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