xref: /dflybsd-src/sys/kern/kern_sig.c (revision bc76a771df54af7e361532b257cecc26227736b4)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. 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  * $FreeBSD: src/sys/kern/kern_sig.c,v 1.72.2.17 2003/05/16 16:34:34 obrien Exp $
40  * $DragonFly: src/sys/kern/kern_sig.c,v 1.29 2004/04/10 20:55:23 dillon Exp $
41  */
42 
43 #include "opt_ktrace.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
48 #include <sys/sysproto.h>
49 #include <sys/signalvar.h>
50 #include <sys/resourcevar.h>
51 #include <sys/vnode.h>
52 #include <sys/event.h>
53 #include <sys/proc.h>
54 #include <sys/namei.h>
55 #include <sys/pioctl.h>
56 #include <sys/systm.h>
57 #include <sys/acct.h>
58 #include <sys/fcntl.h>
59 #include <sys/wait.h>
60 #include <sys/ktrace.h>
61 #include <sys/syslog.h>
62 #include <sys/stat.h>
63 #include <sys/sysent.h>
64 #include <sys/sysctl.h>
65 #include <sys/malloc.h>
66 #include <sys/unistd.h>
67 #include <sys/kern_syscall.h>
68 
69 
70 #include <machine/ipl.h>
71 #include <machine/cpu.h>
72 #include <machine/smp.h>
73 
74 static int	coredump(struct proc *);
75 static char	*expand_name(const char *, uid_t, pid_t);
76 static int	killpg(int sig, int pgid, int all);
77 static int	sig_ffs(sigset_t *set);
78 static int	sigprop(int sig);
79 static void	stop(struct proc *);
80 #ifdef SMP
81 static void	signotify_remote(void *arg);
82 #endif
83 
84 static int	filt_sigattach(struct knote *kn);
85 static void	filt_sigdetach(struct knote *kn);
86 static int	filt_signal(struct knote *kn, long hint);
87 
88 struct filterops sig_filtops =
89 	{ 0, filt_sigattach, filt_sigdetach, filt_signal };
90 
91 static int	kern_logsigexit = 1;
92 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
93     &kern_logsigexit, 0,
94     "Log processes quitting on abnormal signals to syslog(3)");
95 
96 /*
97  * Can process p, with pcred pc, send the signal sig to process q?
98  */
99 #define CANSIGNAL(q, sig) \
100 	(!p_trespass(curproc->p_ucred, (q)->p_ucred) || \
101 	((sig) == SIGCONT && (q)->p_session == curproc->p_session))
102 
103 /*
104  * Policy -- Can real uid ruid with ucred uc send a signal to process q?
105  */
106 #define CANSIGIO(ruid, uc, q) \
107 	((uc)->cr_uid == 0 || \
108 	    (ruid) == (q)->p_ucred->cr_ruid || \
109 	    (uc)->cr_uid == (q)->p_ucred->cr_ruid || \
110 	    (ruid) == (q)->p_ucred->cr_uid || \
111 	    (uc)->cr_uid == (q)->p_ucred->cr_uid)
112 
113 int sugid_coredump;
114 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RW,
115 	&sugid_coredump, 0, "Enable coredumping set user/group ID processes");
116 
117 static int	do_coredump = 1;
118 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
119 	&do_coredump, 0, "Enable/Disable coredumps");
120 
121 /*
122  * Signal properties and actions.
123  * The array below categorizes the signals and their default actions
124  * according to the following properties:
125  */
126 #define	SA_KILL		0x01		/* terminates process by default */
127 #define	SA_CORE		0x02		/* ditto and coredumps */
128 #define	SA_STOP		0x04		/* suspend process */
129 #define	SA_TTYSTOP	0x08		/* ditto, from tty */
130 #define	SA_IGNORE	0x10		/* ignore by default */
131 #define	SA_CONT		0x20		/* continue if suspended */
132 #define	SA_CANTMASK	0x40		/* non-maskable, catchable */
133 #define SA_CKPT         0x80            /* checkpoint process */
134 
135 
136 static int sigproptbl[NSIG] = {
137         SA_KILL,                /* SIGHUP */
138         SA_KILL,                /* SIGINT */
139         SA_KILL|SA_CORE,        /* SIGQUIT */
140         SA_KILL|SA_CORE,        /* SIGILL */
141         SA_KILL|SA_CORE,        /* SIGTRAP */
142         SA_KILL|SA_CORE,        /* SIGABRT */
143         SA_KILL|SA_CORE,        /* SIGEMT */
144         SA_KILL|SA_CORE,        /* SIGFPE */
145         SA_KILL,                /* SIGKILL */
146         SA_KILL|SA_CORE,        /* SIGBUS */
147         SA_KILL|SA_CORE,        /* SIGSEGV */
148         SA_KILL|SA_CORE,        /* SIGSYS */
149         SA_KILL,                /* SIGPIPE */
150         SA_KILL,                /* SIGALRM */
151         SA_KILL,                /* SIGTERM */
152         SA_IGNORE,              /* SIGURG */
153         SA_STOP,                /* SIGSTOP */
154         SA_STOP|SA_TTYSTOP,     /* SIGTSTP */
155         SA_IGNORE|SA_CONT,      /* SIGCONT */
156         SA_IGNORE,              /* SIGCHLD */
157         SA_STOP|SA_TTYSTOP,     /* SIGTTIN */
158         SA_STOP|SA_TTYSTOP,     /* SIGTTOU */
159         SA_IGNORE,              /* SIGIO */
160         SA_KILL,                /* SIGXCPU */
161         SA_KILL,                /* SIGXFSZ */
162         SA_KILL,                /* SIGVTALRM */
163         SA_KILL,                /* SIGPROF */
164         SA_IGNORE,              /* SIGWINCH  */
165         SA_IGNORE,              /* SIGINFO */
166         SA_KILL,                /* SIGUSR1 */
167         SA_KILL,                /* SIGUSR2 */
168 	SA_IGNORE,              /* SIGTHR */
169 	SA_CKPT,                /* SIGCKPT */
170 	SA_KILL|SA_CKPT,        /* SIGCKPTEXIT */
171 	SA_IGNORE,
172 	SA_IGNORE,
173 	SA_IGNORE,
174 	SA_IGNORE,
175 	SA_IGNORE,
176 	SA_IGNORE,
177 	SA_IGNORE,
178 	SA_IGNORE,
179 	SA_IGNORE,
180 	SA_IGNORE,
181 	SA_IGNORE,
182 	SA_IGNORE,
183 	SA_IGNORE,
184 	SA_IGNORE,
185 	SA_IGNORE,
186 	SA_IGNORE,
187 	SA_IGNORE,
188 	SA_IGNORE,
189 	SA_IGNORE,
190 	SA_IGNORE,
191 	SA_IGNORE,
192 	SA_IGNORE,
193 	SA_IGNORE,
194 	SA_IGNORE,
195 	SA_IGNORE,
196 	SA_IGNORE,
197 	SA_IGNORE,
198 	SA_IGNORE,
199 	SA_IGNORE,
200 	SA_IGNORE,
201 
202 };
203 
204 static __inline int
205 sigprop(int sig)
206 {
207 
208 	if (sig > 0 && sig < NSIG)
209 		return (sigproptbl[_SIG_IDX(sig)]);
210 	return (0);
211 }
212 
213 static __inline int
214 sig_ffs(sigset_t *set)
215 {
216 	int i;
217 
218 	for (i = 0; i < _SIG_WORDS; i++)
219 		if (set->__bits[i])
220 			return (ffs(set->__bits[i]) + (i * 32));
221 	return (0);
222 }
223 
224 int
225 kern_sigaction(int sig, struct sigaction *act, struct sigaction *oact)
226 {
227 	struct thread *td = curthread;
228 	struct proc *p = td->td_proc;
229 	struct sigacts *ps = p->p_sigacts;
230 
231 	if (sig <= 0 || sig > _SIG_MAXSIG)
232 		return (EINVAL);
233 
234 	if (oact) {
235 		oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
236 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
237 		oact->sa_flags = 0;
238 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
239 			oact->sa_flags |= SA_ONSTACK;
240 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
241 			oact->sa_flags |= SA_RESTART;
242 		if (SIGISMEMBER(ps->ps_sigreset, sig))
243 			oact->sa_flags |= SA_RESETHAND;
244 		if (SIGISMEMBER(ps->ps_signodefer, sig))
245 			oact->sa_flags |= SA_NODEFER;
246 		if (SIGISMEMBER(ps->ps_siginfo, sig))
247 			oact->sa_flags |= SA_SIGINFO;
248 		if (sig == SIGCHLD && p->p_procsig->ps_flag & PS_NOCLDSTOP)
249 			oact->sa_flags |= SA_NOCLDSTOP;
250 		if (sig == SIGCHLD && p->p_procsig->ps_flag & PS_NOCLDWAIT)
251 			oact->sa_flags |= SA_NOCLDWAIT;
252 	}
253 	if (act) {
254 		if ((sig == SIGKILL || sig == SIGSTOP) &&
255 		    act->sa_handler != SIG_DFL)
256 			return (EINVAL);
257 
258 		/*
259 		 * Change setting atomically.
260 		 */
261 		splhigh();
262 
263 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
264 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
265 		if (act->sa_flags & SA_SIGINFO) {
266 			ps->ps_sigact[_SIG_IDX(sig)] =
267 			    (__sighandler_t *)act->sa_sigaction;
268 			SIGADDSET(ps->ps_siginfo, sig);
269 		} else {
270 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
271 			SIGDELSET(ps->ps_siginfo, sig);
272 		}
273 		if (!(act->sa_flags & SA_RESTART))
274 			SIGADDSET(ps->ps_sigintr, sig);
275 		else
276 			SIGDELSET(ps->ps_sigintr, sig);
277 		if (act->sa_flags & SA_ONSTACK)
278 			SIGADDSET(ps->ps_sigonstack, sig);
279 		else
280 			SIGDELSET(ps->ps_sigonstack, sig);
281 		if (act->sa_flags & SA_RESETHAND)
282 			SIGADDSET(ps->ps_sigreset, sig);
283 		else
284 			SIGDELSET(ps->ps_sigreset, sig);
285 		if (act->sa_flags & SA_NODEFER)
286 			SIGADDSET(ps->ps_signodefer, sig);
287 		else
288 			SIGDELSET(ps->ps_signodefer, sig);
289 		if (sig == SIGCHLD) {
290 			if (act->sa_flags & SA_NOCLDSTOP)
291 				p->p_procsig->ps_flag |= PS_NOCLDSTOP;
292 			else
293 				p->p_procsig->ps_flag &= ~PS_NOCLDSTOP;
294 			if (act->sa_flags & SA_NOCLDWAIT) {
295 				/*
296 				 * Paranoia: since SA_NOCLDWAIT is implemented
297 				 * by reparenting the dying child to PID 1 (and
298 				 * trust it to reap the zombie), PID 1 itself
299 				 * is forbidden to set SA_NOCLDWAIT.
300 				 */
301 				if (p->p_pid == 1)
302 					p->p_procsig->ps_flag &= ~PS_NOCLDWAIT;
303 				else
304 					p->p_procsig->ps_flag |= PS_NOCLDWAIT;
305 			} else {
306 				p->p_procsig->ps_flag &= ~PS_NOCLDWAIT;
307 			}
308 		}
309 		/*
310 		 * Set bit in p_sigignore for signals that are set to SIG_IGN,
311 		 * and for signals set to SIG_DFL where the default is to
312 		 * ignore. However, don't put SIGCONT in p_sigignore, as we
313 		 * have to restart the process.
314 		 */
315 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
316 		    (sigprop(sig) & SA_IGNORE &&
317 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
318 			/* never to be seen again */
319 			SIGDELSET(p->p_siglist, sig);
320 			if (sig != SIGCONT)
321 				/* easier in psignal */
322 				SIGADDSET(p->p_sigignore, sig);
323 			SIGDELSET(p->p_sigcatch, sig);
324 		} else {
325 			SIGDELSET(p->p_sigignore, sig);
326 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
327 				SIGDELSET(p->p_sigcatch, sig);
328 			else
329 				SIGADDSET(p->p_sigcatch, sig);
330 		}
331 
332 		spl0();
333 	}
334 	return (0);
335 }
336 
337 int
338 sigaction(struct sigaction_args *uap)
339 {
340 	struct sigaction act, oact;
341 	struct sigaction *actp, *oactp;
342 	int error;
343 
344 	actp = (uap->act != NULL) ? &act : NULL;
345 	oactp = (uap->oact != NULL) ? &oact : NULL;
346 	if (actp) {
347 		error = copyin(uap->act, actp, sizeof(act));
348 		if (error)
349 			return (error);
350 	}
351 	error = kern_sigaction(uap->sig, actp, oactp);
352 	if (oactp && !error) {
353 		error = copyout(oactp, uap->oact, sizeof(oact));
354 	}
355 	return (error);
356 }
357 
358 /*
359  * Initialize signal state for process 0;
360  * set to ignore signals that are ignored by default.
361  */
362 void
363 siginit(struct proc *p)
364 {
365 	int i;
366 
367 	for (i = 1; i <= NSIG; i++)
368 		if (sigprop(i) & SA_IGNORE && i != SIGCONT)
369 			SIGADDSET(p->p_sigignore, i);
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 = p->p_sigacts;
379 	int sig;
380 
381 	/*
382 	 * Reset caught signals.  Held signals remain held
383 	 * through p_sigmask (unless they were caught,
384 	 * and are now ignored by default).
385 	 */
386 	while (SIGNOTEMPTY(p->p_sigcatch)) {
387 		sig = sig_ffs(&p->p_sigcatch);
388 		SIGDELSET(p->p_sigcatch, sig);
389 		if (sigprop(sig) & SA_IGNORE) {
390 			if (sig != SIGCONT)
391 				SIGADDSET(p->p_sigignore, sig);
392 			SIGDELSET(p->p_siglist, sig);
393 		}
394 		ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
395 	}
396 	/*
397 	 * Reset stack state to the user stack.
398 	 * Clear set of signals caught on the signal stack.
399 	 */
400 	p->p_sigstk.ss_flags = SS_DISABLE;
401 	p->p_sigstk.ss_size = 0;
402 	p->p_sigstk.ss_sp = 0;
403 	p->p_flag &= ~P_ALTSTACK;
404 	/*
405 	 * Reset no zombies if child dies flag as Solaris does.
406 	 */
407 	p->p_procsig->ps_flag &= ~PS_NOCLDWAIT;
408 }
409 
410 /*
411  * kern_sigprocmask() - MP SAFE ONLY IF p == curproc
412  *
413  *	Manipulate signal mask.  This routine is MP SAFE *ONLY* if
414  *	p == curproc.  Also remember that in order to remain MP SAFE
415  *	no spl*() calls may be made.
416  */
417 int
418 kern_sigprocmask(int how, sigset_t *set, sigset_t *oset)
419 {
420 	struct thread *td = curthread;
421 	struct proc *p = td->td_proc;
422 	int error;
423 
424 	if (oset != NULL)
425 		*oset = p->p_sigmask;
426 
427 	error = 0;
428 	if (set != NULL) {
429 		switch (how) {
430 		case SIG_BLOCK:
431 			SIG_CANTMASK(*set);
432 			SIGSETOR(p->p_sigmask, *set);
433 			break;
434 		case SIG_UNBLOCK:
435 			SIGSETNAND(p->p_sigmask, *set);
436 			break;
437 		case SIG_SETMASK:
438 			SIG_CANTMASK(*set);
439 			p->p_sigmask = *set;
440 			break;
441 		default:
442 			error = EINVAL;
443 			break;
444 		}
445 	}
446 	return (error);
447 }
448 
449 /*
450  * sigprocmask() - MP SAFE
451  */
452 int
453 sigprocmask(struct sigprocmask_args *uap)
454 {
455 	sigset_t set, oset;
456 	sigset_t *setp, *osetp;
457 	int error;
458 
459 	setp = (uap->set != NULL) ? &set : NULL;
460 	osetp = (uap->oset != NULL) ? &oset : NULL;
461 	if (setp) {
462 		error = copyin(uap->set, setp, sizeof(set));
463 		if (error)
464 			return (error);
465 	}
466 	error = kern_sigprocmask(uap->how, setp, osetp);
467 	if (osetp && !error) {
468 		error = copyout(osetp, uap->oset, sizeof(oset));
469 	}
470 	return (error);
471 }
472 
473 int
474 kern_sigpending(struct __sigset *set)
475 {
476 	struct thread *td = curthread;
477 	struct proc *p = td->td_proc;
478 
479 	*set = p->p_siglist;
480 
481 	return (0);
482 }
483 
484 int
485 sigpending(struct sigpending_args *uap)
486 {
487 	sigset_t set;
488 	int error;
489 
490 	error = kern_sigpending(&set);
491 
492 	if (error == 0)
493 		error = copyout(&set, uap->set, sizeof(set));
494 	return (error);
495 }
496 
497 /*
498  * Suspend process until signal, providing mask to be set
499  * in the meantime.
500  */
501 int
502 kern_sigsuspend(struct __sigset *set)
503 {
504 	struct thread *td = curthread;
505 	struct proc *p = td->td_proc;
506 	struct sigacts *ps = p->p_sigacts;
507 
508 	/*
509 	 * When returning from sigsuspend, we want
510 	 * the old mask to be restored after the
511 	 * signal handler has finished.  Thus, we
512 	 * save it here and mark the sigacts structure
513 	 * to indicate this.
514 	 */
515 	p->p_oldsigmask = p->p_sigmask;
516 	p->p_flag |= P_OLDMASK;
517 
518 	SIG_CANTMASK(*set);
519 	p->p_sigmask = *set;
520 	while (tsleep(ps, PCATCH, "pause", 0) == 0)
521 		/* void */;
522 	/* always return EINTR rather than ERESTART... */
523 	return (EINTR);
524 }
525 
526 /*
527  * Note nonstandard calling convention: libc stub passes mask, not
528  * pointer, to save a copyin.
529  */
530 int
531 sigsuspend(struct sigsuspend_args *uap)
532 {
533 	sigset_t mask;
534 	int error;
535 
536 	error = copyin(uap->sigmask, &mask, sizeof(mask));
537 	if (error)
538 		return (error);
539 
540 	error = kern_sigsuspend(&mask);
541 
542 	return (error);
543 }
544 
545 int
546 kern_sigaltstack(struct sigaltstack *ss, struct sigaltstack *oss)
547 {
548 	struct thread *td = curthread;
549 	struct proc *p = td->td_proc;
550 
551 	if ((p->p_flag & P_ALTSTACK) == 0)
552 		p->p_sigstk.ss_flags |= SS_DISABLE;
553 
554 	if (oss)
555 		*oss = p->p_sigstk;
556 
557 	if (ss) {
558 		if (ss->ss_flags & SS_DISABLE) {
559 			if (p->p_sigstk.ss_flags & SS_ONSTACK)
560 				return (EINVAL);
561 			p->p_flag &= ~P_ALTSTACK;
562 			p->p_sigstk.ss_flags = ss->ss_flags;
563 		} else {
564 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
565 				return (ENOMEM);
566 			p->p_flag |= P_ALTSTACK;
567 			p->p_sigstk = *ss;
568 		}
569 	}
570 
571 	return (0);
572 }
573 
574 int
575 sigaltstack(struct sigaltstack_args *uap)
576 {
577 	stack_t ss, oss;
578 	int error;
579 
580 	if (uap->ss) {
581 		error = copyin(uap->ss, &ss, sizeof(ss));
582 		if (error)
583 			return (error);
584 	}
585 
586 	error = kern_sigaltstack(uap->ss ? &ss : NULL,
587 	    uap->oss ? &oss : NULL);
588 
589 	if (error == 0 && uap->oss)
590 		error = copyout(&oss, uap->oss, sizeof(*uap->oss));
591 	return (error);
592 }
593 
594 /*
595  * Common code for kill process group/broadcast kill.
596  * cp is calling process.
597  */
598 static int
599 killpg(int sig, int pgid, int all)
600 {
601 	struct proc *cp = curproc;
602 	struct proc *p;
603 	struct pgrp *pgrp;
604 	int nfound = 0;
605 
606 	if (all) {
607 		/*
608 		 * broadcast
609 		 */
610 		FOREACH_PROC_IN_SYSTEM(p) {
611 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
612 			    p == cp || !CANSIGNAL(p, sig))
613 				continue;
614 			nfound++;
615 			if (sig)
616 				psignal(p, sig);
617 		}
618 	} else {
619 		if (pgid == 0) {
620 			/*
621 			 * zero pgid means send to my process group.
622 			 */
623 			pgrp = cp->p_pgrp;
624 		} else {
625 			pgrp = pgfind(pgid);
626 			if (pgrp == NULL)
627 				return (ESRCH);
628 		}
629 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
630 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
631 			    p->p_stat == SZOMB ||
632 			    !CANSIGNAL(p, sig))
633 				continue;
634 			nfound++;
635 			if (sig)
636 				psignal(p, sig);
637 		}
638 	}
639 	return (nfound ? 0 : ESRCH);
640 }
641 
642 int
643 kern_kill(int sig, int pid)
644 {
645 	struct thread *td = curthread;
646 	struct proc *p = td->td_proc;
647 
648 	if ((u_int)sig > _SIG_MAXSIG)
649 		return (EINVAL);
650 	if (pid > 0) {
651 		/* kill single process */
652 		if ((p = pfind(pid)) == NULL)
653 			return (ESRCH);
654 		if (!CANSIGNAL(p, sig))
655 			return (EPERM);
656 		if (sig)
657 			psignal(p, sig);
658 		return (0);
659 	}
660 	switch (pid) {
661 	case -1:		/* broadcast signal */
662 		return (killpg(sig, 0, 1));
663 	case 0:			/* signal own process group */
664 		return (killpg(sig, 0, 0));
665 	default:		/* negative explicit process group */
666 		return (killpg(sig, -pid, 0));
667 	}
668 	/* NOTREACHED */
669 }
670 
671 int
672 kill(struct kill_args *uap)
673 {
674 	int error;
675 
676 	error = kern_kill(uap->signum, uap->pid);
677 
678 	return (error);
679 }
680 
681 /*
682  * Send a signal to a process group.
683  */
684 void
685 gsignal(int pgid, int sig)
686 {
687 	struct pgrp *pgrp;
688 
689 	if (pgid && (pgrp = pgfind(pgid)))
690 		pgsignal(pgrp, sig, 0);
691 }
692 
693 /*
694  * Send a signal to a process group.  If checktty is 1,
695  * limit to members which have a controlling terminal.
696  */
697 void
698 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
699 {
700 	struct proc *p;
701 
702 	if (pgrp)
703 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist)
704 			if (checkctty == 0 || p->p_flag & P_CONTROLT)
705 				psignal(p, sig);
706 }
707 
708 /*
709  * Send a signal caused by a trap to the current process.
710  * If it will be caught immediately, deliver it with correct code.
711  * Otherwise, post it normally.
712  */
713 void
714 trapsignal(struct proc *p, int sig, u_long code)
715 {
716 	struct sigacts *ps = p->p_sigacts;
717 
718 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(p->p_sigcatch, sig) &&
719 	    !SIGISMEMBER(p->p_sigmask, sig)) {
720 		p->p_stats->p_ru.ru_nsignals++;
721 #ifdef KTRACE
722 		if (KTRPOINT(p->p_thread, KTR_PSIG))
723 			ktrpsig(p->p_tracep, sig, ps->ps_sigact[_SIG_IDX(sig)],
724 				&p->p_sigmask, code);
725 #endif
726 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], sig,
727 						&p->p_sigmask, code);
728 		SIGSETOR(p->p_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
729 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
730 			SIGADDSET(p->p_sigmask, sig);
731 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
732 			/*
733 			 * See kern_sigaction() for origin of this code.
734 			 */
735 			SIGDELSET(p->p_sigcatch, sig);
736 			if (sig != SIGCONT &&
737 			    sigprop(sig) & SA_IGNORE)
738 				SIGADDSET(p->p_sigignore, sig);
739 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
740 		}
741 	} else {
742 		p->p_code = code;	/* XXX for core dump/debugger */
743 		p->p_sig = sig;		/* XXX to verify code */
744 		psignal(p, sig);
745 	}
746 }
747 
748 /*
749  * Send the signal to the process.  If the signal has an action, the action
750  * is usually performed by the target process rather than the caller; we add
751  * the signal to the set of pending signals for the process.
752  *
753  * Exceptions:
754  *   o When a stop signal is sent to a sleeping process that takes the
755  *     default action, the process is stopped without awakening it.
756  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
757  *     regardless of the signal action (eg, blocked or ignored).
758  *
759  * Other ignored signals are discarded immediately.
760  */
761 
762 /*
763  * temporary hack to allow checkpoint code to continue to
764  * be in a module for the moment
765  */
766 
767 static proc_func_t ckpt_func;
768 
769 proc_func_t
770 register_ckpt_func(proc_func_t func)
771 {
772 	proc_func_t old_func;
773 
774 	old_func = ckpt_func;
775 	ckpt_func = func;
776 	return (old_func);
777 }
778 
779 void
780 psignal(struct proc *p, int sig)
781 {
782 	int s, prop;
783 	sig_t action;
784 
785 	if (sig > _SIG_MAXSIG || sig <= 0) {
786 		printf("psignal: signal %d\n", sig);
787 		panic("psignal signal number");
788 	}
789 
790 	s = splhigh();
791 	KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
792 	splx(s);
793 
794 	prop = sigprop(sig);
795 
796 	/*
797 	 * If proc is traced, always give parent a chance;
798 	 * if signal event is tracked by procfs, give *that*
799 	 * a chance, as well.
800 	 */
801 	if ((p->p_flag & P_TRACED) || (p->p_stops & S_SIG)) {
802 		action = SIG_DFL;
803 	} else {
804 		/*
805 		 * If the signal is being ignored,
806 		 * then we forget about it immediately.
807 		 * (Note: we don't set SIGCONT in p_sigignore,
808 		 * and if it is set to SIG_IGN,
809 		 * action will be SIG_DFL here.)
810 		 */
811 		if (SIGISMEMBER(p->p_sigignore, sig) || (p->p_flag & P_WEXIT))
812 			return;
813 		if (SIGISMEMBER(p->p_sigmask, sig))
814 			action = SIG_HOLD;
815 		else if (SIGISMEMBER(p->p_sigcatch, sig))
816 			action = SIG_CATCH;
817 		else
818 			action = SIG_DFL;
819 	}
820 
821 	if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
822 	    (p->p_flag & P_TRACED) == 0) {
823 		p->p_nice = NZERO;
824 	}
825 
826 	if (prop & SA_CONT)
827 		SIG_STOPSIGMASK(p->p_siglist);
828 
829 
830 	if (prop & SA_STOP) {
831 		/*
832 		 * If sending a tty stop signal to a member of an orphaned
833 		 * process group, discard the signal here if the action
834 		 * is default; don't stop the process below if sleeping,
835 		 * and don't clear any pending SIGCONT.
836 		 */
837 		if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0 &&
838 		    action == SIG_DFL) {
839 		        return;
840 		}
841 		SIG_CONTSIGMASK(p->p_siglist);
842 	}
843 	SIGADDSET(p->p_siglist, sig);
844 
845 	/*
846 	 * Defer further processing for signals which are held,
847 	 * except that stopped processes must be continued by SIGCONT.
848 	 */
849 	if (action == SIG_HOLD && (!(prop & SA_CONT) || p->p_stat != SSTOP))
850 		return;
851 	s = splhigh();
852 	switch (p->p_stat) {
853 	case SSLEEP:
854 		/*
855 		 * If process is sleeping uninterruptibly
856 		 * we can't interrupt the sleep... the signal will
857 		 * be noticed when the process returns through
858 		 * trap() or syscall().
859 		 */
860 		if ((p->p_flag & P_SINTR) == 0)
861 			goto out;
862 		/*
863 		 * Process is sleeping and traced... make it runnable
864 		 * so it can discover the signal in issignal() and stop
865 		 * for the parent.
866 		 */
867 		if (p->p_flag & P_TRACED)
868 			goto run;
869 		/*
870 		 * If SIGCONT is default (or ignored) and process is
871 		 * asleep, we are finished; the process should not
872 		 * be awakened.
873 		 */
874 		if ((prop & SA_CONT) && action == SIG_DFL) {
875 			SIGDELSET(p->p_siglist, sig);
876 			goto out;
877 		}
878 		/*
879 		 * When a sleeping process receives a stop
880 		 * signal, process immediately if possible.
881 		 * All other (caught or default) signals
882 		 * cause the process to run.
883 		 */
884 		if (prop & SA_STOP) {
885 			if (action != SIG_DFL)
886 				goto run;
887 			/*
888 			 * If a child holding parent blocked,
889 			 * stopping could cause deadlock.
890 			 */
891 			if (p->p_flag & P_PPWAIT)
892 				goto out;
893 			SIGDELSET(p->p_siglist, sig);
894 			p->p_xstat = sig;
895 			if ((p->p_pptr->p_procsig->ps_flag & PS_NOCLDSTOP) == 0)
896 				psignal(p->p_pptr, SIGCHLD);
897 			stop(p);
898 			goto out;
899 		} else {
900 			goto run;
901 		}
902 		/*NOTREACHED*/
903 	case SSTOP:
904 		/*
905 		 * If traced process is already stopped,
906 		 * then no further action is necessary.
907 		 */
908 		if (p->p_flag & P_TRACED)
909 			goto out;
910 
911 		/*
912 		 * Kill signal always sets processes running.
913 		 */
914 		if (sig == SIGKILL)
915 			goto run;
916 
917 		if (prop & SA_CONT) {
918 			/*
919 			 * If SIGCONT is default (or ignored), we continue the
920 			 * process but don't leave the signal in p_siglist, as
921 			 * it has no further action.  If SIGCONT is held, we
922 			 * continue the process and leave the signal in
923 			 * p_siglist.  If the process catches SIGCONT, let it
924 			 * handle the signal itself.  If it isn't waiting on
925 			 * an event, then it goes back to run state.
926 			 * Otherwise, process goes back to sleep state.
927 			 */
928 			if (action == SIG_DFL)
929 				SIGDELSET(p->p_siglist, sig);
930 			if (action == SIG_CATCH)
931 				goto run;
932 			if (p->p_wchan == 0)
933 				goto run;
934 			clrrunnable(p, SSLEEP);
935 			goto out;
936 		}
937 
938 		if (prop & SA_STOP) {
939 			/*
940 			 * Already stopped, don't need to stop again.
941 			 * (If we did the shell could get confused.)
942 			 */
943 			SIGDELSET(p->p_siglist, sig);
944 			goto out;
945 		}
946 
947 		/*
948 		 * If process is sleeping interruptibly, then simulate a
949 		 * wakeup so that when it is continued, it will be made
950 		 * runnable and can look at the signal.  But don't make
951 		 * the process runnable, leave it stopped.
952 		 */
953 		if (p->p_wchan && (p->p_flag & P_SINTR))
954 			unsleep(p->p_thread);
955 		goto out;
956 	default:
957 		/*
958 		 * SRUN, SIDL, SZOMB do nothing with the signal,
959 		 * other than kicking ourselves if we are running.
960 		 * It will either never be noticed, or noticed very soon.
961 		 *
962 		 * For SMP we may have to forward the request to another cpu.
963 		 * YYY the MP lock prevents the target process from moving
964 		 * to another cpu, see kern/kern_switch.c
965 		 */
966 #ifdef SMP
967 		if (p == lwkt_preempted_proc()) {
968 			signotify();
969 		} else {
970 			struct thread *td = p->p_thread;
971 
972 			if (td->td_gd != mycpu)
973 				lwkt_send_ipiq(td->td_gd, signotify_remote, p);
974 		}
975 #else
976 		if (p == lwkt_preempted_proc())
977 			signotify();
978 #endif
979 		goto out;
980 	}
981 	/*NOTREACHED*/
982 run:
983 	setrunnable(p);
984 out:
985 	splx(s);
986 }
987 
988 #ifdef SMP
989 
990 /*
991  * This function is called via an IPI.  We will be in a critical section but
992  * the MP lock will NOT be held.  Also note that by the time the ipi message
993  * gets to us the process 'p' (arg) may no longer be scheduled or even valid.
994  */
995 static void
996 signotify_remote(void *arg)
997 {
998 	struct proc *p = arg;
999 	if (p == lwkt_preempted_proc())
1000 		signotify();
1001 }
1002 
1003 #endif
1004 
1005 /*
1006  * If the current process has received a signal that would interrupt a
1007  * system call, return EINTR or ERESTART as appropriate.
1008  */
1009 int
1010 iscaught(struct proc *p)
1011 {
1012 	int sig;
1013 
1014 	if (p) {
1015 		if ((sig = CURSIG(p)) != 0) {
1016 			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1017 				return (EINTR);
1018 			return (ERESTART);
1019 		}
1020 	}
1021 	return(EWOULDBLOCK);
1022 }
1023 
1024 /*
1025  * If the current process has received a signal (should be caught or cause
1026  * termination, should interrupt current syscall), return the signal number.
1027  * Stop signals with default action are processed immediately, then cleared;
1028  * they aren't returned.  This is checked after each entry to the system for
1029  * a syscall or trap (though this can usually be done without calling issignal
1030  * by checking the pending signal masks in the CURSIG macro.) The normal call
1031  * sequence is
1032  *
1033  *	while (sig = CURSIG(curproc))
1034  *		postsig(sig);
1035  */
1036 int
1037 issignal(struct proc *p)
1038 {
1039 	sigset_t mask;
1040 	int sig, prop;
1041 
1042 	for (;;) {
1043 		int traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
1044 
1045 		mask = p->p_siglist;
1046 		SIGSETNAND(mask, p->p_sigmask);
1047 		if (p->p_flag & P_PPWAIT)
1048 			SIG_STOPSIGMASK(mask);
1049 		if (!SIGNOTEMPTY(mask))	 	/* no signal to send */
1050 			return (0);
1051 		sig = sig_ffs(&mask);
1052 
1053 		STOPEVENT(p, S_SIG, sig);
1054 
1055 		/*
1056 		 * We should see pending but ignored signals
1057 		 * only if P_TRACED was on when they were posted.
1058 		 */
1059 		if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
1060 			SIGDELSET(p->p_siglist, sig);
1061 			continue;
1062 		}
1063 		if (p->p_flag & P_TRACED && (p->p_flag & P_PPWAIT) == 0) {
1064 			/*
1065 			 * If traced, always stop, and stay
1066 			 * stopped until released by the parent.
1067 			 */
1068 			p->p_xstat = sig;
1069 			psignal(p->p_pptr, SIGCHLD);
1070 			do {
1071 				stop(p);
1072 				mi_switch(p);
1073 			} while (!trace_req(p) && p->p_flag & P_TRACED);
1074 
1075 			/*
1076 			 * If parent wants us to take the signal,
1077 			 * then it will leave it in p->p_xstat;
1078 			 * otherwise we just look for signals again.
1079 			 */
1080 			SIGDELSET(p->p_siglist, sig);	/* clear old signal */
1081 			sig = p->p_xstat;
1082 			if (sig == 0)
1083 				continue;
1084 
1085 			/*
1086 			 * Put the new signal into p_siglist.  If the
1087 			 * signal is being masked, look for other signals.
1088 			 */
1089 			SIGADDSET(p->p_siglist, sig);
1090 			if (SIGISMEMBER(p->p_sigmask, sig))
1091 				continue;
1092 
1093 			/*
1094 			 * If the traced bit got turned off, go back up
1095 			 * to the top to rescan signals.  This ensures
1096 			 * that p_sig* and ps_sigact are consistent.
1097 			 */
1098 			if ((p->p_flag & P_TRACED) == 0)
1099 				continue;
1100 		}
1101 
1102 		prop = sigprop(sig);
1103 
1104 		/*
1105 		 * Decide whether the signal should be returned.
1106 		 * Return the signal's number, or fall through
1107 		 * to clear it from the pending mask.
1108 		 */
1109 		switch ((int)(intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
1110 
1111 		case (int)SIG_DFL:
1112 			/*
1113 			 * Don't take default actions on system processes.
1114 			 */
1115 			if (p->p_pid <= 1) {
1116 #ifdef DIAGNOSTIC
1117 				/*
1118 				 * Are you sure you want to ignore SIGSEGV
1119 				 * in init? XXX
1120 				 */
1121 				printf("Process (pid %lu) got signal %d\n",
1122 					(u_long)p->p_pid, sig);
1123 #endif
1124 				break;		/* == ignore */
1125 			}
1126 
1127 			/*
1128 			 * Handle the in-kernel checkpoint action
1129 			 */
1130 			if (prop & SA_CKPT) {
1131 				if (ckpt_func)
1132 					ckpt_func(p);
1133 				break;
1134 			}
1135 
1136 			/*
1137 			 * If there is a pending stop signal to process
1138 			 * with default action, stop here,
1139 			 * then clear the signal.  However,
1140 			 * if process is member of an orphaned
1141 			 * process group, ignore tty stop signals.
1142 			 */
1143 			if (prop & SA_STOP) {
1144 				if (p->p_flag & P_TRACED ||
1145 		    		    (p->p_pgrp->pg_jobc == 0 &&
1146 				    prop & SA_TTYSTOP))
1147 					break;	/* == ignore */
1148 				p->p_xstat = sig;
1149 				stop(p);
1150 				if ((p->p_pptr->p_procsig->ps_flag & PS_NOCLDSTOP) == 0)
1151 					psignal(p->p_pptr, SIGCHLD);
1152 				mi_switch(p);
1153 				break;
1154 			} else if (prop & SA_IGNORE) {
1155 				/*
1156 				 * Except for SIGCONT, shouldn't get here.
1157 				 * Default action is to ignore; drop it.
1158 				 */
1159 				break;		/* == ignore */
1160 			} else {
1161 				return (sig);
1162 			}
1163 
1164 			/*NOTREACHED*/
1165 
1166 		case (int)SIG_IGN:
1167 			/*
1168 			 * Masking above should prevent us ever trying
1169 			 * to take action on an ignored signal other
1170 			 * than SIGCONT, unless process is traced.
1171 			 */
1172 			if ((prop & SA_CONT) == 0 &&
1173 			    (p->p_flag & P_TRACED) == 0)
1174 				printf("issignal\n");
1175 			break;		/* == ignore */
1176 
1177 		default:
1178 			/*
1179 			 * This signal has an action, let
1180 			 * postsig() process it.
1181 			 */
1182 			return (sig);
1183 		}
1184 		SIGDELSET(p->p_siglist, sig);		/* take the signal! */
1185 	}
1186 	/* NOTREACHED */
1187 }
1188 
1189 /*
1190  * Put the argument process into the stopped state and notify the parent
1191  * via wakeup.  Signals are handled elsewhere.  The process must not be
1192  * on the run queue.
1193  */
1194 void
1195 stop(struct proc *p)
1196 {
1197 	p->p_stat = SSTOP;
1198 	p->p_flag &= ~P_WAITED;
1199 	wakeup((caddr_t)p->p_pptr);
1200 }
1201 
1202 /*
1203  * Take the action for the specified signal
1204  * from the current set of pending signals.
1205  */
1206 void
1207 postsig(int sig)
1208 {
1209 	struct proc *p = curproc;
1210 	struct sigacts *ps = p->p_sigacts;
1211 	sig_t action;
1212 	sigset_t returnmask;
1213 	int code;
1214 
1215 	KASSERT(sig != 0, ("postsig"));
1216 
1217 	SIGDELSET(p->p_siglist, sig);
1218 	action = ps->ps_sigact[_SIG_IDX(sig)];
1219 #ifdef KTRACE
1220 	if (KTRPOINT(p->p_thread, KTR_PSIG))
1221 		ktrpsig(p->p_tracep, sig, action, p->p_flag & P_OLDMASK ?
1222 		    &p->p_oldsigmask : &p->p_sigmask, 0);
1223 #endif
1224 	STOPEVENT(p, S_SIG, sig);
1225 
1226 	if (action == SIG_DFL) {
1227 		/*
1228 		 * Default action, where the default is to kill
1229 		 * the process.  (Other cases were ignored above.)
1230 		 */
1231 		sigexit(p, sig);
1232 		/* NOTREACHED */
1233 	} else {
1234 		/*
1235 		 * If we get here, the signal must be caught.
1236 		 */
1237 		KASSERT(action != SIG_IGN && !SIGISMEMBER(p->p_sigmask, sig),
1238 		    ("postsig action"));
1239 		/*
1240 		 * Set the new mask value and also defer further
1241 		 * occurrences of this signal.
1242 		 *
1243 		 * Special case: user has done a sigsuspend.  Here the
1244 		 * current mask is not of interest, but rather the
1245 		 * mask from before the sigsuspend is what we want
1246 		 * restored after the signal processing is completed.
1247 		 */
1248 		splhigh();
1249 		if (p->p_flag & P_OLDMASK) {
1250 			returnmask = p->p_oldsigmask;
1251 			p->p_flag &= ~P_OLDMASK;
1252 		} else {
1253 			returnmask = p->p_sigmask;
1254 		}
1255 
1256 		SIGSETOR(p->p_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
1257 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
1258 			SIGADDSET(p->p_sigmask, sig);
1259 
1260 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
1261 			/*
1262 			 * See kern_sigaction() for origin of this code.
1263 			 */
1264 			SIGDELSET(p->p_sigcatch, sig);
1265 			if (sig != SIGCONT &&
1266 			    sigprop(sig) & SA_IGNORE)
1267 				SIGADDSET(p->p_sigignore, sig);
1268 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1269 		}
1270 		spl0();
1271 		p->p_stats->p_ru.ru_nsignals++;
1272 		if (p->p_sig != sig) {
1273 			code = 0;
1274 		} else {
1275 			code = p->p_code;
1276 			p->p_code = 0;
1277 			p->p_sig = 0;
1278 		}
1279 		(*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
1280 	}
1281 }
1282 
1283 /*
1284  * Kill the current process for stated reason.
1285  */
1286 void
1287 killproc(struct proc *p, char *why)
1288 {
1289 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n", p->p_pid, p->p_comm,
1290 		p->p_ucred ? p->p_ucred->cr_uid : -1, why);
1291 	psignal(p, SIGKILL);
1292 }
1293 
1294 /*
1295  * Force the current process to exit with the specified signal, dumping core
1296  * if appropriate.  We bypass the normal tests for masked and caught signals,
1297  * allowing unrecoverable failures to terminate the process without changing
1298  * signal state.  Mark the accounting record with the signal termination.
1299  * If dumping core, save the signal number for the debugger.  Calls exit and
1300  * does not return.
1301  */
1302 void
1303 sigexit(struct proc *p, int sig)
1304 {
1305 	p->p_acflag |= AXSIG;
1306 	if (sigprop(sig) & SA_CORE) {
1307 		p->p_sig = sig;
1308 		/*
1309 		 * Log signals which would cause core dumps
1310 		 * (Log as LOG_INFO to appease those who don't want
1311 		 * these messages.)
1312 		 * XXX : Todo, as well as euid, write out ruid too
1313 		 */
1314 		if (coredump(p) == 0)
1315 			sig |= WCOREFLAG;
1316 		if (kern_logsigexit)
1317 			log(LOG_INFO,
1318 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
1319 			    p->p_pid, p->p_comm,
1320 			    p->p_ucred ? p->p_ucred->cr_uid : -1,
1321 			    sig &~ WCOREFLAG,
1322 			    sig & WCOREFLAG ? " (core dumped)" : "");
1323 	}
1324 	exit1(W_EXITCODE(0, sig));
1325 	/* NOTREACHED */
1326 }
1327 
1328 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
1329 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
1330 	      sizeof(corefilename), "process corefile name format string");
1331 
1332 /*
1333  * expand_name(name, uid, pid)
1334  * Expand the name described in corefilename, using name, uid, and pid.
1335  * corefilename is a printf-like string, with three format specifiers:
1336  *	%N	name of process ("name")
1337  *	%P	process id (pid)
1338  *	%U	user id (uid)
1339  * For example, "%N.core" is the default; they can be disabled completely
1340  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
1341  * This is controlled by the sysctl variable kern.corefile (see above).
1342  */
1343 
1344 static char *
1345 expand_name(const char *name, uid_t uid, pid_t pid)
1346 {
1347 	char *temp;
1348 	char buf[11];		/* Buffer for pid/uid -- max 4B */
1349 	int i, n;
1350 	char *format = corefilename;
1351 	size_t namelen;
1352 
1353 	temp = malloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
1354 	if (temp == NULL)
1355 		return NULL;
1356 	namelen = strlen(name);
1357 	for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
1358 		int l;
1359 		switch (format[i]) {
1360 		case '%':	/* Format character */
1361 			i++;
1362 			switch (format[i]) {
1363 			case '%':
1364 				temp[n++] = '%';
1365 				break;
1366 			case 'N':	/* process name */
1367 				if ((n + namelen) > MAXPATHLEN) {
1368 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
1369 					    pid, name, uid, temp, name);
1370 					free(temp, M_TEMP);
1371 					return NULL;
1372 				}
1373 				memcpy(temp+n, name, namelen);
1374 				n += namelen;
1375 				break;
1376 			case 'P':	/* process id */
1377 				l = sprintf(buf, "%u", pid);
1378 				if ((n + l) > MAXPATHLEN) {
1379 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
1380 					    pid, name, uid, temp, name);
1381 					free(temp, M_TEMP);
1382 					return NULL;
1383 				}
1384 				memcpy(temp+n, buf, l);
1385 				n += l;
1386 				break;
1387 			case 'U':	/* user id */
1388 				l = sprintf(buf, "%u", uid);
1389 				if ((n + l) > MAXPATHLEN) {
1390 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
1391 					    pid, name, uid, temp, name);
1392 					free(temp, M_TEMP);
1393 					return NULL;
1394 				}
1395 				memcpy(temp+n, buf, l);
1396 				n += l;
1397 				break;
1398 			default:
1399 			  	log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
1400 			}
1401 			break;
1402 		default:
1403 			temp[n++] = format[i];
1404 		}
1405 	}
1406 	temp[n] = '\0';
1407 	return temp;
1408 }
1409 
1410 /*
1411  * Dump a process' core.  The main routine does some
1412  * policy checking, and creates the name of the coredump;
1413  * then it passes on a vnode and a size limit to the process-specific
1414  * coredump routine if there is one; if there _is not_ one, it returns
1415  * ENOSYS; otherwise it returns the error from the process-specific routine.
1416  */
1417 
1418 static int
1419 coredump(struct proc *p)
1420 {
1421 	struct vnode *vp;
1422 	struct ucred *cred = p->p_ucred;
1423 	struct thread *td = p->p_thread;
1424 	struct flock lf;
1425 	struct nameidata nd;
1426 	struct vattr vattr;
1427 	int error, error1;
1428 	char *name;			/* name of corefile */
1429 	off_t limit;
1430 
1431 	STOPEVENT(p, S_CORE, 0);
1432 
1433 	if (((sugid_coredump == 0) && p->p_flag & P_SUGID) || do_coredump == 0)
1434 		return (EFAULT);
1435 
1436 	/*
1437 	 * Note that the bulk of limit checking is done after
1438 	 * the corefile is created.  The exception is if the limit
1439 	 * for corefiles is 0, in which case we don't bother
1440 	 * creating the corefile at all.  This layout means that
1441 	 * a corefile is truncated instead of not being created,
1442 	 * if it is larger than the limit.
1443 	 */
1444 	limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
1445 	if (limit == 0)
1446 		return EFBIG;
1447 
1448 	name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
1449 	if (name == NULL)
1450 		return (EINVAL);
1451 	NDINIT(&nd, NAMEI_LOOKUP, 0, UIO_SYSSPACE, name, td);
1452 	error = vn_open(&nd, O_CREAT | FWRITE | O_NOFOLLOW, S_IRUSR | S_IWUSR);
1453 	free(name, M_TEMP);
1454 	if (error)
1455 		return (error);
1456 	NDFREE(&nd, NDF_ONLY_PNBUF);
1457 	vp = nd.ni_vp;
1458 
1459 	VOP_UNLOCK(vp, NULL, 0, td);
1460 	lf.l_whence = SEEK_SET;
1461 	lf.l_start = 0;
1462 	lf.l_len = 0;
1463 	lf.l_type = F_WRLCK;
1464 	error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK);
1465 	if (error)
1466 		goto out2;
1467 
1468 	/* Don't dump to non-regular files or files with links. */
1469 	if (vp->v_type != VREG ||
1470 	    VOP_GETATTR(vp, &vattr, td) || vattr.va_nlink != 1) {
1471 		error = EFAULT;
1472 		goto out1;
1473 	}
1474 
1475 	VATTR_NULL(&vattr);
1476 	vn_lock(vp, NULL, LK_EXCLUSIVE | LK_RETRY, td);
1477 	vattr.va_size = 0;
1478 	VOP_LEASE(vp, td, cred, LEASE_WRITE);
1479 	VOP_SETATTR(vp, &vattr, cred, td);
1480 	p->p_acflag |= ACORE;
1481 	VOP_UNLOCK(vp, NULL, 0, td);
1482 
1483 	error = p->p_sysent->sv_coredump ?
1484 	  p->p_sysent->sv_coredump(p, vp, limit) :
1485 	  ENOSYS;
1486 
1487 out1:
1488 	lf.l_type = F_UNLCK;
1489 	VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
1490 out2:
1491 	error1 = vn_close(vp, FWRITE, td);
1492 	if (error == 0)
1493 		error = error1;
1494 	return (error);
1495 }
1496 
1497 /*
1498  * Nonexistent system call-- signal process (may want to handle it).
1499  * Flag error in case process won't see signal immediately (blocked or ignored).
1500  */
1501 /* ARGSUSED */
1502 int
1503 nosys(struct nosys_args *args)
1504 {
1505 	psignal(curproc, SIGSYS);
1506 	return (EINVAL);
1507 }
1508 
1509 /*
1510  * Send a SIGIO or SIGURG signal to a process or process group using
1511  * stored credentials rather than those of the current process.
1512  */
1513 void
1514 pgsigio(struct sigio *sigio, int sig, int checkctty)
1515 {
1516 	if (sigio == NULL)
1517 		return;
1518 
1519 	if (sigio->sio_pgid > 0) {
1520 		if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
1521 		             sigio->sio_proc))
1522 			psignal(sigio->sio_proc, sig);
1523 	} else if (sigio->sio_pgid < 0) {
1524 		struct proc *p;
1525 
1526 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist)
1527 			if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
1528 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
1529 				psignal(p, sig);
1530 	}
1531 }
1532 
1533 static int
1534 filt_sigattach(struct knote *kn)
1535 {
1536 	struct proc *p = curproc;
1537 
1538 	kn->kn_ptr.p_proc = p;
1539 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
1540 
1541 	/* XXX lock the proc here while adding to the list? */
1542 	SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext);
1543 
1544 	return (0);
1545 }
1546 
1547 static void
1548 filt_sigdetach(struct knote *kn)
1549 {
1550 	struct proc *p = kn->kn_ptr.p_proc;
1551 
1552 	SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
1553 }
1554 
1555 /*
1556  * signal knotes are shared with proc knotes, so we apply a mask to
1557  * the hint in order to differentiate them from process hints.  This
1558  * could be avoided by using a signal-specific knote list, but probably
1559  * isn't worth the trouble.
1560  */
1561 static int
1562 filt_signal(struct knote *kn, long hint)
1563 {
1564 	if (hint & NOTE_SIGNAL) {
1565 		hint &= ~NOTE_SIGNAL;
1566 
1567 		if (kn->kn_id == hint)
1568 			kn->kn_data++;
1569 	}
1570 	return (kn->kn_data != 0);
1571 }
1572