Merge tag 'for_v5.11-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-fs
[linux-2.6-microblaze.git] / net / rfkill / core.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (C) 2006 - 2007 Ivo van Doorn
4  * Copyright (C) 2007 Dmitry Torokhov
5  * Copyright 2009 Johannes Berg <johannes@sipsolutions.net>
6  */
7
8 #include <linux/kernel.h>
9 #include <linux/module.h>
10 #include <linux/init.h>
11 #include <linux/workqueue.h>
12 #include <linux/capability.h>
13 #include <linux/list.h>
14 #include <linux/mutex.h>
15 #include <linux/rfkill.h>
16 #include <linux/sched.h>
17 #include <linux/spinlock.h>
18 #include <linux/device.h>
19 #include <linux/miscdevice.h>
20 #include <linux/wait.h>
21 #include <linux/poll.h>
22 #include <linux/fs.h>
23 #include <linux/slab.h>
24
25 #include "rfkill.h"
26
27 #define POLL_INTERVAL           (5 * HZ)
28
29 #define RFKILL_BLOCK_HW         BIT(0)
30 #define RFKILL_BLOCK_SW         BIT(1)
31 #define RFKILL_BLOCK_SW_PREV    BIT(2)
32 #define RFKILL_BLOCK_ANY        (RFKILL_BLOCK_HW |\
33                                  RFKILL_BLOCK_SW |\
34                                  RFKILL_BLOCK_SW_PREV)
35 #define RFKILL_BLOCK_SW_SETCALL BIT(31)
36
37 struct rfkill {
38         spinlock_t              lock;
39
40         enum rfkill_type        type;
41
42         unsigned long           state;
43         unsigned long           hard_block_reasons;
44
45         u32                     idx;
46
47         bool                    registered;
48         bool                    persistent;
49         bool                    polling_paused;
50         bool                    suspended;
51
52         const struct rfkill_ops *ops;
53         void                    *data;
54
55 #ifdef CONFIG_RFKILL_LEDS
56         struct led_trigger      led_trigger;
57         const char              *ledtrigname;
58 #endif
59
60         struct device           dev;
61         struct list_head        node;
62
63         struct delayed_work     poll_work;
64         struct work_struct      uevent_work;
65         struct work_struct      sync_work;
66         char                    name[];
67 };
68 #define to_rfkill(d)    container_of(d, struct rfkill, dev)
69
70 struct rfkill_int_event {
71         struct list_head        list;
72         struct rfkill_event     ev;
73 };
74
75 struct rfkill_data {
76         struct list_head        list;
77         struct list_head        events;
78         struct mutex            mtx;
79         wait_queue_head_t       read_wait;
80         bool                    input_handler;
81 };
82
83
84 MODULE_AUTHOR("Ivo van Doorn <IvDoorn@gmail.com>");
85 MODULE_AUTHOR("Johannes Berg <johannes@sipsolutions.net>");
86 MODULE_DESCRIPTION("RF switch support");
87 MODULE_LICENSE("GPL");
88
89
90 /*
91  * The locking here should be made much smarter, we currently have
92  * a bit of a stupid situation because drivers might want to register
93  * the rfkill struct under their own lock, and take this lock during
94  * rfkill method calls -- which will cause an AB-BA deadlock situation.
95  *
96  * To fix that, we need to rework this code here to be mostly lock-free
97  * and only use the mutex for list manipulations, not to protect the
98  * various other global variables. Then we can avoid holding the mutex
99  * around driver operations, and all is happy.
100  */
101 static LIST_HEAD(rfkill_list);  /* list of registered rf switches */
102 static DEFINE_MUTEX(rfkill_global_mutex);
103 static LIST_HEAD(rfkill_fds);   /* list of open fds of /dev/rfkill */
104
105 static unsigned int rfkill_default_state = 1;
106 module_param_named(default_state, rfkill_default_state, uint, 0444);
107 MODULE_PARM_DESC(default_state,
108                  "Default initial state for all radio types, 0 = radio off");
109
110 static struct {
111         bool cur, sav;
112 } rfkill_global_states[NUM_RFKILL_TYPES];
113
114 static bool rfkill_epo_lock_active;
115
116
117 #ifdef CONFIG_RFKILL_LEDS
118 static void rfkill_led_trigger_event(struct rfkill *rfkill)
119 {
120         struct led_trigger *trigger;
121
122         if (!rfkill->registered)
123                 return;
124
125         trigger = &rfkill->led_trigger;
126
127         if (rfkill->state & RFKILL_BLOCK_ANY)
128                 led_trigger_event(trigger, LED_OFF);
129         else
130                 led_trigger_event(trigger, LED_FULL);
131 }
132
133 static int rfkill_led_trigger_activate(struct led_classdev *led)
134 {
135         struct rfkill *rfkill;
136
137         rfkill = container_of(led->trigger, struct rfkill, led_trigger);
138
139         rfkill_led_trigger_event(rfkill);
140
141         return 0;
142 }
143
144 const char *rfkill_get_led_trigger_name(struct rfkill *rfkill)
145 {
146         return rfkill->led_trigger.name;
147 }
148 EXPORT_SYMBOL(rfkill_get_led_trigger_name);
149
150 void rfkill_set_led_trigger_name(struct rfkill *rfkill, const char *name)
151 {
152         BUG_ON(!rfkill);
153
154         rfkill->ledtrigname = name;
155 }
156 EXPORT_SYMBOL(rfkill_set_led_trigger_name);
157
158 static int rfkill_led_trigger_register(struct rfkill *rfkill)
159 {
160         rfkill->led_trigger.name = rfkill->ledtrigname
161                                         ? : dev_name(&rfkill->dev);
162         rfkill->led_trigger.activate = rfkill_led_trigger_activate;
163         return led_trigger_register(&rfkill->led_trigger);
164 }
165
166 static void rfkill_led_trigger_unregister(struct rfkill *rfkill)
167 {
168         led_trigger_unregister(&rfkill->led_trigger);
169 }
170
171 static struct led_trigger rfkill_any_led_trigger;
172 static struct led_trigger rfkill_none_led_trigger;
173 static struct work_struct rfkill_global_led_trigger_work;
174
175 static void rfkill_global_led_trigger_worker(struct work_struct *work)
176 {
177         enum led_brightness brightness = LED_OFF;
178         struct rfkill *rfkill;
179
180         mutex_lock(&rfkill_global_mutex);
181         list_for_each_entry(rfkill, &rfkill_list, node) {
182                 if (!(rfkill->state & RFKILL_BLOCK_ANY)) {
183                         brightness = LED_FULL;
184                         break;
185                 }
186         }
187         mutex_unlock(&rfkill_global_mutex);
188
189         led_trigger_event(&rfkill_any_led_trigger, brightness);
190         led_trigger_event(&rfkill_none_led_trigger,
191                           brightness == LED_OFF ? LED_FULL : LED_OFF);
192 }
193
194 static void rfkill_global_led_trigger_event(void)
195 {
196         schedule_work(&rfkill_global_led_trigger_work);
197 }
198
199 static int rfkill_global_led_trigger_register(void)
200 {
201         int ret;
202
203         INIT_WORK(&rfkill_global_led_trigger_work,
204                         rfkill_global_led_trigger_worker);
205
206         rfkill_any_led_trigger.name = "rfkill-any";
207         ret = led_trigger_register(&rfkill_any_led_trigger);
208         if (ret)
209                 return ret;
210
211         rfkill_none_led_trigger.name = "rfkill-none";
212         ret = led_trigger_register(&rfkill_none_led_trigger);
213         if (ret)
214                 led_trigger_unregister(&rfkill_any_led_trigger);
215         else
216                 /* Delay activation until all global triggers are registered */
217                 rfkill_global_led_trigger_event();
218
219         return ret;
220 }
221
222 static void rfkill_global_led_trigger_unregister(void)
223 {
224         led_trigger_unregister(&rfkill_none_led_trigger);
225         led_trigger_unregister(&rfkill_any_led_trigger);
226         cancel_work_sync(&rfkill_global_led_trigger_work);
227 }
228 #else
229 static void rfkill_led_trigger_event(struct rfkill *rfkill)
230 {
231 }
232
233 static inline int rfkill_led_trigger_register(struct rfkill *rfkill)
234 {
235         return 0;
236 }
237
238 static inline void rfkill_led_trigger_unregister(struct rfkill *rfkill)
239 {
240 }
241
242 static void rfkill_global_led_trigger_event(void)
243 {
244 }
245
246 static int rfkill_global_led_trigger_register(void)
247 {
248         return 0;
249 }
250
251 static void rfkill_global_led_trigger_unregister(void)
252 {
253 }
254 #endif /* CONFIG_RFKILL_LEDS */
255
256 static void rfkill_fill_event(struct rfkill_event *ev, struct rfkill *rfkill,
257                               enum rfkill_operation op)
258 {
259         unsigned long flags;
260
261         ev->idx = rfkill->idx;
262         ev->type = rfkill->type;
263         ev->op = op;
264
265         spin_lock_irqsave(&rfkill->lock, flags);
266         ev->hard = !!(rfkill->state & RFKILL_BLOCK_HW);
267         ev->soft = !!(rfkill->state & (RFKILL_BLOCK_SW |
268                                         RFKILL_BLOCK_SW_PREV));
269         ev->hard_block_reasons = rfkill->hard_block_reasons;
270         spin_unlock_irqrestore(&rfkill->lock, flags);
271 }
272
273 static void rfkill_send_events(struct rfkill *rfkill, enum rfkill_operation op)
274 {
275         struct rfkill_data *data;
276         struct rfkill_int_event *ev;
277
278         list_for_each_entry(data, &rfkill_fds, list) {
279                 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
280                 if (!ev)
281                         continue;
282                 rfkill_fill_event(&ev->ev, rfkill, op);
283                 mutex_lock(&data->mtx);
284                 list_add_tail(&ev->list, &data->events);
285                 mutex_unlock(&data->mtx);
286                 wake_up_interruptible(&data->read_wait);
287         }
288 }
289
290 static void rfkill_event(struct rfkill *rfkill)
291 {
292         if (!rfkill->registered)
293                 return;
294
295         kobject_uevent(&rfkill->dev.kobj, KOBJ_CHANGE);
296
297         /* also send event to /dev/rfkill */
298         rfkill_send_events(rfkill, RFKILL_OP_CHANGE);
299 }
300
301 /**
302  * rfkill_set_block - wrapper for set_block method
303  *
304  * @rfkill: the rfkill struct to use
305  * @blocked: the new software state
306  *
307  * Calls the set_block method (when applicable) and handles notifications
308  * etc. as well.
309  */
310 static void rfkill_set_block(struct rfkill *rfkill, bool blocked)
311 {
312         unsigned long flags;
313         bool prev, curr;
314         int err;
315
316         if (unlikely(rfkill->dev.power.power_state.event & PM_EVENT_SLEEP))
317                 return;
318
319         /*
320          * Some platforms (...!) generate input events which affect the
321          * _hard_ kill state -- whenever something tries to change the
322          * current software state query the hardware state too.
323          */
324         if (rfkill->ops->query)
325                 rfkill->ops->query(rfkill, rfkill->data);
326
327         spin_lock_irqsave(&rfkill->lock, flags);
328         prev = rfkill->state & RFKILL_BLOCK_SW;
329
330         if (prev)
331                 rfkill->state |= RFKILL_BLOCK_SW_PREV;
332         else
333                 rfkill->state &= ~RFKILL_BLOCK_SW_PREV;
334
335         if (blocked)
336                 rfkill->state |= RFKILL_BLOCK_SW;
337         else
338                 rfkill->state &= ~RFKILL_BLOCK_SW;
339
340         rfkill->state |= RFKILL_BLOCK_SW_SETCALL;
341         spin_unlock_irqrestore(&rfkill->lock, flags);
342
343         err = rfkill->ops->set_block(rfkill->data, blocked);
344
345         spin_lock_irqsave(&rfkill->lock, flags);
346         if (err) {
347                 /*
348                  * Failed -- reset status to _PREV, which may be different
349                  * from what we have set _PREV to earlier in this function
350                  * if rfkill_set_sw_state was invoked.
351                  */
352                 if (rfkill->state & RFKILL_BLOCK_SW_PREV)
353                         rfkill->state |= RFKILL_BLOCK_SW;
354                 else
355                         rfkill->state &= ~RFKILL_BLOCK_SW;
356         }
357         rfkill->state &= ~RFKILL_BLOCK_SW_SETCALL;
358         rfkill->state &= ~RFKILL_BLOCK_SW_PREV;
359         curr = rfkill->state & RFKILL_BLOCK_SW;
360         spin_unlock_irqrestore(&rfkill->lock, flags);
361
362         rfkill_led_trigger_event(rfkill);
363         rfkill_global_led_trigger_event();
364
365         if (prev != curr)
366                 rfkill_event(rfkill);
367 }
368
369 static void rfkill_update_global_state(enum rfkill_type type, bool blocked)
370 {
371         int i;
372
373         if (type != RFKILL_TYPE_ALL) {
374                 rfkill_global_states[type].cur = blocked;
375                 return;
376         }
377
378         for (i = 0; i < NUM_RFKILL_TYPES; i++)
379                 rfkill_global_states[i].cur = blocked;
380 }
381
382 #ifdef CONFIG_RFKILL_INPUT
383 static atomic_t rfkill_input_disabled = ATOMIC_INIT(0);
384
385 /**
386  * __rfkill_switch_all - Toggle state of all switches of given type
387  * @type: type of interfaces to be affected
388  * @blocked: the new state
389  *
390  * This function sets the state of all switches of given type,
391  * unless a specific switch is suspended.
392  *
393  * Caller must have acquired rfkill_global_mutex.
394  */
395 static void __rfkill_switch_all(const enum rfkill_type type, bool blocked)
396 {
397         struct rfkill *rfkill;
398
399         rfkill_update_global_state(type, blocked);
400         list_for_each_entry(rfkill, &rfkill_list, node) {
401                 if (rfkill->type != type && type != RFKILL_TYPE_ALL)
402                         continue;
403
404                 rfkill_set_block(rfkill, blocked);
405         }
406 }
407
408 /**
409  * rfkill_switch_all - Toggle state of all switches of given type
410  * @type: type of interfaces to be affected
411  * @blocked: the new state
412  *
413  * Acquires rfkill_global_mutex and calls __rfkill_switch_all(@type, @state).
414  * Please refer to __rfkill_switch_all() for details.
415  *
416  * Does nothing if the EPO lock is active.
417  */
418 void rfkill_switch_all(enum rfkill_type type, bool blocked)
419 {
420         if (atomic_read(&rfkill_input_disabled))
421                 return;
422
423         mutex_lock(&rfkill_global_mutex);
424
425         if (!rfkill_epo_lock_active)
426                 __rfkill_switch_all(type, blocked);
427
428         mutex_unlock(&rfkill_global_mutex);
429 }
430
431 /**
432  * rfkill_epo - emergency power off all transmitters
433  *
434  * This kicks all non-suspended rfkill devices to RFKILL_STATE_SOFT_BLOCKED,
435  * ignoring everything in its path but rfkill_global_mutex and rfkill->mutex.
436  *
437  * The global state before the EPO is saved and can be restored later
438  * using rfkill_restore_states().
439  */
440 void rfkill_epo(void)
441 {
442         struct rfkill *rfkill;
443         int i;
444
445         if (atomic_read(&rfkill_input_disabled))
446                 return;
447
448         mutex_lock(&rfkill_global_mutex);
449
450         rfkill_epo_lock_active = true;
451         list_for_each_entry(rfkill, &rfkill_list, node)
452                 rfkill_set_block(rfkill, true);
453
454         for (i = 0; i < NUM_RFKILL_TYPES; i++) {
455                 rfkill_global_states[i].sav = rfkill_global_states[i].cur;
456                 rfkill_global_states[i].cur = true;
457         }
458
459         mutex_unlock(&rfkill_global_mutex);
460 }
461
462 /**
463  * rfkill_restore_states - restore global states
464  *
465  * Restore (and sync switches to) the global state from the
466  * states in rfkill_default_states.  This can undo the effects of
467  * a call to rfkill_epo().
468  */
469 void rfkill_restore_states(void)
470 {
471         int i;
472
473         if (atomic_read(&rfkill_input_disabled))
474                 return;
475
476         mutex_lock(&rfkill_global_mutex);
477
478         rfkill_epo_lock_active = false;
479         for (i = 0; i < NUM_RFKILL_TYPES; i++)
480                 __rfkill_switch_all(i, rfkill_global_states[i].sav);
481         mutex_unlock(&rfkill_global_mutex);
482 }
483
484 /**
485  * rfkill_remove_epo_lock - unlock state changes
486  *
487  * Used by rfkill-input manually unlock state changes, when
488  * the EPO switch is deactivated.
489  */
490 void rfkill_remove_epo_lock(void)
491 {
492         if (atomic_read(&rfkill_input_disabled))
493                 return;
494
495         mutex_lock(&rfkill_global_mutex);
496         rfkill_epo_lock_active = false;
497         mutex_unlock(&rfkill_global_mutex);
498 }
499
500 /**
501  * rfkill_is_epo_lock_active - returns true EPO is active
502  *
503  * Returns 0 (false) if there is NOT an active EPO condition,
504  * and 1 (true) if there is an active EPO condition, which
505  * locks all radios in one of the BLOCKED states.
506  *
507  * Can be called in atomic context.
508  */
509 bool rfkill_is_epo_lock_active(void)
510 {
511         return rfkill_epo_lock_active;
512 }
513
514 /**
515  * rfkill_get_global_sw_state - returns global state for a type
516  * @type: the type to get the global state of
517  *
518  * Returns the current global state for a given wireless
519  * device type.
520  */
521 bool rfkill_get_global_sw_state(const enum rfkill_type type)
522 {
523         return rfkill_global_states[type].cur;
524 }
525 #endif
526
527 bool rfkill_set_hw_state_reason(struct rfkill *rfkill,
528                                 bool blocked, unsigned long reason)
529 {
530         unsigned long flags;
531         bool ret, prev;
532
533         BUG_ON(!rfkill);
534
535         if (WARN(reason &
536             ~(RFKILL_HARD_BLOCK_SIGNAL | RFKILL_HARD_BLOCK_NOT_OWNER),
537             "hw_state reason not supported: 0x%lx", reason))
538                 return blocked;
539
540         spin_lock_irqsave(&rfkill->lock, flags);
541         prev = !!(rfkill->hard_block_reasons & reason);
542         if (blocked) {
543                 rfkill->state |= RFKILL_BLOCK_HW;
544                 rfkill->hard_block_reasons |= reason;
545         } else {
546                 rfkill->hard_block_reasons &= ~reason;
547                 if (!rfkill->hard_block_reasons)
548                         rfkill->state &= ~RFKILL_BLOCK_HW;
549         }
550         ret = !!(rfkill->state & RFKILL_BLOCK_ANY);
551         spin_unlock_irqrestore(&rfkill->lock, flags);
552
553         rfkill_led_trigger_event(rfkill);
554         rfkill_global_led_trigger_event();
555
556         if (rfkill->registered && prev != blocked)
557                 schedule_work(&rfkill->uevent_work);
558
559         return ret;
560 }
561 EXPORT_SYMBOL(rfkill_set_hw_state_reason);
562
563 static void __rfkill_set_sw_state(struct rfkill *rfkill, bool blocked)
564 {
565         u32 bit = RFKILL_BLOCK_SW;
566
567         /* if in a ops->set_block right now, use other bit */
568         if (rfkill->state & RFKILL_BLOCK_SW_SETCALL)
569                 bit = RFKILL_BLOCK_SW_PREV;
570
571         if (blocked)
572                 rfkill->state |= bit;
573         else
574                 rfkill->state &= ~bit;
575 }
576
577 bool rfkill_set_sw_state(struct rfkill *rfkill, bool blocked)
578 {
579         unsigned long flags;
580         bool prev, hwblock;
581
582         BUG_ON(!rfkill);
583
584         spin_lock_irqsave(&rfkill->lock, flags);
585         prev = !!(rfkill->state & RFKILL_BLOCK_SW);
586         __rfkill_set_sw_state(rfkill, blocked);
587         hwblock = !!(rfkill->state & RFKILL_BLOCK_HW);
588         blocked = blocked || hwblock;
589         spin_unlock_irqrestore(&rfkill->lock, flags);
590
591         if (!rfkill->registered)
592                 return blocked;
593
594         if (prev != blocked && !hwblock)
595                 schedule_work(&rfkill->uevent_work);
596
597         rfkill_led_trigger_event(rfkill);
598         rfkill_global_led_trigger_event();
599
600         return blocked;
601 }
602 EXPORT_SYMBOL(rfkill_set_sw_state);
603
604 void rfkill_init_sw_state(struct rfkill *rfkill, bool blocked)
605 {
606         unsigned long flags;
607
608         BUG_ON(!rfkill);
609         BUG_ON(rfkill->registered);
610
611         spin_lock_irqsave(&rfkill->lock, flags);
612         __rfkill_set_sw_state(rfkill, blocked);
613         rfkill->persistent = true;
614         spin_unlock_irqrestore(&rfkill->lock, flags);
615 }
616 EXPORT_SYMBOL(rfkill_init_sw_state);
617
618 void rfkill_set_states(struct rfkill *rfkill, bool sw, bool hw)
619 {
620         unsigned long flags;
621         bool swprev, hwprev;
622
623         BUG_ON(!rfkill);
624
625         spin_lock_irqsave(&rfkill->lock, flags);
626
627         /*
628          * No need to care about prev/setblock ... this is for uevent only
629          * and that will get triggered by rfkill_set_block anyway.
630          */
631         swprev = !!(rfkill->state & RFKILL_BLOCK_SW);
632         hwprev = !!(rfkill->state & RFKILL_BLOCK_HW);
633         __rfkill_set_sw_state(rfkill, sw);
634         if (hw)
635                 rfkill->state |= RFKILL_BLOCK_HW;
636         else
637                 rfkill->state &= ~RFKILL_BLOCK_HW;
638
639         spin_unlock_irqrestore(&rfkill->lock, flags);
640
641         if (!rfkill->registered) {
642                 rfkill->persistent = true;
643         } else {
644                 if (swprev != sw || hwprev != hw)
645                         schedule_work(&rfkill->uevent_work);
646
647                 rfkill_led_trigger_event(rfkill);
648                 rfkill_global_led_trigger_event();
649         }
650 }
651 EXPORT_SYMBOL(rfkill_set_states);
652
653 static const char * const rfkill_types[] = {
654         NULL, /* RFKILL_TYPE_ALL */
655         "wlan",
656         "bluetooth",
657         "ultrawideband",
658         "wimax",
659         "wwan",
660         "gps",
661         "fm",
662         "nfc",
663 };
664
665 enum rfkill_type rfkill_find_type(const char *name)
666 {
667         int i;
668
669         BUILD_BUG_ON(ARRAY_SIZE(rfkill_types) != NUM_RFKILL_TYPES);
670
671         if (!name)
672                 return RFKILL_TYPE_ALL;
673
674         for (i = 1; i < NUM_RFKILL_TYPES; i++)
675                 if (!strcmp(name, rfkill_types[i]))
676                         return i;
677         return RFKILL_TYPE_ALL;
678 }
679 EXPORT_SYMBOL(rfkill_find_type);
680
681 static ssize_t name_show(struct device *dev, struct device_attribute *attr,
682                          char *buf)
683 {
684         struct rfkill *rfkill = to_rfkill(dev);
685
686         return sprintf(buf, "%s\n", rfkill->name);
687 }
688 static DEVICE_ATTR_RO(name);
689
690 static ssize_t type_show(struct device *dev, struct device_attribute *attr,
691                          char *buf)
692 {
693         struct rfkill *rfkill = to_rfkill(dev);
694
695         return sprintf(buf, "%s\n", rfkill_types[rfkill->type]);
696 }
697 static DEVICE_ATTR_RO(type);
698
699 static ssize_t index_show(struct device *dev, struct device_attribute *attr,
700                           char *buf)
701 {
702         struct rfkill *rfkill = to_rfkill(dev);
703
704         return sprintf(buf, "%d\n", rfkill->idx);
705 }
706 static DEVICE_ATTR_RO(index);
707
708 static ssize_t persistent_show(struct device *dev,
709                                struct device_attribute *attr, char *buf)
710 {
711         struct rfkill *rfkill = to_rfkill(dev);
712
713         return sprintf(buf, "%d\n", rfkill->persistent);
714 }
715 static DEVICE_ATTR_RO(persistent);
716
717 static ssize_t hard_show(struct device *dev, struct device_attribute *attr,
718                          char *buf)
719 {
720         struct rfkill *rfkill = to_rfkill(dev);
721
722         return sprintf(buf, "%d\n", (rfkill->state & RFKILL_BLOCK_HW) ? 1 : 0 );
723 }
724 static DEVICE_ATTR_RO(hard);
725
726 static ssize_t soft_show(struct device *dev, struct device_attribute *attr,
727                          char *buf)
728 {
729         struct rfkill *rfkill = to_rfkill(dev);
730
731         return sprintf(buf, "%d\n", (rfkill->state & RFKILL_BLOCK_SW) ? 1 : 0 );
732 }
733
734 static ssize_t soft_store(struct device *dev, struct device_attribute *attr,
735                           const char *buf, size_t count)
736 {
737         struct rfkill *rfkill = to_rfkill(dev);
738         unsigned long state;
739         int err;
740
741         if (!capable(CAP_NET_ADMIN))
742                 return -EPERM;
743
744         err = kstrtoul(buf, 0, &state);
745         if (err)
746                 return err;
747
748         if (state > 1 )
749                 return -EINVAL;
750
751         mutex_lock(&rfkill_global_mutex);
752         rfkill_set_block(rfkill, state);
753         mutex_unlock(&rfkill_global_mutex);
754
755         return count;
756 }
757 static DEVICE_ATTR_RW(soft);
758
759 static ssize_t hard_block_reasons_show(struct device *dev,
760                                        struct device_attribute *attr,
761                                        char *buf)
762 {
763         struct rfkill *rfkill = to_rfkill(dev);
764
765         return sprintf(buf, "0x%lx\n", rfkill->hard_block_reasons);
766 }
767 static DEVICE_ATTR_RO(hard_block_reasons);
768
769 static u8 user_state_from_blocked(unsigned long state)
770 {
771         if (state & RFKILL_BLOCK_HW)
772                 return RFKILL_USER_STATE_HARD_BLOCKED;
773         if (state & RFKILL_BLOCK_SW)
774                 return RFKILL_USER_STATE_SOFT_BLOCKED;
775
776         return RFKILL_USER_STATE_UNBLOCKED;
777 }
778
779 static ssize_t state_show(struct device *dev, struct device_attribute *attr,
780                           char *buf)
781 {
782         struct rfkill *rfkill = to_rfkill(dev);
783
784         return sprintf(buf, "%d\n", user_state_from_blocked(rfkill->state));
785 }
786
787 static ssize_t state_store(struct device *dev, struct device_attribute *attr,
788                            const char *buf, size_t count)
789 {
790         struct rfkill *rfkill = to_rfkill(dev);
791         unsigned long state;
792         int err;
793
794         if (!capable(CAP_NET_ADMIN))
795                 return -EPERM;
796
797         err = kstrtoul(buf, 0, &state);
798         if (err)
799                 return err;
800
801         if (state != RFKILL_USER_STATE_SOFT_BLOCKED &&
802             state != RFKILL_USER_STATE_UNBLOCKED)
803                 return -EINVAL;
804
805         mutex_lock(&rfkill_global_mutex);
806         rfkill_set_block(rfkill, state == RFKILL_USER_STATE_SOFT_BLOCKED);
807         mutex_unlock(&rfkill_global_mutex);
808
809         return count;
810 }
811 static DEVICE_ATTR_RW(state);
812
813 static struct attribute *rfkill_dev_attrs[] = {
814         &dev_attr_name.attr,
815         &dev_attr_type.attr,
816         &dev_attr_index.attr,
817         &dev_attr_persistent.attr,
818         &dev_attr_state.attr,
819         &dev_attr_soft.attr,
820         &dev_attr_hard.attr,
821         &dev_attr_hard_block_reasons.attr,
822         NULL,
823 };
824 ATTRIBUTE_GROUPS(rfkill_dev);
825
826 static void rfkill_release(struct device *dev)
827 {
828         struct rfkill *rfkill = to_rfkill(dev);
829
830         kfree(rfkill);
831 }
832
833 static int rfkill_dev_uevent(struct device *dev, struct kobj_uevent_env *env)
834 {
835         struct rfkill *rfkill = to_rfkill(dev);
836         unsigned long flags;
837         unsigned long reasons;
838         u32 state;
839         int error;
840
841         error = add_uevent_var(env, "RFKILL_NAME=%s", rfkill->name);
842         if (error)
843                 return error;
844         error = add_uevent_var(env, "RFKILL_TYPE=%s",
845                                rfkill_types[rfkill->type]);
846         if (error)
847                 return error;
848         spin_lock_irqsave(&rfkill->lock, flags);
849         state = rfkill->state;
850         reasons = rfkill->hard_block_reasons;
851         spin_unlock_irqrestore(&rfkill->lock, flags);
852         error = add_uevent_var(env, "RFKILL_STATE=%d",
853                                user_state_from_blocked(state));
854         if (error)
855                 return error;
856         return add_uevent_var(env, "RFKILL_HW_BLOCK_REASON=0x%lx", reasons);
857 }
858
859 void rfkill_pause_polling(struct rfkill *rfkill)
860 {
861         BUG_ON(!rfkill);
862
863         if (!rfkill->ops->poll)
864                 return;
865
866         rfkill->polling_paused = true;
867         cancel_delayed_work_sync(&rfkill->poll_work);
868 }
869 EXPORT_SYMBOL(rfkill_pause_polling);
870
871 void rfkill_resume_polling(struct rfkill *rfkill)
872 {
873         BUG_ON(!rfkill);
874
875         if (!rfkill->ops->poll)
876                 return;
877
878         rfkill->polling_paused = false;
879
880         if (rfkill->suspended)
881                 return;
882
883         queue_delayed_work(system_power_efficient_wq,
884                            &rfkill->poll_work, 0);
885 }
886 EXPORT_SYMBOL(rfkill_resume_polling);
887
888 #ifdef CONFIG_PM_SLEEP
889 static int rfkill_suspend(struct device *dev)
890 {
891         struct rfkill *rfkill = to_rfkill(dev);
892
893         rfkill->suspended = true;
894         cancel_delayed_work_sync(&rfkill->poll_work);
895
896         return 0;
897 }
898
899 static int rfkill_resume(struct device *dev)
900 {
901         struct rfkill *rfkill = to_rfkill(dev);
902         bool cur;
903
904         rfkill->suspended = false;
905
906         if (!rfkill->registered)
907                 return 0;
908
909         if (!rfkill->persistent) {
910                 cur = !!(rfkill->state & RFKILL_BLOCK_SW);
911                 rfkill_set_block(rfkill, cur);
912         }
913
914         if (rfkill->ops->poll && !rfkill->polling_paused)
915                 queue_delayed_work(system_power_efficient_wq,
916                                    &rfkill->poll_work, 0);
917
918         return 0;
919 }
920
921 static SIMPLE_DEV_PM_OPS(rfkill_pm_ops, rfkill_suspend, rfkill_resume);
922 #define RFKILL_PM_OPS (&rfkill_pm_ops)
923 #else
924 #define RFKILL_PM_OPS NULL
925 #endif
926
927 static struct class rfkill_class = {
928         .name           = "rfkill",
929         .dev_release    = rfkill_release,
930         .dev_groups     = rfkill_dev_groups,
931         .dev_uevent     = rfkill_dev_uevent,
932         .pm             = RFKILL_PM_OPS,
933 };
934
935 bool rfkill_blocked(struct rfkill *rfkill)
936 {
937         unsigned long flags;
938         u32 state;
939
940         spin_lock_irqsave(&rfkill->lock, flags);
941         state = rfkill->state;
942         spin_unlock_irqrestore(&rfkill->lock, flags);
943
944         return !!(state & RFKILL_BLOCK_ANY);
945 }
946 EXPORT_SYMBOL(rfkill_blocked);
947
948
949 struct rfkill * __must_check rfkill_alloc(const char *name,
950                                           struct device *parent,
951                                           const enum rfkill_type type,
952                                           const struct rfkill_ops *ops,
953                                           void *ops_data)
954 {
955         struct rfkill *rfkill;
956         struct device *dev;
957
958         if (WARN_ON(!ops))
959                 return NULL;
960
961         if (WARN_ON(!ops->set_block))
962                 return NULL;
963
964         if (WARN_ON(!name))
965                 return NULL;
966
967         if (WARN_ON(type == RFKILL_TYPE_ALL || type >= NUM_RFKILL_TYPES))
968                 return NULL;
969
970         rfkill = kzalloc(sizeof(*rfkill) + strlen(name) + 1, GFP_KERNEL);
971         if (!rfkill)
972                 return NULL;
973
974         spin_lock_init(&rfkill->lock);
975         INIT_LIST_HEAD(&rfkill->node);
976         rfkill->type = type;
977         strcpy(rfkill->name, name);
978         rfkill->ops = ops;
979         rfkill->data = ops_data;
980
981         dev = &rfkill->dev;
982         dev->class = &rfkill_class;
983         dev->parent = parent;
984         device_initialize(dev);
985
986         return rfkill;
987 }
988 EXPORT_SYMBOL(rfkill_alloc);
989
990 static void rfkill_poll(struct work_struct *work)
991 {
992         struct rfkill *rfkill;
993
994         rfkill = container_of(work, struct rfkill, poll_work.work);
995
996         /*
997          * Poll hardware state -- driver will use one of the
998          * rfkill_set{,_hw,_sw}_state functions and use its
999          * return value to update the current status.
1000          */
1001         rfkill->ops->poll(rfkill, rfkill->data);
1002
1003         queue_delayed_work(system_power_efficient_wq,
1004                 &rfkill->poll_work,
1005                 round_jiffies_relative(POLL_INTERVAL));
1006 }
1007
1008 static void rfkill_uevent_work(struct work_struct *work)
1009 {
1010         struct rfkill *rfkill;
1011
1012         rfkill = container_of(work, struct rfkill, uevent_work);
1013
1014         mutex_lock(&rfkill_global_mutex);
1015         rfkill_event(rfkill);
1016         mutex_unlock(&rfkill_global_mutex);
1017 }
1018
1019 static void rfkill_sync_work(struct work_struct *work)
1020 {
1021         struct rfkill *rfkill;
1022         bool cur;
1023
1024         rfkill = container_of(work, struct rfkill, sync_work);
1025
1026         mutex_lock(&rfkill_global_mutex);
1027         cur = rfkill_global_states[rfkill->type].cur;
1028         rfkill_set_block(rfkill, cur);
1029         mutex_unlock(&rfkill_global_mutex);
1030 }
1031
1032 int __must_check rfkill_register(struct rfkill *rfkill)
1033 {
1034         static unsigned long rfkill_no;
1035         struct device *dev;
1036         int error;
1037
1038         if (!rfkill)
1039                 return -EINVAL;
1040
1041         dev = &rfkill->dev;
1042
1043         mutex_lock(&rfkill_global_mutex);
1044
1045         if (rfkill->registered) {
1046                 error = -EALREADY;
1047                 goto unlock;
1048         }
1049
1050         rfkill->idx = rfkill_no;
1051         dev_set_name(dev, "rfkill%lu", rfkill_no);
1052         rfkill_no++;
1053
1054         list_add_tail(&rfkill->node, &rfkill_list);
1055
1056         error = device_add(dev);
1057         if (error)
1058                 goto remove;
1059
1060         error = rfkill_led_trigger_register(rfkill);
1061         if (error)
1062                 goto devdel;
1063
1064         rfkill->registered = true;
1065
1066         INIT_DELAYED_WORK(&rfkill->poll_work, rfkill_poll);
1067         INIT_WORK(&rfkill->uevent_work, rfkill_uevent_work);
1068         INIT_WORK(&rfkill->sync_work, rfkill_sync_work);
1069
1070         if (rfkill->ops->poll)
1071                 queue_delayed_work(system_power_efficient_wq,
1072                         &rfkill->poll_work,
1073                         round_jiffies_relative(POLL_INTERVAL));
1074
1075         if (!rfkill->persistent || rfkill_epo_lock_active) {
1076                 schedule_work(&rfkill->sync_work);
1077         } else {
1078 #ifdef CONFIG_RFKILL_INPUT
1079                 bool soft_blocked = !!(rfkill->state & RFKILL_BLOCK_SW);
1080
1081                 if (!atomic_read(&rfkill_input_disabled))
1082                         __rfkill_switch_all(rfkill->type, soft_blocked);
1083 #endif
1084         }
1085
1086         rfkill_global_led_trigger_event();
1087         rfkill_send_events(rfkill, RFKILL_OP_ADD);
1088
1089         mutex_unlock(&rfkill_global_mutex);
1090         return 0;
1091
1092  devdel:
1093         device_del(&rfkill->dev);
1094  remove:
1095         list_del_init(&rfkill->node);
1096  unlock:
1097         mutex_unlock(&rfkill_global_mutex);
1098         return error;
1099 }
1100 EXPORT_SYMBOL(rfkill_register);
1101
1102 void rfkill_unregister(struct rfkill *rfkill)
1103 {
1104         BUG_ON(!rfkill);
1105
1106         if (rfkill->ops->poll)
1107                 cancel_delayed_work_sync(&rfkill->poll_work);
1108
1109         cancel_work_sync(&rfkill->uevent_work);
1110         cancel_work_sync(&rfkill->sync_work);
1111
1112         rfkill->registered = false;
1113
1114         device_del(&rfkill->dev);
1115
1116         mutex_lock(&rfkill_global_mutex);
1117         rfkill_send_events(rfkill, RFKILL_OP_DEL);
1118         list_del_init(&rfkill->node);
1119         rfkill_global_led_trigger_event();
1120         mutex_unlock(&rfkill_global_mutex);
1121
1122         rfkill_led_trigger_unregister(rfkill);
1123 }
1124 EXPORT_SYMBOL(rfkill_unregister);
1125
1126 void rfkill_destroy(struct rfkill *rfkill)
1127 {
1128         if (rfkill)
1129                 put_device(&rfkill->dev);
1130 }
1131 EXPORT_SYMBOL(rfkill_destroy);
1132
1133 static int rfkill_fop_open(struct inode *inode, struct file *file)
1134 {
1135         struct rfkill_data *data;
1136         struct rfkill *rfkill;
1137         struct rfkill_int_event *ev, *tmp;
1138
1139         data = kzalloc(sizeof(*data), GFP_KERNEL);
1140         if (!data)
1141                 return -ENOMEM;
1142
1143         INIT_LIST_HEAD(&data->events);
1144         mutex_init(&data->mtx);
1145         init_waitqueue_head(&data->read_wait);
1146
1147         mutex_lock(&rfkill_global_mutex);
1148         mutex_lock(&data->mtx);
1149         /*
1150          * start getting events from elsewhere but hold mtx to get
1151          * startup events added first
1152          */
1153
1154         list_for_each_entry(rfkill, &rfkill_list, node) {
1155                 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1156                 if (!ev)
1157                         goto free;
1158                 rfkill_fill_event(&ev->ev, rfkill, RFKILL_OP_ADD);
1159                 list_add_tail(&ev->list, &data->events);
1160         }
1161         list_add(&data->list, &rfkill_fds);
1162         mutex_unlock(&data->mtx);
1163         mutex_unlock(&rfkill_global_mutex);
1164
1165         file->private_data = data;
1166
1167         return stream_open(inode, file);
1168
1169  free:
1170         mutex_unlock(&data->mtx);
1171         mutex_unlock(&rfkill_global_mutex);
1172         mutex_destroy(&data->mtx);
1173         list_for_each_entry_safe(ev, tmp, &data->events, list)
1174                 kfree(ev);
1175         kfree(data);
1176         return -ENOMEM;
1177 }
1178
1179 static __poll_t rfkill_fop_poll(struct file *file, poll_table *wait)
1180 {
1181         struct rfkill_data *data = file->private_data;
1182         __poll_t res = EPOLLOUT | EPOLLWRNORM;
1183
1184         poll_wait(file, &data->read_wait, wait);
1185
1186         mutex_lock(&data->mtx);
1187         if (!list_empty(&data->events))
1188                 res = EPOLLIN | EPOLLRDNORM;
1189         mutex_unlock(&data->mtx);
1190
1191         return res;
1192 }
1193
1194 static ssize_t rfkill_fop_read(struct file *file, char __user *buf,
1195                                size_t count, loff_t *pos)
1196 {
1197         struct rfkill_data *data = file->private_data;
1198         struct rfkill_int_event *ev;
1199         unsigned long sz;
1200         int ret;
1201
1202         mutex_lock(&data->mtx);
1203
1204         while (list_empty(&data->events)) {
1205                 if (file->f_flags & O_NONBLOCK) {
1206                         ret = -EAGAIN;
1207                         goto out;
1208                 }
1209                 mutex_unlock(&data->mtx);
1210                 /* since we re-check and it just compares pointers,
1211                  * using !list_empty() without locking isn't a problem
1212                  */
1213                 ret = wait_event_interruptible(data->read_wait,
1214                                                !list_empty(&data->events));
1215                 mutex_lock(&data->mtx);
1216
1217                 if (ret)
1218                         goto out;
1219         }
1220
1221         ev = list_first_entry(&data->events, struct rfkill_int_event,
1222                                 list);
1223
1224         sz = min_t(unsigned long, sizeof(ev->ev), count);
1225         ret = sz;
1226         if (copy_to_user(buf, &ev->ev, sz))
1227                 ret = -EFAULT;
1228
1229         list_del(&ev->list);
1230         kfree(ev);
1231  out:
1232         mutex_unlock(&data->mtx);
1233         return ret;
1234 }
1235
1236 static ssize_t rfkill_fop_write(struct file *file, const char __user *buf,
1237                                 size_t count, loff_t *pos)
1238 {
1239         struct rfkill *rfkill;
1240         struct rfkill_event ev;
1241         int ret;
1242
1243         /* we don't need the 'hard' variable but accept it */
1244         if (count < RFKILL_EVENT_SIZE_V1 - 1)
1245                 return -EINVAL;
1246
1247         /*
1248          * Copy as much data as we can accept into our 'ev' buffer,
1249          * but tell userspace how much we've copied so it can determine
1250          * our API version even in a write() call, if it cares.
1251          */
1252         count = min(count, sizeof(ev));
1253         if (copy_from_user(&ev, buf, count))
1254                 return -EFAULT;
1255
1256         if (ev.type >= NUM_RFKILL_TYPES)
1257                 return -EINVAL;
1258
1259         mutex_lock(&rfkill_global_mutex);
1260
1261         switch (ev.op) {
1262         case RFKILL_OP_CHANGE_ALL:
1263                 rfkill_update_global_state(ev.type, ev.soft);
1264                 list_for_each_entry(rfkill, &rfkill_list, node)
1265                         if (rfkill->type == ev.type ||
1266                             ev.type == RFKILL_TYPE_ALL)
1267                                 rfkill_set_block(rfkill, ev.soft);
1268                 ret = 0;
1269                 break;
1270         case RFKILL_OP_CHANGE:
1271                 list_for_each_entry(rfkill, &rfkill_list, node)
1272                         if (rfkill->idx == ev.idx &&
1273                             (rfkill->type == ev.type ||
1274                              ev.type == RFKILL_TYPE_ALL))
1275                                 rfkill_set_block(rfkill, ev.soft);
1276                 ret = 0;
1277                 break;
1278         default:
1279                 ret = -EINVAL;
1280                 break;
1281         }
1282
1283         mutex_unlock(&rfkill_global_mutex);
1284
1285         return ret ?: count;
1286 }
1287
1288 static int rfkill_fop_release(struct inode *inode, struct file *file)
1289 {
1290         struct rfkill_data *data = file->private_data;
1291         struct rfkill_int_event *ev, *tmp;
1292
1293         mutex_lock(&rfkill_global_mutex);
1294         list_del(&data->list);
1295         mutex_unlock(&rfkill_global_mutex);
1296
1297         mutex_destroy(&data->mtx);
1298         list_for_each_entry_safe(ev, tmp, &data->events, list)
1299                 kfree(ev);
1300
1301 #ifdef CONFIG_RFKILL_INPUT
1302         if (data->input_handler)
1303                 if (atomic_dec_return(&rfkill_input_disabled) == 0)
1304                         printk(KERN_DEBUG "rfkill: input handler enabled\n");
1305 #endif
1306
1307         kfree(data);
1308
1309         return 0;
1310 }
1311
1312 #ifdef CONFIG_RFKILL_INPUT
1313 static long rfkill_fop_ioctl(struct file *file, unsigned int cmd,
1314                              unsigned long arg)
1315 {
1316         struct rfkill_data *data = file->private_data;
1317
1318         if (_IOC_TYPE(cmd) != RFKILL_IOC_MAGIC)
1319                 return -ENOSYS;
1320
1321         if (_IOC_NR(cmd) != RFKILL_IOC_NOINPUT)
1322                 return -ENOSYS;
1323
1324         mutex_lock(&data->mtx);
1325
1326         if (!data->input_handler) {
1327                 if (atomic_inc_return(&rfkill_input_disabled) == 1)
1328                         printk(KERN_DEBUG "rfkill: input handler disabled\n");
1329                 data->input_handler = true;
1330         }
1331
1332         mutex_unlock(&data->mtx);
1333
1334         return 0;
1335 }
1336 #endif
1337
1338 static const struct file_operations rfkill_fops = {
1339         .owner          = THIS_MODULE,
1340         .open           = rfkill_fop_open,
1341         .read           = rfkill_fop_read,
1342         .write          = rfkill_fop_write,
1343         .poll           = rfkill_fop_poll,
1344         .release        = rfkill_fop_release,
1345 #ifdef CONFIG_RFKILL_INPUT
1346         .unlocked_ioctl = rfkill_fop_ioctl,
1347         .compat_ioctl   = compat_ptr_ioctl,
1348 #endif
1349         .llseek         = no_llseek,
1350 };
1351
1352 #define RFKILL_NAME "rfkill"
1353
1354 static struct miscdevice rfkill_miscdev = {
1355         .fops   = &rfkill_fops,
1356         .name   = RFKILL_NAME,
1357         .minor  = RFKILL_MINOR,
1358 };
1359
1360 static int __init rfkill_init(void)
1361 {
1362         int error;
1363
1364         rfkill_update_global_state(RFKILL_TYPE_ALL, !rfkill_default_state);
1365
1366         error = class_register(&rfkill_class);
1367         if (error)
1368                 goto error_class;
1369
1370         error = misc_register(&rfkill_miscdev);
1371         if (error)
1372                 goto error_misc;
1373
1374         error = rfkill_global_led_trigger_register();
1375         if (error)
1376                 goto error_led_trigger;
1377
1378 #ifdef CONFIG_RFKILL_INPUT
1379         error = rfkill_handler_init();
1380         if (error)
1381                 goto error_input;
1382 #endif
1383
1384         return 0;
1385
1386 #ifdef CONFIG_RFKILL_INPUT
1387 error_input:
1388         rfkill_global_led_trigger_unregister();
1389 #endif
1390 error_led_trigger:
1391         misc_deregister(&rfkill_miscdev);
1392 error_misc:
1393         class_unregister(&rfkill_class);
1394 error_class:
1395         return error;
1396 }
1397 subsys_initcall(rfkill_init);
1398
1399 static void __exit rfkill_exit(void)
1400 {
1401 #ifdef CONFIG_RFKILL_INPUT
1402         rfkill_handler_exit();
1403 #endif
1404         rfkill_global_led_trigger_unregister();
1405         misc_deregister(&rfkill_miscdev);
1406         class_unregister(&rfkill_class);
1407 }
1408 module_exit(rfkill_exit);
1409
1410 MODULE_ALIAS_MISCDEV(RFKILL_MINOR);
1411 MODULE_ALIAS("devname:" RFKILL_NAME);