1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Implement CPU time clocks for the POSIX clock interface.
4 */
5
6 #include <linux/sched/signal.h>
7 #include <linux/sched/cputime.h>
8 #include <linux/posix-timers.h>
9 #include <linux/errno.h>
10 #include <linux/math64.h>
11 #include <linux/uaccess.h>
12 #include <linux/kernel_stat.h>
13 #include <trace/events/timer.h>
14 #include <linux/tick.h>
15 #include <linux/workqueue.h>
16 #include <linux/compat.h>
17 #include <linux/sched/deadline.h>
18
19 #include "posix-timers.h"
20
21 static void posix_cpu_timer_rearm(struct k_itimer *timer);
22
posix_cputimers_group_init(struct posix_cputimers * pct,u64 cpu_limit)23 void posix_cputimers_group_init(struct posix_cputimers *pct, u64 cpu_limit)
24 {
25 posix_cputimers_init(pct);
26 if (cpu_limit != RLIM_INFINITY) {
27 pct->bases[CPUCLOCK_PROF].nextevt = cpu_limit * NSEC_PER_SEC;
28 pct->timers_active = true;
29 }
30 }
31
32 /*
33 * Called after updating RLIMIT_CPU to run cpu timer and update
34 * tsk->signal->posix_cputimers.bases[clock].nextevt expiration cache if
35 * necessary. Needs siglock protection since other code may update the
36 * expiration cache as well.
37 */
update_rlimit_cpu(struct task_struct * task,unsigned long rlim_new)38 void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
39 {
40 u64 nsecs = rlim_new * NSEC_PER_SEC;
41
42 spin_lock_irq(&task->sighand->siglock);
43 set_process_cpu_timer(task, CPUCLOCK_PROF, &nsecs, NULL);
44 spin_unlock_irq(&task->sighand->siglock);
45 }
46
47 /*
48 * Functions for validating access to tasks.
49 */
pid_for_clock(const clockid_t clock,bool gettime)50 static struct pid *pid_for_clock(const clockid_t clock, bool gettime)
51 {
52 const bool thread = !!CPUCLOCK_PERTHREAD(clock);
53 const pid_t upid = CPUCLOCK_PID(clock);
54 struct pid *pid;
55
56 if (CPUCLOCK_WHICH(clock) >= CPUCLOCK_MAX)
57 return NULL;
58
59 /*
60 * If the encoded PID is 0, then the timer is targeted at current
61 * or the process to which current belongs.
62 */
63 if (upid == 0)
64 return thread ? task_pid(current) : task_tgid(current);
65
66 pid = find_vpid(upid);
67 if (!pid)
68 return NULL;
69
70 if (thread) {
71 struct task_struct *tsk = pid_task(pid, PIDTYPE_PID);
72 return (tsk && same_thread_group(tsk, current)) ? pid : NULL;
73 }
74
75 /*
76 * For clock_gettime(PROCESS) allow finding the process by
77 * with the pid of the current task. The code needs the tgid
78 * of the process so that pid_task(pid, PIDTYPE_TGID) can be
79 * used to find the process.
80 */
81 if (gettime && (pid == task_pid(current)))
82 return task_tgid(current);
83
84 /*
85 * For processes require that pid identifies a process.
86 */
87 return pid_has_task(pid, PIDTYPE_TGID) ? pid : NULL;
88 }
89
validate_clock_permissions(const clockid_t clock)90 static inline int validate_clock_permissions(const clockid_t clock)
91 {
92 int ret;
93
94 rcu_read_lock();
95 ret = pid_for_clock(clock, false) ? 0 : -EINVAL;
96 rcu_read_unlock();
97
98 return ret;
99 }
100
clock_pid_type(const clockid_t clock)101 static inline enum pid_type clock_pid_type(const clockid_t clock)
102 {
103 return CPUCLOCK_PERTHREAD(clock) ? PIDTYPE_PID : PIDTYPE_TGID;
104 }
105
cpu_timer_task_rcu(struct k_itimer * timer)106 static inline struct task_struct *cpu_timer_task_rcu(struct k_itimer *timer)
107 {
108 return pid_task(timer->it.cpu.pid, clock_pid_type(timer->it_clock));
109 }
110
111 /*
112 * Update expiry time from increment, and increase overrun count,
113 * given the current clock sample.
114 */
bump_cpu_timer(struct k_itimer * timer,u64 now)115 static u64 bump_cpu_timer(struct k_itimer *timer, u64 now)
116 {
117 u64 delta, incr, expires = timer->it.cpu.node.expires;
118 int i;
119
120 if (!timer->it_interval)
121 return expires;
122
123 if (now < expires)
124 return expires;
125
126 incr = timer->it_interval;
127 delta = now + incr - expires;
128
129 /* Don't use (incr*2 < delta), incr*2 might overflow. */
130 for (i = 0; incr < delta - incr; i++)
131 incr = incr << 1;
132
133 for (; i >= 0; incr >>= 1, i--) {
134 if (delta < incr)
135 continue;
136
137 timer->it.cpu.node.expires += incr;
138 timer->it_overrun += 1LL << i;
139 delta -= incr;
140 }
141 return timer->it.cpu.node.expires;
142 }
143
144 /* Check whether all cache entries contain U64_MAX, i.e. eternal expiry time */
expiry_cache_is_inactive(const struct posix_cputimers * pct)145 static inline bool expiry_cache_is_inactive(const struct posix_cputimers *pct)
146 {
147 return !(~pct->bases[CPUCLOCK_PROF].nextevt |
148 ~pct->bases[CPUCLOCK_VIRT].nextevt |
149 ~pct->bases[CPUCLOCK_SCHED].nextevt);
150 }
151
152 static int
posix_cpu_clock_getres(const clockid_t which_clock,struct timespec64 * tp)153 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec64 *tp)
154 {
155 int error = validate_clock_permissions(which_clock);
156
157 if (!error) {
158 tp->tv_sec = 0;
159 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
160 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
161 /*
162 * If sched_clock is using a cycle counter, we
163 * don't have any idea of its true resolution
164 * exported, but it is much more than 1s/HZ.
165 */
166 tp->tv_nsec = 1;
167 }
168 }
169 return error;
170 }
171
172 static int
posix_cpu_clock_set(const clockid_t clock,const struct timespec64 * tp)173 posix_cpu_clock_set(const clockid_t clock, const struct timespec64 *tp)
174 {
175 int error = validate_clock_permissions(clock);
176
177 /*
178 * You can never reset a CPU clock, but we check for other errors
179 * in the call before failing with EPERM.
180 */
181 return error ? : -EPERM;
182 }
183
184 /*
185 * Sample a per-thread clock for the given task. clkid is validated.
186 */
cpu_clock_sample(const clockid_t clkid,struct task_struct * p)187 static u64 cpu_clock_sample(const clockid_t clkid, struct task_struct *p)
188 {
189 u64 utime, stime;
190
191 if (clkid == CPUCLOCK_SCHED)
192 return task_sched_runtime(p);
193
194 task_cputime(p, &utime, &stime);
195
196 switch (clkid) {
197 case CPUCLOCK_PROF:
198 return utime + stime;
199 case CPUCLOCK_VIRT:
200 return utime;
201 default:
202 WARN_ON_ONCE(1);
203 }
204 return 0;
205 }
206
store_samples(u64 * samples,u64 stime,u64 utime,u64 rtime)207 static inline void store_samples(u64 *samples, u64 stime, u64 utime, u64 rtime)
208 {
209 samples[CPUCLOCK_PROF] = stime + utime;
210 samples[CPUCLOCK_VIRT] = utime;
211 samples[CPUCLOCK_SCHED] = rtime;
212 }
213
task_sample_cputime(struct task_struct * p,u64 * samples)214 static void task_sample_cputime(struct task_struct *p, u64 *samples)
215 {
216 u64 stime, utime;
217
218 task_cputime(p, &utime, &stime);
219 store_samples(samples, stime, utime, p->se.sum_exec_runtime);
220 }
221
proc_sample_cputime_atomic(struct task_cputime_atomic * at,u64 * samples)222 static void proc_sample_cputime_atomic(struct task_cputime_atomic *at,
223 u64 *samples)
224 {
225 u64 stime, utime, rtime;
226
227 utime = atomic64_read(&at->utime);
228 stime = atomic64_read(&at->stime);
229 rtime = atomic64_read(&at->sum_exec_runtime);
230 store_samples(samples, stime, utime, rtime);
231 }
232
233 /*
234 * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg
235 * to avoid race conditions with concurrent updates to cputime.
236 */
__update_gt_cputime(atomic64_t * cputime,u64 sum_cputime)237 static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime)
238 {
239 u64 curr_cputime;
240 retry:
241 curr_cputime = atomic64_read(cputime);
242 if (sum_cputime > curr_cputime) {
243 if (atomic64_cmpxchg(cputime, curr_cputime, sum_cputime) != curr_cputime)
244 goto retry;
245 }
246 }
247
update_gt_cputime(struct task_cputime_atomic * cputime_atomic,struct task_cputime * sum)248 static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic,
249 struct task_cputime *sum)
250 {
251 __update_gt_cputime(&cputime_atomic->utime, sum->utime);
252 __update_gt_cputime(&cputime_atomic->stime, sum->stime);
253 __update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime);
254 }
255
256 /**
257 * thread_group_sample_cputime - Sample cputime for a given task
258 * @tsk: Task for which cputime needs to be started
259 * @samples: Storage for time samples
260 *
261 * Called from sys_getitimer() to calculate the expiry time of an active
262 * timer. That means group cputime accounting is already active. Called
263 * with task sighand lock held.
264 *
265 * Updates @times with an uptodate sample of the thread group cputimes.
266 */
thread_group_sample_cputime(struct task_struct * tsk,u64 * samples)267 void thread_group_sample_cputime(struct task_struct *tsk, u64 *samples)
268 {
269 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
270 struct posix_cputimers *pct = &tsk->signal->posix_cputimers;
271
272 WARN_ON_ONCE(!pct->timers_active);
273
274 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
275 }
276
277 /**
278 * thread_group_start_cputime - Start cputime and return a sample
279 * @tsk: Task for which cputime needs to be started
280 * @samples: Storage for time samples
281 *
282 * The thread group cputime accounting is avoided when there are no posix
283 * CPU timers armed. Before starting a timer it's required to check whether
284 * the time accounting is active. If not, a full update of the atomic
285 * accounting store needs to be done and the accounting enabled.
286 *
287 * Updates @times with an uptodate sample of the thread group cputimes.
288 */
thread_group_start_cputime(struct task_struct * tsk,u64 * samples)289 static void thread_group_start_cputime(struct task_struct *tsk, u64 *samples)
290 {
291 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
292 struct posix_cputimers *pct = &tsk->signal->posix_cputimers;
293
294 lockdep_assert_task_sighand_held(tsk);
295
296 /* Check if cputimer isn't running. This is accessed without locking. */
297 if (!READ_ONCE(pct->timers_active)) {
298 struct task_cputime sum;
299
300 /*
301 * The POSIX timer interface allows for absolute time expiry
302 * values through the TIMER_ABSTIME flag, therefore we have
303 * to synchronize the timer to the clock every time we start it.
304 */
305 thread_group_cputime(tsk, &sum);
306 update_gt_cputime(&cputimer->cputime_atomic, &sum);
307
308 /*
309 * We're setting timers_active without a lock. Ensure this
310 * only gets written to in one operation. We set it after
311 * update_gt_cputime() as a small optimization, but
312 * barriers are not required because update_gt_cputime()
313 * can handle concurrent updates.
314 */
315 WRITE_ONCE(pct->timers_active, true);
316 }
317 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
318 }
319
__thread_group_cputime(struct task_struct * tsk,u64 * samples)320 static void __thread_group_cputime(struct task_struct *tsk, u64 *samples)
321 {
322 struct task_cputime ct;
323
324 thread_group_cputime(tsk, &ct);
325 store_samples(samples, ct.stime, ct.utime, ct.sum_exec_runtime);
326 }
327
328 /*
329 * Sample a process (thread group) clock for the given task clkid. If the
330 * group's cputime accounting is already enabled, read the atomic
331 * store. Otherwise a full update is required. clkid is already validated.
332 */
cpu_clock_sample_group(const clockid_t clkid,struct task_struct * p,bool start)333 static u64 cpu_clock_sample_group(const clockid_t clkid, struct task_struct *p,
334 bool start)
335 {
336 struct thread_group_cputimer *cputimer = &p->signal->cputimer;
337 struct posix_cputimers *pct = &p->signal->posix_cputimers;
338 u64 samples[CPUCLOCK_MAX];
339
340 if (!READ_ONCE(pct->timers_active)) {
341 if (start)
342 thread_group_start_cputime(p, samples);
343 else
344 __thread_group_cputime(p, samples);
345 } else {
346 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
347 }
348
349 return samples[clkid];
350 }
351
posix_cpu_clock_get(const clockid_t clock,struct timespec64 * tp)352 static int posix_cpu_clock_get(const clockid_t clock, struct timespec64 *tp)
353 {
354 const clockid_t clkid = CPUCLOCK_WHICH(clock);
355 struct task_struct *tsk;
356 u64 t;
357
358 rcu_read_lock();
359 tsk = pid_task(pid_for_clock(clock, true), clock_pid_type(clock));
360 if (!tsk) {
361 rcu_read_unlock();
362 return -EINVAL;
363 }
364
365 if (CPUCLOCK_PERTHREAD(clock))
366 t = cpu_clock_sample(clkid, tsk);
367 else
368 t = cpu_clock_sample_group(clkid, tsk, false);
369 rcu_read_unlock();
370
371 *tp = ns_to_timespec64(t);
372 return 0;
373 }
374
375 /*
376 * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
377 * This is called from sys_timer_create() and do_cpu_nanosleep() with the
378 * new timer already all-zeros initialized.
379 */
posix_cpu_timer_create(struct k_itimer * new_timer)380 static int posix_cpu_timer_create(struct k_itimer *new_timer)
381 {
382 static struct lock_class_key posix_cpu_timers_key;
383 struct pid *pid;
384
385 rcu_read_lock();
386 pid = pid_for_clock(new_timer->it_clock, false);
387 if (!pid) {
388 rcu_read_unlock();
389 return -EINVAL;
390 }
391
392 /*
393 * If posix timer expiry is handled in task work context then
394 * timer::it_lock can be taken without disabling interrupts as all
395 * other locking happens in task context. This requires a separate
396 * lock class key otherwise regular posix timer expiry would record
397 * the lock class being taken in interrupt context and generate a
398 * false positive warning.
399 */
400 if (IS_ENABLED(CONFIG_POSIX_CPU_TIMERS_TASK_WORK))
401 lockdep_set_class(&new_timer->it_lock, &posix_cpu_timers_key);
402
403 new_timer->kclock = &clock_posix_cpu;
404 timerqueue_init(&new_timer->it.cpu.node);
405 new_timer->it.cpu.pid = get_pid(pid);
406 rcu_read_unlock();
407 return 0;
408 }
409
timer_base(struct k_itimer * timer,struct task_struct * tsk)410 static struct posix_cputimer_base *timer_base(struct k_itimer *timer,
411 struct task_struct *tsk)
412 {
413 int clkidx = CPUCLOCK_WHICH(timer->it_clock);
414
415 if (CPUCLOCK_PERTHREAD(timer->it_clock))
416 return tsk->posix_cputimers.bases + clkidx;
417 else
418 return tsk->signal->posix_cputimers.bases + clkidx;
419 }
420
421 /*
422 * Force recalculating the base earliest expiration on the next tick.
423 * This will also re-evaluate the need to keep around the process wide
424 * cputime counter and tick dependency and eventually shut these down
425 * if necessary.
426 */
trigger_base_recalc_expires(struct k_itimer * timer,struct task_struct * tsk)427 static void trigger_base_recalc_expires(struct k_itimer *timer,
428 struct task_struct *tsk)
429 {
430 struct posix_cputimer_base *base = timer_base(timer, tsk);
431
432 base->nextevt = 0;
433 }
434
435 /*
436 * Dequeue the timer and reset the base if it was its earliest expiration.
437 * It makes sure the next tick recalculates the base next expiration so we
438 * don't keep the costly process wide cputime counter around for a random
439 * amount of time, along with the tick dependency.
440 *
441 * If another timer gets queued between this and the next tick, its
442 * expiration will update the base next event if necessary on the next
443 * tick.
444 */
disarm_timer(struct k_itimer * timer,struct task_struct * p)445 static void disarm_timer(struct k_itimer *timer, struct task_struct *p)
446 {
447 struct cpu_timer *ctmr = &timer->it.cpu;
448 struct posix_cputimer_base *base;
449
450 if (!cpu_timer_dequeue(ctmr))
451 return;
452
453 base = timer_base(timer, p);
454 if (cpu_timer_getexpires(ctmr) == base->nextevt)
455 trigger_base_recalc_expires(timer, p);
456 }
457
458
459 /*
460 * Clean up a CPU-clock timer that is about to be destroyed.
461 * This is called from timer deletion with the timer already locked.
462 * If we return TIMER_RETRY, it's necessary to release the timer's lock
463 * and try again. (This happens when the timer is in the middle of firing.)
464 */
posix_cpu_timer_del(struct k_itimer * timer)465 static int posix_cpu_timer_del(struct k_itimer *timer)
466 {
467 struct cpu_timer *ctmr = &timer->it.cpu;
468 struct sighand_struct *sighand;
469 struct task_struct *p;
470 unsigned long flags;
471 int ret = 0;
472
473 rcu_read_lock();
474 p = cpu_timer_task_rcu(timer);
475 if (!p)
476 goto out;
477
478 /*
479 * Protect against sighand release/switch in exit/exec and process/
480 * thread timer list entry concurrent read/writes.
481 */
482 sighand = lock_task_sighand(p, &flags);
483 if (unlikely(sighand == NULL)) {
484 /*
485 * This raced with the reaping of the task. The exit cleanup
486 * should have removed this timer from the timer queue.
487 */
488 WARN_ON_ONCE(ctmr->head || timerqueue_node_queued(&ctmr->node));
489 } else {
490 if (timer->it.cpu.firing)
491 ret = TIMER_RETRY;
492 else
493 disarm_timer(timer, p);
494
495 unlock_task_sighand(p, &flags);
496 }
497
498 out:
499 rcu_read_unlock();
500 if (!ret)
501 put_pid(ctmr->pid);
502
503 return ret;
504 }
505
cleanup_timerqueue(struct timerqueue_head * head)506 static void cleanup_timerqueue(struct timerqueue_head *head)
507 {
508 struct timerqueue_node *node;
509 struct cpu_timer *ctmr;
510
511 while ((node = timerqueue_getnext(head))) {
512 timerqueue_del(head, node);
513 ctmr = container_of(node, struct cpu_timer, node);
514 ctmr->head = NULL;
515 }
516 }
517
518 /*
519 * Clean out CPU timers which are still armed when a thread exits. The
520 * timers are only removed from the list. No other updates are done. The
521 * corresponding posix timers are still accessible, but cannot be rearmed.
522 *
523 * This must be called with the siglock held.
524 */
cleanup_timers(struct posix_cputimers * pct)525 static void cleanup_timers(struct posix_cputimers *pct)
526 {
527 cleanup_timerqueue(&pct->bases[CPUCLOCK_PROF].tqhead);
528 cleanup_timerqueue(&pct->bases[CPUCLOCK_VIRT].tqhead);
529 cleanup_timerqueue(&pct->bases[CPUCLOCK_SCHED].tqhead);
530 }
531
532 /*
533 * These are both called with the siglock held, when the current thread
534 * is being reaped. When the final (leader) thread in the group is reaped,
535 * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
536 */
posix_cpu_timers_exit(struct task_struct * tsk)537 void posix_cpu_timers_exit(struct task_struct *tsk)
538 {
539 cleanup_timers(&tsk->posix_cputimers);
540 }
posix_cpu_timers_exit_group(struct task_struct * tsk)541 void posix_cpu_timers_exit_group(struct task_struct *tsk)
542 {
543 cleanup_timers(&tsk->signal->posix_cputimers);
544 }
545
546 /*
547 * Insert the timer on the appropriate list before any timers that
548 * expire later. This must be called with the sighand lock held.
549 */
arm_timer(struct k_itimer * timer,struct task_struct * p)550 static void arm_timer(struct k_itimer *timer, struct task_struct *p)
551 {
552 struct posix_cputimer_base *base = timer_base(timer, p);
553 struct cpu_timer *ctmr = &timer->it.cpu;
554 u64 newexp = cpu_timer_getexpires(ctmr);
555
556 if (!cpu_timer_enqueue(&base->tqhead, ctmr))
557 return;
558
559 /*
560 * We are the new earliest-expiring POSIX 1.b timer, hence
561 * need to update expiration cache. Take into account that
562 * for process timers we share expiration cache with itimers
563 * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
564 */
565 if (newexp < base->nextevt)
566 base->nextevt = newexp;
567
568 if (CPUCLOCK_PERTHREAD(timer->it_clock))
569 tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER);
570 else
571 tick_dep_set_signal(p, TICK_DEP_BIT_POSIX_TIMER);
572 }
573
574 /*
575 * The timer is locked, fire it and arrange for its reload.
576 */
cpu_timer_fire(struct k_itimer * timer)577 static void cpu_timer_fire(struct k_itimer *timer)
578 {
579 struct cpu_timer *ctmr = &timer->it.cpu;
580
581 if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
582 /*
583 * User don't want any signal.
584 */
585 cpu_timer_setexpires(ctmr, 0);
586 } else if (unlikely(timer->sigq == NULL)) {
587 /*
588 * This a special case for clock_nanosleep,
589 * not a normal timer from sys_timer_create.
590 */
591 wake_up_process(timer->it_process);
592 cpu_timer_setexpires(ctmr, 0);
593 } else if (!timer->it_interval) {
594 /*
595 * One-shot timer. Clear it as soon as it's fired.
596 */
597 posix_timer_event(timer, 0);
598 cpu_timer_setexpires(ctmr, 0);
599 } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
600 /*
601 * The signal did not get queued because the signal
602 * was ignored, so we won't get any callback to
603 * reload the timer. But we need to keep it
604 * ticking in case the signal is deliverable next time.
605 */
606 posix_cpu_timer_rearm(timer);
607 ++timer->it_requeue_pending;
608 }
609 }
610
611 /*
612 * Guts of sys_timer_settime for CPU timers.
613 * This is called with the timer locked and interrupts disabled.
614 * If we return TIMER_RETRY, it's necessary to release the timer's lock
615 * and try again. (This happens when the timer is in the middle of firing.)
616 */
posix_cpu_timer_set(struct k_itimer * timer,int timer_flags,struct itimerspec64 * new,struct itimerspec64 * old)617 static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
618 struct itimerspec64 *new, struct itimerspec64 *old)
619 {
620 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
621 u64 old_expires, new_expires, old_incr, val;
622 struct cpu_timer *ctmr = &timer->it.cpu;
623 struct sighand_struct *sighand;
624 struct task_struct *p;
625 unsigned long flags;
626 int ret = 0;
627
628 rcu_read_lock();
629 p = cpu_timer_task_rcu(timer);
630 if (!p) {
631 /*
632 * If p has just been reaped, we can no
633 * longer get any information about it at all.
634 */
635 rcu_read_unlock();
636 return -ESRCH;
637 }
638
639 /*
640 * Use the to_ktime conversion because that clamps the maximum
641 * value to KTIME_MAX and avoid multiplication overflows.
642 */
643 new_expires = ktime_to_ns(timespec64_to_ktime(new->it_value));
644
645 /*
646 * Protect against sighand release/switch in exit/exec and p->cpu_timers
647 * and p->signal->cpu_timers read/write in arm_timer()
648 */
649 sighand = lock_task_sighand(p, &flags);
650 /*
651 * If p has just been reaped, we can no
652 * longer get any information about it at all.
653 */
654 if (unlikely(sighand == NULL)) {
655 rcu_read_unlock();
656 return -ESRCH;
657 }
658
659 /*
660 * Disarm any old timer after extracting its expiry time.
661 */
662 old_incr = timer->it_interval;
663 old_expires = cpu_timer_getexpires(ctmr);
664
665 if (unlikely(timer->it.cpu.firing)) {
666 timer->it.cpu.firing = -1;
667 ret = TIMER_RETRY;
668 } else {
669 cpu_timer_dequeue(ctmr);
670 }
671
672 /*
673 * We need to sample the current value to convert the new
674 * value from to relative and absolute, and to convert the
675 * old value from absolute to relative. To set a process
676 * timer, we need a sample to balance the thread expiry
677 * times (in arm_timer). With an absolute time, we must
678 * check if it's already passed. In short, we need a sample.
679 */
680 if (CPUCLOCK_PERTHREAD(timer->it_clock))
681 val = cpu_clock_sample(clkid, p);
682 else
683 val = cpu_clock_sample_group(clkid, p, true);
684
685 if (old) {
686 if (old_expires == 0) {
687 old->it_value.tv_sec = 0;
688 old->it_value.tv_nsec = 0;
689 } else {
690 /*
691 * Update the timer in case it has overrun already.
692 * If it has, we'll report it as having overrun and
693 * with the next reloaded timer already ticking,
694 * though we are swallowing that pending
695 * notification here to install the new setting.
696 */
697 u64 exp = bump_cpu_timer(timer, val);
698
699 if (val < exp) {
700 old_expires = exp - val;
701 old->it_value = ns_to_timespec64(old_expires);
702 } else {
703 old->it_value.tv_nsec = 1;
704 old->it_value.tv_sec = 0;
705 }
706 }
707 }
708
709 if (unlikely(ret)) {
710 /*
711 * We are colliding with the timer actually firing.
712 * Punt after filling in the timer's old value, and
713 * disable this firing since we are already reporting
714 * it as an overrun (thanks to bump_cpu_timer above).
715 */
716 unlock_task_sighand(p, &flags);
717 goto out;
718 }
719
720 if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
721 new_expires += val;
722 }
723
724 /*
725 * Install the new expiry time (or zero).
726 * For a timer with no notification action, we don't actually
727 * arm the timer (we'll just fake it for timer_gettime).
728 */
729 cpu_timer_setexpires(ctmr, new_expires);
730 if (new_expires != 0 && val < new_expires) {
731 arm_timer(timer, p);
732 }
733
734 unlock_task_sighand(p, &flags);
735 /*
736 * Install the new reload setting, and
737 * set up the signal and overrun bookkeeping.
738 */
739 timer->it_interval = timespec64_to_ktime(new->it_interval);
740
741 /*
742 * This acts as a modification timestamp for the timer,
743 * so any automatic reload attempt will punt on seeing
744 * that we have reset the timer manually.
745 */
746 timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
747 ~REQUEUE_PENDING;
748 timer->it_overrun_last = 0;
749 timer->it_overrun = -1;
750
751 if (val >= new_expires) {
752 if (new_expires != 0) {
753 /*
754 * The designated time already passed, so we notify
755 * immediately, even if the thread never runs to
756 * accumulate more time on this clock.
757 */
758 cpu_timer_fire(timer);
759 }
760
761 /*
762 * Make sure we don't keep around the process wide cputime
763 * counter or the tick dependency if they are not necessary.
764 */
765 sighand = lock_task_sighand(p, &flags);
766 if (!sighand)
767 goto out;
768
769 if (!cpu_timer_queued(ctmr))
770 trigger_base_recalc_expires(timer, p);
771
772 unlock_task_sighand(p, &flags);
773 }
774 out:
775 rcu_read_unlock();
776 if (old)
777 old->it_interval = ns_to_timespec64(old_incr);
778
779 return ret;
780 }
781
posix_cpu_timer_get(struct k_itimer * timer,struct itimerspec64 * itp)782 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp)
783 {
784 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
785 struct cpu_timer *ctmr = &timer->it.cpu;
786 u64 now, expires = cpu_timer_getexpires(ctmr);
787 struct task_struct *p;
788
789 rcu_read_lock();
790 p = cpu_timer_task_rcu(timer);
791 if (!p)
792 goto out;
793
794 /*
795 * Easy part: convert the reload time.
796 */
797 itp->it_interval = ktime_to_timespec64(timer->it_interval);
798
799 if (!expires)
800 goto out;
801
802 /*
803 * Sample the clock to take the difference with the expiry time.
804 */
805 if (CPUCLOCK_PERTHREAD(timer->it_clock))
806 now = cpu_clock_sample(clkid, p);
807 else
808 now = cpu_clock_sample_group(clkid, p, false);
809
810 if (now < expires) {
811 itp->it_value = ns_to_timespec64(expires - now);
812 } else {
813 /*
814 * The timer should have expired already, but the firing
815 * hasn't taken place yet. Say it's just about to expire.
816 */
817 itp->it_value.tv_nsec = 1;
818 itp->it_value.tv_sec = 0;
819 }
820 out:
821 rcu_read_unlock();
822 }
823
824 #define MAX_COLLECTED 20
825
collect_timerqueue(struct timerqueue_head * head,struct list_head * firing,u64 now)826 static u64 collect_timerqueue(struct timerqueue_head *head,
827 struct list_head *firing, u64 now)
828 {
829 struct timerqueue_node *next;
830 int i = 0;
831
832 while ((next = timerqueue_getnext(head))) {
833 struct cpu_timer *ctmr;
834 u64 expires;
835
836 ctmr = container_of(next, struct cpu_timer, node);
837 expires = cpu_timer_getexpires(ctmr);
838 /* Limit the number of timers to expire at once */
839 if (++i == MAX_COLLECTED || now < expires)
840 return expires;
841
842 ctmr->firing = 1;
843 /* See posix_cpu_timer_wait_running() */
844 rcu_assign_pointer(ctmr->handling, current);
845 cpu_timer_dequeue(ctmr);
846 list_add_tail(&ctmr->elist, firing);
847 }
848
849 return U64_MAX;
850 }
851
collect_posix_cputimers(struct posix_cputimers * pct,u64 * samples,struct list_head * firing)852 static void collect_posix_cputimers(struct posix_cputimers *pct, u64 *samples,
853 struct list_head *firing)
854 {
855 struct posix_cputimer_base *base = pct->bases;
856 int i;
857
858 for (i = 0; i < CPUCLOCK_MAX; i++, base++) {
859 base->nextevt = collect_timerqueue(&base->tqhead, firing,
860 samples[i]);
861 }
862 }
863
check_dl_overrun(struct task_struct * tsk)864 static inline void check_dl_overrun(struct task_struct *tsk)
865 {
866 if (tsk->dl.dl_overrun) {
867 tsk->dl.dl_overrun = 0;
868 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
869 }
870 }
871
check_rlimit(u64 time,u64 limit,int signo,bool rt,bool hard)872 static bool check_rlimit(u64 time, u64 limit, int signo, bool rt, bool hard)
873 {
874 if (time < limit)
875 return false;
876
877 if (print_fatal_signals) {
878 pr_info("%s Watchdog Timeout (%s): %s[%d]\n",
879 rt ? "RT" : "CPU", hard ? "hard" : "soft",
880 current->comm, task_pid_nr(current));
881 }
882 __group_send_sig_info(signo, SEND_SIG_PRIV, current);
883 return true;
884 }
885
886 /*
887 * Check for any per-thread CPU timers that have fired and move them off
888 * the tsk->cpu_timers[N] list onto the firing list. Here we update the
889 * tsk->it_*_expires values to reflect the remaining thread CPU timers.
890 */
check_thread_timers(struct task_struct * tsk,struct list_head * firing)891 static void check_thread_timers(struct task_struct *tsk,
892 struct list_head *firing)
893 {
894 struct posix_cputimers *pct = &tsk->posix_cputimers;
895 u64 samples[CPUCLOCK_MAX];
896 unsigned long soft;
897
898 if (dl_task(tsk))
899 check_dl_overrun(tsk);
900
901 if (expiry_cache_is_inactive(pct))
902 return;
903
904 task_sample_cputime(tsk, samples);
905 collect_posix_cputimers(pct, samples, firing);
906
907 /*
908 * Check for the special case thread timers.
909 */
910 soft = task_rlimit(tsk, RLIMIT_RTTIME);
911 if (soft != RLIM_INFINITY) {
912 /* Task RT timeout is accounted in jiffies. RTTIME is usec */
913 unsigned long rttime = tsk->rt.timeout * (USEC_PER_SEC / HZ);
914 unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME);
915
916 /* At the hard limit, send SIGKILL. No further action. */
917 if (hard != RLIM_INFINITY &&
918 check_rlimit(rttime, hard, SIGKILL, true, true))
919 return;
920
921 /* At the soft limit, send a SIGXCPU every second */
922 if (check_rlimit(rttime, soft, SIGXCPU, true, false)) {
923 soft += USEC_PER_SEC;
924 tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur = soft;
925 }
926 }
927
928 if (expiry_cache_is_inactive(pct))
929 tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER);
930 }
931
stop_process_timers(struct signal_struct * sig)932 static inline void stop_process_timers(struct signal_struct *sig)
933 {
934 struct posix_cputimers *pct = &sig->posix_cputimers;
935
936 /* Turn off the active flag. This is done without locking. */
937 WRITE_ONCE(pct->timers_active, false);
938 tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER);
939 }
940
check_cpu_itimer(struct task_struct * tsk,struct cpu_itimer * it,u64 * expires,u64 cur_time,int signo)941 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
942 u64 *expires, u64 cur_time, int signo)
943 {
944 if (!it->expires)
945 return;
946
947 if (cur_time >= it->expires) {
948 if (it->incr)
949 it->expires += it->incr;
950 else
951 it->expires = 0;
952
953 trace_itimer_expire(signo == SIGPROF ?
954 ITIMER_PROF : ITIMER_VIRTUAL,
955 task_tgid(tsk), cur_time);
956 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
957 }
958
959 if (it->expires && it->expires < *expires)
960 *expires = it->expires;
961 }
962
963 /*
964 * Check for any per-thread CPU timers that have fired and move them
965 * off the tsk->*_timers list onto the firing list. Per-thread timers
966 * have already been taken off.
967 */
check_process_timers(struct task_struct * tsk,struct list_head * firing)968 static void check_process_timers(struct task_struct *tsk,
969 struct list_head *firing)
970 {
971 struct signal_struct *const sig = tsk->signal;
972 struct posix_cputimers *pct = &sig->posix_cputimers;
973 u64 samples[CPUCLOCK_MAX];
974 unsigned long soft;
975
976 /*
977 * If there are no active process wide timers (POSIX 1.b, itimers,
978 * RLIMIT_CPU) nothing to check. Also skip the process wide timer
979 * processing when there is already another task handling them.
980 */
981 if (!READ_ONCE(pct->timers_active) || pct->expiry_active)
982 return;
983
984 /*
985 * Signify that a thread is checking for process timers.
986 * Write access to this field is protected by the sighand lock.
987 */
988 pct->expiry_active = true;
989
990 /*
991 * Collect the current process totals. Group accounting is active
992 * so the sample can be taken directly.
993 */
994 proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic, samples);
995 collect_posix_cputimers(pct, samples, firing);
996
997 /*
998 * Check for the special case process timers.
999 */
1000 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF],
1001 &pct->bases[CPUCLOCK_PROF].nextevt,
1002 samples[CPUCLOCK_PROF], SIGPROF);
1003 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT],
1004 &pct->bases[CPUCLOCK_VIRT].nextevt,
1005 samples[CPUCLOCK_VIRT], SIGVTALRM);
1006
1007 soft = task_rlimit(tsk, RLIMIT_CPU);
1008 if (soft != RLIM_INFINITY) {
1009 /* RLIMIT_CPU is in seconds. Samples are nanoseconds */
1010 unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU);
1011 u64 ptime = samples[CPUCLOCK_PROF];
1012 u64 softns = (u64)soft * NSEC_PER_SEC;
1013 u64 hardns = (u64)hard * NSEC_PER_SEC;
1014
1015 /* At the hard limit, send SIGKILL. No further action. */
1016 if (hard != RLIM_INFINITY &&
1017 check_rlimit(ptime, hardns, SIGKILL, false, true))
1018 return;
1019
1020 /* At the soft limit, send a SIGXCPU every second */
1021 if (check_rlimit(ptime, softns, SIGXCPU, false, false)) {
1022 sig->rlim[RLIMIT_CPU].rlim_cur = soft + 1;
1023 softns += NSEC_PER_SEC;
1024 }
1025
1026 /* Update the expiry cache */
1027 if (softns < pct->bases[CPUCLOCK_PROF].nextevt)
1028 pct->bases[CPUCLOCK_PROF].nextevt = softns;
1029 }
1030
1031 if (expiry_cache_is_inactive(pct))
1032 stop_process_timers(sig);
1033
1034 pct->expiry_active = false;
1035 }
1036
1037 /*
1038 * This is called from the signal code (via posixtimer_rearm)
1039 * when the last timer signal was delivered and we have to reload the timer.
1040 */
posix_cpu_timer_rearm(struct k_itimer * timer)1041 static void posix_cpu_timer_rearm(struct k_itimer *timer)
1042 {
1043 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
1044 struct task_struct *p;
1045 struct sighand_struct *sighand;
1046 unsigned long flags;
1047 u64 now;
1048
1049 rcu_read_lock();
1050 p = cpu_timer_task_rcu(timer);
1051 if (!p)
1052 goto out;
1053
1054 /* Protect timer list r/w in arm_timer() */
1055 sighand = lock_task_sighand(p, &flags);
1056 if (unlikely(sighand == NULL))
1057 goto out;
1058
1059 /*
1060 * Fetch the current sample and update the timer's expiry time.
1061 */
1062 if (CPUCLOCK_PERTHREAD(timer->it_clock))
1063 now = cpu_clock_sample(clkid, p);
1064 else
1065 now = cpu_clock_sample_group(clkid, p, true);
1066
1067 bump_cpu_timer(timer, now);
1068
1069 /*
1070 * Now re-arm for the new expiry time.
1071 */
1072 arm_timer(timer, p);
1073 unlock_task_sighand(p, &flags);
1074 out:
1075 rcu_read_unlock();
1076 }
1077
1078 /**
1079 * task_cputimers_expired - Check whether posix CPU timers are expired
1080 *
1081 * @samples: Array of current samples for the CPUCLOCK clocks
1082 * @pct: Pointer to a posix_cputimers container
1083 *
1084 * Returns true if any member of @samples is greater than the corresponding
1085 * member of @pct->bases[CLK].nextevt. False otherwise
1086 */
1087 static inline bool
task_cputimers_expired(const u64 * samples,struct posix_cputimers * pct)1088 task_cputimers_expired(const u64 *samples, struct posix_cputimers *pct)
1089 {
1090 int i;
1091
1092 for (i = 0; i < CPUCLOCK_MAX; i++) {
1093 if (samples[i] >= pct->bases[i].nextevt)
1094 return true;
1095 }
1096 return false;
1097 }
1098
1099 /**
1100 * fastpath_timer_check - POSIX CPU timers fast path.
1101 *
1102 * @tsk: The task (thread) being checked.
1103 *
1104 * Check the task and thread group timers. If both are zero (there are no
1105 * timers set) return false. Otherwise snapshot the task and thread group
1106 * timers and compare them with the corresponding expiration times. Return
1107 * true if a timer has expired, else return false.
1108 */
fastpath_timer_check(struct task_struct * tsk)1109 static inline bool fastpath_timer_check(struct task_struct *tsk)
1110 {
1111 struct posix_cputimers *pct = &tsk->posix_cputimers;
1112 struct signal_struct *sig;
1113
1114 if (!expiry_cache_is_inactive(pct)) {
1115 u64 samples[CPUCLOCK_MAX];
1116
1117 task_sample_cputime(tsk, samples);
1118 if (task_cputimers_expired(samples, pct))
1119 return true;
1120 }
1121
1122 sig = tsk->signal;
1123 pct = &sig->posix_cputimers;
1124 /*
1125 * Check if thread group timers expired when timers are active and
1126 * no other thread in the group is already handling expiry for
1127 * thread group cputimers. These fields are read without the
1128 * sighand lock. However, this is fine because this is meant to be
1129 * a fastpath heuristic to determine whether we should try to
1130 * acquire the sighand lock to handle timer expiry.
1131 *
1132 * In the worst case scenario, if concurrently timers_active is set
1133 * or expiry_active is cleared, but the current thread doesn't see
1134 * the change yet, the timer checks are delayed until the next
1135 * thread in the group gets a scheduler interrupt to handle the
1136 * timer. This isn't an issue in practice because these types of
1137 * delays with signals actually getting sent are expected.
1138 */
1139 if (READ_ONCE(pct->timers_active) && !READ_ONCE(pct->expiry_active)) {
1140 u64 samples[CPUCLOCK_MAX];
1141
1142 proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic,
1143 samples);
1144
1145 if (task_cputimers_expired(samples, pct))
1146 return true;
1147 }
1148
1149 if (dl_task(tsk) && tsk->dl.dl_overrun)
1150 return true;
1151
1152 return false;
1153 }
1154
1155 static void handle_posix_cpu_timers(struct task_struct *tsk);
1156
1157 #ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
posix_cpu_timers_work(struct callback_head * work)1158 static void posix_cpu_timers_work(struct callback_head *work)
1159 {
1160 struct posix_cputimers_work *cw = container_of(work, typeof(*cw), work);
1161
1162 mutex_lock(&cw->mutex);
1163 handle_posix_cpu_timers(current);
1164 mutex_unlock(&cw->mutex);
1165 }
1166
1167 /*
1168 * Invoked from the posix-timer core when a cancel operation failed because
1169 * the timer is marked firing. The caller holds rcu_read_lock(), which
1170 * protects the timer and the task which is expiring it from being freed.
1171 */
posix_cpu_timer_wait_running(struct k_itimer * timr)1172 static void posix_cpu_timer_wait_running(struct k_itimer *timr)
1173 {
1174 struct task_struct *tsk = rcu_dereference(timr->it.cpu.handling);
1175
1176 /* Has the handling task completed expiry already? */
1177 if (!tsk)
1178 return;
1179
1180 /* Ensure that the task cannot go away */
1181 get_task_struct(tsk);
1182 /* Now drop the RCU protection so the mutex can be locked */
1183 rcu_read_unlock();
1184 /* Wait on the expiry mutex */
1185 mutex_lock(&tsk->posix_cputimers_work.mutex);
1186 /* Release it immediately again. */
1187 mutex_unlock(&tsk->posix_cputimers_work.mutex);
1188 /* Drop the task reference. */
1189 put_task_struct(tsk);
1190 /* Relock RCU so the callsite is balanced */
1191 rcu_read_lock();
1192 }
1193
posix_cpu_timer_wait_running_nsleep(struct k_itimer * timr)1194 static void posix_cpu_timer_wait_running_nsleep(struct k_itimer *timr)
1195 {
1196 /* Ensure that timr->it.cpu.handling task cannot go away */
1197 rcu_read_lock();
1198 spin_unlock_irq(&timr->it_lock);
1199 posix_cpu_timer_wait_running(timr);
1200 rcu_read_unlock();
1201 /* @timr is on stack and is valid */
1202 spin_lock_irq(&timr->it_lock);
1203 }
1204
1205 /*
1206 * Clear existing posix CPU timers task work.
1207 */
clear_posix_cputimers_work(struct task_struct * p)1208 void clear_posix_cputimers_work(struct task_struct *p)
1209 {
1210 /*
1211 * A copied work entry from the old task is not meaningful, clear it.
1212 * N.B. init_task_work will not do this.
1213 */
1214 memset(&p->posix_cputimers_work.work, 0,
1215 sizeof(p->posix_cputimers_work.work));
1216 init_task_work(&p->posix_cputimers_work.work,
1217 posix_cpu_timers_work);
1218 mutex_init(&p->posix_cputimers_work.mutex);
1219 p->posix_cputimers_work.scheduled = false;
1220 }
1221
1222 /*
1223 * Initialize posix CPU timers task work in init task. Out of line to
1224 * keep the callback static and to avoid header recursion hell.
1225 */
posix_cputimers_init_work(void)1226 void __init posix_cputimers_init_work(void)
1227 {
1228 clear_posix_cputimers_work(current);
1229 }
1230
1231 /*
1232 * Note: All operations on tsk->posix_cputimer_work.scheduled happen either
1233 * in hard interrupt context or in task context with interrupts
1234 * disabled. Aside of that the writer/reader interaction is always in the
1235 * context of the current task, which means they are strict per CPU.
1236 */
posix_cpu_timers_work_scheduled(struct task_struct * tsk)1237 static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk)
1238 {
1239 return tsk->posix_cputimers_work.scheduled;
1240 }
1241
__run_posix_cpu_timers(struct task_struct * tsk)1242 static inline void __run_posix_cpu_timers(struct task_struct *tsk)
1243 {
1244 if (WARN_ON_ONCE(tsk->posix_cputimers_work.scheduled))
1245 return;
1246
1247 /* Schedule task work to actually expire the timers */
1248 tsk->posix_cputimers_work.scheduled = true;
1249 task_work_add(tsk, &tsk->posix_cputimers_work.work, TWA_RESUME);
1250 }
1251
posix_cpu_timers_enable_work(struct task_struct * tsk,unsigned long start)1252 static inline bool posix_cpu_timers_enable_work(struct task_struct *tsk,
1253 unsigned long start)
1254 {
1255 bool ret = true;
1256
1257 /*
1258 * On !RT kernels interrupts are disabled while collecting expired
1259 * timers, so no tick can happen and the fast path check can be
1260 * reenabled without further checks.
1261 */
1262 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
1263 tsk->posix_cputimers_work.scheduled = false;
1264 return true;
1265 }
1266
1267 /*
1268 * On RT enabled kernels ticks can happen while the expired timers
1269 * are collected under sighand lock. But any tick which observes
1270 * the CPUTIMERS_WORK_SCHEDULED bit set, does not run the fastpath
1271 * checks. So reenabling the tick work has do be done carefully:
1272 *
1273 * Disable interrupts and run the fast path check if jiffies have
1274 * advanced since the collecting of expired timers started. If
1275 * jiffies have not advanced or the fast path check did not find
1276 * newly expired timers, reenable the fast path check in the timer
1277 * interrupt. If there are newly expired timers, return false and
1278 * let the collection loop repeat.
1279 */
1280 local_irq_disable();
1281 if (start != jiffies && fastpath_timer_check(tsk))
1282 ret = false;
1283 else
1284 tsk->posix_cputimers_work.scheduled = false;
1285 local_irq_enable();
1286
1287 return ret;
1288 }
1289 #else /* CONFIG_POSIX_CPU_TIMERS_TASK_WORK */
__run_posix_cpu_timers(struct task_struct * tsk)1290 static inline void __run_posix_cpu_timers(struct task_struct *tsk)
1291 {
1292 lockdep_posixtimer_enter();
1293 handle_posix_cpu_timers(tsk);
1294 lockdep_posixtimer_exit();
1295 }
1296
posix_cpu_timer_wait_running(struct k_itimer * timr)1297 static void posix_cpu_timer_wait_running(struct k_itimer *timr)
1298 {
1299 cpu_relax();
1300 }
1301
posix_cpu_timer_wait_running_nsleep(struct k_itimer * timr)1302 static void posix_cpu_timer_wait_running_nsleep(struct k_itimer *timr)
1303 {
1304 spin_unlock_irq(&timr->it_lock);
1305 cpu_relax();
1306 spin_lock_irq(&timr->it_lock);
1307 }
1308
posix_cpu_timers_work_scheduled(struct task_struct * tsk)1309 static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk)
1310 {
1311 return false;
1312 }
1313
posix_cpu_timers_enable_work(struct task_struct * tsk,unsigned long start)1314 static inline bool posix_cpu_timers_enable_work(struct task_struct *tsk,
1315 unsigned long start)
1316 {
1317 return true;
1318 }
1319 #endif /* CONFIG_POSIX_CPU_TIMERS_TASK_WORK */
1320
handle_posix_cpu_timers(struct task_struct * tsk)1321 static void handle_posix_cpu_timers(struct task_struct *tsk)
1322 {
1323 struct k_itimer *timer, *next;
1324 unsigned long flags, start;
1325 LIST_HEAD(firing);
1326
1327 if (!lock_task_sighand(tsk, &flags))
1328 return;
1329
1330 do {
1331 /*
1332 * On RT locking sighand lock does not disable interrupts,
1333 * so this needs to be careful vs. ticks. Store the current
1334 * jiffies value.
1335 */
1336 start = READ_ONCE(jiffies);
1337 barrier();
1338
1339 /*
1340 * Here we take off tsk->signal->cpu_timers[N] and
1341 * tsk->cpu_timers[N] all the timers that are firing, and
1342 * put them on the firing list.
1343 */
1344 check_thread_timers(tsk, &firing);
1345
1346 check_process_timers(tsk, &firing);
1347
1348 /*
1349 * The above timer checks have updated the expiry cache and
1350 * because nothing can have queued or modified timers after
1351 * sighand lock was taken above it is guaranteed to be
1352 * consistent. So the next timer interrupt fastpath check
1353 * will find valid data.
1354 *
1355 * If timer expiry runs in the timer interrupt context then
1356 * the loop is not relevant as timers will be directly
1357 * expired in interrupt context. The stub function below
1358 * returns always true which allows the compiler to
1359 * optimize the loop out.
1360 *
1361 * If timer expiry is deferred to task work context then
1362 * the following rules apply:
1363 *
1364 * - On !RT kernels no tick can have happened on this CPU
1365 * after sighand lock was acquired because interrupts are
1366 * disabled. So reenabling task work before dropping
1367 * sighand lock and reenabling interrupts is race free.
1368 *
1369 * - On RT kernels ticks might have happened but the tick
1370 * work ignored posix CPU timer handling because the
1371 * CPUTIMERS_WORK_SCHEDULED bit is set. Reenabling work
1372 * must be done very carefully including a check whether
1373 * ticks have happened since the start of the timer
1374 * expiry checks. posix_cpu_timers_enable_work() takes
1375 * care of that and eventually lets the expiry checks
1376 * run again.
1377 */
1378 } while (!posix_cpu_timers_enable_work(tsk, start));
1379
1380 /*
1381 * We must release sighand lock before taking any timer's lock.
1382 * There is a potential race with timer deletion here, as the
1383 * siglock now protects our private firing list. We have set
1384 * the firing flag in each timer, so that a deletion attempt
1385 * that gets the timer lock before we do will give it up and
1386 * spin until we've taken care of that timer below.
1387 */
1388 unlock_task_sighand(tsk, &flags);
1389
1390 /*
1391 * Now that all the timers on our list have the firing flag,
1392 * no one will touch their list entries but us. We'll take
1393 * each timer's lock before clearing its firing flag, so no
1394 * timer call will interfere.
1395 */
1396 list_for_each_entry_safe(timer, next, &firing, it.cpu.elist) {
1397 int cpu_firing;
1398
1399 /*
1400 * spin_lock() is sufficient here even independent of the
1401 * expiry context. If expiry happens in hard interrupt
1402 * context it's obvious. For task work context it's safe
1403 * because all other operations on timer::it_lock happen in
1404 * task context (syscall or exit).
1405 */
1406 spin_lock(&timer->it_lock);
1407 list_del_init(&timer->it.cpu.elist);
1408 cpu_firing = timer->it.cpu.firing;
1409 timer->it.cpu.firing = 0;
1410 /*
1411 * The firing flag is -1 if we collided with a reset
1412 * of the timer, which already reported this
1413 * almost-firing as an overrun. So don't generate an event.
1414 */
1415 if (likely(cpu_firing >= 0))
1416 cpu_timer_fire(timer);
1417 /* See posix_cpu_timer_wait_running() */
1418 rcu_assign_pointer(timer->it.cpu.handling, NULL);
1419 spin_unlock(&timer->it_lock);
1420 }
1421 }
1422
1423 /*
1424 * This is called from the timer interrupt handler. The irq handler has
1425 * already updated our counts. We need to check if any timers fire now.
1426 * Interrupts are disabled.
1427 */
run_posix_cpu_timers(void)1428 void run_posix_cpu_timers(void)
1429 {
1430 struct task_struct *tsk = current;
1431
1432 lockdep_assert_irqs_disabled();
1433
1434 /*
1435 * If the actual expiry is deferred to task work context and the
1436 * work is already scheduled there is no point to do anything here.
1437 */
1438 if (posix_cpu_timers_work_scheduled(tsk))
1439 return;
1440
1441 /*
1442 * The fast path checks that there are no expired thread or thread
1443 * group timers. If that's so, just return.
1444 */
1445 if (!fastpath_timer_check(tsk))
1446 return;
1447
1448 __run_posix_cpu_timers(tsk);
1449 }
1450
1451 /*
1452 * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1453 * The tsk->sighand->siglock must be held by the caller.
1454 */
set_process_cpu_timer(struct task_struct * tsk,unsigned int clkid,u64 * newval,u64 * oldval)1455 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clkid,
1456 u64 *newval, u64 *oldval)
1457 {
1458 u64 now, *nextevt;
1459
1460 if (WARN_ON_ONCE(clkid >= CPUCLOCK_SCHED))
1461 return;
1462
1463 nextevt = &tsk->signal->posix_cputimers.bases[clkid].nextevt;
1464 now = cpu_clock_sample_group(clkid, tsk, true);
1465
1466 if (oldval) {
1467 /*
1468 * We are setting itimer. The *oldval is absolute and we update
1469 * it to be relative, *newval argument is relative and we update
1470 * it to be absolute.
1471 */
1472 if (*oldval) {
1473 if (*oldval <= now) {
1474 /* Just about to fire. */
1475 *oldval = TICK_NSEC;
1476 } else {
1477 *oldval -= now;
1478 }
1479 }
1480
1481 if (*newval)
1482 *newval += now;
1483 }
1484
1485 /*
1486 * Update expiration cache if this is the earliest timer. CPUCLOCK_PROF
1487 * expiry cache is also used by RLIMIT_CPU!.
1488 */
1489 if (*newval < *nextevt)
1490 *nextevt = *newval;
1491
1492 tick_dep_set_signal(tsk, TICK_DEP_BIT_POSIX_TIMER);
1493 }
1494
do_cpu_nanosleep(const clockid_t which_clock,int flags,const struct timespec64 * rqtp)1495 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1496 const struct timespec64 *rqtp)
1497 {
1498 struct itimerspec64 it;
1499 struct k_itimer timer;
1500 u64 expires;
1501 int error;
1502
1503 /*
1504 * Set up a temporary timer and then wait for it to go off.
1505 */
1506 memset(&timer, 0, sizeof timer);
1507 spin_lock_init(&timer.it_lock);
1508 timer.it_clock = which_clock;
1509 timer.it_overrun = -1;
1510 error = posix_cpu_timer_create(&timer);
1511 timer.it_process = current;
1512
1513 if (!error) {
1514 static struct itimerspec64 zero_it;
1515 struct restart_block *restart;
1516
1517 memset(&it, 0, sizeof(it));
1518 it.it_value = *rqtp;
1519
1520 spin_lock_irq(&timer.it_lock);
1521 error = posix_cpu_timer_set(&timer, flags, &it, NULL);
1522 if (error) {
1523 spin_unlock_irq(&timer.it_lock);
1524 return error;
1525 }
1526
1527 while (!signal_pending(current)) {
1528 if (!cpu_timer_getexpires(&timer.it.cpu)) {
1529 /*
1530 * Our timer fired and was reset, below
1531 * deletion can not fail.
1532 */
1533 posix_cpu_timer_del(&timer);
1534 spin_unlock_irq(&timer.it_lock);
1535 return 0;
1536 }
1537
1538 /*
1539 * Block until cpu_timer_fire (or a signal) wakes us.
1540 */
1541 __set_current_state(TASK_INTERRUPTIBLE);
1542 spin_unlock_irq(&timer.it_lock);
1543 schedule();
1544 spin_lock_irq(&timer.it_lock);
1545 }
1546
1547 /*
1548 * We were interrupted by a signal.
1549 */
1550 expires = cpu_timer_getexpires(&timer.it.cpu);
1551 error = posix_cpu_timer_set(&timer, 0, &zero_it, &it);
1552 if (!error) {
1553 /* Timer is now unarmed, deletion can not fail. */
1554 posix_cpu_timer_del(&timer);
1555 } else {
1556 while (error == TIMER_RETRY) {
1557 posix_cpu_timer_wait_running_nsleep(&timer);
1558 error = posix_cpu_timer_del(&timer);
1559 }
1560 }
1561
1562 spin_unlock_irq(&timer.it_lock);
1563
1564 if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) {
1565 /*
1566 * It actually did fire already.
1567 */
1568 return 0;
1569 }
1570
1571 error = -ERESTART_RESTARTBLOCK;
1572 /*
1573 * Report back to the user the time still remaining.
1574 */
1575 restart = ¤t->restart_block;
1576 restart->nanosleep.expires = expires;
1577 if (restart->nanosleep.type != TT_NONE)
1578 error = nanosleep_copyout(restart, &it.it_value);
1579 }
1580
1581 return error;
1582 }
1583
1584 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1585
posix_cpu_nsleep(const clockid_t which_clock,int flags,const struct timespec64 * rqtp)1586 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1587 const struct timespec64 *rqtp)
1588 {
1589 struct restart_block *restart_block = ¤t->restart_block;
1590 int error;
1591
1592 /*
1593 * Diagnose required errors first.
1594 */
1595 if (CPUCLOCK_PERTHREAD(which_clock) &&
1596 (CPUCLOCK_PID(which_clock) == 0 ||
1597 CPUCLOCK_PID(which_clock) == task_pid_vnr(current)))
1598 return -EINVAL;
1599
1600 error = do_cpu_nanosleep(which_clock, flags, rqtp);
1601
1602 if (error == -ERESTART_RESTARTBLOCK) {
1603
1604 if (flags & TIMER_ABSTIME)
1605 return -ERESTARTNOHAND;
1606
1607 restart_block->nanosleep.clockid = which_clock;
1608 set_restart_fn(restart_block, posix_cpu_nsleep_restart);
1609 }
1610 return error;
1611 }
1612
posix_cpu_nsleep_restart(struct restart_block * restart_block)1613 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1614 {
1615 clockid_t which_clock = restart_block->nanosleep.clockid;
1616 struct timespec64 t;
1617
1618 t = ns_to_timespec64(restart_block->nanosleep.expires);
1619
1620 return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t);
1621 }
1622
1623 #define PROCESS_CLOCK make_process_cpuclock(0, CPUCLOCK_SCHED)
1624 #define THREAD_CLOCK make_thread_cpuclock(0, CPUCLOCK_SCHED)
1625
process_cpu_clock_getres(const clockid_t which_clock,struct timespec64 * tp)1626 static int process_cpu_clock_getres(const clockid_t which_clock,
1627 struct timespec64 *tp)
1628 {
1629 return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1630 }
process_cpu_clock_get(const clockid_t which_clock,struct timespec64 * tp)1631 static int process_cpu_clock_get(const clockid_t which_clock,
1632 struct timespec64 *tp)
1633 {
1634 return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1635 }
process_cpu_timer_create(struct k_itimer * timer)1636 static int process_cpu_timer_create(struct k_itimer *timer)
1637 {
1638 timer->it_clock = PROCESS_CLOCK;
1639 return posix_cpu_timer_create(timer);
1640 }
process_cpu_nsleep(const clockid_t which_clock,int flags,const struct timespec64 * rqtp)1641 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1642 const struct timespec64 *rqtp)
1643 {
1644 return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp);
1645 }
thread_cpu_clock_getres(const clockid_t which_clock,struct timespec64 * tp)1646 static int thread_cpu_clock_getres(const clockid_t which_clock,
1647 struct timespec64 *tp)
1648 {
1649 return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1650 }
thread_cpu_clock_get(const clockid_t which_clock,struct timespec64 * tp)1651 static int thread_cpu_clock_get(const clockid_t which_clock,
1652 struct timespec64 *tp)
1653 {
1654 return posix_cpu_clock_get(THREAD_CLOCK, tp);
1655 }
thread_cpu_timer_create(struct k_itimer * timer)1656 static int thread_cpu_timer_create(struct k_itimer *timer)
1657 {
1658 timer->it_clock = THREAD_CLOCK;
1659 return posix_cpu_timer_create(timer);
1660 }
1661
1662 const struct k_clock clock_posix_cpu = {
1663 .clock_getres = posix_cpu_clock_getres,
1664 .clock_set = posix_cpu_clock_set,
1665 .clock_get_timespec = posix_cpu_clock_get,
1666 .timer_create = posix_cpu_timer_create,
1667 .nsleep = posix_cpu_nsleep,
1668 .timer_set = posix_cpu_timer_set,
1669 .timer_del = posix_cpu_timer_del,
1670 .timer_get = posix_cpu_timer_get,
1671 .timer_rearm = posix_cpu_timer_rearm,
1672 .timer_wait_running = posix_cpu_timer_wait_running,
1673 };
1674
1675 const struct k_clock clock_process = {
1676 .clock_getres = process_cpu_clock_getres,
1677 .clock_get_timespec = process_cpu_clock_get,
1678 .timer_create = process_cpu_timer_create,
1679 .nsleep = process_cpu_nsleep,
1680 };
1681
1682 const struct k_clock clock_thread = {
1683 .clock_getres = thread_cpu_clock_getres,
1684 .clock_get_timespec = thread_cpu_clock_get,
1685 .timer_create = thread_cpu_timer_create,
1686 };
1687