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 | ||
1da177e4 | 205 | /* |
becf8b5d | 206 | * Return nonzero if we know a priori this clockid_t value is bogus. |
1da177e4 | 207 | */ |
a924b04d | 208 | static inline int invalid_clockid(const clockid_t which_clock) |
1da177e4 LT |
209 | { |
210 | if (which_clock < 0) /* CPU clock, posix_cpu_* will check it */ | |
211 | return 0; | |
212 | if ((unsigned) which_clock >= MAX_CLOCKS) | |
213 | return 1; | |
214 | if (posix_clocks[which_clock].clock_getres != NULL) | |
215 | return 0; | |
1da177e4 LT |
216 | if (posix_clocks[which_clock].res != 0) |
217 | return 0; | |
1da177e4 LT |
218 | return 1; |
219 | } | |
220 | ||
becf8b5d TG |
221 | /* |
222 | * Get monotonic time for posix timers | |
223 | */ | |
224 | static int posix_ktime_get_ts(clockid_t which_clock, struct timespec *tp) | |
225 | { | |
226 | ktime_get_ts(tp); | |
227 | return 0; | |
228 | } | |
1da177e4 | 229 | |
2d42244a JS |
230 | /* |
231 | * Get monotonic time for posix timers | |
232 | */ | |
233 | static int posix_get_monotonic_raw(clockid_t which_clock, struct timespec *tp) | |
234 | { | |
235 | getrawmonotonic(tp); | |
236 | return 0; | |
237 | } | |
238 | ||
1da177e4 LT |
239 | /* |
240 | * Initialize everything, well, just everything in Posix clocks/timers ;) | |
241 | */ | |
242 | static __init int init_posix_timers(void) | |
243 | { | |
becf8b5d TG |
244 | struct k_clock clock_realtime = { |
245 | .clock_getres = hrtimer_get_res, | |
1da177e4 | 246 | }; |
becf8b5d TG |
247 | struct k_clock clock_monotonic = { |
248 | .clock_getres = hrtimer_get_res, | |
249 | .clock_get = posix_ktime_get_ts, | |
250 | .clock_set = do_posix_clock_nosettime, | |
1da177e4 | 251 | }; |
2d42244a JS |
252 | struct k_clock clock_monotonic_raw = { |
253 | .clock_getres = hrtimer_get_res, | |
254 | .clock_get = posix_get_monotonic_raw, | |
255 | .clock_set = do_posix_clock_nosettime, | |
3d44cc3e | 256 | .timer_create = no_timer_create, |
2d42244a | 257 | }; |
1da177e4 LT |
258 | |
259 | register_posix_clock(CLOCK_REALTIME, &clock_realtime); | |
260 | register_posix_clock(CLOCK_MONOTONIC, &clock_monotonic); | |
2d42244a | 261 | register_posix_clock(CLOCK_MONOTONIC_RAW, &clock_monotonic_raw); |
1da177e4 LT |
262 | |
263 | posix_timers_cache = kmem_cache_create("posix_timers_cache", | |
040b5c6f AD |
264 | sizeof (struct k_itimer), 0, SLAB_PANIC, |
265 | NULL); | |
1da177e4 LT |
266 | idr_init(&posix_timers_id); |
267 | return 0; | |
268 | } | |
269 | ||
270 | __initcall(init_posix_timers); | |
271 | ||
1da177e4 LT |
272 | static void schedule_next_timer(struct k_itimer *timr) |
273 | { | |
44f21475 RZ |
274 | struct hrtimer *timer = &timr->it.real.timer; |
275 | ||
becf8b5d | 276 | if (timr->it.real.interval.tv64 == 0) |
1da177e4 LT |
277 | return; |
278 | ||
4d672e7a DL |
279 | timr->it_overrun += (unsigned int) hrtimer_forward(timer, |
280 | timer->base->get_time(), | |
281 | timr->it.real.interval); | |
44f21475 | 282 | |
1da177e4 LT |
283 | timr->it_overrun_last = timr->it_overrun; |
284 | timr->it_overrun = -1; | |
285 | ++timr->it_requeue_pending; | |
44f21475 | 286 | hrtimer_restart(timer); |
1da177e4 LT |
287 | } |
288 | ||
289 | /* | |
290 | * This function is exported for use by the signal deliver code. It is | |
291 | * called just prior to the info block being released and passes that | |
292 | * block to us. It's function is to update the overrun entry AND to | |
293 | * restart the timer. It should only be called if the timer is to be | |
294 | * restarted (i.e. we have flagged this in the sys_private entry of the | |
295 | * info block). | |
296 | * | |
297 | * To protect aginst the timer going away while the interrupt is queued, | |
298 | * we require that the it_requeue_pending flag be set. | |
299 | */ | |
300 | void do_schedule_next_timer(struct siginfo *info) | |
301 | { | |
302 | struct k_itimer *timr; | |
303 | unsigned long flags; | |
304 | ||
305 | timr = lock_timer(info->si_tid, &flags); | |
306 | ||
becf8b5d TG |
307 | if (timr && timr->it_requeue_pending == info->si_sys_private) { |
308 | if (timr->it_clock < 0) | |
309 | posix_cpu_timer_schedule(timr); | |
310 | else | |
311 | schedule_next_timer(timr); | |
1da177e4 | 312 | |
54da1174 | 313 | info->si_overrun += timr->it_overrun_last; |
becf8b5d TG |
314 | } |
315 | ||
b6557fbc TG |
316 | if (timr) |
317 | unlock_timer(timr, flags); | |
1da177e4 LT |
318 | } |
319 | ||
ba661292 | 320 | int posix_timer_event(struct k_itimer *timr, int si_private) |
1da177e4 | 321 | { |
27af4245 ON |
322 | struct task_struct *task; |
323 | int shared, ret = -1; | |
ba661292 ON |
324 | /* |
325 | * FIXME: if ->sigq is queued we can race with | |
326 | * dequeue_signal()->do_schedule_next_timer(). | |
327 | * | |
328 | * If dequeue_signal() sees the "right" value of | |
329 | * si_sys_private it calls do_schedule_next_timer(). | |
330 | * We re-queue ->sigq and drop ->it_lock(). | |
331 | * do_schedule_next_timer() locks the timer | |
332 | * and re-schedules it while ->sigq is pending. | |
333 | * Not really bad, but not that we want. | |
334 | */ | |
1da177e4 | 335 | timr->sigq->info.si_sys_private = si_private; |
1da177e4 | 336 | |
27af4245 ON |
337 | rcu_read_lock(); |
338 | task = pid_task(timr->it_pid, PIDTYPE_PID); | |
339 | if (task) { | |
340 | shared = !(timr->it_sigev_notify & SIGEV_THREAD_ID); | |
341 | ret = send_sigqueue(timr->sigq, task, shared); | |
342 | } | |
343 | rcu_read_unlock(); | |
4aa73611 ON |
344 | /* If we failed to send the signal the timer stops. */ |
345 | return ret > 0; | |
1da177e4 LT |
346 | } |
347 | EXPORT_SYMBOL_GPL(posix_timer_event); | |
348 | ||
349 | /* | |
350 | * This function gets called when a POSIX.1b interval timer expires. It | |
351 | * is used as a callback from the kernel internal timer. The | |
352 | * run_timer_list code ALWAYS calls with interrupts on. | |
353 | ||
354 | * This code is for CLOCK_REALTIME* and CLOCK_MONOTONIC* timers. | |
355 | */ | |
c9cb2e3d | 356 | static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer) |
1da177e4 | 357 | { |
05cfb614 | 358 | struct k_itimer *timr; |
1da177e4 | 359 | unsigned long flags; |
becf8b5d | 360 | int si_private = 0; |
c9cb2e3d | 361 | enum hrtimer_restart ret = HRTIMER_NORESTART; |
1da177e4 | 362 | |
05cfb614 | 363 | timr = container_of(timer, struct k_itimer, it.real.timer); |
1da177e4 | 364 | spin_lock_irqsave(&timr->it_lock, flags); |
1da177e4 | 365 | |
becf8b5d TG |
366 | if (timr->it.real.interval.tv64 != 0) |
367 | si_private = ++timr->it_requeue_pending; | |
1da177e4 | 368 | |
becf8b5d TG |
369 | if (posix_timer_event(timr, si_private)) { |
370 | /* | |
371 | * signal was not sent because of sig_ignor | |
372 | * we will not get a call back to restart it AND | |
373 | * it should be restarted. | |
374 | */ | |
375 | if (timr->it.real.interval.tv64 != 0) { | |
58229a18 TG |
376 | ktime_t now = hrtimer_cb_get_time(timer); |
377 | ||
378 | /* | |
379 | * FIXME: What we really want, is to stop this | |
380 | * timer completely and restart it in case the | |
381 | * SIG_IGN is removed. This is a non trivial | |
382 | * change which involves sighand locking | |
383 | * (sigh !), which we don't want to do late in | |
384 | * the release cycle. | |
385 | * | |
386 | * For now we just let timers with an interval | |
387 | * less than a jiffie expire every jiffie to | |
388 | * avoid softirq starvation in case of SIG_IGN | |
389 | * and a very small interval, which would put | |
390 | * the timer right back on the softirq pending | |
391 | * list. By moving now ahead of time we trick | |
392 | * hrtimer_forward() to expire the timer | |
393 | * later, while we still maintain the overrun | |
394 | * accuracy, but have some inconsistency in | |
395 | * the timer_gettime() case. This is at least | |
396 | * better than a starved softirq. A more | |
397 | * complex fix which solves also another related | |
398 | * inconsistency is already in the pipeline. | |
399 | */ | |
400 | #ifdef CONFIG_HIGH_RES_TIMERS | |
401 | { | |
402 | ktime_t kj = ktime_set(0, NSEC_PER_SEC / HZ); | |
403 | ||
404 | if (timr->it.real.interval.tv64 < kj.tv64) | |
405 | now = ktime_add(now, kj); | |
406 | } | |
407 | #endif | |
4d672e7a | 408 | timr->it_overrun += (unsigned int) |
58229a18 | 409 | hrtimer_forward(timer, now, |
becf8b5d TG |
410 | timr->it.real.interval); |
411 | ret = HRTIMER_RESTART; | |
a0a0c28c | 412 | ++timr->it_requeue_pending; |
1da177e4 | 413 | } |
1da177e4 | 414 | } |
1da177e4 | 415 | |
becf8b5d TG |
416 | unlock_timer(timr, flags); |
417 | return ret; | |
418 | } | |
1da177e4 | 419 | |
27af4245 | 420 | static struct pid *good_sigevent(sigevent_t * event) |
1da177e4 LT |
421 | { |
422 | struct task_struct *rtn = current->group_leader; | |
423 | ||
424 | if ((event->sigev_notify & SIGEV_THREAD_ID ) && | |
8dc86af0 | 425 | (!(rtn = find_task_by_vpid(event->sigev_notify_thread_id)) || |
bac0abd6 | 426 | !same_thread_group(rtn, current) || |
1da177e4 LT |
427 | (event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_SIGNAL)) |
428 | return NULL; | |
429 | ||
430 | if (((event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) && | |
431 | ((event->sigev_signo <= 0) || (event->sigev_signo > SIGRTMAX))) | |
432 | return NULL; | |
433 | ||
27af4245 | 434 | return task_pid(rtn); |
1da177e4 LT |
435 | } |
436 | ||
a924b04d | 437 | void register_posix_clock(const clockid_t clock_id, struct k_clock *new_clock) |
1da177e4 LT |
438 | { |
439 | if ((unsigned) clock_id >= MAX_CLOCKS) { | |
440 | printk("POSIX clock register failed for clock_id %d\n", | |
441 | clock_id); | |
442 | return; | |
443 | } | |
444 | ||
445 | posix_clocks[clock_id] = *new_clock; | |
446 | } | |
447 | EXPORT_SYMBOL_GPL(register_posix_clock); | |
448 | ||
449 | static struct k_itimer * alloc_posix_timer(void) | |
450 | { | |
451 | struct k_itimer *tmr; | |
c3762229 | 452 | tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL); |
1da177e4 LT |
453 | if (!tmr) |
454 | return tmr; | |
1da177e4 LT |
455 | if (unlikely(!(tmr->sigq = sigqueue_alloc()))) { |
456 | kmem_cache_free(posix_timers_cache, tmr); | |
aa94fbd5 | 457 | return NULL; |
1da177e4 | 458 | } |
ba661292 | 459 | memset(&tmr->sigq->info, 0, sizeof(siginfo_t)); |
1da177e4 LT |
460 | return tmr; |
461 | } | |
462 | ||
463 | #define IT_ID_SET 1 | |
464 | #define IT_ID_NOT_SET 0 | |
465 | static void release_posix_timer(struct k_itimer *tmr, int it_id_set) | |
466 | { | |
467 | if (it_id_set) { | |
468 | unsigned long flags; | |
469 | spin_lock_irqsave(&idr_lock, flags); | |
470 | idr_remove(&posix_timers_id, tmr->it_id); | |
471 | spin_unlock_irqrestore(&idr_lock, flags); | |
472 | } | |
89992102 | 473 | put_pid(tmr->it_pid); |
1da177e4 | 474 | sigqueue_free(tmr->sigq); |
1da177e4 LT |
475 | kmem_cache_free(posix_timers_cache, tmr); |
476 | } | |
477 | ||
478 | /* Create a POSIX.1b interval timer. */ | |
479 | ||
480 | asmlinkage long | |
a924b04d | 481 | sys_timer_create(const clockid_t which_clock, |
1da177e4 LT |
482 | struct sigevent __user *timer_event_spec, |
483 | timer_t __user * created_timer_id) | |
484 | { | |
2cd499e3 | 485 | struct k_itimer *new_timer; |
ef864c95 | 486 | int error, new_timer_id; |
1da177e4 LT |
487 | sigevent_t event; |
488 | int it_id_set = IT_ID_NOT_SET; | |
489 | ||
490 | if (invalid_clockid(which_clock)) | |
491 | return -EINVAL; | |
492 | ||
493 | new_timer = alloc_posix_timer(); | |
494 | if (unlikely(!new_timer)) | |
495 | return -EAGAIN; | |
496 | ||
497 | spin_lock_init(&new_timer->it_lock); | |
498 | retry: | |
499 | if (unlikely(!idr_pre_get(&posix_timers_id, GFP_KERNEL))) { | |
500 | error = -EAGAIN; | |
501 | goto out; | |
502 | } | |
503 | spin_lock_irq(&idr_lock); | |
5a51b713 | 504 | error = idr_get_new(&posix_timers_id, new_timer, &new_timer_id); |
1da177e4 | 505 | spin_unlock_irq(&idr_lock); |
ef864c95 ON |
506 | if (error) { |
507 | if (error == -EAGAIN) | |
508 | goto retry; | |
1da177e4 | 509 | /* |
0b0a3e7b | 510 | * Weird looking, but we return EAGAIN if the IDR is |
1da177e4 LT |
511 | * full (proper POSIX return value for this) |
512 | */ | |
513 | error = -EAGAIN; | |
514 | goto out; | |
515 | } | |
516 | ||
517 | it_id_set = IT_ID_SET; | |
518 | new_timer->it_id = (timer_t) new_timer_id; | |
519 | new_timer->it_clock = which_clock; | |
520 | new_timer->it_overrun = -1; | |
521 | error = CLOCK_DISPATCH(which_clock, timer_create, (new_timer)); | |
522 | if (error) | |
523 | goto out; | |
524 | ||
525 | /* | |
526 | * return the timer_id now. The next step is hard to | |
527 | * back out if there is an error. | |
528 | */ | |
529 | if (copy_to_user(created_timer_id, | |
530 | &new_timer_id, sizeof (new_timer_id))) { | |
531 | error = -EFAULT; | |
532 | goto out; | |
533 | } | |
534 | if (timer_event_spec) { | |
535 | if (copy_from_user(&event, timer_event_spec, sizeof (event))) { | |
536 | error = -EFAULT; | |
537 | goto out; | |
538 | } | |
36b2f046 | 539 | rcu_read_lock(); |
89992102 | 540 | new_timer->it_pid = get_pid(good_sigevent(&event)); |
36b2f046 | 541 | rcu_read_unlock(); |
89992102 | 542 | if (!new_timer->it_pid) { |
1da177e4 LT |
543 | error = -EINVAL; |
544 | goto out; | |
545 | } | |
546 | } else { | |
5a9fa730 ON |
547 | event.sigev_notify = SIGEV_SIGNAL; |
548 | event.sigev_signo = SIGALRM; | |
549 | event.sigev_value.sival_int = new_timer->it_id; | |
89992102 | 550 | new_timer->it_pid = get_pid(task_tgid(current)); |
1da177e4 LT |
551 | } |
552 | ||
5a9fa730 ON |
553 | new_timer->it_sigev_notify = event.sigev_notify; |
554 | new_timer->sigq->info.si_signo = event.sigev_signo; | |
555 | new_timer->sigq->info.si_value = event.sigev_value; | |
717835d9 | 556 | new_timer->sigq->info.si_tid = new_timer->it_id; |
5a9fa730 | 557 | new_timer->sigq->info.si_code = SI_TIMER; |
717835d9 | 558 | |
36b2f046 | 559 | spin_lock_irq(¤t->sighand->siglock); |
27af4245 | 560 | new_timer->it_signal = current->signal; |
36b2f046 ON |
561 | list_add(&new_timer->list, ¤t->signal->posix_timers); |
562 | spin_unlock_irq(¤t->sighand->siglock); | |
ef864c95 ON |
563 | |
564 | return 0; | |
1da177e4 LT |
565 | /* |
566 | * In the case of the timer belonging to another task, after | |
567 | * the task is unlocked, the timer is owned by the other task | |
568 | * and may cease to exist at any time. Don't use or modify | |
569 | * new_timer after the unlock call. | |
570 | */ | |
1da177e4 | 571 | out: |
ef864c95 | 572 | release_posix_timer(new_timer, it_id_set); |
1da177e4 LT |
573 | return error; |
574 | } | |
575 | ||
1da177e4 LT |
576 | /* |
577 | * Locking issues: We need to protect the result of the id look up until | |
578 | * we get the timer locked down so it is not deleted under us. The | |
579 | * removal is done under the idr spinlock so we use that here to bridge | |
580 | * the find to the timer lock. To avoid a dead lock, the timer id MUST | |
581 | * be release with out holding the timer lock. | |
582 | */ | |
31d92845 | 583 | static struct k_itimer *lock_timer(timer_t timer_id, unsigned long *flags) |
1da177e4 LT |
584 | { |
585 | struct k_itimer *timr; | |
586 | /* | |
587 | * Watch out here. We do a irqsave on the idr_lock and pass the | |
588 | * flags part over to the timer lock. Must not let interrupts in | |
589 | * while we are moving the lock. | |
590 | */ | |
1da177e4 | 591 | spin_lock_irqsave(&idr_lock, *flags); |
31d92845 | 592 | timr = idr_find(&posix_timers_id, (int)timer_id); |
1da177e4 LT |
593 | if (timr) { |
594 | spin_lock(&timr->it_lock); | |
89992102 | 595 | if (timr->it_signal == current->signal) { |
179394af | 596 | spin_unlock(&idr_lock); |
31d92845 ON |
597 | return timr; |
598 | } | |
599 | spin_unlock(&timr->it_lock); | |
600 | } | |
601 | spin_unlock_irqrestore(&idr_lock, *flags); | |
1da177e4 | 602 | |
31d92845 | 603 | return NULL; |
1da177e4 LT |
604 | } |
605 | ||
606 | /* | |
607 | * Get the time remaining on a POSIX.1b interval timer. This function | |
608 | * is ALWAYS called with spin_lock_irq on the timer, thus it must not | |
609 | * mess with irq. | |
610 | * | |
611 | * We have a couple of messes to clean up here. First there is the case | |
612 | * of a timer that has a requeue pending. These timers should appear to | |
613 | * be in the timer list with an expiry as if we were to requeue them | |
614 | * now. | |
615 | * | |
616 | * The second issue is the SIGEV_NONE timer which may be active but is | |
617 | * not really ever put in the timer list (to save system resources). | |
618 | * This timer may be expired, and if so, we will do it here. Otherwise | |
619 | * it is the same as a requeue pending timer WRT to what we should | |
620 | * report. | |
621 | */ | |
622 | static void | |
623 | common_timer_get(struct k_itimer *timr, struct itimerspec *cur_setting) | |
624 | { | |
3b98a532 | 625 | ktime_t now, remaining, iv; |
becf8b5d | 626 | struct hrtimer *timer = &timr->it.real.timer; |
1da177e4 | 627 | |
becf8b5d | 628 | memset(cur_setting, 0, sizeof(struct itimerspec)); |
becf8b5d | 629 | |
3b98a532 RZ |
630 | iv = timr->it.real.interval; |
631 | ||
becf8b5d | 632 | /* interval timer ? */ |
3b98a532 RZ |
633 | if (iv.tv64) |
634 | cur_setting->it_interval = ktime_to_timespec(iv); | |
635 | else if (!hrtimer_active(timer) && | |
636 | (timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) | |
becf8b5d | 637 | return; |
3b98a532 RZ |
638 | |
639 | now = timer->base->get_time(); | |
640 | ||
becf8b5d | 641 | /* |
3b98a532 RZ |
642 | * When a requeue is pending or this is a SIGEV_NONE |
643 | * timer move the expiry time forward by intervals, so | |
644 | * expiry is > now. | |
becf8b5d | 645 | */ |
3b98a532 RZ |
646 | if (iv.tv64 && (timr->it_requeue_pending & REQUEUE_PENDING || |
647 | (timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE)) | |
4d672e7a | 648 | timr->it_overrun += (unsigned int) hrtimer_forward(timer, now, iv); |
3b98a532 | 649 | |
cc584b21 | 650 | remaining = ktime_sub(hrtimer_get_expires(timer), now); |
becf8b5d | 651 | /* Return 0 only, when the timer is expired and not pending */ |
3b98a532 RZ |
652 | if (remaining.tv64 <= 0) { |
653 | /* | |
654 | * A single shot SIGEV_NONE timer must return 0, when | |
655 | * it is expired ! | |
656 | */ | |
657 | if ((timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) | |
658 | cur_setting->it_value.tv_nsec = 1; | |
659 | } else | |
becf8b5d | 660 | cur_setting->it_value = ktime_to_timespec(remaining); |
1da177e4 LT |
661 | } |
662 | ||
663 | /* Get the time remaining on a POSIX.1b interval timer. */ | |
664 | asmlinkage long | |
665 | sys_timer_gettime(timer_t timer_id, struct itimerspec __user *setting) | |
666 | { | |
667 | struct k_itimer *timr; | |
668 | struct itimerspec cur_setting; | |
669 | unsigned long flags; | |
670 | ||
671 | timr = lock_timer(timer_id, &flags); | |
672 | if (!timr) | |
673 | return -EINVAL; | |
674 | ||
675 | CLOCK_DISPATCH(timr->it_clock, timer_get, (timr, &cur_setting)); | |
676 | ||
677 | unlock_timer(timr, flags); | |
678 | ||
679 | if (copy_to_user(setting, &cur_setting, sizeof (cur_setting))) | |
680 | return -EFAULT; | |
681 | ||
682 | return 0; | |
683 | } | |
becf8b5d | 684 | |
1da177e4 LT |
685 | /* |
686 | * Get the number of overruns of a POSIX.1b interval timer. This is to | |
687 | * be the overrun of the timer last delivered. At the same time we are | |
688 | * accumulating overruns on the next timer. The overrun is frozen when | |
689 | * the signal is delivered, either at the notify time (if the info block | |
690 | * is not queued) or at the actual delivery time (as we are informed by | |
691 | * the call back to do_schedule_next_timer(). So all we need to do is | |
692 | * to pick up the frozen overrun. | |
693 | */ | |
1da177e4 LT |
694 | asmlinkage long |
695 | sys_timer_getoverrun(timer_t timer_id) | |
696 | { | |
697 | struct k_itimer *timr; | |
698 | int overrun; | |
5ba25331 | 699 | unsigned long flags; |
1da177e4 LT |
700 | |
701 | timr = lock_timer(timer_id, &flags); | |
702 | if (!timr) | |
703 | return -EINVAL; | |
704 | ||
705 | overrun = timr->it_overrun_last; | |
706 | unlock_timer(timr, flags); | |
707 | ||
708 | return overrun; | |
709 | } | |
1da177e4 LT |
710 | |
711 | /* Set a POSIX.1b interval timer. */ | |
712 | /* timr->it_lock is taken. */ | |
858119e1 | 713 | static int |
1da177e4 LT |
714 | common_timer_set(struct k_itimer *timr, int flags, |
715 | struct itimerspec *new_setting, struct itimerspec *old_setting) | |
716 | { | |
becf8b5d | 717 | struct hrtimer *timer = &timr->it.real.timer; |
7978672c | 718 | enum hrtimer_mode mode; |
1da177e4 LT |
719 | |
720 | if (old_setting) | |
721 | common_timer_get(timr, old_setting); | |
722 | ||
723 | /* disable the timer */ | |
becf8b5d | 724 | timr->it.real.interval.tv64 = 0; |
1da177e4 LT |
725 | /* |
726 | * careful here. If smp we could be in the "fire" routine which will | |
727 | * be spinning as we hold the lock. But this is ONLY an SMP issue. | |
728 | */ | |
becf8b5d | 729 | if (hrtimer_try_to_cancel(timer) < 0) |
1da177e4 | 730 | return TIMER_RETRY; |
1da177e4 LT |
731 | |
732 | timr->it_requeue_pending = (timr->it_requeue_pending + 2) & | |
733 | ~REQUEUE_PENDING; | |
734 | timr->it_overrun_last = 0; | |
1da177e4 | 735 | |
becf8b5d TG |
736 | /* switch off the timer when it_value is zero */ |
737 | if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec) | |
738 | return 0; | |
1da177e4 | 739 | |
c9cb2e3d | 740 | mode = flags & TIMER_ABSTIME ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL; |
7978672c | 741 | hrtimer_init(&timr->it.real.timer, timr->it_clock, mode); |
7978672c | 742 | timr->it.real.timer.function = posix_timer_fn; |
becf8b5d | 743 | |
cc584b21 | 744 | hrtimer_set_expires(timer, timespec_to_ktime(new_setting->it_value)); |
becf8b5d TG |
745 | |
746 | /* Convert interval */ | |
747 | timr->it.real.interval = timespec_to_ktime(new_setting->it_interval); | |
748 | ||
749 | /* SIGEV_NONE timers are not queued ! See common_timer_get */ | |
952bbc87 TG |
750 | if (((timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE)) { |
751 | /* Setup correct expiry time for relative timers */ | |
5a7780e7 | 752 | if (mode == HRTIMER_MODE_REL) { |
cc584b21 | 753 | hrtimer_add_expires(timer, timer->base->get_time()); |
5a7780e7 | 754 | } |
becf8b5d | 755 | return 0; |
952bbc87 | 756 | } |
becf8b5d | 757 | |
cc584b21 | 758 | hrtimer_start_expires(timer, mode); |
1da177e4 LT |
759 | return 0; |
760 | } | |
761 | ||
762 | /* Set a POSIX.1b interval timer */ | |
763 | asmlinkage long | |
764 | sys_timer_settime(timer_t timer_id, int flags, | |
765 | const struct itimerspec __user *new_setting, | |
766 | struct itimerspec __user *old_setting) | |
767 | { | |
768 | struct k_itimer *timr; | |
769 | struct itimerspec new_spec, old_spec; | |
770 | int error = 0; | |
5ba25331 | 771 | unsigned long flag; |
1da177e4 LT |
772 | struct itimerspec *rtn = old_setting ? &old_spec : NULL; |
773 | ||
774 | if (!new_setting) | |
775 | return -EINVAL; | |
776 | ||
777 | if (copy_from_user(&new_spec, new_setting, sizeof (new_spec))) | |
778 | return -EFAULT; | |
779 | ||
becf8b5d TG |
780 | if (!timespec_valid(&new_spec.it_interval) || |
781 | !timespec_valid(&new_spec.it_value)) | |
1da177e4 LT |
782 | return -EINVAL; |
783 | retry: | |
784 | timr = lock_timer(timer_id, &flag); | |
785 | if (!timr) | |
786 | return -EINVAL; | |
787 | ||
788 | error = CLOCK_DISPATCH(timr->it_clock, timer_set, | |
789 | (timr, flags, &new_spec, rtn)); | |
790 | ||
791 | unlock_timer(timr, flag); | |
792 | if (error == TIMER_RETRY) { | |
793 | rtn = NULL; // We already got the old time... | |
794 | goto retry; | |
795 | } | |
796 | ||
becf8b5d TG |
797 | if (old_setting && !error && |
798 | copy_to_user(old_setting, &old_spec, sizeof (old_spec))) | |
1da177e4 LT |
799 | error = -EFAULT; |
800 | ||
801 | return error; | |
802 | } | |
803 | ||
804 | static inline int common_timer_del(struct k_itimer *timer) | |
805 | { | |
becf8b5d | 806 | timer->it.real.interval.tv64 = 0; |
f972be33 | 807 | |
becf8b5d | 808 | if (hrtimer_try_to_cancel(&timer->it.real.timer) < 0) |
1da177e4 | 809 | return TIMER_RETRY; |
1da177e4 LT |
810 | return 0; |
811 | } | |
812 | ||
813 | static inline int timer_delete_hook(struct k_itimer *timer) | |
814 | { | |
815 | return CLOCK_DISPATCH(timer->it_clock, timer_del, (timer)); | |
816 | } | |
817 | ||
818 | /* Delete a POSIX.1b interval timer. */ | |
819 | asmlinkage long | |
820 | sys_timer_delete(timer_t timer_id) | |
821 | { | |
822 | struct k_itimer *timer; | |
5ba25331 | 823 | unsigned long flags; |
1da177e4 | 824 | |
1da177e4 | 825 | retry_delete: |
1da177e4 LT |
826 | timer = lock_timer(timer_id, &flags); |
827 | if (!timer) | |
828 | return -EINVAL; | |
829 | ||
becf8b5d | 830 | if (timer_delete_hook(timer) == TIMER_RETRY) { |
1da177e4 LT |
831 | unlock_timer(timer, flags); |
832 | goto retry_delete; | |
833 | } | |
becf8b5d | 834 | |
1da177e4 LT |
835 | spin_lock(¤t->sighand->siglock); |
836 | list_del(&timer->list); | |
837 | spin_unlock(¤t->sighand->siglock); | |
838 | /* | |
839 | * This keeps any tasks waiting on the spin lock from thinking | |
840 | * they got something (see the lock code above). | |
841 | */ | |
89992102 | 842 | timer->it_signal = NULL; |
4b7a1304 | 843 | |
1da177e4 LT |
844 | unlock_timer(timer, flags); |
845 | release_posix_timer(timer, IT_ID_SET); | |
846 | return 0; | |
847 | } | |
becf8b5d | 848 | |
1da177e4 LT |
849 | /* |
850 | * return timer owned by the process, used by exit_itimers | |
851 | */ | |
858119e1 | 852 | static void itimer_delete(struct k_itimer *timer) |
1da177e4 LT |
853 | { |
854 | unsigned long flags; | |
855 | ||
1da177e4 | 856 | retry_delete: |
1da177e4 LT |
857 | spin_lock_irqsave(&timer->it_lock, flags); |
858 | ||
becf8b5d | 859 | if (timer_delete_hook(timer) == TIMER_RETRY) { |
1da177e4 LT |
860 | unlock_timer(timer, flags); |
861 | goto retry_delete; | |
862 | } | |
1da177e4 LT |
863 | list_del(&timer->list); |
864 | /* | |
865 | * This keeps any tasks waiting on the spin lock from thinking | |
866 | * they got something (see the lock code above). | |
867 | */ | |
89992102 | 868 | timer->it_signal = NULL; |
4b7a1304 | 869 | |
1da177e4 LT |
870 | unlock_timer(timer, flags); |
871 | release_posix_timer(timer, IT_ID_SET); | |
872 | } | |
873 | ||
874 | /* | |
25f407f0 | 875 | * This is called by do_exit or de_thread, only when there are no more |
1da177e4 LT |
876 | * references to the shared signal_struct. |
877 | */ | |
878 | void exit_itimers(struct signal_struct *sig) | |
879 | { | |
880 | struct k_itimer *tmr; | |
881 | ||
882 | while (!list_empty(&sig->posix_timers)) { | |
883 | tmr = list_entry(sig->posix_timers.next, struct k_itimer, list); | |
884 | itimer_delete(tmr); | |
885 | } | |
886 | } | |
887 | ||
becf8b5d | 888 | /* Not available / possible... functions */ |
a924b04d | 889 | int do_posix_clock_nosettime(const clockid_t clockid, struct timespec *tp) |
1da177e4 LT |
890 | { |
891 | return -EINVAL; | |
892 | } | |
893 | EXPORT_SYMBOL_GPL(do_posix_clock_nosettime); | |
894 | ||
a924b04d | 895 | int do_posix_clock_nonanosleep(const clockid_t clock, int flags, |
97735f25 | 896 | struct timespec *t, struct timespec __user *r) |
1da177e4 LT |
897 | { |
898 | #ifndef ENOTSUP | |
899 | return -EOPNOTSUPP; /* aka ENOTSUP in userland for POSIX */ | |
900 | #else /* parisc does define it separately. */ | |
901 | return -ENOTSUP; | |
902 | #endif | |
903 | } | |
904 | EXPORT_SYMBOL_GPL(do_posix_clock_nonanosleep); | |
905 | ||
a924b04d TG |
906 | asmlinkage long sys_clock_settime(const clockid_t which_clock, |
907 | const struct timespec __user *tp) | |
1da177e4 LT |
908 | { |
909 | struct timespec new_tp; | |
910 | ||
911 | if (invalid_clockid(which_clock)) | |
912 | return -EINVAL; | |
913 | if (copy_from_user(&new_tp, tp, sizeof (*tp))) | |
914 | return -EFAULT; | |
915 | ||
916 | return CLOCK_DISPATCH(which_clock, clock_set, (which_clock, &new_tp)); | |
917 | } | |
918 | ||
919 | asmlinkage long | |
a924b04d | 920 | sys_clock_gettime(const clockid_t which_clock, struct timespec __user *tp) |
1da177e4 LT |
921 | { |
922 | struct timespec kernel_tp; | |
923 | int error; | |
924 | ||
925 | if (invalid_clockid(which_clock)) | |
926 | return -EINVAL; | |
927 | error = CLOCK_DISPATCH(which_clock, clock_get, | |
928 | (which_clock, &kernel_tp)); | |
929 | if (!error && copy_to_user(tp, &kernel_tp, sizeof (kernel_tp))) | |
930 | error = -EFAULT; | |
931 | ||
932 | return error; | |
933 | ||
934 | } | |
935 | ||
936 | asmlinkage long | |
a924b04d | 937 | sys_clock_getres(const clockid_t which_clock, struct timespec __user *tp) |
1da177e4 LT |
938 | { |
939 | struct timespec rtn_tp; | |
940 | int error; | |
941 | ||
942 | if (invalid_clockid(which_clock)) | |
943 | return -EINVAL; | |
944 | ||
945 | error = CLOCK_DISPATCH(which_clock, clock_getres, | |
946 | (which_clock, &rtn_tp)); | |
947 | ||
948 | if (!error && tp && copy_to_user(tp, &rtn_tp, sizeof (rtn_tp))) { | |
949 | error = -EFAULT; | |
950 | } | |
951 | ||
952 | return error; | |
953 | } | |
954 | ||
97735f25 TG |
955 | /* |
956 | * nanosleep for monotonic and realtime clocks | |
957 | */ | |
958 | static int common_nsleep(const clockid_t which_clock, int flags, | |
959 | struct timespec *tsave, struct timespec __user *rmtp) | |
960 | { | |
080344b9 ON |
961 | return hrtimer_nanosleep(tsave, rmtp, flags & TIMER_ABSTIME ? |
962 | HRTIMER_MODE_ABS : HRTIMER_MODE_REL, | |
963 | which_clock); | |
97735f25 | 964 | } |
1da177e4 LT |
965 | |
966 | asmlinkage long | |
a924b04d | 967 | sys_clock_nanosleep(const clockid_t which_clock, int flags, |
1da177e4 LT |
968 | const struct timespec __user *rqtp, |
969 | struct timespec __user *rmtp) | |
970 | { | |
971 | struct timespec t; | |
1da177e4 LT |
972 | |
973 | if (invalid_clockid(which_clock)) | |
974 | return -EINVAL; | |
975 | ||
976 | if (copy_from_user(&t, rqtp, sizeof (struct timespec))) | |
977 | return -EFAULT; | |
978 | ||
5f82b2b7 | 979 | if (!timespec_valid(&t)) |
1da177e4 LT |
980 | return -EINVAL; |
981 | ||
97735f25 TG |
982 | return CLOCK_DISPATCH(which_clock, nsleep, |
983 | (which_clock, flags, &t, rmtp)); | |
1da177e4 | 984 | } |
1711ef38 TA |
985 | |
986 | /* | |
987 | * nanosleep_restart for monotonic and realtime clocks | |
988 | */ | |
989 | static int common_nsleep_restart(struct restart_block *restart_block) | |
990 | { | |
991 | return hrtimer_nanosleep_restart(restart_block); | |
992 | } | |
993 | ||
994 | /* | |
995 | * This will restart clock_nanosleep. This is required only by | |
996 | * compat_clock_nanosleep_restart for now. | |
997 | */ | |
998 | long | |
999 | clock_nanosleep_restart(struct restart_block *restart_block) | |
1000 | { | |
1001 | clockid_t which_clock = restart_block->arg0; | |
1002 | ||
1003 | return CLOCK_DISPATCH(which_clock, nsleep_restart, | |
1004 | (restart_block)); | |
1005 | } |