projects/linux·blob master · refs

1737 lines · 48.8 KiB

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#include <linux/task_work.h>
19
20#include "posix-timers.h"
21
22static bool posix_cpu_timer_rearm(struct k_itimer *timer);
23
24void posix_cputimers_group_init(struct posix_cputimers *pct, u64 cpu_limit)
25{
26 posix_cputimers_init(pct);
27 if (cpu_limit != RLIM_INFINITY) {
28 pct->bases[CPUCLOCK_PROF].nextevt = cpu_limit * NSEC_PER_SEC;
29 pct->timers_active = true;
30 }
31}
32
33/*
34 * Called after updating RLIMIT_CPU to run cpu timer and update
35 * tsk->signal->posix_cputimers.bases[clock].nextevt expiration cache if
36 * necessary. Needs siglock protection since other code may update the
37 * expiration cache as well.
38 *
39 * Returns 0 on success, -ESRCH on failure. Can fail if the task is exiting and
40 * we cannot lock_task_sighand. Cannot fail if task is current.
41 */
42int update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
43{
44 u64 nsecs = (u64)rlim_new * NSEC_PER_SEC;
45 unsigned long irq_fl;
46
47 if (!lock_task_sighand(task, &irq_fl))
48 return -ESRCH;
49 set_process_cpu_timer(task, CPUCLOCK_PROF, &nsecs, NULL);
50 unlock_task_sighand(task, &irq_fl);
51 return 0;
52}
53
54/*
55 * Functions for validating access to tasks.
56 */
57static struct pid *pid_for_clock(const clockid_t clock, bool gettime)
58{
59 const bool thread = !!CPUCLOCK_PERTHREAD(clock);
60 const pid_t upid = CPUCLOCK_PID(clock);
61 struct pid *pid;
62
63 if (CPUCLOCK_WHICH(clock) >= CPUCLOCK_MAX)
64 return NULL;
65
66 /*
67 * If the encoded PID is 0, then the timer is targeted at current
68 * or the process to which current belongs.
69 */
70 if (upid == 0)
71 return thread ? task_pid(current) : task_tgid(current);
72
73 pid = find_vpid(upid);
74 if (!pid)
75 return NULL;
76
77 if (thread) {
78 struct task_struct *tsk = pid_task(pid, PIDTYPE_PID);
79 return (tsk && same_thread_group(tsk, current)) ? pid : NULL;
80 }
81
82 /*
83 * For clock_gettime(PROCESS) allow finding the process by
84 * with the pid of the current task. The code needs the tgid
85 * of the process so that pid_task(pid, PIDTYPE_TGID) can be
86 * used to find the process.
87 */
88 if (gettime && (pid == task_pid(current)))
89 return task_tgid(current);
90
91 /*
92 * For processes require that pid identifies a process.
93 */
94 return pid_has_task(pid, PIDTYPE_TGID) ? pid : NULL;
95}
96
97static inline int validate_clock_permissions(const clockid_t clock)
98{
99 int ret;
100
101 rcu_read_lock();
102 ret = pid_for_clock(clock, false) ? 0 : -EINVAL;
103 rcu_read_unlock();
104
105 return ret;
106}
107
108static inline enum pid_type clock_pid_type(const clockid_t clock)
109{
110 return CPUCLOCK_PERTHREAD(clock) ? PIDTYPE_PID : PIDTYPE_TGID;
111}
112
113static inline struct task_struct *cpu_timer_task_rcu(struct k_itimer *timer)
114{
115 return pid_task(timer->it.cpu.pid, clock_pid_type(timer->it_clock));
116}
117
118/*
119 * Update expiry time from increment, and increase overrun count,
120 * given the current clock sample.
121 */
122static u64 bump_cpu_timer(struct k_itimer *timer, u64 now)
123{
124 u64 delta, incr, expires = timer->it.cpu.node.expires;
125 int i;
126
127 if (!timer->it_interval)
128 return expires;
129
130 if (now < expires)
131 return expires;
132
133 incr = timer->it_interval;
134 delta = now + incr - expires;
135
136 /* Don't use (incr*2 < delta), incr*2 might overflow. */
137 for (i = 0; incr < delta - incr; i++)
138 incr = incr << 1;
139
140 for (; i >= 0; incr >>= 1, i--) {
141 if (delta < incr)
142 continue;
143
144 timer->it.cpu.node.expires += incr;
145 timer->it_overrun += 1LL << i;
146 delta -= incr;
147 }
148 return timer->it.cpu.node.expires;
149}
150
151/* Check whether all cache entries contain U64_MAX, i.e. eternal expiry time */
152static inline bool expiry_cache_is_inactive(const struct posix_cputimers *pct)
153{
154 return !(~pct->bases[CPUCLOCK_PROF].nextevt |
155 ~pct->bases[CPUCLOCK_VIRT].nextevt |
156 ~pct->bases[CPUCLOCK_SCHED].nextevt);
157}
158
159static int
160posix_cpu_clock_getres(const clockid_t which_clock, struct timespec64 *tp)
161{
162 int error = validate_clock_permissions(which_clock);
163
164 if (!error) {
165 tp->tv_sec = 0;
166 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
167 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
168 /*
169 * If sched_clock is using a cycle counter, we
170 * don't have any idea of its true resolution
171 * exported, but it is much more than 1s/HZ.
172 */
173 tp->tv_nsec = 1;
174 }
175 }
176 return error;
177}
178
179static int
180posix_cpu_clock_set(const clockid_t clock, const struct timespec64 *tp)
181{
182 int error = validate_clock_permissions(clock);
183
184 /*
185 * You can never reset a CPU clock, but we check for other errors
186 * in the call before failing with EPERM.
187 */
188 return error ? : -EPERM;
189}
190
191/*
192 * Sample a per-thread clock for the given task. clkid is validated.
193 */
194static u64 cpu_clock_sample(const clockid_t clkid, struct task_struct *p)
195{
196 u64 utime, stime;
197
198 if (clkid == CPUCLOCK_SCHED)
199 return task_sched_runtime(p);
200
201 task_cputime(p, &utime, &stime);
202
203 switch (clkid) {
204 case CPUCLOCK_PROF:
205 return utime + stime;
206 case CPUCLOCK_VIRT:
207 return utime;
208 default:
209 WARN_ON_ONCE(1);
210 }
211 return 0;
212}
213
214static inline void store_samples(u64 *samples, u64 stime, u64 utime, u64 rtime)
215{
216 samples[CPUCLOCK_PROF] = stime + utime;
217 samples[CPUCLOCK_VIRT] = utime;
218 samples[CPUCLOCK_SCHED] = rtime;
219}
220
221static void task_sample_cputime(struct task_struct *p, u64 *samples)
222{
223 u64 stime, utime;
224
225 task_cputime(p, &utime, &stime);
226 store_samples(samples, stime, utime, p->se.sum_exec_runtime);
227}
228
229static void proc_sample_cputime_atomic(struct task_cputime_atomic *at,
230 u64 *samples)
231{
232 u64 stime, utime, rtime;
233
234 utime = atomic64_read(&at->utime);
235 stime = atomic64_read(&at->stime);
236 rtime = atomic64_read(&at->sum_exec_runtime);
237 store_samples(samples, stime, utime, rtime);
238}
239
240/*
241 * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg
242 * to avoid race conditions with concurrent updates to cputime.
243 */
244static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime)
245{
246 u64 curr_cputime = atomic64_read(cputime);
247
248 do {
249 if (sum_cputime <= curr_cputime)
250 return;
251 } while (!atomic64_try_cmpxchg(cputime, &curr_cputime, sum_cputime));
252}
253
254static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic,
255 struct task_cputime *sum)
256{
257 __update_gt_cputime(&cputime_atomic->utime, sum->utime);
258 __update_gt_cputime(&cputime_atomic->stime, sum->stime);
259 __update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime);
260}
261
262/**
263 * thread_group_sample_cputime - Sample cputime for a given task
264 * @tsk: Task for which cputime needs to be started
265 * @samples: Storage for time samples
266 *
267 * Called from sys_getitimer() to calculate the expiry time of an active
268 * timer. That means group cputime accounting is already active. Called
269 * with task sighand lock held.
270 *
271 * Updates @times with an uptodate sample of the thread group cputimes.
272 */
273void thread_group_sample_cputime(struct task_struct *tsk, u64 *samples)
274{
275 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
276 struct posix_cputimers *pct = &tsk->signal->posix_cputimers;
277
278 WARN_ON_ONCE(!pct->timers_active);
279
280 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
281}
282
283/**
284 * thread_group_start_cputime - Start cputime and return a sample
285 * @tsk: Task for which cputime needs to be started
286 * @samples: Storage for time samples
287 *
288 * The thread group cputime accounting is avoided when there are no posix
289 * CPU timers armed. Before starting a timer it's required to check whether
290 * the time accounting is active. If not, a full update of the atomic
291 * accounting store needs to be done and the accounting enabled.
292 *
293 * Updates @times with an uptodate sample of the thread group cputimes.
294 */
295static void thread_group_start_cputime(struct task_struct *tsk, u64 *samples)
296{
297 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
298 struct posix_cputimers *pct = &tsk->signal->posix_cputimers;
299
300 lockdep_assert_task_sighand_held(tsk);
301
302 /* Check if cputimer isn't running. This is accessed without locking. */
303 if (!READ_ONCE(pct->timers_active)) {
304 struct task_cputime sum;
305
306 /*
307 * The POSIX timer interface allows for absolute time expiry
308 * values through the TIMER_ABSTIME flag, therefore we have
309 * to synchronize the timer to the clock every time we start it.
310 */
311 thread_group_cputime(tsk, &sum);
312 update_gt_cputime(&cputimer->cputime_atomic, &sum);
313
314 /*
315 * We're setting timers_active without a lock. Ensure this
316 * only gets written to in one operation. We set it after
317 * update_gt_cputime() as a small optimization, but
318 * barriers are not required because update_gt_cputime()
319 * can handle concurrent updates.
320 */
321 WRITE_ONCE(pct->timers_active, true);
322 }
323 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
324}
325
326static void __thread_group_cputime(struct task_struct *tsk, u64 *samples)
327{
328 struct task_cputime ct;
329
330 thread_group_cputime(tsk, &ct);
331 store_samples(samples, ct.stime, ct.utime, ct.sum_exec_runtime);
332}
333
334/*
335 * Sample a process (thread group) clock for the given task clkid. If the
336 * group's cputime accounting is already enabled, read the atomic
337 * store. Otherwise a full update is required. clkid is already validated.
338 */
339static u64 cpu_clock_sample_group(const clockid_t clkid, struct task_struct *p,
340 bool start)
341{
342 struct thread_group_cputimer *cputimer = &p->signal->cputimer;
343 struct posix_cputimers *pct = &p->signal->posix_cputimers;
344 u64 samples[CPUCLOCK_MAX];
345
346 if (!READ_ONCE(pct->timers_active)) {
347 if (start)
348 thread_group_start_cputime(p, samples);
349 else
350 __thread_group_cputime(p, samples);
351 } else {
352 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
353 }
354
355 return samples[clkid];
356}
357
358static int posix_cpu_clock_get(const clockid_t clock, struct timespec64 *tp)
359{
360 const clockid_t clkid = CPUCLOCK_WHICH(clock);
361 struct task_struct *tsk;
362 u64 t;
363
364 rcu_read_lock();
365 tsk = pid_task(pid_for_clock(clock, true), clock_pid_type(clock));
366 if (!tsk) {
367 rcu_read_unlock();
368 return -EINVAL;
369 }
370
371 if (CPUCLOCK_PERTHREAD(clock))
372 t = cpu_clock_sample(clkid, tsk);
373 else
374 t = cpu_clock_sample_group(clkid, tsk, false);
375 rcu_read_unlock();
376
377 *tp = ns_to_timespec64(t);
378 return 0;
379}
380
381/*
382 * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
383 * This is called from sys_timer_create() and do_cpu_nanosleep() with the
384 * new timer already all-zeros initialized.
385 */
386static int posix_cpu_timer_create(struct k_itimer *new_timer)
387{
388 static struct lock_class_key posix_cpu_timers_key;
389 struct pid *pid;
390
391 rcu_read_lock();
392 pid = pid_for_clock(new_timer->it_clock, false);
393 if (!pid) {
394 rcu_read_unlock();
395 return -EINVAL;
396 }
397
398 /*
399 * If posix timer expiry is handled in task work context then
400 * timer::it_lock can be taken without disabling interrupts as all
401 * other locking happens in task context. This requires a separate
402 * lock class key otherwise regular posix timer expiry would record
403 * the lock class being taken in interrupt context and generate a
404 * false positive warning.
405 */
406 if (IS_ENABLED(CONFIG_POSIX_CPU_TIMERS_TASK_WORK))
407 lockdep_set_class(&new_timer->it_lock, &posix_cpu_timers_key);
408
409 new_timer->kclock = &clock_posix_cpu;
410 timerqueue_init(&new_timer->it.cpu.node);
411 INIT_LIST_HEAD(&new_timer->it.cpu.elist);
412 new_timer->it.cpu.pid = get_pid(pid);
413 rcu_read_unlock();
414 return 0;
415}
416
417static struct posix_cputimer_base *timer_base(struct k_itimer *timer,
418 struct task_struct *tsk)
419{
420 int clkidx = CPUCLOCK_WHICH(timer->it_clock);
421
422 if (CPUCLOCK_PERTHREAD(timer->it_clock))
423 return tsk->posix_cputimers.bases + clkidx;
424 else
425 return tsk->signal->posix_cputimers.bases + clkidx;
426}
427
428/*
429 * Force recalculating the base earliest expiration on the next tick.
430 * This will also re-evaluate the need to keep around the process wide
431 * cputime counter and tick dependency and eventually shut these down
432 * if necessary.
433 */
434static void trigger_base_recalc_expires(struct k_itimer *timer,
435 struct task_struct *tsk)
436{
437 struct posix_cputimer_base *base = timer_base(timer, tsk);
438
439 base->nextevt = 0;
440}
441
442/*
443 * Dequeue the timer and reset the base if it was its earliest expiration.
444 * It makes sure the next tick recalculates the base next expiration so we
445 * don't keep the costly process wide cputime counter around for a random
446 * amount of time, along with the tick dependency.
447 *
448 * If another timer gets queued between this and the next tick, its
449 * expiration will update the base next event if necessary on the next
450 * tick.
451 */
452static void disarm_timer(struct k_itimer *timer, struct task_struct *p)
453{
454 struct cpu_timer *ctmr = &timer->it.cpu;
455 struct posix_cputimer_base *base;
456
457 if (!cpu_timer_dequeue(ctmr))
458 return;
459
460 base = timer_base(timer, p);
461 if (cpu_timer_getexpires(ctmr) == base->nextevt)
462 trigger_base_recalc_expires(timer, p);
463}
464
465/*
466 * Lookup the task via timer->it.cpu.pid and attempt to lock the task's sighand.
467 *
468 * This can race with the reaping of the task:
469 *
470 * CPU0 CPU1
471 *
472 * // Finds task
473 * p = pid_task(pid, pid_type); __exit_signal(p)
474 * lock(p, sighand);
475 * posix_cpu_timers*_exit();
476 * sighand = lock_task_sighand(p); unhash_task(p);
477 * p->sighand = NULL;
478 * unlock(sighand);
479 *
480 * In this case sighand is NULL, which means the task and the associated timer
481 * queue cannot be longer accessed safely.
482 *
483 * __exit_signal() invokes posix_cpu_timers_exit() and if the thread group is
484 * dead it also invokes posix_cpu_timers_group_exit(). These functions delete
485 * all pending timers from the related timer queues. The POSIX timers (k_itimer)
486 * themself are still accessible, but not longer connected to the task.
487 *
488 * exec() works slightly differently. The task which exec()'s terminates all
489 * other threads in the thread group and runs __exit_signal() on them. As the
490 * thread group is not dead they only clean up the per task timers via
491 * posix_cpu_timers_exit().
492 *
493 * As the TGID on exec() stays the same per process timers stay queued, if they
494 * are armed. This works without a problem when exec() is done by the thread
495 * group leader. If a non-leader thread exec()'s this can end up in the
496 * following scenario:
497 *
498 * CPU0 CPU1
499 * // Returns old leader
500 * p = pid_task(pid, pid_type); de_thread()
501 * switch_leader()
502 * release_task(old leader)
503 * __exit_signal()
504 * old_leader->sighand = NULL;
505 * // Returns NULL
506 * sighand = lock_task_sighand(p)
507 *
508 * That's problematic for several functions:
509 *
510 * - posix_cpu_timer_del(): If the timer is still enqueued on the task the
511 * underlying k_itimer will be freed which results in a UAF in
512 * run_posix_cpu_timers() or on timerqueue related add/delete operations.
513 * If the timer is not enqueued, the failure is harmless
514 *
515 * - posix_cpu_timer_set(): Independent of the enqueued state that results in a
516 * transient failure which is user space visible (-ESRCH) for regular posix
517 * timers. But for the use case in do_cpu_nanosleep() it's the same UAF
518 * problem just that the timer is allocated on the stack.
519 *
520 * - posix_cpu_timer_rearm(): Timer is not enqueued at that point, but this
521 * silently ignores the rearm request, which is a functional problem as the
522 * timer wont expire anymore.
523 */
524static struct task_struct *timer_lock_sighand(struct k_itimer *timer, unsigned long *flags)
525{
526 enum pid_type type = clock_pid_type(timer->it_clock);
527 struct cpu_timer *ctmr = &timer->it.cpu;
528
529 guard(rcu)();
530
531 for (;;) {
532 struct task_struct *t = pid_task(timer->it.cpu.pid, type);
533
534 /* Fail if the task cannot be found. */
535 if (!t)
536 break;
537
538 /* Try to lock the task's sighand */
539 if (lock_task_sighand(t, flags))
540 return t;
541
542 /*
543 * The next PID lookup might either fail or return the new
544 * leader. This is correct for both exit() and exec().
545 */
546 }
547
548 /*
549 * If the timer is still enqueued, warn. There is nothing safe to do
550 * here as there might be two timers in there which are removed in
551 * parallel and that will cause more damage than good. This should never
552 * happen!
553 *
554 * Ensure that the stores to the timer and timerqueue are visible:
555 *
556 * __exit_signal()
557 * posix_cpu_timers*_exit()
558 * write_seqlock(seqlock)
559 * smp_wmb(); <-------
560 * __unhash_process() | !pid_task()
561 * ----> smp_rmb();
562 * WARN_ON_ONCE(...)
563 */
564 smp_rmb();
565 WARN_ON_ONCE(ctmr->head || timerqueue_node_queued(&ctmr->node));
566 return NULL;
567}
568
569/*
570 * If the timer is queued on the expiry list, then it cannot be dequeued because
571 * the firing list is not protected by sighand->lock. The delivery path is
572 * waiting for the timer lock. So go back, unlock and retry.
573 */
574static bool posix_cpu_timer_on_expiry_list(struct k_itimer *timer)
575{
576 if (list_empty(&timer->it.cpu.elist))
577 return false;
578
579 /*
580 * Prevent signal delivery as there is no point in delivering a signal
581 * which is made obsolete right away.
582 */
583 timer->it.cpu.firing = false;
584 return true;
585}
586
587/*
588 * Clean up a CPU-clock timer that is about to be destroyed.
589 * This is called from timer deletion with the timer already locked.
590 * If we return TIMER_RETRY, it's necessary to release the timer's lock
591 * and try again. (This happens when the timer is in the middle of firing.)
592 */
593static int posix_cpu_timer_del(struct k_itimer *timer)
594{
595 struct task_struct *p;
596 unsigned long flags;
597 int ret = 0;
598
599 p = timer_lock_sighand(timer, &flags);
600
601 if (likely(p)) {
602 if (posix_cpu_timer_on_expiry_list(timer))
603 ret = TIMER_RETRY;
604 else
605 disarm_timer(timer, p);
606 unlock_task_sighand(p, &flags);
607 }
608
609 if (!ret) {
610 put_pid(timer->it.cpu.pid);
611 timer->it_status = POSIX_TIMER_DISARMED;
612 }
613 return ret;
614}
615
616static void cleanup_timerqueue(struct timerqueue_head *head)
617{
618 struct timerqueue_node *node;
619 struct cpu_timer *ctmr;
620
621 while ((node = timerqueue_getnext(head))) {
622 timerqueue_del(head, node);
623 ctmr = container_of(node, struct cpu_timer, node);
624 ctmr->head = NULL;
625 }
626}
627
628/*
629 * Clean out CPU timers which are still armed when a thread exits. The
630 * timers are only removed from the list. No other updates are done. The
631 * corresponding posix timers are still accessible, but cannot be rearmed.
632 *
633 * This must be called with the siglock held.
634 */
635static void cleanup_timers(struct posix_cputimers *pct)
636{
637 cleanup_timerqueue(&pct->bases[CPUCLOCK_PROF].tqhead);
638 cleanup_timerqueue(&pct->bases[CPUCLOCK_VIRT].tqhead);
639 cleanup_timerqueue(&pct->bases[CPUCLOCK_SCHED].tqhead);
640}
641
642/*
643 * These are both called with the siglock held, when the current thread
644 * is being reaped. When the final (leader) thread in the group is reaped,
645 * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
646 */
647void posix_cpu_timers_exit(struct task_struct *tsk)
648{
649 cleanup_timers(&tsk->posix_cputimers);
650}
651void posix_cpu_timers_exit_group(struct task_struct *tsk)
652{
653 cleanup_timers(&tsk->signal->posix_cputimers);
654}
655
656/*
657 * Insert the timer on the appropriate list before any timers that
658 * expire later. This must be called with the sighand lock held.
659 */
660static void arm_timer(struct k_itimer *timer, struct task_struct *p)
661{
662 struct posix_cputimer_base *base = timer_base(timer, p);
663 struct cpu_timer *ctmr = &timer->it.cpu;
664 u64 newexp = cpu_timer_getexpires(ctmr);
665
666 timer->it_status = POSIX_TIMER_ARMED;
667 if (!cpu_timer_enqueue(&base->tqhead, ctmr))
668 return;
669
670 /*
671 * We are the new earliest-expiring POSIX 1.b timer, hence
672 * need to update expiration cache. Take into account that
673 * for process timers we share expiration cache with itimers
674 * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
675 */
676 if (newexp < base->nextevt)
677 base->nextevt = newexp;
678
679 if (CPUCLOCK_PERTHREAD(timer->it_clock))
680 tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER);
681 else
682 tick_dep_set_signal(p, TICK_DEP_BIT_POSIX_TIMER);
683}
684
685/*
686 * The timer is locked, fire it and arrange for its reload.
687 */
688static void cpu_timer_fire(struct k_itimer *timer)
689{
690 struct cpu_timer *ctmr = &timer->it.cpu;
691
692 timer->it_status = POSIX_TIMER_DISARMED;
693
694 if (unlikely(ctmr->nanosleep)) {
695 /*
696 * This a special case for clock_nanosleep,
697 * not a normal timer from sys_timer_create.
698 */
699 wake_up_process(timer->it_process);
700 cpu_timer_setexpires(ctmr, 0);
701 } else {
702 posix_timer_queue_signal(timer);
703 /* Disable oneshot timers */
704 if (!timer->it_interval)
705 cpu_timer_setexpires(ctmr, 0);
706 }
707}
708
709static void __posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp, u64 now);
710
711/*
712 * Guts of sys_timer_settime for CPU timers.
713 * This is called with the timer locked and interrupts disabled.
714 * If we return TIMER_RETRY, it's necessary to release the timer's lock
715 * and try again. (This happens when the timer is in the middle of firing.)
716 */
717static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
718 struct itimerspec64 *new, struct itimerspec64 *old)
719{
720 bool sigev_none = timer->it_sigev_notify == SIGEV_NONE;
721 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
722 struct cpu_timer *ctmr = &timer->it.cpu;
723 u64 old_expires, new_expires, now;
724 struct task_struct *p;
725 unsigned long flags;
726 int ret = 0;
727
728 p = timer_lock_sighand(timer, &flags);
729 /*
730 * If p has just been reaped, we can no longer get any information about
731 * it at all.
732 */
733 if (!p)
734 return -ESRCH;
735
736 /*
737 * Use the to_ktime conversion because that clamps the maximum
738 * value to KTIME_MAX and avoid multiplication overflows.
739 */
740 new_expires = ktime_to_ns(timespec64_to_ktime(new->it_value));
741
742 /* Retrieve the current expiry time before disarming the timer */
743 old_expires = cpu_timer_getexpires(ctmr);
744
745 if (posix_cpu_timer_on_expiry_list(timer)) {
746 ret = TIMER_RETRY;
747 } else {
748 cpu_timer_dequeue(ctmr);
749 timer->it_status = POSIX_TIMER_DISARMED;
750 }
751
752 /*
753 * Sample the current clock for saving the previous setting
754 * and for rearming the timer.
755 */
756 if (CPUCLOCK_PERTHREAD(timer->it_clock))
757 now = cpu_clock_sample(clkid, p);
758 else
759 now = cpu_clock_sample_group(clkid, p, !sigev_none);
760
761 /* Retrieve the previous expiry value if requested. */
762 if (old) {
763 old->it_value = (struct timespec64){ };
764 if (old_expires)
765 __posix_cpu_timer_get(timer, old, now);
766 }
767
768 /* Retry if the timer expiry is running concurrently */
769 if (unlikely(ret)) {
770 unlock_task_sighand(p, &flags);
771 return ret;
772 }
773
774 /* Convert relative expiry time to absolute */
775 if (new_expires && !(timer_flags & TIMER_ABSTIME))
776 new_expires += now;
777
778 /* Set the new expiry time (might be 0) */
779 cpu_timer_setexpires(ctmr, new_expires);
780
781 /*
782 * Arm the timer if it is not disabled, the new expiry value has
783 * not yet expired and the timer requires signal delivery.
784 * SIGEV_NONE timers are never armed. In case the timer is not
785 * armed, enforce the reevaluation of the timer base so that the
786 * process wide cputime counter can be disabled eventually.
787 */
788 if (likely(!sigev_none)) {
789 if (new_expires && now < new_expires)
790 arm_timer(timer, p);
791 else
792 trigger_base_recalc_expires(timer, p);
793 }
794
795 unlock_task_sighand(p, &flags);
796
797 posix_timer_set_common(timer, new);
798
799 /*
800 * If the new expiry time was already in the past the timer was not
801 * queued. Fire it immediately even if the thread never runs to
802 * accumulate more time on this clock.
803 */
804 if (!sigev_none && new_expires && now >= new_expires)
805 cpu_timer_fire(timer);
806 return ret;
807}
808
809static void __posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp, u64 now)
810{
811 bool sigev_none = timer->it_sigev_notify == SIGEV_NONE;
812 u64 expires, iv = timer->it_interval;
813
814 /*
815 * Make sure that interval timers are moved forward for the
816 * following cases:
817 * - SIGEV_NONE timers which are never armed
818 * - Timers which expired, but the signal has not yet been
819 * delivered
820 */
821 if (iv && timer->it_status != POSIX_TIMER_ARMED)
822 expires = bump_cpu_timer(timer, now);
823 else
824 expires = cpu_timer_getexpires(&timer->it.cpu);
825
826 /*
827 * Expired interval timers cannot have a remaining time <= 0.
828 * The kernel has to move them forward so that the next
829 * timer expiry is > @now.
830 */
831 if (now < expires) {
832 itp->it_value = ns_to_timespec64(expires - now);
833 } else {
834 /*
835 * A single shot SIGEV_NONE timer must return 0, when it is
836 * expired! Timers which have a real signal delivery mode
837 * must return a remaining time greater than 0 because the
838 * signal has not yet been delivered.
839 */
840 if (!sigev_none)
841 itp->it_value.tv_nsec = 1;
842 }
843}
844
845static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp)
846{
847 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
848 struct task_struct *p;
849 u64 now;
850
851 rcu_read_lock();
852 p = cpu_timer_task_rcu(timer);
853 if (p && cpu_timer_getexpires(&timer->it.cpu)) {
854 itp->it_interval = ktime_to_timespec64(timer->it_interval);
855
856 if (CPUCLOCK_PERTHREAD(timer->it_clock))
857 now = cpu_clock_sample(clkid, p);
858 else
859 now = cpu_clock_sample_group(clkid, p, false);
860
861 __posix_cpu_timer_get(timer, itp, now);
862 }
863 rcu_read_unlock();
864}
865
866#define MAX_COLLECTED 20
867
868static u64 collect_timerqueue(struct timerqueue_head *head,
869 struct list_head *firing, u64 now)
870{
871 struct timerqueue_node *next;
872 int i = 0;
873
874 while ((next = timerqueue_getnext(head))) {
875 struct cpu_timer *ctmr;
876 u64 expires;
877
878 ctmr = container_of(next, struct cpu_timer, node);
879 expires = cpu_timer_getexpires(ctmr);
880 /* Limit the number of timers to expire at once */
881 if (++i == MAX_COLLECTED || now < expires)
882 return expires;
883
884 ctmr->firing = true;
885 /* See posix_cpu_timer_wait_running() */
886 rcu_assign_pointer(ctmr->handling, current);
887 cpu_timer_dequeue(ctmr);
888 list_add_tail(&ctmr->elist, firing);
889 }
890
891 return U64_MAX;
892}
893
894static void collect_posix_cputimers(struct posix_cputimers *pct, u64 *samples,
895 struct list_head *firing)
896{
897 struct posix_cputimer_base *base = pct->bases;
898 int i;
899
900 for (i = 0; i < CPUCLOCK_MAX; i++, base++) {
901 base->nextevt = collect_timerqueue(&base->tqhead, firing,
902 samples[i]);
903 }
904}
905
906static inline void check_dl_overrun(struct task_struct *tsk)
907{
908 if (tsk->dl.dl_overrun) {
909 tsk->dl.dl_overrun = 0;
910 send_signal_locked(SIGXCPU, SEND_SIG_PRIV, tsk, PIDTYPE_TGID);
911 }
912}
913
914static bool check_rlimit(u64 time, u64 limit, int signo, bool rt, bool hard)
915{
916 if (time < limit)
917 return false;
918
919 if (print_fatal_signals) {
920 pr_info("%s Watchdog Timeout (%s): %s[%d]\n",
921 rt ? "RT" : "CPU", hard ? "hard" : "soft",
922 current->comm, task_pid_nr(current));
923 }
924 send_signal_locked(signo, SEND_SIG_PRIV, current, PIDTYPE_TGID);
925 return true;
926}
927
928/*
929 * Check for any per-thread CPU timers that have fired and move them off
930 * the tsk->cpu_timers[N] list onto the firing list. Here we update the
931 * tsk->it_*_expires values to reflect the remaining thread CPU timers.
932 */
933static void check_thread_timers(struct task_struct *tsk,
934 struct list_head *firing)
935{
936 struct posix_cputimers *pct = &tsk->posix_cputimers;
937 u64 samples[CPUCLOCK_MAX];
938 unsigned long soft;
939
940 if (dl_task(tsk))
941 check_dl_overrun(tsk);
942
943 if (expiry_cache_is_inactive(pct))
944 return;
945
946 task_sample_cputime(tsk, samples);
947 collect_posix_cputimers(pct, samples, firing);
948
949 /*
950 * Check for the special case thread timers.
951 */
952 soft = task_rlimit(tsk, RLIMIT_RTTIME);
953 if (soft != RLIM_INFINITY) {
954 /* Task RT timeout is accounted in jiffies. RTTIME is usec */
955 unsigned long rttime = tsk->rt.timeout * (USEC_PER_SEC / HZ);
956 unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME);
957
958 /* At the hard limit, send SIGKILL. No further action. */
959 if (hard != RLIM_INFINITY &&
960 check_rlimit(rttime, hard, SIGKILL, true, true))
961 return;
962
963 /* At the soft limit, send a SIGXCPU every second */
964 if (check_rlimit(rttime, soft, SIGXCPU, true, false)) {
965 soft += USEC_PER_SEC;
966 tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur = soft;
967 }
968 }
969
970 if (expiry_cache_is_inactive(pct))
971 tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER);
972}
973
974static inline void stop_process_timers(struct signal_struct *sig)
975{
976 struct posix_cputimers *pct = &sig->posix_cputimers;
977
978 /* Turn off the active flag. This is done without locking. */
979 WRITE_ONCE(pct->timers_active, false);
980 tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER);
981}
982
983static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
984 u64 *expires, u64 cur_time, int signo)
985{
986 if (!it->expires)
987 return;
988
989 if (cur_time >= it->expires) {
990 if (it->incr)
991 it->expires += it->incr;
992 else
993 it->expires = 0;
994
995 trace_itimer_expire(signo == SIGPROF ?
996 ITIMER_PROF : ITIMER_VIRTUAL,
997 task_tgid(tsk), cur_time);
998 send_signal_locked(signo, SEND_SIG_PRIV, tsk, PIDTYPE_TGID);
999 }
1000
1001 if (it->expires && it->expires < *expires)
1002 *expires = it->expires;
1003}
1004
1005/*
1006 * Check for any per-thread CPU timers that have fired and move them
1007 * off the tsk->*_timers list onto the firing list. Per-thread timers
1008 * have already been taken off.
1009 */
1010static void check_process_timers(struct task_struct *tsk,
1011 struct list_head *firing)
1012{
1013 struct signal_struct *const sig = tsk->signal;
1014 struct posix_cputimers *pct = &sig->posix_cputimers;
1015 u64 samples[CPUCLOCK_MAX];
1016 unsigned long soft;
1017
1018 /*
1019 * If there are no active process wide timers (POSIX 1.b, itimers,
1020 * RLIMIT_CPU) nothing to check. Also skip the process wide timer
1021 * processing when there is already another task handling them.
1022 */
1023 if (!READ_ONCE(pct->timers_active) || pct->expiry_active)
1024 return;
1025
1026 /*
1027 * Signify that a thread is checking for process timers.
1028 * Write access to this field is protected by the sighand lock.
1029 */
1030 pct->expiry_active = true;
1031
1032 /*
1033 * Collect the current process totals. Group accounting is active
1034 * so the sample can be taken directly.
1035 */
1036 proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic, samples);
1037 collect_posix_cputimers(pct, samples, firing);
1038
1039 /*
1040 * Check for the special case process timers.
1041 */
1042 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF],
1043 &pct->bases[CPUCLOCK_PROF].nextevt,
1044 samples[CPUCLOCK_PROF], SIGPROF);
1045 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT],
1046 &pct->bases[CPUCLOCK_VIRT].nextevt,
1047 samples[CPUCLOCK_VIRT], SIGVTALRM);
1048
1049 soft = task_rlimit(tsk, RLIMIT_CPU);
1050 if (soft != RLIM_INFINITY) {
1051 /* RLIMIT_CPU is in seconds. Samples are nanoseconds */
1052 unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU);
1053 u64 ptime = samples[CPUCLOCK_PROF];
1054 u64 softns = (u64)soft * NSEC_PER_SEC;
1055 u64 hardns = (u64)hard * NSEC_PER_SEC;
1056
1057 /* At the hard limit, send SIGKILL. No further action. */
1058 if (hard != RLIM_INFINITY &&
1059 check_rlimit(ptime, hardns, SIGKILL, false, true))
1060 return;
1061
1062 /* At the soft limit, send a SIGXCPU every second */
1063 if (check_rlimit(ptime, softns, SIGXCPU, false, false)) {
1064 sig->rlim[RLIMIT_CPU].rlim_cur = soft + 1;
1065 softns += NSEC_PER_SEC;
1066 }
1067
1068 /* Update the expiry cache */
1069 if (softns < pct->bases[CPUCLOCK_PROF].nextevt)
1070 pct->bases[CPUCLOCK_PROF].nextevt = softns;
1071 }
1072
1073 if (expiry_cache_is_inactive(pct))
1074 stop_process_timers(sig);
1075
1076 pct->expiry_active = false;
1077}
1078
1079/*
1080 * This is called from the signal code (via posixtimer_rearm)
1081 * when the last timer signal was delivered and we have to reload the timer.
1082 *
1083 * Return true unconditionally so the core code assumes the timer to be
1084 * armed. Otherwise it would requeue the signal.
1085 */
1086static bool posix_cpu_timer_rearm(struct k_itimer *timer)
1087{
1088 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
1089 struct task_struct *p;
1090 unsigned long flags;
1091 u64 now;
1092
1093 p = timer_lock_sighand(timer, &flags);
1094 if (unlikely(!p))
1095 return true;
1096
1097 /*
1098 * Fetch the current sample and update the timer's expiry time.
1099 */
1100 if (CPUCLOCK_PERTHREAD(timer->it_clock))
1101 now = cpu_clock_sample(clkid, p);
1102 else
1103 now = cpu_clock_sample_group(clkid, p, true);
1104
1105 bump_cpu_timer(timer, now);
1106
1107 /*
1108 * Now re-arm for the new expiry time.
1109 */
1110 arm_timer(timer, p);
1111 unlock_task_sighand(p, &flags);
1112 return true;
1113}
1114
1115/**
1116 * task_cputimers_expired - Check whether posix CPU timers are expired
1117 *
1118 * @samples: Array of current samples for the CPUCLOCK clocks
1119 * @pct: Pointer to a posix_cputimers container
1120 *
1121 * Returns true if any member of @samples is greater than the corresponding
1122 * member of @pct->bases[CLK].nextevt. False otherwise
1123 */
1124static inline bool
1125task_cputimers_expired(const u64 *samples, struct posix_cputimers *pct)
1126{
1127 int i;
1128
1129 for (i = 0; i < CPUCLOCK_MAX; i++) {
1130 if (samples[i] >= pct->bases[i].nextevt)
1131 return true;
1132 }
1133 return false;
1134}
1135
1136/**
1137 * fastpath_timer_check - POSIX CPU timers fast path.
1138 *
1139 * @tsk: The task (thread) being checked.
1140 *
1141 * Check the task and thread group timers. If both are zero (there are no
1142 * timers set) return false. Otherwise snapshot the task and thread group
1143 * timers and compare them with the corresponding expiration times. Return
1144 * true if a timer has expired, else return false.
1145 */
1146static inline bool fastpath_timer_check(struct task_struct *tsk)
1147{
1148 struct posix_cputimers *pct = &tsk->posix_cputimers;
1149 struct signal_struct *sig;
1150
1151 if (!expiry_cache_is_inactive(pct)) {
1152 u64 samples[CPUCLOCK_MAX];
1153
1154 task_sample_cputime(tsk, samples);
1155 if (task_cputimers_expired(samples, pct))
1156 return true;
1157 }
1158
1159 sig = tsk->signal;
1160 pct = &sig->posix_cputimers;
1161 /*
1162 * Check if thread group timers expired when timers are active and
1163 * no other thread in the group is already handling expiry for
1164 * thread group cputimers. These fields are read without the
1165 * sighand lock. However, this is fine because this is meant to be
1166 * a fastpath heuristic to determine whether we should try to
1167 * acquire the sighand lock to handle timer expiry.
1168 *
1169 * In the worst case scenario, if concurrently timers_active is set
1170 * or expiry_active is cleared, but the current thread doesn't see
1171 * the change yet, the timer checks are delayed until the next
1172 * thread in the group gets a scheduler interrupt to handle the
1173 * timer. This isn't an issue in practice because these types of
1174 * delays with signals actually getting sent are expected.
1175 */
1176 if (READ_ONCE(pct->timers_active) && !READ_ONCE(pct->expiry_active)) {
1177 u64 samples[CPUCLOCK_MAX];
1178
1179 proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic,
1180 samples);
1181
1182 if (task_cputimers_expired(samples, pct))
1183 return true;
1184 }
1185
1186 if (dl_task(tsk) && tsk->dl.dl_overrun)
1187 return true;
1188
1189 return false;
1190}
1191
1192static void handle_posix_cpu_timers(struct task_struct *tsk);
1193
1194#ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
1195static void posix_cpu_timers_work(struct callback_head *work)
1196{
1197 struct posix_cputimers_work *cw = container_of(work, typeof(*cw), work);
1198
1199 mutex_lock(&cw->mutex);
1200 handle_posix_cpu_timers(current);
1201 mutex_unlock(&cw->mutex);
1202}
1203
1204/*
1205 * Invoked from the posix-timer core when a cancel operation failed because
1206 * the timer is marked firing. The caller holds rcu_read_lock(), which
1207 * protects the timer and the task which is expiring it from being freed.
1208 */
1209static void posix_cpu_timer_wait_running(struct k_itimer *timr)
1210{
1211 struct task_struct *tsk = rcu_dereference(timr->it.cpu.handling);
1212
1213 /* Has the handling task completed expiry already? */
1214 if (!tsk)
1215 return;
1216
1217 /* Ensure that the task cannot go away */
1218 get_task_struct(tsk);
1219 /* Now drop the RCU protection so the mutex can be locked */
1220 rcu_read_unlock();
1221 /* Wait on the expiry mutex */
1222 mutex_lock(&tsk->posix_cputimers_work.mutex);
1223 /* Release it immediately again. */
1224 mutex_unlock(&tsk->posix_cputimers_work.mutex);
1225 /* Drop the task reference. */
1226 put_task_struct(tsk);
1227 /* Relock RCU so the callsite is balanced */
1228 rcu_read_lock();
1229}
1230
1231static void posix_cpu_timer_wait_running_nsleep(struct k_itimer *timr)
1232{
1233 /* Ensure that timr->it.cpu.handling task cannot go away */
1234 rcu_read_lock();
1235 spin_unlock_irq(&timr->it_lock);
1236 posix_cpu_timer_wait_running(timr);
1237 rcu_read_unlock();
1238 /* @timr is on stack and is valid */
1239 spin_lock_irq(&timr->it_lock);
1240}
1241
1242/*
1243 * Clear existing posix CPU timers task work.
1244 */
1245void clear_posix_cputimers_work(struct task_struct *p)
1246{
1247 /*
1248 * A copied work entry from the old task is not meaningful, clear it.
1249 * N.B. init_task_work will not do this.
1250 */
1251 memset(&p->posix_cputimers_work.work, 0,
1252 sizeof(p->posix_cputimers_work.work));
1253 init_task_work(&p->posix_cputimers_work.work,
1254 posix_cpu_timers_work);
1255 mutex_init(&p->posix_cputimers_work.mutex);
1256 p->posix_cputimers_work.scheduled = false;
1257}
1258
1259/*
1260 * Initialize posix CPU timers task work in init task. Out of line to
1261 * keep the callback static and to avoid header recursion hell.
1262 */
1263void __init posix_cputimers_init_work(void)
1264{
1265 clear_posix_cputimers_work(current);
1266}
1267
1268/*
1269 * Note: All operations on tsk->posix_cputimer_work.scheduled happen either
1270 * in hard interrupt context or in task context with interrupts
1271 * disabled. Aside of that the writer/reader interaction is always in the
1272 * context of the current task, which means they are strict per CPU.
1273 */
1274static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk)
1275{
1276 return tsk->posix_cputimers_work.scheduled;
1277}
1278
1279static inline void __run_posix_cpu_timers(struct task_struct *tsk)
1280{
1281 if (WARN_ON_ONCE(tsk->posix_cputimers_work.scheduled))
1282 return;
1283
1284 /* Schedule task work to actually expire the timers */
1285 tsk->posix_cputimers_work.scheduled = true;
1286 task_work_add(tsk, &tsk->posix_cputimers_work.work, TWA_RESUME);
1287}
1288
1289static inline bool posix_cpu_timers_enable_work(struct task_struct *tsk,
1290 unsigned long start)
1291{
1292 bool ret = true;
1293
1294 /*
1295 * On !RT kernels interrupts are disabled while collecting expired
1296 * timers, so no tick can happen and the fast path check can be
1297 * reenabled without further checks.
1298 */
1299 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
1300 tsk->posix_cputimers_work.scheduled = false;
1301 return true;
1302 }
1303
1304 /*
1305 * On RT enabled kernels ticks can happen while the expired timers
1306 * are collected under sighand lock. But any tick which observes
1307 * the CPUTIMERS_WORK_SCHEDULED bit set, does not run the fastpath
1308 * checks. So reenabling the tick work has do be done carefully:
1309 *
1310 * Disable interrupts and run the fast path check if jiffies have
1311 * advanced since the collecting of expired timers started. If
1312 * jiffies have not advanced or the fast path check did not find
1313 * newly expired timers, reenable the fast path check in the timer
1314 * interrupt. If there are newly expired timers, return false and
1315 * let the collection loop repeat.
1316 */
1317 local_irq_disable();
1318 if (start != jiffies && fastpath_timer_check(tsk))
1319 ret = false;
1320 else
1321 tsk->posix_cputimers_work.scheduled = false;
1322 local_irq_enable();
1323
1324 return ret;
1325}
1326#else /* CONFIG_POSIX_CPU_TIMERS_TASK_WORK */
1327static inline void __run_posix_cpu_timers(struct task_struct *tsk)
1328{
1329 lockdep_posixtimer_enter();
1330 handle_posix_cpu_timers(tsk);
1331 lockdep_posixtimer_exit();
1332}
1333
1334static void posix_cpu_timer_wait_running(struct k_itimer *timr)
1335{
1336 cpu_relax();
1337}
1338
1339static void posix_cpu_timer_wait_running_nsleep(struct k_itimer *timr)
1340{
1341 spin_unlock_irq(&timr->it_lock);
1342 cpu_relax();
1343 spin_lock_irq(&timr->it_lock);
1344}
1345
1346static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk)
1347{
1348 return false;
1349}
1350
1351static inline bool posix_cpu_timers_enable_work(struct task_struct *tsk,
1352 unsigned long start)
1353{
1354 return true;
1355}
1356#endif /* CONFIG_POSIX_CPU_TIMERS_TASK_WORK */
1357
1358static void handle_posix_cpu_timers(struct task_struct *tsk)
1359{
1360 struct k_itimer *timer, *next;
1361 unsigned long flags, start;
1362 LIST_HEAD(firing);
1363
1364 /*
1365 * tsk is current and ->sighand is stable, see the
1366 * tsk->exit_state check in run_posix_cpu_timers()
1367 */
1368 spin_lock_irqsave(&tsk->sighand->siglock, flags);
1369
1370 do {
1371 /*
1372 * On RT locking sighand lock does not disable interrupts,
1373 * so this needs to be careful vs. ticks. Store the current
1374 * jiffies value.
1375 */
1376 start = READ_ONCE(jiffies);
1377 barrier();
1378
1379 /*
1380 * Here we take off tsk->signal->cpu_timers[N] and
1381 * tsk->cpu_timers[N] all the timers that are firing, and
1382 * put them on the firing list.
1383 */
1384 check_thread_timers(tsk, &firing);
1385
1386 check_process_timers(tsk, &firing);
1387
1388 /*
1389 * The above timer checks have updated the expiry cache and
1390 * because nothing can have queued or modified timers after
1391 * sighand lock was taken above it is guaranteed to be
1392 * consistent. So the next timer interrupt fastpath check
1393 * will find valid data.
1394 *
1395 * If timer expiry runs in the timer interrupt context then
1396 * the loop is not relevant as timers will be directly
1397 * expired in interrupt context. The stub function below
1398 * returns always true which allows the compiler to
1399 * optimize the loop out.
1400 *
1401 * If timer expiry is deferred to task work context then
1402 * the following rules apply:
1403 *
1404 * - On !RT kernels no tick can have happened on this CPU
1405 * after sighand lock was acquired because interrupts are
1406 * disabled. So reenabling task work before dropping
1407 * sighand lock and reenabling interrupts is race free.
1408 *
1409 * - On RT kernels ticks might have happened but the tick
1410 * work ignored posix CPU timer handling because the
1411 * CPUTIMERS_WORK_SCHEDULED bit is set. Reenabling work
1412 * must be done very carefully including a check whether
1413 * ticks have happened since the start of the timer
1414 * expiry checks. posix_cpu_timers_enable_work() takes
1415 * care of that and eventually lets the expiry checks
1416 * run again.
1417 */
1418 } while (!posix_cpu_timers_enable_work(tsk, start));
1419
1420 /*
1421 * We must release sighand lock before taking any timer's lock.
1422 * There is a potential race with timer deletion here, as the
1423 * siglock now protects our private firing list. We have set
1424 * the firing flag in each timer, so that a deletion attempt
1425 * that gets the timer lock before we do will give it up and
1426 * spin until we've taken care of that timer below.
1427 */
1428 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
1429
1430 /*
1431 * Now that all the timers on our list have the firing flag,
1432 * no one will touch their list entries but us. We'll take
1433 * each timer's lock before clearing its firing flag, so no
1434 * timer call will interfere.
1435 */
1436 list_for_each_entry_safe(timer, next, &firing, it.cpu.elist) {
1437 bool cpu_firing;
1438
1439 /*
1440 * spin_lock() is sufficient here even independent of the
1441 * expiry context. If expiry happens in hard interrupt
1442 * context it's obvious. For task work context it's safe
1443 * because all other operations on timer::it_lock happen in
1444 * task context (syscall or exit).
1445 */
1446 spin_lock(&timer->it_lock);
1447 list_del_init(&timer->it.cpu.elist);
1448 cpu_firing = timer->it.cpu.firing;
1449 timer->it.cpu.firing = false;
1450 /*
1451 * If the firing flag is cleared then this raced with a
1452 * timer rearm/delete operation. So don't generate an
1453 * event.
1454 */
1455 if (likely(cpu_firing))
1456 cpu_timer_fire(timer);
1457 /* See posix_cpu_timer_wait_running() */
1458 rcu_assign_pointer(timer->it.cpu.handling, NULL);
1459 spin_unlock(&timer->it_lock);
1460 }
1461}
1462
1463/*
1464 * This is called from the timer interrupt handler. The irq handler has
1465 * already updated our counts. We need to check if any timers fire now.
1466 * Interrupts are disabled.
1467 */
1468void run_posix_cpu_timers(void)
1469{
1470 struct task_struct *tsk = current;
1471
1472 lockdep_assert_irqs_disabled();
1473
1474 /*
1475 * Ensure that release_task(tsk) can't happen while
1476 * handle_posix_cpu_timers() is running. Otherwise, a concurrent
1477 * posix_cpu_timer_del() may fail to lock_task_sighand(tsk) and
1478 * miss timer->it.cpu.firing != 0.
1479 */
1480 if (tsk->exit_state)
1481 return;
1482
1483 /*
1484 * If the actual expiry is deferred to task work context and the
1485 * work is already scheduled there is no point to do anything here.
1486 */
1487 if (posix_cpu_timers_work_scheduled(tsk))
1488 return;
1489
1490 /*
1491 * The fast path checks that there are no expired thread or thread
1492 * group timers. If that's so, just return.
1493 */
1494 if (!fastpath_timer_check(tsk))
1495 return;
1496
1497 __run_posix_cpu_timers(tsk);
1498}
1499
1500/*
1501 * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1502 * The tsk->sighand->siglock must be held by the caller.
1503 */
1504void set_process_cpu_timer(struct task_struct *tsk, unsigned int clkid,
1505 u64 *newval, u64 *oldval)
1506{
1507 u64 now, *nextevt;
1508
1509 if (WARN_ON_ONCE(clkid >= CPUCLOCK_SCHED))
1510 return;
1511
1512 nextevt = &tsk->signal->posix_cputimers.bases[clkid].nextevt;
1513 now = cpu_clock_sample_group(clkid, tsk, true);
1514
1515 if (oldval) {
1516 /*
1517 * We are setting itimer. The *oldval is absolute and we update
1518 * it to be relative, *newval argument is relative and we update
1519 * it to be absolute.
1520 */
1521 if (*oldval) {
1522 if (*oldval <= now) {
1523 /* Just about to fire. */
1524 *oldval = TICK_NSEC;
1525 } else {
1526 *oldval -= now;
1527 }
1528 }
1529
1530 if (*newval)
1531 *newval += now;
1532 }
1533
1534 /*
1535 * Update expiration cache if this is the earliest timer. CPUCLOCK_PROF
1536 * expiry cache is also used by RLIMIT_CPU!.
1537 */
1538 if (*newval < *nextevt)
1539 *nextevt = *newval;
1540
1541 tick_dep_set_signal(tsk, TICK_DEP_BIT_POSIX_TIMER);
1542}
1543
1544static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1545 const struct timespec64 *rqtp)
1546{
1547 struct itimerspec64 it;
1548 struct k_itimer timer;
1549 u64 expires;
1550 int error;
1551
1552 /*
1553 * Set up a temporary timer and then wait for it to go off.
1554 */
1555 memset(&timer, 0, sizeof timer);
1556 spin_lock_init(&timer.it_lock);
1557 timer.it_clock = which_clock;
1558 timer.it_overrun = -1;
1559 error = posix_cpu_timer_create(&timer);
1560 timer.it_process = current;
1561 timer.it.cpu.nanosleep = true;
1562
1563 if (!error) {
1564 static struct itimerspec64 zero_it;
1565 struct restart_block *restart;
1566
1567 memset(&it, 0, sizeof(it));
1568 it.it_value = *rqtp;
1569
1570 spin_lock_irq(&timer.it_lock);
1571 error = posix_cpu_timer_set(&timer, flags, &it, NULL);
1572 if (error) {
1573 posix_cpu_timer_del(&timer);
1574 spin_unlock_irq(&timer.it_lock);
1575 return error;
1576 }
1577
1578 while (!signal_pending(current)) {
1579 if (!cpu_timer_getexpires(&timer.it.cpu)) {
1580 /*
1581 * Our timer fired and was reset, below
1582 * deletion can not fail.
1583 */
1584 posix_cpu_timer_del(&timer);
1585 spin_unlock_irq(&timer.it_lock);
1586 return 0;
1587 }
1588
1589 /*
1590 * Block until cpu_timer_fire (or a signal) wakes us.
1591 */
1592 __set_current_state(TASK_INTERRUPTIBLE);
1593 spin_unlock_irq(&timer.it_lock);
1594 schedule();
1595 spin_lock_irq(&timer.it_lock);
1596 }
1597
1598 /*
1599 * We were interrupted by a signal.
1600 */
1601 expires = cpu_timer_getexpires(&timer.it.cpu);
1602 error = posix_cpu_timer_set(&timer, 0, &zero_it, &it);
1603 if (!error) {
1604 /* Timer is now unarmed, deletion can not fail. */
1605 posix_cpu_timer_del(&timer);
1606 } else {
1607 while (error == TIMER_RETRY) {
1608 posix_cpu_timer_wait_running_nsleep(&timer);
1609 error = posix_cpu_timer_del(&timer);
1610 }
1611 }
1612
1613 spin_unlock_irq(&timer.it_lock);
1614
1615 if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) {
1616 /*
1617 * It actually did fire already.
1618 */
1619 return 0;
1620 }
1621
1622 error = -ERESTART_RESTARTBLOCK;
1623 /*
1624 * Report back to the user the time still remaining.
1625 */
1626 restart = &current->restart_block;
1627 restart->nanosleep.expires = ns_to_ktime(expires);
1628 if (restart->nanosleep.type != TT_NONE)
1629 error = nanosleep_copyout(restart, &it.it_value);
1630 }
1631
1632 return error;
1633}
1634
1635static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1636
1637static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1638 const struct timespec64 *rqtp)
1639{
1640 struct restart_block *restart_block = &current->restart_block;
1641 int error;
1642
1643 /*
1644 * Diagnose required errors first.
1645 */
1646 if (CPUCLOCK_PERTHREAD(which_clock) &&
1647 (CPUCLOCK_PID(which_clock) == 0 ||
1648 CPUCLOCK_PID(which_clock) == task_pid_vnr(current)))
1649 return -EINVAL;
1650
1651 error = do_cpu_nanosleep(which_clock, flags, rqtp);
1652
1653 if (error == -ERESTART_RESTARTBLOCK) {
1654
1655 if (flags & TIMER_ABSTIME)
1656 return -ERESTARTNOHAND;
1657
1658 restart_block->nanosleep.clockid = which_clock;
1659 set_restart_fn(restart_block, posix_cpu_nsleep_restart);
1660 }
1661 return error;
1662}
1663
1664static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1665{
1666 clockid_t which_clock = restart_block->nanosleep.clockid;
1667 struct timespec64 t;
1668
1669 t = ktime_to_timespec64(restart_block->nanosleep.expires);
1670
1671 return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t);
1672}
1673
1674#define PROCESS_CLOCK make_process_cpuclock(0, CPUCLOCK_SCHED)
1675#define THREAD_CLOCK make_thread_cpuclock(0, CPUCLOCK_SCHED)
1676
1677static int process_cpu_clock_getres(const clockid_t which_clock,
1678 struct timespec64 *tp)
1679{
1680 return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1681}
1682static int process_cpu_clock_get(const clockid_t which_clock,
1683 struct timespec64 *tp)
1684{
1685 return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1686}
1687static int process_cpu_timer_create(struct k_itimer *timer)
1688{
1689 timer->it_clock = PROCESS_CLOCK;
1690 return posix_cpu_timer_create(timer);
1691}
1692static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1693 const struct timespec64 *rqtp)
1694{
1695 return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp);
1696}
1697static int thread_cpu_clock_getres(const clockid_t which_clock,
1698 struct timespec64 *tp)
1699{
1700 return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1701}
1702static int thread_cpu_clock_get(const clockid_t which_clock,
1703 struct timespec64 *tp)
1704{
1705 return posix_cpu_clock_get(THREAD_CLOCK, tp);
1706}
1707static int thread_cpu_timer_create(struct k_itimer *timer)
1708{
1709 timer->it_clock = THREAD_CLOCK;
1710 return posix_cpu_timer_create(timer);
1711}
1712
1713const struct k_clock clock_posix_cpu = {
1714 .clock_getres = posix_cpu_clock_getres,
1715 .clock_set = posix_cpu_clock_set,
1716 .clock_get_timespec = posix_cpu_clock_get,
1717 .timer_create = posix_cpu_timer_create,
1718 .nsleep = posix_cpu_nsleep,
1719 .timer_set = posix_cpu_timer_set,
1720 .timer_del = posix_cpu_timer_del,
1721 .timer_get = posix_cpu_timer_get,
1722 .timer_rearm = posix_cpu_timer_rearm,
1723 .timer_wait_running = posix_cpu_timer_wait_running,
1724};
1725
1726const struct k_clock clock_process = {
1727 .clock_getres = process_cpu_clock_getres,
1728 .clock_get_timespec = process_cpu_clock_get,
1729 .timer_create = process_cpu_timer_create,
1730 .nsleep = process_cpu_nsleep,
1731};
1732
1733const struct k_clock clock_thread = {
1734 .clock_getres = thread_cpu_clock_getres,
1735 .clock_get_timespec = thread_cpu_clock_get,
1736 .timer_create = thread_cpu_timer_create,
1737};