core.c (33382B)
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 37struct 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 70struct rfkill_int_event { 71 struct list_head list; 72 struct rfkill_event_ext ev; 73}; 74 75struct 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 u8 max_size; 82}; 83 84 85MODULE_AUTHOR("Ivo van Doorn <IvDoorn@gmail.com>"); 86MODULE_AUTHOR("Johannes Berg <johannes@sipsolutions.net>"); 87MODULE_DESCRIPTION("RF switch support"); 88MODULE_LICENSE("GPL"); 89 90 91/* 92 * The locking here should be made much smarter, we currently have 93 * a bit of a stupid situation because drivers might want to register 94 * the rfkill struct under their own lock, and take this lock during 95 * rfkill method calls -- which will cause an AB-BA deadlock situation. 96 * 97 * To fix that, we need to rework this code here to be mostly lock-free 98 * and only use the mutex for list manipulations, not to protect the 99 * various other global variables. Then we can avoid holding the mutex 100 * around driver operations, and all is happy. 101 */ 102static LIST_HEAD(rfkill_list); /* list of registered rf switches */ 103static DEFINE_MUTEX(rfkill_global_mutex); 104static LIST_HEAD(rfkill_fds); /* list of open fds of /dev/rfkill */ 105 106static unsigned int rfkill_default_state = 1; 107module_param_named(default_state, rfkill_default_state, uint, 0444); 108MODULE_PARM_DESC(default_state, 109 "Default initial state for all radio types, 0 = radio off"); 110 111static struct { 112 bool cur, sav; 113} rfkill_global_states[NUM_RFKILL_TYPES]; 114 115static bool rfkill_epo_lock_active; 116 117 118#ifdef CONFIG_RFKILL_LEDS 119static void rfkill_led_trigger_event(struct rfkill *rfkill) 120{ 121 struct led_trigger *trigger; 122 123 if (!rfkill->registered) 124 return; 125 126 trigger = &rfkill->led_trigger; 127 128 if (rfkill->state & RFKILL_BLOCK_ANY) 129 led_trigger_event(trigger, LED_OFF); 130 else 131 led_trigger_event(trigger, LED_FULL); 132} 133 134static int rfkill_led_trigger_activate(struct led_classdev *led) 135{ 136 struct rfkill *rfkill; 137 138 rfkill = container_of(led->trigger, struct rfkill, led_trigger); 139 140 rfkill_led_trigger_event(rfkill); 141 142 return 0; 143} 144 145const char *rfkill_get_led_trigger_name(struct rfkill *rfkill) 146{ 147 return rfkill->led_trigger.name; 148} 149EXPORT_SYMBOL(rfkill_get_led_trigger_name); 150 151void rfkill_set_led_trigger_name(struct rfkill *rfkill, const char *name) 152{ 153 BUG_ON(!rfkill); 154 155 rfkill->ledtrigname = name; 156} 157EXPORT_SYMBOL(rfkill_set_led_trigger_name); 158 159static int rfkill_led_trigger_register(struct rfkill *rfkill) 160{ 161 rfkill->led_trigger.name = rfkill->ledtrigname 162 ? : dev_name(&rfkill->dev); 163 rfkill->led_trigger.activate = rfkill_led_trigger_activate; 164 return led_trigger_register(&rfkill->led_trigger); 165} 166 167static void rfkill_led_trigger_unregister(struct rfkill *rfkill) 168{ 169 led_trigger_unregister(&rfkill->led_trigger); 170} 171 172static struct led_trigger rfkill_any_led_trigger; 173static struct led_trigger rfkill_none_led_trigger; 174static struct work_struct rfkill_global_led_trigger_work; 175 176static void rfkill_global_led_trigger_worker(struct work_struct *work) 177{ 178 enum led_brightness brightness = LED_OFF; 179 struct rfkill *rfkill; 180 181 mutex_lock(&rfkill_global_mutex); 182 list_for_each_entry(rfkill, &rfkill_list, node) { 183 if (!(rfkill->state & RFKILL_BLOCK_ANY)) { 184 brightness = LED_FULL; 185 break; 186 } 187 } 188 mutex_unlock(&rfkill_global_mutex); 189 190 led_trigger_event(&rfkill_any_led_trigger, brightness); 191 led_trigger_event(&rfkill_none_led_trigger, 192 brightness == LED_OFF ? LED_FULL : LED_OFF); 193} 194 195static void rfkill_global_led_trigger_event(void) 196{ 197 schedule_work(&rfkill_global_led_trigger_work); 198} 199 200static int rfkill_global_led_trigger_register(void) 201{ 202 int ret; 203 204 INIT_WORK(&rfkill_global_led_trigger_work, 205 rfkill_global_led_trigger_worker); 206 207 rfkill_any_led_trigger.name = "rfkill-any"; 208 ret = led_trigger_register(&rfkill_any_led_trigger); 209 if (ret) 210 return ret; 211 212 rfkill_none_led_trigger.name = "rfkill-none"; 213 ret = led_trigger_register(&rfkill_none_led_trigger); 214 if (ret) 215 led_trigger_unregister(&rfkill_any_led_trigger); 216 else 217 /* Delay activation until all global triggers are registered */ 218 rfkill_global_led_trigger_event(); 219 220 return ret; 221} 222 223static void rfkill_global_led_trigger_unregister(void) 224{ 225 led_trigger_unregister(&rfkill_none_led_trigger); 226 led_trigger_unregister(&rfkill_any_led_trigger); 227 cancel_work_sync(&rfkill_global_led_trigger_work); 228} 229#else 230static void rfkill_led_trigger_event(struct rfkill *rfkill) 231{ 232} 233 234static inline int rfkill_led_trigger_register(struct rfkill *rfkill) 235{ 236 return 0; 237} 238 239static inline void rfkill_led_trigger_unregister(struct rfkill *rfkill) 240{ 241} 242 243static void rfkill_global_led_trigger_event(void) 244{ 245} 246 247static int rfkill_global_led_trigger_register(void) 248{ 249 return 0; 250} 251 252static void rfkill_global_led_trigger_unregister(void) 253{ 254} 255#endif /* CONFIG_RFKILL_LEDS */ 256 257static void rfkill_fill_event(struct rfkill_event_ext *ev, 258 struct rfkill *rfkill, 259 enum rfkill_operation op) 260{ 261 unsigned long flags; 262 263 ev->idx = rfkill->idx; 264 ev->type = rfkill->type; 265 ev->op = op; 266 267 spin_lock_irqsave(&rfkill->lock, flags); 268 ev->hard = !!(rfkill->state & RFKILL_BLOCK_HW); 269 ev->soft = !!(rfkill->state & (RFKILL_BLOCK_SW | 270 RFKILL_BLOCK_SW_PREV)); 271 ev->hard_block_reasons = rfkill->hard_block_reasons; 272 spin_unlock_irqrestore(&rfkill->lock, flags); 273} 274 275static void rfkill_send_events(struct rfkill *rfkill, enum rfkill_operation op) 276{ 277 struct rfkill_data *data; 278 struct rfkill_int_event *ev; 279 280 list_for_each_entry(data, &rfkill_fds, list) { 281 ev = kzalloc(sizeof(*ev), GFP_KERNEL); 282 if (!ev) 283 continue; 284 rfkill_fill_event(&ev->ev, rfkill, op); 285 mutex_lock(&data->mtx); 286 list_add_tail(&ev->list, &data->events); 287 mutex_unlock(&data->mtx); 288 wake_up_interruptible(&data->read_wait); 289 } 290} 291 292static void rfkill_event(struct rfkill *rfkill) 293{ 294 if (!rfkill->registered) 295 return; 296 297 kobject_uevent(&rfkill->dev.kobj, KOBJ_CHANGE); 298 299 /* also send event to /dev/rfkill */ 300 rfkill_send_events(rfkill, RFKILL_OP_CHANGE); 301} 302 303/** 304 * rfkill_set_block - wrapper for set_block method 305 * 306 * @rfkill: the rfkill struct to use 307 * @blocked: the new software state 308 * 309 * Calls the set_block method (when applicable) and handles notifications 310 * etc. as well. 311 */ 312static void rfkill_set_block(struct rfkill *rfkill, bool blocked) 313{ 314 unsigned long flags; 315 bool prev, curr; 316 int err; 317 318 if (unlikely(rfkill->dev.power.power_state.event & PM_EVENT_SLEEP)) 319 return; 320 321 /* 322 * Some platforms (...!) generate input events which affect the 323 * _hard_ kill state -- whenever something tries to change the 324 * current software state query the hardware state too. 325 */ 326 if (rfkill->ops->query) 327 rfkill->ops->query(rfkill, rfkill->data); 328 329 spin_lock_irqsave(&rfkill->lock, flags); 330 prev = rfkill->state & RFKILL_BLOCK_SW; 331 332 if (prev) 333 rfkill->state |= RFKILL_BLOCK_SW_PREV; 334 else 335 rfkill->state &= ~RFKILL_BLOCK_SW_PREV; 336 337 if (blocked) 338 rfkill->state |= RFKILL_BLOCK_SW; 339 else 340 rfkill->state &= ~RFKILL_BLOCK_SW; 341 342 rfkill->state |= RFKILL_BLOCK_SW_SETCALL; 343 spin_unlock_irqrestore(&rfkill->lock, flags); 344 345 err = rfkill->ops->set_block(rfkill->data, blocked); 346 347 spin_lock_irqsave(&rfkill->lock, flags); 348 if (err) { 349 /* 350 * Failed -- reset status to _PREV, which may be different 351 * from what we have set _PREV to earlier in this function 352 * if rfkill_set_sw_state was invoked. 353 */ 354 if (rfkill->state & RFKILL_BLOCK_SW_PREV) 355 rfkill->state |= RFKILL_BLOCK_SW; 356 else 357 rfkill->state &= ~RFKILL_BLOCK_SW; 358 } 359 rfkill->state &= ~RFKILL_BLOCK_SW_SETCALL; 360 rfkill->state &= ~RFKILL_BLOCK_SW_PREV; 361 curr = rfkill->state & RFKILL_BLOCK_SW; 362 spin_unlock_irqrestore(&rfkill->lock, flags); 363 364 rfkill_led_trigger_event(rfkill); 365 rfkill_global_led_trigger_event(); 366 367 if (prev != curr) 368 rfkill_event(rfkill); 369} 370 371static void rfkill_update_global_state(enum rfkill_type type, bool blocked) 372{ 373 int i; 374 375 if (type != RFKILL_TYPE_ALL) { 376 rfkill_global_states[type].cur = blocked; 377 return; 378 } 379 380 for (i = 0; i < NUM_RFKILL_TYPES; i++) 381 rfkill_global_states[i].cur = blocked; 382} 383 384#ifdef CONFIG_RFKILL_INPUT 385static atomic_t rfkill_input_disabled = ATOMIC_INIT(0); 386 387/** 388 * __rfkill_switch_all - Toggle state of all switches of given type 389 * @type: type of interfaces to be affected 390 * @blocked: the new state 391 * 392 * This function sets the state of all switches of given type, 393 * unless a specific switch is suspended. 394 * 395 * Caller must have acquired rfkill_global_mutex. 396 */ 397static void __rfkill_switch_all(const enum rfkill_type type, bool blocked) 398{ 399 struct rfkill *rfkill; 400 401 rfkill_update_global_state(type, blocked); 402 list_for_each_entry(rfkill, &rfkill_list, node) { 403 if (rfkill->type != type && type != RFKILL_TYPE_ALL) 404 continue; 405 406 rfkill_set_block(rfkill, blocked); 407 } 408} 409 410/** 411 * rfkill_switch_all - Toggle state of all switches of given type 412 * @type: type of interfaces to be affected 413 * @blocked: the new state 414 * 415 * Acquires rfkill_global_mutex and calls __rfkill_switch_all(@type, @state). 416 * Please refer to __rfkill_switch_all() for details. 417 * 418 * Does nothing if the EPO lock is active. 419 */ 420void rfkill_switch_all(enum rfkill_type type, bool blocked) 421{ 422 if (atomic_read(&rfkill_input_disabled)) 423 return; 424 425 mutex_lock(&rfkill_global_mutex); 426 427 if (!rfkill_epo_lock_active) 428 __rfkill_switch_all(type, blocked); 429 430 mutex_unlock(&rfkill_global_mutex); 431} 432 433/** 434 * rfkill_epo - emergency power off all transmitters 435 * 436 * This kicks all non-suspended rfkill devices to RFKILL_STATE_SOFT_BLOCKED, 437 * ignoring everything in its path but rfkill_global_mutex and rfkill->mutex. 438 * 439 * The global state before the EPO is saved and can be restored later 440 * using rfkill_restore_states(). 441 */ 442void rfkill_epo(void) 443{ 444 struct rfkill *rfkill; 445 int i; 446 447 if (atomic_read(&rfkill_input_disabled)) 448 return; 449 450 mutex_lock(&rfkill_global_mutex); 451 452 rfkill_epo_lock_active = true; 453 list_for_each_entry(rfkill, &rfkill_list, node) 454 rfkill_set_block(rfkill, true); 455 456 for (i = 0; i < NUM_RFKILL_TYPES; i++) { 457 rfkill_global_states[i].sav = rfkill_global_states[i].cur; 458 rfkill_global_states[i].cur = true; 459 } 460 461 mutex_unlock(&rfkill_global_mutex); 462} 463 464/** 465 * rfkill_restore_states - restore global states 466 * 467 * Restore (and sync switches to) the global state from the 468 * states in rfkill_default_states. This can undo the effects of 469 * a call to rfkill_epo(). 470 */ 471void rfkill_restore_states(void) 472{ 473 int i; 474 475 if (atomic_read(&rfkill_input_disabled)) 476 return; 477 478 mutex_lock(&rfkill_global_mutex); 479 480 rfkill_epo_lock_active = false; 481 for (i = 0; i < NUM_RFKILL_TYPES; i++) 482 __rfkill_switch_all(i, rfkill_global_states[i].sav); 483 mutex_unlock(&rfkill_global_mutex); 484} 485 486/** 487 * rfkill_remove_epo_lock - unlock state changes 488 * 489 * Used by rfkill-input manually unlock state changes, when 490 * the EPO switch is deactivated. 491 */ 492void rfkill_remove_epo_lock(void) 493{ 494 if (atomic_read(&rfkill_input_disabled)) 495 return; 496 497 mutex_lock(&rfkill_global_mutex); 498 rfkill_epo_lock_active = false; 499 mutex_unlock(&rfkill_global_mutex); 500} 501 502/** 503 * rfkill_is_epo_lock_active - returns true EPO is active 504 * 505 * Returns 0 (false) if there is NOT an active EPO condition, 506 * and 1 (true) if there is an active EPO condition, which 507 * locks all radios in one of the BLOCKED states. 508 * 509 * Can be called in atomic context. 510 */ 511bool rfkill_is_epo_lock_active(void) 512{ 513 return rfkill_epo_lock_active; 514} 515 516/** 517 * rfkill_get_global_sw_state - returns global state for a type 518 * @type: the type to get the global state of 519 * 520 * Returns the current global state for a given wireless 521 * device type. 522 */ 523bool rfkill_get_global_sw_state(const enum rfkill_type type) 524{ 525 return rfkill_global_states[type].cur; 526} 527#endif 528 529bool rfkill_set_hw_state_reason(struct rfkill *rfkill, 530 bool blocked, unsigned long reason) 531{ 532 unsigned long flags; 533 bool ret, prev; 534 535 BUG_ON(!rfkill); 536 537 if (WARN(reason & 538 ~(RFKILL_HARD_BLOCK_SIGNAL | RFKILL_HARD_BLOCK_NOT_OWNER), 539 "hw_state reason not supported: 0x%lx", reason)) 540 return blocked; 541 542 spin_lock_irqsave(&rfkill->lock, flags); 543 prev = !!(rfkill->hard_block_reasons & reason); 544 if (blocked) { 545 rfkill->state |= RFKILL_BLOCK_HW; 546 rfkill->hard_block_reasons |= reason; 547 } else { 548 rfkill->hard_block_reasons &= ~reason; 549 if (!rfkill->hard_block_reasons) 550 rfkill->state &= ~RFKILL_BLOCK_HW; 551 } 552 ret = !!(rfkill->state & RFKILL_BLOCK_ANY); 553 spin_unlock_irqrestore(&rfkill->lock, flags); 554 555 rfkill_led_trigger_event(rfkill); 556 rfkill_global_led_trigger_event(); 557 558 if (rfkill->registered && prev != blocked) 559 schedule_work(&rfkill->uevent_work); 560 561 return ret; 562} 563EXPORT_SYMBOL(rfkill_set_hw_state_reason); 564 565static void __rfkill_set_sw_state(struct rfkill *rfkill, bool blocked) 566{ 567 u32 bit = RFKILL_BLOCK_SW; 568 569 /* if in a ops->set_block right now, use other bit */ 570 if (rfkill->state & RFKILL_BLOCK_SW_SETCALL) 571 bit = RFKILL_BLOCK_SW_PREV; 572 573 if (blocked) 574 rfkill->state |= bit; 575 else 576 rfkill->state &= ~bit; 577} 578 579bool rfkill_set_sw_state(struct rfkill *rfkill, bool blocked) 580{ 581 unsigned long flags; 582 bool prev, hwblock; 583 584 BUG_ON(!rfkill); 585 586 spin_lock_irqsave(&rfkill->lock, flags); 587 prev = !!(rfkill->state & RFKILL_BLOCK_SW); 588 __rfkill_set_sw_state(rfkill, blocked); 589 hwblock = !!(rfkill->state & RFKILL_BLOCK_HW); 590 blocked = blocked || hwblock; 591 spin_unlock_irqrestore(&rfkill->lock, flags); 592 593 if (!rfkill->registered) 594 return blocked; 595 596 if (prev != blocked && !hwblock) 597 schedule_work(&rfkill->uevent_work); 598 599 rfkill_led_trigger_event(rfkill); 600 rfkill_global_led_trigger_event(); 601 602 return blocked; 603} 604EXPORT_SYMBOL(rfkill_set_sw_state); 605 606void rfkill_init_sw_state(struct rfkill *rfkill, bool blocked) 607{ 608 unsigned long flags; 609 610 BUG_ON(!rfkill); 611 BUG_ON(rfkill->registered); 612 613 spin_lock_irqsave(&rfkill->lock, flags); 614 __rfkill_set_sw_state(rfkill, blocked); 615 rfkill->persistent = true; 616 spin_unlock_irqrestore(&rfkill->lock, flags); 617} 618EXPORT_SYMBOL(rfkill_init_sw_state); 619 620void rfkill_set_states(struct rfkill *rfkill, bool sw, bool hw) 621{ 622 unsigned long flags; 623 bool swprev, hwprev; 624 625 BUG_ON(!rfkill); 626 627 spin_lock_irqsave(&rfkill->lock, flags); 628 629 /* 630 * No need to care about prev/setblock ... this is for uevent only 631 * and that will get triggered by rfkill_set_block anyway. 632 */ 633 swprev = !!(rfkill->state & RFKILL_BLOCK_SW); 634 hwprev = !!(rfkill->state & RFKILL_BLOCK_HW); 635 __rfkill_set_sw_state(rfkill, sw); 636 if (hw) 637 rfkill->state |= RFKILL_BLOCK_HW; 638 else 639 rfkill->state &= ~RFKILL_BLOCK_HW; 640 641 spin_unlock_irqrestore(&rfkill->lock, flags); 642 643 if (!rfkill->registered) { 644 rfkill->persistent = true; 645 } else { 646 if (swprev != sw || hwprev != hw) 647 schedule_work(&rfkill->uevent_work); 648 649 rfkill_led_trigger_event(rfkill); 650 rfkill_global_led_trigger_event(); 651 } 652} 653EXPORT_SYMBOL(rfkill_set_states); 654 655static const char * const rfkill_types[] = { 656 NULL, /* RFKILL_TYPE_ALL */ 657 "wlan", 658 "bluetooth", 659 "ultrawideband", 660 "wimax", 661 "wwan", 662 "gps", 663 "fm", 664 "nfc", 665}; 666 667enum rfkill_type rfkill_find_type(const char *name) 668{ 669 int i; 670 671 BUILD_BUG_ON(ARRAY_SIZE(rfkill_types) != NUM_RFKILL_TYPES); 672 673 if (!name) 674 return RFKILL_TYPE_ALL; 675 676 for (i = 1; i < NUM_RFKILL_TYPES; i++) 677 if (!strcmp(name, rfkill_types[i])) 678 return i; 679 return RFKILL_TYPE_ALL; 680} 681EXPORT_SYMBOL(rfkill_find_type); 682 683static ssize_t name_show(struct device *dev, struct device_attribute *attr, 684 char *buf) 685{ 686 struct rfkill *rfkill = to_rfkill(dev); 687 688 return sprintf(buf, "%s\n", rfkill->name); 689} 690static DEVICE_ATTR_RO(name); 691 692static ssize_t type_show(struct device *dev, struct device_attribute *attr, 693 char *buf) 694{ 695 struct rfkill *rfkill = to_rfkill(dev); 696 697 return sprintf(buf, "%s\n", rfkill_types[rfkill->type]); 698} 699static DEVICE_ATTR_RO(type); 700 701static ssize_t index_show(struct device *dev, struct device_attribute *attr, 702 char *buf) 703{ 704 struct rfkill *rfkill = to_rfkill(dev); 705 706 return sprintf(buf, "%d\n", rfkill->idx); 707} 708static DEVICE_ATTR_RO(index); 709 710static ssize_t persistent_show(struct device *dev, 711 struct device_attribute *attr, char *buf) 712{ 713 struct rfkill *rfkill = to_rfkill(dev); 714 715 return sprintf(buf, "%d\n", rfkill->persistent); 716} 717static DEVICE_ATTR_RO(persistent); 718 719static ssize_t hard_show(struct device *dev, struct device_attribute *attr, 720 char *buf) 721{ 722 struct rfkill *rfkill = to_rfkill(dev); 723 724 return sprintf(buf, "%d\n", (rfkill->state & RFKILL_BLOCK_HW) ? 1 : 0 ); 725} 726static DEVICE_ATTR_RO(hard); 727 728static ssize_t soft_show(struct device *dev, struct device_attribute *attr, 729 char *buf) 730{ 731 struct rfkill *rfkill = to_rfkill(dev); 732 733 return sprintf(buf, "%d\n", (rfkill->state & RFKILL_BLOCK_SW) ? 1 : 0 ); 734} 735 736static ssize_t soft_store(struct device *dev, struct device_attribute *attr, 737 const char *buf, size_t count) 738{ 739 struct rfkill *rfkill = to_rfkill(dev); 740 unsigned long state; 741 int err; 742 743 if (!capable(CAP_NET_ADMIN)) 744 return -EPERM; 745 746 err = kstrtoul(buf, 0, &state); 747 if (err) 748 return err; 749 750 if (state > 1 ) 751 return -EINVAL; 752 753 mutex_lock(&rfkill_global_mutex); 754 rfkill_set_block(rfkill, state); 755 mutex_unlock(&rfkill_global_mutex); 756 757 return count; 758} 759static DEVICE_ATTR_RW(soft); 760 761static ssize_t hard_block_reasons_show(struct device *dev, 762 struct device_attribute *attr, 763 char *buf) 764{ 765 struct rfkill *rfkill = to_rfkill(dev); 766 767 return sprintf(buf, "0x%lx\n", rfkill->hard_block_reasons); 768} 769static DEVICE_ATTR_RO(hard_block_reasons); 770 771static u8 user_state_from_blocked(unsigned long state) 772{ 773 if (state & RFKILL_BLOCK_HW) 774 return RFKILL_USER_STATE_HARD_BLOCKED; 775 if (state & RFKILL_BLOCK_SW) 776 return RFKILL_USER_STATE_SOFT_BLOCKED; 777 778 return RFKILL_USER_STATE_UNBLOCKED; 779} 780 781static ssize_t state_show(struct device *dev, struct device_attribute *attr, 782 char *buf) 783{ 784 struct rfkill *rfkill = to_rfkill(dev); 785 786 return sprintf(buf, "%d\n", user_state_from_blocked(rfkill->state)); 787} 788 789static ssize_t state_store(struct device *dev, struct device_attribute *attr, 790 const char *buf, size_t count) 791{ 792 struct rfkill *rfkill = to_rfkill(dev); 793 unsigned long state; 794 int err; 795 796 if (!capable(CAP_NET_ADMIN)) 797 return -EPERM; 798 799 err = kstrtoul(buf, 0, &state); 800 if (err) 801 return err; 802 803 if (state != RFKILL_USER_STATE_SOFT_BLOCKED && 804 state != RFKILL_USER_STATE_UNBLOCKED) 805 return -EINVAL; 806 807 mutex_lock(&rfkill_global_mutex); 808 rfkill_set_block(rfkill, state == RFKILL_USER_STATE_SOFT_BLOCKED); 809 mutex_unlock(&rfkill_global_mutex); 810 811 return count; 812} 813static DEVICE_ATTR_RW(state); 814 815static struct attribute *rfkill_dev_attrs[] = { 816 &dev_attr_name.attr, 817 &dev_attr_type.attr, 818 &dev_attr_index.attr, 819 &dev_attr_persistent.attr, 820 &dev_attr_state.attr, 821 &dev_attr_soft.attr, 822 &dev_attr_hard.attr, 823 &dev_attr_hard_block_reasons.attr, 824 NULL, 825}; 826ATTRIBUTE_GROUPS(rfkill_dev); 827 828static void rfkill_release(struct device *dev) 829{ 830 struct rfkill *rfkill = to_rfkill(dev); 831 832 kfree(rfkill); 833} 834 835static int rfkill_dev_uevent(struct device *dev, struct kobj_uevent_env *env) 836{ 837 struct rfkill *rfkill = to_rfkill(dev); 838 unsigned long flags; 839 unsigned long reasons; 840 u32 state; 841 int error; 842 843 error = add_uevent_var(env, "RFKILL_NAME=%s", rfkill->name); 844 if (error) 845 return error; 846 error = add_uevent_var(env, "RFKILL_TYPE=%s", 847 rfkill_types[rfkill->type]); 848 if (error) 849 return error; 850 spin_lock_irqsave(&rfkill->lock, flags); 851 state = rfkill->state; 852 reasons = rfkill->hard_block_reasons; 853 spin_unlock_irqrestore(&rfkill->lock, flags); 854 error = add_uevent_var(env, "RFKILL_STATE=%d", 855 user_state_from_blocked(state)); 856 if (error) 857 return error; 858 return add_uevent_var(env, "RFKILL_HW_BLOCK_REASON=0x%lx", reasons); 859} 860 861void rfkill_pause_polling(struct rfkill *rfkill) 862{ 863 BUG_ON(!rfkill); 864 865 if (!rfkill->ops->poll) 866 return; 867 868 rfkill->polling_paused = true; 869 cancel_delayed_work_sync(&rfkill->poll_work); 870} 871EXPORT_SYMBOL(rfkill_pause_polling); 872 873void rfkill_resume_polling(struct rfkill *rfkill) 874{ 875 BUG_ON(!rfkill); 876 877 if (!rfkill->ops->poll) 878 return; 879 880 rfkill->polling_paused = false; 881 882 if (rfkill->suspended) 883 return; 884 885 queue_delayed_work(system_power_efficient_wq, 886 &rfkill->poll_work, 0); 887} 888EXPORT_SYMBOL(rfkill_resume_polling); 889 890#ifdef CONFIG_PM_SLEEP 891static int rfkill_suspend(struct device *dev) 892{ 893 struct rfkill *rfkill = to_rfkill(dev); 894 895 rfkill->suspended = true; 896 cancel_delayed_work_sync(&rfkill->poll_work); 897 898 return 0; 899} 900 901static int rfkill_resume(struct device *dev) 902{ 903 struct rfkill *rfkill = to_rfkill(dev); 904 bool cur; 905 906 rfkill->suspended = false; 907 908 if (!rfkill->registered) 909 return 0; 910 911 if (!rfkill->persistent) { 912 cur = !!(rfkill->state & RFKILL_BLOCK_SW); 913 rfkill_set_block(rfkill, cur); 914 } 915 916 if (rfkill->ops->poll && !rfkill->polling_paused) 917 queue_delayed_work(system_power_efficient_wq, 918 &rfkill->poll_work, 0); 919 920 return 0; 921} 922 923static SIMPLE_DEV_PM_OPS(rfkill_pm_ops, rfkill_suspend, rfkill_resume); 924#define RFKILL_PM_OPS (&rfkill_pm_ops) 925#else 926#define RFKILL_PM_OPS NULL 927#endif 928 929static struct class rfkill_class = { 930 .name = "rfkill", 931 .dev_release = rfkill_release, 932 .dev_groups = rfkill_dev_groups, 933 .dev_uevent = rfkill_dev_uevent, 934 .pm = RFKILL_PM_OPS, 935}; 936 937bool rfkill_blocked(struct rfkill *rfkill) 938{ 939 unsigned long flags; 940 u32 state; 941 942 spin_lock_irqsave(&rfkill->lock, flags); 943 state = rfkill->state; 944 spin_unlock_irqrestore(&rfkill->lock, flags); 945 946 return !!(state & RFKILL_BLOCK_ANY); 947} 948EXPORT_SYMBOL(rfkill_blocked); 949 950bool rfkill_soft_blocked(struct rfkill *rfkill) 951{ 952 unsigned long flags; 953 u32 state; 954 955 spin_lock_irqsave(&rfkill->lock, flags); 956 state = rfkill->state; 957 spin_unlock_irqrestore(&rfkill->lock, flags); 958 959 return !!(state & RFKILL_BLOCK_SW); 960} 961EXPORT_SYMBOL(rfkill_soft_blocked); 962 963struct rfkill * __must_check rfkill_alloc(const char *name, 964 struct device *parent, 965 const enum rfkill_type type, 966 const struct rfkill_ops *ops, 967 void *ops_data) 968{ 969 struct rfkill *rfkill; 970 struct device *dev; 971 972 if (WARN_ON(!ops)) 973 return NULL; 974 975 if (WARN_ON(!ops->set_block)) 976 return NULL; 977 978 if (WARN_ON(!name)) 979 return NULL; 980 981 if (WARN_ON(type == RFKILL_TYPE_ALL || type >= NUM_RFKILL_TYPES)) 982 return NULL; 983 984 rfkill = kzalloc(sizeof(*rfkill) + strlen(name) + 1, GFP_KERNEL); 985 if (!rfkill) 986 return NULL; 987 988 spin_lock_init(&rfkill->lock); 989 INIT_LIST_HEAD(&rfkill->node); 990 rfkill->type = type; 991 strcpy(rfkill->name, name); 992 rfkill->ops = ops; 993 rfkill->data = ops_data; 994 995 dev = &rfkill->dev; 996 dev->class = &rfkill_class; 997 dev->parent = parent; 998 device_initialize(dev); 999 1000 return rfkill; 1001} 1002EXPORT_SYMBOL(rfkill_alloc); 1003 1004static void rfkill_poll(struct work_struct *work) 1005{ 1006 struct rfkill *rfkill; 1007 1008 rfkill = container_of(work, struct rfkill, poll_work.work); 1009 1010 /* 1011 * Poll hardware state -- driver will use one of the 1012 * rfkill_set{,_hw,_sw}_state functions and use its 1013 * return value to update the current status. 1014 */ 1015 rfkill->ops->poll(rfkill, rfkill->data); 1016 1017 queue_delayed_work(system_power_efficient_wq, 1018 &rfkill->poll_work, 1019 round_jiffies_relative(POLL_INTERVAL)); 1020} 1021 1022static void rfkill_uevent_work(struct work_struct *work) 1023{ 1024 struct rfkill *rfkill; 1025 1026 rfkill = container_of(work, struct rfkill, uevent_work); 1027 1028 mutex_lock(&rfkill_global_mutex); 1029 rfkill_event(rfkill); 1030 mutex_unlock(&rfkill_global_mutex); 1031} 1032 1033static void rfkill_sync_work(struct work_struct *work) 1034{ 1035 struct rfkill *rfkill; 1036 bool cur; 1037 1038 rfkill = container_of(work, struct rfkill, sync_work); 1039 1040 mutex_lock(&rfkill_global_mutex); 1041 cur = rfkill_global_states[rfkill->type].cur; 1042 rfkill_set_block(rfkill, cur); 1043 mutex_unlock(&rfkill_global_mutex); 1044} 1045 1046int __must_check rfkill_register(struct rfkill *rfkill) 1047{ 1048 static unsigned long rfkill_no; 1049 struct device *dev; 1050 int error; 1051 1052 if (!rfkill) 1053 return -EINVAL; 1054 1055 dev = &rfkill->dev; 1056 1057 mutex_lock(&rfkill_global_mutex); 1058 1059 if (rfkill->registered) { 1060 error = -EALREADY; 1061 goto unlock; 1062 } 1063 1064 rfkill->idx = rfkill_no; 1065 dev_set_name(dev, "rfkill%lu", rfkill_no); 1066 rfkill_no++; 1067 1068 list_add_tail(&rfkill->node, &rfkill_list); 1069 1070 error = device_add(dev); 1071 if (error) 1072 goto remove; 1073 1074 error = rfkill_led_trigger_register(rfkill); 1075 if (error) 1076 goto devdel; 1077 1078 rfkill->registered = true; 1079 1080 INIT_DELAYED_WORK(&rfkill->poll_work, rfkill_poll); 1081 INIT_WORK(&rfkill->uevent_work, rfkill_uevent_work); 1082 INIT_WORK(&rfkill->sync_work, rfkill_sync_work); 1083 1084 if (rfkill->ops->poll) 1085 queue_delayed_work(system_power_efficient_wq, 1086 &rfkill->poll_work, 1087 round_jiffies_relative(POLL_INTERVAL)); 1088 1089 if (!rfkill->persistent || rfkill_epo_lock_active) { 1090 schedule_work(&rfkill->sync_work); 1091 } else { 1092#ifdef CONFIG_RFKILL_INPUT 1093 bool soft_blocked = !!(rfkill->state & RFKILL_BLOCK_SW); 1094 1095 if (!atomic_read(&rfkill_input_disabled)) 1096 __rfkill_switch_all(rfkill->type, soft_blocked); 1097#endif 1098 } 1099 1100 rfkill_global_led_trigger_event(); 1101 rfkill_send_events(rfkill, RFKILL_OP_ADD); 1102 1103 mutex_unlock(&rfkill_global_mutex); 1104 return 0; 1105 1106 devdel: 1107 device_del(&rfkill->dev); 1108 remove: 1109 list_del_init(&rfkill->node); 1110 unlock: 1111 mutex_unlock(&rfkill_global_mutex); 1112 return error; 1113} 1114EXPORT_SYMBOL(rfkill_register); 1115 1116void rfkill_unregister(struct rfkill *rfkill) 1117{ 1118 BUG_ON(!rfkill); 1119 1120 if (rfkill->ops->poll) 1121 cancel_delayed_work_sync(&rfkill->poll_work); 1122 1123 cancel_work_sync(&rfkill->uevent_work); 1124 cancel_work_sync(&rfkill->sync_work); 1125 1126 rfkill->registered = false; 1127 1128 device_del(&rfkill->dev); 1129 1130 mutex_lock(&rfkill_global_mutex); 1131 rfkill_send_events(rfkill, RFKILL_OP_DEL); 1132 list_del_init(&rfkill->node); 1133 rfkill_global_led_trigger_event(); 1134 mutex_unlock(&rfkill_global_mutex); 1135 1136 rfkill_led_trigger_unregister(rfkill); 1137} 1138EXPORT_SYMBOL(rfkill_unregister); 1139 1140void rfkill_destroy(struct rfkill *rfkill) 1141{ 1142 if (rfkill) 1143 put_device(&rfkill->dev); 1144} 1145EXPORT_SYMBOL(rfkill_destroy); 1146 1147static int rfkill_fop_open(struct inode *inode, struct file *file) 1148{ 1149 struct rfkill_data *data; 1150 struct rfkill *rfkill; 1151 struct rfkill_int_event *ev, *tmp; 1152 1153 data = kzalloc(sizeof(*data), GFP_KERNEL); 1154 if (!data) 1155 return -ENOMEM; 1156 1157 data->max_size = RFKILL_EVENT_SIZE_V1; 1158 1159 INIT_LIST_HEAD(&data->events); 1160 mutex_init(&data->mtx); 1161 init_waitqueue_head(&data->read_wait); 1162 1163 mutex_lock(&rfkill_global_mutex); 1164 mutex_lock(&data->mtx); 1165 /* 1166 * start getting events from elsewhere but hold mtx to get 1167 * startup events added first 1168 */ 1169 1170 list_for_each_entry(rfkill, &rfkill_list, node) { 1171 ev = kzalloc(sizeof(*ev), GFP_KERNEL); 1172 if (!ev) 1173 goto free; 1174 rfkill_fill_event(&ev->ev, rfkill, RFKILL_OP_ADD); 1175 list_add_tail(&ev->list, &data->events); 1176 } 1177 list_add(&data->list, &rfkill_fds); 1178 mutex_unlock(&data->mtx); 1179 mutex_unlock(&rfkill_global_mutex); 1180 1181 file->private_data = data; 1182 1183 return stream_open(inode, file); 1184 1185 free: 1186 mutex_unlock(&data->mtx); 1187 mutex_unlock(&rfkill_global_mutex); 1188 mutex_destroy(&data->mtx); 1189 list_for_each_entry_safe(ev, tmp, &data->events, list) 1190 kfree(ev); 1191 kfree(data); 1192 return -ENOMEM; 1193} 1194 1195static __poll_t rfkill_fop_poll(struct file *file, poll_table *wait) 1196{ 1197 struct rfkill_data *data = file->private_data; 1198 __poll_t res = EPOLLOUT | EPOLLWRNORM; 1199 1200 poll_wait(file, &data->read_wait, wait); 1201 1202 mutex_lock(&data->mtx); 1203 if (!list_empty(&data->events)) 1204 res = EPOLLIN | EPOLLRDNORM; 1205 mutex_unlock(&data->mtx); 1206 1207 return res; 1208} 1209 1210static ssize_t rfkill_fop_read(struct file *file, char __user *buf, 1211 size_t count, loff_t *pos) 1212{ 1213 struct rfkill_data *data = file->private_data; 1214 struct rfkill_int_event *ev; 1215 unsigned long sz; 1216 int ret; 1217 1218 mutex_lock(&data->mtx); 1219 1220 while (list_empty(&data->events)) { 1221 if (file->f_flags & O_NONBLOCK) { 1222 ret = -EAGAIN; 1223 goto out; 1224 } 1225 mutex_unlock(&data->mtx); 1226 /* since we re-check and it just compares pointers, 1227 * using !list_empty() without locking isn't a problem 1228 */ 1229 ret = wait_event_interruptible(data->read_wait, 1230 !list_empty(&data->events)); 1231 mutex_lock(&data->mtx); 1232 1233 if (ret) 1234 goto out; 1235 } 1236 1237 ev = list_first_entry(&data->events, struct rfkill_int_event, 1238 list); 1239 1240 sz = min_t(unsigned long, sizeof(ev->ev), count); 1241 sz = min_t(unsigned long, sz, data->max_size); 1242 ret = sz; 1243 if (copy_to_user(buf, &ev->ev, sz)) 1244 ret = -EFAULT; 1245 1246 list_del(&ev->list); 1247 kfree(ev); 1248 out: 1249 mutex_unlock(&data->mtx); 1250 return ret; 1251} 1252 1253static ssize_t rfkill_fop_write(struct file *file, const char __user *buf, 1254 size_t count, loff_t *pos) 1255{ 1256 struct rfkill_data *data = file->private_data; 1257 struct rfkill *rfkill; 1258 struct rfkill_event_ext ev; 1259 int ret; 1260 1261 /* we don't need the 'hard' variable but accept it */ 1262 if (count < RFKILL_EVENT_SIZE_V1 - 1) 1263 return -EINVAL; 1264 1265 /* 1266 * Copy as much data as we can accept into our 'ev' buffer, 1267 * but tell userspace how much we've copied so it can determine 1268 * our API version even in a write() call, if it cares. 1269 */ 1270 count = min(count, sizeof(ev)); 1271 count = min_t(size_t, count, data->max_size); 1272 if (copy_from_user(&ev, buf, count)) 1273 return -EFAULT; 1274 1275 if (ev.type >= NUM_RFKILL_TYPES) 1276 return -EINVAL; 1277 1278 mutex_lock(&rfkill_global_mutex); 1279 1280 switch (ev.op) { 1281 case RFKILL_OP_CHANGE_ALL: 1282 rfkill_update_global_state(ev.type, ev.soft); 1283 list_for_each_entry(rfkill, &rfkill_list, node) 1284 if (rfkill->type == ev.type || 1285 ev.type == RFKILL_TYPE_ALL) 1286 rfkill_set_block(rfkill, ev.soft); 1287 ret = 0; 1288 break; 1289 case RFKILL_OP_CHANGE: 1290 list_for_each_entry(rfkill, &rfkill_list, node) 1291 if (rfkill->idx == ev.idx && 1292 (rfkill->type == ev.type || 1293 ev.type == RFKILL_TYPE_ALL)) 1294 rfkill_set_block(rfkill, ev.soft); 1295 ret = 0; 1296 break; 1297 default: 1298 ret = -EINVAL; 1299 break; 1300 } 1301 1302 mutex_unlock(&rfkill_global_mutex); 1303 1304 return ret ?: count; 1305} 1306 1307static int rfkill_fop_release(struct inode *inode, struct file *file) 1308{ 1309 struct rfkill_data *data = file->private_data; 1310 struct rfkill_int_event *ev, *tmp; 1311 1312 mutex_lock(&rfkill_global_mutex); 1313 list_del(&data->list); 1314 mutex_unlock(&rfkill_global_mutex); 1315 1316 mutex_destroy(&data->mtx); 1317 list_for_each_entry_safe(ev, tmp, &data->events, list) 1318 kfree(ev); 1319 1320#ifdef CONFIG_RFKILL_INPUT 1321 if (data->input_handler) 1322 if (atomic_dec_return(&rfkill_input_disabled) == 0) 1323 printk(KERN_DEBUG "rfkill: input handler enabled\n"); 1324#endif 1325 1326 kfree(data); 1327 1328 return 0; 1329} 1330 1331static long rfkill_fop_ioctl(struct file *file, unsigned int cmd, 1332 unsigned long arg) 1333{ 1334 struct rfkill_data *data = file->private_data; 1335 int ret = -ENOSYS; 1336 u32 size; 1337 1338 if (_IOC_TYPE(cmd) != RFKILL_IOC_MAGIC) 1339 return -ENOSYS; 1340 1341 mutex_lock(&data->mtx); 1342 switch (_IOC_NR(cmd)) { 1343#ifdef CONFIG_RFKILL_INPUT 1344 case RFKILL_IOC_NOINPUT: 1345 if (!data->input_handler) { 1346 if (atomic_inc_return(&rfkill_input_disabled) == 1) 1347 printk(KERN_DEBUG "rfkill: input handler disabled\n"); 1348 data->input_handler = true; 1349 } 1350 ret = 0; 1351 break; 1352#endif 1353 case RFKILL_IOC_MAX_SIZE: 1354 if (get_user(size, (__u32 __user *)arg)) { 1355 ret = -EFAULT; 1356 break; 1357 } 1358 if (size < RFKILL_EVENT_SIZE_V1 || size > U8_MAX) { 1359 ret = -EINVAL; 1360 break; 1361 } 1362 data->max_size = size; 1363 ret = 0; 1364 break; 1365 default: 1366 break; 1367 } 1368 mutex_unlock(&data->mtx); 1369 1370 return ret; 1371} 1372 1373static const struct file_operations rfkill_fops = { 1374 .owner = THIS_MODULE, 1375 .open = rfkill_fop_open, 1376 .read = rfkill_fop_read, 1377 .write = rfkill_fop_write, 1378 .poll = rfkill_fop_poll, 1379 .release = rfkill_fop_release, 1380 .unlocked_ioctl = rfkill_fop_ioctl, 1381 .compat_ioctl = compat_ptr_ioctl, 1382 .llseek = no_llseek, 1383}; 1384 1385#define RFKILL_NAME "rfkill" 1386 1387static struct miscdevice rfkill_miscdev = { 1388 .fops = &rfkill_fops, 1389 .name = RFKILL_NAME, 1390 .minor = RFKILL_MINOR, 1391}; 1392 1393static int __init rfkill_init(void) 1394{ 1395 int error; 1396 1397 rfkill_update_global_state(RFKILL_TYPE_ALL, !rfkill_default_state); 1398 1399 error = class_register(&rfkill_class); 1400 if (error) 1401 goto error_class; 1402 1403 error = misc_register(&rfkill_miscdev); 1404 if (error) 1405 goto error_misc; 1406 1407 error = rfkill_global_led_trigger_register(); 1408 if (error) 1409 goto error_led_trigger; 1410 1411#ifdef CONFIG_RFKILL_INPUT 1412 error = rfkill_handler_init(); 1413 if (error) 1414 goto error_input; 1415#endif 1416 1417 return 0; 1418 1419#ifdef CONFIG_RFKILL_INPUT 1420error_input: 1421 rfkill_global_led_trigger_unregister(); 1422#endif 1423error_led_trigger: 1424 misc_deregister(&rfkill_miscdev); 1425error_misc: 1426 class_unregister(&rfkill_class); 1427error_class: 1428 return error; 1429} 1430subsys_initcall(rfkill_init); 1431 1432static void __exit rfkill_exit(void) 1433{ 1434#ifdef CONFIG_RFKILL_INPUT 1435 rfkill_handler_exit(); 1436#endif 1437 rfkill_global_led_trigger_unregister(); 1438 misc_deregister(&rfkill_miscdev); 1439 class_unregister(&rfkill_class); 1440} 1441module_exit(rfkill_exit); 1442 1443MODULE_ALIAS_MISCDEV(RFKILL_MINOR); 1444MODULE_ALIAS("devname:" RFKILL_NAME);