Loading...
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
21static void posix_cpu_timer_rearm(struct k_itimer *timer);
22
23void 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 */
38void 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 */
50static struct task_struct *lookup_task(const pid_t pid, bool thread,
51 bool gettime)
52{
53 struct task_struct *p;
54
55 /*
56 * If the encoded PID is 0, then the timer is targeted at current
57 * or the process to which current belongs.
58 */
59 if (!pid)
60 return thread ? current : current->group_leader;
61
62 p = find_task_by_vpid(pid);
63 if (!p)
64 return p;
65
66 if (thread)
67 return same_thread_group(p, current) ? p : NULL;
68
69 if (gettime) {
70 /*
71 * For clock_gettime(PROCESS) the task does not need to be
72 * the actual group leader. tsk->sighand gives
73 * access to the group's clock.
74 *
75 * Timers need the group leader because they take a
76 * reference on it and store the task pointer until the
77 * timer is destroyed.
78 */
79 return (p == current || thread_group_leader(p)) ? p : NULL;
80 }
81
82 /*
83 * For processes require that p is group leader.
84 */
85 return has_group_leader_pid(p) ? p : NULL;
86}
87
88static struct task_struct *__get_task_for_clock(const clockid_t clock,
89 bool getref, bool gettime)
90{
91 const bool thread = !!CPUCLOCK_PERTHREAD(clock);
92 const pid_t pid = CPUCLOCK_PID(clock);
93 struct task_struct *p;
94
95 if (CPUCLOCK_WHICH(clock) >= CPUCLOCK_MAX)
96 return NULL;
97
98 rcu_read_lock();
99 p = lookup_task(pid, thread, gettime);
100 if (p && getref)
101 get_task_struct(p);
102 rcu_read_unlock();
103 return p;
104}
105
106static inline struct task_struct *get_task_for_clock(const clockid_t clock)
107{
108 return __get_task_for_clock(clock, true, false);
109}
110
111static inline struct task_struct *get_task_for_clock_get(const clockid_t clock)
112{
113 return __get_task_for_clock(clock, true, true);
114}
115
116static inline int validate_clock_permissions(const clockid_t clock)
117{
118 return __get_task_for_clock(clock, false, false) ? 0 : -EINVAL;
119}
120
121/*
122 * Update expiry time from increment, and increase overrun count,
123 * given the current clock sample.
124 */
125static u64 bump_cpu_timer(struct k_itimer *timer, u64 now)
126{
127 u64 delta, incr, expires = timer->it.cpu.node.expires;
128 int i;
129
130 if (!timer->it_interval)
131 return expires;
132
133 if (now < expires)
134 return expires;
135
136 incr = timer->it_interval;
137 delta = now + incr - expires;
138
139 /* Don't use (incr*2 < delta), incr*2 might overflow. */
140 for (i = 0; incr < delta - incr; i++)
141 incr = incr << 1;
142
143 for (; i >= 0; incr >>= 1, i--) {
144 if (delta < incr)
145 continue;
146
147 timer->it.cpu.node.expires += incr;
148 timer->it_overrun += 1LL << i;
149 delta -= incr;
150 }
151 return timer->it.cpu.node.expires;
152}
153
154/* Check whether all cache entries contain U64_MAX, i.e. eternal expiry time */
155static inline bool expiry_cache_is_inactive(const struct posix_cputimers *pct)
156{
157 return !(~pct->bases[CPUCLOCK_PROF].nextevt |
158 ~pct->bases[CPUCLOCK_VIRT].nextevt |
159 ~pct->bases[CPUCLOCK_SCHED].nextevt);
160}
161
162static int
163posix_cpu_clock_getres(const clockid_t which_clock, struct timespec64 *tp)
164{
165 int error = validate_clock_permissions(which_clock);
166
167 if (!error) {
168 tp->tv_sec = 0;
169 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
170 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
171 /*
172 * If sched_clock is using a cycle counter, we
173 * don't have any idea of its true resolution
174 * exported, but it is much more than 1s/HZ.
175 */
176 tp->tv_nsec = 1;
177 }
178 }
179 return error;
180}
181
182static int
183posix_cpu_clock_set(const clockid_t clock, const struct timespec64 *tp)
184{
185 int error = validate_clock_permissions(clock);
186
187 /*
188 * You can never reset a CPU clock, but we check for other errors
189 * in the call before failing with EPERM.
190 */
191 return error ? : -EPERM;
192}
193
194/*
195 * Sample a per-thread clock for the given task. clkid is validated.
196 */
197static u64 cpu_clock_sample(const clockid_t clkid, struct task_struct *p)
198{
199 u64 utime, stime;
200
201 if (clkid == CPUCLOCK_SCHED)
202 return task_sched_runtime(p);
203
204 task_cputime(p, &utime, &stime);
205
206 switch (clkid) {
207 case CPUCLOCK_PROF:
208 return utime + stime;
209 case CPUCLOCK_VIRT:
210 return utime;
211 default:
212 WARN_ON_ONCE(1);
213 }
214 return 0;
215}
216
217static inline void store_samples(u64 *samples, u64 stime, u64 utime, u64 rtime)
218{
219 samples[CPUCLOCK_PROF] = stime + utime;
220 samples[CPUCLOCK_VIRT] = utime;
221 samples[CPUCLOCK_SCHED] = rtime;
222}
223
224static void task_sample_cputime(struct task_struct *p, u64 *samples)
225{
226 u64 stime, utime;
227
228 task_cputime(p, &utime, &stime);
229 store_samples(samples, stime, utime, p->se.sum_exec_runtime);
230}
231
232static void proc_sample_cputime_atomic(struct task_cputime_atomic *at,
233 u64 *samples)
234{
235 u64 stime, utime, rtime;
236
237 utime = atomic64_read(&at->utime);
238 stime = atomic64_read(&at->stime);
239 rtime = atomic64_read(&at->sum_exec_runtime);
240 store_samples(samples, stime, utime, rtime);
241}
242
243/*
244 * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg
245 * to avoid race conditions with concurrent updates to cputime.
246 */
247static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime)
248{
249 u64 curr_cputime;
250retry:
251 curr_cputime = atomic64_read(cputime);
252 if (sum_cputime > curr_cputime) {
253 if (atomic64_cmpxchg(cputime, curr_cputime, sum_cputime) != curr_cputime)
254 goto retry;
255 }
256}
257
258static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic,
259 struct task_cputime *sum)
260{
261 __update_gt_cputime(&cputime_atomic->utime, sum->utime);
262 __update_gt_cputime(&cputime_atomic->stime, sum->stime);
263 __update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime);
264}
265
266/**
267 * thread_group_sample_cputime - Sample cputime for a given task
268 * @tsk: Task for which cputime needs to be started
269 * @samples: Storage for time samples
270 *
271 * Called from sys_getitimer() to calculate the expiry time of an active
272 * timer. That means group cputime accounting is already active. Called
273 * with task sighand lock held.
274 *
275 * Updates @times with an uptodate sample of the thread group cputimes.
276 */
277void thread_group_sample_cputime(struct task_struct *tsk, u64 *samples)
278{
279 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
280 struct posix_cputimers *pct = &tsk->signal->posix_cputimers;
281
282 WARN_ON_ONCE(!pct->timers_active);
283
284 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
285}
286
287/**
288 * thread_group_start_cputime - Start cputime and return a sample
289 * @tsk: Task for which cputime needs to be started
290 * @samples: Storage for time samples
291 *
292 * The thread group cputime accouting is avoided when there are no posix
293 * CPU timers armed. Before starting a timer it's required to check whether
294 * the time accounting is active. If not, a full update of the atomic
295 * accounting store needs to be done and the accounting enabled.
296 *
297 * Updates @times with an uptodate sample of the thread group cputimes.
298 */
299static void thread_group_start_cputime(struct task_struct *tsk, u64 *samples)
300{
301 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
302 struct posix_cputimers *pct = &tsk->signal->posix_cputimers;
303
304 /* Check if cputimer isn't running. This is accessed without locking. */
305 if (!READ_ONCE(pct->timers_active)) {
306 struct task_cputime sum;
307
308 /*
309 * The POSIX timer interface allows for absolute time expiry
310 * values through the TIMER_ABSTIME flag, therefore we have
311 * to synchronize the timer to the clock every time we start it.
312 */
313 thread_group_cputime(tsk, &sum);
314 update_gt_cputime(&cputimer->cputime_atomic, &sum);
315
316 /*
317 * We're setting timers_active without a lock. Ensure this
318 * only gets written to in one operation. We set it after
319 * update_gt_cputime() as a small optimization, but
320 * barriers are not required because update_gt_cputime()
321 * can handle concurrent updates.
322 */
323 WRITE_ONCE(pct->timers_active, true);
324 }
325 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
326}
327
328static void __thread_group_cputime(struct task_struct *tsk, u64 *samples)
329{
330 struct task_cputime ct;
331
332 thread_group_cputime(tsk, &ct);
333 store_samples(samples, ct.stime, ct.utime, ct.sum_exec_runtime);
334}
335
336/*
337 * Sample a process (thread group) clock for the given task clkid. If the
338 * group's cputime accounting is already enabled, read the atomic
339 * store. Otherwise a full update is required. Task's sighand lock must be
340 * held to protect the task traversal on a full update. clkid is already
341 * validated.
342 */
343static u64 cpu_clock_sample_group(const clockid_t clkid, struct task_struct *p,
344 bool start)
345{
346 struct thread_group_cputimer *cputimer = &p->signal->cputimer;
347 struct posix_cputimers *pct = &p->signal->posix_cputimers;
348 u64 samples[CPUCLOCK_MAX];
349
350 if (!READ_ONCE(pct->timers_active)) {
351 if (start)
352 thread_group_start_cputime(p, samples);
353 else
354 __thread_group_cputime(p, samples);
355 } else {
356 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
357 }
358
359 return samples[clkid];
360}
361
362static int posix_cpu_clock_get(const clockid_t clock, struct timespec64 *tp)
363{
364 const clockid_t clkid = CPUCLOCK_WHICH(clock);
365 struct task_struct *tsk;
366 u64 t;
367
368 tsk = get_task_for_clock_get(clock);
369 if (!tsk)
370 return -EINVAL;
371
372 if (CPUCLOCK_PERTHREAD(clock))
373 t = cpu_clock_sample(clkid, tsk);
374 else
375 t = cpu_clock_sample_group(clkid, tsk, false);
376 put_task_struct(tsk);
377
378 *tp = ns_to_timespec64(t);
379 return 0;
380}
381
382/*
383 * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
384 * This is called from sys_timer_create() and do_cpu_nanosleep() with the
385 * new timer already all-zeros initialized.
386 */
387static int posix_cpu_timer_create(struct k_itimer *new_timer)
388{
389 struct task_struct *p = get_task_for_clock(new_timer->it_clock);
390
391 if (!p)
392 return -EINVAL;
393
394 new_timer->kclock = &clock_posix_cpu;
395 timerqueue_init(&new_timer->it.cpu.node);
396 new_timer->it.cpu.task = p;
397 return 0;
398}
399
400/*
401 * Clean up a CPU-clock timer that is about to be destroyed.
402 * This is called from timer deletion with the timer already locked.
403 * If we return TIMER_RETRY, it's necessary to release the timer's lock
404 * and try again. (This happens when the timer is in the middle of firing.)
405 */
406static int posix_cpu_timer_del(struct k_itimer *timer)
407{
408 struct cpu_timer *ctmr = &timer->it.cpu;
409 struct task_struct *p = ctmr->task;
410 struct sighand_struct *sighand;
411 unsigned long flags;
412 int ret = 0;
413
414 if (WARN_ON_ONCE(!p))
415 return -EINVAL;
416
417 /*
418 * Protect against sighand release/switch in exit/exec and process/
419 * thread timer list entry concurrent read/writes.
420 */
421 sighand = lock_task_sighand(p, &flags);
422 if (unlikely(sighand == NULL)) {
423 /*
424 * This raced with the reaping of the task. The exit cleanup
425 * should have removed this timer from the timer queue.
426 */
427 WARN_ON_ONCE(ctmr->head || timerqueue_node_queued(&ctmr->node));
428 } else {
429 if (timer->it.cpu.firing)
430 ret = TIMER_RETRY;
431 else
432 cpu_timer_dequeue(ctmr);
433
434 unlock_task_sighand(p, &flags);
435 }
436
437 if (!ret)
438 put_task_struct(p);
439
440 return ret;
441}
442
443static void cleanup_timerqueue(struct timerqueue_head *head)
444{
445 struct timerqueue_node *node;
446 struct cpu_timer *ctmr;
447
448 while ((node = timerqueue_getnext(head))) {
449 timerqueue_del(head, node);
450 ctmr = container_of(node, struct cpu_timer, node);
451 ctmr->head = NULL;
452 }
453}
454
455/*
456 * Clean out CPU timers which are still armed when a thread exits. The
457 * timers are only removed from the list. No other updates are done. The
458 * corresponding posix timers are still accessible, but cannot be rearmed.
459 *
460 * This must be called with the siglock held.
461 */
462static void cleanup_timers(struct posix_cputimers *pct)
463{
464 cleanup_timerqueue(&pct->bases[CPUCLOCK_PROF].tqhead);
465 cleanup_timerqueue(&pct->bases[CPUCLOCK_VIRT].tqhead);
466 cleanup_timerqueue(&pct->bases[CPUCLOCK_SCHED].tqhead);
467}
468
469/*
470 * These are both called with the siglock held, when the current thread
471 * is being reaped. When the final (leader) thread in the group is reaped,
472 * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
473 */
474void posix_cpu_timers_exit(struct task_struct *tsk)
475{
476 cleanup_timers(&tsk->posix_cputimers);
477}
478void posix_cpu_timers_exit_group(struct task_struct *tsk)
479{
480 cleanup_timers(&tsk->signal->posix_cputimers);
481}
482
483/*
484 * Insert the timer on the appropriate list before any timers that
485 * expire later. This must be called with the sighand lock held.
486 */
487static void arm_timer(struct k_itimer *timer)
488{
489 int clkidx = CPUCLOCK_WHICH(timer->it_clock);
490 struct cpu_timer *ctmr = &timer->it.cpu;
491 u64 newexp = cpu_timer_getexpires(ctmr);
492 struct task_struct *p = ctmr->task;
493 struct posix_cputimer_base *base;
494
495 if (CPUCLOCK_PERTHREAD(timer->it_clock))
496 base = p->posix_cputimers.bases + clkidx;
497 else
498 base = p->signal->posix_cputimers.bases + clkidx;
499
500 if (!cpu_timer_enqueue(&base->tqhead, ctmr))
501 return;
502
503 /*
504 * We are the new earliest-expiring POSIX 1.b timer, hence
505 * need to update expiration cache. Take into account that
506 * for process timers we share expiration cache with itimers
507 * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
508 */
509 if (newexp < base->nextevt)
510 base->nextevt = newexp;
511
512 if (CPUCLOCK_PERTHREAD(timer->it_clock))
513 tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER);
514 else
515 tick_dep_set_signal(p->signal, TICK_DEP_BIT_POSIX_TIMER);
516}
517
518/*
519 * The timer is locked, fire it and arrange for its reload.
520 */
521static void cpu_timer_fire(struct k_itimer *timer)
522{
523 struct cpu_timer *ctmr = &timer->it.cpu;
524
525 if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
526 /*
527 * User don't want any signal.
528 */
529 cpu_timer_setexpires(ctmr, 0);
530 } else if (unlikely(timer->sigq == NULL)) {
531 /*
532 * This a special case for clock_nanosleep,
533 * not a normal timer from sys_timer_create.
534 */
535 wake_up_process(timer->it_process);
536 cpu_timer_setexpires(ctmr, 0);
537 } else if (!timer->it_interval) {
538 /*
539 * One-shot timer. Clear it as soon as it's fired.
540 */
541 posix_timer_event(timer, 0);
542 cpu_timer_setexpires(ctmr, 0);
543 } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
544 /*
545 * The signal did not get queued because the signal
546 * was ignored, so we won't get any callback to
547 * reload the timer. But we need to keep it
548 * ticking in case the signal is deliverable next time.
549 */
550 posix_cpu_timer_rearm(timer);
551 ++timer->it_requeue_pending;
552 }
553}
554
555/*
556 * Guts of sys_timer_settime for CPU timers.
557 * This is called with the timer locked and interrupts disabled.
558 * If we return TIMER_RETRY, it's necessary to release the timer's lock
559 * and try again. (This happens when the timer is in the middle of firing.)
560 */
561static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
562 struct itimerspec64 *new, struct itimerspec64 *old)
563{
564 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
565 u64 old_expires, new_expires, old_incr, val;
566 struct cpu_timer *ctmr = &timer->it.cpu;
567 struct task_struct *p = ctmr->task;
568 struct sighand_struct *sighand;
569 unsigned long flags;
570 int ret = 0;
571
572 if (WARN_ON_ONCE(!p))
573 return -EINVAL;
574
575 /*
576 * Use the to_ktime conversion because that clamps the maximum
577 * value to KTIME_MAX and avoid multiplication overflows.
578 */
579 new_expires = ktime_to_ns(timespec64_to_ktime(new->it_value));
580
581 /*
582 * Protect against sighand release/switch in exit/exec and p->cpu_timers
583 * and p->signal->cpu_timers read/write in arm_timer()
584 */
585 sighand = lock_task_sighand(p, &flags);
586 /*
587 * If p has just been reaped, we can no
588 * longer get any information about it at all.
589 */
590 if (unlikely(sighand == NULL))
591 return -ESRCH;
592
593 /*
594 * Disarm any old timer after extracting its expiry time.
595 */
596 old_incr = timer->it_interval;
597 old_expires = cpu_timer_getexpires(ctmr);
598
599 if (unlikely(timer->it.cpu.firing)) {
600 timer->it.cpu.firing = -1;
601 ret = TIMER_RETRY;
602 } else {
603 cpu_timer_dequeue(ctmr);
604 }
605
606 /*
607 * We need to sample the current value to convert the new
608 * value from to relative and absolute, and to convert the
609 * old value from absolute to relative. To set a process
610 * timer, we need a sample to balance the thread expiry
611 * times (in arm_timer). With an absolute time, we must
612 * check if it's already passed. In short, we need a sample.
613 */
614 if (CPUCLOCK_PERTHREAD(timer->it_clock))
615 val = cpu_clock_sample(clkid, p);
616 else
617 val = cpu_clock_sample_group(clkid, p, true);
618
619 if (old) {
620 if (old_expires == 0) {
621 old->it_value.tv_sec = 0;
622 old->it_value.tv_nsec = 0;
623 } else {
624 /*
625 * Update the timer in case it has overrun already.
626 * If it has, we'll report it as having overrun and
627 * with the next reloaded timer already ticking,
628 * though we are swallowing that pending
629 * notification here to install the new setting.
630 */
631 u64 exp = bump_cpu_timer(timer, val);
632
633 if (val < exp) {
634 old_expires = exp - val;
635 old->it_value = ns_to_timespec64(old_expires);
636 } else {
637 old->it_value.tv_nsec = 1;
638 old->it_value.tv_sec = 0;
639 }
640 }
641 }
642
643 if (unlikely(ret)) {
644 /*
645 * We are colliding with the timer actually firing.
646 * Punt after filling in the timer's old value, and
647 * disable this firing since we are already reporting
648 * it as an overrun (thanks to bump_cpu_timer above).
649 */
650 unlock_task_sighand(p, &flags);
651 goto out;
652 }
653
654 if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
655 new_expires += val;
656 }
657
658 /*
659 * Install the new expiry time (or zero).
660 * For a timer with no notification action, we don't actually
661 * arm the timer (we'll just fake it for timer_gettime).
662 */
663 cpu_timer_setexpires(ctmr, new_expires);
664 if (new_expires != 0 && val < new_expires) {
665 arm_timer(timer);
666 }
667
668 unlock_task_sighand(p, &flags);
669 /*
670 * Install the new reload setting, and
671 * set up the signal and overrun bookkeeping.
672 */
673 timer->it_interval = timespec64_to_ktime(new->it_interval);
674
675 /*
676 * This acts as a modification timestamp for the timer,
677 * so any automatic reload attempt will punt on seeing
678 * that we have reset the timer manually.
679 */
680 timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
681 ~REQUEUE_PENDING;
682 timer->it_overrun_last = 0;
683 timer->it_overrun = -1;
684
685 if (new_expires != 0 && !(val < new_expires)) {
686 /*
687 * The designated time already passed, so we notify
688 * immediately, even if the thread never runs to
689 * accumulate more time on this clock.
690 */
691 cpu_timer_fire(timer);
692 }
693
694 ret = 0;
695 out:
696 if (old)
697 old->it_interval = ns_to_timespec64(old_incr);
698
699 return ret;
700}
701
702static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp)
703{
704 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
705 struct cpu_timer *ctmr = &timer->it.cpu;
706 u64 now, expires = cpu_timer_getexpires(ctmr);
707 struct task_struct *p = ctmr->task;
708
709 if (WARN_ON_ONCE(!p))
710 return;
711
712 /*
713 * Easy part: convert the reload time.
714 */
715 itp->it_interval = ktime_to_timespec64(timer->it_interval);
716
717 if (!expires)
718 return;
719
720 /*
721 * Sample the clock to take the difference with the expiry time.
722 */
723 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
724 now = cpu_clock_sample(clkid, p);
725 } else {
726 struct sighand_struct *sighand;
727 unsigned long flags;
728
729 /*
730 * Protect against sighand release/switch in exit/exec and
731 * also make timer sampling safe if it ends up calling
732 * thread_group_cputime().
733 */
734 sighand = lock_task_sighand(p, &flags);
735 if (unlikely(sighand == NULL)) {
736 /*
737 * The process has been reaped.
738 * We can't even collect a sample any more.
739 * Disarm the timer, nothing else to do.
740 */
741 cpu_timer_setexpires(ctmr, 0);
742 return;
743 } else {
744 now = cpu_clock_sample_group(clkid, p, false);
745 unlock_task_sighand(p, &flags);
746 }
747 }
748
749 if (now < expires) {
750 itp->it_value = ns_to_timespec64(expires - now);
751 } else {
752 /*
753 * The timer should have expired already, but the firing
754 * hasn't taken place yet. Say it's just about to expire.
755 */
756 itp->it_value.tv_nsec = 1;
757 itp->it_value.tv_sec = 0;
758 }
759}
760
761#define MAX_COLLECTED 20
762
763static u64 collect_timerqueue(struct timerqueue_head *head,
764 struct list_head *firing, u64 now)
765{
766 struct timerqueue_node *next;
767 int i = 0;
768
769 while ((next = timerqueue_getnext(head))) {
770 struct cpu_timer *ctmr;
771 u64 expires;
772
773 ctmr = container_of(next, struct cpu_timer, node);
774 expires = cpu_timer_getexpires(ctmr);
775 /* Limit the number of timers to expire at once */
776 if (++i == MAX_COLLECTED || now < expires)
777 return expires;
778
779 ctmr->firing = 1;
780 cpu_timer_dequeue(ctmr);
781 list_add_tail(&ctmr->elist, firing);
782 }
783
784 return U64_MAX;
785}
786
787static void collect_posix_cputimers(struct posix_cputimers *pct, u64 *samples,
788 struct list_head *firing)
789{
790 struct posix_cputimer_base *base = pct->bases;
791 int i;
792
793 for (i = 0; i < CPUCLOCK_MAX; i++, base++) {
794 base->nextevt = collect_timerqueue(&base->tqhead, firing,
795 samples[i]);
796 }
797}
798
799static inline void check_dl_overrun(struct task_struct *tsk)
800{
801 if (tsk->dl.dl_overrun) {
802 tsk->dl.dl_overrun = 0;
803 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
804 }
805}
806
807static bool check_rlimit(u64 time, u64 limit, int signo, bool rt, bool hard)
808{
809 if (time < limit)
810 return false;
811
812 if (print_fatal_signals) {
813 pr_info("%s Watchdog Timeout (%s): %s[%d]\n",
814 rt ? "RT" : "CPU", hard ? "hard" : "soft",
815 current->comm, task_pid_nr(current));
816 }
817 __group_send_sig_info(signo, SEND_SIG_PRIV, current);
818 return true;
819}
820
821/*
822 * Check for any per-thread CPU timers that have fired and move them off
823 * the tsk->cpu_timers[N] list onto the firing list. Here we update the
824 * tsk->it_*_expires values to reflect the remaining thread CPU timers.
825 */
826static void check_thread_timers(struct task_struct *tsk,
827 struct list_head *firing)
828{
829 struct posix_cputimers *pct = &tsk->posix_cputimers;
830 u64 samples[CPUCLOCK_MAX];
831 unsigned long soft;
832
833 if (dl_task(tsk))
834 check_dl_overrun(tsk);
835
836 if (expiry_cache_is_inactive(pct))
837 return;
838
839 task_sample_cputime(tsk, samples);
840 collect_posix_cputimers(pct, samples, firing);
841
842 /*
843 * Check for the special case thread timers.
844 */
845 soft = task_rlimit(tsk, RLIMIT_RTTIME);
846 if (soft != RLIM_INFINITY) {
847 /* Task RT timeout is accounted in jiffies. RTTIME is usec */
848 unsigned long rttime = tsk->rt.timeout * (USEC_PER_SEC / HZ);
849 unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME);
850
851 /* At the hard limit, send SIGKILL. No further action. */
852 if (hard != RLIM_INFINITY &&
853 check_rlimit(rttime, hard, SIGKILL, true, true))
854 return;
855
856 /* At the soft limit, send a SIGXCPU every second */
857 if (check_rlimit(rttime, soft, SIGXCPU, true, false)) {
858 soft += USEC_PER_SEC;
859 tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur = soft;
860 }
861 }
862
863 if (expiry_cache_is_inactive(pct))
864 tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER);
865}
866
867static inline void stop_process_timers(struct signal_struct *sig)
868{
869 struct posix_cputimers *pct = &sig->posix_cputimers;
870
871 /* Turn off the active flag. This is done without locking. */
872 WRITE_ONCE(pct->timers_active, false);
873 tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER);
874}
875
876static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
877 u64 *expires, u64 cur_time, int signo)
878{
879 if (!it->expires)
880 return;
881
882 if (cur_time >= it->expires) {
883 if (it->incr)
884 it->expires += it->incr;
885 else
886 it->expires = 0;
887
888 trace_itimer_expire(signo == SIGPROF ?
889 ITIMER_PROF : ITIMER_VIRTUAL,
890 task_tgid(tsk), cur_time);
891 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
892 }
893
894 if (it->expires && it->expires < *expires)
895 *expires = it->expires;
896}
897
898/*
899 * Check for any per-thread CPU timers that have fired and move them
900 * off the tsk->*_timers list onto the firing list. Per-thread timers
901 * have already been taken off.
902 */
903static void check_process_timers(struct task_struct *tsk,
904 struct list_head *firing)
905{
906 struct signal_struct *const sig = tsk->signal;
907 struct posix_cputimers *pct = &sig->posix_cputimers;
908 u64 samples[CPUCLOCK_MAX];
909 unsigned long soft;
910
911 /*
912 * If there are no active process wide timers (POSIX 1.b, itimers,
913 * RLIMIT_CPU) nothing to check. Also skip the process wide timer
914 * processing when there is already another task handling them.
915 */
916 if (!READ_ONCE(pct->timers_active) || pct->expiry_active)
917 return;
918
919 /*
920 * Signify that a thread is checking for process timers.
921 * Write access to this field is protected by the sighand lock.
922 */
923 pct->expiry_active = true;
924
925 /*
926 * Collect the current process totals. Group accounting is active
927 * so the sample can be taken directly.
928 */
929 proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic, samples);
930 collect_posix_cputimers(pct, samples, firing);
931
932 /*
933 * Check for the special case process timers.
934 */
935 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF],
936 &pct->bases[CPUCLOCK_PROF].nextevt,
937 samples[CPUCLOCK_PROF], SIGPROF);
938 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT],
939 &pct->bases[CPUCLOCK_VIRT].nextevt,
940 samples[CPUCLOCK_VIRT], SIGVTALRM);
941
942 soft = task_rlimit(tsk, RLIMIT_CPU);
943 if (soft != RLIM_INFINITY) {
944 /* RLIMIT_CPU is in seconds. Samples are nanoseconds */
945 unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU);
946 u64 ptime = samples[CPUCLOCK_PROF];
947 u64 softns = (u64)soft * NSEC_PER_SEC;
948 u64 hardns = (u64)hard * NSEC_PER_SEC;
949
950 /* At the hard limit, send SIGKILL. No further action. */
951 if (hard != RLIM_INFINITY &&
952 check_rlimit(ptime, hardns, SIGKILL, false, true))
953 return;
954
955 /* At the soft limit, send a SIGXCPU every second */
956 if (check_rlimit(ptime, softns, SIGXCPU, false, false)) {
957 sig->rlim[RLIMIT_CPU].rlim_cur = soft + 1;
958 softns += NSEC_PER_SEC;
959 }
960
961 /* Update the expiry cache */
962 if (softns < pct->bases[CPUCLOCK_PROF].nextevt)
963 pct->bases[CPUCLOCK_PROF].nextevt = softns;
964 }
965
966 if (expiry_cache_is_inactive(pct))
967 stop_process_timers(sig);
968
969 pct->expiry_active = false;
970}
971
972/*
973 * This is called from the signal code (via posixtimer_rearm)
974 * when the last timer signal was delivered and we have to reload the timer.
975 */
976static void posix_cpu_timer_rearm(struct k_itimer *timer)
977{
978 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
979 struct cpu_timer *ctmr = &timer->it.cpu;
980 struct task_struct *p = ctmr->task;
981 struct sighand_struct *sighand;
982 unsigned long flags;
983 u64 now;
984
985 if (WARN_ON_ONCE(!p))
986 return;
987
988 /*
989 * Fetch the current sample and update the timer's expiry time.
990 */
991 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
992 now = cpu_clock_sample(clkid, p);
993 bump_cpu_timer(timer, now);
994 if (unlikely(p->exit_state))
995 return;
996
997 /* Protect timer list r/w in arm_timer() */
998 sighand = lock_task_sighand(p, &flags);
999 if (!sighand)
1000 return;
1001 } else {
1002 /*
1003 * Protect arm_timer() and timer sampling in case of call to
1004 * thread_group_cputime().
1005 */
1006 sighand = lock_task_sighand(p, &flags);
1007 if (unlikely(sighand == NULL)) {
1008 /*
1009 * The process has been reaped.
1010 * We can't even collect a sample any more.
1011 */
1012 cpu_timer_setexpires(ctmr, 0);
1013 return;
1014 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1015 /* If the process is dying, no need to rearm */
1016 goto unlock;
1017 }
1018 now = cpu_clock_sample_group(clkid, p, true);
1019 bump_cpu_timer(timer, now);
1020 /* Leave the sighand locked for the call below. */
1021 }
1022
1023 /*
1024 * Now re-arm for the new expiry time.
1025 */
1026 arm_timer(timer);
1027unlock:
1028 unlock_task_sighand(p, &flags);
1029}
1030
1031/**
1032 * task_cputimers_expired - Check whether posix CPU timers are expired
1033 *
1034 * @samples: Array of current samples for the CPUCLOCK clocks
1035 * @pct: Pointer to a posix_cputimers container
1036 *
1037 * Returns true if any member of @samples is greater than the corresponding
1038 * member of @pct->bases[CLK].nextevt. False otherwise
1039 */
1040static inline bool
1041task_cputimers_expired(const u64 *samples, struct posix_cputimers *pct)
1042{
1043 int i;
1044
1045 for (i = 0; i < CPUCLOCK_MAX; i++) {
1046 if (samples[i] >= pct->bases[i].nextevt)
1047 return true;
1048 }
1049 return false;
1050}
1051
1052/**
1053 * fastpath_timer_check - POSIX CPU timers fast path.
1054 *
1055 * @tsk: The task (thread) being checked.
1056 *
1057 * Check the task and thread group timers. If both are zero (there are no
1058 * timers set) return false. Otherwise snapshot the task and thread group
1059 * timers and compare them with the corresponding expiration times. Return
1060 * true if a timer has expired, else return false.
1061 */
1062static inline bool fastpath_timer_check(struct task_struct *tsk)
1063{
1064 struct posix_cputimers *pct = &tsk->posix_cputimers;
1065 struct signal_struct *sig;
1066
1067 if (!expiry_cache_is_inactive(pct)) {
1068 u64 samples[CPUCLOCK_MAX];
1069
1070 task_sample_cputime(tsk, samples);
1071 if (task_cputimers_expired(samples, pct))
1072 return true;
1073 }
1074
1075 sig = tsk->signal;
1076 pct = &sig->posix_cputimers;
1077 /*
1078 * Check if thread group timers expired when timers are active and
1079 * no other thread in the group is already handling expiry for
1080 * thread group cputimers. These fields are read without the
1081 * sighand lock. However, this is fine because this is meant to be
1082 * a fastpath heuristic to determine whether we should try to
1083 * acquire the sighand lock to handle timer expiry.
1084 *
1085 * In the worst case scenario, if concurrently timers_active is set
1086 * or expiry_active is cleared, but the current thread doesn't see
1087 * the change yet, the timer checks are delayed until the next
1088 * thread in the group gets a scheduler interrupt to handle the
1089 * timer. This isn't an issue in practice because these types of
1090 * delays with signals actually getting sent are expected.
1091 */
1092 if (READ_ONCE(pct->timers_active) && !READ_ONCE(pct->expiry_active)) {
1093 u64 samples[CPUCLOCK_MAX];
1094
1095 proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic,
1096 samples);
1097
1098 if (task_cputimers_expired(samples, pct))
1099 return true;
1100 }
1101
1102 if (dl_task(tsk) && tsk->dl.dl_overrun)
1103 return true;
1104
1105 return false;
1106}
1107
1108/*
1109 * This is called from the timer interrupt handler. The irq handler has
1110 * already updated our counts. We need to check if any timers fire now.
1111 * Interrupts are disabled.
1112 */
1113void run_posix_cpu_timers(void)
1114{
1115 struct task_struct *tsk = current;
1116 struct k_itimer *timer, *next;
1117 unsigned long flags;
1118 LIST_HEAD(firing);
1119
1120 lockdep_assert_irqs_disabled();
1121
1122 /*
1123 * The fast path checks that there are no expired thread or thread
1124 * group timers. If that's so, just return.
1125 */
1126 if (!fastpath_timer_check(tsk))
1127 return;
1128
1129 if (!lock_task_sighand(tsk, &flags))
1130 return;
1131 /*
1132 * Here we take off tsk->signal->cpu_timers[N] and
1133 * tsk->cpu_timers[N] all the timers that are firing, and
1134 * put them on the firing list.
1135 */
1136 check_thread_timers(tsk, &firing);
1137
1138 check_process_timers(tsk, &firing);
1139
1140 /*
1141 * We must release these locks before taking any timer's lock.
1142 * There is a potential race with timer deletion here, as the
1143 * siglock now protects our private firing list. We have set
1144 * the firing flag in each timer, so that a deletion attempt
1145 * that gets the timer lock before we do will give it up and
1146 * spin until we've taken care of that timer below.
1147 */
1148 unlock_task_sighand(tsk, &flags);
1149
1150 /*
1151 * Now that all the timers on our list have the firing flag,
1152 * no one will touch their list entries but us. We'll take
1153 * each timer's lock before clearing its firing flag, so no
1154 * timer call will interfere.
1155 */
1156 list_for_each_entry_safe(timer, next, &firing, it.cpu.elist) {
1157 int cpu_firing;
1158
1159 spin_lock(&timer->it_lock);
1160 list_del_init(&timer->it.cpu.elist);
1161 cpu_firing = timer->it.cpu.firing;
1162 timer->it.cpu.firing = 0;
1163 /*
1164 * The firing flag is -1 if we collided with a reset
1165 * of the timer, which already reported this
1166 * almost-firing as an overrun. So don't generate an event.
1167 */
1168 if (likely(cpu_firing >= 0))
1169 cpu_timer_fire(timer);
1170 spin_unlock(&timer->it_lock);
1171 }
1172}
1173
1174/*
1175 * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1176 * The tsk->sighand->siglock must be held by the caller.
1177 */
1178void set_process_cpu_timer(struct task_struct *tsk, unsigned int clkid,
1179 u64 *newval, u64 *oldval)
1180{
1181 u64 now, *nextevt;
1182
1183 if (WARN_ON_ONCE(clkid >= CPUCLOCK_SCHED))
1184 return;
1185
1186 nextevt = &tsk->signal->posix_cputimers.bases[clkid].nextevt;
1187 now = cpu_clock_sample_group(clkid, tsk, true);
1188
1189 if (oldval) {
1190 /*
1191 * We are setting itimer. The *oldval is absolute and we update
1192 * it to be relative, *newval argument is relative and we update
1193 * it to be absolute.
1194 */
1195 if (*oldval) {
1196 if (*oldval <= now) {
1197 /* Just about to fire. */
1198 *oldval = TICK_NSEC;
1199 } else {
1200 *oldval -= now;
1201 }
1202 }
1203
1204 if (!*newval)
1205 return;
1206 *newval += now;
1207 }
1208
1209 /*
1210 * Update expiration cache if this is the earliest timer. CPUCLOCK_PROF
1211 * expiry cache is also used by RLIMIT_CPU!.
1212 */
1213 if (*newval < *nextevt)
1214 *nextevt = *newval;
1215
1216 tick_dep_set_signal(tsk->signal, TICK_DEP_BIT_POSIX_TIMER);
1217}
1218
1219static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1220 const struct timespec64 *rqtp)
1221{
1222 struct itimerspec64 it;
1223 struct k_itimer timer;
1224 u64 expires;
1225 int error;
1226
1227 /*
1228 * Set up a temporary timer and then wait for it to go off.
1229 */
1230 memset(&timer, 0, sizeof timer);
1231 spin_lock_init(&timer.it_lock);
1232 timer.it_clock = which_clock;
1233 timer.it_overrun = -1;
1234 error = posix_cpu_timer_create(&timer);
1235 timer.it_process = current;
1236
1237 if (!error) {
1238 static struct itimerspec64 zero_it;
1239 struct restart_block *restart;
1240
1241 memset(&it, 0, sizeof(it));
1242 it.it_value = *rqtp;
1243
1244 spin_lock_irq(&timer.it_lock);
1245 error = posix_cpu_timer_set(&timer, flags, &it, NULL);
1246 if (error) {
1247 spin_unlock_irq(&timer.it_lock);
1248 return error;
1249 }
1250
1251 while (!signal_pending(current)) {
1252 if (!cpu_timer_getexpires(&timer.it.cpu)) {
1253 /*
1254 * Our timer fired and was reset, below
1255 * deletion can not fail.
1256 */
1257 posix_cpu_timer_del(&timer);
1258 spin_unlock_irq(&timer.it_lock);
1259 return 0;
1260 }
1261
1262 /*
1263 * Block until cpu_timer_fire (or a signal) wakes us.
1264 */
1265 __set_current_state(TASK_INTERRUPTIBLE);
1266 spin_unlock_irq(&timer.it_lock);
1267 schedule();
1268 spin_lock_irq(&timer.it_lock);
1269 }
1270
1271 /*
1272 * We were interrupted by a signal.
1273 */
1274 expires = cpu_timer_getexpires(&timer.it.cpu);
1275 error = posix_cpu_timer_set(&timer, 0, &zero_it, &it);
1276 if (!error) {
1277 /*
1278 * Timer is now unarmed, deletion can not fail.
1279 */
1280 posix_cpu_timer_del(&timer);
1281 }
1282 spin_unlock_irq(&timer.it_lock);
1283
1284 while (error == TIMER_RETRY) {
1285 /*
1286 * We need to handle case when timer was or is in the
1287 * middle of firing. In other cases we already freed
1288 * resources.
1289 */
1290 spin_lock_irq(&timer.it_lock);
1291 error = posix_cpu_timer_del(&timer);
1292 spin_unlock_irq(&timer.it_lock);
1293 }
1294
1295 if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) {
1296 /*
1297 * It actually did fire already.
1298 */
1299 return 0;
1300 }
1301
1302 error = -ERESTART_RESTARTBLOCK;
1303 /*
1304 * Report back to the user the time still remaining.
1305 */
1306 restart = ¤t->restart_block;
1307 restart->nanosleep.expires = expires;
1308 if (restart->nanosleep.type != TT_NONE)
1309 error = nanosleep_copyout(restart, &it.it_value);
1310 }
1311
1312 return error;
1313}
1314
1315static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1316
1317static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1318 const struct timespec64 *rqtp)
1319{
1320 struct restart_block *restart_block = ¤t->restart_block;
1321 int error;
1322
1323 /*
1324 * Diagnose required errors first.
1325 */
1326 if (CPUCLOCK_PERTHREAD(which_clock) &&
1327 (CPUCLOCK_PID(which_clock) == 0 ||
1328 CPUCLOCK_PID(which_clock) == task_pid_vnr(current)))
1329 return -EINVAL;
1330
1331 error = do_cpu_nanosleep(which_clock, flags, rqtp);
1332
1333 if (error == -ERESTART_RESTARTBLOCK) {
1334
1335 if (flags & TIMER_ABSTIME)
1336 return -ERESTARTNOHAND;
1337
1338 restart_block->fn = posix_cpu_nsleep_restart;
1339 restart_block->nanosleep.clockid = which_clock;
1340 }
1341 return error;
1342}
1343
1344static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1345{
1346 clockid_t which_clock = restart_block->nanosleep.clockid;
1347 struct timespec64 t;
1348
1349 t = ns_to_timespec64(restart_block->nanosleep.expires);
1350
1351 return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t);
1352}
1353
1354#define PROCESS_CLOCK make_process_cpuclock(0, CPUCLOCK_SCHED)
1355#define THREAD_CLOCK make_thread_cpuclock(0, CPUCLOCK_SCHED)
1356
1357static int process_cpu_clock_getres(const clockid_t which_clock,
1358 struct timespec64 *tp)
1359{
1360 return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1361}
1362static int process_cpu_clock_get(const clockid_t which_clock,
1363 struct timespec64 *tp)
1364{
1365 return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1366}
1367static int process_cpu_timer_create(struct k_itimer *timer)
1368{
1369 timer->it_clock = PROCESS_CLOCK;
1370 return posix_cpu_timer_create(timer);
1371}
1372static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1373 const struct timespec64 *rqtp)
1374{
1375 return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp);
1376}
1377static int thread_cpu_clock_getres(const clockid_t which_clock,
1378 struct timespec64 *tp)
1379{
1380 return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1381}
1382static int thread_cpu_clock_get(const clockid_t which_clock,
1383 struct timespec64 *tp)
1384{
1385 return posix_cpu_clock_get(THREAD_CLOCK, tp);
1386}
1387static int thread_cpu_timer_create(struct k_itimer *timer)
1388{
1389 timer->it_clock = THREAD_CLOCK;
1390 return posix_cpu_timer_create(timer);
1391}
1392
1393const struct k_clock clock_posix_cpu = {
1394 .clock_getres = posix_cpu_clock_getres,
1395 .clock_set = posix_cpu_clock_set,
1396 .clock_get = posix_cpu_clock_get,
1397 .timer_create = posix_cpu_timer_create,
1398 .nsleep = posix_cpu_nsleep,
1399 .timer_set = posix_cpu_timer_set,
1400 .timer_del = posix_cpu_timer_del,
1401 .timer_get = posix_cpu_timer_get,
1402 .timer_rearm = posix_cpu_timer_rearm,
1403};
1404
1405const struct k_clock clock_process = {
1406 .clock_getres = process_cpu_clock_getres,
1407 .clock_get = process_cpu_clock_get,
1408 .timer_create = process_cpu_timer_create,
1409 .nsleep = process_cpu_nsleep,
1410};
1411
1412const struct k_clock clock_thread = {
1413 .clock_getres = thread_cpu_clock_getres,
1414 .clock_get = thread_cpu_clock_get,
1415 .timer_create = thread_cpu_timer_create,
1416};
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
21static void posix_cpu_timer_rearm(struct k_itimer *timer);
22
23/*
24 * Called after updating RLIMIT_CPU to run cpu timer and update
25 * tsk->signal->cputime_expires expiration cache if necessary. Needs
26 * siglock protection since other code may update expiration cache as
27 * well.
28 */
29void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
30{
31 u64 nsecs = rlim_new * NSEC_PER_SEC;
32
33 spin_lock_irq(&task->sighand->siglock);
34 set_process_cpu_timer(task, CPUCLOCK_PROF, &nsecs, NULL);
35 spin_unlock_irq(&task->sighand->siglock);
36}
37
38static int check_clock(const clockid_t which_clock)
39{
40 int error = 0;
41 struct task_struct *p;
42 const pid_t pid = CPUCLOCK_PID(which_clock);
43
44 if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
45 return -EINVAL;
46
47 if (pid == 0)
48 return 0;
49
50 rcu_read_lock();
51 p = find_task_by_vpid(pid);
52 if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
53 same_thread_group(p, current) : has_group_leader_pid(p))) {
54 error = -EINVAL;
55 }
56 rcu_read_unlock();
57
58 return error;
59}
60
61/*
62 * Update expiry time from increment, and increase overrun count,
63 * given the current clock sample.
64 */
65static void bump_cpu_timer(struct k_itimer *timer, u64 now)
66{
67 int i;
68 u64 delta, incr;
69
70 if (timer->it.cpu.incr == 0)
71 return;
72
73 if (now < timer->it.cpu.expires)
74 return;
75
76 incr = timer->it.cpu.incr;
77 delta = now + incr - timer->it.cpu.expires;
78
79 /* Don't use (incr*2 < delta), incr*2 might overflow. */
80 for (i = 0; incr < delta - incr; i++)
81 incr = incr << 1;
82
83 for (; i >= 0; incr >>= 1, i--) {
84 if (delta < incr)
85 continue;
86
87 timer->it.cpu.expires += incr;
88 timer->it_overrun += 1 << i;
89 delta -= incr;
90 }
91}
92
93/**
94 * task_cputime_zero - Check a task_cputime struct for all zero fields.
95 *
96 * @cputime: The struct to compare.
97 *
98 * Checks @cputime to see if all fields are zero. Returns true if all fields
99 * are zero, false if any field is nonzero.
100 */
101static inline int task_cputime_zero(const struct task_cputime *cputime)
102{
103 if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime)
104 return 1;
105 return 0;
106}
107
108static inline u64 prof_ticks(struct task_struct *p)
109{
110 u64 utime, stime;
111
112 task_cputime(p, &utime, &stime);
113
114 return utime + stime;
115}
116static inline u64 virt_ticks(struct task_struct *p)
117{
118 u64 utime, stime;
119
120 task_cputime(p, &utime, &stime);
121
122 return utime;
123}
124
125static int
126posix_cpu_clock_getres(const clockid_t which_clock, struct timespec64 *tp)
127{
128 int error = check_clock(which_clock);
129 if (!error) {
130 tp->tv_sec = 0;
131 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
132 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
133 /*
134 * If sched_clock is using a cycle counter, we
135 * don't have any idea of its true resolution
136 * exported, but it is much more than 1s/HZ.
137 */
138 tp->tv_nsec = 1;
139 }
140 }
141 return error;
142}
143
144static int
145posix_cpu_clock_set(const clockid_t which_clock, const struct timespec64 *tp)
146{
147 /*
148 * You can never reset a CPU clock, but we check for other errors
149 * in the call before failing with EPERM.
150 */
151 int error = check_clock(which_clock);
152 if (error == 0) {
153 error = -EPERM;
154 }
155 return error;
156}
157
158
159/*
160 * Sample a per-thread clock for the given task.
161 */
162static int cpu_clock_sample(const clockid_t which_clock,
163 struct task_struct *p, u64 *sample)
164{
165 switch (CPUCLOCK_WHICH(which_clock)) {
166 default:
167 return -EINVAL;
168 case CPUCLOCK_PROF:
169 *sample = prof_ticks(p);
170 break;
171 case CPUCLOCK_VIRT:
172 *sample = virt_ticks(p);
173 break;
174 case CPUCLOCK_SCHED:
175 *sample = task_sched_runtime(p);
176 break;
177 }
178 return 0;
179}
180
181/*
182 * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg
183 * to avoid race conditions with concurrent updates to cputime.
184 */
185static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime)
186{
187 u64 curr_cputime;
188retry:
189 curr_cputime = atomic64_read(cputime);
190 if (sum_cputime > curr_cputime) {
191 if (atomic64_cmpxchg(cputime, curr_cputime, sum_cputime) != curr_cputime)
192 goto retry;
193 }
194}
195
196static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic, struct task_cputime *sum)
197{
198 __update_gt_cputime(&cputime_atomic->utime, sum->utime);
199 __update_gt_cputime(&cputime_atomic->stime, sum->stime);
200 __update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime);
201}
202
203/* Sample task_cputime_atomic values in "atomic_timers", store results in "times". */
204static inline void sample_cputime_atomic(struct task_cputime *times,
205 struct task_cputime_atomic *atomic_times)
206{
207 times->utime = atomic64_read(&atomic_times->utime);
208 times->stime = atomic64_read(&atomic_times->stime);
209 times->sum_exec_runtime = atomic64_read(&atomic_times->sum_exec_runtime);
210}
211
212void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
213{
214 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
215 struct task_cputime sum;
216
217 /* Check if cputimer isn't running. This is accessed without locking. */
218 if (!READ_ONCE(cputimer->running)) {
219 /*
220 * The POSIX timer interface allows for absolute time expiry
221 * values through the TIMER_ABSTIME flag, therefore we have
222 * to synchronize the timer to the clock every time we start it.
223 */
224 thread_group_cputime(tsk, &sum);
225 update_gt_cputime(&cputimer->cputime_atomic, &sum);
226
227 /*
228 * We're setting cputimer->running without a lock. Ensure
229 * this only gets written to in one operation. We set
230 * running after update_gt_cputime() as a small optimization,
231 * but barriers are not required because update_gt_cputime()
232 * can handle concurrent updates.
233 */
234 WRITE_ONCE(cputimer->running, true);
235 }
236 sample_cputime_atomic(times, &cputimer->cputime_atomic);
237}
238
239/*
240 * Sample a process (thread group) clock for the given group_leader task.
241 * Must be called with task sighand lock held for safe while_each_thread()
242 * traversal.
243 */
244static int cpu_clock_sample_group(const clockid_t which_clock,
245 struct task_struct *p,
246 u64 *sample)
247{
248 struct task_cputime cputime;
249
250 switch (CPUCLOCK_WHICH(which_clock)) {
251 default:
252 return -EINVAL;
253 case CPUCLOCK_PROF:
254 thread_group_cputime(p, &cputime);
255 *sample = cputime.utime + cputime.stime;
256 break;
257 case CPUCLOCK_VIRT:
258 thread_group_cputime(p, &cputime);
259 *sample = cputime.utime;
260 break;
261 case CPUCLOCK_SCHED:
262 thread_group_cputime(p, &cputime);
263 *sample = cputime.sum_exec_runtime;
264 break;
265 }
266 return 0;
267}
268
269static int posix_cpu_clock_get_task(struct task_struct *tsk,
270 const clockid_t which_clock,
271 struct timespec64 *tp)
272{
273 int err = -EINVAL;
274 u64 rtn;
275
276 if (CPUCLOCK_PERTHREAD(which_clock)) {
277 if (same_thread_group(tsk, current))
278 err = cpu_clock_sample(which_clock, tsk, &rtn);
279 } else {
280 if (tsk == current || thread_group_leader(tsk))
281 err = cpu_clock_sample_group(which_clock, tsk, &rtn);
282 }
283
284 if (!err)
285 *tp = ns_to_timespec64(rtn);
286
287 return err;
288}
289
290
291static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec64 *tp)
292{
293 const pid_t pid = CPUCLOCK_PID(which_clock);
294 int err = -EINVAL;
295
296 if (pid == 0) {
297 /*
298 * Special case constant value for our own clocks.
299 * We don't have to do any lookup to find ourselves.
300 */
301 err = posix_cpu_clock_get_task(current, which_clock, tp);
302 } else {
303 /*
304 * Find the given PID, and validate that the caller
305 * should be able to see it.
306 */
307 struct task_struct *p;
308 rcu_read_lock();
309 p = find_task_by_vpid(pid);
310 if (p)
311 err = posix_cpu_clock_get_task(p, which_clock, tp);
312 rcu_read_unlock();
313 }
314
315 return err;
316}
317
318/*
319 * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
320 * This is called from sys_timer_create() and do_cpu_nanosleep() with the
321 * new timer already all-zeros initialized.
322 */
323static int posix_cpu_timer_create(struct k_itimer *new_timer)
324{
325 int ret = 0;
326 const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
327 struct task_struct *p;
328
329 if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
330 return -EINVAL;
331
332 new_timer->kclock = &clock_posix_cpu;
333
334 INIT_LIST_HEAD(&new_timer->it.cpu.entry);
335
336 rcu_read_lock();
337 if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
338 if (pid == 0) {
339 p = current;
340 } else {
341 p = find_task_by_vpid(pid);
342 if (p && !same_thread_group(p, current))
343 p = NULL;
344 }
345 } else {
346 if (pid == 0) {
347 p = current->group_leader;
348 } else {
349 p = find_task_by_vpid(pid);
350 if (p && !has_group_leader_pid(p))
351 p = NULL;
352 }
353 }
354 new_timer->it.cpu.task = p;
355 if (p) {
356 get_task_struct(p);
357 } else {
358 ret = -EINVAL;
359 }
360 rcu_read_unlock();
361
362 return ret;
363}
364
365/*
366 * Clean up a CPU-clock timer that is about to be destroyed.
367 * This is called from timer deletion with the timer already locked.
368 * If we return TIMER_RETRY, it's necessary to release the timer's lock
369 * and try again. (This happens when the timer is in the middle of firing.)
370 */
371static int posix_cpu_timer_del(struct k_itimer *timer)
372{
373 int ret = 0;
374 unsigned long flags;
375 struct sighand_struct *sighand;
376 struct task_struct *p = timer->it.cpu.task;
377
378 WARN_ON_ONCE(p == NULL);
379
380 /*
381 * Protect against sighand release/switch in exit/exec and process/
382 * thread timer list entry concurrent read/writes.
383 */
384 sighand = lock_task_sighand(p, &flags);
385 if (unlikely(sighand == NULL)) {
386 /*
387 * We raced with the reaping of the task.
388 * The deletion should have cleared us off the list.
389 */
390 WARN_ON_ONCE(!list_empty(&timer->it.cpu.entry));
391 } else {
392 if (timer->it.cpu.firing)
393 ret = TIMER_RETRY;
394 else
395 list_del(&timer->it.cpu.entry);
396
397 unlock_task_sighand(p, &flags);
398 }
399
400 if (!ret)
401 put_task_struct(p);
402
403 return ret;
404}
405
406static void cleanup_timers_list(struct list_head *head)
407{
408 struct cpu_timer_list *timer, *next;
409
410 list_for_each_entry_safe(timer, next, head, entry)
411 list_del_init(&timer->entry);
412}
413
414/*
415 * Clean out CPU timers still ticking when a thread exited. The task
416 * pointer is cleared, and the expiry time is replaced with the residual
417 * time for later timer_gettime calls to return.
418 * This must be called with the siglock held.
419 */
420static void cleanup_timers(struct list_head *head)
421{
422 cleanup_timers_list(head);
423 cleanup_timers_list(++head);
424 cleanup_timers_list(++head);
425}
426
427/*
428 * These are both called with the siglock held, when the current thread
429 * is being reaped. When the final (leader) thread in the group is reaped,
430 * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
431 */
432void posix_cpu_timers_exit(struct task_struct *tsk)
433{
434 cleanup_timers(tsk->cpu_timers);
435}
436void posix_cpu_timers_exit_group(struct task_struct *tsk)
437{
438 cleanup_timers(tsk->signal->cpu_timers);
439}
440
441static inline int expires_gt(u64 expires, u64 new_exp)
442{
443 return expires == 0 || expires > new_exp;
444}
445
446/*
447 * Insert the timer on the appropriate list before any timers that
448 * expire later. This must be called with the sighand lock held.
449 */
450static void arm_timer(struct k_itimer *timer)
451{
452 struct task_struct *p = timer->it.cpu.task;
453 struct list_head *head, *listpos;
454 struct task_cputime *cputime_expires;
455 struct cpu_timer_list *const nt = &timer->it.cpu;
456 struct cpu_timer_list *next;
457
458 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
459 head = p->cpu_timers;
460 cputime_expires = &p->cputime_expires;
461 } else {
462 head = p->signal->cpu_timers;
463 cputime_expires = &p->signal->cputime_expires;
464 }
465 head += CPUCLOCK_WHICH(timer->it_clock);
466
467 listpos = head;
468 list_for_each_entry(next, head, entry) {
469 if (nt->expires < next->expires)
470 break;
471 listpos = &next->entry;
472 }
473 list_add(&nt->entry, listpos);
474
475 if (listpos == head) {
476 u64 exp = nt->expires;
477
478 /*
479 * We are the new earliest-expiring POSIX 1.b timer, hence
480 * need to update expiration cache. Take into account that
481 * for process timers we share expiration cache with itimers
482 * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
483 */
484
485 switch (CPUCLOCK_WHICH(timer->it_clock)) {
486 case CPUCLOCK_PROF:
487 if (expires_gt(cputime_expires->prof_exp, exp))
488 cputime_expires->prof_exp = exp;
489 break;
490 case CPUCLOCK_VIRT:
491 if (expires_gt(cputime_expires->virt_exp, exp))
492 cputime_expires->virt_exp = exp;
493 break;
494 case CPUCLOCK_SCHED:
495 if (expires_gt(cputime_expires->sched_exp, exp))
496 cputime_expires->sched_exp = exp;
497 break;
498 }
499 if (CPUCLOCK_PERTHREAD(timer->it_clock))
500 tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER);
501 else
502 tick_dep_set_signal(p->signal, TICK_DEP_BIT_POSIX_TIMER);
503 }
504}
505
506/*
507 * The timer is locked, fire it and arrange for its reload.
508 */
509static void cpu_timer_fire(struct k_itimer *timer)
510{
511 if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
512 /*
513 * User don't want any signal.
514 */
515 timer->it.cpu.expires = 0;
516 } else if (unlikely(timer->sigq == NULL)) {
517 /*
518 * This a special case for clock_nanosleep,
519 * not a normal timer from sys_timer_create.
520 */
521 wake_up_process(timer->it_process);
522 timer->it.cpu.expires = 0;
523 } else if (timer->it.cpu.incr == 0) {
524 /*
525 * One-shot timer. Clear it as soon as it's fired.
526 */
527 posix_timer_event(timer, 0);
528 timer->it.cpu.expires = 0;
529 } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
530 /*
531 * The signal did not get queued because the signal
532 * was ignored, so we won't get any callback to
533 * reload the timer. But we need to keep it
534 * ticking in case the signal is deliverable next time.
535 */
536 posix_cpu_timer_rearm(timer);
537 ++timer->it_requeue_pending;
538 }
539}
540
541/*
542 * Sample a process (thread group) timer for the given group_leader task.
543 * Must be called with task sighand lock held for safe while_each_thread()
544 * traversal.
545 */
546static int cpu_timer_sample_group(const clockid_t which_clock,
547 struct task_struct *p, u64 *sample)
548{
549 struct task_cputime cputime;
550
551 thread_group_cputimer(p, &cputime);
552 switch (CPUCLOCK_WHICH(which_clock)) {
553 default:
554 return -EINVAL;
555 case CPUCLOCK_PROF:
556 *sample = cputime.utime + cputime.stime;
557 break;
558 case CPUCLOCK_VIRT:
559 *sample = cputime.utime;
560 break;
561 case CPUCLOCK_SCHED:
562 *sample = cputime.sum_exec_runtime;
563 break;
564 }
565 return 0;
566}
567
568/*
569 * Guts of sys_timer_settime for CPU timers.
570 * This is called with the timer locked and interrupts disabled.
571 * If we return TIMER_RETRY, it's necessary to release the timer's lock
572 * and try again. (This happens when the timer is in the middle of firing.)
573 */
574static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
575 struct itimerspec64 *new, struct itimerspec64 *old)
576{
577 unsigned long flags;
578 struct sighand_struct *sighand;
579 struct task_struct *p = timer->it.cpu.task;
580 u64 old_expires, new_expires, old_incr, val;
581 int ret;
582
583 WARN_ON_ONCE(p == NULL);
584
585 /*
586 * Use the to_ktime conversion because that clamps the maximum
587 * value to KTIME_MAX and avoid multiplication overflows.
588 */
589 new_expires = ktime_to_ns(timespec64_to_ktime(new->it_value));
590
591 /*
592 * Protect against sighand release/switch in exit/exec and p->cpu_timers
593 * and p->signal->cpu_timers read/write in arm_timer()
594 */
595 sighand = lock_task_sighand(p, &flags);
596 /*
597 * If p has just been reaped, we can no
598 * longer get any information about it at all.
599 */
600 if (unlikely(sighand == NULL)) {
601 return -ESRCH;
602 }
603
604 /*
605 * Disarm any old timer after extracting its expiry time.
606 */
607 lockdep_assert_irqs_disabled();
608
609 ret = 0;
610 old_incr = timer->it.cpu.incr;
611 old_expires = timer->it.cpu.expires;
612 if (unlikely(timer->it.cpu.firing)) {
613 timer->it.cpu.firing = -1;
614 ret = TIMER_RETRY;
615 } else
616 list_del_init(&timer->it.cpu.entry);
617
618 /*
619 * We need to sample the current value to convert the new
620 * value from to relative and absolute, and to convert the
621 * old value from absolute to relative. To set a process
622 * timer, we need a sample to balance the thread expiry
623 * times (in arm_timer). With an absolute time, we must
624 * check if it's already passed. In short, we need a sample.
625 */
626 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
627 cpu_clock_sample(timer->it_clock, p, &val);
628 } else {
629 cpu_timer_sample_group(timer->it_clock, p, &val);
630 }
631
632 if (old) {
633 if (old_expires == 0) {
634 old->it_value.tv_sec = 0;
635 old->it_value.tv_nsec = 0;
636 } else {
637 /*
638 * Update the timer in case it has
639 * overrun already. If it has,
640 * we'll report it as having overrun
641 * and with the next reloaded timer
642 * already ticking, though we are
643 * swallowing that pending
644 * notification here to install the
645 * new setting.
646 */
647 bump_cpu_timer(timer, val);
648 if (val < timer->it.cpu.expires) {
649 old_expires = timer->it.cpu.expires - val;
650 old->it_value = ns_to_timespec64(old_expires);
651 } else {
652 old->it_value.tv_nsec = 1;
653 old->it_value.tv_sec = 0;
654 }
655 }
656 }
657
658 if (unlikely(ret)) {
659 /*
660 * We are colliding with the timer actually firing.
661 * Punt after filling in the timer's old value, and
662 * disable this firing since we are already reporting
663 * it as an overrun (thanks to bump_cpu_timer above).
664 */
665 unlock_task_sighand(p, &flags);
666 goto out;
667 }
668
669 if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
670 new_expires += val;
671 }
672
673 /*
674 * Install the new expiry time (or zero).
675 * For a timer with no notification action, we don't actually
676 * arm the timer (we'll just fake it for timer_gettime).
677 */
678 timer->it.cpu.expires = new_expires;
679 if (new_expires != 0 && val < new_expires) {
680 arm_timer(timer);
681 }
682
683 unlock_task_sighand(p, &flags);
684 /*
685 * Install the new reload setting, and
686 * set up the signal and overrun bookkeeping.
687 */
688 timer->it.cpu.incr = timespec64_to_ns(&new->it_interval);
689
690 /*
691 * This acts as a modification timestamp for the timer,
692 * so any automatic reload attempt will punt on seeing
693 * that we have reset the timer manually.
694 */
695 timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
696 ~REQUEUE_PENDING;
697 timer->it_overrun_last = 0;
698 timer->it_overrun = -1;
699
700 if (new_expires != 0 && !(val < new_expires)) {
701 /*
702 * The designated time already passed, so we notify
703 * immediately, even if the thread never runs to
704 * accumulate more time on this clock.
705 */
706 cpu_timer_fire(timer);
707 }
708
709 ret = 0;
710 out:
711 if (old)
712 old->it_interval = ns_to_timespec64(old_incr);
713
714 return ret;
715}
716
717static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp)
718{
719 u64 now;
720 struct task_struct *p = timer->it.cpu.task;
721
722 WARN_ON_ONCE(p == NULL);
723
724 /*
725 * Easy part: convert the reload time.
726 */
727 itp->it_interval = ns_to_timespec64(timer->it.cpu.incr);
728
729 if (!timer->it.cpu.expires)
730 return;
731
732 /*
733 * Sample the clock to take the difference with the expiry time.
734 */
735 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
736 cpu_clock_sample(timer->it_clock, p, &now);
737 } else {
738 struct sighand_struct *sighand;
739 unsigned long flags;
740
741 /*
742 * Protect against sighand release/switch in exit/exec and
743 * also make timer sampling safe if it ends up calling
744 * thread_group_cputime().
745 */
746 sighand = lock_task_sighand(p, &flags);
747 if (unlikely(sighand == NULL)) {
748 /*
749 * The process has been reaped.
750 * We can't even collect a sample any more.
751 * Call the timer disarmed, nothing else to do.
752 */
753 timer->it.cpu.expires = 0;
754 return;
755 } else {
756 cpu_timer_sample_group(timer->it_clock, p, &now);
757 unlock_task_sighand(p, &flags);
758 }
759 }
760
761 if (now < timer->it.cpu.expires) {
762 itp->it_value = ns_to_timespec64(timer->it.cpu.expires - now);
763 } else {
764 /*
765 * The timer should have expired already, but the firing
766 * hasn't taken place yet. Say it's just about to expire.
767 */
768 itp->it_value.tv_nsec = 1;
769 itp->it_value.tv_sec = 0;
770 }
771}
772
773static unsigned long long
774check_timers_list(struct list_head *timers,
775 struct list_head *firing,
776 unsigned long long curr)
777{
778 int maxfire = 20;
779
780 while (!list_empty(timers)) {
781 struct cpu_timer_list *t;
782
783 t = list_first_entry(timers, struct cpu_timer_list, entry);
784
785 if (!--maxfire || curr < t->expires)
786 return t->expires;
787
788 t->firing = 1;
789 list_move_tail(&t->entry, firing);
790 }
791
792 return 0;
793}
794
795static inline void check_dl_overrun(struct task_struct *tsk)
796{
797 if (tsk->dl.dl_overrun) {
798 tsk->dl.dl_overrun = 0;
799 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
800 }
801}
802
803/*
804 * Check for any per-thread CPU timers that have fired and move them off
805 * the tsk->cpu_timers[N] list onto the firing list. Here we update the
806 * tsk->it_*_expires values to reflect the remaining thread CPU timers.
807 */
808static void check_thread_timers(struct task_struct *tsk,
809 struct list_head *firing)
810{
811 struct list_head *timers = tsk->cpu_timers;
812 struct task_cputime *tsk_expires = &tsk->cputime_expires;
813 u64 expires;
814 unsigned long soft;
815
816 if (dl_task(tsk))
817 check_dl_overrun(tsk);
818
819 /*
820 * If cputime_expires is zero, then there are no active
821 * per thread CPU timers.
822 */
823 if (task_cputime_zero(&tsk->cputime_expires))
824 return;
825
826 expires = check_timers_list(timers, firing, prof_ticks(tsk));
827 tsk_expires->prof_exp = expires;
828
829 expires = check_timers_list(++timers, firing, virt_ticks(tsk));
830 tsk_expires->virt_exp = expires;
831
832 tsk_expires->sched_exp = check_timers_list(++timers, firing,
833 tsk->se.sum_exec_runtime);
834
835 /*
836 * Check for the special case thread timers.
837 */
838 soft = task_rlimit(tsk, RLIMIT_RTTIME);
839 if (soft != RLIM_INFINITY) {
840 unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME);
841
842 if (hard != RLIM_INFINITY &&
843 tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
844 /*
845 * At the hard limit, we just die.
846 * No need to calculate anything else now.
847 */
848 if (print_fatal_signals) {
849 pr_info("CPU Watchdog Timeout (hard): %s[%d]\n",
850 tsk->comm, task_pid_nr(tsk));
851 }
852 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
853 return;
854 }
855 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
856 /*
857 * At the soft limit, send a SIGXCPU every second.
858 */
859 if (soft < hard) {
860 soft += USEC_PER_SEC;
861 tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur =
862 soft;
863 }
864 if (print_fatal_signals) {
865 pr_info("RT Watchdog Timeout (soft): %s[%d]\n",
866 tsk->comm, task_pid_nr(tsk));
867 }
868 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
869 }
870 }
871 if (task_cputime_zero(tsk_expires))
872 tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER);
873}
874
875static inline void stop_process_timers(struct signal_struct *sig)
876{
877 struct thread_group_cputimer *cputimer = &sig->cputimer;
878
879 /* Turn off cputimer->running. This is done without locking. */
880 WRITE_ONCE(cputimer->running, false);
881 tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER);
882}
883
884static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
885 u64 *expires, u64 cur_time, int signo)
886{
887 if (!it->expires)
888 return;
889
890 if (cur_time >= it->expires) {
891 if (it->incr)
892 it->expires += it->incr;
893 else
894 it->expires = 0;
895
896 trace_itimer_expire(signo == SIGPROF ?
897 ITIMER_PROF : ITIMER_VIRTUAL,
898 tsk->signal->leader_pid, cur_time);
899 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
900 }
901
902 if (it->expires && (!*expires || it->expires < *expires))
903 *expires = it->expires;
904}
905
906/*
907 * Check for any per-thread CPU timers that have fired and move them
908 * off the tsk->*_timers list onto the firing list. Per-thread timers
909 * have already been taken off.
910 */
911static void check_process_timers(struct task_struct *tsk,
912 struct list_head *firing)
913{
914 struct signal_struct *const sig = tsk->signal;
915 u64 utime, ptime, virt_expires, prof_expires;
916 u64 sum_sched_runtime, sched_expires;
917 struct list_head *timers = sig->cpu_timers;
918 struct task_cputime cputime;
919 unsigned long soft;
920
921 if (dl_task(tsk))
922 check_dl_overrun(tsk);
923
924 /*
925 * If cputimer is not running, then there are no active
926 * process wide timers (POSIX 1.b, itimers, RLIMIT_CPU).
927 */
928 if (!READ_ONCE(tsk->signal->cputimer.running))
929 return;
930
931 /*
932 * Signify that a thread is checking for process timers.
933 * Write access to this field is protected by the sighand lock.
934 */
935 sig->cputimer.checking_timer = true;
936
937 /*
938 * Collect the current process totals.
939 */
940 thread_group_cputimer(tsk, &cputime);
941 utime = cputime.utime;
942 ptime = utime + cputime.stime;
943 sum_sched_runtime = cputime.sum_exec_runtime;
944
945 prof_expires = check_timers_list(timers, firing, ptime);
946 virt_expires = check_timers_list(++timers, firing, utime);
947 sched_expires = check_timers_list(++timers, firing, sum_sched_runtime);
948
949 /*
950 * Check for the special case process timers.
951 */
952 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
953 SIGPROF);
954 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
955 SIGVTALRM);
956 soft = task_rlimit(tsk, RLIMIT_CPU);
957 if (soft != RLIM_INFINITY) {
958 unsigned long psecs = div_u64(ptime, NSEC_PER_SEC);
959 unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU);
960 u64 x;
961 if (psecs >= hard) {
962 /*
963 * At the hard limit, we just die.
964 * No need to calculate anything else now.
965 */
966 if (print_fatal_signals) {
967 pr_info("RT Watchdog Timeout (hard): %s[%d]\n",
968 tsk->comm, task_pid_nr(tsk));
969 }
970 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
971 return;
972 }
973 if (psecs >= soft) {
974 /*
975 * At the soft limit, send a SIGXCPU every second.
976 */
977 if (print_fatal_signals) {
978 pr_info("CPU Watchdog Timeout (soft): %s[%d]\n",
979 tsk->comm, task_pid_nr(tsk));
980 }
981 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
982 if (soft < hard) {
983 soft++;
984 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
985 }
986 }
987 x = soft * NSEC_PER_SEC;
988 if (!prof_expires || x < prof_expires)
989 prof_expires = x;
990 }
991
992 sig->cputime_expires.prof_exp = prof_expires;
993 sig->cputime_expires.virt_exp = virt_expires;
994 sig->cputime_expires.sched_exp = sched_expires;
995 if (task_cputime_zero(&sig->cputime_expires))
996 stop_process_timers(sig);
997
998 sig->cputimer.checking_timer = false;
999}
1000
1001/*
1002 * This is called from the signal code (via posixtimer_rearm)
1003 * when the last timer signal was delivered and we have to reload the timer.
1004 */
1005static void posix_cpu_timer_rearm(struct k_itimer *timer)
1006{
1007 struct sighand_struct *sighand;
1008 unsigned long flags;
1009 struct task_struct *p = timer->it.cpu.task;
1010 u64 now;
1011
1012 WARN_ON_ONCE(p == NULL);
1013
1014 /*
1015 * Fetch the current sample and update the timer's expiry time.
1016 */
1017 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1018 cpu_clock_sample(timer->it_clock, p, &now);
1019 bump_cpu_timer(timer, now);
1020 if (unlikely(p->exit_state))
1021 return;
1022
1023 /* Protect timer list r/w in arm_timer() */
1024 sighand = lock_task_sighand(p, &flags);
1025 if (!sighand)
1026 return;
1027 } else {
1028 /*
1029 * Protect arm_timer() and timer sampling in case of call to
1030 * thread_group_cputime().
1031 */
1032 sighand = lock_task_sighand(p, &flags);
1033 if (unlikely(sighand == NULL)) {
1034 /*
1035 * The process has been reaped.
1036 * We can't even collect a sample any more.
1037 */
1038 timer->it.cpu.expires = 0;
1039 return;
1040 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1041 /* If the process is dying, no need to rearm */
1042 goto unlock;
1043 }
1044 cpu_timer_sample_group(timer->it_clock, p, &now);
1045 bump_cpu_timer(timer, now);
1046 /* Leave the sighand locked for the call below. */
1047 }
1048
1049 /*
1050 * Now re-arm for the new expiry time.
1051 */
1052 lockdep_assert_irqs_disabled();
1053 arm_timer(timer);
1054unlock:
1055 unlock_task_sighand(p, &flags);
1056}
1057
1058/**
1059 * task_cputime_expired - Compare two task_cputime entities.
1060 *
1061 * @sample: The task_cputime structure to be checked for expiration.
1062 * @expires: Expiration times, against which @sample will be checked.
1063 *
1064 * Checks @sample against @expires to see if any field of @sample has expired.
1065 * Returns true if any field of the former is greater than the corresponding
1066 * field of the latter if the latter field is set. Otherwise returns false.
1067 */
1068static inline int task_cputime_expired(const struct task_cputime *sample,
1069 const struct task_cputime *expires)
1070{
1071 if (expires->utime && sample->utime >= expires->utime)
1072 return 1;
1073 if (expires->stime && sample->utime + sample->stime >= expires->stime)
1074 return 1;
1075 if (expires->sum_exec_runtime != 0 &&
1076 sample->sum_exec_runtime >= expires->sum_exec_runtime)
1077 return 1;
1078 return 0;
1079}
1080
1081/**
1082 * fastpath_timer_check - POSIX CPU timers fast path.
1083 *
1084 * @tsk: The task (thread) being checked.
1085 *
1086 * Check the task and thread group timers. If both are zero (there are no
1087 * timers set) return false. Otherwise snapshot the task and thread group
1088 * timers and compare them with the corresponding expiration times. Return
1089 * true if a timer has expired, else return false.
1090 */
1091static inline int fastpath_timer_check(struct task_struct *tsk)
1092{
1093 struct signal_struct *sig;
1094
1095 if (!task_cputime_zero(&tsk->cputime_expires)) {
1096 struct task_cputime task_sample;
1097
1098 task_cputime(tsk, &task_sample.utime, &task_sample.stime);
1099 task_sample.sum_exec_runtime = tsk->se.sum_exec_runtime;
1100 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1101 return 1;
1102 }
1103
1104 sig = tsk->signal;
1105 /*
1106 * Check if thread group timers expired when the cputimer is
1107 * running and no other thread in the group is already checking
1108 * for thread group cputimers. These fields are read without the
1109 * sighand lock. However, this is fine because this is meant to
1110 * be a fastpath heuristic to determine whether we should try to
1111 * acquire the sighand lock to check/handle timers.
1112 *
1113 * In the worst case scenario, if 'running' or 'checking_timer' gets
1114 * set but the current thread doesn't see the change yet, we'll wait
1115 * until the next thread in the group gets a scheduler interrupt to
1116 * handle the timer. This isn't an issue in practice because these
1117 * types of delays with signals actually getting sent are expected.
1118 */
1119 if (READ_ONCE(sig->cputimer.running) &&
1120 !READ_ONCE(sig->cputimer.checking_timer)) {
1121 struct task_cputime group_sample;
1122
1123 sample_cputime_atomic(&group_sample, &sig->cputimer.cputime_atomic);
1124
1125 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1126 return 1;
1127 }
1128
1129 if (dl_task(tsk) && tsk->dl.dl_overrun)
1130 return 1;
1131
1132 return 0;
1133}
1134
1135/*
1136 * This is called from the timer interrupt handler. The irq handler has
1137 * already updated our counts. We need to check if any timers fire now.
1138 * Interrupts are disabled.
1139 */
1140void run_posix_cpu_timers(struct task_struct *tsk)
1141{
1142 LIST_HEAD(firing);
1143 struct k_itimer *timer, *next;
1144 unsigned long flags;
1145
1146 lockdep_assert_irqs_disabled();
1147
1148 /*
1149 * The fast path checks that there are no expired thread or thread
1150 * group timers. If that's so, just return.
1151 */
1152 if (!fastpath_timer_check(tsk))
1153 return;
1154
1155 if (!lock_task_sighand(tsk, &flags))
1156 return;
1157 /*
1158 * Here we take off tsk->signal->cpu_timers[N] and
1159 * tsk->cpu_timers[N] all the timers that are firing, and
1160 * put them on the firing list.
1161 */
1162 check_thread_timers(tsk, &firing);
1163
1164 check_process_timers(tsk, &firing);
1165
1166 /*
1167 * We must release these locks before taking any timer's lock.
1168 * There is a potential race with timer deletion here, as the
1169 * siglock now protects our private firing list. We have set
1170 * the firing flag in each timer, so that a deletion attempt
1171 * that gets the timer lock before we do will give it up and
1172 * spin until we've taken care of that timer below.
1173 */
1174 unlock_task_sighand(tsk, &flags);
1175
1176 /*
1177 * Now that all the timers on our list have the firing flag,
1178 * no one will touch their list entries but us. We'll take
1179 * each timer's lock before clearing its firing flag, so no
1180 * timer call will interfere.
1181 */
1182 list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1183 int cpu_firing;
1184
1185 spin_lock(&timer->it_lock);
1186 list_del_init(&timer->it.cpu.entry);
1187 cpu_firing = timer->it.cpu.firing;
1188 timer->it.cpu.firing = 0;
1189 /*
1190 * The firing flag is -1 if we collided with a reset
1191 * of the timer, which already reported this
1192 * almost-firing as an overrun. So don't generate an event.
1193 */
1194 if (likely(cpu_firing >= 0))
1195 cpu_timer_fire(timer);
1196 spin_unlock(&timer->it_lock);
1197 }
1198}
1199
1200/*
1201 * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1202 * The tsk->sighand->siglock must be held by the caller.
1203 */
1204void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1205 u64 *newval, u64 *oldval)
1206{
1207 u64 now;
1208 int ret;
1209
1210 WARN_ON_ONCE(clock_idx == CPUCLOCK_SCHED);
1211 ret = cpu_timer_sample_group(clock_idx, tsk, &now);
1212
1213 if (oldval && ret != -EINVAL) {
1214 /*
1215 * We are setting itimer. The *oldval is absolute and we update
1216 * it to be relative, *newval argument is relative and we update
1217 * it to be absolute.
1218 */
1219 if (*oldval) {
1220 if (*oldval <= now) {
1221 /* Just about to fire. */
1222 *oldval = TICK_NSEC;
1223 } else {
1224 *oldval -= now;
1225 }
1226 }
1227
1228 if (!*newval)
1229 return;
1230 *newval += now;
1231 }
1232
1233 /*
1234 * Update expiration cache if we are the earliest timer, or eventually
1235 * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1236 */
1237 switch (clock_idx) {
1238 case CPUCLOCK_PROF:
1239 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1240 tsk->signal->cputime_expires.prof_exp = *newval;
1241 break;
1242 case CPUCLOCK_VIRT:
1243 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1244 tsk->signal->cputime_expires.virt_exp = *newval;
1245 break;
1246 }
1247
1248 tick_dep_set_signal(tsk->signal, TICK_DEP_BIT_POSIX_TIMER);
1249}
1250
1251static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1252 const struct timespec64 *rqtp)
1253{
1254 struct itimerspec64 it;
1255 struct k_itimer timer;
1256 u64 expires;
1257 int error;
1258
1259 /*
1260 * Set up a temporary timer and then wait for it to go off.
1261 */
1262 memset(&timer, 0, sizeof timer);
1263 spin_lock_init(&timer.it_lock);
1264 timer.it_clock = which_clock;
1265 timer.it_overrun = -1;
1266 error = posix_cpu_timer_create(&timer);
1267 timer.it_process = current;
1268 if (!error) {
1269 static struct itimerspec64 zero_it;
1270 struct restart_block *restart;
1271
1272 memset(&it, 0, sizeof(it));
1273 it.it_value = *rqtp;
1274
1275 spin_lock_irq(&timer.it_lock);
1276 error = posix_cpu_timer_set(&timer, flags, &it, NULL);
1277 if (error) {
1278 spin_unlock_irq(&timer.it_lock);
1279 return error;
1280 }
1281
1282 while (!signal_pending(current)) {
1283 if (timer.it.cpu.expires == 0) {
1284 /*
1285 * Our timer fired and was reset, below
1286 * deletion can not fail.
1287 */
1288 posix_cpu_timer_del(&timer);
1289 spin_unlock_irq(&timer.it_lock);
1290 return 0;
1291 }
1292
1293 /*
1294 * Block until cpu_timer_fire (or a signal) wakes us.
1295 */
1296 __set_current_state(TASK_INTERRUPTIBLE);
1297 spin_unlock_irq(&timer.it_lock);
1298 schedule();
1299 spin_lock_irq(&timer.it_lock);
1300 }
1301
1302 /*
1303 * We were interrupted by a signal.
1304 */
1305 expires = timer.it.cpu.expires;
1306 error = posix_cpu_timer_set(&timer, 0, &zero_it, &it);
1307 if (!error) {
1308 /*
1309 * Timer is now unarmed, deletion can not fail.
1310 */
1311 posix_cpu_timer_del(&timer);
1312 }
1313 spin_unlock_irq(&timer.it_lock);
1314
1315 while (error == TIMER_RETRY) {
1316 /*
1317 * We need to handle case when timer was or is in the
1318 * middle of firing. In other cases we already freed
1319 * resources.
1320 */
1321 spin_lock_irq(&timer.it_lock);
1322 error = posix_cpu_timer_del(&timer);
1323 spin_unlock_irq(&timer.it_lock);
1324 }
1325
1326 if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) {
1327 /*
1328 * It actually did fire already.
1329 */
1330 return 0;
1331 }
1332
1333 error = -ERESTART_RESTARTBLOCK;
1334 /*
1335 * Report back to the user the time still remaining.
1336 */
1337 restart = ¤t->restart_block;
1338 restart->nanosleep.expires = expires;
1339 if (restart->nanosleep.type != TT_NONE)
1340 error = nanosleep_copyout(restart, &it.it_value);
1341 }
1342
1343 return error;
1344}
1345
1346static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1347
1348static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1349 const struct timespec64 *rqtp)
1350{
1351 struct restart_block *restart_block = ¤t->restart_block;
1352 int error;
1353
1354 /*
1355 * Diagnose required errors first.
1356 */
1357 if (CPUCLOCK_PERTHREAD(which_clock) &&
1358 (CPUCLOCK_PID(which_clock) == 0 ||
1359 CPUCLOCK_PID(which_clock) == task_pid_vnr(current)))
1360 return -EINVAL;
1361
1362 error = do_cpu_nanosleep(which_clock, flags, rqtp);
1363
1364 if (error == -ERESTART_RESTARTBLOCK) {
1365
1366 if (flags & TIMER_ABSTIME)
1367 return -ERESTARTNOHAND;
1368
1369 restart_block->fn = posix_cpu_nsleep_restart;
1370 restart_block->nanosleep.clockid = which_clock;
1371 }
1372 return error;
1373}
1374
1375static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1376{
1377 clockid_t which_clock = restart_block->nanosleep.clockid;
1378 struct timespec64 t;
1379
1380 t = ns_to_timespec64(restart_block->nanosleep.expires);
1381
1382 return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t);
1383}
1384
1385#define PROCESS_CLOCK make_process_cpuclock(0, CPUCLOCK_SCHED)
1386#define THREAD_CLOCK make_thread_cpuclock(0, CPUCLOCK_SCHED)
1387
1388static int process_cpu_clock_getres(const clockid_t which_clock,
1389 struct timespec64 *tp)
1390{
1391 return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1392}
1393static int process_cpu_clock_get(const clockid_t which_clock,
1394 struct timespec64 *tp)
1395{
1396 return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1397}
1398static int process_cpu_timer_create(struct k_itimer *timer)
1399{
1400 timer->it_clock = PROCESS_CLOCK;
1401 return posix_cpu_timer_create(timer);
1402}
1403static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1404 const struct timespec64 *rqtp)
1405{
1406 return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp);
1407}
1408static int thread_cpu_clock_getres(const clockid_t which_clock,
1409 struct timespec64 *tp)
1410{
1411 return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1412}
1413static int thread_cpu_clock_get(const clockid_t which_clock,
1414 struct timespec64 *tp)
1415{
1416 return posix_cpu_clock_get(THREAD_CLOCK, tp);
1417}
1418static int thread_cpu_timer_create(struct k_itimer *timer)
1419{
1420 timer->it_clock = THREAD_CLOCK;
1421 return posix_cpu_timer_create(timer);
1422}
1423
1424const struct k_clock clock_posix_cpu = {
1425 .clock_getres = posix_cpu_clock_getres,
1426 .clock_set = posix_cpu_clock_set,
1427 .clock_get = posix_cpu_clock_get,
1428 .timer_create = posix_cpu_timer_create,
1429 .nsleep = posix_cpu_nsleep,
1430 .timer_set = posix_cpu_timer_set,
1431 .timer_del = posix_cpu_timer_del,
1432 .timer_get = posix_cpu_timer_get,
1433 .timer_rearm = posix_cpu_timer_rearm,
1434};
1435
1436const struct k_clock clock_process = {
1437 .clock_getres = process_cpu_clock_getres,
1438 .clock_get = process_cpu_clock_get,
1439 .timer_create = process_cpu_timer_create,
1440 .nsleep = process_cpu_nsleep,
1441};
1442
1443const struct k_clock clock_thread = {
1444 .clock_getres = thread_cpu_clock_getres,
1445 .clock_get = thread_cpu_clock_get,
1446 .timer_create = thread_cpu_timer_create,
1447};