xref: /netbsd-src/sys/kern/kern_sig.c (revision b8c616269f5ebf18ab2e35cb8099d683130a177c)
1 /*	$NetBSD: kern_sig.c,v 1.131 2003/02/03 22:56:23 jdolecek Exp $	*/
2 
3 /*
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *	This product includes software developed by the University of
23  *	California, Berkeley and its contributors.
24  * 4. Neither the name of the University nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  *
40  *	@(#)kern_sig.c	8.14 (Berkeley) 5/14/95
41  */
42 
43 #include <sys/cdefs.h>
44 __KERNEL_RCSID(0, "$NetBSD: kern_sig.c,v 1.131 2003/02/03 22:56:23 jdolecek Exp $");
45 
46 #include "opt_ktrace.h"
47 #include "opt_compat_sunos.h"
48 #include "opt_compat_netbsd32.h"
49 
50 #define	SIGPROP		/* include signal properties table */
51 #include <sys/param.h>
52 #include <sys/signalvar.h>
53 #include <sys/resourcevar.h>
54 #include <sys/namei.h>
55 #include <sys/vnode.h>
56 #include <sys/proc.h>
57 #include <sys/systm.h>
58 #include <sys/timeb.h>
59 #include <sys/times.h>
60 #include <sys/buf.h>
61 #include <sys/acct.h>
62 #include <sys/file.h>
63 #include <sys/kernel.h>
64 #include <sys/wait.h>
65 #include <sys/ktrace.h>
66 #include <sys/syslog.h>
67 #include <sys/stat.h>
68 #include <sys/core.h>
69 #include <sys/filedesc.h>
70 #include <sys/malloc.h>
71 #include <sys/pool.h>
72 #include <sys/ucontext.h>
73 #include <sys/sa.h>
74 #include <sys/savar.h>
75 #include <sys/exec.h>
76 
77 #include <sys/mount.h>
78 #include <sys/syscallargs.h>
79 
80 #include <machine/cpu.h>
81 
82 #include <sys/user.h>		/* for coredump */
83 
84 #include <uvm/uvm_extern.h>
85 
86 static void	proc_stop(struct proc *p);
87 static int	build_corename(struct proc *, char [MAXPATHLEN]);
88 sigset_t	contsigmask, stopsigmask, sigcantmask;
89 
90 struct pool	sigacts_pool;	/* memory pool for sigacts structures */
91 struct pool	siginfo_pool;	/* memory pool for siginfo structures */
92 
93 /*
94  * Can process p, with pcred pc, send the signal signum to process q?
95  */
96 #define	CANSIGNAL(p, pc, q, signum) \
97 	((pc)->pc_ucred->cr_uid == 0 || \
98 	    (pc)->p_ruid == (q)->p_cred->p_ruid || \
99 	    (pc)->pc_ucred->cr_uid == (q)->p_cred->p_ruid || \
100 	    (pc)->p_ruid == (q)->p_ucred->cr_uid || \
101 	    (pc)->pc_ucred->cr_uid == (q)->p_ucred->cr_uid || \
102 	    ((signum) == SIGCONT && (q)->p_session == (p)->p_session))
103 
104 /*
105  * Initialize signal-related data structures.
106  */
107 void
108 signal_init(void)
109 {
110 
111 	pool_init(&sigacts_pool, sizeof(struct sigacts), 0, 0, 0, "sigapl",
112 	    &pool_allocator_nointr);
113 	pool_init(&siginfo_pool, sizeof(siginfo_t), 0, 0, 0, "siginfo",
114 	    &pool_allocator_nointr);
115 }
116 
117 /*
118  * Create an initial sigctx structure, using the same signal state
119  * as p. If 'share' is set, share the sigctx_proc part, otherwise just
120  * copy it from parent.
121  */
122 void
123 sigactsinit(struct proc *np, struct proc *pp, int share)
124 {
125 	struct sigacts *ps;
126 
127 	if (share) {
128 		np->p_sigacts = pp->p_sigacts;
129 		pp->p_sigacts->sa_refcnt++;
130 	} else {
131 		ps = pool_get(&sigacts_pool, PR_WAITOK);
132 		if (pp)
133 			memcpy(ps, pp->p_sigacts, sizeof(struct sigacts));
134 		else
135 			memset(ps, '\0', sizeof(struct sigacts));
136 		ps->sa_refcnt = 1;
137 		np->p_sigacts = ps;
138 	}
139 }
140 
141 /*
142  * Make this process not share its sigctx, maintaining all
143  * signal state.
144  */
145 void
146 sigactsunshare(struct proc *p)
147 {
148 	struct sigacts *oldps;
149 
150 	if (p->p_sigacts->sa_refcnt == 1)
151 		return;
152 
153 	oldps = p->p_sigacts;
154 	sigactsinit(p, NULL, 0);
155 
156 	if (--oldps->sa_refcnt == 0)
157 		pool_put(&sigacts_pool, oldps);
158 }
159 
160 /*
161  * Release a sigctx structure.
162  */
163 void
164 sigactsfree(struct proc *p)
165 {
166 	struct sigacts *ps;
167 
168 	ps = p->p_sigacts;
169 	if (--ps->sa_refcnt > 0)
170 		return;
171 
172 	pool_put(&sigacts_pool, ps);
173 }
174 
175 int
176 sigaction1(struct proc *p, int signum, const struct sigaction *nsa,
177 	struct sigaction *osa, void *tramp, int vers)
178 {
179 	struct sigacts	*ps;
180 	int		prop;
181 
182 	ps = p->p_sigacts;
183 	if (signum <= 0 || signum >= NSIG)
184 		return (EINVAL);
185 
186 	/*
187 	 * Trampoline ABI version 0 is reserved for the legacy
188 	 * kernel-provided on-stack trampoline.  Conversely, if
189 	 * we are using a non-0 ABI version, we must have a
190 	 * trampoline.
191 	 */
192 	if ((vers != 0 && tramp == NULL) ||
193 	    (vers == 0 && tramp != NULL))
194 		return (EINVAL);
195 
196 	if (osa)
197 		*osa = SIGACTION_PS(ps, signum);
198 
199 	if (nsa) {
200 		if (nsa->sa_flags & ~SA_ALLBITS)
201 			return (EINVAL);
202 
203 		prop = sigprop[signum];
204 		if (prop & SA_CANTMASK)
205 			return (EINVAL);
206 
207 		(void) splsched();	/* XXXSMP */
208 		SIGACTION_PS(ps, signum) = *nsa;
209 		ps->sa_sigdesc[signum].sd_tramp = tramp;
210 		ps->sa_sigdesc[signum].sd_vers = vers;
211 		sigminusset(&sigcantmask, &SIGACTION_PS(ps, signum).sa_mask);
212 		if ((prop & SA_NORESET) != 0)
213 			SIGACTION_PS(ps, signum).sa_flags &= ~SA_RESETHAND;
214 		if (signum == SIGCHLD) {
215 			if (nsa->sa_flags & SA_NOCLDSTOP)
216 				p->p_flag |= P_NOCLDSTOP;
217 			else
218 				p->p_flag &= ~P_NOCLDSTOP;
219 			if (nsa->sa_flags & SA_NOCLDWAIT) {
220 				/*
221 				 * Paranoia: since SA_NOCLDWAIT is implemented
222 				 * by reparenting the dying child to PID 1 (and
223 				 * trust it to reap the zombie), PID 1 itself
224 				 * is forbidden to set SA_NOCLDWAIT.
225 				 */
226 				if (p->p_pid == 1)
227 					p->p_flag &= ~P_NOCLDWAIT;
228 				else
229 					p->p_flag |= P_NOCLDWAIT;
230 			} else
231 				p->p_flag &= ~P_NOCLDWAIT;
232 		}
233 		if ((nsa->sa_flags & SA_NODEFER) == 0)
234 			sigaddset(&SIGACTION_PS(ps, signum).sa_mask, signum);
235 		else
236 			sigdelset(&SIGACTION_PS(ps, signum).sa_mask, signum);
237 		/*
238 	 	 * Set bit in p_sigctx.ps_sigignore for signals that are set to
239 		 * SIG_IGN, and for signals set to SIG_DFL where the default is
240 		 * to ignore. However, don't put SIGCONT in
241 		 * p_sigctx.ps_sigignore, as we have to restart the process.
242 	 	 */
243 		if (nsa->sa_handler == SIG_IGN ||
244 		    (nsa->sa_handler == SIG_DFL && (prop & SA_IGNORE) != 0)) {
245 						/* never to be seen again */
246 			sigdelset(&p->p_sigctx.ps_siglist, signum);
247 			if (signum != SIGCONT) {
248 						/* easier in psignal */
249 				sigaddset(&p->p_sigctx.ps_sigignore, signum);
250 			}
251 			sigdelset(&p->p_sigctx.ps_sigcatch, signum);
252 		} else {
253 			sigdelset(&p->p_sigctx.ps_sigignore, signum);
254 			if (nsa->sa_handler == SIG_DFL)
255 				sigdelset(&p->p_sigctx.ps_sigcatch, signum);
256 			else
257 				sigaddset(&p->p_sigctx.ps_sigcatch, signum);
258 		}
259 		(void) spl0();
260 	}
261 
262 	return (0);
263 }
264 
265 /* ARGSUSED */
266 int
267 sys___sigaction14(struct lwp *l, void *v, register_t *retval)
268 {
269 	struct sys___sigaction14_args /* {
270 		syscallarg(int)				signum;
271 		syscallarg(const struct sigaction *)	nsa;
272 		syscallarg(struct sigaction *)		osa;
273 	} */ *uap = v;
274 	struct proc		*p;
275 	struct sigaction	nsa, osa;
276 	int			error;
277 
278 	if (SCARG(uap, nsa)) {
279 		error = copyin(SCARG(uap, nsa), &nsa, sizeof(nsa));
280 		if (error)
281 			return (error);
282 	}
283 	p = l->l_proc;
284 	error = sigaction1(p, SCARG(uap, signum),
285 	    SCARG(uap, nsa) ? &nsa : 0, SCARG(uap, osa) ? &osa : 0,
286 	    NULL, 0);
287 	if (error)
288 		return (error);
289 	if (SCARG(uap, osa)) {
290 		error = copyout(&osa, SCARG(uap, osa), sizeof(osa));
291 		if (error)
292 			return (error);
293 	}
294 	return (0);
295 }
296 
297 /* ARGSUSED */
298 int
299 sys___sigaction_sigtramp(struct lwp *l, void *v, register_t *retval)
300 {
301 	struct sys___sigaction_sigtramp_args /* {
302 		syscallarg(int)				signum;
303 		syscallarg(const struct sigaction *)	nsa;
304 		syscallarg(struct sigaction *)		osa;
305 		syscallarg(void *)			tramp;
306 		syscallarg(int)				vers;
307 	} */ *uap = v;
308 	struct proc *p = l->l_proc;
309 	struct sigaction nsa, osa;
310 	int error;
311 
312 	if (SCARG(uap, nsa)) {
313 		error = copyin(SCARG(uap, nsa), &nsa, sizeof(nsa));
314 		if (error)
315 			return (error);
316 	}
317 	error = sigaction1(p, SCARG(uap, signum),
318 	    SCARG(uap, nsa) ? &nsa : 0, SCARG(uap, osa) ? &osa : 0,
319 	    SCARG(uap, tramp), SCARG(uap, vers));
320 	if (error)
321 		return (error);
322 	if (SCARG(uap, osa)) {
323 		error = copyout(&osa, SCARG(uap, osa), sizeof(osa));
324 		if (error)
325 			return (error);
326 	}
327 	return (0);
328 }
329 
330 /*
331  * Initialize signal state for process 0;
332  * set to ignore signals that are ignored by default and disable the signal
333  * stack.
334  */
335 void
336 siginit(struct proc *p)
337 {
338 	struct sigacts	*ps;
339 	int		signum, prop;
340 
341 	ps = p->p_sigacts;
342 	sigemptyset(&contsigmask);
343 	sigemptyset(&stopsigmask);
344 	sigemptyset(&sigcantmask);
345 	for (signum = 1; signum < NSIG; signum++) {
346 		prop = sigprop[signum];
347 		if (prop & SA_CONT)
348 			sigaddset(&contsigmask, signum);
349 		if (prop & SA_STOP)
350 			sigaddset(&stopsigmask, signum);
351 		if (prop & SA_CANTMASK)
352 			sigaddset(&sigcantmask, signum);
353 		if (prop & SA_IGNORE && signum != SIGCONT)
354 			sigaddset(&p->p_sigctx.ps_sigignore, signum);
355 		sigemptyset(&SIGACTION_PS(ps, signum).sa_mask);
356 		SIGACTION_PS(ps, signum).sa_flags = SA_RESTART;
357 	}
358 	sigemptyset(&p->p_sigctx.ps_sigcatch);
359 	p->p_flag &= ~P_NOCLDSTOP;
360 
361 	/*
362 	 * Reset stack state to the user stack.
363 	 */
364 	p->p_sigctx.ps_sigstk.ss_flags = SS_DISABLE;
365 	p->p_sigctx.ps_sigstk.ss_size = 0;
366 	p->p_sigctx.ps_sigstk.ss_sp = 0;
367 
368 	/* One reference. */
369 	ps->sa_refcnt = 1;
370 }
371 
372 /*
373  * Reset signals for an exec of the specified process.
374  */
375 void
376 execsigs(struct proc *p)
377 {
378 	struct sigacts	*ps;
379 	int		signum, prop;
380 
381 	sigactsunshare(p);
382 
383 	ps = p->p_sigacts;
384 
385 	/*
386 	 * Reset caught signals.  Held signals remain held
387 	 * through p_sigctx.ps_sigmask (unless they were caught,
388 	 * and are now ignored by default).
389 	 */
390 	for (signum = 1; signum < NSIG; signum++) {
391 		if (sigismember(&p->p_sigctx.ps_sigcatch, signum)) {
392 			prop = sigprop[signum];
393 			if (prop & SA_IGNORE) {
394 				if ((prop & SA_CONT) == 0)
395 					sigaddset(&p->p_sigctx.ps_sigignore,
396 					    signum);
397 				sigdelset(&p->p_sigctx.ps_siglist, signum);
398 			}
399 			SIGACTION_PS(ps, signum).sa_handler = SIG_DFL;
400 		}
401 		sigemptyset(&SIGACTION_PS(ps, signum).sa_mask);
402 		SIGACTION_PS(ps, signum).sa_flags = SA_RESTART;
403 	}
404 	sigemptyset(&p->p_sigctx.ps_sigcatch);
405 	p->p_flag &= ~P_NOCLDSTOP;
406 
407 	/*
408 	 * Reset stack state to the user stack.
409 	 */
410 	p->p_sigctx.ps_sigstk.ss_flags = SS_DISABLE;
411 	p->p_sigctx.ps_sigstk.ss_size = 0;
412 	p->p_sigctx.ps_sigstk.ss_sp = 0;
413 }
414 
415 int
416 sigprocmask1(struct proc *p, int how, const sigset_t *nss, sigset_t *oss)
417 {
418 
419 	if (oss)
420 		*oss = p->p_sigctx.ps_sigmask;
421 
422 	if (nss) {
423 		(void)splsched();	/* XXXSMP */
424 		switch (how) {
425 		case SIG_BLOCK:
426 			sigplusset(nss, &p->p_sigctx.ps_sigmask);
427 			break;
428 		case SIG_UNBLOCK:
429 			sigminusset(nss, &p->p_sigctx.ps_sigmask);
430 			CHECKSIGS(p);
431 			break;
432 		case SIG_SETMASK:
433 			p->p_sigctx.ps_sigmask = *nss;
434 			CHECKSIGS(p);
435 			break;
436 		default:
437 			(void)spl0();	/* XXXSMP */
438 			return (EINVAL);
439 		}
440 		sigminusset(&sigcantmask, &p->p_sigctx.ps_sigmask);
441 		(void)spl0();		/* XXXSMP */
442 	}
443 
444 	return (0);
445 }
446 
447 /*
448  * Manipulate signal mask.
449  * Note that we receive new mask, not pointer,
450  * and return old mask as return value;
451  * the library stub does the rest.
452  */
453 int
454 sys___sigprocmask14(struct lwp *l, void *v, register_t *retval)
455 {
456 	struct sys___sigprocmask14_args /* {
457 		syscallarg(int)			how;
458 		syscallarg(const sigset_t *)	set;
459 		syscallarg(sigset_t *)		oset;
460 	} */ *uap = v;
461 	struct proc	*p;
462 	sigset_t	nss, oss;
463 	int		error;
464 
465 	if (SCARG(uap, set)) {
466 		error = copyin(SCARG(uap, set), &nss, sizeof(nss));
467 		if (error)
468 			return (error);
469 	}
470 	p = l->l_proc;
471 	error = sigprocmask1(p, SCARG(uap, how),
472 	    SCARG(uap, set) ? &nss : 0, SCARG(uap, oset) ? &oss : 0);
473 	if (error)
474 		return (error);
475 	if (SCARG(uap, oset)) {
476 		error = copyout(&oss, SCARG(uap, oset), sizeof(oss));
477 		if (error)
478 			return (error);
479 	}
480 	return (0);
481 }
482 
483 void
484 sigpending1(struct proc *p, sigset_t *ss)
485 {
486 
487 	*ss = p->p_sigctx.ps_siglist;
488 	sigminusset(&p->p_sigctx.ps_sigmask, ss);
489 }
490 
491 /* ARGSUSED */
492 int
493 sys___sigpending14(struct lwp *l, void *v, register_t *retval)
494 {
495 	struct sys___sigpending14_args /* {
496 		syscallarg(sigset_t *)	set;
497 	} */ *uap = v;
498 	struct proc	*p;
499 	sigset_t	ss;
500 
501 	p = l->l_proc;
502 	sigpending1(p, &ss);
503 	return (copyout(&ss, SCARG(uap, set), sizeof(ss)));
504 }
505 
506 int
507 sigsuspend1(struct proc *p, const sigset_t *ss)
508 {
509 	struct sigacts *ps;
510 
511 	ps = p->p_sigacts;
512 	if (ss) {
513 		/*
514 		 * When returning from sigpause, we want
515 		 * the old mask to be restored after the
516 		 * signal handler has finished.  Thus, we
517 		 * save it here and mark the sigctx structure
518 		 * to indicate this.
519 		 */
520 		p->p_sigctx.ps_oldmask = p->p_sigctx.ps_sigmask;
521 		p->p_sigctx.ps_flags |= SAS_OLDMASK;
522 		(void) splsched();	/* XXXSMP */
523 		p->p_sigctx.ps_sigmask = *ss;
524 		CHECKSIGS(p);
525 		sigminusset(&sigcantmask, &p->p_sigctx.ps_sigmask);
526 		(void) spl0();		/* XXXSMP */
527 	}
528 
529 	while (tsleep((caddr_t) ps, PPAUSE|PCATCH, "pause", 0) == 0)
530 		/* void */;
531 	/* always return EINTR rather than ERESTART... */
532 	return (EINTR);
533 }
534 
535 /*
536  * Suspend process until signal, providing mask to be set
537  * in the meantime.  Note nonstandard calling convention:
538  * libc stub passes mask, not pointer, to save a copyin.
539  */
540 /* ARGSUSED */
541 int
542 sys___sigsuspend14(struct lwp *l, void *v, register_t *retval)
543 {
544 	struct sys___sigsuspend14_args /* {
545 		syscallarg(const sigset_t *)	set;
546 	} */ *uap = v;
547 	struct proc	*p;
548 	sigset_t	ss;
549 	int		error;
550 
551 	if (SCARG(uap, set)) {
552 		error = copyin(SCARG(uap, set), &ss, sizeof(ss));
553 		if (error)
554 			return (error);
555 	}
556 
557 	p = l->l_proc;
558 	return (sigsuspend1(p, SCARG(uap, set) ? &ss : 0));
559 }
560 
561 int
562 sigaltstack1(struct proc *p, const struct sigaltstack *nss,
563 	struct sigaltstack *oss)
564 {
565 
566 	if (oss)
567 		*oss = p->p_sigctx.ps_sigstk;
568 
569 	if (nss) {
570 		if (nss->ss_flags & ~SS_ALLBITS)
571 			return (EINVAL);
572 
573 		if (nss->ss_flags & SS_DISABLE) {
574 			if (p->p_sigctx.ps_sigstk.ss_flags & SS_ONSTACK)
575 				return (EINVAL);
576 		} else {
577 			if (nss->ss_size < MINSIGSTKSZ)
578 				return (ENOMEM);
579 		}
580 		p->p_sigctx.ps_sigstk = *nss;
581 	}
582 
583 	return (0);
584 }
585 
586 /* ARGSUSED */
587 int
588 sys___sigaltstack14(struct lwp *l, void *v, register_t *retval)
589 {
590 	struct sys___sigaltstack14_args /* {
591 		syscallarg(const struct sigaltstack *)	nss;
592 		syscallarg(struct sigaltstack *)	oss;
593 	} */ *uap = v;
594 	struct proc		*p;
595 	struct sigaltstack	nss, oss;
596 	int			error;
597 
598 	if (SCARG(uap, nss)) {
599 		error = copyin(SCARG(uap, nss), &nss, sizeof(nss));
600 		if (error)
601 			return (error);
602 	}
603 	p = l->l_proc;
604 	error = sigaltstack1(p,
605 	    SCARG(uap, nss) ? &nss : 0, SCARG(uap, oss) ? &oss : 0);
606 	if (error)
607 		return (error);
608 	if (SCARG(uap, oss)) {
609 		error = copyout(&oss, SCARG(uap, oss), sizeof(oss));
610 		if (error)
611 			return (error);
612 	}
613 	return (0);
614 }
615 
616 /* ARGSUSED */
617 int
618 sys_kill(struct lwp *l, void *v, register_t *retval)
619 {
620 	struct sys_kill_args /* {
621 		syscallarg(int)	pid;
622 		syscallarg(int)	signum;
623 	} */ *uap = v;
624 	struct proc	*cp, *p;
625 	struct pcred	*pc;
626 
627 	cp = l->l_proc;
628 	pc = cp->p_cred;
629 	if ((u_int)SCARG(uap, signum) >= NSIG)
630 		return (EINVAL);
631 	if (SCARG(uap, pid) > 0) {
632 		/* kill single process */
633 		if ((p = pfind(SCARG(uap, pid))) == NULL)
634 			return (ESRCH);
635 		if (!CANSIGNAL(cp, pc, p, SCARG(uap, signum)))
636 			return (EPERM);
637 		if (SCARG(uap, signum))
638 			psignal(p, SCARG(uap, signum));
639 		return (0);
640 	}
641 	switch (SCARG(uap, pid)) {
642 	case -1:		/* broadcast signal */
643 		return (killpg1(cp, SCARG(uap, signum), 0, 1));
644 	case 0:			/* signal own process group */
645 		return (killpg1(cp, SCARG(uap, signum), 0, 0));
646 	default:		/* negative explicit process group */
647 		return (killpg1(cp, SCARG(uap, signum), -SCARG(uap, pid), 0));
648 	}
649 	/* NOTREACHED */
650 }
651 
652 /*
653  * Common code for kill process group/broadcast kill.
654  * cp is calling process.
655  */
656 int
657 killpg1(struct proc *cp, int signum, int pgid, int all)
658 {
659 	struct proc	*p;
660 	struct pcred	*pc;
661 	struct pgrp	*pgrp;
662 	int		nfound;
663 
664 	pc = cp->p_cred;
665 	nfound = 0;
666 	if (all) {
667 		/*
668 		 * broadcast
669 		 */
670 		proclist_lock_read();
671 		LIST_FOREACH(p, &allproc, p_list) {
672 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
673 			    p == cp || !CANSIGNAL(cp, pc, p, signum))
674 				continue;
675 			nfound++;
676 			if (signum)
677 				psignal(p, signum);
678 		}
679 		proclist_unlock_read();
680 	} else {
681 		if (pgid == 0)
682 			/*
683 			 * zero pgid means send to my process group.
684 			 */
685 			pgrp = cp->p_pgrp;
686 		else {
687 			pgrp = pgfind(pgid);
688 			if (pgrp == NULL)
689 				return (ESRCH);
690 		}
691 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
692 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
693 			    !CANSIGNAL(cp, pc, p, signum))
694 				continue;
695 			nfound++;
696 			if (signum && P_ZOMBIE(p) == 0)
697 				psignal(p, signum);
698 		}
699 	}
700 	return (nfound ? 0 : ESRCH);
701 }
702 
703 /*
704  * Send a signal to a process group.
705  */
706 void
707 gsignal(int pgid, int signum)
708 {
709 	struct pgrp *pgrp;
710 
711 	if (pgid && (pgrp = pgfind(pgid)))
712 		pgsignal(pgrp, signum, 0);
713 }
714 
715 /*
716  * Send a signal to a process group. If checktty is 1,
717  * limit to members which have a controlling terminal.
718  */
719 void
720 pgsignal(struct pgrp *pgrp, int signum, int checkctty)
721 {
722 	struct proc *p;
723 
724 	if (pgrp)
725 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist)
726 			if (checkctty == 0 || p->p_flag & P_CONTROLT)
727 				psignal(p, signum);
728 }
729 
730 /*
731  * Send a signal caused by a trap to the current process.
732  * If it will be caught immediately, deliver it with correct code.
733  * Otherwise, post it normally.
734  */
735 void
736 trapsignal(struct lwp *l, int signum, u_long code)
737 {
738 	struct proc	*p;
739 	struct sigacts	*ps;
740 
741 	p = l->l_proc;
742 	ps = p->p_sigacts;
743 	if ((p->p_flag & P_TRACED) == 0 &&
744 	    sigismember(&p->p_sigctx.ps_sigcatch, signum) &&
745 	    !sigismember(&p->p_sigctx.ps_sigmask, signum)) {
746 		p->p_stats->p_ru.ru_nsignals++;
747 #ifdef KTRACE
748 		if (KTRPOINT(p, KTR_PSIG))
749 			ktrpsig(p, signum,
750 			    SIGACTION_PS(ps, signum).sa_handler,
751 			    &p->p_sigctx.ps_sigmask, code);
752 #endif
753 		psendsig(l, signum, &p->p_sigctx.ps_sigmask, code);
754 		(void) splsched();	/* XXXSMP */
755 		sigplusset(&SIGACTION_PS(ps, signum).sa_mask,
756 		    &p->p_sigctx.ps_sigmask);
757 		if (SIGACTION_PS(ps, signum).sa_flags & SA_RESETHAND) {
758 			sigdelset(&p->p_sigctx.ps_sigcatch, signum);
759 			if (signum != SIGCONT && sigprop[signum] & SA_IGNORE)
760 				sigaddset(&p->p_sigctx.ps_sigignore, signum);
761 			SIGACTION_PS(ps, signum).sa_handler = SIG_DFL;
762 		}
763 		(void) spl0();		/* XXXSMP */
764 	} else {
765 		p->p_sigctx.ps_code = code;	/* XXX for core dump/debugger */
766 		p->p_sigctx.ps_sig = signum;	/* XXX to verify code */
767 		psignal(p, signum);
768 	}
769 }
770 
771 /*
772  * Send the signal to the process.  If the signal has an action, the action
773  * is usually performed by the target process rather than the caller; we add
774  * the signal to the set of pending signals for the process.
775  *
776  * Exceptions:
777  *   o When a stop signal is sent to a sleeping process that takes the
778  *     default action, the process is stopped without awakening it.
779  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
780  *     regardless of the signal action (eg, blocked or ignored).
781  *
782  * Other ignored signals are discarded immediately.
783  *
784  * XXXSMP: Invoked as psignal() or sched_psignal().
785  */
786 void
787 psignal1(struct proc *p, int signum,
788 	int dolock)		/* XXXSMP: works, but icky */
789 {
790 	struct lwp *l, *suspended;
791 	int	s = 0, prop, allsusp;
792 	sig_t	action;
793 
794 #ifdef DIAGNOSTIC
795 	if (signum <= 0 || signum >= NSIG)
796 		panic("psignal signal number");
797 
798 	/* XXXSMP: works, but icky */
799 	if (dolock)
800 		SCHED_ASSERT_UNLOCKED();
801 	else
802 		SCHED_ASSERT_LOCKED();
803 #endif
804 	/*
805 	 * Notify any interested parties in the signal.
806 	 */
807 	KNOTE(&p->p_klist, NOTE_SIGNAL | signum);
808 
809 	prop = sigprop[signum];
810 
811 	/*
812 	 * If proc is traced, always give parent a chance.
813 	 */
814 	if (p->p_flag & P_TRACED)
815 		action = SIG_DFL;
816 	else {
817 		/*
818 		 * If the signal is being ignored,
819 		 * then we forget about it immediately.
820 		 * (Note: we don't set SIGCONT in p_sigctx.ps_sigignore,
821 		 * and if it is set to SIG_IGN,
822 		 * action will be SIG_DFL here.)
823 		 */
824 		if (sigismember(&p->p_sigctx.ps_sigignore, signum))
825 			return;
826 		if (sigismember(&p->p_sigctx.ps_sigmask, signum))
827 			action = SIG_HOLD;
828 		else if (sigismember(&p->p_sigctx.ps_sigcatch, signum))
829 			action = SIG_CATCH;
830 		else {
831 			action = SIG_DFL;
832 
833 			if (prop & SA_KILL && p->p_nice > NZERO)
834 				p->p_nice = NZERO;
835 
836 			/*
837 			 * If sending a tty stop signal to a member of an
838 			 * orphaned process group, discard the signal here if
839 			 * the action is default; don't stop the process below
840 			 * if sleeping, and don't clear any pending SIGCONT.
841 			 */
842 			if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0)
843 				return;
844 		}
845 	}
846 
847 	if (prop & SA_CONT)
848 		sigminusset(&stopsigmask, &p->p_sigctx.ps_siglist);
849 
850 	if (prop & SA_STOP)
851 		sigminusset(&contsigmask, &p->p_sigctx.ps_siglist);
852 
853 	sigaddset(&p->p_sigctx.ps_siglist, signum);
854 
855 	/* CHECKSIGS() is "inlined" here. */
856 	p->p_sigctx.ps_sigcheck = 1;
857 
858 	/*
859 	 * Defer further processing for signals which are held,
860 	 * except that stopped processes must be continued by SIGCONT.
861 	 */
862 	if (action == SIG_HOLD && ((prop & SA_CONT) == 0 || p->p_stat != SSTOP))
863 		return;
864 	/* XXXSMP: works, but icky */
865 	if (dolock)
866 		SCHED_LOCK(s);
867 
868 	if (p->p_nrlwps > 0) {
869 		/*
870 		 * At least one LWP is running or on a run queue.
871 		 * The signal will be noticed when one of them returns
872 		 * to userspace.
873 		 */
874 		signotify(p);
875 		/*
876 		 * The signal will be noticed very soon.
877 		 */
878 		goto out;
879 	} else {
880 		/* Process is sleeping or stopped */
881 		if (p->p_flag & P_SA) {
882 			l = p->p_sa->sa_idle;
883 		} else {
884 			/*
885 			 * Find out if any of the sleeps are interruptable,
886 			 * and if all the live LWPs remaining are suspended.
887 			 */
888 			allsusp = 1;
889 			LIST_FOREACH(l, &p->p_lwps, l_sibling) {
890 				if (l->l_stat == LSSLEEP &&
891 				    l->l_flag & L_SINTR)
892 					break;
893 				if (l->l_stat == LSSUSPENDED)
894 					suspended = l;
895 				else if ((l->l_stat != LSZOMB) &&
896 				         (l->l_stat != LSDEAD))
897 					allsusp = 0;
898 			}
899 		}
900 		if (p->p_stat == SACTIVE) {
901 			/* All LWPs must be sleeping */
902 			KDASSERT(((p->p_flag & P_SA) == 0) || (l != NULL));
903 
904 			if (l != NULL && (p->p_flag & P_TRACED))
905 				goto run;
906 
907 			/*
908 			 * If SIGCONT is default (or ignored) and process is
909 			 * asleep, we are finished; the process should not
910 			 * be awakened.
911 			 */
912 			if ((prop & SA_CONT) && action == SIG_DFL) {
913 				sigdelset(&p->p_sigctx.ps_siglist, signum);
914 				goto out;
915 			}
916 
917 			/*
918 			 * When a sleeping process receives a stop
919 			 * signal, process immediately if possible.
920 			 */
921 			if ((prop & SA_STOP) && action == SIG_DFL) {
922 				/*
923 				 * If a child holding parent blocked,
924 				 * stopping could cause deadlock.
925 				 */
926 				if (p->p_flag & P_PPWAIT)
927 					goto out;
928 				sigdelset(&p->p_sigctx.ps_siglist, signum);
929 				p->p_xstat = signum;
930 				if ((p->p_pptr->p_flag & P_NOCLDSTOP) == 0) {
931 					/*
932 					 * XXXSMP: recursive call; don't lock
933 					 * the second time around.
934 					 */
935 					sched_psignal(p->p_pptr, SIGCHLD);
936 				}
937 				proc_stop(p);	/* XXXSMP: recurse? */
938 				goto out;
939 			}
940 
941 			if (l == NULL) {
942 				/*
943 				 * Special case: SIGKILL of a process
944 				 * which is entirely composed of
945 				 * suspended LWPs should succeed. We
946 				 * make this happen by unsuspending one of
947 				 * them.
948 				 */
949 				if (allsusp && (signum == SIGKILL))
950 					lwp_continue(suspended);
951 				goto out;
952 			}
953 			/*
954 			 * All other (caught or default) signals
955 			 * cause the process to run.
956 			 */
957 			goto runfast;
958 			/*NOTREACHED*/
959 		} else if (p->p_stat == SSTOP) {
960 			/* Process is stopped */
961 			/*
962 			 * If traced process is already stopped,
963 			 * then no further action is necessary.
964 			 */
965 			if (p->p_flag & P_TRACED)
966 				goto out;
967 
968 			/*
969 			 * Kill signal always sets processes running,
970 			 * if possible.
971 			 */
972 			if (signum == SIGKILL) {
973 				l = proc_unstop(p);
974 				if (l)
975 					goto runfast;
976 				/* XXX should this be possible? */
977 				goto out;
978 			}
979 
980 			if (prop & SA_CONT) {
981 				/*
982 				 * If SIGCONT is default (or ignored),
983 				 * we continue the process but don't
984 				 * leave the signal in ps_siglist, as
985 				 * it has no further action.  If
986 				 * SIGCONT is held, we continue the
987 				 * process and leave the signal in
988 				 * ps_siglist.  If the process catches
989 				 * SIGCONT, let it handle the signal
990 				 * itself.  If it isn't waiting on an
991 				 * event, then it goes back to run
992 				 * state.  Otherwise, process goes
993 				 * back to sleep state.
994 				 */
995 				if (action == SIG_DFL)
996 					sigdelset(&p->p_sigctx.ps_siglist,
997 					signum);
998 				l = proc_unstop(p);
999 				/*
1000 				 * XXX see note in proc_unstop(). SIGKILL
1001 				 * XXX and SIGCONT have conflicting needs.
1002 				 */
1003 				if (l && (l->l_stat == LSSLEEP))
1004 					l = NULL;
1005 				if (l && (action == SIG_CATCH))
1006 					goto runfast;
1007 				if (l)
1008 					goto run;
1009 				goto out;
1010 			}
1011 
1012 			if (prop & SA_STOP) {
1013 				/*
1014 				 * Already stopped, don't need to stop again.
1015 				 * (If we did the shell could get confused.)
1016 				 */
1017 				sigdelset(&p->p_sigctx.ps_siglist, signum);
1018 				goto out;
1019 			}
1020 
1021 			/*
1022 			 * If process is sleeping interruptibly, then
1023 			 * simulate a wakeup so that when it is
1024 			 * continued, it will be made runnable and can
1025 			 * look at the signal.  But don't make the
1026 			 * process runnable, leave it stopped.
1027 			 */
1028 			if (l)
1029 				unsleep(l);
1030 			goto out;
1031 		} else {
1032 			/* Else what? */
1033 			panic("psignal: Invalid process state %d.",
1034 				p->p_stat);
1035 		}
1036 	}
1037 	/*NOTREACHED*/
1038 
1039  runfast:
1040 	/*
1041 	 * Raise priority to at least PUSER.
1042 	 */
1043 	if (l->l_priority > PUSER)
1044 		l->l_priority = PUSER;
1045  run:
1046 	setrunnable(l);		/* XXXSMP: recurse? */
1047  out:
1048 	/* XXXSMP: works, but icky */
1049 	if (dolock)
1050 		SCHED_UNLOCK(s);
1051 }
1052 
1053 void
1054 psendsig(struct lwp *l, int sig, sigset_t *mask, u_long code)
1055 {
1056 	struct proc *p = l->l_proc;
1057 	struct lwp *le, *li;
1058 	siginfo_t *si;
1059 
1060 	if (p->p_flag & P_SA) {
1061 		si = pool_get(&siginfo_pool, PR_WAITOK);
1062 		si->si_signo = sig;
1063 		si->si_errno = 0;
1064 		si->si_code = code;
1065 		le = li = NULL;
1066 		if (code)
1067 			le = l;
1068 		else
1069 			li = l;
1070 
1071 		sa_upcall(l, SA_UPCALL_SIGNAL | SA_UPCALL_DEFER, le, li,
1072 			    sizeof(siginfo_t), si);
1073 		return;
1074 	}
1075 
1076 	(*p->p_emul->e_sendsig)(sig, mask, code);
1077 }
1078 
1079 static __inline int firstsig(const sigset_t *);
1080 
1081 static __inline int
1082 firstsig(const sigset_t *ss)
1083 {
1084 	int sig;
1085 
1086 	sig = ffs(ss->__bits[0]);
1087 	if (sig != 0)
1088 		return (sig);
1089 #if NSIG > 33
1090 	sig = ffs(ss->__bits[1]);
1091 	if (sig != 0)
1092 		return (sig + 32);
1093 #endif
1094 #if NSIG > 65
1095 	sig = ffs(ss->__bits[2]);
1096 	if (sig != 0)
1097 		return (sig + 64);
1098 #endif
1099 #if NSIG > 97
1100 	sig = ffs(ss->__bits[3]);
1101 	if (sig != 0)
1102 		return (sig + 96);
1103 #endif
1104 	return (0);
1105 }
1106 
1107 /*
1108  * If the current process has received a signal (should be caught or cause
1109  * termination, should interrupt current syscall), return the signal number.
1110  * Stop signals with default action are processed immediately, then cleared;
1111  * they aren't returned.  This is checked after each entry to the system for
1112  * a syscall or trap (though this can usually be done without calling issignal
1113  * by checking the pending signal masks in the CURSIG macro.) The normal call
1114  * sequence is
1115  *
1116  *	while (signum = CURSIG(curlwp))
1117  *		postsig(signum);
1118  */
1119 int
1120 issignal(struct lwp *l)
1121 {
1122 	struct proc	*p = l->l_proc;
1123 	int		s = 0, signum, prop;
1124 	int		dolock = (l->l_flag & L_SINTR) == 0, locked = !dolock;
1125 	sigset_t	ss;
1126 
1127 	if (p->p_stat == SSTOP) {
1128 		/*
1129 		 * The process is stopped/stopping. Stop ourselves now that
1130 		 * we're on the kernel/userspace boundary.
1131 		 */
1132 		if (dolock)
1133 			SCHED_LOCK(s);
1134 		l->l_stat = LSSTOP;
1135 		p->p_nrlwps--;
1136 		if (p->p_flag & P_TRACED)
1137 			goto sigtraceswitch;
1138 		else
1139 			goto sigswitch;
1140 	}
1141 	for (;;) {
1142 		sigpending1(p, &ss);
1143 		if (p->p_flag & P_PPWAIT)
1144 			sigminusset(&stopsigmask, &ss);
1145 		signum = firstsig(&ss);
1146 		if (signum == 0) {		 	/* no signal to send */
1147 			p->p_sigctx.ps_sigcheck = 0;
1148 			if (locked && dolock)
1149 				SCHED_LOCK(s);
1150 			return (0);
1151 		}
1152 							/* take the signal! */
1153 		sigdelset(&p->p_sigctx.ps_siglist, signum);
1154 
1155 		/*
1156 		 * We should see pending but ignored signals
1157 		 * only if P_TRACED was on when they were posted.
1158 		 */
1159 		if (sigismember(&p->p_sigctx.ps_sigignore, signum) &&
1160 		    (p->p_flag & P_TRACED) == 0)
1161 			continue;
1162 
1163 		if (p->p_flag & P_TRACED && (p->p_flag & P_PPWAIT) == 0) {
1164 			/*
1165 			 * If traced, always stop, and stay
1166 			 * stopped until released by the debugger.
1167 			 */
1168 			p->p_xstat = signum;
1169 			if ((p->p_flag & P_FSTRACE) == 0)
1170 				psignal1(p->p_pptr, SIGCHLD, dolock);
1171 			if (dolock)
1172 				SCHED_LOCK(s);
1173 			proc_stop(p);
1174 		sigtraceswitch:
1175 			mi_switch(l, NULL);
1176 			SCHED_ASSERT_UNLOCKED();
1177 			if (dolock)
1178 				splx(s);
1179 			else
1180 				dolock = 1;
1181 
1182 			/*
1183 			 * If we are no longer being traced, or the parent
1184 			 * didn't give us a signal, look for more signals.
1185 			 */
1186 			if ((p->p_flag & P_TRACED) == 0 || p->p_xstat == 0)
1187 				continue;
1188 
1189 			/*
1190 			 * If the new signal is being masked, look for other
1191 			 * signals.
1192 			 */
1193 			signum = p->p_xstat;
1194 			p->p_xstat = 0;
1195 			/*
1196 			 * `p->p_sigctx.ps_siglist |= mask' is done
1197 			 * in setrunnable().
1198 			 */
1199 			if (sigismember(&p->p_sigctx.ps_sigmask, signum))
1200 				continue;
1201 							/* take the signal! */
1202 			sigdelset(&p->p_sigctx.ps_siglist, signum);
1203 		}
1204 
1205 		prop = sigprop[signum];
1206 
1207 		/*
1208 		 * Decide whether the signal should be returned.
1209 		 * Return the signal's number, or fall through
1210 		 * to clear it from the pending mask.
1211 		 */
1212 		switch ((long)SIGACTION(p, signum).sa_handler) {
1213 
1214 		case (long)SIG_DFL:
1215 			/*
1216 			 * Don't take default actions on system processes.
1217 			 */
1218 			if (p->p_pid <= 1) {
1219 #ifdef DIAGNOSTIC
1220 				/*
1221 				 * Are you sure you want to ignore SIGSEGV
1222 				 * in init? XXX
1223 				 */
1224 				printf("Process (pid %d) got signal %d\n",
1225 				    p->p_pid, signum);
1226 #endif
1227 				break;		/* == ignore */
1228 			}
1229 			/*
1230 			 * If there is a pending stop signal to process
1231 			 * with default action, stop here,
1232 			 * then clear the signal.  However,
1233 			 * if process is member of an orphaned
1234 			 * process group, ignore tty stop signals.
1235 			 */
1236 			if (prop & SA_STOP) {
1237 				if (p->p_flag & P_TRACED ||
1238 		    		    (p->p_pgrp->pg_jobc == 0 &&
1239 				    prop & SA_TTYSTOP))
1240 					break;	/* == ignore */
1241 				p->p_xstat = signum;
1242 				if ((p->p_pptr->p_flag & P_NOCLDSTOP) == 0)
1243 					psignal1(p->p_pptr, SIGCHLD, dolock);
1244 				if (dolock)
1245 					SCHED_LOCK(s);
1246 				proc_stop(p);
1247 			sigswitch:
1248 				mi_switch(l, NULL);
1249 				SCHED_ASSERT_UNLOCKED();
1250 				if (dolock)
1251 					splx(s);
1252 				else
1253 					dolock = 1;
1254 				break;
1255 			} else if (prop & SA_IGNORE) {
1256 				/*
1257 				 * Except for SIGCONT, shouldn't get here.
1258 				 * Default action is to ignore; drop it.
1259 				 */
1260 				break;		/* == ignore */
1261 			} else
1262 				goto keep;
1263 			/*NOTREACHED*/
1264 
1265 		case (long)SIG_IGN:
1266 			/*
1267 			 * Masking above should prevent us ever trying
1268 			 * to take action on an ignored signal other
1269 			 * than SIGCONT, unless process is traced.
1270 			 */
1271 #ifdef DEBUG_ISSIGNAL
1272 			if ((prop & SA_CONT) == 0 &&
1273 			    (p->p_flag & P_TRACED) == 0)
1274 				printf("issignal\n");
1275 #endif
1276 			break;		/* == ignore */
1277 
1278 		default:
1279 			/*
1280 			 * This signal has an action, let
1281 			 * postsig() process it.
1282 			 */
1283 			goto keep;
1284 		}
1285 	}
1286 	/* NOTREACHED */
1287 
1288  keep:
1289 						/* leave the signal for later */
1290 	sigaddset(&p->p_sigctx.ps_siglist, signum);
1291 	CHECKSIGS(p);
1292 	if (locked && dolock)
1293 		SCHED_LOCK(s);
1294 	return (signum);
1295 }
1296 
1297 /*
1298  * Put the argument process into the stopped state and notify the parent
1299  * via wakeup.  Signals are handled elsewhere.  The process must not be
1300  * on the run queue.
1301  */
1302 static void
1303 proc_stop(struct proc *p)
1304 {
1305 	struct lwp *l;
1306 
1307 	SCHED_ASSERT_LOCKED();
1308 
1309 	/* XXX lock process LWP state */
1310 	p->p_stat = SSTOP;
1311 	p->p_flag &= ~P_WAITED;
1312 
1313 	/*
1314 	 * Put as many LWP's as possible in stopped state.
1315 	 * Sleeping ones will notice the stopped state as they try to
1316 	 * return to userspace.
1317 	 */
1318 
1319 	for (l = LIST_FIRST(&p->p_lwps); l != NULL;
1320 	     l = LIST_NEXT(l, l_sibling)) {
1321 		if (l->l_stat == LSONPROC) {
1322 			/* XXX SMP this assumes that a LWP that is LSONPROC
1323 			 * is curlwp and hence is about to be mi_switched
1324 			 * away; the only callers of proc_stop() are:
1325 			 * - psignal
1326 			 * - issignal()
1327 			 * For the former, proc_stop() is only called when
1328 			 * no processes are running, so we don't worry.
1329 			 * For the latter, proc_stop() is called right
1330 			 * before mi_switch().
1331 			 */
1332 			l->l_stat = LSSTOP;
1333 			p->p_nrlwps--;
1334 		} else if (l->l_stat == LSRUN) {
1335 			/* Remove LWP from the run queue */
1336 			remrunqueue(l);
1337 			l->l_stat = LSSTOP;
1338 			p->p_nrlwps--;
1339 		} else if ((l->l_stat == LSSLEEP) ||
1340 		    (l->l_stat == LSSUSPENDED) ||
1341 		    (l->l_stat == LSZOMB) ||
1342 		    (l->l_stat == LSDEAD)) {
1343 			/*
1344 			 * Don't do anything; let sleeping LWPs
1345 			 * discover the stopped state of the process
1346 			 * on their way out of the kernel; otherwise,
1347 			 * things like NFS threads that sleep with
1348 			 * locks will block the rest of the system
1349 			 * from getting any work done.
1350 			 *
1351 			 * Suspended/dead/zombie LWPs aren't going
1352 			 * anywhere, so we don't need to touch them.
1353 			 */
1354 		}
1355 #ifdef DIAGNOSTIC
1356 		else {
1357 			panic("proc_stop: process %d lwp %d "
1358 			      "in unstoppable state %d.\n",
1359 			    p->p_pid, l->l_lid, l->l_stat);
1360 		}
1361 #endif
1362 	}
1363 	/* XXX unlock process LWP state */
1364 
1365 	sched_wakeup((caddr_t)p->p_pptr);
1366 }
1367 
1368 struct lwp *
1369 proc_unstop(p)
1370 	struct proc *p;
1371 {
1372 	struct lwp *l, *lr = NULL;
1373 
1374 	SCHED_ASSERT_LOCKED();
1375 
1376 	/*
1377 	 * Our caller will want to invoke setrunnable() on whatever we return,
1378 	 * provided that it isn't NULL.
1379 	 */
1380 
1381 	p->p_stat = SACTIVE;
1382 	/*
1383 	 * For the benefit of SIGKILL, return the idle LWP if there's
1384 	 * nothing better (and if there is an idle LWP, there
1385 	 * shouldn't be anything better.
1386 	 * XXX This is bad for SIGCONT; SIGSTOP/SIGCONT shouldn't
1387 	 * XXX noticably affect the state of a process, idling or
1388 	 * XXX not. We work around this in the SIGCONT handling in
1389 	 * XXX psignal().
1390 	 */
1391 	if (p->p_flag & P_SA)
1392 		lr = p->p_sa->sa_idle; /* OK if this is NULL. */
1393 	for (l = LIST_FIRST(&p->p_lwps); l != NULL;
1394 	     l = LIST_NEXT(l, l_sibling))
1395 		if (l->l_stat == LSSTOP) {
1396 			if (l->l_wchan == 0) {
1397 				if (lr == NULL || l == lr)
1398 					lr = l;
1399 				else
1400 					setrunnable(l);
1401 			} else
1402 				l->l_stat = LSSLEEP;
1403 		}
1404 
1405 	return lr;
1406 }
1407 
1408 /*
1409  * Take the action for the specified signal
1410  * from the current set of pending signals.
1411  */
1412 void
1413 postsig(int signum)
1414 {
1415 	struct lwp *l;
1416 	struct proc	*p;
1417 	struct sigacts	*ps;
1418 	sig_t		action;
1419 	u_long		code;
1420 	sigset_t	*returnmask;
1421 
1422 	l = curlwp;
1423 	p = l->l_proc;
1424 	ps = p->p_sigacts;
1425 #ifdef DIAGNOSTIC
1426 	if (signum == 0)
1427 		panic("postsig");
1428 #endif
1429 
1430 	KERNEL_PROC_LOCK(l);
1431 
1432 	sigdelset(&p->p_sigctx.ps_siglist, signum);
1433 	action = SIGACTION_PS(ps, signum).sa_handler;
1434 #ifdef KTRACE
1435 	if (KTRPOINT(p, KTR_PSIG))
1436 		ktrpsig(p,
1437 		    signum, action, p->p_sigctx.ps_flags & SAS_OLDMASK ?
1438 		    &p->p_sigctx.ps_oldmask : &p->p_sigctx.ps_sigmask, 0);
1439 #endif
1440 	if (action == SIG_DFL) {
1441 		/*
1442 		 * Default action, where the default is to kill
1443 		 * the process.  (Other cases were ignored above.)
1444 		 */
1445 		sigexit(l, signum);
1446 		/* NOTREACHED */
1447 	} else {
1448 		/*
1449 		 * If we get here, the signal must be caught.
1450 		 */
1451 #ifdef DIAGNOSTIC
1452 		if (action == SIG_IGN ||
1453 		    sigismember(&p->p_sigctx.ps_sigmask, signum))
1454 			panic("postsig action");
1455 #endif
1456 		/*
1457 		 * Set the new mask value and also defer further
1458 		 * occurences of this signal.
1459 		 *
1460 		 * Special case: user has done a sigpause.  Here the
1461 		 * current mask is not of interest, but rather the
1462 		 * mask from before the sigpause is what we want
1463 		 * restored after the signal processing is completed.
1464 		 */
1465 		if (p->p_sigctx.ps_flags & SAS_OLDMASK) {
1466 			returnmask = &p->p_sigctx.ps_oldmask;
1467 			p->p_sigctx.ps_flags &= ~SAS_OLDMASK;
1468 		} else
1469 			returnmask = &p->p_sigctx.ps_sigmask;
1470 		p->p_stats->p_ru.ru_nsignals++;
1471 		if (p->p_sigctx.ps_sig != signum) {
1472 			code = 0;
1473 		} else {
1474 			code = p->p_sigctx.ps_code;
1475 			p->p_sigctx.ps_code = 0;
1476 			p->p_sigctx.ps_sig = 0;
1477 		}
1478 		psendsig(l, signum, returnmask, code);
1479 		(void) splsched();	/* XXXSMP */
1480 		sigplusset(&SIGACTION_PS(ps, signum).sa_mask,
1481 		    &p->p_sigctx.ps_sigmask);
1482 		if (SIGACTION_PS(ps, signum).sa_flags & SA_RESETHAND) {
1483 			sigdelset(&p->p_sigctx.ps_sigcatch, signum);
1484 			if (signum != SIGCONT && sigprop[signum] & SA_IGNORE)
1485 				sigaddset(&p->p_sigctx.ps_sigignore, signum);
1486 			SIGACTION_PS(ps, signum).sa_handler = SIG_DFL;
1487 		}
1488 		(void) spl0();		/* XXXSMP */
1489 	}
1490 
1491 	KERNEL_PROC_UNLOCK(l);
1492 }
1493 
1494 /*
1495  * Kill the current process for stated reason.
1496  */
1497 void
1498 killproc(struct proc *p, const char *why)
1499 {
1500 
1501 	log(LOG_ERR, "pid %d was killed: %s\n", p->p_pid, why);
1502 	uprintf("sorry, pid %d was killed: %s\n", p->p_pid, why);
1503 	psignal(p, SIGKILL);
1504 }
1505 
1506 /*
1507  * Force the current process to exit with the specified signal, dumping core
1508  * if appropriate.  We bypass the normal tests for masked and caught signals,
1509  * allowing unrecoverable failures to terminate the process without changing
1510  * signal state.  Mark the accounting record with the signal termination.
1511  * If dumping core, save the signal number for the debugger.  Calls exit and
1512  * does not return.
1513  */
1514 
1515 #if defined(DEBUG)
1516 int	kern_logsigexit = 1;	/* not static to make public for sysctl */
1517 #else
1518 int	kern_logsigexit = 0;	/* not static to make public for sysctl */
1519 #endif
1520 
1521 static	const char logcoredump[] =
1522 	"pid %d (%s), uid %d: exited on signal %d (core dumped)\n";
1523 static	const char lognocoredump[] =
1524 	"pid %d (%s), uid %d: exited on signal %d (core not dumped, err = %d)\n";
1525 
1526 /* Wrapper function for use in p_userret */
1527 static void
1528 lwp_coredump_hook(struct lwp *l, void *arg)
1529 {
1530 	int s;
1531 
1532 	/*
1533 	 * Suspend ourselves, so that the kernel stack and therefore
1534 	 * the userland registers saved in the trapframe are around
1535 	 * for coredump() to write them out.
1536 	 */
1537 	KERNEL_PROC_LOCK(l);
1538 	l->l_flag &= ~L_DETACHED;
1539 	SCHED_LOCK(s);
1540 	l->l_stat = LSSUSPENDED;
1541 	l->l_proc->p_nrlwps--;
1542 	/* XXX NJWLWP check if this makes sense here: */
1543 	l->l_proc->p_stats->p_ru.ru_nvcsw++;
1544 	mi_switch(l, NULL);
1545 	SCHED_ASSERT_UNLOCKED();
1546 	splx(s);
1547 
1548 	lwp_exit(l);
1549 }
1550 
1551 void
1552 sigexit(struct lwp *l, int signum)
1553 {
1554 	struct proc	*p;
1555 	int		error, exitsig;
1556 
1557 	p = l->l_proc;
1558 
1559 	/*
1560 	 * Don't permit coredump() or exit1() multiple times
1561 	 * in the same process.
1562 	 */
1563 	if (p->p_flag & P_WEXIT)
1564 		(*p->p_userret)(l, p->p_userret_arg);
1565 	p->p_flag |= P_WEXIT;
1566 	/* We don't want to switch away from exiting. */
1567 	/* XXX multiprocessor: stop LWPs on other processors. */
1568 	if (l->l_flag & L_SA) {
1569 		l->l_flag &= ~L_SA;
1570 		p->p_flag &= ~P_SA;
1571 	}
1572 
1573 	/* Make other LWPs stick around long enough to be dumped */
1574 	p->p_userret = lwp_coredump_hook;
1575 	p->p_userret_arg = NULL;
1576 
1577 	exitsig = signum;
1578 	p->p_acflag |= AXSIG;
1579 	if (sigprop[signum] & SA_CORE) {
1580 		p->p_sigctx.ps_sig = signum;
1581 		if ((error = coredump(l)) == 0)
1582 			exitsig |= WCOREFLAG;
1583 
1584 		if (kern_logsigexit) {
1585 			/* XXX What if we ever have really large UIDs? */
1586 			int uid = p->p_cred && p->p_ucred ?
1587 				(int) p->p_ucred->cr_uid : -1;
1588 
1589 			if (error)
1590 				log(LOG_INFO, lognocoredump, p->p_pid,
1591 				    p->p_comm, uid, signum, error);
1592 			else
1593 				log(LOG_INFO, logcoredump, p->p_pid,
1594 				    p->p_comm, uid, signum);
1595 		}
1596 
1597 	}
1598 
1599 	exit1(l, W_EXITCODE(0, exitsig));
1600 	/* NOTREACHED */
1601 }
1602 
1603 /*
1604  * Dump core, into a file named "progname.core" or "core" (depending on the
1605  * value of shortcorename), unless the process was setuid/setgid.
1606  */
1607 int
1608 coredump(struct lwp *l)
1609 {
1610 	struct vnode		*vp;
1611 	struct proc		*p;
1612 	struct vmspace		*vm;
1613 	struct ucred		*cred;
1614 	struct nameidata	nd;
1615 	struct vattr		vattr;
1616 	int			error, error1;
1617 	char			name[MAXPATHLEN];
1618 
1619 	p = l->l_proc;
1620 	vm = p->p_vmspace;
1621 	cred = p->p_cred->pc_ucred;
1622 
1623 	/*
1624 	 * Make sure the process has not set-id, to prevent data leaks.
1625 	 */
1626 	if (p->p_flag & P_SUGID)
1627 		return (EPERM);
1628 
1629 	/*
1630 	 * Refuse to core if the data + stack + user size is larger than
1631 	 * the core dump limit.  XXX THIS IS WRONG, because of mapped
1632 	 * data.
1633 	 */
1634 	if (USPACE + ctob(vm->vm_dsize + vm->vm_ssize) >=
1635 	    p->p_rlimit[RLIMIT_CORE].rlim_cur)
1636 		return (EFBIG);		/* better error code? */
1637 
1638 	/*
1639 	 * The core dump will go in the current working directory.  Make
1640 	 * sure that the directory is still there and that the mount flags
1641 	 * allow us to write core dumps there.
1642 	 */
1643 	vp = p->p_cwdi->cwdi_cdir;
1644 	if (vp->v_mount == NULL ||
1645 	    (vp->v_mount->mnt_flag & MNT_NOCOREDUMP) != 0)
1646 		return (EPERM);
1647 
1648 	error = build_corename(p, name);
1649 	if (error)
1650 		return error;
1651 
1652 	NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, p);
1653 	error = vn_open(&nd, O_CREAT | O_NOFOLLOW | FWRITE, S_IRUSR | S_IWUSR);
1654 	if (error)
1655 		return (error);
1656 	vp = nd.ni_vp;
1657 
1658 	/* Don't dump to non-regular files or files with links. */
1659 	if (vp->v_type != VREG ||
1660 	    VOP_GETATTR(vp, &vattr, cred, p) || vattr.va_nlink != 1) {
1661 		error = EINVAL;
1662 		goto out;
1663 	}
1664 	VATTR_NULL(&vattr);
1665 	vattr.va_size = 0;
1666 	VOP_LEASE(vp, p, cred, LEASE_WRITE);
1667 	VOP_SETATTR(vp, &vattr, cred, p);
1668 	p->p_acflag |= ACORE;
1669 
1670 	/* Now dump the actual core file. */
1671 	error = (*p->p_execsw->es_coredump)(l, vp, cred);
1672  out:
1673 	VOP_UNLOCK(vp, 0);
1674 	error1 = vn_close(vp, FWRITE, cred, p);
1675 	if (error == 0)
1676 		error = error1;
1677 	return (error);
1678 }
1679 
1680 /*
1681  * Nonexistent system call-- signal process (may want to handle it).
1682  * Flag error in case process won't see signal immediately (blocked or ignored).
1683  */
1684 /* ARGSUSED */
1685 int
1686 sys_nosys(struct lwp *l, void *v, register_t *retval)
1687 {
1688 	struct proc 	*p;
1689 
1690 	p = l->l_proc;
1691 	psignal(p, SIGSYS);
1692 	return (ENOSYS);
1693 }
1694 
1695 static int
1696 build_corename(struct proc *p, char dst[MAXPATHLEN])
1697 {
1698 	const char	*s;
1699 	char		*d, *end;
1700 	int		i;
1701 
1702 	for (s = p->p_limit->pl_corename, d = dst, end = d + MAXPATHLEN;
1703 	    *s != '\0'; s++) {
1704 		if (*s == '%') {
1705 			switch (*(s + 1)) {
1706 			case 'n':
1707 				i = snprintf(d, end - d, "%s", p->p_comm);
1708 				break;
1709 			case 'p':
1710 				i = snprintf(d, end - d, "%d", p->p_pid);
1711 				break;
1712 			case 'u':
1713 				i = snprintf(d, end - d, "%s",
1714 				    p->p_pgrp->pg_session->s_login);
1715 				break;
1716 			case 't':
1717 				i = snprintf(d, end - d, "%ld",
1718 				    p->p_stats->p_start.tv_sec);
1719 				break;
1720 			default:
1721 				goto copy;
1722 			}
1723 			d += i;
1724 			s++;
1725 		} else {
1726  copy:			*d = *s;
1727 			d++;
1728 		}
1729 		if (d >= end)
1730 			return (ENAMETOOLONG);
1731 	}
1732 	*d = '\0';
1733 	return 0;
1734 }
1735 
1736 void
1737 getucontext(struct lwp *l, ucontext_t *ucp)
1738 {
1739 	struct proc	*p;
1740 
1741 	p = l->l_proc;
1742 
1743 	ucp->uc_flags = 0;
1744 	ucp->uc_link = l->l_ctxlink;
1745 
1746 	(void)sigprocmask1(p, 0, NULL, &ucp->uc_sigmask);
1747 	ucp->uc_flags |= _UC_SIGMASK;
1748 
1749 	/*
1750 	 * The (unsupplied) definition of the `current execution stack'
1751 	 * in the System V Interface Definition appears to allow returning
1752 	 * the main context stack.
1753 	 */
1754 	if ((p->p_sigctx.ps_sigstk.ss_flags & SS_ONSTACK) == 0) {
1755 		ucp->uc_stack.ss_sp = (void *)USRSTACK;
1756 		ucp->uc_stack.ss_size = ctob(p->p_vmspace->vm_ssize);
1757 		ucp->uc_stack.ss_flags = 0;	/* XXX, def. is Very Fishy */
1758 	} else {
1759 		/* Simply copy alternate signal execution stack. */
1760 		ucp->uc_stack = p->p_sigctx.ps_sigstk;
1761 	}
1762 	ucp->uc_flags |= _UC_STACK;
1763 
1764 	cpu_getmcontext(l, &ucp->uc_mcontext, &ucp->uc_flags);
1765 }
1766 
1767 /* ARGSUSED */
1768 int
1769 sys_getcontext(struct lwp *l, void *v, register_t *retval)
1770 {
1771 	struct sys_getcontext_args /* {
1772 		syscallarg(struct __ucontext *) ucp;
1773 	} */ *uap = v;
1774 	ucontext_t uc;
1775 
1776 	getucontext(l, &uc);
1777 
1778 	return (copyout(&uc, SCARG(uap, ucp), sizeof (*SCARG(uap, ucp))));
1779 }
1780 
1781 int
1782 setucontext(struct lwp *l, const ucontext_t *ucp)
1783 {
1784 	struct proc	*p;
1785 	int		error;
1786 
1787 	p = l->l_proc;
1788 	if ((error = cpu_setmcontext(l, &ucp->uc_mcontext, ucp->uc_flags)) != 0)
1789 		return (error);
1790 	l->l_ctxlink = ucp->uc_link;
1791 	/*
1792 	 * We might want to take care of the stack portion here but currently
1793 	 * don't; see the comment in getucontext().
1794 	 */
1795 	if ((ucp->uc_flags & _UC_SIGMASK) != 0)
1796 		sigprocmask1(p, SIG_SETMASK, &ucp->uc_sigmask, NULL);
1797 
1798 	return 0;
1799 }
1800 
1801 /* ARGSUSED */
1802 int
1803 sys_setcontext(struct lwp *l, void *v, register_t *retval)
1804 {
1805 	struct sys_setcontext_args /* {
1806 		syscallarg(const ucontext_t *) ucp;
1807 	} */ *uap = v;
1808 	ucontext_t uc;
1809 	int error;
1810 
1811 	if (SCARG(uap, ucp) == NULL)	/* i.e. end of uc_link chain */
1812 		exit1(l, W_EXITCODE(0, 0));
1813 	else if ((error = copyin(SCARG(uap, ucp), &uc, sizeof (uc))) != 0 ||
1814 	    (error = setucontext(l, &uc)) != 0)
1815 		return (error);
1816 
1817 	return (EJUSTRETURN);
1818 }
1819 
1820 
1821 /*
1822  * Returns true if signal is ignored or masked for passed process.
1823  */
1824 int
1825 sigismasked(struct proc *p, int sig)
1826 {
1827 
1828 	return (sigismember(&p->p_sigctx.ps_sigignore, sig) ||
1829 	    sigismember(&p->p_sigctx.ps_sigmask, sig));
1830 }
1831 
1832 static int
1833 filt_sigattach(struct knote *kn)
1834 {
1835 	struct proc *p = curproc;
1836 
1837 	kn->kn_ptr.p_proc = p;
1838 	kn->kn_flags |= EV_CLEAR;               /* automatically set */
1839 
1840 	SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext);
1841 
1842 	return (0);
1843 }
1844 
1845 static void
1846 filt_sigdetach(struct knote *kn)
1847 {
1848 	struct proc *p = kn->kn_ptr.p_proc;
1849 
1850 	SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
1851 }
1852 
1853 /*
1854  * signal knotes are shared with proc knotes, so we apply a mask to
1855  * the hint in order to differentiate them from process hints.  This
1856  * could be avoided by using a signal-specific knote list, but probably
1857  * isn't worth the trouble.
1858  */
1859 static int
1860 filt_signal(struct knote *kn, long hint)
1861 {
1862 
1863 	if (hint & NOTE_SIGNAL) {
1864 		hint &= ~NOTE_SIGNAL;
1865 
1866 		if (kn->kn_id == hint)
1867 			kn->kn_data++;
1868 	}
1869 	return (kn->kn_data != 0);
1870 }
1871 
1872 const struct filterops sig_filtops = {
1873 	0, filt_sigattach, filt_sigdetach, filt_signal
1874 };
1875