xref: /linux/net/rfkill/core.c (revision 5bdef865eb358b6f3760e25e591ae115e9eeddef)
1 /*
2  * Copyright (C) 2006 - 2007 Ivo van Doorn
3  * Copyright (C) 2007 Dmitry Torokhov
4  * Copyright 2009 Johannes Berg <johannes@sipsolutions.net>
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation; either version 2 of the License, or
9  * (at your option) any later version.
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the
18  * Free Software Foundation, Inc.,
19  * 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
20  */
21 
22 #include <linux/kernel.h>
23 #include <linux/module.h>
24 #include <linux/init.h>
25 #include <linux/workqueue.h>
26 #include <linux/capability.h>
27 #include <linux/list.h>
28 #include <linux/mutex.h>
29 #include <linux/rfkill.h>
30 #include <linux/spinlock.h>
31 #include <linux/miscdevice.h>
32 #include <linux/wait.h>
33 #include <linux/poll.h>
34 #include <linux/fs.h>
35 
36 #include "rfkill.h"
37 
38 #define POLL_INTERVAL		(5 * HZ)
39 
40 #define RFKILL_BLOCK_HW		BIT(0)
41 #define RFKILL_BLOCK_SW		BIT(1)
42 #define RFKILL_BLOCK_SW_PREV	BIT(2)
43 #define RFKILL_BLOCK_ANY	(RFKILL_BLOCK_HW |\
44 				 RFKILL_BLOCK_SW |\
45 				 RFKILL_BLOCK_SW_PREV)
46 #define RFKILL_BLOCK_SW_SETCALL	BIT(31)
47 
48 struct rfkill {
49 	spinlock_t		lock;
50 
51 	const char		*name;
52 	enum rfkill_type	type;
53 
54 	unsigned long		state;
55 
56 	u32			idx;
57 
58 	bool			registered;
59 	bool			persistent;
60 
61 	const struct rfkill_ops	*ops;
62 	void			*data;
63 
64 #ifdef CONFIG_RFKILL_LEDS
65 	struct led_trigger	led_trigger;
66 	const char		*ledtrigname;
67 #endif
68 
69 	struct device		dev;
70 	struct list_head	node;
71 
72 	struct delayed_work	poll_work;
73 	struct work_struct	uevent_work;
74 	struct work_struct	sync_work;
75 };
76 #define to_rfkill(d)	container_of(d, struct rfkill, dev)
77 
78 struct rfkill_int_event {
79 	struct list_head	list;
80 	struct rfkill_event	ev;
81 };
82 
83 struct rfkill_data {
84 	struct list_head	list;
85 	struct list_head	events;
86 	struct mutex		mtx;
87 	wait_queue_head_t	read_wait;
88 	bool			input_handler;
89 };
90 
91 
92 MODULE_AUTHOR("Ivo van Doorn <IvDoorn@gmail.com>");
93 MODULE_AUTHOR("Johannes Berg <johannes@sipsolutions.net>");
94 MODULE_DESCRIPTION("RF switch support");
95 MODULE_LICENSE("GPL");
96 
97 
98 /*
99  * The locking here should be made much smarter, we currently have
100  * a bit of a stupid situation because drivers might want to register
101  * the rfkill struct under their own lock, and take this lock during
102  * rfkill method calls -- which will cause an AB-BA deadlock situation.
103  *
104  * To fix that, we need to rework this code here to be mostly lock-free
105  * and only use the mutex for list manipulations, not to protect the
106  * various other global variables. Then we can avoid holding the mutex
107  * around driver operations, and all is happy.
108  */
109 static LIST_HEAD(rfkill_list);	/* list of registered rf switches */
110 static DEFINE_MUTEX(rfkill_global_mutex);
111 static LIST_HEAD(rfkill_fds);	/* list of open fds of /dev/rfkill */
112 
113 static unsigned int rfkill_default_state = 1;
114 module_param_named(default_state, rfkill_default_state, uint, 0444);
115 MODULE_PARM_DESC(default_state,
116 		 "Default initial state for all radio types, 0 = radio off");
117 
118 static struct {
119 	bool cur, sav;
120 } rfkill_global_states[NUM_RFKILL_TYPES];
121 
122 static bool rfkill_epo_lock_active;
123 
124 
125 #ifdef CONFIG_RFKILL_LEDS
126 static void rfkill_led_trigger_event(struct rfkill *rfkill)
127 {
128 	struct led_trigger *trigger;
129 
130 	if (!rfkill->registered)
131 		return;
132 
133 	trigger = &rfkill->led_trigger;
134 
135 	if (rfkill->state & RFKILL_BLOCK_ANY)
136 		led_trigger_event(trigger, LED_OFF);
137 	else
138 		led_trigger_event(trigger, LED_FULL);
139 }
140 
141 static void rfkill_led_trigger_activate(struct led_classdev *led)
142 {
143 	struct rfkill *rfkill;
144 
145 	rfkill = container_of(led->trigger, struct rfkill, led_trigger);
146 
147 	rfkill_led_trigger_event(rfkill);
148 }
149 
150 const char *rfkill_get_led_trigger_name(struct rfkill *rfkill)
151 {
152 	return rfkill->led_trigger.name;
153 }
154 EXPORT_SYMBOL(rfkill_get_led_trigger_name);
155 
156 void rfkill_set_led_trigger_name(struct rfkill *rfkill, const char *name)
157 {
158 	BUG_ON(!rfkill);
159 
160 	rfkill->ledtrigname = name;
161 }
162 EXPORT_SYMBOL(rfkill_set_led_trigger_name);
163 
164 static int rfkill_led_trigger_register(struct rfkill *rfkill)
165 {
166 	rfkill->led_trigger.name = rfkill->ledtrigname
167 					? : dev_name(&rfkill->dev);
168 	rfkill->led_trigger.activate = rfkill_led_trigger_activate;
169 	return led_trigger_register(&rfkill->led_trigger);
170 }
171 
172 static void rfkill_led_trigger_unregister(struct rfkill *rfkill)
173 {
174 	led_trigger_unregister(&rfkill->led_trigger);
175 }
176 #else
177 static void rfkill_led_trigger_event(struct rfkill *rfkill)
178 {
179 }
180 
181 static inline int rfkill_led_trigger_register(struct rfkill *rfkill)
182 {
183 	return 0;
184 }
185 
186 static inline void rfkill_led_trigger_unregister(struct rfkill *rfkill)
187 {
188 }
189 #endif /* CONFIG_RFKILL_LEDS */
190 
191 static void rfkill_fill_event(struct rfkill_event *ev, struct rfkill *rfkill,
192 			      enum rfkill_operation op)
193 {
194 	unsigned long flags;
195 
196 	ev->idx = rfkill->idx;
197 	ev->type = rfkill->type;
198 	ev->op = op;
199 
200 	spin_lock_irqsave(&rfkill->lock, flags);
201 	ev->hard = !!(rfkill->state & RFKILL_BLOCK_HW);
202 	ev->soft = !!(rfkill->state & (RFKILL_BLOCK_SW |
203 					RFKILL_BLOCK_SW_PREV));
204 	spin_unlock_irqrestore(&rfkill->lock, flags);
205 }
206 
207 static void rfkill_send_events(struct rfkill *rfkill, enum rfkill_operation op)
208 {
209 	struct rfkill_data *data;
210 	struct rfkill_int_event *ev;
211 
212 	list_for_each_entry(data, &rfkill_fds, list) {
213 		ev = kzalloc(sizeof(*ev), GFP_KERNEL);
214 		if (!ev)
215 			continue;
216 		rfkill_fill_event(&ev->ev, rfkill, op);
217 		mutex_lock(&data->mtx);
218 		list_add_tail(&ev->list, &data->events);
219 		mutex_unlock(&data->mtx);
220 		wake_up_interruptible(&data->read_wait);
221 	}
222 }
223 
224 static void rfkill_event(struct rfkill *rfkill)
225 {
226 	if (!rfkill->registered)
227 		return;
228 
229 	kobject_uevent(&rfkill->dev.kobj, KOBJ_CHANGE);
230 
231 	/* also send event to /dev/rfkill */
232 	rfkill_send_events(rfkill, RFKILL_OP_CHANGE);
233 }
234 
235 static bool __rfkill_set_hw_state(struct rfkill *rfkill,
236 				  bool blocked, bool *change)
237 {
238 	unsigned long flags;
239 	bool prev, any;
240 
241 	BUG_ON(!rfkill);
242 
243 	spin_lock_irqsave(&rfkill->lock, flags);
244 	prev = !!(rfkill->state & RFKILL_BLOCK_HW);
245 	if (blocked)
246 		rfkill->state |= RFKILL_BLOCK_HW;
247 	else
248 		rfkill->state &= ~RFKILL_BLOCK_HW;
249 	*change = prev != blocked;
250 	any = rfkill->state & RFKILL_BLOCK_ANY;
251 	spin_unlock_irqrestore(&rfkill->lock, flags);
252 
253 	rfkill_led_trigger_event(rfkill);
254 
255 	return any;
256 }
257 
258 /**
259  * rfkill_set_block - wrapper for set_block method
260  *
261  * @rfkill: the rfkill struct to use
262  * @blocked: the new software state
263  *
264  * Calls the set_block method (when applicable) and handles notifications
265  * etc. as well.
266  */
267 static void rfkill_set_block(struct rfkill *rfkill, bool blocked)
268 {
269 	unsigned long flags;
270 	int err;
271 
272 	if (unlikely(rfkill->dev.power.power_state.event & PM_EVENT_SLEEP))
273 		return;
274 
275 	/*
276 	 * Some platforms (...!) generate input events which affect the
277 	 * _hard_ kill state -- whenever something tries to change the
278 	 * current software state query the hardware state too.
279 	 */
280 	if (rfkill->ops->query)
281 		rfkill->ops->query(rfkill, rfkill->data);
282 
283 	spin_lock_irqsave(&rfkill->lock, flags);
284 	if (rfkill->state & RFKILL_BLOCK_SW)
285 		rfkill->state |= RFKILL_BLOCK_SW_PREV;
286 	else
287 		rfkill->state &= ~RFKILL_BLOCK_SW_PREV;
288 
289 	if (blocked)
290 		rfkill->state |= RFKILL_BLOCK_SW;
291 	else
292 		rfkill->state &= ~RFKILL_BLOCK_SW;
293 
294 	rfkill->state |= RFKILL_BLOCK_SW_SETCALL;
295 	spin_unlock_irqrestore(&rfkill->lock, flags);
296 
297 	err = rfkill->ops->set_block(rfkill->data, blocked);
298 
299 	spin_lock_irqsave(&rfkill->lock, flags);
300 	if (err) {
301 		/*
302 		 * Failed -- reset status to _prev, this may be different
303 		 * from what set set _PREV to earlier in this function
304 		 * if rfkill_set_sw_state was invoked.
305 		 */
306 		if (rfkill->state & RFKILL_BLOCK_SW_PREV)
307 			rfkill->state |= RFKILL_BLOCK_SW;
308 		else
309 			rfkill->state &= ~RFKILL_BLOCK_SW;
310 	}
311 	rfkill->state &= ~RFKILL_BLOCK_SW_SETCALL;
312 	rfkill->state &= ~RFKILL_BLOCK_SW_PREV;
313 	spin_unlock_irqrestore(&rfkill->lock, flags);
314 
315 	rfkill_led_trigger_event(rfkill);
316 	rfkill_event(rfkill);
317 }
318 
319 #ifdef CONFIG_RFKILL_INPUT
320 static atomic_t rfkill_input_disabled = ATOMIC_INIT(0);
321 
322 /**
323  * __rfkill_switch_all - Toggle state of all switches of given type
324  * @type: type of interfaces to be affected
325  * @state: the new state
326  *
327  * This function sets the state of all switches of given type,
328  * unless a specific switch is claimed by userspace (in which case,
329  * that switch is left alone) or suspended.
330  *
331  * Caller must have acquired rfkill_global_mutex.
332  */
333 static void __rfkill_switch_all(const enum rfkill_type type, bool blocked)
334 {
335 	struct rfkill *rfkill;
336 
337 	rfkill_global_states[type].cur = blocked;
338 	list_for_each_entry(rfkill, &rfkill_list, node) {
339 		if (rfkill->type != type)
340 			continue;
341 
342 		rfkill_set_block(rfkill, blocked);
343 	}
344 }
345 
346 /**
347  * rfkill_switch_all - Toggle state of all switches of given type
348  * @type: type of interfaces to be affected
349  * @state: the new state
350  *
351  * Acquires rfkill_global_mutex and calls __rfkill_switch_all(@type, @state).
352  * Please refer to __rfkill_switch_all() for details.
353  *
354  * Does nothing if the EPO lock is active.
355  */
356 void rfkill_switch_all(enum rfkill_type type, bool blocked)
357 {
358 	if (atomic_read(&rfkill_input_disabled))
359 		return;
360 
361 	mutex_lock(&rfkill_global_mutex);
362 
363 	if (!rfkill_epo_lock_active)
364 		__rfkill_switch_all(type, blocked);
365 
366 	mutex_unlock(&rfkill_global_mutex);
367 }
368 
369 /**
370  * rfkill_epo - emergency power off all transmitters
371  *
372  * This kicks all non-suspended rfkill devices to RFKILL_STATE_SOFT_BLOCKED,
373  * ignoring everything in its path but rfkill_global_mutex and rfkill->mutex.
374  *
375  * The global state before the EPO is saved and can be restored later
376  * using rfkill_restore_states().
377  */
378 void rfkill_epo(void)
379 {
380 	struct rfkill *rfkill;
381 	int i;
382 
383 	if (atomic_read(&rfkill_input_disabled))
384 		return;
385 
386 	mutex_lock(&rfkill_global_mutex);
387 
388 	rfkill_epo_lock_active = true;
389 	list_for_each_entry(rfkill, &rfkill_list, node)
390 		rfkill_set_block(rfkill, true);
391 
392 	for (i = 0; i < NUM_RFKILL_TYPES; i++) {
393 		rfkill_global_states[i].sav = rfkill_global_states[i].cur;
394 		rfkill_global_states[i].cur = true;
395 	}
396 
397 	mutex_unlock(&rfkill_global_mutex);
398 }
399 
400 /**
401  * rfkill_restore_states - restore global states
402  *
403  * Restore (and sync switches to) the global state from the
404  * states in rfkill_default_states.  This can undo the effects of
405  * a call to rfkill_epo().
406  */
407 void rfkill_restore_states(void)
408 {
409 	int i;
410 
411 	if (atomic_read(&rfkill_input_disabled))
412 		return;
413 
414 	mutex_lock(&rfkill_global_mutex);
415 
416 	rfkill_epo_lock_active = false;
417 	for (i = 0; i < NUM_RFKILL_TYPES; i++)
418 		__rfkill_switch_all(i, rfkill_global_states[i].sav);
419 	mutex_unlock(&rfkill_global_mutex);
420 }
421 
422 /**
423  * rfkill_remove_epo_lock - unlock state changes
424  *
425  * Used by rfkill-input manually unlock state changes, when
426  * the EPO switch is deactivated.
427  */
428 void rfkill_remove_epo_lock(void)
429 {
430 	if (atomic_read(&rfkill_input_disabled))
431 		return;
432 
433 	mutex_lock(&rfkill_global_mutex);
434 	rfkill_epo_lock_active = false;
435 	mutex_unlock(&rfkill_global_mutex);
436 }
437 
438 /**
439  * rfkill_is_epo_lock_active - returns true EPO is active
440  *
441  * Returns 0 (false) if there is NOT an active EPO contidion,
442  * and 1 (true) if there is an active EPO contition, which
443  * locks all radios in one of the BLOCKED states.
444  *
445  * Can be called in atomic context.
446  */
447 bool rfkill_is_epo_lock_active(void)
448 {
449 	return rfkill_epo_lock_active;
450 }
451 
452 /**
453  * rfkill_get_global_sw_state - returns global state for a type
454  * @type: the type to get the global state of
455  *
456  * Returns the current global state for a given wireless
457  * device type.
458  */
459 bool rfkill_get_global_sw_state(const enum rfkill_type type)
460 {
461 	return rfkill_global_states[type].cur;
462 }
463 #endif
464 
465 
466 bool rfkill_set_hw_state(struct rfkill *rfkill, bool blocked)
467 {
468 	bool ret, change;
469 
470 	ret = __rfkill_set_hw_state(rfkill, blocked, &change);
471 
472 	if (!rfkill->registered)
473 		return ret;
474 
475 	if (change)
476 		schedule_work(&rfkill->uevent_work);
477 
478 	return ret;
479 }
480 EXPORT_SYMBOL(rfkill_set_hw_state);
481 
482 static void __rfkill_set_sw_state(struct rfkill *rfkill, bool blocked)
483 {
484 	u32 bit = RFKILL_BLOCK_SW;
485 
486 	/* if in a ops->set_block right now, use other bit */
487 	if (rfkill->state & RFKILL_BLOCK_SW_SETCALL)
488 		bit = RFKILL_BLOCK_SW_PREV;
489 
490 	if (blocked)
491 		rfkill->state |= bit;
492 	else
493 		rfkill->state &= ~bit;
494 }
495 
496 bool rfkill_set_sw_state(struct rfkill *rfkill, bool blocked)
497 {
498 	unsigned long flags;
499 	bool prev, hwblock;
500 
501 	BUG_ON(!rfkill);
502 
503 	spin_lock_irqsave(&rfkill->lock, flags);
504 	prev = !!(rfkill->state & RFKILL_BLOCK_SW);
505 	__rfkill_set_sw_state(rfkill, blocked);
506 	hwblock = !!(rfkill->state & RFKILL_BLOCK_HW);
507 	blocked = blocked || hwblock;
508 	spin_unlock_irqrestore(&rfkill->lock, flags);
509 
510 	if (!rfkill->registered)
511 		return blocked;
512 
513 	if (prev != blocked && !hwblock)
514 		schedule_work(&rfkill->uevent_work);
515 
516 	rfkill_led_trigger_event(rfkill);
517 
518 	return blocked;
519 }
520 EXPORT_SYMBOL(rfkill_set_sw_state);
521 
522 void rfkill_init_sw_state(struct rfkill *rfkill, bool blocked)
523 {
524 	unsigned long flags;
525 
526 	BUG_ON(!rfkill);
527 	BUG_ON(rfkill->registered);
528 
529 	spin_lock_irqsave(&rfkill->lock, flags);
530 	__rfkill_set_sw_state(rfkill, blocked);
531 	rfkill->persistent = true;
532 	spin_unlock_irqrestore(&rfkill->lock, flags);
533 }
534 EXPORT_SYMBOL(rfkill_init_sw_state);
535 
536 void rfkill_set_states(struct rfkill *rfkill, bool sw, bool hw)
537 {
538 	unsigned long flags;
539 	bool swprev, hwprev;
540 
541 	BUG_ON(!rfkill);
542 
543 	spin_lock_irqsave(&rfkill->lock, flags);
544 
545 	/*
546 	 * No need to care about prev/setblock ... this is for uevent only
547 	 * and that will get triggered by rfkill_set_block anyway.
548 	 */
549 	swprev = !!(rfkill->state & RFKILL_BLOCK_SW);
550 	hwprev = !!(rfkill->state & RFKILL_BLOCK_HW);
551 	__rfkill_set_sw_state(rfkill, sw);
552 
553 	spin_unlock_irqrestore(&rfkill->lock, flags);
554 
555 	if (!rfkill->registered) {
556 		rfkill->persistent = true;
557 	} else {
558 		if (swprev != sw || hwprev != hw)
559 			schedule_work(&rfkill->uevent_work);
560 
561 		rfkill_led_trigger_event(rfkill);
562 	}
563 }
564 EXPORT_SYMBOL(rfkill_set_states);
565 
566 static ssize_t rfkill_name_show(struct device *dev,
567 				struct device_attribute *attr,
568 				char *buf)
569 {
570 	struct rfkill *rfkill = to_rfkill(dev);
571 
572 	return sprintf(buf, "%s\n", rfkill->name);
573 }
574 
575 static const char *rfkill_get_type_str(enum rfkill_type type)
576 {
577 	switch (type) {
578 	case RFKILL_TYPE_WLAN:
579 		return "wlan";
580 	case RFKILL_TYPE_BLUETOOTH:
581 		return "bluetooth";
582 	case RFKILL_TYPE_UWB:
583 		return "ultrawideband";
584 	case RFKILL_TYPE_WIMAX:
585 		return "wimax";
586 	case RFKILL_TYPE_WWAN:
587 		return "wwan";
588 	default:
589 		BUG();
590 	}
591 
592 	BUILD_BUG_ON(NUM_RFKILL_TYPES != RFKILL_TYPE_WWAN + 1);
593 }
594 
595 static ssize_t rfkill_type_show(struct device *dev,
596 				struct device_attribute *attr,
597 				char *buf)
598 {
599 	struct rfkill *rfkill = to_rfkill(dev);
600 
601 	return sprintf(buf, "%s\n", rfkill_get_type_str(rfkill->type));
602 }
603 
604 static ssize_t rfkill_idx_show(struct device *dev,
605 			       struct device_attribute *attr,
606 			       char *buf)
607 {
608 	struct rfkill *rfkill = to_rfkill(dev);
609 
610 	return sprintf(buf, "%d\n", rfkill->idx);
611 }
612 
613 static ssize_t rfkill_persistent_show(struct device *dev,
614 			       struct device_attribute *attr,
615 			       char *buf)
616 {
617 	struct rfkill *rfkill = to_rfkill(dev);
618 
619 	return sprintf(buf, "%d\n", rfkill->persistent);
620 }
621 
622 static u8 user_state_from_blocked(unsigned long state)
623 {
624 	if (state & RFKILL_BLOCK_HW)
625 		return RFKILL_USER_STATE_HARD_BLOCKED;
626 	if (state & RFKILL_BLOCK_SW)
627 		return RFKILL_USER_STATE_SOFT_BLOCKED;
628 
629 	return RFKILL_USER_STATE_UNBLOCKED;
630 }
631 
632 static ssize_t rfkill_state_show(struct device *dev,
633 				 struct device_attribute *attr,
634 				 char *buf)
635 {
636 	struct rfkill *rfkill = to_rfkill(dev);
637 	unsigned long flags;
638 	u32 state;
639 
640 	spin_lock_irqsave(&rfkill->lock, flags);
641 	state = rfkill->state;
642 	spin_unlock_irqrestore(&rfkill->lock, flags);
643 
644 	return sprintf(buf, "%d\n", user_state_from_blocked(state));
645 }
646 
647 static ssize_t rfkill_state_store(struct device *dev,
648 				  struct device_attribute *attr,
649 				  const char *buf, size_t count)
650 {
651 	/*
652 	 * The intention was that userspace can only take control over
653 	 * a given device when/if rfkill-input doesn't control it due
654 	 * to user_claim. Since user_claim is currently unsupported,
655 	 * we never support changing the state from userspace -- this
656 	 * can be implemented again later.
657 	 */
658 
659 	return -EPERM;
660 }
661 
662 static ssize_t rfkill_claim_show(struct device *dev,
663 				 struct device_attribute *attr,
664 				 char *buf)
665 {
666 	return sprintf(buf, "%d\n", 0);
667 }
668 
669 static ssize_t rfkill_claim_store(struct device *dev,
670 				  struct device_attribute *attr,
671 				  const char *buf, size_t count)
672 {
673 	return -EOPNOTSUPP;
674 }
675 
676 static struct device_attribute rfkill_dev_attrs[] = {
677 	__ATTR(name, S_IRUGO, rfkill_name_show, NULL),
678 	__ATTR(type, S_IRUGO, rfkill_type_show, NULL),
679 	__ATTR(index, S_IRUGO, rfkill_idx_show, NULL),
680 	__ATTR(persistent, S_IRUGO, rfkill_persistent_show, NULL),
681 	__ATTR(state, S_IRUGO|S_IWUSR, rfkill_state_show, rfkill_state_store),
682 	__ATTR(claim, S_IRUGO|S_IWUSR, rfkill_claim_show, rfkill_claim_store),
683 	__ATTR_NULL
684 };
685 
686 static void rfkill_release(struct device *dev)
687 {
688 	struct rfkill *rfkill = to_rfkill(dev);
689 
690 	kfree(rfkill);
691 }
692 
693 static int rfkill_dev_uevent(struct device *dev, struct kobj_uevent_env *env)
694 {
695 	struct rfkill *rfkill = to_rfkill(dev);
696 	unsigned long flags;
697 	u32 state;
698 	int error;
699 
700 	error = add_uevent_var(env, "RFKILL_NAME=%s", rfkill->name);
701 	if (error)
702 		return error;
703 	error = add_uevent_var(env, "RFKILL_TYPE=%s",
704 			       rfkill_get_type_str(rfkill->type));
705 	if (error)
706 		return error;
707 	spin_lock_irqsave(&rfkill->lock, flags);
708 	state = rfkill->state;
709 	spin_unlock_irqrestore(&rfkill->lock, flags);
710 	error = add_uevent_var(env, "RFKILL_STATE=%d",
711 			       user_state_from_blocked(state));
712 	return error;
713 }
714 
715 void rfkill_pause_polling(struct rfkill *rfkill)
716 {
717 	BUG_ON(!rfkill);
718 
719 	if (!rfkill->ops->poll)
720 		return;
721 
722 	cancel_delayed_work_sync(&rfkill->poll_work);
723 }
724 EXPORT_SYMBOL(rfkill_pause_polling);
725 
726 void rfkill_resume_polling(struct rfkill *rfkill)
727 {
728 	BUG_ON(!rfkill);
729 
730 	if (!rfkill->ops->poll)
731 		return;
732 
733 	schedule_work(&rfkill->poll_work.work);
734 }
735 EXPORT_SYMBOL(rfkill_resume_polling);
736 
737 static int rfkill_suspend(struct device *dev, pm_message_t state)
738 {
739 	struct rfkill *rfkill = to_rfkill(dev);
740 
741 	rfkill_pause_polling(rfkill);
742 
743 	return 0;
744 }
745 
746 static int rfkill_resume(struct device *dev)
747 {
748 	struct rfkill *rfkill = to_rfkill(dev);
749 	bool cur;
750 
751 	if (!rfkill->persistent) {
752 		cur = !!(rfkill->state & RFKILL_BLOCK_SW);
753 		rfkill_set_block(rfkill, cur);
754 	}
755 
756 	rfkill_resume_polling(rfkill);
757 
758 	return 0;
759 }
760 
761 static struct class rfkill_class = {
762 	.name		= "rfkill",
763 	.dev_release	= rfkill_release,
764 	.dev_attrs	= rfkill_dev_attrs,
765 	.dev_uevent	= rfkill_dev_uevent,
766 	.suspend	= rfkill_suspend,
767 	.resume		= rfkill_resume,
768 };
769 
770 bool rfkill_blocked(struct rfkill *rfkill)
771 {
772 	unsigned long flags;
773 	u32 state;
774 
775 	spin_lock_irqsave(&rfkill->lock, flags);
776 	state = rfkill->state;
777 	spin_unlock_irqrestore(&rfkill->lock, flags);
778 
779 	return !!(state & RFKILL_BLOCK_ANY);
780 }
781 EXPORT_SYMBOL(rfkill_blocked);
782 
783 
784 struct rfkill * __must_check rfkill_alloc(const char *name,
785 					  struct device *parent,
786 					  const enum rfkill_type type,
787 					  const struct rfkill_ops *ops,
788 					  void *ops_data)
789 {
790 	struct rfkill *rfkill;
791 	struct device *dev;
792 
793 	if (WARN_ON(!ops))
794 		return NULL;
795 
796 	if (WARN_ON(!ops->set_block))
797 		return NULL;
798 
799 	if (WARN_ON(!name))
800 		return NULL;
801 
802 	if (WARN_ON(type == RFKILL_TYPE_ALL || type >= NUM_RFKILL_TYPES))
803 		return NULL;
804 
805 	rfkill = kzalloc(sizeof(*rfkill), GFP_KERNEL);
806 	if (!rfkill)
807 		return NULL;
808 
809 	spin_lock_init(&rfkill->lock);
810 	INIT_LIST_HEAD(&rfkill->node);
811 	rfkill->type = type;
812 	rfkill->name = name;
813 	rfkill->ops = ops;
814 	rfkill->data = ops_data;
815 
816 	dev = &rfkill->dev;
817 	dev->class = &rfkill_class;
818 	dev->parent = parent;
819 	device_initialize(dev);
820 
821 	return rfkill;
822 }
823 EXPORT_SYMBOL(rfkill_alloc);
824 
825 static void rfkill_poll(struct work_struct *work)
826 {
827 	struct rfkill *rfkill;
828 
829 	rfkill = container_of(work, struct rfkill, poll_work.work);
830 
831 	/*
832 	 * Poll hardware state -- driver will use one of the
833 	 * rfkill_set{,_hw,_sw}_state functions and use its
834 	 * return value to update the current status.
835 	 */
836 	rfkill->ops->poll(rfkill, rfkill->data);
837 
838 	schedule_delayed_work(&rfkill->poll_work,
839 		round_jiffies_relative(POLL_INTERVAL));
840 }
841 
842 static void rfkill_uevent_work(struct work_struct *work)
843 {
844 	struct rfkill *rfkill;
845 
846 	rfkill = container_of(work, struct rfkill, uevent_work);
847 
848 	mutex_lock(&rfkill_global_mutex);
849 	rfkill_event(rfkill);
850 	mutex_unlock(&rfkill_global_mutex);
851 }
852 
853 static void rfkill_sync_work(struct work_struct *work)
854 {
855 	struct rfkill *rfkill;
856 	bool cur;
857 
858 	rfkill = container_of(work, struct rfkill, sync_work);
859 
860 	mutex_lock(&rfkill_global_mutex);
861 	cur = rfkill_global_states[rfkill->type].cur;
862 	rfkill_set_block(rfkill, cur);
863 	mutex_unlock(&rfkill_global_mutex);
864 }
865 
866 int __must_check rfkill_register(struct rfkill *rfkill)
867 {
868 	static unsigned long rfkill_no;
869 	struct device *dev = &rfkill->dev;
870 	int error;
871 
872 	BUG_ON(!rfkill);
873 
874 	mutex_lock(&rfkill_global_mutex);
875 
876 	if (rfkill->registered) {
877 		error = -EALREADY;
878 		goto unlock;
879 	}
880 
881 	rfkill->idx = rfkill_no;
882 	dev_set_name(dev, "rfkill%lu", rfkill_no);
883 	rfkill_no++;
884 
885 	list_add_tail(&rfkill->node, &rfkill_list);
886 
887 	error = device_add(dev);
888 	if (error)
889 		goto remove;
890 
891 	error = rfkill_led_trigger_register(rfkill);
892 	if (error)
893 		goto devdel;
894 
895 	rfkill->registered = true;
896 
897 	INIT_DELAYED_WORK(&rfkill->poll_work, rfkill_poll);
898 	INIT_WORK(&rfkill->uevent_work, rfkill_uevent_work);
899 	INIT_WORK(&rfkill->sync_work, rfkill_sync_work);
900 
901 	if (rfkill->ops->poll)
902 		schedule_delayed_work(&rfkill->poll_work,
903 			round_jiffies_relative(POLL_INTERVAL));
904 
905 	if (!rfkill->persistent || rfkill_epo_lock_active) {
906 		schedule_work(&rfkill->sync_work);
907 	} else {
908 #ifdef CONFIG_RFKILL_INPUT
909 		bool soft_blocked = !!(rfkill->state & RFKILL_BLOCK_SW);
910 
911 		if (!atomic_read(&rfkill_input_disabled))
912 			__rfkill_switch_all(rfkill->type, soft_blocked);
913 #endif
914 	}
915 
916 	rfkill_send_events(rfkill, RFKILL_OP_ADD);
917 
918 	mutex_unlock(&rfkill_global_mutex);
919 	return 0;
920 
921  devdel:
922 	device_del(&rfkill->dev);
923  remove:
924 	list_del_init(&rfkill->node);
925  unlock:
926 	mutex_unlock(&rfkill_global_mutex);
927 	return error;
928 }
929 EXPORT_SYMBOL(rfkill_register);
930 
931 void rfkill_unregister(struct rfkill *rfkill)
932 {
933 	BUG_ON(!rfkill);
934 
935 	if (rfkill->ops->poll)
936 		cancel_delayed_work_sync(&rfkill->poll_work);
937 
938 	cancel_work_sync(&rfkill->uevent_work);
939 	cancel_work_sync(&rfkill->sync_work);
940 
941 	rfkill->registered = false;
942 
943 	device_del(&rfkill->dev);
944 
945 	mutex_lock(&rfkill_global_mutex);
946 	rfkill_send_events(rfkill, RFKILL_OP_DEL);
947 	list_del_init(&rfkill->node);
948 	mutex_unlock(&rfkill_global_mutex);
949 
950 	rfkill_led_trigger_unregister(rfkill);
951 }
952 EXPORT_SYMBOL(rfkill_unregister);
953 
954 void rfkill_destroy(struct rfkill *rfkill)
955 {
956 	if (rfkill)
957 		put_device(&rfkill->dev);
958 }
959 EXPORT_SYMBOL(rfkill_destroy);
960 
961 static int rfkill_fop_open(struct inode *inode, struct file *file)
962 {
963 	struct rfkill_data *data;
964 	struct rfkill *rfkill;
965 	struct rfkill_int_event *ev, *tmp;
966 
967 	data = kzalloc(sizeof(*data), GFP_KERNEL);
968 	if (!data)
969 		return -ENOMEM;
970 
971 	INIT_LIST_HEAD(&data->events);
972 	mutex_init(&data->mtx);
973 	init_waitqueue_head(&data->read_wait);
974 
975 	mutex_lock(&rfkill_global_mutex);
976 	mutex_lock(&data->mtx);
977 	/*
978 	 * start getting events from elsewhere but hold mtx to get
979 	 * startup events added first
980 	 */
981 	list_add(&data->list, &rfkill_fds);
982 
983 	list_for_each_entry(rfkill, &rfkill_list, node) {
984 		ev = kzalloc(sizeof(*ev), GFP_KERNEL);
985 		if (!ev)
986 			goto free;
987 		rfkill_fill_event(&ev->ev, rfkill, RFKILL_OP_ADD);
988 		list_add_tail(&ev->list, &data->events);
989 	}
990 	mutex_unlock(&data->mtx);
991 	mutex_unlock(&rfkill_global_mutex);
992 
993 	file->private_data = data;
994 
995 	return nonseekable_open(inode, file);
996 
997  free:
998 	mutex_unlock(&data->mtx);
999 	mutex_unlock(&rfkill_global_mutex);
1000 	mutex_destroy(&data->mtx);
1001 	list_for_each_entry_safe(ev, tmp, &data->events, list)
1002 		kfree(ev);
1003 	kfree(data);
1004 	return -ENOMEM;
1005 }
1006 
1007 static unsigned int rfkill_fop_poll(struct file *file, poll_table *wait)
1008 {
1009 	struct rfkill_data *data = file->private_data;
1010 	unsigned int res = POLLOUT | POLLWRNORM;
1011 
1012 	poll_wait(file, &data->read_wait, wait);
1013 
1014 	mutex_lock(&data->mtx);
1015 	if (!list_empty(&data->events))
1016 		res = POLLIN | POLLRDNORM;
1017 	mutex_unlock(&data->mtx);
1018 
1019 	return res;
1020 }
1021 
1022 static bool rfkill_readable(struct rfkill_data *data)
1023 {
1024 	bool r;
1025 
1026 	mutex_lock(&data->mtx);
1027 	r = !list_empty(&data->events);
1028 	mutex_unlock(&data->mtx);
1029 
1030 	return r;
1031 }
1032 
1033 static ssize_t rfkill_fop_read(struct file *file, char __user *buf,
1034 			       size_t count, loff_t *pos)
1035 {
1036 	struct rfkill_data *data = file->private_data;
1037 	struct rfkill_int_event *ev;
1038 	unsigned long sz;
1039 	int ret;
1040 
1041 	mutex_lock(&data->mtx);
1042 
1043 	while (list_empty(&data->events)) {
1044 		if (file->f_flags & O_NONBLOCK) {
1045 			ret = -EAGAIN;
1046 			goto out;
1047 		}
1048 		mutex_unlock(&data->mtx);
1049 		ret = wait_event_interruptible(data->read_wait,
1050 					       rfkill_readable(data));
1051 		mutex_lock(&data->mtx);
1052 
1053 		if (ret)
1054 			goto out;
1055 	}
1056 
1057 	ev = list_first_entry(&data->events, struct rfkill_int_event,
1058 				list);
1059 
1060 	sz = min_t(unsigned long, sizeof(ev->ev), count);
1061 	ret = sz;
1062 	if (copy_to_user(buf, &ev->ev, sz))
1063 		ret = -EFAULT;
1064 
1065 	list_del(&ev->list);
1066 	kfree(ev);
1067  out:
1068 	mutex_unlock(&data->mtx);
1069 	return ret;
1070 }
1071 
1072 static ssize_t rfkill_fop_write(struct file *file, const char __user *buf,
1073 				size_t count, loff_t *pos)
1074 {
1075 	struct rfkill *rfkill;
1076 	struct rfkill_event ev;
1077 
1078 	/* we don't need the 'hard' variable but accept it */
1079 	if (count < sizeof(ev) - 1)
1080 		return -EINVAL;
1081 
1082 	if (copy_from_user(&ev, buf, sizeof(ev) - 1))
1083 		return -EFAULT;
1084 
1085 	if (ev.op != RFKILL_OP_CHANGE && ev.op != RFKILL_OP_CHANGE_ALL)
1086 		return -EINVAL;
1087 
1088 	if (ev.type >= NUM_RFKILL_TYPES)
1089 		return -EINVAL;
1090 
1091 	mutex_lock(&rfkill_global_mutex);
1092 
1093 	if (ev.op == RFKILL_OP_CHANGE_ALL) {
1094 		if (ev.type == RFKILL_TYPE_ALL) {
1095 			enum rfkill_type i;
1096 			for (i = 0; i < NUM_RFKILL_TYPES; i++)
1097 				rfkill_global_states[i].cur = ev.soft;
1098 		} else {
1099 			rfkill_global_states[ev.type].cur = ev.soft;
1100 		}
1101 	}
1102 
1103 	list_for_each_entry(rfkill, &rfkill_list, node) {
1104 		if (rfkill->idx != ev.idx && ev.op != RFKILL_OP_CHANGE_ALL)
1105 			continue;
1106 
1107 		if (rfkill->type != ev.type && ev.type != RFKILL_TYPE_ALL)
1108 			continue;
1109 
1110 		rfkill_set_block(rfkill, ev.soft);
1111 	}
1112 	mutex_unlock(&rfkill_global_mutex);
1113 
1114 	return count;
1115 }
1116 
1117 static int rfkill_fop_release(struct inode *inode, struct file *file)
1118 {
1119 	struct rfkill_data *data = file->private_data;
1120 	struct rfkill_int_event *ev, *tmp;
1121 
1122 	mutex_lock(&rfkill_global_mutex);
1123 	list_del(&data->list);
1124 	mutex_unlock(&rfkill_global_mutex);
1125 
1126 	mutex_destroy(&data->mtx);
1127 	list_for_each_entry_safe(ev, tmp, &data->events, list)
1128 		kfree(ev);
1129 
1130 #ifdef CONFIG_RFKILL_INPUT
1131 	if (data->input_handler)
1132 		if (atomic_dec_return(&rfkill_input_disabled) == 0)
1133 			printk(KERN_DEBUG "rfkill: input handler enabled\n");
1134 #endif
1135 
1136 	kfree(data);
1137 
1138 	return 0;
1139 }
1140 
1141 #ifdef CONFIG_RFKILL_INPUT
1142 static long rfkill_fop_ioctl(struct file *file, unsigned int cmd,
1143 			     unsigned long arg)
1144 {
1145 	struct rfkill_data *data = file->private_data;
1146 
1147 	if (_IOC_TYPE(cmd) != RFKILL_IOC_MAGIC)
1148 		return -ENOSYS;
1149 
1150 	if (_IOC_NR(cmd) != RFKILL_IOC_NOINPUT)
1151 		return -ENOSYS;
1152 
1153 	mutex_lock(&data->mtx);
1154 
1155 	if (!data->input_handler) {
1156 		if (atomic_inc_return(&rfkill_input_disabled) == 1)
1157 			printk(KERN_DEBUG "rfkill: input handler disabled\n");
1158 		data->input_handler = true;
1159 	}
1160 
1161 	mutex_unlock(&data->mtx);
1162 
1163 	return 0;
1164 }
1165 #endif
1166 
1167 static const struct file_operations rfkill_fops = {
1168 	.open		= rfkill_fop_open,
1169 	.read		= rfkill_fop_read,
1170 	.write		= rfkill_fop_write,
1171 	.poll		= rfkill_fop_poll,
1172 	.release	= rfkill_fop_release,
1173 #ifdef CONFIG_RFKILL_INPUT
1174 	.unlocked_ioctl	= rfkill_fop_ioctl,
1175 	.compat_ioctl	= rfkill_fop_ioctl,
1176 #endif
1177 };
1178 
1179 static struct miscdevice rfkill_miscdev = {
1180 	.name	= "rfkill",
1181 	.fops	= &rfkill_fops,
1182 	.minor	= MISC_DYNAMIC_MINOR,
1183 };
1184 
1185 static int __init rfkill_init(void)
1186 {
1187 	int error;
1188 	int i;
1189 
1190 	for (i = 0; i < NUM_RFKILL_TYPES; i++)
1191 		rfkill_global_states[i].cur = !rfkill_default_state;
1192 
1193 	error = class_register(&rfkill_class);
1194 	if (error)
1195 		goto out;
1196 
1197 	error = misc_register(&rfkill_miscdev);
1198 	if (error) {
1199 		class_unregister(&rfkill_class);
1200 		goto out;
1201 	}
1202 
1203 #ifdef CONFIG_RFKILL_INPUT
1204 	error = rfkill_handler_init();
1205 	if (error) {
1206 		misc_deregister(&rfkill_miscdev);
1207 		class_unregister(&rfkill_class);
1208 		goto out;
1209 	}
1210 #endif
1211 
1212  out:
1213 	return error;
1214 }
1215 subsys_initcall(rfkill_init);
1216 
1217 static void __exit rfkill_exit(void)
1218 {
1219 #ifdef CONFIG_RFKILL_INPUT
1220 	rfkill_handler_exit();
1221 #endif
1222 	misc_deregister(&rfkill_miscdev);
1223 	class_unregister(&rfkill_class);
1224 }
1225 module_exit(rfkill_exit);
1226