xref: /linux/kernel/signal.c (revision a6021aa24f6417416d93318bbfa022ab229c33c8)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/kernel/signal.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  *
7  *  1997-11-02  Modified for POSIX.1b signals by Richard Henderson
8  *
9  *  2003-06-02  Jim Houston - Concurrent Computer Corp.
10  *		Changes to use preallocated sigqueue structures
11  *		to allow signals to be sent reliably.
12  */
13 
14 #include <linux/slab.h>
15 #include <linux/export.h>
16 #include <linux/init.h>
17 #include <linux/sched/mm.h>
18 #include <linux/sched/user.h>
19 #include <linux/sched/debug.h>
20 #include <linux/sched/task.h>
21 #include <linux/sched/task_stack.h>
22 #include <linux/sched/cputime.h>
23 #include <linux/file.h>
24 #include <linux/fs.h>
25 #include <linux/mm.h>
26 #include <linux/proc_fs.h>
27 #include <linux/tty.h>
28 #include <linux/binfmts.h>
29 #include <linux/coredump.h>
30 #include <linux/security.h>
31 #include <linux/syscalls.h>
32 #include <linux/ptrace.h>
33 #include <linux/signal.h>
34 #include <linux/signalfd.h>
35 #include <linux/ratelimit.h>
36 #include <linux/task_work.h>
37 #include <linux/capability.h>
38 #include <linux/freezer.h>
39 #include <linux/pid_namespace.h>
40 #include <linux/nsproxy.h>
41 #include <linux/user_namespace.h>
42 #include <linux/uprobes.h>
43 #include <linux/compat.h>
44 #include <linux/cn_proc.h>
45 #include <linux/compiler.h>
46 #include <linux/posix-timers.h>
47 #include <linux/cgroup.h>
48 #include <linux/audit.h>
49 #include <linux/sysctl.h>
50 #include <uapi/linux/pidfd.h>
51 
52 #define CREATE_TRACE_POINTS
53 #include <trace/events/signal.h>
54 
55 #include <asm/param.h>
56 #include <linux/uaccess.h>
57 #include <asm/unistd.h>
58 #include <asm/siginfo.h>
59 #include <asm/cacheflush.h>
60 #include <asm/syscall.h>	/* for syscall_get_* */
61 
62 /*
63  * SLAB caches for signal bits.
64  */
65 
66 static struct kmem_cache *sigqueue_cachep;
67 
68 int print_fatal_signals __read_mostly;
69 
70 static void __user *sig_handler(struct task_struct *t, int sig)
71 {
72 	return t->sighand->action[sig - 1].sa.sa_handler;
73 }
74 
75 static inline bool sig_handler_ignored(void __user *handler, int sig)
76 {
77 	/* Is it explicitly or implicitly ignored? */
78 	return handler == SIG_IGN ||
79 	       (handler == SIG_DFL && sig_kernel_ignore(sig));
80 }
81 
82 static bool sig_task_ignored(struct task_struct *t, int sig, bool force)
83 {
84 	void __user *handler;
85 
86 	handler = sig_handler(t, sig);
87 
88 	/* SIGKILL and SIGSTOP may not be sent to the global init */
89 	if (unlikely(is_global_init(t) && sig_kernel_only(sig)))
90 		return true;
91 
92 	if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
93 	    handler == SIG_DFL && !(force && sig_kernel_only(sig)))
94 		return true;
95 
96 	/* Only allow kernel generated signals to this kthread */
97 	if (unlikely((t->flags & PF_KTHREAD) &&
98 		     (handler == SIG_KTHREAD_KERNEL) && !force))
99 		return true;
100 
101 	return sig_handler_ignored(handler, sig);
102 }
103 
104 static bool sig_ignored(struct task_struct *t, int sig, bool force)
105 {
106 	/*
107 	 * Blocked signals are never ignored, since the
108 	 * signal handler may change by the time it is
109 	 * unblocked.
110 	 */
111 	if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
112 		return false;
113 
114 	/*
115 	 * Tracers may want to know about even ignored signal unless it
116 	 * is SIGKILL which can't be reported anyway but can be ignored
117 	 * by SIGNAL_UNKILLABLE task.
118 	 */
119 	if (t->ptrace && sig != SIGKILL)
120 		return false;
121 
122 	return sig_task_ignored(t, sig, force);
123 }
124 
125 /*
126  * Re-calculate pending state from the set of locally pending
127  * signals, globally pending signals, and blocked signals.
128  */
129 static inline bool has_pending_signals(sigset_t *signal, sigset_t *blocked)
130 {
131 	unsigned long ready;
132 	long i;
133 
134 	switch (_NSIG_WORDS) {
135 	default:
136 		for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
137 			ready |= signal->sig[i] &~ blocked->sig[i];
138 		break;
139 
140 	case 4: ready  = signal->sig[3] &~ blocked->sig[3];
141 		ready |= signal->sig[2] &~ blocked->sig[2];
142 		ready |= signal->sig[1] &~ blocked->sig[1];
143 		ready |= signal->sig[0] &~ blocked->sig[0];
144 		break;
145 
146 	case 2: ready  = signal->sig[1] &~ blocked->sig[1];
147 		ready |= signal->sig[0] &~ blocked->sig[0];
148 		break;
149 
150 	case 1: ready  = signal->sig[0] &~ blocked->sig[0];
151 	}
152 	return ready !=	0;
153 }
154 
155 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
156 
157 static bool recalc_sigpending_tsk(struct task_struct *t)
158 {
159 	if ((t->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) ||
160 	    PENDING(&t->pending, &t->blocked) ||
161 	    PENDING(&t->signal->shared_pending, &t->blocked) ||
162 	    cgroup_task_frozen(t)) {
163 		set_tsk_thread_flag(t, TIF_SIGPENDING);
164 		return true;
165 	}
166 
167 	/*
168 	 * We must never clear the flag in another thread, or in current
169 	 * when it's possible the current syscall is returning -ERESTART*.
170 	 * So we don't clear it here, and only callers who know they should do.
171 	 */
172 	return false;
173 }
174 
175 void recalc_sigpending(void)
176 {
177 	if (!recalc_sigpending_tsk(current) && !freezing(current))
178 		clear_thread_flag(TIF_SIGPENDING);
179 
180 }
181 EXPORT_SYMBOL(recalc_sigpending);
182 
183 void calculate_sigpending(void)
184 {
185 	/* Have any signals or users of TIF_SIGPENDING been delayed
186 	 * until after fork?
187 	 */
188 	spin_lock_irq(&current->sighand->siglock);
189 	set_tsk_thread_flag(current, TIF_SIGPENDING);
190 	recalc_sigpending();
191 	spin_unlock_irq(&current->sighand->siglock);
192 }
193 
194 /* Given the mask, find the first available signal that should be serviced. */
195 
196 #define SYNCHRONOUS_MASK \
197 	(sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
198 	 sigmask(SIGTRAP) | sigmask(SIGFPE) | sigmask(SIGSYS))
199 
200 int next_signal(struct sigpending *pending, sigset_t *mask)
201 {
202 	unsigned long i, *s, *m, x;
203 	int sig = 0;
204 
205 	s = pending->signal.sig;
206 	m = mask->sig;
207 
208 	/*
209 	 * Handle the first word specially: it contains the
210 	 * synchronous signals that need to be dequeued first.
211 	 */
212 	x = *s &~ *m;
213 	if (x) {
214 		if (x & SYNCHRONOUS_MASK)
215 			x &= SYNCHRONOUS_MASK;
216 		sig = ffz(~x) + 1;
217 		return sig;
218 	}
219 
220 	switch (_NSIG_WORDS) {
221 	default:
222 		for (i = 1; i < _NSIG_WORDS; ++i) {
223 			x = *++s &~ *++m;
224 			if (!x)
225 				continue;
226 			sig = ffz(~x) + i*_NSIG_BPW + 1;
227 			break;
228 		}
229 		break;
230 
231 	case 2:
232 		x = s[1] &~ m[1];
233 		if (!x)
234 			break;
235 		sig = ffz(~x) + _NSIG_BPW + 1;
236 		break;
237 
238 	case 1:
239 		/* Nothing to do */
240 		break;
241 	}
242 
243 	return sig;
244 }
245 
246 static inline void print_dropped_signal(int sig)
247 {
248 	static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
249 
250 	if (!print_fatal_signals)
251 		return;
252 
253 	if (!__ratelimit(&ratelimit_state))
254 		return;
255 
256 	pr_info("%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
257 				current->comm, current->pid, sig);
258 }
259 
260 /**
261  * task_set_jobctl_pending - set jobctl pending bits
262  * @task: target task
263  * @mask: pending bits to set
264  *
265  * Clear @mask from @task->jobctl.  @mask must be subset of
266  * %JOBCTL_PENDING_MASK | %JOBCTL_STOP_CONSUME | %JOBCTL_STOP_SIGMASK |
267  * %JOBCTL_TRAPPING.  If stop signo is being set, the existing signo is
268  * cleared.  If @task is already being killed or exiting, this function
269  * becomes noop.
270  *
271  * CONTEXT:
272  * Must be called with @task->sighand->siglock held.
273  *
274  * RETURNS:
275  * %true if @mask is set, %false if made noop because @task was dying.
276  */
277 bool task_set_jobctl_pending(struct task_struct *task, unsigned long mask)
278 {
279 	BUG_ON(mask & ~(JOBCTL_PENDING_MASK | JOBCTL_STOP_CONSUME |
280 			JOBCTL_STOP_SIGMASK | JOBCTL_TRAPPING));
281 	BUG_ON((mask & JOBCTL_TRAPPING) && !(mask & JOBCTL_PENDING_MASK));
282 
283 	if (unlikely(fatal_signal_pending(task) || (task->flags & PF_EXITING)))
284 		return false;
285 
286 	if (mask & JOBCTL_STOP_SIGMASK)
287 		task->jobctl &= ~JOBCTL_STOP_SIGMASK;
288 
289 	task->jobctl |= mask;
290 	return true;
291 }
292 
293 /**
294  * task_clear_jobctl_trapping - clear jobctl trapping bit
295  * @task: target task
296  *
297  * If JOBCTL_TRAPPING is set, a ptracer is waiting for us to enter TRACED.
298  * Clear it and wake up the ptracer.  Note that we don't need any further
299  * locking.  @task->siglock guarantees that @task->parent points to the
300  * ptracer.
301  *
302  * CONTEXT:
303  * Must be called with @task->sighand->siglock held.
304  */
305 void task_clear_jobctl_trapping(struct task_struct *task)
306 {
307 	if (unlikely(task->jobctl & JOBCTL_TRAPPING)) {
308 		task->jobctl &= ~JOBCTL_TRAPPING;
309 		smp_mb();	/* advised by wake_up_bit() */
310 		wake_up_bit(&task->jobctl, JOBCTL_TRAPPING_BIT);
311 	}
312 }
313 
314 /**
315  * task_clear_jobctl_pending - clear jobctl pending bits
316  * @task: target task
317  * @mask: pending bits to clear
318  *
319  * Clear @mask from @task->jobctl.  @mask must be subset of
320  * %JOBCTL_PENDING_MASK.  If %JOBCTL_STOP_PENDING is being cleared, other
321  * STOP bits are cleared together.
322  *
323  * If clearing of @mask leaves no stop or trap pending, this function calls
324  * task_clear_jobctl_trapping().
325  *
326  * CONTEXT:
327  * Must be called with @task->sighand->siglock held.
328  */
329 void task_clear_jobctl_pending(struct task_struct *task, unsigned long mask)
330 {
331 	BUG_ON(mask & ~JOBCTL_PENDING_MASK);
332 
333 	if (mask & JOBCTL_STOP_PENDING)
334 		mask |= JOBCTL_STOP_CONSUME | JOBCTL_STOP_DEQUEUED;
335 
336 	task->jobctl &= ~mask;
337 
338 	if (!(task->jobctl & JOBCTL_PENDING_MASK))
339 		task_clear_jobctl_trapping(task);
340 }
341 
342 /**
343  * task_participate_group_stop - participate in a group stop
344  * @task: task participating in a group stop
345  *
346  * @task has %JOBCTL_STOP_PENDING set and is participating in a group stop.
347  * Group stop states are cleared and the group stop count is consumed if
348  * %JOBCTL_STOP_CONSUME was set.  If the consumption completes the group
349  * stop, the appropriate `SIGNAL_*` flags are set.
350  *
351  * CONTEXT:
352  * Must be called with @task->sighand->siglock held.
353  *
354  * RETURNS:
355  * %true if group stop completion should be notified to the parent, %false
356  * otherwise.
357  */
358 static bool task_participate_group_stop(struct task_struct *task)
359 {
360 	struct signal_struct *sig = task->signal;
361 	bool consume = task->jobctl & JOBCTL_STOP_CONSUME;
362 
363 	WARN_ON_ONCE(!(task->jobctl & JOBCTL_STOP_PENDING));
364 
365 	task_clear_jobctl_pending(task, JOBCTL_STOP_PENDING);
366 
367 	if (!consume)
368 		return false;
369 
370 	if (!WARN_ON_ONCE(sig->group_stop_count == 0))
371 		sig->group_stop_count--;
372 
373 	/*
374 	 * Tell the caller to notify completion iff we are entering into a
375 	 * fresh group stop.  Read comment in do_signal_stop() for details.
376 	 */
377 	if (!sig->group_stop_count && !(sig->flags & SIGNAL_STOP_STOPPED)) {
378 		signal_set_stop_flags(sig, SIGNAL_STOP_STOPPED);
379 		return true;
380 	}
381 	return false;
382 }
383 
384 void task_join_group_stop(struct task_struct *task)
385 {
386 	unsigned long mask = current->jobctl & JOBCTL_STOP_SIGMASK;
387 	struct signal_struct *sig = current->signal;
388 
389 	if (sig->group_stop_count) {
390 		sig->group_stop_count++;
391 		mask |= JOBCTL_STOP_CONSUME;
392 	} else if (!(sig->flags & SIGNAL_STOP_STOPPED))
393 		return;
394 
395 	/* Have the new thread join an on-going signal group stop */
396 	task_set_jobctl_pending(task, mask | JOBCTL_STOP_PENDING);
397 }
398 
399 /*
400  * allocate a new signal queue record
401  * - this may be called without locks if and only if t == current, otherwise an
402  *   appropriate lock must be held to stop the target task from exiting
403  */
404 static struct sigqueue *
405 __sigqueue_alloc(int sig, struct task_struct *t, gfp_t gfp_flags,
406 		 int override_rlimit, const unsigned int sigqueue_flags)
407 {
408 	struct sigqueue *q = NULL;
409 	struct ucounts *ucounts;
410 	long sigpending;
411 
412 	/*
413 	 * Protect access to @t credentials. This can go away when all
414 	 * callers hold rcu read lock.
415 	 *
416 	 * NOTE! A pending signal will hold on to the user refcount,
417 	 * and we get/put the refcount only when the sigpending count
418 	 * changes from/to zero.
419 	 */
420 	rcu_read_lock();
421 	ucounts = task_ucounts(t);
422 	sigpending = inc_rlimit_get_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
423 	rcu_read_unlock();
424 	if (!sigpending)
425 		return NULL;
426 
427 	if (override_rlimit || likely(sigpending <= task_rlimit(t, RLIMIT_SIGPENDING))) {
428 		q = kmem_cache_alloc(sigqueue_cachep, gfp_flags);
429 	} else {
430 		print_dropped_signal(sig);
431 	}
432 
433 	if (unlikely(q == NULL)) {
434 		dec_rlimit_put_ucounts(ucounts, UCOUNT_RLIMIT_SIGPENDING);
435 	} else {
436 		INIT_LIST_HEAD(&q->list);
437 		q->flags = sigqueue_flags;
438 		q->ucounts = ucounts;
439 	}
440 	return q;
441 }
442 
443 static void __sigqueue_free(struct sigqueue *q)
444 {
445 	if (q->flags & SIGQUEUE_PREALLOC)
446 		return;
447 	if (q->ucounts) {
448 		dec_rlimit_put_ucounts(q->ucounts, UCOUNT_RLIMIT_SIGPENDING);
449 		q->ucounts = NULL;
450 	}
451 	kmem_cache_free(sigqueue_cachep, q);
452 }
453 
454 void flush_sigqueue(struct sigpending *queue)
455 {
456 	struct sigqueue *q;
457 
458 	sigemptyset(&queue->signal);
459 	while (!list_empty(&queue->list)) {
460 		q = list_entry(queue->list.next, struct sigqueue , list);
461 		list_del_init(&q->list);
462 		__sigqueue_free(q);
463 	}
464 }
465 
466 /*
467  * Flush all pending signals for this kthread.
468  */
469 void flush_signals(struct task_struct *t)
470 {
471 	unsigned long flags;
472 
473 	spin_lock_irqsave(&t->sighand->siglock, flags);
474 	clear_tsk_thread_flag(t, TIF_SIGPENDING);
475 	flush_sigqueue(&t->pending);
476 	flush_sigqueue(&t->signal->shared_pending);
477 	spin_unlock_irqrestore(&t->sighand->siglock, flags);
478 }
479 EXPORT_SYMBOL(flush_signals);
480 
481 #ifdef CONFIG_POSIX_TIMERS
482 static void __flush_itimer_signals(struct sigpending *pending)
483 {
484 	sigset_t signal, retain;
485 	struct sigqueue *q, *n;
486 
487 	signal = pending->signal;
488 	sigemptyset(&retain);
489 
490 	list_for_each_entry_safe(q, n, &pending->list, list) {
491 		int sig = q->info.si_signo;
492 
493 		if (likely(q->info.si_code != SI_TIMER)) {
494 			sigaddset(&retain, sig);
495 		} else {
496 			sigdelset(&signal, sig);
497 			list_del_init(&q->list);
498 			__sigqueue_free(q);
499 		}
500 	}
501 
502 	sigorsets(&pending->signal, &signal, &retain);
503 }
504 
505 void flush_itimer_signals(void)
506 {
507 	struct task_struct *tsk = current;
508 	unsigned long flags;
509 
510 	spin_lock_irqsave(&tsk->sighand->siglock, flags);
511 	__flush_itimer_signals(&tsk->pending);
512 	__flush_itimer_signals(&tsk->signal->shared_pending);
513 	spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
514 }
515 #endif
516 
517 void ignore_signals(struct task_struct *t)
518 {
519 	int i;
520 
521 	for (i = 0; i < _NSIG; ++i)
522 		t->sighand->action[i].sa.sa_handler = SIG_IGN;
523 
524 	flush_signals(t);
525 }
526 
527 /*
528  * Flush all handlers for a task.
529  */
530 
531 void
532 flush_signal_handlers(struct task_struct *t, int force_default)
533 {
534 	int i;
535 	struct k_sigaction *ka = &t->sighand->action[0];
536 	for (i = _NSIG ; i != 0 ; i--) {
537 		if (force_default || ka->sa.sa_handler != SIG_IGN)
538 			ka->sa.sa_handler = SIG_DFL;
539 		ka->sa.sa_flags = 0;
540 #ifdef __ARCH_HAS_SA_RESTORER
541 		ka->sa.sa_restorer = NULL;
542 #endif
543 		sigemptyset(&ka->sa.sa_mask);
544 		ka++;
545 	}
546 }
547 
548 bool unhandled_signal(struct task_struct *tsk, int sig)
549 {
550 	void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
551 	if (is_global_init(tsk))
552 		return true;
553 
554 	if (handler != SIG_IGN && handler != SIG_DFL)
555 		return false;
556 
557 	/* If dying, we handle all new signals by ignoring them */
558 	if (fatal_signal_pending(tsk))
559 		return false;
560 
561 	/* if ptraced, let the tracer determine */
562 	return !tsk->ptrace;
563 }
564 
565 static void collect_signal(int sig, struct sigpending *list, kernel_siginfo_t *info,
566 			   bool *resched_timer)
567 {
568 	struct sigqueue *q, *first = NULL;
569 
570 	/*
571 	 * Collect the siginfo appropriate to this signal.  Check if
572 	 * there is another siginfo for the same signal.
573 	*/
574 	list_for_each_entry(q, &list->list, list) {
575 		if (q->info.si_signo == sig) {
576 			if (first)
577 				goto still_pending;
578 			first = q;
579 		}
580 	}
581 
582 	sigdelset(&list->signal, sig);
583 
584 	if (first) {
585 still_pending:
586 		list_del_init(&first->list);
587 		copy_siginfo(info, &first->info);
588 
589 		*resched_timer =
590 			(first->flags & SIGQUEUE_PREALLOC) &&
591 			(info->si_code == SI_TIMER) &&
592 			(info->si_sys_private);
593 
594 		__sigqueue_free(first);
595 	} else {
596 		/*
597 		 * Ok, it wasn't in the queue.  This must be
598 		 * a fast-pathed signal or we must have been
599 		 * out of queue space.  So zero out the info.
600 		 */
601 		clear_siginfo(info);
602 		info->si_signo = sig;
603 		info->si_errno = 0;
604 		info->si_code = SI_USER;
605 		info->si_pid = 0;
606 		info->si_uid = 0;
607 	}
608 }
609 
610 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
611 			kernel_siginfo_t *info, bool *resched_timer)
612 {
613 	int sig = next_signal(pending, mask);
614 
615 	if (sig)
616 		collect_signal(sig, pending, info, resched_timer);
617 	return sig;
618 }
619 
620 /*
621  * Try to dequeue a signal. If a deliverable signal is found fill in the
622  * caller provided siginfo and return the signal number. Otherwise return
623  * 0.
624  */
625 int dequeue_signal(sigset_t *mask, kernel_siginfo_t *info, enum pid_type *type)
626 {
627 	struct task_struct *tsk = current;
628 	bool resched_timer = false;
629 	int signr;
630 
631 	lockdep_assert_held(&tsk->sighand->siglock);
632 
633 	*type = PIDTYPE_PID;
634 	signr = __dequeue_signal(&tsk->pending, mask, info, &resched_timer);
635 	if (!signr) {
636 		*type = PIDTYPE_TGID;
637 		signr = __dequeue_signal(&tsk->signal->shared_pending,
638 					 mask, info, &resched_timer);
639 #ifdef CONFIG_POSIX_TIMERS
640 		/*
641 		 * itimer signal ?
642 		 *
643 		 * itimers are process shared and we restart periodic
644 		 * itimers in the signal delivery path to prevent DoS
645 		 * attacks in the high resolution timer case. This is
646 		 * compliant with the old way of self-restarting
647 		 * itimers, as the SIGALRM is a legacy signal and only
648 		 * queued once. Changing the restart behaviour to
649 		 * restart the timer in the signal dequeue path is
650 		 * reducing the timer noise on heavy loaded !highres
651 		 * systems too.
652 		 */
653 		if (unlikely(signr == SIGALRM)) {
654 			struct hrtimer *tmr = &tsk->signal->real_timer;
655 
656 			if (!hrtimer_is_queued(tmr) &&
657 			    tsk->signal->it_real_incr != 0) {
658 				hrtimer_forward(tmr, tmr->base->get_time(),
659 						tsk->signal->it_real_incr);
660 				hrtimer_restart(tmr);
661 			}
662 		}
663 #endif
664 	}
665 
666 	recalc_sigpending();
667 	if (!signr)
668 		return 0;
669 
670 	if (unlikely(sig_kernel_stop(signr))) {
671 		/*
672 		 * Set a marker that we have dequeued a stop signal.  Our
673 		 * caller might release the siglock and then the pending
674 		 * stop signal it is about to process is no longer in the
675 		 * pending bitmasks, but must still be cleared by a SIGCONT
676 		 * (and overruled by a SIGKILL).  So those cases clear this
677 		 * shared flag after we've set it.  Note that this flag may
678 		 * remain set after the signal we return is ignored or
679 		 * handled.  That doesn't matter because its only purpose
680 		 * is to alert stop-signal processing code when another
681 		 * processor has come along and cleared the flag.
682 		 */
683 		current->jobctl |= JOBCTL_STOP_DEQUEUED;
684 	}
685 #ifdef CONFIG_POSIX_TIMERS
686 	if (resched_timer) {
687 		/*
688 		 * Release the siglock to ensure proper locking order
689 		 * of timer locks outside of siglocks.  Note, we leave
690 		 * irqs disabled here, since the posix-timers code is
691 		 * about to disable them again anyway.
692 		 */
693 		spin_unlock(&tsk->sighand->siglock);
694 		posixtimer_rearm(info);
695 		spin_lock(&tsk->sighand->siglock);
696 
697 		/* Don't expose the si_sys_private value to userspace */
698 		info->si_sys_private = 0;
699 	}
700 #endif
701 	return signr;
702 }
703 EXPORT_SYMBOL_GPL(dequeue_signal);
704 
705 static int dequeue_synchronous_signal(kernel_siginfo_t *info)
706 {
707 	struct task_struct *tsk = current;
708 	struct sigpending *pending = &tsk->pending;
709 	struct sigqueue *q, *sync = NULL;
710 
711 	/*
712 	 * Might a synchronous signal be in the queue?
713 	 */
714 	if (!((pending->signal.sig[0] & ~tsk->blocked.sig[0]) & SYNCHRONOUS_MASK))
715 		return 0;
716 
717 	/*
718 	 * Return the first synchronous signal in the queue.
719 	 */
720 	list_for_each_entry(q, &pending->list, list) {
721 		/* Synchronous signals have a positive si_code */
722 		if ((q->info.si_code > SI_USER) &&
723 		    (sigmask(q->info.si_signo) & SYNCHRONOUS_MASK)) {
724 			sync = q;
725 			goto next;
726 		}
727 	}
728 	return 0;
729 next:
730 	/*
731 	 * Check if there is another siginfo for the same signal.
732 	 */
733 	list_for_each_entry_continue(q, &pending->list, list) {
734 		if (q->info.si_signo == sync->info.si_signo)
735 			goto still_pending;
736 	}
737 
738 	sigdelset(&pending->signal, sync->info.si_signo);
739 	recalc_sigpending();
740 still_pending:
741 	list_del_init(&sync->list);
742 	copy_siginfo(info, &sync->info);
743 	__sigqueue_free(sync);
744 	return info->si_signo;
745 }
746 
747 /*
748  * Tell a process that it has a new active signal..
749  *
750  * NOTE! we rely on the previous spin_lock to
751  * lock interrupts for us! We can only be called with
752  * "siglock" held, and the local interrupt must
753  * have been disabled when that got acquired!
754  *
755  * No need to set need_resched since signal event passing
756  * goes through ->blocked
757  */
758 void signal_wake_up_state(struct task_struct *t, unsigned int state)
759 {
760 	lockdep_assert_held(&t->sighand->siglock);
761 
762 	set_tsk_thread_flag(t, TIF_SIGPENDING);
763 
764 	/*
765 	 * TASK_WAKEKILL also means wake it up in the stopped/traced/killable
766 	 * case. We don't check t->state here because there is a race with it
767 	 * executing another processor and just now entering stopped state.
768 	 * By using wake_up_state, we ensure the process will wake up and
769 	 * handle its death signal.
770 	 */
771 	if (!wake_up_state(t, state | TASK_INTERRUPTIBLE))
772 		kick_process(t);
773 }
774 
775 /*
776  * Remove signals in mask from the pending set and queue.
777  * Returns 1 if any signals were found.
778  *
779  * All callers must be holding the siglock.
780  */
781 static void flush_sigqueue_mask(sigset_t *mask, struct sigpending *s)
782 {
783 	struct sigqueue *q, *n;
784 	sigset_t m;
785 
786 	sigandsets(&m, mask, &s->signal);
787 	if (sigisemptyset(&m))
788 		return;
789 
790 	sigandnsets(&s->signal, &s->signal, mask);
791 	list_for_each_entry_safe(q, n, &s->list, list) {
792 		if (sigismember(mask, q->info.si_signo)) {
793 			list_del_init(&q->list);
794 			__sigqueue_free(q);
795 		}
796 	}
797 }
798 
799 static inline int is_si_special(const struct kernel_siginfo *info)
800 {
801 	return info <= SEND_SIG_PRIV;
802 }
803 
804 static inline bool si_fromuser(const struct kernel_siginfo *info)
805 {
806 	return info == SEND_SIG_NOINFO ||
807 		(!is_si_special(info) && SI_FROMUSER(info));
808 }
809 
810 /*
811  * called with RCU read lock from check_kill_permission()
812  */
813 static bool kill_ok_by_cred(struct task_struct *t)
814 {
815 	const struct cred *cred = current_cred();
816 	const struct cred *tcred = __task_cred(t);
817 
818 	return uid_eq(cred->euid, tcred->suid) ||
819 	       uid_eq(cred->euid, tcred->uid) ||
820 	       uid_eq(cred->uid, tcred->suid) ||
821 	       uid_eq(cred->uid, tcred->uid) ||
822 	       ns_capable(tcred->user_ns, CAP_KILL);
823 }
824 
825 /*
826  * Bad permissions for sending the signal
827  * - the caller must hold the RCU read lock
828  */
829 static int check_kill_permission(int sig, struct kernel_siginfo *info,
830 				 struct task_struct *t)
831 {
832 	struct pid *sid;
833 	int error;
834 
835 	if (!valid_signal(sig))
836 		return -EINVAL;
837 
838 	if (!si_fromuser(info))
839 		return 0;
840 
841 	error = audit_signal_info(sig, t); /* Let audit system see the signal */
842 	if (error)
843 		return error;
844 
845 	if (!same_thread_group(current, t) &&
846 	    !kill_ok_by_cred(t)) {
847 		switch (sig) {
848 		case SIGCONT:
849 			sid = task_session(t);
850 			/*
851 			 * We don't return the error if sid == NULL. The
852 			 * task was unhashed, the caller must notice this.
853 			 */
854 			if (!sid || sid == task_session(current))
855 				break;
856 			fallthrough;
857 		default:
858 			return -EPERM;
859 		}
860 	}
861 
862 	return security_task_kill(t, info, sig, NULL);
863 }
864 
865 /**
866  * ptrace_trap_notify - schedule trap to notify ptracer
867  * @t: tracee wanting to notify tracer
868  *
869  * This function schedules sticky ptrace trap which is cleared on the next
870  * TRAP_STOP to notify ptracer of an event.  @t must have been seized by
871  * ptracer.
872  *
873  * If @t is running, STOP trap will be taken.  If trapped for STOP and
874  * ptracer is listening for events, tracee is woken up so that it can
875  * re-trap for the new event.  If trapped otherwise, STOP trap will be
876  * eventually taken without returning to userland after the existing traps
877  * are finished by PTRACE_CONT.
878  *
879  * CONTEXT:
880  * Must be called with @task->sighand->siglock held.
881  */
882 static void ptrace_trap_notify(struct task_struct *t)
883 {
884 	WARN_ON_ONCE(!(t->ptrace & PT_SEIZED));
885 	lockdep_assert_held(&t->sighand->siglock);
886 
887 	task_set_jobctl_pending(t, JOBCTL_TRAP_NOTIFY);
888 	ptrace_signal_wake_up(t, t->jobctl & JOBCTL_LISTENING);
889 }
890 
891 /*
892  * Handle magic process-wide effects of stop/continue signals. Unlike
893  * the signal actions, these happen immediately at signal-generation
894  * time regardless of blocking, ignoring, or handling.  This does the
895  * actual continuing for SIGCONT, but not the actual stopping for stop
896  * signals. The process stop is done as a signal action for SIG_DFL.
897  *
898  * Returns true if the signal should be actually delivered, otherwise
899  * it should be dropped.
900  */
901 static bool prepare_signal(int sig, struct task_struct *p, bool force)
902 {
903 	struct signal_struct *signal = p->signal;
904 	struct task_struct *t;
905 	sigset_t flush;
906 
907 	if (signal->flags & SIGNAL_GROUP_EXIT) {
908 		if (signal->core_state)
909 			return sig == SIGKILL;
910 		/*
911 		 * The process is in the middle of dying, drop the signal.
912 		 */
913 		return false;
914 	} else if (sig_kernel_stop(sig)) {
915 		/*
916 		 * This is a stop signal.  Remove SIGCONT from all queues.
917 		 */
918 		siginitset(&flush, sigmask(SIGCONT));
919 		flush_sigqueue_mask(&flush, &signal->shared_pending);
920 		for_each_thread(p, t)
921 			flush_sigqueue_mask(&flush, &t->pending);
922 	} else if (sig == SIGCONT) {
923 		unsigned int why;
924 		/*
925 		 * Remove all stop signals from all queues, wake all threads.
926 		 */
927 		siginitset(&flush, SIG_KERNEL_STOP_MASK);
928 		flush_sigqueue_mask(&flush, &signal->shared_pending);
929 		for_each_thread(p, t) {
930 			flush_sigqueue_mask(&flush, &t->pending);
931 			task_clear_jobctl_pending(t, JOBCTL_STOP_PENDING);
932 			if (likely(!(t->ptrace & PT_SEIZED))) {
933 				t->jobctl &= ~JOBCTL_STOPPED;
934 				wake_up_state(t, __TASK_STOPPED);
935 			} else
936 				ptrace_trap_notify(t);
937 		}
938 
939 		/*
940 		 * Notify the parent with CLD_CONTINUED if we were stopped.
941 		 *
942 		 * If we were in the middle of a group stop, we pretend it
943 		 * was already finished, and then continued. Since SIGCHLD
944 		 * doesn't queue we report only CLD_STOPPED, as if the next
945 		 * CLD_CONTINUED was dropped.
946 		 */
947 		why = 0;
948 		if (signal->flags & SIGNAL_STOP_STOPPED)
949 			why |= SIGNAL_CLD_CONTINUED;
950 		else if (signal->group_stop_count)
951 			why |= SIGNAL_CLD_STOPPED;
952 
953 		if (why) {
954 			/*
955 			 * The first thread which returns from do_signal_stop()
956 			 * will take ->siglock, notice SIGNAL_CLD_MASK, and
957 			 * notify its parent. See get_signal().
958 			 */
959 			signal_set_stop_flags(signal, why | SIGNAL_STOP_CONTINUED);
960 			signal->group_stop_count = 0;
961 			signal->group_exit_code = 0;
962 		}
963 	}
964 
965 	return !sig_ignored(p, sig, force);
966 }
967 
968 /*
969  * Test if P wants to take SIG.  After we've checked all threads with this,
970  * it's equivalent to finding no threads not blocking SIG.  Any threads not
971  * blocking SIG were ruled out because they are not running and already
972  * have pending signals.  Such threads will dequeue from the shared queue
973  * as soon as they're available, so putting the signal on the shared queue
974  * will be equivalent to sending it to one such thread.
975  */
976 static inline bool wants_signal(int sig, struct task_struct *p)
977 {
978 	if (sigismember(&p->blocked, sig))
979 		return false;
980 
981 	if (p->flags & PF_EXITING)
982 		return false;
983 
984 	if (sig == SIGKILL)
985 		return true;
986 
987 	if (task_is_stopped_or_traced(p))
988 		return false;
989 
990 	return task_curr(p) || !task_sigpending(p);
991 }
992 
993 static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
994 {
995 	struct signal_struct *signal = p->signal;
996 	struct task_struct *t;
997 
998 	/*
999 	 * Now find a thread we can wake up to take the signal off the queue.
1000 	 *
1001 	 * Try the suggested task first (may or may not be the main thread).
1002 	 */
1003 	if (wants_signal(sig, p))
1004 		t = p;
1005 	else if ((type == PIDTYPE_PID) || thread_group_empty(p))
1006 		/*
1007 		 * There is just one thread and it does not need to be woken.
1008 		 * It will dequeue unblocked signals before it runs again.
1009 		 */
1010 		return;
1011 	else {
1012 		/*
1013 		 * Otherwise try to find a suitable thread.
1014 		 */
1015 		t = signal->curr_target;
1016 		while (!wants_signal(sig, t)) {
1017 			t = next_thread(t);
1018 			if (t == signal->curr_target)
1019 				/*
1020 				 * No thread needs to be woken.
1021 				 * Any eligible threads will see
1022 				 * the signal in the queue soon.
1023 				 */
1024 				return;
1025 		}
1026 		signal->curr_target = t;
1027 	}
1028 
1029 	/*
1030 	 * Found a killable thread.  If the signal will be fatal,
1031 	 * then start taking the whole group down immediately.
1032 	 */
1033 	if (sig_fatal(p, sig) &&
1034 	    (signal->core_state || !(signal->flags & SIGNAL_GROUP_EXIT)) &&
1035 	    !sigismember(&t->real_blocked, sig) &&
1036 	    (sig == SIGKILL || !p->ptrace)) {
1037 		/*
1038 		 * This signal will be fatal to the whole group.
1039 		 */
1040 		if (!sig_kernel_coredump(sig)) {
1041 			/*
1042 			 * Start a group exit and wake everybody up.
1043 			 * This way we don't have other threads
1044 			 * running and doing things after a slower
1045 			 * thread has the fatal signal pending.
1046 			 */
1047 			signal->flags = SIGNAL_GROUP_EXIT;
1048 			signal->group_exit_code = sig;
1049 			signal->group_stop_count = 0;
1050 			__for_each_thread(signal, t) {
1051 				task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1052 				sigaddset(&t->pending.signal, SIGKILL);
1053 				signal_wake_up(t, 1);
1054 			}
1055 			return;
1056 		}
1057 	}
1058 
1059 	/*
1060 	 * The signal is already in the shared-pending queue.
1061 	 * Tell the chosen thread to wake up and dequeue it.
1062 	 */
1063 	signal_wake_up(t, sig == SIGKILL);
1064 	return;
1065 }
1066 
1067 static inline bool legacy_queue(struct sigpending *signals, int sig)
1068 {
1069 	return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
1070 }
1071 
1072 static int __send_signal_locked(int sig, struct kernel_siginfo *info,
1073 				struct task_struct *t, enum pid_type type, bool force)
1074 {
1075 	struct sigpending *pending;
1076 	struct sigqueue *q;
1077 	int override_rlimit;
1078 	int ret = 0, result;
1079 
1080 	lockdep_assert_held(&t->sighand->siglock);
1081 
1082 	result = TRACE_SIGNAL_IGNORED;
1083 	if (!prepare_signal(sig, t, force))
1084 		goto ret;
1085 
1086 	pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
1087 	/*
1088 	 * Short-circuit ignored signals and support queuing
1089 	 * exactly one non-rt signal, so that we can get more
1090 	 * detailed information about the cause of the signal.
1091 	 */
1092 	result = TRACE_SIGNAL_ALREADY_PENDING;
1093 	if (legacy_queue(pending, sig))
1094 		goto ret;
1095 
1096 	result = TRACE_SIGNAL_DELIVERED;
1097 	/*
1098 	 * Skip useless siginfo allocation for SIGKILL and kernel threads.
1099 	 */
1100 	if ((sig == SIGKILL) || (t->flags & PF_KTHREAD))
1101 		goto out_set;
1102 
1103 	/*
1104 	 * Real-time signals must be queued if sent by sigqueue, or
1105 	 * some other real-time mechanism.  It is implementation
1106 	 * defined whether kill() does so.  We attempt to do so, on
1107 	 * the principle of least surprise, but since kill is not
1108 	 * allowed to fail with EAGAIN when low on memory we just
1109 	 * make sure at least one signal gets delivered and don't
1110 	 * pass on the info struct.
1111 	 */
1112 	if (sig < SIGRTMIN)
1113 		override_rlimit = (is_si_special(info) || info->si_code >= 0);
1114 	else
1115 		override_rlimit = 0;
1116 
1117 	q = __sigqueue_alloc(sig, t, GFP_ATOMIC, override_rlimit, 0);
1118 
1119 	if (q) {
1120 		list_add_tail(&q->list, &pending->list);
1121 		switch ((unsigned long) info) {
1122 		case (unsigned long) SEND_SIG_NOINFO:
1123 			clear_siginfo(&q->info);
1124 			q->info.si_signo = sig;
1125 			q->info.si_errno = 0;
1126 			q->info.si_code = SI_USER;
1127 			q->info.si_pid = task_tgid_nr_ns(current,
1128 							task_active_pid_ns(t));
1129 			rcu_read_lock();
1130 			q->info.si_uid =
1131 				from_kuid_munged(task_cred_xxx(t, user_ns),
1132 						 current_uid());
1133 			rcu_read_unlock();
1134 			break;
1135 		case (unsigned long) SEND_SIG_PRIV:
1136 			clear_siginfo(&q->info);
1137 			q->info.si_signo = sig;
1138 			q->info.si_errno = 0;
1139 			q->info.si_code = SI_KERNEL;
1140 			q->info.si_pid = 0;
1141 			q->info.si_uid = 0;
1142 			break;
1143 		default:
1144 			copy_siginfo(&q->info, info);
1145 			break;
1146 		}
1147 	} else if (!is_si_special(info) &&
1148 		   sig >= SIGRTMIN && info->si_code != SI_USER) {
1149 		/*
1150 		 * Queue overflow, abort.  We may abort if the
1151 		 * signal was rt and sent by user using something
1152 		 * other than kill().
1153 		 */
1154 		result = TRACE_SIGNAL_OVERFLOW_FAIL;
1155 		ret = -EAGAIN;
1156 		goto ret;
1157 	} else {
1158 		/*
1159 		 * This is a silent loss of information.  We still
1160 		 * send the signal, but the *info bits are lost.
1161 		 */
1162 		result = TRACE_SIGNAL_LOSE_INFO;
1163 	}
1164 
1165 out_set:
1166 	signalfd_notify(t, sig);
1167 	sigaddset(&pending->signal, sig);
1168 
1169 	/* Let multiprocess signals appear after on-going forks */
1170 	if (type > PIDTYPE_TGID) {
1171 		struct multiprocess_signals *delayed;
1172 		hlist_for_each_entry(delayed, &t->signal->multiprocess, node) {
1173 			sigset_t *signal = &delayed->signal;
1174 			/* Can't queue both a stop and a continue signal */
1175 			if (sig == SIGCONT)
1176 				sigdelsetmask(signal, SIG_KERNEL_STOP_MASK);
1177 			else if (sig_kernel_stop(sig))
1178 				sigdelset(signal, SIGCONT);
1179 			sigaddset(signal, sig);
1180 		}
1181 	}
1182 
1183 	complete_signal(sig, t, type);
1184 ret:
1185 	trace_signal_generate(sig, info, t, type != PIDTYPE_PID, result);
1186 	return ret;
1187 }
1188 
1189 static inline bool has_si_pid_and_uid(struct kernel_siginfo *info)
1190 {
1191 	bool ret = false;
1192 	switch (siginfo_layout(info->si_signo, info->si_code)) {
1193 	case SIL_KILL:
1194 	case SIL_CHLD:
1195 	case SIL_RT:
1196 		ret = true;
1197 		break;
1198 	case SIL_TIMER:
1199 	case SIL_POLL:
1200 	case SIL_FAULT:
1201 	case SIL_FAULT_TRAPNO:
1202 	case SIL_FAULT_MCEERR:
1203 	case SIL_FAULT_BNDERR:
1204 	case SIL_FAULT_PKUERR:
1205 	case SIL_FAULT_PERF_EVENT:
1206 	case SIL_SYS:
1207 		ret = false;
1208 		break;
1209 	}
1210 	return ret;
1211 }
1212 
1213 int send_signal_locked(int sig, struct kernel_siginfo *info,
1214 		       struct task_struct *t, enum pid_type type)
1215 {
1216 	/* Should SIGKILL or SIGSTOP be received by a pid namespace init? */
1217 	bool force = false;
1218 
1219 	if (info == SEND_SIG_NOINFO) {
1220 		/* Force if sent from an ancestor pid namespace */
1221 		force = !task_pid_nr_ns(current, task_active_pid_ns(t));
1222 	} else if (info == SEND_SIG_PRIV) {
1223 		/* Don't ignore kernel generated signals */
1224 		force = true;
1225 	} else if (has_si_pid_and_uid(info)) {
1226 		/* SIGKILL and SIGSTOP is special or has ids */
1227 		struct user_namespace *t_user_ns;
1228 
1229 		rcu_read_lock();
1230 		t_user_ns = task_cred_xxx(t, user_ns);
1231 		if (current_user_ns() != t_user_ns) {
1232 			kuid_t uid = make_kuid(current_user_ns(), info->si_uid);
1233 			info->si_uid = from_kuid_munged(t_user_ns, uid);
1234 		}
1235 		rcu_read_unlock();
1236 
1237 		/* A kernel generated signal? */
1238 		force = (info->si_code == SI_KERNEL);
1239 
1240 		/* From an ancestor pid namespace? */
1241 		if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
1242 			info->si_pid = 0;
1243 			force = true;
1244 		}
1245 	}
1246 	return __send_signal_locked(sig, info, t, type, force);
1247 }
1248 
1249 static void print_fatal_signal(int signr)
1250 {
1251 	struct pt_regs *regs = task_pt_regs(current);
1252 	struct file *exe_file;
1253 
1254 	exe_file = get_task_exe_file(current);
1255 	if (exe_file) {
1256 		pr_info("%pD: %s: potentially unexpected fatal signal %d.\n",
1257 			exe_file, current->comm, signr);
1258 		fput(exe_file);
1259 	} else {
1260 		pr_info("%s: potentially unexpected fatal signal %d.\n",
1261 			current->comm, signr);
1262 	}
1263 
1264 #if defined(__i386__) && !defined(__arch_um__)
1265 	pr_info("code at %08lx: ", regs->ip);
1266 	{
1267 		int i;
1268 		for (i = 0; i < 16; i++) {
1269 			unsigned char insn;
1270 
1271 			if (get_user(insn, (unsigned char *)(regs->ip + i)))
1272 				break;
1273 			pr_cont("%02x ", insn);
1274 		}
1275 	}
1276 	pr_cont("\n");
1277 #endif
1278 	preempt_disable();
1279 	show_regs(regs);
1280 	preempt_enable();
1281 }
1282 
1283 static int __init setup_print_fatal_signals(char *str)
1284 {
1285 	get_option (&str, &print_fatal_signals);
1286 
1287 	return 1;
1288 }
1289 
1290 __setup("print-fatal-signals=", setup_print_fatal_signals);
1291 
1292 int do_send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p,
1293 			enum pid_type type)
1294 {
1295 	unsigned long flags;
1296 	int ret = -ESRCH;
1297 
1298 	if (lock_task_sighand(p, &flags)) {
1299 		ret = send_signal_locked(sig, info, p, type);
1300 		unlock_task_sighand(p, &flags);
1301 	}
1302 
1303 	return ret;
1304 }
1305 
1306 enum sig_handler {
1307 	HANDLER_CURRENT, /* If reachable use the current handler */
1308 	HANDLER_SIG_DFL, /* Always use SIG_DFL handler semantics */
1309 	HANDLER_EXIT,	 /* Only visible as the process exit code */
1310 };
1311 
1312 /*
1313  * Force a signal that the process can't ignore: if necessary
1314  * we unblock the signal and change any SIG_IGN to SIG_DFL.
1315  *
1316  * Note: If we unblock the signal, we always reset it to SIG_DFL,
1317  * since we do not want to have a signal handler that was blocked
1318  * be invoked when user space had explicitly blocked it.
1319  *
1320  * We don't want to have recursive SIGSEGV's etc, for example,
1321  * that is why we also clear SIGNAL_UNKILLABLE.
1322  */
1323 static int
1324 force_sig_info_to_task(struct kernel_siginfo *info, struct task_struct *t,
1325 	enum sig_handler handler)
1326 {
1327 	unsigned long int flags;
1328 	int ret, blocked, ignored;
1329 	struct k_sigaction *action;
1330 	int sig = info->si_signo;
1331 
1332 	spin_lock_irqsave(&t->sighand->siglock, flags);
1333 	action = &t->sighand->action[sig-1];
1334 	ignored = action->sa.sa_handler == SIG_IGN;
1335 	blocked = sigismember(&t->blocked, sig);
1336 	if (blocked || ignored || (handler != HANDLER_CURRENT)) {
1337 		action->sa.sa_handler = SIG_DFL;
1338 		if (handler == HANDLER_EXIT)
1339 			action->sa.sa_flags |= SA_IMMUTABLE;
1340 		if (blocked)
1341 			sigdelset(&t->blocked, sig);
1342 	}
1343 	/*
1344 	 * Don't clear SIGNAL_UNKILLABLE for traced tasks, users won't expect
1345 	 * debugging to leave init killable. But HANDLER_EXIT is always fatal.
1346 	 */
1347 	if (action->sa.sa_handler == SIG_DFL &&
1348 	    (!t->ptrace || (handler == HANDLER_EXIT)))
1349 		t->signal->flags &= ~SIGNAL_UNKILLABLE;
1350 	ret = send_signal_locked(sig, info, t, PIDTYPE_PID);
1351 	/* This can happen if the signal was already pending and blocked */
1352 	if (!task_sigpending(t))
1353 		signal_wake_up(t, 0);
1354 	spin_unlock_irqrestore(&t->sighand->siglock, flags);
1355 
1356 	return ret;
1357 }
1358 
1359 int force_sig_info(struct kernel_siginfo *info)
1360 {
1361 	return force_sig_info_to_task(info, current, HANDLER_CURRENT);
1362 }
1363 
1364 /*
1365  * Nuke all other threads in the group.
1366  */
1367 int zap_other_threads(struct task_struct *p)
1368 {
1369 	struct task_struct *t;
1370 	int count = 0;
1371 
1372 	p->signal->group_stop_count = 0;
1373 
1374 	for_other_threads(p, t) {
1375 		task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1376 		count++;
1377 
1378 		/* Don't bother with already dead threads */
1379 		if (t->exit_state)
1380 			continue;
1381 		sigaddset(&t->pending.signal, SIGKILL);
1382 		signal_wake_up(t, 1);
1383 	}
1384 
1385 	return count;
1386 }
1387 
1388 struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1389 					   unsigned long *flags)
1390 {
1391 	struct sighand_struct *sighand;
1392 
1393 	rcu_read_lock();
1394 	for (;;) {
1395 		sighand = rcu_dereference(tsk->sighand);
1396 		if (unlikely(sighand == NULL))
1397 			break;
1398 
1399 		/*
1400 		 * This sighand can be already freed and even reused, but
1401 		 * we rely on SLAB_TYPESAFE_BY_RCU and sighand_ctor() which
1402 		 * initializes ->siglock: this slab can't go away, it has
1403 		 * the same object type, ->siglock can't be reinitialized.
1404 		 *
1405 		 * We need to ensure that tsk->sighand is still the same
1406 		 * after we take the lock, we can race with de_thread() or
1407 		 * __exit_signal(). In the latter case the next iteration
1408 		 * must see ->sighand == NULL.
1409 		 */
1410 		spin_lock_irqsave(&sighand->siglock, *flags);
1411 		if (likely(sighand == rcu_access_pointer(tsk->sighand)))
1412 			break;
1413 		spin_unlock_irqrestore(&sighand->siglock, *flags);
1414 	}
1415 	rcu_read_unlock();
1416 
1417 	return sighand;
1418 }
1419 
1420 #ifdef CONFIG_LOCKDEP
1421 void lockdep_assert_task_sighand_held(struct task_struct *task)
1422 {
1423 	struct sighand_struct *sighand;
1424 
1425 	rcu_read_lock();
1426 	sighand = rcu_dereference(task->sighand);
1427 	if (sighand)
1428 		lockdep_assert_held(&sighand->siglock);
1429 	else
1430 		WARN_ON_ONCE(1);
1431 	rcu_read_unlock();
1432 }
1433 #endif
1434 
1435 /*
1436  * send signal info to all the members of a thread group or to the
1437  * individual thread if type == PIDTYPE_PID.
1438  */
1439 int group_send_sig_info(int sig, struct kernel_siginfo *info,
1440 			struct task_struct *p, enum pid_type type)
1441 {
1442 	int ret;
1443 
1444 	rcu_read_lock();
1445 	ret = check_kill_permission(sig, info, p);
1446 	rcu_read_unlock();
1447 
1448 	if (!ret && sig)
1449 		ret = do_send_sig_info(sig, info, p, type);
1450 
1451 	return ret;
1452 }
1453 
1454 /*
1455  * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1456  * control characters do (^C, ^Z etc)
1457  * - the caller must hold at least a readlock on tasklist_lock
1458  */
1459 int __kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1460 {
1461 	struct task_struct *p = NULL;
1462 	int ret = -ESRCH;
1463 
1464 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1465 		int err = group_send_sig_info(sig, info, p, PIDTYPE_PGID);
1466 		/*
1467 		 * If group_send_sig_info() succeeds at least once ret
1468 		 * becomes 0 and after that the code below has no effect.
1469 		 * Otherwise we return the last err or -ESRCH if this
1470 		 * process group is empty.
1471 		 */
1472 		if (ret)
1473 			ret = err;
1474 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1475 
1476 	return ret;
1477 }
1478 
1479 static int kill_pid_info_type(int sig, struct kernel_siginfo *info,
1480 				struct pid *pid, enum pid_type type)
1481 {
1482 	int error = -ESRCH;
1483 	struct task_struct *p;
1484 
1485 	for (;;) {
1486 		rcu_read_lock();
1487 		p = pid_task(pid, PIDTYPE_PID);
1488 		if (p)
1489 			error = group_send_sig_info(sig, info, p, type);
1490 		rcu_read_unlock();
1491 		if (likely(!p || error != -ESRCH))
1492 			return error;
1493 		/*
1494 		 * The task was unhashed in between, try again.  If it
1495 		 * is dead, pid_task() will return NULL, if we race with
1496 		 * de_thread() it will find the new leader.
1497 		 */
1498 	}
1499 }
1500 
1501 int kill_pid_info(int sig, struct kernel_siginfo *info, struct pid *pid)
1502 {
1503 	return kill_pid_info_type(sig, info, pid, PIDTYPE_TGID);
1504 }
1505 
1506 static int kill_proc_info(int sig, struct kernel_siginfo *info, pid_t pid)
1507 {
1508 	int error;
1509 	rcu_read_lock();
1510 	error = kill_pid_info(sig, info, find_vpid(pid));
1511 	rcu_read_unlock();
1512 	return error;
1513 }
1514 
1515 static inline bool kill_as_cred_perm(const struct cred *cred,
1516 				     struct task_struct *target)
1517 {
1518 	const struct cred *pcred = __task_cred(target);
1519 
1520 	return uid_eq(cred->euid, pcred->suid) ||
1521 	       uid_eq(cred->euid, pcred->uid) ||
1522 	       uid_eq(cred->uid, pcred->suid) ||
1523 	       uid_eq(cred->uid, pcred->uid);
1524 }
1525 
1526 /*
1527  * The usb asyncio usage of siginfo is wrong.  The glibc support
1528  * for asyncio which uses SI_ASYNCIO assumes the layout is SIL_RT.
1529  * AKA after the generic fields:
1530  *	kernel_pid_t	si_pid;
1531  *	kernel_uid32_t	si_uid;
1532  *	sigval_t	si_value;
1533  *
1534  * Unfortunately when usb generates SI_ASYNCIO it assumes the layout
1535  * after the generic fields is:
1536  *	void __user 	*si_addr;
1537  *
1538  * This is a practical problem when there is a 64bit big endian kernel
1539  * and a 32bit userspace.  As the 32bit address will encoded in the low
1540  * 32bits of the pointer.  Those low 32bits will be stored at higher
1541  * address than appear in a 32 bit pointer.  So userspace will not
1542  * see the address it was expecting for it's completions.
1543  *
1544  * There is nothing in the encoding that can allow
1545  * copy_siginfo_to_user32 to detect this confusion of formats, so
1546  * handle this by requiring the caller of kill_pid_usb_asyncio to
1547  * notice when this situration takes place and to store the 32bit
1548  * pointer in sival_int, instead of sival_addr of the sigval_t addr
1549  * parameter.
1550  */
1551 int kill_pid_usb_asyncio(int sig, int errno, sigval_t addr,
1552 			 struct pid *pid, const struct cred *cred)
1553 {
1554 	struct kernel_siginfo info;
1555 	struct task_struct *p;
1556 	unsigned long flags;
1557 	int ret = -EINVAL;
1558 
1559 	if (!valid_signal(sig))
1560 		return ret;
1561 
1562 	clear_siginfo(&info);
1563 	info.si_signo = sig;
1564 	info.si_errno = errno;
1565 	info.si_code = SI_ASYNCIO;
1566 	*((sigval_t *)&info.si_pid) = addr;
1567 
1568 	rcu_read_lock();
1569 	p = pid_task(pid, PIDTYPE_PID);
1570 	if (!p) {
1571 		ret = -ESRCH;
1572 		goto out_unlock;
1573 	}
1574 	if (!kill_as_cred_perm(cred, p)) {
1575 		ret = -EPERM;
1576 		goto out_unlock;
1577 	}
1578 	ret = security_task_kill(p, &info, sig, cred);
1579 	if (ret)
1580 		goto out_unlock;
1581 
1582 	if (sig) {
1583 		if (lock_task_sighand(p, &flags)) {
1584 			ret = __send_signal_locked(sig, &info, p, PIDTYPE_TGID, false);
1585 			unlock_task_sighand(p, &flags);
1586 		} else
1587 			ret = -ESRCH;
1588 	}
1589 out_unlock:
1590 	rcu_read_unlock();
1591 	return ret;
1592 }
1593 EXPORT_SYMBOL_GPL(kill_pid_usb_asyncio);
1594 
1595 /*
1596  * kill_something_info() interprets pid in interesting ways just like kill(2).
1597  *
1598  * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1599  * is probably wrong.  Should make it like BSD or SYSV.
1600  */
1601 
1602 static int kill_something_info(int sig, struct kernel_siginfo *info, pid_t pid)
1603 {
1604 	int ret;
1605 
1606 	if (pid > 0)
1607 		return kill_proc_info(sig, info, pid);
1608 
1609 	/* -INT_MIN is undefined.  Exclude this case to avoid a UBSAN warning */
1610 	if (pid == INT_MIN)
1611 		return -ESRCH;
1612 
1613 	read_lock(&tasklist_lock);
1614 	if (pid != -1) {
1615 		ret = __kill_pgrp_info(sig, info,
1616 				pid ? find_vpid(-pid) : task_pgrp(current));
1617 	} else {
1618 		int retval = 0, count = 0;
1619 		struct task_struct * p;
1620 
1621 		for_each_process(p) {
1622 			if (task_pid_vnr(p) > 1 &&
1623 					!same_thread_group(p, current)) {
1624 				int err = group_send_sig_info(sig, info, p,
1625 							      PIDTYPE_MAX);
1626 				++count;
1627 				if (err != -EPERM)
1628 					retval = err;
1629 			}
1630 		}
1631 		ret = count ? retval : -ESRCH;
1632 	}
1633 	read_unlock(&tasklist_lock);
1634 
1635 	return ret;
1636 }
1637 
1638 /*
1639  * These are for backward compatibility with the rest of the kernel source.
1640  */
1641 
1642 int send_sig_info(int sig, struct kernel_siginfo *info, struct task_struct *p)
1643 {
1644 	/*
1645 	 * Make sure legacy kernel users don't send in bad values
1646 	 * (normal paths check this in check_kill_permission).
1647 	 */
1648 	if (!valid_signal(sig))
1649 		return -EINVAL;
1650 
1651 	return do_send_sig_info(sig, info, p, PIDTYPE_PID);
1652 }
1653 EXPORT_SYMBOL(send_sig_info);
1654 
1655 #define __si_special(priv) \
1656 	((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1657 
1658 int
1659 send_sig(int sig, struct task_struct *p, int priv)
1660 {
1661 	return send_sig_info(sig, __si_special(priv), p);
1662 }
1663 EXPORT_SYMBOL(send_sig);
1664 
1665 void force_sig(int sig)
1666 {
1667 	struct kernel_siginfo info;
1668 
1669 	clear_siginfo(&info);
1670 	info.si_signo = sig;
1671 	info.si_errno = 0;
1672 	info.si_code = SI_KERNEL;
1673 	info.si_pid = 0;
1674 	info.si_uid = 0;
1675 	force_sig_info(&info);
1676 }
1677 EXPORT_SYMBOL(force_sig);
1678 
1679 void force_fatal_sig(int sig)
1680 {
1681 	struct kernel_siginfo info;
1682 
1683 	clear_siginfo(&info);
1684 	info.si_signo = sig;
1685 	info.si_errno = 0;
1686 	info.si_code = SI_KERNEL;
1687 	info.si_pid = 0;
1688 	info.si_uid = 0;
1689 	force_sig_info_to_task(&info, current, HANDLER_SIG_DFL);
1690 }
1691 
1692 void force_exit_sig(int sig)
1693 {
1694 	struct kernel_siginfo info;
1695 
1696 	clear_siginfo(&info);
1697 	info.si_signo = sig;
1698 	info.si_errno = 0;
1699 	info.si_code = SI_KERNEL;
1700 	info.si_pid = 0;
1701 	info.si_uid = 0;
1702 	force_sig_info_to_task(&info, current, HANDLER_EXIT);
1703 }
1704 
1705 /*
1706  * When things go south during signal handling, we
1707  * will force a SIGSEGV. And if the signal that caused
1708  * the problem was already a SIGSEGV, we'll want to
1709  * make sure we don't even try to deliver the signal..
1710  */
1711 void force_sigsegv(int sig)
1712 {
1713 	if (sig == SIGSEGV)
1714 		force_fatal_sig(SIGSEGV);
1715 	else
1716 		force_sig(SIGSEGV);
1717 }
1718 
1719 int force_sig_fault_to_task(int sig, int code, void __user *addr,
1720 			    struct task_struct *t)
1721 {
1722 	struct kernel_siginfo info;
1723 
1724 	clear_siginfo(&info);
1725 	info.si_signo = sig;
1726 	info.si_errno = 0;
1727 	info.si_code  = code;
1728 	info.si_addr  = addr;
1729 	return force_sig_info_to_task(&info, t, HANDLER_CURRENT);
1730 }
1731 
1732 int force_sig_fault(int sig, int code, void __user *addr)
1733 {
1734 	return force_sig_fault_to_task(sig, code, addr, current);
1735 }
1736 
1737 int send_sig_fault(int sig, int code, void __user *addr, struct task_struct *t)
1738 {
1739 	struct kernel_siginfo info;
1740 
1741 	clear_siginfo(&info);
1742 	info.si_signo = sig;
1743 	info.si_errno = 0;
1744 	info.si_code  = code;
1745 	info.si_addr  = addr;
1746 	return send_sig_info(info.si_signo, &info, t);
1747 }
1748 
1749 int force_sig_mceerr(int code, void __user *addr, short lsb)
1750 {
1751 	struct kernel_siginfo info;
1752 
1753 	WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1754 	clear_siginfo(&info);
1755 	info.si_signo = SIGBUS;
1756 	info.si_errno = 0;
1757 	info.si_code = code;
1758 	info.si_addr = addr;
1759 	info.si_addr_lsb = lsb;
1760 	return force_sig_info(&info);
1761 }
1762 
1763 int send_sig_mceerr(int code, void __user *addr, short lsb, struct task_struct *t)
1764 {
1765 	struct kernel_siginfo info;
1766 
1767 	WARN_ON((code != BUS_MCEERR_AO) && (code != BUS_MCEERR_AR));
1768 	clear_siginfo(&info);
1769 	info.si_signo = SIGBUS;
1770 	info.si_errno = 0;
1771 	info.si_code = code;
1772 	info.si_addr = addr;
1773 	info.si_addr_lsb = lsb;
1774 	return send_sig_info(info.si_signo, &info, t);
1775 }
1776 EXPORT_SYMBOL(send_sig_mceerr);
1777 
1778 int force_sig_bnderr(void __user *addr, void __user *lower, void __user *upper)
1779 {
1780 	struct kernel_siginfo info;
1781 
1782 	clear_siginfo(&info);
1783 	info.si_signo = SIGSEGV;
1784 	info.si_errno = 0;
1785 	info.si_code  = SEGV_BNDERR;
1786 	info.si_addr  = addr;
1787 	info.si_lower = lower;
1788 	info.si_upper = upper;
1789 	return force_sig_info(&info);
1790 }
1791 
1792 #ifdef SEGV_PKUERR
1793 int force_sig_pkuerr(void __user *addr, u32 pkey)
1794 {
1795 	struct kernel_siginfo info;
1796 
1797 	clear_siginfo(&info);
1798 	info.si_signo = SIGSEGV;
1799 	info.si_errno = 0;
1800 	info.si_code  = SEGV_PKUERR;
1801 	info.si_addr  = addr;
1802 	info.si_pkey  = pkey;
1803 	return force_sig_info(&info);
1804 }
1805 #endif
1806 
1807 int send_sig_perf(void __user *addr, u32 type, u64 sig_data)
1808 {
1809 	struct kernel_siginfo info;
1810 
1811 	clear_siginfo(&info);
1812 	info.si_signo     = SIGTRAP;
1813 	info.si_errno     = 0;
1814 	info.si_code      = TRAP_PERF;
1815 	info.si_addr      = addr;
1816 	info.si_perf_data = sig_data;
1817 	info.si_perf_type = type;
1818 
1819 	/*
1820 	 * Signals generated by perf events should not terminate the whole
1821 	 * process if SIGTRAP is blocked, however, delivering the signal
1822 	 * asynchronously is better than not delivering at all. But tell user
1823 	 * space if the signal was asynchronous, so it can clearly be
1824 	 * distinguished from normal synchronous ones.
1825 	 */
1826 	info.si_perf_flags = sigismember(&current->blocked, info.si_signo) ?
1827 				     TRAP_PERF_FLAG_ASYNC :
1828 				     0;
1829 
1830 	return send_sig_info(info.si_signo, &info, current);
1831 }
1832 
1833 /**
1834  * force_sig_seccomp - signals the task to allow in-process syscall emulation
1835  * @syscall: syscall number to send to userland
1836  * @reason: filter-supplied reason code to send to userland (via si_errno)
1837  * @force_coredump: true to trigger a coredump
1838  *
1839  * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
1840  */
1841 int force_sig_seccomp(int syscall, int reason, bool force_coredump)
1842 {
1843 	struct kernel_siginfo info;
1844 
1845 	clear_siginfo(&info);
1846 	info.si_signo = SIGSYS;
1847 	info.si_code = SYS_SECCOMP;
1848 	info.si_call_addr = (void __user *)KSTK_EIP(current);
1849 	info.si_errno = reason;
1850 	info.si_arch = syscall_get_arch(current);
1851 	info.si_syscall = syscall;
1852 	return force_sig_info_to_task(&info, current,
1853 		force_coredump ? HANDLER_EXIT : HANDLER_CURRENT);
1854 }
1855 
1856 /* For the crazy architectures that include trap information in
1857  * the errno field, instead of an actual errno value.
1858  */
1859 int force_sig_ptrace_errno_trap(int errno, void __user *addr)
1860 {
1861 	struct kernel_siginfo info;
1862 
1863 	clear_siginfo(&info);
1864 	info.si_signo = SIGTRAP;
1865 	info.si_errno = errno;
1866 	info.si_code  = TRAP_HWBKPT;
1867 	info.si_addr  = addr;
1868 	return force_sig_info(&info);
1869 }
1870 
1871 /* For the rare architectures that include trap information using
1872  * si_trapno.
1873  */
1874 int force_sig_fault_trapno(int sig, int code, void __user *addr, int trapno)
1875 {
1876 	struct kernel_siginfo info;
1877 
1878 	clear_siginfo(&info);
1879 	info.si_signo = sig;
1880 	info.si_errno = 0;
1881 	info.si_code  = code;
1882 	info.si_addr  = addr;
1883 	info.si_trapno = trapno;
1884 	return force_sig_info(&info);
1885 }
1886 
1887 /* For the rare architectures that include trap information using
1888  * si_trapno.
1889  */
1890 int send_sig_fault_trapno(int sig, int code, void __user *addr, int trapno,
1891 			  struct task_struct *t)
1892 {
1893 	struct kernel_siginfo info;
1894 
1895 	clear_siginfo(&info);
1896 	info.si_signo = sig;
1897 	info.si_errno = 0;
1898 	info.si_code  = code;
1899 	info.si_addr  = addr;
1900 	info.si_trapno = trapno;
1901 	return send_sig_info(info.si_signo, &info, t);
1902 }
1903 
1904 static int kill_pgrp_info(int sig, struct kernel_siginfo *info, struct pid *pgrp)
1905 {
1906 	int ret;
1907 	read_lock(&tasklist_lock);
1908 	ret = __kill_pgrp_info(sig, info, pgrp);
1909 	read_unlock(&tasklist_lock);
1910 	return ret;
1911 }
1912 
1913 int kill_pgrp(struct pid *pid, int sig, int priv)
1914 {
1915 	return kill_pgrp_info(sig, __si_special(priv), pid);
1916 }
1917 EXPORT_SYMBOL(kill_pgrp);
1918 
1919 int kill_pid(struct pid *pid, int sig, int priv)
1920 {
1921 	return kill_pid_info(sig, __si_special(priv), pid);
1922 }
1923 EXPORT_SYMBOL(kill_pid);
1924 
1925 /*
1926  * These functions support sending signals using preallocated sigqueue
1927  * structures.  This is needed "because realtime applications cannot
1928  * afford to lose notifications of asynchronous events, like timer
1929  * expirations or I/O completions".  In the case of POSIX Timers
1930  * we allocate the sigqueue structure from the timer_create.  If this
1931  * allocation fails we are able to report the failure to the application
1932  * with an EAGAIN error.
1933  */
1934 struct sigqueue *sigqueue_alloc(void)
1935 {
1936 	return __sigqueue_alloc(-1, current, GFP_KERNEL, 0, SIGQUEUE_PREALLOC);
1937 }
1938 
1939 void sigqueue_free(struct sigqueue *q)
1940 {
1941 	spinlock_t *lock = &current->sighand->siglock;
1942 	unsigned long flags;
1943 
1944 	if (WARN_ON_ONCE(!(q->flags & SIGQUEUE_PREALLOC)))
1945 		return;
1946 	/*
1947 	 * We must hold ->siglock while testing q->list
1948 	 * to serialize with collect_signal() or with
1949 	 * __exit_signal()->flush_sigqueue().
1950 	 */
1951 	spin_lock_irqsave(lock, flags);
1952 	q->flags &= ~SIGQUEUE_PREALLOC;
1953 	/*
1954 	 * If it is queued it will be freed when dequeued,
1955 	 * like the "regular" sigqueue.
1956 	 */
1957 	if (!list_empty(&q->list))
1958 		q = NULL;
1959 	spin_unlock_irqrestore(lock, flags);
1960 
1961 	if (q)
1962 		__sigqueue_free(q);
1963 }
1964 
1965 int send_sigqueue(struct sigqueue *q, struct pid *pid, enum pid_type type)
1966 {
1967 	int sig = q->info.si_signo;
1968 	struct sigpending *pending;
1969 	struct task_struct *t;
1970 	unsigned long flags;
1971 	int ret, result;
1972 
1973 	if (WARN_ON_ONCE(!(q->flags & SIGQUEUE_PREALLOC)))
1974 		return 0;
1975 	if (WARN_ON_ONCE(q->info.si_code != SI_TIMER))
1976 		return 0;
1977 
1978 	ret = -1;
1979 	rcu_read_lock();
1980 
1981 	/*
1982 	 * This function is used by POSIX timers to deliver a timer signal.
1983 	 * Where type is PIDTYPE_PID (such as for timers with SIGEV_THREAD_ID
1984 	 * set), the signal must be delivered to the specific thread (queues
1985 	 * into t->pending).
1986 	 *
1987 	 * Where type is not PIDTYPE_PID, signals must be delivered to the
1988 	 * process. In this case, prefer to deliver to current if it is in
1989 	 * the same thread group as the target process, which avoids
1990 	 * unnecessarily waking up a potentially idle task.
1991 	 */
1992 	t = pid_task(pid, type);
1993 	if (!t)
1994 		goto ret;
1995 	if (type != PIDTYPE_PID && same_thread_group(t, current))
1996 		t = current;
1997 	if (!likely(lock_task_sighand(t, &flags)))
1998 		goto ret;
1999 
2000 	ret = 1; /* the signal is ignored */
2001 	result = TRACE_SIGNAL_IGNORED;
2002 	if (!prepare_signal(sig, t, false))
2003 		goto out;
2004 
2005 	ret = 0;
2006 	if (unlikely(!list_empty(&q->list))) {
2007 		/*
2008 		 * If an SI_TIMER entry is already queue just increment
2009 		 * the overrun count.
2010 		 */
2011 		q->info.si_overrun++;
2012 		result = TRACE_SIGNAL_ALREADY_PENDING;
2013 		goto out;
2014 	}
2015 	q->info.si_overrun = 0;
2016 
2017 	signalfd_notify(t, sig);
2018 	pending = (type != PIDTYPE_PID) ? &t->signal->shared_pending : &t->pending;
2019 	list_add_tail(&q->list, &pending->list);
2020 	sigaddset(&pending->signal, sig);
2021 	complete_signal(sig, t, type);
2022 	result = TRACE_SIGNAL_DELIVERED;
2023 out:
2024 	trace_signal_generate(sig, &q->info, t, type != PIDTYPE_PID, result);
2025 	unlock_task_sighand(t, &flags);
2026 ret:
2027 	rcu_read_unlock();
2028 	return ret;
2029 }
2030 
2031 void do_notify_pidfd(struct task_struct *task)
2032 {
2033 	struct pid *pid = task_pid(task);
2034 
2035 	WARN_ON(task->exit_state == 0);
2036 
2037 	__wake_up(&pid->wait_pidfd, TASK_NORMAL, 0,
2038 			poll_to_key(EPOLLIN | EPOLLRDNORM));
2039 }
2040 
2041 /*
2042  * Let a parent know about the death of a child.
2043  * For a stopped/continued status change, use do_notify_parent_cldstop instead.
2044  *
2045  * Returns true if our parent ignored us and so we've switched to
2046  * self-reaping.
2047  */
2048 bool do_notify_parent(struct task_struct *tsk, int sig)
2049 {
2050 	struct kernel_siginfo info;
2051 	unsigned long flags;
2052 	struct sighand_struct *psig;
2053 	bool autoreap = false;
2054 	u64 utime, stime;
2055 
2056 	WARN_ON_ONCE(sig == -1);
2057 
2058 	/* do_notify_parent_cldstop should have been called instead.  */
2059 	WARN_ON_ONCE(task_is_stopped_or_traced(tsk));
2060 
2061 	WARN_ON_ONCE(!tsk->ptrace &&
2062 	       (tsk->group_leader != tsk || !thread_group_empty(tsk)));
2063 	/*
2064 	 * tsk is a group leader and has no threads, wake up the
2065 	 * non-PIDFD_THREAD waiters.
2066 	 */
2067 	if (thread_group_empty(tsk))
2068 		do_notify_pidfd(tsk);
2069 
2070 	if (sig != SIGCHLD) {
2071 		/*
2072 		 * This is only possible if parent == real_parent.
2073 		 * Check if it has changed security domain.
2074 		 */
2075 		if (tsk->parent_exec_id != READ_ONCE(tsk->parent->self_exec_id))
2076 			sig = SIGCHLD;
2077 	}
2078 
2079 	clear_siginfo(&info);
2080 	info.si_signo = sig;
2081 	info.si_errno = 0;
2082 	/*
2083 	 * We are under tasklist_lock here so our parent is tied to
2084 	 * us and cannot change.
2085 	 *
2086 	 * task_active_pid_ns will always return the same pid namespace
2087 	 * until a task passes through release_task.
2088 	 *
2089 	 * write_lock() currently calls preempt_disable() which is the
2090 	 * same as rcu_read_lock(), but according to Oleg, this is not
2091 	 * correct to rely on this
2092 	 */
2093 	rcu_read_lock();
2094 	info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(tsk->parent));
2095 	info.si_uid = from_kuid_munged(task_cred_xxx(tsk->parent, user_ns),
2096 				       task_uid(tsk));
2097 	rcu_read_unlock();
2098 
2099 	task_cputime(tsk, &utime, &stime);
2100 	info.si_utime = nsec_to_clock_t(utime + tsk->signal->utime);
2101 	info.si_stime = nsec_to_clock_t(stime + tsk->signal->stime);
2102 
2103 	info.si_status = tsk->exit_code & 0x7f;
2104 	if (tsk->exit_code & 0x80)
2105 		info.si_code = CLD_DUMPED;
2106 	else if (tsk->exit_code & 0x7f)
2107 		info.si_code = CLD_KILLED;
2108 	else {
2109 		info.si_code = CLD_EXITED;
2110 		info.si_status = tsk->exit_code >> 8;
2111 	}
2112 
2113 	psig = tsk->parent->sighand;
2114 	spin_lock_irqsave(&psig->siglock, flags);
2115 	if (!tsk->ptrace && sig == SIGCHLD &&
2116 	    (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
2117 	     (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
2118 		/*
2119 		 * We are exiting and our parent doesn't care.  POSIX.1
2120 		 * defines special semantics for setting SIGCHLD to SIG_IGN
2121 		 * or setting the SA_NOCLDWAIT flag: we should be reaped
2122 		 * automatically and not left for our parent's wait4 call.
2123 		 * Rather than having the parent do it as a magic kind of
2124 		 * signal handler, we just set this to tell do_exit that we
2125 		 * can be cleaned up without becoming a zombie.  Note that
2126 		 * we still call __wake_up_parent in this case, because a
2127 		 * blocked sys_wait4 might now return -ECHILD.
2128 		 *
2129 		 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
2130 		 * is implementation-defined: we do (if you don't want
2131 		 * it, just use SIG_IGN instead).
2132 		 */
2133 		autoreap = true;
2134 		if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
2135 			sig = 0;
2136 	}
2137 	/*
2138 	 * Send with __send_signal as si_pid and si_uid are in the
2139 	 * parent's namespaces.
2140 	 */
2141 	if (valid_signal(sig) && sig)
2142 		__send_signal_locked(sig, &info, tsk->parent, PIDTYPE_TGID, false);
2143 	__wake_up_parent(tsk, tsk->parent);
2144 	spin_unlock_irqrestore(&psig->siglock, flags);
2145 
2146 	return autoreap;
2147 }
2148 
2149 /**
2150  * do_notify_parent_cldstop - notify parent of stopped/continued state change
2151  * @tsk: task reporting the state change
2152  * @for_ptracer: the notification is for ptracer
2153  * @why: CLD_{CONTINUED|STOPPED|TRAPPED} to report
2154  *
2155  * Notify @tsk's parent that the stopped/continued state has changed.  If
2156  * @for_ptracer is %false, @tsk's group leader notifies to its real parent.
2157  * If %true, @tsk reports to @tsk->parent which should be the ptracer.
2158  *
2159  * CONTEXT:
2160  * Must be called with tasklist_lock at least read locked.
2161  */
2162 static void do_notify_parent_cldstop(struct task_struct *tsk,
2163 				     bool for_ptracer, int why)
2164 {
2165 	struct kernel_siginfo info;
2166 	unsigned long flags;
2167 	struct task_struct *parent;
2168 	struct sighand_struct *sighand;
2169 	u64 utime, stime;
2170 
2171 	if (for_ptracer) {
2172 		parent = tsk->parent;
2173 	} else {
2174 		tsk = tsk->group_leader;
2175 		parent = tsk->real_parent;
2176 	}
2177 
2178 	clear_siginfo(&info);
2179 	info.si_signo = SIGCHLD;
2180 	info.si_errno = 0;
2181 	/*
2182 	 * see comment in do_notify_parent() about the following 4 lines
2183 	 */
2184 	rcu_read_lock();
2185 	info.si_pid = task_pid_nr_ns(tsk, task_active_pid_ns(parent));
2186 	info.si_uid = from_kuid_munged(task_cred_xxx(parent, user_ns), task_uid(tsk));
2187 	rcu_read_unlock();
2188 
2189 	task_cputime(tsk, &utime, &stime);
2190 	info.si_utime = nsec_to_clock_t(utime);
2191 	info.si_stime = nsec_to_clock_t(stime);
2192 
2193  	info.si_code = why;
2194  	switch (why) {
2195  	case CLD_CONTINUED:
2196  		info.si_status = SIGCONT;
2197  		break;
2198  	case CLD_STOPPED:
2199  		info.si_status = tsk->signal->group_exit_code & 0x7f;
2200  		break;
2201  	case CLD_TRAPPED:
2202  		info.si_status = tsk->exit_code & 0x7f;
2203  		break;
2204  	default:
2205  		BUG();
2206  	}
2207 
2208 	sighand = parent->sighand;
2209 	spin_lock_irqsave(&sighand->siglock, flags);
2210 	if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
2211 	    !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
2212 		send_signal_locked(SIGCHLD, &info, parent, PIDTYPE_TGID);
2213 	/*
2214 	 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
2215 	 */
2216 	__wake_up_parent(tsk, parent);
2217 	spin_unlock_irqrestore(&sighand->siglock, flags);
2218 }
2219 
2220 /*
2221  * This must be called with current->sighand->siglock held.
2222  *
2223  * This should be the path for all ptrace stops.
2224  * We always set current->last_siginfo while stopped here.
2225  * That makes it a way to test a stopped process for
2226  * being ptrace-stopped vs being job-control-stopped.
2227  *
2228  * Returns the signal the ptracer requested the code resume
2229  * with.  If the code did not stop because the tracer is gone,
2230  * the stop signal remains unchanged unless clear_code.
2231  */
2232 static int ptrace_stop(int exit_code, int why, unsigned long message,
2233 		       kernel_siginfo_t *info)
2234 	__releases(&current->sighand->siglock)
2235 	__acquires(&current->sighand->siglock)
2236 {
2237 	bool gstop_done = false;
2238 
2239 	if (arch_ptrace_stop_needed()) {
2240 		/*
2241 		 * The arch code has something special to do before a
2242 		 * ptrace stop.  This is allowed to block, e.g. for faults
2243 		 * on user stack pages.  We can't keep the siglock while
2244 		 * calling arch_ptrace_stop, so we must release it now.
2245 		 * To preserve proper semantics, we must do this before
2246 		 * any signal bookkeeping like checking group_stop_count.
2247 		 */
2248 		spin_unlock_irq(&current->sighand->siglock);
2249 		arch_ptrace_stop();
2250 		spin_lock_irq(&current->sighand->siglock);
2251 	}
2252 
2253 	/*
2254 	 * After this point ptrace_signal_wake_up or signal_wake_up
2255 	 * will clear TASK_TRACED if ptrace_unlink happens or a fatal
2256 	 * signal comes in.  Handle previous ptrace_unlinks and fatal
2257 	 * signals here to prevent ptrace_stop sleeping in schedule.
2258 	 */
2259 	if (!current->ptrace || __fatal_signal_pending(current))
2260 		return exit_code;
2261 
2262 	set_special_state(TASK_TRACED);
2263 	current->jobctl |= JOBCTL_TRACED;
2264 
2265 	/*
2266 	 * We're committing to trapping.  TRACED should be visible before
2267 	 * TRAPPING is cleared; otherwise, the tracer might fail do_wait().
2268 	 * Also, transition to TRACED and updates to ->jobctl should be
2269 	 * atomic with respect to siglock and should be done after the arch
2270 	 * hook as siglock is released and regrabbed across it.
2271 	 *
2272 	 *     TRACER				    TRACEE
2273 	 *
2274 	 *     ptrace_attach()
2275 	 * [L]   wait_on_bit(JOBCTL_TRAPPING)	[S] set_special_state(TRACED)
2276 	 *     do_wait()
2277 	 *       set_current_state()                smp_wmb();
2278 	 *       ptrace_do_wait()
2279 	 *         wait_task_stopped()
2280 	 *           task_stopped_code()
2281 	 * [L]         task_is_traced()		[S] task_clear_jobctl_trapping();
2282 	 */
2283 	smp_wmb();
2284 
2285 	current->ptrace_message = message;
2286 	current->last_siginfo = info;
2287 	current->exit_code = exit_code;
2288 
2289 	/*
2290 	 * If @why is CLD_STOPPED, we're trapping to participate in a group
2291 	 * stop.  Do the bookkeeping.  Note that if SIGCONT was delievered
2292 	 * across siglock relocks since INTERRUPT was scheduled, PENDING
2293 	 * could be clear now.  We act as if SIGCONT is received after
2294 	 * TASK_TRACED is entered - ignore it.
2295 	 */
2296 	if (why == CLD_STOPPED && (current->jobctl & JOBCTL_STOP_PENDING))
2297 		gstop_done = task_participate_group_stop(current);
2298 
2299 	/* any trap clears pending STOP trap, STOP trap clears NOTIFY */
2300 	task_clear_jobctl_pending(current, JOBCTL_TRAP_STOP);
2301 	if (info && info->si_code >> 8 == PTRACE_EVENT_STOP)
2302 		task_clear_jobctl_pending(current, JOBCTL_TRAP_NOTIFY);
2303 
2304 	/* entering a trap, clear TRAPPING */
2305 	task_clear_jobctl_trapping(current);
2306 
2307 	spin_unlock_irq(&current->sighand->siglock);
2308 	read_lock(&tasklist_lock);
2309 	/*
2310 	 * Notify parents of the stop.
2311 	 *
2312 	 * While ptraced, there are two parents - the ptracer and
2313 	 * the real_parent of the group_leader.  The ptracer should
2314 	 * know about every stop while the real parent is only
2315 	 * interested in the completion of group stop.  The states
2316 	 * for the two don't interact with each other.  Notify
2317 	 * separately unless they're gonna be duplicates.
2318 	 */
2319 	if (current->ptrace)
2320 		do_notify_parent_cldstop(current, true, why);
2321 	if (gstop_done && (!current->ptrace || ptrace_reparented(current)))
2322 		do_notify_parent_cldstop(current, false, why);
2323 
2324 	/*
2325 	 * The previous do_notify_parent_cldstop() invocation woke ptracer.
2326 	 * One a PREEMPTION kernel this can result in preemption requirement
2327 	 * which will be fulfilled after read_unlock() and the ptracer will be
2328 	 * put on the CPU.
2329 	 * The ptracer is in wait_task_inactive(, __TASK_TRACED) waiting for
2330 	 * this task wait in schedule(). If this task gets preempted then it
2331 	 * remains enqueued on the runqueue. The ptracer will observe this and
2332 	 * then sleep for a delay of one HZ tick. In the meantime this task
2333 	 * gets scheduled, enters schedule() and will wait for the ptracer.
2334 	 *
2335 	 * This preemption point is not bad from a correctness point of
2336 	 * view but extends the runtime by one HZ tick time due to the
2337 	 * ptracer's sleep.  The preempt-disable section ensures that there
2338 	 * will be no preemption between unlock and schedule() and so
2339 	 * improving the performance since the ptracer will observe that
2340 	 * the tracee is scheduled out once it gets on the CPU.
2341 	 *
2342 	 * On PREEMPT_RT locking tasklist_lock does not disable preemption.
2343 	 * Therefore the task can be preempted after do_notify_parent_cldstop()
2344 	 * before unlocking tasklist_lock so there is no benefit in doing this.
2345 	 *
2346 	 * In fact disabling preemption is harmful on PREEMPT_RT because
2347 	 * the spinlock_t in cgroup_enter_frozen() must not be acquired
2348 	 * with preemption disabled due to the 'sleeping' spinlock
2349 	 * substitution of RT.
2350 	 */
2351 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
2352 		preempt_disable();
2353 	read_unlock(&tasklist_lock);
2354 	cgroup_enter_frozen();
2355 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
2356 		preempt_enable_no_resched();
2357 	schedule();
2358 	cgroup_leave_frozen(true);
2359 
2360 	/*
2361 	 * We are back.  Now reacquire the siglock before touching
2362 	 * last_siginfo, so that we are sure to have synchronized with
2363 	 * any signal-sending on another CPU that wants to examine it.
2364 	 */
2365 	spin_lock_irq(&current->sighand->siglock);
2366 	exit_code = current->exit_code;
2367 	current->last_siginfo = NULL;
2368 	current->ptrace_message = 0;
2369 	current->exit_code = 0;
2370 
2371 	/* LISTENING can be set only during STOP traps, clear it */
2372 	current->jobctl &= ~(JOBCTL_LISTENING | JOBCTL_PTRACE_FROZEN);
2373 
2374 	/*
2375 	 * Queued signals ignored us while we were stopped for tracing.
2376 	 * So check for any that we should take before resuming user mode.
2377 	 * This sets TIF_SIGPENDING, but never clears it.
2378 	 */
2379 	recalc_sigpending_tsk(current);
2380 	return exit_code;
2381 }
2382 
2383 static int ptrace_do_notify(int signr, int exit_code, int why, unsigned long message)
2384 {
2385 	kernel_siginfo_t info;
2386 
2387 	clear_siginfo(&info);
2388 	info.si_signo = signr;
2389 	info.si_code = exit_code;
2390 	info.si_pid = task_pid_vnr(current);
2391 	info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
2392 
2393 	/* Let the debugger run.  */
2394 	return ptrace_stop(exit_code, why, message, &info);
2395 }
2396 
2397 int ptrace_notify(int exit_code, unsigned long message)
2398 {
2399 	int signr;
2400 
2401 	BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
2402 	if (unlikely(task_work_pending(current)))
2403 		task_work_run();
2404 
2405 	spin_lock_irq(&current->sighand->siglock);
2406 	signr = ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED, message);
2407 	spin_unlock_irq(&current->sighand->siglock);
2408 	return signr;
2409 }
2410 
2411 /**
2412  * do_signal_stop - handle group stop for SIGSTOP and other stop signals
2413  * @signr: signr causing group stop if initiating
2414  *
2415  * If %JOBCTL_STOP_PENDING is not set yet, initiate group stop with @signr
2416  * and participate in it.  If already set, participate in the existing
2417  * group stop.  If participated in a group stop (and thus slept), %true is
2418  * returned with siglock released.
2419  *
2420  * If ptraced, this function doesn't handle stop itself.  Instead,
2421  * %JOBCTL_TRAP_STOP is scheduled and %false is returned with siglock
2422  * untouched.  The caller must ensure that INTERRUPT trap handling takes
2423  * places afterwards.
2424  *
2425  * CONTEXT:
2426  * Must be called with @current->sighand->siglock held, which is released
2427  * on %true return.
2428  *
2429  * RETURNS:
2430  * %false if group stop is already cancelled or ptrace trap is scheduled.
2431  * %true if participated in group stop.
2432  */
2433 static bool do_signal_stop(int signr)
2434 	__releases(&current->sighand->siglock)
2435 {
2436 	struct signal_struct *sig = current->signal;
2437 
2438 	if (!(current->jobctl & JOBCTL_STOP_PENDING)) {
2439 		unsigned long gstop = JOBCTL_STOP_PENDING | JOBCTL_STOP_CONSUME;
2440 		struct task_struct *t;
2441 
2442 		/* signr will be recorded in task->jobctl for retries */
2443 		WARN_ON_ONCE(signr & ~JOBCTL_STOP_SIGMASK);
2444 
2445 		if (!likely(current->jobctl & JOBCTL_STOP_DEQUEUED) ||
2446 		    unlikely(sig->flags & SIGNAL_GROUP_EXIT) ||
2447 		    unlikely(sig->group_exec_task))
2448 			return false;
2449 		/*
2450 		 * There is no group stop already in progress.  We must
2451 		 * initiate one now.
2452 		 *
2453 		 * While ptraced, a task may be resumed while group stop is
2454 		 * still in effect and then receive a stop signal and
2455 		 * initiate another group stop.  This deviates from the
2456 		 * usual behavior as two consecutive stop signals can't
2457 		 * cause two group stops when !ptraced.  That is why we
2458 		 * also check !task_is_stopped(t) below.
2459 		 *
2460 		 * The condition can be distinguished by testing whether
2461 		 * SIGNAL_STOP_STOPPED is already set.  Don't generate
2462 		 * group_exit_code in such case.
2463 		 *
2464 		 * This is not necessary for SIGNAL_STOP_CONTINUED because
2465 		 * an intervening stop signal is required to cause two
2466 		 * continued events regardless of ptrace.
2467 		 */
2468 		if (!(sig->flags & SIGNAL_STOP_STOPPED))
2469 			sig->group_exit_code = signr;
2470 
2471 		sig->group_stop_count = 0;
2472 		if (task_set_jobctl_pending(current, signr | gstop))
2473 			sig->group_stop_count++;
2474 
2475 		for_other_threads(current, t) {
2476 			/*
2477 			 * Setting state to TASK_STOPPED for a group
2478 			 * stop is always done with the siglock held,
2479 			 * so this check has no races.
2480 			 */
2481 			if (!task_is_stopped(t) &&
2482 			    task_set_jobctl_pending(t, signr | gstop)) {
2483 				sig->group_stop_count++;
2484 				if (likely(!(t->ptrace & PT_SEIZED)))
2485 					signal_wake_up(t, 0);
2486 				else
2487 					ptrace_trap_notify(t);
2488 			}
2489 		}
2490 	}
2491 
2492 	if (likely(!current->ptrace)) {
2493 		int notify = 0;
2494 
2495 		/*
2496 		 * If there are no other threads in the group, or if there
2497 		 * is a group stop in progress and we are the last to stop,
2498 		 * report to the parent.
2499 		 */
2500 		if (task_participate_group_stop(current))
2501 			notify = CLD_STOPPED;
2502 
2503 		current->jobctl |= JOBCTL_STOPPED;
2504 		set_special_state(TASK_STOPPED);
2505 		spin_unlock_irq(&current->sighand->siglock);
2506 
2507 		/*
2508 		 * Notify the parent of the group stop completion.  Because
2509 		 * we're not holding either the siglock or tasklist_lock
2510 		 * here, ptracer may attach inbetween; however, this is for
2511 		 * group stop and should always be delivered to the real
2512 		 * parent of the group leader.  The new ptracer will get
2513 		 * its notification when this task transitions into
2514 		 * TASK_TRACED.
2515 		 */
2516 		if (notify) {
2517 			read_lock(&tasklist_lock);
2518 			do_notify_parent_cldstop(current, false, notify);
2519 			read_unlock(&tasklist_lock);
2520 		}
2521 
2522 		/* Now we don't run again until woken by SIGCONT or SIGKILL */
2523 		cgroup_enter_frozen();
2524 		schedule();
2525 		return true;
2526 	} else {
2527 		/*
2528 		 * While ptraced, group stop is handled by STOP trap.
2529 		 * Schedule it and let the caller deal with it.
2530 		 */
2531 		task_set_jobctl_pending(current, JOBCTL_TRAP_STOP);
2532 		return false;
2533 	}
2534 }
2535 
2536 /**
2537  * do_jobctl_trap - take care of ptrace jobctl traps
2538  *
2539  * When PT_SEIZED, it's used for both group stop and explicit
2540  * SEIZE/INTERRUPT traps.  Both generate PTRACE_EVENT_STOP trap with
2541  * accompanying siginfo.  If stopped, lower eight bits of exit_code contain
2542  * the stop signal; otherwise, %SIGTRAP.
2543  *
2544  * When !PT_SEIZED, it's used only for group stop trap with stop signal
2545  * number as exit_code and no siginfo.
2546  *
2547  * CONTEXT:
2548  * Must be called with @current->sighand->siglock held, which may be
2549  * released and re-acquired before returning with intervening sleep.
2550  */
2551 static void do_jobctl_trap(void)
2552 {
2553 	struct signal_struct *signal = current->signal;
2554 	int signr = current->jobctl & JOBCTL_STOP_SIGMASK;
2555 
2556 	if (current->ptrace & PT_SEIZED) {
2557 		if (!signal->group_stop_count &&
2558 		    !(signal->flags & SIGNAL_STOP_STOPPED))
2559 			signr = SIGTRAP;
2560 		WARN_ON_ONCE(!signr);
2561 		ptrace_do_notify(signr, signr | (PTRACE_EVENT_STOP << 8),
2562 				 CLD_STOPPED, 0);
2563 	} else {
2564 		WARN_ON_ONCE(!signr);
2565 		ptrace_stop(signr, CLD_STOPPED, 0, NULL);
2566 	}
2567 }
2568 
2569 /**
2570  * do_freezer_trap - handle the freezer jobctl trap
2571  *
2572  * Puts the task into frozen state, if only the task is not about to quit.
2573  * In this case it drops JOBCTL_TRAP_FREEZE.
2574  *
2575  * CONTEXT:
2576  * Must be called with @current->sighand->siglock held,
2577  * which is always released before returning.
2578  */
2579 static void do_freezer_trap(void)
2580 	__releases(&current->sighand->siglock)
2581 {
2582 	/*
2583 	 * If there are other trap bits pending except JOBCTL_TRAP_FREEZE,
2584 	 * let's make another loop to give it a chance to be handled.
2585 	 * In any case, we'll return back.
2586 	 */
2587 	if ((current->jobctl & (JOBCTL_PENDING_MASK | JOBCTL_TRAP_FREEZE)) !=
2588 	     JOBCTL_TRAP_FREEZE) {
2589 		spin_unlock_irq(&current->sighand->siglock);
2590 		return;
2591 	}
2592 
2593 	/*
2594 	 * Now we're sure that there is no pending fatal signal and no
2595 	 * pending traps. Clear TIF_SIGPENDING to not get out of schedule()
2596 	 * immediately (if there is a non-fatal signal pending), and
2597 	 * put the task into sleep.
2598 	 */
2599 	__set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
2600 	clear_thread_flag(TIF_SIGPENDING);
2601 	spin_unlock_irq(&current->sighand->siglock);
2602 	cgroup_enter_frozen();
2603 	schedule();
2604 
2605 	/*
2606 	 * We could've been woken by task_work, run it to clear
2607 	 * TIF_NOTIFY_SIGNAL. The caller will retry if necessary.
2608 	 */
2609 	clear_notify_signal();
2610 	if (unlikely(task_work_pending(current)))
2611 		task_work_run();
2612 }
2613 
2614 static int ptrace_signal(int signr, kernel_siginfo_t *info, enum pid_type type)
2615 {
2616 	/*
2617 	 * We do not check sig_kernel_stop(signr) but set this marker
2618 	 * unconditionally because we do not know whether debugger will
2619 	 * change signr. This flag has no meaning unless we are going
2620 	 * to stop after return from ptrace_stop(). In this case it will
2621 	 * be checked in do_signal_stop(), we should only stop if it was
2622 	 * not cleared by SIGCONT while we were sleeping. See also the
2623 	 * comment in dequeue_signal().
2624 	 */
2625 	current->jobctl |= JOBCTL_STOP_DEQUEUED;
2626 	signr = ptrace_stop(signr, CLD_TRAPPED, 0, info);
2627 
2628 	/* We're back.  Did the debugger cancel the sig?  */
2629 	if (signr == 0)
2630 		return signr;
2631 
2632 	/*
2633 	 * Update the siginfo structure if the signal has
2634 	 * changed.  If the debugger wanted something
2635 	 * specific in the siginfo structure then it should
2636 	 * have updated *info via PTRACE_SETSIGINFO.
2637 	 */
2638 	if (signr != info->si_signo) {
2639 		clear_siginfo(info);
2640 		info->si_signo = signr;
2641 		info->si_errno = 0;
2642 		info->si_code = SI_USER;
2643 		rcu_read_lock();
2644 		info->si_pid = task_pid_vnr(current->parent);
2645 		info->si_uid = from_kuid_munged(current_user_ns(),
2646 						task_uid(current->parent));
2647 		rcu_read_unlock();
2648 	}
2649 
2650 	/* If the (new) signal is now blocked, requeue it.  */
2651 	if (sigismember(&current->blocked, signr) ||
2652 	    fatal_signal_pending(current)) {
2653 		send_signal_locked(signr, info, current, type);
2654 		signr = 0;
2655 	}
2656 
2657 	return signr;
2658 }
2659 
2660 static void hide_si_addr_tag_bits(struct ksignal *ksig)
2661 {
2662 	switch (siginfo_layout(ksig->sig, ksig->info.si_code)) {
2663 	case SIL_FAULT:
2664 	case SIL_FAULT_TRAPNO:
2665 	case SIL_FAULT_MCEERR:
2666 	case SIL_FAULT_BNDERR:
2667 	case SIL_FAULT_PKUERR:
2668 	case SIL_FAULT_PERF_EVENT:
2669 		ksig->info.si_addr = arch_untagged_si_addr(
2670 			ksig->info.si_addr, ksig->sig, ksig->info.si_code);
2671 		break;
2672 	case SIL_KILL:
2673 	case SIL_TIMER:
2674 	case SIL_POLL:
2675 	case SIL_CHLD:
2676 	case SIL_RT:
2677 	case SIL_SYS:
2678 		break;
2679 	}
2680 }
2681 
2682 bool get_signal(struct ksignal *ksig)
2683 {
2684 	struct sighand_struct *sighand = current->sighand;
2685 	struct signal_struct *signal = current->signal;
2686 	int signr;
2687 
2688 	clear_notify_signal();
2689 	if (unlikely(task_work_pending(current)))
2690 		task_work_run();
2691 
2692 	if (!task_sigpending(current))
2693 		return false;
2694 
2695 	if (unlikely(uprobe_deny_signal()))
2696 		return false;
2697 
2698 	/*
2699 	 * Do this once, we can't return to user-mode if freezing() == T.
2700 	 * do_signal_stop() and ptrace_stop() do freezable_schedule() and
2701 	 * thus do not need another check after return.
2702 	 */
2703 	try_to_freeze();
2704 
2705 relock:
2706 	spin_lock_irq(&sighand->siglock);
2707 
2708 	/*
2709 	 * Every stopped thread goes here after wakeup. Check to see if
2710 	 * we should notify the parent, prepare_signal(SIGCONT) encodes
2711 	 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
2712 	 */
2713 	if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
2714 		int why;
2715 
2716 		if (signal->flags & SIGNAL_CLD_CONTINUED)
2717 			why = CLD_CONTINUED;
2718 		else
2719 			why = CLD_STOPPED;
2720 
2721 		signal->flags &= ~SIGNAL_CLD_MASK;
2722 
2723 		spin_unlock_irq(&sighand->siglock);
2724 
2725 		/*
2726 		 * Notify the parent that we're continuing.  This event is
2727 		 * always per-process and doesn't make whole lot of sense
2728 		 * for ptracers, who shouldn't consume the state via
2729 		 * wait(2) either, but, for backward compatibility, notify
2730 		 * the ptracer of the group leader too unless it's gonna be
2731 		 * a duplicate.
2732 		 */
2733 		read_lock(&tasklist_lock);
2734 		do_notify_parent_cldstop(current, false, why);
2735 
2736 		if (ptrace_reparented(current->group_leader))
2737 			do_notify_parent_cldstop(current->group_leader,
2738 						true, why);
2739 		read_unlock(&tasklist_lock);
2740 
2741 		goto relock;
2742 	}
2743 
2744 	for (;;) {
2745 		struct k_sigaction *ka;
2746 		enum pid_type type;
2747 
2748 		/* Has this task already been marked for death? */
2749 		if ((signal->flags & SIGNAL_GROUP_EXIT) ||
2750 		     signal->group_exec_task) {
2751 			signr = SIGKILL;
2752 			sigdelset(&current->pending.signal, SIGKILL);
2753 			trace_signal_deliver(SIGKILL, SEND_SIG_NOINFO,
2754 					     &sighand->action[SIGKILL-1]);
2755 			recalc_sigpending();
2756 			/*
2757 			 * implies do_group_exit() or return to PF_USER_WORKER,
2758 			 * no need to initialize ksig->info/etc.
2759 			 */
2760 			goto fatal;
2761 		}
2762 
2763 		if (unlikely(current->jobctl & JOBCTL_STOP_PENDING) &&
2764 		    do_signal_stop(0))
2765 			goto relock;
2766 
2767 		if (unlikely(current->jobctl &
2768 			     (JOBCTL_TRAP_MASK | JOBCTL_TRAP_FREEZE))) {
2769 			if (current->jobctl & JOBCTL_TRAP_MASK) {
2770 				do_jobctl_trap();
2771 				spin_unlock_irq(&sighand->siglock);
2772 			} else if (current->jobctl & JOBCTL_TRAP_FREEZE)
2773 				do_freezer_trap();
2774 
2775 			goto relock;
2776 		}
2777 
2778 		/*
2779 		 * If the task is leaving the frozen state, let's update
2780 		 * cgroup counters and reset the frozen bit.
2781 		 */
2782 		if (unlikely(cgroup_task_frozen(current))) {
2783 			spin_unlock_irq(&sighand->siglock);
2784 			cgroup_leave_frozen(false);
2785 			goto relock;
2786 		}
2787 
2788 		/*
2789 		 * Signals generated by the execution of an instruction
2790 		 * need to be delivered before any other pending signals
2791 		 * so that the instruction pointer in the signal stack
2792 		 * frame points to the faulting instruction.
2793 		 */
2794 		type = PIDTYPE_PID;
2795 		signr = dequeue_synchronous_signal(&ksig->info);
2796 		if (!signr)
2797 			signr = dequeue_signal(&current->blocked, &ksig->info, &type);
2798 
2799 		if (!signr)
2800 			break; /* will return 0 */
2801 
2802 		if (unlikely(current->ptrace) && (signr != SIGKILL) &&
2803 		    !(sighand->action[signr -1].sa.sa_flags & SA_IMMUTABLE)) {
2804 			signr = ptrace_signal(signr, &ksig->info, type);
2805 			if (!signr)
2806 				continue;
2807 		}
2808 
2809 		ka = &sighand->action[signr-1];
2810 
2811 		/* Trace actually delivered signals. */
2812 		trace_signal_deliver(signr, &ksig->info, ka);
2813 
2814 		if (ka->sa.sa_handler == SIG_IGN) /* Do nothing.  */
2815 			continue;
2816 		if (ka->sa.sa_handler != SIG_DFL) {
2817 			/* Run the handler.  */
2818 			ksig->ka = *ka;
2819 
2820 			if (ka->sa.sa_flags & SA_ONESHOT)
2821 				ka->sa.sa_handler = SIG_DFL;
2822 
2823 			break; /* will return non-zero "signr" value */
2824 		}
2825 
2826 		/*
2827 		 * Now we are doing the default action for this signal.
2828 		 */
2829 		if (sig_kernel_ignore(signr)) /* Default is nothing. */
2830 			continue;
2831 
2832 		/*
2833 		 * Global init gets no signals it doesn't want.
2834 		 * Container-init gets no signals it doesn't want from same
2835 		 * container.
2836 		 *
2837 		 * Note that if global/container-init sees a sig_kernel_only()
2838 		 * signal here, the signal must have been generated internally
2839 		 * or must have come from an ancestor namespace. In either
2840 		 * case, the signal cannot be dropped.
2841 		 */
2842 		if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
2843 				!sig_kernel_only(signr))
2844 			continue;
2845 
2846 		if (sig_kernel_stop(signr)) {
2847 			/*
2848 			 * The default action is to stop all threads in
2849 			 * the thread group.  The job control signals
2850 			 * do nothing in an orphaned pgrp, but SIGSTOP
2851 			 * always works.  Note that siglock needs to be
2852 			 * dropped during the call to is_orphaned_pgrp()
2853 			 * because of lock ordering with tasklist_lock.
2854 			 * This allows an intervening SIGCONT to be posted.
2855 			 * We need to check for that and bail out if necessary.
2856 			 */
2857 			if (signr != SIGSTOP) {
2858 				spin_unlock_irq(&sighand->siglock);
2859 
2860 				/* signals can be posted during this window */
2861 
2862 				if (is_current_pgrp_orphaned())
2863 					goto relock;
2864 
2865 				spin_lock_irq(&sighand->siglock);
2866 			}
2867 
2868 			if (likely(do_signal_stop(signr))) {
2869 				/* It released the siglock.  */
2870 				goto relock;
2871 			}
2872 
2873 			/*
2874 			 * We didn't actually stop, due to a race
2875 			 * with SIGCONT or something like that.
2876 			 */
2877 			continue;
2878 		}
2879 
2880 	fatal:
2881 		spin_unlock_irq(&sighand->siglock);
2882 		if (unlikely(cgroup_task_frozen(current)))
2883 			cgroup_leave_frozen(true);
2884 
2885 		/*
2886 		 * Anything else is fatal, maybe with a core dump.
2887 		 */
2888 		current->flags |= PF_SIGNALED;
2889 
2890 		if (sig_kernel_coredump(signr)) {
2891 			if (print_fatal_signals)
2892 				print_fatal_signal(signr);
2893 			proc_coredump_connector(current);
2894 			/*
2895 			 * If it was able to dump core, this kills all
2896 			 * other threads in the group and synchronizes with
2897 			 * their demise.  If we lost the race with another
2898 			 * thread getting here, it set group_exit_code
2899 			 * first and our do_group_exit call below will use
2900 			 * that value and ignore the one we pass it.
2901 			 */
2902 			do_coredump(&ksig->info);
2903 		}
2904 
2905 		/*
2906 		 * PF_USER_WORKER threads will catch and exit on fatal signals
2907 		 * themselves. They have cleanup that must be performed, so we
2908 		 * cannot call do_exit() on their behalf. Note that ksig won't
2909 		 * be properly initialized, PF_USER_WORKER's shouldn't use it.
2910 		 */
2911 		if (current->flags & PF_USER_WORKER)
2912 			goto out;
2913 
2914 		/*
2915 		 * Death signals, no core dump.
2916 		 */
2917 		do_group_exit(signr);
2918 		/* NOTREACHED */
2919 	}
2920 	spin_unlock_irq(&sighand->siglock);
2921 
2922 	ksig->sig = signr;
2923 
2924 	if (signr && !(ksig->ka.sa.sa_flags & SA_EXPOSE_TAGBITS))
2925 		hide_si_addr_tag_bits(ksig);
2926 out:
2927 	return signr > 0;
2928 }
2929 
2930 /**
2931  * signal_delivered - called after signal delivery to update blocked signals
2932  * @ksig:		kernel signal struct
2933  * @stepping:		nonzero if debugger single-step or block-step in use
2934  *
2935  * This function should be called when a signal has successfully been
2936  * delivered. It updates the blocked signals accordingly (@ksig->ka.sa.sa_mask
2937  * is always blocked), and the signal itself is blocked unless %SA_NODEFER
2938  * is set in @ksig->ka.sa.sa_flags.  Tracing is notified.
2939  */
2940 static void signal_delivered(struct ksignal *ksig, int stepping)
2941 {
2942 	sigset_t blocked;
2943 
2944 	/* A signal was successfully delivered, and the
2945 	   saved sigmask was stored on the signal frame,
2946 	   and will be restored by sigreturn.  So we can
2947 	   simply clear the restore sigmask flag.  */
2948 	clear_restore_sigmask();
2949 
2950 	sigorsets(&blocked, &current->blocked, &ksig->ka.sa.sa_mask);
2951 	if (!(ksig->ka.sa.sa_flags & SA_NODEFER))
2952 		sigaddset(&blocked, ksig->sig);
2953 	set_current_blocked(&blocked);
2954 	if (current->sas_ss_flags & SS_AUTODISARM)
2955 		sas_ss_reset(current);
2956 	if (stepping)
2957 		ptrace_notify(SIGTRAP, 0);
2958 }
2959 
2960 void signal_setup_done(int failed, struct ksignal *ksig, int stepping)
2961 {
2962 	if (failed)
2963 		force_sigsegv(ksig->sig);
2964 	else
2965 		signal_delivered(ksig, stepping);
2966 }
2967 
2968 /*
2969  * It could be that complete_signal() picked us to notify about the
2970  * group-wide signal. Other threads should be notified now to take
2971  * the shared signals in @which since we will not.
2972  */
2973 static void retarget_shared_pending(struct task_struct *tsk, sigset_t *which)
2974 {
2975 	sigset_t retarget;
2976 	struct task_struct *t;
2977 
2978 	sigandsets(&retarget, &tsk->signal->shared_pending.signal, which);
2979 	if (sigisemptyset(&retarget))
2980 		return;
2981 
2982 	for_other_threads(tsk, t) {
2983 		if (t->flags & PF_EXITING)
2984 			continue;
2985 
2986 		if (!has_pending_signals(&retarget, &t->blocked))
2987 			continue;
2988 		/* Remove the signals this thread can handle. */
2989 		sigandsets(&retarget, &retarget, &t->blocked);
2990 
2991 		if (!task_sigpending(t))
2992 			signal_wake_up(t, 0);
2993 
2994 		if (sigisemptyset(&retarget))
2995 			break;
2996 	}
2997 }
2998 
2999 void exit_signals(struct task_struct *tsk)
3000 {
3001 	int group_stop = 0;
3002 	sigset_t unblocked;
3003 
3004 	/*
3005 	 * @tsk is about to have PF_EXITING set - lock out users which
3006 	 * expect stable threadgroup.
3007 	 */
3008 	cgroup_threadgroup_change_begin(tsk);
3009 
3010 	if (thread_group_empty(tsk) || (tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
3011 		sched_mm_cid_exit_signals(tsk);
3012 		tsk->flags |= PF_EXITING;
3013 		cgroup_threadgroup_change_end(tsk);
3014 		return;
3015 	}
3016 
3017 	spin_lock_irq(&tsk->sighand->siglock);
3018 	/*
3019 	 * From now this task is not visible for group-wide signals,
3020 	 * see wants_signal(), do_signal_stop().
3021 	 */
3022 	sched_mm_cid_exit_signals(tsk);
3023 	tsk->flags |= PF_EXITING;
3024 
3025 	cgroup_threadgroup_change_end(tsk);
3026 
3027 	if (!task_sigpending(tsk))
3028 		goto out;
3029 
3030 	unblocked = tsk->blocked;
3031 	signotset(&unblocked);
3032 	retarget_shared_pending(tsk, &unblocked);
3033 
3034 	if (unlikely(tsk->jobctl & JOBCTL_STOP_PENDING) &&
3035 	    task_participate_group_stop(tsk))
3036 		group_stop = CLD_STOPPED;
3037 out:
3038 	spin_unlock_irq(&tsk->sighand->siglock);
3039 
3040 	/*
3041 	 * If group stop has completed, deliver the notification.  This
3042 	 * should always go to the real parent of the group leader.
3043 	 */
3044 	if (unlikely(group_stop)) {
3045 		read_lock(&tasklist_lock);
3046 		do_notify_parent_cldstop(tsk, false, group_stop);
3047 		read_unlock(&tasklist_lock);
3048 	}
3049 }
3050 
3051 /*
3052  * System call entry points.
3053  */
3054 
3055 /**
3056  *  sys_restart_syscall - restart a system call
3057  */
3058 SYSCALL_DEFINE0(restart_syscall)
3059 {
3060 	struct restart_block *restart = &current->restart_block;
3061 	return restart->fn(restart);
3062 }
3063 
3064 long do_no_restart_syscall(struct restart_block *param)
3065 {
3066 	return -EINTR;
3067 }
3068 
3069 static void __set_task_blocked(struct task_struct *tsk, const sigset_t *newset)
3070 {
3071 	if (task_sigpending(tsk) && !thread_group_empty(tsk)) {
3072 		sigset_t newblocked;
3073 		/* A set of now blocked but previously unblocked signals. */
3074 		sigandnsets(&newblocked, newset, &current->blocked);
3075 		retarget_shared_pending(tsk, &newblocked);
3076 	}
3077 	tsk->blocked = *newset;
3078 	recalc_sigpending();
3079 }
3080 
3081 /**
3082  * set_current_blocked - change current->blocked mask
3083  * @newset: new mask
3084  *
3085  * It is wrong to change ->blocked directly, this helper should be used
3086  * to ensure the process can't miss a shared signal we are going to block.
3087  */
3088 void set_current_blocked(sigset_t *newset)
3089 {
3090 	sigdelsetmask(newset, sigmask(SIGKILL) | sigmask(SIGSTOP));
3091 	__set_current_blocked(newset);
3092 }
3093 
3094 void __set_current_blocked(const sigset_t *newset)
3095 {
3096 	struct task_struct *tsk = current;
3097 
3098 	/*
3099 	 * In case the signal mask hasn't changed, there is nothing we need
3100 	 * to do. The current->blocked shouldn't be modified by other task.
3101 	 */
3102 	if (sigequalsets(&tsk->blocked, newset))
3103 		return;
3104 
3105 	spin_lock_irq(&tsk->sighand->siglock);
3106 	__set_task_blocked(tsk, newset);
3107 	spin_unlock_irq(&tsk->sighand->siglock);
3108 }
3109 
3110 /*
3111  * This is also useful for kernel threads that want to temporarily
3112  * (or permanently) block certain signals.
3113  *
3114  * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
3115  * interface happily blocks "unblockable" signals like SIGKILL
3116  * and friends.
3117  */
3118 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
3119 {
3120 	struct task_struct *tsk = current;
3121 	sigset_t newset;
3122 
3123 	/* Lockless, only current can change ->blocked, never from irq */
3124 	if (oldset)
3125 		*oldset = tsk->blocked;
3126 
3127 	switch (how) {
3128 	case SIG_BLOCK:
3129 		sigorsets(&newset, &tsk->blocked, set);
3130 		break;
3131 	case SIG_UNBLOCK:
3132 		sigandnsets(&newset, &tsk->blocked, set);
3133 		break;
3134 	case SIG_SETMASK:
3135 		newset = *set;
3136 		break;
3137 	default:
3138 		return -EINVAL;
3139 	}
3140 
3141 	__set_current_blocked(&newset);
3142 	return 0;
3143 }
3144 EXPORT_SYMBOL(sigprocmask);
3145 
3146 /*
3147  * The api helps set app-provided sigmasks.
3148  *
3149  * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
3150  * epoll_pwait where a new sigmask is passed from userland for the syscalls.
3151  *
3152  * Note that it does set_restore_sigmask() in advance, so it must be always
3153  * paired with restore_saved_sigmask_unless() before return from syscall.
3154  */
3155 int set_user_sigmask(const sigset_t __user *umask, size_t sigsetsize)
3156 {
3157 	sigset_t kmask;
3158 
3159 	if (!umask)
3160 		return 0;
3161 	if (sigsetsize != sizeof(sigset_t))
3162 		return -EINVAL;
3163 	if (copy_from_user(&kmask, umask, sizeof(sigset_t)))
3164 		return -EFAULT;
3165 
3166 	set_restore_sigmask();
3167 	current->saved_sigmask = current->blocked;
3168 	set_current_blocked(&kmask);
3169 
3170 	return 0;
3171 }
3172 
3173 #ifdef CONFIG_COMPAT
3174 int set_compat_user_sigmask(const compat_sigset_t __user *umask,
3175 			    size_t sigsetsize)
3176 {
3177 	sigset_t kmask;
3178 
3179 	if (!umask)
3180 		return 0;
3181 	if (sigsetsize != sizeof(compat_sigset_t))
3182 		return -EINVAL;
3183 	if (get_compat_sigset(&kmask, umask))
3184 		return -EFAULT;
3185 
3186 	set_restore_sigmask();
3187 	current->saved_sigmask = current->blocked;
3188 	set_current_blocked(&kmask);
3189 
3190 	return 0;
3191 }
3192 #endif
3193 
3194 /**
3195  *  sys_rt_sigprocmask - change the list of currently blocked signals
3196  *  @how: whether to add, remove, or set signals
3197  *  @nset: stores pending signals
3198  *  @oset: previous value of signal mask if non-null
3199  *  @sigsetsize: size of sigset_t type
3200  */
3201 SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, nset,
3202 		sigset_t __user *, oset, size_t, sigsetsize)
3203 {
3204 	sigset_t old_set, new_set;
3205 	int error;
3206 
3207 	/* XXX: Don't preclude handling different sized sigset_t's.  */
3208 	if (sigsetsize != sizeof(sigset_t))
3209 		return -EINVAL;
3210 
3211 	old_set = current->blocked;
3212 
3213 	if (nset) {
3214 		if (copy_from_user(&new_set, nset, sizeof(sigset_t)))
3215 			return -EFAULT;
3216 		sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3217 
3218 		error = sigprocmask(how, &new_set, NULL);
3219 		if (error)
3220 			return error;
3221 	}
3222 
3223 	if (oset) {
3224 		if (copy_to_user(oset, &old_set, sizeof(sigset_t)))
3225 			return -EFAULT;
3226 	}
3227 
3228 	return 0;
3229 }
3230 
3231 #ifdef CONFIG_COMPAT
3232 COMPAT_SYSCALL_DEFINE4(rt_sigprocmask, int, how, compat_sigset_t __user *, nset,
3233 		compat_sigset_t __user *, oset, compat_size_t, sigsetsize)
3234 {
3235 	sigset_t old_set = current->blocked;
3236 
3237 	/* XXX: Don't preclude handling different sized sigset_t's.  */
3238 	if (sigsetsize != sizeof(sigset_t))
3239 		return -EINVAL;
3240 
3241 	if (nset) {
3242 		sigset_t new_set;
3243 		int error;
3244 		if (get_compat_sigset(&new_set, nset))
3245 			return -EFAULT;
3246 		sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
3247 
3248 		error = sigprocmask(how, &new_set, NULL);
3249 		if (error)
3250 			return error;
3251 	}
3252 	return oset ? put_compat_sigset(oset, &old_set, sizeof(*oset)) : 0;
3253 }
3254 #endif
3255 
3256 static void do_sigpending(sigset_t *set)
3257 {
3258 	spin_lock_irq(&current->sighand->siglock);
3259 	sigorsets(set, &current->pending.signal,
3260 		  &current->signal->shared_pending.signal);
3261 	spin_unlock_irq(&current->sighand->siglock);
3262 
3263 	/* Outside the lock because only this thread touches it.  */
3264 	sigandsets(set, &current->blocked, set);
3265 }
3266 
3267 /**
3268  *  sys_rt_sigpending - examine a pending signal that has been raised
3269  *			while blocked
3270  *  @uset: stores pending signals
3271  *  @sigsetsize: size of sigset_t type or larger
3272  */
3273 SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, uset, size_t, sigsetsize)
3274 {
3275 	sigset_t set;
3276 
3277 	if (sigsetsize > sizeof(*uset))
3278 		return -EINVAL;
3279 
3280 	do_sigpending(&set);
3281 
3282 	if (copy_to_user(uset, &set, sigsetsize))
3283 		return -EFAULT;
3284 
3285 	return 0;
3286 }
3287 
3288 #ifdef CONFIG_COMPAT
3289 COMPAT_SYSCALL_DEFINE2(rt_sigpending, compat_sigset_t __user *, uset,
3290 		compat_size_t, sigsetsize)
3291 {
3292 	sigset_t set;
3293 
3294 	if (sigsetsize > sizeof(*uset))
3295 		return -EINVAL;
3296 
3297 	do_sigpending(&set);
3298 
3299 	return put_compat_sigset(uset, &set, sigsetsize);
3300 }
3301 #endif
3302 
3303 static const struct {
3304 	unsigned char limit, layout;
3305 } sig_sicodes[] = {
3306 	[SIGILL]  = { NSIGILL,  SIL_FAULT },
3307 	[SIGFPE]  = { NSIGFPE,  SIL_FAULT },
3308 	[SIGSEGV] = { NSIGSEGV, SIL_FAULT },
3309 	[SIGBUS]  = { NSIGBUS,  SIL_FAULT },
3310 	[SIGTRAP] = { NSIGTRAP, SIL_FAULT },
3311 #if defined(SIGEMT)
3312 	[SIGEMT]  = { NSIGEMT,  SIL_FAULT },
3313 #endif
3314 	[SIGCHLD] = { NSIGCHLD, SIL_CHLD },
3315 	[SIGPOLL] = { NSIGPOLL, SIL_POLL },
3316 	[SIGSYS]  = { NSIGSYS,  SIL_SYS },
3317 };
3318 
3319 static bool known_siginfo_layout(unsigned sig, int si_code)
3320 {
3321 	if (si_code == SI_KERNEL)
3322 		return true;
3323 	else if ((si_code > SI_USER)) {
3324 		if (sig_specific_sicodes(sig)) {
3325 			if (si_code <= sig_sicodes[sig].limit)
3326 				return true;
3327 		}
3328 		else if (si_code <= NSIGPOLL)
3329 			return true;
3330 	}
3331 	else if (si_code >= SI_DETHREAD)
3332 		return true;
3333 	else if (si_code == SI_ASYNCNL)
3334 		return true;
3335 	return false;
3336 }
3337 
3338 enum siginfo_layout siginfo_layout(unsigned sig, int si_code)
3339 {
3340 	enum siginfo_layout layout = SIL_KILL;
3341 	if ((si_code > SI_USER) && (si_code < SI_KERNEL)) {
3342 		if ((sig < ARRAY_SIZE(sig_sicodes)) &&
3343 		    (si_code <= sig_sicodes[sig].limit)) {
3344 			layout = sig_sicodes[sig].layout;
3345 			/* Handle the exceptions */
3346 			if ((sig == SIGBUS) &&
3347 			    (si_code >= BUS_MCEERR_AR) && (si_code <= BUS_MCEERR_AO))
3348 				layout = SIL_FAULT_MCEERR;
3349 			else if ((sig == SIGSEGV) && (si_code == SEGV_BNDERR))
3350 				layout = SIL_FAULT_BNDERR;
3351 #ifdef SEGV_PKUERR
3352 			else if ((sig == SIGSEGV) && (si_code == SEGV_PKUERR))
3353 				layout = SIL_FAULT_PKUERR;
3354 #endif
3355 			else if ((sig == SIGTRAP) && (si_code == TRAP_PERF))
3356 				layout = SIL_FAULT_PERF_EVENT;
3357 			else if (IS_ENABLED(CONFIG_SPARC) &&
3358 				 (sig == SIGILL) && (si_code == ILL_ILLTRP))
3359 				layout = SIL_FAULT_TRAPNO;
3360 			else if (IS_ENABLED(CONFIG_ALPHA) &&
3361 				 ((sig == SIGFPE) ||
3362 				  ((sig == SIGTRAP) && (si_code == TRAP_UNK))))
3363 				layout = SIL_FAULT_TRAPNO;
3364 		}
3365 		else if (si_code <= NSIGPOLL)
3366 			layout = SIL_POLL;
3367 	} else {
3368 		if (si_code == SI_TIMER)
3369 			layout = SIL_TIMER;
3370 		else if (si_code == SI_SIGIO)
3371 			layout = SIL_POLL;
3372 		else if (si_code < 0)
3373 			layout = SIL_RT;
3374 	}
3375 	return layout;
3376 }
3377 
3378 static inline char __user *si_expansion(const siginfo_t __user *info)
3379 {
3380 	return ((char __user *)info) + sizeof(struct kernel_siginfo);
3381 }
3382 
3383 int copy_siginfo_to_user(siginfo_t __user *to, const kernel_siginfo_t *from)
3384 {
3385 	char __user *expansion = si_expansion(to);
3386 	if (copy_to_user(to, from , sizeof(struct kernel_siginfo)))
3387 		return -EFAULT;
3388 	if (clear_user(expansion, SI_EXPANSION_SIZE))
3389 		return -EFAULT;
3390 	return 0;
3391 }
3392 
3393 static int post_copy_siginfo_from_user(kernel_siginfo_t *info,
3394 				       const siginfo_t __user *from)
3395 {
3396 	if (unlikely(!known_siginfo_layout(info->si_signo, info->si_code))) {
3397 		char __user *expansion = si_expansion(from);
3398 		char buf[SI_EXPANSION_SIZE];
3399 		int i;
3400 		/*
3401 		 * An unknown si_code might need more than
3402 		 * sizeof(struct kernel_siginfo) bytes.  Verify all of the
3403 		 * extra bytes are 0.  This guarantees copy_siginfo_to_user
3404 		 * will return this data to userspace exactly.
3405 		 */
3406 		if (copy_from_user(&buf, expansion, SI_EXPANSION_SIZE))
3407 			return -EFAULT;
3408 		for (i = 0; i < SI_EXPANSION_SIZE; i++) {
3409 			if (buf[i] != 0)
3410 				return -E2BIG;
3411 		}
3412 	}
3413 	return 0;
3414 }
3415 
3416 static int __copy_siginfo_from_user(int signo, kernel_siginfo_t *to,
3417 				    const siginfo_t __user *from)
3418 {
3419 	if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3420 		return -EFAULT;
3421 	to->si_signo = signo;
3422 	return post_copy_siginfo_from_user(to, from);
3423 }
3424 
3425 int copy_siginfo_from_user(kernel_siginfo_t *to, const siginfo_t __user *from)
3426 {
3427 	if (copy_from_user(to, from, sizeof(struct kernel_siginfo)))
3428 		return -EFAULT;
3429 	return post_copy_siginfo_from_user(to, from);
3430 }
3431 
3432 #ifdef CONFIG_COMPAT
3433 /**
3434  * copy_siginfo_to_external32 - copy a kernel siginfo into a compat user siginfo
3435  * @to: compat siginfo destination
3436  * @from: kernel siginfo source
3437  *
3438  * Note: This function does not work properly for the SIGCHLD on x32, but
3439  * fortunately it doesn't have to.  The only valid callers for this function are
3440  * copy_siginfo_to_user32, which is overriden for x32 and the coredump code.
3441  * The latter does not care because SIGCHLD will never cause a coredump.
3442  */
3443 void copy_siginfo_to_external32(struct compat_siginfo *to,
3444 		const struct kernel_siginfo *from)
3445 {
3446 	memset(to, 0, sizeof(*to));
3447 
3448 	to->si_signo = from->si_signo;
3449 	to->si_errno = from->si_errno;
3450 	to->si_code  = from->si_code;
3451 	switch(siginfo_layout(from->si_signo, from->si_code)) {
3452 	case SIL_KILL:
3453 		to->si_pid = from->si_pid;
3454 		to->si_uid = from->si_uid;
3455 		break;
3456 	case SIL_TIMER:
3457 		to->si_tid     = from->si_tid;
3458 		to->si_overrun = from->si_overrun;
3459 		to->si_int     = from->si_int;
3460 		break;
3461 	case SIL_POLL:
3462 		to->si_band = from->si_band;
3463 		to->si_fd   = from->si_fd;
3464 		break;
3465 	case SIL_FAULT:
3466 		to->si_addr = ptr_to_compat(from->si_addr);
3467 		break;
3468 	case SIL_FAULT_TRAPNO:
3469 		to->si_addr = ptr_to_compat(from->si_addr);
3470 		to->si_trapno = from->si_trapno;
3471 		break;
3472 	case SIL_FAULT_MCEERR:
3473 		to->si_addr = ptr_to_compat(from->si_addr);
3474 		to->si_addr_lsb = from->si_addr_lsb;
3475 		break;
3476 	case SIL_FAULT_BNDERR:
3477 		to->si_addr = ptr_to_compat(from->si_addr);
3478 		to->si_lower = ptr_to_compat(from->si_lower);
3479 		to->si_upper = ptr_to_compat(from->si_upper);
3480 		break;
3481 	case SIL_FAULT_PKUERR:
3482 		to->si_addr = ptr_to_compat(from->si_addr);
3483 		to->si_pkey = from->si_pkey;
3484 		break;
3485 	case SIL_FAULT_PERF_EVENT:
3486 		to->si_addr = ptr_to_compat(from->si_addr);
3487 		to->si_perf_data = from->si_perf_data;
3488 		to->si_perf_type = from->si_perf_type;
3489 		to->si_perf_flags = from->si_perf_flags;
3490 		break;
3491 	case SIL_CHLD:
3492 		to->si_pid = from->si_pid;
3493 		to->si_uid = from->si_uid;
3494 		to->si_status = from->si_status;
3495 		to->si_utime = from->si_utime;
3496 		to->si_stime = from->si_stime;
3497 		break;
3498 	case SIL_RT:
3499 		to->si_pid = from->si_pid;
3500 		to->si_uid = from->si_uid;
3501 		to->si_int = from->si_int;
3502 		break;
3503 	case SIL_SYS:
3504 		to->si_call_addr = ptr_to_compat(from->si_call_addr);
3505 		to->si_syscall   = from->si_syscall;
3506 		to->si_arch      = from->si_arch;
3507 		break;
3508 	}
3509 }
3510 
3511 int __copy_siginfo_to_user32(struct compat_siginfo __user *to,
3512 			   const struct kernel_siginfo *from)
3513 {
3514 	struct compat_siginfo new;
3515 
3516 	copy_siginfo_to_external32(&new, from);
3517 	if (copy_to_user(to, &new, sizeof(struct compat_siginfo)))
3518 		return -EFAULT;
3519 	return 0;
3520 }
3521 
3522 static int post_copy_siginfo_from_user32(kernel_siginfo_t *to,
3523 					 const struct compat_siginfo *from)
3524 {
3525 	clear_siginfo(to);
3526 	to->si_signo = from->si_signo;
3527 	to->si_errno = from->si_errno;
3528 	to->si_code  = from->si_code;
3529 	switch(siginfo_layout(from->si_signo, from->si_code)) {
3530 	case SIL_KILL:
3531 		to->si_pid = from->si_pid;
3532 		to->si_uid = from->si_uid;
3533 		break;
3534 	case SIL_TIMER:
3535 		to->si_tid     = from->si_tid;
3536 		to->si_overrun = from->si_overrun;
3537 		to->si_int     = from->si_int;
3538 		break;
3539 	case SIL_POLL:
3540 		to->si_band = from->si_band;
3541 		to->si_fd   = from->si_fd;
3542 		break;
3543 	case SIL_FAULT:
3544 		to->si_addr = compat_ptr(from->si_addr);
3545 		break;
3546 	case SIL_FAULT_TRAPNO:
3547 		to->si_addr = compat_ptr(from->si_addr);
3548 		to->si_trapno = from->si_trapno;
3549 		break;
3550 	case SIL_FAULT_MCEERR:
3551 		to->si_addr = compat_ptr(from->si_addr);
3552 		to->si_addr_lsb = from->si_addr_lsb;
3553 		break;
3554 	case SIL_FAULT_BNDERR:
3555 		to->si_addr = compat_ptr(from->si_addr);
3556 		to->si_lower = compat_ptr(from->si_lower);
3557 		to->si_upper = compat_ptr(from->si_upper);
3558 		break;
3559 	case SIL_FAULT_PKUERR:
3560 		to->si_addr = compat_ptr(from->si_addr);
3561 		to->si_pkey = from->si_pkey;
3562 		break;
3563 	case SIL_FAULT_PERF_EVENT:
3564 		to->si_addr = compat_ptr(from->si_addr);
3565 		to->si_perf_data = from->si_perf_data;
3566 		to->si_perf_type = from->si_perf_type;
3567 		to->si_perf_flags = from->si_perf_flags;
3568 		break;
3569 	case SIL_CHLD:
3570 		to->si_pid    = from->si_pid;
3571 		to->si_uid    = from->si_uid;
3572 		to->si_status = from->si_status;
3573 #ifdef CONFIG_X86_X32_ABI
3574 		if (in_x32_syscall()) {
3575 			to->si_utime = from->_sifields._sigchld_x32._utime;
3576 			to->si_stime = from->_sifields._sigchld_x32._stime;
3577 		} else
3578 #endif
3579 		{
3580 			to->si_utime = from->si_utime;
3581 			to->si_stime = from->si_stime;
3582 		}
3583 		break;
3584 	case SIL_RT:
3585 		to->si_pid = from->si_pid;
3586 		to->si_uid = from->si_uid;
3587 		to->si_int = from->si_int;
3588 		break;
3589 	case SIL_SYS:
3590 		to->si_call_addr = compat_ptr(from->si_call_addr);
3591 		to->si_syscall   = from->si_syscall;
3592 		to->si_arch      = from->si_arch;
3593 		break;
3594 	}
3595 	return 0;
3596 }
3597 
3598 static int __copy_siginfo_from_user32(int signo, struct kernel_siginfo *to,
3599 				      const struct compat_siginfo __user *ufrom)
3600 {
3601 	struct compat_siginfo from;
3602 
3603 	if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3604 		return -EFAULT;
3605 
3606 	from.si_signo = signo;
3607 	return post_copy_siginfo_from_user32(to, &from);
3608 }
3609 
3610 int copy_siginfo_from_user32(struct kernel_siginfo *to,
3611 			     const struct compat_siginfo __user *ufrom)
3612 {
3613 	struct compat_siginfo from;
3614 
3615 	if (copy_from_user(&from, ufrom, sizeof(struct compat_siginfo)))
3616 		return -EFAULT;
3617 
3618 	return post_copy_siginfo_from_user32(to, &from);
3619 }
3620 #endif /* CONFIG_COMPAT */
3621 
3622 /**
3623  *  do_sigtimedwait - wait for queued signals specified in @which
3624  *  @which: queued signals to wait for
3625  *  @info: if non-null, the signal's siginfo is returned here
3626  *  @ts: upper bound on process time suspension
3627  */
3628 static int do_sigtimedwait(const sigset_t *which, kernel_siginfo_t *info,
3629 		    const struct timespec64 *ts)
3630 {
3631 	ktime_t *to = NULL, timeout = KTIME_MAX;
3632 	struct task_struct *tsk = current;
3633 	sigset_t mask = *which;
3634 	enum pid_type type;
3635 	int sig, ret = 0;
3636 
3637 	if (ts) {
3638 		if (!timespec64_valid(ts))
3639 			return -EINVAL;
3640 		timeout = timespec64_to_ktime(*ts);
3641 		to = &timeout;
3642 	}
3643 
3644 	/*
3645 	 * Invert the set of allowed signals to get those we want to block.
3646 	 */
3647 	sigdelsetmask(&mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
3648 	signotset(&mask);
3649 
3650 	spin_lock_irq(&tsk->sighand->siglock);
3651 	sig = dequeue_signal(&mask, info, &type);
3652 	if (!sig && timeout) {
3653 		/*
3654 		 * None ready, temporarily unblock those we're interested
3655 		 * while we are sleeping in so that we'll be awakened when
3656 		 * they arrive. Unblocking is always fine, we can avoid
3657 		 * set_current_blocked().
3658 		 */
3659 		tsk->real_blocked = tsk->blocked;
3660 		sigandsets(&tsk->blocked, &tsk->blocked, &mask);
3661 		recalc_sigpending();
3662 		spin_unlock_irq(&tsk->sighand->siglock);
3663 
3664 		__set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
3665 		ret = schedule_hrtimeout_range(to, tsk->timer_slack_ns,
3666 					       HRTIMER_MODE_REL);
3667 		spin_lock_irq(&tsk->sighand->siglock);
3668 		__set_task_blocked(tsk, &tsk->real_blocked);
3669 		sigemptyset(&tsk->real_blocked);
3670 		sig = dequeue_signal(&mask, info, &type);
3671 	}
3672 	spin_unlock_irq(&tsk->sighand->siglock);
3673 
3674 	if (sig)
3675 		return sig;
3676 	return ret ? -EINTR : -EAGAIN;
3677 }
3678 
3679 /**
3680  *  sys_rt_sigtimedwait - synchronously wait for queued signals specified
3681  *			in @uthese
3682  *  @uthese: queued signals to wait for
3683  *  @uinfo: if non-null, the signal's siginfo is returned here
3684  *  @uts: upper bound on process time suspension
3685  *  @sigsetsize: size of sigset_t type
3686  */
3687 SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
3688 		siginfo_t __user *, uinfo,
3689 		const struct __kernel_timespec __user *, uts,
3690 		size_t, sigsetsize)
3691 {
3692 	sigset_t these;
3693 	struct timespec64 ts;
3694 	kernel_siginfo_t info;
3695 	int ret;
3696 
3697 	/* XXX: Don't preclude handling different sized sigset_t's.  */
3698 	if (sigsetsize != sizeof(sigset_t))
3699 		return -EINVAL;
3700 
3701 	if (copy_from_user(&these, uthese, sizeof(these)))
3702 		return -EFAULT;
3703 
3704 	if (uts) {
3705 		if (get_timespec64(&ts, uts))
3706 			return -EFAULT;
3707 	}
3708 
3709 	ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3710 
3711 	if (ret > 0 && uinfo) {
3712 		if (copy_siginfo_to_user(uinfo, &info))
3713 			ret = -EFAULT;
3714 	}
3715 
3716 	return ret;
3717 }
3718 
3719 #ifdef CONFIG_COMPAT_32BIT_TIME
3720 SYSCALL_DEFINE4(rt_sigtimedwait_time32, const sigset_t __user *, uthese,
3721 		siginfo_t __user *, uinfo,
3722 		const struct old_timespec32 __user *, uts,
3723 		size_t, sigsetsize)
3724 {
3725 	sigset_t these;
3726 	struct timespec64 ts;
3727 	kernel_siginfo_t info;
3728 	int ret;
3729 
3730 	if (sigsetsize != sizeof(sigset_t))
3731 		return -EINVAL;
3732 
3733 	if (copy_from_user(&these, uthese, sizeof(these)))
3734 		return -EFAULT;
3735 
3736 	if (uts) {
3737 		if (get_old_timespec32(&ts, uts))
3738 			return -EFAULT;
3739 	}
3740 
3741 	ret = do_sigtimedwait(&these, &info, uts ? &ts : NULL);
3742 
3743 	if (ret > 0 && uinfo) {
3744 		if (copy_siginfo_to_user(uinfo, &info))
3745 			ret = -EFAULT;
3746 	}
3747 
3748 	return ret;
3749 }
3750 #endif
3751 
3752 #ifdef CONFIG_COMPAT
3753 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time64, compat_sigset_t __user *, uthese,
3754 		struct compat_siginfo __user *, uinfo,
3755 		struct __kernel_timespec __user *, uts, compat_size_t, sigsetsize)
3756 {
3757 	sigset_t s;
3758 	struct timespec64 t;
3759 	kernel_siginfo_t info;
3760 	long ret;
3761 
3762 	if (sigsetsize != sizeof(sigset_t))
3763 		return -EINVAL;
3764 
3765 	if (get_compat_sigset(&s, uthese))
3766 		return -EFAULT;
3767 
3768 	if (uts) {
3769 		if (get_timespec64(&t, uts))
3770 			return -EFAULT;
3771 	}
3772 
3773 	ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3774 
3775 	if (ret > 0 && uinfo) {
3776 		if (copy_siginfo_to_user32(uinfo, &info))
3777 			ret = -EFAULT;
3778 	}
3779 
3780 	return ret;
3781 }
3782 
3783 #ifdef CONFIG_COMPAT_32BIT_TIME
3784 COMPAT_SYSCALL_DEFINE4(rt_sigtimedwait_time32, compat_sigset_t __user *, uthese,
3785 		struct compat_siginfo __user *, uinfo,
3786 		struct old_timespec32 __user *, uts, compat_size_t, sigsetsize)
3787 {
3788 	sigset_t s;
3789 	struct timespec64 t;
3790 	kernel_siginfo_t info;
3791 	long ret;
3792 
3793 	if (sigsetsize != sizeof(sigset_t))
3794 		return -EINVAL;
3795 
3796 	if (get_compat_sigset(&s, uthese))
3797 		return -EFAULT;
3798 
3799 	if (uts) {
3800 		if (get_old_timespec32(&t, uts))
3801 			return -EFAULT;
3802 	}
3803 
3804 	ret = do_sigtimedwait(&s, &info, uts ? &t : NULL);
3805 
3806 	if (ret > 0 && uinfo) {
3807 		if (copy_siginfo_to_user32(uinfo, &info))
3808 			ret = -EFAULT;
3809 	}
3810 
3811 	return ret;
3812 }
3813 #endif
3814 #endif
3815 
3816 static void prepare_kill_siginfo(int sig, struct kernel_siginfo *info,
3817 				 enum pid_type type)
3818 {
3819 	clear_siginfo(info);
3820 	info->si_signo = sig;
3821 	info->si_errno = 0;
3822 	info->si_code = (type == PIDTYPE_PID) ? SI_TKILL : SI_USER;
3823 	info->si_pid = task_tgid_vnr(current);
3824 	info->si_uid = from_kuid_munged(current_user_ns(), current_uid());
3825 }
3826 
3827 /**
3828  *  sys_kill - send a signal to a process
3829  *  @pid: the PID of the process
3830  *  @sig: signal to be sent
3831  */
3832 SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
3833 {
3834 	struct kernel_siginfo info;
3835 
3836 	prepare_kill_siginfo(sig, &info, PIDTYPE_TGID);
3837 
3838 	return kill_something_info(sig, &info, pid);
3839 }
3840 
3841 /*
3842  * Verify that the signaler and signalee either are in the same pid namespace
3843  * or that the signaler's pid namespace is an ancestor of the signalee's pid
3844  * namespace.
3845  */
3846 static bool access_pidfd_pidns(struct pid *pid)
3847 {
3848 	struct pid_namespace *active = task_active_pid_ns(current);
3849 	struct pid_namespace *p = ns_of_pid(pid);
3850 
3851 	for (;;) {
3852 		if (!p)
3853 			return false;
3854 		if (p == active)
3855 			break;
3856 		p = p->parent;
3857 	}
3858 
3859 	return true;
3860 }
3861 
3862 static int copy_siginfo_from_user_any(kernel_siginfo_t *kinfo,
3863 		siginfo_t __user *info)
3864 {
3865 #ifdef CONFIG_COMPAT
3866 	/*
3867 	 * Avoid hooking up compat syscalls and instead handle necessary
3868 	 * conversions here. Note, this is a stop-gap measure and should not be
3869 	 * considered a generic solution.
3870 	 */
3871 	if (in_compat_syscall())
3872 		return copy_siginfo_from_user32(
3873 			kinfo, (struct compat_siginfo __user *)info);
3874 #endif
3875 	return copy_siginfo_from_user(kinfo, info);
3876 }
3877 
3878 static struct pid *pidfd_to_pid(const struct file *file)
3879 {
3880 	struct pid *pid;
3881 
3882 	pid = pidfd_pid(file);
3883 	if (!IS_ERR(pid))
3884 		return pid;
3885 
3886 	return tgid_pidfd_to_pid(file);
3887 }
3888 
3889 #define PIDFD_SEND_SIGNAL_FLAGS                            \
3890 	(PIDFD_SIGNAL_THREAD | PIDFD_SIGNAL_THREAD_GROUP | \
3891 	 PIDFD_SIGNAL_PROCESS_GROUP)
3892 
3893 /**
3894  * sys_pidfd_send_signal - Signal a process through a pidfd
3895  * @pidfd:  file descriptor of the process
3896  * @sig:    signal to send
3897  * @info:   signal info
3898  * @flags:  future flags
3899  *
3900  * Send the signal to the thread group or to the individual thread depending
3901  * on PIDFD_THREAD.
3902  * In the future extension to @flags may be used to override the default scope
3903  * of @pidfd.
3904  *
3905  * Return: 0 on success, negative errno on failure
3906  */
3907 SYSCALL_DEFINE4(pidfd_send_signal, int, pidfd, int, sig,
3908 		siginfo_t __user *, info, unsigned int, flags)
3909 {
3910 	int ret;
3911 	struct fd f;
3912 	struct pid *pid;
3913 	kernel_siginfo_t kinfo;
3914 	enum pid_type type;
3915 
3916 	/* Enforce flags be set to 0 until we add an extension. */
3917 	if (flags & ~PIDFD_SEND_SIGNAL_FLAGS)
3918 		return -EINVAL;
3919 
3920 	/* Ensure that only a single signal scope determining flag is set. */
3921 	if (hweight32(flags & PIDFD_SEND_SIGNAL_FLAGS) > 1)
3922 		return -EINVAL;
3923 
3924 	f = fdget(pidfd);
3925 	if (!fd_file(f))
3926 		return -EBADF;
3927 
3928 	/* Is this a pidfd? */
3929 	pid = pidfd_to_pid(fd_file(f));
3930 	if (IS_ERR(pid)) {
3931 		ret = PTR_ERR(pid);
3932 		goto err;
3933 	}
3934 
3935 	ret = -EINVAL;
3936 	if (!access_pidfd_pidns(pid))
3937 		goto err;
3938 
3939 	switch (flags) {
3940 	case 0:
3941 		/* Infer scope from the type of pidfd. */
3942 		if (fd_file(f)->f_flags & PIDFD_THREAD)
3943 			type = PIDTYPE_PID;
3944 		else
3945 			type = PIDTYPE_TGID;
3946 		break;
3947 	case PIDFD_SIGNAL_THREAD:
3948 		type = PIDTYPE_PID;
3949 		break;
3950 	case PIDFD_SIGNAL_THREAD_GROUP:
3951 		type = PIDTYPE_TGID;
3952 		break;
3953 	case PIDFD_SIGNAL_PROCESS_GROUP:
3954 		type = PIDTYPE_PGID;
3955 		break;
3956 	}
3957 
3958 	if (info) {
3959 		ret = copy_siginfo_from_user_any(&kinfo, info);
3960 		if (unlikely(ret))
3961 			goto err;
3962 
3963 		ret = -EINVAL;
3964 		if (unlikely(sig != kinfo.si_signo))
3965 			goto err;
3966 
3967 		/* Only allow sending arbitrary signals to yourself. */
3968 		ret = -EPERM;
3969 		if ((task_pid(current) != pid || type > PIDTYPE_TGID) &&
3970 		    (kinfo.si_code >= 0 || kinfo.si_code == SI_TKILL))
3971 			goto err;
3972 	} else {
3973 		prepare_kill_siginfo(sig, &kinfo, type);
3974 	}
3975 
3976 	if (type == PIDTYPE_PGID)
3977 		ret = kill_pgrp_info(sig, &kinfo, pid);
3978 	else
3979 		ret = kill_pid_info_type(sig, &kinfo, pid, type);
3980 err:
3981 	fdput(f);
3982 	return ret;
3983 }
3984 
3985 static int
3986 do_send_specific(pid_t tgid, pid_t pid, int sig, struct kernel_siginfo *info)
3987 {
3988 	struct task_struct *p;
3989 	int error = -ESRCH;
3990 
3991 	rcu_read_lock();
3992 	p = find_task_by_vpid(pid);
3993 	if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
3994 		error = check_kill_permission(sig, info, p);
3995 		/*
3996 		 * The null signal is a permissions and process existence
3997 		 * probe.  No signal is actually delivered.
3998 		 */
3999 		if (!error && sig) {
4000 			error = do_send_sig_info(sig, info, p, PIDTYPE_PID);
4001 			/*
4002 			 * If lock_task_sighand() failed we pretend the task
4003 			 * dies after receiving the signal. The window is tiny,
4004 			 * and the signal is private anyway.
4005 			 */
4006 			if (unlikely(error == -ESRCH))
4007 				error = 0;
4008 		}
4009 	}
4010 	rcu_read_unlock();
4011 
4012 	return error;
4013 }
4014 
4015 static int do_tkill(pid_t tgid, pid_t pid, int sig)
4016 {
4017 	struct kernel_siginfo info;
4018 
4019 	prepare_kill_siginfo(sig, &info, PIDTYPE_PID);
4020 
4021 	return do_send_specific(tgid, pid, sig, &info);
4022 }
4023 
4024 /**
4025  *  sys_tgkill - send signal to one specific thread
4026  *  @tgid: the thread group ID of the thread
4027  *  @pid: the PID of the thread
4028  *  @sig: signal to be sent
4029  *
4030  *  This syscall also checks the @tgid and returns -ESRCH even if the PID
4031  *  exists but it's not belonging to the target process anymore. This
4032  *  method solves the problem of threads exiting and PIDs getting reused.
4033  */
4034 SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
4035 {
4036 	/* This is only valid for single tasks */
4037 	if (pid <= 0 || tgid <= 0)
4038 		return -EINVAL;
4039 
4040 	return do_tkill(tgid, pid, sig);
4041 }
4042 
4043 /**
4044  *  sys_tkill - send signal to one specific task
4045  *  @pid: the PID of the task
4046  *  @sig: signal to be sent
4047  *
4048  *  Send a signal to only one task, even if it's a CLONE_THREAD task.
4049  */
4050 SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
4051 {
4052 	/* This is only valid for single tasks */
4053 	if (pid <= 0)
4054 		return -EINVAL;
4055 
4056 	return do_tkill(0, pid, sig);
4057 }
4058 
4059 static int do_rt_sigqueueinfo(pid_t pid, int sig, kernel_siginfo_t *info)
4060 {
4061 	/* Not even root can pretend to send signals from the kernel.
4062 	 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4063 	 */
4064 	if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4065 	    (task_pid_vnr(current) != pid))
4066 		return -EPERM;
4067 
4068 	/* POSIX.1b doesn't mention process groups.  */
4069 	return kill_proc_info(sig, info, pid);
4070 }
4071 
4072 /**
4073  *  sys_rt_sigqueueinfo - send signal information to a signal
4074  *  @pid: the PID of the thread
4075  *  @sig: signal to be sent
4076  *  @uinfo: signal info to be sent
4077  */
4078 SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
4079 		siginfo_t __user *, uinfo)
4080 {
4081 	kernel_siginfo_t info;
4082 	int ret = __copy_siginfo_from_user(sig, &info, uinfo);
4083 	if (unlikely(ret))
4084 		return ret;
4085 	return do_rt_sigqueueinfo(pid, sig, &info);
4086 }
4087 
4088 #ifdef CONFIG_COMPAT
4089 COMPAT_SYSCALL_DEFINE3(rt_sigqueueinfo,
4090 			compat_pid_t, pid,
4091 			int, sig,
4092 			struct compat_siginfo __user *, uinfo)
4093 {
4094 	kernel_siginfo_t info;
4095 	int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4096 	if (unlikely(ret))
4097 		return ret;
4098 	return do_rt_sigqueueinfo(pid, sig, &info);
4099 }
4100 #endif
4101 
4102 static int do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, kernel_siginfo_t *info)
4103 {
4104 	/* This is only valid for single tasks */
4105 	if (pid <= 0 || tgid <= 0)
4106 		return -EINVAL;
4107 
4108 	/* Not even root can pretend to send signals from the kernel.
4109 	 * Nor can they impersonate a kill()/tgkill(), which adds source info.
4110 	 */
4111 	if ((info->si_code >= 0 || info->si_code == SI_TKILL) &&
4112 	    (task_pid_vnr(current) != pid))
4113 		return -EPERM;
4114 
4115 	return do_send_specific(tgid, pid, sig, info);
4116 }
4117 
4118 SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
4119 		siginfo_t __user *, uinfo)
4120 {
4121 	kernel_siginfo_t info;
4122 	int ret = __copy_siginfo_from_user(sig, &info, uinfo);
4123 	if (unlikely(ret))
4124 		return ret;
4125 	return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4126 }
4127 
4128 #ifdef CONFIG_COMPAT
4129 COMPAT_SYSCALL_DEFINE4(rt_tgsigqueueinfo,
4130 			compat_pid_t, tgid,
4131 			compat_pid_t, pid,
4132 			int, sig,
4133 			struct compat_siginfo __user *, uinfo)
4134 {
4135 	kernel_siginfo_t info;
4136 	int ret = __copy_siginfo_from_user32(sig, &info, uinfo);
4137 	if (unlikely(ret))
4138 		return ret;
4139 	return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
4140 }
4141 #endif
4142 
4143 /*
4144  * For kthreads only, must not be used if cloned with CLONE_SIGHAND
4145  */
4146 void kernel_sigaction(int sig, __sighandler_t action)
4147 {
4148 	spin_lock_irq(&current->sighand->siglock);
4149 	current->sighand->action[sig - 1].sa.sa_handler = action;
4150 	if (action == SIG_IGN) {
4151 		sigset_t mask;
4152 
4153 		sigemptyset(&mask);
4154 		sigaddset(&mask, sig);
4155 
4156 		flush_sigqueue_mask(&mask, &current->signal->shared_pending);
4157 		flush_sigqueue_mask(&mask, &current->pending);
4158 		recalc_sigpending();
4159 	}
4160 	spin_unlock_irq(&current->sighand->siglock);
4161 }
4162 EXPORT_SYMBOL(kernel_sigaction);
4163 
4164 void __weak sigaction_compat_abi(struct k_sigaction *act,
4165 		struct k_sigaction *oact)
4166 {
4167 }
4168 
4169 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
4170 {
4171 	struct task_struct *p = current, *t;
4172 	struct k_sigaction *k;
4173 	sigset_t mask;
4174 
4175 	if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
4176 		return -EINVAL;
4177 
4178 	k = &p->sighand->action[sig-1];
4179 
4180 	spin_lock_irq(&p->sighand->siglock);
4181 	if (k->sa.sa_flags & SA_IMMUTABLE) {
4182 		spin_unlock_irq(&p->sighand->siglock);
4183 		return -EINVAL;
4184 	}
4185 	if (oact)
4186 		*oact = *k;
4187 
4188 	/*
4189 	 * Make sure that we never accidentally claim to support SA_UNSUPPORTED,
4190 	 * e.g. by having an architecture use the bit in their uapi.
4191 	 */
4192 	BUILD_BUG_ON(UAPI_SA_FLAGS & SA_UNSUPPORTED);
4193 
4194 	/*
4195 	 * Clear unknown flag bits in order to allow userspace to detect missing
4196 	 * support for flag bits and to allow the kernel to use non-uapi bits
4197 	 * internally.
4198 	 */
4199 	if (act)
4200 		act->sa.sa_flags &= UAPI_SA_FLAGS;
4201 	if (oact)
4202 		oact->sa.sa_flags &= UAPI_SA_FLAGS;
4203 
4204 	sigaction_compat_abi(act, oact);
4205 
4206 	if (act) {
4207 		sigdelsetmask(&act->sa.sa_mask,
4208 			      sigmask(SIGKILL) | sigmask(SIGSTOP));
4209 		*k = *act;
4210 		/*
4211 		 * POSIX 3.3.1.3:
4212 		 *  "Setting a signal action to SIG_IGN for a signal that is
4213 		 *   pending shall cause the pending signal to be discarded,
4214 		 *   whether or not it is blocked."
4215 		 *
4216 		 *  "Setting a signal action to SIG_DFL for a signal that is
4217 		 *   pending and whose default action is to ignore the signal
4218 		 *   (for example, SIGCHLD), shall cause the pending signal to
4219 		 *   be discarded, whether or not it is blocked"
4220 		 */
4221 		if (sig_handler_ignored(sig_handler(p, sig), sig)) {
4222 			sigemptyset(&mask);
4223 			sigaddset(&mask, sig);
4224 			flush_sigqueue_mask(&mask, &p->signal->shared_pending);
4225 			for_each_thread(p, t)
4226 				flush_sigqueue_mask(&mask, &t->pending);
4227 		}
4228 	}
4229 
4230 	spin_unlock_irq(&p->sighand->siglock);
4231 	return 0;
4232 }
4233 
4234 #ifdef CONFIG_DYNAMIC_SIGFRAME
4235 static inline void sigaltstack_lock(void)
4236 	__acquires(&current->sighand->siglock)
4237 {
4238 	spin_lock_irq(&current->sighand->siglock);
4239 }
4240 
4241 static inline void sigaltstack_unlock(void)
4242 	__releases(&current->sighand->siglock)
4243 {
4244 	spin_unlock_irq(&current->sighand->siglock);
4245 }
4246 #else
4247 static inline void sigaltstack_lock(void) { }
4248 static inline void sigaltstack_unlock(void) { }
4249 #endif
4250 
4251 static int
4252 do_sigaltstack (const stack_t *ss, stack_t *oss, unsigned long sp,
4253 		size_t min_ss_size)
4254 {
4255 	struct task_struct *t = current;
4256 	int ret = 0;
4257 
4258 	if (oss) {
4259 		memset(oss, 0, sizeof(stack_t));
4260 		oss->ss_sp = (void __user *) t->sas_ss_sp;
4261 		oss->ss_size = t->sas_ss_size;
4262 		oss->ss_flags = sas_ss_flags(sp) |
4263 			(current->sas_ss_flags & SS_FLAG_BITS);
4264 	}
4265 
4266 	if (ss) {
4267 		void __user *ss_sp = ss->ss_sp;
4268 		size_t ss_size = ss->ss_size;
4269 		unsigned ss_flags = ss->ss_flags;
4270 		int ss_mode;
4271 
4272 		if (unlikely(on_sig_stack(sp)))
4273 			return -EPERM;
4274 
4275 		ss_mode = ss_flags & ~SS_FLAG_BITS;
4276 		if (unlikely(ss_mode != SS_DISABLE && ss_mode != SS_ONSTACK &&
4277 				ss_mode != 0))
4278 			return -EINVAL;
4279 
4280 		/*
4281 		 * Return before taking any locks if no actual
4282 		 * sigaltstack changes were requested.
4283 		 */
4284 		if (t->sas_ss_sp == (unsigned long)ss_sp &&
4285 		    t->sas_ss_size == ss_size &&
4286 		    t->sas_ss_flags == ss_flags)
4287 			return 0;
4288 
4289 		sigaltstack_lock();
4290 		if (ss_mode == SS_DISABLE) {
4291 			ss_size = 0;
4292 			ss_sp = NULL;
4293 		} else {
4294 			if (unlikely(ss_size < min_ss_size))
4295 				ret = -ENOMEM;
4296 			if (!sigaltstack_size_valid(ss_size))
4297 				ret = -ENOMEM;
4298 		}
4299 		if (!ret) {
4300 			t->sas_ss_sp = (unsigned long) ss_sp;
4301 			t->sas_ss_size = ss_size;
4302 			t->sas_ss_flags = ss_flags;
4303 		}
4304 		sigaltstack_unlock();
4305 	}
4306 	return ret;
4307 }
4308 
4309 SYSCALL_DEFINE2(sigaltstack,const stack_t __user *,uss, stack_t __user *,uoss)
4310 {
4311 	stack_t new, old;
4312 	int err;
4313 	if (uss && copy_from_user(&new, uss, sizeof(stack_t)))
4314 		return -EFAULT;
4315 	err = do_sigaltstack(uss ? &new : NULL, uoss ? &old : NULL,
4316 			      current_user_stack_pointer(),
4317 			      MINSIGSTKSZ);
4318 	if (!err && uoss && copy_to_user(uoss, &old, sizeof(stack_t)))
4319 		err = -EFAULT;
4320 	return err;
4321 }
4322 
4323 int restore_altstack(const stack_t __user *uss)
4324 {
4325 	stack_t new;
4326 	if (copy_from_user(&new, uss, sizeof(stack_t)))
4327 		return -EFAULT;
4328 	(void)do_sigaltstack(&new, NULL, current_user_stack_pointer(),
4329 			     MINSIGSTKSZ);
4330 	/* squash all but EFAULT for now */
4331 	return 0;
4332 }
4333 
4334 int __save_altstack(stack_t __user *uss, unsigned long sp)
4335 {
4336 	struct task_struct *t = current;
4337 	int err = __put_user((void __user *)t->sas_ss_sp, &uss->ss_sp) |
4338 		__put_user(t->sas_ss_flags, &uss->ss_flags) |
4339 		__put_user(t->sas_ss_size, &uss->ss_size);
4340 	return err;
4341 }
4342 
4343 #ifdef CONFIG_COMPAT
4344 static int do_compat_sigaltstack(const compat_stack_t __user *uss_ptr,
4345 				 compat_stack_t __user *uoss_ptr)
4346 {
4347 	stack_t uss, uoss;
4348 	int ret;
4349 
4350 	if (uss_ptr) {
4351 		compat_stack_t uss32;
4352 		if (copy_from_user(&uss32, uss_ptr, sizeof(compat_stack_t)))
4353 			return -EFAULT;
4354 		uss.ss_sp = compat_ptr(uss32.ss_sp);
4355 		uss.ss_flags = uss32.ss_flags;
4356 		uss.ss_size = uss32.ss_size;
4357 	}
4358 	ret = do_sigaltstack(uss_ptr ? &uss : NULL, &uoss,
4359 			     compat_user_stack_pointer(),
4360 			     COMPAT_MINSIGSTKSZ);
4361 	if (ret >= 0 && uoss_ptr)  {
4362 		compat_stack_t old;
4363 		memset(&old, 0, sizeof(old));
4364 		old.ss_sp = ptr_to_compat(uoss.ss_sp);
4365 		old.ss_flags = uoss.ss_flags;
4366 		old.ss_size = uoss.ss_size;
4367 		if (copy_to_user(uoss_ptr, &old, sizeof(compat_stack_t)))
4368 			ret = -EFAULT;
4369 	}
4370 	return ret;
4371 }
4372 
4373 COMPAT_SYSCALL_DEFINE2(sigaltstack,
4374 			const compat_stack_t __user *, uss_ptr,
4375 			compat_stack_t __user *, uoss_ptr)
4376 {
4377 	return do_compat_sigaltstack(uss_ptr, uoss_ptr);
4378 }
4379 
4380 int compat_restore_altstack(const compat_stack_t __user *uss)
4381 {
4382 	int err = do_compat_sigaltstack(uss, NULL);
4383 	/* squash all but -EFAULT for now */
4384 	return err == -EFAULT ? err : 0;
4385 }
4386 
4387 int __compat_save_altstack(compat_stack_t __user *uss, unsigned long sp)
4388 {
4389 	int err;
4390 	struct task_struct *t = current;
4391 	err = __put_user(ptr_to_compat((void __user *)t->sas_ss_sp),
4392 			 &uss->ss_sp) |
4393 		__put_user(t->sas_ss_flags, &uss->ss_flags) |
4394 		__put_user(t->sas_ss_size, &uss->ss_size);
4395 	return err;
4396 }
4397 #endif
4398 
4399 #ifdef __ARCH_WANT_SYS_SIGPENDING
4400 
4401 /**
4402  *  sys_sigpending - examine pending signals
4403  *  @uset: where mask of pending signal is returned
4404  */
4405 SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, uset)
4406 {
4407 	sigset_t set;
4408 
4409 	if (sizeof(old_sigset_t) > sizeof(*uset))
4410 		return -EINVAL;
4411 
4412 	do_sigpending(&set);
4413 
4414 	if (copy_to_user(uset, &set, sizeof(old_sigset_t)))
4415 		return -EFAULT;
4416 
4417 	return 0;
4418 }
4419 
4420 #ifdef CONFIG_COMPAT
4421 COMPAT_SYSCALL_DEFINE1(sigpending, compat_old_sigset_t __user *, set32)
4422 {
4423 	sigset_t set;
4424 
4425 	do_sigpending(&set);
4426 
4427 	return put_user(set.sig[0], set32);
4428 }
4429 #endif
4430 
4431 #endif
4432 
4433 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
4434 /**
4435  *  sys_sigprocmask - examine and change blocked signals
4436  *  @how: whether to add, remove, or set signals
4437  *  @nset: signals to add or remove (if non-null)
4438  *  @oset: previous value of signal mask if non-null
4439  *
4440  * Some platforms have their own version with special arguments;
4441  * others support only sys_rt_sigprocmask.
4442  */
4443 
4444 SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, nset,
4445 		old_sigset_t __user *, oset)
4446 {
4447 	old_sigset_t old_set, new_set;
4448 	sigset_t new_blocked;
4449 
4450 	old_set = current->blocked.sig[0];
4451 
4452 	if (nset) {
4453 		if (copy_from_user(&new_set, nset, sizeof(*nset)))
4454 			return -EFAULT;
4455 
4456 		new_blocked = current->blocked;
4457 
4458 		switch (how) {
4459 		case SIG_BLOCK:
4460 			sigaddsetmask(&new_blocked, new_set);
4461 			break;
4462 		case SIG_UNBLOCK:
4463 			sigdelsetmask(&new_blocked, new_set);
4464 			break;
4465 		case SIG_SETMASK:
4466 			new_blocked.sig[0] = new_set;
4467 			break;
4468 		default:
4469 			return -EINVAL;
4470 		}
4471 
4472 		set_current_blocked(&new_blocked);
4473 	}
4474 
4475 	if (oset) {
4476 		if (copy_to_user(oset, &old_set, sizeof(*oset)))
4477 			return -EFAULT;
4478 	}
4479 
4480 	return 0;
4481 }
4482 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
4483 
4484 #ifndef CONFIG_ODD_RT_SIGACTION
4485 /**
4486  *  sys_rt_sigaction - alter an action taken by a process
4487  *  @sig: signal to be sent
4488  *  @act: new sigaction
4489  *  @oact: used to save the previous sigaction
4490  *  @sigsetsize: size of sigset_t type
4491  */
4492 SYSCALL_DEFINE4(rt_sigaction, int, sig,
4493 		const struct sigaction __user *, act,
4494 		struct sigaction __user *, oact,
4495 		size_t, sigsetsize)
4496 {
4497 	struct k_sigaction new_sa, old_sa;
4498 	int ret;
4499 
4500 	/* XXX: Don't preclude handling different sized sigset_t's.  */
4501 	if (sigsetsize != sizeof(sigset_t))
4502 		return -EINVAL;
4503 
4504 	if (act && copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
4505 		return -EFAULT;
4506 
4507 	ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
4508 	if (ret)
4509 		return ret;
4510 
4511 	if (oact && copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
4512 		return -EFAULT;
4513 
4514 	return 0;
4515 }
4516 #ifdef CONFIG_COMPAT
4517 COMPAT_SYSCALL_DEFINE4(rt_sigaction, int, sig,
4518 		const struct compat_sigaction __user *, act,
4519 		struct compat_sigaction __user *, oact,
4520 		compat_size_t, sigsetsize)
4521 {
4522 	struct k_sigaction new_ka, old_ka;
4523 #ifdef __ARCH_HAS_SA_RESTORER
4524 	compat_uptr_t restorer;
4525 #endif
4526 	int ret;
4527 
4528 	/* XXX: Don't preclude handling different sized sigset_t's.  */
4529 	if (sigsetsize != sizeof(compat_sigset_t))
4530 		return -EINVAL;
4531 
4532 	if (act) {
4533 		compat_uptr_t handler;
4534 		ret = get_user(handler, &act->sa_handler);
4535 		new_ka.sa.sa_handler = compat_ptr(handler);
4536 #ifdef __ARCH_HAS_SA_RESTORER
4537 		ret |= get_user(restorer, &act->sa_restorer);
4538 		new_ka.sa.sa_restorer = compat_ptr(restorer);
4539 #endif
4540 		ret |= get_compat_sigset(&new_ka.sa.sa_mask, &act->sa_mask);
4541 		ret |= get_user(new_ka.sa.sa_flags, &act->sa_flags);
4542 		if (ret)
4543 			return -EFAULT;
4544 	}
4545 
4546 	ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4547 	if (!ret && oact) {
4548 		ret = put_user(ptr_to_compat(old_ka.sa.sa_handler),
4549 			       &oact->sa_handler);
4550 		ret |= put_compat_sigset(&oact->sa_mask, &old_ka.sa.sa_mask,
4551 					 sizeof(oact->sa_mask));
4552 		ret |= put_user(old_ka.sa.sa_flags, &oact->sa_flags);
4553 #ifdef __ARCH_HAS_SA_RESTORER
4554 		ret |= put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4555 				&oact->sa_restorer);
4556 #endif
4557 	}
4558 	return ret;
4559 }
4560 #endif
4561 #endif /* !CONFIG_ODD_RT_SIGACTION */
4562 
4563 #ifdef CONFIG_OLD_SIGACTION
4564 SYSCALL_DEFINE3(sigaction, int, sig,
4565 		const struct old_sigaction __user *, act,
4566 	        struct old_sigaction __user *, oact)
4567 {
4568 	struct k_sigaction new_ka, old_ka;
4569 	int ret;
4570 
4571 	if (act) {
4572 		old_sigset_t mask;
4573 		if (!access_ok(act, sizeof(*act)) ||
4574 		    __get_user(new_ka.sa.sa_handler, &act->sa_handler) ||
4575 		    __get_user(new_ka.sa.sa_restorer, &act->sa_restorer) ||
4576 		    __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4577 		    __get_user(mask, &act->sa_mask))
4578 			return -EFAULT;
4579 #ifdef __ARCH_HAS_KA_RESTORER
4580 		new_ka.ka_restorer = NULL;
4581 #endif
4582 		siginitset(&new_ka.sa.sa_mask, mask);
4583 	}
4584 
4585 	ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4586 
4587 	if (!ret && oact) {
4588 		if (!access_ok(oact, sizeof(*oact)) ||
4589 		    __put_user(old_ka.sa.sa_handler, &oact->sa_handler) ||
4590 		    __put_user(old_ka.sa.sa_restorer, &oact->sa_restorer) ||
4591 		    __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4592 		    __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4593 			return -EFAULT;
4594 	}
4595 
4596 	return ret;
4597 }
4598 #endif
4599 #ifdef CONFIG_COMPAT_OLD_SIGACTION
4600 COMPAT_SYSCALL_DEFINE3(sigaction, int, sig,
4601 		const struct compat_old_sigaction __user *, act,
4602 	        struct compat_old_sigaction __user *, oact)
4603 {
4604 	struct k_sigaction new_ka, old_ka;
4605 	int ret;
4606 	compat_old_sigset_t mask;
4607 	compat_uptr_t handler, restorer;
4608 
4609 	if (act) {
4610 		if (!access_ok(act, sizeof(*act)) ||
4611 		    __get_user(handler, &act->sa_handler) ||
4612 		    __get_user(restorer, &act->sa_restorer) ||
4613 		    __get_user(new_ka.sa.sa_flags, &act->sa_flags) ||
4614 		    __get_user(mask, &act->sa_mask))
4615 			return -EFAULT;
4616 
4617 #ifdef __ARCH_HAS_KA_RESTORER
4618 		new_ka.ka_restorer = NULL;
4619 #endif
4620 		new_ka.sa.sa_handler = compat_ptr(handler);
4621 		new_ka.sa.sa_restorer = compat_ptr(restorer);
4622 		siginitset(&new_ka.sa.sa_mask, mask);
4623 	}
4624 
4625 	ret = do_sigaction(sig, act ? &new_ka : NULL, oact ? &old_ka : NULL);
4626 
4627 	if (!ret && oact) {
4628 		if (!access_ok(oact, sizeof(*oact)) ||
4629 		    __put_user(ptr_to_compat(old_ka.sa.sa_handler),
4630 			       &oact->sa_handler) ||
4631 		    __put_user(ptr_to_compat(old_ka.sa.sa_restorer),
4632 			       &oact->sa_restorer) ||
4633 		    __put_user(old_ka.sa.sa_flags, &oact->sa_flags) ||
4634 		    __put_user(old_ka.sa.sa_mask.sig[0], &oact->sa_mask))
4635 			return -EFAULT;
4636 	}
4637 	return ret;
4638 }
4639 #endif
4640 
4641 #ifdef CONFIG_SGETMASK_SYSCALL
4642 
4643 /*
4644  * For backwards compatibility.  Functionality superseded by sigprocmask.
4645  */
4646 SYSCALL_DEFINE0(sgetmask)
4647 {
4648 	/* SMP safe */
4649 	return current->blocked.sig[0];
4650 }
4651 
4652 SYSCALL_DEFINE1(ssetmask, int, newmask)
4653 {
4654 	int old = current->blocked.sig[0];
4655 	sigset_t newset;
4656 
4657 	siginitset(&newset, newmask);
4658 	set_current_blocked(&newset);
4659 
4660 	return old;
4661 }
4662 #endif /* CONFIG_SGETMASK_SYSCALL */
4663 
4664 #ifdef __ARCH_WANT_SYS_SIGNAL
4665 /*
4666  * For backwards compatibility.  Functionality superseded by sigaction.
4667  */
4668 SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
4669 {
4670 	struct k_sigaction new_sa, old_sa;
4671 	int ret;
4672 
4673 	new_sa.sa.sa_handler = handler;
4674 	new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
4675 	sigemptyset(&new_sa.sa.sa_mask);
4676 
4677 	ret = do_sigaction(sig, &new_sa, &old_sa);
4678 
4679 	return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
4680 }
4681 #endif /* __ARCH_WANT_SYS_SIGNAL */
4682 
4683 #ifdef __ARCH_WANT_SYS_PAUSE
4684 
4685 SYSCALL_DEFINE0(pause)
4686 {
4687 	while (!signal_pending(current)) {
4688 		__set_current_state(TASK_INTERRUPTIBLE);
4689 		schedule();
4690 	}
4691 	return -ERESTARTNOHAND;
4692 }
4693 
4694 #endif
4695 
4696 static int sigsuspend(sigset_t *set)
4697 {
4698 	current->saved_sigmask = current->blocked;
4699 	set_current_blocked(set);
4700 
4701 	while (!signal_pending(current)) {
4702 		__set_current_state(TASK_INTERRUPTIBLE);
4703 		schedule();
4704 	}
4705 	set_restore_sigmask();
4706 	return -ERESTARTNOHAND;
4707 }
4708 
4709 /**
4710  *  sys_rt_sigsuspend - replace the signal mask for a value with the
4711  *	@unewset value until a signal is received
4712  *  @unewset: new signal mask value
4713  *  @sigsetsize: size of sigset_t type
4714  */
4715 SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
4716 {
4717 	sigset_t newset;
4718 
4719 	/* XXX: Don't preclude handling different sized sigset_t's.  */
4720 	if (sigsetsize != sizeof(sigset_t))
4721 		return -EINVAL;
4722 
4723 	if (copy_from_user(&newset, unewset, sizeof(newset)))
4724 		return -EFAULT;
4725 	return sigsuspend(&newset);
4726 }
4727 
4728 #ifdef CONFIG_COMPAT
4729 COMPAT_SYSCALL_DEFINE2(rt_sigsuspend, compat_sigset_t __user *, unewset, compat_size_t, sigsetsize)
4730 {
4731 	sigset_t newset;
4732 
4733 	/* XXX: Don't preclude handling different sized sigset_t's.  */
4734 	if (sigsetsize != sizeof(sigset_t))
4735 		return -EINVAL;
4736 
4737 	if (get_compat_sigset(&newset, unewset))
4738 		return -EFAULT;
4739 	return sigsuspend(&newset);
4740 }
4741 #endif
4742 
4743 #ifdef CONFIG_OLD_SIGSUSPEND
4744 SYSCALL_DEFINE1(sigsuspend, old_sigset_t, mask)
4745 {
4746 	sigset_t blocked;
4747 	siginitset(&blocked, mask);
4748 	return sigsuspend(&blocked);
4749 }
4750 #endif
4751 #ifdef CONFIG_OLD_SIGSUSPEND3
4752 SYSCALL_DEFINE3(sigsuspend, int, unused1, int, unused2, old_sigset_t, mask)
4753 {
4754 	sigset_t blocked;
4755 	siginitset(&blocked, mask);
4756 	return sigsuspend(&blocked);
4757 }
4758 #endif
4759 
4760 __weak const char *arch_vma_name(struct vm_area_struct *vma)
4761 {
4762 	return NULL;
4763 }
4764 
4765 static inline void siginfo_buildtime_checks(void)
4766 {
4767 	BUILD_BUG_ON(sizeof(struct siginfo) != SI_MAX_SIZE);
4768 
4769 	/* Verify the offsets in the two siginfos match */
4770 #define CHECK_OFFSET(field) \
4771 	BUILD_BUG_ON(offsetof(siginfo_t, field) != offsetof(kernel_siginfo_t, field))
4772 
4773 	/* kill */
4774 	CHECK_OFFSET(si_pid);
4775 	CHECK_OFFSET(si_uid);
4776 
4777 	/* timer */
4778 	CHECK_OFFSET(si_tid);
4779 	CHECK_OFFSET(si_overrun);
4780 	CHECK_OFFSET(si_value);
4781 
4782 	/* rt */
4783 	CHECK_OFFSET(si_pid);
4784 	CHECK_OFFSET(si_uid);
4785 	CHECK_OFFSET(si_value);
4786 
4787 	/* sigchld */
4788 	CHECK_OFFSET(si_pid);
4789 	CHECK_OFFSET(si_uid);
4790 	CHECK_OFFSET(si_status);
4791 	CHECK_OFFSET(si_utime);
4792 	CHECK_OFFSET(si_stime);
4793 
4794 	/* sigfault */
4795 	CHECK_OFFSET(si_addr);
4796 	CHECK_OFFSET(si_trapno);
4797 	CHECK_OFFSET(si_addr_lsb);
4798 	CHECK_OFFSET(si_lower);
4799 	CHECK_OFFSET(si_upper);
4800 	CHECK_OFFSET(si_pkey);
4801 	CHECK_OFFSET(si_perf_data);
4802 	CHECK_OFFSET(si_perf_type);
4803 	CHECK_OFFSET(si_perf_flags);
4804 
4805 	/* sigpoll */
4806 	CHECK_OFFSET(si_band);
4807 	CHECK_OFFSET(si_fd);
4808 
4809 	/* sigsys */
4810 	CHECK_OFFSET(si_call_addr);
4811 	CHECK_OFFSET(si_syscall);
4812 	CHECK_OFFSET(si_arch);
4813 #undef CHECK_OFFSET
4814 
4815 	/* usb asyncio */
4816 	BUILD_BUG_ON(offsetof(struct siginfo, si_pid) !=
4817 		     offsetof(struct siginfo, si_addr));
4818 	if (sizeof(int) == sizeof(void __user *)) {
4819 		BUILD_BUG_ON(sizeof_field(struct siginfo, si_pid) !=
4820 			     sizeof(void __user *));
4821 	} else {
4822 		BUILD_BUG_ON((sizeof_field(struct siginfo, si_pid) +
4823 			      sizeof_field(struct siginfo, si_uid)) !=
4824 			     sizeof(void __user *));
4825 		BUILD_BUG_ON(offsetofend(struct siginfo, si_pid) !=
4826 			     offsetof(struct siginfo, si_uid));
4827 	}
4828 #ifdef CONFIG_COMPAT
4829 	BUILD_BUG_ON(offsetof(struct compat_siginfo, si_pid) !=
4830 		     offsetof(struct compat_siginfo, si_addr));
4831 	BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4832 		     sizeof(compat_uptr_t));
4833 	BUILD_BUG_ON(sizeof_field(struct compat_siginfo, si_pid) !=
4834 		     sizeof_field(struct siginfo, si_pid));
4835 #endif
4836 }
4837 
4838 #if defined(CONFIG_SYSCTL)
4839 static struct ctl_table signal_debug_table[] = {
4840 #ifdef CONFIG_SYSCTL_EXCEPTION_TRACE
4841 	{
4842 		.procname	= "exception-trace",
4843 		.data		= &show_unhandled_signals,
4844 		.maxlen		= sizeof(int),
4845 		.mode		= 0644,
4846 		.proc_handler	= proc_dointvec
4847 	},
4848 #endif
4849 };
4850 
4851 static int __init init_signal_sysctls(void)
4852 {
4853 	register_sysctl_init("debug", signal_debug_table);
4854 	return 0;
4855 }
4856 early_initcall(init_signal_sysctls);
4857 #endif /* CONFIG_SYSCTL */
4858 
4859 void __init signals_init(void)
4860 {
4861 	siginfo_buildtime_checks();
4862 
4863 	sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC | SLAB_ACCOUNT);
4864 }
4865 
4866 #ifdef CONFIG_KGDB_KDB
4867 #include <linux/kdb.h>
4868 /*
4869  * kdb_send_sig - Allows kdb to send signals without exposing
4870  * signal internals.  This function checks if the required locks are
4871  * available before calling the main signal code, to avoid kdb
4872  * deadlocks.
4873  */
4874 void kdb_send_sig(struct task_struct *t, int sig)
4875 {
4876 	static struct task_struct *kdb_prev_t;
4877 	int new_t, ret;
4878 	if (!spin_trylock(&t->sighand->siglock)) {
4879 		kdb_printf("Can't do kill command now.\n"
4880 			   "The sigmask lock is held somewhere else in "
4881 			   "kernel, try again later\n");
4882 		return;
4883 	}
4884 	new_t = kdb_prev_t != t;
4885 	kdb_prev_t = t;
4886 	if (!task_is_running(t) && new_t) {
4887 		spin_unlock(&t->sighand->siglock);
4888 		kdb_printf("Process is not RUNNING, sending a signal from "
4889 			   "kdb risks deadlock\n"
4890 			   "on the run queue locks. "
4891 			   "The signal has _not_ been sent.\n"
4892 			   "Reissue the kill command if you want to risk "
4893 			   "the deadlock.\n");
4894 		return;
4895 	}
4896 	ret = send_signal_locked(sig, SEND_SIG_PRIV, t, PIDTYPE_PID);
4897 	spin_unlock(&t->sighand->siglock);
4898 	if (ret)
4899 		kdb_printf("Fail to deliver Signal %d to process %d.\n",
4900 			   sig, t->pid);
4901 	else
4902 		kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
4903 }
4904 #endif	/* CONFIG_KGDB_KDB */
4905