Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
f30c2269 | 2 | * linux/kernel/posix-timers.c |
1da177e4 LT |
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> | |
1da177e4 LT |
34 | #include <linux/interrupt.h> |
35 | #include <linux/slab.h> | |
36 | #include <linux/time.h> | |
97d1f15b | 37 | #include <linux/mutex.h> |
1da177e4 LT |
38 | |
39 | #include <asm/uaccess.h> | |
1da177e4 LT |
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 | ||
1da177e4 LT |
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 | */ | |
e18b890b | 71 | static struct kmem_cache *posix_timers_cache; |
1da177e4 LT |
72 | static struct idr posix_timers_id; |
73 | static DEFINE_SPINLOCK(idr_lock); | |
74 | ||
1da177e4 LT |
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_process | |
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]; | |
becf8b5d | 135 | |
1da177e4 | 136 | /* |
becf8b5d | 137 | * These ones are defined below. |
1da177e4 | 138 | */ |
becf8b5d TG |
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); | |
1da177e4 | 145 | |
c9cb2e3d | 146 | static enum hrtimer_restart posix_timer_fn(struct hrtimer *data); |
1da177e4 LT |
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 | ||
a924b04d | 171 | static inline int common_clock_getres(const clockid_t which_clock, |
1da177e4 LT |
172 | struct timespec *tp) |
173 | { | |
174 | tp->tv_sec = 0; | |
175 | tp->tv_nsec = posix_clocks[which_clock].res; | |
176 | return 0; | |
177 | } | |
178 | ||
becf8b5d TG |
179 | /* |
180 | * Get real time for posix timers | |
181 | */ | |
182 | static int common_clock_get(clockid_t which_clock, struct timespec *tp) | |
1da177e4 | 183 | { |
becf8b5d | 184 | ktime_get_real_ts(tp); |
1da177e4 LT |
185 | return 0; |
186 | } | |
187 | ||
a924b04d TG |
188 | static inline int common_clock_set(const clockid_t which_clock, |
189 | struct timespec *tp) | |
1da177e4 LT |
190 | { |
191 | return do_sys_settimeofday(tp, NULL); | |
192 | } | |
193 | ||
858119e1 | 194 | static int common_timer_create(struct k_itimer *new_timer) |
1da177e4 | 195 | { |
7978672c | 196 | hrtimer_init(&new_timer->it.real.timer, new_timer->it_clock, 0); |
1da177e4 LT |
197 | return 0; |
198 | } | |
199 | ||
200 | /* | |
becf8b5d | 201 | * Return nonzero if we know a priori this clockid_t value is bogus. |
1da177e4 | 202 | */ |
a924b04d | 203 | static inline int invalid_clockid(const clockid_t which_clock) |
1da177e4 LT |
204 | { |
205 | if (which_clock < 0) /* CPU clock, posix_cpu_* will check it */ | |
206 | return 0; | |
207 | if ((unsigned) which_clock >= MAX_CLOCKS) | |
208 | return 1; | |
209 | if (posix_clocks[which_clock].clock_getres != NULL) | |
210 | return 0; | |
1da177e4 LT |
211 | if (posix_clocks[which_clock].res != 0) |
212 | return 0; | |
1da177e4 LT |
213 | return 1; |
214 | } | |
215 | ||
becf8b5d TG |
216 | /* |
217 | * Get monotonic time for posix timers | |
218 | */ | |
219 | static int posix_ktime_get_ts(clockid_t which_clock, struct timespec *tp) | |
220 | { | |
221 | ktime_get_ts(tp); | |
222 | return 0; | |
223 | } | |
1da177e4 LT |
224 | |
225 | /* | |
226 | * Initialize everything, well, just everything in Posix clocks/timers ;) | |
227 | */ | |
228 | static __init int init_posix_timers(void) | |
229 | { | |
becf8b5d TG |
230 | struct k_clock clock_realtime = { |
231 | .clock_getres = hrtimer_get_res, | |
1da177e4 | 232 | }; |
becf8b5d TG |
233 | struct k_clock clock_monotonic = { |
234 | .clock_getres = hrtimer_get_res, | |
235 | .clock_get = posix_ktime_get_ts, | |
236 | .clock_set = do_posix_clock_nosettime, | |
1da177e4 LT |
237 | }; |
238 | ||
239 | register_posix_clock(CLOCK_REALTIME, &clock_realtime); | |
240 | register_posix_clock(CLOCK_MONOTONIC, &clock_monotonic); | |
241 | ||
242 | posix_timers_cache = kmem_cache_create("posix_timers_cache", | |
040b5c6f AD |
243 | sizeof (struct k_itimer), 0, SLAB_PANIC, |
244 | NULL); | |
1da177e4 LT |
245 | idr_init(&posix_timers_id); |
246 | return 0; | |
247 | } | |
248 | ||
249 | __initcall(init_posix_timers); | |
250 | ||
1da177e4 LT |
251 | static void schedule_next_timer(struct k_itimer *timr) |
252 | { | |
44f21475 RZ |
253 | struct hrtimer *timer = &timr->it.real.timer; |
254 | ||
becf8b5d | 255 | if (timr->it.real.interval.tv64 == 0) |
1da177e4 LT |
256 | return; |
257 | ||
4d672e7a DL |
258 | timr->it_overrun += (unsigned int) hrtimer_forward(timer, |
259 | timer->base->get_time(), | |
260 | timr->it.real.interval); | |
44f21475 | 261 | |
1da177e4 LT |
262 | timr->it_overrun_last = timr->it_overrun; |
263 | timr->it_overrun = -1; | |
264 | ++timr->it_requeue_pending; | |
44f21475 | 265 | hrtimer_restart(timer); |
1da177e4 LT |
266 | } |
267 | ||
268 | /* | |
269 | * This function is exported for use by the signal deliver code. It is | |
270 | * called just prior to the info block being released and passes that | |
271 | * block to us. It's function is to update the overrun entry AND to | |
272 | * restart the timer. It should only be called if the timer is to be | |
273 | * restarted (i.e. we have flagged this in the sys_private entry of the | |
274 | * info block). | |
275 | * | |
276 | * To protect aginst the timer going away while the interrupt is queued, | |
277 | * we require that the it_requeue_pending flag be set. | |
278 | */ | |
279 | void do_schedule_next_timer(struct siginfo *info) | |
280 | { | |
281 | struct k_itimer *timr; | |
282 | unsigned long flags; | |
283 | ||
284 | timr = lock_timer(info->si_tid, &flags); | |
285 | ||
becf8b5d TG |
286 | if (timr && timr->it_requeue_pending == info->si_sys_private) { |
287 | if (timr->it_clock < 0) | |
288 | posix_cpu_timer_schedule(timr); | |
289 | else | |
290 | schedule_next_timer(timr); | |
1da177e4 | 291 | |
54da1174 | 292 | info->si_overrun += timr->it_overrun_last; |
becf8b5d TG |
293 | } |
294 | ||
b6557fbc TG |
295 | if (timr) |
296 | unlock_timer(timr, flags); | |
1da177e4 LT |
297 | } |
298 | ||
ba661292 | 299 | int posix_timer_event(struct k_itimer *timr, int si_private) |
1da177e4 | 300 | { |
4aa73611 | 301 | int shared, ret; |
ba661292 ON |
302 | /* |
303 | * FIXME: if ->sigq is queued we can race with | |
304 | * dequeue_signal()->do_schedule_next_timer(). | |
305 | * | |
306 | * If dequeue_signal() sees the "right" value of | |
307 | * si_sys_private it calls do_schedule_next_timer(). | |
308 | * We re-queue ->sigq and drop ->it_lock(). | |
309 | * do_schedule_next_timer() locks the timer | |
310 | * and re-schedules it while ->sigq is pending. | |
311 | * Not really bad, but not that we want. | |
312 | */ | |
1da177e4 | 313 | timr->sigq->info.si_sys_private = si_private; |
1da177e4 LT |
314 | |
315 | timr->sigq->info.si_signo = timr->it_sigev_signo; | |
1da177e4 LT |
316 | timr->sigq->info.si_code = SI_TIMER; |
317 | timr->sigq->info.si_tid = timr->it_id; | |
318 | timr->sigq->info.si_value = timr->it_sigev_value; | |
e752dd6c | 319 | |
4aa73611 ON |
320 | shared = !(timr->it_sigev_notify & SIGEV_THREAD_ID); |
321 | ret = send_sigqueue(timr->sigq, timr->it_process, shared); | |
322 | /* If we failed to send the signal the timer stops. */ | |
323 | return ret > 0; | |
1da177e4 LT |
324 | } |
325 | EXPORT_SYMBOL_GPL(posix_timer_event); | |
326 | ||
327 | /* | |
328 | * This function gets called when a POSIX.1b interval timer expires. It | |
329 | * is used as a callback from the kernel internal timer. The | |
330 | * run_timer_list code ALWAYS calls with interrupts on. | |
331 | ||
332 | * This code is for CLOCK_REALTIME* and CLOCK_MONOTONIC* timers. | |
333 | */ | |
c9cb2e3d | 334 | static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer) |
1da177e4 | 335 | { |
05cfb614 | 336 | struct k_itimer *timr; |
1da177e4 | 337 | unsigned long flags; |
becf8b5d | 338 | int si_private = 0; |
c9cb2e3d | 339 | enum hrtimer_restart ret = HRTIMER_NORESTART; |
1da177e4 | 340 | |
05cfb614 | 341 | timr = container_of(timer, struct k_itimer, it.real.timer); |
1da177e4 | 342 | spin_lock_irqsave(&timr->it_lock, flags); |
1da177e4 | 343 | |
becf8b5d TG |
344 | if (timr->it.real.interval.tv64 != 0) |
345 | si_private = ++timr->it_requeue_pending; | |
1da177e4 | 346 | |
becf8b5d TG |
347 | if (posix_timer_event(timr, si_private)) { |
348 | /* | |
349 | * signal was not sent because of sig_ignor | |
350 | * we will not get a call back to restart it AND | |
351 | * it should be restarted. | |
352 | */ | |
353 | if (timr->it.real.interval.tv64 != 0) { | |
58229a18 TG |
354 | ktime_t now = hrtimer_cb_get_time(timer); |
355 | ||
356 | /* | |
357 | * FIXME: What we really want, is to stop this | |
358 | * timer completely and restart it in case the | |
359 | * SIG_IGN is removed. This is a non trivial | |
360 | * change which involves sighand locking | |
361 | * (sigh !), which we don't want to do late in | |
362 | * the release cycle. | |
363 | * | |
364 | * For now we just let timers with an interval | |
365 | * less than a jiffie expire every jiffie to | |
366 | * avoid softirq starvation in case of SIG_IGN | |
367 | * and a very small interval, which would put | |
368 | * the timer right back on the softirq pending | |
369 | * list. By moving now ahead of time we trick | |
370 | * hrtimer_forward() to expire the timer | |
371 | * later, while we still maintain the overrun | |
372 | * accuracy, but have some inconsistency in | |
373 | * the timer_gettime() case. This is at least | |
374 | * better than a starved softirq. A more | |
375 | * complex fix which solves also another related | |
376 | * inconsistency is already in the pipeline. | |
377 | */ | |
378 | #ifdef CONFIG_HIGH_RES_TIMERS | |
379 | { | |
380 | ktime_t kj = ktime_set(0, NSEC_PER_SEC / HZ); | |
381 | ||
382 | if (timr->it.real.interval.tv64 < kj.tv64) | |
383 | now = ktime_add(now, kj); | |
384 | } | |
385 | #endif | |
4d672e7a | 386 | timr->it_overrun += (unsigned int) |
58229a18 | 387 | hrtimer_forward(timer, now, |
becf8b5d TG |
388 | timr->it.real.interval); |
389 | ret = HRTIMER_RESTART; | |
a0a0c28c | 390 | ++timr->it_requeue_pending; |
1da177e4 | 391 | } |
1da177e4 | 392 | } |
1da177e4 | 393 | |
becf8b5d TG |
394 | unlock_timer(timr, flags); |
395 | return ret; | |
396 | } | |
1da177e4 | 397 | |
858119e1 | 398 | static struct task_struct * good_sigevent(sigevent_t * event) |
1da177e4 LT |
399 | { |
400 | struct task_struct *rtn = current->group_leader; | |
401 | ||
402 | if ((event->sigev_notify & SIGEV_THREAD_ID ) && | |
8dc86af0 | 403 | (!(rtn = find_task_by_vpid(event->sigev_notify_thread_id)) || |
bac0abd6 | 404 | !same_thread_group(rtn, current) || |
1da177e4 LT |
405 | (event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_SIGNAL)) |
406 | return NULL; | |
407 | ||
408 | if (((event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) && | |
409 | ((event->sigev_signo <= 0) || (event->sigev_signo > SIGRTMAX))) | |
410 | return NULL; | |
411 | ||
412 | return rtn; | |
413 | } | |
414 | ||
a924b04d | 415 | void register_posix_clock(const clockid_t clock_id, struct k_clock *new_clock) |
1da177e4 LT |
416 | { |
417 | if ((unsigned) clock_id >= MAX_CLOCKS) { | |
418 | printk("POSIX clock register failed for clock_id %d\n", | |
419 | clock_id); | |
420 | return; | |
421 | } | |
422 | ||
423 | posix_clocks[clock_id] = *new_clock; | |
424 | } | |
425 | EXPORT_SYMBOL_GPL(register_posix_clock); | |
426 | ||
427 | static struct k_itimer * alloc_posix_timer(void) | |
428 | { | |
429 | struct k_itimer *tmr; | |
c3762229 | 430 | tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL); |
1da177e4 LT |
431 | if (!tmr) |
432 | return tmr; | |
1da177e4 LT |
433 | if (unlikely(!(tmr->sigq = sigqueue_alloc()))) { |
434 | kmem_cache_free(posix_timers_cache, tmr); | |
435 | tmr = NULL; | |
436 | } | |
ba661292 | 437 | memset(&tmr->sigq->info, 0, sizeof(siginfo_t)); |
1da177e4 LT |
438 | return tmr; |
439 | } | |
440 | ||
441 | #define IT_ID_SET 1 | |
442 | #define IT_ID_NOT_SET 0 | |
443 | static void release_posix_timer(struct k_itimer *tmr, int it_id_set) | |
444 | { | |
445 | if (it_id_set) { | |
446 | unsigned long flags; | |
447 | spin_lock_irqsave(&idr_lock, flags); | |
448 | idr_remove(&posix_timers_id, tmr->it_id); | |
449 | spin_unlock_irqrestore(&idr_lock, flags); | |
450 | } | |
451 | sigqueue_free(tmr->sigq); | |
1da177e4 LT |
452 | kmem_cache_free(posix_timers_cache, tmr); |
453 | } | |
454 | ||
455 | /* Create a POSIX.1b interval timer. */ | |
456 | ||
457 | asmlinkage long | |
a924b04d | 458 | sys_timer_create(const clockid_t which_clock, |
1da177e4 LT |
459 | struct sigevent __user *timer_event_spec, |
460 | timer_t __user * created_timer_id) | |
461 | { | |
462 | int error = 0; | |
463 | struct k_itimer *new_timer = NULL; | |
464 | int new_timer_id; | |
465 | struct task_struct *process = NULL; | |
466 | unsigned long flags; | |
467 | sigevent_t event; | |
468 | int it_id_set = IT_ID_NOT_SET; | |
469 | ||
470 | if (invalid_clockid(which_clock)) | |
471 | return -EINVAL; | |
472 | ||
473 | new_timer = alloc_posix_timer(); | |
474 | if (unlikely(!new_timer)) | |
475 | return -EAGAIN; | |
476 | ||
477 | spin_lock_init(&new_timer->it_lock); | |
478 | retry: | |
479 | if (unlikely(!idr_pre_get(&posix_timers_id, GFP_KERNEL))) { | |
480 | error = -EAGAIN; | |
481 | goto out; | |
482 | } | |
483 | spin_lock_irq(&idr_lock); | |
becf8b5d | 484 | error = idr_get_new(&posix_timers_id, (void *) new_timer, |
1da177e4 LT |
485 | &new_timer_id); |
486 | spin_unlock_irq(&idr_lock); | |
487 | if (error == -EAGAIN) | |
488 | goto retry; | |
489 | else if (error) { | |
490 | /* | |
0b0a3e7b | 491 | * Weird looking, but we return EAGAIN if the IDR is |
1da177e4 LT |
492 | * full (proper POSIX return value for this) |
493 | */ | |
494 | error = -EAGAIN; | |
495 | goto out; | |
496 | } | |
497 | ||
498 | it_id_set = IT_ID_SET; | |
499 | new_timer->it_id = (timer_t) new_timer_id; | |
500 | new_timer->it_clock = which_clock; | |
501 | new_timer->it_overrun = -1; | |
502 | error = CLOCK_DISPATCH(which_clock, timer_create, (new_timer)); | |
503 | if (error) | |
504 | goto out; | |
505 | ||
506 | /* | |
507 | * return the timer_id now. The next step is hard to | |
508 | * back out if there is an error. | |
509 | */ | |
510 | if (copy_to_user(created_timer_id, | |
511 | &new_timer_id, sizeof (new_timer_id))) { | |
512 | error = -EFAULT; | |
513 | goto out; | |
514 | } | |
515 | if (timer_event_spec) { | |
516 | if (copy_from_user(&event, timer_event_spec, sizeof (event))) { | |
517 | error = -EFAULT; | |
518 | goto out; | |
519 | } | |
520 | new_timer->it_sigev_notify = event.sigev_notify; | |
521 | new_timer->it_sigev_signo = event.sigev_signo; | |
522 | new_timer->it_sigev_value = event.sigev_value; | |
523 | ||
524 | read_lock(&tasklist_lock); | |
525 | if ((process = good_sigevent(&event))) { | |
526 | /* | |
527 | * We may be setting up this process for another | |
528 | * thread. It may be exiting. To catch this | |
529 | * case the we check the PF_EXITING flag. If | |
530 | * the flag is not set, the siglock will catch | |
531 | * him before it is too late (in exit_itimers). | |
532 | * | |
533 | * The exec case is a bit more invloved but easy | |
534 | * to code. If the process is in our thread | |
535 | * group (and it must be or we would not allow | |
536 | * it here) and is doing an exec, it will cause | |
537 | * us to be killed. In this case it will wait | |
538 | * for us to die which means we can finish this | |
539 | * linkage with our last gasp. I.e. no code :) | |
540 | */ | |
541 | spin_lock_irqsave(&process->sighand->siglock, flags); | |
542 | if (!(process->flags & PF_EXITING)) { | |
918fc037 | 543 | get_task_struct(process); |
1da177e4 LT |
544 | new_timer->it_process = process; |
545 | list_add(&new_timer->list, | |
546 | &process->signal->posix_timers); | |
d02479bd | 547 | spin_unlock_irqrestore(&process->sighand->siglock, flags); |
1da177e4 LT |
548 | } else { |
549 | spin_unlock_irqrestore(&process->sighand->siglock, flags); | |
550 | process = NULL; | |
551 | } | |
552 | } | |
553 | read_unlock(&tasklist_lock); | |
554 | if (!process) { | |
555 | error = -EINVAL; | |
556 | goto out; | |
557 | } | |
558 | } else { | |
559 | new_timer->it_sigev_notify = SIGEV_SIGNAL; | |
560 | new_timer->it_sigev_signo = SIGALRM; | |
561 | new_timer->it_sigev_value.sival_int = new_timer->it_id; | |
562 | process = current->group_leader; | |
918fc037 | 563 | get_task_struct(process); |
1da177e4 LT |
564 | spin_lock_irqsave(&process->sighand->siglock, flags); |
565 | new_timer->it_process = process; | |
566 | list_add(&new_timer->list, &process->signal->posix_timers); | |
567 | spin_unlock_irqrestore(&process->sighand->siglock, flags); | |
568 | } | |
569 | ||
570 | /* | |
571 | * In the case of the timer belonging to another task, after | |
572 | * the task is unlocked, the timer is owned by the other task | |
573 | * and may cease to exist at any time. Don't use or modify | |
574 | * new_timer after the unlock call. | |
575 | */ | |
576 | ||
577 | out: | |
578 | if (error) | |
579 | release_posix_timer(new_timer, it_id_set); | |
580 | ||
581 | return error; | |
582 | } | |
583 | ||
1da177e4 LT |
584 | /* |
585 | * Locking issues: We need to protect the result of the id look up until | |
586 | * we get the timer locked down so it is not deleted under us. The | |
587 | * removal is done under the idr spinlock so we use that here to bridge | |
588 | * the find to the timer lock. To avoid a dead lock, the timer id MUST | |
589 | * be release with out holding the timer lock. | |
590 | */ | |
591 | static struct k_itimer * lock_timer(timer_t timer_id, unsigned long *flags) | |
592 | { | |
593 | struct k_itimer *timr; | |
594 | /* | |
595 | * Watch out here. We do a irqsave on the idr_lock and pass the | |
596 | * flags part over to the timer lock. Must not let interrupts in | |
597 | * while we are moving the lock. | |
598 | */ | |
599 | ||
600 | spin_lock_irqsave(&idr_lock, *flags); | |
601 | timr = (struct k_itimer *) idr_find(&posix_timers_id, (int) timer_id); | |
602 | if (timr) { | |
603 | spin_lock(&timr->it_lock); | |
1da177e4 LT |
604 | |
605 | if ((timr->it_id != timer_id) || !(timr->it_process) || | |
bac0abd6 | 606 | !same_thread_group(timr->it_process, current)) { |
179394af TG |
607 | spin_unlock(&timr->it_lock); |
608 | spin_unlock_irqrestore(&idr_lock, *flags); | |
1da177e4 | 609 | timr = NULL; |
179394af TG |
610 | } else |
611 | spin_unlock(&idr_lock); | |
1da177e4 LT |
612 | } else |
613 | spin_unlock_irqrestore(&idr_lock, *flags); | |
614 | ||
615 | return timr; | |
616 | } | |
617 | ||
618 | /* | |
619 | * Get the time remaining on a POSIX.1b interval timer. This function | |
620 | * is ALWAYS called with spin_lock_irq on the timer, thus it must not | |
621 | * mess with irq. | |
622 | * | |
623 | * We have a couple of messes to clean up here. First there is the case | |
624 | * of a timer that has a requeue pending. These timers should appear to | |
625 | * be in the timer list with an expiry as if we were to requeue them | |
626 | * now. | |
627 | * | |
628 | * The second issue is the SIGEV_NONE timer which may be active but is | |
629 | * not really ever put in the timer list (to save system resources). | |
630 | * This timer may be expired, and if so, we will do it here. Otherwise | |
631 | * it is the same as a requeue pending timer WRT to what we should | |
632 | * report. | |
633 | */ | |
634 | static void | |
635 | common_timer_get(struct k_itimer *timr, struct itimerspec *cur_setting) | |
636 | { | |
3b98a532 | 637 | ktime_t now, remaining, iv; |
becf8b5d | 638 | struct hrtimer *timer = &timr->it.real.timer; |
1da177e4 | 639 | |
becf8b5d | 640 | memset(cur_setting, 0, sizeof(struct itimerspec)); |
becf8b5d | 641 | |
3b98a532 RZ |
642 | iv = timr->it.real.interval; |
643 | ||
becf8b5d | 644 | /* interval timer ? */ |
3b98a532 RZ |
645 | if (iv.tv64) |
646 | cur_setting->it_interval = ktime_to_timespec(iv); | |
647 | else if (!hrtimer_active(timer) && | |
648 | (timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) | |
becf8b5d | 649 | return; |
3b98a532 RZ |
650 | |
651 | now = timer->base->get_time(); | |
652 | ||
becf8b5d | 653 | /* |
3b98a532 RZ |
654 | * When a requeue is pending or this is a SIGEV_NONE |
655 | * timer move the expiry time forward by intervals, so | |
656 | * expiry is > now. | |
becf8b5d | 657 | */ |
3b98a532 RZ |
658 | if (iv.tv64 && (timr->it_requeue_pending & REQUEUE_PENDING || |
659 | (timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE)) | |
4d672e7a | 660 | timr->it_overrun += (unsigned int) hrtimer_forward(timer, now, iv); |
3b98a532 RZ |
661 | |
662 | remaining = ktime_sub(timer->expires, now); | |
becf8b5d | 663 | /* Return 0 only, when the timer is expired and not pending */ |
3b98a532 RZ |
664 | if (remaining.tv64 <= 0) { |
665 | /* | |
666 | * A single shot SIGEV_NONE timer must return 0, when | |
667 | * it is expired ! | |
668 | */ | |
669 | if ((timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) | |
670 | cur_setting->it_value.tv_nsec = 1; | |
671 | } else | |
becf8b5d | 672 | cur_setting->it_value = ktime_to_timespec(remaining); |
1da177e4 LT |
673 | } |
674 | ||
675 | /* Get the time remaining on a POSIX.1b interval timer. */ | |
676 | asmlinkage long | |
677 | sys_timer_gettime(timer_t timer_id, struct itimerspec __user *setting) | |
678 | { | |
679 | struct k_itimer *timr; | |
680 | struct itimerspec cur_setting; | |
681 | unsigned long flags; | |
682 | ||
683 | timr = lock_timer(timer_id, &flags); | |
684 | if (!timr) | |
685 | return -EINVAL; | |
686 | ||
687 | CLOCK_DISPATCH(timr->it_clock, timer_get, (timr, &cur_setting)); | |
688 | ||
689 | unlock_timer(timr, flags); | |
690 | ||
691 | if (copy_to_user(setting, &cur_setting, sizeof (cur_setting))) | |
692 | return -EFAULT; | |
693 | ||
694 | return 0; | |
695 | } | |
becf8b5d | 696 | |
1da177e4 LT |
697 | /* |
698 | * Get the number of overruns of a POSIX.1b interval timer. This is to | |
699 | * be the overrun of the timer last delivered. At the same time we are | |
700 | * accumulating overruns on the next timer. The overrun is frozen when | |
701 | * the signal is delivered, either at the notify time (if the info block | |
702 | * is not queued) or at the actual delivery time (as we are informed by | |
703 | * the call back to do_schedule_next_timer(). So all we need to do is | |
704 | * to pick up the frozen overrun. | |
705 | */ | |
1da177e4 LT |
706 | asmlinkage long |
707 | sys_timer_getoverrun(timer_t timer_id) | |
708 | { | |
709 | struct k_itimer *timr; | |
710 | int overrun; | |
5ba25331 | 711 | unsigned long flags; |
1da177e4 LT |
712 | |
713 | timr = lock_timer(timer_id, &flags); | |
714 | if (!timr) | |
715 | return -EINVAL; | |
716 | ||
717 | overrun = timr->it_overrun_last; | |
718 | unlock_timer(timr, flags); | |
719 | ||
720 | return overrun; | |
721 | } | |
1da177e4 LT |
722 | |
723 | /* Set a POSIX.1b interval timer. */ | |
724 | /* timr->it_lock is taken. */ | |
858119e1 | 725 | static int |
1da177e4 LT |
726 | common_timer_set(struct k_itimer *timr, int flags, |
727 | struct itimerspec *new_setting, struct itimerspec *old_setting) | |
728 | { | |
becf8b5d | 729 | struct hrtimer *timer = &timr->it.real.timer; |
7978672c | 730 | enum hrtimer_mode mode; |
1da177e4 LT |
731 | |
732 | if (old_setting) | |
733 | common_timer_get(timr, old_setting); | |
734 | ||
735 | /* disable the timer */ | |
becf8b5d | 736 | timr->it.real.interval.tv64 = 0; |
1da177e4 LT |
737 | /* |
738 | * careful here. If smp we could be in the "fire" routine which will | |
739 | * be spinning as we hold the lock. But this is ONLY an SMP issue. | |
740 | */ | |
becf8b5d | 741 | if (hrtimer_try_to_cancel(timer) < 0) |
1da177e4 | 742 | return TIMER_RETRY; |
1da177e4 LT |
743 | |
744 | timr->it_requeue_pending = (timr->it_requeue_pending + 2) & | |
745 | ~REQUEUE_PENDING; | |
746 | timr->it_overrun_last = 0; | |
1da177e4 | 747 | |
becf8b5d TG |
748 | /* switch off the timer when it_value is zero */ |
749 | if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec) | |
750 | return 0; | |
1da177e4 | 751 | |
c9cb2e3d | 752 | mode = flags & TIMER_ABSTIME ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL; |
7978672c | 753 | hrtimer_init(&timr->it.real.timer, timr->it_clock, mode); |
7978672c | 754 | timr->it.real.timer.function = posix_timer_fn; |
becf8b5d TG |
755 | |
756 | timer->expires = timespec_to_ktime(new_setting->it_value); | |
757 | ||
758 | /* Convert interval */ | |
759 | timr->it.real.interval = timespec_to_ktime(new_setting->it_interval); | |
760 | ||
761 | /* SIGEV_NONE timers are not queued ! See common_timer_get */ | |
952bbc87 TG |
762 | if (((timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE)) { |
763 | /* Setup correct expiry time for relative timers */ | |
5a7780e7 TG |
764 | if (mode == HRTIMER_MODE_REL) { |
765 | timer->expires = | |
766 | ktime_add_safe(timer->expires, | |
767 | timer->base->get_time()); | |
768 | } | |
becf8b5d | 769 | return 0; |
952bbc87 | 770 | } |
becf8b5d | 771 | |
7978672c | 772 | hrtimer_start(timer, timer->expires, mode); |
1da177e4 LT |
773 | return 0; |
774 | } | |
775 | ||
776 | /* Set a POSIX.1b interval timer */ | |
777 | asmlinkage long | |
778 | sys_timer_settime(timer_t timer_id, int flags, | |
779 | const struct itimerspec __user *new_setting, | |
780 | struct itimerspec __user *old_setting) | |
781 | { | |
782 | struct k_itimer *timr; | |
783 | struct itimerspec new_spec, old_spec; | |
784 | int error = 0; | |
5ba25331 | 785 | unsigned long flag; |
1da177e4 LT |
786 | struct itimerspec *rtn = old_setting ? &old_spec : NULL; |
787 | ||
788 | if (!new_setting) | |
789 | return -EINVAL; | |
790 | ||
791 | if (copy_from_user(&new_spec, new_setting, sizeof (new_spec))) | |
792 | return -EFAULT; | |
793 | ||
becf8b5d TG |
794 | if (!timespec_valid(&new_spec.it_interval) || |
795 | !timespec_valid(&new_spec.it_value)) | |
1da177e4 LT |
796 | return -EINVAL; |
797 | retry: | |
798 | timr = lock_timer(timer_id, &flag); | |
799 | if (!timr) | |
800 | return -EINVAL; | |
801 | ||
802 | error = CLOCK_DISPATCH(timr->it_clock, timer_set, | |
803 | (timr, flags, &new_spec, rtn)); | |
804 | ||
805 | unlock_timer(timr, flag); | |
806 | if (error == TIMER_RETRY) { | |
807 | rtn = NULL; // We already got the old time... | |
808 | goto retry; | |
809 | } | |
810 | ||
becf8b5d TG |
811 | if (old_setting && !error && |
812 | copy_to_user(old_setting, &old_spec, sizeof (old_spec))) | |
1da177e4 LT |
813 | error = -EFAULT; |
814 | ||
815 | return error; | |
816 | } | |
817 | ||
818 | static inline int common_timer_del(struct k_itimer *timer) | |
819 | { | |
becf8b5d | 820 | timer->it.real.interval.tv64 = 0; |
f972be33 | 821 | |
becf8b5d | 822 | if (hrtimer_try_to_cancel(&timer->it.real.timer) < 0) |
1da177e4 | 823 | return TIMER_RETRY; |
1da177e4 LT |
824 | return 0; |
825 | } | |
826 | ||
827 | static inline int timer_delete_hook(struct k_itimer *timer) | |
828 | { | |
829 | return CLOCK_DISPATCH(timer->it_clock, timer_del, (timer)); | |
830 | } | |
831 | ||
832 | /* Delete a POSIX.1b interval timer. */ | |
833 | asmlinkage long | |
834 | sys_timer_delete(timer_t timer_id) | |
835 | { | |
836 | struct k_itimer *timer; | |
5ba25331 | 837 | unsigned long flags; |
1da177e4 | 838 | |
1da177e4 | 839 | retry_delete: |
1da177e4 LT |
840 | timer = lock_timer(timer_id, &flags); |
841 | if (!timer) | |
842 | return -EINVAL; | |
843 | ||
becf8b5d | 844 | if (timer_delete_hook(timer) == TIMER_RETRY) { |
1da177e4 LT |
845 | unlock_timer(timer, flags); |
846 | goto retry_delete; | |
847 | } | |
becf8b5d | 848 | |
1da177e4 LT |
849 | spin_lock(¤t->sighand->siglock); |
850 | list_del(&timer->list); | |
851 | spin_unlock(¤t->sighand->siglock); | |
852 | /* | |
853 | * This keeps any tasks waiting on the spin lock from thinking | |
854 | * they got something (see the lock code above). | |
855 | */ | |
918fc037 | 856 | put_task_struct(timer->it_process); |
4b7a1304 ON |
857 | timer->it_process = NULL; |
858 | ||
1da177e4 LT |
859 | unlock_timer(timer, flags); |
860 | release_posix_timer(timer, IT_ID_SET); | |
861 | return 0; | |
862 | } | |
becf8b5d | 863 | |
1da177e4 LT |
864 | /* |
865 | * return timer owned by the process, used by exit_itimers | |
866 | */ | |
858119e1 | 867 | static void itimer_delete(struct k_itimer *timer) |
1da177e4 LT |
868 | { |
869 | unsigned long flags; | |
870 | ||
1da177e4 | 871 | retry_delete: |
1da177e4 LT |
872 | spin_lock_irqsave(&timer->it_lock, flags); |
873 | ||
becf8b5d | 874 | if (timer_delete_hook(timer) == TIMER_RETRY) { |
1da177e4 LT |
875 | unlock_timer(timer, flags); |
876 | goto retry_delete; | |
877 | } | |
1da177e4 LT |
878 | list_del(&timer->list); |
879 | /* | |
880 | * This keeps any tasks waiting on the spin lock from thinking | |
881 | * they got something (see the lock code above). | |
882 | */ | |
918fc037 | 883 | put_task_struct(timer->it_process); |
4b7a1304 ON |
884 | timer->it_process = NULL; |
885 | ||
1da177e4 LT |
886 | unlock_timer(timer, flags); |
887 | release_posix_timer(timer, IT_ID_SET); | |
888 | } | |
889 | ||
890 | /* | |
25f407f0 | 891 | * This is called by do_exit or de_thread, only when there are no more |
1da177e4 LT |
892 | * references to the shared signal_struct. |
893 | */ | |
894 | void exit_itimers(struct signal_struct *sig) | |
895 | { | |
896 | struct k_itimer *tmr; | |
897 | ||
898 | while (!list_empty(&sig->posix_timers)) { | |
899 | tmr = list_entry(sig->posix_timers.next, struct k_itimer, list); | |
900 | itimer_delete(tmr); | |
901 | } | |
902 | } | |
903 | ||
becf8b5d | 904 | /* Not available / possible... functions */ |
a924b04d | 905 | int do_posix_clock_nosettime(const clockid_t clockid, struct timespec *tp) |
1da177e4 LT |
906 | { |
907 | return -EINVAL; | |
908 | } | |
909 | EXPORT_SYMBOL_GPL(do_posix_clock_nosettime); | |
910 | ||
a924b04d | 911 | int do_posix_clock_nonanosleep(const clockid_t clock, int flags, |
97735f25 | 912 | struct timespec *t, struct timespec __user *r) |
1da177e4 LT |
913 | { |
914 | #ifndef ENOTSUP | |
915 | return -EOPNOTSUPP; /* aka ENOTSUP in userland for POSIX */ | |
916 | #else /* parisc does define it separately. */ | |
917 | return -ENOTSUP; | |
918 | #endif | |
919 | } | |
920 | EXPORT_SYMBOL_GPL(do_posix_clock_nonanosleep); | |
921 | ||
a924b04d TG |
922 | asmlinkage long sys_clock_settime(const clockid_t which_clock, |
923 | const struct timespec __user *tp) | |
1da177e4 LT |
924 | { |
925 | struct timespec new_tp; | |
926 | ||
927 | if (invalid_clockid(which_clock)) | |
928 | return -EINVAL; | |
929 | if (copy_from_user(&new_tp, tp, sizeof (*tp))) | |
930 | return -EFAULT; | |
931 | ||
932 | return CLOCK_DISPATCH(which_clock, clock_set, (which_clock, &new_tp)); | |
933 | } | |
934 | ||
935 | asmlinkage long | |
a924b04d | 936 | sys_clock_gettime(const clockid_t which_clock, struct timespec __user *tp) |
1da177e4 LT |
937 | { |
938 | struct timespec kernel_tp; | |
939 | int error; | |
940 | ||
941 | if (invalid_clockid(which_clock)) | |
942 | return -EINVAL; | |
943 | error = CLOCK_DISPATCH(which_clock, clock_get, | |
944 | (which_clock, &kernel_tp)); | |
945 | if (!error && copy_to_user(tp, &kernel_tp, sizeof (kernel_tp))) | |
946 | error = -EFAULT; | |
947 | ||
948 | return error; | |
949 | ||
950 | } | |
951 | ||
952 | asmlinkage long | |
a924b04d | 953 | sys_clock_getres(const clockid_t which_clock, struct timespec __user *tp) |
1da177e4 LT |
954 | { |
955 | struct timespec rtn_tp; | |
956 | int error; | |
957 | ||
958 | if (invalid_clockid(which_clock)) | |
959 | return -EINVAL; | |
960 | ||
961 | error = CLOCK_DISPATCH(which_clock, clock_getres, | |
962 | (which_clock, &rtn_tp)); | |
963 | ||
964 | if (!error && tp && copy_to_user(tp, &rtn_tp, sizeof (rtn_tp))) { | |
965 | error = -EFAULT; | |
966 | } | |
967 | ||
968 | return error; | |
969 | } | |
970 | ||
97735f25 TG |
971 | /* |
972 | * nanosleep for monotonic and realtime clocks | |
973 | */ | |
974 | static int common_nsleep(const clockid_t which_clock, int flags, | |
975 | struct timespec *tsave, struct timespec __user *rmtp) | |
976 | { | |
080344b9 ON |
977 | return hrtimer_nanosleep(tsave, rmtp, flags & TIMER_ABSTIME ? |
978 | HRTIMER_MODE_ABS : HRTIMER_MODE_REL, | |
979 | which_clock); | |
97735f25 | 980 | } |
1da177e4 LT |
981 | |
982 | asmlinkage long | |
a924b04d | 983 | sys_clock_nanosleep(const clockid_t which_clock, int flags, |
1da177e4 LT |
984 | const struct timespec __user *rqtp, |
985 | struct timespec __user *rmtp) | |
986 | { | |
987 | struct timespec t; | |
1da177e4 LT |
988 | |
989 | if (invalid_clockid(which_clock)) | |
990 | return -EINVAL; | |
991 | ||
992 | if (copy_from_user(&t, rqtp, sizeof (struct timespec))) | |
993 | return -EFAULT; | |
994 | ||
5f82b2b7 | 995 | if (!timespec_valid(&t)) |
1da177e4 LT |
996 | return -EINVAL; |
997 | ||
97735f25 TG |
998 | return CLOCK_DISPATCH(which_clock, nsleep, |
999 | (which_clock, flags, &t, rmtp)); | |
1da177e4 | 1000 | } |
1711ef38 TA |
1001 | |
1002 | /* | |
1003 | * nanosleep_restart for monotonic and realtime clocks | |
1004 | */ | |
1005 | static int common_nsleep_restart(struct restart_block *restart_block) | |
1006 | { | |
1007 | return hrtimer_nanosleep_restart(restart_block); | |
1008 | } | |
1009 | ||
1010 | /* | |
1011 | * This will restart clock_nanosleep. This is required only by | |
1012 | * compat_clock_nanosleep_restart for now. | |
1013 | */ | |
1014 | long | |
1015 | clock_nanosleep_restart(struct restart_block *restart_block) | |
1016 | { | |
1017 | clockid_t which_clock = restart_block->arg0; | |
1018 | ||
1019 | return CLOCK_DISPATCH(which_clock, nsleep_restart, | |
1020 | (restart_block)); | |
1021 | } |