xref: /dflybsd-src/sys/kern/kern_sig.c (revision c0d274d062fd959993bf623f25f7cb6a8a676c4e)
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/systm.h>
56 #include <sys/acct.h>
57 #include <sys/fcntl.h>
58 #include <sys/lock.h>
59 #include <sys/wait.h>
60 #include <sys/ktrace.h>
61 #include <sys/syslog.h>
62 #include <sys/stat.h>
63 #include <sys/sysent.h>
64 #include <sys/sysctl.h>
65 #include <sys/malloc.h>
66 #include <sys/interrupt.h>
67 #include <sys/unistd.h>
68 #include <sys/kern_syscall.h>
69 #include <sys/vkernel.h>
70 
71 #include <sys/signal2.h>
72 #include <sys/thread2.h>
73 
74 #include <machine/cpu.h>
75 #include <machine/smp.h>
76 
77 static int	coredump(struct lwp *, int);
78 static char	*expand_name(const char *, uid_t, pid_t);
79 static int	dokillpg(int sig, int pgid, int all);
80 static int	sig_ffs(sigset_t *set);
81 static int	sigprop(int sig);
82 static void	lwp_signotify(struct lwp *lp);
83 #ifdef SMP
84 static void	signotify_remote(void *arg);
85 #endif
86 static int	kern_sigtimedwait(sigset_t set, siginfo_t *info,
87 		    struct timespec *timeout);
88 
89 static int	filt_sigattach(struct knote *kn);
90 static void	filt_sigdetach(struct knote *kn);
91 static int	filt_signal(struct knote *kn, long hint);
92 
93 struct filterops sig_filtops =
94 	{ 0, filt_sigattach, filt_sigdetach, filt_signal };
95 
96 static int	kern_logsigexit = 1;
97 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
98     &kern_logsigexit, 0,
99     "Log processes quitting on abnormal signals to syslog(3)");
100 
101 /*
102  * Can process p, with pcred pc, send the signal sig to process q?
103  */
104 #define CANSIGNAL(q, sig) \
105 	(!p_trespass(curproc->p_ucred, (q)->p_ucred) || \
106 	((sig) == SIGCONT && (q)->p_session == curproc->p_session))
107 
108 /*
109  * Policy -- Can real uid ruid with ucred uc send a signal to process q?
110  */
111 #define CANSIGIO(ruid, uc, q) \
112 	((uc)->cr_uid == 0 || \
113 	    (ruid) == (q)->p_ucred->cr_ruid || \
114 	    (uc)->cr_uid == (q)->p_ucred->cr_ruid || \
115 	    (ruid) == (q)->p_ucred->cr_uid || \
116 	    (uc)->cr_uid == (q)->p_ucred->cr_uid)
117 
118 int sugid_coredump;
119 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RW,
120 	&sugid_coredump, 0, "Enable coredumping set user/group ID processes");
121 
122 static int	do_coredump = 1;
123 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
124 	&do_coredump, 0, "Enable/Disable coredumps");
125 
126 /*
127  * Signal properties and actions.
128  * The array below categorizes the signals and their default actions
129  * according to the following properties:
130  */
131 #define	SA_KILL		0x01		/* terminates process by default */
132 #define	SA_CORE		0x02		/* ditto and coredumps */
133 #define	SA_STOP		0x04		/* suspend process */
134 #define	SA_TTYSTOP	0x08		/* ditto, from tty */
135 #define	SA_IGNORE	0x10		/* ignore by default */
136 #define	SA_CONT		0x20		/* continue if suspended */
137 #define	SA_CANTMASK	0x40		/* non-maskable, catchable */
138 #define SA_CKPT         0x80            /* checkpoint process */
139 
140 
141 static int sigproptbl[NSIG] = {
142         SA_KILL,                /* SIGHUP */
143         SA_KILL,                /* SIGINT */
144         SA_KILL|SA_CORE,        /* SIGQUIT */
145         SA_KILL|SA_CORE,        /* SIGILL */
146         SA_KILL|SA_CORE,        /* SIGTRAP */
147         SA_KILL|SA_CORE,        /* SIGABRT */
148         SA_KILL|SA_CORE,        /* SIGEMT */
149         SA_KILL|SA_CORE,        /* SIGFPE */
150         SA_KILL,                /* SIGKILL */
151         SA_KILL|SA_CORE,        /* SIGBUS */
152         SA_KILL|SA_CORE,        /* SIGSEGV */
153         SA_KILL|SA_CORE,        /* SIGSYS */
154         SA_KILL,                /* SIGPIPE */
155         SA_KILL,                /* SIGALRM */
156         SA_KILL,                /* SIGTERM */
157         SA_IGNORE,              /* SIGURG */
158         SA_STOP,                /* SIGSTOP */
159         SA_STOP|SA_TTYSTOP,     /* SIGTSTP */
160         SA_IGNORE|SA_CONT,      /* SIGCONT */
161         SA_IGNORE,              /* SIGCHLD */
162         SA_STOP|SA_TTYSTOP,     /* SIGTTIN */
163         SA_STOP|SA_TTYSTOP,     /* SIGTTOU */
164         SA_IGNORE,              /* SIGIO */
165         SA_KILL,                /* SIGXCPU */
166         SA_KILL,                /* SIGXFSZ */
167         SA_KILL,                /* SIGVTALRM */
168         SA_KILL,                /* SIGPROF */
169         SA_IGNORE,              /* SIGWINCH  */
170         SA_IGNORE,              /* SIGINFO */
171         SA_KILL,                /* SIGUSR1 */
172         SA_KILL,                /* SIGUSR2 */
173 	SA_IGNORE,              /* SIGTHR */
174 	SA_CKPT,                /* SIGCKPT */
175 	SA_KILL|SA_CKPT,        /* SIGCKPTEXIT */
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 	SA_IGNORE,
206 
207 };
208 
209 static __inline int
210 sigprop(int sig)
211 {
212 
213 	if (sig > 0 && sig < NSIG)
214 		return (sigproptbl[_SIG_IDX(sig)]);
215 	return (0);
216 }
217 
218 static __inline int
219 sig_ffs(sigset_t *set)
220 {
221 	int i;
222 
223 	for (i = 0; i < _SIG_WORDS; i++)
224 		if (set->__bits[i])
225 			return (ffs(set->__bits[i]) + (i * 32));
226 	return (0);
227 }
228 
229 /*
230  * No requirements.
231  */
232 int
233 kern_sigaction(int sig, struct sigaction *act, struct sigaction *oact)
234 {
235 	struct thread *td = curthread;
236 	struct proc *p = td->td_proc;
237 	struct lwp *lp;
238 	struct sigacts *ps = p->p_sigacts;
239 
240 	if (sig <= 0 || sig > _SIG_MAXSIG)
241 		return (EINVAL);
242 
243 	lwkt_gettoken(&p->p_token);
244 
245 	if (oact) {
246 		oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
247 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
248 		oact->sa_flags = 0;
249 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
250 			oact->sa_flags |= SA_ONSTACK;
251 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
252 			oact->sa_flags |= SA_RESTART;
253 		if (SIGISMEMBER(ps->ps_sigreset, sig))
254 			oact->sa_flags |= SA_RESETHAND;
255 		if (SIGISMEMBER(ps->ps_signodefer, sig))
256 			oact->sa_flags |= SA_NODEFER;
257 		if (SIGISMEMBER(ps->ps_siginfo, sig))
258 			oact->sa_flags |= SA_SIGINFO;
259 		if (SIGISMEMBER(ps->ps_sigmailbox, sig))
260 			oact->sa_flags |= SA_MAILBOX;
261 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDSTOP)
262 			oact->sa_flags |= SA_NOCLDSTOP;
263 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDWAIT)
264 			oact->sa_flags |= SA_NOCLDWAIT;
265 	}
266 	if (act) {
267 		/*
268 		 * Check for invalid requests.  KILL and STOP cannot be
269 		 * caught.
270 		 */
271 		if (sig == SIGKILL || sig == SIGSTOP) {
272 			if (act->sa_handler != SIG_DFL) {
273 				lwkt_reltoken(&p->p_token);
274 				return (EINVAL);
275 			}
276 #if 0
277 			/* (not needed, SIG_DFL forces action to occur) */
278 			if (act->sa_flags & SA_MAILBOX) {
279 				lwkt_reltoken(&p->p_token);
280 				return (EINVAL);
281 			}
282 #endif
283 		}
284 
285 		/*
286 		 * Change setting atomically.
287 		 */
288 		crit_enter();
289 
290 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
291 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
292 		if (act->sa_flags & SA_SIGINFO) {
293 			ps->ps_sigact[_SIG_IDX(sig)] =
294 			    (__sighandler_t *)act->sa_sigaction;
295 			SIGADDSET(ps->ps_siginfo, sig);
296 		} else {
297 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
298 			SIGDELSET(ps->ps_siginfo, sig);
299 		}
300 		if (!(act->sa_flags & SA_RESTART))
301 			SIGADDSET(ps->ps_sigintr, sig);
302 		else
303 			SIGDELSET(ps->ps_sigintr, sig);
304 		if (act->sa_flags & SA_ONSTACK)
305 			SIGADDSET(ps->ps_sigonstack, sig);
306 		else
307 			SIGDELSET(ps->ps_sigonstack, sig);
308 		if (act->sa_flags & SA_RESETHAND)
309 			SIGADDSET(ps->ps_sigreset, sig);
310 		else
311 			SIGDELSET(ps->ps_sigreset, sig);
312 		if (act->sa_flags & SA_NODEFER)
313 			SIGADDSET(ps->ps_signodefer, sig);
314 		else
315 			SIGDELSET(ps->ps_signodefer, sig);
316 		if (act->sa_flags & SA_MAILBOX)
317 			SIGADDSET(ps->ps_sigmailbox, sig);
318 		else
319 			SIGDELSET(ps->ps_sigmailbox, sig);
320 		if (sig == SIGCHLD) {
321 			if (act->sa_flags & SA_NOCLDSTOP)
322 				p->p_sigacts->ps_flag |= PS_NOCLDSTOP;
323 			else
324 				p->p_sigacts->ps_flag &= ~PS_NOCLDSTOP;
325 			if (act->sa_flags & SA_NOCLDWAIT) {
326 				/*
327 				 * Paranoia: since SA_NOCLDWAIT is implemented
328 				 * by reparenting the dying child to PID 1 (and
329 				 * trust it to reap the zombie), PID 1 itself
330 				 * is forbidden to set SA_NOCLDWAIT.
331 				 */
332 				if (p->p_pid == 1)
333 					p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
334 				else
335 					p->p_sigacts->ps_flag |= PS_NOCLDWAIT;
336 			} else {
337 				p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
338 			}
339 		}
340 		/*
341 		 * Set bit in p_sigignore for signals that are set to SIG_IGN,
342 		 * and for signals set to SIG_DFL where the default is to
343 		 * ignore. However, don't put SIGCONT in p_sigignore, as we
344 		 * have to restart the process.
345 		 */
346 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
347 		    (sigprop(sig) & SA_IGNORE &&
348 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
349 			/* never to be seen again */
350 			SIGDELSET(p->p_siglist, sig);
351 			/*
352 			 * Remove the signal also from the thread lists.
353 			 */
354 			FOREACH_LWP_IN_PROC(lp, p) {
355 				SIGDELSET(lp->lwp_siglist, sig);
356 			}
357 			if (sig != SIGCONT) {
358 				/* easier in ksignal */
359 				SIGADDSET(p->p_sigignore, sig);
360 			}
361 			SIGDELSET(p->p_sigcatch, sig);
362 		} else {
363 			SIGDELSET(p->p_sigignore, sig);
364 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
365 				SIGDELSET(p->p_sigcatch, sig);
366 			else
367 				SIGADDSET(p->p_sigcatch, sig);
368 		}
369 
370 		crit_exit();
371 	}
372 	lwkt_reltoken(&p->p_token);
373 	return (0);
374 }
375 
376 int
377 sys_sigaction(struct sigaction_args *uap)
378 {
379 	struct sigaction act, oact;
380 	struct sigaction *actp, *oactp;
381 	int error;
382 
383 	actp = (uap->act != NULL) ? &act : NULL;
384 	oactp = (uap->oact != NULL) ? &oact : NULL;
385 	if (actp) {
386 		error = copyin(uap->act, actp, sizeof(act));
387 		if (error)
388 			return (error);
389 	}
390 	error = kern_sigaction(uap->sig, actp, oactp);
391 	if (oactp && !error) {
392 		error = copyout(oactp, uap->oact, sizeof(oact));
393 	}
394 	return (error);
395 }
396 
397 /*
398  * Initialize signal state for process 0;
399  * set to ignore signals that are ignored by default.
400  */
401 void
402 siginit(struct proc *p)
403 {
404 	int i;
405 
406 	for (i = 1; i <= NSIG; i++)
407 		if (sigprop(i) & SA_IGNORE && i != SIGCONT)
408 			SIGADDSET(p->p_sigignore, i);
409 }
410 
411 /*
412  * Reset signals for an exec of the specified process.
413  */
414 void
415 execsigs(struct proc *p)
416 {
417 	struct sigacts *ps = p->p_sigacts;
418 	struct lwp *lp;
419 	int sig;
420 
421 	lp = ONLY_LWP_IN_PROC(p);
422 
423 	/*
424 	 * Reset caught signals.  Held signals remain held
425 	 * through p_sigmask (unless they were caught,
426 	 * and are now ignored by default).
427 	 */
428 	while (SIGNOTEMPTY(p->p_sigcatch)) {
429 		sig = sig_ffs(&p->p_sigcatch);
430 		SIGDELSET(p->p_sigcatch, sig);
431 		if (sigprop(sig) & SA_IGNORE) {
432 			if (sig != SIGCONT)
433 				SIGADDSET(p->p_sigignore, sig);
434 			SIGDELSET(p->p_siglist, sig);
435 			SIGDELSET(lp->lwp_siglist, sig);
436 		}
437 		ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
438 	}
439 
440 	/*
441 	 * Reset stack state to the user stack.
442 	 * Clear set of signals caught on the signal stack.
443 	 */
444 	lp->lwp_sigstk.ss_flags = SS_DISABLE;
445 	lp->lwp_sigstk.ss_size = 0;
446 	lp->lwp_sigstk.ss_sp = 0;
447 	lp->lwp_flag &= ~LWP_ALTSTACK;
448 	/*
449 	 * Reset no zombies if child dies flag as Solaris does.
450 	 */
451 	p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
452 }
453 
454 /*
455  * kern_sigprocmask() - MP SAFE ONLY IF p == curproc
456  *
457  *	Manipulate signal mask.  This routine is MP SAFE *ONLY* if
458  *	p == curproc.
459  */
460 int
461 kern_sigprocmask(int how, sigset_t *set, sigset_t *oset)
462 {
463 	struct thread *td = curthread;
464 	struct lwp *lp = td->td_lwp;
465 	struct proc *p = td->td_proc;
466 	int error;
467 
468 	lwkt_gettoken(&p->p_token);
469 
470 	if (oset != NULL)
471 		*oset = lp->lwp_sigmask;
472 
473 	error = 0;
474 	if (set != NULL) {
475 		switch (how) {
476 		case SIG_BLOCK:
477 			SIG_CANTMASK(*set);
478 			SIGSETOR(lp->lwp_sigmask, *set);
479 			break;
480 		case SIG_UNBLOCK:
481 			SIGSETNAND(lp->lwp_sigmask, *set);
482 			break;
483 		case SIG_SETMASK:
484 			SIG_CANTMASK(*set);
485 			lp->lwp_sigmask = *set;
486 			break;
487 		default:
488 			error = EINVAL;
489 			break;
490 		}
491 	}
492 
493 	lwkt_reltoken(&p->p_token);
494 
495 	return (error);
496 }
497 
498 /*
499  * sigprocmask()
500  *
501  * MPSAFE
502  */
503 int
504 sys_sigprocmask(struct sigprocmask_args *uap)
505 {
506 	sigset_t set, oset;
507 	sigset_t *setp, *osetp;
508 	int error;
509 
510 	setp = (uap->set != NULL) ? &set : NULL;
511 	osetp = (uap->oset != NULL) ? &oset : NULL;
512 	if (setp) {
513 		error = copyin(uap->set, setp, sizeof(set));
514 		if (error)
515 			return (error);
516 	}
517 	error = kern_sigprocmask(uap->how, setp, osetp);
518 	if (osetp && !error) {
519 		error = copyout(osetp, uap->oset, sizeof(oset));
520 	}
521 	return (error);
522 }
523 
524 /*
525  * MPSAFE
526  */
527 int
528 kern_sigpending(struct __sigset *set)
529 {
530 	struct lwp *lp = curthread->td_lwp;
531 
532 	*set = lwp_sigpend(lp);
533 
534 	return (0);
535 }
536 
537 /*
538  * MPSAFE
539  */
540 int
541 sys_sigpending(struct sigpending_args *uap)
542 {
543 	sigset_t set;
544 	int error;
545 
546 	error = kern_sigpending(&set);
547 
548 	if (error == 0)
549 		error = copyout(&set, uap->set, sizeof(set));
550 	return (error);
551 }
552 
553 /*
554  * Suspend process until signal, providing mask to be set
555  * in the meantime.
556  *
557  * MPSAFE
558  */
559 int
560 kern_sigsuspend(struct __sigset *set)
561 {
562 	struct thread *td = curthread;
563 	struct lwp *lp = td->td_lwp;
564 	struct proc *p = td->td_proc;
565 	struct sigacts *ps = p->p_sigacts;
566 
567 	/*
568 	 * When returning from sigsuspend, we want
569 	 * the old mask to be restored after the
570 	 * signal handler has finished.  Thus, we
571 	 * save it here and mark the sigacts structure
572 	 * to indicate this.
573 	 */
574 	lp->lwp_oldsigmask = lp->lwp_sigmask;
575 	lp->lwp_flag |= LWP_OLDMASK;
576 
577 	SIG_CANTMASK(*set);
578 	lp->lwp_sigmask = *set;
579 	while (tsleep(ps, PCATCH, "pause", 0) == 0)
580 		/* void */;
581 	/* always return EINTR rather than ERESTART... */
582 	return (EINTR);
583 }
584 
585 /*
586  * Note nonstandard calling convention: libc stub passes mask, not
587  * pointer, to save a copyin.
588  *
589  * MPSAFE
590  */
591 int
592 sys_sigsuspend(struct sigsuspend_args *uap)
593 {
594 	sigset_t mask;
595 	int error;
596 
597 	error = copyin(uap->sigmask, &mask, sizeof(mask));
598 	if (error)
599 		return (error);
600 
601 	error = kern_sigsuspend(&mask);
602 
603 	return (error);
604 }
605 
606 /*
607  * MPSAFE
608  */
609 int
610 kern_sigaltstack(struct sigaltstack *ss, struct sigaltstack *oss)
611 {
612 	struct thread *td = curthread;
613 	struct lwp *lp = td->td_lwp;
614 	struct proc *p = td->td_proc;
615 
616 	if ((lp->lwp_flag & LWP_ALTSTACK) == 0)
617 		lp->lwp_sigstk.ss_flags |= SS_DISABLE;
618 
619 	if (oss)
620 		*oss = lp->lwp_sigstk;
621 
622 	if (ss) {
623 		if (ss->ss_flags & SS_DISABLE) {
624 			if (lp->lwp_sigstk.ss_flags & SS_ONSTACK)
625 				return (EINVAL);
626 			lp->lwp_flag &= ~LWP_ALTSTACK;
627 			lp->lwp_sigstk.ss_flags = ss->ss_flags;
628 		} else {
629 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
630 				return (ENOMEM);
631 			lp->lwp_flag |= LWP_ALTSTACK;
632 			lp->lwp_sigstk = *ss;
633 		}
634 	}
635 
636 	return (0);
637 }
638 
639 /*
640  * MPSAFE
641  */
642 int
643 sys_sigaltstack(struct sigaltstack_args *uap)
644 {
645 	stack_t ss, oss;
646 	int error;
647 
648 	if (uap->ss) {
649 		error = copyin(uap->ss, &ss, sizeof(ss));
650 		if (error)
651 			return (error);
652 	}
653 
654 	error = kern_sigaltstack(uap->ss ? &ss : NULL,
655 	    uap->oss ? &oss : NULL);
656 
657 	if (error == 0 && uap->oss)
658 		error = copyout(&oss, uap->oss, sizeof(*uap->oss));
659 	return (error);
660 }
661 
662 /*
663  * Common code for kill process group/broadcast kill.
664  * cp is calling process.
665  */
666 struct killpg_info {
667 	int nfound;
668 	int sig;
669 };
670 
671 static int killpg_all_callback(struct proc *p, void *data);
672 
673 static int
674 dokillpg(int sig, int pgid, int all)
675 {
676 	struct killpg_info info;
677 	struct proc *cp = curproc;
678 	struct proc *p;
679 	struct pgrp *pgrp;
680 
681 	info.nfound = 0;
682 	info.sig = sig;
683 
684 	if (all) {
685 		/*
686 		 * broadcast
687 		 */
688 		allproc_scan(killpg_all_callback, &info);
689 	} else {
690 		if (pgid == 0) {
691 			/*
692 			 * zero pgid means send to my process group.
693 			 */
694 			pgrp = cp->p_pgrp;
695 			pgref(pgrp);
696 		} else {
697 			pgrp = pgfind(pgid);
698 			if (pgrp == NULL)
699 				return (ESRCH);
700 		}
701 
702 		/*
703 		 * Must interlock all signals against fork
704 		 */
705 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
706 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
707 			if (p->p_pid <= 1 ||
708 			    p->p_stat == SZOMB ||
709 			    (p->p_flag & P_SYSTEM) ||
710 			    !CANSIGNAL(p, sig)) {
711 				continue;
712 			}
713 			++info.nfound;
714 			if (sig)
715 				ksignal(p, sig);
716 		}
717 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
718 		pgrel(pgrp);
719 	}
720 	return (info.nfound ? 0 : ESRCH);
721 }
722 
723 static int
724 killpg_all_callback(struct proc *p, void *data)
725 {
726 	struct killpg_info *info = data;
727 
728 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) ||
729 	    p == curproc || !CANSIGNAL(p, info->sig)) {
730 		return (0);
731 	}
732 	++info->nfound;
733 	if (info->sig)
734 		ksignal(p, info->sig);
735 	return(0);
736 }
737 
738 /*
739  * Send a general signal to a process or LWPs within that process.  Note
740  * that new signals cannot be sent if a process is exiting.
741  *
742  * No requirements.
743  */
744 int
745 kern_kill(int sig, pid_t pid, lwpid_t tid)
746 {
747 	int t;
748 
749 	if ((u_int)sig > _SIG_MAXSIG)
750 		return (EINVAL);
751 
752 	lwkt_gettoken(&proc_token);
753 
754 	if (pid > 0) {
755 		struct proc *p;
756 		struct lwp *lp = NULL;
757 
758 		/* kill single process */
759 		if ((p = pfind(pid)) == NULL) {
760 			lwkt_reltoken(&proc_token);
761 			return (ESRCH);
762 		}
763 		lwkt_gettoken(&p->p_token);
764 		if (!CANSIGNAL(p, sig)) {
765 			lwkt_reltoken(&p->p_token);
766 			PRELE(p);
767 			lwkt_reltoken(&proc_token);
768 			return (EPERM);
769 		}
770 
771 		/*
772 		 * NOP if the process is exiting.  Note that lwpsignal() is
773 		 * called directly with P_WEXIT set to kill individual LWPs
774 		 * during exit, which is allowed.
775 		 */
776 		if (p->p_flag & P_WEXIT) {
777 			lwkt_reltoken(&p->p_token);
778 			PRELE(p);
779 			lwkt_reltoken(&proc_token);
780 			return (0);
781 		}
782 		if (tid != -1) {
783 			lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, tid);
784 			if (lp == NULL) {
785 				lwkt_reltoken(&p->p_token);
786 				PRELE(p);
787 				lwkt_reltoken(&proc_token);
788 				return (ESRCH);
789 			}
790 		}
791 		if (sig)
792 			lwpsignal(p, lp, sig);
793 		lwkt_reltoken(&p->p_token);
794 		PRELE(p);
795 		lwkt_reltoken(&proc_token);
796 		return (0);
797 	}
798 
799 	/*
800 	 * If we come here, pid is a special broadcast pid.
801 	 * This doesn't mix with a tid.
802 	 */
803 	if (tid != -1) {
804 		lwkt_reltoken(&proc_token);
805 		return (EINVAL);
806 	}
807 	switch (pid) {
808 	case -1:		/* broadcast signal */
809 		t = (dokillpg(sig, 0, 1));
810 		break;
811 	case 0:			/* signal own process group */
812 		t = (dokillpg(sig, 0, 0));
813 		break;
814 	default:		/* negative explicit process group */
815 		t = (dokillpg(sig, -pid, 0));
816 		break;
817 	}
818 	lwkt_reltoken(&proc_token);
819 	return t;
820 }
821 
822 int
823 sys_kill(struct kill_args *uap)
824 {
825 	int error;
826 
827 	error = kern_kill(uap->signum, uap->pid, -1);
828 	return (error);
829 }
830 
831 int
832 sys_lwp_kill(struct lwp_kill_args *uap)
833 {
834 	int error;
835 	pid_t pid = uap->pid;
836 
837 	/*
838 	 * A tid is mandatory for lwp_kill(), otherwise
839 	 * you could simply use kill().
840 	 */
841 	if (uap->tid == -1)
842 		return (EINVAL);
843 
844 	/*
845 	 * To save on a getpid() function call for intra-process
846 	 * signals, pid == -1 means current process.
847 	 */
848 	if (pid == -1)
849 		pid = curproc->p_pid;
850 
851 	error = kern_kill(uap->signum, pid, uap->tid);
852 	return (error);
853 }
854 
855 /*
856  * Send a signal to a process group.
857  */
858 void
859 gsignal(int pgid, int sig)
860 {
861 	struct pgrp *pgrp;
862 
863 	if (pgid && (pgrp = pgfind(pgid)))
864 		pgsignal(pgrp, sig, 0);
865 }
866 
867 /*
868  * Send a signal to a process group.  If checktty is 1,
869  * limit to members which have a controlling terminal.
870  *
871  * pg_lock interlocks against a fork that might be in progress, to
872  * ensure that the new child process picks up the signal.
873  */
874 void
875 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
876 {
877 	struct proc *p;
878 
879 	/*
880 	 * Must interlock all signals against fork
881 	 */
882 	if (pgrp) {
883 		pgref(pgrp);
884 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
885 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
886 			if (checkctty == 0 || p->p_flag & P_CONTROLT)
887 				ksignal(p, sig);
888 		}
889 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
890 		pgrel(pgrp);
891 	}
892 }
893 
894 /*
895  * Send a signal caused by a trap to the current lwp.  If it will be caught
896  * immediately, deliver it with correct code.  Otherwise, post it normally.
897  *
898  * These signals may ONLY be delivered to the specified lwp and may never
899  * be delivered to the process generically.
900  */
901 void
902 trapsignal(struct lwp *lp, int sig, u_long code)
903 {
904 	struct proc *p = lp->lwp_proc;
905 	struct sigacts *ps = p->p_sigacts;
906 
907 	/*
908 	 * If we are a virtual kernel running an emulated user process
909 	 * context, switch back to the virtual kernel context before
910 	 * trying to post the signal.
911 	 */
912 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
913 		struct trapframe *tf = lp->lwp_md.md_regs;
914 		tf->tf_trapno = 0;
915 		vkernel_trap(lp, tf);
916 	}
917 
918 
919 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(p->p_sigcatch, sig) &&
920 	    !SIGISMEMBER(lp->lwp_sigmask, sig)) {
921 		lp->lwp_ru.ru_nsignals++;
922 #ifdef KTRACE
923 		if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
924 			ktrpsig(lp, sig, ps->ps_sigact[_SIG_IDX(sig)],
925 				&lp->lwp_sigmask, code);
926 #endif
927 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], sig,
928 						&lp->lwp_sigmask, code);
929 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
930 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
931 			SIGADDSET(lp->lwp_sigmask, sig);
932 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
933 			/*
934 			 * See kern_sigaction() for origin of this code.
935 			 */
936 			SIGDELSET(p->p_sigcatch, sig);
937 			if (sig != SIGCONT &&
938 			    sigprop(sig) & SA_IGNORE)
939 				SIGADDSET(p->p_sigignore, sig);
940 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
941 		}
942 	} else {
943 		lp->lwp_code = code;	/* XXX for core dump/debugger */
944 		lp->lwp_sig = sig;	/* XXX to verify code */
945 		lwpsignal(p, lp, sig);
946 	}
947 }
948 
949 /*
950  * Find a suitable lwp to deliver the signal to.  Returns NULL if all
951  * lwps hold the signal blocked.
952  *
953  * Caller must hold p->p_token.
954  */
955 static struct lwp *
956 find_lwp_for_signal(struct proc *p, int sig)
957 {
958 	struct lwp *lp;
959 	struct lwp *run, *sleep, *stop;
960 
961 	/*
962 	 * If the running/preempted thread belongs to the proc to which
963 	 * the signal is being delivered and this thread does not block
964 	 * the signal, then we can avoid a context switch by delivering
965 	 * the signal to this thread, because it will return to userland
966 	 * soon anyways.
967 	 */
968 	lp = lwkt_preempted_proc();
969 	if (lp != NULL && lp->lwp_proc == p &&
970 	    !SIGISMEMBER(lp->lwp_sigmask, sig)) {
971 		return (lp);
972 	}
973 
974 	run = sleep = stop = NULL;
975 	FOREACH_LWP_IN_PROC(lp, p) {
976 		/*
977 		 * If the signal is being blocked by the lwp, then this
978 		 * lwp is not eligible for receiving the signal.
979 		 */
980 		if (SIGISMEMBER(lp->lwp_sigmask, sig))
981 			continue;
982 
983 		switch (lp->lwp_stat) {
984 		case LSRUN:
985 			run = lp;
986 			break;
987 
988 		case LSSTOP:
989 			stop = lp;
990 			break;
991 
992 		case LSSLEEP:
993 			if (lp->lwp_flag & LWP_SINTR)
994 				sleep = lp;
995 			break;
996 		}
997 	}
998 
999 	if (run != NULL)
1000 		return (run);
1001 	else if (sleep != NULL)
1002 		return (sleep);
1003 	else
1004 		return (stop);
1005 }
1006 
1007 /*
1008  * Send the signal to the process.  If the signal has an action, the action
1009  * is usually performed by the target process rather than the caller; we add
1010  * the signal to the set of pending signals for the process.
1011  *
1012  * Exceptions:
1013  *   o When a stop signal is sent to a sleeping process that takes the
1014  *     default action, the process is stopped without awakening it.
1015  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
1016  *     regardless of the signal action (eg, blocked or ignored).
1017  *
1018  * Other ignored signals are discarded immediately.
1019  *
1020  * If the caller wishes to call this function from a hard code section the
1021  * caller must already hold p->p_token (see kern_clock.c).
1022  *
1023  * No requirements.
1024  */
1025 void
1026 ksignal(struct proc *p, int sig)
1027 {
1028 	lwpsignal(p, NULL, sig);
1029 }
1030 
1031 /*
1032  * The core for ksignal.  lp may be NULL, then a suitable thread
1033  * will be chosen.  If not, lp MUST be a member of p.
1034  *
1035  * If the caller wishes to call this function from a hard code section the
1036  * caller must already hold p->p_token.
1037  *
1038  * No requirements.
1039  */
1040 void
1041 lwpsignal(struct proc *p, struct lwp *lp, int sig)
1042 {
1043 	int prop;
1044 	sig_t action;
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 			lwkt_reltoken(&p->p_token);
1084 			PRELE(p);
1085 			return;
1086 		}
1087 		if (SIGISMEMBER(p->p_sigcatch, sig))
1088 			action = SIG_CATCH;
1089 		else
1090 			action = SIG_DFL;
1091 	}
1092 
1093 	/*
1094 	 * If continuing, clear any pending STOP signals.
1095 	 */
1096 	if (prop & SA_CONT)
1097 		SIG_STOPSIGMASK(p->p_siglist);
1098 
1099 	if (prop & SA_STOP) {
1100 		/*
1101 		 * If sending a tty stop signal to a member of an orphaned
1102 		 * process group, discard the signal here if the action
1103 		 * is default; don't stop the process below if sleeping,
1104 		 * and don't clear any pending SIGCONT.
1105 		 */
1106 		if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0 &&
1107 		    action == SIG_DFL) {
1108 			lwkt_reltoken(&p->p_token);
1109 			PRELE(p);
1110 		        return;
1111 		}
1112 		SIG_CONTSIGMASK(p->p_siglist);
1113 		p->p_flag &= ~P_CONTINUED;
1114 	}
1115 
1116 	crit_enter();
1117 
1118 	if (p->p_stat == SSTOP) {
1119 		/*
1120 		 * Nobody can handle this signal, add it to the lwp or
1121 		 * process pending list
1122 		 */
1123 		if (lp)
1124 			SIGADDSET(lp->lwp_siglist, sig);
1125 		else
1126 			SIGADDSET(p->p_siglist, sig);
1127 
1128 		/*
1129 		 * If the process is stopped and is being traced, then no
1130 		 * further action is necessary.
1131 		 */
1132 		if (p->p_flag & P_TRACED)
1133 			goto out;
1134 
1135 		/*
1136 		 * If the process is stopped and receives a KILL signal,
1137 		 * make the process runnable.
1138 		 */
1139 		if (sig == SIGKILL) {
1140 			proc_unstop(p);
1141 			goto active_process;
1142 		}
1143 
1144 		/*
1145 		 * If the process is stopped and receives a CONT signal,
1146 		 * then try to make the process runnable again.
1147 		 */
1148 		if (prop & SA_CONT) {
1149 			/*
1150 			 * If SIGCONT is default (or ignored), we continue the
1151 			 * process but don't leave the signal in p_siglist, as
1152 			 * it has no further action.  If SIGCONT is held, we
1153 			 * continue the process and leave the signal in
1154 			 * p_siglist.  If the process catches SIGCONT, let it
1155 			 * handle the signal itself.
1156 			 */
1157 			/* XXX what if the signal is being held blocked? */
1158 			p->p_flag |= P_CONTINUED;
1159 			wakeup(p->p_pptr);
1160 			if (action == SIG_DFL)
1161 				SIGDELSET(p->p_siglist, sig);
1162 			proc_unstop(p);
1163 			if (action == SIG_CATCH)
1164 				goto active_process;
1165 			goto out;
1166 		}
1167 
1168 		/*
1169 		 * If the process is stopped and receives another STOP
1170 		 * signal, we do not need to stop it again.  If we did
1171 		 * the shell could get confused.
1172 		 *
1173 		 * However, if the current/preempted lwp is part of the
1174 		 * process receiving the signal, we need to keep it,
1175 		 * so that this lwp can stop in issignal() later, as
1176 		 * we don't want to wait until it reaches userret!
1177 		 */
1178 		if (prop & SA_STOP) {
1179 			if (lwkt_preempted_proc() == NULL ||
1180 			    lwkt_preempted_proc()->lwp_proc != p)
1181 				SIGDELSET(p->p_siglist, sig);
1182 		}
1183 
1184 		/*
1185 		 * Otherwise the process is stopped and it received some
1186 		 * signal, which does not change its stopped state.
1187 		 *
1188 		 * We have to select one thread to set LWP_BREAKTSLEEP,
1189 		 * so that the current signal will break the sleep
1190 		 * as soon as a SA_CONT signal will unstop the process.
1191 		 */
1192 		if (lp == NULL)
1193 			lp = find_lwp_for_signal(p, sig);
1194 		if (lp != NULL &&
1195 		    (lp->lwp_stat == LSSLEEP || lp->lwp_stat == LSSTOP))
1196 			lp->lwp_flag |= LWP_BREAKTSLEEP;
1197 		goto out;
1198 
1199 		/* NOTREACHED */
1200 	}
1201 	/* else not stopped */
1202 active_process:
1203 
1204 	/*
1205 	 * Never deliver a lwp-specific signal to a random lwp.
1206 	 */
1207 	if (lp == NULL) {
1208 		lp = find_lwp_for_signal(p, sig);
1209 		if (lp && SIGISMEMBER(lp->lwp_sigmask, sig))
1210 			lp = NULL;
1211 	}
1212 
1213 	/*
1214 	 * Deliver to the process generically if (1) the signal is being
1215 	 * sent to any thread or (2) we could not find a thread to deliver
1216 	 * it to.
1217 	 */
1218 	if (lp == NULL) {
1219 		SIGADDSET(p->p_siglist, sig);
1220 		goto out;
1221 	}
1222 
1223 	/*
1224 	 * Deliver to a specific LWP whether it masks it or not.  It will
1225 	 * not be dispatched if masked but we must still deliver it.
1226 	 */
1227 	if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
1228 	    (p->p_flag & P_TRACED) == 0) {
1229 		p->p_nice = NZERO;
1230 	}
1231 
1232 	/*
1233 	 * If the process receives a STOP signal which indeed needs to
1234 	 * stop the process, do so.  If the process chose to catch the
1235 	 * signal, it will be treated like any other signal.
1236 	 */
1237 	if ((prop & SA_STOP) && action == SIG_DFL) {
1238 		/*
1239 		 * If a child holding parent blocked, stopping
1240 		 * could cause deadlock.  Take no action at this
1241 		 * time.
1242 		 */
1243 		if (p->p_flag & P_PPWAIT) {
1244 			SIGADDSET(p->p_siglist, sig);
1245 			goto out;
1246 		}
1247 
1248 		/*
1249 		 * Do not actually try to manipulate the process, but simply
1250 		 * stop it.  Lwps will stop as soon as they safely can.
1251 		 */
1252 		p->p_xstat = sig;
1253 		proc_stop(p);
1254 		goto out;
1255 	}
1256 
1257 	/*
1258 	 * If it is a CONT signal with default action, just ignore it.
1259 	 */
1260 	if ((prop & SA_CONT) && action == SIG_DFL)
1261 		goto out;
1262 
1263 	/*
1264 	 * Mark signal pending at this specific thread.
1265 	 */
1266 	SIGADDSET(lp->lwp_siglist, sig);
1267 
1268 	lwp_signotify(lp);
1269 
1270 out:
1271 	lwkt_reltoken(&p->p_token);
1272 	PRELE(p);
1273 	crit_exit();
1274 }
1275 
1276 /*
1277  * p->p_token must be held
1278  */
1279 static void
1280 lwp_signotify(struct lwp *lp)
1281 {
1282 	ASSERT_LWKT_TOKEN_HELD(&lp->lwp_proc->p_token);
1283 	crit_enter();
1284 
1285 	if (lp->lwp_stat == LSSLEEP || lp->lwp_stat == LSSTOP) {
1286 		/*
1287 		 * Thread is in tsleep.
1288 		 */
1289 
1290 		/*
1291 		 * If the thread is sleeping uninterruptibly
1292 		 * we can't interrupt the sleep... the signal will
1293 		 * be noticed when the lwp returns through
1294 		 * trap() or syscall().
1295 		 *
1296 		 * Otherwise the signal can interrupt the sleep.
1297 		 *
1298 		 * If the process is traced, the lwp will handle the
1299 		 * tracing in issignal() when it returns to userland.
1300 		 */
1301 		if (lp->lwp_flag & LWP_SINTR) {
1302 			/*
1303 			 * Make runnable and break out of any tsleep as well.
1304 			 */
1305 			lp->lwp_flag |= LWP_BREAKTSLEEP;
1306 			setrunnable(lp);
1307 		}
1308 	} else {
1309 		/*
1310 		 * Otherwise the thread is running
1311 		 *
1312 		 * LSRUN does nothing with the signal, other than kicking
1313 		 * ourselves if we are running.
1314 		 * SZOMB and SIDL mean that it will either never be noticed,
1315 		 * or noticed very soon.
1316 		 *
1317 		 * Note that lwp_thread may be NULL or may not be completely
1318 		 * initialized if the process is in the SIDL or SZOMB state.
1319 		 *
1320 		 * For SMP we may have to forward the request to another cpu.
1321 		 * YYY the MP lock prevents the target process from moving
1322 		 * to another cpu, see kern/kern_switch.c
1323 		 *
1324 		 * If the target thread is waiting on its message port,
1325 		 * wakeup the target thread so it can check (or ignore)
1326 		 * the new signal.  YYY needs cleanup.
1327 		 */
1328 		if (lp == lwkt_preempted_proc()) {
1329 			signotify();
1330 		} else if (lp->lwp_stat == LSRUN) {
1331 			struct thread *td = lp->lwp_thread;
1332 			struct proc *p __debugvar = lp->lwp_proc;
1333 
1334 			KASSERT(td != NULL,
1335 			    ("pid %d/%d NULL lwp_thread stat %d flags %08x/%08x",
1336 			    p->p_pid, lp->lwp_tid, lp->lwp_stat,
1337 			    p->p_flag, lp->lwp_flag));
1338 
1339 			/*
1340 			 * To prevent a MP race with TDF_SINTR we must
1341 			 * schedule the thread on the correct cpu.
1342 			 */
1343 #ifdef SMP
1344 			if (td->td_gd != mycpu) {
1345 				LWPHOLD(lp);
1346 				lwkt_send_ipiq(td->td_gd, signotify_remote, lp);
1347 			} else
1348 #endif
1349 			if (td->td_flags & TDF_SINTR)
1350 				lwkt_schedule(td);
1351 		}
1352 	}
1353 	crit_exit();
1354 }
1355 
1356 #ifdef SMP
1357 
1358 /*
1359  * This function is called via an IPI.  We will be in a critical section but
1360  * the MP lock will NOT be held.  The passed lp will be held.
1361  *
1362  * We must essentially repeat the code at the end of lwp_signotify(),
1363  * in particular rechecking all races.  If we are still not on the
1364  * correct cpu we leave the lwp ref intact and continue the chase.
1365  *
1366  * XXX this may still not be entirely correct, since we are checking
1367  *     lwp_stat asynchronously.
1368  */
1369 static void
1370 signotify_remote(void *arg)
1371 {
1372 	struct lwp *lp = arg;
1373 	thread_t td;
1374 
1375 	if (lp == lwkt_preempted_proc()) {
1376 		signotify();
1377 	} else if (lp->lwp_stat == LSRUN) {
1378 		/*
1379 		 * To prevent a MP race with TDF_SINTR we must
1380 		 * schedule the thread on the correct cpu.
1381 		 */
1382 		td = lp->lwp_thread;
1383 		if (td->td_gd != mycpu) {
1384 			lwkt_send_ipiq(td->td_gd, signotify_remote, lp);
1385 			return;
1386 			/* NOT REACHED */
1387 		}
1388 		if (td->td_flags & TDF_SINTR)
1389 			lwkt_schedule(td);
1390 	}
1391 	LWPRELE(lp);
1392 }
1393 
1394 #endif
1395 
1396 /*
1397  * Caller must hold p->p_token
1398  */
1399 void
1400 proc_stop(struct proc *p)
1401 {
1402 	struct lwp *lp;
1403 
1404 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1405 	crit_enter();
1406 
1407 	/* If somebody raced us, be happy with it */
1408 	if (p->p_stat == SSTOP || p->p_stat == SZOMB) {
1409 		crit_exit();
1410 		return;
1411 	}
1412 	p->p_stat = SSTOP;
1413 
1414 	FOREACH_LWP_IN_PROC(lp, p) {
1415 		switch (lp->lwp_stat) {
1416 		case LSSTOP:
1417 			/*
1418 			 * Do nothing, we are already counted in
1419 			 * p_nstopped.
1420 			 */
1421 			break;
1422 
1423 		case LSSLEEP:
1424 			/*
1425 			 * We're sleeping, but we will stop before
1426 			 * returning to userspace, so count us
1427 			 * as stopped as well.  We set LWP_WSTOP
1428 			 * to signal the lwp that it should not
1429 			 * increase p_nstopped when reaching tstop().
1430 			 */
1431 			if ((lp->lwp_flag & LWP_WSTOP) == 0) {
1432 				lp->lwp_flag |= LWP_WSTOP;
1433 				++p->p_nstopped;
1434 			}
1435 			break;
1436 
1437 		case LSRUN:
1438 			/*
1439 			 * We might notify ourself, but that's not
1440 			 * a problem.
1441 			 */
1442 			lwp_signotify(lp);
1443 			break;
1444 		}
1445 	}
1446 
1447 	if (p->p_nstopped == p->p_nthreads) {
1448 		p->p_flag &= ~P_WAITED;
1449 		wakeup(p->p_pptr);
1450 		if ((p->p_pptr->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1451 			ksignal(p->p_pptr, SIGCHLD);
1452 	}
1453 	crit_exit();
1454 }
1455 
1456 /*
1457  * Caller must hold proc_token
1458  */
1459 void
1460 proc_unstop(struct proc *p)
1461 {
1462 	struct lwp *lp;
1463 
1464 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1465 	crit_enter();
1466 
1467 	if (p->p_stat != SSTOP) {
1468 		crit_exit();
1469 		return;
1470 	}
1471 
1472 	p->p_stat = SACTIVE;
1473 
1474 	FOREACH_LWP_IN_PROC(lp, p) {
1475 		switch (lp->lwp_stat) {
1476 		case LSRUN:
1477 			/*
1478 			 * Uh?  Not stopped?  Well, I guess that's okay.
1479 			 */
1480 			if (bootverbose)
1481 				kprintf("proc_unstop: lwp %d/%d not sleeping\n",
1482 					p->p_pid, lp->lwp_tid);
1483 			break;
1484 
1485 		case LSSLEEP:
1486 			/*
1487 			 * Still sleeping.  Don't bother waking it up.
1488 			 * However, if this thread was counted as
1489 			 * stopped, undo this.
1490 			 *
1491 			 * Nevertheless we call setrunnable() so that it
1492 			 * will wake up in case a signal or timeout arrived
1493 			 * in the meantime.
1494 			 */
1495 			if (lp->lwp_flag & LWP_WSTOP) {
1496 				lp->lwp_flag &= ~LWP_WSTOP;
1497 				--p->p_nstopped;
1498 			} else {
1499 				if (bootverbose)
1500 					kprintf("proc_unstop: lwp %d/%d sleeping, not stopped\n",
1501 						p->p_pid, lp->lwp_tid);
1502 			}
1503 			/* FALLTHROUGH */
1504 
1505 		case LSSTOP:
1506 			setrunnable(lp);
1507 			break;
1508 
1509 		}
1510 	}
1511 	crit_exit();
1512 }
1513 
1514 /*
1515  * No requirements.
1516  */
1517 static int
1518 kern_sigtimedwait(sigset_t waitset, siginfo_t *info, struct timespec *timeout)
1519 {
1520 	sigset_t savedmask, set;
1521 	struct proc *p = curproc;
1522 	struct lwp *lp = curthread->td_lwp;
1523 	int error, sig, hz, timevalid = 0;
1524 	struct timespec rts, ets, ts;
1525 	struct timeval tv;
1526 
1527 	error = 0;
1528 	sig = 0;
1529 	ets.tv_sec = 0;		/* silence compiler warning */
1530 	ets.tv_nsec = 0;	/* silence compiler warning */
1531 	SIG_CANTMASK(waitset);
1532 	savedmask = lp->lwp_sigmask;
1533 
1534 	if (timeout) {
1535 		if (timeout->tv_sec >= 0 && timeout->tv_nsec >= 0 &&
1536 		    timeout->tv_nsec < 1000000000) {
1537 			timevalid = 1;
1538 			getnanouptime(&rts);
1539 		 	ets = rts;
1540 			timespecadd(&ets, timeout);
1541 		}
1542 	}
1543 
1544 	for (;;) {
1545 		set = lwp_sigpend(lp);
1546 		SIGSETAND(set, waitset);
1547 		if ((sig = sig_ffs(&set)) != 0) {
1548 			SIGFILLSET(lp->lwp_sigmask);
1549 			SIGDELSET(lp->lwp_sigmask, sig);
1550 			SIG_CANTMASK(lp->lwp_sigmask);
1551 			sig = issignal(lp, 1);
1552 			/*
1553 			 * It may be a STOP signal, in the case, issignal
1554 			 * returns 0, because we may stop there, and new
1555 			 * signal can come in, we should restart if we got
1556 			 * nothing.
1557 			 */
1558 			if (sig == 0)
1559 				continue;
1560 			else
1561 				break;
1562 		}
1563 
1564 		/*
1565 		 * Previous checking got nothing, and we retried but still
1566 		 * got nothing, we should return the error status.
1567 		 */
1568 		if (error)
1569 			break;
1570 
1571 		/*
1572 		 * POSIX says this must be checked after looking for pending
1573 		 * signals.
1574 		 */
1575 		if (timeout) {
1576 			if (timevalid == 0) {
1577 				error = EINVAL;
1578 				break;
1579 			}
1580 			getnanouptime(&rts);
1581 			if (timespeccmp(&rts, &ets, >=)) {
1582 				error = EAGAIN;
1583 				break;
1584 			}
1585 			ts = ets;
1586 			timespecsub(&ts, &rts);
1587 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1588 			hz = tvtohz_high(&tv);
1589 		} else {
1590 			hz = 0;
1591 		}
1592 
1593 		lp->lwp_sigmask = savedmask;
1594 		SIGSETNAND(lp->lwp_sigmask, waitset);
1595 		/*
1596 		 * We won't ever be woken up.  Instead, our sleep will
1597 		 * be broken in lwpsignal().
1598 		 */
1599 		error = tsleep(&p->p_sigacts, PCATCH, "sigwt", hz);
1600 		if (timeout) {
1601 			if (error == ERESTART) {
1602 				/* can not restart a timeout wait. */
1603 				error = EINTR;
1604 			} else if (error == EAGAIN) {
1605 				/* will calculate timeout by ourself. */
1606 				error = 0;
1607 			}
1608 		}
1609 		/* Retry ... */
1610 	}
1611 
1612 	lp->lwp_sigmask = savedmask;
1613 	if (sig) {
1614 		error = 0;
1615 		bzero(info, sizeof(*info));
1616 		info->si_signo = sig;
1617 		lwp_delsig(lp, sig);	/* take the signal! */
1618 
1619 		if (sig == SIGKILL) {
1620 			sigexit(lp, sig);
1621 			/* NOT REACHED */
1622 		}
1623 	}
1624 
1625 	return (error);
1626 }
1627 
1628 /*
1629  * MPALMOSTSAFE
1630  */
1631 int
1632 sys_sigtimedwait(struct sigtimedwait_args *uap)
1633 {
1634 	struct timespec ts;
1635 	struct timespec *timeout;
1636 	sigset_t set;
1637 	siginfo_t info;
1638 	int error;
1639 
1640 	if (uap->timeout) {
1641 		error = copyin(uap->timeout, &ts, sizeof(ts));
1642 		if (error)
1643 			return (error);
1644 		timeout = &ts;
1645 	} else {
1646 		timeout = NULL;
1647 	}
1648 	error = copyin(uap->set, &set, sizeof(set));
1649 	if (error)
1650 		return (error);
1651 	error = kern_sigtimedwait(set, &info, timeout);
1652 	if (error)
1653 		return (error);
1654  	if (uap->info)
1655 		error = copyout(&info, uap->info, sizeof(info));
1656 	/* Repost if we got an error. */
1657 	/*
1658 	 * XXX lwp
1659 	 *
1660 	 * This could transform a thread-specific signal to another
1661 	 * thread / process pending signal.
1662 	 */
1663 	if (error) {
1664 		ksignal(curproc, info.si_signo);
1665 	} else {
1666 		uap->sysmsg_result = info.si_signo;
1667 	}
1668 	return (error);
1669 }
1670 
1671 /*
1672  * MPALMOSTSAFE
1673  */
1674 int
1675 sys_sigwaitinfo(struct sigwaitinfo_args *uap)
1676 {
1677 	siginfo_t info;
1678 	sigset_t set;
1679 	int error;
1680 
1681 	error = copyin(uap->set, &set, sizeof(set));
1682 	if (error)
1683 		return (error);
1684 	error = kern_sigtimedwait(set, &info, NULL);
1685 	if (error)
1686 		return (error);
1687 	if (uap->info)
1688 		error = copyout(&info, uap->info, sizeof(info));
1689 	/* Repost if we got an error. */
1690 	/*
1691 	 * XXX lwp
1692 	 *
1693 	 * This could transform a thread-specific signal to another
1694 	 * thread / process pending signal.
1695 	 */
1696 	if (error) {
1697 		ksignal(curproc, info.si_signo);
1698 	} else {
1699 		uap->sysmsg_result = info.si_signo;
1700 	}
1701 	return (error);
1702 }
1703 
1704 /*
1705  * If the current process has received a signal that would interrupt a
1706  * system call, return EINTR or ERESTART as appropriate.
1707  */
1708 int
1709 iscaught(struct lwp *lp)
1710 {
1711 	struct proc *p = lp->lwp_proc;
1712 	int sig;
1713 
1714 	if (p) {
1715 		if ((sig = CURSIG(lp)) != 0) {
1716 			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1717 				return (EINTR);
1718 			return (ERESTART);
1719 		}
1720 	}
1721 	return(EWOULDBLOCK);
1722 }
1723 
1724 /*
1725  * If the current process has received a signal (should be caught or cause
1726  * termination, should interrupt current syscall), return the signal number.
1727  * Stop signals with default action are processed immediately, then cleared;
1728  * they aren't returned.  This is checked after each entry to the system for
1729  * a syscall or trap (though this can usually be done without calling issignal
1730  * by checking the pending signal masks in the CURSIG macro).
1731  *
1732  * This routine is called via CURSIG/__cursig.  We will acquire and release
1733  * p->p_token but if the caller needs to interlock the test the caller must
1734  * also hold p->p_token.
1735  *
1736  *	while (sig = CURSIG(curproc))
1737  *		postsig(sig);
1738  *
1739  * MPSAFE
1740  */
1741 int
1742 issignal(struct lwp *lp, int maytrace)
1743 {
1744 	struct proc *p = lp->lwp_proc;
1745 	sigset_t mask;
1746 	int sig, prop;
1747 
1748 	lwkt_gettoken(&p->p_token);
1749 
1750 	for (;;) {
1751 		int traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
1752 
1753 		/*
1754 		 * If this process is supposed to stop, stop this thread.
1755 		 */
1756 		if (p->p_stat == SSTOP)
1757 			tstop();
1758 
1759 		mask = lwp_sigpend(lp);
1760 		SIGSETNAND(mask, lp->lwp_sigmask);
1761 		if (p->p_flag & P_PPWAIT)
1762 			SIG_STOPSIGMASK(mask);
1763 		if (SIGISEMPTY(mask)) {		/* no signal to send */
1764 			lwkt_reltoken(&p->p_token);
1765 			return (0);
1766 		}
1767 		sig = sig_ffs(&mask);
1768 
1769 		STOPEVENT(p, S_SIG, sig);
1770 
1771 		/*
1772 		 * We should see pending but ignored signals
1773 		 * only if P_TRACED was on when they were posted.
1774 		 */
1775 		if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
1776 			lwp_delsig(lp, sig);
1777 			continue;
1778 		}
1779 		if (maytrace && (p->p_flag & P_TRACED) && (p->p_flag & P_PPWAIT) == 0) {
1780 			/*
1781 			 * If traced, always stop, and stay stopped until
1782 			 * released by the parent.
1783 			 *
1784 			 * NOTE: SSTOP may get cleared during the loop,
1785 			 * but we do not re-notify the parent if we have
1786 			 * to loop several times waiting for the parent
1787 			 * to let us continue.
1788 			 *
1789 			 * XXX not sure if this is still true
1790 			 */
1791 			p->p_xstat = sig;
1792 			proc_stop(p);
1793 			do {
1794 				tstop();
1795 			} while (!trace_req(p) && (p->p_flag & P_TRACED));
1796 
1797 			/*
1798 			 * If parent wants us to take the signal,
1799 			 * then it will leave it in p->p_xstat;
1800 			 * otherwise we just look for signals again.
1801 			 */
1802 			lwp_delsig(lp, sig);	/* clear old signal */
1803 			sig = p->p_xstat;
1804 			if (sig == 0)
1805 				continue;
1806 
1807 			/*
1808 			 * Put the new signal into p_siglist.  If the
1809 			 * signal is being masked, look for other signals.
1810 			 *
1811 			 * XXX lwp might need a call to ksignal()
1812 			 */
1813 			SIGADDSET(p->p_siglist, sig);
1814 			if (SIGISMEMBER(lp->lwp_sigmask, sig))
1815 				continue;
1816 
1817 			/*
1818 			 * If the traced bit got turned off, go back up
1819 			 * to the top to rescan signals.  This ensures
1820 			 * that p_sig* and ps_sigact are consistent.
1821 			 */
1822 			if ((p->p_flag & P_TRACED) == 0)
1823 				continue;
1824 		}
1825 
1826 		prop = sigprop(sig);
1827 
1828 		/*
1829 		 * Decide whether the signal should be returned.
1830 		 * Return the signal's number, or fall through
1831 		 * to clear it from the pending mask.
1832 		 */
1833 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
1834 		case (intptr_t)SIG_DFL:
1835 			/*
1836 			 * Don't take default actions on system processes.
1837 			 */
1838 			if (p->p_pid <= 1) {
1839 #ifdef DIAGNOSTIC
1840 				/*
1841 				 * Are you sure you want to ignore SIGSEGV
1842 				 * in init? XXX
1843 				 */
1844 				kprintf("Process (pid %lu) got signal %d\n",
1845 					(u_long)p->p_pid, sig);
1846 #endif
1847 				break;		/* == ignore */
1848 			}
1849 
1850 			/*
1851 			 * Handle the in-kernel checkpoint action
1852 			 */
1853 			if (prop & SA_CKPT) {
1854 				checkpoint_signal_handler(lp);
1855 				break;
1856 			}
1857 
1858 			/*
1859 			 * If there is a pending stop signal to process
1860 			 * with default action, stop here,
1861 			 * then clear the signal.  However,
1862 			 * if process is member of an orphaned
1863 			 * process group, ignore tty stop signals.
1864 			 */
1865 			if (prop & SA_STOP) {
1866 				if (p->p_flag & P_TRACED ||
1867 		    		    (p->p_pgrp->pg_jobc == 0 &&
1868 				    prop & SA_TTYSTOP))
1869 					break;	/* == ignore */
1870 				p->p_xstat = sig;
1871 				proc_stop(p);
1872 				tstop();
1873 				break;
1874 			} else if (prop & SA_IGNORE) {
1875 				/*
1876 				 * Except for SIGCONT, shouldn't get here.
1877 				 * Default action is to ignore; drop it.
1878 				 */
1879 				break;		/* == ignore */
1880 			} else {
1881 				lwkt_reltoken(&p->p_token);
1882 				return (sig);
1883 			}
1884 
1885 			/*NOTREACHED*/
1886 
1887 		case (intptr_t)SIG_IGN:
1888 			/*
1889 			 * Masking above should prevent us ever trying
1890 			 * to take action on an ignored signal other
1891 			 * than SIGCONT, unless process is traced.
1892 			 */
1893 			if ((prop & SA_CONT) == 0 &&
1894 			    (p->p_flag & P_TRACED) == 0)
1895 				kprintf("issignal\n");
1896 			break;		/* == ignore */
1897 
1898 		default:
1899 			/*
1900 			 * This signal has an action, let
1901 			 * postsig() process it.
1902 			 */
1903 			lwkt_reltoken(&p->p_token);
1904 			return (sig);
1905 		}
1906 		lwp_delsig(lp, sig);		/* take the signal! */
1907 	}
1908 	/* NOTREACHED */
1909 }
1910 
1911 /*
1912  * Take the action for the specified signal
1913  * from the current set of pending signals.
1914  */
1915 void
1916 postsig(int sig)
1917 {
1918 	struct lwp *lp = curthread->td_lwp;
1919 	struct proc *p = lp->lwp_proc;
1920 	struct sigacts *ps = p->p_sigacts;
1921 	sig_t action;
1922 	sigset_t returnmask;
1923 	int code;
1924 
1925 	KASSERT(sig != 0, ("postsig"));
1926 
1927 	KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1928 
1929 	/*
1930 	 * If we are a virtual kernel running an emulated user process
1931 	 * context, switch back to the virtual kernel context before
1932 	 * trying to post the signal.
1933 	 */
1934 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1935 		struct trapframe *tf = lp->lwp_md.md_regs;
1936 		tf->tf_trapno = 0;
1937 		vkernel_trap(lp, tf);
1938 	}
1939 
1940 	lwp_delsig(lp, sig);
1941 	action = ps->ps_sigact[_SIG_IDX(sig)];
1942 #ifdef KTRACE
1943 	if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
1944 		ktrpsig(lp, sig, action, lp->lwp_flag & LWP_OLDMASK ?
1945 			&lp->lwp_oldsigmask : &lp->lwp_sigmask, 0);
1946 #endif
1947 	STOPEVENT(p, S_SIG, sig);
1948 
1949 	if (action == SIG_DFL) {
1950 		/*
1951 		 * Default action, where the default is to kill
1952 		 * the process.  (Other cases were ignored above.)
1953 		 */
1954 		sigexit(lp, sig);
1955 		/* NOTREACHED */
1956 	} else {
1957 		/*
1958 		 * If we get here, the signal must be caught.
1959 		 */
1960 		KASSERT(action != SIG_IGN && !SIGISMEMBER(lp->lwp_sigmask, sig),
1961 		    ("postsig action"));
1962 
1963 		crit_enter();
1964 
1965 		/*
1966 		 * Reset the signal handler if asked to
1967 		 */
1968 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
1969 			/*
1970 			 * See kern_sigaction() for origin of this code.
1971 			 */
1972 			SIGDELSET(p->p_sigcatch, sig);
1973 			if (sig != SIGCONT &&
1974 			    sigprop(sig) & SA_IGNORE)
1975 				SIGADDSET(p->p_sigignore, sig);
1976 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1977 		}
1978 
1979 		/*
1980 		 * Handle the mailbox case.  Copyout to the appropriate
1981 		 * location but do not generate a signal frame.  The system
1982 		 * call simply returns EINTR and the user is responsible for
1983 		 * polling the mailbox.
1984 		 */
1985 		if (SIGISMEMBER(ps->ps_sigmailbox, sig)) {
1986 			int sig_copy = sig;
1987 			copyout(&sig_copy, (void *)action, sizeof(int));
1988 			curproc->p_flag |= P_MAILBOX;
1989 			crit_exit();
1990 			goto done;
1991 		}
1992 
1993 		/*
1994 		 * Set the signal mask and calculate the mask to restore
1995 		 * when the signal function returns.
1996 		 *
1997 		 * Special case: user has done a sigsuspend.  Here the
1998 		 * current mask is not of interest, but rather the
1999 		 * mask from before the sigsuspend is what we want
2000 		 * restored after the signal processing is completed.
2001 		 */
2002 		if (lp->lwp_flag & LWP_OLDMASK) {
2003 			returnmask = lp->lwp_oldsigmask;
2004 			lp->lwp_flag &= ~LWP_OLDMASK;
2005 		} else {
2006 			returnmask = lp->lwp_sigmask;
2007 		}
2008 
2009 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
2010 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
2011 			SIGADDSET(lp->lwp_sigmask, sig);
2012 
2013 		crit_exit();
2014 		lp->lwp_ru.ru_nsignals++;
2015 		if (lp->lwp_sig != sig) {
2016 			code = 0;
2017 		} else {
2018 			code = lp->lwp_code;
2019 			lp->lwp_code = 0;
2020 			lp->lwp_sig = 0;
2021 		}
2022 		(*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
2023 	}
2024 done:
2025 	;
2026 }
2027 
2028 /*
2029  * Kill the current process for stated reason.
2030  */
2031 void
2032 killproc(struct proc *p, char *why)
2033 {
2034 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n",
2035 		p->p_pid, p->p_comm,
2036 		p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2037 	ksignal(p, SIGKILL);
2038 }
2039 
2040 /*
2041  * Force the current process to exit with the specified signal, dumping core
2042  * if appropriate.  We bypass the normal tests for masked and caught signals,
2043  * allowing unrecoverable failures to terminate the process without changing
2044  * signal state.  Mark the accounting record with the signal termination.
2045  * If dumping core, save the signal number for the debugger.  Calls exit and
2046  * does not return.
2047  *
2048  * This routine does not return.
2049  */
2050 void
2051 sigexit(struct lwp *lp, int sig)
2052 {
2053 	struct proc *p = lp->lwp_proc;
2054 
2055 	lwkt_gettoken(&p->p_token);
2056 	p->p_acflag |= AXSIG;
2057 	if (sigprop(sig) & SA_CORE) {
2058 		lp->lwp_sig = sig;
2059 		/*
2060 		 * Log signals which would cause core dumps
2061 		 * (Log as LOG_INFO to appease those who don't want
2062 		 * these messages.)
2063 		 * XXX : Todo, as well as euid, write out ruid too
2064 		 */
2065 		if (coredump(lp, sig) == 0)
2066 			sig |= WCOREFLAG;
2067 		if (kern_logsigexit)
2068 			log(LOG_INFO,
2069 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2070 			    p->p_pid, p->p_comm,
2071 			    p->p_ucred ? p->p_ucred->cr_uid : -1,
2072 			    sig &~ WCOREFLAG,
2073 			    sig & WCOREFLAG ? " (core dumped)" : "");
2074 	}
2075 	lwkt_reltoken(&p->p_token);
2076 	exit1(W_EXITCODE(0, sig));
2077 	/* NOTREACHED */
2078 }
2079 
2080 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
2081 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
2082 	      sizeof(corefilename), "process corefile name format string");
2083 
2084 /*
2085  * expand_name(name, uid, pid)
2086  * Expand the name described in corefilename, using name, uid, and pid.
2087  * corefilename is a kprintf-like string, with three format specifiers:
2088  *	%N	name of process ("name")
2089  *	%P	process id (pid)
2090  *	%U	user id (uid)
2091  * For example, "%N.core" is the default; they can be disabled completely
2092  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
2093  * This is controlled by the sysctl variable kern.corefile (see above).
2094  */
2095 
2096 static char *
2097 expand_name(const char *name, uid_t uid, pid_t pid)
2098 {
2099 	char *temp;
2100 	char buf[11];		/* Buffer for pid/uid -- max 4B */
2101 	int i, n;
2102 	char *format = corefilename;
2103 	size_t namelen;
2104 
2105 	temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
2106 	if (temp == NULL)
2107 		return NULL;
2108 	namelen = strlen(name);
2109 	for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
2110 		int l;
2111 		switch (format[i]) {
2112 		case '%':	/* Format character */
2113 			i++;
2114 			switch (format[i]) {
2115 			case '%':
2116 				temp[n++] = '%';
2117 				break;
2118 			case 'N':	/* process name */
2119 				if ((n + namelen) > MAXPATHLEN) {
2120 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2121 					    pid, name, uid, temp, name);
2122 					kfree(temp, M_TEMP);
2123 					return NULL;
2124 				}
2125 				memcpy(temp+n, name, namelen);
2126 				n += namelen;
2127 				break;
2128 			case 'P':	/* process id */
2129 				l = ksprintf(buf, "%u", pid);
2130 				if ((n + l) > MAXPATHLEN) {
2131 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2132 					    pid, name, uid, temp, name);
2133 					kfree(temp, M_TEMP);
2134 					return NULL;
2135 				}
2136 				memcpy(temp+n, buf, l);
2137 				n += l;
2138 				break;
2139 			case 'U':	/* user id */
2140 				l = ksprintf(buf, "%u", uid);
2141 				if ((n + l) > MAXPATHLEN) {
2142 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2143 					    pid, name, uid, temp, name);
2144 					kfree(temp, M_TEMP);
2145 					return NULL;
2146 				}
2147 				memcpy(temp+n, buf, l);
2148 				n += l;
2149 				break;
2150 			default:
2151 			  	log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
2152 			}
2153 			break;
2154 		default:
2155 			temp[n++] = format[i];
2156 		}
2157 	}
2158 	temp[n] = '\0';
2159 	return temp;
2160 }
2161 
2162 /*
2163  * Dump a process' core.  The main routine does some
2164  * policy checking, and creates the name of the coredump;
2165  * then it passes on a vnode and a size limit to the process-specific
2166  * coredump routine if there is one; if there _is not_ one, it returns
2167  * ENOSYS; otherwise it returns the error from the process-specific routine.
2168  *
2169  * The parameter `lp' is the lwp which triggered the coredump.
2170  */
2171 
2172 static int
2173 coredump(struct lwp *lp, int sig)
2174 {
2175 	struct proc *p = lp->lwp_proc;
2176 	struct vnode *vp;
2177 	struct ucred *cred = p->p_ucred;
2178 	struct flock lf;
2179 	struct nlookupdata nd;
2180 	struct vattr vattr;
2181 	int error, error1;
2182 	char *name;			/* name of corefile */
2183 	off_t limit;
2184 
2185 	STOPEVENT(p, S_CORE, 0);
2186 
2187 	if (((sugid_coredump == 0) && p->p_flag & P_SUGID) || do_coredump == 0)
2188 		return (EFAULT);
2189 
2190 	/*
2191 	 * Note that the bulk of limit checking is done after
2192 	 * the corefile is created.  The exception is if the limit
2193 	 * for corefiles is 0, in which case we don't bother
2194 	 * creating the corefile at all.  This layout means that
2195 	 * a corefile is truncated instead of not being created,
2196 	 * if it is larger than the limit.
2197 	 */
2198 	limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
2199 	if (limit == 0)
2200 		return EFBIG;
2201 
2202 	name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
2203 	if (name == NULL)
2204 		return (EINVAL);
2205 	error = nlookup_init(&nd, name, UIO_SYSSPACE, NLC_LOCKVP);
2206 	if (error == 0)
2207 		error = vn_open(&nd, NULL, O_CREAT | FWRITE | O_NOFOLLOW, S_IRUSR | S_IWUSR);
2208 	kfree(name, M_TEMP);
2209 	if (error) {
2210 		nlookup_done(&nd);
2211 		return (error);
2212 	}
2213 	vp = nd.nl_open_vp;
2214 	nd.nl_open_vp = NULL;
2215 	nlookup_done(&nd);
2216 
2217 	vn_unlock(vp);
2218 	lf.l_whence = SEEK_SET;
2219 	lf.l_start = 0;
2220 	lf.l_len = 0;
2221 	lf.l_type = F_WRLCK;
2222 	error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, 0);
2223 	if (error)
2224 		goto out2;
2225 
2226 	/* Don't dump to non-regular files or files with links. */
2227 	if (vp->v_type != VREG ||
2228 	    VOP_GETATTR(vp, &vattr) || vattr.va_nlink != 1) {
2229 		error = EFAULT;
2230 		goto out1;
2231 	}
2232 
2233 	/* Don't dump to files current user does not own */
2234 	if (vattr.va_uid != p->p_ucred->cr_uid) {
2235 		error = EFAULT;
2236 		goto out1;
2237 	}
2238 
2239 	VATTR_NULL(&vattr);
2240 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2241 	vattr.va_size = 0;
2242 	VOP_SETATTR(vp, &vattr, cred);
2243 	p->p_acflag |= ACORE;
2244 	vn_unlock(vp);
2245 
2246 	error = p->p_sysent->sv_coredump ?
2247 		  p->p_sysent->sv_coredump(lp, sig, vp, limit) : ENOSYS;
2248 
2249 out1:
2250 	lf.l_type = F_UNLCK;
2251 	VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, 0);
2252 out2:
2253 	error1 = vn_close(vp, FWRITE);
2254 	if (error == 0)
2255 		error = error1;
2256 	return (error);
2257 }
2258 
2259 /*
2260  * Nonexistent system call-- signal process (may want to handle it).
2261  * Flag error in case process won't see signal immediately (blocked or ignored).
2262  *
2263  * MPALMOSTSAFE
2264  */
2265 /* ARGSUSED */
2266 int
2267 sys_nosys(struct nosys_args *args)
2268 {
2269 	lwpsignal(curproc, curthread->td_lwp, SIGSYS);
2270 	return (EINVAL);
2271 }
2272 
2273 /*
2274  * Send a SIGIO or SIGURG signal to a process or process group using
2275  * stored credentials rather than those of the current process.
2276  */
2277 void
2278 pgsigio(struct sigio *sigio, int sig, int checkctty)
2279 {
2280 	if (sigio == NULL)
2281 		return;
2282 
2283 	if (sigio->sio_pgid > 0) {
2284 		if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
2285 		             sigio->sio_proc))
2286 			ksignal(sigio->sio_proc, sig);
2287 	} else if (sigio->sio_pgid < 0) {
2288 		struct proc *p;
2289 		struct pgrp *pg = sigio->sio_pgrp;
2290 
2291 		/*
2292 		 * Must interlock all signals against fork
2293 		 */
2294 		pgref(pg);
2295 		lockmgr(&pg->pg_lock, LK_EXCLUSIVE);
2296 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
2297 			if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
2298 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
2299 				ksignal(p, sig);
2300 		}
2301 		lockmgr(&pg->pg_lock, LK_RELEASE);
2302 		pgrel(pg);
2303 	}
2304 }
2305 
2306 static int
2307 filt_sigattach(struct knote *kn)
2308 {
2309 	struct proc *p = curproc;
2310 
2311 	kn->kn_ptr.p_proc = p;
2312 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
2313 
2314 	/* XXX lock the proc here while adding to the list? */
2315 	knote_insert(&p->p_klist, kn);
2316 
2317 	return (0);
2318 }
2319 
2320 static void
2321 filt_sigdetach(struct knote *kn)
2322 {
2323 	struct proc *p = kn->kn_ptr.p_proc;
2324 
2325 	knote_remove(&p->p_klist, kn);
2326 }
2327 
2328 /*
2329  * signal knotes are shared with proc knotes, so we apply a mask to
2330  * the hint in order to differentiate them from process hints.  This
2331  * could be avoided by using a signal-specific knote list, but probably
2332  * isn't worth the trouble.
2333  */
2334 static int
2335 filt_signal(struct knote *kn, long hint)
2336 {
2337 	if (hint & NOTE_SIGNAL) {
2338 		hint &= ~NOTE_SIGNAL;
2339 
2340 		if (kn->kn_id == hint)
2341 			kn->kn_data++;
2342 	}
2343 	return (kn->kn_data != 0);
2344 }
2345