells.
268	*/
269	cell = alloc_cell(name, uuid, md);
270	if (!cell)
271	return -ENOMEM;
272
273	/*
274	* Insert the cell into both hash tables.
275	*/
276	down_write(sem: &_hash_lock);
277	hc = __get_name_cell(str: name);
278	if (hc) {
279	dm_put(md: hc->md);
280	goto bad;
281	}
282
283	__link_name(new_hc: cell);
284
285	if (uuid) {
286	hc = __get_uuid_cell(str: uuid);
287	if (hc) {
288	__unlink_name(hc: cell);
289	dm_put(md: hc->md);
290	goto bad;
291	}
292	__link_uuid(new_hc: cell);
293	}
294	dm_get(md);
295	mutex_lock(&dm_hash_cells_mutex);
296	dm_set_mdptr(md, ptr: cell);
297	mutex_unlock(lock: &dm_hash_cells_mutex);
298	up_write(sem: &_hash_lock);
299
300	return 0;
301
302	bad:
303	up_write(sem: &_hash_lock);
304	free_cell(hc: cell);
305	return -EBUSY;
306	}
307
308	static struct dm_table *__hash_remove(struct hash_cell *hc)
309	{
310	struct dm_table *table;
311	int srcu_idx;
312
313	lockdep_assert_held(&_hash_lock);
314
315	/* remove from the dev trees */
316	__unlink_name(hc);
317	__unlink_uuid(hc);
318	mutex_lock(&dm_hash_cells_mutex);
319	dm_set_mdptr(md: hc->md, NULL);
320	mutex_unlock(lock: &dm_hash_cells_mutex);
321
322	table = dm_get_live_table(md: hc->md, srcu_idx: &srcu_idx);
323	if (table)
324	dm_table_event(t: table);
325	dm_put_live_table(md: hc->md, srcu_idx);
326
327	table = NULL;
328	if (hc->new_map)
329	table = hc->new_map;
330	dm_put(md: hc->md);
331	free_cell(hc);
332
333	return table;
334	}
335
336	static void dm_hash_remove_all(bool keep_open_devices, bool mark_deferred, bool only_deferred)
337	{
338	int dev_skipped;
339	struct rb_node *n;
340	struct hash_cell *hc;
341	struct mapped_device *md;
342	struct dm_table *t;
343
344	retry:
345	dev_skipped = 0;
346
347	down_write(sem: &_hash_lock);
348
349	for (n = rb_first(&name_rb_tree); n; n = rb_next(n)) {
350	hc = container_of(n, struct hash_cell, name_node);
351	md = hc->md;
352	dm_get(md);
353
354	if (keep_open_devices &&
355	dm_lock_for_deletion(md, mark_deferred, only_deferred)) {
356	dm_put(md);
357	dev_skipped++;
358	continue;
359	}
360
361	t = __hash_remove(hc);
362
363	up_write(sem: &_hash_lock);
364
365	if (t) {
366	dm_sync_table(md);
367	dm_table_destroy(t);
368	}
369	dm_ima_measure_on_device_remove(md, remove_all: true);
370	dm_put(md);
371	if (likely(keep_open_devices))
372	dm_destroy(md);
373	else
374	dm_destroy_immediate(md);
375
376	/*
377	* Some mapped devices may be using other mapped
378	* devices, so repeat until we make no further
379	* progress. If a new mapped device is created
380	* here it will also get removed.
381	*/
382	goto retry;
383	}
384
385	up_write(sem: &_hash_lock);
386
387	if (dev_skipped)
388	DMWARN("remove_all left %d open device(s)", dev_skipped);
389	}
390
391	/*
392	* Set the uuid of a hash_cell that isn't already set.
393	*/
394	static void __set_cell_uuid(struct hash_cell *hc, char *new_uuid)
395	{
396	mutex_lock(&dm_hash_cells_mutex);
397	hc->uuid = new_uuid;
398	mutex_unlock(lock: &dm_hash_cells_mutex);
399
400	__link_uuid(new_hc: hc);
401	}
402
403	/*
404	* Changes the name of a hash_cell and returns the old name for
405	* the caller to free.
406	*/
407	static char *__change_cell_name(struct hash_cell *hc, char *new_name)
408	{
409	char *old_name;
410
411	/*
412	* Rename and move the name cell.
413	*/
414	__unlink_name(hc);
415	old_name = hc->name;
416
417	mutex_lock(&dm_hash_cells_mutex);
418	hc->name = new_name;
419	mutex_unlock(lock: &dm_hash_cells_mutex);
420
421	__link_name(new_hc: hc);
422
423	return old_name;
424	}
425
426	static struct mapped_device *dm_hash_rename(struct dm_ioctl *param,
427	const char *new)
428	{
429	char *new_data, *old_name = NULL;
430	struct hash_cell *hc;
431	struct dm_table *table;
432	struct mapped_device *md;
433	unsigned int change_uuid = (param->flags & DM_UUID_FLAG) ? 1 : 0;
434	int srcu_idx;
435
436	/*
437	* duplicate new.
438	*/
439	new_data = kstrdup(s: new, GFP_KERNEL);
440	if (!new_data)
441	return ERR_PTR(error: -ENOMEM);
442
443	down_write(sem: &_hash_lock);
444
445	/*
446	* Is new free ?
447	*/
448	if (change_uuid)
449	hc = __get_uuid_cell(str: new);
450	else
451	hc = __get_name_cell(str: new);
452
453	if (hc) {
454	DMERR("Unable to change %s on mapped device %s to one that already exists: %s",
455	change_uuid ? "uuid" : "name",
456	param->name, new);
457	dm_put(md: hc->md);
458	up_write(sem: &_hash_lock);
459	kfree(objp: new_data);
460	return ERR_PTR(error: -EBUSY);
461	}
462
463	/*
464	* Is there such a device as 'old' ?
465	*/
466	hc = __get_name_cell(str: param->name);
467	if (!hc) {
468	DMERR("Unable to rename non-existent device, %s to %s%s",
469	param->name, change_uuid ? "uuid " : "", new);
470	up_write(sem: &_hash_lock);
471	kfree(objp: new_data);
472	return ERR_PTR(error: -ENXIO);
473	}
474
475	/*
476	* Does this device already have a uuid?
477	*/
478	if (change_uuid && hc->uuid) {
479	DMERR("Unable to change uuid of mapped device %s to %s "
480	"because uuid is already set to %s",
481	param->name, new, hc->uuid);
482	dm_put(md: hc->md);
483	up_write(sem: &_hash_lock);
484	kfree(objp: new_data);
485	return ERR_PTR(error: -EINVAL);
486	}
487
488	if (change_uuid)
489	__set_cell_uuid(hc, new_uuid: new_data);
490	else
491	old_name = __change_cell_name(hc, new_name: new_data);
492
493	/*
494	* Wake up any dm event waiters.
495	*/
496	table = dm_get_live_table(md: hc->md, srcu_idx: &srcu_idx);
497	if (table)
498	dm_table_event(t: table);
499	dm_put_live_table(md: hc->md, srcu_idx);
500
501	if (!dm_kobject_uevent(md: hc->md, action: KOBJ_CHANGE, cookie: param->event_nr, need_resize_uevent: false))
502	param->flags |= DM_UEVENT_GENERATED_FLAG;
503
504	md = hc->md;
505
506	dm_ima_measure_on_device_rename(md);
507
508	up_write(sem: &_hash_lock);
509	kfree(objp: old_name);
510
511	return md;
512	}
513
514	void dm_deferred_remove(void)
515	{
516	dm_hash_remove_all(keep_open_devices: true, mark_deferred: false, only_deferred: true);
517	}
518
519	/*
520	*---------------------------------------------------------------
521	* Implementation of the ioctl commands
522	*---------------------------------------------------------------
523	*/
524	/*
525	* All the ioctl commands get dispatched to functions with this
526	* prototype.
527	*/
528	typedef int (*ioctl_fn)(struct file *filp, struct dm_ioctl *param, size_t param_size);
529
530	static int remove_all(struct file *filp, struct dm_ioctl *param, size_t param_size)
531	{
532	dm_hash_remove_all(keep_open_devices: true, mark_deferred: !!(param->flags & DM_DEFERRED_REMOVE), only_deferred: false);
533	param->data_size = 0;
534	return 0;
535	}
536
537	/*
538	* Round up the ptr to an 8-byte boundary.
539	*/
540	#define ALIGN_MASK 7
541	static inline size_t align_val(size_t val)
542	{
543	return (val + ALIGN_MASK) & ~ALIGN_MASK;
544	}
545	static inline void *align_ptr(void *ptr)
546	{
547	return (void *)align_val(val: (size_t)ptr);
548	}
549
550	/*
551	* Retrieves the data payload buffer from an already allocated
552	* struct dm_ioctl.
553	*/
554	static void *get_result_buffer(struct dm_ioctl *param, size_t param_size,
555	size_t *len)
556	{
557	param->data_start = align_ptr(ptr: param + 1) - (void *) param;
558
559	if (param->data_start < param_size)
560	*len = param_size - param->data_start;
561	else
562	*len = 0;
563
564	return ((void *) param) + param->data_start;
565	}
566
567	static bool filter_device(struct hash_cell *hc, const char *pfx_name, const char *pfx_uuid)
568	{
569	const char *val;
570	size_t val_len, pfx_len;
571
572	val = hc->name;
573	val_len = strlen(val);
574	pfx_len = strnlen(p: pfx_name, DM_NAME_LEN);
575	if (pfx_len > val_len)
576	return false;
577	if (memcmp(p: val, q: pfx_name, size: pfx_len))
578	return false;
579
580	val = hc->uuid ? hc->uuid : "";
581	val_len = strlen(val);
582	pfx_len = strnlen(p: pfx_uuid, DM_UUID_LEN);
583	if (pfx_len > val_len)
584	return false;
585	if (memcmp(p: val, q: pfx_uuid, size: pfx_len))
586	return false;
587
588	return true;
589	}
590
591	static int list_devices(struct file *filp, struct dm_ioctl *param, size_t param_size)
592	{
593	struct rb_node *n;
594	struct hash_cell *hc;
595	size_t len, needed = 0;
596	struct gendisk *disk;
597	struct dm_name_list *orig_nl, *nl, *old_nl = NULL;
598	uint32_t *event_nr;
599
600	down_write(sem: &_hash_lock);
601
602	/*
603	* Loop through all the devices working out how much
604	* space we need.
605	*/
606	for (n = rb_first(&name_rb_tree); n; n = rb_next(n)) {
607	hc = container_of(n, struct hash_cell, name_node);
608	if (!filter_device(hc, pfx_name: param->name, pfx_uuid: param->uuid))
609	continue;
610	needed += align_val(offsetof(struct dm_name_list, name) + strlen(hc->name) + 1);
611	needed += align_val(val: sizeof(uint32_t) * 2);
612	if (param->flags & DM_UUID_FLAG && hc->uuid)
613	needed += align_val(strlen(hc->uuid) + 1);
614	}
615
616	/*
617	* Grab our output buffer.
618	*/
619	nl = orig_nl = get_result_buffer(param, param_size, len: &len);
620	if (len < needed || len < sizeof(nl->dev)) {
621	param->flags |= DM_BUFFER_FULL_FLAG;
622	goto out;
623	}
624	param->data_size = param->data_start + needed;
625
626	nl->dev = 0; /* Flags no data */
627
628	/*
629	* Now loop through filling out the names.
630	*/
631	for (n = rb_first(&name_rb_tree); n; n = rb_next(n)) {
632	void *uuid_ptr;
633
634	hc = container_of(n, struct hash_cell, name_node);
635	if (!filter_device(hc, pfx_name: param->name, pfx_uuid: param->uuid))
636	continue;
637	if (old_nl)
638	old_nl->next = (uint32_t) ((void *) nl -
639	(void *) old_nl);
640	disk = dm_disk(md: hc->md);
641	nl->dev = huge_encode_dev(dev: disk_devt(disk));
642	nl->next = 0;
643	strcpy(p: nl->name, q: hc->name);
644
645	old_nl = nl;
646	event_nr = align_ptr(ptr: nl->name + strlen(hc->name) + 1);
647	event_nr[0] = dm_get_event_nr(md: hc->md);
648	event_nr[1] = 0;
649	uuid_ptr = align_ptr(ptr: event_nr + 2);
650	if (param->flags & DM_UUID_FLAG) {
651	if (hc->uuid) {
652	event_nr[1] |= DM_NAME_LIST_FLAG_HAS_UUID;
653	strcpy(p: uuid_ptr, q: hc->uuid);
654	uuid_ptr = align_ptr(ptr: uuid_ptr + strlen(hc->uuid) + 1);
655	} else {
656	event_nr[1] |= DM_NAME_LIST_FLAG_DOESNT_HAVE_UUID;
657	}
658	}
659	nl = uuid_ptr;
660	}
661	/*
662	* If mismatch happens, security may be compromised due to buffer
663	* overflow, so it's better to crash.
664	*/
665	BUG_ON((char *)nl - (char *)orig_nl != needed);
666
667	out:
668	up_write(sem: &_hash_lock);
669	return 0;
670	}
671
672	static void list_version_get_needed(struct target_type *tt, void *needed_param)
673	{
674	size_t *needed = needed_param;
675
676	*needed += sizeof(struct dm_target_versions);
677	*needed += strlen(tt->name) + 1;
678	*needed += ALIGN_MASK;
679	}
680
681	static void list_version_get_info(struct target_type *tt, void *param)
682	{
683	struct vers_iter *info = param;
684
685	/* Check space - it might have changed since the first iteration */
686	if ((char *)info->vers + sizeof(tt->version) + strlen(tt->name) + 1 > info->end) {
687	info->flags = DM_BUFFER_FULL_FLAG;
688	return;
689	}
690
691	if (info->old_vers)
692	info->old_vers->next = (uint32_t) ((void *)info->vers - (void *)info->old_vers);
693
694	info->vers->version[0] = tt->version[0];
695	info->vers->version[1] = tt->version[1];
696	info->vers->version[2] = tt->version[2];
697	info->vers->next = 0;
698	strcpy(p: info->vers->name, q: tt->name);
699
700	info->old_vers = info->vers;
701	info->vers = align_ptr(ptr: (void *)(info->vers + 1) + strlen(tt->name) + 1);
702	}
703
704	static int __list_versions(struct dm_ioctl *param, size_t param_size, const char *name)
705	{
706	size_t len, needed = 0;
707	struct dm_target_versions *vers;
708	struct vers_iter iter_info;
709	struct target_type *tt = NULL;
710
711	if (name) {
712	tt = dm_get_target_type(name);
713	if (!tt)
714	return -EINVAL;
715	}
716
717	/*
718	* Loop through all the devices working out how much
719	* space we need.
720	*/
721	if (!tt)
722	dm_target_iterate(iter_func: list_version_get_needed, param: &needed);
723	else
724	list_version_get_needed(tt, needed_param: &needed);
725
726	/*
727	* Grab our output buffer.
728	*/
729	vers = get_result_buffer(param, param_size, len: &len);
730	if (len < needed) {
731	param->flags |= DM_BUFFER_FULL_FLAG;
732	goto out;
733	}
734	param->data_size = param->data_start + needed;
735
736	iter_info.param_size = param_size;
737	iter_info.old_vers = NULL;
738	iter_info.vers = vers;
739	iter_info.flags = 0;
740	iter_info.end = (char *)vers + needed;
741
742	/*
743	* Now loop through filling out the names & versions.
744	*/
745	if (!tt)
746	dm_target_iterate(iter_func: list_version_get_info, param: &iter_info);
747	else
748	list_version_get_info(tt, param: &iter_info);
749	param->flags |= iter_info.flags;
750
751	out:
752	if (tt)
753	dm_put_target_type(tt);
754	return 0;
755	}
756
757	static int list_versions(struct file *filp, struct dm_ioctl *param, size_t param_size)
758	{
759	return __list_versions(param, param_size, NULL);
760	}
761
762	static int get_target_version(struct file *filp, struct dm_ioctl *param, size_t param_size)
763	{
764	return __list_versions(param, param_size, name: param->name);
765	}
766
767	static int check_name(const char *name)
768	{
769	if (strchr(name, '/')) {
770	DMERR("device name cannot contain '/'");
771	return -EINVAL;
772	}
773
774	if (strcmp(name, DM_CONTROL_NODE) == 0 ||
775	strcmp(name, ".") == 0 ||
776	strcmp(name, "..") == 0) {
777	DMERR("device name cannot be \"%s\", \".\", or \"..\"", DM_CONTROL_NODE);
778	return -EINVAL;
779	}
780
781	return 0;
782	}
783
784	/*
785	* On successful return, the caller must not attempt to acquire
786	* _hash_lock without first calling dm_put_live_table, because dm_table_destroy
787	* waits for this dm_put_live_table and could be called under this lock.
788	*/
789	static struct dm_table *dm_get_inactive_table(struct mapped_device *md, int *srcu_idx)
790	{
791	struct hash_cell *hc;
792	struct dm_table *table = NULL;
793
794	/* increment rcu count, we don't care about the table pointer */
795	dm_get_live_table(md, srcu_idx);
796
797	down_read(sem: &_hash_lock);
798	hc = dm_get_mdptr(md);
799	if (!hc) {
800	DMERR("device has been removed from the dev hash table.");
801	goto out;
802	}
803
804	table = hc->new_map;
805
806	out:
807	up_read(sem: &_hash_lock);
808
809	return table;
810	}
811
812	static struct dm_table *dm_get_live_or_inactive_table(struct mapped_device *md,
813	struct dm_ioctl *param,
814	int *srcu_idx)
815	{
816	return (param->flags & DM_QUERY_INACTIVE_TABLE_FLAG) ?
817	dm_get_inactive_table(md, srcu_idx) : dm_get_live_table(md, srcu_idx);
818	}
819
820	/*
821	* Fills in a dm_ioctl structure, ready for sending back to
822	* userland.
823	*/
824	static void __dev_status(struct mapped_device *md, struct dm_ioctl *param)
825	{
826	struct gendisk *disk = dm_disk(md);
827	struct dm_table *table;
828	int srcu_idx;
829
830	param->flags &= ~(DM_SUSPEND_FLAG | DM_READONLY_FLAG |
831	DM_ACTIVE_PRESENT_FLAG | DM_INTERNAL_SUSPEND_FLAG);
832
833	if (dm_suspended_md(md))
834	param->flags |= DM_SUSPEND_FLAG;
835
836	if (dm_suspended_internally_md(md))
837	param->flags |= DM_INTERNAL_SUSPEND_FLAG;
838
839	if (dm_test_deferred_remove_flag(md))
840	param->flags |= DM_DEFERRED_REMOVE;
841
842	param->dev = huge_encode_dev(dev: disk_devt(disk));
843
844	/*
845	* Yes, this will be out of date by the time it gets back
846	* to userland, but it is still very useful for
847	* debugging.
848	*/
849	param->open_count = dm_open_count(md);
850
851	param->event_nr = dm_get_event_nr(md);
852	param->target_count = 0;
853
854	table = dm_get_live_table(md, srcu_idx: &srcu_idx);
855	if (table) {
856	if (!(param->flags & DM_QUERY_INACTIVE_TABLE_FLAG)) {
857	if (get_disk_ro(disk))
858	param->flags |= DM_READONLY_FLAG;
859	param->target_count = table->num_targets;
860	}
861
862	param->flags |= DM_ACTIVE_PRESENT_FLAG;
863	}
864	dm_put_live_table(md, srcu_idx);
865
866	if (param->flags & DM_QUERY_INACTIVE_TABLE_FLAG) {
867	int srcu_idx;
868
869	table = dm_get_inactive_table(md, srcu_idx: &srcu_idx);
870	if (table) {
871	if (!(dm_table_get_mode(t: table) & BLK_OPEN_WRITE))
872	param->flags |= DM_READONLY_FLAG;
873	param->target_count = table->num_targets;
874	}
875	dm_put_live_table(md, srcu_idx);
876	}
877	}
878
879	static int dev_create(struct file *filp, struct dm_ioctl *param, size_t param_size)
880	{
881	int r, m = DM_ANY_MINOR;
882	struct mapped_device *md;
883
884	r = check_name(name: param->name);
885	if (r)
886	return r;
887
888	if (param->flags & DM_PERSISTENT_DEV_FLAG)
889	m = MINOR(huge_decode_dev(param->dev));
890
891	r = dm_create(minor: m, md: &md);
892	if (r)
893	return r;
894
895	r = dm_hash_insert(name: param->name, uuid: *param->uuid ? param->uuid : NULL, md);
896	if (r) {
897	dm_put(md);
898	dm_destroy(md);
899	return r;
900	}
901
902	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
903
904	__dev_status(md, param);
905
906	dm_put(md);
907
908	return 0;
909	}
910
911	/*
912	* Always use UUID for lookups if it's present, otherwise use name or dev.
913	*/
914	static struct hash_cell *__find_device_hash_cell(struct dm_ioctl *param)
915	{
916	struct hash_cell *hc = NULL;
917
918	if (*param->uuid) {
919	if (*param->name || param->dev) {
920	DMERR("Invalid ioctl structure: uuid %s, name %s, dev %llx",
921	param->uuid, param->name, (unsigned long long)param->dev);
922	return NULL;
923	}
924
925	hc = __get_uuid_cell(str: param->uuid);
926	if (!hc)
927	return NULL;
928	} else if (*param->name) {
929	if (param->dev) {
930	DMERR("Invalid ioctl structure: name %s, dev %llx",
931	param->name, (unsigned long long)param->dev);
932	return NULL;
933	}
934
935	hc = __get_name_cell(str: param->name);
936	if (!hc)
937	return NULL;
938	} else if (param->dev) {
939	hc = __get_dev_cell(dev: param->dev);
940	if (!hc)
941	return NULL;
942	} else
943	return NULL;
944
945	/*
946	* Sneakily write in both the name and the uuid
947	* while we have the cell.
948	*/
949	strscpy(p: param->name, q: hc->name, size: sizeof(param->name));
950	if (hc->uuid)
951	strscpy(p: param->uuid, q: hc->uuid, size: sizeof(param->uuid));
952	else
953	param->uuid[0] = '\0';
954
955	if (hc->new_map)
956	param->flags |= DM_INACTIVE_PRESENT_FLAG;
957	else
958	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
959
960	return hc;
961	}
962
963	static struct mapped_device *find_device(struct dm_ioctl *param)
964	{
965	struct hash_cell *hc;
966	struct mapped_device *md = NULL;
967
968	down_read(sem: &_hash_lock);
969	hc = __find_device_hash_cell(param);
970	if (hc)
971	md = hc->md;
972	up_read(sem: &_hash_lock);
973
974	return md;
975	}
976
977	static int dev_remove(struct file *filp, struct dm_ioctl *param, size_t param_size)
978	{
979	struct hash_cell *hc;
980	struct mapped_device *md;
981	int r;
982	struct dm_table *t;
983
984	down_write(sem: &_hash_lock);
985	hc = __find_device_hash_cell(param);
986
987	if (!hc) {
988	DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
989	up_write(sem: &_hash_lock);
990	return -ENXIO;
991	}
992
993	md = hc->md;
994
995	/*
996	* Ensure the device is not open and nothing further can open it.
997	*/
998	r = dm_lock_for_deletion(md, mark_deferred: !!(param->flags & DM_DEFERRED_REMOVE), only_deferred: false);
999	if (r) {
1000	if (r == -EBUSY && param->flags & DM_DEFERRED_REMOVE) {
1001	up_write(sem: &_hash_lock);
1002	dm_put(md);
1003	return 0;
1004	}
1005	DMDEBUG_LIMIT("unable to remove open device %s", hc->name);
1006	up_write(sem: &_hash_lock);
1007	dm_put(md);
1008	return r;
1009	}
1010
1011	t = __hash_remove(hc);
1012	up_write(sem: &_hash_lock);
1013
1014	if (t) {
1015	dm_sync_table(md);
1016	dm_table_destroy(t);
1017	}
1018
1019	param->flags &= ~DM_DEFERRED_REMOVE;
1020
1021	dm_ima_measure_on_device_remove(md, remove_all: false);
1022
1023	if (!dm_kobject_uevent(md, action: KOBJ_REMOVE, cookie: param->event_nr, need_resize_uevent: false))
1024	param->flags |= DM_UEVENT_GENERATED_FLAG;
1025
1026	dm_put(md);
1027	dm_destroy(md);
1028	return 0;
1029	}
1030
1031	/*
1032	* Check a string doesn't overrun the chunk of
1033	* memory we copied from userland.
1034	*/
1035	static int invalid_str(char *str, void *end)
1036	{
1037	while ((void *) str < end)
1038	if (!*str++)
1039	return 0;
1040
1041	return -EINVAL;
1042	}
1043
1044	static int dev_rename(struct file *filp, struct dm_ioctl *param, size_t param_size)
1045	{
1046	int r;
1047	char *new_data = (char *) param + param->data_start;
1048	struct mapped_device *md;
1049	unsigned int change_uuid = (param->flags & DM_UUID_FLAG) ? 1 : 0;
1050
1051	if (new_data < param->data ||
1052	invalid_str(str: new_data, end: (void *) param + param_size) || !*new_data ||
1053	strlen(new_data) > (change_uuid ? DM_UUID_LEN - 1 : DM_NAME_LEN - 1)) {
1054	DMERR("Invalid new mapped device name or uuid string supplied.");
1055	return -EINVAL;
1056	}
1057
1058	if (!change_uuid) {
1059	r = check_name(name: new_data);
1060	if (r)
1061	return r;
1062	}
1063
1064	md = dm_hash_rename(param, new: new_data);
1065	if (IS_ERR(ptr: md))
1066	return PTR_ERR(ptr: md);
1067
1068	__dev_status(md, param);
1069	dm_put(md);
1070
1071	return 0;
1072	}
1073
1074	static int dev_set_geometry(struct file *filp, struct dm_ioctl *param, size_t param_size)
1075	{
1076	int r = -EINVAL, x;
1077	struct mapped_device *md;
1078	struct hd_geometry geometry;
1079	unsigned long indata[4];
1080	char *geostr = (char *) param + param->data_start;
1081	char dummy;
1082
1083	md = find_device(param);
1084	if (!md)
1085	return -ENXIO;
1086
1087	if (geostr < param->data ||
1088	invalid_str(str: geostr, end: (void *) param + param_size)) {
1089	DMERR("Invalid geometry supplied.");
1090	goto out;
1091	}
1092
1093	x = sscanf(geostr, "%lu %lu %lu %lu%c", indata,
1094	indata + 1, indata + 2, indata + 3, &dummy);
1095
1096	if (x != 4) {
1097	DMERR("Unable to interpret geometry settings.");
1098	goto out;
1099	}
1100
1101	if (indata[0] > 65535 || indata[1] > 255 || indata[2] > 255) {
1102	DMERR("Geometry exceeds range limits.");
1103	goto out;
1104	}
1105
1106	geometry.cylinders = indata[0];
1107	geometry.heads = indata[1];
1108	geometry.sectors = indata[2];
1109	geometry.start = indata[3];
1110
1111	r = dm_set_geometry(md, geo: &geometry);
1112
1113	param->data_size = 0;
1114
1115	out:
1116	dm_put(md);
1117	return r;
1118	}
1119
1120	static int do_suspend(struct dm_ioctl *param)
1121	{
1122	int r = 0;
1123	unsigned int suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
1124	struct mapped_device *md;
1125
1126	md = find_device(param);
1127	if (!md)
1128	return -ENXIO;
1129
1130	if (param->flags & DM_SKIP_LOCKFS_FLAG)
1131	suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
1132	if (param->flags & DM_NOFLUSH_FLAG)
1133	suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
1134
1135	if (!dm_suspended_md(md)) {
1136	r = dm_suspend(md, suspend_flags);
1137	if (r)
1138	goto out;
1139	}
1140
1141	__dev_status(md, param);
1142
1143	out:
1144	dm_put(md);
1145
1146	return r;
1147	}
1148
1149	static int do_resume(struct dm_ioctl *param)
1150	{
1151	int r = 0;
1152	unsigned int suspend_flags = DM_SUSPEND_LOCKFS_FLAG;
1153	struct hash_cell *hc;
1154	struct mapped_device *md;
1155	struct dm_table *new_map, *old_map = NULL;
1156	bool need_resize_uevent = false;
1157
1158	down_write(sem: &_hash_lock);
1159
1160	hc = __find_device_hash_cell(param);
1161	if (!hc) {
1162	DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
1163	up_write(sem: &_hash_lock);
1164	return -ENXIO;
1165	}
1166
1167	md = hc->md;
1168
1169	new_map = hc->new_map;
1170	hc->new_map = NULL;
1171	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1172
1173	up_write(sem: &_hash_lock);
1174
1175	/* Do we need to load a new map ? */
1176	if (new_map) {
1177	sector_t old_size, new_size;
1178
1179	/* Suspend if it isn't already suspended */
1180	if (param->flags & DM_SKIP_LOCKFS_FLAG)
1181	suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
1182	if (param->flags & DM_NOFLUSH_FLAG)
1183	suspend_flags |= DM_SUSPEND_NOFLUSH_FLAG;
1184	if (!dm_suspended_md(md))
1185	dm_suspend(md, suspend_flags);
1186
1187	old_size = dm_get_size(md);
1188	old_map = dm_swap_table(md, t: new_map);
1189	if (IS_ERR(ptr: old_map)) {
1190	dm_sync_table(md);
1191	dm_table_destroy(t: new_map);
1192	dm_put(md);
1193	return PTR_ERR(ptr: old_map);
1194	}
1195	new_size = dm_get_size(md);
1196	if (old_size && new_size && old_size != new_size)
1197	need_resize_uevent = true;
1198
1199	if (dm_table_get_mode(t: new_map) & BLK_OPEN_WRITE)
1200	set_disk_ro(disk: dm_disk(md), read_only: 0);
1201	else
1202	set_disk_ro(disk: dm_disk(md), read_only: 1);
1203	}
1204
1205	if (dm_suspended_md(md)) {
1206	r = dm_resume(md);
1207	if (!r) {
1208	dm_ima_measure_on_device_resume(md, swap: new_map ? true : false);
1209
1210	if (!dm_kobject_uevent(md, action: KOBJ_CHANGE, cookie: param->event_nr, need_resize_uevent))
1211	param->flags |= DM_UEVENT_GENERATED_FLAG;
1212	}
1213	}
1214
1215	/*
1216	* Since dm_swap_table synchronizes RCU, nobody should be in
1217	* read-side critical section already.
1218	*/
1219	if (old_map)
1220	dm_table_destroy(t: old_map);
1221
1222	if (!r)
1223	__dev_status(md, param);
1224
1225	dm_put(md);
1226	return r;
1227	}
1228
1229	/*
1230	* Set or unset the suspension state of a device.
1231	* If the device already is in the requested state we just return its status.
1232	*/
1233	static int dev_suspend(struct file *filp, struct dm_ioctl *param, size_t param_size)
1234	{
1235	if (param->flags & DM_SUSPEND_FLAG)
1236	return do_suspend(param);
1237
1238	return do_resume(param);
1239	}
1240
1241	/*
1242	* Copies device info back to user space, used by
1243	* the create and info ioctls.
1244	*/
1245	static int dev_status(struct file *filp, struct dm_ioctl *param, size_t param_size)
1246	{
1247	struct mapped_device *md;
1248
1249	md = find_device(param);
1250	if (!md)
1251	return -ENXIO;
1252
1253	__dev_status(md, param);
1254	dm_put(md);
1255
1256	return 0;
1257	}
1258
1259	/*
1260	* Build up the status struct for each target
1261	*/
1262	static void retrieve_status(struct dm_table *table,
1263	struct dm_ioctl *param, size_t param_size)
1264	{
1265	unsigned int i, num_targets;
1266	struct dm_target_spec *spec;
1267	char *outbuf, *outptr;
1268	status_type_t type;
1269	size_t remaining, len, used = 0;
1270	unsigned int status_flags = 0;
1271
1272	outptr = outbuf = get_result_buffer(param, param_size, len: &len);
1273
1274	if (param->flags & DM_STATUS_TABLE_FLAG)
1275	type = STATUSTYPE_TABLE;
1276	else if (param->flags & DM_IMA_MEASUREMENT_FLAG)
1277	type = STATUSTYPE_IMA;
1278	else
1279	type = STATUSTYPE_INFO;
1280
1281	/* Get all the target info */
1282	num_targets = table->num_targets;
1283	for (i = 0; i < num_targets; i++) {
1284	struct dm_target *ti = dm_table_get_target(t: table, index: i);
1285	size_t l;
1286
1287	remaining = len - (outptr - outbuf);
1288	if (remaining <= sizeof(struct dm_target_spec)) {
1289	param->flags |= DM_BUFFER_FULL_FLAG;
1290	break;
1291	}
1292
1293	spec = (struct dm_target_spec *) outptr;
1294
1295	spec->status = 0;
1296	spec->sector_start = ti->begin;
1297	spec->length = ti->len;
1298	strscpy_pad(dest: spec->target_type, src: ti->type->name,
1299	count: sizeof(spec->target_type));
1300
1301	outptr += sizeof(struct dm_target_spec);
1302	remaining = len - (outptr - outbuf);
1303	if (remaining <= 0) {
1304	param->flags |= DM_BUFFER_FULL_FLAG;
1305	break;
1306	}
1307
1308	/* Get the status/table string from the target driver */
1309	if (ti->type->status) {
1310	if (param->flags & DM_NOFLUSH_FLAG)
1311	status_flags |= DM_STATUS_NOFLUSH_FLAG;
1312	ti->type->status(ti, type, status_flags, outptr, remaining);
1313	} else
1314	outptr[0] = '\0';
1315
1316	l = strlen(outptr) + 1;
1317	if (l == remaining) {
1318	param->flags |= DM_BUFFER_FULL_FLAG;
1319	break;
1320	}
1321
1322	outptr += l;
1323	used = param->data_start + (outptr - outbuf);
1324
1325	outptr = align_ptr(ptr: outptr);
1326	spec->next = outptr - outbuf;
1327	}
1328
1329	if (used)
1330	param->data_size = used;
1331
1332	param->target_count = num_targets;
1333	}
1334
1335	/*
1336	* Wait for a device to report an event
1337	*/
1338	static int dev_wait(struct file *filp, struct dm_ioctl *param, size_t param_size)
1339	{
1340	int r = 0;
1341	struct mapped_device *md;
1342	struct dm_table *table;
1343	int srcu_idx;
1344
1345	md = find_device(param);
1346	if (!md)
1347	return -ENXIO;
1348
1349	/*
1350	* Wait for a notification event
1351	*/
1352	if (dm_wait_event(md, event_nr: param->event_nr)) {
1353	r = -ERESTARTSYS;
1354	goto out;
1355	}
1356
1357	/*
1358	* The userland program is going to want to know what
1359	* changed to trigger the event, so we may as well tell
1360	* him and save an ioctl.
1361	*/
1362	__dev_status(md, param);
1363
1364	table = dm_get_live_or_inactive_table(md, param, srcu_idx: &srcu_idx);
1365	if (table)
1366	retrieve_status(table, param, param_size);
1367	dm_put_live_table(md, srcu_idx);
1368
1369	out:
1370	dm_put(md);
1371
1372	return r;
1373	}
1374
1375	/*
1376	* Remember the global event number and make it possible to poll
1377	* for further events.
1378	*/
1379	static int dev_arm_poll(struct file *filp, struct dm_ioctl *param, size_t param_size)
1380	{
1381	struct dm_file *priv = filp->private_data;
1382
1383	priv->global_event_nr = atomic_read(v: &dm_global_event_nr);
1384
1385	return 0;
1386	}
1387
1388	static inline blk_mode_t get_mode(struct dm_ioctl *param)
1389	{
1390	blk_mode_t mode = BLK_OPEN_READ | BLK_OPEN_WRITE;
1391
1392	if (param->flags & DM_READONLY_FLAG)
1393	mode = BLK_OPEN_READ;
1394
1395	return mode;
1396	}
1397
1398	static int next_target(struct dm_target_spec *last, uint32_t next, const char *end,
1399	struct dm_target_spec **spec, char **target_params)
1400	{
1401	static_assert(__alignof__(struct dm_target_spec) <= 8,
1402	"struct dm_target_spec must not require more than 8-byte alignment");
1403
1404	/*
1405	* Number of bytes remaining, starting with last. This is always
1406	* sizeof(struct dm_target_spec) or more, as otherwise *last was
1407	* out of bounds already.
1408	*/
1409	size_t remaining = end - (char *)last;
1410
1411	/*
1412	* There must be room for both the next target spec and the
1413	* NUL-terminator of the target itself.
1414	*/
1415	if (remaining - sizeof(struct dm_target_spec) <= next) {
1416	DMERR("Target spec extends beyond end of parameters");
1417	return -EINVAL;
1418	}
1419
1420	if (next % __alignof__(struct dm_target_spec)) {
1421	DMERR("Next dm_target_spec (offset %u) is not %zu-byte aligned",
1422	next, __alignof__(struct dm_target_spec));
1423	return -EINVAL;
1424	}
1425
1426	*spec = (struct dm_target_spec *) ((unsigned char *) last + next);
1427	*target_params = (char *) (*spec + 1);
1428
1429	return 0;
1430	}
1431
1432	static int populate_table(struct dm_table *table,
1433	struct dm_ioctl *param, size_t param_size)
1434	{
1435	int r;
1436	unsigned int i = 0;
1437	struct dm_target_spec *spec = (struct dm_target_spec *) param;
1438	uint32_t next = param->data_start;
1439	const char *const end = (const char *) param + param_size;
1440	char *target_params;
1441	size_t min_size = sizeof(struct dm_ioctl);
1442
1443	if (!param->target_count) {
1444	DMERR("%s: no targets specified", __func__);
1445	return -EINVAL;
1446	}
1447
1448	for (i = 0; i < param->target_count; i++) {
1449	const char *nul_terminator;
1450
1451	if (next < min_size) {
1452	DMERR("%s: next target spec (offset %u) overlaps %s",
1453	__func__, next, i ? "previous target" : "'struct dm_ioctl'");
1454	return -EINVAL;
1455	}
1456
1457	r = next_target(last: spec, next, end, spec: &spec, target_params: &target_params);
1458	if (r) {
1459	DMERR("unable to find target");
1460	return r;
1461	}
1462
1463	nul_terminator = memchr(p: target_params, c: 0, size: (size_t)(end - target_params));
1464	if (nul_terminator == NULL) {
1465	DMERR("%s: target parameters not NUL-terminated", __func__);
1466	return -EINVAL;
1467	}
1468
1469	/* Add 1 for NUL terminator */
1470	min_size = (size_t)(nul_terminator - (const char *)spec) + 1;
1471
1472	r = dm_table_add_target(t: table, type: spec->target_type,
1473	start: (sector_t) spec->sector_start,
1474	len: (sector_t) spec->length,
1475	params: target_params);
1476	if (r) {
1477	DMERR("error adding target to table");
1478	return r;
1479	}
1480
1481	next = spec->next;
1482	}
1483
1484	return dm_table_complete(t: table);
1485	}
1486
1487	static bool is_valid_type(enum dm_queue_mode cur, enum dm_queue_mode new)
1488	{
1489	if (cur == new ||
1490	(cur == DM_TYPE_BIO_BASED && new == DM_TYPE_DAX_BIO_BASED))
1491	return true;
1492
1493	return false;
1494	}
1495
1496	static int table_load(struct file *filp, struct dm_ioctl *param, size_t param_size)
1497	{
1498	int r;
1499	struct hash_cell *hc;
1500	struct dm_table *t, *old_map = NULL;
1501	struct mapped_device *md;
1502	struct target_type *immutable_target_type;
1503
1504	md = find_device(param);
1505	if (!md)
1506	return -ENXIO;
1507
1508	r = dm_table_create(result: &t, mode: get_mode(param), num_targets: param->target_count, md);
1509	if (r)
1510	goto err;
1511
1512	/* Protect md->type and md->queue against concurrent table loads. */
1513	dm_lock_md_type(md);
1514	r = populate_table(table: t, param, param_size);
1515	if (r)
1516	goto err_unlock_md_type;
1517
1518	dm_ima_measure_on_table_load(table: t, status_flags: STATUSTYPE_IMA);
1519
1520	immutable_target_type = dm_get_immutable_target_type(md);
1521	if (immutable_target_type &&
1522	(immutable_target_type != dm_table_get_immutable_target_type(t)) &&
1523	!dm_table_get_wildcard_target(t)) {
1524	DMERR("can't replace immutable target type %s",
1525	immutable_target_type->name);
1526	r = -EINVAL;
1527	goto err_unlock_md_type;
1528	}
1529
1530	if (dm_get_md_type(md) == DM_TYPE_NONE) {
1531	/* setup md->queue to reflect md's type (may block) */
1532	r = dm_setup_md_queue(md, t);
1533	if (r) {
1534	DMERR("unable to set up device queue for new table.");
1535	goto err_unlock_md_type;
1536	}
1537	} else if (!is_valid_type(cur: dm_get_md_type(md), new: dm_table_get_type(t))) {
1538	DMERR("can't change device type (old=%u vs new=%u) after initial table load.",
1539	dm_get_md_type(md), dm_table_get_type(t));
1540	r = -EINVAL;
1541	goto err_unlock_md_type;
1542	}
1543
1544	dm_unlock_md_type(md);
1545
1546	/* stage inactive table */
1547	down_write(sem: &_hash_lock);
1548	hc = dm_get_mdptr(md);
1549	if (!hc) {
1550	DMERR("device has been removed from the dev hash table.");
1551	up_write(sem: &_hash_lock);
1552	r = -ENXIO;
1553	goto err_destroy_table;
1554	}
1555
1556	if (hc->new_map)
1557	old_map = hc->new_map;
1558	hc->new_map = t;
1559	up_write(sem: &_hash_lock);
1560
1561	param->flags |= DM_INACTIVE_PRESENT_FLAG;
1562	__dev_status(md, param);
1563
1564	if (old_map) {
1565	dm_sync_table(md);
1566	dm_table_destroy(t: old_map);
1567	}
1568
1569	dm_put(md);
1570
1571	return 0;
1572
1573	err_unlock_md_type:
1574	dm_unlock_md_type(md);
1575	err_destroy_table:
1576	dm_table_destroy(t);
1577	err:
1578	dm_put(md);
1579
1580	return r;
1581	}
1582
1583	static int table_clear(struct file *filp, struct dm_ioctl *param, size_t param_size)
1584	{
1585	struct hash_cell *hc;
1586	struct mapped_device *md;
1587	struct dm_table *old_map = NULL;
1588	bool has_new_map = false;
1589
1590	down_write(sem: &_hash_lock);
1591
1592	hc = __find_device_hash_cell(param);
1593	if (!hc) {
1594	DMDEBUG_LIMIT("device doesn't appear to be in the dev hash table.");
1595	up_write(sem: &_hash_lock);
1596	return -ENXIO;
1597	}
1598
1599	if (hc->new_map) {
1600	old_map = hc->new_map;
1601	hc->new_map = NULL;
1602	has_new_map = true;
1603	}
1604
1605	md = hc->md;
1606	up_write(sem: &_hash_lock);
1607
1608	param->flags &= ~DM_INACTIVE_PRESENT_FLAG;
1609	__dev_status(md, param);
1610
1611	if (old_map) {
1612	dm_sync_table(md);
1613	dm_table_destroy(t: old_map);
1614	}
1615	dm_ima_measure_on_table_clear(md, new_map: has_new_map);
1616	dm_put(md);
1617
1618	return 0;
1619	}
1620
1621	/*
1622	* Retrieves a list of devices used by a particular dm device.
1623	*/
1624	static void retrieve_deps(struct dm_table *table,
1625	struct dm_ioctl *param, size_t param_size)
1626	{
1627	unsigned int count = 0;
1628	struct list_head *tmp;
1629	size_t len, needed;
1630	struct dm_dev_internal *dd;
1631	struct dm_target_deps *deps;
1632
1633	down_read(sem: &table->devices_lock);
1634
1635	deps = get_result_buffer(param, param_size, len: &len);
1636
1637	/*
1638	* Count the devices.
1639	*/
1640	list_for_each(tmp, dm_table_get_devices(table))
1641	count++;
1642
1643	/*
1644	* Check we have enough space.
1645	*/
1646	needed = struct_size(deps, dev, count);
1647	if (len < needed) {
1648	param->flags |= DM_BUFFER_FULL_FLAG;
1649	goto out;
1650	}
1651
1652	/*
1653	* Fill in the devices.
1654	*/
1655	deps->count = count;
1656	count = 0;
1657	list_for_each_entry(dd, dm_table_get_devices(table), list)
1658	deps->dev[count++] = huge_encode_dev(dev: dd->dm_dev->bdev->bd_dev);
1659
1660	param->data_size = param->data_start + needed;
1661
1662	out:
1663	up_read(sem: &table->devices_lock);
1664	}
1665
1666	static int table_deps(struct file *filp, struct dm_ioctl *param, size_t param_size)
1667	{
1668	struct mapped_device *md;
1669	struct dm_table *table;
1670	int srcu_idx;
1671
1672	md = find_device(param);
1673	if (!md)
1674	return -ENXIO;
1675
1676	__dev_status(md, param);
1677
1678	table = dm_get_live_or_inactive_table(md, param, srcu_idx: &srcu_idx);
1679	if (table)
1680	retrieve_deps(table, param, param_size);
1681	dm_put_live_table(md, srcu_idx);
1682
1683	dm_put(md);
1684
1685	return 0;
1686	}
1687
1688	/*
1689	* Return the status of a device as a text string for each
1690	* target.
1691	*/
1692	static int table_status(struct file *filp, struct dm_ioctl *param, size_t param_size)
1693	{
1694	struct mapped_device *md;
1695	struct dm_table *table;
1696	int srcu_idx;
1697
1698	md = find_device(param);
1699	if (!md)
1700	return -ENXIO;
1701
1702	__dev_status(md, param);
1703
1704	table = dm_get_live_or_inactive_table(md, param, srcu_idx: &srcu_idx);
1705	if (table)
1706	retrieve_status(table, param, param_size);
1707	dm_put_live_table(md, srcu_idx);
1708
1709	dm_put(md);
1710
1711	return 0;
1712	}
1713
1714	/*
1715	* Process device-mapper dependent messages. Messages prefixed with '@'
1716	* are processed by the DM core. All others are delivered to the target.
1717	* Returns a number <= 1 if message was processed by device mapper.
1718	* Returns 2 if message should be delivered to the target.
1719	*/
1720	static int message_for_md(struct mapped_device *md, unsigned int argc, char **argv,
1721	char *result, unsigned int maxlen)
1722	{
1723	int r;
1724
1725	if (**argv != '@')
1726	return 2; /* no '@' prefix, deliver to target */
1727
1728	if (!strcasecmp(s1: argv[0], s2: "@cancel_deferred_remove")) {
1729	if (argc != 1) {
1730	DMERR("Invalid arguments for @cancel_deferred_remove");
1731	return -EINVAL;
1732	}
1733	return dm_cancel_deferred_remove(md);
1734	}
1735
1736	r = dm_stats_message(md, argc, argv, result, maxlen);
1737	if (r < 2)
1738	return r;
1739
1740	DMERR("Unsupported message sent to DM core: %s", argv[0]);
1741	return -EINVAL;
1742	}
1743
1744	/*
1745	* Pass a message to the target that's at the supplied device offset.
1746	*/
1747	static int target_message(struct file *filp, struct dm_ioctl *param, size_t param_size)
1748	{
1749	int r, argc;
1750	char **argv;
1751	struct mapped_device *md;
1752	struct dm_table *table;
1753	struct dm_target *ti;
1754	struct dm_target_msg *tmsg = (void *) param + param->data_start;
1755	size_t maxlen;
1756	char *result = get_result_buffer(param, param_size, len: &maxlen);
1757	int srcu_idx;
1758
1759	md = find_device(param);
1760	if (!md)
1761	return -ENXIO;
1762
1763	if (tmsg < (struct dm_target_msg *) param->data ||
1764	invalid_str(str: tmsg->message, end: (void *) param + param_size)) {
1765	DMERR("Invalid target message parameters.");
1766	r = -EINVAL;
1767	goto out;
1768	}
1769
1770	r = dm_split_args(argc: &argc, argvp: &argv, input: tmsg->message);
1771	if (r) {
1772	DMERR("Failed to split target message parameters");
1773	goto out;
1774	}
1775
1776	if (!argc) {
1777	DMERR("Empty message received.");
1778	r = -EINVAL;
1779	goto out_argv;
1780	}
1781
1782	r = message_for_md(md, argc, argv, result, maxlen);
1783	if (r <= 1)
1784	goto out_argv;
1785
1786	table = dm_get_live_table(md, srcu_idx: &srcu_idx);
1787	if (!table)
1788	goto out_table;
1789
1790	if (dm_deleting_md(md)) {
1791	r = -ENXIO;
1792	goto out_table;
1793	}
1794
1795	ti = dm_table_find_target(t: table, sector: tmsg->sector);
1796	if (!ti) {
1797	DMERR("Target message sector outside device.");
1798	r = -EINVAL;
1799	} else if (ti->type->message)
1800	r = ti->type->message(ti, argc, argv, result, maxlen);
1801	else {
1802	DMERR("Target type does not support messages");
1803	r = -EINVAL;
1804	}
1805
1806	out_table:
1807	dm_put_live_table(md, srcu_idx);
1808	out_argv:
1809	kfree(objp: argv);
1810	out:
1811	if (r >= 0)
1812	__dev_status(md, param);
1813
1814	if (r == 1) {
1815	param->flags |= DM_DATA_OUT_FLAG;
1816	if (dm_message_test_buffer_overflow(result, maxlen))
1817	param->flags |= DM_BUFFER_FULL_FLAG;
1818	else
1819	param->data_size = param->data_start + strlen(result) + 1;
1820	r = 0;
1821	}
1822
1823	dm_put(md);
1824	return r;
1825	}
1826
1827	/*
1828	* The ioctl parameter block consists of two parts, a dm_ioctl struct
1829	* followed by a data buffer. This flag is set if the second part,
1830	* which has a variable size, is not used by the function processing
1831	* the ioctl.
1832	*/
1833	#define IOCTL_FLAGS_NO_PARAMS 1
1834	#define IOCTL_FLAGS_ISSUE_GLOBAL_EVENT 2
1835
1836	/*
1837	*---------------------------------------------------------------
1838	* Implementation of open/close/ioctl on the special char device.
1839	*---------------------------------------------------------------
1840	*/
1841	static ioctl_fn lookup_ioctl(unsigned int cmd, int *ioctl_flags)
1842	{
1843	static const struct {
1844	int cmd;
1845	int flags;
1846	ioctl_fn fn;
1847	} _ioctls[] = {
1848	{DM_VERSION_CMD, 0, NULL}, /* version is dealt with elsewhere */
1849	{DM_REMOVE_ALL_CMD, IOCTL_FLAGS_NO_PARAMS | IOCTL_FLAGS_ISSUE_GLOBAL_EVENT, remove_all},
1850	{DM_LIST_DEVICES_CMD, 0, list_devices},
1851
1852	{DM_DEV_CREATE_CMD, IOCTL_FLAGS_NO_PARAMS | IOCTL_FLAGS_ISSUE_GLOBAL_EVENT, dev_create},
1853	{DM_DEV_REMOVE_CMD, IOCTL_FLAGS_NO_PARAMS | IOCTL_FLAGS_ISSUE_GLOBAL_EVENT, dev_remove},
1854	{DM_DEV_RENAME_CMD, IOCTL_FLAGS_ISSUE_GLOBAL_EVENT, dev_rename},
1855	{DM_DEV_SUSPEND_CMD, IOCTL_FLAGS_NO_PARAMS, dev_suspend},
1856	{DM_DEV_STATUS_CMD, IOCTL_FLAGS_NO_PARAMS, dev_status},
1857	{DM_DEV_WAIT_CMD, 0, dev_wait},
1858
1859	{DM_TABLE_LOAD_CMD, 0, table_load},
1860	{DM_TABLE_CLEAR_CMD, IOCTL_FLAGS_NO_PARAMS, table_clear},
1861	{DM_TABLE_DEPS_CMD, 0, table_deps},
1862	{DM_TABLE_STATUS_CMD, 0, table_status},
1863
1864	{DM_LIST_VERSIONS_CMD, 0, list_versions},
1865
1866	{DM_TARGET_MSG_CMD, 0, target_message},
1867	{DM_DEV_SET_GEOMETRY_CMD, 0, dev_set_geometry},
1868	{DM_DEV_ARM_POLL_CMD, IOCTL_FLAGS_NO_PARAMS, dev_arm_poll},
1869	{DM_GET_TARGET_VERSION_CMD, 0, get_target_version},
1870	};
1871
1872	if (unlikely(cmd >= ARRAY_SIZE(_ioctls)))
1873	return NULL;
1874
1875	cmd = array_index_nospec(cmd, ARRAY_SIZE(_ioctls));
1876	*ioctl_flags = _ioctls[cmd].flags;
1877	return _ioctls[cmd].fn;
1878	}
1879
1880	/*
1881	* As well as checking the version compatibility this always
1882	* copies the kernel interface version out.
1883	*/
1884	static int check_version(unsigned int cmd, struct dm_ioctl __user *user,
1885	struct dm_ioctl *kernel_params)
1886	{
1887	int r = 0;
1888
1889	/* Make certain version is first member of dm_ioctl struct */
1890	BUILD_BUG_ON(offsetof(struct dm_ioctl, version) != 0);
1891
1892	if (copy_from_user(to: kernel_params->version, from: user->version, n: sizeof(kernel_params->version)))
1893	return -EFAULT;
1894
1895	if ((kernel_params->version[0] != DM_VERSION_MAJOR) ||
1896	(kernel_params->version[1] > DM_VERSION_MINOR)) {
1897	DMERR("ioctl interface mismatch: kernel(%u.%u.%u), user(%u.%u.%u), cmd(%d)",
1898	DM_VERSION_MAJOR, DM_VERSION_MINOR,
1899	DM_VERSION_PATCHLEVEL,
1900	kernel_params->version[0],
1901	kernel_params->version[1],
1902	kernel_params->version[2],
1903	cmd);
1904	r = -EINVAL;
1905	}
1906
1907	/*
1908	* Fill in the kernel version.
1909	*/
1910	kernel_params->version[0] = DM_VERSION_MAJOR;
1911	kernel_params->version[1] = DM_VERSION_MINOR;
1912	kernel_params->version[2] = DM_VERSION_PATCHLEVEL;
1913	if (copy_to_user(to: user->version, from: kernel_params->version, n: sizeof(kernel_params->version)))
1914	return -EFAULT;
1915
1916	return r;
1917	}
1918
1919	#define DM_PARAMS_MALLOC 0x0001 /* Params allocated with kvmalloc() */
1920	#define DM_WIPE_BUFFER 0x0010 /* Wipe input buffer before returning from ioctl */
1921
1922	static void free_params(struct dm_ioctl *param, size_t param_size, int param_flags)
1923	{
1924	if (param_flags & DM_WIPE_BUFFER)
1925	memset(param, 0, param_size);
1926
1927	if (param_flags & DM_PARAMS_MALLOC)
1928	kvfree(addr: param);
1929	}
1930
1931	static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl *param_kernel,
1932	int ioctl_flags, struct dm_ioctl **param, int *param_flags)
1933	{
1934	struct dm_ioctl *dmi;
1935	int secure_data;
1936	const size_t minimum_data_size = offsetof(struct dm_ioctl, data);
1937
1938	/* check_version() already copied version from userspace, avoid TOCTOU */
1939	if (copy_from_user(to: (char *)param_kernel + sizeof(param_kernel->version),
1940	from: (char __user *)user + sizeof(param_kernel->version),
1941	n: minimum_data_size - sizeof(param_kernel->version)))
1942	return -EFAULT;
1943
1944	if (param_kernel->data_size < minimum_data_size) {
1945	DMERR("Invalid data size in the ioctl structure: %u",
1946	param_kernel->data_size);
1947	return -EINVAL;
1948	}
1949
1950	secure_data = param_kernel->flags & DM_SECURE_DATA_FLAG;
1951
1952	*param_flags = secure_data ? DM_WIPE_BUFFER : 0;
1953
1954	if (ioctl_flags & IOCTL_FLAGS_NO_PARAMS) {
1955	dmi = param_kernel;
1956	dmi->data_size = minimum_data_size;
1957	goto data_copied;
1958	}
1959
1960	/*
1961	* Use __GFP_HIGH to avoid low memory issues when a device is
1962	* suspended and the ioctl is needed to resume it.
1963	* Use kmalloc() rather than vmalloc() when we can.
1964	*/
1965	dmi = NULL;
1966	dmi = kvmalloc(size: param_kernel->data_size, GFP_NOIO | __GFP_HIGH);
1967
1968	if (!dmi) {
1969	if (secure_data && clear_user(to: user, n: param_kernel->data_size))
1970	return -EFAULT;
1971	return -ENOMEM;
1972	}
1973
1974	*param_flags |= DM_PARAMS_MALLOC;
1975
1976	/* Copy from param_kernel (which was already copied from user) */
1977	memcpy(dmi, param_kernel, minimum_data_size);
1978
1979	if (copy_from_user(to: &dmi->data, from: (char __user *)user + minimum_data_size,
1980	n: param_kernel->data_size - minimum_data_size))
1981	goto bad;
1982	data_copied:
1983	/* Wipe the user buffer so we do not return it to userspace */
1984	if (secure_data && clear_user(to: user, n: param_kernel->data_size))
1985	goto bad;
1986
1987	*param = dmi;
1988	return 0;
1989
1990	bad:
1991	free_params(param: dmi, param_size: param_kernel->data_size, param_flags: *param_flags);
1992
1993	return -EFAULT;
1994	}
1995
1996	static int validate_params(uint cmd, struct dm_ioctl *param)
1997	{
1998	/* Always clear this flag */
1999	param->flags &= ~DM_BUFFER_FULL_FLAG;
2000	param->flags &= ~DM_UEVENT_GENERATED_FLAG;
2001	param->flags &= ~DM_SECURE_DATA_FLAG;
2002	param->flags &= ~DM_DATA_OUT_FLAG;
2003
2004	/* Ignores parameters */
2005	if (cmd == DM_REMOVE_ALL_CMD ||
2006	cmd == DM_LIST_DEVICES_CMD ||
2007	cmd == DM_LIST_VERSIONS_CMD)
2008	return 0;
2009
2010	if (cmd == DM_DEV_CREATE_CMD) {
2011	if (!*param->name) {
2012	DMERR("name not supplied when creating device");
2013	return -EINVAL;
2014	}
2015	} else if (*param->uuid && *param->name) {
2016	DMERR("only supply one of name or uuid, cmd(%u)", cmd);
2017	return -EINVAL;
2018	}
2019
2020	/* Ensure strings are terminated */
2021	param->name[DM_NAME_LEN - 1] = '\0';
2022	param->uuid[DM_UUID_LEN - 1] = '\0';
2023
2024	return 0;
2025	}
2026
2027	static int ctl_ioctl(struct file *file, uint command, struct dm_ioctl __user *user)
2028	{
2029	int r = 0;
2030	int ioctl_flags;
2031	int param_flags;
2032	unsigned int cmd;
2033	struct dm_ioctl *param;
2034	ioctl_fn fn = NULL;
2035	size_t input_param_size;
2036	struct dm_ioctl param_kernel;
2037
2038	/* only root can play with this */
2039	if (!capable(CAP_SYS_ADMIN))
2040	return -EACCES;
2041
2042	if (_IOC_TYPE(command) != DM_IOCTL)
2043	return -ENOTTY;
2044
2045	cmd = _IOC_NR(command);
2046
2047	/*
2048	* Check the interface version passed in. This also
2049	* writes out the kernel's interface version.
2050	*/
2051	r = check_version(cmd, user, kernel_params: &param_kernel);
2052	if (r)
2053	return r;
2054
2055	/*
2056	* Nothing more to do for the version command.
2057	*/
2058	if (cmd == DM_VERSION_CMD)
2059	return 0;
2060
2061	fn = lookup_ioctl(cmd, ioctl_flags: &ioctl_flags);
2062	if (!fn) {
2063	DMERR("dm_ctl_ioctl: unknown command 0x%x", command);
2064	return -ENOTTY;
2065	}
2066
2067	/*
2068	* Copy the parameters into kernel space.
2069	*/
2070	r = copy_params(user, param_kernel: &param_kernel, ioctl_flags, param: &param, param_flags: &param_flags);
2071
2072	if (r)
2073	return r;
2074
2075	input_param_size = param->data_size;
2076	r = validate_params(cmd, param);
2077	if (r)
2078	goto out;
2079
2080	param->data_size = offsetof(struct dm_ioctl, data);
2081	r = fn(file, param, input_param_size);
2082
2083	if (unlikely(param->flags & DM_BUFFER_FULL_FLAG) &&
2084	unlikely(ioctl_flags & IOCTL_FLAGS_NO_PARAMS))
2085	DMERR("ioctl %d tried to output some data but has IOCTL_FLAGS_NO_PARAMS set", cmd);
2086
2087	if (!r && ioctl_flags & IOCTL_FLAGS_ISSUE_GLOBAL_EVENT)
2088	dm_issue_global_event();
2089
2090	/*
2091	* Copy the results back to userland.
2092	*/
2093	if (!r && copy_to_user(to: user, from: param, n: param->data_size))
2094	r = -EFAULT;
2095
2096	out:
2097	free_params(param, param_size: input_param_size, param_flags);
2098	return r;
2099	}
2100
2101	static long dm_ctl_ioctl(struct file *file, uint command, ulong u)
2102	{
2103	return (long)ctl_ioctl(file, command, user: (struct dm_ioctl __user *)u);
2104	}
2105
2106	#ifdef CONFIG_COMPAT
2107	static long dm_compat_ctl_ioctl(struct file *file, uint command, ulong u)
2108	{
2109	return (long)dm_ctl_ioctl(file, command, u: (ulong) compat_ptr(uptr: u));
2110	}
2111	#else
2112	#define dm_compat_ctl_ioctl NULL
2113	#endif
2114
2115	static int dm_open(struct inode *inode, struct file *filp)
2116	{
2117	int r;
2118	struct dm_file *priv;
2119
2120	r = nonseekable_open(inode, filp);
2121	if (unlikely(r))
2122	return r;
2123
2124	priv = filp->private_data = kmalloc(size: sizeof(struct dm_file), GFP_KERNEL);
2125	if (!priv)
2126	return -ENOMEM;
2127
2128	priv->global_event_nr = atomic_read(v: &dm_global_event_nr);
2129
2130	return 0;
2131	}
2132
2133	static int dm_release(struct inode *inode, struct file *filp)
2134	{
2135	kfree(objp: filp->private_data);
2136	return 0;
2137	}
2138
2139	static __poll_t dm_poll(struct file *filp, poll_table *wait)
2140	{
2141	struct dm_file *priv = filp->private_data;
2142	__poll_t mask = 0;
2143
2144	poll_wait(filp, wait_address: &dm_global_eventq, p: wait);
2145
2146	if ((int)(atomic_read(v: &dm_global_event_nr) - priv->global_event_nr) > 0)
2147	mask |= EPOLLIN;
2148
2149	return mask;
2150	}
2151
2152	static const struct file_operations _ctl_fops = {
2153	.open = dm_open,
2154	.release = dm_release,
2155	.poll = dm_poll,
2156	.unlocked_ioctl = dm_ctl_ioctl,
2157	.compat_ioctl = dm_compat_ctl_ioctl,
2158	.owner = THIS_MODULE,
2159	.llseek = noop_llseek,
2160	};
2161
2162	static struct miscdevice _dm_misc = {
2163	.minor = MAPPER_CTRL_MINOR,
2164	.name = DM_NAME,
2165	.nodename = DM_DIR "/" DM_CONTROL_NODE,
2166	.fops = &_ctl_fops
2167	};
2168
2169	MODULE_ALIAS_MISCDEV(MAPPER_CTRL_MINOR);
2170	MODULE_ALIAS("devname:" DM_DIR "/" DM_CONTROL_NODE);
2171
2172	/*
2173	* Create misc character device and link to DM_DIR/control.
2174	*/
2175	int __init dm_interface_init(void)
2176	{
2177	int r;
2178
2179	r = misc_register(misc: &_dm_misc);
2180	if (r) {
2181	DMERR("misc_register failed for control device");
2182	return r;
2183	}
2184
2185	DMINFO("%d.%d.%d%s initialised: %s", DM_VERSION_MAJOR,
2186	DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL, DM_VERSION_EXTRA,
2187	DM_DRIVER_EMAIL);
2188	return 0;
2189	}
2190
2191	void dm_interface_exit(void)
2192	{
2193	misc_deregister(misc: &_dm_misc);
2194	dm_hash_exit();
2195	}
2196
2197	/**
2198	* dm_copy_name_and_uuid - Copy mapped device name & uuid into supplied buffers
2199	* @md: Pointer to mapped_device
2200	* @name: Buffer (size DM_NAME_LEN) for name
2201	* @uuid: Buffer (size DM_UUID_LEN) for uuid or empty string if uuid not defined
2202	*/
2203	int dm_copy_name_and_uuid(struct mapped_device *md, char *name, char *uuid)
2204	{
2205	int r = 0;
2206	struct hash_cell *hc;
2207
2208	if (!md)
2209	return -ENXIO;
2210
2211	mutex_lock(&dm_hash_cells_mutex);
2212	hc = dm_get_mdptr(md);
2213	if (!hc) {
2214	r = -ENXIO;
2215	goto out;
2216	}
2217
2218	if (name)
2219	strcpy(p: name, q: hc->name);
2220	if (uuid)
2221	strcpy(p: uuid, q: hc->uuid ? : "");
2222
2223	out:
2224	mutex_unlock(lock: &dm_hash_cells_mutex);
2225
2226	return r;
2227	}
2228	EXPORT_SYMBOL_GPL(dm_copy_name_and_uuid);
2229
2230	/**
2231	* dm_early_create - create a mapped device in early boot.
2232	*
2233	* @dmi: Contains main information of the device mapping to be created.
2234	* @spec_array: array of pointers to struct dm_target_spec. Describes the
2235	* mapping table of the device.
2236	* @target_params_array: array of strings with the parameters to a specific
2237	* target.
2238	*
2239	* Instead of having the struct dm_target_spec and the parameters for every
2240	* target embedded at the end of struct dm_ioctl (as performed in a normal
2241	* ioctl), pass them as arguments, so the caller doesn't need to serialize them.
2242	* The size of the spec_array and target_params_array is given by
2243	* @dmi->target_count.
2244	* This function is supposed to be called in early boot, so locking mechanisms
2245	* to protect against concurrent loads are not required.
2246	*/
2247	int __init dm_early_create(struct dm_ioctl *dmi,
2248	struct dm_target_spec **spec_array,
2249	char **target_params_array)
2250	{
2251	int r, m = DM_ANY_MINOR;
2252	struct dm_table *t, *old_map;
2253	struct mapped_device *md;
2254	unsigned int i;
2255
2256	if (!dmi->target_count)
2257	return -EINVAL;
2258
2259	r = check_name(name: dmi->name);
2260	if (r)
2261	return r;
2262
2263	if (dmi->flags & DM_PERSISTENT_DEV_FLAG)
2264	m = MINOR(huge_decode_dev(dmi->dev));
2265
2266	/* alloc dm device */
2267	r = dm_create(minor: m, md: &md);
2268	if (r)
2269	return r;
2270
2271	/* hash insert */
2272	r = dm_hash_insert(name: dmi->name, uuid: *dmi->uuid ? dmi->uuid : NULL, md);
2273	if (r)
2274	goto err_destroy_dm;
2275
2276	/* alloc table */
2277	r = dm_table_create(result: &t, mode: get_mode(param: dmi), num_targets: dmi->target_count, md);
2278	if (r)
2279	goto err_hash_remove;
2280
2281	/* add targets */
2282	for (i = 0; i < dmi->target_count; i++) {
2283	r = dm_table_add_target(t, type: spec_array[i]->target_type,
2284	start: (sector_t) spec_array[i]->sector_start,
2285	len: (sector_t) spec_array[i]->length,
2286	params: target_params_array[i]);
2287	if (r) {
2288	DMERR("error adding target to table");
2289	goto err_destroy_table;
2290	}
2291	}
2292
2293	/* finish table */
2294	r = dm_table_complete(t);
2295	if (r)
2296	goto err_destroy_table;
2297
2298	/* setup md->queue to reflect md's type (may block) */
2299	r = dm_setup_md_queue(md, t);
2300	if (r) {
2301	DMERR("unable to set up device queue for new table.");
2302	goto err_destroy_table;
2303	}
2304
2305	/* Set new map */
2306	dm_suspend(md, suspend_flags: 0);
2307	old_map = dm_swap_table(md, t);
2308	if (IS_ERR(ptr: old_map)) {
2309	r = PTR_ERR(ptr: old_map);
2310	goto err_destroy_table;
2311	}
2312	set_disk_ro(disk: dm_disk(md), read_only: !!(dmi->flags & DM_READONLY_FLAG));
2313
2314	/* resume device */
2315	r = dm_resume(md);
2316	if (r)
2317	goto err_destroy_table;
2318
2319	DMINFO("%s (%s) is ready", md->disk->disk_name, dmi->name);
2320	dm_put(md);
2321	return 0;
2322
2323	err_destroy_table:
2324	dm_table_destroy(t);
2325	err_hash_remove:
2326	down_write(sem: &_hash_lock);
2327	(void) __hash_remove(hc: __get_name_cell(str: dmi->name));
2328	up_write(sem: &_hash_lock);
2329	/* release reference from __get_name_cell */
2330	dm_put(md);
2331	err_destroy_dm:
2332	dm_put(md);
2333	dm_destroy(md);
2334	return r;
2335	}
2336