xref: /openbsd-src/sys/kern/kern_sig.c (revision edc99bcd885f4004d6cffdbe82f15aa2eb27ee9c)
1 /*	$OpenBSD: kern_sig.c,v 1.231 2019/06/21 09:39:48 visa 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 #define	SIGPROP		/* include signal properties table */
42 #include <sys/param.h>
43 #include <sys/signalvar.h>
44 #include <sys/resourcevar.h>
45 #include <sys/queue.h>
46 #include <sys/namei.h>
47 #include <sys/vnode.h>
48 #include <sys/event.h>
49 #include <sys/proc.h>
50 #include <sys/systm.h>
51 #include <sys/acct.h>
52 #include <sys/fcntl.h>
53 #include <sys/filedesc.h>
54 #include <sys/kernel.h>
55 #include <sys/wait.h>
56 #include <sys/ktrace.h>
57 #include <sys/stat.h>
58 #include <sys/core.h>
59 #include <sys/malloc.h>
60 #include <sys/pool.h>
61 #include <sys/ptrace.h>
62 #include <sys/sched.h>
63 #include <sys/user.h>
64 #include <sys/syslog.h>
65 #include <sys/pledge.h>
66 #include <sys/witness.h>
67 
68 #include <sys/mount.h>
69 #include <sys/syscallargs.h>
70 
71 #include <uvm/uvm_extern.h>
72 #include <machine/tcb.h>
73 
74 int	filt_sigattach(struct knote *kn);
75 void	filt_sigdetach(struct knote *kn);
76 int	filt_signal(struct knote *kn, long hint);
77 
78 struct filterops sig_filtops =
79 	{ 0, filt_sigattach, filt_sigdetach, filt_signal };
80 
81 void proc_stop(struct proc *p, int);
82 void proc_stop_sweep(void *);
83 struct timeout proc_stop_to;
84 
85 void postsig(struct proc *, int);
86 int cansignal(struct proc *, struct process *, int);
87 
88 struct pool sigacts_pool;	/* memory pool for sigacts structures */
89 
90 void sigio_del(struct sigiolst *);
91 void sigio_unlink(struct sigio_ref *, struct sigiolst *);
92 struct mutex sigio_lock = MUTEX_INITIALIZER(IPL_HIGH);
93 
94 /*
95  * Can thread p, send the signal signum to process qr?
96  */
97 int
98 cansignal(struct proc *p, struct process *qr, int signum)
99 {
100 	struct process *pr = p->p_p;
101 	struct ucred *uc = p->p_ucred;
102 	struct ucred *quc = qr->ps_ucred;
103 
104 	if (uc->cr_uid == 0)
105 		return (1);		/* root can always signal */
106 
107 	if (pr == qr)
108 		return (1);		/* process can always signal itself */
109 
110 	/* optimization: if the same creds then the tests below will pass */
111 	if (uc == quc)
112 		return (1);
113 
114 	if (signum == SIGCONT && qr->ps_session == pr->ps_session)
115 		return (1);		/* SIGCONT in session */
116 
117 	/*
118 	 * Using kill(), only certain signals can be sent to setugid
119 	 * child processes
120 	 */
121 	if (qr->ps_flags & PS_SUGID) {
122 		switch (signum) {
123 		case 0:
124 		case SIGKILL:
125 		case SIGINT:
126 		case SIGTERM:
127 		case SIGALRM:
128 		case SIGSTOP:
129 		case SIGTTIN:
130 		case SIGTTOU:
131 		case SIGTSTP:
132 		case SIGHUP:
133 		case SIGUSR1:
134 		case SIGUSR2:
135 			if (uc->cr_ruid == quc->cr_ruid ||
136 			    uc->cr_uid == quc->cr_ruid)
137 				return (1);
138 		}
139 		return (0);
140 	}
141 
142 	if (uc->cr_ruid == quc->cr_ruid ||
143 	    uc->cr_ruid == quc->cr_svuid ||
144 	    uc->cr_uid == quc->cr_ruid ||
145 	    uc->cr_uid == quc->cr_svuid)
146 		return (1);
147 	return (0);
148 }
149 
150 /*
151  * Initialize signal-related data structures.
152  */
153 void
154 signal_init(void)
155 {
156 	timeout_set(&proc_stop_to, proc_stop_sweep, NULL);
157 
158 	pool_init(&sigacts_pool, sizeof(struct sigacts), 0, IPL_NONE,
159 	    PR_WAITOK, "sigapl", NULL);
160 }
161 
162 /*
163  * Create an initial sigacts structure, using the same signal state
164  * as p.
165  */
166 struct sigacts *
167 sigactsinit(struct process *pr)
168 {
169 	struct sigacts *ps;
170 
171 	ps = pool_get(&sigacts_pool, PR_WAITOK);
172 	memcpy(ps, pr->ps_sigacts, sizeof(struct sigacts));
173 	ps->ps_refcnt = 1;
174 	return (ps);
175 }
176 
177 /*
178  * Share a sigacts structure.
179  */
180 struct sigacts *
181 sigactsshare(struct process *pr)
182 {
183 	struct sigacts *ps = pr->ps_sigacts;
184 
185 	ps->ps_refcnt++;
186 	return ps;
187 }
188 
189 /*
190  * Initialize a new sigaltstack structure.
191  */
192 void
193 sigstkinit(struct sigaltstack *ss)
194 {
195 	ss->ss_flags = SS_DISABLE;
196 	ss->ss_size = 0;
197 	ss->ss_sp = 0;
198 }
199 
200 /*
201  * Make this process not share its sigacts, maintaining all
202  * signal state.
203  */
204 void
205 sigactsunshare(struct process *pr)
206 {
207 	struct sigacts *newps;
208 
209 	if (pr->ps_sigacts->ps_refcnt == 1)
210 		return;
211 
212 	newps = sigactsinit(pr);
213 	sigactsfree(pr);
214 	pr->ps_sigacts = newps;
215 }
216 
217 /*
218  * Release a sigacts structure.
219  */
220 void
221 sigactsfree(struct process *pr)
222 {
223 	struct sigacts *ps = pr->ps_sigacts;
224 
225 	if (--ps->ps_refcnt > 0)
226 		return;
227 
228 	pr->ps_sigacts = NULL;
229 
230 	pool_put(&sigacts_pool, ps);
231 }
232 
233 int
234 sys_sigaction(struct proc *p, void *v, register_t *retval)
235 {
236 	struct sys_sigaction_args /* {
237 		syscallarg(int) signum;
238 		syscallarg(const struct sigaction *) nsa;
239 		syscallarg(struct sigaction *) osa;
240 	} */ *uap = v;
241 	struct sigaction vec;
242 #ifdef KTRACE
243 	struct sigaction ovec;
244 #endif
245 	struct sigaction *sa;
246 	const struct sigaction *nsa;
247 	struct sigaction *osa;
248 	struct sigacts *ps = p->p_p->ps_sigacts;
249 	int signum;
250 	int bit, error;
251 
252 	signum = SCARG(uap, signum);
253 	nsa = SCARG(uap, nsa);
254 	osa = SCARG(uap, osa);
255 
256 	if (signum <= 0 || signum >= NSIG ||
257 	    (nsa && (signum == SIGKILL || signum == SIGSTOP)))
258 		return (EINVAL);
259 	sa = &vec;
260 	if (osa) {
261 		sa->sa_handler = ps->ps_sigact[signum];
262 		sa->sa_mask = ps->ps_catchmask[signum];
263 		bit = sigmask(signum);
264 		sa->sa_flags = 0;
265 		if ((ps->ps_sigonstack & bit) != 0)
266 			sa->sa_flags |= SA_ONSTACK;
267 		if ((ps->ps_sigintr & bit) == 0)
268 			sa->sa_flags |= SA_RESTART;
269 		if ((ps->ps_sigreset & bit) != 0)
270 			sa->sa_flags |= SA_RESETHAND;
271 		if ((ps->ps_siginfo & bit) != 0)
272 			sa->sa_flags |= SA_SIGINFO;
273 		if (signum == SIGCHLD) {
274 			if ((ps->ps_flags & SAS_NOCLDSTOP) != 0)
275 				sa->sa_flags |= SA_NOCLDSTOP;
276 			if ((ps->ps_flags & SAS_NOCLDWAIT) != 0)
277 				sa->sa_flags |= SA_NOCLDWAIT;
278 		}
279 		if ((sa->sa_mask & bit) == 0)
280 			sa->sa_flags |= SA_NODEFER;
281 		sa->sa_mask &= ~bit;
282 		error = copyout(sa, osa, sizeof (vec));
283 		if (error)
284 			return (error);
285 #ifdef KTRACE
286 		if (KTRPOINT(p, KTR_STRUCT))
287 			ovec = vec;
288 #endif
289 	}
290 	if (nsa) {
291 		error = copyin(nsa, sa, sizeof (vec));
292 		if (error)
293 			return (error);
294 #ifdef KTRACE
295 		if (KTRPOINT(p, KTR_STRUCT))
296 			ktrsigaction(p, sa);
297 #endif
298 		setsigvec(p, signum, sa);
299 	}
300 #ifdef KTRACE
301 	if (osa && KTRPOINT(p, KTR_STRUCT))
302 		ktrsigaction(p, &ovec);
303 #endif
304 	return (0);
305 }
306 
307 void
308 setsigvec(struct proc *p, int signum, struct sigaction *sa)
309 {
310 	struct sigacts *ps = p->p_p->ps_sigacts;
311 	int bit;
312 	int s;
313 
314 	bit = sigmask(signum);
315 	/*
316 	 * Change setting atomically.
317 	 */
318 	s = splhigh();
319 	ps->ps_sigact[signum] = sa->sa_handler;
320 	if ((sa->sa_flags & SA_NODEFER) == 0)
321 		sa->sa_mask |= sigmask(signum);
322 	ps->ps_catchmask[signum] = sa->sa_mask &~ sigcantmask;
323 	if (signum == SIGCHLD) {
324 		if (sa->sa_flags & SA_NOCLDSTOP)
325 			atomic_setbits_int(&ps->ps_flags, SAS_NOCLDSTOP);
326 		else
327 			atomic_clearbits_int(&ps->ps_flags, SAS_NOCLDSTOP);
328 		/*
329 		 * If the SA_NOCLDWAIT flag is set or the handler
330 		 * is SIG_IGN we reparent the dying child to PID 1
331 		 * (init) which will reap the zombie.  Because we use
332 		 * init to do our dirty work we never set SAS_NOCLDWAIT
333 		 * for PID 1.
334 		 * XXX exit1 rework means this is unnecessary?
335 		 */
336 		if (initprocess->ps_sigacts != ps &&
337 		    ((sa->sa_flags & SA_NOCLDWAIT) ||
338 		    sa->sa_handler == SIG_IGN))
339 			atomic_setbits_int(&ps->ps_flags, SAS_NOCLDWAIT);
340 		else
341 			atomic_clearbits_int(&ps->ps_flags, SAS_NOCLDWAIT);
342 	}
343 	if ((sa->sa_flags & SA_RESETHAND) != 0)
344 		ps->ps_sigreset |= bit;
345 	else
346 		ps->ps_sigreset &= ~bit;
347 	if ((sa->sa_flags & SA_SIGINFO) != 0)
348 		ps->ps_siginfo |= bit;
349 	else
350 		ps->ps_siginfo &= ~bit;
351 	if ((sa->sa_flags & SA_RESTART) == 0)
352 		ps->ps_sigintr |= bit;
353 	else
354 		ps->ps_sigintr &= ~bit;
355 	if ((sa->sa_flags & SA_ONSTACK) != 0)
356 		ps->ps_sigonstack |= bit;
357 	else
358 		ps->ps_sigonstack &= ~bit;
359 	/*
360 	 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
361 	 * and for signals set to SIG_DFL where the default is to ignore.
362 	 * However, don't put SIGCONT in ps_sigignore,
363 	 * as we have to restart the process.
364 	 */
365 	if (sa->sa_handler == SIG_IGN ||
366 	    (sigprop[signum] & SA_IGNORE && sa->sa_handler == SIG_DFL)) {
367 		atomic_clearbits_int(&p->p_siglist, bit);
368 		atomic_clearbits_int(&p->p_p->ps_siglist, bit);
369 		if (signum != SIGCONT)
370 			ps->ps_sigignore |= bit;	/* easier in psignal */
371 		ps->ps_sigcatch &= ~bit;
372 	} else {
373 		ps->ps_sigignore &= ~bit;
374 		if (sa->sa_handler == SIG_DFL)
375 			ps->ps_sigcatch &= ~bit;
376 		else
377 			ps->ps_sigcatch |= bit;
378 	}
379 	splx(s);
380 }
381 
382 /*
383  * Initialize signal state for process 0;
384  * set to ignore signals that are ignored by default.
385  */
386 void
387 siginit(struct process *pr)
388 {
389 	struct sigacts *ps = pr->ps_sigacts;
390 	int i;
391 
392 	for (i = 0; i < NSIG; i++)
393 		if (sigprop[i] & SA_IGNORE && i != SIGCONT)
394 			ps->ps_sigignore |= sigmask(i);
395 	ps->ps_flags = SAS_NOCLDWAIT | SAS_NOCLDSTOP;
396 }
397 
398 /*
399  * Reset signals for an exec by the specified thread.
400  */
401 void
402 execsigs(struct proc *p)
403 {
404 	struct sigacts *ps;
405 	int nc, mask;
406 
407 	sigactsunshare(p->p_p);
408 	ps = p->p_p->ps_sigacts;
409 
410 	/*
411 	 * Reset caught signals.  Held signals remain held
412 	 * through p_sigmask (unless they were caught,
413 	 * and are now ignored by default).
414 	 */
415 	while (ps->ps_sigcatch) {
416 		nc = ffs((long)ps->ps_sigcatch);
417 		mask = sigmask(nc);
418 		ps->ps_sigcatch &= ~mask;
419 		if (sigprop[nc] & SA_IGNORE) {
420 			if (nc != SIGCONT)
421 				ps->ps_sigignore |= mask;
422 			atomic_clearbits_int(&p->p_siglist, mask);
423 			atomic_clearbits_int(&p->p_p->ps_siglist, mask);
424 		}
425 		ps->ps_sigact[nc] = SIG_DFL;
426 	}
427 	/*
428 	 * Reset stack state to the user stack.
429 	 * Clear set of signals caught on the signal stack.
430 	 */
431 	sigstkinit(&p->p_sigstk);
432 	ps->ps_flags &= ~SAS_NOCLDWAIT;
433 	if (ps->ps_sigact[SIGCHLD] == SIG_IGN)
434 		ps->ps_sigact[SIGCHLD] = SIG_DFL;
435 }
436 
437 /*
438  * Manipulate signal mask.
439  * Note that we receive new mask, not pointer,
440  * and return old mask as return value;
441  * the library stub does the rest.
442  */
443 int
444 sys_sigprocmask(struct proc *p, void *v, register_t *retval)
445 {
446 	struct sys_sigprocmask_args /* {
447 		syscallarg(int) how;
448 		syscallarg(sigset_t) mask;
449 	} */ *uap = v;
450 	int error = 0;
451 	sigset_t mask;
452 
453 	*retval = p->p_sigmask;
454 	mask = SCARG(uap, mask) &~ sigcantmask;
455 
456 	switch (SCARG(uap, how)) {
457 	case SIG_BLOCK:
458 		atomic_setbits_int(&p->p_sigmask, mask);
459 		break;
460 	case SIG_UNBLOCK:
461 		atomic_clearbits_int(&p->p_sigmask, mask);
462 		break;
463 	case SIG_SETMASK:
464 		p->p_sigmask = mask;
465 		break;
466 	default:
467 		error = EINVAL;
468 		break;
469 	}
470 	return (error);
471 }
472 
473 int
474 sys_sigpending(struct proc *p, void *v, register_t *retval)
475 {
476 
477 	*retval = p->p_siglist | p->p_p->ps_siglist;
478 	return (0);
479 }
480 
481 /*
482  * Temporarily replace calling proc's signal mask for the duration of a
483  * system call.  Original signal mask will be restored by userret().
484  */
485 void
486 dosigsuspend(struct proc *p, sigset_t newmask)
487 {
488 	KASSERT(p == curproc);
489 
490 	p->p_oldmask = p->p_sigmask;
491 	atomic_setbits_int(&p->p_flag, P_SIGSUSPEND);
492 	p->p_sigmask = newmask;
493 }
494 
495 /*
496  * Suspend process until signal, providing mask to be set
497  * in the meantime.  Note nonstandard calling convention:
498  * libc stub passes mask, not pointer, to save a copyin.
499  */
500 int
501 sys_sigsuspend(struct proc *p, void *v, register_t *retval)
502 {
503 	struct sys_sigsuspend_args /* {
504 		syscallarg(int) mask;
505 	} */ *uap = v;
506 	struct process *pr = p->p_p;
507 	struct sigacts *ps = pr->ps_sigacts;
508 
509 	dosigsuspend(p, SCARG(uap, mask) &~ sigcantmask);
510 	while (tsleep(ps, PPAUSE|PCATCH, "pause", 0) == 0)
511 		/* void */;
512 	/* always return EINTR rather than ERESTART... */
513 	return (EINTR);
514 }
515 
516 int
517 sigonstack(size_t stack)
518 {
519 	const struct sigaltstack *ss = &curproc->p_sigstk;
520 
521 	return (ss->ss_flags & SS_DISABLE ? 0 :
522 	    (stack - (size_t)ss->ss_sp < ss->ss_size));
523 }
524 
525 int
526 sys_sigaltstack(struct proc *p, void *v, register_t *retval)
527 {
528 	struct sys_sigaltstack_args /* {
529 		syscallarg(const struct sigaltstack *) nss;
530 		syscallarg(struct sigaltstack *) oss;
531 	} */ *uap = v;
532 	struct sigaltstack ss;
533 	const struct sigaltstack *nss;
534 	struct sigaltstack *oss;
535 	int onstack = sigonstack(PROC_STACK(p));
536 	int error;
537 
538 	nss = SCARG(uap, nss);
539 	oss = SCARG(uap, oss);
540 
541 	if (oss != NULL) {
542 		ss = p->p_sigstk;
543 		if (onstack)
544 			ss.ss_flags |= SS_ONSTACK;
545 		if ((error = copyout(&ss, oss, sizeof(ss))))
546 			return (error);
547 	}
548 	if (nss == NULL)
549 		return (0);
550 	error = copyin(nss, &ss, sizeof(ss));
551 	if (error)
552 		return (error);
553 	if (onstack)
554 		return (EPERM);
555 	if (ss.ss_flags & ~SS_DISABLE)
556 		return (EINVAL);
557 	if (ss.ss_flags & SS_DISABLE) {
558 		p->p_sigstk.ss_flags = ss.ss_flags;
559 		return (0);
560 	}
561 	if (ss.ss_size < MINSIGSTKSZ)
562 		return (ENOMEM);
563 
564 	error = uvm_map_remap_as_stack(p, (vaddr_t)ss.ss_sp, ss.ss_size);
565 	if (error)
566 		return (error);
567 
568 	p->p_sigstk = ss;
569 	return (0);
570 }
571 
572 int
573 sys_kill(struct proc *cp, void *v, register_t *retval)
574 {
575 	struct sys_kill_args /* {
576 		syscallarg(int) pid;
577 		syscallarg(int) signum;
578 	} */ *uap = v;
579 	struct process *pr;
580 	int pid = SCARG(uap, pid);
581 	int signum = SCARG(uap, signum);
582 	int error;
583 	int zombie = 0;
584 
585 	if ((error = pledge_kill(cp, pid)) != 0)
586 		return (error);
587 	if (((u_int)signum) >= NSIG)
588 		return (EINVAL);
589 	if (pid > 0) {
590 		if ((pr = prfind(pid)) == NULL) {
591 			if ((pr = zombiefind(pid)) == NULL)
592 				return (ESRCH);
593 			else
594 				zombie = 1;
595 		}
596 		if (!cansignal(cp, pr, signum))
597 			return (EPERM);
598 
599 		/* kill single process */
600 		if (signum && !zombie)
601 			prsignal(pr, signum);
602 		return (0);
603 	}
604 	switch (pid) {
605 	case -1:		/* broadcast signal */
606 		return (killpg1(cp, signum, 0, 1));
607 	case 0:			/* signal own process group */
608 		return (killpg1(cp, signum, 0, 0));
609 	default:		/* negative explicit process group */
610 		return (killpg1(cp, signum, -pid, 0));
611 	}
612 }
613 
614 int
615 sys_thrkill(struct proc *cp, void *v, register_t *retval)
616 {
617 	struct sys_thrkill_args /* {
618 		syscallarg(pid_t) tid;
619 		syscallarg(int) signum;
620 		syscallarg(void *) tcb;
621 	} */ *uap = v;
622 	struct proc *p;
623 	int tid = SCARG(uap, tid);
624 	int signum = SCARG(uap, signum);
625 	void *tcb;
626 
627 	if (((u_int)signum) >= NSIG)
628 		return (EINVAL);
629 	if (tid > THREAD_PID_OFFSET) {
630 		if ((p = tfind(tid - THREAD_PID_OFFSET)) == NULL)
631 			return (ESRCH);
632 
633 		/* can only kill threads in the same process */
634 		if (p->p_p != cp->p_p)
635 			return (ESRCH);
636 	} else if (tid == 0)
637 		p = cp;
638 	else
639 		return (EINVAL);
640 
641 	/* optionally require the target thread to have the given tcb addr */
642 	tcb = SCARG(uap, tcb);
643 	if (tcb != NULL && tcb != TCB_GET(p))
644 		return (ESRCH);
645 
646 	if (signum)
647 		ptsignal(p, signum, STHREAD);
648 	return (0);
649 }
650 
651 /*
652  * Common code for kill process group/broadcast kill.
653  * cp is calling process.
654  */
655 int
656 killpg1(struct proc *cp, int signum, int pgid, int all)
657 {
658 	struct process *pr;
659 	struct pgrp *pgrp;
660 	int nfound = 0;
661 
662 	if (all) {
663 		/*
664 		 * broadcast
665 		 */
666 		LIST_FOREACH(pr, &allprocess, ps_list) {
667 			if (pr->ps_pid <= 1 ||
668 			    pr->ps_flags & (PS_SYSTEM | PS_NOBROADCASTKILL) ||
669 			    pr == cp->p_p || !cansignal(cp, pr, signum))
670 				continue;
671 			nfound++;
672 			if (signum)
673 				prsignal(pr, signum);
674 		}
675 	} else {
676 		if (pgid == 0)
677 			/*
678 			 * zero pgid means send to my process group.
679 			 */
680 			pgrp = cp->p_p->ps_pgrp;
681 		else {
682 			pgrp = pgfind(pgid);
683 			if (pgrp == NULL)
684 				return (ESRCH);
685 		}
686 		LIST_FOREACH(pr, &pgrp->pg_members, ps_pglist) {
687 			if (pr->ps_pid <= 1 || pr->ps_flags & PS_SYSTEM ||
688 			    !cansignal(cp, pr, signum))
689 				continue;
690 			nfound++;
691 			if (signum)
692 				prsignal(pr, signum);
693 		}
694 	}
695 	return (nfound ? 0 : ESRCH);
696 }
697 
698 #define CANDELIVER(uid, euid, pr) \
699 	(euid == 0 || \
700 	(uid) == (pr)->ps_ucred->cr_ruid || \
701 	(uid) == (pr)->ps_ucred->cr_svuid || \
702 	(uid) == (pr)->ps_ucred->cr_uid || \
703 	(euid) == (pr)->ps_ucred->cr_ruid || \
704 	(euid) == (pr)->ps_ucred->cr_svuid || \
705 	(euid) == (pr)->ps_ucred->cr_uid)
706 
707 #define CANSIGIO(cr, pr) \
708 	CANDELIVER((cr)->cr_ruid, (cr)->cr_uid, (pr))
709 
710 /*
711  * Deliver signum to pgid, but first check uid/euid against each
712  * process and see if it is permitted.
713  */
714 void
715 csignal(pid_t pgid, int signum, uid_t uid, uid_t euid)
716 {
717 	struct pgrp *pgrp;
718 	struct process *pr;
719 
720 	if (pgid == 0)
721 		return;
722 	if (pgid < 0) {
723 		pgid = -pgid;
724 		if ((pgrp = pgfind(pgid)) == NULL)
725 			return;
726 		LIST_FOREACH(pr, &pgrp->pg_members, ps_pglist)
727 			if (CANDELIVER(uid, euid, pr))
728 				prsignal(pr, signum);
729 	} else {
730 		if ((pr = prfind(pgid)) == NULL)
731 			return;
732 		if (CANDELIVER(uid, euid, pr))
733 			prsignal(pr, signum);
734 	}
735 }
736 
737 /*
738  * Send a signal to a process group.  If checktty is 1,
739  * limit to members which have a controlling terminal.
740  */
741 void
742 pgsignal(struct pgrp *pgrp, int signum, int checkctty)
743 {
744 	struct process *pr;
745 
746 	if (pgrp)
747 		LIST_FOREACH(pr, &pgrp->pg_members, ps_pglist)
748 			if (checkctty == 0 || pr->ps_flags & PS_CONTROLT)
749 				prsignal(pr, signum);
750 }
751 
752 /*
753  * Send a SIGIO or SIGURG signal to a process or process group using stored
754  * credentials rather than those of the current process.
755  */
756 void
757 pgsigio(struct sigio_ref *sir, int sig, int checkctty)
758 {
759 	struct process *pr;
760 	struct sigio *sigio;
761 
762 	if (sir->sir_sigio == NULL)
763 		return;
764 
765 	mtx_enter(&sigio_lock);
766 	sigio = sir->sir_sigio;
767 	if (sigio == NULL)
768 		goto out;
769 	if (sigio->sio_pgid > 0) {
770 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc))
771 			prsignal(sigio->sio_proc, sig);
772 	} else if (sigio->sio_pgid < 0) {
773 		LIST_FOREACH(pr, &sigio->sio_pgrp->pg_members, ps_pglist) {
774 			if (CANSIGIO(sigio->sio_ucred, pr) &&
775 			    (checkctty == 0 || (pr->ps_flags & PS_CONTROLT)))
776 				prsignal(pr, sig);
777 		}
778 	}
779 out:
780 	mtx_leave(&sigio_lock);
781 }
782 
783 /*
784  * Recalculate the signal mask and reset the signal disposition after
785  * usermode frame for delivery is formed.
786  */
787 void
788 postsig_done(struct proc *p, int signum, struct sigacts *ps)
789 {
790 	int mask = sigmask(signum);
791 
792 	KERNEL_ASSERT_LOCKED();
793 
794 	p->p_ru.ru_nsignals++;
795 	atomic_setbits_int(&p->p_sigmask, ps->ps_catchmask[signum]);
796 	if ((ps->ps_sigreset & mask) != 0) {
797 		ps->ps_sigcatch &= ~mask;
798 		if (signum != SIGCONT && sigprop[signum] & SA_IGNORE)
799 			ps->ps_sigignore |= mask;
800 		ps->ps_sigact[signum] = SIG_DFL;
801 	}
802 }
803 
804 /*
805  * Send a signal caused by a trap to the current thread
806  * If it will be caught immediately, deliver it with correct code.
807  * Otherwise, post it normally.
808  */
809 void
810 trapsignal(struct proc *p, int signum, u_long trapno, int code,
811     union sigval sigval)
812 {
813 	struct process *pr = p->p_p;
814 	struct sigacts *ps = pr->ps_sigacts;
815 	int mask;
816 
817 	switch (signum) {
818 	case SIGILL:
819 	case SIGBUS:
820 	case SIGSEGV:
821 		pr->ps_acflag |= ATRAP;
822 		break;
823 	}
824 
825 	mask = sigmask(signum);
826 	if ((pr->ps_flags & PS_TRACED) == 0 &&
827 	    (ps->ps_sigcatch & mask) != 0 &&
828 	    (p->p_sigmask & mask) == 0) {
829 		siginfo_t si;
830 		initsiginfo(&si, signum, trapno, code, sigval);
831 #ifdef KTRACE
832 		if (KTRPOINT(p, KTR_PSIG)) {
833 			ktrpsig(p, signum, ps->ps_sigact[signum],
834 			    p->p_sigmask, code, &si);
835 		}
836 #endif
837 		sendsig(ps->ps_sigact[signum], signum, p->p_sigmask, &si);
838 		postsig_done(p, signum, ps);
839 	} else {
840 		p->p_sisig = signum;
841 		p->p_sitrapno = trapno;	/* XXX for core dump/debugger */
842 		p->p_sicode = code;
843 		p->p_sigval = sigval;
844 
845 		/*
846 		 * Signals like SIGBUS and SIGSEGV should not, when
847 		 * generated by the kernel, be ignorable or blockable.
848 		 * If it is and we're not being traced, then just kill
849 		 * the process.
850 		 */
851 		if ((pr->ps_flags & PS_TRACED) == 0 &&
852 		    (sigprop[signum] & SA_KILL) &&
853 		    ((p->p_sigmask & mask) || (ps->ps_sigignore & mask)))
854 			sigexit(p, signum);
855 		ptsignal(p, signum, STHREAD);
856 	}
857 }
858 
859 /*
860  * Send the signal to the process.  If the signal has an action, the action
861  * is usually performed by the target process rather than the caller; we add
862  * the signal to the set of pending signals for the process.
863  *
864  * Exceptions:
865  *   o When a stop signal is sent to a sleeping process that takes the
866  *     default action, the process is stopped without awakening it.
867  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
868  *     regardless of the signal action (eg, blocked or ignored).
869  *
870  * Other ignored signals are discarded immediately.
871  */
872 void
873 psignal(struct proc *p, int signum)
874 {
875 	ptsignal(p, signum, SPROCESS);
876 }
877 
878 /*
879  * type = SPROCESS	process signal, can be diverted (sigwait())
880  * type = STHREAD	thread signal, but should be propagated if unhandled
881  * type = SPROPAGATED	propagated to this thread, so don't propagate again
882  */
883 void
884 ptsignal(struct proc *p, int signum, enum signal_type type)
885 {
886 	int s, prop;
887 	sig_t action;
888 	int mask;
889 	int *siglist;
890 	struct process *pr = p->p_p;
891 	struct proc *q;
892 	int wakeparent = 0;
893 
894 	KERNEL_ASSERT_LOCKED();
895 
896 #ifdef DIAGNOSTIC
897 	if ((u_int)signum >= NSIG || signum == 0)
898 		panic("psignal signal number");
899 #endif
900 
901 	/* Ignore signal if the target process is exiting */
902 	if (pr->ps_flags & PS_EXITING)
903 		return;
904 
905 	mask = sigmask(signum);
906 
907 	if (type == SPROCESS) {
908 		/* Accept SIGKILL to coredumping processes */
909 		if (pr->ps_flags & PS_COREDUMP && signum == SIGKILL) {
910 			if (pr->ps_single != NULL)
911 				p = pr->ps_single;
912 			atomic_setbits_int(&p->p_p->ps_siglist, mask);
913 			return;
914 		}
915 
916 		/*
917 		 * If the current thread can process the signal
918 		 * immediately (it's unblocked) then have it take it.
919 		 */
920 		q = curproc;
921 		if (q != NULL && q->p_p == pr && (q->p_flag & P_WEXIT) == 0 &&
922 		    (q->p_sigmask & mask) == 0)
923 			p = q;
924 		else {
925 			/*
926 			 * A process-wide signal can be diverted to a
927 			 * different thread that's in sigwait() for this
928 			 * signal.  If there isn't such a thread, then
929 			 * pick a thread that doesn't have it blocked so
930 			 * that the stop/kill consideration isn't
931 			 * delayed.  Otherwise, mark it pending on the
932 			 * main thread.
933 			 */
934 			TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
935 				/* ignore exiting threads */
936 				if (q->p_flag & P_WEXIT)
937 					continue;
938 
939 				/* skip threads that have the signal blocked */
940 				if ((q->p_sigmask & mask) != 0)
941 					continue;
942 
943 				/* okay, could send to this thread */
944 				p = q;
945 
946 				/*
947 				 * sigsuspend, sigwait, ppoll/pselect, etc?
948 				 * Definitely go to this thread, as it's
949 				 * already blocked in the kernel.
950 				 */
951 				if (q->p_flag & P_SIGSUSPEND)
952 					break;
953 			}
954 		}
955 	}
956 
957 	if (type != SPROPAGATED)
958 		KNOTE(&pr->ps_klist, NOTE_SIGNAL | signum);
959 
960 	prop = sigprop[signum];
961 
962 	/*
963 	 * If proc is traced, always give parent a chance.
964 	 */
965 	if (pr->ps_flags & PS_TRACED) {
966 		action = SIG_DFL;
967 	} else {
968 		/*
969 		 * If the signal is being ignored,
970 		 * then we forget about it immediately.
971 		 * (Note: we don't set SIGCONT in ps_sigignore,
972 		 * and if it is set to SIG_IGN,
973 		 * action will be SIG_DFL here.)
974 		 */
975 		if (pr->ps_sigacts->ps_sigignore & mask)
976 			return;
977 		if (p->p_sigmask & mask) {
978 			action = SIG_HOLD;
979 		} else if (pr->ps_sigacts->ps_sigcatch & mask) {
980 			action = SIG_CATCH;
981 		} else {
982 			action = SIG_DFL;
983 
984 			if (prop & SA_KILL && pr->ps_nice > NZERO)
985 				 pr->ps_nice = NZERO;
986 
987 			/*
988 			 * If sending a tty stop signal to a member of an
989 			 * orphaned process group, discard the signal here if
990 			 * the action is default; don't stop the process below
991 			 * if sleeping, and don't clear any pending SIGCONT.
992 			 */
993 			if (prop & SA_TTYSTOP && pr->ps_pgrp->pg_jobc == 0)
994 				return;
995 		}
996 	}
997 	/*
998 	 * If delivered to process, mark as pending there.  Continue and stop
999 	 * signals will be propagated to all threads.  So they are always
1000 	 * marked at thread level.
1001 	 */
1002 	siglist = (type == SPROCESS) ? &pr->ps_siglist : &p->p_siglist;
1003 	if (prop & SA_CONT) {
1004 		siglist = &p->p_siglist;
1005 		atomic_clearbits_int(siglist, stopsigmask);
1006 	}
1007 	if (prop & SA_STOP) {
1008 		siglist = &p->p_siglist;
1009 		atomic_clearbits_int(siglist, contsigmask);
1010 		atomic_clearbits_int(&p->p_flag, P_CONTINUED);
1011 	}
1012 	atomic_setbits_int(siglist, mask);
1013 
1014 	/*
1015 	 * XXX delay processing of SA_STOP signals unless action == SIG_DFL?
1016 	 */
1017 	if (prop & (SA_CONT | SA_STOP) && type != SPROPAGATED)
1018 		TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link)
1019 			if (q != p)
1020 				ptsignal(q, signum, SPROPAGATED);
1021 
1022 	/*
1023 	 * Defer further processing for signals which are held,
1024 	 * except that stopped processes must be continued by SIGCONT.
1025 	 */
1026 	if (action == SIG_HOLD && ((prop & SA_CONT) == 0 || p->p_stat != SSTOP))
1027 		return;
1028 
1029 	SCHED_LOCK(s);
1030 
1031 	switch (p->p_stat) {
1032 
1033 	case SSLEEP:
1034 		/*
1035 		 * If process is sleeping uninterruptibly
1036 		 * we can't interrupt the sleep... the signal will
1037 		 * be noticed when the process returns through
1038 		 * trap() or syscall().
1039 		 */
1040 		if ((p->p_flag & P_SINTR) == 0)
1041 			goto out;
1042 		/*
1043 		 * Process is sleeping and traced... make it runnable
1044 		 * so it can discover the signal in issignal() and stop
1045 		 * for the parent.
1046 		 */
1047 		if (pr->ps_flags & PS_TRACED)
1048 			goto run;
1049 		/*
1050 		 * If SIGCONT is default (or ignored) and process is
1051 		 * asleep, we are finished; the process should not
1052 		 * be awakened.
1053 		 */
1054 		if ((prop & SA_CONT) && action == SIG_DFL) {
1055 			atomic_clearbits_int(siglist, mask);
1056 			goto out;
1057 		}
1058 		/*
1059 		 * When a sleeping process receives a stop
1060 		 * signal, process immediately if possible.
1061 		 */
1062 		if ((prop & SA_STOP) && action == SIG_DFL) {
1063 			/*
1064 			 * If a child holding parent blocked,
1065 			 * stopping could cause deadlock.
1066 			 */
1067 			if (pr->ps_flags & PS_PPWAIT)
1068 				goto out;
1069 			atomic_clearbits_int(siglist, mask);
1070 			p->p_xstat = signum;
1071 			proc_stop(p, 0);
1072 			goto out;
1073 		}
1074 		/*
1075 		 * All other (caught or default) signals
1076 		 * cause the process to run.
1077 		 */
1078 		goto runfast;
1079 		/*NOTREACHED*/
1080 
1081 	case SSTOP:
1082 		/*
1083 		 * If traced process is already stopped,
1084 		 * then no further action is necessary.
1085 		 */
1086 		if (pr->ps_flags & PS_TRACED)
1087 			goto out;
1088 
1089 		/*
1090 		 * Kill signal always sets processes running.
1091 		 */
1092 		if (signum == SIGKILL) {
1093 			atomic_clearbits_int(&p->p_flag, P_SUSPSIG);
1094 			goto runfast;
1095 		}
1096 
1097 		if (prop & SA_CONT) {
1098 			/*
1099 			 * If SIGCONT is default (or ignored), we continue the
1100 			 * process but don't leave the signal in p_siglist, as
1101 			 * it has no further action.  If SIGCONT is held, we
1102 			 * continue the process and leave the signal in
1103 			 * p_siglist.  If the process catches SIGCONT, let it
1104 			 * handle the signal itself.  If it isn't waiting on
1105 			 * an event, then it goes back to run state.
1106 			 * Otherwise, process goes back to sleep state.
1107 			 */
1108 			atomic_setbits_int(&p->p_flag, P_CONTINUED);
1109 			atomic_clearbits_int(&p->p_flag, P_SUSPSIG);
1110 			wakeparent = 1;
1111 			if (action == SIG_DFL)
1112 				atomic_clearbits_int(siglist, mask);
1113 			if (action == SIG_CATCH)
1114 				goto runfast;
1115 			if (p->p_wchan == 0)
1116 				goto run;
1117 			p->p_stat = SSLEEP;
1118 			goto out;
1119 		}
1120 
1121 		if (prop & SA_STOP) {
1122 			/*
1123 			 * Already stopped, don't need to stop again.
1124 			 * (If we did the shell could get confused.)
1125 			 */
1126 			atomic_clearbits_int(siglist, mask);
1127 			goto out;
1128 		}
1129 
1130 		/*
1131 		 * If process is sleeping interruptibly, then simulate a
1132 		 * wakeup so that when it is continued, it will be made
1133 		 * runnable and can look at the signal.  But don't make
1134 		 * the process runnable, leave it stopped.
1135 		 */
1136 		if (p->p_wchan && p->p_flag & P_SINTR)
1137 			unsleep(p);
1138 		goto out;
1139 
1140 	case SONPROC:
1141 		signotify(p);
1142 		/* FALLTHROUGH */
1143 	default:
1144 		/*
1145 		 * SRUN, SIDL, SDEAD do nothing with the signal,
1146 		 * other than kicking ourselves if we are running.
1147 		 * It will either never be noticed, or noticed very soon.
1148 		 */
1149 		goto out;
1150 	}
1151 	/*NOTREACHED*/
1152 
1153 runfast:
1154 	/*
1155 	 * Raise priority to at least PUSER.
1156 	 */
1157 	if (p->p_priority > PUSER)
1158 		p->p_priority = PUSER;
1159 run:
1160 	setrunnable(p);
1161 out:
1162 	SCHED_UNLOCK(s);
1163 	if (wakeparent)
1164 		wakeup(pr->ps_pptr);
1165 }
1166 
1167 /*
1168  * If the current process has received a signal (should be caught or cause
1169  * termination, should interrupt current syscall), return the signal number.
1170  * Stop signals with default action are processed immediately, then cleared;
1171  * they aren't returned.  This is checked after each entry to the system for
1172  * a syscall or trap (though this can usually be done without calling issignal
1173  * by checking the pending signal masks in the CURSIG macro.) The normal call
1174  * sequence is
1175  *
1176  *	while (signum = CURSIG(curproc))
1177  *		postsig(signum);
1178  *
1179  * Assumes that if the P_SINTR flag is set, we're holding both the
1180  * kernel and scheduler locks.
1181  */
1182 int
1183 issignal(struct proc *p)
1184 {
1185 	struct process *pr = p->p_p;
1186 	int signum, mask, prop;
1187 	int dolock = (p->p_flag & P_SINTR) == 0;
1188 	int s;
1189 
1190 	for (;;) {
1191 		mask = SIGPENDING(p);
1192 		if (pr->ps_flags & PS_PPWAIT)
1193 			mask &= ~stopsigmask;
1194 		if (mask == 0)	 	/* no signal to send */
1195 			return (0);
1196 		signum = ffs((long)mask);
1197 		mask = sigmask(signum);
1198 		atomic_clearbits_int(&p->p_siglist, mask);
1199 		atomic_clearbits_int(&p->p_p->ps_siglist, mask);
1200 
1201 		/*
1202 		 * We should see pending but ignored signals
1203 		 * only if PS_TRACED was on when they were posted.
1204 		 */
1205 		if (mask & pr->ps_sigacts->ps_sigignore &&
1206 		    (pr->ps_flags & PS_TRACED) == 0)
1207 			continue;
1208 
1209 		/*
1210 		 * If traced, always stop, and stay stopped until released
1211 		 * by the debugger.  If our parent process is waiting for
1212 		 * us, don't hang as we could deadlock.
1213 		 */
1214 		if (((pr->ps_flags & (PS_TRACED | PS_PPWAIT)) == PS_TRACED) &&
1215 		    signum != SIGKILL) {
1216 			p->p_xstat = signum;
1217 
1218 			if (dolock)
1219 				KERNEL_LOCK();
1220 			single_thread_set(p, SINGLE_PTRACE, 0);
1221 			if (dolock)
1222 				KERNEL_UNLOCK();
1223 
1224 			if (dolock)
1225 				SCHED_LOCK(s);
1226 			proc_stop(p, 1);
1227 			if (dolock)
1228 				SCHED_UNLOCK(s);
1229 
1230 			if (dolock)
1231 				KERNEL_LOCK();
1232 			single_thread_clear(p, 0);
1233 			if (dolock)
1234 				KERNEL_UNLOCK();
1235 
1236 			/*
1237 			 * If we are no longer being traced, or the parent
1238 			 * didn't give us a signal, look for more signals.
1239 			 */
1240 			if ((pr->ps_flags & PS_TRACED) == 0 || p->p_xstat == 0)
1241 				continue;
1242 
1243 			/*
1244 			 * If the new signal is being masked, look for other
1245 			 * signals.
1246 			 */
1247 			signum = p->p_xstat;
1248 			mask = sigmask(signum);
1249 			if ((p->p_sigmask & mask) != 0)
1250 				continue;
1251 
1252 			/* take the signal! */
1253 			atomic_clearbits_int(&p->p_siglist, mask);
1254 			atomic_clearbits_int(&p->p_p->ps_siglist, mask);
1255 		}
1256 
1257 		prop = sigprop[signum];
1258 
1259 		/*
1260 		 * Decide whether the signal should be returned.
1261 		 * Return the signal's number, or fall through
1262 		 * to clear it from the pending mask.
1263 		 */
1264 		switch ((long)pr->ps_sigacts->ps_sigact[signum]) {
1265 		case (long)SIG_DFL:
1266 			/*
1267 			 * Don't take default actions on system processes.
1268 			 */
1269 			if (pr->ps_pid <= 1) {
1270 #ifdef DIAGNOSTIC
1271 				/*
1272 				 * Are you sure you want to ignore SIGSEGV
1273 				 * in init? XXX
1274 				 */
1275 				printf("Process (pid %d) got signal"
1276 				    " %d\n", pr->ps_pid, signum);
1277 #endif
1278 				break;		/* == ignore */
1279 			}
1280 			/*
1281 			 * If there is a pending stop signal to process
1282 			 * with default action, stop here,
1283 			 * then clear the signal.  However,
1284 			 * if process is member of an orphaned
1285 			 * process group, ignore tty stop signals.
1286 			 */
1287 			if (prop & SA_STOP) {
1288 				if (pr->ps_flags & PS_TRACED ||
1289 		    		    (pr->ps_pgrp->pg_jobc == 0 &&
1290 				    prop & SA_TTYSTOP))
1291 					break;	/* == ignore */
1292 				p->p_xstat = signum;
1293 				if (dolock)
1294 					SCHED_LOCK(s);
1295 				proc_stop(p, 1);
1296 				if (dolock)
1297 					SCHED_UNLOCK(s);
1298 				break;
1299 			} else if (prop & SA_IGNORE) {
1300 				/*
1301 				 * Except for SIGCONT, shouldn't get here.
1302 				 * Default action is to ignore; drop it.
1303 				 */
1304 				break;		/* == ignore */
1305 			} else
1306 				goto keep;
1307 			/*NOTREACHED*/
1308 		case (long)SIG_IGN:
1309 			/*
1310 			 * Masking above should prevent us ever trying
1311 			 * to take action on an ignored signal other
1312 			 * than SIGCONT, unless process is traced.
1313 			 */
1314 			if ((prop & SA_CONT) == 0 &&
1315 			    (pr->ps_flags & PS_TRACED) == 0)
1316 				printf("issignal\n");
1317 			break;		/* == ignore */
1318 		default:
1319 			/*
1320 			 * This signal has an action, let
1321 			 * postsig() process it.
1322 			 */
1323 			goto keep;
1324 		}
1325 	}
1326 	/* NOTREACHED */
1327 
1328 keep:
1329 	atomic_setbits_int(&p->p_siglist, mask); /*leave the signal for later */
1330 	return (signum);
1331 }
1332 
1333 /*
1334  * Put the argument process into the stopped state and notify the parent
1335  * via wakeup.  Signals are handled elsewhere.  The process must not be
1336  * on the run queue.
1337  */
1338 void
1339 proc_stop(struct proc *p, int sw)
1340 {
1341 	struct process *pr = p->p_p;
1342 	extern void *softclock_si;
1343 
1344 #ifdef MULTIPROCESSOR
1345 	SCHED_ASSERT_LOCKED();
1346 #endif
1347 
1348 	p->p_stat = SSTOP;
1349 	atomic_clearbits_int(&pr->ps_flags, PS_WAITED);
1350 	atomic_setbits_int(&pr->ps_flags, PS_STOPPED);
1351 	atomic_setbits_int(&p->p_flag, P_SUSPSIG);
1352 	if (!timeout_pending(&proc_stop_to)) {
1353 		timeout_add(&proc_stop_to, 0);
1354 		/*
1355 		 * We need this soft interrupt to be handled fast.
1356 		 * Extra calls to softclock don't hurt.
1357 		 */
1358                 softintr_schedule(softclock_si);
1359 	}
1360 	if (sw)
1361 		mi_switch();
1362 }
1363 
1364 /*
1365  * Called from a timeout to send signals to the parents of stopped processes.
1366  * We can't do this in proc_stop because it's called with nasty locks held
1367  * and we would need recursive scheduler lock to deal with that.
1368  */
1369 void
1370 proc_stop_sweep(void *v)
1371 {
1372 	struct process *pr;
1373 
1374 	LIST_FOREACH(pr, &allprocess, ps_list) {
1375 		if ((pr->ps_flags & PS_STOPPED) == 0)
1376 			continue;
1377 		atomic_clearbits_int(&pr->ps_flags, PS_STOPPED);
1378 
1379 		if ((pr->ps_pptr->ps_sigacts->ps_flags & SAS_NOCLDSTOP) == 0)
1380 			prsignal(pr->ps_pptr, SIGCHLD);
1381 		wakeup(pr->ps_pptr);
1382 	}
1383 }
1384 
1385 /*
1386  * Take the action for the specified signal
1387  * from the current set of pending signals.
1388  */
1389 void
1390 postsig(struct proc *p, int signum)
1391 {
1392 	struct process *pr = p->p_p;
1393 	struct sigacts *ps = pr->ps_sigacts;
1394 	sig_t action;
1395 	u_long trapno;
1396 	int mask, returnmask;
1397 	siginfo_t si;
1398 	union sigval sigval;
1399 	int s, code;
1400 
1401 	KASSERT(signum != 0);
1402 	KERNEL_ASSERT_LOCKED();
1403 
1404 	mask = sigmask(signum);
1405 	atomic_clearbits_int(&p->p_siglist, mask);
1406 	action = ps->ps_sigact[signum];
1407 	sigval.sival_ptr = 0;
1408 
1409 	if (p->p_sisig != signum) {
1410 		trapno = 0;
1411 		code = SI_USER;
1412 		sigval.sival_ptr = 0;
1413 	} else {
1414 		trapno = p->p_sitrapno;
1415 		code = p->p_sicode;
1416 		sigval = p->p_sigval;
1417 	}
1418 	initsiginfo(&si, signum, trapno, code, sigval);
1419 
1420 #ifdef KTRACE
1421 	if (KTRPOINT(p, KTR_PSIG)) {
1422 		ktrpsig(p, signum, action, p->p_flag & P_SIGSUSPEND ?
1423 		    p->p_oldmask : p->p_sigmask, code, &si);
1424 	}
1425 #endif
1426 	if (action == SIG_DFL) {
1427 		/*
1428 		 * Default action, where the default is to kill
1429 		 * the process.  (Other cases were ignored above.)
1430 		 */
1431 		sigexit(p, signum);
1432 		/* NOTREACHED */
1433 	} else {
1434 		/*
1435 		 * If we get here, the signal must be caught.
1436 		 */
1437 #ifdef DIAGNOSTIC
1438 		if (action == SIG_IGN || (p->p_sigmask & mask))
1439 			panic("postsig action");
1440 #endif
1441 		/*
1442 		 * Set the new mask value and also defer further
1443 		 * occurrences of this signal.
1444 		 *
1445 		 * Special case: user has done a sigpause.  Here the
1446 		 * current mask is not of interest, but rather the
1447 		 * mask from before the sigpause is what we want
1448 		 * restored after the signal processing is completed.
1449 		 */
1450 #ifdef MULTIPROCESSOR
1451 		s = splsched();
1452 #else
1453 		s = splhigh();
1454 #endif
1455 		if (p->p_flag & P_SIGSUSPEND) {
1456 			atomic_clearbits_int(&p->p_flag, P_SIGSUSPEND);
1457 			returnmask = p->p_oldmask;
1458 		} else {
1459 			returnmask = p->p_sigmask;
1460 		}
1461 		if (p->p_sisig == signum) {
1462 			p->p_sisig = 0;
1463 			p->p_sitrapno = 0;
1464 			p->p_sicode = SI_USER;
1465 			p->p_sigval.sival_ptr = NULL;
1466 		}
1467 
1468 		sendsig(action, signum, returnmask, &si);
1469 		postsig_done(p, signum, ps);
1470 		splx(s);
1471 	}
1472 }
1473 
1474 /*
1475  * Force the current process to exit with the specified signal, dumping core
1476  * if appropriate.  We bypass the normal tests for masked and caught signals,
1477  * allowing unrecoverable failures to terminate the process without changing
1478  * signal state.  Mark the accounting record with the signal termination.
1479  * If dumping core, save the signal number for the debugger.  Calls exit and
1480  * does not return.
1481  */
1482 void
1483 sigexit(struct proc *p, int signum)
1484 {
1485 	/* Mark process as going away */
1486 	atomic_setbits_int(&p->p_flag, P_WEXIT);
1487 
1488 	p->p_p->ps_acflag |= AXSIG;
1489 	if (sigprop[signum] & SA_CORE) {
1490 		p->p_sisig = signum;
1491 
1492 		/* if there are other threads, pause them */
1493 		if (P_HASSIBLING(p))
1494 			single_thread_set(p, SINGLE_SUSPEND, 0);
1495 
1496 		if (coredump(p) == 0)
1497 			signum |= WCOREFLAG;
1498 	}
1499 	exit1(p, W_EXITCODE(0, signum), EXIT_NORMAL);
1500 	/* NOTREACHED */
1501 }
1502 
1503 int nosuidcoredump = 1;
1504 
1505 struct coredump_iostate {
1506 	struct proc *io_proc;
1507 	struct vnode *io_vp;
1508 	struct ucred *io_cred;
1509 	off_t io_offset;
1510 };
1511 
1512 /*
1513  * Dump core, into a file named "progname.core", unless the process was
1514  * setuid/setgid.
1515  */
1516 int
1517 coredump(struct proc *p)
1518 {
1519 #ifdef SMALL_KERNEL
1520 	return EPERM;
1521 #else
1522 	struct process *pr = p->p_p;
1523 	struct vnode *vp;
1524 	struct ucred *cred = p->p_ucred;
1525 	struct vmspace *vm = p->p_vmspace;
1526 	struct nameidata nd;
1527 	struct vattr vattr;
1528 	struct coredump_iostate	io;
1529 	int error, len, incrash = 0;
1530 	char name[MAXPATHLEN];
1531 	const char *dir = "/var/crash";
1532 
1533 	if (pr->ps_emul->e_coredump == NULL)
1534 		return (EINVAL);
1535 
1536 	pr->ps_flags |= PS_COREDUMP;
1537 
1538 	/*
1539 	 * If the process has inconsistent uids, nosuidcoredump
1540 	 * determines coredump placement policy.
1541 	 */
1542 	if (((pr->ps_flags & PS_SUGID) && (error = suser(p))) ||
1543 	   ((pr->ps_flags & PS_SUGID) && nosuidcoredump)) {
1544 		if (nosuidcoredump == 3 || nosuidcoredump == 2)
1545 			incrash = 1;
1546 		else
1547 			return (EPERM);
1548 	}
1549 
1550 	/* Don't dump if will exceed file size limit. */
1551 	if (USPACE + ptoa(vm->vm_dsize + vm->vm_ssize) >= lim_cur(RLIMIT_CORE))
1552 		return (EFBIG);
1553 
1554 	if (incrash && nosuidcoredump == 3) {
1555 		/*
1556 		 * If the program directory does not exist, dumps of
1557 		 * that core will silently fail.
1558 		 */
1559 		len = snprintf(name, sizeof(name), "%s/%s/%u.core",
1560 		    dir, pr->ps_comm, pr->ps_pid);
1561 	} else if (incrash && nosuidcoredump == 2)
1562 		len = snprintf(name, sizeof(name), "%s/%s.core",
1563 		    dir, pr->ps_comm);
1564 	else
1565 		len = snprintf(name, sizeof(name), "%s.core", pr->ps_comm);
1566 	if (len >= sizeof(name))
1567 		return (EACCES);
1568 
1569 	/*
1570 	 * Control the UID used to write out.  The normal case uses
1571 	 * the real UID.  If the sugid case is going to write into the
1572 	 * controlled directory, we do so as root.
1573 	 */
1574 	if (incrash == 0) {
1575 		cred = crdup(cred);
1576 		cred->cr_uid = cred->cr_ruid;
1577 		cred->cr_gid = cred->cr_rgid;
1578 	} else {
1579 		if (p->p_fd->fd_rdir) {
1580 			vrele(p->p_fd->fd_rdir);
1581 			p->p_fd->fd_rdir = NULL;
1582 		}
1583 		p->p_ucred = crdup(p->p_ucred);
1584 		crfree(cred);
1585 		cred = p->p_ucred;
1586 		crhold(cred);
1587 		cred->cr_uid = 0;
1588 		cred->cr_gid = 0;
1589 	}
1590 
1591 	NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, p);
1592 
1593 	error = vn_open(&nd, O_CREAT | FWRITE | O_NOFOLLOW | O_NONBLOCK,
1594 	    S_IRUSR | S_IWUSR);
1595 
1596 	if (error)
1597 		goto out;
1598 
1599 	/*
1600 	 * Don't dump to non-regular files, files with links, or files
1601 	 * owned by someone else.
1602 	 */
1603 	vp = nd.ni_vp;
1604 	if ((error = VOP_GETATTR(vp, &vattr, cred, p)) != 0) {
1605 		VOP_UNLOCK(vp);
1606 		vn_close(vp, FWRITE, cred, p);
1607 		goto out;
1608 	}
1609 	if (vp->v_type != VREG || vattr.va_nlink != 1 ||
1610 	    vattr.va_mode & ((VREAD | VWRITE) >> 3 | (VREAD | VWRITE) >> 6) ||
1611 	    vattr.va_uid != cred->cr_uid) {
1612 		error = EACCES;
1613 		VOP_UNLOCK(vp);
1614 		vn_close(vp, FWRITE, cred, p);
1615 		goto out;
1616 	}
1617 	VATTR_NULL(&vattr);
1618 	vattr.va_size = 0;
1619 	VOP_SETATTR(vp, &vattr, cred, p);
1620 	pr->ps_acflag |= ACORE;
1621 
1622 	io.io_proc = p;
1623 	io.io_vp = vp;
1624 	io.io_cred = cred;
1625 	io.io_offset = 0;
1626 	VOP_UNLOCK(vp);
1627 	vref(vp);
1628 	error = vn_close(vp, FWRITE, cred, p);
1629 	if (error == 0)
1630 		error = (*pr->ps_emul->e_coredump)(p, &io);
1631 	vrele(vp);
1632 out:
1633 	crfree(cred);
1634 	return (error);
1635 #endif
1636 }
1637 
1638 #ifndef SMALL_KERNEL
1639 int
1640 coredump_write(void *cookie, enum uio_seg segflg, const void *data, size_t len)
1641 {
1642 	struct coredump_iostate *io = cookie;
1643 	off_t coffset = 0;
1644 	size_t csize;
1645 	int chunk, error;
1646 
1647 	csize = len;
1648 	do {
1649 		if (sigmask(SIGKILL) &
1650 		    (io->io_proc->p_siglist | io->io_proc->p_p->ps_siglist))
1651 			return (EINTR);
1652 
1653 		/* Rest of the loop sleeps with lock held, so... */
1654 		yield();
1655 
1656 		chunk = MIN(csize, MAXPHYS);
1657 		error = vn_rdwr(UIO_WRITE, io->io_vp,
1658 		    (caddr_t)data + coffset, chunk,
1659 		    io->io_offset + coffset, segflg,
1660 		    IO_UNIT, io->io_cred, NULL, io->io_proc);
1661 		if (error) {
1662 			struct process *pr = io->io_proc->p_p;
1663 
1664 			if (error == ENOSPC)
1665 				log(LOG_ERR,
1666 				    "coredump of %s(%d) failed, filesystem full\n",
1667 				    pr->ps_comm, pr->ps_pid);
1668 			else
1669 				log(LOG_ERR,
1670 				    "coredump of %s(%d), write failed: errno %d\n",
1671 				    pr->ps_comm, pr->ps_pid, error);
1672 			return (error);
1673 		}
1674 
1675 		coffset += chunk;
1676 		csize -= chunk;
1677 	} while (csize > 0);
1678 
1679 	io->io_offset += len;
1680 	return (0);
1681 }
1682 
1683 void
1684 coredump_unmap(void *cookie, vaddr_t start, vaddr_t end)
1685 {
1686 	struct coredump_iostate *io = cookie;
1687 
1688 	uvm_unmap(&io->io_proc->p_vmspace->vm_map, start, end);
1689 }
1690 
1691 #endif	/* !SMALL_KERNEL */
1692 
1693 /*
1694  * Nonexistent system call-- signal process (may want to handle it).
1695  * Flag error in case process won't see signal immediately (blocked or ignored).
1696  */
1697 int
1698 sys_nosys(struct proc *p, void *v, register_t *retval)
1699 {
1700 
1701 	ptsignal(p, SIGSYS, STHREAD);
1702 	return (ENOSYS);
1703 }
1704 
1705 int
1706 sys___thrsigdivert(struct proc *p, void *v, register_t *retval)
1707 {
1708 	static int sigwaitsleep;
1709 	struct sys___thrsigdivert_args /* {
1710 		syscallarg(sigset_t) sigmask;
1711 		syscallarg(siginfo_t *) info;
1712 		syscallarg(const struct timespec *) timeout;
1713 	} */ *uap = v;
1714 	struct process *pr = p->p_p;
1715 	sigset_t *m;
1716 	sigset_t mask = SCARG(uap, sigmask) &~ sigcantmask;
1717 	siginfo_t si;
1718 	uint64_t to_ticks = 0;
1719 	int timeinvalid = 0;
1720 	int error = 0;
1721 
1722 	memset(&si, 0, sizeof(si));
1723 
1724 	if (SCARG(uap, timeout) != NULL) {
1725 		struct timespec ts;
1726 		if ((error = copyin(SCARG(uap, timeout), &ts, sizeof(ts))) != 0)
1727 			return (error);
1728 #ifdef KTRACE
1729 		if (KTRPOINT(p, KTR_STRUCT))
1730 			ktrreltimespec(p, &ts);
1731 #endif
1732 		if (!timespecisvalid(&ts))
1733 			timeinvalid = 1;
1734 		else {
1735 			to_ticks = (uint64_t)hz * ts.tv_sec +
1736 			    ts.tv_nsec / (tick * 1000);
1737 			if (to_ticks > INT_MAX)
1738 				to_ticks = INT_MAX;
1739 			if (to_ticks == 0 && ts.tv_nsec)
1740 				to_ticks = 1;
1741 		}
1742 	}
1743 
1744 	dosigsuspend(p, p->p_sigmask &~ mask);
1745 	for (;;) {
1746 		si.si_signo = CURSIG(p);
1747 		if (si.si_signo != 0) {
1748 			sigset_t smask = sigmask(si.si_signo);
1749 			if (smask & mask) {
1750 				if (p->p_siglist & smask)
1751 					m = &p->p_siglist;
1752 				else if (pr->ps_siglist & smask)
1753 					m = &pr->ps_siglist;
1754 				else {
1755 					/* signal got eaten by someone else? */
1756 					continue;
1757 				}
1758 				atomic_clearbits_int(m, smask);
1759 				error = 0;
1760 				break;
1761 			}
1762 		}
1763 
1764 		/* per-POSIX, delay this error until after the above */
1765 		if (timeinvalid)
1766 			error = EINVAL;
1767 
1768 		if (SCARG(uap, timeout) != NULL && to_ticks == 0)
1769 			error = EAGAIN;
1770 
1771 		if (error != 0)
1772 			break;
1773 
1774 		error = tsleep(&sigwaitsleep, PPAUSE|PCATCH, "sigwait",
1775 		    (int)to_ticks);
1776 	}
1777 
1778 	if (error == 0) {
1779 		*retval = si.si_signo;
1780 		if (SCARG(uap, info) != NULL)
1781 			error = copyout(&si, SCARG(uap, info), sizeof(si));
1782 	} else if (error == ERESTART && SCARG(uap, timeout) != NULL) {
1783 		/*
1784 		 * Restarting is wrong if there's a timeout, as it'll be
1785 		 * for the same interval again
1786 		 */
1787 		error = EINTR;
1788 	}
1789 
1790 	return (error);
1791 }
1792 
1793 void
1794 initsiginfo(siginfo_t *si, int sig, u_long trapno, int code, union sigval val)
1795 {
1796 	memset(si, 0, sizeof(*si));
1797 
1798 	si->si_signo = sig;
1799 	si->si_code = code;
1800 	if (code == SI_USER) {
1801 		si->si_value = val;
1802 	} else {
1803 		switch (sig) {
1804 		case SIGSEGV:
1805 		case SIGILL:
1806 		case SIGBUS:
1807 		case SIGFPE:
1808 			si->si_addr = val.sival_ptr;
1809 			si->si_trapno = trapno;
1810 			break;
1811 		case SIGXFSZ:
1812 			break;
1813 		}
1814 	}
1815 }
1816 
1817 int
1818 filt_sigattach(struct knote *kn)
1819 {
1820 	struct process *pr = curproc->p_p;
1821 
1822 	if (kn->kn_id >= NSIG)
1823 		return EINVAL;
1824 
1825 	kn->kn_ptr.p_process = pr;
1826 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
1827 
1828 	/* XXX lock the proc here while adding to the list? */
1829 	SLIST_INSERT_HEAD(&pr->ps_klist, kn, kn_selnext);
1830 
1831 	return (0);
1832 }
1833 
1834 void
1835 filt_sigdetach(struct knote *kn)
1836 {
1837 	struct process *pr = kn->kn_ptr.p_process;
1838 
1839 	SLIST_REMOVE(&pr->ps_klist, kn, knote, kn_selnext);
1840 }
1841 
1842 /*
1843  * signal knotes are shared with proc knotes, so we apply a mask to
1844  * the hint in order to differentiate them from process hints.  This
1845  * could be avoided by using a signal-specific knote list, but probably
1846  * isn't worth the trouble.
1847  */
1848 int
1849 filt_signal(struct knote *kn, long hint)
1850 {
1851 
1852 	if (hint & NOTE_SIGNAL) {
1853 		hint &= ~NOTE_SIGNAL;
1854 
1855 		if (kn->kn_id == hint)
1856 			kn->kn_data++;
1857 	}
1858 	return (kn->kn_data != 0);
1859 }
1860 
1861 void
1862 userret(struct proc *p)
1863 {
1864 	int signum;
1865 
1866 	/* send SIGPROF or SIGVTALRM if their timers interrupted this thread */
1867 	if (p->p_flag & P_PROFPEND) {
1868 		atomic_clearbits_int(&p->p_flag, P_PROFPEND);
1869 		KERNEL_LOCK();
1870 		psignal(p, SIGPROF);
1871 		KERNEL_UNLOCK();
1872 	}
1873 	if (p->p_flag & P_ALRMPEND) {
1874 		atomic_clearbits_int(&p->p_flag, P_ALRMPEND);
1875 		KERNEL_LOCK();
1876 		psignal(p, SIGVTALRM);
1877 		KERNEL_UNLOCK();
1878 	}
1879 
1880 	if (SIGPENDING(p) != 0) {
1881 		KERNEL_LOCK();
1882 		while ((signum = CURSIG(p)) != 0)
1883 			postsig(p, signum);
1884 		KERNEL_UNLOCK();
1885 	}
1886 
1887 	/*
1888 	 * If P_SIGSUSPEND is still set here, then we still need to restore
1889 	 * the original sigmask before returning to userspace.  Also, this
1890 	 * might unmask some pending signals, so we need to check a second
1891 	 * time for signals to post.
1892 	 */
1893 	if (p->p_flag & P_SIGSUSPEND) {
1894 		atomic_clearbits_int(&p->p_flag, P_SIGSUSPEND);
1895 		p->p_sigmask = p->p_oldmask;
1896 
1897 		KERNEL_LOCK();
1898 		while ((signum = CURSIG(p)) != 0)
1899 			postsig(p, signum);
1900 		KERNEL_UNLOCK();
1901 	}
1902 
1903 	if (p->p_flag & P_SUSPSINGLE) {
1904 		KERNEL_LOCK();
1905 		single_thread_check(p, 0);
1906 		KERNEL_UNLOCK();
1907 	}
1908 
1909 	WITNESS_WARN(WARN_PANIC, NULL, "userret: returning");
1910 
1911 	p->p_cpu->ci_schedstate.spc_curpriority = p->p_priority = p->p_usrpri;
1912 }
1913 
1914 int
1915 single_thread_check(struct proc *p, int deep)
1916 {
1917 	struct process *pr = p->p_p;
1918 
1919 	if (pr->ps_single != NULL && pr->ps_single != p) {
1920 		do {
1921 			int s;
1922 
1923 			/* if we're in deep, we need to unwind to the edge */
1924 			if (deep) {
1925 				if (pr->ps_flags & PS_SINGLEUNWIND)
1926 					return (ERESTART);
1927 				if (pr->ps_flags & PS_SINGLEEXIT)
1928 					return (EINTR);
1929 			}
1930 
1931 			if (--pr->ps_singlecount == 0)
1932 				wakeup(&pr->ps_singlecount);
1933 			if (pr->ps_flags & PS_SINGLEEXIT)
1934 				exit1(p, 0, EXIT_THREAD_NOCHECK);
1935 
1936 			/* not exiting and don't need to unwind, so suspend */
1937 			SCHED_LOCK(s);
1938 			p->p_stat = SSTOP;
1939 			mi_switch();
1940 			SCHED_UNLOCK(s);
1941 		} while (pr->ps_single != NULL);
1942 	}
1943 
1944 	return (0);
1945 }
1946 
1947 /*
1948  * Stop other threads in the process.  The mode controls how and
1949  * where the other threads should stop:
1950  *  - SINGLE_SUSPEND: stop wherever they are, will later either be told to exit
1951  *    (by setting to SINGLE_EXIT) or be released (via single_thread_clear())
1952  *  - SINGLE_PTRACE: stop wherever they are, will wait for them to stop
1953  *    later (via single_thread_wait()) and released as with SINGLE_SUSPEND
1954  *  - SINGLE_UNWIND: just unwind to kernel boundary, will be told to exit
1955  *    or released as with SINGLE_SUSPEND
1956  *  - SINGLE_EXIT: unwind to kernel boundary and exit
1957  */
1958 int
1959 single_thread_set(struct proc *p, enum single_thread_mode mode, int deep)
1960 {
1961 	struct process *pr = p->p_p;
1962 	struct proc *q;
1963 	int error;
1964 
1965 	KERNEL_ASSERT_LOCKED();
1966 
1967 	if ((error = single_thread_check(p, deep)))
1968 		return error;
1969 
1970 	switch (mode) {
1971 	case SINGLE_SUSPEND:
1972 	case SINGLE_PTRACE:
1973 		break;
1974 	case SINGLE_UNWIND:
1975 		atomic_setbits_int(&pr->ps_flags, PS_SINGLEUNWIND);
1976 		break;
1977 	case SINGLE_EXIT:
1978 		atomic_setbits_int(&pr->ps_flags, PS_SINGLEEXIT);
1979 		atomic_clearbits_int(&pr->ps_flags, PS_SINGLEUNWIND);
1980 		break;
1981 #ifdef DIAGNOSTIC
1982 	default:
1983 		panic("single_thread_mode = %d", mode);
1984 #endif
1985 	}
1986 	pr->ps_single = p;
1987 	pr->ps_singlecount = 0;
1988 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
1989 		int s;
1990 
1991 		if (q == p)
1992 			continue;
1993 		if (q->p_flag & P_WEXIT) {
1994 			if (mode == SINGLE_EXIT) {
1995 				SCHED_LOCK(s);
1996 				if (q->p_stat == SSTOP) {
1997 					setrunnable(q);
1998 					pr->ps_singlecount++;
1999 				}
2000 				SCHED_UNLOCK(s);
2001 			}
2002 			continue;
2003 		}
2004 		SCHED_LOCK(s);
2005 		atomic_setbits_int(&q->p_flag, P_SUSPSINGLE);
2006 		switch (q->p_stat) {
2007 		case SIDL:
2008 		case SRUN:
2009 			pr->ps_singlecount++;
2010 			break;
2011 		case SSLEEP:
2012 			/* if it's not interruptible, then just have to wait */
2013 			if (q->p_flag & P_SINTR) {
2014 				/* merely need to suspend?  just stop it */
2015 				if (mode == SINGLE_SUSPEND ||
2016 				    mode == SINGLE_PTRACE) {
2017 					q->p_stat = SSTOP;
2018 					break;
2019 				}
2020 				/* need to unwind or exit, so wake it */
2021 				setrunnable(q);
2022 			}
2023 			pr->ps_singlecount++;
2024 			break;
2025 		case SSTOP:
2026 			if (mode == SINGLE_EXIT) {
2027 				setrunnable(q);
2028 				pr->ps_singlecount++;
2029 			}
2030 			break;
2031 		case SDEAD:
2032 			break;
2033 		case SONPROC:
2034 			pr->ps_singlecount++;
2035 			signotify(q);
2036 			break;
2037 		}
2038 		SCHED_UNLOCK(s);
2039 	}
2040 
2041 	if (mode != SINGLE_PTRACE)
2042 		single_thread_wait(pr);
2043 
2044 	return 0;
2045 }
2046 
2047 void
2048 single_thread_wait(struct process *pr)
2049 {
2050 	/* wait until they're all suspended */
2051 	while (pr->ps_singlecount > 0)
2052 		tsleep(&pr->ps_singlecount, PUSER, "suspend", 0);
2053 }
2054 
2055 void
2056 single_thread_clear(struct proc *p, int flag)
2057 {
2058 	struct process *pr = p->p_p;
2059 	struct proc *q;
2060 
2061 	KASSERT(pr->ps_single == p);
2062 	KERNEL_ASSERT_LOCKED();
2063 
2064 	pr->ps_single = NULL;
2065 	atomic_clearbits_int(&pr->ps_flags, PS_SINGLEUNWIND | PS_SINGLEEXIT);
2066 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
2067 		int s;
2068 
2069 		if (q == p || (q->p_flag & P_SUSPSINGLE) == 0)
2070 			continue;
2071 		atomic_clearbits_int(&q->p_flag, P_SUSPSINGLE);
2072 
2073 		/*
2074 		 * if the thread was only stopped for single threading
2075 		 * then clearing that either makes it runnable or puts
2076 		 * it back into some sleep queue
2077 		 */
2078 		SCHED_LOCK(s);
2079 		if (q->p_stat == SSTOP && (q->p_flag & flag) == 0) {
2080 			if (q->p_wchan == 0)
2081 				setrunnable(q);
2082 			else
2083 				q->p_stat = SSLEEP;
2084 		}
2085 		SCHED_UNLOCK(s);
2086 	}
2087 }
2088 
2089 void
2090 sigio_del(struct sigiolst *rmlist)
2091 {
2092 	struct sigio *sigio;
2093 
2094 	while ((sigio = LIST_FIRST(rmlist)) != NULL) {
2095 		LIST_REMOVE(sigio, sio_pgsigio);
2096 		crfree(sigio->sio_ucred);
2097 		free(sigio, M_SIGIO, sizeof(*sigio));
2098 	}
2099 }
2100 
2101 void
2102 sigio_unlink(struct sigio_ref *sir, struct sigiolst *rmlist)
2103 {
2104 	struct sigio *sigio;
2105 
2106 	MUTEX_ASSERT_LOCKED(&sigio_lock);
2107 
2108 	sigio = sir->sir_sigio;
2109 	if (sigio != NULL) {
2110 		KASSERT(sigio->sio_myref == sir);
2111 		sir->sir_sigio = NULL;
2112 
2113 		if (sigio->sio_pgid > 0)
2114 			sigio->sio_proc = NULL;
2115 		else
2116 			sigio->sio_pgrp = NULL;
2117 		LIST_REMOVE(sigio, sio_pgsigio);
2118 
2119 		LIST_INSERT_HEAD(rmlist, sigio, sio_pgsigio);
2120 	}
2121 }
2122 
2123 void
2124 sigio_free(struct sigio_ref *sir)
2125 {
2126 	struct sigiolst rmlist;
2127 
2128 	if (sir->sir_sigio == NULL)
2129 		return;
2130 
2131 	LIST_INIT(&rmlist);
2132 
2133 	mtx_enter(&sigio_lock);
2134 	sigio_unlink(sir, &rmlist);
2135 	mtx_leave(&sigio_lock);
2136 
2137 	sigio_del(&rmlist);
2138 }
2139 
2140 void
2141 sigio_freelist(struct sigiolst *sigiolst)
2142 {
2143 	struct sigiolst rmlist;
2144 	struct sigio *sigio;
2145 
2146 	if (LIST_EMPTY(sigiolst))
2147 		return;
2148 
2149 	LIST_INIT(&rmlist);
2150 
2151 	mtx_enter(&sigio_lock);
2152 	while ((sigio = LIST_FIRST(sigiolst)) != NULL)
2153 		sigio_unlink(sigio->sio_myref, &rmlist);
2154 	mtx_leave(&sigio_lock);
2155 
2156 	sigio_del(&rmlist);
2157 }
2158 
2159 int
2160 sigio_setown(struct sigio_ref *sir, pid_t pgid)
2161 {
2162 	struct sigiolst rmlist;
2163 	struct proc *p = curproc;
2164 	struct pgrp *pgrp = NULL;
2165 	struct process *pr = NULL;
2166 	struct sigio *sigio;
2167 	int error;
2168 
2169 	if (pgid == 0) {
2170 		sigio_free(sir);
2171 		return (0);
2172 	}
2173 
2174 	sigio = malloc(sizeof(*sigio), M_SIGIO, M_WAITOK);
2175 	sigio->sio_pgid = pgid;
2176 	sigio->sio_ucred = crhold(p->p_ucred);
2177 	sigio->sio_myref = sir;
2178 
2179 	LIST_INIT(&rmlist);
2180 
2181 	/*
2182 	 * The kernel lock, and not sleeping between prfind()/pgfind() and
2183 	 * linking of the sigio ensure that the process or process group does
2184 	 * not disappear unexpectedly.
2185 	 */
2186 	KERNEL_LOCK();
2187 	mtx_enter(&sigio_lock);
2188 
2189 	if (pgid > 0) {
2190 		pr = prfind(pgid);
2191 		if (pr == NULL) {
2192 			error = ESRCH;
2193 			goto fail;
2194 		}
2195 
2196 		/*
2197 		 * Policy - Don't allow a process to FSETOWN a process
2198 		 * in another session.
2199 		 *
2200 		 * Remove this test to allow maximum flexibility or
2201 		 * restrict FSETOWN to the current process or process
2202 		 * group for maximum safety.
2203 		 */
2204 		if (pr->ps_session != p->p_p->ps_session) {
2205 			error = EPERM;
2206 			goto fail;
2207 		}
2208 
2209 		if ((pr->ps_flags & PS_EXITING) != 0) {
2210 			error = ESRCH;
2211 			goto fail;
2212 		}
2213 	} else /* if (pgid < 0) */ {
2214 		pgrp = pgfind(-pgid);
2215 		if (pgrp == NULL) {
2216 			error = ESRCH;
2217 			goto fail;
2218 		}
2219 
2220 		/*
2221 		 * Policy - Don't allow a process to FSETOWN a process
2222 		 * in another session.
2223 		 *
2224 		 * Remove this test to allow maximum flexibility or
2225 		 * restrict FSETOWN to the current process or process
2226 		 * group for maximum safety.
2227 		 */
2228 		if (pgrp->pg_session != p->p_p->ps_session) {
2229 			error = EPERM;
2230 			goto fail;
2231 		}
2232 	}
2233 
2234 	if (pgid > 0) {
2235 		sigio->sio_proc = pr;
2236 		LIST_INSERT_HEAD(&pr->ps_sigiolst, sigio, sio_pgsigio);
2237 	} else {
2238 		sigio->sio_pgrp = pgrp;
2239 		LIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
2240 	}
2241 
2242 	sigio_unlink(sir, &rmlist);
2243 	sir->sir_sigio = sigio;
2244 
2245 	mtx_leave(&sigio_lock);
2246 	KERNEL_UNLOCK();
2247 
2248 	sigio_del(&rmlist);
2249 
2250 	return (0);
2251 
2252 fail:
2253 	mtx_leave(&sigio_lock);
2254 	KERNEL_UNLOCK();
2255 
2256 	crfree(sigio->sio_ucred);
2257 	free(sigio, M_SIGIO, sizeof(*sigio));
2258 
2259 	return (error);
2260 }
2261 
2262 pid_t
2263 sigio_getown(struct sigio_ref *sir)
2264 {
2265 	struct sigio *sigio;
2266 	pid_t pgid = 0;
2267 
2268 	mtx_enter(&sigio_lock);
2269 	sigio = sir->sir_sigio;
2270 	if (sigio != NULL)
2271 		pgid = sigio->sio_pgid;
2272 	mtx_leave(&sigio_lock);
2273 
2274 	return (pgid);
2275 }
2276 
2277 void
2278 sigio_copy(struct sigio_ref *dst, struct sigio_ref *src)
2279 {
2280 	struct sigiolst rmlist;
2281 	struct sigio *newsigio, *sigio;
2282 
2283 	sigio_free(dst);
2284 
2285 	if (src->sir_sigio == NULL)
2286 		return;
2287 
2288 	newsigio = malloc(sizeof(*newsigio), M_SIGIO, M_WAITOK);
2289 	LIST_INIT(&rmlist);
2290 
2291 	mtx_enter(&sigio_lock);
2292 
2293 	sigio = src->sir_sigio;
2294 	if (sigio == NULL) {
2295 		mtx_leave(&sigio_lock);
2296 		free(newsigio, M_SIGIO, sizeof(*newsigio));
2297 		return;
2298 	}
2299 
2300 	newsigio->sio_pgid = sigio->sio_pgid;
2301 	newsigio->sio_ucred = crhold(sigio->sio_ucred);
2302 	newsigio->sio_myref = dst;
2303 	if (newsigio->sio_pgid > 0) {
2304 		newsigio->sio_proc = sigio->sio_proc;
2305 		LIST_INSERT_HEAD(&newsigio->sio_proc->ps_sigiolst, newsigio,
2306 		    sio_pgsigio);
2307 	} else {
2308 		newsigio->sio_pgrp = sigio->sio_pgrp;
2309 		LIST_INSERT_HEAD(&newsigio->sio_pgrp->pg_sigiolst, newsigio,
2310 		    sio_pgsigio);
2311 	}
2312 
2313 	sigio_unlink(dst, &rmlist);
2314 	dst->sir_sigio = newsigio;
2315 
2316 	mtx_leave(&sigio_lock);
2317 
2318 	sigio_del(&rmlist);
2319 }
2320