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> | |
40 | #include <asm/semaphore.h> | |
41 | #include <linux/list.h> | |
42 | #include <linux/init.h> | |
43 | #include <linux/compiler.h> | |
44 | #include <linux/idr.h> | |
45 | #include <linux/posix-timers.h> | |
46 | #include <linux/syscalls.h> | |
47 | #include <linux/wait.h> | |
48 | #include <linux/workqueue.h> | |
49 | #include <linux/module.h> | |
50 | ||
1da177e4 LT |
51 | /* |
52 | * Management arrays for POSIX timers. Timers are kept in slab memory | |
53 | * Timer ids are allocated by an external routine that keeps track of the | |
54 | * id and the timer. The external interface is: | |
55 | * | |
56 | * void *idr_find(struct idr *idp, int id); to find timer_id <id> | |
57 | * int idr_get_new(struct idr *idp, void *ptr); to get a new id and | |
58 | * related it to <ptr> | |
59 | * void idr_remove(struct idr *idp, int id); to release <id> | |
60 | * void idr_init(struct idr *idp); to initialize <idp> | |
61 | * which we supply. | |
62 | * The idr_get_new *may* call slab for more memory so it must not be | |
63 | * called under a spin lock. Likewise idr_remore may release memory | |
64 | * (but it may be ok to do this under a lock...). | |
65 | * idr_find is just a memory look up and is quite fast. A -1 return | |
66 | * indicates that the requested id does not exist. | |
67 | */ | |
68 | ||
69 | /* | |
70 | * Lets keep our timers in a slab cache :-) | |
71 | */ | |
e18b890b | 72 | static struct kmem_cache *posix_timers_cache; |
1da177e4 LT |
73 | static struct idr posix_timers_id; |
74 | static DEFINE_SPINLOCK(idr_lock); | |
75 | ||
1da177e4 LT |
76 | /* |
77 | * we assume that the new SIGEV_THREAD_ID shares no bits with the other | |
78 | * SIGEV values. Here we put out an error if this assumption fails. | |
79 | */ | |
80 | #if SIGEV_THREAD_ID != (SIGEV_THREAD_ID & \ | |
81 | ~(SIGEV_SIGNAL | SIGEV_NONE | SIGEV_THREAD)) | |
82 | #error "SIGEV_THREAD_ID must not share bit with other SIGEV values!" | |
83 | #endif | |
84 | ||
85 | ||
86 | /* | |
87 | * The timer ID is turned into a timer address by idr_find(). | |
88 | * Verifying a valid ID consists of: | |
89 | * | |
90 | * a) checking that idr_find() returns other than -1. | |
91 | * b) checking that the timer id matches the one in the timer itself. | |
92 | * c) that the timer owner is in the callers thread group. | |
93 | */ | |
94 | ||
95 | /* | |
96 | * CLOCKs: The POSIX standard calls for a couple of clocks and allows us | |
97 | * to implement others. This structure defines the various | |
98 | * clocks and allows the possibility of adding others. We | |
99 | * provide an interface to add clocks to the table and expect | |
100 | * the "arch" code to add at least one clock that is high | |
101 | * resolution. Here we define the standard CLOCK_REALTIME as a | |
102 | * 1/HZ resolution clock. | |
103 | * | |
104 | * RESOLUTION: Clock resolution is used to round up timer and interval | |
105 | * times, NOT to report clock times, which are reported with as | |
106 | * much resolution as the system can muster. In some cases this | |
107 | * resolution may depend on the underlying clock hardware and | |
108 | * may not be quantifiable until run time, and only then is the | |
109 | * necessary code is written. The standard says we should say | |
110 | * something about this issue in the documentation... | |
111 | * | |
112 | * FUNCTIONS: The CLOCKs structure defines possible functions to handle | |
113 | * various clock functions. For clocks that use the standard | |
114 | * system timer code these entries should be NULL. This will | |
115 | * allow dispatch without the overhead of indirect function | |
116 | * calls. CLOCKS that depend on other sources (e.g. WWV or GPS) | |
117 | * must supply functions here, even if the function just returns | |
118 | * ENOSYS. The standard POSIX timer management code assumes the | |
119 | * following: 1.) The k_itimer struct (sched.h) is used for the | |
120 | * timer. 2.) The list, it_lock, it_clock, it_id and it_process | |
121 | * fields are not modified by timer code. | |
122 | * | |
123 | * At this time all functions EXCEPT clock_nanosleep can be | |
124 | * redirected by the CLOCKS structure. Clock_nanosleep is in | |
125 | * there, but the code ignores it. | |
126 | * | |
127 | * Permissions: It is assumed that the clock_settime() function defined | |
128 | * for each clock will take care of permission checks. Some | |
129 | * clocks may be set able by any user (i.e. local process | |
130 | * clocks) others not. Currently the only set able clock we | |
131 | * have is CLOCK_REALTIME and its high res counter part, both of | |
132 | * which we beg off on and pass to do_sys_settimeofday(). | |
133 | */ | |
134 | ||
135 | static struct k_clock posix_clocks[MAX_CLOCKS]; | |
becf8b5d | 136 | |
1da177e4 | 137 | /* |
becf8b5d | 138 | * These ones are defined below. |
1da177e4 | 139 | */ |
becf8b5d TG |
140 | static int common_nsleep(const clockid_t, int flags, struct timespec *t, |
141 | struct timespec __user *rmtp); | |
142 | static void common_timer_get(struct k_itimer *, struct itimerspec *); | |
143 | static int common_timer_set(struct k_itimer *, int, | |
144 | struct itimerspec *, struct itimerspec *); | |
145 | static int common_timer_del(struct k_itimer *timer); | |
1da177e4 | 146 | |
c9cb2e3d | 147 | static enum hrtimer_restart posix_timer_fn(struct hrtimer *data); |
1da177e4 LT |
148 | |
149 | static struct k_itimer *lock_timer(timer_t timer_id, unsigned long *flags); | |
150 | ||
151 | static inline void unlock_timer(struct k_itimer *timr, unsigned long flags) | |
152 | { | |
153 | spin_unlock_irqrestore(&timr->it_lock, flags); | |
154 | } | |
155 | ||
156 | /* | |
157 | * Call the k_clock hook function if non-null, or the default function. | |
158 | */ | |
159 | #define CLOCK_DISPATCH(clock, call, arglist) \ | |
160 | ((clock) < 0 ? posix_cpu_##call arglist : \ | |
161 | (posix_clocks[clock].call != NULL \ | |
162 | ? (*posix_clocks[clock].call) arglist : common_##call arglist)) | |
163 | ||
164 | /* | |
165 | * Default clock hook functions when the struct k_clock passed | |
166 | * to register_posix_clock leaves a function pointer null. | |
167 | * | |
168 | * The function common_CALL is the default implementation for | |
169 | * the function pointer CALL in struct k_clock. | |
170 | */ | |
171 | ||
a924b04d | 172 | static inline int common_clock_getres(const clockid_t which_clock, |
1da177e4 LT |
173 | struct timespec *tp) |
174 | { | |
175 | tp->tv_sec = 0; | |
176 | tp->tv_nsec = posix_clocks[which_clock].res; | |
177 | return 0; | |
178 | } | |
179 | ||
becf8b5d TG |
180 | /* |
181 | * Get real time for posix timers | |
182 | */ | |
183 | static int common_clock_get(clockid_t which_clock, struct timespec *tp) | |
1da177e4 | 184 | { |
becf8b5d | 185 | ktime_get_real_ts(tp); |
1da177e4 LT |
186 | return 0; |
187 | } | |
188 | ||
a924b04d TG |
189 | static inline int common_clock_set(const clockid_t which_clock, |
190 | struct timespec *tp) | |
1da177e4 LT |
191 | { |
192 | return do_sys_settimeofday(tp, NULL); | |
193 | } | |
194 | ||
858119e1 | 195 | static int common_timer_create(struct k_itimer *new_timer) |
1da177e4 | 196 | { |
7978672c | 197 | hrtimer_init(&new_timer->it.real.timer, new_timer->it_clock, 0); |
1da177e4 LT |
198 | return 0; |
199 | } | |
200 | ||
201 | /* | |
becf8b5d | 202 | * Return nonzero if we know a priori this clockid_t value is bogus. |
1da177e4 | 203 | */ |
a924b04d | 204 | static inline int invalid_clockid(const clockid_t which_clock) |
1da177e4 LT |
205 | { |
206 | if (which_clock < 0) /* CPU clock, posix_cpu_* will check it */ | |
207 | return 0; | |
208 | if ((unsigned) which_clock >= MAX_CLOCKS) | |
209 | return 1; | |
210 | if (posix_clocks[which_clock].clock_getres != NULL) | |
211 | return 0; | |
1da177e4 LT |
212 | if (posix_clocks[which_clock].res != 0) |
213 | return 0; | |
1da177e4 LT |
214 | return 1; |
215 | } | |
216 | ||
becf8b5d TG |
217 | /* |
218 | * Get monotonic time for posix timers | |
219 | */ | |
220 | static int posix_ktime_get_ts(clockid_t which_clock, struct timespec *tp) | |
221 | { | |
222 | ktime_get_ts(tp); | |
223 | return 0; | |
224 | } | |
1da177e4 LT |
225 | |
226 | /* | |
227 | * Initialize everything, well, just everything in Posix clocks/timers ;) | |
228 | */ | |
229 | static __init int init_posix_timers(void) | |
230 | { | |
becf8b5d TG |
231 | struct k_clock clock_realtime = { |
232 | .clock_getres = hrtimer_get_res, | |
1da177e4 | 233 | }; |
becf8b5d TG |
234 | struct k_clock clock_monotonic = { |
235 | .clock_getres = hrtimer_get_res, | |
236 | .clock_get = posix_ktime_get_ts, | |
237 | .clock_set = do_posix_clock_nosettime, | |
1da177e4 LT |
238 | }; |
239 | ||
240 | register_posix_clock(CLOCK_REALTIME, &clock_realtime); | |
241 | register_posix_clock(CLOCK_MONOTONIC, &clock_monotonic); | |
242 | ||
243 | posix_timers_cache = kmem_cache_create("posix_timers_cache", | |
040b5c6f AD |
244 | sizeof (struct k_itimer), 0, SLAB_PANIC, |
245 | NULL); | |
1da177e4 LT |
246 | idr_init(&posix_timers_id); |
247 | return 0; | |
248 | } | |
249 | ||
250 | __initcall(init_posix_timers); | |
251 | ||
1da177e4 LT |
252 | static void schedule_next_timer(struct k_itimer *timr) |
253 | { | |
44f21475 RZ |
254 | struct hrtimer *timer = &timr->it.real.timer; |
255 | ||
becf8b5d | 256 | if (timr->it.real.interval.tv64 == 0) |
1da177e4 LT |
257 | return; |
258 | ||
4d672e7a DL |
259 | timr->it_overrun += (unsigned int) hrtimer_forward(timer, |
260 | timer->base->get_time(), | |
261 | timr->it.real.interval); | |
44f21475 | 262 | |
1da177e4 LT |
263 | timr->it_overrun_last = timr->it_overrun; |
264 | timr->it_overrun = -1; | |
265 | ++timr->it_requeue_pending; | |
44f21475 | 266 | hrtimer_restart(timer); |
1da177e4 LT |
267 | } |
268 | ||
269 | /* | |
270 | * This function is exported for use by the signal deliver code. It is | |
271 | * called just prior to the info block being released and passes that | |
272 | * block to us. It's function is to update the overrun entry AND to | |
273 | * restart the timer. It should only be called if the timer is to be | |
274 | * restarted (i.e. we have flagged this in the sys_private entry of the | |
275 | * info block). | |
276 | * | |
277 | * To protect aginst the timer going away while the interrupt is queued, | |
278 | * we require that the it_requeue_pending flag be set. | |
279 | */ | |
280 | void do_schedule_next_timer(struct siginfo *info) | |
281 | { | |
282 | struct k_itimer *timr; | |
283 | unsigned long flags; | |
284 | ||
285 | timr = lock_timer(info->si_tid, &flags); | |
286 | ||
becf8b5d TG |
287 | if (timr && timr->it_requeue_pending == info->si_sys_private) { |
288 | if (timr->it_clock < 0) | |
289 | posix_cpu_timer_schedule(timr); | |
290 | else | |
291 | schedule_next_timer(timr); | |
1da177e4 | 292 | |
becf8b5d TG |
293 | info->si_overrun = timr->it_overrun_last; |
294 | } | |
295 | ||
b6557fbc TG |
296 | if (timr) |
297 | unlock_timer(timr, flags); | |
1da177e4 LT |
298 | } |
299 | ||
300 | int posix_timer_event(struct k_itimer *timr,int si_private) | |
301 | { | |
302 | memset(&timr->sigq->info, 0, sizeof(siginfo_t)); | |
303 | timr->sigq->info.si_sys_private = si_private; | |
becf8b5d | 304 | /* Send signal to the process that owns this timer.*/ |
1da177e4 LT |
305 | |
306 | timr->sigq->info.si_signo = timr->it_sigev_signo; | |
307 | timr->sigq->info.si_errno = 0; | |
308 | timr->sigq->info.si_code = SI_TIMER; | |
309 | timr->sigq->info.si_tid = timr->it_id; | |
310 | timr->sigq->info.si_value = timr->it_sigev_value; | |
e752dd6c | 311 | |
1da177e4 | 312 | if (timr->it_sigev_notify & SIGEV_THREAD_ID) { |
e752dd6c ON |
313 | struct task_struct *leader; |
314 | int ret = send_sigqueue(timr->it_sigev_signo, timr->sigq, | |
315 | timr->it_process); | |
316 | ||
317 | if (likely(ret >= 0)) | |
318 | return ret; | |
319 | ||
320 | timr->it_sigev_notify = SIGEV_SIGNAL; | |
321 | leader = timr->it_process->group_leader; | |
322 | put_task_struct(timr->it_process); | |
323 | timr->it_process = leader; | |
1da177e4 | 324 | } |
e752dd6c ON |
325 | |
326 | return send_group_sigqueue(timr->it_sigev_signo, timr->sigq, | |
327 | timr->it_process); | |
1da177e4 LT |
328 | } |
329 | EXPORT_SYMBOL_GPL(posix_timer_event); | |
330 | ||
331 | /* | |
332 | * This function gets called when a POSIX.1b interval timer expires. It | |
333 | * is used as a callback from the kernel internal timer. The | |
334 | * run_timer_list code ALWAYS calls with interrupts on. | |
335 | ||
336 | * This code is for CLOCK_REALTIME* and CLOCK_MONOTONIC* timers. | |
337 | */ | |
c9cb2e3d | 338 | static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer) |
1da177e4 | 339 | { |
05cfb614 | 340 | struct k_itimer *timr; |
1da177e4 | 341 | unsigned long flags; |
becf8b5d | 342 | int si_private = 0; |
c9cb2e3d | 343 | enum hrtimer_restart ret = HRTIMER_NORESTART; |
1da177e4 | 344 | |
05cfb614 | 345 | timr = container_of(timer, struct k_itimer, it.real.timer); |
1da177e4 | 346 | spin_lock_irqsave(&timr->it_lock, flags); |
1da177e4 | 347 | |
becf8b5d TG |
348 | if (timr->it.real.interval.tv64 != 0) |
349 | si_private = ++timr->it_requeue_pending; | |
1da177e4 | 350 | |
becf8b5d TG |
351 | if (posix_timer_event(timr, si_private)) { |
352 | /* | |
353 | * signal was not sent because of sig_ignor | |
354 | * we will not get a call back to restart it AND | |
355 | * it should be restarted. | |
356 | */ | |
357 | if (timr->it.real.interval.tv64 != 0) { | |
58229a18 TG |
358 | ktime_t now = hrtimer_cb_get_time(timer); |
359 | ||
360 | /* | |
361 | * FIXME: What we really want, is to stop this | |
362 | * timer completely and restart it in case the | |
363 | * SIG_IGN is removed. This is a non trivial | |
364 | * change which involves sighand locking | |
365 | * (sigh !), which we don't want to do late in | |
366 | * the release cycle. | |
367 | * | |
368 | * For now we just let timers with an interval | |
369 | * less than a jiffie expire every jiffie to | |
370 | * avoid softirq starvation in case of SIG_IGN | |
371 | * and a very small interval, which would put | |
372 | * the timer right back on the softirq pending | |
373 | * list. By moving now ahead of time we trick | |
374 | * hrtimer_forward() to expire the timer | |
375 | * later, while we still maintain the overrun | |
376 | * accuracy, but have some inconsistency in | |
377 | * the timer_gettime() case. This is at least | |
378 | * better than a starved softirq. A more | |
379 | * complex fix which solves also another related | |
380 | * inconsistency is already in the pipeline. | |
381 | */ | |
382 | #ifdef CONFIG_HIGH_RES_TIMERS | |
383 | { | |
384 | ktime_t kj = ktime_set(0, NSEC_PER_SEC / HZ); | |
385 | ||
386 | if (timr->it.real.interval.tv64 < kj.tv64) | |
387 | now = ktime_add(now, kj); | |
388 | } | |
389 | #endif | |
4d672e7a | 390 | timr->it_overrun += (unsigned int) |
58229a18 | 391 | hrtimer_forward(timer, now, |
becf8b5d TG |
392 | timr->it.real.interval); |
393 | ret = HRTIMER_RESTART; | |
a0a0c28c | 394 | ++timr->it_requeue_pending; |
1da177e4 | 395 | } |
1da177e4 | 396 | } |
1da177e4 | 397 | |
becf8b5d TG |
398 | unlock_timer(timr, flags); |
399 | return ret; | |
400 | } | |
1da177e4 | 401 | |
858119e1 | 402 | static struct task_struct * good_sigevent(sigevent_t * event) |
1da177e4 LT |
403 | { |
404 | struct task_struct *rtn = current->group_leader; | |
405 | ||
406 | if ((event->sigev_notify & SIGEV_THREAD_ID ) && | |
8dc86af0 | 407 | (!(rtn = find_task_by_vpid(event->sigev_notify_thread_id)) || |
bac0abd6 | 408 | !same_thread_group(rtn, current) || |
1da177e4 LT |
409 | (event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_SIGNAL)) |
410 | return NULL; | |
411 | ||
412 | if (((event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) && | |
413 | ((event->sigev_signo <= 0) || (event->sigev_signo > SIGRTMAX))) | |
414 | return NULL; | |
415 | ||
416 | return rtn; | |
417 | } | |
418 | ||
a924b04d | 419 | void register_posix_clock(const clockid_t clock_id, struct k_clock *new_clock) |
1da177e4 LT |
420 | { |
421 | if ((unsigned) clock_id >= MAX_CLOCKS) { | |
422 | printk("POSIX clock register failed for clock_id %d\n", | |
423 | clock_id); | |
424 | return; | |
425 | } | |
426 | ||
427 | posix_clocks[clock_id] = *new_clock; | |
428 | } | |
429 | EXPORT_SYMBOL_GPL(register_posix_clock); | |
430 | ||
431 | static struct k_itimer * alloc_posix_timer(void) | |
432 | { | |
433 | struct k_itimer *tmr; | |
c3762229 | 434 | tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL); |
1da177e4 LT |
435 | if (!tmr) |
436 | return tmr; | |
1da177e4 LT |
437 | if (unlikely(!(tmr->sigq = sigqueue_alloc()))) { |
438 | kmem_cache_free(posix_timers_cache, tmr); | |
439 | tmr = NULL; | |
440 | } | |
441 | return tmr; | |
442 | } | |
443 | ||
444 | #define IT_ID_SET 1 | |
445 | #define IT_ID_NOT_SET 0 | |
446 | static void release_posix_timer(struct k_itimer *tmr, int it_id_set) | |
447 | { | |
448 | if (it_id_set) { | |
449 | unsigned long flags; | |
450 | spin_lock_irqsave(&idr_lock, flags); | |
451 | idr_remove(&posix_timers_id, tmr->it_id); | |
452 | spin_unlock_irqrestore(&idr_lock, flags); | |
453 | } | |
454 | sigqueue_free(tmr->sigq); | |
455 | if (unlikely(tmr->it_process) && | |
456 | tmr->it_sigev_notify == (SIGEV_SIGNAL|SIGEV_THREAD_ID)) | |
457 | put_task_struct(tmr->it_process); | |
458 | kmem_cache_free(posix_timers_cache, tmr); | |
459 | } | |
460 | ||
461 | /* Create a POSIX.1b interval timer. */ | |
462 | ||
463 | asmlinkage long | |
a924b04d | 464 | sys_timer_create(const clockid_t which_clock, |
1da177e4 LT |
465 | struct sigevent __user *timer_event_spec, |
466 | timer_t __user * created_timer_id) | |
467 | { | |
468 | int error = 0; | |
469 | struct k_itimer *new_timer = NULL; | |
470 | int new_timer_id; | |
471 | struct task_struct *process = NULL; | |
472 | unsigned long flags; | |
473 | sigevent_t event; | |
474 | int it_id_set = IT_ID_NOT_SET; | |
475 | ||
476 | if (invalid_clockid(which_clock)) | |
477 | return -EINVAL; | |
478 | ||
479 | new_timer = alloc_posix_timer(); | |
480 | if (unlikely(!new_timer)) | |
481 | return -EAGAIN; | |
482 | ||
483 | spin_lock_init(&new_timer->it_lock); | |
484 | retry: | |
485 | if (unlikely(!idr_pre_get(&posix_timers_id, GFP_KERNEL))) { | |
486 | error = -EAGAIN; | |
487 | goto out; | |
488 | } | |
489 | spin_lock_irq(&idr_lock); | |
becf8b5d | 490 | error = idr_get_new(&posix_timers_id, (void *) new_timer, |
1da177e4 LT |
491 | &new_timer_id); |
492 | spin_unlock_irq(&idr_lock); | |
493 | if (error == -EAGAIN) | |
494 | goto retry; | |
495 | else if (error) { | |
496 | /* | |
0b0a3e7b | 497 | * Weird looking, but we return EAGAIN if the IDR is |
1da177e4 LT |
498 | * full (proper POSIX return value for this) |
499 | */ | |
500 | error = -EAGAIN; | |
501 | goto out; | |
502 | } | |
503 | ||
504 | it_id_set = IT_ID_SET; | |
505 | new_timer->it_id = (timer_t) new_timer_id; | |
506 | new_timer->it_clock = which_clock; | |
507 | new_timer->it_overrun = -1; | |
508 | error = CLOCK_DISPATCH(which_clock, timer_create, (new_timer)); | |
509 | if (error) | |
510 | goto out; | |
511 | ||
512 | /* | |
513 | * return the timer_id now. The next step is hard to | |
514 | * back out if there is an error. | |
515 | */ | |
516 | if (copy_to_user(created_timer_id, | |
517 | &new_timer_id, sizeof (new_timer_id))) { | |
518 | error = -EFAULT; | |
519 | goto out; | |
520 | } | |
521 | if (timer_event_spec) { | |
522 | if (copy_from_user(&event, timer_event_spec, sizeof (event))) { | |
523 | error = -EFAULT; | |
524 | goto out; | |
525 | } | |
526 | new_timer->it_sigev_notify = event.sigev_notify; | |
527 | new_timer->it_sigev_signo = event.sigev_signo; | |
528 | new_timer->it_sigev_value = event.sigev_value; | |
529 | ||
530 | read_lock(&tasklist_lock); | |
531 | if ((process = good_sigevent(&event))) { | |
532 | /* | |
533 | * We may be setting up this process for another | |
534 | * thread. It may be exiting. To catch this | |
535 | * case the we check the PF_EXITING flag. If | |
536 | * the flag is not set, the siglock will catch | |
537 | * him before it is too late (in exit_itimers). | |
538 | * | |
539 | * The exec case is a bit more invloved but easy | |
540 | * to code. If the process is in our thread | |
541 | * group (and it must be or we would not allow | |
542 | * it here) and is doing an exec, it will cause | |
543 | * us to be killed. In this case it will wait | |
544 | * for us to die which means we can finish this | |
545 | * linkage with our last gasp. I.e. no code :) | |
546 | */ | |
547 | spin_lock_irqsave(&process->sighand->siglock, flags); | |
548 | if (!(process->flags & PF_EXITING)) { | |
549 | new_timer->it_process = process; | |
550 | list_add(&new_timer->list, | |
551 | &process->signal->posix_timers); | |
1da177e4 LT |
552 | if (new_timer->it_sigev_notify == (SIGEV_SIGNAL|SIGEV_THREAD_ID)) |
553 | get_task_struct(process); | |
d02479bd | 554 | spin_unlock_irqrestore(&process->sighand->siglock, flags); |
1da177e4 LT |
555 | } else { |
556 | spin_unlock_irqrestore(&process->sighand->siglock, flags); | |
557 | process = NULL; | |
558 | } | |
559 | } | |
560 | read_unlock(&tasklist_lock); | |
561 | if (!process) { | |
562 | error = -EINVAL; | |
563 | goto out; | |
564 | } | |
565 | } else { | |
566 | new_timer->it_sigev_notify = SIGEV_SIGNAL; | |
567 | new_timer->it_sigev_signo = SIGALRM; | |
568 | new_timer->it_sigev_value.sival_int = new_timer->it_id; | |
569 | process = current->group_leader; | |
570 | spin_lock_irqsave(&process->sighand->siglock, flags); | |
571 | new_timer->it_process = process; | |
572 | list_add(&new_timer->list, &process->signal->posix_timers); | |
573 | spin_unlock_irqrestore(&process->sighand->siglock, flags); | |
574 | } | |
575 | ||
576 | /* | |
577 | * In the case of the timer belonging to another task, after | |
578 | * the task is unlocked, the timer is owned by the other task | |
579 | * and may cease to exist at any time. Don't use or modify | |
580 | * new_timer after the unlock call. | |
581 | */ | |
582 | ||
583 | out: | |
584 | if (error) | |
585 | release_posix_timer(new_timer, it_id_set); | |
586 | ||
587 | return error; | |
588 | } | |
589 | ||
1da177e4 LT |
590 | /* |
591 | * Locking issues: We need to protect the result of the id look up until | |
592 | * we get the timer locked down so it is not deleted under us. The | |
593 | * removal is done under the idr spinlock so we use that here to bridge | |
594 | * the find to the timer lock. To avoid a dead lock, the timer id MUST | |
595 | * be release with out holding the timer lock. | |
596 | */ | |
597 | static struct k_itimer * lock_timer(timer_t timer_id, unsigned long *flags) | |
598 | { | |
599 | struct k_itimer *timr; | |
600 | /* | |
601 | * Watch out here. We do a irqsave on the idr_lock and pass the | |
602 | * flags part over to the timer lock. Must not let interrupts in | |
603 | * while we are moving the lock. | |
604 | */ | |
605 | ||
606 | spin_lock_irqsave(&idr_lock, *flags); | |
607 | timr = (struct k_itimer *) idr_find(&posix_timers_id, (int) timer_id); | |
608 | if (timr) { | |
609 | spin_lock(&timr->it_lock); | |
1da177e4 LT |
610 | |
611 | if ((timr->it_id != timer_id) || !(timr->it_process) || | |
bac0abd6 | 612 | !same_thread_group(timr->it_process, current)) { |
179394af TG |
613 | spin_unlock(&timr->it_lock); |
614 | spin_unlock_irqrestore(&idr_lock, *flags); | |
1da177e4 | 615 | timr = NULL; |
179394af TG |
616 | } else |
617 | spin_unlock(&idr_lock); | |
1da177e4 LT |
618 | } else |
619 | spin_unlock_irqrestore(&idr_lock, *flags); | |
620 | ||
621 | return timr; | |
622 | } | |
623 | ||
624 | /* | |
625 | * Get the time remaining on a POSIX.1b interval timer. This function | |
626 | * is ALWAYS called with spin_lock_irq on the timer, thus it must not | |
627 | * mess with irq. | |
628 | * | |
629 | * We have a couple of messes to clean up here. First there is the case | |
630 | * of a timer that has a requeue pending. These timers should appear to | |
631 | * be in the timer list with an expiry as if we were to requeue them | |
632 | * now. | |
633 | * | |
634 | * The second issue is the SIGEV_NONE timer which may be active but is | |
635 | * not really ever put in the timer list (to save system resources). | |
636 | * This timer may be expired, and if so, we will do it here. Otherwise | |
637 | * it is the same as a requeue pending timer WRT to what we should | |
638 | * report. | |
639 | */ | |
640 | static void | |
641 | common_timer_get(struct k_itimer *timr, struct itimerspec *cur_setting) | |
642 | { | |
3b98a532 | 643 | ktime_t now, remaining, iv; |
becf8b5d | 644 | struct hrtimer *timer = &timr->it.real.timer; |
1da177e4 | 645 | |
becf8b5d | 646 | memset(cur_setting, 0, sizeof(struct itimerspec)); |
becf8b5d | 647 | |
3b98a532 RZ |
648 | iv = timr->it.real.interval; |
649 | ||
becf8b5d | 650 | /* interval timer ? */ |
3b98a532 RZ |
651 | if (iv.tv64) |
652 | cur_setting->it_interval = ktime_to_timespec(iv); | |
653 | else if (!hrtimer_active(timer) && | |
654 | (timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) | |
becf8b5d | 655 | return; |
3b98a532 RZ |
656 | |
657 | now = timer->base->get_time(); | |
658 | ||
becf8b5d | 659 | /* |
3b98a532 RZ |
660 | * When a requeue is pending or this is a SIGEV_NONE |
661 | * timer move the expiry time forward by intervals, so | |
662 | * expiry is > now. | |
becf8b5d | 663 | */ |
3b98a532 RZ |
664 | if (iv.tv64 && (timr->it_requeue_pending & REQUEUE_PENDING || |
665 | (timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE)) | |
4d672e7a | 666 | timr->it_overrun += (unsigned int) hrtimer_forward(timer, now, iv); |
3b98a532 RZ |
667 | |
668 | remaining = ktime_sub(timer->expires, now); | |
becf8b5d | 669 | /* Return 0 only, when the timer is expired and not pending */ |
3b98a532 RZ |
670 | if (remaining.tv64 <= 0) { |
671 | /* | |
672 | * A single shot SIGEV_NONE timer must return 0, when | |
673 | * it is expired ! | |
674 | */ | |
675 | if ((timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) | |
676 | cur_setting->it_value.tv_nsec = 1; | |
677 | } else | |
becf8b5d | 678 | cur_setting->it_value = ktime_to_timespec(remaining); |
1da177e4 LT |
679 | } |
680 | ||
681 | /* Get the time remaining on a POSIX.1b interval timer. */ | |
682 | asmlinkage long | |
683 | sys_timer_gettime(timer_t timer_id, struct itimerspec __user *setting) | |
684 | { | |
685 | struct k_itimer *timr; | |
686 | struct itimerspec cur_setting; | |
687 | unsigned long flags; | |
688 | ||
689 | timr = lock_timer(timer_id, &flags); | |
690 | if (!timr) | |
691 | return -EINVAL; | |
692 | ||
693 | CLOCK_DISPATCH(timr->it_clock, timer_get, (timr, &cur_setting)); | |
694 | ||
695 | unlock_timer(timr, flags); | |
696 | ||
697 | if (copy_to_user(setting, &cur_setting, sizeof (cur_setting))) | |
698 | return -EFAULT; | |
699 | ||
700 | return 0; | |
701 | } | |
becf8b5d | 702 | |
1da177e4 LT |
703 | /* |
704 | * Get the number of overruns of a POSIX.1b interval timer. This is to | |
705 | * be the overrun of the timer last delivered. At the same time we are | |
706 | * accumulating overruns on the next timer. The overrun is frozen when | |
707 | * the signal is delivered, either at the notify time (if the info block | |
708 | * is not queued) or at the actual delivery time (as we are informed by | |
709 | * the call back to do_schedule_next_timer(). So all we need to do is | |
710 | * to pick up the frozen overrun. | |
711 | */ | |
1da177e4 LT |
712 | asmlinkage long |
713 | sys_timer_getoverrun(timer_t timer_id) | |
714 | { | |
715 | struct k_itimer *timr; | |
716 | int overrun; | |
5ba25331 | 717 | unsigned long flags; |
1da177e4 LT |
718 | |
719 | timr = lock_timer(timer_id, &flags); | |
720 | if (!timr) | |
721 | return -EINVAL; | |
722 | ||
723 | overrun = timr->it_overrun_last; | |
724 | unlock_timer(timr, flags); | |
725 | ||
726 | return overrun; | |
727 | } | |
1da177e4 LT |
728 | |
729 | /* Set a POSIX.1b interval timer. */ | |
730 | /* timr->it_lock is taken. */ | |
858119e1 | 731 | static int |
1da177e4 LT |
732 | common_timer_set(struct k_itimer *timr, int flags, |
733 | struct itimerspec *new_setting, struct itimerspec *old_setting) | |
734 | { | |
becf8b5d | 735 | struct hrtimer *timer = &timr->it.real.timer; |
7978672c | 736 | enum hrtimer_mode mode; |
1da177e4 LT |
737 | |
738 | if (old_setting) | |
739 | common_timer_get(timr, old_setting); | |
740 | ||
741 | /* disable the timer */ | |
becf8b5d | 742 | timr->it.real.interval.tv64 = 0; |
1da177e4 LT |
743 | /* |
744 | * careful here. If smp we could be in the "fire" routine which will | |
745 | * be spinning as we hold the lock. But this is ONLY an SMP issue. | |
746 | */ | |
becf8b5d | 747 | if (hrtimer_try_to_cancel(timer) < 0) |
1da177e4 | 748 | return TIMER_RETRY; |
1da177e4 LT |
749 | |
750 | timr->it_requeue_pending = (timr->it_requeue_pending + 2) & | |
751 | ~REQUEUE_PENDING; | |
752 | timr->it_overrun_last = 0; | |
1da177e4 | 753 | |
becf8b5d TG |
754 | /* switch off the timer when it_value is zero */ |
755 | if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec) | |
756 | return 0; | |
1da177e4 | 757 | |
c9cb2e3d | 758 | mode = flags & TIMER_ABSTIME ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL; |
7978672c | 759 | hrtimer_init(&timr->it.real.timer, timr->it_clock, mode); |
7978672c | 760 | timr->it.real.timer.function = posix_timer_fn; |
becf8b5d TG |
761 | |
762 | timer->expires = timespec_to_ktime(new_setting->it_value); | |
763 | ||
764 | /* Convert interval */ | |
765 | timr->it.real.interval = timespec_to_ktime(new_setting->it_interval); | |
766 | ||
767 | /* SIGEV_NONE timers are not queued ! See common_timer_get */ | |
952bbc87 TG |
768 | if (((timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE)) { |
769 | /* Setup correct expiry time for relative timers */ | |
5a7780e7 TG |
770 | if (mode == HRTIMER_MODE_REL) { |
771 | timer->expires = | |
772 | ktime_add_safe(timer->expires, | |
773 | timer->base->get_time()); | |
774 | } | |
becf8b5d | 775 | return 0; |
952bbc87 | 776 | } |
becf8b5d | 777 | |
7978672c | 778 | hrtimer_start(timer, timer->expires, mode); |
1da177e4 LT |
779 | return 0; |
780 | } | |
781 | ||
782 | /* Set a POSIX.1b interval timer */ | |
783 | asmlinkage long | |
784 | sys_timer_settime(timer_t timer_id, int flags, | |
785 | const struct itimerspec __user *new_setting, | |
786 | struct itimerspec __user *old_setting) | |
787 | { | |
788 | struct k_itimer *timr; | |
789 | struct itimerspec new_spec, old_spec; | |
790 | int error = 0; | |
5ba25331 | 791 | unsigned long flag; |
1da177e4 LT |
792 | struct itimerspec *rtn = old_setting ? &old_spec : NULL; |
793 | ||
794 | if (!new_setting) | |
795 | return -EINVAL; | |
796 | ||
797 | if (copy_from_user(&new_spec, new_setting, sizeof (new_spec))) | |
798 | return -EFAULT; | |
799 | ||
becf8b5d TG |
800 | if (!timespec_valid(&new_spec.it_interval) || |
801 | !timespec_valid(&new_spec.it_value)) | |
1da177e4 LT |
802 | return -EINVAL; |
803 | retry: | |
804 | timr = lock_timer(timer_id, &flag); | |
805 | if (!timr) | |
806 | return -EINVAL; | |
807 | ||
808 | error = CLOCK_DISPATCH(timr->it_clock, timer_set, | |
809 | (timr, flags, &new_spec, rtn)); | |
810 | ||
811 | unlock_timer(timr, flag); | |
812 | if (error == TIMER_RETRY) { | |
813 | rtn = NULL; // We already got the old time... | |
814 | goto retry; | |
815 | } | |
816 | ||
becf8b5d TG |
817 | if (old_setting && !error && |
818 | copy_to_user(old_setting, &old_spec, sizeof (old_spec))) | |
1da177e4 LT |
819 | error = -EFAULT; |
820 | ||
821 | return error; | |
822 | } | |
823 | ||
824 | static inline int common_timer_del(struct k_itimer *timer) | |
825 | { | |
becf8b5d | 826 | timer->it.real.interval.tv64 = 0; |
f972be33 | 827 | |
becf8b5d | 828 | if (hrtimer_try_to_cancel(&timer->it.real.timer) < 0) |
1da177e4 | 829 | return TIMER_RETRY; |
1da177e4 LT |
830 | return 0; |
831 | } | |
832 | ||
833 | static inline int timer_delete_hook(struct k_itimer *timer) | |
834 | { | |
835 | return CLOCK_DISPATCH(timer->it_clock, timer_del, (timer)); | |
836 | } | |
837 | ||
838 | /* Delete a POSIX.1b interval timer. */ | |
839 | asmlinkage long | |
840 | sys_timer_delete(timer_t timer_id) | |
841 | { | |
842 | struct k_itimer *timer; | |
5ba25331 | 843 | unsigned long flags; |
1da177e4 | 844 | |
1da177e4 | 845 | retry_delete: |
1da177e4 LT |
846 | timer = lock_timer(timer_id, &flags); |
847 | if (!timer) | |
848 | return -EINVAL; | |
849 | ||
becf8b5d | 850 | if (timer_delete_hook(timer) == TIMER_RETRY) { |
1da177e4 LT |
851 | unlock_timer(timer, flags); |
852 | goto retry_delete; | |
853 | } | |
becf8b5d | 854 | |
1da177e4 LT |
855 | spin_lock(¤t->sighand->siglock); |
856 | list_del(&timer->list); | |
857 | spin_unlock(¤t->sighand->siglock); | |
858 | /* | |
859 | * This keeps any tasks waiting on the spin lock from thinking | |
860 | * they got something (see the lock code above). | |
861 | */ | |
862 | if (timer->it_process) { | |
863 | if (timer->it_sigev_notify == (SIGEV_SIGNAL|SIGEV_THREAD_ID)) | |
864 | put_task_struct(timer->it_process); | |
865 | timer->it_process = NULL; | |
866 | } | |
867 | unlock_timer(timer, flags); | |
868 | release_posix_timer(timer, IT_ID_SET); | |
869 | return 0; | |
870 | } | |
becf8b5d | 871 | |
1da177e4 LT |
872 | /* |
873 | * return timer owned by the process, used by exit_itimers | |
874 | */ | |
858119e1 | 875 | static void itimer_delete(struct k_itimer *timer) |
1da177e4 LT |
876 | { |
877 | unsigned long flags; | |
878 | ||
1da177e4 | 879 | retry_delete: |
1da177e4 LT |
880 | spin_lock_irqsave(&timer->it_lock, flags); |
881 | ||
becf8b5d | 882 | if (timer_delete_hook(timer) == TIMER_RETRY) { |
1da177e4 LT |
883 | unlock_timer(timer, flags); |
884 | goto retry_delete; | |
885 | } | |
1da177e4 LT |
886 | list_del(&timer->list); |
887 | /* | |
888 | * This keeps any tasks waiting on the spin lock from thinking | |
889 | * they got something (see the lock code above). | |
890 | */ | |
891 | if (timer->it_process) { | |
892 | if (timer->it_sigev_notify == (SIGEV_SIGNAL|SIGEV_THREAD_ID)) | |
893 | put_task_struct(timer->it_process); | |
894 | timer->it_process = NULL; | |
895 | } | |
896 | unlock_timer(timer, flags); | |
897 | release_posix_timer(timer, IT_ID_SET); | |
898 | } | |
899 | ||
900 | /* | |
25f407f0 | 901 | * This is called by do_exit or de_thread, only when there are no more |
1da177e4 LT |
902 | * references to the shared signal_struct. |
903 | */ | |
904 | void exit_itimers(struct signal_struct *sig) | |
905 | { | |
906 | struct k_itimer *tmr; | |
907 | ||
908 | while (!list_empty(&sig->posix_timers)) { | |
909 | tmr = list_entry(sig->posix_timers.next, struct k_itimer, list); | |
910 | itimer_delete(tmr); | |
911 | } | |
912 | } | |
913 | ||
becf8b5d | 914 | /* Not available / possible... functions */ |
a924b04d | 915 | int do_posix_clock_nosettime(const clockid_t clockid, struct timespec *tp) |
1da177e4 LT |
916 | { |
917 | return -EINVAL; | |
918 | } | |
919 | EXPORT_SYMBOL_GPL(do_posix_clock_nosettime); | |
920 | ||
a924b04d | 921 | int do_posix_clock_nonanosleep(const clockid_t clock, int flags, |
97735f25 | 922 | struct timespec *t, struct timespec __user *r) |
1da177e4 LT |
923 | { |
924 | #ifndef ENOTSUP | |
925 | return -EOPNOTSUPP; /* aka ENOTSUP in userland for POSIX */ | |
926 | #else /* parisc does define it separately. */ | |
927 | return -ENOTSUP; | |
928 | #endif | |
929 | } | |
930 | EXPORT_SYMBOL_GPL(do_posix_clock_nonanosleep); | |
931 | ||
a924b04d TG |
932 | asmlinkage long sys_clock_settime(const clockid_t which_clock, |
933 | const struct timespec __user *tp) | |
1da177e4 LT |
934 | { |
935 | struct timespec new_tp; | |
936 | ||
937 | if (invalid_clockid(which_clock)) | |
938 | return -EINVAL; | |
939 | if (copy_from_user(&new_tp, tp, sizeof (*tp))) | |
940 | return -EFAULT; | |
941 | ||
942 | return CLOCK_DISPATCH(which_clock, clock_set, (which_clock, &new_tp)); | |
943 | } | |
944 | ||
945 | asmlinkage long | |
a924b04d | 946 | sys_clock_gettime(const clockid_t which_clock, struct timespec __user *tp) |
1da177e4 LT |
947 | { |
948 | struct timespec kernel_tp; | |
949 | int error; | |
950 | ||
951 | if (invalid_clockid(which_clock)) | |
952 | return -EINVAL; | |
953 | error = CLOCK_DISPATCH(which_clock, clock_get, | |
954 | (which_clock, &kernel_tp)); | |
955 | if (!error && copy_to_user(tp, &kernel_tp, sizeof (kernel_tp))) | |
956 | error = -EFAULT; | |
957 | ||
958 | return error; | |
959 | ||
960 | } | |
961 | ||
962 | asmlinkage long | |
a924b04d | 963 | sys_clock_getres(const clockid_t which_clock, struct timespec __user *tp) |
1da177e4 LT |
964 | { |
965 | struct timespec rtn_tp; | |
966 | int error; | |
967 | ||
968 | if (invalid_clockid(which_clock)) | |
969 | return -EINVAL; | |
970 | ||
971 | error = CLOCK_DISPATCH(which_clock, clock_getres, | |
972 | (which_clock, &rtn_tp)); | |
973 | ||
974 | if (!error && tp && copy_to_user(tp, &rtn_tp, sizeof (rtn_tp))) { | |
975 | error = -EFAULT; | |
976 | } | |
977 | ||
978 | return error; | |
979 | } | |
980 | ||
97735f25 TG |
981 | /* |
982 | * nanosleep for monotonic and realtime clocks | |
983 | */ | |
984 | static int common_nsleep(const clockid_t which_clock, int flags, | |
985 | struct timespec *tsave, struct timespec __user *rmtp) | |
986 | { | |
080344b9 ON |
987 | return hrtimer_nanosleep(tsave, rmtp, flags & TIMER_ABSTIME ? |
988 | HRTIMER_MODE_ABS : HRTIMER_MODE_REL, | |
989 | which_clock); | |
97735f25 | 990 | } |
1da177e4 LT |
991 | |
992 | asmlinkage long | |
a924b04d | 993 | sys_clock_nanosleep(const clockid_t which_clock, int flags, |
1da177e4 LT |
994 | const struct timespec __user *rqtp, |
995 | struct timespec __user *rmtp) | |
996 | { | |
997 | struct timespec t; | |
1da177e4 LT |
998 | |
999 | if (invalid_clockid(which_clock)) | |
1000 | return -EINVAL; | |
1001 | ||
1002 | if (copy_from_user(&t, rqtp, sizeof (struct timespec))) | |
1003 | return -EFAULT; | |
1004 | ||
5f82b2b7 | 1005 | if (!timespec_valid(&t)) |
1da177e4 LT |
1006 | return -EINVAL; |
1007 | ||
97735f25 TG |
1008 | return CLOCK_DISPATCH(which_clock, nsleep, |
1009 | (which_clock, flags, &t, rmtp)); | |
1da177e4 | 1010 | } |
1711ef38 TA |
1011 | |
1012 | /* | |
1013 | * nanosleep_restart for monotonic and realtime clocks | |
1014 | */ | |
1015 | static int common_nsleep_restart(struct restart_block *restart_block) | |
1016 | { | |
1017 | return hrtimer_nanosleep_restart(restart_block); | |
1018 | } | |
1019 | ||
1020 | /* | |
1021 | * This will restart clock_nanosleep. This is required only by | |
1022 | * compat_clock_nanosleep_restart for now. | |
1023 | */ | |
1024 | long | |
1025 | clock_nanosleep_restart(struct restart_block *restart_block) | |
1026 | { | |
1027 | clockid_t which_clock = restart_block->arg0; | |
1028 | ||
1029 | return CLOCK_DISPATCH(which_clock, nsleep_restart, | |
1030 | (restart_block)); | |
1031 | } |