xref: /dflybsd-src/sys/kern/kern_sig.c (revision 99512ac4b46c423d7a70b99c3d0c20769afaf083)
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 		} else {
696 			pgrp = pgfind(pgid);
697 			if (pgrp == NULL)
698 				return (ESRCH);
699 		}
700 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
701 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
702 			if (p->p_pid <= 1 ||
703 			    p->p_stat == SZOMB ||
704 			    (p->p_flag & P_SYSTEM) ||
705 			    !CANSIGNAL(p, sig)) {
706 				continue;
707 			}
708 			++info.nfound;
709 			if (sig)
710 				ksignal(p, sig);
711 		}
712 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
713 	}
714 	return (info.nfound ? 0 : ESRCH);
715 }
716 
717 static int
718 killpg_all_callback(struct proc *p, void *data)
719 {
720 	struct killpg_info *info = data;
721 
722 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) ||
723 	    p == curproc || !CANSIGNAL(p, info->sig)) {
724 		return (0);
725 	}
726 	++info->nfound;
727 	if (info->sig)
728 		ksignal(p, info->sig);
729 	return(0);
730 }
731 
732 /*
733  * Send a general signal to a process or LWPs within that process.  Note
734  * that new signals cannot be sent if a process is exiting.
735  *
736  * No requirements.
737  */
738 int
739 kern_kill(int sig, pid_t pid, lwpid_t tid)
740 {
741 	int t;
742 
743 	if ((u_int)sig > _SIG_MAXSIG)
744 		return (EINVAL);
745 
746 	lwkt_gettoken(&proc_token);
747 
748 	if (pid > 0) {
749 		struct proc *p;
750 		struct lwp *lp = NULL;
751 
752 		/* kill single process */
753 		if ((p = pfind(pid)) == NULL) {
754 			lwkt_reltoken(&proc_token);
755 			return (ESRCH);
756 		}
757 		PHOLD(p);
758 		lwkt_gettoken(&p->p_token);
759 		if (!CANSIGNAL(p, sig)) {
760 			lwkt_reltoken(&p->p_token);
761 			PRELE(p);
762 			lwkt_reltoken(&proc_token);
763 			return (EPERM);
764 		}
765 
766 		/*
767 		 * NOP if the process is exiting.  Note that lwpsignal() is
768 		 * called directly with P_WEXIT set to kill individual LWPs
769 		 * during exit, which is allowed.
770 		 */
771 		if (p->p_flag & P_WEXIT) {
772 			lwkt_reltoken(&p->p_token);
773 			PRELE(p);
774 			lwkt_reltoken(&proc_token);
775 			return (0);
776 		}
777 		if (tid != -1) {
778 			lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, tid);
779 			if (lp == NULL) {
780 				lwkt_reltoken(&p->p_token);
781 				PRELE(p);
782 				lwkt_reltoken(&proc_token);
783 				return (ESRCH);
784 			}
785 		}
786 		if (sig)
787 			lwpsignal(p, lp, sig);
788 		lwkt_reltoken(&p->p_token);
789 		PRELE(p);
790 		lwkt_reltoken(&proc_token);
791 		return (0);
792 	}
793 
794 	/*
795 	 * If we come here, pid is a special broadcast pid.
796 	 * This doesn't mix with a tid.
797 	 */
798 	if (tid != -1) {
799 		lwkt_reltoken(&proc_token);
800 		return (EINVAL);
801 	}
802 	switch (pid) {
803 	case -1:		/* broadcast signal */
804 		t = (dokillpg(sig, 0, 1));
805 		break;
806 	case 0:			/* signal own process group */
807 		t = (dokillpg(sig, 0, 0));
808 		break;
809 	default:		/* negative explicit process group */
810 		t = (dokillpg(sig, -pid, 0));
811 		break;
812 	}
813 	lwkt_reltoken(&proc_token);
814 	return t;
815 }
816 
817 int
818 sys_kill(struct kill_args *uap)
819 {
820 	int error;
821 
822 	error = kern_kill(uap->signum, uap->pid, -1);
823 	return (error);
824 }
825 
826 int
827 sys_lwp_kill(struct lwp_kill_args *uap)
828 {
829 	int error;
830 	pid_t pid = uap->pid;
831 
832 	/*
833 	 * A tid is mandatory for lwp_kill(), otherwise
834 	 * you could simply use kill().
835 	 */
836 	if (uap->tid == -1)
837 		return (EINVAL);
838 
839 	/*
840 	 * To save on a getpid() function call for intra-process
841 	 * signals, pid == -1 means current process.
842 	 */
843 	if (pid == -1)
844 		pid = curproc->p_pid;
845 
846 	error = kern_kill(uap->signum, pid, uap->tid);
847 	return (error);
848 }
849 
850 /*
851  * Send a signal to a process group.
852  */
853 void
854 gsignal(int pgid, int sig)
855 {
856 	struct pgrp *pgrp;
857 
858 	if (pgid && (pgrp = pgfind(pgid)))
859 		pgsignal(pgrp, sig, 0);
860 }
861 
862 /*
863  * Send a signal to a process group.  If checktty is 1,
864  * limit to members which have a controlling terminal.
865  *
866  * pg_lock interlocks against a fork that might be in progress, to
867  * ensure that the new child process picks up the signal.
868  */
869 void
870 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
871 {
872 	struct proc *p;
873 
874 	if (pgrp) {
875 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
876 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
877 			if (checkctty == 0 || p->p_flag & P_CONTROLT)
878 				ksignal(p, sig);
879 		}
880 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
881 	}
882 }
883 
884 /*
885  * Send a signal caused by a trap to the current lwp.  If it will be caught
886  * immediately, deliver it with correct code.  Otherwise, post it normally.
887  *
888  * These signals may ONLY be delivered to the specified lwp and may never
889  * be delivered to the process generically.
890  */
891 void
892 trapsignal(struct lwp *lp, int sig, u_long code)
893 {
894 	struct proc *p = lp->lwp_proc;
895 	struct sigacts *ps = p->p_sigacts;
896 
897 	/*
898 	 * If we are a virtual kernel running an emulated user process
899 	 * context, switch back to the virtual kernel context before
900 	 * trying to post the signal.
901 	 */
902 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
903 		struct trapframe *tf = lp->lwp_md.md_regs;
904 		tf->tf_trapno = 0;
905 		vkernel_trap(lp, tf);
906 	}
907 
908 
909 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(p->p_sigcatch, sig) &&
910 	    !SIGISMEMBER(lp->lwp_sigmask, sig)) {
911 		lp->lwp_ru.ru_nsignals++;
912 #ifdef KTRACE
913 		if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
914 			ktrpsig(lp, sig, ps->ps_sigact[_SIG_IDX(sig)],
915 				&lp->lwp_sigmask, code);
916 #endif
917 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], sig,
918 						&lp->lwp_sigmask, code);
919 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
920 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
921 			SIGADDSET(lp->lwp_sigmask, sig);
922 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
923 			/*
924 			 * See kern_sigaction() for origin of this code.
925 			 */
926 			SIGDELSET(p->p_sigcatch, sig);
927 			if (sig != SIGCONT &&
928 			    sigprop(sig) & SA_IGNORE)
929 				SIGADDSET(p->p_sigignore, sig);
930 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
931 		}
932 	} else {
933 		lp->lwp_code = code;	/* XXX for core dump/debugger */
934 		lp->lwp_sig = sig;	/* XXX to verify code */
935 		lwpsignal(p, lp, sig);
936 	}
937 }
938 
939 /*
940  * Find a suitable lwp to deliver the signal to.
941  *
942  * Returns NULL if all lwps hold the signal blocked.
943  */
944 static struct lwp *
945 find_lwp_for_signal(struct proc *p, int sig)
946 {
947 	struct lwp *lp;
948 	struct lwp *run, *sleep, *stop;
949 
950 	/*
951 	 * If the running/preempted thread belongs to the proc to which
952 	 * the signal is being delivered and this thread does not block
953 	 * the signal, then we can avoid a context switch by delivering
954 	 * the signal to this thread, because it will return to userland
955 	 * soon anyways.
956 	 */
957 	lp = lwkt_preempted_proc();
958 	if (lp != NULL && lp->lwp_proc == p && !SIGISMEMBER(lp->lwp_sigmask, sig))
959 		return (lp);
960 
961 	run = sleep = stop = NULL;
962 	FOREACH_LWP_IN_PROC(lp, p) {
963 		/*
964 		 * If the signal is being blocked by the lwp, then this
965 		 * lwp is not eligible for receiving the signal.
966 		 */
967 		if (SIGISMEMBER(lp->lwp_sigmask, sig))
968 			continue;
969 
970 		switch (lp->lwp_stat) {
971 		case LSRUN:
972 			run = lp;
973 			break;
974 
975 		case LSSTOP:
976 			stop = lp;
977 			break;
978 
979 		case LSSLEEP:
980 			if (lp->lwp_flag & LWP_SINTR)
981 				sleep = lp;
982 			break;
983 		}
984 	}
985 
986 	if (run != NULL)
987 		return (run);
988 	else if (sleep != NULL)
989 		return (sleep);
990 	else
991 		return (stop);
992 }
993 
994 /*
995  * Send the signal to the process.  If the signal has an action, the action
996  * is usually performed by the target process rather than the caller; we add
997  * the signal to the set of pending signals for the process.
998  *
999  * Exceptions:
1000  *   o When a stop signal is sent to a sleeping process that takes the
1001  *     default action, the process is stopped without awakening it.
1002  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
1003  *     regardless of the signal action (eg, blocked or ignored).
1004  *
1005  * Other ignored signals are discarded immediately.
1006  *
1007  * No requirements.
1008  */
1009 void
1010 ksignal(struct proc *p, int sig)
1011 {
1012 	lwpsignal(p, NULL, sig);
1013 }
1014 
1015 /*
1016  * The core for ksignal.  lp may be NULL, then a suitable thread
1017  * will be chosen.  If not, lp MUST be a member of p.
1018  *
1019  * No requirements.
1020  */
1021 void
1022 lwpsignal(struct proc *p, struct lwp *lp, int sig)
1023 {
1024 	int prop;
1025 	sig_t action;
1026 
1027 	if (sig > _SIG_MAXSIG || sig <= 0) {
1028 		kprintf("lwpsignal: signal %d\n", sig);
1029 		panic("lwpsignal signal number");
1030 	}
1031 
1032 	KKASSERT(lp == NULL || lp->lwp_proc == p);
1033 
1034 	lwkt_gettoken(&p->p_token);
1035 
1036 	prop = sigprop(sig);
1037 
1038 	/*
1039 	 * If proc is traced, always give parent a chance;
1040 	 * if signal event is tracked by procfs, give *that*
1041 	 * a chance, as well.
1042 	 */
1043 	if ((p->p_flag & P_TRACED) || (p->p_stops & S_SIG)) {
1044 		action = SIG_DFL;
1045 	} else {
1046 		/*
1047 		 * Do not try to deliver signals to an exiting lwp.  Note
1048 		 * that we must still deliver the signal if P_WEXIT is set
1049 		 * in the process flags.
1050 		 */
1051 		if (lp && (lp->lwp_flag & LWP_WEXIT)) {
1052 			lwkt_reltoken(&p->p_token);
1053 			return;
1054 		}
1055 
1056 		/*
1057 		 * If the signal is being ignored, then we forget about
1058 		 * it immediately.  NOTE: We don't set SIGCONT in p_sigignore,
1059 		 * and if it is set to SIG_IGN, action will be SIG_DFL here.
1060 		 */
1061 		if (SIGISMEMBER(p->p_sigignore, sig)) {
1062 			lwkt_reltoken(&p->p_token);
1063 			return;
1064 		}
1065 		if (SIGISMEMBER(p->p_sigcatch, sig))
1066 			action = SIG_CATCH;
1067 		else
1068 			action = SIG_DFL;
1069 	}
1070 
1071 	/*
1072 	 * If continuing, clear any pending STOP signals.
1073 	 */
1074 	if (prop & SA_CONT)
1075 		SIG_STOPSIGMASK(p->p_siglist);
1076 
1077 	if (prop & SA_STOP) {
1078 		/*
1079 		 * If sending a tty stop signal to a member of an orphaned
1080 		 * process group, discard the signal here if the action
1081 		 * is default; don't stop the process below if sleeping,
1082 		 * and don't clear any pending SIGCONT.
1083 		 */
1084 		if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0 &&
1085 		    action == SIG_DFL) {
1086 			lwkt_reltoken(&p->p_token);
1087 		        return;
1088 		}
1089 		SIG_CONTSIGMASK(p->p_siglist);
1090 		p->p_flag &= ~P_CONTINUED;
1091 	}
1092 
1093 	crit_enter();
1094 
1095 	if (p->p_stat == SSTOP) {
1096 		/*
1097 		 * Nobody can handle this signal, add it to the lwp or
1098 		 * process pending list
1099 		 */
1100 		if (lp)
1101 			SIGADDSET(lp->lwp_siglist, sig);
1102 		else
1103 			SIGADDSET(p->p_siglist, sig);
1104 
1105 		/*
1106 		 * If the process is stopped and is being traced, then no
1107 		 * further action is necessary.
1108 		 */
1109 		if (p->p_flag & P_TRACED)
1110 			goto out;
1111 
1112 		/*
1113 		 * If the process is stopped and receives a KILL signal,
1114 		 * make the process runnable.
1115 		 */
1116 		if (sig == SIGKILL) {
1117 			proc_unstop(p);
1118 			goto active_process;
1119 		}
1120 
1121 		/*
1122 		 * If the process is stopped and receives a CONT signal,
1123 		 * then try to make the process runnable again.
1124 		 */
1125 		if (prop & SA_CONT) {
1126 			/*
1127 			 * If SIGCONT is default (or ignored), we continue the
1128 			 * process but don't leave the signal in p_siglist, as
1129 			 * it has no further action.  If SIGCONT is held, we
1130 			 * continue the process and leave the signal in
1131 			 * p_siglist.  If the process catches SIGCONT, let it
1132 			 * handle the signal itself.
1133 			 */
1134 			/* XXX what if the signal is being held blocked? */
1135 			p->p_flag |= P_CONTINUED;
1136 			wakeup(p->p_pptr);
1137 			if (action == SIG_DFL)
1138 				SIGDELSET(p->p_siglist, sig);
1139 			proc_unstop(p);
1140 			if (action == SIG_CATCH)
1141 				goto active_process;
1142 			goto out;
1143 		}
1144 
1145 		/*
1146 		 * If the process is stopped and receives another STOP
1147 		 * signal, we do not need to stop it again.  If we did
1148 		 * the shell could get confused.
1149 		 *
1150 		 * However, if the current/preempted lwp is part of the
1151 		 * process receiving the signal, we need to keep it,
1152 		 * so that this lwp can stop in issignal() later, as
1153 		 * we don't want to wait until it reaches userret!
1154 		 */
1155 		if (prop & SA_STOP) {
1156 			if (lwkt_preempted_proc() == NULL ||
1157 			    lwkt_preempted_proc()->lwp_proc != p)
1158 				SIGDELSET(p->p_siglist, sig);
1159 		}
1160 
1161 		/*
1162 		 * Otherwise the process is stopped and it received some
1163 		 * signal, which does not change its stopped state.
1164 		 *
1165 		 * We have to select one thread to set LWP_BREAKTSLEEP,
1166 		 * so that the current signal will break the sleep
1167 		 * as soon as a SA_CONT signal will unstop the process.
1168 		 */
1169 		if (lp == NULL)
1170 			lp = find_lwp_for_signal(p, sig);
1171 		if (lp != NULL &&
1172 		    (lp->lwp_stat == LSSLEEP || lp->lwp_stat == LSSTOP))
1173 			lp->lwp_flag |= LWP_BREAKTSLEEP;
1174 		goto out;
1175 
1176 		/* NOTREACHED */
1177 	}
1178 	/* else not stopped */
1179 active_process:
1180 
1181 	/*
1182 	 * Never deliver a lwp-specific signal to a random lwp.
1183 	 */
1184 	if (lp == NULL) {
1185 		lp = find_lwp_for_signal(p, sig);
1186 		if (lp && SIGISMEMBER(lp->lwp_sigmask, sig))
1187 			lp = NULL;
1188 	}
1189 
1190 	/*
1191 	 * Deliver to the process generically if (1) the signal is being
1192 	 * sent to any thread or (2) we could not find a thread to deliver
1193 	 * it to.
1194 	 */
1195 	if (lp == NULL) {
1196 		SIGADDSET(p->p_siglist, sig);
1197 		goto out;
1198 	}
1199 
1200 	/*
1201 	 * Deliver to a specific LWP whether it masks it or not.  It will
1202 	 * not be dispatched if masked but we must still deliver it.
1203 	 */
1204 	if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
1205 	    (p->p_flag & P_TRACED) == 0) {
1206 		p->p_nice = NZERO;
1207 	}
1208 
1209 	/*
1210 	 * If the process receives a STOP signal which indeed needs to
1211 	 * stop the process, do so.  If the process chose to catch the
1212 	 * signal, it will be treated like any other signal.
1213 	 */
1214 	if ((prop & SA_STOP) && action == SIG_DFL) {
1215 		/*
1216 		 * If a child holding parent blocked, stopping
1217 		 * could cause deadlock.  Take no action at this
1218 		 * time.
1219 		 */
1220 		if (p->p_flag & P_PPWAIT) {
1221 			SIGADDSET(p->p_siglist, sig);
1222 			goto out;
1223 		}
1224 
1225 		/*
1226 		 * Do not actually try to manipulate the process, but simply
1227 		 * stop it.  Lwps will stop as soon as they safely can.
1228 		 */
1229 		p->p_xstat = sig;
1230 		proc_stop(p);
1231 		goto out;
1232 	}
1233 
1234 	/*
1235 	 * If it is a CONT signal with default action, just ignore it.
1236 	 */
1237 	if ((prop & SA_CONT) && action == SIG_DFL)
1238 		goto out;
1239 
1240 	/*
1241 	 * Mark signal pending at this specific thread.
1242 	 */
1243 	SIGADDSET(lp->lwp_siglist, sig);
1244 
1245 	lwp_signotify(lp);
1246 
1247 out:
1248 	lwkt_reltoken(&p->p_token);
1249 	crit_exit();
1250 }
1251 
1252 /*
1253  * p->p_token must be held
1254  */
1255 static void
1256 lwp_signotify(struct lwp *lp)
1257 {
1258 	ASSERT_LWKT_TOKEN_HELD(&lp->lwp_proc->p_token);
1259 	crit_enter();
1260 
1261 	if (lp->lwp_stat == LSSLEEP || lp->lwp_stat == LSSTOP) {
1262 		/*
1263 		 * Thread is in tsleep.
1264 		 */
1265 
1266 		/*
1267 		 * If the thread is sleeping uninterruptibly
1268 		 * we can't interrupt the sleep... the signal will
1269 		 * be noticed when the lwp returns through
1270 		 * trap() or syscall().
1271 		 *
1272 		 * Otherwise the signal can interrupt the sleep.
1273 		 *
1274 		 * If the process is traced, the lwp will handle the
1275 		 * tracing in issignal() when it returns to userland.
1276 		 */
1277 		if (lp->lwp_flag & LWP_SINTR) {
1278 			/*
1279 			 * Make runnable and break out of any tsleep as well.
1280 			 */
1281 			lp->lwp_flag |= LWP_BREAKTSLEEP;
1282 			setrunnable(lp);
1283 		}
1284 	} else {
1285 		/*
1286 		 * Otherwise the thread is running
1287 		 *
1288 		 * LSRUN does nothing with the signal, other than kicking
1289 		 * ourselves if we are running.
1290 		 * SZOMB and SIDL mean that it will either never be noticed,
1291 		 * or noticed very soon.
1292 		 *
1293 		 * Note that lwp_thread may be NULL or may not be completely
1294 		 * initialized if the process is in the SIDL or SZOMB state.
1295 		 *
1296 		 * For SMP we may have to forward the request to another cpu.
1297 		 * YYY the MP lock prevents the target process from moving
1298 		 * to another cpu, see kern/kern_switch.c
1299 		 *
1300 		 * If the target thread is waiting on its message port,
1301 		 * wakeup the target thread so it can check (or ignore)
1302 		 * the new signal.  YYY needs cleanup.
1303 		 */
1304 		if (lp == lwkt_preempted_proc()) {
1305 			signotify();
1306 		} else if (lp->lwp_stat == LSRUN) {
1307 			struct thread *td = lp->lwp_thread;
1308 			struct proc *p __debugvar = lp->lwp_proc;
1309 
1310 			KASSERT(td != NULL,
1311 			    ("pid %d/%d NULL lwp_thread stat %d flags %08x/%08x",
1312 			    p->p_pid, lp->lwp_tid, lp->lwp_stat,
1313 			    p->p_flag, lp->lwp_flag));
1314 
1315 			/*
1316 			 * To prevent a MP race with TDF_SINTR we must
1317 			 * schedule the thread on the correct cpu.
1318 			 */
1319 #ifdef SMP
1320 			if (td->td_gd != mycpu) {
1321 				LWPHOLD(lp);
1322 				lwkt_send_ipiq(td->td_gd, signotify_remote, lp);
1323 			} else
1324 #endif
1325 			if (td->td_flags & TDF_SINTR)
1326 				lwkt_schedule(td);
1327 		}
1328 	}
1329 	crit_exit();
1330 }
1331 
1332 #ifdef SMP
1333 
1334 /*
1335  * This function is called via an IPI.  We will be in a critical section but
1336  * the MP lock will NOT be held.  The passed lp will be held.
1337  *
1338  * We must essentially repeat the code at the end of lwp_signotify(),
1339  * in particular rechecking all races.  If we are still not on the
1340  * correct cpu we leave the lwp ref intact and continue the chase.
1341  *
1342  * XXX this may still not be entirely correct, since we are checking
1343  *     lwp_stat asynchronously.
1344  */
1345 static void
1346 signotify_remote(void *arg)
1347 {
1348 	struct lwp *lp = arg;
1349 	thread_t td;
1350 
1351 	if (lp == lwkt_preempted_proc()) {
1352 		signotify();
1353 	} else if (lp->lwp_stat == LSRUN) {
1354 		/*
1355 		 * To prevent a MP race with TDF_SINTR we must
1356 		 * schedule the thread on the correct cpu.
1357 		 */
1358 		td = lp->lwp_thread;
1359 		if (td->td_gd != mycpu) {
1360 			lwkt_send_ipiq(td->td_gd, signotify_remote, lp);
1361 			return;
1362 			/* NOT REACHED */
1363 		}
1364 		if (td->td_flags & TDF_SINTR)
1365 			lwkt_schedule(td);
1366 	}
1367 	LWPRELE(lp);
1368 }
1369 
1370 #endif
1371 
1372 /*
1373  * Caller must hold p->p_token
1374  */
1375 void
1376 proc_stop(struct proc *p)
1377 {
1378 	struct lwp *lp;
1379 
1380 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1381 	crit_enter();
1382 
1383 	/* If somebody raced us, be happy with it */
1384 	if (p->p_stat == SSTOP || p->p_stat == SZOMB) {
1385 		crit_exit();
1386 		return;
1387 	}
1388 	p->p_stat = SSTOP;
1389 
1390 	FOREACH_LWP_IN_PROC(lp, p) {
1391 		switch (lp->lwp_stat) {
1392 		case LSSTOP:
1393 			/*
1394 			 * Do nothing, we are already counted in
1395 			 * p_nstopped.
1396 			 */
1397 			break;
1398 
1399 		case LSSLEEP:
1400 			/*
1401 			 * We're sleeping, but we will stop before
1402 			 * returning to userspace, so count us
1403 			 * as stopped as well.  We set LWP_WSTOP
1404 			 * to signal the lwp that it should not
1405 			 * increase p_nstopped when reaching tstop().
1406 			 */
1407 			if ((lp->lwp_flag & LWP_WSTOP) == 0) {
1408 				lp->lwp_flag |= LWP_WSTOP;
1409 				++p->p_nstopped;
1410 			}
1411 			break;
1412 
1413 		case LSRUN:
1414 			/*
1415 			 * We might notify ourself, but that's not
1416 			 * a problem.
1417 			 */
1418 			lwp_signotify(lp);
1419 			break;
1420 		}
1421 	}
1422 
1423 	if (p->p_nstopped == p->p_nthreads) {
1424 		p->p_flag &= ~P_WAITED;
1425 		wakeup(p->p_pptr);
1426 		if ((p->p_pptr->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1427 			ksignal(p->p_pptr, SIGCHLD);
1428 	}
1429 	crit_exit();
1430 }
1431 
1432 /*
1433  * Caller must hold proc_token
1434  */
1435 void
1436 proc_unstop(struct proc *p)
1437 {
1438 	struct lwp *lp;
1439 
1440 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1441 	crit_enter();
1442 
1443 	if (p->p_stat != SSTOP) {
1444 		crit_exit();
1445 		return;
1446 	}
1447 
1448 	p->p_stat = SACTIVE;
1449 
1450 	FOREACH_LWP_IN_PROC(lp, p) {
1451 		switch (lp->lwp_stat) {
1452 		case LSRUN:
1453 			/*
1454 			 * Uh?  Not stopped?  Well, I guess that's okay.
1455 			 */
1456 			if (bootverbose)
1457 				kprintf("proc_unstop: lwp %d/%d not sleeping\n",
1458 					p->p_pid, lp->lwp_tid);
1459 			break;
1460 
1461 		case LSSLEEP:
1462 			/*
1463 			 * Still sleeping.  Don't bother waking it up.
1464 			 * However, if this thread was counted as
1465 			 * stopped, undo this.
1466 			 *
1467 			 * Nevertheless we call setrunnable() so that it
1468 			 * will wake up in case a signal or timeout arrived
1469 			 * in the meantime.
1470 			 */
1471 			if (lp->lwp_flag & LWP_WSTOP) {
1472 				lp->lwp_flag &= ~LWP_WSTOP;
1473 				--p->p_nstopped;
1474 			} else {
1475 				if (bootverbose)
1476 					kprintf("proc_unstop: lwp %d/%d sleeping, not stopped\n",
1477 						p->p_pid, lp->lwp_tid);
1478 			}
1479 			/* FALLTHROUGH */
1480 
1481 		case LSSTOP:
1482 			setrunnable(lp);
1483 			break;
1484 
1485 		}
1486 	}
1487 	crit_exit();
1488 }
1489 
1490 /*
1491  * No requirements.
1492  */
1493 static int
1494 kern_sigtimedwait(sigset_t waitset, siginfo_t *info, struct timespec *timeout)
1495 {
1496 	sigset_t savedmask, set;
1497 	struct proc *p = curproc;
1498 	struct lwp *lp = curthread->td_lwp;
1499 	int error, sig, hz, timevalid = 0;
1500 	struct timespec rts, ets, ts;
1501 	struct timeval tv;
1502 
1503 	error = 0;
1504 	sig = 0;
1505 	ets.tv_sec = 0;		/* silence compiler warning */
1506 	ets.tv_nsec = 0;	/* silence compiler warning */
1507 	SIG_CANTMASK(waitset);
1508 	savedmask = lp->lwp_sigmask;
1509 
1510 	if (timeout) {
1511 		if (timeout->tv_sec >= 0 && timeout->tv_nsec >= 0 &&
1512 		    timeout->tv_nsec < 1000000000) {
1513 			timevalid = 1;
1514 			getnanouptime(&rts);
1515 		 	ets = rts;
1516 			timespecadd(&ets, timeout);
1517 		}
1518 	}
1519 
1520 	for (;;) {
1521 		set = lwp_sigpend(lp);
1522 		SIGSETAND(set, waitset);
1523 		if ((sig = sig_ffs(&set)) != 0) {
1524 			SIGFILLSET(lp->lwp_sigmask);
1525 			SIGDELSET(lp->lwp_sigmask, sig);
1526 			SIG_CANTMASK(lp->lwp_sigmask);
1527 			sig = issignal(lp, 1);
1528 			/*
1529 			 * It may be a STOP signal, in the case, issignal
1530 			 * returns 0, because we may stop there, and new
1531 			 * signal can come in, we should restart if we got
1532 			 * nothing.
1533 			 */
1534 			if (sig == 0)
1535 				continue;
1536 			else
1537 				break;
1538 		}
1539 
1540 		/*
1541 		 * Previous checking got nothing, and we retried but still
1542 		 * got nothing, we should return the error status.
1543 		 */
1544 		if (error)
1545 			break;
1546 
1547 		/*
1548 		 * POSIX says this must be checked after looking for pending
1549 		 * signals.
1550 		 */
1551 		if (timeout) {
1552 			if (timevalid == 0) {
1553 				error = EINVAL;
1554 				break;
1555 			}
1556 			getnanouptime(&rts);
1557 			if (timespeccmp(&rts, &ets, >=)) {
1558 				error = EAGAIN;
1559 				break;
1560 			}
1561 			ts = ets;
1562 			timespecsub(&ts, &rts);
1563 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1564 			hz = tvtohz_high(&tv);
1565 		} else {
1566 			hz = 0;
1567 		}
1568 
1569 		lp->lwp_sigmask = savedmask;
1570 		SIGSETNAND(lp->lwp_sigmask, waitset);
1571 		/*
1572 		 * We won't ever be woken up.  Instead, our sleep will
1573 		 * be broken in lwpsignal().
1574 		 */
1575 		error = tsleep(&p->p_sigacts, PCATCH, "sigwt", hz);
1576 		if (timeout) {
1577 			if (error == ERESTART) {
1578 				/* can not restart a timeout wait. */
1579 				error = EINTR;
1580 			} else if (error == EAGAIN) {
1581 				/* will calculate timeout by ourself. */
1582 				error = 0;
1583 			}
1584 		}
1585 		/* Retry ... */
1586 	}
1587 
1588 	lp->lwp_sigmask = savedmask;
1589 	if (sig) {
1590 		error = 0;
1591 		bzero(info, sizeof(*info));
1592 		info->si_signo = sig;
1593 		lwp_delsig(lp, sig);	/* take the signal! */
1594 
1595 		if (sig == SIGKILL) {
1596 			sigexit(lp, sig);
1597 			/* NOT REACHED */
1598 		}
1599 	}
1600 
1601 	return (error);
1602 }
1603 
1604 /*
1605  * MPALMOSTSAFE
1606  */
1607 int
1608 sys_sigtimedwait(struct sigtimedwait_args *uap)
1609 {
1610 	struct timespec ts;
1611 	struct timespec *timeout;
1612 	sigset_t set;
1613 	siginfo_t info;
1614 	int error;
1615 
1616 	if (uap->timeout) {
1617 		error = copyin(uap->timeout, &ts, sizeof(ts));
1618 		if (error)
1619 			return (error);
1620 		timeout = &ts;
1621 	} else {
1622 		timeout = NULL;
1623 	}
1624 	error = copyin(uap->set, &set, sizeof(set));
1625 	if (error)
1626 		return (error);
1627 	error = kern_sigtimedwait(set, &info, timeout);
1628 	if (error)
1629 		return (error);
1630  	if (uap->info)
1631 		error = copyout(&info, uap->info, sizeof(info));
1632 	/* Repost if we got an error. */
1633 	/*
1634 	 * XXX lwp
1635 	 *
1636 	 * This could transform a thread-specific signal to another
1637 	 * thread / process pending signal.
1638 	 */
1639 	if (error) {
1640 		ksignal(curproc, info.si_signo);
1641 	} else {
1642 		uap->sysmsg_result = info.si_signo;
1643 	}
1644 	return (error);
1645 }
1646 
1647 /*
1648  * MPALMOSTSAFE
1649  */
1650 int
1651 sys_sigwaitinfo(struct sigwaitinfo_args *uap)
1652 {
1653 	siginfo_t info;
1654 	sigset_t set;
1655 	int error;
1656 
1657 	error = copyin(uap->set, &set, sizeof(set));
1658 	if (error)
1659 		return (error);
1660 	error = kern_sigtimedwait(set, &info, NULL);
1661 	if (error)
1662 		return (error);
1663 	if (uap->info)
1664 		error = copyout(&info, uap->info, sizeof(info));
1665 	/* Repost if we got an error. */
1666 	/*
1667 	 * XXX lwp
1668 	 *
1669 	 * This could transform a thread-specific signal to another
1670 	 * thread / process pending signal.
1671 	 */
1672 	if (error) {
1673 		ksignal(curproc, info.si_signo);
1674 	} else {
1675 		uap->sysmsg_result = info.si_signo;
1676 	}
1677 	return (error);
1678 }
1679 
1680 /*
1681  * If the current process has received a signal that would interrupt a
1682  * system call, return EINTR or ERESTART as appropriate.
1683  */
1684 int
1685 iscaught(struct lwp *lp)
1686 {
1687 	struct proc *p = lp->lwp_proc;
1688 	int sig;
1689 
1690 	if (p) {
1691 		if ((sig = CURSIG(lp)) != 0) {
1692 			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1693 				return (EINTR);
1694 			return (ERESTART);
1695 		}
1696 	}
1697 	return(EWOULDBLOCK);
1698 }
1699 
1700 /*
1701  * If the current process has received a signal (should be caught or cause
1702  * termination, should interrupt current syscall), return the signal number.
1703  * Stop signals with default action are processed immediately, then cleared;
1704  * they aren't returned.  This is checked after each entry to the system for
1705  * a syscall or trap (though this can usually be done without calling issignal
1706  * by checking the pending signal masks in the CURSIG macro).
1707  *
1708  * This routine is called via CURSIG/__cursig.  We will acquire and release
1709  * p->p_token but if the caller needs to interlock the test the caller must
1710  * also hold p->p_token.
1711  *
1712  *	while (sig = CURSIG(curproc))
1713  *		postsig(sig);
1714  *
1715  * MPSAFE
1716  */
1717 int
1718 issignal(struct lwp *lp, int maytrace)
1719 {
1720 	struct proc *p = lp->lwp_proc;
1721 	sigset_t mask;
1722 	int sig, prop;
1723 
1724 	lwkt_gettoken(&p->p_token);
1725 
1726 	for (;;) {
1727 		int traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
1728 
1729 		/*
1730 		 * If this process is supposed to stop, stop this thread.
1731 		 */
1732 		if (p->p_stat == SSTOP)
1733 			tstop();
1734 
1735 		mask = lwp_sigpend(lp);
1736 		SIGSETNAND(mask, lp->lwp_sigmask);
1737 		if (p->p_flag & P_PPWAIT)
1738 			SIG_STOPSIGMASK(mask);
1739 		if (SIGISEMPTY(mask)) {		/* no signal to send */
1740 			lwkt_reltoken(&p->p_token);
1741 			return (0);
1742 		}
1743 		sig = sig_ffs(&mask);
1744 
1745 		STOPEVENT(p, S_SIG, sig);
1746 
1747 		/*
1748 		 * We should see pending but ignored signals
1749 		 * only if P_TRACED was on when they were posted.
1750 		 */
1751 		if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
1752 			lwp_delsig(lp, sig);
1753 			continue;
1754 		}
1755 		if (maytrace && (p->p_flag & P_TRACED) && (p->p_flag & P_PPWAIT) == 0) {
1756 			/*
1757 			 * If traced, always stop, and stay stopped until
1758 			 * released by the parent.
1759 			 *
1760 			 * NOTE: SSTOP may get cleared during the loop,
1761 			 * but we do not re-notify the parent if we have
1762 			 * to loop several times waiting for the parent
1763 			 * to let us continue.
1764 			 *
1765 			 * XXX not sure if this is still true
1766 			 */
1767 			p->p_xstat = sig;
1768 			proc_stop(p);
1769 			do {
1770 				tstop();
1771 			} while (!trace_req(p) && (p->p_flag & P_TRACED));
1772 
1773 			/*
1774 			 * If parent wants us to take the signal,
1775 			 * then it will leave it in p->p_xstat;
1776 			 * otherwise we just look for signals again.
1777 			 */
1778 			lwp_delsig(lp, sig);	/* clear old signal */
1779 			sig = p->p_xstat;
1780 			if (sig == 0)
1781 				continue;
1782 
1783 			/*
1784 			 * Put the new signal into p_siglist.  If the
1785 			 * signal is being masked, look for other signals.
1786 			 *
1787 			 * XXX lwp might need a call to ksignal()
1788 			 */
1789 			SIGADDSET(p->p_siglist, sig);
1790 			if (SIGISMEMBER(lp->lwp_sigmask, sig))
1791 				continue;
1792 
1793 			/*
1794 			 * If the traced bit got turned off, go back up
1795 			 * to the top to rescan signals.  This ensures
1796 			 * that p_sig* and ps_sigact are consistent.
1797 			 */
1798 			if ((p->p_flag & P_TRACED) == 0)
1799 				continue;
1800 		}
1801 
1802 		prop = sigprop(sig);
1803 
1804 		/*
1805 		 * Decide whether the signal should be returned.
1806 		 * Return the signal's number, or fall through
1807 		 * to clear it from the pending mask.
1808 		 */
1809 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
1810 		case (intptr_t)SIG_DFL:
1811 			/*
1812 			 * Don't take default actions on system processes.
1813 			 */
1814 			if (p->p_pid <= 1) {
1815 #ifdef DIAGNOSTIC
1816 				/*
1817 				 * Are you sure you want to ignore SIGSEGV
1818 				 * in init? XXX
1819 				 */
1820 				kprintf("Process (pid %lu) got signal %d\n",
1821 					(u_long)p->p_pid, sig);
1822 #endif
1823 				break;		/* == ignore */
1824 			}
1825 
1826 			/*
1827 			 * Handle the in-kernel checkpoint action
1828 			 */
1829 			if (prop & SA_CKPT) {
1830 				checkpoint_signal_handler(lp);
1831 				break;
1832 			}
1833 
1834 			/*
1835 			 * If there is a pending stop signal to process
1836 			 * with default action, stop here,
1837 			 * then clear the signal.  However,
1838 			 * if process is member of an orphaned
1839 			 * process group, ignore tty stop signals.
1840 			 */
1841 			if (prop & SA_STOP) {
1842 				if (p->p_flag & P_TRACED ||
1843 		    		    (p->p_pgrp->pg_jobc == 0 &&
1844 				    prop & SA_TTYSTOP))
1845 					break;	/* == ignore */
1846 				p->p_xstat = sig;
1847 				proc_stop(p);
1848 				tstop();
1849 				break;
1850 			} else if (prop & SA_IGNORE) {
1851 				/*
1852 				 * Except for SIGCONT, shouldn't get here.
1853 				 * Default action is to ignore; drop it.
1854 				 */
1855 				break;		/* == ignore */
1856 			} else {
1857 				lwkt_reltoken(&p->p_token);
1858 				return (sig);
1859 			}
1860 
1861 			/*NOTREACHED*/
1862 
1863 		case (intptr_t)SIG_IGN:
1864 			/*
1865 			 * Masking above should prevent us ever trying
1866 			 * to take action on an ignored signal other
1867 			 * than SIGCONT, unless process is traced.
1868 			 */
1869 			if ((prop & SA_CONT) == 0 &&
1870 			    (p->p_flag & P_TRACED) == 0)
1871 				kprintf("issignal\n");
1872 			break;		/* == ignore */
1873 
1874 		default:
1875 			/*
1876 			 * This signal has an action, let
1877 			 * postsig() process it.
1878 			 */
1879 			lwkt_reltoken(&p->p_token);
1880 			return (sig);
1881 		}
1882 		lwp_delsig(lp, sig);		/* take the signal! */
1883 	}
1884 	/* NOTREACHED */
1885 }
1886 
1887 /*
1888  * Take the action for the specified signal
1889  * from the current set of pending signals.
1890  */
1891 void
1892 postsig(int sig)
1893 {
1894 	struct lwp *lp = curthread->td_lwp;
1895 	struct proc *p = lp->lwp_proc;
1896 	struct sigacts *ps = p->p_sigacts;
1897 	sig_t action;
1898 	sigset_t returnmask;
1899 	int code;
1900 
1901 	KASSERT(sig != 0, ("postsig"));
1902 
1903 	KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1904 
1905 	/*
1906 	 * If we are a virtual kernel running an emulated user process
1907 	 * context, switch back to the virtual kernel context before
1908 	 * trying to post the signal.
1909 	 */
1910 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1911 		struct trapframe *tf = lp->lwp_md.md_regs;
1912 		tf->tf_trapno = 0;
1913 		vkernel_trap(lp, tf);
1914 	}
1915 
1916 	lwp_delsig(lp, sig);
1917 	action = ps->ps_sigact[_SIG_IDX(sig)];
1918 #ifdef KTRACE
1919 	if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
1920 		ktrpsig(lp, sig, action, lp->lwp_flag & LWP_OLDMASK ?
1921 			&lp->lwp_oldsigmask : &lp->lwp_sigmask, 0);
1922 #endif
1923 	STOPEVENT(p, S_SIG, sig);
1924 
1925 	if (action == SIG_DFL) {
1926 		/*
1927 		 * Default action, where the default is to kill
1928 		 * the process.  (Other cases were ignored above.)
1929 		 */
1930 		sigexit(lp, sig);
1931 		/* NOTREACHED */
1932 	} else {
1933 		/*
1934 		 * If we get here, the signal must be caught.
1935 		 */
1936 		KASSERT(action != SIG_IGN && !SIGISMEMBER(lp->lwp_sigmask, sig),
1937 		    ("postsig action"));
1938 
1939 		crit_enter();
1940 
1941 		/*
1942 		 * Reset the signal handler if asked to
1943 		 */
1944 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
1945 			/*
1946 			 * See kern_sigaction() for origin of this code.
1947 			 */
1948 			SIGDELSET(p->p_sigcatch, sig);
1949 			if (sig != SIGCONT &&
1950 			    sigprop(sig) & SA_IGNORE)
1951 				SIGADDSET(p->p_sigignore, sig);
1952 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1953 		}
1954 
1955 		/*
1956 		 * Handle the mailbox case.  Copyout to the appropriate
1957 		 * location but do not generate a signal frame.  The system
1958 		 * call simply returns EINTR and the user is responsible for
1959 		 * polling the mailbox.
1960 		 */
1961 		if (SIGISMEMBER(ps->ps_sigmailbox, sig)) {
1962 			int sig_copy = sig;
1963 			copyout(&sig_copy, (void *)action, sizeof(int));
1964 			curproc->p_flag |= P_MAILBOX;
1965 			crit_exit();
1966 			goto done;
1967 		}
1968 
1969 		/*
1970 		 * Set the signal mask and calculate the mask to restore
1971 		 * when the signal function returns.
1972 		 *
1973 		 * Special case: user has done a sigsuspend.  Here the
1974 		 * current mask is not of interest, but rather the
1975 		 * mask from before the sigsuspend is what we want
1976 		 * restored after the signal processing is completed.
1977 		 */
1978 		if (lp->lwp_flag & LWP_OLDMASK) {
1979 			returnmask = lp->lwp_oldsigmask;
1980 			lp->lwp_flag &= ~LWP_OLDMASK;
1981 		} else {
1982 			returnmask = lp->lwp_sigmask;
1983 		}
1984 
1985 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
1986 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
1987 			SIGADDSET(lp->lwp_sigmask, sig);
1988 
1989 		crit_exit();
1990 		lp->lwp_ru.ru_nsignals++;
1991 		if (lp->lwp_sig != sig) {
1992 			code = 0;
1993 		} else {
1994 			code = lp->lwp_code;
1995 			lp->lwp_code = 0;
1996 			lp->lwp_sig = 0;
1997 		}
1998 		(*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
1999 	}
2000 done:
2001 	;
2002 }
2003 
2004 /*
2005  * Kill the current process for stated reason.
2006  */
2007 void
2008 killproc(struct proc *p, char *why)
2009 {
2010 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n",
2011 		p->p_pid, p->p_comm,
2012 		p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2013 	ksignal(p, SIGKILL);
2014 }
2015 
2016 /*
2017  * Force the current process to exit with the specified signal, dumping core
2018  * if appropriate.  We bypass the normal tests for masked and caught signals,
2019  * allowing unrecoverable failures to terminate the process without changing
2020  * signal state.  Mark the accounting record with the signal termination.
2021  * If dumping core, save the signal number for the debugger.  Calls exit and
2022  * does not return.
2023  *
2024  * This routine does not return.
2025  */
2026 void
2027 sigexit(struct lwp *lp, int sig)
2028 {
2029 	struct proc *p = lp->lwp_proc;
2030 
2031 	p->p_acflag |= AXSIG;
2032 	if (sigprop(sig) & SA_CORE) {
2033 		lp->lwp_sig = sig;
2034 		/*
2035 		 * Log signals which would cause core dumps
2036 		 * (Log as LOG_INFO to appease those who don't want
2037 		 * these messages.)
2038 		 * XXX : Todo, as well as euid, write out ruid too
2039 		 */
2040 		if (coredump(lp, sig) == 0)
2041 			sig |= WCOREFLAG;
2042 		if (kern_logsigexit)
2043 			log(LOG_INFO,
2044 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2045 			    p->p_pid, p->p_comm,
2046 			    p->p_ucred ? p->p_ucred->cr_uid : -1,
2047 			    sig &~ WCOREFLAG,
2048 			    sig & WCOREFLAG ? " (core dumped)" : "");
2049 	}
2050 	exit1(W_EXITCODE(0, sig));
2051 	/* NOTREACHED */
2052 }
2053 
2054 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
2055 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
2056 	      sizeof(corefilename), "process corefile name format string");
2057 
2058 /*
2059  * expand_name(name, uid, pid)
2060  * Expand the name described in corefilename, using name, uid, and pid.
2061  * corefilename is a kprintf-like string, with three format specifiers:
2062  *	%N	name of process ("name")
2063  *	%P	process id (pid)
2064  *	%U	user id (uid)
2065  * For example, "%N.core" is the default; they can be disabled completely
2066  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
2067  * This is controlled by the sysctl variable kern.corefile (see above).
2068  */
2069 
2070 static char *
2071 expand_name(const char *name, uid_t uid, pid_t pid)
2072 {
2073 	char *temp;
2074 	char buf[11];		/* Buffer for pid/uid -- max 4B */
2075 	int i, n;
2076 	char *format = corefilename;
2077 	size_t namelen;
2078 
2079 	temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
2080 	if (temp == NULL)
2081 		return NULL;
2082 	namelen = strlen(name);
2083 	for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
2084 		int l;
2085 		switch (format[i]) {
2086 		case '%':	/* Format character */
2087 			i++;
2088 			switch (format[i]) {
2089 			case '%':
2090 				temp[n++] = '%';
2091 				break;
2092 			case 'N':	/* process name */
2093 				if ((n + namelen) > MAXPATHLEN) {
2094 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2095 					    pid, name, uid, temp, name);
2096 					kfree(temp, M_TEMP);
2097 					return NULL;
2098 				}
2099 				memcpy(temp+n, name, namelen);
2100 				n += namelen;
2101 				break;
2102 			case 'P':	/* process id */
2103 				l = ksprintf(buf, "%u", pid);
2104 				if ((n + l) > MAXPATHLEN) {
2105 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2106 					    pid, name, uid, temp, name);
2107 					kfree(temp, M_TEMP);
2108 					return NULL;
2109 				}
2110 				memcpy(temp+n, buf, l);
2111 				n += l;
2112 				break;
2113 			case 'U':	/* user id */
2114 				l = ksprintf(buf, "%u", uid);
2115 				if ((n + l) > MAXPATHLEN) {
2116 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2117 					    pid, name, uid, temp, name);
2118 					kfree(temp, M_TEMP);
2119 					return NULL;
2120 				}
2121 				memcpy(temp+n, buf, l);
2122 				n += l;
2123 				break;
2124 			default:
2125 			  	log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
2126 			}
2127 			break;
2128 		default:
2129 			temp[n++] = format[i];
2130 		}
2131 	}
2132 	temp[n] = '\0';
2133 	return temp;
2134 }
2135 
2136 /*
2137  * Dump a process' core.  The main routine does some
2138  * policy checking, and creates the name of the coredump;
2139  * then it passes on a vnode and a size limit to the process-specific
2140  * coredump routine if there is one; if there _is not_ one, it returns
2141  * ENOSYS; otherwise it returns the error from the process-specific routine.
2142  *
2143  * The parameter `lp' is the lwp which triggered the coredump.
2144  */
2145 
2146 static int
2147 coredump(struct lwp *lp, int sig)
2148 {
2149 	struct proc *p = lp->lwp_proc;
2150 	struct vnode *vp;
2151 	struct ucred *cred = p->p_ucred;
2152 	struct flock lf;
2153 	struct nlookupdata nd;
2154 	struct vattr vattr;
2155 	int error, error1;
2156 	char *name;			/* name of corefile */
2157 	off_t limit;
2158 
2159 	STOPEVENT(p, S_CORE, 0);
2160 
2161 	if (((sugid_coredump == 0) && p->p_flag & P_SUGID) || do_coredump == 0)
2162 		return (EFAULT);
2163 
2164 	/*
2165 	 * Note that the bulk of limit checking is done after
2166 	 * the corefile is created.  The exception is if the limit
2167 	 * for corefiles is 0, in which case we don't bother
2168 	 * creating the corefile at all.  This layout means that
2169 	 * a corefile is truncated instead of not being created,
2170 	 * if it is larger than the limit.
2171 	 */
2172 	limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
2173 	if (limit == 0)
2174 		return EFBIG;
2175 
2176 	name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
2177 	if (name == NULL)
2178 		return (EINVAL);
2179 	error = nlookup_init(&nd, name, UIO_SYSSPACE, NLC_LOCKVP);
2180 	if (error == 0)
2181 		error = vn_open(&nd, NULL, O_CREAT | FWRITE | O_NOFOLLOW, S_IRUSR | S_IWUSR);
2182 	kfree(name, M_TEMP);
2183 	if (error) {
2184 		nlookup_done(&nd);
2185 		return (error);
2186 	}
2187 	vp = nd.nl_open_vp;
2188 	nd.nl_open_vp = NULL;
2189 	nlookup_done(&nd);
2190 
2191 	vn_unlock(vp);
2192 	lf.l_whence = SEEK_SET;
2193 	lf.l_start = 0;
2194 	lf.l_len = 0;
2195 	lf.l_type = F_WRLCK;
2196 	error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, 0);
2197 	if (error)
2198 		goto out2;
2199 
2200 	/* Don't dump to non-regular files or files with links. */
2201 	if (vp->v_type != VREG ||
2202 	    VOP_GETATTR(vp, &vattr) || vattr.va_nlink != 1) {
2203 		error = EFAULT;
2204 		goto out1;
2205 	}
2206 
2207 	/* Don't dump to files current user does not own */
2208 	if (vattr.va_uid != p->p_ucred->cr_uid) {
2209 		error = EFAULT;
2210 		goto out1;
2211 	}
2212 
2213 	VATTR_NULL(&vattr);
2214 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2215 	vattr.va_size = 0;
2216 	VOP_SETATTR(vp, &vattr, cred);
2217 	p->p_acflag |= ACORE;
2218 	vn_unlock(vp);
2219 
2220 	error = p->p_sysent->sv_coredump ?
2221 		  p->p_sysent->sv_coredump(lp, sig, vp, limit) : ENOSYS;
2222 
2223 out1:
2224 	lf.l_type = F_UNLCK;
2225 	VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, 0);
2226 out2:
2227 	error1 = vn_close(vp, FWRITE);
2228 	if (error == 0)
2229 		error = error1;
2230 	return (error);
2231 }
2232 
2233 /*
2234  * Nonexistent system call-- signal process (may want to handle it).
2235  * Flag error in case process won't see signal immediately (blocked or ignored).
2236  *
2237  * MPALMOSTSAFE
2238  */
2239 /* ARGSUSED */
2240 int
2241 sys_nosys(struct nosys_args *args)
2242 {
2243 	lwpsignal(curproc, curthread->td_lwp, SIGSYS);
2244 	return (EINVAL);
2245 }
2246 
2247 /*
2248  * Send a SIGIO or SIGURG signal to a process or process group using
2249  * stored credentials rather than those of the current process.
2250  */
2251 void
2252 pgsigio(struct sigio *sigio, int sig, int checkctty)
2253 {
2254 	if (sigio == NULL)
2255 		return;
2256 
2257 	if (sigio->sio_pgid > 0) {
2258 		if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
2259 		             sigio->sio_proc))
2260 			ksignal(sigio->sio_proc, sig);
2261 	} else if (sigio->sio_pgid < 0) {
2262 		struct proc *p;
2263 
2264 		lockmgr(&sigio->sio_pgrp->pg_lock, LK_EXCLUSIVE);
2265 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
2266 			if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
2267 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
2268 				ksignal(p, sig);
2269 		}
2270 		lockmgr(&sigio->sio_pgrp->pg_lock, LK_RELEASE);
2271 	}
2272 }
2273 
2274 static int
2275 filt_sigattach(struct knote *kn)
2276 {
2277 	struct proc *p = curproc;
2278 
2279 	kn->kn_ptr.p_proc = p;
2280 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
2281 
2282 	/* XXX lock the proc here while adding to the list? */
2283 	knote_insert(&p->p_klist, kn);
2284 
2285 	return (0);
2286 }
2287 
2288 static void
2289 filt_sigdetach(struct knote *kn)
2290 {
2291 	struct proc *p = kn->kn_ptr.p_proc;
2292 
2293 	knote_remove(&p->p_klist, kn);
2294 }
2295 
2296 /*
2297  * signal knotes are shared with proc knotes, so we apply a mask to
2298  * the hint in order to differentiate them from process hints.  This
2299  * could be avoided by using a signal-specific knote list, but probably
2300  * isn't worth the trouble.
2301  */
2302 static int
2303 filt_signal(struct knote *kn, long hint)
2304 {
2305 	if (hint & NOTE_SIGNAL) {
2306 		hint &= ~NOTE_SIGNAL;
2307 
2308 		if (kn->kn_id == hint)
2309 			kn->kn_data++;
2310 	}
2311 	return (kn->kn_data != 0);
2312 }
2313