xref: /openbsd/sys/kern/kern_sig.c (revision d415bd75)
1 /*	$OpenBSD: kern_sig.c,v 1.320 2023/10/06 08:58:13 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 int	filt_sigattach(struct knote *kn);
74 void	filt_sigdetach(struct knote *kn);
75 int	filt_signal(struct knote *kn, long hint);
76 
77 const struct filterops sig_filtops = {
78 	.f_flags	= 0,
79 	.f_attach	= filt_sigattach,
80 	.f_detach	= filt_sigdetach,
81 	.f_event	= filt_signal,
82 };
83 
84 /*
85  * The array below categorizes the signals and their default actions.
86  */
87 const int sigprop[NSIG] = {
88 	0,			/* unused */
89 	SA_KILL,		/* SIGHUP */
90 	SA_KILL,		/* SIGINT */
91 	SA_KILL|SA_CORE,	/* SIGQUIT */
92 	SA_KILL|SA_CORE,	/* SIGILL */
93 	SA_KILL|SA_CORE,	/* SIGTRAP */
94 	SA_KILL|SA_CORE,	/* SIGABRT */
95 	SA_KILL|SA_CORE,	/* SIGEMT */
96 	SA_KILL|SA_CORE,	/* SIGFPE */
97 	SA_KILL,		/* SIGKILL */
98 	SA_KILL|SA_CORE,	/* SIGBUS */
99 	SA_KILL|SA_CORE,	/* SIGSEGV */
100 	SA_KILL|SA_CORE,	/* SIGSYS */
101 	SA_KILL,		/* SIGPIPE */
102 	SA_KILL,		/* SIGALRM */
103 	SA_KILL,		/* SIGTERM */
104 	SA_IGNORE,		/* SIGURG */
105 	SA_STOP,		/* SIGSTOP */
106 	SA_STOP|SA_TTYSTOP,	/* SIGTSTP */
107 	SA_IGNORE|SA_CONT,	/* SIGCONT */
108 	SA_IGNORE,		/* SIGCHLD */
109 	SA_STOP|SA_TTYSTOP,	/* SIGTTIN */
110 	SA_STOP|SA_TTYSTOP,	/* SIGTTOU */
111 	SA_IGNORE,		/* SIGIO */
112 	SA_KILL,		/* SIGXCPU */
113 	SA_KILL,		/* SIGXFSZ */
114 	SA_KILL,		/* SIGVTALRM */
115 	SA_KILL,		/* SIGPROF */
116 	SA_IGNORE,		/* SIGWINCH  */
117 	SA_IGNORE,		/* SIGINFO */
118 	SA_KILL,		/* SIGUSR1 */
119 	SA_KILL,		/* SIGUSR2 */
120 	SA_IGNORE,		/* SIGTHR */
121 };
122 
123 #define	CONTSIGMASK	(sigmask(SIGCONT))
124 #define	STOPSIGMASK	(sigmask(SIGSTOP) | sigmask(SIGTSTP) | \
125 			    sigmask(SIGTTIN) | sigmask(SIGTTOU))
126 
127 void setsigvec(struct proc *, int, struct sigaction *);
128 
129 void proc_stop(struct proc *p, int);
130 void proc_stop_sweep(void *);
131 void *proc_stop_si;
132 
133 void setsigctx(struct proc *, int, struct sigctx *);
134 void postsig_done(struct proc *, int, sigset_t, int);
135 void postsig(struct proc *, int, struct sigctx *);
136 int cansignal(struct proc *, struct process *, int);
137 
138 struct pool sigacts_pool;	/* memory pool for sigacts structures */
139 
140 void sigio_del(struct sigiolst *);
141 void sigio_unlink(struct sigio_ref *, struct sigiolst *);
142 struct mutex sigio_lock = MUTEX_INITIALIZER(IPL_HIGH);
143 
144 /*
145  * Can thread p, send the signal signum to process qr?
146  */
147 int
148 cansignal(struct proc *p, struct process *qr, int signum)
149 {
150 	struct process *pr = p->p_p;
151 	struct ucred *uc = p->p_ucred;
152 	struct ucred *quc = qr->ps_ucred;
153 
154 	if (uc->cr_uid == 0)
155 		return (1);		/* root can always signal */
156 
157 	if (pr == qr)
158 		return (1);		/* process can always signal itself */
159 
160 	/* optimization: if the same creds then the tests below will pass */
161 	if (uc == quc)
162 		return (1);
163 
164 	if (signum == SIGCONT && qr->ps_session == pr->ps_session)
165 		return (1);		/* SIGCONT in session */
166 
167 	/*
168 	 * Using kill(), only certain signals can be sent to setugid
169 	 * child processes
170 	 */
171 	if (qr->ps_flags & PS_SUGID) {
172 		switch (signum) {
173 		case 0:
174 		case SIGKILL:
175 		case SIGINT:
176 		case SIGTERM:
177 		case SIGALRM:
178 		case SIGSTOP:
179 		case SIGTTIN:
180 		case SIGTTOU:
181 		case SIGTSTP:
182 		case SIGHUP:
183 		case SIGUSR1:
184 		case SIGUSR2:
185 			if (uc->cr_ruid == quc->cr_ruid ||
186 			    uc->cr_uid == quc->cr_ruid)
187 				return (1);
188 		}
189 		return (0);
190 	}
191 
192 	if (uc->cr_ruid == quc->cr_ruid ||
193 	    uc->cr_ruid == quc->cr_svuid ||
194 	    uc->cr_uid == quc->cr_ruid ||
195 	    uc->cr_uid == quc->cr_svuid)
196 		return (1);
197 	return (0);
198 }
199 
200 /*
201  * Initialize signal-related data structures.
202  */
203 void
204 signal_init(void)
205 {
206 	proc_stop_si = softintr_establish(IPL_SOFTCLOCK, proc_stop_sweep,
207 	    NULL);
208 	if (proc_stop_si == NULL)
209 		panic("signal_init failed to register softintr");
210 
211 	pool_init(&sigacts_pool, sizeof(struct sigacts), 0, IPL_NONE,
212 	    PR_WAITOK, "sigapl", NULL);
213 }
214 
215 /*
216  * Initialize a new sigaltstack structure.
217  */
218 void
219 sigstkinit(struct sigaltstack *ss)
220 {
221 	ss->ss_flags = SS_DISABLE;
222 	ss->ss_size = 0;
223 	ss->ss_sp = NULL;
224 }
225 
226 /*
227  * Create an initial sigacts structure, using the same signal state
228  * as pr.
229  */
230 struct sigacts *
231 sigactsinit(struct process *pr)
232 {
233 	struct sigacts *ps;
234 
235 	ps = pool_get(&sigacts_pool, PR_WAITOK);
236 	memcpy(ps, pr->ps_sigacts, sizeof(struct sigacts));
237 	return (ps);
238 }
239 
240 /*
241  * Release a sigacts structure.
242  */
243 void
244 sigactsfree(struct sigacts *ps)
245 {
246 	pool_put(&sigacts_pool, ps);
247 }
248 
249 int
250 sys_sigaction(struct proc *p, void *v, register_t *retval)
251 {
252 	struct sys_sigaction_args /* {
253 		syscallarg(int) signum;
254 		syscallarg(const struct sigaction *) nsa;
255 		syscallarg(struct sigaction *) osa;
256 	} */ *uap = v;
257 	struct sigaction vec;
258 #ifdef KTRACE
259 	struct sigaction ovec;
260 #endif
261 	struct sigaction *sa;
262 	const struct sigaction *nsa;
263 	struct sigaction *osa;
264 	struct sigacts *ps = p->p_p->ps_sigacts;
265 	int signum;
266 	int bit, error;
267 
268 	signum = SCARG(uap, signum);
269 	nsa = SCARG(uap, nsa);
270 	osa = SCARG(uap, osa);
271 
272 	if (signum <= 0 || signum >= NSIG ||
273 	    (nsa && (signum == SIGKILL || signum == SIGSTOP)))
274 		return (EINVAL);
275 	sa = &vec;
276 	if (osa) {
277 		mtx_enter(&p->p_p->ps_mtx);
278 		sa->sa_handler = ps->ps_sigact[signum];
279 		sa->sa_mask = ps->ps_catchmask[signum];
280 		bit = sigmask(signum);
281 		sa->sa_flags = 0;
282 		if ((ps->ps_sigonstack & bit) != 0)
283 			sa->sa_flags |= SA_ONSTACK;
284 		if ((ps->ps_sigintr & bit) == 0)
285 			sa->sa_flags |= SA_RESTART;
286 		if ((ps->ps_sigreset & bit) != 0)
287 			sa->sa_flags |= SA_RESETHAND;
288 		if ((ps->ps_siginfo & bit) != 0)
289 			sa->sa_flags |= SA_SIGINFO;
290 		if (signum == SIGCHLD) {
291 			if ((ps->ps_sigflags & SAS_NOCLDSTOP) != 0)
292 				sa->sa_flags |= SA_NOCLDSTOP;
293 			if ((ps->ps_sigflags & SAS_NOCLDWAIT) != 0)
294 				sa->sa_flags |= SA_NOCLDWAIT;
295 		}
296 		mtx_leave(&p->p_p->ps_mtx);
297 		if ((sa->sa_mask & bit) == 0)
298 			sa->sa_flags |= SA_NODEFER;
299 		sa->sa_mask &= ~bit;
300 		error = copyout(sa, osa, sizeof (vec));
301 		if (error)
302 			return (error);
303 #ifdef KTRACE
304 		if (KTRPOINT(p, KTR_STRUCT))
305 			ovec = vec;
306 #endif
307 	}
308 	if (nsa) {
309 		error = copyin(nsa, sa, sizeof (vec));
310 		if (error)
311 			return (error);
312 #ifdef KTRACE
313 		if (KTRPOINT(p, KTR_STRUCT))
314 			ktrsigaction(p, sa);
315 #endif
316 		setsigvec(p, signum, sa);
317 	}
318 #ifdef KTRACE
319 	if (osa && KTRPOINT(p, KTR_STRUCT))
320 		ktrsigaction(p, &ovec);
321 #endif
322 	return (0);
323 }
324 
325 void
326 setsigvec(struct proc *p, int signum, struct sigaction *sa)
327 {
328 	struct sigacts *ps = p->p_p->ps_sigacts;
329 	int bit;
330 
331 	bit = sigmask(signum);
332 
333 	mtx_enter(&p->p_p->ps_mtx);
334 	ps->ps_sigact[signum] = sa->sa_handler;
335 	if ((sa->sa_flags & SA_NODEFER) == 0)
336 		sa->sa_mask |= sigmask(signum);
337 	ps->ps_catchmask[signum] = sa->sa_mask &~ sigcantmask;
338 	if (signum == SIGCHLD) {
339 		if (sa->sa_flags & SA_NOCLDSTOP)
340 			atomic_setbits_int(&ps->ps_sigflags, SAS_NOCLDSTOP);
341 		else
342 			atomic_clearbits_int(&ps->ps_sigflags, SAS_NOCLDSTOP);
343 		/*
344 		 * If the SA_NOCLDWAIT flag is set or the handler
345 		 * is SIG_IGN we reparent the dying child to PID 1
346 		 * (init) which will reap the zombie.  Because we use
347 		 * init to do our dirty work we never set SAS_NOCLDWAIT
348 		 * for PID 1.
349 		 * XXX exit1 rework means this is unnecessary?
350 		 */
351 		if (initprocess->ps_sigacts != ps &&
352 		    ((sa->sa_flags & SA_NOCLDWAIT) ||
353 		    sa->sa_handler == SIG_IGN))
354 			atomic_setbits_int(&ps->ps_sigflags, SAS_NOCLDWAIT);
355 		else
356 			atomic_clearbits_int(&ps->ps_sigflags, SAS_NOCLDWAIT);
357 	}
358 	if ((sa->sa_flags & SA_RESETHAND) != 0)
359 		ps->ps_sigreset |= bit;
360 	else
361 		ps->ps_sigreset &= ~bit;
362 	if ((sa->sa_flags & SA_SIGINFO) != 0)
363 		ps->ps_siginfo |= bit;
364 	else
365 		ps->ps_siginfo &= ~bit;
366 	if ((sa->sa_flags & SA_RESTART) == 0)
367 		ps->ps_sigintr |= bit;
368 	else
369 		ps->ps_sigintr &= ~bit;
370 	if ((sa->sa_flags & SA_ONSTACK) != 0)
371 		ps->ps_sigonstack |= bit;
372 	else
373 		ps->ps_sigonstack &= ~bit;
374 	/*
375 	 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
376 	 * and for signals set to SIG_DFL where the default is to ignore.
377 	 * However, don't put SIGCONT in ps_sigignore,
378 	 * as we have to restart the process.
379 	 */
380 	if (sa->sa_handler == SIG_IGN ||
381 	    (sigprop[signum] & SA_IGNORE && sa->sa_handler == SIG_DFL)) {
382 		atomic_clearbits_int(&p->p_siglist, bit);
383 		atomic_clearbits_int(&p->p_p->ps_siglist, bit);
384 		if (signum != SIGCONT)
385 			ps->ps_sigignore |= bit;	/* easier in psignal */
386 		ps->ps_sigcatch &= ~bit;
387 	} else {
388 		ps->ps_sigignore &= ~bit;
389 		if (sa->sa_handler == SIG_DFL)
390 			ps->ps_sigcatch &= ~bit;
391 		else
392 			ps->ps_sigcatch |= bit;
393 	}
394 	mtx_leave(&p->p_p->ps_mtx);
395 }
396 
397 /*
398  * Initialize signal state for process 0;
399  * set to ignore signals that are ignored by default.
400  */
401 void
402 siginit(struct sigacts *ps)
403 {
404 	int i;
405 
406 	for (i = 0; i < NSIG; i++)
407 		if (sigprop[i] & SA_IGNORE && i != SIGCONT)
408 			ps->ps_sigignore |= sigmask(i);
409 	ps->ps_sigflags = SAS_NOCLDWAIT | SAS_NOCLDSTOP;
410 }
411 
412 /*
413  * Reset signals for an exec by the specified thread.
414  */
415 void
416 execsigs(struct proc *p)
417 {
418 	struct sigacts *ps;
419 	int nc, mask;
420 
421 	ps = p->p_p->ps_sigacts;
422 	mtx_enter(&p->p_p->ps_mtx);
423 
424 	/*
425 	 * Reset caught signals.  Held signals remain held
426 	 * through p_sigmask (unless they were caught,
427 	 * and are now ignored by default).
428 	 */
429 	while (ps->ps_sigcatch) {
430 		nc = ffs((long)ps->ps_sigcatch);
431 		mask = sigmask(nc);
432 		ps->ps_sigcatch &= ~mask;
433 		if (sigprop[nc] & SA_IGNORE) {
434 			if (nc != SIGCONT)
435 				ps->ps_sigignore |= mask;
436 			atomic_clearbits_int(&p->p_siglist, mask);
437 			atomic_clearbits_int(&p->p_p->ps_siglist, mask);
438 		}
439 		ps->ps_sigact[nc] = SIG_DFL;
440 	}
441 	/*
442 	 * Reset stack state to the user stack.
443 	 * Clear set of signals caught on the signal stack.
444 	 */
445 	sigstkinit(&p->p_sigstk);
446 	atomic_clearbits_int(&ps->ps_sigflags, SAS_NOCLDWAIT);
447 	if (ps->ps_sigact[SIGCHLD] == SIG_IGN)
448 		ps->ps_sigact[SIGCHLD] = SIG_DFL;
449 	mtx_leave(&p->p_p->ps_mtx);
450 }
451 
452 /*
453  * Manipulate signal mask.
454  * Note that we receive new mask, not pointer,
455  * and return old mask as return value;
456  * the library stub does the rest.
457  */
458 int
459 sys_sigprocmask(struct proc *p, void *v, register_t *retval)
460 {
461 	struct sys_sigprocmask_args /* {
462 		syscallarg(int) how;
463 		syscallarg(sigset_t) mask;
464 	} */ *uap = v;
465 	int error = 0;
466 	sigset_t mask;
467 
468 	KASSERT(p == curproc);
469 
470 	*retval = p->p_sigmask;
471 	mask = SCARG(uap, mask) &~ sigcantmask;
472 
473 	switch (SCARG(uap, how)) {
474 	case SIG_BLOCK:
475 		atomic_setbits_int(&p->p_sigmask, mask);
476 		break;
477 	case SIG_UNBLOCK:
478 		atomic_clearbits_int(&p->p_sigmask, mask);
479 		break;
480 	case SIG_SETMASK:
481 		p->p_sigmask = mask;
482 		break;
483 	default:
484 		error = EINVAL;
485 		break;
486 	}
487 	return (error);
488 }
489 
490 int
491 sys_sigpending(struct proc *p, void *v, register_t *retval)
492 {
493 	*retval = p->p_siglist | p->p_p->ps_siglist;
494 	return (0);
495 }
496 
497 /*
498  * Temporarily replace calling proc's signal mask for the duration of a
499  * system call.  Original signal mask will be restored by userret().
500  */
501 void
502 dosigsuspend(struct proc *p, sigset_t newmask)
503 {
504 	KASSERT(p == curproc);
505 
506 	p->p_oldmask = p->p_sigmask;
507 	atomic_setbits_int(&p->p_flag, P_SIGSUSPEND);
508 	p->p_sigmask = newmask;
509 }
510 
511 /*
512  * Suspend thread until signal, providing mask to be set
513  * in the meantime.  Note nonstandard calling convention:
514  * libc stub passes mask, not pointer, to save a copyin.
515  */
516 int
517 sys_sigsuspend(struct proc *p, void *v, register_t *retval)
518 {
519 	struct sys_sigsuspend_args /* {
520 		syscallarg(int) mask;
521 	} */ *uap = v;
522 
523 	dosigsuspend(p, SCARG(uap, mask) &~ sigcantmask);
524 	while (tsleep_nsec(&nowake, PPAUSE|PCATCH, "sigsusp", INFSLP) == 0)
525 		continue;
526 	/* always return EINTR rather than ERESTART... */
527 	return (EINTR);
528 }
529 
530 int
531 sigonstack(size_t stack)
532 {
533 	const struct sigaltstack *ss = &curproc->p_sigstk;
534 
535 	return (ss->ss_flags & SS_DISABLE ? 0 :
536 	    (stack - (size_t)ss->ss_sp < ss->ss_size));
537 }
538 
539 int
540 sys_sigaltstack(struct proc *p, void *v, register_t *retval)
541 {
542 	struct sys_sigaltstack_args /* {
543 		syscallarg(const struct sigaltstack *) nss;
544 		syscallarg(struct sigaltstack *) oss;
545 	} */ *uap = v;
546 	struct sigaltstack ss;
547 	const struct sigaltstack *nss;
548 	struct sigaltstack *oss;
549 	int onstack = sigonstack(PROC_STACK(p));
550 	int error;
551 
552 	nss = SCARG(uap, nss);
553 	oss = SCARG(uap, oss);
554 
555 	if (oss != NULL) {
556 		ss = p->p_sigstk;
557 		if (onstack)
558 			ss.ss_flags |= SS_ONSTACK;
559 		if ((error = copyout(&ss, oss, sizeof(ss))))
560 			return (error);
561 	}
562 	if (nss == NULL)
563 		return (0);
564 	error = copyin(nss, &ss, sizeof(ss));
565 	if (error)
566 		return (error);
567 	if (onstack)
568 		return (EPERM);
569 	if (ss.ss_flags & ~SS_DISABLE)
570 		return (EINVAL);
571 	if (ss.ss_flags & SS_DISABLE) {
572 		p->p_sigstk.ss_flags = ss.ss_flags;
573 		return (0);
574 	}
575 	if (ss.ss_size < MINSIGSTKSZ)
576 		return (ENOMEM);
577 
578 	error = uvm_map_remap_as_stack(p, (vaddr_t)ss.ss_sp, ss.ss_size);
579 	if (error)
580 		return (error);
581 
582 	p->p_sigstk = ss;
583 	return (0);
584 }
585 
586 int
587 sys_kill(struct proc *cp, void *v, register_t *retval)
588 {
589 	struct sys_kill_args /* {
590 		syscallarg(int) pid;
591 		syscallarg(int) signum;
592 	} */ *uap = v;
593 	struct process *pr;
594 	int pid = SCARG(uap, pid);
595 	int signum = SCARG(uap, signum);
596 	int error;
597 	int zombie = 0;
598 
599 	if ((error = pledge_kill(cp, pid)) != 0)
600 		return (error);
601 	if (((u_int)signum) >= NSIG)
602 		return (EINVAL);
603 	if (pid > 0) {
604 		if ((pr = prfind(pid)) == NULL) {
605 			if ((pr = zombiefind(pid)) == NULL)
606 				return (ESRCH);
607 			else
608 				zombie = 1;
609 		}
610 		if (!cansignal(cp, pr, signum))
611 			return (EPERM);
612 
613 		/* kill single process */
614 		if (signum && !zombie)
615 			prsignal(pr, signum);
616 		return (0);
617 	}
618 	switch (pid) {
619 	case -1:		/* broadcast signal */
620 		return (killpg1(cp, signum, 0, 1));
621 	case 0:			/* signal own process group */
622 		return (killpg1(cp, signum, 0, 0));
623 	default:		/* negative explicit process group */
624 		return (killpg1(cp, signum, -pid, 0));
625 	}
626 }
627 
628 int
629 sys_thrkill(struct proc *cp, void *v, register_t *retval)
630 {
631 	struct sys_thrkill_args /* {
632 		syscallarg(pid_t) tid;
633 		syscallarg(int) signum;
634 		syscallarg(void *) tcb;
635 	} */ *uap = v;
636 	struct proc *p;
637 	int tid = SCARG(uap, tid);
638 	int signum = SCARG(uap, signum);
639 	void *tcb;
640 
641 	if (((u_int)signum) >= NSIG)
642 		return (EINVAL);
643 
644 	p = tid ? tfind_user(tid, cp->p_p) : cp;
645 	if (p == NULL)
646 		return (ESRCH);
647 
648 	/* optionally require the target thread to have the given tcb addr */
649 	tcb = SCARG(uap, tcb);
650 	if (tcb != NULL && tcb != TCB_GET(p))
651 		return (ESRCH);
652 
653 	if (signum)
654 		ptsignal(p, signum, STHREAD);
655 	return (0);
656 }
657 
658 /*
659  * Common code for kill process group/broadcast kill.
660  * cp is calling process.
661  */
662 int
663 killpg1(struct proc *cp, int signum, int pgid, int all)
664 {
665 	struct process *pr;
666 	struct pgrp *pgrp;
667 	int nfound = 0;
668 
669 	if (all) {
670 		/*
671 		 * broadcast
672 		 */
673 		LIST_FOREACH(pr, &allprocess, ps_list) {
674 			if (pr->ps_pid <= 1 ||
675 			    pr->ps_flags & (PS_SYSTEM | PS_NOBROADCASTKILL) ||
676 			    pr == cp->p_p || !cansignal(cp, pr, signum))
677 				continue;
678 			nfound++;
679 			if (signum)
680 				prsignal(pr, signum);
681 		}
682 	} else {
683 		if (pgid == 0)
684 			/*
685 			 * zero pgid means send to my process group.
686 			 */
687 			pgrp = cp->p_p->ps_pgrp;
688 		else {
689 			pgrp = pgfind(pgid);
690 			if (pgrp == NULL)
691 				return (ESRCH);
692 		}
693 		LIST_FOREACH(pr, &pgrp->pg_members, ps_pglist) {
694 			if (pr->ps_pid <= 1 || pr->ps_flags & PS_SYSTEM ||
695 			    !cansignal(cp, pr, signum))
696 				continue;
697 			nfound++;
698 			if (signum)
699 				prsignal(pr, signum);
700 		}
701 	}
702 	return (nfound ? 0 : ESRCH);
703 }
704 
705 #define CANDELIVER(uid, euid, pr) \
706 	(euid == 0 || \
707 	(uid) == (pr)->ps_ucred->cr_ruid || \
708 	(uid) == (pr)->ps_ucred->cr_svuid || \
709 	(uid) == (pr)->ps_ucred->cr_uid || \
710 	(euid) == (pr)->ps_ucred->cr_ruid || \
711 	(euid) == (pr)->ps_ucred->cr_svuid || \
712 	(euid) == (pr)->ps_ucred->cr_uid)
713 
714 #define CANSIGIO(cr, pr) \
715 	CANDELIVER((cr)->cr_ruid, (cr)->cr_uid, (pr))
716 
717 /*
718  * Send a signal to a process group.  If checktty is 1,
719  * limit to members which have a controlling terminal.
720  */
721 void
722 pgsignal(struct pgrp *pgrp, int signum, int checkctty)
723 {
724 	struct process *pr;
725 
726 	if (pgrp)
727 		LIST_FOREACH(pr, &pgrp->pg_members, ps_pglist)
728 			if (checkctty == 0 || pr->ps_flags & PS_CONTROLT)
729 				prsignal(pr, signum);
730 }
731 
732 /*
733  * Send a SIGIO or SIGURG signal to a process or process group using stored
734  * credentials rather than those of the current process.
735  */
736 void
737 pgsigio(struct sigio_ref *sir, int sig, int checkctty)
738 {
739 	struct process *pr;
740 	struct sigio *sigio;
741 
742 	if (sir->sir_sigio == NULL)
743 		return;
744 
745 	KERNEL_LOCK();
746 	mtx_enter(&sigio_lock);
747 	sigio = sir->sir_sigio;
748 	if (sigio == NULL)
749 		goto out;
750 	if (sigio->sio_pgid > 0) {
751 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc))
752 			prsignal(sigio->sio_proc, sig);
753 	} else if (sigio->sio_pgid < 0) {
754 		LIST_FOREACH(pr, &sigio->sio_pgrp->pg_members, ps_pglist) {
755 			if (CANSIGIO(sigio->sio_ucred, pr) &&
756 			    (checkctty == 0 || (pr->ps_flags & PS_CONTROLT)))
757 				prsignal(pr, sig);
758 		}
759 	}
760 out:
761 	mtx_leave(&sigio_lock);
762 	KERNEL_UNLOCK();
763 }
764 
765 /*
766  * Recalculate the signal mask and reset the signal disposition after
767  * usermode frame for delivery is formed.
768  */
769 void
770 postsig_done(struct proc *p, int signum, sigset_t catchmask, int reset)
771 {
772 	p->p_ru.ru_nsignals++;
773 	atomic_setbits_int(&p->p_sigmask, catchmask);
774 	if (reset != 0) {
775 		sigset_t mask = sigmask(signum);
776 		struct sigacts *ps = p->p_p->ps_sigacts;
777 
778 		mtx_enter(&p->p_p->ps_mtx);
779 		ps->ps_sigcatch &= ~mask;
780 		if (signum != SIGCONT && sigprop[signum] & SA_IGNORE)
781 			ps->ps_sigignore |= mask;
782 		ps->ps_sigact[signum] = SIG_DFL;
783 		mtx_leave(&p->p_p->ps_mtx);
784 	}
785 }
786 
787 /*
788  * Send a signal caused by a trap to the current thread
789  * If it will be caught immediately, deliver it with correct code.
790  * Otherwise, post it normally.
791  */
792 void
793 trapsignal(struct proc *p, int signum, u_long trapno, int code,
794     union sigval sigval)
795 {
796 	struct process *pr = p->p_p;
797 	struct sigctx ctx;
798 	int mask;
799 
800 	switch (signum) {
801 	case SIGILL:
802 	case SIGBUS:
803 	case SIGSEGV:
804 		pr->ps_acflag |= ATRAP;
805 		break;
806 	}
807 
808 	mask = sigmask(signum);
809 	setsigctx(p, signum, &ctx);
810 	if ((pr->ps_flags & PS_TRACED) == 0 && ctx.sig_catch != 0 &&
811 	    (p->p_sigmask & mask) == 0) {
812 		siginfo_t si;
813 
814 		initsiginfo(&si, signum, trapno, code, sigval);
815 #ifdef KTRACE
816 		if (KTRPOINT(p, KTR_PSIG)) {
817 			ktrpsig(p, signum, ctx.sig_action,
818 			    p->p_sigmask, code, &si);
819 		}
820 #endif
821 		if (sendsig(ctx.sig_action, signum, p->p_sigmask, &si,
822 		    ctx.sig_info, ctx.sig_onstack)) {
823 			KERNEL_LOCK();
824 			sigexit(p, SIGILL);
825 			/* NOTREACHED */
826 		}
827 		postsig_done(p, signum, ctx.sig_catchmask, ctx.sig_reset);
828 	} else {
829 		p->p_sisig = signum;
830 		p->p_sitrapno = trapno;	/* XXX for core dump/debugger */
831 		p->p_sicode = code;
832 		p->p_sigval = sigval;
833 
834 		/*
835 		 * If traced, stop if signal is masked, and stay stopped
836 		 * until released by the debugger.  If our parent process
837 		 * is waiting for us, don't hang as we could deadlock.
838 		 */
839 		if (((pr->ps_flags & (PS_TRACED | PS_PPWAIT)) == PS_TRACED) &&
840 		    signum != SIGKILL && (p->p_sigmask & mask) != 0) {
841 			int s;
842 
843 			single_thread_set(p, SINGLE_SUSPEND | SINGLE_NOWAIT);
844 			pr->ps_xsig = signum;
845 
846 			SCHED_LOCK(s);
847 			proc_stop(p, 1);
848 			SCHED_UNLOCK(s);
849 
850 			signum = pr->ps_xsig;
851 			single_thread_clear(p, 0);
852 
853 			/*
854 			 * If we are no longer being traced, or the parent
855 			 * didn't give us a signal, skip sending the signal.
856 			 */
857 			if ((pr->ps_flags & PS_TRACED) == 0 ||
858 			    signum == 0)
859 				return;
860 
861 			/* update signal info */
862 			p->p_sisig = signum;
863 			mask = sigmask(signum);
864 		}
865 
866 		/*
867 		 * Signals like SIGBUS and SIGSEGV should not, when
868 		 * generated by the kernel, be ignorable or blockable.
869 		 * If it is and we're not being traced, then just kill
870 		 * the process.
871 		 * After vfs_shutdown(9), init(8) cannot receive signals
872 		 * because new code pages of the signal handler cannot be
873 		 * mapped from halted storage.  init(8) may not die or the
874 		 * kernel panics.  Better loop between signal handler and
875 		 * page fault trap until the machine is halted.
876 		 */
877 		if ((pr->ps_flags & PS_TRACED) == 0 &&
878 		    (sigprop[signum] & SA_KILL) &&
879 		    ((p->p_sigmask & mask) || ctx.sig_ignore) &&
880 		    pr->ps_pid != 1) {
881 			KERNEL_LOCK();
882 			sigexit(p, signum);
883 			/* NOTREACHED */
884 		}
885 		KERNEL_LOCK();
886 		ptsignal(p, signum, STHREAD);
887 		KERNEL_UNLOCK();
888 	}
889 }
890 
891 /*
892  * Send the signal to the process.  If the signal has an action, the action
893  * is usually performed by the target process rather than the caller; we add
894  * the signal to the set of pending signals for the process.
895  *
896  * Exceptions:
897  *   o When a stop signal is sent to a sleeping process that takes the
898  *     default action, the process is stopped without awakening it.
899  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
900  *     regardless of the signal action (eg, blocked or ignored).
901  *
902  * Other ignored signals are discarded immediately.
903  */
904 void
905 psignal(struct proc *p, int signum)
906 {
907 	ptsignal(p, signum, SPROCESS);
908 }
909 
910 /*
911  * type = SPROCESS	process signal, can be diverted (sigwait())
912  * type = STHREAD	thread signal, but should be propagated if unhandled
913  * type = SPROPAGATED	propagated to this thread, so don't propagate again
914  */
915 void
916 ptsignal(struct proc *p, int signum, enum signal_type type)
917 {
918 	int s, prop;
919 	sig_t action;
920 	int mask;
921 	int *siglist;
922 	struct process *pr = p->p_p;
923 	struct proc *q;
924 	int wakeparent = 0;
925 
926 	KERNEL_ASSERT_LOCKED();
927 
928 #ifdef DIAGNOSTIC
929 	if ((u_int)signum >= NSIG || signum == 0)
930 		panic("psignal signal number");
931 #endif
932 
933 	/* Ignore signal if the target process is exiting */
934 	if (pr->ps_flags & PS_EXITING)
935 		return;
936 
937 	mask = sigmask(signum);
938 
939 	if (type == SPROCESS) {
940 		/* Accept SIGKILL to coredumping processes */
941 		if (pr->ps_flags & PS_COREDUMP && signum == SIGKILL) {
942 			atomic_setbits_int(&pr->ps_siglist, mask);
943 			return;
944 		}
945 
946 		/*
947 		 * If the current thread can process the signal
948 		 * immediately (it's unblocked) then have it take it.
949 		 */
950 		q = curproc;
951 		if (q != NULL && q->p_p == pr && (q->p_flag & P_WEXIT) == 0 &&
952 		    (q->p_sigmask & mask) == 0)
953 			p = q;
954 		else {
955 			/*
956 			 * A process-wide signal can be diverted to a
957 			 * different thread that's in sigwait() for this
958 			 * signal.  If there isn't such a thread, then
959 			 * pick a thread that doesn't have it blocked so
960 			 * that the stop/kill consideration isn't
961 			 * delayed.  Otherwise, mark it pending on the
962 			 * main thread.
963 			 */
964 			TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
965 				/* ignore exiting threads */
966 				if (q->p_flag & P_WEXIT)
967 					continue;
968 
969 				/* skip threads that have the signal blocked */
970 				if ((q->p_sigmask & mask) != 0)
971 					continue;
972 
973 				/* okay, could send to this thread */
974 				p = q;
975 
976 				/*
977 				 * sigsuspend, sigwait, ppoll/pselect, etc?
978 				 * Definitely go to this thread, as it's
979 				 * already blocked in the kernel.
980 				 */
981 				if (q->p_flag & P_SIGSUSPEND)
982 					break;
983 			}
984 		}
985 	}
986 
987 	if (type != SPROPAGATED)
988 		knote_locked(&pr->ps_klist, NOTE_SIGNAL | signum);
989 
990 	prop = sigprop[signum];
991 
992 	/*
993 	 * If proc is traced, always give parent a chance.
994 	 */
995 	if (pr->ps_flags & PS_TRACED) {
996 		action = SIG_DFL;
997 	} else {
998 		sigset_t sigcatch, sigignore;
999 
1000 		/*
1001 		 * If the signal is being ignored,
1002 		 * then we forget about it immediately.
1003 		 * (Note: we don't set SIGCONT in ps_sigignore,
1004 		 * and if it is set to SIG_IGN,
1005 		 * action will be SIG_DFL here.)
1006 		 */
1007 		mtx_enter(&pr->ps_mtx);
1008 		sigignore = pr->ps_sigacts->ps_sigignore;
1009 		sigcatch = pr->ps_sigacts->ps_sigcatch;
1010 		mtx_leave(&pr->ps_mtx);
1011 
1012 		if (sigignore & mask)
1013 			return;
1014 		if (p->p_sigmask & mask) {
1015 			action = SIG_HOLD;
1016 		} else if (sigcatch & mask) {
1017 			action = SIG_CATCH;
1018 		} else {
1019 			action = SIG_DFL;
1020 
1021 			if (prop & SA_KILL && pr->ps_nice > NZERO)
1022 				 pr->ps_nice = NZERO;
1023 
1024 			/*
1025 			 * If sending a tty stop signal to a member of an
1026 			 * orphaned process group, discard the signal here if
1027 			 * the action is default; don't stop the process below
1028 			 * if sleeping, and don't clear any pending SIGCONT.
1029 			 */
1030 			if (prop & SA_TTYSTOP && pr->ps_pgrp->pg_jobc == 0)
1031 				return;
1032 		}
1033 	}
1034 	/*
1035 	 * If delivered to process, mark as pending there.  Continue and stop
1036 	 * signals will be propagated to all threads.  So they are always
1037 	 * marked at thread level.
1038 	 */
1039 	siglist = (type == SPROCESS) ? &pr->ps_siglist : &p->p_siglist;
1040 	if (prop & SA_CONT) {
1041 		siglist = &p->p_siglist;
1042 		atomic_clearbits_int(siglist, STOPSIGMASK);
1043 	}
1044 	if (prop & SA_STOP) {
1045 		siglist = &p->p_siglist;
1046 		atomic_clearbits_int(siglist, CONTSIGMASK);
1047 		atomic_clearbits_int(&p->p_flag, P_CONTINUED);
1048 	}
1049 
1050 	/*
1051 	 * XXX delay processing of SA_STOP signals unless action == SIG_DFL?
1052 	 */
1053 	if (prop & (SA_CONT | SA_STOP) && type != SPROPAGATED)
1054 		TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link)
1055 			if (q != p)
1056 				ptsignal(q, signum, SPROPAGATED);
1057 
1058 	/*
1059 	 * Defer further processing for signals which are held,
1060 	 * except that stopped processes must be continued by SIGCONT.
1061 	 */
1062 	if (action == SIG_HOLD && ((prop & SA_CONT) == 0 ||
1063 	    p->p_stat != SSTOP)) {
1064 		atomic_setbits_int(siglist, mask);
1065 		return;
1066 	}
1067 
1068 	SCHED_LOCK(s);
1069 
1070 	switch (p->p_stat) {
1071 
1072 	case SSLEEP:
1073 		/*
1074 		 * If process is sleeping uninterruptibly
1075 		 * we can't interrupt the sleep... the signal will
1076 		 * be noticed when the process returns through
1077 		 * trap() or syscall().
1078 		 */
1079 		if ((p->p_flag & P_SINTR) == 0)
1080 			goto out;
1081 		/*
1082 		 * Process is sleeping and traced... make it runnable
1083 		 * so it can discover the signal in cursig() and stop
1084 		 * for the parent.
1085 		 */
1086 		if (pr->ps_flags & PS_TRACED)
1087 			goto run;
1088 		/*
1089 		 * If SIGCONT is default (or ignored) and process is
1090 		 * asleep, we are finished; the process should not
1091 		 * be awakened.
1092 		 */
1093 		if ((prop & SA_CONT) && action == SIG_DFL) {
1094 			mask = 0;
1095 			goto out;
1096 		}
1097 		/*
1098 		 * When a sleeping process receives a stop
1099 		 * signal, process immediately if possible.
1100 		 */
1101 		if ((prop & SA_STOP) && action == SIG_DFL) {
1102 			/*
1103 			 * If a child holding parent blocked,
1104 			 * stopping could cause deadlock.
1105 			 */
1106 			if (pr->ps_flags & PS_PPWAIT)
1107 				goto out;
1108 			mask = 0;
1109 			pr->ps_xsig = signum;
1110 			proc_stop(p, 0);
1111 			goto out;
1112 		}
1113 		/*
1114 		 * All other (caught or default) signals
1115 		 * cause the process to run.
1116 		 */
1117 		goto runfast;
1118 		/* NOTREACHED */
1119 
1120 	case SSTOP:
1121 		/*
1122 		 * If traced process is already stopped,
1123 		 * then no further action is necessary.
1124 		 */
1125 		if (pr->ps_flags & PS_TRACED)
1126 			goto out;
1127 
1128 		/*
1129 		 * Kill signal always sets processes running.
1130 		 */
1131 		if (signum == SIGKILL) {
1132 			atomic_clearbits_int(&p->p_flag, P_SUSPSIG);
1133 			goto runfast;
1134 		}
1135 
1136 		if (prop & SA_CONT) {
1137 			/*
1138 			 * If SIGCONT is default (or ignored), we continue the
1139 			 * process but don't leave the signal in p_siglist, as
1140 			 * it has no further action.  If SIGCONT is held, we
1141 			 * continue the process and leave the signal in
1142 			 * p_siglist.  If the process catches SIGCONT, let it
1143 			 * handle the signal itself.  If it isn't waiting on
1144 			 * an event, then it goes back to run state.
1145 			 * Otherwise, process goes back to sleep state.
1146 			 */
1147 			atomic_setbits_int(&p->p_flag, P_CONTINUED);
1148 			atomic_clearbits_int(&p->p_flag, P_SUSPSIG);
1149 			wakeparent = 1;
1150 			if (action == SIG_DFL)
1151 				atomic_clearbits_int(siglist, mask);
1152 			if (action == SIG_CATCH)
1153 				goto runfast;
1154 			if (p->p_wchan == NULL)
1155 				goto run;
1156 			atomic_clearbits_int(&p->p_flag, P_WSLEEP);
1157 			p->p_stat = SSLEEP;
1158 			goto out;
1159 		}
1160 
1161 		if (prop & SA_STOP) {
1162 			/*
1163 			 * Already stopped, don't need to stop again.
1164 			 * (If we did the shell could get confused.)
1165 			 */
1166 			mask = 0;
1167 			goto out;
1168 		}
1169 
1170 		/*
1171 		 * If process is sleeping interruptibly, then simulate a
1172 		 * wakeup so that when it is continued, it will be made
1173 		 * runnable and can look at the signal.  But don't make
1174 		 * the process runnable, leave it stopped.
1175 		 */
1176 		if (p->p_flag & P_SINTR)
1177 			unsleep(p);
1178 		goto out;
1179 
1180 	case SONPROC:
1181 		/* set siglist before issuing the ast */
1182 		atomic_setbits_int(siglist, mask);
1183 		mask = 0;
1184 		signotify(p);
1185 		/* FALLTHROUGH */
1186 	default:
1187 		/*
1188 		 * SRUN, SIDL, SDEAD do nothing with the signal,
1189 		 * other than kicking ourselves if we are running.
1190 		 * It will either never be noticed, or noticed very soon.
1191 		 */
1192 		goto out;
1193 	}
1194 	/* NOTREACHED */
1195 
1196 runfast:
1197 	/*
1198 	 * Raise priority to at least PUSER.
1199 	 */
1200 	if (p->p_usrpri > PUSER)
1201 		p->p_usrpri = PUSER;
1202 run:
1203 	setrunnable(p);
1204 out:
1205 	/* finally adjust siglist */
1206 	if (mask)
1207 		atomic_setbits_int(siglist, mask);
1208 	SCHED_UNLOCK(s);
1209 	if (wakeparent)
1210 		wakeup(pr->ps_pptr);
1211 }
1212 
1213 /* fill the signal context which should be used by postsig() and issignal() */
1214 void
1215 setsigctx(struct proc *p, int signum, struct sigctx *sctx)
1216 {
1217 	struct sigacts *ps = p->p_p->ps_sigacts;
1218 	sigset_t mask;
1219 
1220 	mtx_enter(&p->p_p->ps_mtx);
1221 	mask = sigmask(signum);
1222 	sctx->sig_action = ps->ps_sigact[signum];
1223 	sctx->sig_catchmask = ps->ps_catchmask[signum];
1224 	sctx->sig_reset = (ps->ps_sigreset & mask) != 0;
1225 	sctx->sig_info = (ps->ps_siginfo & mask) != 0;
1226 	sctx->sig_intr = (ps->ps_sigintr & mask) != 0;
1227 	sctx->sig_onstack = (ps->ps_sigonstack & mask) != 0;
1228 	sctx->sig_ignore = (ps->ps_sigignore & mask) != 0;
1229 	sctx->sig_catch = (ps->ps_sigcatch & mask) != 0;
1230 	mtx_leave(&p->p_p->ps_mtx);
1231 }
1232 
1233 /*
1234  * Determine signal that should be delivered to process p, the current
1235  * process, 0 if none.
1236  *
1237  * If the current process has received a signal (should be caught or cause
1238  * termination, should interrupt current syscall), return the signal number.
1239  * Stop signals with default action are processed immediately, then cleared;
1240  * they aren't returned.  This is checked after each entry to the system for
1241  * a syscall or trap. The normal call sequence is
1242  *
1243  *	while (signum = cursig(curproc, &ctx))
1244  *		postsig(signum, &ctx);
1245  *
1246  * Assumes that if the P_SINTR flag is set, we're holding both the
1247  * kernel and scheduler locks.
1248  */
1249 int
1250 cursig(struct proc *p, struct sigctx *sctx)
1251 {
1252 	struct process *pr = p->p_p;
1253 	int signum, mask, prop;
1254 	sigset_t ps_siglist;
1255 	int s;
1256 
1257 	KASSERT(p == curproc);
1258 
1259 	for (;;) {
1260 		ps_siglist = READ_ONCE(pr->ps_siglist);
1261 		membar_consumer();
1262 		mask = SIGPENDING(p);
1263 		if (pr->ps_flags & PS_PPWAIT)
1264 			mask &= ~STOPSIGMASK;
1265 		if (mask == 0)	 	/* no signal to send */
1266 			return (0);
1267 		signum = ffs((long)mask);
1268 		mask = sigmask(signum);
1269 
1270 		/* take the signal! */
1271 		if (atomic_cas_uint(&pr->ps_siglist, ps_siglist,
1272 		    ps_siglist & ~mask) != ps_siglist) {
1273 			/* lost race taking the process signal, restart */
1274 			continue;
1275 		}
1276 		atomic_clearbits_int(&p->p_siglist, mask);
1277 		setsigctx(p, signum, sctx);
1278 
1279 		/*
1280 		 * We should see pending but ignored signals
1281 		 * only if PS_TRACED was on when they were posted.
1282 		 */
1283 		if (sctx->sig_ignore && (pr->ps_flags & PS_TRACED) == 0)
1284 			continue;
1285 
1286 		/*
1287 		 * If traced, always stop, and stay stopped until released
1288 		 * by the debugger.  If our parent process is waiting for
1289 		 * us, don't hang as we could deadlock.
1290 		 */
1291 		if (((pr->ps_flags & (PS_TRACED | PS_PPWAIT)) == PS_TRACED) &&
1292 		    signum != SIGKILL) {
1293 			single_thread_set(p, SINGLE_SUSPEND | SINGLE_NOWAIT);
1294 			pr->ps_xsig = signum;
1295 
1296 			SCHED_LOCK(s);
1297 			proc_stop(p, 1);
1298 			SCHED_UNLOCK(s);
1299 
1300 			/*
1301 			 * re-take the signal before releasing
1302 			 * the other threads. Must check the continue
1303 			 * conditions below and only take the signal if
1304 			 * those are not true.
1305 			 */
1306 			signum = pr->ps_xsig;
1307 			mask = sigmask(signum);
1308 			setsigctx(p, signum, sctx);
1309 			if (!((pr->ps_flags & PS_TRACED) == 0 ||
1310 			    signum == 0 ||
1311 			    (p->p_sigmask & mask) != 0)) {
1312 				atomic_clearbits_int(&p->p_siglist, mask);
1313 				atomic_clearbits_int(&pr->ps_siglist, mask);
1314 			}
1315 
1316 			single_thread_clear(p, 0);
1317 
1318 			/*
1319 			 * If we are no longer being traced, or the parent
1320 			 * didn't give us a signal, look for more signals.
1321 			 */
1322 			if ((pr->ps_flags & PS_TRACED) == 0 ||
1323 			    signum == 0)
1324 				continue;
1325 
1326 			/*
1327 			 * If the new signal is being masked, look for other
1328 			 * signals.
1329 			 */
1330 			if ((p->p_sigmask & mask) != 0)
1331 				continue;
1332 
1333 		}
1334 
1335 		prop = sigprop[signum];
1336 
1337 		/*
1338 		 * Decide whether the signal should be returned.
1339 		 * Return the signal's number, or fall through
1340 		 * to clear it from the pending mask.
1341 		 */
1342 		switch ((long)sctx->sig_action) {
1343 		case (long)SIG_DFL:
1344 			/*
1345 			 * Don't take default actions on system processes.
1346 			 */
1347 			if (pr->ps_pid <= 1) {
1348 #ifdef DIAGNOSTIC
1349 				/*
1350 				 * Are you sure you want to ignore SIGSEGV
1351 				 * in init? XXX
1352 				 */
1353 				printf("Process (pid %d) got signal"
1354 				    " %d\n", pr->ps_pid, signum);
1355 #endif
1356 				break;		/* == ignore */
1357 			}
1358 			/*
1359 			 * If there is a pending stop signal to process
1360 			 * with default action, stop here,
1361 			 * then clear the signal.  However,
1362 			 * if process is member of an orphaned
1363 			 * process group, ignore tty stop signals.
1364 			 */
1365 			if (prop & SA_STOP) {
1366 				if (pr->ps_flags & PS_TRACED ||
1367 		    		    (pr->ps_pgrp->pg_jobc == 0 &&
1368 				    prop & SA_TTYSTOP))
1369 					break;	/* == ignore */
1370 				pr->ps_xsig = signum;
1371 				SCHED_LOCK(s);
1372 				proc_stop(p, 1);
1373 				SCHED_UNLOCK(s);
1374 				break;
1375 			} else if (prop & SA_IGNORE) {
1376 				/*
1377 				 * Except for SIGCONT, shouldn't get here.
1378 				 * Default action is to ignore; drop it.
1379 				 */
1380 				break;		/* == ignore */
1381 			} else
1382 				goto keep;
1383 			/* NOTREACHED */
1384 		case (long)SIG_IGN:
1385 			/*
1386 			 * Masking above should prevent us ever trying
1387 			 * to take action on an ignored signal other
1388 			 * than SIGCONT, unless process is traced.
1389 			 */
1390 			if ((prop & SA_CONT) == 0 &&
1391 			    (pr->ps_flags & PS_TRACED) == 0)
1392 				printf("%s\n", __func__);
1393 			break;		/* == ignore */
1394 		default:
1395 			/*
1396 			 * This signal has an action, let
1397 			 * postsig() process it.
1398 			 */
1399 			goto keep;
1400 		}
1401 	}
1402 	/* NOTREACHED */
1403 
1404 keep:
1405 	atomic_setbits_int(&p->p_siglist, mask); /*leave the signal for later */
1406 	return (signum);
1407 }
1408 
1409 /*
1410  * Put the argument process into the stopped state and notify the parent
1411  * via wakeup.  Signals are handled elsewhere.  The process must not be
1412  * on the run queue.
1413  */
1414 void
1415 proc_stop(struct proc *p, int sw)
1416 {
1417 	struct process *pr = p->p_p;
1418 
1419 #ifdef MULTIPROCESSOR
1420 	SCHED_ASSERT_LOCKED();
1421 #endif
1422 
1423 	p->p_stat = SSTOP;
1424 	atomic_clearbits_int(&pr->ps_flags, PS_WAITED);
1425 	atomic_setbits_int(&pr->ps_flags, PS_STOPPED);
1426 	atomic_setbits_int(&p->p_flag, P_SUSPSIG);
1427 	/*
1428 	 * We need this soft interrupt to be handled fast.
1429 	 * Extra calls to softclock don't hurt.
1430 	 */
1431 	softintr_schedule(proc_stop_si);
1432 	if (sw)
1433 		mi_switch();
1434 }
1435 
1436 /*
1437  * Called from a soft interrupt to send signals to the parents of stopped
1438  * processes.
1439  * We can't do this in proc_stop because it's called with nasty locks held
1440  * and we would need recursive scheduler lock to deal with that.
1441  */
1442 void
1443 proc_stop_sweep(void *v)
1444 {
1445 	struct process *pr;
1446 
1447 	LIST_FOREACH(pr, &allprocess, ps_list) {
1448 		if ((pr->ps_flags & PS_STOPPED) == 0)
1449 			continue;
1450 		atomic_clearbits_int(&pr->ps_flags, PS_STOPPED);
1451 
1452 		if ((pr->ps_pptr->ps_sigacts->ps_sigflags & SAS_NOCLDSTOP) == 0)
1453 			prsignal(pr->ps_pptr, SIGCHLD);
1454 		wakeup(pr->ps_pptr);
1455 	}
1456 }
1457 
1458 /*
1459  * Take the action for the specified signal
1460  * from the current set of pending signals.
1461  */
1462 void
1463 postsig(struct proc *p, int signum, struct sigctx *sctx)
1464 {
1465 	u_long trapno;
1466 	int mask, returnmask;
1467 	siginfo_t si;
1468 	union sigval sigval;
1469 	int code;
1470 
1471 	KASSERT(signum != 0);
1472 
1473 	mask = sigmask(signum);
1474 	atomic_clearbits_int(&p->p_siglist, mask);
1475 	sigval.sival_ptr = NULL;
1476 
1477 	if (p->p_sisig != signum) {
1478 		trapno = 0;
1479 		code = SI_USER;
1480 		sigval.sival_ptr = NULL;
1481 	} else {
1482 		trapno = p->p_sitrapno;
1483 		code = p->p_sicode;
1484 		sigval = p->p_sigval;
1485 	}
1486 	initsiginfo(&si, signum, trapno, code, sigval);
1487 
1488 #ifdef KTRACE
1489 	if (KTRPOINT(p, KTR_PSIG)) {
1490 		ktrpsig(p, signum, sctx->sig_action, p->p_flag & P_SIGSUSPEND ?
1491 		    p->p_oldmask : p->p_sigmask, code, &si);
1492 	}
1493 #endif
1494 	if (sctx->sig_action == SIG_DFL) {
1495 		/*
1496 		 * Default action, where the default is to kill
1497 		 * the process.  (Other cases were ignored above.)
1498 		 */
1499 		KERNEL_LOCK();
1500 		sigexit(p, signum);
1501 		/* NOTREACHED */
1502 	} else {
1503 		/*
1504 		 * If we get here, the signal must be caught.
1505 		 */
1506 #ifdef DIAGNOSTIC
1507 		if (sctx->sig_action == SIG_IGN || (p->p_sigmask & mask))
1508 			panic("postsig action");
1509 #endif
1510 		/*
1511 		 * Set the new mask value and also defer further
1512 		 * occurrences of this signal.
1513 		 *
1514 		 * Special case: user has done a sigpause.  Here the
1515 		 * current mask is not of interest, but rather the
1516 		 * mask from before the sigpause is what we want
1517 		 * restored after the signal processing is completed.
1518 		 */
1519 		if (p->p_flag & P_SIGSUSPEND) {
1520 			atomic_clearbits_int(&p->p_flag, P_SIGSUSPEND);
1521 			returnmask = p->p_oldmask;
1522 		} else {
1523 			returnmask = p->p_sigmask;
1524 		}
1525 		if (p->p_sisig == signum) {
1526 			p->p_sisig = 0;
1527 			p->p_sitrapno = 0;
1528 			p->p_sicode = SI_USER;
1529 			p->p_sigval.sival_ptr = NULL;
1530 		}
1531 
1532 		if (sendsig(sctx->sig_action, signum, returnmask, &si,
1533 		    sctx->sig_info, sctx->sig_onstack)) {
1534 			KERNEL_LOCK();
1535 			sigexit(p, SIGILL);
1536 			/* NOTREACHED */
1537 		}
1538 		postsig_done(p, signum, sctx->sig_catchmask, sctx->sig_reset);
1539 	}
1540 }
1541 
1542 /*
1543  * Force the current process to exit with the specified signal, dumping core
1544  * if appropriate.  We bypass the normal tests for masked and caught signals,
1545  * allowing unrecoverable failures to terminate the process without changing
1546  * signal state.  Mark the accounting record with the signal termination.
1547  * If dumping core, save the signal number for the debugger.  Calls exit and
1548  * does not return.
1549  */
1550 void
1551 sigexit(struct proc *p, int signum)
1552 {
1553 	/* Mark process as going away */
1554 	atomic_setbits_int(&p->p_flag, P_WEXIT);
1555 
1556 	p->p_p->ps_acflag |= AXSIG;
1557 	if (sigprop[signum] & SA_CORE) {
1558 		p->p_sisig = signum;
1559 
1560 		/* if there are other threads, pause them */
1561 		if (P_HASSIBLING(p))
1562 			single_thread_set(p, SINGLE_UNWIND);
1563 
1564 		if (coredump(p) == 0)
1565 			signum |= WCOREFLAG;
1566 	}
1567 	exit1(p, 0, signum, EXIT_NORMAL);
1568 	/* NOTREACHED */
1569 }
1570 
1571 /*
1572  * Send uncatchable SIGABRT for coredump.
1573  */
1574 void
1575 sigabort(struct proc *p)
1576 {
1577 	struct sigaction sa;
1578 
1579 	memset(&sa, 0, sizeof sa);
1580 	sa.sa_handler = SIG_DFL;
1581 	setsigvec(p, SIGABRT, &sa);
1582 	atomic_clearbits_int(&p->p_sigmask, sigmask(SIGABRT));
1583 	psignal(p, SIGABRT);
1584 }
1585 
1586 /*
1587  * Return 1 if `sig', a given signal, is ignored or masked for `p', a given
1588  * thread, and 0 otherwise.
1589  */
1590 int
1591 sigismasked(struct proc *p, int sig)
1592 {
1593 	struct process *pr = p->p_p;
1594 	int rv;
1595 
1596 	mtx_enter(&pr->ps_mtx);
1597 	rv = (pr->ps_sigacts->ps_sigignore & sigmask(sig)) ||
1598 	    (p->p_sigmask & sigmask(sig));
1599 	mtx_leave(&pr->ps_mtx);
1600 
1601 	return !!rv;
1602 }
1603 
1604 struct coredump_iostate {
1605 	struct proc *io_proc;
1606 	struct vnode *io_vp;
1607 	struct ucred *io_cred;
1608 	off_t io_offset;
1609 };
1610 
1611 /*
1612  * Dump core, into a file named "progname.core", unless the process was
1613  * setuid/setgid.
1614  */
1615 int
1616 coredump(struct proc *p)
1617 {
1618 #ifdef SMALL_KERNEL
1619 	return EPERM;
1620 #else
1621 	struct process *pr = p->p_p;
1622 	struct vnode *vp;
1623 	struct ucred *cred = p->p_ucred;
1624 	struct vmspace *vm = p->p_vmspace;
1625 	struct nameidata nd;
1626 	struct vattr vattr;
1627 	struct coredump_iostate	io;
1628 	int error, len, incrash = 0;
1629 	char *name;
1630 	const char *dir = "/var/crash";
1631 
1632 	atomic_setbits_int(&pr->ps_flags, PS_COREDUMP);
1633 
1634 #ifdef PMAP_CHECK_COPYIN
1635 	/* disable copyin checks, so we can write out text sections if needed */
1636 	p->p_vmspace->vm_map.check_copyin_count = 0;
1637 #endif
1638 
1639 	/* Don't dump if will exceed file size limit. */
1640 	if (USPACE + ptoa(vm->vm_dsize + vm->vm_ssize) >= lim_cur(RLIMIT_CORE))
1641 		return (EFBIG);
1642 
1643 	name = pool_get(&namei_pool, PR_WAITOK);
1644 
1645 	/*
1646 	 * If the process has inconsistent uids, nosuidcoredump
1647 	 * determines coredump placement policy.
1648 	 */
1649 	if (((pr->ps_flags & PS_SUGID) && (error = suser(p))) ||
1650 	   ((pr->ps_flags & PS_SUGID) && nosuidcoredump)) {
1651 		if (nosuidcoredump == 3) {
1652 			/*
1653 			 * If the program directory does not exist, dumps of
1654 			 * that core will silently fail.
1655 			 */
1656 			len = snprintf(name, MAXPATHLEN, "%s/%s/%u.core",
1657 			    dir, pr->ps_comm, pr->ps_pid);
1658 			incrash = KERNELPATH;
1659 		} else if (nosuidcoredump == 2) {
1660 			len = snprintf(name, MAXPATHLEN, "%s/%s.core",
1661 			    dir, pr->ps_comm);
1662 			incrash = KERNELPATH;
1663 		} else {
1664 			pool_put(&namei_pool, name);
1665 			return (EPERM);
1666 		}
1667 	} else
1668 		len = snprintf(name, MAXPATHLEN, "%s.core", pr->ps_comm);
1669 
1670 	if (len >= MAXPATHLEN) {
1671 		pool_put(&namei_pool, name);
1672 		return (EACCES);
1673 	}
1674 
1675 	/*
1676 	 * Control the UID used to write out.  The normal case uses
1677 	 * the real UID.  If the sugid case is going to write into the
1678 	 * controlled directory, we do so as root.
1679 	 */
1680 	if (incrash == 0) {
1681 		cred = crdup(cred);
1682 		cred->cr_uid = cred->cr_ruid;
1683 		cred->cr_gid = cred->cr_rgid;
1684 	} else {
1685 		if (p->p_fd->fd_rdir) {
1686 			vrele(p->p_fd->fd_rdir);
1687 			p->p_fd->fd_rdir = NULL;
1688 		}
1689 		p->p_ucred = crdup(p->p_ucred);
1690 		crfree(cred);
1691 		cred = p->p_ucred;
1692 		crhold(cred);
1693 		cred->cr_uid = 0;
1694 		cred->cr_gid = 0;
1695 	}
1696 
1697 	/* incrash should be 0 or KERNELPATH only */
1698 	NDINIT(&nd, 0, BYPASSUNVEIL | incrash, UIO_SYSSPACE, name, p);
1699 
1700 	error = vn_open(&nd, O_CREAT | FWRITE | O_NOFOLLOW | O_NONBLOCK,
1701 	    S_IRUSR | S_IWUSR);
1702 
1703 	if (error)
1704 		goto out;
1705 
1706 	/*
1707 	 * Don't dump to non-regular files, files with links, or files
1708 	 * owned by someone else.
1709 	 */
1710 	vp = nd.ni_vp;
1711 	if ((error = VOP_GETATTR(vp, &vattr, cred, p)) != 0) {
1712 		VOP_UNLOCK(vp);
1713 		vn_close(vp, FWRITE, cred, p);
1714 		goto out;
1715 	}
1716 	if (vp->v_type != VREG || vattr.va_nlink != 1 ||
1717 	    vattr.va_mode & ((VREAD | VWRITE) >> 3 | (VREAD | VWRITE) >> 6) ||
1718 	    vattr.va_uid != cred->cr_uid) {
1719 		error = EACCES;
1720 		VOP_UNLOCK(vp);
1721 		vn_close(vp, FWRITE, cred, p);
1722 		goto out;
1723 	}
1724 	VATTR_NULL(&vattr);
1725 	vattr.va_size = 0;
1726 	VOP_SETATTR(vp, &vattr, cred, p);
1727 	pr->ps_acflag |= ACORE;
1728 
1729 	io.io_proc = p;
1730 	io.io_vp = vp;
1731 	io.io_cred = cred;
1732 	io.io_offset = 0;
1733 	VOP_UNLOCK(vp);
1734 	vref(vp);
1735 	error = vn_close(vp, FWRITE, cred, p);
1736 	if (error == 0)
1737 		error = coredump_elf(p, &io);
1738 	vrele(vp);
1739 out:
1740 	crfree(cred);
1741 	pool_put(&namei_pool, name);
1742 	return (error);
1743 #endif
1744 }
1745 
1746 #ifndef SMALL_KERNEL
1747 int
1748 coredump_write(void *cookie, enum uio_seg segflg, const void *data, size_t len)
1749 {
1750 	struct coredump_iostate *io = cookie;
1751 	off_t coffset = 0;
1752 	size_t csize;
1753 	int chunk, error;
1754 
1755 	csize = len;
1756 	do {
1757 		if (sigmask(SIGKILL) &
1758 		    (io->io_proc->p_siglist | io->io_proc->p_p->ps_siglist))
1759 			return (EINTR);
1760 
1761 		/* Rest of the loop sleeps with lock held, so... */
1762 		yield();
1763 
1764 		chunk = MIN(csize, MAXPHYS);
1765 		error = vn_rdwr(UIO_WRITE, io->io_vp,
1766 		    (caddr_t)data + coffset, chunk,
1767 		    io->io_offset + coffset, segflg,
1768 		    IO_UNIT, io->io_cred, NULL, io->io_proc);
1769 		if (error) {
1770 			struct process *pr = io->io_proc->p_p;
1771 
1772 			if (error == ENOSPC)
1773 				log(LOG_ERR,
1774 				    "coredump of %s(%d) failed, filesystem full\n",
1775 				    pr->ps_comm, pr->ps_pid);
1776 			else
1777 				log(LOG_ERR,
1778 				    "coredump of %s(%d), write failed: errno %d\n",
1779 				    pr->ps_comm, pr->ps_pid, error);
1780 			return (error);
1781 		}
1782 
1783 		coffset += chunk;
1784 		csize -= chunk;
1785 	} while (csize > 0);
1786 
1787 	io->io_offset += len;
1788 	return (0);
1789 }
1790 
1791 void
1792 coredump_unmap(void *cookie, vaddr_t start, vaddr_t end)
1793 {
1794 	struct coredump_iostate *io = cookie;
1795 
1796 	uvm_unmap(&io->io_proc->p_vmspace->vm_map, start, end);
1797 }
1798 
1799 #endif	/* !SMALL_KERNEL */
1800 
1801 /*
1802  * Nonexistent system call-- signal process (may want to handle it).
1803  * Flag error in case process won't see signal immediately (blocked or ignored).
1804  */
1805 int
1806 sys_nosys(struct proc *p, void *v, register_t *retval)
1807 {
1808 	ptsignal(p, SIGSYS, STHREAD);
1809 	return (ENOSYS);
1810 }
1811 
1812 int
1813 sys___thrsigdivert(struct proc *p, void *v, register_t *retval)
1814 {
1815 	struct sys___thrsigdivert_args /* {
1816 		syscallarg(sigset_t) sigmask;
1817 		syscallarg(siginfo_t *) info;
1818 		syscallarg(const struct timespec *) timeout;
1819 	} */ *uap = v;
1820 	struct sigctx ctx;
1821 	sigset_t mask = SCARG(uap, sigmask) &~ sigcantmask;
1822 	siginfo_t si;
1823 	uint64_t nsecs = INFSLP;
1824 	int timeinvalid = 0;
1825 	int error = 0;
1826 
1827 	memset(&si, 0, sizeof(si));
1828 
1829 	if (SCARG(uap, timeout) != NULL) {
1830 		struct timespec ts;
1831 		if ((error = copyin(SCARG(uap, timeout), &ts, sizeof(ts))) != 0)
1832 			return (error);
1833 #ifdef KTRACE
1834 		if (KTRPOINT(p, KTR_STRUCT))
1835 			ktrreltimespec(p, &ts);
1836 #endif
1837 		if (!timespecisvalid(&ts))
1838 			timeinvalid = 1;
1839 		else
1840 			nsecs = TIMESPEC_TO_NSEC(&ts);
1841 	}
1842 
1843 	dosigsuspend(p, p->p_sigmask &~ mask);
1844 	for (;;) {
1845 		si.si_signo = cursig(p, &ctx);
1846 		if (si.si_signo != 0) {
1847 			sigset_t smask = sigmask(si.si_signo);
1848 			if (smask & mask) {
1849 				atomic_clearbits_int(&p->p_siglist, smask);
1850 				error = 0;
1851 				break;
1852 			}
1853 		}
1854 
1855 		/* per-POSIX, delay this error until after the above */
1856 		if (timeinvalid)
1857 			error = EINVAL;
1858 		/* per-POSIX, return immediately if timeout is zero-valued */
1859 		if (nsecs == 0)
1860 			error = EAGAIN;
1861 
1862 		if (error != 0)
1863 			break;
1864 
1865 		error = tsleep_nsec(&nowake, PPAUSE|PCATCH, "sigwait", nsecs);
1866 	}
1867 
1868 	if (error == 0) {
1869 		*retval = si.si_signo;
1870 		if (SCARG(uap, info) != NULL) {
1871 			error = copyout(&si, SCARG(uap, info), sizeof(si));
1872 #ifdef KTRACE
1873 			if (error == 0 && KTRPOINT(p, KTR_STRUCT))
1874 				ktrsiginfo(p, &si);
1875 #endif
1876 		}
1877 	} else if (error == ERESTART && SCARG(uap, timeout) != NULL) {
1878 		/*
1879 		 * Restarting is wrong if there's a timeout, as it'll be
1880 		 * for the same interval again
1881 		 */
1882 		error = EINTR;
1883 	}
1884 
1885 	return (error);
1886 }
1887 
1888 void
1889 initsiginfo(siginfo_t *si, int sig, u_long trapno, int code, union sigval val)
1890 {
1891 	memset(si, 0, sizeof(*si));
1892 
1893 	si->si_signo = sig;
1894 	si->si_code = code;
1895 	if (code == SI_USER) {
1896 		si->si_value = val;
1897 	} else {
1898 		switch (sig) {
1899 		case SIGSEGV:
1900 		case SIGILL:
1901 		case SIGBUS:
1902 		case SIGFPE:
1903 			si->si_addr = val.sival_ptr;
1904 			si->si_trapno = trapno;
1905 			break;
1906 		case SIGXFSZ:
1907 			break;
1908 		}
1909 	}
1910 }
1911 
1912 int
1913 filt_sigattach(struct knote *kn)
1914 {
1915 	struct process *pr = curproc->p_p;
1916 	int s;
1917 
1918 	if (kn->kn_id >= NSIG)
1919 		return EINVAL;
1920 
1921 	kn->kn_ptr.p_process = pr;
1922 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
1923 
1924 	s = splhigh();
1925 	klist_insert_locked(&pr->ps_klist, kn);
1926 	splx(s);
1927 
1928 	return (0);
1929 }
1930 
1931 void
1932 filt_sigdetach(struct knote *kn)
1933 {
1934 	struct process *pr = kn->kn_ptr.p_process;
1935 	int s;
1936 
1937 	s = splhigh();
1938 	klist_remove_locked(&pr->ps_klist, kn);
1939 	splx(s);
1940 }
1941 
1942 /*
1943  * signal knotes are shared with proc knotes, so we apply a mask to
1944  * the hint in order to differentiate them from process hints.  This
1945  * could be avoided by using a signal-specific knote list, but probably
1946  * isn't worth the trouble.
1947  */
1948 int
1949 filt_signal(struct knote *kn, long hint)
1950 {
1951 
1952 	if (hint & NOTE_SIGNAL) {
1953 		hint &= ~NOTE_SIGNAL;
1954 
1955 		if (kn->kn_id == hint)
1956 			kn->kn_data++;
1957 	}
1958 	return (kn->kn_data != 0);
1959 }
1960 
1961 void
1962 userret(struct proc *p)
1963 {
1964 	struct sigctx ctx;
1965 	int signum;
1966 
1967 	if (p->p_flag & P_SUSPSINGLE)
1968 		single_thread_check(p, 0);
1969 
1970 	/* send SIGPROF or SIGVTALRM if their timers interrupted this thread */
1971 	if (p->p_flag & P_PROFPEND) {
1972 		atomic_clearbits_int(&p->p_flag, P_PROFPEND);
1973 		KERNEL_LOCK();
1974 		psignal(p, SIGPROF);
1975 		KERNEL_UNLOCK();
1976 	}
1977 	if (p->p_flag & P_ALRMPEND) {
1978 		atomic_clearbits_int(&p->p_flag, P_ALRMPEND);
1979 		KERNEL_LOCK();
1980 		psignal(p, SIGVTALRM);
1981 		KERNEL_UNLOCK();
1982 	}
1983 
1984 	if (SIGPENDING(p) != 0) {
1985 		while ((signum = cursig(p, &ctx)) != 0)
1986 			postsig(p, signum, &ctx);
1987 	}
1988 
1989 	/*
1990 	 * If P_SIGSUSPEND is still set here, then we still need to restore
1991 	 * the original sigmask before returning to userspace.  Also, this
1992 	 * might unmask some pending signals, so we need to check a second
1993 	 * time for signals to post.
1994 	 */
1995 	if (p->p_flag & P_SIGSUSPEND) {
1996 		atomic_clearbits_int(&p->p_flag, P_SIGSUSPEND);
1997 		p->p_sigmask = p->p_oldmask;
1998 
1999 		while ((signum = cursig(p, &ctx)) != 0)
2000 			postsig(p, signum, &ctx);
2001 	}
2002 
2003 	WITNESS_WARN(WARN_PANIC, NULL, "userret: returning");
2004 
2005 	p->p_cpu->ci_schedstate.spc_curpriority = p->p_usrpri;
2006 }
2007 
2008 int
2009 single_thread_check_locked(struct proc *p, int deep, int s)
2010 {
2011 	struct process *pr = p->p_p;
2012 
2013 	SCHED_ASSERT_LOCKED();
2014 
2015 	if (pr->ps_single == NULL || pr->ps_single == p)
2016 		return (0);
2017 
2018 	do {
2019 		/* if we're in deep, we need to unwind to the edge */
2020 		if (deep) {
2021 			if (pr->ps_flags & PS_SINGLEUNWIND)
2022 				return (ERESTART);
2023 			if (pr->ps_flags & PS_SINGLEEXIT)
2024 				return (EINTR);
2025 		}
2026 
2027 		if (atomic_dec_int_nv(&pr->ps_singlecount) == 0)
2028 			wakeup(&pr->ps_singlecount);
2029 
2030 		if (pr->ps_flags & PS_SINGLEEXIT) {
2031 			SCHED_UNLOCK(s);
2032 			KERNEL_LOCK();
2033 			exit1(p, 0, 0, EXIT_THREAD_NOCHECK);
2034 			/* NOTREACHED */
2035 		}
2036 
2037 		/* not exiting and don't need to unwind, so suspend */
2038 		p->p_stat = SSTOP;
2039 		mi_switch();
2040 	} while (pr->ps_single != NULL);
2041 
2042 	return (0);
2043 }
2044 
2045 int
2046 single_thread_check(struct proc *p, int deep)
2047 {
2048 	int s, error;
2049 
2050 	SCHED_LOCK(s);
2051 	error = single_thread_check_locked(p, deep, s);
2052 	SCHED_UNLOCK(s);
2053 
2054 	return error;
2055 }
2056 
2057 /*
2058  * Stop other threads in the process.  The mode controls how and
2059  * where the other threads should stop:
2060  *  - SINGLE_SUSPEND: stop wherever they are, will later be released (via
2061  *    single_thread_clear())
2062  *  - SINGLE_UNWIND: just unwind to kernel boundary, will be told to exit
2063  *    (by setting to SINGLE_EXIT) or released as with SINGLE_SUSPEND
2064  *  - SINGLE_EXIT: unwind to kernel boundary and exit
2065  */
2066 int
2067 single_thread_set(struct proc *p, int flags)
2068 {
2069 	struct process *pr = p->p_p;
2070 	struct proc *q;
2071 	int error, s, mode = flags & SINGLE_MASK;
2072 
2073 	KASSERT(curproc == p);
2074 
2075 	SCHED_LOCK(s);
2076 	error = single_thread_check_locked(p, flags & SINGLE_DEEP, s);
2077 	if (error) {
2078 		SCHED_UNLOCK(s);
2079 		return error;
2080 	}
2081 
2082 	switch (mode) {
2083 	case SINGLE_SUSPEND:
2084 		break;
2085 	case SINGLE_UNWIND:
2086 		atomic_setbits_int(&pr->ps_flags, PS_SINGLEUNWIND);
2087 		break;
2088 	case SINGLE_EXIT:
2089 		atomic_setbits_int(&pr->ps_flags, PS_SINGLEEXIT);
2090 		atomic_clearbits_int(&pr->ps_flags, PS_SINGLEUNWIND);
2091 		break;
2092 #ifdef DIAGNOSTIC
2093 	default:
2094 		panic("single_thread_mode = %d", mode);
2095 #endif
2096 	}
2097 	pr->ps_singlecount = 0;
2098 	membar_producer();
2099 	pr->ps_single = p;
2100 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
2101 		if (q == p)
2102 			continue;
2103 		if (q->p_flag & P_WEXIT) {
2104 			if (mode == SINGLE_EXIT) {
2105 				if (q->p_stat == SSTOP) {
2106 					setrunnable(q);
2107 					atomic_inc_int(&pr->ps_singlecount);
2108 				}
2109 			}
2110 			continue;
2111 		}
2112 		atomic_setbits_int(&q->p_flag, P_SUSPSINGLE);
2113 		switch (q->p_stat) {
2114 		case SIDL:
2115 		case SRUN:
2116 			atomic_inc_int(&pr->ps_singlecount);
2117 			break;
2118 		case SSLEEP:
2119 			/* if it's not interruptible, then just have to wait */
2120 			if (q->p_flag & P_SINTR) {
2121 				/* merely need to suspend?  just stop it */
2122 				if (mode == SINGLE_SUSPEND) {
2123 					q->p_stat = SSTOP;
2124 					break;
2125 				}
2126 				/* need to unwind or exit, so wake it */
2127 				setrunnable(q);
2128 			}
2129 			atomic_inc_int(&pr->ps_singlecount);
2130 			break;
2131 		case SSTOP:
2132 			if (mode == SINGLE_EXIT) {
2133 				setrunnable(q);
2134 				atomic_inc_int(&pr->ps_singlecount);
2135 			}
2136 			break;
2137 		case SDEAD:
2138 			break;
2139 		case SONPROC:
2140 			atomic_inc_int(&pr->ps_singlecount);
2141 			signotify(q);
2142 			break;
2143 		}
2144 	}
2145 	SCHED_UNLOCK(s);
2146 
2147 	if ((flags & SINGLE_NOWAIT) == 0)
2148 		single_thread_wait(pr, 1);
2149 
2150 	return 0;
2151 }
2152 
2153 /*
2154  * Wait for other threads to stop. If recheck is false then the function
2155  * returns non-zero if the caller needs to restart the check else 0 is
2156  * returned. If recheck is true the return value is always 0.
2157  */
2158 int
2159 single_thread_wait(struct process *pr, int recheck)
2160 {
2161 	int wait;
2162 
2163 	/* wait until they're all suspended */
2164 	wait = pr->ps_singlecount > 0;
2165 	while (wait) {
2166 		sleep_setup(&pr->ps_singlecount, PWAIT, "suspend");
2167 		wait = pr->ps_singlecount > 0;
2168 		sleep_finish(0, wait);
2169 		if (!recheck)
2170 			break;
2171 	}
2172 
2173 	return wait;
2174 }
2175 
2176 void
2177 single_thread_clear(struct proc *p, int flag)
2178 {
2179 	struct process *pr = p->p_p;
2180 	struct proc *q;
2181 	int s;
2182 
2183 	KASSERT(pr->ps_single == p);
2184 	KASSERT(curproc == p);
2185 
2186 	SCHED_LOCK(s);
2187 	pr->ps_single = NULL;
2188 	atomic_clearbits_int(&pr->ps_flags, PS_SINGLEUNWIND | PS_SINGLEEXIT);
2189 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
2190 		if (q == p || (q->p_flag & P_SUSPSINGLE) == 0)
2191 			continue;
2192 		atomic_clearbits_int(&q->p_flag, P_SUSPSINGLE);
2193 
2194 		/*
2195 		 * if the thread was only stopped for single threading
2196 		 * then clearing that either makes it runnable or puts
2197 		 * it back into some sleep queue
2198 		 */
2199 		if (q->p_stat == SSTOP && (q->p_flag & flag) == 0) {
2200 			if (q->p_wchan == NULL)
2201 				setrunnable(q);
2202 			else {
2203 				atomic_clearbits_int(&q->p_flag, P_WSLEEP);
2204 				q->p_stat = SSLEEP;
2205 			}
2206 		}
2207 	}
2208 	SCHED_UNLOCK(s);
2209 }
2210 
2211 void
2212 sigio_del(struct sigiolst *rmlist)
2213 {
2214 	struct sigio *sigio;
2215 
2216 	while ((sigio = LIST_FIRST(rmlist)) != NULL) {
2217 		LIST_REMOVE(sigio, sio_pgsigio);
2218 		crfree(sigio->sio_ucred);
2219 		free(sigio, M_SIGIO, sizeof(*sigio));
2220 	}
2221 }
2222 
2223 void
2224 sigio_unlink(struct sigio_ref *sir, struct sigiolst *rmlist)
2225 {
2226 	struct sigio *sigio;
2227 
2228 	MUTEX_ASSERT_LOCKED(&sigio_lock);
2229 
2230 	sigio = sir->sir_sigio;
2231 	if (sigio != NULL) {
2232 		KASSERT(sigio->sio_myref == sir);
2233 		sir->sir_sigio = NULL;
2234 
2235 		if (sigio->sio_pgid > 0)
2236 			sigio->sio_proc = NULL;
2237 		else
2238 			sigio->sio_pgrp = NULL;
2239 		LIST_REMOVE(sigio, sio_pgsigio);
2240 
2241 		LIST_INSERT_HEAD(rmlist, sigio, sio_pgsigio);
2242 	}
2243 }
2244 
2245 void
2246 sigio_free(struct sigio_ref *sir)
2247 {
2248 	struct sigiolst rmlist;
2249 
2250 	if (sir->sir_sigio == NULL)
2251 		return;
2252 
2253 	LIST_INIT(&rmlist);
2254 
2255 	mtx_enter(&sigio_lock);
2256 	sigio_unlink(sir, &rmlist);
2257 	mtx_leave(&sigio_lock);
2258 
2259 	sigio_del(&rmlist);
2260 }
2261 
2262 void
2263 sigio_freelist(struct sigiolst *sigiolst)
2264 {
2265 	struct sigiolst rmlist;
2266 	struct sigio *sigio;
2267 
2268 	if (LIST_EMPTY(sigiolst))
2269 		return;
2270 
2271 	LIST_INIT(&rmlist);
2272 
2273 	mtx_enter(&sigio_lock);
2274 	while ((sigio = LIST_FIRST(sigiolst)) != NULL)
2275 		sigio_unlink(sigio->sio_myref, &rmlist);
2276 	mtx_leave(&sigio_lock);
2277 
2278 	sigio_del(&rmlist);
2279 }
2280 
2281 int
2282 sigio_setown(struct sigio_ref *sir, u_long cmd, caddr_t data)
2283 {
2284 	struct sigiolst rmlist;
2285 	struct proc *p = curproc;
2286 	struct pgrp *pgrp = NULL;
2287 	struct process *pr = NULL;
2288 	struct sigio *sigio;
2289 	int error;
2290 	pid_t pgid = *(int *)data;
2291 
2292 	if (pgid == 0) {
2293 		sigio_free(sir);
2294 		return (0);
2295 	}
2296 
2297 	if (cmd == TIOCSPGRP) {
2298 		if (pgid < 0)
2299 			return (EINVAL);
2300 		pgid = -pgid;
2301 	}
2302 
2303 	sigio = malloc(sizeof(*sigio), M_SIGIO, M_WAITOK);
2304 	sigio->sio_pgid = pgid;
2305 	sigio->sio_ucred = crhold(p->p_ucred);
2306 	sigio->sio_myref = sir;
2307 
2308 	LIST_INIT(&rmlist);
2309 
2310 	/*
2311 	 * The kernel lock, and not sleeping between prfind()/pgfind() and
2312 	 * linking of the sigio ensure that the process or process group does
2313 	 * not disappear unexpectedly.
2314 	 */
2315 	KERNEL_LOCK();
2316 	mtx_enter(&sigio_lock);
2317 
2318 	if (pgid > 0) {
2319 		pr = prfind(pgid);
2320 		if (pr == NULL) {
2321 			error = ESRCH;
2322 			goto fail;
2323 		}
2324 
2325 		/*
2326 		 * Policy - Don't allow a process to FSETOWN a process
2327 		 * in another session.
2328 		 *
2329 		 * Remove this test to allow maximum flexibility or
2330 		 * restrict FSETOWN to the current process or process
2331 		 * group for maximum safety.
2332 		 */
2333 		if (pr->ps_session != p->p_p->ps_session) {
2334 			error = EPERM;
2335 			goto fail;
2336 		}
2337 
2338 		if ((pr->ps_flags & PS_EXITING) != 0) {
2339 			error = ESRCH;
2340 			goto fail;
2341 		}
2342 	} else /* if (pgid < 0) */ {
2343 		pgrp = pgfind(-pgid);
2344 		if (pgrp == NULL) {
2345 			error = ESRCH;
2346 			goto fail;
2347 		}
2348 
2349 		/*
2350 		 * Policy - Don't allow a process to FSETOWN a process
2351 		 * in another session.
2352 		 *
2353 		 * Remove this test to allow maximum flexibility or
2354 		 * restrict FSETOWN to the current process or process
2355 		 * group for maximum safety.
2356 		 */
2357 		if (pgrp->pg_session != p->p_p->ps_session) {
2358 			error = EPERM;
2359 			goto fail;
2360 		}
2361 	}
2362 
2363 	if (pgid > 0) {
2364 		sigio->sio_proc = pr;
2365 		LIST_INSERT_HEAD(&pr->ps_sigiolst, sigio, sio_pgsigio);
2366 	} else {
2367 		sigio->sio_pgrp = pgrp;
2368 		LIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
2369 	}
2370 
2371 	sigio_unlink(sir, &rmlist);
2372 	sir->sir_sigio = sigio;
2373 
2374 	mtx_leave(&sigio_lock);
2375 	KERNEL_UNLOCK();
2376 
2377 	sigio_del(&rmlist);
2378 
2379 	return (0);
2380 
2381 fail:
2382 	mtx_leave(&sigio_lock);
2383 	KERNEL_UNLOCK();
2384 
2385 	crfree(sigio->sio_ucred);
2386 	free(sigio, M_SIGIO, sizeof(*sigio));
2387 
2388 	return (error);
2389 }
2390 
2391 void
2392 sigio_getown(struct sigio_ref *sir, u_long cmd, caddr_t data)
2393 {
2394 	struct sigio *sigio;
2395 	pid_t pgid = 0;
2396 
2397 	mtx_enter(&sigio_lock);
2398 	sigio = sir->sir_sigio;
2399 	if (sigio != NULL)
2400 		pgid = sigio->sio_pgid;
2401 	mtx_leave(&sigio_lock);
2402 
2403 	if (cmd == TIOCGPGRP)
2404 		pgid = -pgid;
2405 
2406 	*(int *)data = pgid;
2407 }
2408 
2409 void
2410 sigio_copy(struct sigio_ref *dst, struct sigio_ref *src)
2411 {
2412 	struct sigiolst rmlist;
2413 	struct sigio *newsigio, *sigio;
2414 
2415 	sigio_free(dst);
2416 
2417 	if (src->sir_sigio == NULL)
2418 		return;
2419 
2420 	newsigio = malloc(sizeof(*newsigio), M_SIGIO, M_WAITOK);
2421 	LIST_INIT(&rmlist);
2422 
2423 	mtx_enter(&sigio_lock);
2424 
2425 	sigio = src->sir_sigio;
2426 	if (sigio == NULL) {
2427 		mtx_leave(&sigio_lock);
2428 		free(newsigio, M_SIGIO, sizeof(*newsigio));
2429 		return;
2430 	}
2431 
2432 	newsigio->sio_pgid = sigio->sio_pgid;
2433 	newsigio->sio_ucred = crhold(sigio->sio_ucred);
2434 	newsigio->sio_myref = dst;
2435 	if (newsigio->sio_pgid > 0) {
2436 		newsigio->sio_proc = sigio->sio_proc;
2437 		LIST_INSERT_HEAD(&newsigio->sio_proc->ps_sigiolst, newsigio,
2438 		    sio_pgsigio);
2439 	} else {
2440 		newsigio->sio_pgrp = sigio->sio_pgrp;
2441 		LIST_INSERT_HEAD(&newsigio->sio_pgrp->pg_sigiolst, newsigio,
2442 		    sio_pgsigio);
2443 	}
2444 
2445 	sigio_unlink(dst, &rmlist);
2446 	dst->sir_sigio = newsigio;
2447 
2448 	mtx_leave(&sigio_lock);
2449 
2450 	sigio_del(&rmlist);
2451 }
2452