xref: /openbsd-src/sys/kern/kern_sig.c (revision 315d09dbf7c4a4f4f843436e6ff775334abae2c0)
1 /*	$OpenBSD: kern_sig.c,v 1.351 2024/11/20 10:21:22 claudio Exp $	*/
2 /*	$NetBSD: kern_sig.c,v 1.54 1996/04/22 01:38:32 christos Exp $	*/
3 
4 /*
5  * Copyright (c) 1997 Theo de Raadt. All rights reserved.
6  * Copyright (c) 1982, 1986, 1989, 1991, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  * (c) UNIX System Laboratories, Inc.
9  * All or some portions of this file are derived from material licensed
10  * to the University of California by American Telephone and Telegraph
11  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
12  * the permission of UNIX System Laboratories, Inc.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_sig.c	8.7 (Berkeley) 4/18/94
39  */
40 
41 #include <sys/param.h>
42 #include <sys/signalvar.h>
43 #include <sys/queue.h>
44 #include <sys/namei.h>
45 #include <sys/vnode.h>
46 #include <sys/event.h>
47 #include <sys/proc.h>
48 #include <sys/systm.h>
49 #include <sys/acct.h>
50 #include <sys/fcntl.h>
51 #include <sys/filedesc.h>
52 #include <sys/wait.h>
53 #include <sys/ktrace.h>
54 #include <sys/stat.h>
55 #include <sys/malloc.h>
56 #include <sys/pool.h>
57 #include <sys/sched.h>
58 #include <sys/user.h>
59 #include <sys/syslog.h>
60 #include <sys/ttycom.h>
61 #include <sys/pledge.h>
62 #include <sys/witness.h>
63 #include <sys/exec_elf.h>
64 
65 #include <sys/mount.h>
66 #include <sys/syscallargs.h>
67 
68 #include <uvm/uvm_extern.h>
69 #include <machine/tcb.h>
70 
71 int nosuidcoredump = 1;
72 
73 /*
74  * The array below categorizes the signals and their default actions.
75  */
76 const int sigprop[NSIG] = {
77 	0,			/* unused */
78 	SA_KILL,		/* SIGHUP */
79 	SA_KILL,		/* SIGINT */
80 	SA_KILL|SA_CORE,	/* SIGQUIT */
81 	SA_KILL|SA_CORE,	/* SIGILL */
82 	SA_KILL|SA_CORE,	/* SIGTRAP */
83 	SA_KILL|SA_CORE,	/* SIGABRT */
84 	SA_KILL|SA_CORE,	/* SIGEMT */
85 	SA_KILL|SA_CORE,	/* SIGFPE */
86 	SA_KILL,		/* SIGKILL */
87 	SA_KILL|SA_CORE,	/* SIGBUS */
88 	SA_KILL|SA_CORE,	/* SIGSEGV */
89 	SA_KILL|SA_CORE,	/* SIGSYS */
90 	SA_KILL,		/* SIGPIPE */
91 	SA_KILL,		/* SIGALRM */
92 	SA_KILL,		/* SIGTERM */
93 	SA_IGNORE,		/* SIGURG */
94 	SA_STOP,		/* SIGSTOP */
95 	SA_STOP|SA_TTYSTOP,	/* SIGTSTP */
96 	SA_IGNORE|SA_CONT,	/* SIGCONT */
97 	SA_IGNORE,		/* SIGCHLD */
98 	SA_STOP|SA_TTYSTOP,	/* SIGTTIN */
99 	SA_STOP|SA_TTYSTOP,	/* SIGTTOU */
100 	SA_IGNORE,		/* SIGIO */
101 	SA_KILL,		/* SIGXCPU */
102 	SA_KILL,		/* SIGXFSZ */
103 	SA_KILL,		/* SIGVTALRM */
104 	SA_KILL,		/* SIGPROF */
105 	SA_IGNORE,		/* SIGWINCH  */
106 	SA_IGNORE,		/* SIGINFO */
107 	SA_KILL,		/* SIGUSR1 */
108 	SA_KILL,		/* SIGUSR2 */
109 	SA_IGNORE,		/* SIGTHR */
110 };
111 
112 #define	CONTSIGMASK	(sigmask(SIGCONT))
113 #define	STOPSIGMASK	(sigmask(SIGSTOP) | sigmask(SIGTSTP) | \
114 			    sigmask(SIGTTIN) | sigmask(SIGTTOU))
115 
116 void setsigvec(struct proc *, int, struct sigaction *);
117 
118 int proc_trap(struct proc *, int);
119 void proc_stop(struct proc *p, int);
120 void proc_stop_sweep(void *);
121 void *proc_stop_si;
122 
123 void setsigctx(struct proc *, int, struct sigctx *);
124 void postsig_done(struct proc *, int, sigset_t, int);
125 void postsig(struct proc *, int, struct sigctx *);
126 int cansignal(struct proc *, struct process *, int);
127 
128 void ptsignal_locked(struct proc *, int, enum signal_type);
129 
130 struct pool sigacts_pool;	/* memory pool for sigacts structures */
131 
132 void sigio_del(struct sigiolst *);
133 void sigio_unlink(struct sigio_ref *, struct sigiolst *);
134 struct mutex sigio_lock = MUTEX_INITIALIZER(IPL_HIGH);
135 
136 /*
137  * Can thread p, send the signal signum to process qr?
138  */
139 int
140 cansignal(struct proc *p, struct process *qr, int signum)
141 {
142 	struct process *pr = p->p_p;
143 	struct ucred *uc = p->p_ucred;
144 	struct ucred *quc = qr->ps_ucred;
145 
146 	if (uc->cr_uid == 0)
147 		return (1);		/* root can always signal */
148 
149 	if (pr == qr)
150 		return (1);		/* process can always signal itself */
151 
152 	/* optimization: if the same creds then the tests below will pass */
153 	if (uc == quc)
154 		return (1);
155 
156 	if (signum == SIGCONT && qr->ps_session == pr->ps_session)
157 		return (1);		/* SIGCONT in session */
158 
159 	/*
160 	 * Using kill(), only certain signals can be sent to setugid
161 	 * child processes
162 	 */
163 	if (qr->ps_flags & PS_SUGID) {
164 		switch (signum) {
165 		case 0:
166 		case SIGKILL:
167 		case SIGINT:
168 		case SIGTERM:
169 		case SIGALRM:
170 		case SIGSTOP:
171 		case SIGTTIN:
172 		case SIGTTOU:
173 		case SIGTSTP:
174 		case SIGHUP:
175 		case SIGUSR1:
176 		case SIGUSR2:
177 			if (uc->cr_ruid == quc->cr_ruid ||
178 			    uc->cr_uid == quc->cr_ruid)
179 				return (1);
180 		}
181 		return (0);
182 	}
183 
184 	if (uc->cr_ruid == quc->cr_ruid ||
185 	    uc->cr_ruid == quc->cr_svuid ||
186 	    uc->cr_uid == quc->cr_ruid ||
187 	    uc->cr_uid == quc->cr_svuid)
188 		return (1);
189 	return (0);
190 }
191 
192 /*
193  * Initialize signal-related data structures.
194  */
195 void
196 signal_init(void)
197 {
198 	proc_stop_si = softintr_establish(IPL_SOFTCLOCK, proc_stop_sweep,
199 	    NULL);
200 	if (proc_stop_si == NULL)
201 		panic("signal_init failed to register softintr");
202 
203 	pool_init(&sigacts_pool, sizeof(struct sigacts), 0, IPL_NONE,
204 	    PR_WAITOK, "sigapl", NULL);
205 }
206 
207 /*
208  * Initialize a new sigaltstack structure.
209  */
210 void
211 sigstkinit(struct sigaltstack *ss)
212 {
213 	ss->ss_flags = SS_DISABLE;
214 	ss->ss_size = 0;
215 	ss->ss_sp = NULL;
216 }
217 
218 /*
219  * Create an initial sigacts structure, using the same signal state
220  * as pr.
221  */
222 struct sigacts *
223 sigactsinit(struct process *pr)
224 {
225 	struct sigacts *ps;
226 
227 	ps = pool_get(&sigacts_pool, PR_WAITOK);
228 	memcpy(ps, pr->ps_sigacts, sizeof(struct sigacts));
229 	return (ps);
230 }
231 
232 /*
233  * Release a sigacts structure.
234  */
235 void
236 sigactsfree(struct sigacts *ps)
237 {
238 	pool_put(&sigacts_pool, ps);
239 }
240 
241 int
242 sys_sigaction(struct proc *p, void *v, register_t *retval)
243 {
244 	struct sys_sigaction_args /* {
245 		syscallarg(int) signum;
246 		syscallarg(const struct sigaction *) nsa;
247 		syscallarg(struct sigaction *) osa;
248 	} */ *uap = v;
249 	struct sigaction vec;
250 #ifdef KTRACE
251 	struct sigaction ovec;
252 #endif
253 	struct sigaction *sa;
254 	const struct sigaction *nsa;
255 	struct sigaction *osa;
256 	struct sigacts *ps = p->p_p->ps_sigacts;
257 	int signum;
258 	int bit, error;
259 
260 	signum = SCARG(uap, signum);
261 	nsa = SCARG(uap, nsa);
262 	osa = SCARG(uap, osa);
263 
264 	if (signum <= 0 || signum >= NSIG ||
265 	    (nsa && (signum == SIGKILL || signum == SIGSTOP)))
266 		return (EINVAL);
267 	sa = &vec;
268 	if (osa) {
269 		mtx_enter(&p->p_p->ps_mtx);
270 		sa->sa_handler = ps->ps_sigact[signum];
271 		sa->sa_mask = ps->ps_catchmask[signum];
272 		bit = sigmask(signum);
273 		sa->sa_flags = 0;
274 		if ((ps->ps_sigonstack & bit) != 0)
275 			sa->sa_flags |= SA_ONSTACK;
276 		if ((ps->ps_sigintr & bit) == 0)
277 			sa->sa_flags |= SA_RESTART;
278 		if ((ps->ps_sigreset & bit) != 0)
279 			sa->sa_flags |= SA_RESETHAND;
280 		if ((ps->ps_siginfo & bit) != 0)
281 			sa->sa_flags |= SA_SIGINFO;
282 		if (signum == SIGCHLD) {
283 			if ((ps->ps_sigflags & SAS_NOCLDSTOP) != 0)
284 				sa->sa_flags |= SA_NOCLDSTOP;
285 			if ((ps->ps_sigflags & SAS_NOCLDWAIT) != 0)
286 				sa->sa_flags |= SA_NOCLDWAIT;
287 		}
288 		mtx_leave(&p->p_p->ps_mtx);
289 		if ((sa->sa_mask & bit) == 0)
290 			sa->sa_flags |= SA_NODEFER;
291 		sa->sa_mask &= ~bit;
292 		error = copyout(sa, osa, sizeof (vec));
293 		if (error)
294 			return (error);
295 #ifdef KTRACE
296 		if (KTRPOINT(p, KTR_STRUCT))
297 			ovec = vec;
298 #endif
299 	}
300 	if (nsa) {
301 		error = copyin(nsa, sa, sizeof (vec));
302 		if (error)
303 			return (error);
304 #ifdef KTRACE
305 		if (KTRPOINT(p, KTR_STRUCT))
306 			ktrsigaction(p, sa);
307 #endif
308 		setsigvec(p, signum, sa);
309 	}
310 #ifdef KTRACE
311 	if (osa && KTRPOINT(p, KTR_STRUCT))
312 		ktrsigaction(p, &ovec);
313 #endif
314 	return (0);
315 }
316 
317 void
318 setsigvec(struct proc *p, int signum, struct sigaction *sa)
319 {
320 	struct sigacts *ps = p->p_p->ps_sigacts;
321 	int bit;
322 
323 	bit = sigmask(signum);
324 
325 	mtx_enter(&p->p_p->ps_mtx);
326 	ps->ps_sigact[signum] = sa->sa_handler;
327 	if ((sa->sa_flags & SA_NODEFER) == 0)
328 		sa->sa_mask |= sigmask(signum);
329 	ps->ps_catchmask[signum] = sa->sa_mask &~ sigcantmask;
330 	if (signum == SIGCHLD) {
331 		if (sa->sa_flags & SA_NOCLDSTOP)
332 			atomic_setbits_int(&ps->ps_sigflags, SAS_NOCLDSTOP);
333 		else
334 			atomic_clearbits_int(&ps->ps_sigflags, SAS_NOCLDSTOP);
335 		/*
336 		 * If the SA_NOCLDWAIT flag is set or the handler
337 		 * is SIG_IGN we reparent the dying child to PID 1
338 		 * (init) which will reap the zombie.  Because we use
339 		 * init to do our dirty work we never set SAS_NOCLDWAIT
340 		 * for PID 1.
341 		 * XXX exit1 rework means this is unnecessary?
342 		 */
343 		if (initprocess->ps_sigacts != ps &&
344 		    ((sa->sa_flags & SA_NOCLDWAIT) ||
345 		    sa->sa_handler == SIG_IGN))
346 			atomic_setbits_int(&ps->ps_sigflags, SAS_NOCLDWAIT);
347 		else
348 			atomic_clearbits_int(&ps->ps_sigflags, SAS_NOCLDWAIT);
349 	}
350 	if ((sa->sa_flags & SA_RESETHAND) != 0)
351 		ps->ps_sigreset |= bit;
352 	else
353 		ps->ps_sigreset &= ~bit;
354 	if ((sa->sa_flags & SA_SIGINFO) != 0)
355 		ps->ps_siginfo |= bit;
356 	else
357 		ps->ps_siginfo &= ~bit;
358 	if ((sa->sa_flags & SA_RESTART) == 0)
359 		ps->ps_sigintr |= bit;
360 	else
361 		ps->ps_sigintr &= ~bit;
362 	if ((sa->sa_flags & SA_ONSTACK) != 0)
363 		ps->ps_sigonstack |= bit;
364 	else
365 		ps->ps_sigonstack &= ~bit;
366 	/*
367 	 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
368 	 * and for signals set to SIG_DFL where the default is to ignore.
369 	 * However, don't put SIGCONT in ps_sigignore,
370 	 * as we have to restart the process.
371 	 */
372 	if (sa->sa_handler == SIG_IGN ||
373 	    (sigprop[signum] & SA_IGNORE && sa->sa_handler == SIG_DFL)) {
374 		atomic_clearbits_int(&p->p_siglist, bit);
375 		atomic_clearbits_int(&p->p_p->ps_siglist, bit);
376 		if (signum != SIGCONT)
377 			ps->ps_sigignore |= bit;	/* easier in psignal */
378 		ps->ps_sigcatch &= ~bit;
379 	} else {
380 		ps->ps_sigignore &= ~bit;
381 		if (sa->sa_handler == SIG_DFL)
382 			ps->ps_sigcatch &= ~bit;
383 		else
384 			ps->ps_sigcatch |= bit;
385 	}
386 	mtx_leave(&p->p_p->ps_mtx);
387 }
388 
389 /*
390  * Initialize signal state for process 0;
391  * set to ignore signals that are ignored by default.
392  */
393 void
394 siginit(struct sigacts *ps)
395 {
396 	int i;
397 
398 	for (i = 0; i < NSIG; i++)
399 		if (sigprop[i] & SA_IGNORE && i != SIGCONT)
400 			ps->ps_sigignore |= sigmask(i);
401 	ps->ps_sigflags = SAS_NOCLDWAIT | SAS_NOCLDSTOP;
402 }
403 
404 /*
405  * Reset signals for an exec by the specified thread.
406  */
407 void
408 execsigs(struct proc *p)
409 {
410 	struct sigacts *ps;
411 	int nc, mask;
412 
413 	ps = p->p_p->ps_sigacts;
414 	mtx_enter(&p->p_p->ps_mtx);
415 
416 	/*
417 	 * Reset caught signals.  Held signals remain held
418 	 * through p_sigmask (unless they were caught,
419 	 * and are now ignored by default).
420 	 */
421 	while (ps->ps_sigcatch) {
422 		nc = ffs((long)ps->ps_sigcatch);
423 		mask = sigmask(nc);
424 		ps->ps_sigcatch &= ~mask;
425 		if (sigprop[nc] & SA_IGNORE) {
426 			if (nc != SIGCONT)
427 				ps->ps_sigignore |= mask;
428 			atomic_clearbits_int(&p->p_siglist, mask);
429 			atomic_clearbits_int(&p->p_p->ps_siglist, mask);
430 		}
431 		ps->ps_sigact[nc] = SIG_DFL;
432 	}
433 	/*
434 	 * Reset stack state to the user stack.
435 	 * Clear set of signals caught on the signal stack.
436 	 */
437 	sigstkinit(&p->p_sigstk);
438 	atomic_clearbits_int(&ps->ps_sigflags, SAS_NOCLDWAIT);
439 	if (ps->ps_sigact[SIGCHLD] == SIG_IGN)
440 		ps->ps_sigact[SIGCHLD] = SIG_DFL;
441 	mtx_leave(&p->p_p->ps_mtx);
442 }
443 
444 /*
445  * Manipulate signal mask.
446  * Note that we receive new mask, not pointer,
447  * and return old mask as return value;
448  * the library stub does the rest.
449  */
450 int
451 sys_sigprocmask(struct proc *p, void *v, register_t *retval)
452 {
453 	struct sys_sigprocmask_args /* {
454 		syscallarg(int) how;
455 		syscallarg(sigset_t) mask;
456 	} */ *uap = v;
457 	int error = 0;
458 	sigset_t mask;
459 
460 	KASSERT(p == curproc);
461 
462 	*retval = p->p_sigmask;
463 	mask = SCARG(uap, mask) &~ sigcantmask;
464 
465 	switch (SCARG(uap, how)) {
466 	case SIG_BLOCK:
467 		SET(p->p_sigmask, mask);
468 		break;
469 	case SIG_UNBLOCK:
470 		CLR(p->p_sigmask, mask);
471 		break;
472 	case SIG_SETMASK:
473 		p->p_sigmask = mask;
474 		break;
475 	default:
476 		error = EINVAL;
477 		break;
478 	}
479 	return (error);
480 }
481 
482 int
483 sys_sigpending(struct proc *p, void *v, register_t *retval)
484 {
485 	*retval = p->p_siglist | p->p_p->ps_siglist;
486 	return (0);
487 }
488 
489 /*
490  * Temporarily replace calling proc's signal mask for the duration of a
491  * system call.  Original signal mask will be restored by userret().
492  */
493 void
494 dosigsuspend(struct proc *p, sigset_t newmask)
495 {
496 	KASSERT(p == curproc);
497 
498 	p->p_oldmask = p->p_sigmask;
499 	p->p_sigmask = newmask;
500 	atomic_setbits_int(&p->p_flag, P_SIGSUSPEND);
501 }
502 
503 /*
504  * Suspend thread until signal, providing mask to be set
505  * in the meantime.  Note nonstandard calling convention:
506  * libc stub passes mask, not pointer, to save a copyin.
507  */
508 int
509 sys_sigsuspend(struct proc *p, void *v, register_t *retval)
510 {
511 	struct sys_sigsuspend_args /* {
512 		syscallarg(int) mask;
513 	} */ *uap = v;
514 
515 	dosigsuspend(p, SCARG(uap, mask) &~ sigcantmask);
516 	while (tsleep_nsec(&nowake, PPAUSE|PCATCH, "sigsusp", INFSLP) == 0)
517 		continue;
518 	/* always return EINTR rather than ERESTART... */
519 	return (EINTR);
520 }
521 
522 int
523 sigonstack(size_t stack)
524 {
525 	const struct sigaltstack *ss = &curproc->p_sigstk;
526 
527 	return (ss->ss_flags & SS_DISABLE ? 0 :
528 	    (stack - (size_t)ss->ss_sp < ss->ss_size));
529 }
530 
531 int
532 sys_sigaltstack(struct proc *p, void *v, register_t *retval)
533 {
534 	struct sys_sigaltstack_args /* {
535 		syscallarg(const struct sigaltstack *) nss;
536 		syscallarg(struct sigaltstack *) oss;
537 	} */ *uap = v;
538 	struct sigaltstack ss;
539 	const struct sigaltstack *nss;
540 	struct sigaltstack *oss;
541 	int onstack = sigonstack(PROC_STACK(p));
542 	int error;
543 
544 	nss = SCARG(uap, nss);
545 	oss = SCARG(uap, oss);
546 
547 	if (oss != NULL) {
548 		ss = p->p_sigstk;
549 		if (onstack)
550 			ss.ss_flags |= SS_ONSTACK;
551 		if ((error = copyout(&ss, oss, sizeof(ss))))
552 			return (error);
553 	}
554 	if (nss == NULL)
555 		return (0);
556 	error = copyin(nss, &ss, sizeof(ss));
557 	if (error)
558 		return (error);
559 	if (onstack)
560 		return (EPERM);
561 	if (ss.ss_flags & ~SS_DISABLE)
562 		return (EINVAL);
563 	if (ss.ss_flags & SS_DISABLE) {
564 		p->p_sigstk.ss_flags = ss.ss_flags;
565 		return (0);
566 	}
567 	if (ss.ss_size < MINSIGSTKSZ)
568 		return (ENOMEM);
569 
570 	error = uvm_map_remap_as_stack(p, (vaddr_t)ss.ss_sp, ss.ss_size);
571 	if (error)
572 		return (error);
573 
574 	p->p_sigstk = ss;
575 	return (0);
576 }
577 
578 int
579 sys_kill(struct proc *cp, void *v, register_t *retval)
580 {
581 	struct sys_kill_args /* {
582 		syscallarg(int) pid;
583 		syscallarg(int) signum;
584 	} */ *uap = v;
585 	struct process *pr;
586 	int pid = SCARG(uap, pid);
587 	int signum = SCARG(uap, signum);
588 	int error;
589 	int zombie = 0;
590 
591 	if ((error = pledge_kill(cp, pid)) != 0)
592 		return (error);
593 	if (((u_int)signum) >= NSIG)
594 		return (EINVAL);
595 	if (pid > 0) {
596 		if ((pr = prfind(pid)) == NULL) {
597 			if ((pr = zombiefind(pid)) == NULL)
598 				return (ESRCH);
599 			else
600 				zombie = 1;
601 		}
602 		if (!cansignal(cp, pr, signum))
603 			return (EPERM);
604 
605 		/* kill single process */
606 		if (signum && !zombie)
607 			prsignal(pr, signum);
608 		return (0);
609 	}
610 	switch (pid) {
611 	case -1:		/* broadcast signal */
612 		return (killpg1(cp, signum, 0, 1));
613 	case 0:			/* signal own process group */
614 		return (killpg1(cp, signum, 0, 0));
615 	default:		/* negative explicit process group */
616 		return (killpg1(cp, signum, -pid, 0));
617 	}
618 }
619 
620 int
621 sys_thrkill(struct proc *cp, void *v, register_t *retval)
622 {
623 	struct sys_thrkill_args /* {
624 		syscallarg(pid_t) tid;
625 		syscallarg(int) signum;
626 		syscallarg(void *) tcb;
627 	} */ *uap = v;
628 	struct proc *p;
629 	int tid = SCARG(uap, tid);
630 	int signum = SCARG(uap, signum);
631 	void *tcb;
632 
633 	if (((u_int)signum) >= NSIG)
634 		return (EINVAL);
635 
636 	p = tid ? tfind_user(tid, cp->p_p) : cp;
637 	if (p == NULL)
638 		return (ESRCH);
639 
640 	/* optionally require the target thread to have the given tcb addr */
641 	tcb = SCARG(uap, tcb);
642 	if (tcb != NULL && tcb != TCB_GET(p))
643 		return (ESRCH);
644 
645 	if (signum)
646 		ptsignal(p, signum, STHREAD);
647 	return (0);
648 }
649 
650 /*
651  * Common code for kill process group/broadcast kill.
652  * cp is calling process.
653  */
654 int
655 killpg1(struct proc *cp, int signum, int pgid, int all)
656 {
657 	struct process *pr;
658 	struct pgrp *pgrp;
659 	int nfound = 0;
660 
661 	if (all) {
662 		/*
663 		 * broadcast
664 		 */
665 		LIST_FOREACH(pr, &allprocess, ps_list) {
666 			if (pr->ps_pid <= 1 ||
667 			    pr->ps_flags & (PS_SYSTEM | PS_NOBROADCASTKILL) ||
668 			    pr == cp->p_p || !cansignal(cp, pr, signum))
669 				continue;
670 			nfound++;
671 			if (signum)
672 				prsignal(pr, signum);
673 		}
674 	} else {
675 		if (pgid == 0)
676 			/*
677 			 * zero pgid means send to my process group.
678 			 */
679 			pgrp = cp->p_p->ps_pgrp;
680 		else {
681 			pgrp = pgfind(pgid);
682 			if (pgrp == NULL)
683 				return (ESRCH);
684 		}
685 		LIST_FOREACH(pr, &pgrp->pg_members, ps_pglist) {
686 			if (pr->ps_pid <= 1 || pr->ps_flags & PS_SYSTEM ||
687 			    !cansignal(cp, pr, signum))
688 				continue;
689 			nfound++;
690 			if (signum)
691 				prsignal(pr, signum);
692 		}
693 	}
694 	return (nfound ? 0 : ESRCH);
695 }
696 
697 #define CANDELIVER(uid, euid, pr) \
698 	(euid == 0 || \
699 	(uid) == (pr)->ps_ucred->cr_ruid || \
700 	(uid) == (pr)->ps_ucred->cr_svuid || \
701 	(uid) == (pr)->ps_ucred->cr_uid || \
702 	(euid) == (pr)->ps_ucred->cr_ruid || \
703 	(euid) == (pr)->ps_ucred->cr_svuid || \
704 	(euid) == (pr)->ps_ucred->cr_uid)
705 
706 #define CANSIGIO(cr, pr) \
707 	CANDELIVER((cr)->cr_ruid, (cr)->cr_uid, (pr))
708 
709 /*
710  * Send a signal to a process group.  If checktty is 1,
711  * limit to members which have a controlling terminal.
712  */
713 void
714 pgsignal(struct pgrp *pgrp, int signum, int checkctty)
715 {
716 	struct process *pr;
717 
718 	if (pgrp)
719 		LIST_FOREACH(pr, &pgrp->pg_members, ps_pglist)
720 			if (checkctty == 0 || pr->ps_flags & PS_CONTROLT)
721 				prsignal(pr, signum);
722 }
723 
724 /*
725  * Send a SIGIO or SIGURG signal to a process or process group using stored
726  * credentials rather than those of the current process.
727  */
728 void
729 pgsigio(struct sigio_ref *sir, int sig, int checkctty)
730 {
731 	struct process *pr;
732 	struct sigio *sigio;
733 
734 	if (sir->sir_sigio == NULL)
735 		return;
736 
737 	KERNEL_LOCK();
738 	mtx_enter(&sigio_lock);
739 	sigio = sir->sir_sigio;
740 	if (sigio == NULL)
741 		goto out;
742 	if (sigio->sio_pgid > 0) {
743 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc))
744 			prsignal(sigio->sio_proc, sig);
745 	} else if (sigio->sio_pgid < 0) {
746 		LIST_FOREACH(pr, &sigio->sio_pgrp->pg_members, ps_pglist) {
747 			if (CANSIGIO(sigio->sio_ucred, pr) &&
748 			    (checkctty == 0 || (pr->ps_flags & PS_CONTROLT)))
749 				prsignal(pr, sig);
750 		}
751 	}
752 out:
753 	mtx_leave(&sigio_lock);
754 	KERNEL_UNLOCK();
755 }
756 
757 /*
758  * Recalculate the signal mask and reset the signal disposition after
759  * usermode frame for delivery is formed.
760  */
761 void
762 postsig_done(struct proc *p, int signum, sigset_t catchmask, int reset)
763 {
764 	p->p_ru.ru_nsignals++;
765 	SET(p->p_sigmask, catchmask);
766 	if (reset != 0) {
767 		sigset_t mask = sigmask(signum);
768 		struct sigacts *ps = p->p_p->ps_sigacts;
769 
770 		mtx_enter(&p->p_p->ps_mtx);
771 		ps->ps_sigcatch &= ~mask;
772 		if (signum != SIGCONT && sigprop[signum] & SA_IGNORE)
773 			ps->ps_sigignore |= mask;
774 		ps->ps_sigact[signum] = SIG_DFL;
775 		mtx_leave(&p->p_p->ps_mtx);
776 	}
777 }
778 
779 /*
780  * Send a signal caused by a trap to the current thread
781  * If it will be caught immediately, deliver it with correct code.
782  * Otherwise, post it normally.
783  */
784 void
785 trapsignal(struct proc *p, int signum, u_long trapno, int code,
786     union sigval sigval)
787 {
788 	struct process *pr = p->p_p;
789 	struct sigctx ctx;
790 	int mask;
791 
792 	switch (signum) {
793 	case SIGILL:
794 		if (code == ILL_BTCFI) {
795 			pr->ps_acflag |= ABTCFI;
796 			break;
797 		}
798 		/* FALLTHROUGH */
799 	case SIGBUS:
800 	case SIGSEGV:
801 		pr->ps_acflag |= ATRAP;
802 		break;
803 	}
804 
805 	mask = sigmask(signum);
806 	setsigctx(p, signum, &ctx);
807 	if ((pr->ps_flags & PS_TRACED) == 0 && ctx.sig_catch != 0 &&
808 	    (p->p_sigmask & mask) == 0) {
809 		siginfo_t si;
810 
811 		initsiginfo(&si, signum, trapno, code, sigval);
812 #ifdef KTRACE
813 		if (KTRPOINT(p, KTR_PSIG)) {
814 			ktrpsig(p, signum, ctx.sig_action,
815 			    p->p_sigmask, code, &si);
816 		}
817 #endif
818 		if (sendsig(ctx.sig_action, signum, p->p_sigmask, &si,
819 		    ctx.sig_info, ctx.sig_onstack)) {
820 			KERNEL_LOCK();
821 			sigexit(p, SIGILL);
822 			/* NOTREACHED */
823 		}
824 		postsig_done(p, signum, ctx.sig_catchmask, ctx.sig_reset);
825 	} else {
826 		p->p_sisig = signum;
827 		p->p_sitrapno = trapno;	/* XXX for core dump/debugger */
828 		p->p_sicode = code;
829 		p->p_sigval = sigval;
830 
831 		/*
832 		 * If traced, stop if signal is masked, and stay stopped
833 		 * until released by the debugger.  If our parent process
834 		 * is waiting for us, don't hang as we could deadlock.
835 		 */
836 		if (((pr->ps_flags & (PS_TRACED | PS_PPWAIT)) == PS_TRACED) &&
837 		    signum != SIGKILL && (p->p_sigmask & mask) != 0) {
838 			signum = proc_trap(p, signum);
839 
840 			mask = sigmask(signum);
841 			setsigctx(p, signum, &ctx);
842 
843 			/*
844 			 * If we are no longer being traced, or the parent
845 			 * didn't give us a signal, skip sending the signal.
846 			 */
847 			if ((pr->ps_flags & PS_TRACED) == 0 || signum == 0)
848 				return;
849 
850 			/* update signal info */
851 			p->p_sisig = signum;
852 		}
853 
854 		/*
855 		 * Signals like SIGBUS and SIGSEGV should not, when
856 		 * generated by the kernel, be ignorable or blockable.
857 		 * If it is and we're not being traced, then just kill
858 		 * the process.
859 		 * After vfs_shutdown(9), init(8) cannot receive signals
860 		 * because new code pages of the signal handler cannot be
861 		 * mapped from halted storage.  init(8) may not die or the
862 		 * kernel panics.  Better loop between signal handler and
863 		 * page fault trap until the machine is halted.
864 		 */
865 		if ((pr->ps_flags & PS_TRACED) == 0 &&
866 		    (sigprop[signum] & SA_KILL) &&
867 		    ((p->p_sigmask & mask) || ctx.sig_ignore) &&
868 		    pr->ps_pid != 1) {
869 			KERNEL_LOCK();
870 			sigexit(p, signum);
871 			/* NOTREACHED */
872 		}
873 		ptsignal(p, signum, STHREAD);
874 	}
875 }
876 
877 /*
878  * Send the signal to the process.  If the signal has an action, the action
879  * is usually performed by the target process rather than the caller; we add
880  * the signal to the set of pending signals for the process.
881  *
882  * Exceptions:
883  *   o When a stop signal is sent to a sleeping process that takes the
884  *     default action, the process is stopped without awakening it.
885  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
886  *     regardless of the signal action (eg, blocked or ignored).
887  *
888  * Other ignored signals are discarded immediately.
889  */
890 void
891 psignal(struct proc *p, int signum)
892 {
893 	ptsignal(p, signum, SPROCESS);
894 }
895 
896 void
897 prsignal(struct process *pr, int signum)
898 {
899 	mtx_enter(&pr->ps_mtx);
900 	/* Ignore signal if the target process is exiting */
901 	if (pr->ps_flags & PS_EXITING) {
902 		mtx_leave(&pr->ps_mtx);
903 		return;
904 	}
905 	ptsignal_locked(TAILQ_FIRST(&pr->ps_threads), signum, SPROCESS);
906 	mtx_leave(&pr->ps_mtx);
907 }
908 
909 /*
910  * type = SPROCESS	process signal, can be diverted (sigwait())
911  * type = STHREAD	thread signal, but should be propagated if unhandled
912  * type = SPROPAGATED	propagated to this thread, so don't propagate again
913  */
914 void
915 ptsignal(struct proc *p, int signum, enum signal_type type)
916 {
917 	struct process *pr = p->p_p;
918 
919 	mtx_enter(&pr->ps_mtx);
920 	ptsignal_locked(p, signum, type);
921 	mtx_leave(&pr->ps_mtx);
922 }
923 
924 void
925 ptsignal_locked(struct proc *p, int signum, enum signal_type type)
926 {
927 	int prop;
928 	sig_t action, altaction = SIG_DFL;
929 	sigset_t mask, sigmask;
930 	int *siglist;
931 	struct process *pr = p->p_p;
932 	struct proc *q;
933 	int wakeparent = 0;
934 
935 	MUTEX_ASSERT_LOCKED(&pr->ps_mtx);
936 
937 #ifdef DIAGNOSTIC
938 	if ((u_int)signum >= NSIG || signum == 0)
939 		panic("psignal signal number");
940 #endif
941 
942 	/* Ignore signal if the target process is exiting */
943 	if (pr->ps_flags & PS_EXITING)
944 		return;
945 
946 	mask = sigmask(signum);
947 	sigmask = READ_ONCE(p->p_sigmask);
948 
949 	if (type == SPROCESS) {
950 		sigset_t tmpmask;
951 
952 		/* Accept SIGKILL to coredumping processes */
953 		if (pr->ps_flags & PS_COREDUMP && signum == SIGKILL) {
954 			atomic_setbits_int(&pr->ps_siglist, mask);
955 			return;
956 		}
957 
958 		/*
959 		 * If the current thread can process the signal
960 		 * immediately (it's unblocked) then have it take it.
961 		 */
962 		q = curproc;
963 		tmpmask = READ_ONCE(q->p_sigmask);
964 		if (q->p_p == pr && (q->p_flag & P_WEXIT) == 0 &&
965 		    (tmpmask & mask) == 0) {
966 			p = q;
967 			sigmask = tmpmask;
968 		} else {
969 			/*
970 			 * A process-wide signal can be diverted to a
971 			 * different thread that's in sigwait() for this
972 			 * signal.  If there isn't such a thread, then
973 			 * pick a thread that doesn't have it blocked so
974 			 * that the stop/kill consideration isn't
975 			 * delayed.  Otherwise, mark it pending on the
976 			 * main thread.
977 			 */
978 			TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
979 
980 				/* ignore exiting threads */
981 				if (q->p_flag & P_WEXIT)
982 					continue;
983 
984 				/* skip threads that have the signal blocked */
985 				tmpmask = READ_ONCE(q->p_sigmask);
986 				if ((tmpmask & mask) != 0)
987 					continue;
988 
989 				/* okay, could send to this thread */
990 				p = q;
991 				sigmask = tmpmask;
992 
993 				/*
994 				 * sigsuspend, sigwait, ppoll/pselect, etc?
995 				 * Definitely go to this thread, as it's
996 				 * already blocked in the kernel.
997 				 */
998 				if (q->p_flag & P_SIGSUSPEND)
999 					break;
1000 			}
1001 		}
1002 	}
1003 
1004 	if (type != SPROPAGATED)
1005 		knote_locked(&pr->ps_klist, NOTE_SIGNAL | signum);
1006 
1007 	prop = sigprop[signum];
1008 
1009 	/*
1010 	 * If proc is traced, always give parent a chance.
1011 	 */
1012 	if (pr->ps_flags & PS_TRACED) {
1013 		action = SIG_DFL;
1014 	} else {
1015 		sigset_t sigcatch, sigignore;
1016 
1017 		/*
1018 		 * If the signal is being ignored,
1019 		 * then we forget about it immediately.
1020 		 * (Note: we don't set SIGCONT in ps_sigignore,
1021 		 * and if it is set to SIG_IGN,
1022 		 * action will be SIG_DFL here.)
1023 		 */
1024 		sigignore = pr->ps_sigacts->ps_sigignore;
1025 		sigcatch = pr->ps_sigacts->ps_sigcatch;
1026 
1027 		if (sigignore & mask)
1028 			return;
1029 		if (sigmask & mask) {
1030 			action = SIG_HOLD;
1031 			if (sigcatch & mask)
1032 				altaction = SIG_CATCH;
1033 		} else if (sigcatch & mask) {
1034 			action = SIG_CATCH;
1035 		} else {
1036 			action = SIG_DFL;
1037 
1038 			if (prop & SA_KILL && pr->ps_nice > NZERO)
1039 				 pr->ps_nice = NZERO;
1040 
1041 			/*
1042 			 * If sending a tty stop signal to a member of an
1043 			 * orphaned process group, discard the signal here if
1044 			 * the action is default; don't stop the process below
1045 			 * if sleeping, and don't clear any pending SIGCONT.
1046 			 */
1047 			if (prop & SA_TTYSTOP && pr->ps_pgrp->pg_jobc == 0)
1048 				return;
1049 		}
1050 	}
1051 	/*
1052 	 * If delivered to process, mark as pending there.  Continue and stop
1053 	 * signals will be propagated to all threads.  So they are always
1054 	 * marked at thread level.
1055 	 */
1056 	siglist = (type == SPROCESS) ? &pr->ps_siglist : &p->p_siglist;
1057 	if (prop & (SA_CONT | SA_STOP))
1058 		siglist = &p->p_siglist;
1059 
1060 	/*
1061 	 * XXX delay processing of SA_STOP signals unless action == SIG_DFL?
1062 	 */
1063 	if (prop & (SA_CONT | SA_STOP) && type != SPROPAGATED)
1064 		TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link)
1065 			if (q != p)
1066 				ptsignal_locked(q, signum, SPROPAGATED);
1067 
1068 	SCHED_LOCK();
1069 
1070 	switch (p->p_stat) {
1071 
1072 	case SSTOP:
1073 		/*
1074 		 * If traced process is already stopped,
1075 		 * then no further action is necessary.
1076 		 */
1077 		if (pr->ps_flags & PS_TRACED)
1078 			goto out;
1079 
1080 		/*
1081 		 * Kill signal always sets processes running.
1082 		 */
1083 		if (signum == SIGKILL) {
1084 			atomic_clearbits_int(&p->p_flag, P_SUSPSIG);
1085 			/* Raise priority to at least PUSER. */
1086 			if (p->p_usrpri > PUSER)
1087 				p->p_usrpri = PUSER;
1088 			unsleep(p);
1089 			setrunnable(p);
1090 			goto out;
1091 		}
1092 
1093 		if (prop & SA_CONT) {
1094 			/*
1095 			 * If SIGCONT is default (or ignored), we continue the
1096 			 * process but don't leave the signal in p_siglist, as
1097 			 * it has no further action.  If SIGCONT is held, we
1098 			 * continue the process and leave the signal in
1099 			 * p_siglist.  If the process catches SIGCONT, let it
1100 			 * handle the signal itself.  If it isn't waiting on
1101 			 * an event, then it goes back to run state.
1102 			 * Otherwise, process goes back to sleep state.
1103 			 */
1104 			atomic_setbits_int(&pr->ps_flags, PS_CONTINUED);
1105 			atomic_clearbits_int(&pr->ps_flags,
1106 			    PS_WAITED | PS_STOPPED);
1107 			atomic_clearbits_int(&p->p_flag, P_SUSPSIG);
1108 			wakeparent = 1;
1109 			if (action == SIG_DFL)
1110 				mask = 0;
1111 			if (action == SIG_CATCH) {
1112 				/* Raise priority to at least PUSER. */
1113 				if (p->p_usrpri > PUSER)
1114 					p->p_usrpri = PUSER;
1115 				unsleep(p);
1116 				setrunnable(p);
1117 				goto out;
1118 			}
1119 			if (p->p_wchan == NULL) {
1120 				setrunnable(p);
1121 				goto out;
1122 			}
1123 			atomic_clearbits_int(&p->p_flag, P_WSLEEP);
1124 			p->p_stat = SSLEEP;
1125 			goto out;
1126 		}
1127 
1128 		/*
1129 		 * Defer further processing for signals which are held,
1130 		 * except that stopped processes must be continued by SIGCONT.
1131 		 */
1132 		if (action == SIG_HOLD)
1133 			goto out;
1134 
1135 		if (prop & SA_STOP) {
1136 			/*
1137 			 * Already stopped, don't need to stop again.
1138 			 * (If we did the shell could get confused.)
1139 			 */
1140 			mask = 0;
1141 			goto out;
1142 		}
1143 
1144 		/*
1145 		 * If process is sleeping interruptibly, then simulate a
1146 		 * wakeup so that when it is continued, it will be made
1147 		 * runnable and can look at the signal.  But don't make
1148 		 * the process runnable, leave it stopped.
1149 		 */
1150 		if (p->p_flag & P_SINTR)
1151 			unsleep(p);
1152 		goto out;
1153 
1154 	case SSLEEP:
1155 		/*
1156 		 * If process is sleeping uninterruptibly
1157 		 * we can't interrupt the sleep... the signal will
1158 		 * be noticed when the process returns through
1159 		 * trap() or syscall().
1160 		 */
1161 		if ((p->p_flag & P_SINTR) == 0)
1162 			goto out;
1163 		/*
1164 		 * Process is sleeping and traced... make it runnable
1165 		 * so it can discover the signal in cursig() and stop
1166 		 * for the parent.
1167 		 */
1168 		if (pr->ps_flags & PS_TRACED) {
1169 			unsleep(p);
1170 			setrunnable(p);
1171 			goto out;
1172 		}
1173 
1174 		/*
1175 		 * Recheck sigmask before waking up the process,
1176 		 * there is a chance that while sending the signal
1177 		 * the process changed sigmask and went to sleep.
1178 		 */
1179 		sigmask = READ_ONCE(p->p_sigmask);
1180 		if (sigmask & mask)
1181 			goto out;
1182 		else if (action == SIG_HOLD) {
1183 			/* signal got unmasked, get proper action */
1184 			action = altaction;
1185 
1186 			if (action == SIG_DFL) {
1187 				if (prop & SA_KILL && pr->ps_nice > NZERO)
1188 					 pr->ps_nice = NZERO;
1189 
1190 				/*
1191 				 * Discard tty stop signals sent to an
1192 				 * orphaned process group, see above.
1193 				 */
1194 				if (prop & SA_TTYSTOP &&
1195 				    pr->ps_pgrp->pg_jobc == 0) {
1196 					mask = 0;
1197 					prop = 0;
1198 					goto out;
1199 				}
1200 			}
1201 		}
1202 
1203 		/*
1204 		 * If SIGCONT is default (or ignored) and process is
1205 		 * asleep, we are finished; the process should not
1206 		 * be awakened.
1207 		 */
1208 		if ((prop & SA_CONT) && action == SIG_DFL) {
1209 			mask = 0;
1210 			goto out;
1211 		}
1212 		/*
1213 		 * When a sleeping process receives a stop
1214 		 * signal, process immediately if possible.
1215 		 */
1216 		if ((prop & SA_STOP) && action == SIG_DFL) {
1217 			/*
1218 			 * If a child holding parent blocked,
1219 			 * stopping could cause deadlock.
1220 			 */
1221 			if (pr->ps_flags & PS_PPWAIT)
1222 				goto out;
1223 			mask = 0;
1224 			pr->ps_xsig = signum;
1225 			proc_stop(p, 0);
1226 			goto out;
1227 		}
1228 		/*
1229 		 * All other (caught or default) signals
1230 		 * cause the process to run.
1231 		 * Raise priority to at least PUSER.
1232 		 */
1233 		if (p->p_usrpri > PUSER)
1234 			p->p_usrpri = PUSER;
1235 		unsleep(p);
1236 		setrunnable(p);
1237 		goto out;
1238 		/* NOTREACHED */
1239 
1240 	case SONPROC:
1241 		if (action == SIG_HOLD)
1242 			goto out;
1243 
1244 		/* set siglist before issuing the ast */
1245 		atomic_setbits_int(siglist, mask);
1246 		mask = 0;
1247 		signotify(p);
1248 		/* FALLTHROUGH */
1249 	default:
1250 		/*
1251 		 * SRUN, SIDL, SDEAD do nothing with the signal,
1252 		 * other than kicking ourselves if we are running.
1253 		 * It will either never be noticed, or noticed very soon.
1254 		 */
1255 		goto out;
1256 	}
1257 	/* NOTREACHED */
1258 
1259 out:
1260 	/* finally adjust siglist */
1261 	if (mask)
1262 		atomic_setbits_int(siglist, mask);
1263 	if (prop & SA_CONT) {
1264 		atomic_clearbits_int(siglist, STOPSIGMASK);
1265 	}
1266 	if (prop & SA_STOP) {
1267 		atomic_clearbits_int(siglist, CONTSIGMASK);
1268 		atomic_clearbits_int(&pr->ps_flags, PS_CONTINUED);
1269 	}
1270 
1271 	SCHED_UNLOCK();
1272 	if (wakeparent)
1273 		wakeup(pr->ps_pptr);
1274 }
1275 
1276 /* fill the signal context which should be used by postsig() and issignal() */
1277 void
1278 setsigctx(struct proc *p, int signum, struct sigctx *sctx)
1279 {
1280 	struct process *pr = p->p_p;
1281 	struct sigacts *ps = pr->ps_sigacts;
1282 	sigset_t mask;
1283 
1284 	mtx_enter(&pr->ps_mtx);
1285 	mask = sigmask(signum);
1286 	sctx->sig_action = ps->ps_sigact[signum];
1287 	sctx->sig_catchmask = ps->ps_catchmask[signum];
1288 	sctx->sig_reset = (ps->ps_sigreset & mask) != 0;
1289 	sctx->sig_info = (ps->ps_siginfo & mask) != 0;
1290 	sctx->sig_intr = (ps->ps_sigintr & mask) != 0;
1291 	sctx->sig_onstack = (ps->ps_sigonstack & mask) != 0;
1292 	sctx->sig_ignore = (ps->ps_sigignore & mask) != 0;
1293 	sctx->sig_catch = (ps->ps_sigcatch & mask) != 0;
1294 	sctx->sig_stop = sigprop[signum] & SA_STOP &&
1295 	    (long)sctx->sig_action == (long)SIG_DFL;
1296 	if (sctx->sig_stop) {
1297 		/*
1298 		 * If the process is a member of an orphaned
1299 		 * process group, ignore tty stop signals.
1300 		 */
1301 		if (pr->ps_flags & PS_TRACED ||
1302 		    (pr->ps_pgrp->pg_jobc == 0 &&
1303 		    sigprop[signum] & SA_TTYSTOP)) {
1304 			sctx->sig_stop = 0;
1305 			sctx->sig_ignore = 1;
1306 		}
1307 	}
1308 	mtx_leave(&pr->ps_mtx);
1309 }
1310 
1311 /*
1312  * Determine signal that should be delivered to process p, the current
1313  * process, 0 if none.
1314  *
1315  * If the current process has received a signal (should be caught or cause
1316  * termination, should interrupt current syscall), return the signal number.
1317  * Stop signals with default action are processed immediately, then cleared;
1318  * they aren't returned.  This is checked after each entry to the system for
1319  * a syscall or trap. The normal call sequence is
1320  *
1321  *	while (signum = cursig(curproc, &ctx, 0))
1322  *		postsig(signum, &ctx);
1323  *
1324  * Assumes that if the P_SINTR flag is set, we're holding both the
1325  * kernel and scheduler locks.
1326  */
1327 int
1328 cursig(struct proc *p, struct sigctx *sctx, int deep)
1329 {
1330 	struct process *pr = p->p_p;
1331 	int signum, mask, keep = 0, prop;
1332 	sigset_t ps_siglist;
1333 
1334 	KASSERT(p == curproc);
1335 
1336 	for (;;) {
1337 		ps_siglist = READ_ONCE(pr->ps_siglist);
1338 		membar_consumer();
1339 		mask = SIGPENDING(p);
1340 		if (pr->ps_flags & PS_PPWAIT)
1341 			mask &= ~STOPSIGMASK;
1342 		signum = ffs(mask);
1343 		if (signum == 0)	 	/* no signal to send */
1344 			goto keep;
1345 		mask = sigmask(signum);
1346 
1347 		/* take the signal! */
1348 		if (atomic_cas_uint(&pr->ps_siglist, ps_siglist,
1349 		    ps_siglist & ~mask) != ps_siglist) {
1350 			/* lost race taking the process signal, restart */
1351 			continue;
1352 		}
1353 		atomic_clearbits_int(&p->p_siglist, mask);
1354 		setsigctx(p, signum, sctx);
1355 
1356 		/*
1357 		 * We should see pending but ignored signals
1358 		 * only if PS_TRACED was on when they were posted.
1359 		 */
1360 		if (sctx->sig_ignore && (pr->ps_flags & PS_TRACED) == 0)
1361 			continue;
1362 
1363 		/*
1364 		 * If cursig is called while going to sleep, abort now
1365 		 * and stop the sleep. When the call unwinded to userret
1366 		 * cursig is called again and there the signal can be
1367 		 * handled cleanly.
1368 		 */
1369 		if (deep) {
1370 			/*
1371 			 * Do not stop the thread here if multiple
1372 			 * signals are pending and at least one of
1373 			 * them would force an unwind.
1374 			 *
1375 			 * ffs() favors low numbered signals and
1376 			 * so stop signals may be picked up before
1377 			 * other pending signals.
1378 			 */
1379 			if (sctx->sig_stop && SIGPENDING(p)) {
1380 				keep |= mask;
1381 				continue;
1382 			}
1383 			goto keep;
1384 		}
1385 
1386 		/*
1387 		 * If traced, always stop, and stay stopped until released
1388 		 * by the debugger.  If our parent process is waiting for
1389 		 * us, don't hang as we could deadlock.
1390 		 */
1391 		if (((pr->ps_flags & (PS_TRACED | PS_PPWAIT)) == PS_TRACED) &&
1392 		    signum != SIGKILL) {
1393 			signum = proc_trap(p, signum);
1394 
1395 			mask = sigmask(signum);
1396 			setsigctx(p, signum, sctx);
1397 
1398 			/*
1399 			 * If we are no longer being traced, or the parent
1400 			 * didn't give us a signal, or the signal is ignored,
1401 			 * look for more signals.
1402 			 */
1403 			if ((pr->ps_flags & PS_TRACED) == 0 || signum == 0 ||
1404 			    sctx->sig_ignore)
1405 				continue;
1406 
1407 			/*
1408 			 * If the new signal is being masked, look for other
1409 			 * signals but leave it for later.
1410 			 */
1411 			if ((p->p_sigmask & mask) != 0) {
1412 				atomic_setbits_int(&p->p_siglist, mask);
1413 				continue;
1414 			}
1415 
1416 		}
1417 
1418 		prop = sigprop[signum];
1419 
1420 		/*
1421 		 * Decide whether the signal should be returned.
1422 		 * Return the signal's number, or fall through
1423 		 * to clear it from the pending mask.
1424 		 */
1425 		switch ((long)sctx->sig_action) {
1426 		case (long)SIG_DFL:
1427 			/*
1428 			 * Don't take default actions on system processes.
1429 			 */
1430 			if (pr->ps_pid <= 1) {
1431 #ifdef DIAGNOSTIC
1432 				/*
1433 				 * Are you sure you want to ignore SIGSEGV
1434 				 * in init? XXX
1435 				 */
1436 				printf("Process (pid %d) got signal"
1437 				    " %d\n", pr->ps_pid, signum);
1438 #endif
1439 				break;		/* == ignore */
1440 			}
1441 			/*
1442 			 * If there is a pending stop signal to process
1443 			 * with default action, stop here,
1444 			 * then clear the signal.
1445 			 */
1446 			if (sctx->sig_stop) {
1447 				pr->ps_xsig = signum;
1448 				SCHED_LOCK();
1449 				proc_stop(p, 1);
1450 				SCHED_UNLOCK();
1451 				break;
1452 			} else if (prop & SA_IGNORE) {
1453 				/*
1454 				 * Except for SIGCONT, shouldn't get here.
1455 				 * Default action is to ignore; drop it.
1456 				 */
1457 				break;		/* == ignore */
1458 			} else
1459 				goto keep;
1460 			/* NOTREACHED */
1461 		case (long)SIG_IGN:
1462 			/*
1463 			 * Masking above should prevent us ever trying
1464 			 * to take action on an ignored signal other
1465 			 * than SIGCONT, unless process is traced.
1466 			 */
1467 			if ((prop & SA_CONT) == 0 &&
1468 			    (pr->ps_flags & PS_TRACED) == 0)
1469 				printf("%s\n", __func__);
1470 			break;		/* == ignore */
1471 		default:
1472 			/*
1473 			 * This signal has an action, let
1474 			 * postsig() process it.
1475 			 */
1476 			goto keep;
1477 		}
1478 	}
1479 	/* NOTREACHED */
1480 
1481 keep:
1482 	/*
1483 	 * if we stashed a stop signal but no other signal is pending
1484 	 * anymore pick the stop signal up again.
1485 	 */
1486 	if (keep != 0 && signum == 0) {
1487 		signum = ffs(keep);
1488 		setsigctx(p, signum, sctx);
1489 	}
1490 	/* move the signal to p_siglist for later */
1491 	atomic_setbits_int(&p->p_siglist, mask | keep);
1492 	return (signum);
1493 }
1494 
1495 int
1496 proc_trap(struct proc *p, int signum)
1497 {
1498 	struct process *pr = p->p_p;
1499 
1500 	single_thread_set(p, SINGLE_SUSPEND | SINGLE_NOWAIT);
1501 	pr->ps_xsig = signum;
1502 
1503 	SCHED_LOCK();
1504 	proc_stop(p, 1);
1505 	SCHED_UNLOCK();
1506 
1507 	signum = pr->ps_xsig;
1508 	pr->ps_xsig = 0;
1509 	if ((p->p_flag & P_TRACESINGLE) == 0)
1510 		single_thread_clear(p, 0);
1511 	atomic_clearbits_int(&p->p_flag, P_TRACESINGLE);
1512 
1513 	return signum;
1514 }
1515 
1516 /*
1517  * Put the argument process into the stopped state and notify the parent
1518  * via wakeup.  Signals are handled elsewhere.  The process must not be
1519  * on the run queue.
1520  */
1521 void
1522 proc_stop(struct proc *p, int sw)
1523 {
1524 	struct process *pr = p->p_p;
1525 
1526 #ifdef MULTIPROCESSOR
1527 	SCHED_ASSERT_LOCKED();
1528 #endif
1529 	/* do not stop exiting procs */
1530 	if (ISSET(p->p_flag, P_WEXIT))
1531 		return;
1532 
1533 	p->p_stat = SSTOP;
1534 	atomic_clearbits_int(&pr->ps_flags, PS_WAITED);
1535 	atomic_setbits_int(&pr->ps_flags, PS_STOPPING);
1536 	atomic_setbits_int(&p->p_flag, P_SUSPSIG);
1537 	/*
1538 	 * We need this soft interrupt to be handled fast.
1539 	 * Extra calls to softclock don't hurt.
1540 	 */
1541 	softintr_schedule(proc_stop_si);
1542 	if (sw)
1543 		mi_switch();
1544 }
1545 
1546 /*
1547  * Called from a soft interrupt to send signals to the parents of stopped
1548  * processes.
1549  * We can't do this in proc_stop because it's called with nasty locks held
1550  * and we would need recursive scheduler lock to deal with that.
1551  */
1552 void
1553 proc_stop_sweep(void *v)
1554 {
1555 	struct process *pr;
1556 
1557 	LIST_FOREACH(pr, &allprocess, ps_list) {
1558 		if ((pr->ps_flags & PS_STOPPING) == 0)
1559 			continue;
1560 		atomic_setbits_int(&pr->ps_flags, PS_STOPPED);
1561 		atomic_clearbits_int(&pr->ps_flags, PS_STOPPING);
1562 
1563 		if ((pr->ps_pptr->ps_sigacts->ps_sigflags & SAS_NOCLDSTOP) == 0)
1564 			prsignal(pr->ps_pptr, SIGCHLD);
1565 		wakeup(pr->ps_pptr);
1566 	}
1567 }
1568 
1569 /*
1570  * Take the action for the specified signal
1571  * from the current set of pending signals.
1572  */
1573 void
1574 postsig(struct proc *p, int signum, struct sigctx *sctx)
1575 {
1576 	u_long trapno;
1577 	int mask, returnmask;
1578 	siginfo_t si;
1579 	union sigval sigval;
1580 	int code;
1581 
1582 	KASSERT(signum != 0);
1583 
1584 	mask = sigmask(signum);
1585 	atomic_clearbits_int(&p->p_siglist, mask);
1586 	sigval.sival_ptr = NULL;
1587 
1588 	if (p->p_sisig != signum) {
1589 		trapno = 0;
1590 		code = SI_USER;
1591 		sigval.sival_ptr = NULL;
1592 	} else {
1593 		trapno = p->p_sitrapno;
1594 		code = p->p_sicode;
1595 		sigval = p->p_sigval;
1596 	}
1597 	initsiginfo(&si, signum, trapno, code, sigval);
1598 
1599 #ifdef KTRACE
1600 	if (KTRPOINT(p, KTR_PSIG)) {
1601 		ktrpsig(p, signum, sctx->sig_action, p->p_flag & P_SIGSUSPEND ?
1602 		    p->p_oldmask : p->p_sigmask, code, &si);
1603 	}
1604 #endif
1605 	if (sctx->sig_action == SIG_DFL) {
1606 		/*
1607 		 * Default action, where the default is to kill
1608 		 * the process.  (Other cases were ignored above.)
1609 		 */
1610 		KERNEL_LOCK();
1611 		sigexit(p, signum);
1612 		/* NOTREACHED */
1613 	} else {
1614 		/*
1615 		 * If we get here, the signal must be caught.
1616 		 */
1617 #ifdef DIAGNOSTIC
1618 		if (sctx->sig_action == SIG_IGN || (p->p_sigmask & mask))
1619 			panic("postsig action");
1620 #endif
1621 		/*
1622 		 * Set the new mask value and also defer further
1623 		 * occurrences of this signal.
1624 		 *
1625 		 * Special case: user has done a sigpause.  Here the
1626 		 * current mask is not of interest, but rather the
1627 		 * mask from before the sigpause is what we want
1628 		 * restored after the signal processing is completed.
1629 		 */
1630 		if (p->p_flag & P_SIGSUSPEND) {
1631 			atomic_clearbits_int(&p->p_flag, P_SIGSUSPEND);
1632 			returnmask = p->p_oldmask;
1633 		} else {
1634 			returnmask = p->p_sigmask;
1635 		}
1636 		if (p->p_sisig == signum) {
1637 			p->p_sisig = 0;
1638 			p->p_sitrapno = 0;
1639 			p->p_sicode = SI_USER;
1640 			p->p_sigval.sival_ptr = NULL;
1641 		}
1642 
1643 		if (sendsig(sctx->sig_action, signum, returnmask, &si,
1644 		    sctx->sig_info, sctx->sig_onstack)) {
1645 			KERNEL_LOCK();
1646 			sigexit(p, SIGILL);
1647 			/* NOTREACHED */
1648 		}
1649 		postsig_done(p, signum, sctx->sig_catchmask, sctx->sig_reset);
1650 	}
1651 }
1652 
1653 /*
1654  * Force the current process to exit with the specified signal, dumping core
1655  * if appropriate.  We bypass the normal tests for masked and caught signals,
1656  * allowing unrecoverable failures to terminate the process without changing
1657  * signal state.  Mark the accounting record with the signal termination.
1658  * If dumping core, save the signal number for the debugger.  Calls exit and
1659  * does not return.
1660  */
1661 void
1662 sigexit(struct proc *p, int signum)
1663 {
1664 	/* Mark process as going away */
1665 	atomic_setbits_int(&p->p_flag, P_WEXIT);
1666 
1667 	p->p_p->ps_acflag |= AXSIG;
1668 	if (sigprop[signum] & SA_CORE) {
1669 		p->p_sisig = signum;
1670 
1671 		/* if there are other threads, pause them */
1672 		if (P_HASSIBLING(p))
1673 			single_thread_set(p, SINGLE_UNWIND);
1674 
1675 		if (coredump(p) == 0)
1676 			signum |= WCOREFLAG;
1677 	}
1678 	exit1(p, 0, signum, EXIT_NORMAL);
1679 	/* NOTREACHED */
1680 }
1681 
1682 /*
1683  * Send uncatchable SIGABRT for coredump.
1684  */
1685 void
1686 sigabort(struct proc *p)
1687 {
1688 	struct sigaction sa;
1689 
1690 	KASSERT(p == curproc || panicstr || db_active);
1691 
1692 	memset(&sa, 0, sizeof sa);
1693 	sa.sa_handler = SIG_DFL;
1694 	setsigvec(p, SIGABRT, &sa);
1695 	CLR(p->p_sigmask, sigmask(SIGABRT));
1696 	psignal(p, SIGABRT);
1697 }
1698 
1699 /*
1700  * Return 1 if `sig', a given signal, is ignored or masked for `p', a given
1701  * thread, and 0 otherwise.
1702  */
1703 int
1704 sigismasked(struct proc *p, int sig)
1705 {
1706 	struct process *pr = p->p_p;
1707 	int rv;
1708 
1709 	KASSERT(p == curproc);
1710 
1711 	mtx_enter(&pr->ps_mtx);
1712 	rv = (pr->ps_sigacts->ps_sigignore & sigmask(sig)) ||
1713 	    (p->p_sigmask & sigmask(sig));
1714 	mtx_leave(&pr->ps_mtx);
1715 
1716 	return !!rv;
1717 }
1718 
1719 struct coredump_iostate {
1720 	struct proc *io_proc;
1721 	struct vnode *io_vp;
1722 	struct ucred *io_cred;
1723 	off_t io_offset;
1724 };
1725 
1726 /*
1727  * Dump core, into a file named "progname.core", unless the process was
1728  * setuid/setgid.
1729  */
1730 int
1731 coredump(struct proc *p)
1732 {
1733 #ifdef SMALL_KERNEL
1734 	return EPERM;
1735 #else
1736 	struct process *pr = p->p_p;
1737 	struct vnode *vp;
1738 	struct ucred *cred = p->p_ucred;
1739 	struct vmspace *vm = p->p_vmspace;
1740 	struct nameidata nd;
1741 	struct vattr vattr;
1742 	struct coredump_iostate	io;
1743 	int error, len, incrash = 0;
1744 	char *name;
1745 	const char *dir = "/var/crash";
1746 
1747 	atomic_setbits_int(&pr->ps_flags, PS_COREDUMP);
1748 
1749 #ifdef PMAP_CHECK_COPYIN
1750 	/* disable copyin checks, so we can write out text sections if needed */
1751 	p->p_vmspace->vm_map.check_copyin_count = 0;
1752 #endif
1753 
1754 	/* Don't dump if will exceed file size limit. */
1755 	if (USPACE + ptoa(vm->vm_dsize + vm->vm_ssize) >= lim_cur(RLIMIT_CORE))
1756 		return (EFBIG);
1757 
1758 	name = pool_get(&namei_pool, PR_WAITOK);
1759 
1760 	/*
1761 	 * If the process has inconsistent uids, nosuidcoredump
1762 	 * determines coredump placement policy.
1763 	 */
1764 	if (((pr->ps_flags & PS_SUGID) && (error = suser(p))) ||
1765 	   ((pr->ps_flags & PS_SUGID) && nosuidcoredump)) {
1766 		if (nosuidcoredump == 3) {
1767 			/*
1768 			 * If the program directory does not exist, dumps of
1769 			 * that core will silently fail.
1770 			 */
1771 			len = snprintf(name, MAXPATHLEN, "%s/%s/%u.core",
1772 			    dir, pr->ps_comm, pr->ps_pid);
1773 			incrash = KERNELPATH;
1774 		} else if (nosuidcoredump == 2) {
1775 			len = snprintf(name, MAXPATHLEN, "%s/%s.core",
1776 			    dir, pr->ps_comm);
1777 			incrash = KERNELPATH;
1778 		} else {
1779 			pool_put(&namei_pool, name);
1780 			return (EPERM);
1781 		}
1782 	} else
1783 		len = snprintf(name, MAXPATHLEN, "%s.core", pr->ps_comm);
1784 
1785 	if (len >= MAXPATHLEN) {
1786 		pool_put(&namei_pool, name);
1787 		return (EACCES);
1788 	}
1789 
1790 	/*
1791 	 * Control the UID used to write out.  The normal case uses
1792 	 * the real UID.  If the sugid case is going to write into the
1793 	 * controlled directory, we do so as root.
1794 	 */
1795 	if (incrash == 0) {
1796 		cred = crdup(cred);
1797 		cred->cr_uid = cred->cr_ruid;
1798 		cred->cr_gid = cred->cr_rgid;
1799 	} else {
1800 		if (p->p_fd->fd_rdir) {
1801 			vrele(p->p_fd->fd_rdir);
1802 			p->p_fd->fd_rdir = NULL;
1803 		}
1804 		p->p_ucred = crdup(p->p_ucred);
1805 		crfree(cred);
1806 		cred = p->p_ucred;
1807 		crhold(cred);
1808 		cred->cr_uid = 0;
1809 		cred->cr_gid = 0;
1810 	}
1811 
1812 	/* incrash should be 0 or KERNELPATH only */
1813 	NDINIT(&nd, 0, BYPASSUNVEIL | incrash, UIO_SYSSPACE, name, p);
1814 
1815 	error = vn_open(&nd, O_CREAT | FWRITE | O_NOFOLLOW | O_NONBLOCK,
1816 	    S_IRUSR | S_IWUSR);
1817 
1818 	if (error)
1819 		goto out;
1820 
1821 	/*
1822 	 * Don't dump to non-regular files, files with links, or files
1823 	 * owned by someone else.
1824 	 */
1825 	vp = nd.ni_vp;
1826 	if ((error = VOP_GETATTR(vp, &vattr, cred, p)) != 0) {
1827 		VOP_UNLOCK(vp);
1828 		vn_close(vp, FWRITE, cred, p);
1829 		goto out;
1830 	}
1831 	if (vp->v_type != VREG || vattr.va_nlink != 1 ||
1832 	    vattr.va_mode & ((VREAD | VWRITE) >> 3 | (VREAD | VWRITE) >> 6) ||
1833 	    vattr.va_uid != cred->cr_uid) {
1834 		error = EACCES;
1835 		VOP_UNLOCK(vp);
1836 		vn_close(vp, FWRITE, cred, p);
1837 		goto out;
1838 	}
1839 	vattr_null(&vattr);
1840 	vattr.va_size = 0;
1841 	VOP_SETATTR(vp, &vattr, cred, p);
1842 	pr->ps_acflag |= ACORE;
1843 
1844 	io.io_proc = p;
1845 	io.io_vp = vp;
1846 	io.io_cred = cred;
1847 	io.io_offset = 0;
1848 	VOP_UNLOCK(vp);
1849 	vref(vp);
1850 	error = vn_close(vp, FWRITE, cred, p);
1851 	if (error == 0)
1852 		error = coredump_elf(p, &io);
1853 	vrele(vp);
1854 out:
1855 	crfree(cred);
1856 	pool_put(&namei_pool, name);
1857 	return (error);
1858 #endif
1859 }
1860 
1861 #ifndef SMALL_KERNEL
1862 int
1863 coredump_write(void *cookie, enum uio_seg segflg, const void *data, size_t len,
1864     int isvnode)
1865 {
1866 	struct coredump_iostate *io = cookie;
1867 	off_t coffset = 0;
1868 	size_t csize;
1869 	int chunk, error;
1870 
1871 	csize = len;
1872 	do {
1873 		if (sigmask(SIGKILL) &
1874 		    (io->io_proc->p_siglist | io->io_proc->p_p->ps_siglist))
1875 			return (EINTR);
1876 
1877 		/* Rest of the loop sleeps with lock held, so... */
1878 		yield();
1879 
1880 		chunk = MIN(csize, MAXPHYS);
1881 		error = vn_rdwr(UIO_WRITE, io->io_vp,
1882 		    (caddr_t)data + coffset, chunk,
1883 		    io->io_offset + coffset, segflg,
1884 		    IO_UNIT, io->io_cred, NULL, io->io_proc);
1885 		if (error && (error != EFAULT || !isvnode)) {
1886 			struct process *pr = io->io_proc->p_p;
1887 
1888 			if (error == ENOSPC)
1889 				log(LOG_ERR,
1890 				    "coredump of %s(%d) failed, filesystem full\n",
1891 				    pr->ps_comm, pr->ps_pid);
1892 			else
1893 				log(LOG_ERR,
1894 				    "coredump of %s(%d), write failed: errno %d\n",
1895 				    pr->ps_comm, pr->ps_pid, error);
1896 			return (error);
1897 		}
1898 
1899 		coffset += chunk;
1900 		csize -= chunk;
1901 	} while (csize > 0);
1902 
1903 	io->io_offset += len;
1904 	return (0);
1905 }
1906 
1907 void
1908 coredump_unmap(void *cookie, vaddr_t start, vaddr_t end)
1909 {
1910 	struct coredump_iostate *io = cookie;
1911 
1912 	uvm_unmap(&io->io_proc->p_vmspace->vm_map, start, end);
1913 }
1914 
1915 #endif	/* !SMALL_KERNEL */
1916 
1917 /*
1918  * Nonexistent system call-- signal process (may want to handle it).
1919  * Flag error in case process won't see signal immediately (blocked or ignored).
1920  */
1921 int
1922 sys_nosys(struct proc *p, void *v, register_t *retval)
1923 {
1924 	ptsignal(p, SIGSYS, STHREAD);
1925 	return (ENOSYS);
1926 }
1927 
1928 int
1929 sys___thrsigdivert(struct proc *p, void *v, register_t *retval)
1930 {
1931 	struct sys___thrsigdivert_args /* {
1932 		syscallarg(sigset_t) sigmask;
1933 		syscallarg(siginfo_t *) info;
1934 		syscallarg(const struct timespec *) timeout;
1935 	} */ *uap = v;
1936 	struct sigctx ctx;
1937 	sigset_t mask = SCARG(uap, sigmask) &~ sigcantmask;
1938 	siginfo_t si;
1939 	uint64_t nsecs = INFSLP;
1940 	int timeinvalid = 0;
1941 	int error = 0;
1942 
1943 	memset(&si, 0, sizeof(si));
1944 
1945 	if (SCARG(uap, timeout) != NULL) {
1946 		struct timespec ts;
1947 		if ((error = copyin(SCARG(uap, timeout), &ts, sizeof(ts))) != 0)
1948 			return (error);
1949 #ifdef KTRACE
1950 		if (KTRPOINT(p, KTR_STRUCT))
1951 			ktrreltimespec(p, &ts);
1952 #endif
1953 		if (!timespecisvalid(&ts))
1954 			timeinvalid = 1;
1955 		else
1956 			nsecs = TIMESPEC_TO_NSEC(&ts);
1957 	}
1958 
1959 	dosigsuspend(p, p->p_sigmask &~ mask);
1960 	for (;;) {
1961 		si.si_signo = cursig(p, &ctx, 0);
1962 		if (si.si_signo != 0) {
1963 			sigset_t smask = sigmask(si.si_signo);
1964 			if (smask & mask) {
1965 				atomic_clearbits_int(&p->p_siglist, smask);
1966 				error = 0;
1967 				break;
1968 			}
1969 		}
1970 
1971 		/* per-POSIX, delay this error until after the above */
1972 		if (timeinvalid)
1973 			error = EINVAL;
1974 		/* per-POSIX, return immediately if timeout is zero-valued */
1975 		if (nsecs == 0)
1976 			error = EAGAIN;
1977 
1978 		if (error != 0)
1979 			break;
1980 
1981 		error = tsleep_nsec(&nowake, PPAUSE|PCATCH, "sigwait", nsecs);
1982 	}
1983 
1984 	if (error == 0) {
1985 		*retval = si.si_signo;
1986 		if (SCARG(uap, info) != NULL) {
1987 			error = copyout(&si, SCARG(uap, info), sizeof(si));
1988 #ifdef KTRACE
1989 			if (error == 0 && KTRPOINT(p, KTR_STRUCT))
1990 				ktrsiginfo(p, &si);
1991 #endif
1992 		}
1993 	} else if (error == ERESTART && SCARG(uap, timeout) != NULL) {
1994 		/*
1995 		 * Restarting is wrong if there's a timeout, as it'll be
1996 		 * for the same interval again
1997 		 */
1998 		error = EINTR;
1999 	}
2000 
2001 	return (error);
2002 }
2003 
2004 void
2005 initsiginfo(siginfo_t *si, int sig, u_long trapno, int code, union sigval val)
2006 {
2007 	memset(si, 0, sizeof(*si));
2008 
2009 	si->si_signo = sig;
2010 	si->si_code = code;
2011 	if (code == SI_USER) {
2012 		si->si_value = val;
2013 	} else {
2014 		switch (sig) {
2015 		case SIGSEGV:
2016 		case SIGILL:
2017 		case SIGBUS:
2018 		case SIGFPE:
2019 			si->si_addr = val.sival_ptr;
2020 			si->si_trapno = trapno;
2021 			break;
2022 		case SIGXFSZ:
2023 			break;
2024 		}
2025 	}
2026 }
2027 
2028 void
2029 userret(struct proc *p)
2030 {
2031 	struct sigctx ctx;
2032 	int signum;
2033 
2034 	if (p->p_flag & P_SUSPSINGLE)
2035 		single_thread_check(p, 0);
2036 
2037 	/* send SIGPROF or SIGVTALRM if their timers interrupted this thread */
2038 	if (p->p_flag & P_PROFPEND) {
2039 		atomic_clearbits_int(&p->p_flag, P_PROFPEND);
2040 		psignal(p, SIGPROF);
2041 	}
2042 	if (p->p_flag & P_ALRMPEND) {
2043 		atomic_clearbits_int(&p->p_flag, P_ALRMPEND);
2044 		psignal(p, SIGVTALRM);
2045 	}
2046 
2047 	if (SIGPENDING(p) != 0) {
2048 		while ((signum = cursig(p, &ctx, 0)) != 0)
2049 			postsig(p, signum, &ctx);
2050 	}
2051 
2052 	/*
2053 	 * If P_SIGSUSPEND is still set here, then we still need to restore
2054 	 * the original sigmask before returning to userspace.  Also, this
2055 	 * might unmask some pending signals, so we need to check a second
2056 	 * time for signals to post.
2057 	 */
2058 	if (p->p_flag & P_SIGSUSPEND) {
2059 		p->p_sigmask = p->p_oldmask;
2060 		atomic_clearbits_int(&p->p_flag, P_SIGSUSPEND);
2061 
2062 		while ((signum = cursig(p, &ctx, 0)) != 0)
2063 			postsig(p, signum, &ctx);
2064 	}
2065 
2066 	WITNESS_WARN(WARN_PANIC, NULL, "userret: returning");
2067 
2068 	p->p_cpu->ci_schedstate.spc_curpriority = p->p_usrpri;
2069 }
2070 
2071 int
2072 single_thread_check_locked(struct proc *p, int deep)
2073 {
2074 	struct process *pr = p->p_p;
2075 
2076 	MUTEX_ASSERT_LOCKED(&pr->ps_mtx);
2077 
2078 	if (pr->ps_single == NULL || pr->ps_single == p)
2079 		return (0);
2080 
2081 	do {
2082 		/* if we're in deep, we need to unwind to the edge */
2083 		if (deep) {
2084 			if (pr->ps_flags & PS_SINGLEUNWIND)
2085 				return (ERESTART);
2086 			if (pr->ps_flags & PS_SINGLEEXIT)
2087 				return (EINTR);
2088 		}
2089 
2090 		if (pr->ps_flags & PS_SINGLEEXIT) {
2091 			mtx_leave(&pr->ps_mtx);
2092 			KERNEL_LOCK();
2093 			exit1(p, 0, 0, EXIT_THREAD_NOCHECK);
2094 			/* NOTREACHED */
2095 		}
2096 
2097 		if (--pr->ps_singlecnt == 0)
2098 			wakeup(&pr->ps_singlecnt);
2099 
2100 		/* not exiting and don't need to unwind, so suspend */
2101 		mtx_leave(&pr->ps_mtx);
2102 
2103 		SCHED_LOCK();
2104 		p->p_stat = SSTOP;
2105 		mi_switch();
2106 		SCHED_UNLOCK();
2107 		mtx_enter(&pr->ps_mtx);
2108 	} while (pr->ps_single != NULL);
2109 
2110 	return (0);
2111 }
2112 
2113 int
2114 single_thread_check(struct proc *p, int deep)
2115 {
2116 	int error;
2117 
2118 	mtx_enter(&p->p_p->ps_mtx);
2119 	error = single_thread_check_locked(p, deep);
2120 	mtx_leave(&p->p_p->ps_mtx);
2121 
2122 	return error;
2123 }
2124 
2125 /*
2126  * Stop other threads in the process.  The mode controls how and
2127  * where the other threads should stop:
2128  *  - SINGLE_SUSPEND: stop wherever they are, will later be released (via
2129  *    single_thread_clear())
2130  *  - SINGLE_UNWIND: just unwind to kernel boundary, will be told to exit
2131  *    (by setting to SINGLE_EXIT) or released as with SINGLE_SUSPEND
2132  *  - SINGLE_EXIT: unwind to kernel boundary and exit
2133  */
2134 int
2135 single_thread_set(struct proc *p, int flags)
2136 {
2137 	struct process *pr = p->p_p;
2138 	struct proc *q;
2139 	int error, mode = flags & SINGLE_MASK;
2140 
2141 	KASSERT(curproc == p);
2142 
2143 	mtx_enter(&pr->ps_mtx);
2144 	error = single_thread_check_locked(p, flags & SINGLE_DEEP);
2145 	if (error) {
2146 		mtx_leave(&pr->ps_mtx);
2147 		return error;
2148 	}
2149 
2150 	switch (mode) {
2151 	case SINGLE_SUSPEND:
2152 		break;
2153 	case SINGLE_UNWIND:
2154 		atomic_setbits_int(&pr->ps_flags, PS_SINGLEUNWIND);
2155 		break;
2156 	case SINGLE_EXIT:
2157 		atomic_setbits_int(&pr->ps_flags, PS_SINGLEEXIT);
2158 		atomic_clearbits_int(&pr->ps_flags, PS_SINGLEUNWIND);
2159 		break;
2160 #ifdef DIAGNOSTIC
2161 	default:
2162 		panic("single_thread_mode = %d", mode);
2163 #endif
2164 	}
2165 	KASSERT((p->p_flag & P_SUSPSINGLE) == 0);
2166 	pr->ps_single = p;
2167 	pr->ps_singlecnt = pr->ps_threadcnt;
2168 
2169 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
2170 		if (q == p)
2171 			continue;
2172 		SCHED_LOCK();
2173 		atomic_setbits_int(&q->p_flag, P_SUSPSINGLE);
2174 		switch (q->p_stat) {
2175 		case SSTOP:
2176 			if (mode == SINGLE_EXIT) {
2177 				unsleep(q);
2178 				setrunnable(q);
2179 			} else
2180 				--pr->ps_singlecnt;
2181 			break;
2182 		case SSLEEP:
2183 			/* if it's not interruptible, then just have to wait */
2184 			if (q->p_flag & P_SINTR) {
2185 				/* merely need to suspend?  just stop it */
2186 				if (mode == SINGLE_SUSPEND) {
2187 					q->p_stat = SSTOP;
2188 					--pr->ps_singlecnt;
2189 					break;
2190 				}
2191 				/* need to unwind or exit, so wake it */
2192 				unsleep(q);
2193 				setrunnable(q);
2194 			}
2195 			break;
2196 		case SONPROC:
2197 			signotify(q);
2198 			break;
2199 		case SRUN:
2200 		case SIDL:
2201 		case SDEAD:
2202 			break;
2203 		}
2204 		SCHED_UNLOCK();
2205 	}
2206 
2207 	/* count ourself out */
2208 	--pr->ps_singlecnt;
2209 	mtx_leave(&pr->ps_mtx);
2210 
2211 	if ((flags & SINGLE_NOWAIT) == 0)
2212 		single_thread_wait(pr, 1);
2213 
2214 	return 0;
2215 }
2216 
2217 /*
2218  * Wait for other threads to stop. If recheck is false then the function
2219  * returns non-zero if the caller needs to restart the check else 0 is
2220  * returned. If recheck is true the return value is always 0.
2221  */
2222 int
2223 single_thread_wait(struct process *pr, int recheck)
2224 {
2225 	int wait;
2226 
2227 	/* wait until they're all suspended */
2228 	mtx_enter(&pr->ps_mtx);
2229 	while ((wait = pr->ps_singlecnt > 0)) {
2230 		msleep_nsec(&pr->ps_singlecnt, &pr->ps_mtx, PWAIT, "suspend",
2231 		    INFSLP);
2232 		if (!recheck)
2233 			break;
2234 	}
2235 	KASSERT((pr->ps_single->p_flag & P_SUSPSINGLE) == 0);
2236 	mtx_leave(&pr->ps_mtx);
2237 
2238 	return wait;
2239 }
2240 
2241 void
2242 single_thread_clear(struct proc *p, int flag)
2243 {
2244 	struct process *pr = p->p_p;
2245 	struct proc *q;
2246 
2247 	KASSERT(pr->ps_single == p);
2248 	KASSERT(curproc == p);
2249 
2250 	mtx_enter(&pr->ps_mtx);
2251 	pr->ps_single = NULL;
2252 	atomic_clearbits_int(&pr->ps_flags, PS_SINGLEUNWIND | PS_SINGLEEXIT);
2253 
2254 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
2255 		if (q == p || (q->p_flag & P_SUSPSINGLE) == 0)
2256 			continue;
2257 		atomic_clearbits_int(&q->p_flag, P_SUSPSINGLE);
2258 
2259 		/*
2260 		 * if the thread was only stopped for single threading
2261 		 * then clearing that either makes it runnable or puts
2262 		 * it back into some sleep queue
2263 		 */
2264 		SCHED_LOCK();
2265 		if (q->p_stat == SSTOP && (q->p_flag & flag) == 0) {
2266 			if (q->p_wchan == NULL)
2267 				setrunnable(q);
2268 			else {
2269 				atomic_clearbits_int(&q->p_flag, P_WSLEEP);
2270 				q->p_stat = SSLEEP;
2271 			}
2272 		}
2273 		SCHED_UNLOCK();
2274 	}
2275 	mtx_leave(&pr->ps_mtx);
2276 }
2277 
2278 void
2279 sigio_del(struct sigiolst *rmlist)
2280 {
2281 	struct sigio *sigio;
2282 
2283 	while ((sigio = LIST_FIRST(rmlist)) != NULL) {
2284 		LIST_REMOVE(sigio, sio_pgsigio);
2285 		crfree(sigio->sio_ucred);
2286 		free(sigio, M_SIGIO, sizeof(*sigio));
2287 	}
2288 }
2289 
2290 void
2291 sigio_unlink(struct sigio_ref *sir, struct sigiolst *rmlist)
2292 {
2293 	struct sigio *sigio;
2294 
2295 	MUTEX_ASSERT_LOCKED(&sigio_lock);
2296 
2297 	sigio = sir->sir_sigio;
2298 	if (sigio != NULL) {
2299 		KASSERT(sigio->sio_myref == sir);
2300 		sir->sir_sigio = NULL;
2301 
2302 		if (sigio->sio_pgid > 0)
2303 			sigio->sio_proc = NULL;
2304 		else
2305 			sigio->sio_pgrp = NULL;
2306 		LIST_REMOVE(sigio, sio_pgsigio);
2307 
2308 		LIST_INSERT_HEAD(rmlist, sigio, sio_pgsigio);
2309 	}
2310 }
2311 
2312 void
2313 sigio_free(struct sigio_ref *sir)
2314 {
2315 	struct sigiolst rmlist;
2316 
2317 	if (sir->sir_sigio == NULL)
2318 		return;
2319 
2320 	LIST_INIT(&rmlist);
2321 
2322 	mtx_enter(&sigio_lock);
2323 	sigio_unlink(sir, &rmlist);
2324 	mtx_leave(&sigio_lock);
2325 
2326 	sigio_del(&rmlist);
2327 }
2328 
2329 void
2330 sigio_freelist(struct sigiolst *sigiolst)
2331 {
2332 	struct sigiolst rmlist;
2333 	struct sigio *sigio;
2334 
2335 	if (LIST_EMPTY(sigiolst))
2336 		return;
2337 
2338 	LIST_INIT(&rmlist);
2339 
2340 	mtx_enter(&sigio_lock);
2341 	while ((sigio = LIST_FIRST(sigiolst)) != NULL)
2342 		sigio_unlink(sigio->sio_myref, &rmlist);
2343 	mtx_leave(&sigio_lock);
2344 
2345 	sigio_del(&rmlist);
2346 }
2347 
2348 int
2349 sigio_setown(struct sigio_ref *sir, u_long cmd, caddr_t data)
2350 {
2351 	struct sigiolst rmlist;
2352 	struct proc *p = curproc;
2353 	struct pgrp *pgrp = NULL;
2354 	struct process *pr = NULL;
2355 	struct sigio *sigio;
2356 	int error;
2357 	pid_t pgid = *(int *)data;
2358 
2359 	if (pgid == 0) {
2360 		sigio_free(sir);
2361 		return (0);
2362 	}
2363 
2364 	if (cmd == TIOCSPGRP) {
2365 		if (pgid < 0)
2366 			return (EINVAL);
2367 		pgid = -pgid;
2368 	}
2369 
2370 	sigio = malloc(sizeof(*sigio), M_SIGIO, M_WAITOK);
2371 	sigio->sio_pgid = pgid;
2372 	sigio->sio_ucred = crhold(p->p_ucred);
2373 	sigio->sio_myref = sir;
2374 
2375 	LIST_INIT(&rmlist);
2376 
2377 	/*
2378 	 * The kernel lock, and not sleeping between prfind()/pgfind() and
2379 	 * linking of the sigio ensure that the process or process group does
2380 	 * not disappear unexpectedly.
2381 	 */
2382 	KERNEL_LOCK();
2383 	mtx_enter(&sigio_lock);
2384 
2385 	if (pgid > 0) {
2386 		pr = prfind(pgid);
2387 		if (pr == NULL) {
2388 			error = ESRCH;
2389 			goto fail;
2390 		}
2391 
2392 		/*
2393 		 * Policy - Don't allow a process to FSETOWN a process
2394 		 * in another session.
2395 		 *
2396 		 * Remove this test to allow maximum flexibility or
2397 		 * restrict FSETOWN to the current process or process
2398 		 * group for maximum safety.
2399 		 */
2400 		if (pr->ps_session != p->p_p->ps_session) {
2401 			error = EPERM;
2402 			goto fail;
2403 		}
2404 
2405 		if ((pr->ps_flags & PS_EXITING) != 0) {
2406 			error = ESRCH;
2407 			goto fail;
2408 		}
2409 	} else /* if (pgid < 0) */ {
2410 		pgrp = pgfind(-pgid);
2411 		if (pgrp == NULL) {
2412 			error = ESRCH;
2413 			goto fail;
2414 		}
2415 
2416 		/*
2417 		 * Policy - Don't allow a process to FSETOWN a process
2418 		 * in another session.
2419 		 *
2420 		 * Remove this test to allow maximum flexibility or
2421 		 * restrict FSETOWN to the current process or process
2422 		 * group for maximum safety.
2423 		 */
2424 		if (pgrp->pg_session != p->p_p->ps_session) {
2425 			error = EPERM;
2426 			goto fail;
2427 		}
2428 	}
2429 
2430 	if (pgid > 0) {
2431 		sigio->sio_proc = pr;
2432 		LIST_INSERT_HEAD(&pr->ps_sigiolst, sigio, sio_pgsigio);
2433 	} else {
2434 		sigio->sio_pgrp = pgrp;
2435 		LIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
2436 	}
2437 
2438 	sigio_unlink(sir, &rmlist);
2439 	sir->sir_sigio = sigio;
2440 
2441 	mtx_leave(&sigio_lock);
2442 	KERNEL_UNLOCK();
2443 
2444 	sigio_del(&rmlist);
2445 
2446 	return (0);
2447 
2448 fail:
2449 	mtx_leave(&sigio_lock);
2450 	KERNEL_UNLOCK();
2451 
2452 	crfree(sigio->sio_ucred);
2453 	free(sigio, M_SIGIO, sizeof(*sigio));
2454 
2455 	return (error);
2456 }
2457 
2458 void
2459 sigio_getown(struct sigio_ref *sir, u_long cmd, caddr_t data)
2460 {
2461 	struct sigio *sigio;
2462 	pid_t pgid = 0;
2463 
2464 	mtx_enter(&sigio_lock);
2465 	sigio = sir->sir_sigio;
2466 	if (sigio != NULL)
2467 		pgid = sigio->sio_pgid;
2468 	mtx_leave(&sigio_lock);
2469 
2470 	if (cmd == TIOCGPGRP)
2471 		pgid = -pgid;
2472 
2473 	*(int *)data = pgid;
2474 }
2475 
2476 void
2477 sigio_copy(struct sigio_ref *dst, struct sigio_ref *src)
2478 {
2479 	struct sigiolst rmlist;
2480 	struct sigio *newsigio, *sigio;
2481 
2482 	sigio_free(dst);
2483 
2484 	if (src->sir_sigio == NULL)
2485 		return;
2486 
2487 	newsigio = malloc(sizeof(*newsigio), M_SIGIO, M_WAITOK);
2488 	LIST_INIT(&rmlist);
2489 
2490 	mtx_enter(&sigio_lock);
2491 
2492 	sigio = src->sir_sigio;
2493 	if (sigio == NULL) {
2494 		mtx_leave(&sigio_lock);
2495 		free(newsigio, M_SIGIO, sizeof(*newsigio));
2496 		return;
2497 	}
2498 
2499 	newsigio->sio_pgid = sigio->sio_pgid;
2500 	newsigio->sio_ucred = crhold(sigio->sio_ucred);
2501 	newsigio->sio_myref = dst;
2502 	if (newsigio->sio_pgid > 0) {
2503 		newsigio->sio_proc = sigio->sio_proc;
2504 		LIST_INSERT_HEAD(&newsigio->sio_proc->ps_sigiolst, newsigio,
2505 		    sio_pgsigio);
2506 	} else {
2507 		newsigio->sio_pgrp = sigio->sio_pgrp;
2508 		LIST_INSERT_HEAD(&newsigio->sio_pgrp->pg_sigiolst, newsigio,
2509 		    sio_pgsigio);
2510 	}
2511 
2512 	sigio_unlink(dst, &rmlist);
2513 	dst->sir_sigio = newsigio;
2514 
2515 	mtx_leave(&sigio_lock);
2516 
2517 	sigio_del(&rmlist);
2518 }
2519