]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * linux/kernel/posix-timers.c | |
3 | * | |
4 | * | |
5 | * 2002-10-15 Posix Clocks & timers | |
6 | * by George Anzinger george@mvista.com | |
7 | * | |
8 | * Copyright (C) 2002 2003 by MontaVista Software. | |
9 | * | |
10 | * 2004-06-01 Fix CLOCK_REALTIME clock/timer TIMER_ABSTIME bug. | |
11 | * Copyright (C) 2004 Boris Hu | |
12 | * | |
13 | * This program is free software; you can redistribute it and/or modify | |
14 | * it under the terms of the GNU General Public License as published by | |
15 | * the Free Software Foundation; either version 2 of the License, or (at | |
16 | * your option) any later version. | |
17 | * | |
18 | * This program is distributed in the hope that it will be useful, but | |
19 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
20 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
21 | * General Public License for more details. | |
22 | ||
23 | * You should have received a copy of the GNU General Public License | |
24 | * along with this program; if not, write to the Free Software | |
25 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
26 | * | |
27 | * MontaVista Software | 1237 East Arques Avenue | Sunnyvale | CA 94085 | USA | |
28 | */ | |
29 | ||
30 | /* These are all the functions necessary to implement | |
31 | * POSIX clocks & timers | |
32 | */ | |
33 | #include <linux/mm.h> | |
34 | #include <linux/interrupt.h> | |
35 | #include <linux/slab.h> | |
36 | #include <linux/time.h> | |
37 | #include <linux/mutex.h> | |
38 | ||
39 | #include <asm/uaccess.h> | |
40 | #include <linux/list.h> | |
41 | #include <linux/init.h> | |
42 | #include <linux/compiler.h> | |
43 | #include <linux/idr.h> | |
44 | #include <linux/posix-timers.h> | |
45 | #include <linux/syscalls.h> | |
46 | #include <linux/wait.h> | |
47 | #include <linux/workqueue.h> | |
48 | #include <linux/module.h> | |
49 | ||
50 | /* | |
51 | * Management arrays for POSIX timers. Timers are kept in slab memory | |
52 | * Timer ids are allocated by an external routine that keeps track of the | |
53 | * id and the timer. The external interface is: | |
54 | * | |
55 | * void *idr_find(struct idr *idp, int id); to find timer_id <id> | |
56 | * int idr_get_new(struct idr *idp, void *ptr); to get a new id and | |
57 | * related it to <ptr> | |
58 | * void idr_remove(struct idr *idp, int id); to release <id> | |
59 | * void idr_init(struct idr *idp); to initialize <idp> | |
60 | * which we supply. | |
61 | * The idr_get_new *may* call slab for more memory so it must not be | |
62 | * called under a spin lock. Likewise idr_remore may release memory | |
63 | * (but it may be ok to do this under a lock...). | |
64 | * idr_find is just a memory look up and is quite fast. A -1 return | |
65 | * indicates that the requested id does not exist. | |
66 | */ | |
67 | ||
68 | /* | |
69 | * Lets keep our timers in a slab cache :-) | |
70 | */ | |
71 | static struct kmem_cache *posix_timers_cache; | |
72 | static struct idr posix_timers_id; | |
73 | static DEFINE_SPINLOCK(idr_lock); | |
74 | ||
75 | /* | |
76 | * we assume that the new SIGEV_THREAD_ID shares no bits with the other | |
77 | * SIGEV values. Here we put out an error if this assumption fails. | |
78 | */ | |
79 | #if SIGEV_THREAD_ID != (SIGEV_THREAD_ID & \ | |
80 | ~(SIGEV_SIGNAL | SIGEV_NONE | SIGEV_THREAD)) | |
81 | #error "SIGEV_THREAD_ID must not share bit with other SIGEV values!" | |
82 | #endif | |
83 | ||
84 | ||
85 | /* | |
86 | * The timer ID is turned into a timer address by idr_find(). | |
87 | * Verifying a valid ID consists of: | |
88 | * | |
89 | * a) checking that idr_find() returns other than -1. | |
90 | * b) checking that the timer id matches the one in the timer itself. | |
91 | * c) that the timer owner is in the callers thread group. | |
92 | */ | |
93 | ||
94 | /* | |
95 | * CLOCKs: The POSIX standard calls for a couple of clocks and allows us | |
96 | * to implement others. This structure defines the various | |
97 | * clocks and allows the possibility of adding others. We | |
98 | * provide an interface to add clocks to the table and expect | |
99 | * the "arch" code to add at least one clock that is high | |
100 | * resolution. Here we define the standard CLOCK_REALTIME as a | |
101 | * 1/HZ resolution clock. | |
102 | * | |
103 | * RESOLUTION: Clock resolution is used to round up timer and interval | |
104 | * times, NOT to report clock times, which are reported with as | |
105 | * much resolution as the system can muster. In some cases this | |
106 | * resolution may depend on the underlying clock hardware and | |
107 | * may not be quantifiable until run time, and only then is the | |
108 | * necessary code is written. The standard says we should say | |
109 | * something about this issue in the documentation... | |
110 | * | |
111 | * FUNCTIONS: The CLOCKs structure defines possible functions to handle | |
112 | * various clock functions. For clocks that use the standard | |
113 | * system timer code these entries should be NULL. This will | |
114 | * allow dispatch without the overhead of indirect function | |
115 | * calls. CLOCKS that depend on other sources (e.g. WWV or GPS) | |
116 | * must supply functions here, even if the function just returns | |
117 | * ENOSYS. The standard POSIX timer management code assumes the | |
118 | * following: 1.) The k_itimer struct (sched.h) is used for the | |
119 | * timer. 2.) The list, it_lock, it_clock, it_id and it_pid | |
120 | * fields are not modified by timer code. | |
121 | * | |
122 | * At this time all functions EXCEPT clock_nanosleep can be | |
123 | * redirected by the CLOCKS structure. Clock_nanosleep is in | |
124 | * there, but the code ignores it. | |
125 | * | |
126 | * Permissions: It is assumed that the clock_settime() function defined | |
127 | * for each clock will take care of permission checks. Some | |
128 | * clocks may be set able by any user (i.e. local process | |
129 | * clocks) others not. Currently the only set able clock we | |
130 | * have is CLOCK_REALTIME and its high res counter part, both of | |
131 | * which we beg off on and pass to do_sys_settimeofday(). | |
132 | */ | |
133 | ||
134 | static struct k_clock posix_clocks[MAX_CLOCKS]; | |
135 | ||
136 | /* | |
137 | * These ones are defined below. | |
138 | */ | |
139 | static int common_nsleep(const clockid_t, int flags, struct timespec *t, | |
140 | struct timespec __user *rmtp); | |
141 | static void common_timer_get(struct k_itimer *, struct itimerspec *); | |
142 | static int common_timer_set(struct k_itimer *, int, | |
143 | struct itimerspec *, struct itimerspec *); | |
144 | static int common_timer_del(struct k_itimer *timer); | |
145 | ||
146 | static enum hrtimer_restart posix_timer_fn(struct hrtimer *data); | |
147 | ||
148 | static struct k_itimer *lock_timer(timer_t timer_id, unsigned long *flags); | |
149 | ||
150 | static inline void unlock_timer(struct k_itimer *timr, unsigned long flags) | |
151 | { | |
152 | spin_unlock_irqrestore(&timr->it_lock, flags); | |
153 | } | |
154 | ||
155 | /* | |
156 | * Call the k_clock hook function if non-null, or the default function. | |
157 | */ | |
158 | #define CLOCK_DISPATCH(clock, call, arglist) \ | |
159 | ((clock) < 0 ? posix_cpu_##call arglist : \ | |
160 | (posix_clocks[clock].call != NULL \ | |
161 | ? (*posix_clocks[clock].call) arglist : common_##call arglist)) | |
162 | ||
163 | /* | |
164 | * Default clock hook functions when the struct k_clock passed | |
165 | * to register_posix_clock leaves a function pointer null. | |
166 | * | |
167 | * The function common_CALL is the default implementation for | |
168 | * the function pointer CALL in struct k_clock. | |
169 | */ | |
170 | ||
171 | static inline int common_clock_getres(const clockid_t which_clock, | |
172 | struct timespec *tp) | |
173 | { | |
174 | tp->tv_sec = 0; | |
175 | tp->tv_nsec = posix_clocks[which_clock].res; | |
176 | return 0; | |
177 | } | |
178 | ||
179 | /* | |
180 | * Get real time for posix timers | |
181 | */ | |
182 | static int common_clock_get(clockid_t which_clock, struct timespec *tp) | |
183 | { | |
184 | ktime_get_real_ts(tp); | |
185 | return 0; | |
186 | } | |
187 | ||
188 | static inline int common_clock_set(const clockid_t which_clock, | |
189 | struct timespec *tp) | |
190 | { | |
191 | return do_sys_settimeofday(tp, NULL); | |
192 | } | |
193 | ||
194 | static int common_timer_create(struct k_itimer *new_timer) | |
195 | { | |
196 | hrtimer_init(&new_timer->it.real.timer, new_timer->it_clock, 0); | |
197 | return 0; | |
198 | } | |
199 | ||
200 | static int no_timer_create(struct k_itimer *new_timer) | |
201 | { | |
202 | return -EOPNOTSUPP; | |
203 | } | |
204 | ||
205 | static int no_nsleep(const clockid_t which_clock, int flags, | |
206 | struct timespec *tsave, struct timespec __user *rmtp) | |
207 | { | |
208 | return -EOPNOTSUPP; | |
209 | } | |
210 | ||
211 | /* | |
212 | * Return nonzero if we know a priori this clockid_t value is bogus. | |
213 | */ | |
214 | static inline int invalid_clockid(const clockid_t which_clock) | |
215 | { | |
216 | if (which_clock < 0) /* CPU clock, posix_cpu_* will check it */ | |
217 | return 0; | |
218 | if ((unsigned) which_clock >= MAX_CLOCKS) | |
219 | return 1; | |
220 | if (posix_clocks[which_clock].clock_getres != NULL) | |
221 | return 0; | |
222 | if (posix_clocks[which_clock].res != 0) | |
223 | return 0; | |
224 | return 1; | |
225 | } | |
226 | ||
227 | /* | |
228 | * Get monotonic time for posix timers | |
229 | */ | |
230 | static int posix_ktime_get_ts(clockid_t which_clock, struct timespec *tp) | |
231 | { | |
232 | ktime_get_ts(tp); | |
233 | return 0; | |
234 | } | |
235 | ||
236 | /* | |
237 | * Get monotonic time for posix timers | |
238 | */ | |
239 | static int posix_get_monotonic_raw(clockid_t which_clock, struct timespec *tp) | |
240 | { | |
241 | getrawmonotonic(tp); | |
242 | return 0; | |
243 | } | |
244 | ||
245 | ||
246 | static int posix_get_realtime_coarse(clockid_t which_clock, struct timespec *tp) | |
247 | { | |
248 | *tp = current_kernel_time(); | |
249 | return 0; | |
250 | } | |
251 | ||
252 | static int posix_get_monotonic_coarse(clockid_t which_clock, | |
253 | struct timespec *tp) | |
254 | { | |
255 | *tp = get_monotonic_coarse(); | |
256 | return 0; | |
257 | } | |
258 | ||
259 | static int posix_get_coarse_res(const clockid_t which_clock, struct timespec *tp) | |
260 | { | |
261 | *tp = ktime_to_timespec(KTIME_LOW_RES); | |
262 | return 0; | |
263 | } | |
264 | /* | |
265 | * Initialize everything, well, just everything in Posix clocks/timers ;) | |
266 | */ | |
267 | static __init int init_posix_timers(void) | |
268 | { | |
269 | struct k_clock clock_realtime = { | |
270 | .clock_getres = hrtimer_get_res, | |
271 | }; | |
272 | struct k_clock clock_monotonic = { | |
273 | .clock_getres = hrtimer_get_res, | |
274 | .clock_get = posix_ktime_get_ts, | |
275 | .clock_set = do_posix_clock_nosettime, | |
276 | }; | |
277 | struct k_clock clock_monotonic_raw = { | |
278 | .clock_getres = hrtimer_get_res, | |
279 | .clock_get = posix_get_monotonic_raw, | |
280 | .clock_set = do_posix_clock_nosettime, | |
281 | .timer_create = no_timer_create, | |
282 | .nsleep = no_nsleep, | |
283 | }; | |
284 | struct k_clock clock_realtime_coarse = { | |
285 | .clock_getres = posix_get_coarse_res, | |
286 | .clock_get = posix_get_realtime_coarse, | |
287 | .clock_set = do_posix_clock_nosettime, | |
288 | .timer_create = no_timer_create, | |
289 | .nsleep = no_nsleep, | |
290 | }; | |
291 | struct k_clock clock_monotonic_coarse = { | |
292 | .clock_getres = posix_get_coarse_res, | |
293 | .clock_get = posix_get_monotonic_coarse, | |
294 | .clock_set = do_posix_clock_nosettime, | |
295 | .timer_create = no_timer_create, | |
296 | .nsleep = no_nsleep, | |
297 | }; | |
298 | ||
299 | register_posix_clock(CLOCK_REALTIME, &clock_realtime); | |
300 | register_posix_clock(CLOCK_MONOTONIC, &clock_monotonic); | |
301 | register_posix_clock(CLOCK_MONOTONIC_RAW, &clock_monotonic_raw); | |
302 | register_posix_clock(CLOCK_REALTIME_COARSE, &clock_realtime_coarse); | |
303 | register_posix_clock(CLOCK_MONOTONIC_COARSE, &clock_monotonic_coarse); | |
304 | ||
305 | posix_timers_cache = kmem_cache_create("posix_timers_cache", | |
306 | sizeof (struct k_itimer), 0, SLAB_PANIC, | |
307 | NULL); | |
308 | idr_init(&posix_timers_id); | |
309 | return 0; | |
310 | } | |
311 | ||
312 | __initcall(init_posix_timers); | |
313 | ||
314 | static void schedule_next_timer(struct k_itimer *timr) | |
315 | { | |
316 | struct hrtimer *timer = &timr->it.real.timer; | |
317 | ||
318 | if (timr->it.real.interval.tv64 == 0) | |
319 | return; | |
320 | ||
321 | timr->it_overrun += (unsigned int) hrtimer_forward(timer, | |
322 | timer->base->get_time(), | |
323 | timr->it.real.interval); | |
324 | ||
325 | timr->it_overrun_last = timr->it_overrun; | |
326 | timr->it_overrun = -1; | |
327 | ++timr->it_requeue_pending; | |
328 | hrtimer_restart(timer); | |
329 | } | |
330 | ||
331 | /* | |
332 | * This function is exported for use by the signal deliver code. It is | |
333 | * called just prior to the info block being released and passes that | |
334 | * block to us. It's function is to update the overrun entry AND to | |
335 | * restart the timer. It should only be called if the timer is to be | |
336 | * restarted (i.e. we have flagged this in the sys_private entry of the | |
337 | * info block). | |
338 | * | |
339 | * To protect aginst the timer going away while the interrupt is queued, | |
340 | * we require that the it_requeue_pending flag be set. | |
341 | */ | |
342 | void do_schedule_next_timer(struct siginfo *info) | |
343 | { | |
344 | struct k_itimer *timr; | |
345 | unsigned long flags; | |
346 | ||
347 | timr = lock_timer(info->si_tid, &flags); | |
348 | ||
349 | if (timr && timr->it_requeue_pending == info->si_sys_private) { | |
350 | if (timr->it_clock < 0) | |
351 | posix_cpu_timer_schedule(timr); | |
352 | else | |
353 | schedule_next_timer(timr); | |
354 | ||
355 | info->si_overrun += timr->it_overrun_last; | |
356 | } | |
357 | ||
358 | if (timr) | |
359 | unlock_timer(timr, flags); | |
360 | } | |
361 | ||
362 | int posix_timer_event(struct k_itimer *timr, int si_private) | |
363 | { | |
364 | struct task_struct *task; | |
365 | int shared, ret = -1; | |
366 | /* | |
367 | * FIXME: if ->sigq is queued we can race with | |
368 | * dequeue_signal()->do_schedule_next_timer(). | |
369 | * | |
370 | * If dequeue_signal() sees the "right" value of | |
371 | * si_sys_private it calls do_schedule_next_timer(). | |
372 | * We re-queue ->sigq and drop ->it_lock(). | |
373 | * do_schedule_next_timer() locks the timer | |
374 | * and re-schedules it while ->sigq is pending. | |
375 | * Not really bad, but not that we want. | |
376 | */ | |
377 | timr->sigq->info.si_sys_private = si_private; | |
378 | ||
379 | rcu_read_lock(); | |
380 | task = pid_task(timr->it_pid, PIDTYPE_PID); | |
381 | if (task) { | |
382 | shared = !(timr->it_sigev_notify & SIGEV_THREAD_ID); | |
383 | ret = send_sigqueue(timr->sigq, task, shared); | |
384 | } | |
385 | rcu_read_unlock(); | |
386 | /* If we failed to send the signal the timer stops. */ | |
387 | return ret > 0; | |
388 | } | |
389 | EXPORT_SYMBOL_GPL(posix_timer_event); | |
390 | ||
391 | /* | |
392 | * This function gets called when a POSIX.1b interval timer expires. It | |
393 | * is used as a callback from the kernel internal timer. The | |
394 | * run_timer_list code ALWAYS calls with interrupts on. | |
395 | ||
396 | * This code is for CLOCK_REALTIME* and CLOCK_MONOTONIC* timers. | |
397 | */ | |
398 | static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer) | |
399 | { | |
400 | struct k_itimer *timr; | |
401 | unsigned long flags; | |
402 | int si_private = 0; | |
403 | enum hrtimer_restart ret = HRTIMER_NORESTART; | |
404 | ||
405 | timr = container_of(timer, struct k_itimer, it.real.timer); | |
406 | spin_lock_irqsave(&timr->it_lock, flags); | |
407 | ||
408 | if (timr->it.real.interval.tv64 != 0) | |
409 | si_private = ++timr->it_requeue_pending; | |
410 | ||
411 | if (posix_timer_event(timr, si_private)) { | |
412 | /* | |
413 | * signal was not sent because of sig_ignor | |
414 | * we will not get a call back to restart it AND | |
415 | * it should be restarted. | |
416 | */ | |
417 | if (timr->it.real.interval.tv64 != 0) { | |
418 | ktime_t now = hrtimer_cb_get_time(timer); | |
419 | ||
420 | /* | |
421 | * FIXME: What we really want, is to stop this | |
422 | * timer completely and restart it in case the | |
423 | * SIG_IGN is removed. This is a non trivial | |
424 | * change which involves sighand locking | |
425 | * (sigh !), which we don't want to do late in | |
426 | * the release cycle. | |
427 | * | |
428 | * For now we just let timers with an interval | |
429 | * less than a jiffie expire every jiffie to | |
430 | * avoid softirq starvation in case of SIG_IGN | |
431 | * and a very small interval, which would put | |
432 | * the timer right back on the softirq pending | |
433 | * list. By moving now ahead of time we trick | |
434 | * hrtimer_forward() to expire the timer | |
435 | * later, while we still maintain the overrun | |
436 | * accuracy, but have some inconsistency in | |
437 | * the timer_gettime() case. This is at least | |
438 | * better than a starved softirq. A more | |
439 | * complex fix which solves also another related | |
440 | * inconsistency is already in the pipeline. | |
441 | */ | |
442 | #ifdef CONFIG_HIGH_RES_TIMERS | |
443 | { | |
444 | ktime_t kj = ktime_set(0, NSEC_PER_SEC / HZ); | |
445 | ||
446 | if (timr->it.real.interval.tv64 < kj.tv64) | |
447 | now = ktime_add(now, kj); | |
448 | } | |
449 | #endif | |
450 | timr->it_overrun += (unsigned int) | |
451 | hrtimer_forward(timer, now, | |
452 | timr->it.real.interval); | |
453 | ret = HRTIMER_RESTART; | |
454 | ++timr->it_requeue_pending; | |
455 | } | |
456 | } | |
457 | ||
458 | unlock_timer(timr, flags); | |
459 | return ret; | |
460 | } | |
461 | ||
462 | static struct pid *good_sigevent(sigevent_t * event) | |
463 | { | |
464 | struct task_struct *rtn = current->group_leader; | |
465 | ||
466 | if ((event->sigev_notify & SIGEV_THREAD_ID ) && | |
467 | (!(rtn = find_task_by_vpid(event->sigev_notify_thread_id)) || | |
468 | !same_thread_group(rtn, current) || | |
469 | (event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_SIGNAL)) | |
470 | return NULL; | |
471 | ||
472 | if (((event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) && | |
473 | ((event->sigev_signo <= 0) || (event->sigev_signo > SIGRTMAX))) | |
474 | return NULL; | |
475 | ||
476 | return task_pid(rtn); | |
477 | } | |
478 | ||
479 | void register_posix_clock(const clockid_t clock_id, struct k_clock *new_clock) | |
480 | { | |
481 | if ((unsigned) clock_id >= MAX_CLOCKS) { | |
482 | printk("POSIX clock register failed for clock_id %d\n", | |
483 | clock_id); | |
484 | return; | |
485 | } | |
486 | ||
487 | posix_clocks[clock_id] = *new_clock; | |
488 | } | |
489 | EXPORT_SYMBOL_GPL(register_posix_clock); | |
490 | ||
491 | static struct k_itimer * alloc_posix_timer(void) | |
492 | { | |
493 | struct k_itimer *tmr; | |
494 | tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL); | |
495 | if (!tmr) | |
496 | return tmr; | |
497 | if (unlikely(!(tmr->sigq = sigqueue_alloc()))) { | |
498 | kmem_cache_free(posix_timers_cache, tmr); | |
499 | return NULL; | |
500 | } | |
501 | memset(&tmr->sigq->info, 0, sizeof(siginfo_t)); | |
502 | return tmr; | |
503 | } | |
504 | ||
505 | #define IT_ID_SET 1 | |
506 | #define IT_ID_NOT_SET 0 | |
507 | static void release_posix_timer(struct k_itimer *tmr, int it_id_set) | |
508 | { | |
509 | if (it_id_set) { | |
510 | unsigned long flags; | |
511 | spin_lock_irqsave(&idr_lock, flags); | |
512 | idr_remove(&posix_timers_id, tmr->it_id); | |
513 | spin_unlock_irqrestore(&idr_lock, flags); | |
514 | } | |
515 | put_pid(tmr->it_pid); | |
516 | sigqueue_free(tmr->sigq); | |
517 | kmem_cache_free(posix_timers_cache, tmr); | |
518 | } | |
519 | ||
520 | /* Create a POSIX.1b interval timer. */ | |
521 | ||
522 | SYSCALL_DEFINE3(timer_create, const clockid_t, which_clock, | |
523 | struct sigevent __user *, timer_event_spec, | |
524 | timer_t __user *, created_timer_id) | |
525 | { | |
526 | struct k_itimer *new_timer; | |
527 | int error, new_timer_id; | |
528 | sigevent_t event; | |
529 | int it_id_set = IT_ID_NOT_SET; | |
530 | ||
531 | if (invalid_clockid(which_clock)) | |
532 | return -EINVAL; | |
533 | ||
534 | new_timer = alloc_posix_timer(); | |
535 | if (unlikely(!new_timer)) | |
536 | return -EAGAIN; | |
537 | ||
538 | spin_lock_init(&new_timer->it_lock); | |
539 | retry: | |
540 | if (unlikely(!idr_pre_get(&posix_timers_id, GFP_KERNEL))) { | |
541 | error = -EAGAIN; | |
542 | goto out; | |
543 | } | |
544 | spin_lock_irq(&idr_lock); | |
545 | error = idr_get_new(&posix_timers_id, new_timer, &new_timer_id); | |
546 | spin_unlock_irq(&idr_lock); | |
547 | if (error) { | |
548 | if (error == -EAGAIN) | |
549 | goto retry; | |
550 | /* | |
551 | * Weird looking, but we return EAGAIN if the IDR is | |
552 | * full (proper POSIX return value for this) | |
553 | */ | |
554 | error = -EAGAIN; | |
555 | goto out; | |
556 | } | |
557 | ||
558 | it_id_set = IT_ID_SET; | |
559 | new_timer->it_id = (timer_t) new_timer_id; | |
560 | new_timer->it_clock = which_clock; | |
561 | new_timer->it_overrun = -1; | |
562 | ||
563 | if (timer_event_spec) { | |
564 | if (copy_from_user(&event, timer_event_spec, sizeof (event))) { | |
565 | error = -EFAULT; | |
566 | goto out; | |
567 | } | |
568 | rcu_read_lock(); | |
569 | new_timer->it_pid = get_pid(good_sigevent(&event)); | |
570 | rcu_read_unlock(); | |
571 | if (!new_timer->it_pid) { | |
572 | error = -EINVAL; | |
573 | goto out; | |
574 | } | |
575 | } else { | |
576 | event.sigev_notify = SIGEV_SIGNAL; | |
577 | event.sigev_signo = SIGALRM; | |
578 | event.sigev_value.sival_int = new_timer->it_id; | |
579 | new_timer->it_pid = get_pid(task_tgid(current)); | |
580 | } | |
581 | ||
582 | new_timer->it_sigev_notify = event.sigev_notify; | |
583 | new_timer->sigq->info.si_signo = event.sigev_signo; | |
584 | new_timer->sigq->info.si_value = event.sigev_value; | |
585 | new_timer->sigq->info.si_tid = new_timer->it_id; | |
586 | new_timer->sigq->info.si_code = SI_TIMER; | |
587 | ||
588 | if (copy_to_user(created_timer_id, | |
589 | &new_timer_id, sizeof (new_timer_id))) { | |
590 | error = -EFAULT; | |
591 | goto out; | |
592 | } | |
593 | ||
594 | error = CLOCK_DISPATCH(which_clock, timer_create, (new_timer)); | |
595 | if (error) | |
596 | goto out; | |
597 | ||
598 | spin_lock_irq(¤t->sighand->siglock); | |
599 | new_timer->it_signal = current->signal; | |
600 | list_add(&new_timer->list, ¤t->signal->posix_timers); | |
601 | spin_unlock_irq(¤t->sighand->siglock); | |
602 | ||
603 | return 0; | |
604 | /* | |
605 | * In the case of the timer belonging to another task, after | |
606 | * the task is unlocked, the timer is owned by the other task | |
607 | * and may cease to exist at any time. Don't use or modify | |
608 | * new_timer after the unlock call. | |
609 | */ | |
610 | out: | |
611 | release_posix_timer(new_timer, it_id_set); | |
612 | return error; | |
613 | } | |
614 | ||
615 | /* | |
616 | * Locking issues: We need to protect the result of the id look up until | |
617 | * we get the timer locked down so it is not deleted under us. The | |
618 | * removal is done under the idr spinlock so we use that here to bridge | |
619 | * the find to the timer lock. To avoid a dead lock, the timer id MUST | |
620 | * be release with out holding the timer lock. | |
621 | */ | |
622 | static struct k_itimer *lock_timer(timer_t timer_id, unsigned long *flags) | |
623 | { | |
624 | struct k_itimer *timr; | |
625 | /* | |
626 | * Watch out here. We do a irqsave on the idr_lock and pass the | |
627 | * flags part over to the timer lock. Must not let interrupts in | |
628 | * while we are moving the lock. | |
629 | */ | |
630 | spin_lock_irqsave(&idr_lock, *flags); | |
631 | timr = idr_find(&posix_timers_id, (int)timer_id); | |
632 | if (timr) { | |
633 | spin_lock(&timr->it_lock); | |
634 | if (timr->it_signal == current->signal) { | |
635 | spin_unlock(&idr_lock); | |
636 | return timr; | |
637 | } | |
638 | spin_unlock(&timr->it_lock); | |
639 | } | |
640 | spin_unlock_irqrestore(&idr_lock, *flags); | |
641 | ||
642 | return NULL; | |
643 | } | |
644 | ||
645 | /* | |
646 | * Get the time remaining on a POSIX.1b interval timer. This function | |
647 | * is ALWAYS called with spin_lock_irq on the timer, thus it must not | |
648 | * mess with irq. | |
649 | * | |
650 | * We have a couple of messes to clean up here. First there is the case | |
651 | * of a timer that has a requeue pending. These timers should appear to | |
652 | * be in the timer list with an expiry as if we were to requeue them | |
653 | * now. | |
654 | * | |
655 | * The second issue is the SIGEV_NONE timer which may be active but is | |
656 | * not really ever put in the timer list (to save system resources). | |
657 | * This timer may be expired, and if so, we will do it here. Otherwise | |
658 | * it is the same as a requeue pending timer WRT to what we should | |
659 | * report. | |
660 | */ | |
661 | static void | |
662 | common_timer_get(struct k_itimer *timr, struct itimerspec *cur_setting) | |
663 | { | |
664 | ktime_t now, remaining, iv; | |
665 | struct hrtimer *timer = &timr->it.real.timer; | |
666 | ||
667 | memset(cur_setting, 0, sizeof(struct itimerspec)); | |
668 | ||
669 | iv = timr->it.real.interval; | |
670 | ||
671 | /* interval timer ? */ | |
672 | if (iv.tv64) | |
673 | cur_setting->it_interval = ktime_to_timespec(iv); | |
674 | else if (!hrtimer_active(timer) && | |
675 | (timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) | |
676 | return; | |
677 | ||
678 | now = timer->base->get_time(); | |
679 | ||
680 | /* | |
681 | * When a requeue is pending or this is a SIGEV_NONE | |
682 | * timer move the expiry time forward by intervals, so | |
683 | * expiry is > now. | |
684 | */ | |
685 | if (iv.tv64 && (timr->it_requeue_pending & REQUEUE_PENDING || | |
686 | (timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE)) | |
687 | timr->it_overrun += (unsigned int) hrtimer_forward(timer, now, iv); | |
688 | ||
689 | remaining = ktime_sub(hrtimer_get_expires(timer), now); | |
690 | /* Return 0 only, when the timer is expired and not pending */ | |
691 | if (remaining.tv64 <= 0) { | |
692 | /* | |
693 | * A single shot SIGEV_NONE timer must return 0, when | |
694 | * it is expired ! | |
695 | */ | |
696 | if ((timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) | |
697 | cur_setting->it_value.tv_nsec = 1; | |
698 | } else | |
699 | cur_setting->it_value = ktime_to_timespec(remaining); | |
700 | } | |
701 | ||
702 | /* Get the time remaining on a POSIX.1b interval timer. */ | |
703 | SYSCALL_DEFINE2(timer_gettime, timer_t, timer_id, | |
704 | struct itimerspec __user *, setting) | |
705 | { | |
706 | struct k_itimer *timr; | |
707 | struct itimerspec cur_setting; | |
708 | unsigned long flags; | |
709 | ||
710 | timr = lock_timer(timer_id, &flags); | |
711 | if (!timr) | |
712 | return -EINVAL; | |
713 | ||
714 | CLOCK_DISPATCH(timr->it_clock, timer_get, (timr, &cur_setting)); | |
715 | ||
716 | unlock_timer(timr, flags); | |
717 | ||
718 | if (copy_to_user(setting, &cur_setting, sizeof (cur_setting))) | |
719 | return -EFAULT; | |
720 | ||
721 | return 0; | |
722 | } | |
723 | ||
724 | /* | |
725 | * Get the number of overruns of a POSIX.1b interval timer. This is to | |
726 | * be the overrun of the timer last delivered. At the same time we are | |
727 | * accumulating overruns on the next timer. The overrun is frozen when | |
728 | * the signal is delivered, either at the notify time (if the info block | |
729 | * is not queued) or at the actual delivery time (as we are informed by | |
730 | * the call back to do_schedule_next_timer(). So all we need to do is | |
731 | * to pick up the frozen overrun. | |
732 | */ | |
733 | SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id) | |
734 | { | |
735 | struct k_itimer *timr; | |
736 | int overrun; | |
737 | unsigned long flags; | |
738 | ||
739 | timr = lock_timer(timer_id, &flags); | |
740 | if (!timr) | |
741 | return -EINVAL; | |
742 | ||
743 | overrun = timr->it_overrun_last; | |
744 | unlock_timer(timr, flags); | |
745 | ||
746 | return overrun; | |
747 | } | |
748 | ||
749 | /* Set a POSIX.1b interval timer. */ | |
750 | /* timr->it_lock is taken. */ | |
751 | static int | |
752 | common_timer_set(struct k_itimer *timr, int flags, | |
753 | struct itimerspec *new_setting, struct itimerspec *old_setting) | |
754 | { | |
755 | struct hrtimer *timer = &timr->it.real.timer; | |
756 | enum hrtimer_mode mode; | |
757 | ||
758 | if (old_setting) | |
759 | common_timer_get(timr, old_setting); | |
760 | ||
761 | /* disable the timer */ | |
762 | timr->it.real.interval.tv64 = 0; | |
763 | /* | |
764 | * careful here. If smp we could be in the "fire" routine which will | |
765 | * be spinning as we hold the lock. But this is ONLY an SMP issue. | |
766 | */ | |
767 | if (hrtimer_try_to_cancel(timer) < 0) | |
768 | return TIMER_RETRY; | |
769 | ||
770 | timr->it_requeue_pending = (timr->it_requeue_pending + 2) & | |
771 | ~REQUEUE_PENDING; | |
772 | timr->it_overrun_last = 0; | |
773 | ||
774 | /* switch off the timer when it_value is zero */ | |
775 | if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec) | |
776 | return 0; | |
777 | ||
778 | mode = flags & TIMER_ABSTIME ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL; | |
779 | hrtimer_init(&timr->it.real.timer, timr->it_clock, mode); | |
780 | timr->it.real.timer.function = posix_timer_fn; | |
781 | ||
782 | hrtimer_set_expires(timer, timespec_to_ktime(new_setting->it_value)); | |
783 | ||
784 | /* Convert interval */ | |
785 | timr->it.real.interval = timespec_to_ktime(new_setting->it_interval); | |
786 | ||
787 | /* SIGEV_NONE timers are not queued ! See common_timer_get */ | |
788 | if (((timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE)) { | |
789 | /* Setup correct expiry time for relative timers */ | |
790 | if (mode == HRTIMER_MODE_REL) { | |
791 | hrtimer_add_expires(timer, timer->base->get_time()); | |
792 | } | |
793 | return 0; | |
794 | } | |
795 | ||
796 | hrtimer_start_expires(timer, mode); | |
797 | return 0; | |
798 | } | |
799 | ||
800 | /* Set a POSIX.1b interval timer */ | |
801 | SYSCALL_DEFINE4(timer_settime, timer_t, timer_id, int, flags, | |
802 | const struct itimerspec __user *, new_setting, | |
803 | struct itimerspec __user *, old_setting) | |
804 | { | |
805 | struct k_itimer *timr; | |
806 | struct itimerspec new_spec, old_spec; | |
807 | int error = 0; | |
808 | unsigned long flag; | |
809 | struct itimerspec *rtn = old_setting ? &old_spec : NULL; | |
810 | ||
811 | if (!new_setting) | |
812 | return -EINVAL; | |
813 | ||
814 | if (copy_from_user(&new_spec, new_setting, sizeof (new_spec))) | |
815 | return -EFAULT; | |
816 | ||
817 | if (!timespec_valid(&new_spec.it_interval) || | |
818 | !timespec_valid(&new_spec.it_value)) | |
819 | return -EINVAL; | |
820 | retry: | |
821 | timr = lock_timer(timer_id, &flag); | |
822 | if (!timr) | |
823 | return -EINVAL; | |
824 | ||
825 | error = CLOCK_DISPATCH(timr->it_clock, timer_set, | |
826 | (timr, flags, &new_spec, rtn)); | |
827 | ||
828 | unlock_timer(timr, flag); | |
829 | if (error == TIMER_RETRY) { | |
830 | rtn = NULL; // We already got the old time... | |
831 | goto retry; | |
832 | } | |
833 | ||
834 | if (old_setting && !error && | |
835 | copy_to_user(old_setting, &old_spec, sizeof (old_spec))) | |
836 | error = -EFAULT; | |
837 | ||
838 | return error; | |
839 | } | |
840 | ||
841 | static inline int common_timer_del(struct k_itimer *timer) | |
842 | { | |
843 | timer->it.real.interval.tv64 = 0; | |
844 | ||
845 | if (hrtimer_try_to_cancel(&timer->it.real.timer) < 0) | |
846 | return TIMER_RETRY; | |
847 | return 0; | |
848 | } | |
849 | ||
850 | static inline int timer_delete_hook(struct k_itimer *timer) | |
851 | { | |
852 | return CLOCK_DISPATCH(timer->it_clock, timer_del, (timer)); | |
853 | } | |
854 | ||
855 | /* Delete a POSIX.1b interval timer. */ | |
856 | SYSCALL_DEFINE1(timer_delete, timer_t, timer_id) | |
857 | { | |
858 | struct k_itimer *timer; | |
859 | unsigned long flags; | |
860 | ||
861 | retry_delete: | |
862 | timer = lock_timer(timer_id, &flags); | |
863 | if (!timer) | |
864 | return -EINVAL; | |
865 | ||
866 | if (timer_delete_hook(timer) == TIMER_RETRY) { | |
867 | unlock_timer(timer, flags); | |
868 | goto retry_delete; | |
869 | } | |
870 | ||
871 | spin_lock(¤t->sighand->siglock); | |
872 | list_del(&timer->list); | |
873 | spin_unlock(¤t->sighand->siglock); | |
874 | /* | |
875 | * This keeps any tasks waiting on the spin lock from thinking | |
876 | * they got something (see the lock code above). | |
877 | */ | |
878 | timer->it_signal = NULL; | |
879 | ||
880 | unlock_timer(timer, flags); | |
881 | release_posix_timer(timer, IT_ID_SET); | |
882 | return 0; | |
883 | } | |
884 | ||
885 | /* | |
886 | * return timer owned by the process, used by exit_itimers | |
887 | */ | |
888 | static void itimer_delete(struct k_itimer *timer) | |
889 | { | |
890 | unsigned long flags; | |
891 | ||
892 | retry_delete: | |
893 | spin_lock_irqsave(&timer->it_lock, flags); | |
894 | ||
895 | if (timer_delete_hook(timer) == TIMER_RETRY) { | |
896 | unlock_timer(timer, flags); | |
897 | goto retry_delete; | |
898 | } | |
899 | list_del(&timer->list); | |
900 | /* | |
901 | * This keeps any tasks waiting on the spin lock from thinking | |
902 | * they got something (see the lock code above). | |
903 | */ | |
904 | timer->it_signal = NULL; | |
905 | ||
906 | unlock_timer(timer, flags); | |
907 | release_posix_timer(timer, IT_ID_SET); | |
908 | } | |
909 | ||
910 | /* | |
911 | * This is called by do_exit or de_thread, only when there are no more | |
912 | * references to the shared signal_struct. | |
913 | */ | |
914 | void exit_itimers(struct signal_struct *sig) | |
915 | { | |
916 | struct k_itimer *tmr; | |
917 | ||
918 | while (!list_empty(&sig->posix_timers)) { | |
919 | tmr = list_entry(sig->posix_timers.next, struct k_itimer, list); | |
920 | itimer_delete(tmr); | |
921 | } | |
922 | } | |
923 | ||
924 | /* Not available / possible... functions */ | |
925 | int do_posix_clock_nosettime(const clockid_t clockid, struct timespec *tp) | |
926 | { | |
927 | return -EINVAL; | |
928 | } | |
929 | EXPORT_SYMBOL_GPL(do_posix_clock_nosettime); | |
930 | ||
931 | int do_posix_clock_nonanosleep(const clockid_t clock, int flags, | |
932 | struct timespec *t, struct timespec __user *r) | |
933 | { | |
934 | #ifndef ENOTSUP | |
935 | return -EOPNOTSUPP; /* aka ENOTSUP in userland for POSIX */ | |
936 | #else /* parisc does define it separately. */ | |
937 | return -ENOTSUP; | |
938 | #endif | |
939 | } | |
940 | EXPORT_SYMBOL_GPL(do_posix_clock_nonanosleep); | |
941 | ||
942 | SYSCALL_DEFINE2(clock_settime, const clockid_t, which_clock, | |
943 | const struct timespec __user *, tp) | |
944 | { | |
945 | struct timespec new_tp; | |
946 | ||
947 | if (invalid_clockid(which_clock)) | |
948 | return -EINVAL; | |
949 | if (copy_from_user(&new_tp, tp, sizeof (*tp))) | |
950 | return -EFAULT; | |
951 | ||
952 | return CLOCK_DISPATCH(which_clock, clock_set, (which_clock, &new_tp)); | |
953 | } | |
954 | ||
955 | SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock, | |
956 | struct timespec __user *,tp) | |
957 | { | |
958 | struct timespec kernel_tp; | |
959 | int error; | |
960 | ||
961 | if (invalid_clockid(which_clock)) | |
962 | return -EINVAL; | |
963 | error = CLOCK_DISPATCH(which_clock, clock_get, | |
964 | (which_clock, &kernel_tp)); | |
965 | if (!error && copy_to_user(tp, &kernel_tp, sizeof (kernel_tp))) | |
966 | error = -EFAULT; | |
967 | ||
968 | return error; | |
969 | ||
970 | } | |
971 | ||
972 | SYSCALL_DEFINE2(clock_getres, const clockid_t, which_clock, | |
973 | struct timespec __user *, tp) | |
974 | { | |
975 | struct timespec rtn_tp; | |
976 | int error; | |
977 | ||
978 | if (invalid_clockid(which_clock)) | |
979 | return -EINVAL; | |
980 | ||
981 | error = CLOCK_DISPATCH(which_clock, clock_getres, | |
982 | (which_clock, &rtn_tp)); | |
983 | ||
984 | if (!error && tp && copy_to_user(tp, &rtn_tp, sizeof (rtn_tp))) { | |
985 | error = -EFAULT; | |
986 | } | |
987 | ||
988 | return error; | |
989 | } | |
990 | ||
991 | /* | |
992 | * nanosleep for monotonic and realtime clocks | |
993 | */ | |
994 | static int common_nsleep(const clockid_t which_clock, int flags, | |
995 | struct timespec *tsave, struct timespec __user *rmtp) | |
996 | { | |
997 | return hrtimer_nanosleep(tsave, rmtp, flags & TIMER_ABSTIME ? | |
998 | HRTIMER_MODE_ABS : HRTIMER_MODE_REL, | |
999 | which_clock); | |
1000 | } | |
1001 | ||
1002 | SYSCALL_DEFINE4(clock_nanosleep, const clockid_t, which_clock, int, flags, | |
1003 | const struct timespec __user *, rqtp, | |
1004 | struct timespec __user *, rmtp) | |
1005 | { | |
1006 | struct timespec t; | |
1007 | ||
1008 | if (invalid_clockid(which_clock)) | |
1009 | return -EINVAL; | |
1010 | ||
1011 | if (copy_from_user(&t, rqtp, sizeof (struct timespec))) | |
1012 | return -EFAULT; | |
1013 | ||
1014 | if (!timespec_valid(&t)) | |
1015 | return -EINVAL; | |
1016 | ||
1017 | return CLOCK_DISPATCH(which_clock, nsleep, | |
1018 | (which_clock, flags, &t, rmtp)); | |
1019 | } | |
1020 | ||
1021 | /* | |
1022 | * nanosleep_restart for monotonic and realtime clocks | |
1023 | */ | |
1024 | static int common_nsleep_restart(struct restart_block *restart_block) | |
1025 | { | |
1026 | return hrtimer_nanosleep_restart(restart_block); | |
1027 | } | |
1028 | ||
1029 | /* | |
1030 | * This will restart clock_nanosleep. This is required only by | |
1031 | * compat_clock_nanosleep_restart for now. | |
1032 | */ | |
1033 | long | |
1034 | clock_nanosleep_restart(struct restart_block *restart_block) | |
1035 | { | |
1036 | clockid_t which_clock = restart_block->arg0; | |
1037 | ||
1038 | return CLOCK_DISPATCH(which_clock, nsleep_restart, | |
1039 | (restart_block)); | |
1040 | } |