xref: /dflybsd-src/sys/kern/kern_sig.c (revision abf21138fc9d6b9536faf556858abf264f704f26)
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 	struct proc *q;
1044 	sig_t action;
1045 	int prop;
1046 
1047 	if (sig > _SIG_MAXSIG || sig <= 0) {
1048 		kprintf("lwpsignal: signal %d\n", sig);
1049 		panic("lwpsignal signal number");
1050 	}
1051 
1052 	KKASSERT(lp == NULL || lp->lwp_proc == p);
1053 
1054 	PHOLD(p);
1055 	lwkt_gettoken(&p->p_token);
1056 
1057 	prop = sigprop(sig);
1058 
1059 	/*
1060 	 * If proc is traced, always give parent a chance;
1061 	 * if signal event is tracked by procfs, give *that*
1062 	 * a chance, as well.
1063 	 */
1064 	if ((p->p_flag & P_TRACED) || (p->p_stops & S_SIG)) {
1065 		action = SIG_DFL;
1066 	} else {
1067 		/*
1068 		 * Do not try to deliver signals to an exiting lwp.  Note
1069 		 * that we must still deliver the signal if P_WEXIT is set
1070 		 * in the process flags.
1071 		 */
1072 		if (lp && (lp->lwp_flag & LWP_WEXIT)) {
1073 			lwkt_reltoken(&p->p_token);
1074 			PRELE(p);
1075 			return;
1076 		}
1077 
1078 		/*
1079 		 * If the signal is being ignored, then we forget about
1080 		 * it immediately.  NOTE: We don't set SIGCONT in p_sigignore,
1081 		 * and if it is set to SIG_IGN, action will be SIG_DFL here.
1082 		 */
1083 		if (SIGISMEMBER(p->p_sigignore, sig)) {
1084 			/*
1085 			 * Even if a signal is set SIG_IGN, it may still be
1086 			 * lurking in a kqueue.
1087 			 */
1088 			KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1089 			lwkt_reltoken(&p->p_token);
1090 			PRELE(p);
1091 			return;
1092 		}
1093 		if (SIGISMEMBER(p->p_sigcatch, sig))
1094 			action = SIG_CATCH;
1095 		else
1096 			action = SIG_DFL;
1097 	}
1098 
1099 	/*
1100 	 * If continuing, clear any pending STOP signals.
1101 	 */
1102 	if (prop & SA_CONT)
1103 		SIG_STOPSIGMASK(p->p_siglist);
1104 
1105 	if (prop & SA_STOP) {
1106 		/*
1107 		 * If sending a tty stop signal to a member of an orphaned
1108 		 * process group, discard the signal here if the action
1109 		 * is default; don't stop the process below if sleeping,
1110 		 * and don't clear any pending SIGCONT.
1111 		 */
1112 		if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0 &&
1113 		    action == SIG_DFL) {
1114 			lwkt_reltoken(&p->p_token);
1115 			PRELE(p);
1116 		        return;
1117 		}
1118 		SIG_CONTSIGMASK(p->p_siglist);
1119 		p->p_flag &= ~P_CONTINUED;
1120 	}
1121 
1122 	crit_enter();
1123 
1124 	if (p->p_stat == SSTOP) {
1125 		/*
1126 		 * Nobody can handle this signal, add it to the lwp or
1127 		 * process pending list
1128 		 */
1129 		if (lp)
1130 			SIGADDSET(lp->lwp_siglist, sig);
1131 		else
1132 			SIGADDSET(p->p_siglist, sig);
1133 
1134 		/*
1135 		 * If the process is stopped and is being traced, then no
1136 		 * further action is necessary.
1137 		 */
1138 		if (p->p_flag & P_TRACED)
1139 			goto out;
1140 
1141 		/*
1142 		 * If the process is stopped and receives a KILL signal,
1143 		 * make the process runnable.
1144 		 */
1145 		if (sig == SIGKILL) {
1146 			proc_unstop(p);
1147 			goto active_process;
1148 		}
1149 
1150 		/*
1151 		 * If the process is stopped and receives a CONT signal,
1152 		 * then try to make the process runnable again.
1153 		 */
1154 		if (prop & SA_CONT) {
1155 			/*
1156 			 * If SIGCONT is default (or ignored), we continue the
1157 			 * process but don't leave the signal in p_siglist, as
1158 			 * it has no further action.  If SIGCONT is held, we
1159 			 * continue the process and leave the signal in
1160 			 * p_siglist.  If the process catches SIGCONT, let it
1161 			 * handle the signal itself.
1162 			 *
1163 			 * XXX what if the signal is being held blocked?
1164 			 *
1165 			 * Token required to interlock kern_wait().
1166 			 * Reparenting can also cause a race so we have to
1167 			 * hold (q).
1168 			 */
1169 			q = p->p_pptr;
1170 			PHOLD(q);
1171 			lwkt_gettoken(&q->p_token);
1172 			p->p_flag |= P_CONTINUED;
1173 			wakeup(q);
1174 			if (action == SIG_DFL)
1175 				SIGDELSET(p->p_siglist, sig);
1176 			proc_unstop(p);
1177 			lwkt_reltoken(&q->p_token);
1178 			PRELE(q);
1179 			if (action == SIG_CATCH)
1180 				goto active_process;
1181 			goto out;
1182 		}
1183 
1184 		/*
1185 		 * If the process is stopped and receives another STOP
1186 		 * signal, we do not need to stop it again.  If we did
1187 		 * the shell could get confused.
1188 		 *
1189 		 * However, if the current/preempted lwp is part of the
1190 		 * process receiving the signal, we need to keep it,
1191 		 * so that this lwp can stop in issignal() later, as
1192 		 * we don't want to wait until it reaches userret!
1193 		 */
1194 		if (prop & SA_STOP) {
1195 			if (lwkt_preempted_proc() == NULL ||
1196 			    lwkt_preempted_proc()->lwp_proc != p)
1197 				SIGDELSET(p->p_siglist, sig);
1198 		}
1199 
1200 		/*
1201 		 * Otherwise the process is stopped and it received some
1202 		 * signal, which does not change its stopped state.
1203 		 *
1204 		 * We have to select one thread to set LWP_BREAKTSLEEP,
1205 		 * so that the current signal will break the sleep
1206 		 * as soon as a SA_CONT signal will unstop the process.
1207 		 */
1208 		if (lp == NULL)
1209 			lp = find_lwp_for_signal(p, sig);
1210 		if (lp != NULL &&
1211 		    (lp->lwp_stat == LSSLEEP || lp->lwp_stat == LSSTOP))
1212 			lp->lwp_flag |= LWP_BREAKTSLEEP;
1213 		goto out;
1214 
1215 		/* NOTREACHED */
1216 	}
1217 	/* else not stopped */
1218 active_process:
1219 
1220 	/*
1221 	 * Never deliver a lwp-specific signal to a random lwp.
1222 	 */
1223 	if (lp == NULL) {
1224 		lp = find_lwp_for_signal(p, sig);
1225 		if (lp && SIGISMEMBER(lp->lwp_sigmask, sig))
1226 			lp = NULL;
1227 	}
1228 
1229 	/*
1230 	 * Deliver to the process generically if (1) the signal is being
1231 	 * sent to any thread or (2) we could not find a thread to deliver
1232 	 * it to.
1233 	 */
1234 	if (lp == NULL) {
1235 		SIGADDSET(p->p_siglist, sig);
1236 		goto out;
1237 	}
1238 
1239 	/*
1240 	 * Deliver to a specific LWP whether it masks it or not.  It will
1241 	 * not be dispatched if masked but we must still deliver it.
1242 	 */
1243 	if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
1244 	    (p->p_flag & P_TRACED) == 0) {
1245 		p->p_nice = NZERO;
1246 	}
1247 
1248 	/*
1249 	 * If the process receives a STOP signal which indeed needs to
1250 	 * stop the process, do so.  If the process chose to catch the
1251 	 * signal, it will be treated like any other signal.
1252 	 */
1253 	if ((prop & SA_STOP) && action == SIG_DFL) {
1254 		/*
1255 		 * If a child holding parent blocked, stopping
1256 		 * could cause deadlock.  Take no action at this
1257 		 * time.
1258 		 */
1259 		if (p->p_flag & P_PPWAIT) {
1260 			SIGADDSET(p->p_siglist, sig);
1261 			goto out;
1262 		}
1263 
1264 		/*
1265 		 * Do not actually try to manipulate the process, but simply
1266 		 * stop it.  Lwps will stop as soon as they safely can.
1267 		 */
1268 		p->p_xstat = sig;
1269 		proc_stop(p);
1270 		goto out;
1271 	}
1272 
1273 	/*
1274 	 * If it is a CONT signal with default action, just ignore it.
1275 	 */
1276 	if ((prop & SA_CONT) && action == SIG_DFL)
1277 		goto out;
1278 
1279 	/*
1280 	 * Mark signal pending at this specific thread.
1281 	 */
1282 	SIGADDSET(lp->lwp_siglist, sig);
1283 
1284 	lwp_signotify(lp);
1285 
1286 out:
1287 	lwkt_reltoken(&p->p_token);
1288 	PRELE(p);
1289 	crit_exit();
1290 }
1291 
1292 /*
1293  * p->p_token must be held
1294  */
1295 static void
1296 lwp_signotify(struct lwp *lp)
1297 {
1298 	ASSERT_LWKT_TOKEN_HELD(&lp->lwp_proc->p_token);
1299 	crit_enter();
1300 
1301 	if (lp->lwp_stat == LSSLEEP || lp->lwp_stat == LSSTOP) {
1302 		/*
1303 		 * Thread is in tsleep.
1304 		 */
1305 
1306 		/*
1307 		 * If the thread is sleeping uninterruptibly
1308 		 * we can't interrupt the sleep... the signal will
1309 		 * be noticed when the lwp returns through
1310 		 * trap() or syscall().
1311 		 *
1312 		 * Otherwise the signal can interrupt the sleep.
1313 		 *
1314 		 * If the process is traced, the lwp will handle the
1315 		 * tracing in issignal() when it returns to userland.
1316 		 */
1317 		if (lp->lwp_flag & LWP_SINTR) {
1318 			/*
1319 			 * Make runnable and break out of any tsleep as well.
1320 			 */
1321 			lp->lwp_flag |= LWP_BREAKTSLEEP;
1322 			setrunnable(lp);
1323 		}
1324 	} else {
1325 		/*
1326 		 * Otherwise the thread is running
1327 		 *
1328 		 * LSRUN does nothing with the signal, other than kicking
1329 		 * ourselves if we are running.
1330 		 * SZOMB and SIDL mean that it will either never be noticed,
1331 		 * or noticed very soon.
1332 		 *
1333 		 * Note that lwp_thread may be NULL or may not be completely
1334 		 * initialized if the process is in the SIDL or SZOMB state.
1335 		 *
1336 		 * For SMP we may have to forward the request to another cpu.
1337 		 * YYY the MP lock prevents the target process from moving
1338 		 * to another cpu, see kern/kern_switch.c
1339 		 *
1340 		 * If the target thread is waiting on its message port,
1341 		 * wakeup the target thread so it can check (or ignore)
1342 		 * the new signal.  YYY needs cleanup.
1343 		 */
1344 		if (lp == lwkt_preempted_proc()) {
1345 			signotify();
1346 		} else if (lp->lwp_stat == LSRUN) {
1347 			struct thread *td = lp->lwp_thread;
1348 			struct proc *p __debugvar = lp->lwp_proc;
1349 
1350 			KASSERT(td != NULL,
1351 			    ("pid %d/%d NULL lwp_thread stat %d flags %08x/%08x",
1352 			    p->p_pid, lp->lwp_tid, lp->lwp_stat,
1353 			    p->p_flag, lp->lwp_flag));
1354 
1355 			/*
1356 			 * To prevent a MP race with TDF_SINTR we must
1357 			 * schedule the thread on the correct cpu.
1358 			 */
1359 #ifdef SMP
1360 			if (td->td_gd != mycpu) {
1361 				LWPHOLD(lp);
1362 				lwkt_send_ipiq(td->td_gd, signotify_remote, lp);
1363 			} else
1364 #endif
1365 			if (td->td_flags & TDF_SINTR)
1366 				lwkt_schedule(td);
1367 		}
1368 	}
1369 	crit_exit();
1370 }
1371 
1372 #ifdef SMP
1373 
1374 /*
1375  * This function is called via an IPI.  We will be in a critical section but
1376  * the MP lock will NOT be held.  The passed lp will be held.
1377  *
1378  * We must essentially repeat the code at the end of lwp_signotify(),
1379  * in particular rechecking all races.  If we are still not on the
1380  * correct cpu we leave the lwp ref intact and continue the chase.
1381  *
1382  * XXX this may still not be entirely correct, since we are checking
1383  *     lwp_stat asynchronously.
1384  */
1385 static void
1386 signotify_remote(void *arg)
1387 {
1388 	struct lwp *lp = arg;
1389 	thread_t td;
1390 
1391 	if (lp == lwkt_preempted_proc()) {
1392 		signotify();
1393 	} else if (lp->lwp_stat == LSRUN) {
1394 		/*
1395 		 * To prevent a MP race with TDF_SINTR we must
1396 		 * schedule the thread on the correct cpu.
1397 		 */
1398 		td = lp->lwp_thread;
1399 		if (td->td_gd != mycpu) {
1400 			lwkt_send_ipiq(td->td_gd, signotify_remote, lp);
1401 			return;
1402 			/* NOT REACHED */
1403 		}
1404 		if (td->td_flags & TDF_SINTR)
1405 			lwkt_schedule(td);
1406 	}
1407 	LWPRELE(lp);
1408 }
1409 
1410 #endif
1411 
1412 /*
1413  * Caller must hold p->p_token
1414  */
1415 void
1416 proc_stop(struct proc *p)
1417 {
1418 	struct proc *q;
1419 	struct lwp *lp;
1420 
1421 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1422 	crit_enter();
1423 
1424 	/* If somebody raced us, be happy with it */
1425 	if (p->p_stat == SSTOP || p->p_stat == SZOMB) {
1426 		crit_exit();
1427 		return;
1428 	}
1429 	p->p_stat = SSTOP;
1430 
1431 	FOREACH_LWP_IN_PROC(lp, p) {
1432 		switch (lp->lwp_stat) {
1433 		case LSSTOP:
1434 			/*
1435 			 * Do nothing, we are already counted in
1436 			 * p_nstopped.
1437 			 */
1438 			break;
1439 
1440 		case LSSLEEP:
1441 			/*
1442 			 * We're sleeping, but we will stop before
1443 			 * returning to userspace, so count us
1444 			 * as stopped as well.  We set LWP_WSTOP
1445 			 * to signal the lwp that it should not
1446 			 * increase p_nstopped when reaching tstop().
1447 			 */
1448 			if ((lp->lwp_flag & LWP_WSTOP) == 0) {
1449 				lp->lwp_flag |= LWP_WSTOP;
1450 				++p->p_nstopped;
1451 			}
1452 			break;
1453 
1454 		case LSRUN:
1455 			/*
1456 			 * We might notify ourself, but that's not
1457 			 * a problem.
1458 			 */
1459 			lwp_signotify(lp);
1460 			break;
1461 		}
1462 	}
1463 
1464 	if (p->p_nstopped == p->p_nthreads) {
1465 		/*
1466 		 * Token required to interlock kern_wait().  Reparenting can
1467 		 * also cause a race so we have to hold (q).
1468 		 */
1469 		q = p->p_pptr;
1470 		PHOLD(q);
1471 		lwkt_gettoken(&q->p_token);
1472 		p->p_flag &= ~P_WAITED;
1473 		wakeup(q);
1474 		if ((q->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1475 			ksignal(p->p_pptr, SIGCHLD);
1476 		lwkt_reltoken(&q->p_token);
1477 		PRELE(q);
1478 	}
1479 	crit_exit();
1480 }
1481 
1482 /*
1483  * Caller must hold proc_token
1484  */
1485 void
1486 proc_unstop(struct proc *p)
1487 {
1488 	struct lwp *lp;
1489 
1490 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1491 	crit_enter();
1492 
1493 	if (p->p_stat != SSTOP) {
1494 		crit_exit();
1495 		return;
1496 	}
1497 
1498 	p->p_stat = SACTIVE;
1499 
1500 	FOREACH_LWP_IN_PROC(lp, p) {
1501 		switch (lp->lwp_stat) {
1502 		case LSRUN:
1503 			/*
1504 			 * Uh?  Not stopped?  Well, I guess that's okay.
1505 			 */
1506 			if (bootverbose)
1507 				kprintf("proc_unstop: lwp %d/%d not sleeping\n",
1508 					p->p_pid, lp->lwp_tid);
1509 			break;
1510 
1511 		case LSSLEEP:
1512 			/*
1513 			 * Still sleeping.  Don't bother waking it up.
1514 			 * However, if this thread was counted as
1515 			 * stopped, undo this.
1516 			 *
1517 			 * Nevertheless we call setrunnable() so that it
1518 			 * will wake up in case a signal or timeout arrived
1519 			 * in the meantime.
1520 			 */
1521 			if (lp->lwp_flag & LWP_WSTOP) {
1522 				lp->lwp_flag &= ~LWP_WSTOP;
1523 				--p->p_nstopped;
1524 			} else {
1525 				if (bootverbose)
1526 					kprintf("proc_unstop: lwp %d/%d sleeping, not stopped\n",
1527 						p->p_pid, lp->lwp_tid);
1528 			}
1529 			/* FALLTHROUGH */
1530 
1531 		case LSSTOP:
1532 			setrunnable(lp);
1533 			break;
1534 
1535 		}
1536 	}
1537 	crit_exit();
1538 }
1539 
1540 /*
1541  * No requirements.
1542  */
1543 static int
1544 kern_sigtimedwait(sigset_t waitset, siginfo_t *info, struct timespec *timeout)
1545 {
1546 	sigset_t savedmask, set;
1547 	struct proc *p = curproc;
1548 	struct lwp *lp = curthread->td_lwp;
1549 	int error, sig, hz, timevalid = 0;
1550 	struct timespec rts, ets, ts;
1551 	struct timeval tv;
1552 
1553 	error = 0;
1554 	sig = 0;
1555 	ets.tv_sec = 0;		/* silence compiler warning */
1556 	ets.tv_nsec = 0;	/* silence compiler warning */
1557 	SIG_CANTMASK(waitset);
1558 	savedmask = lp->lwp_sigmask;
1559 
1560 	if (timeout) {
1561 		if (timeout->tv_sec >= 0 && timeout->tv_nsec >= 0 &&
1562 		    timeout->tv_nsec < 1000000000) {
1563 			timevalid = 1;
1564 			getnanouptime(&rts);
1565 		 	ets = rts;
1566 			timespecadd(&ets, timeout);
1567 		}
1568 	}
1569 
1570 	for (;;) {
1571 		set = lwp_sigpend(lp);
1572 		SIGSETAND(set, waitset);
1573 		if ((sig = sig_ffs(&set)) != 0) {
1574 			SIGFILLSET(lp->lwp_sigmask);
1575 			SIGDELSET(lp->lwp_sigmask, sig);
1576 			SIG_CANTMASK(lp->lwp_sigmask);
1577 			sig = issignal(lp, 1);
1578 			/*
1579 			 * It may be a STOP signal, in the case, issignal
1580 			 * returns 0, because we may stop there, and new
1581 			 * signal can come in, we should restart if we got
1582 			 * nothing.
1583 			 */
1584 			if (sig == 0)
1585 				continue;
1586 			else
1587 				break;
1588 		}
1589 
1590 		/*
1591 		 * Previous checking got nothing, and we retried but still
1592 		 * got nothing, we should return the error status.
1593 		 */
1594 		if (error)
1595 			break;
1596 
1597 		/*
1598 		 * POSIX says this must be checked after looking for pending
1599 		 * signals.
1600 		 */
1601 		if (timeout) {
1602 			if (timevalid == 0) {
1603 				error = EINVAL;
1604 				break;
1605 			}
1606 			getnanouptime(&rts);
1607 			if (timespeccmp(&rts, &ets, >=)) {
1608 				error = EAGAIN;
1609 				break;
1610 			}
1611 			ts = ets;
1612 			timespecsub(&ts, &rts);
1613 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1614 			hz = tvtohz_high(&tv);
1615 		} else {
1616 			hz = 0;
1617 		}
1618 
1619 		lp->lwp_sigmask = savedmask;
1620 		SIGSETNAND(lp->lwp_sigmask, waitset);
1621 		/*
1622 		 * We won't ever be woken up.  Instead, our sleep will
1623 		 * be broken in lwpsignal().
1624 		 */
1625 		error = tsleep(&p->p_sigacts, PCATCH, "sigwt", hz);
1626 		if (timeout) {
1627 			if (error == ERESTART) {
1628 				/* can not restart a timeout wait. */
1629 				error = EINTR;
1630 			} else if (error == EAGAIN) {
1631 				/* will calculate timeout by ourself. */
1632 				error = 0;
1633 			}
1634 		}
1635 		/* Retry ... */
1636 	}
1637 
1638 	lp->lwp_sigmask = savedmask;
1639 	if (sig) {
1640 		error = 0;
1641 		bzero(info, sizeof(*info));
1642 		info->si_signo = sig;
1643 		lwp_delsig(lp, sig);	/* take the signal! */
1644 
1645 		if (sig == SIGKILL) {
1646 			sigexit(lp, sig);
1647 			/* NOT REACHED */
1648 		}
1649 	}
1650 
1651 	return (error);
1652 }
1653 
1654 /*
1655  * MPALMOSTSAFE
1656  */
1657 int
1658 sys_sigtimedwait(struct sigtimedwait_args *uap)
1659 {
1660 	struct timespec ts;
1661 	struct timespec *timeout;
1662 	sigset_t set;
1663 	siginfo_t info;
1664 	int error;
1665 
1666 	if (uap->timeout) {
1667 		error = copyin(uap->timeout, &ts, sizeof(ts));
1668 		if (error)
1669 			return (error);
1670 		timeout = &ts;
1671 	} else {
1672 		timeout = NULL;
1673 	}
1674 	error = copyin(uap->set, &set, sizeof(set));
1675 	if (error)
1676 		return (error);
1677 	error = kern_sigtimedwait(set, &info, timeout);
1678 	if (error)
1679 		return (error);
1680  	if (uap->info)
1681 		error = copyout(&info, uap->info, sizeof(info));
1682 	/* Repost if we got an error. */
1683 	/*
1684 	 * XXX lwp
1685 	 *
1686 	 * This could transform a thread-specific signal to another
1687 	 * thread / process pending signal.
1688 	 */
1689 	if (error) {
1690 		ksignal(curproc, info.si_signo);
1691 	} else {
1692 		uap->sysmsg_result = info.si_signo;
1693 	}
1694 	return (error);
1695 }
1696 
1697 /*
1698  * MPALMOSTSAFE
1699  */
1700 int
1701 sys_sigwaitinfo(struct sigwaitinfo_args *uap)
1702 {
1703 	siginfo_t info;
1704 	sigset_t set;
1705 	int error;
1706 
1707 	error = copyin(uap->set, &set, sizeof(set));
1708 	if (error)
1709 		return (error);
1710 	error = kern_sigtimedwait(set, &info, NULL);
1711 	if (error)
1712 		return (error);
1713 	if (uap->info)
1714 		error = copyout(&info, uap->info, sizeof(info));
1715 	/* Repost if we got an error. */
1716 	/*
1717 	 * XXX lwp
1718 	 *
1719 	 * This could transform a thread-specific signal to another
1720 	 * thread / process pending signal.
1721 	 */
1722 	if (error) {
1723 		ksignal(curproc, info.si_signo);
1724 	} else {
1725 		uap->sysmsg_result = info.si_signo;
1726 	}
1727 	return (error);
1728 }
1729 
1730 /*
1731  * If the current process has received a signal that would interrupt a
1732  * system call, return EINTR or ERESTART as appropriate.
1733  */
1734 int
1735 iscaught(struct lwp *lp)
1736 {
1737 	struct proc *p = lp->lwp_proc;
1738 	int sig;
1739 
1740 	if (p) {
1741 		if ((sig = CURSIG(lp)) != 0) {
1742 			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1743 				return (EINTR);
1744 			return (ERESTART);
1745 		}
1746 	}
1747 	return(EWOULDBLOCK);
1748 }
1749 
1750 /*
1751  * If the current process has received a signal (should be caught or cause
1752  * termination, should interrupt current syscall), return the signal number.
1753  * Stop signals with default action are processed immediately, then cleared;
1754  * they aren't returned.  This is checked after each entry to the system for
1755  * a syscall or trap (though this can usually be done without calling issignal
1756  * by checking the pending signal masks in the CURSIG macro).
1757  *
1758  * This routine is called via CURSIG/__cursig.  We will acquire and release
1759  * p->p_token but if the caller needs to interlock the test the caller must
1760  * also hold p->p_token.
1761  *
1762  *	while (sig = CURSIG(curproc))
1763  *		postsig(sig);
1764  *
1765  * MPSAFE
1766  */
1767 int
1768 issignal(struct lwp *lp, int maytrace)
1769 {
1770 	struct proc *p = lp->lwp_proc;
1771 	sigset_t mask;
1772 	int sig, prop;
1773 
1774 	lwkt_gettoken(&p->p_token);
1775 
1776 	for (;;) {
1777 		int traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
1778 
1779 		/*
1780 		 * If this process is supposed to stop, stop this thread.
1781 		 */
1782 		if (p->p_stat == SSTOP)
1783 			tstop();
1784 
1785 		mask = lwp_sigpend(lp);
1786 		SIGSETNAND(mask, lp->lwp_sigmask);
1787 		if (p->p_flag & P_PPWAIT)
1788 			SIG_STOPSIGMASK(mask);
1789 		if (SIGISEMPTY(mask)) {		/* no signal to send */
1790 			lwkt_reltoken(&p->p_token);
1791 			return (0);
1792 		}
1793 		sig = sig_ffs(&mask);
1794 
1795 		STOPEVENT(p, S_SIG, sig);
1796 
1797 		/*
1798 		 * We should see pending but ignored signals
1799 		 * only if P_TRACED was on when they were posted.
1800 		 */
1801 		if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
1802 			lwp_delsig(lp, sig);
1803 			continue;
1804 		}
1805 		if (maytrace && (p->p_flag & P_TRACED) && (p->p_flag & P_PPWAIT) == 0) {
1806 			/*
1807 			 * If traced, always stop, and stay stopped until
1808 			 * released by the parent.
1809 			 *
1810 			 * NOTE: SSTOP may get cleared during the loop,
1811 			 * but we do not re-notify the parent if we have
1812 			 * to loop several times waiting for the parent
1813 			 * to let us continue.
1814 			 *
1815 			 * XXX not sure if this is still true
1816 			 */
1817 			p->p_xstat = sig;
1818 			proc_stop(p);
1819 			do {
1820 				tstop();
1821 			} while (!trace_req(p) && (p->p_flag & P_TRACED));
1822 
1823 			/*
1824 			 * If parent wants us to take the signal,
1825 			 * then it will leave it in p->p_xstat;
1826 			 * otherwise we just look for signals again.
1827 			 */
1828 			lwp_delsig(lp, sig);	/* clear old signal */
1829 			sig = p->p_xstat;
1830 			if (sig == 0)
1831 				continue;
1832 
1833 			/*
1834 			 * Put the new signal into p_siglist.  If the
1835 			 * signal is being masked, look for other signals.
1836 			 *
1837 			 * XXX lwp might need a call to ksignal()
1838 			 */
1839 			SIGADDSET(p->p_siglist, sig);
1840 			if (SIGISMEMBER(lp->lwp_sigmask, sig))
1841 				continue;
1842 
1843 			/*
1844 			 * If the traced bit got turned off, go back up
1845 			 * to the top to rescan signals.  This ensures
1846 			 * that p_sig* and ps_sigact are consistent.
1847 			 */
1848 			if ((p->p_flag & P_TRACED) == 0)
1849 				continue;
1850 		}
1851 
1852 		prop = sigprop(sig);
1853 
1854 		/*
1855 		 * Decide whether the signal should be returned.
1856 		 * Return the signal's number, or fall through
1857 		 * to clear it from the pending mask.
1858 		 */
1859 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
1860 		case (intptr_t)SIG_DFL:
1861 			/*
1862 			 * Don't take default actions on system processes.
1863 			 */
1864 			if (p->p_pid <= 1) {
1865 #ifdef DIAGNOSTIC
1866 				/*
1867 				 * Are you sure you want to ignore SIGSEGV
1868 				 * in init? XXX
1869 				 */
1870 				kprintf("Process (pid %lu) got signal %d\n",
1871 					(u_long)p->p_pid, sig);
1872 #endif
1873 				break;		/* == ignore */
1874 			}
1875 
1876 			/*
1877 			 * Handle the in-kernel checkpoint action
1878 			 */
1879 			if (prop & SA_CKPT) {
1880 				checkpoint_signal_handler(lp);
1881 				break;
1882 			}
1883 
1884 			/*
1885 			 * If there is a pending stop signal to process
1886 			 * with default action, stop here,
1887 			 * then clear the signal.  However,
1888 			 * if process is member of an orphaned
1889 			 * process group, ignore tty stop signals.
1890 			 */
1891 			if (prop & SA_STOP) {
1892 				if (p->p_flag & P_TRACED ||
1893 		    		    (p->p_pgrp->pg_jobc == 0 &&
1894 				    prop & SA_TTYSTOP))
1895 					break;	/* == ignore */
1896 				p->p_xstat = sig;
1897 				proc_stop(p);
1898 				tstop();
1899 				break;
1900 			} else if (prop & SA_IGNORE) {
1901 				/*
1902 				 * Except for SIGCONT, shouldn't get here.
1903 				 * Default action is to ignore; drop it.
1904 				 */
1905 				break;		/* == ignore */
1906 			} else {
1907 				lwkt_reltoken(&p->p_token);
1908 				return (sig);
1909 			}
1910 
1911 			/*NOTREACHED*/
1912 
1913 		case (intptr_t)SIG_IGN:
1914 			/*
1915 			 * Masking above should prevent us ever trying
1916 			 * to take action on an ignored signal other
1917 			 * than SIGCONT, unless process is traced.
1918 			 */
1919 			if ((prop & SA_CONT) == 0 &&
1920 			    (p->p_flag & P_TRACED) == 0)
1921 				kprintf("issignal\n");
1922 			break;		/* == ignore */
1923 
1924 		default:
1925 			/*
1926 			 * This signal has an action, let
1927 			 * postsig() process it.
1928 			 */
1929 			lwkt_reltoken(&p->p_token);
1930 			return (sig);
1931 		}
1932 		lwp_delsig(lp, sig);		/* take the signal! */
1933 	}
1934 	/* NOTREACHED */
1935 }
1936 
1937 /*
1938  * Take the action for the specified signal
1939  * from the current set of pending signals.
1940  *
1941  * Caller must hold p->p_token
1942  */
1943 void
1944 postsig(int sig)
1945 {
1946 	struct lwp *lp = curthread->td_lwp;
1947 	struct proc *p = lp->lwp_proc;
1948 	struct sigacts *ps = p->p_sigacts;
1949 	sig_t action;
1950 	sigset_t returnmask;
1951 	int code;
1952 
1953 	KASSERT(sig != 0, ("postsig"));
1954 
1955 	KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1956 
1957 	/*
1958 	 * If we are a virtual kernel running an emulated user process
1959 	 * context, switch back to the virtual kernel context before
1960 	 * trying to post the signal.
1961 	 */
1962 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1963 		struct trapframe *tf = lp->lwp_md.md_regs;
1964 		tf->tf_trapno = 0;
1965 		vkernel_trap(lp, tf);
1966 	}
1967 
1968 	lwp_delsig(lp, sig);
1969 	action = ps->ps_sigact[_SIG_IDX(sig)];
1970 #ifdef KTRACE
1971 	if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
1972 		ktrpsig(lp, sig, action, lp->lwp_flag & LWP_OLDMASK ?
1973 			&lp->lwp_oldsigmask : &lp->lwp_sigmask, 0);
1974 #endif
1975 	STOPEVENT(p, S_SIG, sig);
1976 
1977 	if (action == SIG_DFL) {
1978 		/*
1979 		 * Default action, where the default is to kill
1980 		 * the process.  (Other cases were ignored above.)
1981 		 */
1982 		sigexit(lp, sig);
1983 		/* NOTREACHED */
1984 	} else {
1985 		/*
1986 		 * If we get here, the signal must be caught.
1987 		 */
1988 		KASSERT(action != SIG_IGN && !SIGISMEMBER(lp->lwp_sigmask, sig),
1989 		    ("postsig action"));
1990 
1991 		crit_enter();
1992 
1993 		/*
1994 		 * Reset the signal handler if asked to
1995 		 */
1996 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
1997 			/*
1998 			 * See kern_sigaction() for origin of this code.
1999 			 */
2000 			SIGDELSET(p->p_sigcatch, sig);
2001 			if (sig != SIGCONT &&
2002 			    sigprop(sig) & SA_IGNORE)
2003 				SIGADDSET(p->p_sigignore, sig);
2004 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2005 		}
2006 
2007 		/*
2008 		 * Handle the mailbox case.  Copyout to the appropriate
2009 		 * location but do not generate a signal frame.  The system
2010 		 * call simply returns EINTR and the user is responsible for
2011 		 * polling the mailbox.
2012 		 */
2013 		if (SIGISMEMBER(ps->ps_sigmailbox, sig)) {
2014 			int sig_copy = sig;
2015 			copyout(&sig_copy, (void *)action, sizeof(int));
2016 			curproc->p_flag |= P_MAILBOX;
2017 			crit_exit();
2018 			goto done;
2019 		}
2020 
2021 		/*
2022 		 * Set the signal mask and calculate the mask to restore
2023 		 * when the signal function returns.
2024 		 *
2025 		 * Special case: user has done a sigsuspend.  Here the
2026 		 * current mask is not of interest, but rather the
2027 		 * mask from before the sigsuspend is what we want
2028 		 * restored after the signal processing is completed.
2029 		 */
2030 		if (lp->lwp_flag & LWP_OLDMASK) {
2031 			returnmask = lp->lwp_oldsigmask;
2032 			lp->lwp_flag &= ~LWP_OLDMASK;
2033 		} else {
2034 			returnmask = lp->lwp_sigmask;
2035 		}
2036 
2037 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
2038 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
2039 			SIGADDSET(lp->lwp_sigmask, sig);
2040 
2041 		crit_exit();
2042 		lp->lwp_ru.ru_nsignals++;
2043 		if (lp->lwp_sig != sig) {
2044 			code = 0;
2045 		} else {
2046 			code = lp->lwp_code;
2047 			lp->lwp_code = 0;
2048 			lp->lwp_sig = 0;
2049 		}
2050 		(*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
2051 	}
2052 done:
2053 	;
2054 }
2055 
2056 /*
2057  * Kill the current process for stated reason.
2058  */
2059 void
2060 killproc(struct proc *p, char *why)
2061 {
2062 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n",
2063 		p->p_pid, p->p_comm,
2064 		p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2065 	ksignal(p, SIGKILL);
2066 }
2067 
2068 /*
2069  * Force the current process to exit with the specified signal, dumping core
2070  * if appropriate.  We bypass the normal tests for masked and caught signals,
2071  * allowing unrecoverable failures to terminate the process without changing
2072  * signal state.  Mark the accounting record with the signal termination.
2073  * If dumping core, save the signal number for the debugger.  Calls exit and
2074  * does not return.
2075  *
2076  * This routine does not return.
2077  */
2078 void
2079 sigexit(struct lwp *lp, int sig)
2080 {
2081 	struct proc *p = lp->lwp_proc;
2082 
2083 	lwkt_gettoken(&p->p_token);
2084 	p->p_acflag |= AXSIG;
2085 	if (sigprop(sig) & SA_CORE) {
2086 		lp->lwp_sig = sig;
2087 		/*
2088 		 * Log signals which would cause core dumps
2089 		 * (Log as LOG_INFO to appease those who don't want
2090 		 * these messages.)
2091 		 * XXX : Todo, as well as euid, write out ruid too
2092 		 */
2093 		if (coredump(lp, sig) == 0)
2094 			sig |= WCOREFLAG;
2095 		if (kern_logsigexit)
2096 			log(LOG_INFO,
2097 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2098 			    p->p_pid, p->p_comm,
2099 			    p->p_ucred ? p->p_ucred->cr_uid : -1,
2100 			    sig &~ WCOREFLAG,
2101 			    sig & WCOREFLAG ? " (core dumped)" : "");
2102 	}
2103 	lwkt_reltoken(&p->p_token);
2104 	exit1(W_EXITCODE(0, sig));
2105 	/* NOTREACHED */
2106 }
2107 
2108 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
2109 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
2110 	      sizeof(corefilename), "process corefile name format string");
2111 
2112 /*
2113  * expand_name(name, uid, pid)
2114  * Expand the name described in corefilename, using name, uid, and pid.
2115  * corefilename is a kprintf-like string, with three format specifiers:
2116  *	%N	name of process ("name")
2117  *	%P	process id (pid)
2118  *	%U	user id (uid)
2119  * For example, "%N.core" is the default; they can be disabled completely
2120  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
2121  * This is controlled by the sysctl variable kern.corefile (see above).
2122  */
2123 
2124 static char *
2125 expand_name(const char *name, uid_t uid, pid_t pid)
2126 {
2127 	char *temp;
2128 	char buf[11];		/* Buffer for pid/uid -- max 4B */
2129 	int i, n;
2130 	char *format = corefilename;
2131 	size_t namelen;
2132 
2133 	temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
2134 	if (temp == NULL)
2135 		return NULL;
2136 	namelen = strlen(name);
2137 	for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
2138 		int l;
2139 		switch (format[i]) {
2140 		case '%':	/* Format character */
2141 			i++;
2142 			switch (format[i]) {
2143 			case '%':
2144 				temp[n++] = '%';
2145 				break;
2146 			case 'N':	/* process name */
2147 				if ((n + namelen) > MAXPATHLEN) {
2148 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2149 					    pid, name, uid, temp, name);
2150 					kfree(temp, M_TEMP);
2151 					return NULL;
2152 				}
2153 				memcpy(temp+n, name, namelen);
2154 				n += namelen;
2155 				break;
2156 			case 'P':	/* process id */
2157 				l = ksprintf(buf, "%u", pid);
2158 				if ((n + l) > MAXPATHLEN) {
2159 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2160 					    pid, name, uid, temp, name);
2161 					kfree(temp, M_TEMP);
2162 					return NULL;
2163 				}
2164 				memcpy(temp+n, buf, l);
2165 				n += l;
2166 				break;
2167 			case 'U':	/* user id */
2168 				l = ksprintf(buf, "%u", uid);
2169 				if ((n + l) > MAXPATHLEN) {
2170 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2171 					    pid, name, uid, temp, name);
2172 					kfree(temp, M_TEMP);
2173 					return NULL;
2174 				}
2175 				memcpy(temp+n, buf, l);
2176 				n += l;
2177 				break;
2178 			default:
2179 			  	log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
2180 			}
2181 			break;
2182 		default:
2183 			temp[n++] = format[i];
2184 		}
2185 	}
2186 	temp[n] = '\0';
2187 	return temp;
2188 }
2189 
2190 /*
2191  * Dump a process' core.  The main routine does some
2192  * policy checking, and creates the name of the coredump;
2193  * then it passes on a vnode and a size limit to the process-specific
2194  * coredump routine if there is one; if there _is not_ one, it returns
2195  * ENOSYS; otherwise it returns the error from the process-specific routine.
2196  *
2197  * The parameter `lp' is the lwp which triggered the coredump.
2198  */
2199 
2200 static int
2201 coredump(struct lwp *lp, int sig)
2202 {
2203 	struct proc *p = lp->lwp_proc;
2204 	struct vnode *vp;
2205 	struct ucred *cred = p->p_ucred;
2206 	struct flock lf;
2207 	struct nlookupdata nd;
2208 	struct vattr vattr;
2209 	int error, error1;
2210 	char *name;			/* name of corefile */
2211 	off_t limit;
2212 
2213 	STOPEVENT(p, S_CORE, 0);
2214 
2215 	if (((sugid_coredump == 0) && p->p_flag & P_SUGID) || do_coredump == 0)
2216 		return (EFAULT);
2217 
2218 	/*
2219 	 * Note that the bulk of limit checking is done after
2220 	 * the corefile is created.  The exception is if the limit
2221 	 * for corefiles is 0, in which case we don't bother
2222 	 * creating the corefile at all.  This layout means that
2223 	 * a corefile is truncated instead of not being created,
2224 	 * if it is larger than the limit.
2225 	 */
2226 	limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
2227 	if (limit == 0)
2228 		return EFBIG;
2229 
2230 	name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
2231 	if (name == NULL)
2232 		return (EINVAL);
2233 	error = nlookup_init(&nd, name, UIO_SYSSPACE, NLC_LOCKVP);
2234 	if (error == 0)
2235 		error = vn_open(&nd, NULL, O_CREAT | FWRITE | O_NOFOLLOW, S_IRUSR | S_IWUSR);
2236 	kfree(name, M_TEMP);
2237 	if (error) {
2238 		nlookup_done(&nd);
2239 		return (error);
2240 	}
2241 	vp = nd.nl_open_vp;
2242 	nd.nl_open_vp = NULL;
2243 	nlookup_done(&nd);
2244 
2245 	vn_unlock(vp);
2246 	lf.l_whence = SEEK_SET;
2247 	lf.l_start = 0;
2248 	lf.l_len = 0;
2249 	lf.l_type = F_WRLCK;
2250 	error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, 0);
2251 	if (error)
2252 		goto out2;
2253 
2254 	/* Don't dump to non-regular files or files with links. */
2255 	if (vp->v_type != VREG ||
2256 	    VOP_GETATTR(vp, &vattr) || vattr.va_nlink != 1) {
2257 		error = EFAULT;
2258 		goto out1;
2259 	}
2260 
2261 	/* Don't dump to files current user does not own */
2262 	if (vattr.va_uid != p->p_ucred->cr_uid) {
2263 		error = EFAULT;
2264 		goto out1;
2265 	}
2266 
2267 	VATTR_NULL(&vattr);
2268 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2269 	vattr.va_size = 0;
2270 	VOP_SETATTR(vp, &vattr, cred);
2271 	p->p_acflag |= ACORE;
2272 	vn_unlock(vp);
2273 
2274 	error = p->p_sysent->sv_coredump ?
2275 		  p->p_sysent->sv_coredump(lp, sig, vp, limit) : ENOSYS;
2276 
2277 out1:
2278 	lf.l_type = F_UNLCK;
2279 	VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, 0);
2280 out2:
2281 	error1 = vn_close(vp, FWRITE);
2282 	if (error == 0)
2283 		error = error1;
2284 	return (error);
2285 }
2286 
2287 /*
2288  * Nonexistent system call-- signal process (may want to handle it).
2289  * Flag error in case process won't see signal immediately (blocked or ignored).
2290  *
2291  * MPALMOSTSAFE
2292  */
2293 /* ARGSUSED */
2294 int
2295 sys_nosys(struct nosys_args *args)
2296 {
2297 	lwpsignal(curproc, curthread->td_lwp, SIGSYS);
2298 	return (EINVAL);
2299 }
2300 
2301 /*
2302  * Send a SIGIO or SIGURG signal to a process or process group using
2303  * stored credentials rather than those of the current process.
2304  */
2305 void
2306 pgsigio(struct sigio *sigio, int sig, int checkctty)
2307 {
2308 	if (sigio == NULL)
2309 		return;
2310 
2311 	if (sigio->sio_pgid > 0) {
2312 		if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
2313 		             sigio->sio_proc))
2314 			ksignal(sigio->sio_proc, sig);
2315 	} else if (sigio->sio_pgid < 0) {
2316 		struct proc *p;
2317 		struct pgrp *pg = sigio->sio_pgrp;
2318 
2319 		/*
2320 		 * Must interlock all signals against fork
2321 		 */
2322 		pgref(pg);
2323 		lockmgr(&pg->pg_lock, LK_EXCLUSIVE);
2324 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
2325 			if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
2326 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
2327 				ksignal(p, sig);
2328 		}
2329 		lockmgr(&pg->pg_lock, LK_RELEASE);
2330 		pgrel(pg);
2331 	}
2332 }
2333 
2334 static int
2335 filt_sigattach(struct knote *kn)
2336 {
2337 	struct proc *p = curproc;
2338 
2339 	kn->kn_ptr.p_proc = p;
2340 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
2341 
2342 	/* XXX lock the proc here while adding to the list? */
2343 	knote_insert(&p->p_klist, kn);
2344 
2345 	return (0);
2346 }
2347 
2348 static void
2349 filt_sigdetach(struct knote *kn)
2350 {
2351 	struct proc *p = kn->kn_ptr.p_proc;
2352 
2353 	knote_remove(&p->p_klist, kn);
2354 }
2355 
2356 /*
2357  * signal knotes are shared with proc knotes, so we apply a mask to
2358  * the hint in order to differentiate them from process hints.  This
2359  * could be avoided by using a signal-specific knote list, but probably
2360  * isn't worth the trouble.
2361  */
2362 static int
2363 filt_signal(struct knote *kn, long hint)
2364 {
2365 	if (hint & NOTE_SIGNAL) {
2366 		hint &= ~NOTE_SIGNAL;
2367 
2368 		if (kn->kn_id == hint)
2369 			kn->kn_data++;
2370 	}
2371 	return (kn->kn_data != 0);
2372 }
2373