xref: /dragonfly/sys/kern/kern_sig.c (revision 416d05d7)
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 1.65 2007/02/18 16:15:23 corecode 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/nlookup.h>
55 #include <sys/pioctl.h>
56 #include <sys/systm.h>
57 #include <sys/acct.h>
58 #include <sys/fcntl.h>
59 #include <sys/lock.h>
60 #include <sys/wait.h>
61 #include <sys/ktrace.h>
62 #include <sys/syslog.h>
63 #include <sys/stat.h>
64 #include <sys/sysent.h>
65 #include <sys/sysctl.h>
66 #include <sys/malloc.h>
67 #include <sys/interrupt.h>
68 #include <sys/unistd.h>
69 #include <sys/kern_syscall.h>
70 #include <sys/vkernel.h>
71 #include <sys/thread2.h>
72 
73 #include <machine/cpu.h>
74 #include <machine/smp.h>
75 
76 static int	coredump(struct proc *);
77 static char	*expand_name(const char *, uid_t, pid_t);
78 static int	dokillpg(int sig, int pgid, int all);
79 static int	sig_ffs(sigset_t *set);
80 static int	sigprop(int sig);
81 #ifdef SMP
82 static void	signotify_remote(void *arg);
83 #endif
84 static int	kern_sigtimedwait(sigset_t set, siginfo_t *info,
85 		    struct timespec *timeout);
86 
87 static int	filt_sigattach(struct knote *kn);
88 static void	filt_sigdetach(struct knote *kn);
89 static int	filt_signal(struct knote *kn, long hint);
90 
91 struct filterops sig_filtops =
92 	{ 0, filt_sigattach, filt_sigdetach, filt_signal };
93 
94 static int	kern_logsigexit = 1;
95 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
96     &kern_logsigexit, 0,
97     "Log processes quitting on abnormal signals to syslog(3)");
98 
99 /*
100  * Can process p, with pcred pc, send the signal sig to process q?
101  */
102 #define CANSIGNAL(q, sig) \
103 	(!p_trespass(curproc->p_ucred, (q)->p_ucred) || \
104 	((sig) == SIGCONT && (q)->p_session == curproc->p_session))
105 
106 /*
107  * Policy -- Can real uid ruid with ucred uc send a signal to process q?
108  */
109 #define CANSIGIO(ruid, uc, q) \
110 	((uc)->cr_uid == 0 || \
111 	    (ruid) == (q)->p_ucred->cr_ruid || \
112 	    (uc)->cr_uid == (q)->p_ucred->cr_ruid || \
113 	    (ruid) == (q)->p_ucred->cr_uid || \
114 	    (uc)->cr_uid == (q)->p_ucred->cr_uid)
115 
116 int sugid_coredump;
117 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RW,
118 	&sugid_coredump, 0, "Enable coredumping set user/group ID processes");
119 
120 static int	do_coredump = 1;
121 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
122 	&do_coredump, 0, "Enable/Disable coredumps");
123 
124 /*
125  * Signal properties and actions.
126  * The array below categorizes the signals and their default actions
127  * according to the following properties:
128  */
129 #define	SA_KILL		0x01		/* terminates process by default */
130 #define	SA_CORE		0x02		/* ditto and coredumps */
131 #define	SA_STOP		0x04		/* suspend process */
132 #define	SA_TTYSTOP	0x08		/* ditto, from tty */
133 #define	SA_IGNORE	0x10		/* ignore by default */
134 #define	SA_CONT		0x20		/* continue if suspended */
135 #define	SA_CANTMASK	0x40		/* non-maskable, catchable */
136 #define SA_CKPT         0x80            /* checkpoint process */
137 
138 
139 static int sigproptbl[NSIG] = {
140         SA_KILL,                /* SIGHUP */
141         SA_KILL,                /* SIGINT */
142         SA_KILL|SA_CORE,        /* SIGQUIT */
143         SA_KILL|SA_CORE,        /* SIGILL */
144         SA_KILL|SA_CORE,        /* SIGTRAP */
145         SA_KILL|SA_CORE,        /* SIGABRT */
146         SA_KILL|SA_CORE,        /* SIGEMT */
147         SA_KILL|SA_CORE,        /* SIGFPE */
148         SA_KILL,                /* SIGKILL */
149         SA_KILL|SA_CORE,        /* SIGBUS */
150         SA_KILL|SA_CORE,        /* SIGSEGV */
151         SA_KILL|SA_CORE,        /* SIGSYS */
152         SA_KILL,                /* SIGPIPE */
153         SA_KILL,                /* SIGALRM */
154         SA_KILL,                /* SIGTERM */
155         SA_IGNORE,              /* SIGURG */
156         SA_STOP,                /* SIGSTOP */
157         SA_STOP|SA_TTYSTOP,     /* SIGTSTP */
158         SA_IGNORE|SA_CONT,      /* SIGCONT */
159         SA_IGNORE,              /* SIGCHLD */
160         SA_STOP|SA_TTYSTOP,     /* SIGTTIN */
161         SA_STOP|SA_TTYSTOP,     /* SIGTTOU */
162         SA_IGNORE,              /* SIGIO */
163         SA_KILL,                /* SIGXCPU */
164         SA_KILL,                /* SIGXFSZ */
165         SA_KILL,                /* SIGVTALRM */
166         SA_KILL,                /* SIGPROF */
167         SA_IGNORE,              /* SIGWINCH  */
168         SA_IGNORE,              /* SIGINFO */
169         SA_KILL,                /* SIGUSR1 */
170         SA_KILL,                /* SIGUSR2 */
171 	SA_IGNORE,              /* SIGTHR */
172 	SA_CKPT,                /* SIGCKPT */
173 	SA_KILL|SA_CKPT,        /* SIGCKPTEXIT */
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 	SA_IGNORE,
202 	SA_IGNORE,
203 	SA_IGNORE,
204 
205 };
206 
207 static __inline int
208 sigprop(int sig)
209 {
210 
211 	if (sig > 0 && sig < NSIG)
212 		return (sigproptbl[_SIG_IDX(sig)]);
213 	return (0);
214 }
215 
216 static __inline int
217 sig_ffs(sigset_t *set)
218 {
219 	int i;
220 
221 	for (i = 0; i < _SIG_WORDS; i++)
222 		if (set->__bits[i])
223 			return (ffs(set->__bits[i]) + (i * 32));
224 	return (0);
225 }
226 
227 int
228 kern_sigaction(int sig, struct sigaction *act, struct sigaction *oact)
229 {
230 	struct thread *td = curthread;
231 	struct proc *p = td->td_proc;
232 	struct sigacts *ps = p->p_sigacts;
233 
234 	if (sig <= 0 || sig > _SIG_MAXSIG)
235 		return (EINVAL);
236 
237 	if (oact) {
238 		oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
239 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
240 		oact->sa_flags = 0;
241 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
242 			oact->sa_flags |= SA_ONSTACK;
243 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
244 			oact->sa_flags |= SA_RESTART;
245 		if (SIGISMEMBER(ps->ps_sigreset, sig))
246 			oact->sa_flags |= SA_RESETHAND;
247 		if (SIGISMEMBER(ps->ps_signodefer, sig))
248 			oact->sa_flags |= SA_NODEFER;
249 		if (SIGISMEMBER(ps->ps_siginfo, sig))
250 			oact->sa_flags |= SA_SIGINFO;
251 		if (SIGISMEMBER(ps->ps_sigmailbox, sig))
252 			oact->sa_flags |= SA_MAILBOX;
253 		if (sig == SIGCHLD && p->p_procsig->ps_flag & PS_NOCLDSTOP)
254 			oact->sa_flags |= SA_NOCLDSTOP;
255 		if (sig == SIGCHLD && p->p_procsig->ps_flag & PS_NOCLDWAIT)
256 			oact->sa_flags |= SA_NOCLDWAIT;
257 	}
258 	if (act) {
259 		/*
260 		 * Check for invalid requests.  KILL and STOP cannot be
261 		 * caught.
262 		 */
263 		if (sig == SIGKILL || sig == SIGSTOP) {
264 			if (act->sa_handler != SIG_DFL)
265 				return (EINVAL);
266 #if 0
267 			/* (not needed, SIG_DFL forces action to occur) */
268 			if (act->sa_flags & SA_MAILBOX)
269 				return (EINVAL);
270 #endif
271 		}
272 
273 		/*
274 		 * Change setting atomically.
275 		 */
276 		crit_enter();
277 
278 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
279 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
280 		if (act->sa_flags & SA_SIGINFO) {
281 			ps->ps_sigact[_SIG_IDX(sig)] =
282 			    (__sighandler_t *)act->sa_sigaction;
283 			SIGADDSET(ps->ps_siginfo, sig);
284 		} else {
285 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
286 			SIGDELSET(ps->ps_siginfo, sig);
287 		}
288 		if (!(act->sa_flags & SA_RESTART))
289 			SIGADDSET(ps->ps_sigintr, sig);
290 		else
291 			SIGDELSET(ps->ps_sigintr, sig);
292 		if (act->sa_flags & SA_ONSTACK)
293 			SIGADDSET(ps->ps_sigonstack, sig);
294 		else
295 			SIGDELSET(ps->ps_sigonstack, sig);
296 		if (act->sa_flags & SA_RESETHAND)
297 			SIGADDSET(ps->ps_sigreset, sig);
298 		else
299 			SIGDELSET(ps->ps_sigreset, sig);
300 		if (act->sa_flags & SA_NODEFER)
301 			SIGADDSET(ps->ps_signodefer, sig);
302 		else
303 			SIGDELSET(ps->ps_signodefer, sig);
304 		if (act->sa_flags & SA_MAILBOX)
305 			SIGADDSET(ps->ps_sigmailbox, sig);
306 		else
307 			SIGDELSET(ps->ps_sigmailbox, sig);
308 		if (sig == SIGCHLD) {
309 			if (act->sa_flags & SA_NOCLDSTOP)
310 				p->p_procsig->ps_flag |= PS_NOCLDSTOP;
311 			else
312 				p->p_procsig->ps_flag &= ~PS_NOCLDSTOP;
313 			if (act->sa_flags & SA_NOCLDWAIT) {
314 				/*
315 				 * Paranoia: since SA_NOCLDWAIT is implemented
316 				 * by reparenting the dying child to PID 1 (and
317 				 * trust it to reap the zombie), PID 1 itself
318 				 * is forbidden to set SA_NOCLDWAIT.
319 				 */
320 				if (p->p_pid == 1)
321 					p->p_procsig->ps_flag &= ~PS_NOCLDWAIT;
322 				else
323 					p->p_procsig->ps_flag |= PS_NOCLDWAIT;
324 			} else {
325 				p->p_procsig->ps_flag &= ~PS_NOCLDWAIT;
326 			}
327 		}
328 		/*
329 		 * Set bit in p_sigignore for signals that are set to SIG_IGN,
330 		 * and for signals set to SIG_DFL where the default is to
331 		 * ignore. However, don't put SIGCONT in p_sigignore, as we
332 		 * have to restart the process.
333 		 */
334 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
335 		    (sigprop(sig) & SA_IGNORE &&
336 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
337 			/* never to be seen again */
338 			SIGDELSET(p->p_siglist, sig);
339 			if (sig != SIGCONT)
340 				/* easier in ksignal */
341 				SIGADDSET(p->p_sigignore, sig);
342 			SIGDELSET(p->p_sigcatch, sig);
343 		} else {
344 			SIGDELSET(p->p_sigignore, sig);
345 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
346 				SIGDELSET(p->p_sigcatch, sig);
347 			else
348 				SIGADDSET(p->p_sigcatch, sig);
349 		}
350 
351 		crit_exit();
352 	}
353 	return (0);
354 }
355 
356 int
357 sys_sigaction(struct sigaction_args *uap)
358 {
359 	struct sigaction act, oact;
360 	struct sigaction *actp, *oactp;
361 	int error;
362 
363 	actp = (uap->act != NULL) ? &act : NULL;
364 	oactp = (uap->oact != NULL) ? &oact : NULL;
365 	if (actp) {
366 		error = copyin(uap->act, actp, sizeof(act));
367 		if (error)
368 			return (error);
369 	}
370 	error = kern_sigaction(uap->sig, actp, oactp);
371 	if (oactp && !error) {
372 		error = copyout(oactp, uap->oact, sizeof(oact));
373 	}
374 	return (error);
375 }
376 
377 /*
378  * Initialize signal state for process 0;
379  * set to ignore signals that are ignored by default.
380  */
381 void
382 siginit(struct proc *p)
383 {
384 	int i;
385 
386 	for (i = 1; i <= NSIG; i++)
387 		if (sigprop(i) & SA_IGNORE && i != SIGCONT)
388 			SIGADDSET(p->p_sigignore, i);
389 }
390 
391 /*
392  * Reset signals for an exec of the specified process.
393  */
394 void
395 execsigs(struct proc *p)
396 {
397 	struct sigacts *ps = p->p_sigacts;
398 	struct lwp *lp;
399 	int sig;
400 
401 	/*
402 	 * Reset caught signals.  Held signals remain held
403 	 * through p_sigmask (unless they were caught,
404 	 * and are now ignored by default).
405 	 */
406 	while (SIGNOTEMPTY(p->p_sigcatch)) {
407 		sig = sig_ffs(&p->p_sigcatch);
408 		SIGDELSET(p->p_sigcatch, sig);
409 		if (sigprop(sig) & SA_IGNORE) {
410 			if (sig != SIGCONT)
411 				SIGADDSET(p->p_sigignore, sig);
412 			SIGDELSET(p->p_siglist, sig);
413 		}
414 		ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
415 	}
416 	/*
417 	 * Reset stack state to the user stack.
418 	 * Clear set of signals caught on the signal stack.
419 	 */
420 	lp = ONLY_LWP_IN_PROC(p);
421 	lp->lwp_sigstk.ss_flags = SS_DISABLE;
422 	lp->lwp_sigstk.ss_size = 0;
423 	lp->lwp_sigstk.ss_sp = 0;
424 	lp->lwp_flag &= ~LWP_ALTSTACK;
425 	/*
426 	 * Reset no zombies if child dies flag as Solaris does.
427 	 */
428 	p->p_procsig->ps_flag &= ~PS_NOCLDWAIT;
429 }
430 
431 /*
432  * kern_sigprocmask() - MP SAFE ONLY IF p == curproc
433  *
434  *	Manipulate signal mask.  This routine is MP SAFE *ONLY* if
435  *	p == curproc.
436  */
437 int
438 kern_sigprocmask(int how, sigset_t *set, sigset_t *oset)
439 {
440 	struct thread *td = curthread;
441 	struct lwp *lp = td->td_lwp;
442 	int error;
443 
444 	if (oset != NULL)
445 		*oset = lp->lwp_sigmask;
446 
447 	error = 0;
448 	if (set != NULL) {
449 		switch (how) {
450 		case SIG_BLOCK:
451 			SIG_CANTMASK(*set);
452 			SIGSETOR(lp->lwp_sigmask, *set);
453 			break;
454 		case SIG_UNBLOCK:
455 			SIGSETNAND(lp->lwp_sigmask, *set);
456 			break;
457 		case SIG_SETMASK:
458 			SIG_CANTMASK(*set);
459 			lp->lwp_sigmask = *set;
460 			break;
461 		default:
462 			error = EINVAL;
463 			break;
464 		}
465 	}
466 	return (error);
467 }
468 
469 /*
470  * sigprocmask() - MP SAFE
471  */
472 int
473 sys_sigprocmask(struct sigprocmask_args *uap)
474 {
475 	sigset_t set, oset;
476 	sigset_t *setp, *osetp;
477 	int error;
478 
479 	setp = (uap->set != NULL) ? &set : NULL;
480 	osetp = (uap->oset != NULL) ? &oset : NULL;
481 	if (setp) {
482 		error = copyin(uap->set, setp, sizeof(set));
483 		if (error)
484 			return (error);
485 	}
486 	error = kern_sigprocmask(uap->how, setp, osetp);
487 	if (osetp && !error) {
488 		error = copyout(osetp, uap->oset, sizeof(oset));
489 	}
490 	return (error);
491 }
492 
493 int
494 kern_sigpending(struct __sigset *set)
495 {
496 	struct thread *td = curthread;
497 	struct proc *p = td->td_proc;
498 
499 	*set = p->p_siglist;
500 
501 	return (0);
502 }
503 
504 int
505 sys_sigpending(struct sigpending_args *uap)
506 {
507 	sigset_t set;
508 	int error;
509 
510 	error = kern_sigpending(&set);
511 
512 	if (error == 0)
513 		error = copyout(&set, uap->set, sizeof(set));
514 	return (error);
515 }
516 
517 /*
518  * Suspend process until signal, providing mask to be set
519  * in the meantime.
520  */
521 int
522 kern_sigsuspend(struct __sigset *set)
523 {
524 	struct thread *td = curthread;
525 	struct lwp *lp = td->td_lwp;
526 	struct proc *p = td->td_proc;
527 	struct sigacts *ps = p->p_sigacts;
528 
529 	/*
530 	 * When returning from sigsuspend, we want
531 	 * the old mask to be restored after the
532 	 * signal handler has finished.  Thus, we
533 	 * save it here and mark the sigacts structure
534 	 * to indicate this.
535 	 */
536 	lp->lwp_oldsigmask = lp->lwp_sigmask;
537 	lp->lwp_flag |= LWP_OLDMASK;
538 
539 	SIG_CANTMASK(*set);
540 	lp->lwp_sigmask = *set;
541 	while (tsleep(ps, PCATCH, "pause", 0) == 0)
542 		/* void */;
543 	/* always return EINTR rather than ERESTART... */
544 	return (EINTR);
545 }
546 
547 /*
548  * Note nonstandard calling convention: libc stub passes mask, not
549  * pointer, to save a copyin.
550  */
551 int
552 sys_sigsuspend(struct sigsuspend_args *uap)
553 {
554 	sigset_t mask;
555 	int error;
556 
557 	error = copyin(uap->sigmask, &mask, sizeof(mask));
558 	if (error)
559 		return (error);
560 
561 	error = kern_sigsuspend(&mask);
562 
563 	return (error);
564 }
565 
566 int
567 kern_sigaltstack(struct sigaltstack *ss, struct sigaltstack *oss)
568 {
569 	struct thread *td = curthread;
570 	struct lwp *lp = td->td_lwp;
571 	struct proc *p = td->td_proc;
572 
573 	if ((lp->lwp_flag & LWP_ALTSTACK) == 0)
574 		lp->lwp_sigstk.ss_flags |= SS_DISABLE;
575 
576 	if (oss)
577 		*oss = lp->lwp_sigstk;
578 
579 	if (ss) {
580 		if (ss->ss_flags & SS_DISABLE) {
581 			if (lp->lwp_sigstk.ss_flags & SS_ONSTACK)
582 				return (EINVAL);
583 			lp->lwp_flag &= ~LWP_ALTSTACK;
584 			lp->lwp_sigstk.ss_flags = ss->ss_flags;
585 		} else {
586 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
587 				return (ENOMEM);
588 			lp->lwp_flag |= LWP_ALTSTACK;
589 			lp->lwp_sigstk = *ss;
590 		}
591 	}
592 
593 	return (0);
594 }
595 
596 int
597 sys_sigaltstack(struct sigaltstack_args *uap)
598 {
599 	stack_t ss, oss;
600 	int error;
601 
602 	if (uap->ss) {
603 		error = copyin(uap->ss, &ss, sizeof(ss));
604 		if (error)
605 			return (error);
606 	}
607 
608 	error = kern_sigaltstack(uap->ss ? &ss : NULL,
609 	    uap->oss ? &oss : NULL);
610 
611 	if (error == 0 && uap->oss)
612 		error = copyout(&oss, uap->oss, sizeof(*uap->oss));
613 	return (error);
614 }
615 
616 /*
617  * Common code for kill process group/broadcast kill.
618  * cp is calling process.
619  */
620 struct killpg_info {
621 	int nfound;
622 	int sig;
623 };
624 
625 static int killpg_all_callback(struct proc *p, void *data);
626 
627 static int
628 dokillpg(int sig, int pgid, int all)
629 {
630 	struct killpg_info info;
631 	struct proc *cp = curproc;
632 	struct proc *p;
633 	struct pgrp *pgrp;
634 
635 	info.nfound = 0;
636 	info.sig = sig;
637 
638 	if (all) {
639 		/*
640 		 * broadcast
641 		 */
642 		allproc_scan(killpg_all_callback, &info);
643 	} else {
644 		if (pgid == 0) {
645 			/*
646 			 * zero pgid means send to my process group.
647 			 */
648 			pgrp = cp->p_pgrp;
649 		} else {
650 			pgrp = pgfind(pgid);
651 			if (pgrp == NULL)
652 				return (ESRCH);
653 		}
654 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
655 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
656 			if (p->p_pid <= 1 ||
657 			    p->p_stat == SZOMB ||
658 			    (p->p_flag & P_SYSTEM) ||
659 			    !CANSIGNAL(p, sig)) {
660 				continue;
661 			}
662 			++info.nfound;
663 			if (sig)
664 				ksignal(p, sig);
665 		}
666 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
667 	}
668 	return (info.nfound ? 0 : ESRCH);
669 }
670 
671 static int
672 killpg_all_callback(struct proc *p, void *data)
673 {
674 	struct killpg_info *info = data;
675 
676 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) ||
677 	    p == curproc || !CANSIGNAL(p, info->sig)) {
678 		return (0);
679 	}
680 	++info->nfound;
681 	if (info->sig)
682 		ksignal(p, info->sig);
683 	return(0);
684 }
685 
686 int
687 kern_kill(int sig, int pid)
688 {
689 	struct thread *td = curthread;
690 	struct proc *p = td->td_proc;
691 
692 	if ((u_int)sig > _SIG_MAXSIG)
693 		return (EINVAL);
694 	if (pid > 0) {
695 		/* kill single process */
696 		if ((p = pfind(pid)) == NULL)
697 			return (ESRCH);
698 		if (!CANSIGNAL(p, sig))
699 			return (EPERM);
700 		if (sig)
701 			ksignal(p, sig);
702 		return (0);
703 	}
704 	switch (pid) {
705 	case -1:		/* broadcast signal */
706 		return (dokillpg(sig, 0, 1));
707 	case 0:			/* signal own process group */
708 		return (dokillpg(sig, 0, 0));
709 	default:		/* negative explicit process group */
710 		return (dokillpg(sig, -pid, 0));
711 	}
712 	/* NOTREACHED */
713 }
714 
715 int
716 sys_kill(struct kill_args *uap)
717 {
718 	int error;
719 
720 	error = kern_kill(uap->signum, uap->pid);
721 
722 	return (error);
723 }
724 
725 /*
726  * Send a signal to a process group.
727  */
728 void
729 gsignal(int pgid, int sig)
730 {
731 	struct pgrp *pgrp;
732 
733 	if (pgid && (pgrp = pgfind(pgid)))
734 		pgsignal(pgrp, sig, 0);
735 }
736 
737 /*
738  * Send a signal to a process group.  If checktty is 1,
739  * limit to members which have a controlling terminal.
740  *
741  * pg_lock interlocks against a fork that might be in progress, to
742  * ensure that the new child process picks up the signal.
743  */
744 void
745 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
746 {
747 	struct proc *p;
748 
749 	if (pgrp) {
750 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
751 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
752 			if (checkctty == 0 || p->p_flag & P_CONTROLT)
753 				ksignal(p, sig);
754 		}
755 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
756 	}
757 }
758 
759 /*
760  * Send a signal caused by a trap to the current process.
761  * If it will be caught immediately, deliver it with correct code.
762  * Otherwise, post it normally.
763  */
764 void
765 trapsignal(struct lwp *lp, int sig, u_long code)
766 {
767 	struct proc *p = lp->lwp_proc;
768 	struct sigacts *ps = p->p_sigacts;
769 
770 	/*
771 	 * If we are a virtual kernel running an emulated user process
772 	 * context, switch back to the virtual kernel context before
773 	 * trying to post the signal.
774 	 */
775 	if (p->p_vkernel && p->p_vkernel->vk_current) {
776 		struct trapframe *tf = curthread->td_lwp->lwp_md.md_regs;
777 		tf->tf_trapno = 0;
778 		vkernel_trap(p, tf);
779 	}
780 
781 
782 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(p->p_sigcatch, sig) &&
783 	    !SIGISMEMBER(lp->lwp_sigmask, sig)) {
784 		lp->lwp_ru.ru_nsignals++;
785 #ifdef KTRACE
786 		if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
787 			ktrpsig(p, sig, ps->ps_sigact[_SIG_IDX(sig)],
788 				&lp->lwp_sigmask, code);
789 #endif
790 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], sig,
791 						&lp->lwp_sigmask, code);
792 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
793 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
794 			SIGADDSET(lp->lwp_sigmask, sig);
795 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
796 			/*
797 			 * See kern_sigaction() for origin of this code.
798 			 */
799 			SIGDELSET(p->p_sigcatch, sig);
800 			if (sig != SIGCONT &&
801 			    sigprop(sig) & SA_IGNORE)
802 				SIGADDSET(p->p_sigignore, sig);
803 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
804 		}
805 	} else {
806 		p->p_code = code;	/* XXX for core dump/debugger */
807 		p->p_sig = sig;		/* XXX to verify code */
808 		ksignal(p, sig);
809 	}
810 }
811 
812 /*
813  * Send the signal to the process.  If the signal has an action, the action
814  * is usually performed by the target process rather than the caller; we add
815  * the signal to the set of pending signals for the process.
816  *
817  * Exceptions:
818  *   o When a stop signal is sent to a sleeping process that takes the
819  *     default action, the process is stopped without awakening it.
820  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
821  *     regardless of the signal action (eg, blocked or ignored).
822  *
823  * Other ignored signals are discarded immediately.
824  */
825 void
826 ksignal(struct proc *p, int sig)
827 {
828 	/* XXX lwp more intelligent lwp choice needed */
829 	struct lwp *lp = FIRST_LWP_IN_PROC(p);
830 	int prop;
831 	sig_t action;
832 
833 	if (sig > _SIG_MAXSIG || sig <= 0) {
834 		kprintf("ksignal: signal %d\n", sig);
835 		panic("ksignal signal number");
836 	}
837 
838 	crit_enter();
839 	KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
840 	crit_exit();
841 
842 	prop = sigprop(sig);
843 
844 	/*
845 	 * If proc is traced, always give parent a chance;
846 	 * if signal event is tracked by procfs, give *that*
847 	 * a chance, as well.
848 	 */
849 	if ((p->p_flag & P_TRACED) || (p->p_stops & S_SIG)) {
850 		action = SIG_DFL;
851 	} else {
852 		/*
853 		 * If the signal is being ignored,
854 		 * then we forget about it immediately.
855 		 * (Note: we don't set SIGCONT in p_sigignore,
856 		 * and if it is set to SIG_IGN,
857 		 * action will be SIG_DFL here.)
858 		 */
859 		if (SIGISMEMBER(p->p_sigignore, sig) || (p->p_flag & P_WEXIT))
860 			return;
861 		if (SIGISMEMBER(lp->lwp_sigmask, sig))
862 			action = SIG_HOLD;
863 		else if (SIGISMEMBER(p->p_sigcatch, sig))
864 			action = SIG_CATCH;
865 		else
866 			action = SIG_DFL;
867 	}
868 
869 	if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
870 	    (p->p_flag & P_TRACED) == 0) {
871 		p->p_nice = NZERO;
872 	}
873 
874 	/*
875 	 * If continuing, clear any pending STOP signals.
876 	 */
877 	if (prop & SA_CONT)
878 		SIG_STOPSIGMASK(p->p_siglist);
879 
880 	if (prop & SA_STOP) {
881 		/*
882 		 * If sending a tty stop signal to a member of an orphaned
883 		 * process group, discard the signal here if the action
884 		 * is default; don't stop the process below if sleeping,
885 		 * and don't clear any pending SIGCONT.
886 		 */
887 		if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0 &&
888 		    action == SIG_DFL) {
889 		        return;
890 		}
891 		SIG_CONTSIGMASK(p->p_siglist);
892 	}
893 	SIGADDSET(p->p_siglist, sig);
894 
895 	/*
896 	 * Defer further processing for signals which are held,
897 	 * except that stopped processes must be continued by SIGCONT.
898 	 */
899 	if (action == SIG_HOLD) {
900 		if ((prop & SA_CONT) == 0 || p->p_stat != SSTOP)
901 			return;
902 	}
903 
904 	crit_enter();
905 
906 
907 	/* XXX lwp handle stop/continue */
908 
909 	/*
910 	 * LWP is in tsleep and not stopped
911 	 */
912 	if (lp->lwp_stat == LSSLEEP && p->p_stat != SSTOP) {
913 		/*
914 		 * If the process is sleeping uninterruptibly
915 		 * we can't interrupt the sleep... the signal will
916 		 * be noticed when the process returns through
917 		 * trap() or syscall().
918 		 */
919 		if ((lp->lwp_flag & LWP_SINTR) == 0)
920 			goto out;
921 
922 		/*
923 		 * If the process is sleeping and traced, make it runnable
924 		 * so it can discover the signal in issignal() and stop
925 		 * for the parent.
926 		 *
927 		 * If the process is stopped and traced, no further action
928 		 * is necessary.
929 		 */
930 		if (p->p_flag & P_TRACED)
931 			goto run;
932 
933 		/*
934 		 * If the process is sleeping and SA_CONT, and the signal
935 		 * mode is SIG_DFL, then make the process runnable.
936 		 *
937 		 * However, do *NOT* set LWP_BREAKTSLEEP.  We do not want
938 		 * a SIGCONT to terminate an interruptable tsleep early
939 		 * and generate a spurious EINTR.
940 		 */
941 		if ((prop & SA_CONT) && action == SIG_DFL) {
942 			SIGDELSET(p->p_siglist, sig);
943 			goto run_no_break;
944 		}
945 
946 		/*
947 		 * If the process is sleeping and receives a STOP signal,
948 		 * process immediately if possible.  All other (caught or
949 		 * default) signals cause the process to run.
950 		 */
951 		if (prop & SA_STOP) {
952 			if (action != SIG_DFL)
953 				goto run;
954 
955 			/*
956 			 * If a child holding parent blocked, stopping
957 			 * could cause deadlock.  Take no action at this
958 			 * time.
959 			 */
960 			if (p->p_flag & P_PPWAIT)
961 				goto out;
962 
963 			/*
964 			 * Do not actually try to manipulate the process while
965 			 * it is sleeping, simply set SSTOP to indicate that
966 			 * lwps should stop as soon as they safely can.
967 			 */
968 			SIGDELSET(p->p_siglist, sig);
969 			p->p_xstat = sig;
970 			proc_stop(p, 1);
971 			goto out;
972 		}
973 
974 		/*
975 		 * Otherwise the signal can interrupt the sleep.
976 		 */
977 		goto run;
978 	}
979 
980 	/*
981 	 * Process is in tsleep and is stopped
982 	 */
983 	if (lp->lwp_stat == LSSLEEP && p->p_stat == SSTOP) {
984 		/*
985 		 * If the process is stopped and is being traced, then no
986 		 * further action is necessary.
987 		 */
988 		if (p->p_flag & P_TRACED)
989 			goto out;
990 
991 		/*
992 		 * If the process is stopped and receives a KILL signal,
993 		 * make the process runnable.
994 		 */
995 		if (sig == SIGKILL) {
996 			proc_unstop(p);
997 			goto out;
998 		}
999 
1000 		/*
1001 		 * If the process is stopped and receives a CONT signal,
1002 		 * then try to make the process runnable again.
1003 		 */
1004 		if (prop & SA_CONT) {
1005 			/*
1006 			 * If SIGCONT is default (or ignored), we continue the
1007 			 * process but don't leave the signal in p_siglist, as
1008 			 * it has no further action.  If SIGCONT is held, we
1009 			 * continue the process and leave the signal in
1010 			 * p_siglist.  If the process catches SIGCONT, let it
1011 			 * handle the signal itself.
1012 			 */
1013 			if (action == SIG_DFL)
1014 				SIGDELSET(p->p_siglist, sig);
1015 			proc_unstop(p);
1016 			if (action == SIG_CATCH)
1017 				goto run;
1018 			goto out;
1019 		}
1020 
1021 		/*
1022 		 * If the process is stopped and receives another STOP
1023 		 * signal, we do not need to stop it again.  If we did
1024 		 * the shell could get confused.
1025 		 */
1026 		if (prop & SA_STOP) {
1027 			SIGDELSET(p->p_siglist, sig);
1028 			goto out;
1029 		}
1030 
1031 		/*
1032 		 * Otherwise the process is sleeping interruptably but
1033 		 * is stopped, just set the LWP_BREAKTSLEEP flag and take
1034 		 * no further action.  The next runnable action will wake
1035 		 * the process up.
1036 		 */
1037 		lp->lwp_flag |= LWP_BREAKTSLEEP;
1038 		goto out;
1039 	}
1040 
1041 	/*
1042 	 * Otherwise the process is running
1043 	 *
1044 	 * SRUN, SIDL, SZOMB do nothing with the signal,
1045 	 * other than kicking ourselves if we are running.
1046 	 * It will either never be noticed, or noticed very soon.
1047 	 *
1048 	 * Note that p_thread may be NULL or may not be completely
1049 	 * initialized if the process is in the SIDL or SZOMB state.
1050 	 *
1051 	 * For SMP we may have to forward the request to another cpu.
1052 	 * YYY the MP lock prevents the target process from moving
1053 	 * to another cpu, see kern/kern_switch.c
1054 	 *
1055 	 * If the target thread is waiting on its message port,
1056 	 * wakeup the target thread so it can check (or ignore)
1057 	 * the new signal.  YYY needs cleanup.
1058 	 */
1059 	if (lp == lwkt_preempted_proc()) {
1060 		signotify();
1061 	} else if (lp->lwp_stat == LSRUN) {
1062 		struct thread *td = lp->lwp_thread;
1063 
1064 		KASSERT(td != NULL,
1065 		    ("pid %d/%d NULL lwp_thread stat %d flags %08x/%08x",
1066 		    p->p_pid, lp->lwp_tid, lp->lwp_stat, p->p_flag, lp->lwp_flag));
1067 
1068 #ifdef SMP
1069 		if (td->td_gd != mycpu)
1070 			lwkt_send_ipiq(td->td_gd, signotify_remote, lp);
1071 		else
1072 #endif
1073 		if (td->td_msgport.mp_flags & MSGPORTF_WAITING)
1074 			lwkt_schedule(td);
1075 	}
1076 	goto out;
1077 	/*NOTREACHED*/
1078 run:
1079 	/*
1080 	 * Make runnable and break out of any tsleep as well.
1081 	 */
1082 	lp->lwp_flag |= LWP_BREAKTSLEEP;
1083 run_no_break:
1084 	setrunnable(lp);
1085 out:
1086 	crit_exit();
1087 }
1088 
1089 #ifdef SMP
1090 
1091 /*
1092  * This function is called via an IPI.  We will be in a critical section but
1093  * the MP lock will NOT be held.  Also note that by the time the ipi message
1094  * gets to us the process 'p' (arg) may no longer be scheduled or even valid.
1095  */
1096 static void
1097 signotify_remote(void *arg)
1098 {
1099 	struct lwp *lp = arg;
1100 
1101 	if (lp == lwkt_preempted_proc()) {
1102 		signotify();
1103 	} else {
1104 		struct thread *td = lp->lwp_thread;
1105 		if (td->td_msgport.mp_flags & MSGPORTF_WAITING)
1106 			lwkt_schedule(td);
1107 	}
1108 }
1109 
1110 #endif
1111 
1112 void
1113 proc_stop(struct proc *p, int notify)
1114 {
1115 	/* XXX lwp */
1116 	p->p_stat = SSTOP;
1117 	p->p_flag &= ~P_WAITED;
1118 	wakeup(p->p_pptr);
1119 	if (notify > 1 ||
1120 	    notify && (p->p_pptr->p_procsig->ps_flag & PS_NOCLDSTOP) == 0)
1121 		ksignal(p->p_pptr, SIGCHLD);
1122 }
1123 
1124 void
1125 proc_unstop(struct proc *p)
1126 {
1127 	struct lwp *lp = FIRST_LWP_IN_PROC(p);	/* XXX lwp */
1128 
1129 	p->p_stat = SACTIVE;
1130 	setrunnable(lp);
1131 }
1132 
1133 static int
1134 kern_sigtimedwait(sigset_t waitset, siginfo_t *info, struct timespec *timeout)
1135 {
1136 	sigset_t savedmask, set;
1137 	struct proc *p = curproc;
1138 	struct lwp *lp = curthread->td_lwp;
1139 	int error, sig, hz, timevalid = 0;
1140 	struct timespec rts, ets, ts;
1141 	struct timeval tv;
1142 
1143 	error = 0;
1144 	sig = 0;
1145 	SIG_CANTMASK(waitset);
1146 	savedmask = lp->lwp_sigmask;
1147 
1148 	if (timeout) {
1149 		if (timeout->tv_sec >= 0 && timeout->tv_nsec >= 0 &&
1150 		    timeout->tv_nsec < 1000000000) {
1151 			timevalid = 1;
1152 			getnanouptime(&rts);
1153 		 	ets = rts;
1154 			timespecadd(&ets, timeout);
1155 		}
1156 	}
1157 
1158 	for (;;) {
1159 		set = p->p_siglist;
1160 		SIGSETAND(set, waitset);
1161 		if ((sig = sig_ffs(&set)) != 0) {
1162 			SIGFILLSET(lp->lwp_sigmask);
1163 			SIGDELSET(lp->lwp_sigmask, sig);
1164 			SIG_CANTMASK(lp->lwp_sigmask);
1165 			sig = issignal(lp);
1166 			/*
1167 			 * It may be a STOP signal, in the case, issignal
1168 			 * returns 0, because we may stop there, and new
1169 			 * signal can come in, we should restart if we got
1170 			 * nothing.
1171 			 */
1172 			if (sig == 0)
1173 				continue;
1174 			else
1175 				break;
1176 		}
1177 
1178 		/*
1179 		 * Previous checking got nothing, and we retried but still
1180 		 * got nothing, we should return the error status.
1181 		 */
1182 		if (error)
1183 			break;
1184 
1185 		/*
1186 		 * POSIX says this must be checked after looking for pending
1187 		 * signals.
1188 		 */
1189 		if (timeout) {
1190 			if (!timevalid) {
1191 				error = EINVAL;
1192 				break;
1193 			}
1194 			getnanouptime(&rts);
1195 			if (timespeccmp(&rts, &ets, >=)) {
1196 				error = EAGAIN;
1197 				break;
1198 			}
1199 			ts = ets;
1200 			timespecsub(&ts, &rts);
1201 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1202 			hz = tvtohz_high(&tv);
1203 		} else
1204 			hz = 0;
1205 
1206 		lp->lwp_sigmask = savedmask;
1207 		SIGSETNAND(lp->lwp_sigmask, waitset);
1208 		error = tsleep(&p->p_sigacts, PCATCH, "sigwt", hz);
1209 		if (timeout) {
1210 			if (error == ERESTART) {
1211 				/* can not restart a timeout wait. */
1212 				error = EINTR;
1213 			} else if (error == EAGAIN) {
1214 				/* will calculate timeout by ourself. */
1215 				error = 0;
1216 			}
1217 		}
1218 		/* Retry ... */
1219 	}
1220 
1221 	lp->lwp_sigmask = savedmask;
1222 	if (sig) {
1223 		error = 0;
1224 		bzero(info, sizeof(*info));
1225 		info->si_signo = sig;
1226 		SIGDELSET(p->p_siglist, sig);	/* take the signal! */
1227 
1228 		if (sig == SIGKILL)
1229 			sigexit(p, sig);
1230 	}
1231 	return (error);
1232 }
1233 
1234 int
1235 sys_sigtimedwait(struct sigtimedwait_args *uap)
1236 {
1237 	struct timespec ts;
1238 	struct timespec *timeout;
1239 	sigset_t set;
1240 	siginfo_t info;
1241 	int error;
1242 
1243 	if (uap->timeout) {
1244 		error = copyin(uap->timeout, &ts, sizeof(ts));
1245 		if (error)
1246 			return (error);
1247 		timeout = &ts;
1248 	} else {
1249 		timeout = NULL;
1250 	}
1251 	error = copyin(uap->set, &set, sizeof(set));
1252 	if (error)
1253 		return (error);
1254 	error = kern_sigtimedwait(set, &info, timeout);
1255 	if (error)
1256 		return (error);
1257  	if (uap->info)
1258 		error = copyout(&info, uap->info, sizeof(info));
1259 	/* Repost if we got an error. */
1260 	if (error)
1261 		ksignal(curproc, info.si_signo);
1262 	else
1263 		uap->sysmsg_result = info.si_signo;
1264 	return (error);
1265 }
1266 
1267 int
1268 sys_sigwaitinfo(struct sigwaitinfo_args *uap)
1269 {
1270 	siginfo_t info;
1271 	sigset_t set;
1272 	int error;
1273 
1274 	error = copyin(uap->set, &set, sizeof(set));
1275 	if (error)
1276 		return (error);
1277 	error = kern_sigtimedwait(set, &info, NULL);
1278 	if (error)
1279 		return (error);
1280 	if (uap->info)
1281 		error = copyout(&info, uap->info, sizeof(info));
1282 	/* Repost if we got an error. */
1283 	if (error)
1284 		ksignal(curproc, info.si_signo);
1285 	else
1286 		uap->sysmsg_result = info.si_signo;
1287 	return (error);
1288 }
1289 
1290 /*
1291  * If the current process has received a signal that would interrupt a
1292  * system call, return EINTR or ERESTART as appropriate.
1293  */
1294 int
1295 iscaught(struct lwp *lp)
1296 {
1297 	struct proc *p = lp->lwp_proc;
1298 	int sig;
1299 
1300 	if (p) {
1301 		if ((sig = CURSIG(lp)) != 0) {
1302 			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1303 				return (EINTR);
1304 			return (ERESTART);
1305 		}
1306 	}
1307 	return(EWOULDBLOCK);
1308 }
1309 
1310 /*
1311  * If the current process has received a signal (should be caught or cause
1312  * termination, should interrupt current syscall), return the signal number.
1313  * Stop signals with default action are processed immediately, then cleared;
1314  * they aren't returned.  This is checked after each entry to the system for
1315  * a syscall or trap (though this can usually be done without calling issignal
1316  * by checking the pending signal masks in the CURSIG macro.) The normal call
1317  * sequence is
1318  *
1319  * This routine is called via CURSIG/__cursig and the MP lock might not be
1320  * held.  Obtain the MP lock for the duration of the operation.
1321  *
1322  *	while (sig = CURSIG(curproc))
1323  *		postsig(sig);
1324  */
1325 int
1326 issignal(struct lwp *lp)
1327 {
1328 	struct proc *p = lp->lwp_proc;
1329 	sigset_t mask;
1330 	int sig, prop;
1331 
1332 	get_mplock();
1333 	for (;;) {
1334 		int traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
1335 
1336 		mask = p->p_siglist;
1337 		SIGSETNAND(mask, lp->lwp_sigmask);
1338 		if (p->p_flag & P_PPWAIT)
1339 			SIG_STOPSIGMASK(mask);
1340 		if (!SIGNOTEMPTY(mask)) { 	/* no signal to send */
1341 			rel_mplock();
1342 			return (0);
1343 		}
1344 		sig = sig_ffs(&mask);
1345 
1346 		STOPEVENT(p, S_SIG, sig);
1347 
1348 		/*
1349 		 * We should see pending but ignored signals
1350 		 * only if P_TRACED was on when they were posted.
1351 		 */
1352 		if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
1353 			SIGDELSET(p->p_siglist, sig);
1354 			continue;
1355 		}
1356 		if ((p->p_flag & P_TRACED) && (p->p_flag & P_PPWAIT) == 0) {
1357 			/*
1358 			 * If traced, always stop, and stay stopped until
1359 			 * released by the parent.
1360 			 *
1361 			 * NOTE: SSTOP may get cleared during the loop,
1362 			 * but we do not re-notify the parent if we have
1363 			 * to loop several times waiting for the parent
1364 			 * to let us continue.
1365 			 *
1366 			 * XXX not sure if this is still true
1367 			 */
1368 			p->p_xstat = sig;
1369 			proc_stop(p, 2);
1370 			do {
1371 				tstop();
1372 			} while (!trace_req(p) && (p->p_flag & P_TRACED));
1373 
1374 			/*
1375 			 * If parent wants us to take the signal,
1376 			 * then it will leave it in p->p_xstat;
1377 			 * otherwise we just look for signals again.
1378 			 */
1379 			SIGDELSET(p->p_siglist, sig);	/* clear old signal */
1380 			sig = p->p_xstat;
1381 			if (sig == 0)
1382 				continue;
1383 
1384 			/*
1385 			 * Put the new signal into p_siglist.  If the
1386 			 * signal is being masked, look for other signals.
1387 			 */
1388 			SIGADDSET(p->p_siglist, sig);
1389 			if (SIGISMEMBER(lp->lwp_sigmask, sig))
1390 				continue;
1391 
1392 			/*
1393 			 * If the traced bit got turned off, go back up
1394 			 * to the top to rescan signals.  This ensures
1395 			 * that p_sig* and ps_sigact are consistent.
1396 			 */
1397 			if ((p->p_flag & P_TRACED) == 0)
1398 				continue;
1399 		}
1400 
1401 		prop = sigprop(sig);
1402 
1403 		/*
1404 		 * Decide whether the signal should be returned.
1405 		 * Return the signal's number, or fall through
1406 		 * to clear it from the pending mask.
1407 		 */
1408 		switch ((int)(intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
1409 		case (int)SIG_DFL:
1410 			/*
1411 			 * Don't take default actions on system processes.
1412 			 */
1413 			if (p->p_pid <= 1) {
1414 #ifdef DIAGNOSTIC
1415 				/*
1416 				 * Are you sure you want to ignore SIGSEGV
1417 				 * in init? XXX
1418 				 */
1419 				kprintf("Process (pid %lu) got signal %d\n",
1420 					(u_long)p->p_pid, sig);
1421 #endif
1422 				break;		/* == ignore */
1423 			}
1424 
1425 			/*
1426 			 * Handle the in-kernel checkpoint action
1427 			 */
1428 			if (prop & SA_CKPT) {
1429 				checkpoint_signal_handler(p);
1430 				break;
1431 			}
1432 
1433 			/*
1434 			 * If there is a pending stop signal to process
1435 			 * with default action, stop here,
1436 			 * then clear the signal.  However,
1437 			 * if process is member of an orphaned
1438 			 * process group, ignore tty stop signals.
1439 			 */
1440 			if (prop & SA_STOP) {
1441 				if (p->p_flag & P_TRACED ||
1442 		    		    (p->p_pgrp->pg_jobc == 0 &&
1443 				    prop & SA_TTYSTOP))
1444 					break;	/* == ignore */
1445 				p->p_xstat = sig;
1446 				proc_stop(p, 1);
1447 				while (p->p_stat == SSTOP) {
1448 					tstop();
1449 				}
1450 				break;
1451 			} else if (prop & SA_IGNORE) {
1452 				/*
1453 				 * Except for SIGCONT, shouldn't get here.
1454 				 * Default action is to ignore; drop it.
1455 				 */
1456 				break;		/* == ignore */
1457 			} else {
1458 				rel_mplock();
1459 				return (sig);
1460 			}
1461 
1462 			/*NOTREACHED*/
1463 
1464 		case (int)SIG_IGN:
1465 			/*
1466 			 * Masking above should prevent us ever trying
1467 			 * to take action on an ignored signal other
1468 			 * than SIGCONT, unless process is traced.
1469 			 */
1470 			if ((prop & SA_CONT) == 0 &&
1471 			    (p->p_flag & P_TRACED) == 0)
1472 				kprintf("issignal\n");
1473 			break;		/* == ignore */
1474 
1475 		default:
1476 			/*
1477 			 * This signal has an action, let
1478 			 * postsig() process it.
1479 			 */
1480 			rel_mplock();
1481 			return (sig);
1482 		}
1483 		SIGDELSET(p->p_siglist, sig);		/* take the signal! */
1484 	}
1485 	/* NOTREACHED */
1486 }
1487 
1488 /*
1489  * Take the action for the specified signal
1490  * from the current set of pending signals.
1491  */
1492 void
1493 postsig(int sig)
1494 {
1495 	struct lwp *lp = curthread->td_lwp;
1496 	struct proc *p = lp->lwp_proc;
1497 	struct sigacts *ps = p->p_sigacts;
1498 	sig_t action;
1499 	sigset_t returnmask;
1500 	int code;
1501 
1502 	KASSERT(sig != 0, ("postsig"));
1503 
1504 	/*
1505 	 * If we are a virtual kernel running an emulated user process
1506 	 * context, switch back to the virtual kernel context before
1507 	 * trying to post the signal.
1508 	 */
1509 	if (p->p_vkernel && p->p_vkernel->vk_current) {
1510 		struct trapframe *tf = curthread->td_lwp->lwp_md.md_regs;
1511 		tf->tf_trapno = 0;
1512 		vkernel_trap(p, tf);
1513 	}
1514 
1515 	SIGDELSET(p->p_siglist, sig);
1516 	action = ps->ps_sigact[_SIG_IDX(sig)];
1517 #ifdef KTRACE
1518 	if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
1519 		ktrpsig(p, sig, action, lp->lwp_flag & LWP_OLDMASK ?
1520 			&lp->lwp_oldsigmask : &lp->lwp_sigmask, 0);
1521 #endif
1522 	STOPEVENT(p, S_SIG, sig);
1523 
1524 	if (action == SIG_DFL) {
1525 		/*
1526 		 * Default action, where the default is to kill
1527 		 * the process.  (Other cases were ignored above.)
1528 		 */
1529 		sigexit(p, sig);
1530 		/* NOTREACHED */
1531 	} else {
1532 		/*
1533 		 * If we get here, the signal must be caught.
1534 		 */
1535 		KASSERT(action != SIG_IGN && !SIGISMEMBER(lp->lwp_sigmask, sig),
1536 		    ("postsig action"));
1537 
1538 		crit_enter();
1539 
1540 		/*
1541 		 * Reset the signal handler if asked to
1542 		 */
1543 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
1544 			/*
1545 			 * See kern_sigaction() for origin of this code.
1546 			 */
1547 			SIGDELSET(p->p_sigcatch, sig);
1548 			if (sig != SIGCONT &&
1549 			    sigprop(sig) & SA_IGNORE)
1550 				SIGADDSET(p->p_sigignore, sig);
1551 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1552 		}
1553 
1554 		/*
1555 		 * Handle the mailbox case.  Copyout to the appropriate
1556 		 * location but do not generate a signal frame.  The system
1557 		 * call simply returns EINTR and the user is responsible for
1558 		 * polling the mailbox.
1559 		 */
1560 		if (SIGISMEMBER(ps->ps_sigmailbox, sig)) {
1561 			int sig_copy = sig;
1562 			copyout(&sig_copy, (void *)action, sizeof(int));
1563 			curproc->p_flag |= P_MAILBOX;
1564 			crit_exit();
1565 			goto done;
1566 		}
1567 
1568 		/*
1569 		 * Set the signal mask and calculate the mask to restore
1570 		 * when the signal function returns.
1571 		 *
1572 		 * Special case: user has done a sigsuspend.  Here the
1573 		 * current mask is not of interest, but rather the
1574 		 * mask from before the sigsuspend is what we want
1575 		 * restored after the signal processing is completed.
1576 		 */
1577 		if (lp->lwp_flag & LWP_OLDMASK) {
1578 			returnmask = lp->lwp_oldsigmask;
1579 			lp->lwp_flag &= ~LWP_OLDMASK;
1580 		} else {
1581 			returnmask = lp->lwp_sigmask;
1582 		}
1583 
1584 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
1585 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
1586 			SIGADDSET(lp->lwp_sigmask, sig);
1587 
1588 		crit_exit();
1589 		lp->lwp_ru.ru_nsignals++;
1590 		if (p->p_sig != sig) {
1591 			code = 0;
1592 		} else {
1593 			code = p->p_code;
1594 			p->p_code = 0;
1595 			p->p_sig = 0;
1596 		}
1597 		(*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
1598 	}
1599 done:
1600 	;
1601 }
1602 
1603 /*
1604  * Kill the current process for stated reason.
1605  */
1606 void
1607 killproc(struct proc *p, char *why)
1608 {
1609 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n",
1610 		p->p_pid, p->p_comm,
1611 		p->p_ucred ? p->p_ucred->cr_uid : -1, why);
1612 	ksignal(p, SIGKILL);
1613 }
1614 
1615 /*
1616  * Force the current process to exit with the specified signal, dumping core
1617  * if appropriate.  We bypass the normal tests for masked and caught signals,
1618  * allowing unrecoverable failures to terminate the process without changing
1619  * signal state.  Mark the accounting record with the signal termination.
1620  * If dumping core, save the signal number for the debugger.  Calls exit and
1621  * does not return.
1622  */
1623 void
1624 sigexit(struct proc *p, int sig)
1625 {
1626 	p->p_acflag |= AXSIG;
1627 	if (sigprop(sig) & SA_CORE) {
1628 		p->p_sig = sig;
1629 		/*
1630 		 * Log signals which would cause core dumps
1631 		 * (Log as LOG_INFO to appease those who don't want
1632 		 * these messages.)
1633 		 * XXX : Todo, as well as euid, write out ruid too
1634 		 */
1635 		if (coredump(p) == 0)
1636 			sig |= WCOREFLAG;
1637 		if (kern_logsigexit)
1638 			log(LOG_INFO,
1639 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
1640 			    p->p_pid, p->p_comm,
1641 			    p->p_ucred ? p->p_ucred->cr_uid : -1,
1642 			    sig &~ WCOREFLAG,
1643 			    sig & WCOREFLAG ? " (core dumped)" : "");
1644 	}
1645 	exit1(W_EXITCODE(0, sig));
1646 	/* NOTREACHED */
1647 }
1648 
1649 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
1650 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
1651 	      sizeof(corefilename), "process corefile name format string");
1652 
1653 /*
1654  * expand_name(name, uid, pid)
1655  * Expand the name described in corefilename, using name, uid, and pid.
1656  * corefilename is a kprintf-like string, with three format specifiers:
1657  *	%N	name of process ("name")
1658  *	%P	process id (pid)
1659  *	%U	user id (uid)
1660  * For example, "%N.core" is the default; they can be disabled completely
1661  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
1662  * This is controlled by the sysctl variable kern.corefile (see above).
1663  */
1664 
1665 static char *
1666 expand_name(const char *name, uid_t uid, pid_t pid)
1667 {
1668 	char *temp;
1669 	char buf[11];		/* Buffer for pid/uid -- max 4B */
1670 	int i, n;
1671 	char *format = corefilename;
1672 	size_t namelen;
1673 
1674 	temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
1675 	if (temp == NULL)
1676 		return NULL;
1677 	namelen = strlen(name);
1678 	for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
1679 		int l;
1680 		switch (format[i]) {
1681 		case '%':	/* Format character */
1682 			i++;
1683 			switch (format[i]) {
1684 			case '%':
1685 				temp[n++] = '%';
1686 				break;
1687 			case 'N':	/* process name */
1688 				if ((n + namelen) > MAXPATHLEN) {
1689 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
1690 					    pid, name, uid, temp, name);
1691 					kfree(temp, M_TEMP);
1692 					return NULL;
1693 				}
1694 				memcpy(temp+n, name, namelen);
1695 				n += namelen;
1696 				break;
1697 			case 'P':	/* process id */
1698 				l = ksprintf(buf, "%u", pid);
1699 				if ((n + l) > MAXPATHLEN) {
1700 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
1701 					    pid, name, uid, temp, name);
1702 					kfree(temp, M_TEMP);
1703 					return NULL;
1704 				}
1705 				memcpy(temp+n, buf, l);
1706 				n += l;
1707 				break;
1708 			case 'U':	/* user id */
1709 				l = ksprintf(buf, "%u", uid);
1710 				if ((n + l) > MAXPATHLEN) {
1711 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
1712 					    pid, name, uid, temp, name);
1713 					kfree(temp, M_TEMP);
1714 					return NULL;
1715 				}
1716 				memcpy(temp+n, buf, l);
1717 				n += l;
1718 				break;
1719 			default:
1720 			  	log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
1721 			}
1722 			break;
1723 		default:
1724 			temp[n++] = format[i];
1725 		}
1726 	}
1727 	temp[n] = '\0';
1728 	return temp;
1729 }
1730 
1731 /*
1732  * Dump a process' core.  The main routine does some
1733  * policy checking, and creates the name of the coredump;
1734  * then it passes on a vnode and a size limit to the process-specific
1735  * coredump routine if there is one; if there _is not_ one, it returns
1736  * ENOSYS; otherwise it returns the error from the process-specific routine.
1737  */
1738 
1739 static int
1740 coredump(struct proc *p)
1741 {
1742 	struct vnode *vp;
1743 	struct ucred *cred = p->p_ucred;
1744 	struct flock lf;
1745 	struct nlookupdata nd;
1746 	struct vattr vattr;
1747 	int error, error1;
1748 	char *name;			/* name of corefile */
1749 	off_t limit;
1750 
1751 	STOPEVENT(p, S_CORE, 0);
1752 
1753 	if (((sugid_coredump == 0) && p->p_flag & P_SUGID) || do_coredump == 0)
1754 		return (EFAULT);
1755 
1756 	/*
1757 	 * Note that the bulk of limit checking is done after
1758 	 * the corefile is created.  The exception is if the limit
1759 	 * for corefiles is 0, in which case we don't bother
1760 	 * creating the corefile at all.  This layout means that
1761 	 * a corefile is truncated instead of not being created,
1762 	 * if it is larger than the limit.
1763 	 */
1764 	limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
1765 	if (limit == 0)
1766 		return EFBIG;
1767 
1768 	name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
1769 	if (name == NULL)
1770 		return (EINVAL);
1771 	error = nlookup_init(&nd, name, UIO_SYSSPACE, NLC_LOCKVP);
1772 	if (error == 0)
1773 		error = vn_open(&nd, NULL, O_CREAT | FWRITE | O_NOFOLLOW, S_IRUSR | S_IWUSR);
1774 	kfree(name, M_TEMP);
1775 	if (error) {
1776 		nlookup_done(&nd);
1777 		return (error);
1778 	}
1779 	vp = nd.nl_open_vp;
1780 	nd.nl_open_vp = NULL;
1781 	nlookup_done(&nd);
1782 
1783 	vn_unlock(vp);
1784 	lf.l_whence = SEEK_SET;
1785 	lf.l_start = 0;
1786 	lf.l_len = 0;
1787 	lf.l_type = F_WRLCK;
1788 	error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, 0);
1789 	if (error)
1790 		goto out2;
1791 
1792 	/* Don't dump to non-regular files or files with links. */
1793 	if (vp->v_type != VREG ||
1794 	    VOP_GETATTR(vp, &vattr) || vattr.va_nlink != 1) {
1795 		error = EFAULT;
1796 		goto out1;
1797 	}
1798 
1799 	VATTR_NULL(&vattr);
1800 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1801 	vattr.va_size = 0;
1802 	VOP_SETATTR(vp, &vattr, cred);
1803 	p->p_acflag |= ACORE;
1804 	vn_unlock(vp);
1805 
1806 	error = p->p_sysent->sv_coredump ?
1807 		  p->p_sysent->sv_coredump(p, vp, limit) : ENOSYS;
1808 
1809 out1:
1810 	lf.l_type = F_UNLCK;
1811 	VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, 0);
1812 out2:
1813 	error1 = vn_close(vp, FWRITE);
1814 	if (error == 0)
1815 		error = error1;
1816 	return (error);
1817 }
1818 
1819 /*
1820  * Nonexistent system call-- signal process (may want to handle it).
1821  * Flag error in case process won't see signal immediately (blocked or ignored).
1822  */
1823 /* ARGSUSED */
1824 int
1825 sys_nosys(struct nosys_args *args)
1826 {
1827 	ksignal(curproc, SIGSYS);
1828 	return (EINVAL);
1829 }
1830 
1831 /*
1832  * Send a SIGIO or SIGURG signal to a process or process group using
1833  * stored credentials rather than those of the current process.
1834  */
1835 void
1836 pgsigio(struct sigio *sigio, int sig, int checkctty)
1837 {
1838 	if (sigio == NULL)
1839 		return;
1840 
1841 	if (sigio->sio_pgid > 0) {
1842 		if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
1843 		             sigio->sio_proc))
1844 			ksignal(sigio->sio_proc, sig);
1845 	} else if (sigio->sio_pgid < 0) {
1846 		struct proc *p;
1847 
1848 		lockmgr(&sigio->sio_pgrp->pg_lock, LK_EXCLUSIVE);
1849 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
1850 			if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
1851 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
1852 				ksignal(p, sig);
1853 		}
1854 		lockmgr(&sigio->sio_pgrp->pg_lock, LK_RELEASE);
1855 	}
1856 }
1857 
1858 static int
1859 filt_sigattach(struct knote *kn)
1860 {
1861 	struct proc *p = curproc;
1862 
1863 	kn->kn_ptr.p_proc = p;
1864 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
1865 
1866 	/* XXX lock the proc here while adding to the list? */
1867 	SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext);
1868 
1869 	return (0);
1870 }
1871 
1872 static void
1873 filt_sigdetach(struct knote *kn)
1874 {
1875 	struct proc *p = kn->kn_ptr.p_proc;
1876 
1877 	SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
1878 }
1879 
1880 /*
1881  * signal knotes are shared with proc knotes, so we apply a mask to
1882  * the hint in order to differentiate them from process hints.  This
1883  * could be avoided by using a signal-specific knote list, but probably
1884  * isn't worth the trouble.
1885  */
1886 static int
1887 filt_signal(struct knote *kn, long hint)
1888 {
1889 	if (hint & NOTE_SIGNAL) {
1890 		hint &= ~NOTE_SIGNAL;
1891 
1892 		if (kn->kn_id == hint)
1893 			kn->kn_data++;
1894 	}
1895 	return (kn->kn_data != 0);
1896 }
1897