xref: /dflybsd-src/sys/kern/kern_sig.c (revision 0e1bdf492f13155bc45fb3fcd056ee51d3c763c0)
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  */
41 
42 #include "opt_ktrace.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/sysproto.h>
48 #include <sys/signalvar.h>
49 #include <sys/resourcevar.h>
50 #include <sys/vnode.h>
51 #include <sys/event.h>
52 #include <sys/proc.h>
53 #include <sys/nlookup.h>
54 #include <sys/pioctl.h>
55 #include <sys/acct.h>
56 #include <sys/fcntl.h>
57 #include <sys/lock.h>
58 #include <sys/wait.h>
59 #include <sys/ktrace.h>
60 #include <sys/syslog.h>
61 #include <sys/stat.h>
62 #include <sys/sysent.h>
63 #include <sys/sysctl.h>
64 #include <sys/malloc.h>
65 #include <sys/interrupt.h>
66 #include <sys/unistd.h>
67 #include <sys/kern_syscall.h>
68 #include <sys/vkernel.h>
69 
70 #include <sys/signal2.h>
71 #include <sys/thread2.h>
72 
73 #include <machine/cpu.h>
74 #include <machine/smp.h>
75 
76 static int	coredump(struct lwp *, int);
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 static void	lwp_signotify(struct lwp *lp);
82 #ifdef SMP
83 static void	signotify_remote(void *arg);
84 #endif
85 static int	kern_sigtimedwait(sigset_t set, siginfo_t *info,
86 		    struct timespec *timeout);
87 
88 static int	filt_sigattach(struct knote *kn);
89 static void	filt_sigdetach(struct knote *kn);
90 static int	filt_signal(struct knote *kn, long hint);
91 
92 struct filterops sig_filtops =
93 	{ 0, filt_sigattach, filt_sigdetach, filt_signal };
94 
95 static int	kern_logsigexit = 1;
96 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
97     &kern_logsigexit, 0,
98     "Log processes quitting on abnormal signals to syslog(3)");
99 
100 /*
101  * Can process p, with pcred pc, send the signal sig to process q?
102  */
103 #define CANSIGNAL(q, sig) \
104 	(!p_trespass(curproc->p_ucred, (q)->p_ucred) || \
105 	((sig) == SIGCONT && (q)->p_session == curproc->p_session))
106 
107 /*
108  * Policy -- Can real uid ruid with ucred uc send a signal to process q?
109  */
110 #define CANSIGIO(ruid, uc, q) \
111 	((uc)->cr_uid == 0 || \
112 	    (ruid) == (q)->p_ucred->cr_ruid || \
113 	    (uc)->cr_uid == (q)->p_ucred->cr_ruid || \
114 	    (ruid) == (q)->p_ucred->cr_uid || \
115 	    (uc)->cr_uid == (q)->p_ucred->cr_uid)
116 
117 int sugid_coredump;
118 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RW,
119 	&sugid_coredump, 0, "Enable coredumping set user/group ID processes");
120 
121 static int	do_coredump = 1;
122 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
123 	&do_coredump, 0, "Enable/Disable coredumps");
124 
125 /*
126  * Signal properties and actions.
127  * The array below categorizes the signals and their default actions
128  * according to the following properties:
129  */
130 #define	SA_KILL		0x01		/* terminates process by default */
131 #define	SA_CORE		0x02		/* ditto and coredumps */
132 #define	SA_STOP		0x04		/* suspend process */
133 #define	SA_TTYSTOP	0x08		/* ditto, from tty */
134 #define	SA_IGNORE	0x10		/* ignore by default */
135 #define	SA_CONT		0x20		/* continue if suspended */
136 #define	SA_CANTMASK	0x40		/* non-maskable, catchable */
137 #define SA_CKPT         0x80            /* checkpoint process */
138 
139 
140 static int sigproptbl[NSIG] = {
141         SA_KILL,                /* SIGHUP */
142         SA_KILL,                /* SIGINT */
143         SA_KILL|SA_CORE,        /* SIGQUIT */
144         SA_KILL|SA_CORE,        /* SIGILL */
145         SA_KILL|SA_CORE,        /* SIGTRAP */
146         SA_KILL|SA_CORE,        /* SIGABRT */
147         SA_KILL|SA_CORE,        /* SIGEMT */
148         SA_KILL|SA_CORE,        /* SIGFPE */
149         SA_KILL,                /* SIGKILL */
150         SA_KILL|SA_CORE,        /* SIGBUS */
151         SA_KILL|SA_CORE,        /* SIGSEGV */
152         SA_KILL|SA_CORE,        /* SIGSYS */
153         SA_KILL,                /* SIGPIPE */
154         SA_KILL,                /* SIGALRM */
155         SA_KILL,                /* SIGTERM */
156         SA_IGNORE,              /* SIGURG */
157         SA_STOP,                /* SIGSTOP */
158         SA_STOP|SA_TTYSTOP,     /* SIGTSTP */
159         SA_IGNORE|SA_CONT,      /* SIGCONT */
160         SA_IGNORE,              /* SIGCHLD */
161         SA_STOP|SA_TTYSTOP,     /* SIGTTIN */
162         SA_STOP|SA_TTYSTOP,     /* SIGTTOU */
163         SA_IGNORE,              /* SIGIO */
164         SA_KILL,                /* SIGXCPU */
165         SA_KILL,                /* SIGXFSZ */
166         SA_KILL,                /* SIGVTALRM */
167         SA_KILL,                /* SIGPROF */
168         SA_IGNORE,              /* SIGWINCH  */
169         SA_IGNORE,              /* SIGINFO */
170         SA_KILL,                /* SIGUSR1 */
171         SA_KILL,                /* SIGUSR2 */
172 	SA_IGNORE,              /* SIGTHR */
173 	SA_CKPT,                /* SIGCKPT */
174 	SA_KILL|SA_CKPT,        /* SIGCKPTEXIT */
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 	SA_IGNORE,
205 
206 };
207 
208 static __inline int
209 sigprop(int sig)
210 {
211 
212 	if (sig > 0 && sig < NSIG)
213 		return (sigproptbl[_SIG_IDX(sig)]);
214 	return (0);
215 }
216 
217 static __inline int
218 sig_ffs(sigset_t *set)
219 {
220 	int i;
221 
222 	for (i = 0; i < _SIG_WORDS; i++)
223 		if (set->__bits[i])
224 			return (ffs(set->__bits[i]) + (i * 32));
225 	return (0);
226 }
227 
228 /*
229  * No requirements.
230  */
231 int
232 kern_sigaction(int sig, struct sigaction *act, struct sigaction *oact)
233 {
234 	struct thread *td = curthread;
235 	struct proc *p = td->td_proc;
236 	struct lwp *lp;
237 	struct sigacts *ps = p->p_sigacts;
238 
239 	if (sig <= 0 || sig > _SIG_MAXSIG)
240 		return (EINVAL);
241 
242 	lwkt_gettoken(&p->p_token);
243 
244 	if (oact) {
245 		oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
246 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
247 		oact->sa_flags = 0;
248 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
249 			oact->sa_flags |= SA_ONSTACK;
250 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
251 			oact->sa_flags |= SA_RESTART;
252 		if (SIGISMEMBER(ps->ps_sigreset, sig))
253 			oact->sa_flags |= SA_RESETHAND;
254 		if (SIGISMEMBER(ps->ps_signodefer, sig))
255 			oact->sa_flags |= SA_NODEFER;
256 		if (SIGISMEMBER(ps->ps_siginfo, sig))
257 			oact->sa_flags |= SA_SIGINFO;
258 		if (SIGISMEMBER(ps->ps_sigmailbox, sig))
259 			oact->sa_flags |= SA_MAILBOX;
260 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDSTOP)
261 			oact->sa_flags |= SA_NOCLDSTOP;
262 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDWAIT)
263 			oact->sa_flags |= SA_NOCLDWAIT;
264 	}
265 	if (act) {
266 		/*
267 		 * Check for invalid requests.  KILL and STOP cannot be
268 		 * caught.
269 		 */
270 		if (sig == SIGKILL || sig == SIGSTOP) {
271 			if (act->sa_handler != SIG_DFL) {
272 				lwkt_reltoken(&p->p_token);
273 				return (EINVAL);
274 			}
275 #if 0
276 			/* (not needed, SIG_DFL forces action to occur) */
277 			if (act->sa_flags & SA_MAILBOX) {
278 				lwkt_reltoken(&p->p_token);
279 				return (EINVAL);
280 			}
281 #endif
282 		}
283 
284 		/*
285 		 * Change setting atomically.
286 		 */
287 		crit_enter();
288 
289 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
290 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
291 		if (act->sa_flags & SA_SIGINFO) {
292 			ps->ps_sigact[_SIG_IDX(sig)] =
293 			    (__sighandler_t *)act->sa_sigaction;
294 			SIGADDSET(ps->ps_siginfo, sig);
295 		} else {
296 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
297 			SIGDELSET(ps->ps_siginfo, sig);
298 		}
299 		if (!(act->sa_flags & SA_RESTART))
300 			SIGADDSET(ps->ps_sigintr, sig);
301 		else
302 			SIGDELSET(ps->ps_sigintr, sig);
303 		if (act->sa_flags & SA_ONSTACK)
304 			SIGADDSET(ps->ps_sigonstack, sig);
305 		else
306 			SIGDELSET(ps->ps_sigonstack, sig);
307 		if (act->sa_flags & SA_RESETHAND)
308 			SIGADDSET(ps->ps_sigreset, sig);
309 		else
310 			SIGDELSET(ps->ps_sigreset, sig);
311 		if (act->sa_flags & SA_NODEFER)
312 			SIGADDSET(ps->ps_signodefer, sig);
313 		else
314 			SIGDELSET(ps->ps_signodefer, sig);
315 		if (act->sa_flags & SA_MAILBOX)
316 			SIGADDSET(ps->ps_sigmailbox, sig);
317 		else
318 			SIGDELSET(ps->ps_sigmailbox, sig);
319 		if (sig == SIGCHLD) {
320 			if (act->sa_flags & SA_NOCLDSTOP)
321 				p->p_sigacts->ps_flag |= PS_NOCLDSTOP;
322 			else
323 				p->p_sigacts->ps_flag &= ~PS_NOCLDSTOP;
324 			if (act->sa_flags & SA_NOCLDWAIT) {
325 				/*
326 				 * Paranoia: since SA_NOCLDWAIT is implemented
327 				 * by reparenting the dying child to PID 1 (and
328 				 * trust it to reap the zombie), PID 1 itself
329 				 * is forbidden to set SA_NOCLDWAIT.
330 				 */
331 				if (p->p_pid == 1)
332 					p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
333 				else
334 					p->p_sigacts->ps_flag |= PS_NOCLDWAIT;
335 			} else {
336 				p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
337 			}
338 		}
339 		/*
340 		 * Set bit in p_sigignore for signals that are set to SIG_IGN,
341 		 * and for signals set to SIG_DFL where the default is to
342 		 * ignore. However, don't put SIGCONT in p_sigignore, as we
343 		 * have to restart the process.
344 		 */
345 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
346 		    (sigprop(sig) & SA_IGNORE &&
347 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
348 			/* never to be seen again */
349 			SIGDELSET(p->p_siglist, sig);
350 			/*
351 			 * Remove the signal also from the thread lists.
352 			 */
353 			FOREACH_LWP_IN_PROC(lp, p) {
354 				SIGDELSET(lp->lwp_siglist, sig);
355 			}
356 			if (sig != SIGCONT) {
357 				/* easier in ksignal */
358 				SIGADDSET(p->p_sigignore, sig);
359 			}
360 			SIGDELSET(p->p_sigcatch, sig);
361 		} else {
362 			SIGDELSET(p->p_sigignore, sig);
363 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
364 				SIGDELSET(p->p_sigcatch, sig);
365 			else
366 				SIGADDSET(p->p_sigcatch, sig);
367 		}
368 
369 		crit_exit();
370 	}
371 	lwkt_reltoken(&p->p_token);
372 	return (0);
373 }
374 
375 int
376 sys_sigaction(struct sigaction_args *uap)
377 {
378 	struct sigaction act, oact;
379 	struct sigaction *actp, *oactp;
380 	int error;
381 
382 	actp = (uap->act != NULL) ? &act : NULL;
383 	oactp = (uap->oact != NULL) ? &oact : NULL;
384 	if (actp) {
385 		error = copyin(uap->act, actp, sizeof(act));
386 		if (error)
387 			return (error);
388 	}
389 	error = kern_sigaction(uap->sig, actp, oactp);
390 	if (oactp && !error) {
391 		error = copyout(oactp, uap->oact, sizeof(oact));
392 	}
393 	return (error);
394 }
395 
396 /*
397  * Initialize signal state for process 0;
398  * set to ignore signals that are ignored by default.
399  */
400 void
401 siginit(struct proc *p)
402 {
403 	int i;
404 
405 	for (i = 1; i <= NSIG; i++)
406 		if (sigprop(i) & SA_IGNORE && i != SIGCONT)
407 			SIGADDSET(p->p_sigignore, i);
408 }
409 
410 /*
411  * Reset signals for an exec of the specified process.
412  */
413 void
414 execsigs(struct proc *p)
415 {
416 	struct sigacts *ps = p->p_sigacts;
417 	struct lwp *lp;
418 	int sig;
419 
420 	lp = ONLY_LWP_IN_PROC(p);
421 
422 	/*
423 	 * Reset caught signals.  Held signals remain held
424 	 * through p_sigmask (unless they were caught,
425 	 * and are now ignored by default).
426 	 */
427 	while (SIGNOTEMPTY(p->p_sigcatch)) {
428 		sig = sig_ffs(&p->p_sigcatch);
429 		SIGDELSET(p->p_sigcatch, sig);
430 		if (sigprop(sig) & SA_IGNORE) {
431 			if (sig != SIGCONT)
432 				SIGADDSET(p->p_sigignore, sig);
433 			SIGDELSET(p->p_siglist, sig);
434 			SIGDELSET(lp->lwp_siglist, sig);
435 		}
436 		ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
437 	}
438 
439 	/*
440 	 * Reset stack state to the user stack.
441 	 * Clear set of signals caught on the signal stack.
442 	 */
443 	lp->lwp_sigstk.ss_flags = SS_DISABLE;
444 	lp->lwp_sigstk.ss_size = 0;
445 	lp->lwp_sigstk.ss_sp = 0;
446 	lp->lwp_flag &= ~LWP_ALTSTACK;
447 	/*
448 	 * Reset no zombies if child dies flag as Solaris does.
449 	 */
450 	p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
451 }
452 
453 /*
454  * kern_sigprocmask() - MP SAFE ONLY IF p == curproc
455  *
456  *	Manipulate signal mask.  This routine is MP SAFE *ONLY* if
457  *	p == curproc.
458  */
459 int
460 kern_sigprocmask(int how, sigset_t *set, sigset_t *oset)
461 {
462 	struct thread *td = curthread;
463 	struct lwp *lp = td->td_lwp;
464 	struct proc *p = td->td_proc;
465 	int error;
466 
467 	lwkt_gettoken(&p->p_token);
468 
469 	if (oset != NULL)
470 		*oset = lp->lwp_sigmask;
471 
472 	error = 0;
473 	if (set != NULL) {
474 		switch (how) {
475 		case SIG_BLOCK:
476 			SIG_CANTMASK(*set);
477 			SIGSETOR(lp->lwp_sigmask, *set);
478 			break;
479 		case SIG_UNBLOCK:
480 			SIGSETNAND(lp->lwp_sigmask, *set);
481 			break;
482 		case SIG_SETMASK:
483 			SIG_CANTMASK(*set);
484 			lp->lwp_sigmask = *set;
485 			break;
486 		default:
487 			error = EINVAL;
488 			break;
489 		}
490 	}
491 
492 	lwkt_reltoken(&p->p_token);
493 
494 	return (error);
495 }
496 
497 /*
498  * sigprocmask()
499  *
500  * MPSAFE
501  */
502 int
503 sys_sigprocmask(struct sigprocmask_args *uap)
504 {
505 	sigset_t set, oset;
506 	sigset_t *setp, *osetp;
507 	int error;
508 
509 	setp = (uap->set != NULL) ? &set : NULL;
510 	osetp = (uap->oset != NULL) ? &oset : NULL;
511 	if (setp) {
512 		error = copyin(uap->set, setp, sizeof(set));
513 		if (error)
514 			return (error);
515 	}
516 	error = kern_sigprocmask(uap->how, setp, osetp);
517 	if (osetp && !error) {
518 		error = copyout(osetp, uap->oset, sizeof(oset));
519 	}
520 	return (error);
521 }
522 
523 /*
524  * MPSAFE
525  */
526 int
527 kern_sigpending(struct __sigset *set)
528 {
529 	struct lwp *lp = curthread->td_lwp;
530 
531 	*set = lwp_sigpend(lp);
532 
533 	return (0);
534 }
535 
536 /*
537  * MPSAFE
538  */
539 int
540 sys_sigpending(struct sigpending_args *uap)
541 {
542 	sigset_t set;
543 	int error;
544 
545 	error = kern_sigpending(&set);
546 
547 	if (error == 0)
548 		error = copyout(&set, uap->set, sizeof(set));
549 	return (error);
550 }
551 
552 /*
553  * Suspend process until signal, providing mask to be set
554  * in the meantime.
555  *
556  * MPSAFE
557  */
558 int
559 kern_sigsuspend(struct __sigset *set)
560 {
561 	struct thread *td = curthread;
562 	struct lwp *lp = td->td_lwp;
563 	struct proc *p = td->td_proc;
564 	struct sigacts *ps = p->p_sigacts;
565 
566 	/*
567 	 * When returning from sigsuspend, we want
568 	 * the old mask to be restored after the
569 	 * signal handler has finished.  Thus, we
570 	 * save it here and mark the sigacts structure
571 	 * to indicate this.
572 	 */
573 	lp->lwp_oldsigmask = lp->lwp_sigmask;
574 	lp->lwp_flag |= LWP_OLDMASK;
575 
576 	SIG_CANTMASK(*set);
577 	lp->lwp_sigmask = *set;
578 	while (tsleep(ps, PCATCH, "pause", 0) == 0)
579 		/* void */;
580 	/* always return EINTR rather than ERESTART... */
581 	return (EINTR);
582 }
583 
584 /*
585  * Note nonstandard calling convention: libc stub passes mask, not
586  * pointer, to save a copyin.
587  *
588  * MPSAFE
589  */
590 int
591 sys_sigsuspend(struct sigsuspend_args *uap)
592 {
593 	sigset_t mask;
594 	int error;
595 
596 	error = copyin(uap->sigmask, &mask, sizeof(mask));
597 	if (error)
598 		return (error);
599 
600 	error = kern_sigsuspend(&mask);
601 
602 	return (error);
603 }
604 
605 /*
606  * MPSAFE
607  */
608 int
609 kern_sigaltstack(struct sigaltstack *ss, struct sigaltstack *oss)
610 {
611 	struct thread *td = curthread;
612 	struct lwp *lp = td->td_lwp;
613 	struct proc *p = td->td_proc;
614 
615 	if ((lp->lwp_flag & LWP_ALTSTACK) == 0)
616 		lp->lwp_sigstk.ss_flags |= SS_DISABLE;
617 
618 	if (oss)
619 		*oss = lp->lwp_sigstk;
620 
621 	if (ss) {
622 		if (ss->ss_flags & SS_DISABLE) {
623 			if (lp->lwp_sigstk.ss_flags & SS_ONSTACK)
624 				return (EINVAL);
625 			lp->lwp_flag &= ~LWP_ALTSTACK;
626 			lp->lwp_sigstk.ss_flags = ss->ss_flags;
627 		} else {
628 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
629 				return (ENOMEM);
630 			lp->lwp_flag |= LWP_ALTSTACK;
631 			lp->lwp_sigstk = *ss;
632 		}
633 	}
634 
635 	return (0);
636 }
637 
638 /*
639  * MPSAFE
640  */
641 int
642 sys_sigaltstack(struct sigaltstack_args *uap)
643 {
644 	stack_t ss, oss;
645 	int error;
646 
647 	if (uap->ss) {
648 		error = copyin(uap->ss, &ss, sizeof(ss));
649 		if (error)
650 			return (error);
651 	}
652 
653 	error = kern_sigaltstack(uap->ss ? &ss : NULL,
654 	    uap->oss ? &oss : NULL);
655 
656 	if (error == 0 && uap->oss)
657 		error = copyout(&oss, uap->oss, sizeof(*uap->oss));
658 	return (error);
659 }
660 
661 /*
662  * Common code for kill process group/broadcast kill.
663  * cp is calling process.
664  */
665 struct killpg_info {
666 	int nfound;
667 	int sig;
668 };
669 
670 static int killpg_all_callback(struct proc *p, void *data);
671 
672 static int
673 dokillpg(int sig, int pgid, int all)
674 {
675 	struct killpg_info info;
676 	struct proc *cp = curproc;
677 	struct proc *p;
678 	struct pgrp *pgrp;
679 
680 	info.nfound = 0;
681 	info.sig = sig;
682 
683 	if (all) {
684 		/*
685 		 * broadcast
686 		 */
687 		allproc_scan(killpg_all_callback, &info);
688 	} else {
689 		if (pgid == 0) {
690 			/*
691 			 * zero pgid means send to my process group.
692 			 */
693 			pgrp = cp->p_pgrp;
694 			pgref(pgrp);
695 		} else {
696 			pgrp = pgfind(pgid);
697 			if (pgrp == NULL)
698 				return (ESRCH);
699 		}
700 
701 		/*
702 		 * Must interlock all signals against fork
703 		 */
704 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
705 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
706 			if (p->p_pid <= 1 ||
707 			    p->p_stat == SZOMB ||
708 			    (p->p_flag & P_SYSTEM) ||
709 			    !CANSIGNAL(p, sig)) {
710 				continue;
711 			}
712 			++info.nfound;
713 			if (sig)
714 				ksignal(p, sig);
715 		}
716 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
717 		pgrel(pgrp);
718 	}
719 	return (info.nfound ? 0 : ESRCH);
720 }
721 
722 static int
723 killpg_all_callback(struct proc *p, void *data)
724 {
725 	struct killpg_info *info = data;
726 
727 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) ||
728 	    p == curproc || !CANSIGNAL(p, info->sig)) {
729 		return (0);
730 	}
731 	++info->nfound;
732 	if (info->sig)
733 		ksignal(p, info->sig);
734 	return(0);
735 }
736 
737 /*
738  * Send a general signal to a process or LWPs within that process.  Note
739  * that new signals cannot be sent if a process is exiting.
740  *
741  * No requirements.
742  */
743 int
744 kern_kill(int sig, pid_t pid, lwpid_t tid)
745 {
746 	int t;
747 
748 	if ((u_int)sig > _SIG_MAXSIG)
749 		return (EINVAL);
750 
751 	lwkt_gettoken(&proc_token);
752 
753 	if (pid > 0) {
754 		struct proc *p;
755 		struct lwp *lp = NULL;
756 
757 		/* kill single process */
758 		if ((p = pfind(pid)) == NULL) {
759 			lwkt_reltoken(&proc_token);
760 			return (ESRCH);
761 		}
762 		lwkt_gettoken(&p->p_token);
763 		if (!CANSIGNAL(p, sig)) {
764 			lwkt_reltoken(&p->p_token);
765 			PRELE(p);
766 			lwkt_reltoken(&proc_token);
767 			return (EPERM);
768 		}
769 
770 		/*
771 		 * NOP if the process is exiting.  Note that lwpsignal() is
772 		 * called directly with P_WEXIT set to kill individual LWPs
773 		 * during exit, which is allowed.
774 		 */
775 		if (p->p_flag & P_WEXIT) {
776 			lwkt_reltoken(&p->p_token);
777 			PRELE(p);
778 			lwkt_reltoken(&proc_token);
779 			return (0);
780 		}
781 		if (tid != -1) {
782 			lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, tid);
783 			if (lp == NULL) {
784 				lwkt_reltoken(&p->p_token);
785 				PRELE(p);
786 				lwkt_reltoken(&proc_token);
787 				return (ESRCH);
788 			}
789 		}
790 		if (sig)
791 			lwpsignal(p, lp, sig);
792 		lwkt_reltoken(&p->p_token);
793 		PRELE(p);
794 		lwkt_reltoken(&proc_token);
795 		return (0);
796 	}
797 
798 	/*
799 	 * If we come here, pid is a special broadcast pid.
800 	 * This doesn't mix with a tid.
801 	 */
802 	if (tid != -1) {
803 		lwkt_reltoken(&proc_token);
804 		return (EINVAL);
805 	}
806 	switch (pid) {
807 	case -1:		/* broadcast signal */
808 		t = (dokillpg(sig, 0, 1));
809 		break;
810 	case 0:			/* signal own process group */
811 		t = (dokillpg(sig, 0, 0));
812 		break;
813 	default:		/* negative explicit process group */
814 		t = (dokillpg(sig, -pid, 0));
815 		break;
816 	}
817 	lwkt_reltoken(&proc_token);
818 	return t;
819 }
820 
821 int
822 sys_kill(struct kill_args *uap)
823 {
824 	int error;
825 
826 	error = kern_kill(uap->signum, uap->pid, -1);
827 	return (error);
828 }
829 
830 int
831 sys_lwp_kill(struct lwp_kill_args *uap)
832 {
833 	int error;
834 	pid_t pid = uap->pid;
835 
836 	/*
837 	 * A tid is mandatory for lwp_kill(), otherwise
838 	 * you could simply use kill().
839 	 */
840 	if (uap->tid == -1)
841 		return (EINVAL);
842 
843 	/*
844 	 * To save on a getpid() function call for intra-process
845 	 * signals, pid == -1 means current process.
846 	 */
847 	if (pid == -1)
848 		pid = curproc->p_pid;
849 
850 	error = kern_kill(uap->signum, pid, uap->tid);
851 	return (error);
852 }
853 
854 /*
855  * Send a signal to a process group.
856  */
857 void
858 gsignal(int pgid, int sig)
859 {
860 	struct pgrp *pgrp;
861 
862 	if (pgid && (pgrp = pgfind(pgid)))
863 		pgsignal(pgrp, sig, 0);
864 }
865 
866 /*
867  * Send a signal to a process group.  If checktty is 1,
868  * limit to members which have a controlling terminal.
869  *
870  * pg_lock interlocks against a fork that might be in progress, to
871  * ensure that the new child process picks up the signal.
872  */
873 void
874 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
875 {
876 	struct proc *p;
877 
878 	/*
879 	 * Must interlock all signals against fork
880 	 */
881 	if (pgrp) {
882 		pgref(pgrp);
883 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
884 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
885 			if (checkctty == 0 || p->p_flag & P_CONTROLT)
886 				ksignal(p, sig);
887 		}
888 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
889 		pgrel(pgrp);
890 	}
891 }
892 
893 /*
894  * Send a signal caused by a trap to the current lwp.  If it will be caught
895  * immediately, deliver it with correct code.  Otherwise, post it normally.
896  *
897  * These signals may ONLY be delivered to the specified lwp and may never
898  * be delivered to the process generically.
899  */
900 void
901 trapsignal(struct lwp *lp, int sig, u_long code)
902 {
903 	struct proc *p = lp->lwp_proc;
904 	struct sigacts *ps = p->p_sigacts;
905 
906 	/*
907 	 * If we are a virtual kernel running an emulated user process
908 	 * context, switch back to the virtual kernel context before
909 	 * trying to post the signal.
910 	 */
911 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
912 		struct trapframe *tf = lp->lwp_md.md_regs;
913 		tf->tf_trapno = 0;
914 		vkernel_trap(lp, tf);
915 	}
916 
917 
918 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(p->p_sigcatch, sig) &&
919 	    !SIGISMEMBER(lp->lwp_sigmask, sig)) {
920 		lp->lwp_ru.ru_nsignals++;
921 #ifdef KTRACE
922 		if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
923 			ktrpsig(lp, sig, ps->ps_sigact[_SIG_IDX(sig)],
924 				&lp->lwp_sigmask, code);
925 #endif
926 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], sig,
927 						&lp->lwp_sigmask, code);
928 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
929 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
930 			SIGADDSET(lp->lwp_sigmask, sig);
931 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
932 			/*
933 			 * See kern_sigaction() for origin of this code.
934 			 */
935 			SIGDELSET(p->p_sigcatch, sig);
936 			if (sig != SIGCONT &&
937 			    sigprop(sig) & SA_IGNORE)
938 				SIGADDSET(p->p_sigignore, sig);
939 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
940 		}
941 	} else {
942 		lp->lwp_code = code;	/* XXX for core dump/debugger */
943 		lp->lwp_sig = sig;	/* XXX to verify code */
944 		lwpsignal(p, lp, sig);
945 	}
946 }
947 
948 /*
949  * Find a suitable lwp to deliver the signal to.  Returns NULL if all
950  * lwps hold the signal blocked.
951  *
952  * Caller must hold p->p_token.
953  */
954 static struct lwp *
955 find_lwp_for_signal(struct proc *p, int sig)
956 {
957 	struct lwp *lp;
958 	struct lwp *run, *sleep, *stop;
959 
960 	/*
961 	 * If the running/preempted thread belongs to the proc to which
962 	 * the signal is being delivered and this thread does not block
963 	 * the signal, then we can avoid a context switch by delivering
964 	 * the signal to this thread, because it will return to userland
965 	 * soon anyways.
966 	 */
967 	lp = lwkt_preempted_proc();
968 	if (lp != NULL && lp->lwp_proc == p &&
969 	    !SIGISMEMBER(lp->lwp_sigmask, sig)) {
970 		return (lp);
971 	}
972 
973 	run = sleep = stop = NULL;
974 	FOREACH_LWP_IN_PROC(lp, p) {
975 		/*
976 		 * If the signal is being blocked by the lwp, then this
977 		 * lwp is not eligible for receiving the signal.
978 		 */
979 		if (SIGISMEMBER(lp->lwp_sigmask, sig))
980 			continue;
981 
982 		switch (lp->lwp_stat) {
983 		case LSRUN:
984 			run = lp;
985 			break;
986 
987 		case LSSTOP:
988 			stop = lp;
989 			break;
990 
991 		case LSSLEEP:
992 			if (lp->lwp_flag & LWP_SINTR)
993 				sleep = lp;
994 			break;
995 		}
996 	}
997 
998 	if (run != NULL)
999 		return (run);
1000 	else if (sleep != NULL)
1001 		return (sleep);
1002 	else
1003 		return (stop);
1004 }
1005 
1006 /*
1007  * Send the signal to the process.  If the signal has an action, the action
1008  * is usually performed by the target process rather than the caller; we add
1009  * the signal to the set of pending signals for the process.
1010  *
1011  * Exceptions:
1012  *   o When a stop signal is sent to a sleeping process that takes the
1013  *     default action, the process is stopped without awakening it.
1014  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
1015  *     regardless of the signal action (eg, blocked or ignored).
1016  *
1017  * Other ignored signals are discarded immediately.
1018  *
1019  * If the caller wishes to call this function from a hard code section the
1020  * caller must already hold p->p_token (see kern_clock.c).
1021  *
1022  * No requirements.
1023  */
1024 void
1025 ksignal(struct proc *p, int sig)
1026 {
1027 	lwpsignal(p, NULL, sig);
1028 }
1029 
1030 /*
1031  * The core for ksignal.  lp may be NULL, then a suitable thread
1032  * will be chosen.  If not, lp MUST be a member of p.
1033  *
1034  * If the caller wishes to call this function from a hard code section the
1035  * caller must already hold p->p_token.
1036  *
1037  * No requirements.
1038  */
1039 void
1040 lwpsignal(struct proc *p, struct lwp *lp, int sig)
1041 {
1042 	struct proc *q;
1043 	sig_t action;
1044 	int prop;
1045 
1046 	if (sig > _SIG_MAXSIG || sig <= 0) {
1047 		kprintf("lwpsignal: signal %d\n", sig);
1048 		panic("lwpsignal signal number");
1049 	}
1050 
1051 	KKASSERT(lp == NULL || lp->lwp_proc == p);
1052 
1053 	PHOLD(p);
1054 	lwkt_gettoken(&p->p_token);
1055 
1056 	prop = sigprop(sig);
1057 
1058 	/*
1059 	 * If proc is traced, always give parent a chance;
1060 	 * if signal event is tracked by procfs, give *that*
1061 	 * a chance, as well.
1062 	 */
1063 	if ((p->p_flag & P_TRACED) || (p->p_stops & S_SIG)) {
1064 		action = SIG_DFL;
1065 	} else {
1066 		/*
1067 		 * Do not try to deliver signals to an exiting lwp.  Note
1068 		 * that we must still deliver the signal if P_WEXIT is set
1069 		 * in the process flags.
1070 		 */
1071 		if (lp && (lp->lwp_flag & LWP_WEXIT)) {
1072 			lwkt_reltoken(&p->p_token);
1073 			PRELE(p);
1074 			return;
1075 		}
1076 
1077 		/*
1078 		 * If the signal is being ignored, then we forget about
1079 		 * it immediately.  NOTE: We don't set SIGCONT in p_sigignore,
1080 		 * and if it is set to SIG_IGN, action will be SIG_DFL here.
1081 		 */
1082 		if (SIGISMEMBER(p->p_sigignore, sig)) {
1083 			/*
1084 			 * Even if a signal is set SIG_IGN, it may still be
1085 			 * lurking in a kqueue.
1086 			 */
1087 			KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1088 			lwkt_reltoken(&p->p_token);
1089 			PRELE(p);
1090 			return;
1091 		}
1092 		if (SIGISMEMBER(p->p_sigcatch, sig))
1093 			action = SIG_CATCH;
1094 		else
1095 			action = SIG_DFL;
1096 	}
1097 
1098 	/*
1099 	 * If continuing, clear any pending STOP signals.
1100 	 */
1101 	if (prop & SA_CONT)
1102 		SIG_STOPSIGMASK(p->p_siglist);
1103 
1104 	if (prop & SA_STOP) {
1105 		/*
1106 		 * If sending a tty stop signal to a member of an orphaned
1107 		 * process group, discard the signal here if the action
1108 		 * is default; don't stop the process below if sleeping,
1109 		 * and don't clear any pending SIGCONT.
1110 		 */
1111 		if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0 &&
1112 		    action == SIG_DFL) {
1113 			lwkt_reltoken(&p->p_token);
1114 			PRELE(p);
1115 		        return;
1116 		}
1117 		SIG_CONTSIGMASK(p->p_siglist);
1118 		p->p_flag &= ~P_CONTINUED;
1119 	}
1120 
1121 	crit_enter();
1122 
1123 	if (p->p_stat == SSTOP) {
1124 		/*
1125 		 * Nobody can handle this signal, add it to the lwp or
1126 		 * process pending list
1127 		 */
1128 		if (lp)
1129 			SIGADDSET(lp->lwp_siglist, sig);
1130 		else
1131 			SIGADDSET(p->p_siglist, sig);
1132 
1133 		/*
1134 		 * If the process is stopped and is being traced, then no
1135 		 * further action is necessary.
1136 		 */
1137 		if (p->p_flag & P_TRACED)
1138 			goto out;
1139 
1140 		/*
1141 		 * If the process is stopped and receives a KILL signal,
1142 		 * make the process runnable.
1143 		 */
1144 		if (sig == SIGKILL) {
1145 			proc_unstop(p);
1146 			goto active_process;
1147 		}
1148 
1149 		/*
1150 		 * If the process is stopped and receives a CONT signal,
1151 		 * then try to make the process runnable again.
1152 		 */
1153 		if (prop & SA_CONT) {
1154 			/*
1155 			 * If SIGCONT is default (or ignored), we continue the
1156 			 * process but don't leave the signal in p_siglist, as
1157 			 * it has no further action.  If SIGCONT is held, we
1158 			 * continue the process and leave the signal in
1159 			 * p_siglist.  If the process catches SIGCONT, let it
1160 			 * handle the signal itself.
1161 			 *
1162 			 * XXX what if the signal is being held blocked?
1163 			 *
1164 			 * Token required to interlock kern_wait().
1165 			 * Reparenting can also cause a race so we have to
1166 			 * hold (q).
1167 			 */
1168 			q = p->p_pptr;
1169 			PHOLD(q);
1170 			lwkt_gettoken(&q->p_token);
1171 			p->p_flag |= P_CONTINUED;
1172 			wakeup(q);
1173 			if (action == SIG_DFL)
1174 				SIGDELSET(p->p_siglist, sig);
1175 			proc_unstop(p);
1176 			lwkt_reltoken(&q->p_token);
1177 			PRELE(q);
1178 			if (action == SIG_CATCH)
1179 				goto active_process;
1180 			goto out;
1181 		}
1182 
1183 		/*
1184 		 * If the process is stopped and receives another STOP
1185 		 * signal, we do not need to stop it again.  If we did
1186 		 * the shell could get confused.
1187 		 *
1188 		 * However, if the current/preempted lwp is part of the
1189 		 * process receiving the signal, we need to keep it,
1190 		 * so that this lwp can stop in issignal() later, as
1191 		 * we don't want to wait until it reaches userret!
1192 		 */
1193 		if (prop & SA_STOP) {
1194 			if (lwkt_preempted_proc() == NULL ||
1195 			    lwkt_preempted_proc()->lwp_proc != p)
1196 				SIGDELSET(p->p_siglist, sig);
1197 		}
1198 
1199 		/*
1200 		 * Otherwise the process is stopped and it received some
1201 		 * signal, which does not change its stopped state.
1202 		 *
1203 		 * We have to select one thread to set LWP_BREAKTSLEEP,
1204 		 * so that the current signal will break the sleep
1205 		 * as soon as a SA_CONT signal will unstop the process.
1206 		 */
1207 		if (lp == NULL)
1208 			lp = find_lwp_for_signal(p, sig);
1209 		if (lp != NULL &&
1210 		    (lp->lwp_stat == LSSLEEP || lp->lwp_stat == LSSTOP))
1211 			lp->lwp_flag |= LWP_BREAKTSLEEP;
1212 		goto out;
1213 
1214 		/* NOTREACHED */
1215 	}
1216 	/* else not stopped */
1217 active_process:
1218 
1219 	/*
1220 	 * Never deliver a lwp-specific signal to a random lwp.
1221 	 */
1222 	if (lp == NULL) {
1223 		lp = find_lwp_for_signal(p, sig);
1224 		if (lp && SIGISMEMBER(lp->lwp_sigmask, sig))
1225 			lp = NULL;
1226 	}
1227 
1228 	/*
1229 	 * Deliver to the process generically if (1) the signal is being
1230 	 * sent to any thread or (2) we could not find a thread to deliver
1231 	 * it to.
1232 	 */
1233 	if (lp == NULL) {
1234 		SIGADDSET(p->p_siglist, sig);
1235 		goto out;
1236 	}
1237 
1238 	/*
1239 	 * Deliver to a specific LWP whether it masks it or not.  It will
1240 	 * not be dispatched if masked but we must still deliver it.
1241 	 */
1242 	if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
1243 	    (p->p_flag & P_TRACED) == 0) {
1244 		p->p_nice = NZERO;
1245 	}
1246 
1247 	/*
1248 	 * If the process receives a STOP signal which indeed needs to
1249 	 * stop the process, do so.  If the process chose to catch the
1250 	 * signal, it will be treated like any other signal.
1251 	 */
1252 	if ((prop & SA_STOP) && action == SIG_DFL) {
1253 		/*
1254 		 * If a child holding parent blocked, stopping
1255 		 * could cause deadlock.  Take no action at this
1256 		 * time.
1257 		 */
1258 		if (p->p_flag & P_PPWAIT) {
1259 			SIGADDSET(p->p_siglist, sig);
1260 			goto out;
1261 		}
1262 
1263 		/*
1264 		 * Do not actually try to manipulate the process, but simply
1265 		 * stop it.  Lwps will stop as soon as they safely can.
1266 		 */
1267 		p->p_xstat = sig;
1268 		proc_stop(p);
1269 		goto out;
1270 	}
1271 
1272 	/*
1273 	 * If it is a CONT signal with default action, just ignore it.
1274 	 */
1275 	if ((prop & SA_CONT) && action == SIG_DFL)
1276 		goto out;
1277 
1278 	/*
1279 	 * Mark signal pending at this specific thread.
1280 	 */
1281 	SIGADDSET(lp->lwp_siglist, sig);
1282 
1283 	lwp_signotify(lp);
1284 
1285 out:
1286 	lwkt_reltoken(&p->p_token);
1287 	PRELE(p);
1288 	crit_exit();
1289 }
1290 
1291 /*
1292  * p->p_token must be held
1293  */
1294 static void
1295 lwp_signotify(struct lwp *lp)
1296 {
1297 	ASSERT_LWKT_TOKEN_HELD(&lp->lwp_proc->p_token);
1298 	crit_enter();
1299 
1300 	if (lp->lwp_stat == LSSLEEP || lp->lwp_stat == LSSTOP) {
1301 		/*
1302 		 * Thread is in tsleep.
1303 		 */
1304 
1305 		/*
1306 		 * If the thread is sleeping uninterruptibly
1307 		 * we can't interrupt the sleep... the signal will
1308 		 * be noticed when the lwp returns through
1309 		 * trap() or syscall().
1310 		 *
1311 		 * Otherwise the signal can interrupt the sleep.
1312 		 *
1313 		 * If the process is traced, the lwp will handle the
1314 		 * tracing in issignal() when it returns to userland.
1315 		 */
1316 		if (lp->lwp_flag & LWP_SINTR) {
1317 			/*
1318 			 * Make runnable and break out of any tsleep as well.
1319 			 */
1320 			lp->lwp_flag |= LWP_BREAKTSLEEP;
1321 			setrunnable(lp);
1322 		}
1323 	} else {
1324 		/*
1325 		 * Otherwise the thread is running
1326 		 *
1327 		 * LSRUN does nothing with the signal, other than kicking
1328 		 * ourselves if we are running.
1329 		 * SZOMB and SIDL mean that it will either never be noticed,
1330 		 * or noticed very soon.
1331 		 *
1332 		 * Note that lwp_thread may be NULL or may not be completely
1333 		 * initialized if the process is in the SIDL or SZOMB state.
1334 		 *
1335 		 * For SMP we may have to forward the request to another cpu.
1336 		 * YYY the MP lock prevents the target process from moving
1337 		 * to another cpu, see kern/kern_switch.c
1338 		 *
1339 		 * If the target thread is waiting on its message port,
1340 		 * wakeup the target thread so it can check (or ignore)
1341 		 * the new signal.  YYY needs cleanup.
1342 		 */
1343 		if (lp == lwkt_preempted_proc()) {
1344 			signotify();
1345 		} else if (lp->lwp_stat == LSRUN) {
1346 			struct thread *td = lp->lwp_thread;
1347 			struct proc *p __debugvar = lp->lwp_proc;
1348 
1349 			KASSERT(td != NULL,
1350 			    ("pid %d/%d NULL lwp_thread stat %d flags %08x/%08x",
1351 			    p->p_pid, lp->lwp_tid, lp->lwp_stat,
1352 			    p->p_flag, lp->lwp_flag));
1353 
1354 			/*
1355 			 * To prevent a MP race with TDF_SINTR we must
1356 			 * schedule the thread on the correct cpu.
1357 			 */
1358 #ifdef SMP
1359 			if (td->td_gd != mycpu) {
1360 				LWPHOLD(lp);
1361 				lwkt_send_ipiq(td->td_gd, signotify_remote, lp);
1362 			} else
1363 #endif
1364 			if (td->td_flags & TDF_SINTR)
1365 				lwkt_schedule(td);
1366 		}
1367 	}
1368 	crit_exit();
1369 }
1370 
1371 #ifdef SMP
1372 
1373 /*
1374  * This function is called via an IPI.  We will be in a critical section but
1375  * the MP lock will NOT be held.  The passed lp will be held.
1376  *
1377  * We must essentially repeat the code at the end of lwp_signotify(),
1378  * in particular rechecking all races.  If we are still not on the
1379  * correct cpu we leave the lwp ref intact and continue the chase.
1380  *
1381  * XXX this may still not be entirely correct, since we are checking
1382  *     lwp_stat asynchronously.
1383  */
1384 static void
1385 signotify_remote(void *arg)
1386 {
1387 	struct lwp *lp = arg;
1388 	thread_t td;
1389 
1390 	if (lp == lwkt_preempted_proc()) {
1391 		signotify();
1392 	} else if (lp->lwp_stat == LSRUN) {
1393 		/*
1394 		 * To prevent a MP race with TDF_SINTR we must
1395 		 * schedule the thread on the correct cpu.
1396 		 */
1397 		td = lp->lwp_thread;
1398 		if (td->td_gd != mycpu) {
1399 			lwkt_send_ipiq(td->td_gd, signotify_remote, lp);
1400 			return;
1401 			/* NOT REACHED */
1402 		}
1403 		if (td->td_flags & TDF_SINTR)
1404 			lwkt_schedule(td);
1405 	}
1406 	LWPRELE(lp);
1407 }
1408 
1409 #endif
1410 
1411 /*
1412  * Caller must hold p->p_token
1413  */
1414 void
1415 proc_stop(struct proc *p)
1416 {
1417 	struct proc *q;
1418 	struct lwp *lp;
1419 
1420 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1421 	crit_enter();
1422 
1423 	/* If somebody raced us, be happy with it */
1424 	if (p->p_stat == SSTOP || p->p_stat == SZOMB) {
1425 		crit_exit();
1426 		return;
1427 	}
1428 	p->p_stat = SSTOP;
1429 
1430 	FOREACH_LWP_IN_PROC(lp, p) {
1431 		switch (lp->lwp_stat) {
1432 		case LSSTOP:
1433 			/*
1434 			 * Do nothing, we are already counted in
1435 			 * p_nstopped.
1436 			 */
1437 			break;
1438 
1439 		case LSSLEEP:
1440 			/*
1441 			 * We're sleeping, but we will stop before
1442 			 * returning to userspace, so count us
1443 			 * as stopped as well.  We set LWP_WSTOP
1444 			 * to signal the lwp that it should not
1445 			 * increase p_nstopped when reaching tstop().
1446 			 */
1447 			if ((lp->lwp_flag & LWP_WSTOP) == 0) {
1448 				lp->lwp_flag |= LWP_WSTOP;
1449 				++p->p_nstopped;
1450 			}
1451 			break;
1452 
1453 		case LSRUN:
1454 			/*
1455 			 * We might notify ourself, but that's not
1456 			 * a problem.
1457 			 */
1458 			lwp_signotify(lp);
1459 			break;
1460 		}
1461 	}
1462 
1463 	if (p->p_nstopped == p->p_nthreads) {
1464 		/*
1465 		 * Token required to interlock kern_wait().  Reparenting can
1466 		 * also cause a race so we have to hold (q).
1467 		 */
1468 		q = p->p_pptr;
1469 		PHOLD(q);
1470 		lwkt_gettoken(&q->p_token);
1471 		p->p_flag &= ~P_WAITED;
1472 		wakeup(q);
1473 		if ((q->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1474 			ksignal(p->p_pptr, SIGCHLD);
1475 		lwkt_reltoken(&q->p_token);
1476 		PRELE(q);
1477 	}
1478 	crit_exit();
1479 }
1480 
1481 /*
1482  * Caller must hold proc_token
1483  */
1484 void
1485 proc_unstop(struct proc *p)
1486 {
1487 	struct lwp *lp;
1488 
1489 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1490 	crit_enter();
1491 
1492 	if (p->p_stat != SSTOP) {
1493 		crit_exit();
1494 		return;
1495 	}
1496 
1497 	p->p_stat = SACTIVE;
1498 
1499 	FOREACH_LWP_IN_PROC(lp, p) {
1500 		switch (lp->lwp_stat) {
1501 		case LSRUN:
1502 			/*
1503 			 * Uh?  Not stopped?  Well, I guess that's okay.
1504 			 */
1505 			if (bootverbose)
1506 				kprintf("proc_unstop: lwp %d/%d not sleeping\n",
1507 					p->p_pid, lp->lwp_tid);
1508 			break;
1509 
1510 		case LSSLEEP:
1511 			/*
1512 			 * Still sleeping.  Don't bother waking it up.
1513 			 * However, if this thread was counted as
1514 			 * stopped, undo this.
1515 			 *
1516 			 * Nevertheless we call setrunnable() so that it
1517 			 * will wake up in case a signal or timeout arrived
1518 			 * in the meantime.
1519 			 */
1520 			if (lp->lwp_flag & LWP_WSTOP) {
1521 				lp->lwp_flag &= ~LWP_WSTOP;
1522 				--p->p_nstopped;
1523 			} else {
1524 				if (bootverbose)
1525 					kprintf("proc_unstop: lwp %d/%d sleeping, not stopped\n",
1526 						p->p_pid, lp->lwp_tid);
1527 			}
1528 			/* FALLTHROUGH */
1529 
1530 		case LSSTOP:
1531 			setrunnable(lp);
1532 			break;
1533 
1534 		}
1535 	}
1536 	crit_exit();
1537 }
1538 
1539 /*
1540  * No requirements.
1541  */
1542 static int
1543 kern_sigtimedwait(sigset_t waitset, siginfo_t *info, struct timespec *timeout)
1544 {
1545 	sigset_t savedmask, set;
1546 	struct proc *p = curproc;
1547 	struct lwp *lp = curthread->td_lwp;
1548 	int error, sig, hz, timevalid = 0;
1549 	struct timespec rts, ets, ts;
1550 	struct timeval tv;
1551 
1552 	error = 0;
1553 	sig = 0;
1554 	ets.tv_sec = 0;		/* silence compiler warning */
1555 	ets.tv_nsec = 0;	/* silence compiler warning */
1556 	SIG_CANTMASK(waitset);
1557 	savedmask = lp->lwp_sigmask;
1558 
1559 	if (timeout) {
1560 		if (timeout->tv_sec >= 0 && timeout->tv_nsec >= 0 &&
1561 		    timeout->tv_nsec < 1000000000) {
1562 			timevalid = 1;
1563 			getnanouptime(&rts);
1564 		 	ets = rts;
1565 			timespecadd(&ets, timeout);
1566 		}
1567 	}
1568 
1569 	for (;;) {
1570 		set = lwp_sigpend(lp);
1571 		SIGSETAND(set, waitset);
1572 		if ((sig = sig_ffs(&set)) != 0) {
1573 			SIGFILLSET(lp->lwp_sigmask);
1574 			SIGDELSET(lp->lwp_sigmask, sig);
1575 			SIG_CANTMASK(lp->lwp_sigmask);
1576 			sig = issignal(lp, 1);
1577 			/*
1578 			 * It may be a STOP signal, in the case, issignal
1579 			 * returns 0, because we may stop there, and new
1580 			 * signal can come in, we should restart if we got
1581 			 * nothing.
1582 			 */
1583 			if (sig == 0)
1584 				continue;
1585 			else
1586 				break;
1587 		}
1588 
1589 		/*
1590 		 * Previous checking got nothing, and we retried but still
1591 		 * got nothing, we should return the error status.
1592 		 */
1593 		if (error)
1594 			break;
1595 
1596 		/*
1597 		 * POSIX says this must be checked after looking for pending
1598 		 * signals.
1599 		 */
1600 		if (timeout) {
1601 			if (timevalid == 0) {
1602 				error = EINVAL;
1603 				break;
1604 			}
1605 			getnanouptime(&rts);
1606 			if (timespeccmp(&rts, &ets, >=)) {
1607 				error = EAGAIN;
1608 				break;
1609 			}
1610 			ts = ets;
1611 			timespecsub(&ts, &rts);
1612 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1613 			hz = tvtohz_high(&tv);
1614 		} else {
1615 			hz = 0;
1616 		}
1617 
1618 		lp->lwp_sigmask = savedmask;
1619 		SIGSETNAND(lp->lwp_sigmask, waitset);
1620 		/*
1621 		 * We won't ever be woken up.  Instead, our sleep will
1622 		 * be broken in lwpsignal().
1623 		 */
1624 		error = tsleep(&p->p_sigacts, PCATCH, "sigwt", hz);
1625 		if (timeout) {
1626 			if (error == ERESTART) {
1627 				/* can not restart a timeout wait. */
1628 				error = EINTR;
1629 			} else if (error == EAGAIN) {
1630 				/* will calculate timeout by ourself. */
1631 				error = 0;
1632 			}
1633 		}
1634 		/* Retry ... */
1635 	}
1636 
1637 	lp->lwp_sigmask = savedmask;
1638 	if (sig) {
1639 		error = 0;
1640 		bzero(info, sizeof(*info));
1641 		info->si_signo = sig;
1642 		lwp_delsig(lp, sig);	/* take the signal! */
1643 
1644 		if (sig == SIGKILL) {
1645 			sigexit(lp, sig);
1646 			/* NOT REACHED */
1647 		}
1648 	}
1649 
1650 	return (error);
1651 }
1652 
1653 /*
1654  * MPALMOSTSAFE
1655  */
1656 int
1657 sys_sigtimedwait(struct sigtimedwait_args *uap)
1658 {
1659 	struct timespec ts;
1660 	struct timespec *timeout;
1661 	sigset_t set;
1662 	siginfo_t info;
1663 	int error;
1664 
1665 	if (uap->timeout) {
1666 		error = copyin(uap->timeout, &ts, sizeof(ts));
1667 		if (error)
1668 			return (error);
1669 		timeout = &ts;
1670 	} else {
1671 		timeout = NULL;
1672 	}
1673 	error = copyin(uap->set, &set, sizeof(set));
1674 	if (error)
1675 		return (error);
1676 	error = kern_sigtimedwait(set, &info, timeout);
1677 	if (error)
1678 		return (error);
1679  	if (uap->info)
1680 		error = copyout(&info, uap->info, sizeof(info));
1681 	/* Repost if we got an error. */
1682 	/*
1683 	 * XXX lwp
1684 	 *
1685 	 * This could transform a thread-specific signal to another
1686 	 * thread / process pending signal.
1687 	 */
1688 	if (error) {
1689 		ksignal(curproc, info.si_signo);
1690 	} else {
1691 		uap->sysmsg_result = info.si_signo;
1692 	}
1693 	return (error);
1694 }
1695 
1696 /*
1697  * MPALMOSTSAFE
1698  */
1699 int
1700 sys_sigwaitinfo(struct sigwaitinfo_args *uap)
1701 {
1702 	siginfo_t info;
1703 	sigset_t set;
1704 	int error;
1705 
1706 	error = copyin(uap->set, &set, sizeof(set));
1707 	if (error)
1708 		return (error);
1709 	error = kern_sigtimedwait(set, &info, NULL);
1710 	if (error)
1711 		return (error);
1712 	if (uap->info)
1713 		error = copyout(&info, uap->info, sizeof(info));
1714 	/* Repost if we got an error. */
1715 	/*
1716 	 * XXX lwp
1717 	 *
1718 	 * This could transform a thread-specific signal to another
1719 	 * thread / process pending signal.
1720 	 */
1721 	if (error) {
1722 		ksignal(curproc, info.si_signo);
1723 	} else {
1724 		uap->sysmsg_result = info.si_signo;
1725 	}
1726 	return (error);
1727 }
1728 
1729 /*
1730  * If the current process has received a signal that would interrupt a
1731  * system call, return EINTR or ERESTART as appropriate.
1732  */
1733 int
1734 iscaught(struct lwp *lp)
1735 {
1736 	struct proc *p = lp->lwp_proc;
1737 	int sig;
1738 
1739 	if (p) {
1740 		if ((sig = CURSIG(lp)) != 0) {
1741 			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1742 				return (EINTR);
1743 			return (ERESTART);
1744 		}
1745 	}
1746 	return(EWOULDBLOCK);
1747 }
1748 
1749 /*
1750  * If the current process has received a signal (should be caught or cause
1751  * termination, should interrupt current syscall), return the signal number.
1752  * Stop signals with default action are processed immediately, then cleared;
1753  * they aren't returned.  This is checked after each entry to the system for
1754  * a syscall or trap (though this can usually be done without calling issignal
1755  * by checking the pending signal masks in the CURSIG macro).
1756  *
1757  * This routine is called via CURSIG/__cursig.  We will acquire and release
1758  * p->p_token but if the caller needs to interlock the test the caller must
1759  * also hold p->p_token.
1760  *
1761  *	while (sig = CURSIG(curproc))
1762  *		postsig(sig);
1763  *
1764  * MPSAFE
1765  */
1766 int
1767 issignal(struct lwp *lp, int maytrace)
1768 {
1769 	struct proc *p = lp->lwp_proc;
1770 	sigset_t mask;
1771 	int sig, prop;
1772 
1773 	lwkt_gettoken(&p->p_token);
1774 
1775 	for (;;) {
1776 		int traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
1777 
1778 		/*
1779 		 * If this process is supposed to stop, stop this thread.
1780 		 */
1781 		if (p->p_stat == SSTOP)
1782 			tstop();
1783 
1784 		mask = lwp_sigpend(lp);
1785 		SIGSETNAND(mask, lp->lwp_sigmask);
1786 		if (p->p_flag & P_PPWAIT)
1787 			SIG_STOPSIGMASK(mask);
1788 		if (SIGISEMPTY(mask)) {		/* no signal to send */
1789 			lwkt_reltoken(&p->p_token);
1790 			return (0);
1791 		}
1792 		sig = sig_ffs(&mask);
1793 
1794 		STOPEVENT(p, S_SIG, sig);
1795 
1796 		/*
1797 		 * We should see pending but ignored signals
1798 		 * only if P_TRACED was on when they were posted.
1799 		 */
1800 		if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
1801 			lwp_delsig(lp, sig);
1802 			continue;
1803 		}
1804 		if (maytrace && (p->p_flag & P_TRACED) && (p->p_flag & P_PPWAIT) == 0) {
1805 			/*
1806 			 * If traced, always stop, and stay stopped until
1807 			 * released by the parent.
1808 			 *
1809 			 * NOTE: SSTOP may get cleared during the loop,
1810 			 * but we do not re-notify the parent if we have
1811 			 * to loop several times waiting for the parent
1812 			 * to let us continue.
1813 			 *
1814 			 * XXX not sure if this is still true
1815 			 */
1816 			p->p_xstat = sig;
1817 			proc_stop(p);
1818 			do {
1819 				tstop();
1820 			} while (!trace_req(p) && (p->p_flag & P_TRACED));
1821 
1822 			/*
1823 			 * If parent wants us to take the signal,
1824 			 * then it will leave it in p->p_xstat;
1825 			 * otherwise we just look for signals again.
1826 			 */
1827 			lwp_delsig(lp, sig);	/* clear old signal */
1828 			sig = p->p_xstat;
1829 			if (sig == 0)
1830 				continue;
1831 
1832 			/*
1833 			 * Put the new signal into p_siglist.  If the
1834 			 * signal is being masked, look for other signals.
1835 			 *
1836 			 * XXX lwp might need a call to ksignal()
1837 			 */
1838 			SIGADDSET(p->p_siglist, sig);
1839 			if (SIGISMEMBER(lp->lwp_sigmask, sig))
1840 				continue;
1841 
1842 			/*
1843 			 * If the traced bit got turned off, go back up
1844 			 * to the top to rescan signals.  This ensures
1845 			 * that p_sig* and ps_sigact are consistent.
1846 			 */
1847 			if ((p->p_flag & P_TRACED) == 0)
1848 				continue;
1849 		}
1850 
1851 		prop = sigprop(sig);
1852 
1853 		/*
1854 		 * Decide whether the signal should be returned.
1855 		 * Return the signal's number, or fall through
1856 		 * to clear it from the pending mask.
1857 		 */
1858 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
1859 		case (intptr_t)SIG_DFL:
1860 			/*
1861 			 * Don't take default actions on system processes.
1862 			 */
1863 			if (p->p_pid <= 1) {
1864 #ifdef DIAGNOSTIC
1865 				/*
1866 				 * Are you sure you want to ignore SIGSEGV
1867 				 * in init? XXX
1868 				 */
1869 				kprintf("Process (pid %lu) got signal %d\n",
1870 					(u_long)p->p_pid, sig);
1871 #endif
1872 				break;		/* == ignore */
1873 			}
1874 
1875 			/*
1876 			 * Handle the in-kernel checkpoint action
1877 			 */
1878 			if (prop & SA_CKPT) {
1879 				checkpoint_signal_handler(lp);
1880 				break;
1881 			}
1882 
1883 			/*
1884 			 * If there is a pending stop signal to process
1885 			 * with default action, stop here,
1886 			 * then clear the signal.  However,
1887 			 * if process is member of an orphaned
1888 			 * process group, ignore tty stop signals.
1889 			 */
1890 			if (prop & SA_STOP) {
1891 				if (p->p_flag & P_TRACED ||
1892 		    		    (p->p_pgrp->pg_jobc == 0 &&
1893 				    prop & SA_TTYSTOP))
1894 					break;	/* == ignore */
1895 				p->p_xstat = sig;
1896 				proc_stop(p);
1897 				tstop();
1898 				break;
1899 			} else if (prop & SA_IGNORE) {
1900 				/*
1901 				 * Except for SIGCONT, shouldn't get here.
1902 				 * Default action is to ignore; drop it.
1903 				 */
1904 				break;		/* == ignore */
1905 			} else {
1906 				lwkt_reltoken(&p->p_token);
1907 				return (sig);
1908 			}
1909 
1910 			/*NOTREACHED*/
1911 
1912 		case (intptr_t)SIG_IGN:
1913 			/*
1914 			 * Masking above should prevent us ever trying
1915 			 * to take action on an ignored signal other
1916 			 * than SIGCONT, unless process is traced.
1917 			 */
1918 			if ((prop & SA_CONT) == 0 &&
1919 			    (p->p_flag & P_TRACED) == 0)
1920 				kprintf("issignal\n");
1921 			break;		/* == ignore */
1922 
1923 		default:
1924 			/*
1925 			 * This signal has an action, let
1926 			 * postsig() process it.
1927 			 */
1928 			lwkt_reltoken(&p->p_token);
1929 			return (sig);
1930 		}
1931 		lwp_delsig(lp, sig);		/* take the signal! */
1932 	}
1933 	/* NOTREACHED */
1934 }
1935 
1936 /*
1937  * Take the action for the specified signal
1938  * from the current set of pending signals.
1939  *
1940  * Caller must hold p->p_token
1941  */
1942 void
1943 postsig(int sig)
1944 {
1945 	struct lwp *lp = curthread->td_lwp;
1946 	struct proc *p = lp->lwp_proc;
1947 	struct sigacts *ps = p->p_sigacts;
1948 	sig_t action;
1949 	sigset_t returnmask;
1950 	int code;
1951 
1952 	KASSERT(sig != 0, ("postsig"));
1953 
1954 	KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1955 
1956 	/*
1957 	 * If we are a virtual kernel running an emulated user process
1958 	 * context, switch back to the virtual kernel context before
1959 	 * trying to post the signal.
1960 	 */
1961 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1962 		struct trapframe *tf = lp->lwp_md.md_regs;
1963 		tf->tf_trapno = 0;
1964 		vkernel_trap(lp, tf);
1965 	}
1966 
1967 	lwp_delsig(lp, sig);
1968 	action = ps->ps_sigact[_SIG_IDX(sig)];
1969 #ifdef KTRACE
1970 	if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
1971 		ktrpsig(lp, sig, action, lp->lwp_flag & LWP_OLDMASK ?
1972 			&lp->lwp_oldsigmask : &lp->lwp_sigmask, 0);
1973 #endif
1974 	STOPEVENT(p, S_SIG, sig);
1975 
1976 	if (action == SIG_DFL) {
1977 		/*
1978 		 * Default action, where the default is to kill
1979 		 * the process.  (Other cases were ignored above.)
1980 		 */
1981 		sigexit(lp, sig);
1982 		/* NOTREACHED */
1983 	} else {
1984 		/*
1985 		 * If we get here, the signal must be caught.
1986 		 */
1987 		KASSERT(action != SIG_IGN && !SIGISMEMBER(lp->lwp_sigmask, sig),
1988 		    ("postsig action"));
1989 
1990 		crit_enter();
1991 
1992 		/*
1993 		 * Reset the signal handler if asked to
1994 		 */
1995 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
1996 			/*
1997 			 * See kern_sigaction() for origin of this code.
1998 			 */
1999 			SIGDELSET(p->p_sigcatch, sig);
2000 			if (sig != SIGCONT &&
2001 			    sigprop(sig) & SA_IGNORE)
2002 				SIGADDSET(p->p_sigignore, sig);
2003 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2004 		}
2005 
2006 		/*
2007 		 * Handle the mailbox case.  Copyout to the appropriate
2008 		 * location but do not generate a signal frame.  The system
2009 		 * call simply returns EINTR and the user is responsible for
2010 		 * polling the mailbox.
2011 		 */
2012 		if (SIGISMEMBER(ps->ps_sigmailbox, sig)) {
2013 			int sig_copy = sig;
2014 			copyout(&sig_copy, (void *)action, sizeof(int));
2015 			curproc->p_flag |= P_MAILBOX;
2016 			crit_exit();
2017 			goto done;
2018 		}
2019 
2020 		/*
2021 		 * Set the signal mask and calculate the mask to restore
2022 		 * when the signal function returns.
2023 		 *
2024 		 * Special case: user has done a sigsuspend.  Here the
2025 		 * current mask is not of interest, but rather the
2026 		 * mask from before the sigsuspend is what we want
2027 		 * restored after the signal processing is completed.
2028 		 */
2029 		if (lp->lwp_flag & LWP_OLDMASK) {
2030 			returnmask = lp->lwp_oldsigmask;
2031 			lp->lwp_flag &= ~LWP_OLDMASK;
2032 		} else {
2033 			returnmask = lp->lwp_sigmask;
2034 		}
2035 
2036 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
2037 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
2038 			SIGADDSET(lp->lwp_sigmask, sig);
2039 
2040 		crit_exit();
2041 		lp->lwp_ru.ru_nsignals++;
2042 		if (lp->lwp_sig != sig) {
2043 			code = 0;
2044 		} else {
2045 			code = lp->lwp_code;
2046 			lp->lwp_code = 0;
2047 			lp->lwp_sig = 0;
2048 		}
2049 		(*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
2050 	}
2051 done:
2052 	;
2053 }
2054 
2055 /*
2056  * Kill the current process for stated reason.
2057  */
2058 void
2059 killproc(struct proc *p, char *why)
2060 {
2061 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n",
2062 		p->p_pid, p->p_comm,
2063 		p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2064 	ksignal(p, SIGKILL);
2065 }
2066 
2067 /*
2068  * Force the current process to exit with the specified signal, dumping core
2069  * if appropriate.  We bypass the normal tests for masked and caught signals,
2070  * allowing unrecoverable failures to terminate the process without changing
2071  * signal state.  Mark the accounting record with the signal termination.
2072  * If dumping core, save the signal number for the debugger.  Calls exit and
2073  * does not return.
2074  *
2075  * This routine does not return.
2076  */
2077 void
2078 sigexit(struct lwp *lp, int sig)
2079 {
2080 	struct proc *p = lp->lwp_proc;
2081 
2082 	lwkt_gettoken(&p->p_token);
2083 	p->p_acflag |= AXSIG;
2084 	if (sigprop(sig) & SA_CORE) {
2085 		lp->lwp_sig = sig;
2086 		/*
2087 		 * Log signals which would cause core dumps
2088 		 * (Log as LOG_INFO to appease those who don't want
2089 		 * these messages.)
2090 		 * XXX : Todo, as well as euid, write out ruid too
2091 		 */
2092 		if (coredump(lp, sig) == 0)
2093 			sig |= WCOREFLAG;
2094 		if (kern_logsigexit)
2095 			log(LOG_INFO,
2096 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2097 			    p->p_pid, p->p_comm,
2098 			    p->p_ucred ? p->p_ucred->cr_uid : -1,
2099 			    sig &~ WCOREFLAG,
2100 			    sig & WCOREFLAG ? " (core dumped)" : "");
2101 	}
2102 	lwkt_reltoken(&p->p_token);
2103 	exit1(W_EXITCODE(0, sig));
2104 	/* NOTREACHED */
2105 }
2106 
2107 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
2108 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
2109 	      sizeof(corefilename), "process corefile name format string");
2110 
2111 /*
2112  * expand_name(name, uid, pid)
2113  * Expand the name described in corefilename, using name, uid, and pid.
2114  * corefilename is a kprintf-like string, with three format specifiers:
2115  *	%N	name of process ("name")
2116  *	%P	process id (pid)
2117  *	%U	user id (uid)
2118  * For example, "%N.core" is the default; they can be disabled completely
2119  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
2120  * This is controlled by the sysctl variable kern.corefile (see above).
2121  */
2122 
2123 static char *
2124 expand_name(const char *name, uid_t uid, pid_t pid)
2125 {
2126 	char *temp;
2127 	char buf[11];		/* Buffer for pid/uid -- max 4B */
2128 	int i, n;
2129 	char *format = corefilename;
2130 	size_t namelen;
2131 
2132 	temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
2133 	if (temp == NULL)
2134 		return NULL;
2135 	namelen = strlen(name);
2136 	for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
2137 		int l;
2138 		switch (format[i]) {
2139 		case '%':	/* Format character */
2140 			i++;
2141 			switch (format[i]) {
2142 			case '%':
2143 				temp[n++] = '%';
2144 				break;
2145 			case 'N':	/* process name */
2146 				if ((n + namelen) > MAXPATHLEN) {
2147 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2148 					    pid, name, uid, temp, name);
2149 					kfree(temp, M_TEMP);
2150 					return NULL;
2151 				}
2152 				memcpy(temp+n, name, namelen);
2153 				n += namelen;
2154 				break;
2155 			case 'P':	/* process id */
2156 				l = ksprintf(buf, "%u", pid);
2157 				if ((n + l) > MAXPATHLEN) {
2158 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2159 					    pid, name, uid, temp, name);
2160 					kfree(temp, M_TEMP);
2161 					return NULL;
2162 				}
2163 				memcpy(temp+n, buf, l);
2164 				n += l;
2165 				break;
2166 			case 'U':	/* user id */
2167 				l = ksprintf(buf, "%u", uid);
2168 				if ((n + l) > MAXPATHLEN) {
2169 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2170 					    pid, name, uid, temp, name);
2171 					kfree(temp, M_TEMP);
2172 					return NULL;
2173 				}
2174 				memcpy(temp+n, buf, l);
2175 				n += l;
2176 				break;
2177 			default:
2178 			  	log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
2179 			}
2180 			break;
2181 		default:
2182 			temp[n++] = format[i];
2183 		}
2184 	}
2185 	temp[n] = '\0';
2186 	return temp;
2187 }
2188 
2189 /*
2190  * Dump a process' core.  The main routine does some
2191  * policy checking, and creates the name of the coredump;
2192  * then it passes on a vnode and a size limit to the process-specific
2193  * coredump routine if there is one; if there _is not_ one, it returns
2194  * ENOSYS; otherwise it returns the error from the process-specific routine.
2195  *
2196  * The parameter `lp' is the lwp which triggered the coredump.
2197  */
2198 
2199 static int
2200 coredump(struct lwp *lp, int sig)
2201 {
2202 	struct proc *p = lp->lwp_proc;
2203 	struct vnode *vp;
2204 	struct ucred *cred = p->p_ucred;
2205 	struct flock lf;
2206 	struct nlookupdata nd;
2207 	struct vattr vattr;
2208 	int error, error1;
2209 	char *name;			/* name of corefile */
2210 	off_t limit;
2211 
2212 	STOPEVENT(p, S_CORE, 0);
2213 
2214 	if (((sugid_coredump == 0) && p->p_flag & P_SUGID) || do_coredump == 0)
2215 		return (EFAULT);
2216 
2217 	/*
2218 	 * Note that the bulk of limit checking is done after
2219 	 * the corefile is created.  The exception is if the limit
2220 	 * for corefiles is 0, in which case we don't bother
2221 	 * creating the corefile at all.  This layout means that
2222 	 * a corefile is truncated instead of not being created,
2223 	 * if it is larger than the limit.
2224 	 */
2225 	limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
2226 	if (limit == 0)
2227 		return EFBIG;
2228 
2229 	name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
2230 	if (name == NULL)
2231 		return (EINVAL);
2232 	error = nlookup_init(&nd, name, UIO_SYSSPACE, NLC_LOCKVP);
2233 	if (error == 0)
2234 		error = vn_open(&nd, NULL, O_CREAT | FWRITE | O_NOFOLLOW, S_IRUSR | S_IWUSR);
2235 	kfree(name, M_TEMP);
2236 	if (error) {
2237 		nlookup_done(&nd);
2238 		return (error);
2239 	}
2240 	vp = nd.nl_open_vp;
2241 	nd.nl_open_vp = NULL;
2242 	nlookup_done(&nd);
2243 
2244 	vn_unlock(vp);
2245 	lf.l_whence = SEEK_SET;
2246 	lf.l_start = 0;
2247 	lf.l_len = 0;
2248 	lf.l_type = F_WRLCK;
2249 	error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, 0);
2250 	if (error)
2251 		goto out2;
2252 
2253 	/* Don't dump to non-regular files or files with links. */
2254 	if (vp->v_type != VREG ||
2255 	    VOP_GETATTR(vp, &vattr) || vattr.va_nlink != 1) {
2256 		error = EFAULT;
2257 		goto out1;
2258 	}
2259 
2260 	/* Don't dump to files current user does not own */
2261 	if (vattr.va_uid != p->p_ucred->cr_uid) {
2262 		error = EFAULT;
2263 		goto out1;
2264 	}
2265 
2266 	VATTR_NULL(&vattr);
2267 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2268 	vattr.va_size = 0;
2269 	VOP_SETATTR(vp, &vattr, cred);
2270 	p->p_acflag |= ACORE;
2271 	vn_unlock(vp);
2272 
2273 	error = p->p_sysent->sv_coredump ?
2274 		  p->p_sysent->sv_coredump(lp, sig, vp, limit) : ENOSYS;
2275 
2276 out1:
2277 	lf.l_type = F_UNLCK;
2278 	VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, 0);
2279 out2:
2280 	error1 = vn_close(vp, FWRITE);
2281 	if (error == 0)
2282 		error = error1;
2283 	return (error);
2284 }
2285 
2286 /*
2287  * Nonexistent system call-- signal process (may want to handle it).
2288  * Flag error in case process won't see signal immediately (blocked or ignored).
2289  *
2290  * MPALMOSTSAFE
2291  */
2292 /* ARGSUSED */
2293 int
2294 sys_nosys(struct nosys_args *args)
2295 {
2296 	lwpsignal(curproc, curthread->td_lwp, SIGSYS);
2297 	return (EINVAL);
2298 }
2299 
2300 /*
2301  * Send a SIGIO or SIGURG signal to a process or process group using
2302  * stored credentials rather than those of the current process.
2303  */
2304 void
2305 pgsigio(struct sigio *sigio, int sig, int checkctty)
2306 {
2307 	if (sigio == NULL)
2308 		return;
2309 
2310 	if (sigio->sio_pgid > 0) {
2311 		if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
2312 		             sigio->sio_proc))
2313 			ksignal(sigio->sio_proc, sig);
2314 	} else if (sigio->sio_pgid < 0) {
2315 		struct proc *p;
2316 		struct pgrp *pg = sigio->sio_pgrp;
2317 
2318 		/*
2319 		 * Must interlock all signals against fork
2320 		 */
2321 		pgref(pg);
2322 		lockmgr(&pg->pg_lock, LK_EXCLUSIVE);
2323 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
2324 			if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
2325 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
2326 				ksignal(p, sig);
2327 		}
2328 		lockmgr(&pg->pg_lock, LK_RELEASE);
2329 		pgrel(pg);
2330 	}
2331 }
2332 
2333 static int
2334 filt_sigattach(struct knote *kn)
2335 {
2336 	struct proc *p = curproc;
2337 
2338 	kn->kn_ptr.p_proc = p;
2339 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
2340 
2341 	/* XXX lock the proc here while adding to the list? */
2342 	knote_insert(&p->p_klist, kn);
2343 
2344 	return (0);
2345 }
2346 
2347 static void
2348 filt_sigdetach(struct knote *kn)
2349 {
2350 	struct proc *p = kn->kn_ptr.p_proc;
2351 
2352 	knote_remove(&p->p_klist, kn);
2353 }
2354 
2355 /*
2356  * signal knotes are shared with proc knotes, so we apply a mask to
2357  * the hint in order to differentiate them from process hints.  This
2358  * could be avoided by using a signal-specific knote list, but probably
2359  * isn't worth the trouble.
2360  */
2361 static int
2362 filt_signal(struct knote *kn, long hint)
2363 {
2364 	if (hint & NOTE_SIGNAL) {
2365 		hint &= ~NOTE_SIGNAL;
2366 
2367 		if (kn->kn_id == hint)
2368 			kn->kn_data++;
2369 	}
2370 	return (kn->kn_data != 0);
2371 }
2372