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 | |
27af4245 | 119 | * timer. 2.) The list, it_lock, it_clock, it_id and it_pid |
1da177e4 LT |
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 | ||
3d44cc3e TG |
200 | static int no_timer_create(struct k_itimer *new_timer) |
201 | { | |
202 | return -EOPNOTSUPP; | |
203 | } | |
204 | ||
70d715fd HS |
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 | ||
1da177e4 | 211 | /* |
becf8b5d | 212 | * Return nonzero if we know a priori this clockid_t value is bogus. |
1da177e4 | 213 | */ |
a924b04d | 214 | static inline int invalid_clockid(const clockid_t which_clock) |
1da177e4 LT |
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; | |
1da177e4 LT |
222 | if (posix_clocks[which_clock].res != 0) |
223 | return 0; | |
1da177e4 LT |
224 | return 1; |
225 | } | |
226 | ||
becf8b5d TG |
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 | } | |
1da177e4 | 235 | |
2d42244a JS |
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 | ||
1da177e4 LT |
245 | /* |
246 | * Initialize everything, well, just everything in Posix clocks/timers ;) | |
247 | */ | |
248 | static __init int init_posix_timers(void) | |
249 | { | |
becf8b5d TG |
250 | struct k_clock clock_realtime = { |
251 | .clock_getres = hrtimer_get_res, | |
1da177e4 | 252 | }; |
becf8b5d TG |
253 | struct k_clock clock_monotonic = { |
254 | .clock_getres = hrtimer_get_res, | |
255 | .clock_get = posix_ktime_get_ts, | |
256 | .clock_set = do_posix_clock_nosettime, | |
1da177e4 | 257 | }; |
2d42244a JS |
258 | struct k_clock clock_monotonic_raw = { |
259 | .clock_getres = hrtimer_get_res, | |
260 | .clock_get = posix_get_monotonic_raw, | |
261 | .clock_set = do_posix_clock_nosettime, | |
3d44cc3e | 262 | .timer_create = no_timer_create, |
70d715fd | 263 | .nsleep = no_nsleep, |
2d42244a | 264 | }; |
1da177e4 LT |
265 | |
266 | register_posix_clock(CLOCK_REALTIME, &clock_realtime); | |
267 | register_posix_clock(CLOCK_MONOTONIC, &clock_monotonic); | |
2d42244a | 268 | register_posix_clock(CLOCK_MONOTONIC_RAW, &clock_monotonic_raw); |
1da177e4 LT |
269 | |
270 | posix_timers_cache = kmem_cache_create("posix_timers_cache", | |
040b5c6f AD |
271 | sizeof (struct k_itimer), 0, SLAB_PANIC, |
272 | NULL); | |
1da177e4 LT |
273 | idr_init(&posix_timers_id); |
274 | return 0; | |
275 | } | |
276 | ||
277 | __initcall(init_posix_timers); | |
278 | ||
1da177e4 LT |
279 | static void schedule_next_timer(struct k_itimer *timr) |
280 | { | |
44f21475 RZ |
281 | struct hrtimer *timer = &timr->it.real.timer; |
282 | ||
becf8b5d | 283 | if (timr->it.real.interval.tv64 == 0) |
1da177e4 LT |
284 | return; |
285 | ||
4d672e7a DL |
286 | timr->it_overrun += (unsigned int) hrtimer_forward(timer, |
287 | timer->base->get_time(), | |
288 | timr->it.real.interval); | |
44f21475 | 289 | |
1da177e4 LT |
290 | timr->it_overrun_last = timr->it_overrun; |
291 | timr->it_overrun = -1; | |
292 | ++timr->it_requeue_pending; | |
44f21475 | 293 | hrtimer_restart(timer); |
1da177e4 LT |
294 | } |
295 | ||
296 | /* | |
297 | * This function is exported for use by the signal deliver code. It is | |
298 | * called just prior to the info block being released and passes that | |
299 | * block to us. It's function is to update the overrun entry AND to | |
300 | * restart the timer. It should only be called if the timer is to be | |
301 | * restarted (i.e. we have flagged this in the sys_private entry of the | |
302 | * info block). | |
303 | * | |
304 | * To protect aginst the timer going away while the interrupt is queued, | |
305 | * we require that the it_requeue_pending flag be set. | |
306 | */ | |
307 | void do_schedule_next_timer(struct siginfo *info) | |
308 | { | |
309 | struct k_itimer *timr; | |
310 | unsigned long flags; | |
311 | ||
312 | timr = lock_timer(info->si_tid, &flags); | |
313 | ||
becf8b5d TG |
314 | if (timr && timr->it_requeue_pending == info->si_sys_private) { |
315 | if (timr->it_clock < 0) | |
316 | posix_cpu_timer_schedule(timr); | |
317 | else | |
318 | schedule_next_timer(timr); | |
1da177e4 | 319 | |
54da1174 | 320 | info->si_overrun += timr->it_overrun_last; |
becf8b5d TG |
321 | } |
322 | ||
b6557fbc TG |
323 | if (timr) |
324 | unlock_timer(timr, flags); | |
1da177e4 LT |
325 | } |
326 | ||
ba661292 | 327 | int posix_timer_event(struct k_itimer *timr, int si_private) |
1da177e4 | 328 | { |
27af4245 ON |
329 | struct task_struct *task; |
330 | int shared, ret = -1; | |
ba661292 ON |
331 | /* |
332 | * FIXME: if ->sigq is queued we can race with | |
333 | * dequeue_signal()->do_schedule_next_timer(). | |
334 | * | |
335 | * If dequeue_signal() sees the "right" value of | |
336 | * si_sys_private it calls do_schedule_next_timer(). | |
337 | * We re-queue ->sigq and drop ->it_lock(). | |
338 | * do_schedule_next_timer() locks the timer | |
339 | * and re-schedules it while ->sigq is pending. | |
340 | * Not really bad, but not that we want. | |
341 | */ | |
1da177e4 | 342 | timr->sigq->info.si_sys_private = si_private; |
1da177e4 | 343 | |
27af4245 ON |
344 | rcu_read_lock(); |
345 | task = pid_task(timr->it_pid, PIDTYPE_PID); | |
346 | if (task) { | |
347 | shared = !(timr->it_sigev_notify & SIGEV_THREAD_ID); | |
348 | ret = send_sigqueue(timr->sigq, task, shared); | |
349 | } | |
350 | rcu_read_unlock(); | |
4aa73611 ON |
351 | /* If we failed to send the signal the timer stops. */ |
352 | return ret > 0; | |
1da177e4 LT |
353 | } |
354 | EXPORT_SYMBOL_GPL(posix_timer_event); | |
355 | ||
356 | /* | |
357 | * This function gets called when a POSIX.1b interval timer expires. It | |
358 | * is used as a callback from the kernel internal timer. The | |
359 | * run_timer_list code ALWAYS calls with interrupts on. | |
360 | ||
361 | * This code is for CLOCK_REALTIME* and CLOCK_MONOTONIC* timers. | |
362 | */ | |
c9cb2e3d | 363 | static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer) |
1da177e4 | 364 | { |
05cfb614 | 365 | struct k_itimer *timr; |
1da177e4 | 366 | unsigned long flags; |
becf8b5d | 367 | int si_private = 0; |
c9cb2e3d | 368 | enum hrtimer_restart ret = HRTIMER_NORESTART; |
1da177e4 | 369 | |
05cfb614 | 370 | timr = container_of(timer, struct k_itimer, it.real.timer); |
1da177e4 | 371 | spin_lock_irqsave(&timr->it_lock, flags); |
1da177e4 | 372 | |
becf8b5d TG |
373 | if (timr->it.real.interval.tv64 != 0) |
374 | si_private = ++timr->it_requeue_pending; | |
1da177e4 | 375 | |
becf8b5d TG |
376 | if (posix_timer_event(timr, si_private)) { |
377 | /* | |
378 | * signal was not sent because of sig_ignor | |
379 | * we will not get a call back to restart it AND | |
380 | * it should be restarted. | |
381 | */ | |
382 | if (timr->it.real.interval.tv64 != 0) { | |
58229a18 TG |
383 | ktime_t now = hrtimer_cb_get_time(timer); |
384 | ||
385 | /* | |
386 | * FIXME: What we really want, is to stop this | |
387 | * timer completely and restart it in case the | |
388 | * SIG_IGN is removed. This is a non trivial | |
389 | * change which involves sighand locking | |
390 | * (sigh !), which we don't want to do late in | |
391 | * the release cycle. | |
392 | * | |
393 | * For now we just let timers with an interval | |
394 | * less than a jiffie expire every jiffie to | |
395 | * avoid softirq starvation in case of SIG_IGN | |
396 | * and a very small interval, which would put | |
397 | * the timer right back on the softirq pending | |
398 | * list. By moving now ahead of time we trick | |
399 | * hrtimer_forward() to expire the timer | |
400 | * later, while we still maintain the overrun | |
401 | * accuracy, but have some inconsistency in | |
402 | * the timer_gettime() case. This is at least | |
403 | * better than a starved softirq. A more | |
404 | * complex fix which solves also another related | |
405 | * inconsistency is already in the pipeline. | |
406 | */ | |
407 | #ifdef CONFIG_HIGH_RES_TIMERS | |
408 | { | |
409 | ktime_t kj = ktime_set(0, NSEC_PER_SEC / HZ); | |
410 | ||
411 | if (timr->it.real.interval.tv64 < kj.tv64) | |
412 | now = ktime_add(now, kj); | |
413 | } | |
414 | #endif | |
4d672e7a | 415 | timr->it_overrun += (unsigned int) |
58229a18 | 416 | hrtimer_forward(timer, now, |
becf8b5d TG |
417 | timr->it.real.interval); |
418 | ret = HRTIMER_RESTART; | |
a0a0c28c | 419 | ++timr->it_requeue_pending; |
1da177e4 | 420 | } |
1da177e4 | 421 | } |
1da177e4 | 422 | |
becf8b5d TG |
423 | unlock_timer(timr, flags); |
424 | return ret; | |
425 | } | |
1da177e4 | 426 | |
27af4245 | 427 | static struct pid *good_sigevent(sigevent_t * event) |
1da177e4 LT |
428 | { |
429 | struct task_struct *rtn = current->group_leader; | |
430 | ||
431 | if ((event->sigev_notify & SIGEV_THREAD_ID ) && | |
8dc86af0 | 432 | (!(rtn = find_task_by_vpid(event->sigev_notify_thread_id)) || |
bac0abd6 | 433 | !same_thread_group(rtn, current) || |
1da177e4 LT |
434 | (event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_SIGNAL)) |
435 | return NULL; | |
436 | ||
437 | if (((event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) && | |
438 | ((event->sigev_signo <= 0) || (event->sigev_signo > SIGRTMAX))) | |
439 | return NULL; | |
440 | ||
27af4245 | 441 | return task_pid(rtn); |
1da177e4 LT |
442 | } |
443 | ||
a924b04d | 444 | void register_posix_clock(const clockid_t clock_id, struct k_clock *new_clock) |
1da177e4 LT |
445 | { |
446 | if ((unsigned) clock_id >= MAX_CLOCKS) { | |
447 | printk("POSIX clock register failed for clock_id %d\n", | |
448 | clock_id); | |
449 | return; | |
450 | } | |
451 | ||
452 | posix_clocks[clock_id] = *new_clock; | |
453 | } | |
454 | EXPORT_SYMBOL_GPL(register_posix_clock); | |
455 | ||
456 | static struct k_itimer * alloc_posix_timer(void) | |
457 | { | |
458 | struct k_itimer *tmr; | |
c3762229 | 459 | tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL); |
1da177e4 LT |
460 | if (!tmr) |
461 | return tmr; | |
1da177e4 LT |
462 | if (unlikely(!(tmr->sigq = sigqueue_alloc()))) { |
463 | kmem_cache_free(posix_timers_cache, tmr); | |
aa94fbd5 | 464 | return NULL; |
1da177e4 | 465 | } |
ba661292 | 466 | memset(&tmr->sigq->info, 0, sizeof(siginfo_t)); |
1da177e4 LT |
467 | return tmr; |
468 | } | |
469 | ||
470 | #define IT_ID_SET 1 | |
471 | #define IT_ID_NOT_SET 0 | |
472 | static void release_posix_timer(struct k_itimer *tmr, int it_id_set) | |
473 | { | |
474 | if (it_id_set) { | |
475 | unsigned long flags; | |
476 | spin_lock_irqsave(&idr_lock, flags); | |
477 | idr_remove(&posix_timers_id, tmr->it_id); | |
478 | spin_unlock_irqrestore(&idr_lock, flags); | |
479 | } | |
89992102 | 480 | put_pid(tmr->it_pid); |
1da177e4 | 481 | sigqueue_free(tmr->sigq); |
1da177e4 LT |
482 | kmem_cache_free(posix_timers_cache, tmr); |
483 | } | |
484 | ||
485 | /* Create a POSIX.1b interval timer. */ | |
486 | ||
362e9c07 HC |
487 | SYSCALL_DEFINE3(timer_create, const clockid_t, which_clock, |
488 | struct sigevent __user *, timer_event_spec, | |
489 | timer_t __user *, created_timer_id) | |
1da177e4 | 490 | { |
2cd499e3 | 491 | struct k_itimer *new_timer; |
ef864c95 | 492 | int error, new_timer_id; |
1da177e4 LT |
493 | sigevent_t event; |
494 | int it_id_set = IT_ID_NOT_SET; | |
495 | ||
496 | if (invalid_clockid(which_clock)) | |
497 | return -EINVAL; | |
498 | ||
499 | new_timer = alloc_posix_timer(); | |
500 | if (unlikely(!new_timer)) | |
501 | return -EAGAIN; | |
502 | ||
503 | spin_lock_init(&new_timer->it_lock); | |
504 | retry: | |
505 | if (unlikely(!idr_pre_get(&posix_timers_id, GFP_KERNEL))) { | |
506 | error = -EAGAIN; | |
507 | goto out; | |
508 | } | |
509 | spin_lock_irq(&idr_lock); | |
5a51b713 | 510 | error = idr_get_new(&posix_timers_id, new_timer, &new_timer_id); |
1da177e4 | 511 | spin_unlock_irq(&idr_lock); |
ef864c95 ON |
512 | if (error) { |
513 | if (error == -EAGAIN) | |
514 | goto retry; | |
1da177e4 | 515 | /* |
0b0a3e7b | 516 | * Weird looking, but we return EAGAIN if the IDR is |
1da177e4 LT |
517 | * full (proper POSIX return value for this) |
518 | */ | |
519 | error = -EAGAIN; | |
520 | goto out; | |
521 | } | |
522 | ||
523 | it_id_set = IT_ID_SET; | |
524 | new_timer->it_id = (timer_t) new_timer_id; | |
525 | new_timer->it_clock = which_clock; | |
526 | new_timer->it_overrun = -1; | |
527 | error = CLOCK_DISPATCH(which_clock, timer_create, (new_timer)); | |
528 | if (error) | |
529 | goto out; | |
530 | ||
531 | /* | |
532 | * return the timer_id now. The next step is hard to | |
533 | * back out if there is an error. | |
534 | */ | |
535 | if (copy_to_user(created_timer_id, | |
536 | &new_timer_id, sizeof (new_timer_id))) { | |
537 | error = -EFAULT; | |
538 | goto out; | |
539 | } | |
540 | if (timer_event_spec) { | |
541 | if (copy_from_user(&event, timer_event_spec, sizeof (event))) { | |
542 | error = -EFAULT; | |
543 | goto out; | |
544 | } | |
36b2f046 | 545 | rcu_read_lock(); |
89992102 | 546 | new_timer->it_pid = get_pid(good_sigevent(&event)); |
36b2f046 | 547 | rcu_read_unlock(); |
89992102 | 548 | if (!new_timer->it_pid) { |
1da177e4 LT |
549 | error = -EINVAL; |
550 | goto out; | |
551 | } | |
552 | } else { | |
5a9fa730 ON |
553 | event.sigev_notify = SIGEV_SIGNAL; |
554 | event.sigev_signo = SIGALRM; | |
555 | event.sigev_value.sival_int = new_timer->it_id; | |
89992102 | 556 | new_timer->it_pid = get_pid(task_tgid(current)); |
1da177e4 LT |
557 | } |
558 | ||
5a9fa730 ON |
559 | new_timer->it_sigev_notify = event.sigev_notify; |
560 | new_timer->sigq->info.si_signo = event.sigev_signo; | |
561 | new_timer->sigq->info.si_value = event.sigev_value; | |
717835d9 | 562 | new_timer->sigq->info.si_tid = new_timer->it_id; |
5a9fa730 | 563 | new_timer->sigq->info.si_code = SI_TIMER; |
717835d9 | 564 | |
36b2f046 | 565 | spin_lock_irq(¤t->sighand->siglock); |
27af4245 | 566 | new_timer->it_signal = current->signal; |
36b2f046 ON |
567 | list_add(&new_timer->list, ¤t->signal->posix_timers); |
568 | spin_unlock_irq(¤t->sighand->siglock); | |
ef864c95 ON |
569 | |
570 | return 0; | |
1da177e4 LT |
571 | /* |
572 | * In the case of the timer belonging to another task, after | |
573 | * the task is unlocked, the timer is owned by the other task | |
574 | * and may cease to exist at any time. Don't use or modify | |
575 | * new_timer after the unlock call. | |
576 | */ | |
1da177e4 | 577 | out: |
ef864c95 | 578 | release_posix_timer(new_timer, it_id_set); |
1da177e4 LT |
579 | return error; |
580 | } | |
581 | ||
1da177e4 LT |
582 | /* |
583 | * Locking issues: We need to protect the result of the id look up until | |
584 | * we get the timer locked down so it is not deleted under us. The | |
585 | * removal is done under the idr spinlock so we use that here to bridge | |
586 | * the find to the timer lock. To avoid a dead lock, the timer id MUST | |
587 | * be release with out holding the timer lock. | |
588 | */ | |
31d92845 | 589 | static struct k_itimer *lock_timer(timer_t timer_id, unsigned long *flags) |
1da177e4 LT |
590 | { |
591 | struct k_itimer *timr; | |
592 | /* | |
593 | * Watch out here. We do a irqsave on the idr_lock and pass the | |
594 | * flags part over to the timer lock. Must not let interrupts in | |
595 | * while we are moving the lock. | |
596 | */ | |
1da177e4 | 597 | spin_lock_irqsave(&idr_lock, *flags); |
31d92845 | 598 | timr = idr_find(&posix_timers_id, (int)timer_id); |
1da177e4 LT |
599 | if (timr) { |
600 | spin_lock(&timr->it_lock); | |
89992102 | 601 | if (timr->it_signal == current->signal) { |
179394af | 602 | spin_unlock(&idr_lock); |
31d92845 ON |
603 | return timr; |
604 | } | |
605 | spin_unlock(&timr->it_lock); | |
606 | } | |
607 | spin_unlock_irqrestore(&idr_lock, *flags); | |
1da177e4 | 608 | |
31d92845 | 609 | return NULL; |
1da177e4 LT |
610 | } |
611 | ||
612 | /* | |
613 | * Get the time remaining on a POSIX.1b interval timer. This function | |
614 | * is ALWAYS called with spin_lock_irq on the timer, thus it must not | |
615 | * mess with irq. | |
616 | * | |
617 | * We have a couple of messes to clean up here. First there is the case | |
618 | * of a timer that has a requeue pending. These timers should appear to | |
619 | * be in the timer list with an expiry as if we were to requeue them | |
620 | * now. | |
621 | * | |
622 | * The second issue is the SIGEV_NONE timer which may be active but is | |
623 | * not really ever put in the timer list (to save system resources). | |
624 | * This timer may be expired, and if so, we will do it here. Otherwise | |
625 | * it is the same as a requeue pending timer WRT to what we should | |
626 | * report. | |
627 | */ | |
628 | static void | |
629 | common_timer_get(struct k_itimer *timr, struct itimerspec *cur_setting) | |
630 | { | |
3b98a532 | 631 | ktime_t now, remaining, iv; |
becf8b5d | 632 | struct hrtimer *timer = &timr->it.real.timer; |
1da177e4 | 633 | |
becf8b5d | 634 | memset(cur_setting, 0, sizeof(struct itimerspec)); |
becf8b5d | 635 | |
3b98a532 RZ |
636 | iv = timr->it.real.interval; |
637 | ||
becf8b5d | 638 | /* interval timer ? */ |
3b98a532 RZ |
639 | if (iv.tv64) |
640 | cur_setting->it_interval = ktime_to_timespec(iv); | |
641 | else if (!hrtimer_active(timer) && | |
642 | (timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) | |
becf8b5d | 643 | return; |
3b98a532 RZ |
644 | |
645 | now = timer->base->get_time(); | |
646 | ||
becf8b5d | 647 | /* |
3b98a532 RZ |
648 | * When a requeue is pending or this is a SIGEV_NONE |
649 | * timer move the expiry time forward by intervals, so | |
650 | * expiry is > now. | |
becf8b5d | 651 | */ |
3b98a532 RZ |
652 | if (iv.tv64 && (timr->it_requeue_pending & REQUEUE_PENDING || |
653 | (timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE)) | |
4d672e7a | 654 | timr->it_overrun += (unsigned int) hrtimer_forward(timer, now, iv); |
3b98a532 | 655 | |
cc584b21 | 656 | remaining = ktime_sub(hrtimer_get_expires(timer), now); |
becf8b5d | 657 | /* Return 0 only, when the timer is expired and not pending */ |
3b98a532 RZ |
658 | if (remaining.tv64 <= 0) { |
659 | /* | |
660 | * A single shot SIGEV_NONE timer must return 0, when | |
661 | * it is expired ! | |
662 | */ | |
663 | if ((timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) | |
664 | cur_setting->it_value.tv_nsec = 1; | |
665 | } else | |
becf8b5d | 666 | cur_setting->it_value = ktime_to_timespec(remaining); |
1da177e4 LT |
667 | } |
668 | ||
669 | /* Get the time remaining on a POSIX.1b interval timer. */ | |
362e9c07 HC |
670 | SYSCALL_DEFINE2(timer_gettime, timer_t, timer_id, |
671 | struct itimerspec __user *, setting) | |
1da177e4 LT |
672 | { |
673 | struct k_itimer *timr; | |
674 | struct itimerspec cur_setting; | |
675 | unsigned long flags; | |
676 | ||
677 | timr = lock_timer(timer_id, &flags); | |
678 | if (!timr) | |
679 | return -EINVAL; | |
680 | ||
681 | CLOCK_DISPATCH(timr->it_clock, timer_get, (timr, &cur_setting)); | |
682 | ||
683 | unlock_timer(timr, flags); | |
684 | ||
685 | if (copy_to_user(setting, &cur_setting, sizeof (cur_setting))) | |
686 | return -EFAULT; | |
687 | ||
688 | return 0; | |
689 | } | |
becf8b5d | 690 | |
1da177e4 LT |
691 | /* |
692 | * Get the number of overruns of a POSIX.1b interval timer. This is to | |
693 | * be the overrun of the timer last delivered. At the same time we are | |
694 | * accumulating overruns on the next timer. The overrun is frozen when | |
695 | * the signal is delivered, either at the notify time (if the info block | |
696 | * is not queued) or at the actual delivery time (as we are informed by | |
697 | * the call back to do_schedule_next_timer(). So all we need to do is | |
698 | * to pick up the frozen overrun. | |
699 | */ | |
362e9c07 | 700 | SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id) |
1da177e4 LT |
701 | { |
702 | struct k_itimer *timr; | |
703 | int overrun; | |
5ba25331 | 704 | unsigned long flags; |
1da177e4 LT |
705 | |
706 | timr = lock_timer(timer_id, &flags); | |
707 | if (!timr) | |
708 | return -EINVAL; | |
709 | ||
710 | overrun = timr->it_overrun_last; | |
711 | unlock_timer(timr, flags); | |
712 | ||
713 | return overrun; | |
714 | } | |
1da177e4 LT |
715 | |
716 | /* Set a POSIX.1b interval timer. */ | |
717 | /* timr->it_lock is taken. */ | |
858119e1 | 718 | static int |
1da177e4 LT |
719 | common_timer_set(struct k_itimer *timr, int flags, |
720 | struct itimerspec *new_setting, struct itimerspec *old_setting) | |
721 | { | |
becf8b5d | 722 | struct hrtimer *timer = &timr->it.real.timer; |
7978672c | 723 | enum hrtimer_mode mode; |
1da177e4 LT |
724 | |
725 | if (old_setting) | |
726 | common_timer_get(timr, old_setting); | |
727 | ||
728 | /* disable the timer */ | |
becf8b5d | 729 | timr->it.real.interval.tv64 = 0; |
1da177e4 LT |
730 | /* |
731 | * careful here. If smp we could be in the "fire" routine which will | |
732 | * be spinning as we hold the lock. But this is ONLY an SMP issue. | |
733 | */ | |
becf8b5d | 734 | if (hrtimer_try_to_cancel(timer) < 0) |
1da177e4 | 735 | return TIMER_RETRY; |
1da177e4 LT |
736 | |
737 | timr->it_requeue_pending = (timr->it_requeue_pending + 2) & | |
738 | ~REQUEUE_PENDING; | |
739 | timr->it_overrun_last = 0; | |
1da177e4 | 740 | |
becf8b5d TG |
741 | /* switch off the timer when it_value is zero */ |
742 | if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec) | |
743 | return 0; | |
1da177e4 | 744 | |
c9cb2e3d | 745 | mode = flags & TIMER_ABSTIME ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL; |
7978672c | 746 | hrtimer_init(&timr->it.real.timer, timr->it_clock, mode); |
7978672c | 747 | timr->it.real.timer.function = posix_timer_fn; |
becf8b5d | 748 | |
cc584b21 | 749 | hrtimer_set_expires(timer, timespec_to_ktime(new_setting->it_value)); |
becf8b5d TG |
750 | |
751 | /* Convert interval */ | |
752 | timr->it.real.interval = timespec_to_ktime(new_setting->it_interval); | |
753 | ||
754 | /* SIGEV_NONE timers are not queued ! See common_timer_get */ | |
952bbc87 TG |
755 | if (((timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE)) { |
756 | /* Setup correct expiry time for relative timers */ | |
5a7780e7 | 757 | if (mode == HRTIMER_MODE_REL) { |
cc584b21 | 758 | hrtimer_add_expires(timer, timer->base->get_time()); |
5a7780e7 | 759 | } |
becf8b5d | 760 | return 0; |
952bbc87 | 761 | } |
becf8b5d | 762 | |
cc584b21 | 763 | hrtimer_start_expires(timer, mode); |
1da177e4 LT |
764 | return 0; |
765 | } | |
766 | ||
767 | /* Set a POSIX.1b interval timer */ | |
362e9c07 HC |
768 | SYSCALL_DEFINE4(timer_settime, timer_t, timer_id, int, flags, |
769 | const struct itimerspec __user *, new_setting, | |
770 | struct itimerspec __user *, old_setting) | |
1da177e4 LT |
771 | { |
772 | struct k_itimer *timr; | |
773 | struct itimerspec new_spec, old_spec; | |
774 | int error = 0; | |
5ba25331 | 775 | unsigned long flag; |
1da177e4 LT |
776 | struct itimerspec *rtn = old_setting ? &old_spec : NULL; |
777 | ||
778 | if (!new_setting) | |
779 | return -EINVAL; | |
780 | ||
781 | if (copy_from_user(&new_spec, new_setting, sizeof (new_spec))) | |
782 | return -EFAULT; | |
783 | ||
becf8b5d TG |
784 | if (!timespec_valid(&new_spec.it_interval) || |
785 | !timespec_valid(&new_spec.it_value)) | |
1da177e4 LT |
786 | return -EINVAL; |
787 | retry: | |
788 | timr = lock_timer(timer_id, &flag); | |
789 | if (!timr) | |
790 | return -EINVAL; | |
791 | ||
792 | error = CLOCK_DISPATCH(timr->it_clock, timer_set, | |
793 | (timr, flags, &new_spec, rtn)); | |
794 | ||
795 | unlock_timer(timr, flag); | |
796 | if (error == TIMER_RETRY) { | |
797 | rtn = NULL; // We already got the old time... | |
798 | goto retry; | |
799 | } | |
800 | ||
becf8b5d TG |
801 | if (old_setting && !error && |
802 | copy_to_user(old_setting, &old_spec, sizeof (old_spec))) | |
1da177e4 LT |
803 | error = -EFAULT; |
804 | ||
805 | return error; | |
806 | } | |
807 | ||
808 | static inline int common_timer_del(struct k_itimer *timer) | |
809 | { | |
becf8b5d | 810 | timer->it.real.interval.tv64 = 0; |
f972be33 | 811 | |
becf8b5d | 812 | if (hrtimer_try_to_cancel(&timer->it.real.timer) < 0) |
1da177e4 | 813 | return TIMER_RETRY; |
1da177e4 LT |
814 | return 0; |
815 | } | |
816 | ||
817 | static inline int timer_delete_hook(struct k_itimer *timer) | |
818 | { | |
819 | return CLOCK_DISPATCH(timer->it_clock, timer_del, (timer)); | |
820 | } | |
821 | ||
822 | /* Delete a POSIX.1b interval timer. */ | |
362e9c07 | 823 | SYSCALL_DEFINE1(timer_delete, timer_t, timer_id) |
1da177e4 LT |
824 | { |
825 | struct k_itimer *timer; | |
5ba25331 | 826 | unsigned long flags; |
1da177e4 | 827 | |
1da177e4 | 828 | retry_delete: |
1da177e4 LT |
829 | timer = lock_timer(timer_id, &flags); |
830 | if (!timer) | |
831 | return -EINVAL; | |
832 | ||
becf8b5d | 833 | if (timer_delete_hook(timer) == TIMER_RETRY) { |
1da177e4 LT |
834 | unlock_timer(timer, flags); |
835 | goto retry_delete; | |
836 | } | |
becf8b5d | 837 | |
1da177e4 LT |
838 | spin_lock(¤t->sighand->siglock); |
839 | list_del(&timer->list); | |
840 | spin_unlock(¤t->sighand->siglock); | |
841 | /* | |
842 | * This keeps any tasks waiting on the spin lock from thinking | |
843 | * they got something (see the lock code above). | |
844 | */ | |
89992102 | 845 | timer->it_signal = NULL; |
4b7a1304 | 846 | |
1da177e4 LT |
847 | unlock_timer(timer, flags); |
848 | release_posix_timer(timer, IT_ID_SET); | |
849 | return 0; | |
850 | } | |
becf8b5d | 851 | |
1da177e4 LT |
852 | /* |
853 | * return timer owned by the process, used by exit_itimers | |
854 | */ | |
858119e1 | 855 | static void itimer_delete(struct k_itimer *timer) |
1da177e4 LT |
856 | { |
857 | unsigned long flags; | |
858 | ||
1da177e4 | 859 | retry_delete: |
1da177e4 LT |
860 | spin_lock_irqsave(&timer->it_lock, flags); |
861 | ||
becf8b5d | 862 | if (timer_delete_hook(timer) == TIMER_RETRY) { |
1da177e4 LT |
863 | unlock_timer(timer, flags); |
864 | goto retry_delete; | |
865 | } | |
1da177e4 LT |
866 | list_del(&timer->list); |
867 | /* | |
868 | * This keeps any tasks waiting on the spin lock from thinking | |
869 | * they got something (see the lock code above). | |
870 | */ | |
89992102 | 871 | timer->it_signal = NULL; |
4b7a1304 | 872 | |
1da177e4 LT |
873 | unlock_timer(timer, flags); |
874 | release_posix_timer(timer, IT_ID_SET); | |
875 | } | |
876 | ||
877 | /* | |
25f407f0 | 878 | * This is called by do_exit or de_thread, only when there are no more |
1da177e4 LT |
879 | * references to the shared signal_struct. |
880 | */ | |
881 | void exit_itimers(struct signal_struct *sig) | |
882 | { | |
883 | struct k_itimer *tmr; | |
884 | ||
885 | while (!list_empty(&sig->posix_timers)) { | |
886 | tmr = list_entry(sig->posix_timers.next, struct k_itimer, list); | |
887 | itimer_delete(tmr); | |
888 | } | |
889 | } | |
890 | ||
becf8b5d | 891 | /* Not available / possible... functions */ |
a924b04d | 892 | int do_posix_clock_nosettime(const clockid_t clockid, struct timespec *tp) |
1da177e4 LT |
893 | { |
894 | return -EINVAL; | |
895 | } | |
896 | EXPORT_SYMBOL_GPL(do_posix_clock_nosettime); | |
897 | ||
a924b04d | 898 | int do_posix_clock_nonanosleep(const clockid_t clock, int flags, |
97735f25 | 899 | struct timespec *t, struct timespec __user *r) |
1da177e4 LT |
900 | { |
901 | #ifndef ENOTSUP | |
902 | return -EOPNOTSUPP; /* aka ENOTSUP in userland for POSIX */ | |
903 | #else /* parisc does define it separately. */ | |
904 | return -ENOTSUP; | |
905 | #endif | |
906 | } | |
907 | EXPORT_SYMBOL_GPL(do_posix_clock_nonanosleep); | |
908 | ||
362e9c07 HC |
909 | SYSCALL_DEFINE2(clock_settime, const clockid_t, which_clock, |
910 | const struct timespec __user *, tp) | |
1da177e4 LT |
911 | { |
912 | struct timespec new_tp; | |
913 | ||
914 | if (invalid_clockid(which_clock)) | |
915 | return -EINVAL; | |
916 | if (copy_from_user(&new_tp, tp, sizeof (*tp))) | |
917 | return -EFAULT; | |
918 | ||
919 | return CLOCK_DISPATCH(which_clock, clock_set, (which_clock, &new_tp)); | |
920 | } | |
921 | ||
362e9c07 HC |
922 | SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock, |
923 | struct timespec __user *,tp) | |
1da177e4 LT |
924 | { |
925 | struct timespec kernel_tp; | |
926 | int error; | |
927 | ||
928 | if (invalid_clockid(which_clock)) | |
929 | return -EINVAL; | |
930 | error = CLOCK_DISPATCH(which_clock, clock_get, | |
931 | (which_clock, &kernel_tp)); | |
932 | if (!error && copy_to_user(tp, &kernel_tp, sizeof (kernel_tp))) | |
933 | error = -EFAULT; | |
934 | ||
935 | return error; | |
936 | ||
937 | } | |
938 | ||
362e9c07 HC |
939 | SYSCALL_DEFINE2(clock_getres, const clockid_t, which_clock, |
940 | struct timespec __user *, tp) | |
1da177e4 LT |
941 | { |
942 | struct timespec rtn_tp; | |
943 | int error; | |
944 | ||
945 | if (invalid_clockid(which_clock)) | |
946 | return -EINVAL; | |
947 | ||
948 | error = CLOCK_DISPATCH(which_clock, clock_getres, | |
949 | (which_clock, &rtn_tp)); | |
950 | ||
951 | if (!error && tp && copy_to_user(tp, &rtn_tp, sizeof (rtn_tp))) { | |
952 | error = -EFAULT; | |
953 | } | |
954 | ||
955 | return error; | |
956 | } | |
957 | ||
97735f25 TG |
958 | /* |
959 | * nanosleep for monotonic and realtime clocks | |
960 | */ | |
961 | static int common_nsleep(const clockid_t which_clock, int flags, | |
962 | struct timespec *tsave, struct timespec __user *rmtp) | |
963 | { | |
080344b9 ON |
964 | return hrtimer_nanosleep(tsave, rmtp, flags & TIMER_ABSTIME ? |
965 | HRTIMER_MODE_ABS : HRTIMER_MODE_REL, | |
966 | which_clock); | |
97735f25 | 967 | } |
1da177e4 | 968 | |
362e9c07 HC |
969 | SYSCALL_DEFINE4(clock_nanosleep, const clockid_t, which_clock, int, flags, |
970 | const struct timespec __user *, rqtp, | |
971 | struct timespec __user *, rmtp) | |
1da177e4 LT |
972 | { |
973 | struct timespec t; | |
1da177e4 LT |
974 | |
975 | if (invalid_clockid(which_clock)) | |
976 | return -EINVAL; | |
977 | ||
978 | if (copy_from_user(&t, rqtp, sizeof (struct timespec))) | |
979 | return -EFAULT; | |
980 | ||
5f82b2b7 | 981 | if (!timespec_valid(&t)) |
1da177e4 LT |
982 | return -EINVAL; |
983 | ||
97735f25 TG |
984 | return CLOCK_DISPATCH(which_clock, nsleep, |
985 | (which_clock, flags, &t, rmtp)); | |
1da177e4 | 986 | } |
1711ef38 TA |
987 | |
988 | /* | |
989 | * nanosleep_restart for monotonic and realtime clocks | |
990 | */ | |
991 | static int common_nsleep_restart(struct restart_block *restart_block) | |
992 | { | |
993 | return hrtimer_nanosleep_restart(restart_block); | |
994 | } | |
995 | ||
996 | /* | |
997 | * This will restart clock_nanosleep. This is required only by | |
998 | * compat_clock_nanosleep_restart for now. | |
999 | */ | |
1000 | long | |
1001 | clock_nanosleep_restart(struct restart_block *restart_block) | |
1002 | { | |
1003 | clockid_t which_clock = restart_block->arg0; | |
1004 | ||
1005 | return CLOCK_DISPATCH(which_clock, nsleep_restart, | |
1006 | (restart_block)); | |
1007 | } |