xref: /dragonfly/sys/kern/kern_sig.c (revision 6693db17)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_sig.c	8.7 (Berkeley) 4/18/94
39  * $FreeBSD: src/sys/kern/kern_sig.c,v 1.72.2.17 2003/05/16 16:34:34 obrien Exp $
40  * $DragonFly: src/sys/kern/kern_sig.c,v 1.90 2008/06/09 04:33:08 dillon Exp $
41  */
42 
43 #include "opt_ktrace.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
48 #include <sys/sysproto.h>
49 #include <sys/signalvar.h>
50 #include <sys/resourcevar.h>
51 #include <sys/vnode.h>
52 #include <sys/event.h>
53 #include <sys/proc.h>
54 #include <sys/nlookup.h>
55 #include <sys/pioctl.h>
56 #include <sys/systm.h>
57 #include <sys/acct.h>
58 #include <sys/fcntl.h>
59 #include <sys/lock.h>
60 #include <sys/wait.h>
61 #include <sys/ktrace.h>
62 #include <sys/syslog.h>
63 #include <sys/stat.h>
64 #include <sys/sysent.h>
65 #include <sys/sysctl.h>
66 #include <sys/malloc.h>
67 #include <sys/interrupt.h>
68 #include <sys/unistd.h>
69 #include <sys/kern_syscall.h>
70 #include <sys/vkernel.h>
71 
72 #include <sys/signal2.h>
73 #include <sys/thread2.h>
74 #include <sys/mplock2.h>
75 
76 #include <machine/cpu.h>
77 #include <machine/smp.h>
78 
79 static int	coredump(struct lwp *, int);
80 static char	*expand_name(const char *, uid_t, pid_t);
81 static int	dokillpg(int sig, int pgid, int all);
82 static int	sig_ffs(sigset_t *set);
83 static int	sigprop(int sig);
84 #ifdef SMP
85 static void	signotify_remote(void *arg);
86 #endif
87 static int	kern_sigtimedwait(sigset_t set, siginfo_t *info,
88 		    struct timespec *timeout);
89 
90 static int	filt_sigattach(struct knote *kn);
91 static void	filt_sigdetach(struct knote *kn);
92 static int	filt_signal(struct knote *kn, long hint);
93 
94 struct filterops sig_filtops =
95 	{ 0, filt_sigattach, filt_sigdetach, filt_signal };
96 
97 static int	kern_logsigexit = 1;
98 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
99     &kern_logsigexit, 0,
100     "Log processes quitting on abnormal signals to syslog(3)");
101 
102 /*
103  * Can process p, with pcred pc, send the signal sig to process q?
104  */
105 #define CANSIGNAL(q, sig) \
106 	(!p_trespass(curproc->p_ucred, (q)->p_ucred) || \
107 	((sig) == SIGCONT && (q)->p_session == curproc->p_session))
108 
109 /*
110  * Policy -- Can real uid ruid with ucred uc send a signal to process q?
111  */
112 #define CANSIGIO(ruid, uc, q) \
113 	((uc)->cr_uid == 0 || \
114 	    (ruid) == (q)->p_ucred->cr_ruid || \
115 	    (uc)->cr_uid == (q)->p_ucred->cr_ruid || \
116 	    (ruid) == (q)->p_ucred->cr_uid || \
117 	    (uc)->cr_uid == (q)->p_ucred->cr_uid)
118 
119 int sugid_coredump;
120 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RW,
121 	&sugid_coredump, 0, "Enable coredumping set user/group ID processes");
122 
123 static int	do_coredump = 1;
124 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
125 	&do_coredump, 0, "Enable/Disable coredumps");
126 
127 /*
128  * Signal properties and actions.
129  * The array below categorizes the signals and their default actions
130  * according to the following properties:
131  */
132 #define	SA_KILL		0x01		/* terminates process by default */
133 #define	SA_CORE		0x02		/* ditto and coredumps */
134 #define	SA_STOP		0x04		/* suspend process */
135 #define	SA_TTYSTOP	0x08		/* ditto, from tty */
136 #define	SA_IGNORE	0x10		/* ignore by default */
137 #define	SA_CONT		0x20		/* continue if suspended */
138 #define	SA_CANTMASK	0x40		/* non-maskable, catchable */
139 #define SA_CKPT         0x80            /* checkpoint process */
140 
141 
142 static int sigproptbl[NSIG] = {
143         SA_KILL,                /* SIGHUP */
144         SA_KILL,                /* SIGINT */
145         SA_KILL|SA_CORE,        /* SIGQUIT */
146         SA_KILL|SA_CORE,        /* SIGILL */
147         SA_KILL|SA_CORE,        /* SIGTRAP */
148         SA_KILL|SA_CORE,        /* SIGABRT */
149         SA_KILL|SA_CORE,        /* SIGEMT */
150         SA_KILL|SA_CORE,        /* SIGFPE */
151         SA_KILL,                /* SIGKILL */
152         SA_KILL|SA_CORE,        /* SIGBUS */
153         SA_KILL|SA_CORE,        /* SIGSEGV */
154         SA_KILL|SA_CORE,        /* SIGSYS */
155         SA_KILL,                /* SIGPIPE */
156         SA_KILL,                /* SIGALRM */
157         SA_KILL,                /* SIGTERM */
158         SA_IGNORE,              /* SIGURG */
159         SA_STOP,                /* SIGSTOP */
160         SA_STOP|SA_TTYSTOP,     /* SIGTSTP */
161         SA_IGNORE|SA_CONT,      /* SIGCONT */
162         SA_IGNORE,              /* SIGCHLD */
163         SA_STOP|SA_TTYSTOP,     /* SIGTTIN */
164         SA_STOP|SA_TTYSTOP,     /* SIGTTOU */
165         SA_IGNORE,              /* SIGIO */
166         SA_KILL,                /* SIGXCPU */
167         SA_KILL,                /* SIGXFSZ */
168         SA_KILL,                /* SIGVTALRM */
169         SA_KILL,                /* SIGPROF */
170         SA_IGNORE,              /* SIGWINCH  */
171         SA_IGNORE,              /* SIGINFO */
172         SA_KILL,                /* SIGUSR1 */
173         SA_KILL,                /* SIGUSR2 */
174 	SA_IGNORE,              /* SIGTHR */
175 	SA_CKPT,                /* SIGCKPT */
176 	SA_KILL|SA_CKPT,        /* SIGCKPTEXIT */
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 	SA_IGNORE,
207 
208 };
209 
210 static __inline int
211 sigprop(int sig)
212 {
213 
214 	if (sig > 0 && sig < NSIG)
215 		return (sigproptbl[_SIG_IDX(sig)]);
216 	return (0);
217 }
218 
219 static __inline int
220 sig_ffs(sigset_t *set)
221 {
222 	int i;
223 
224 	for (i = 0; i < _SIG_WORDS; i++)
225 		if (set->__bits[i])
226 			return (ffs(set->__bits[i]) + (i * 32));
227 	return (0);
228 }
229 
230 int
231 kern_sigaction(int sig, struct sigaction *act, struct sigaction *oact)
232 {
233 	struct thread *td = curthread;
234 	struct proc *p = td->td_proc;
235 	struct lwp *lp;
236 	struct sigacts *ps = p->p_sigacts;
237 
238 	if (sig <= 0 || sig > _SIG_MAXSIG)
239 		return (EINVAL);
240 
241 	if (oact) {
242 		oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
243 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
244 		oact->sa_flags = 0;
245 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
246 			oact->sa_flags |= SA_ONSTACK;
247 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
248 			oact->sa_flags |= SA_RESTART;
249 		if (SIGISMEMBER(ps->ps_sigreset, sig))
250 			oact->sa_flags |= SA_RESETHAND;
251 		if (SIGISMEMBER(ps->ps_signodefer, sig))
252 			oact->sa_flags |= SA_NODEFER;
253 		if (SIGISMEMBER(ps->ps_siginfo, sig))
254 			oact->sa_flags |= SA_SIGINFO;
255 		if (SIGISMEMBER(ps->ps_sigmailbox, sig))
256 			oact->sa_flags |= SA_MAILBOX;
257 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDSTOP)
258 			oact->sa_flags |= SA_NOCLDSTOP;
259 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDWAIT)
260 			oact->sa_flags |= SA_NOCLDWAIT;
261 	}
262 	if (act) {
263 		/*
264 		 * Check for invalid requests.  KILL and STOP cannot be
265 		 * caught.
266 		 */
267 		if (sig == SIGKILL || sig == SIGSTOP) {
268 			if (act->sa_handler != SIG_DFL)
269 				return (EINVAL);
270 #if 0
271 			/* (not needed, SIG_DFL forces action to occur) */
272 			if (act->sa_flags & SA_MAILBOX)
273 				return (EINVAL);
274 #endif
275 		}
276 
277 		/*
278 		 * Change setting atomically.
279 		 */
280 		crit_enter();
281 
282 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
283 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
284 		if (act->sa_flags & SA_SIGINFO) {
285 			ps->ps_sigact[_SIG_IDX(sig)] =
286 			    (__sighandler_t *)act->sa_sigaction;
287 			SIGADDSET(ps->ps_siginfo, sig);
288 		} else {
289 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
290 			SIGDELSET(ps->ps_siginfo, sig);
291 		}
292 		if (!(act->sa_flags & SA_RESTART))
293 			SIGADDSET(ps->ps_sigintr, sig);
294 		else
295 			SIGDELSET(ps->ps_sigintr, sig);
296 		if (act->sa_flags & SA_ONSTACK)
297 			SIGADDSET(ps->ps_sigonstack, sig);
298 		else
299 			SIGDELSET(ps->ps_sigonstack, sig);
300 		if (act->sa_flags & SA_RESETHAND)
301 			SIGADDSET(ps->ps_sigreset, sig);
302 		else
303 			SIGDELSET(ps->ps_sigreset, sig);
304 		if (act->sa_flags & SA_NODEFER)
305 			SIGADDSET(ps->ps_signodefer, sig);
306 		else
307 			SIGDELSET(ps->ps_signodefer, sig);
308 		if (act->sa_flags & SA_MAILBOX)
309 			SIGADDSET(ps->ps_sigmailbox, sig);
310 		else
311 			SIGDELSET(ps->ps_sigmailbox, sig);
312 		if (sig == SIGCHLD) {
313 			if (act->sa_flags & SA_NOCLDSTOP)
314 				p->p_sigacts->ps_flag |= PS_NOCLDSTOP;
315 			else
316 				p->p_sigacts->ps_flag &= ~PS_NOCLDSTOP;
317 			if (act->sa_flags & SA_NOCLDWAIT) {
318 				/*
319 				 * Paranoia: since SA_NOCLDWAIT is implemented
320 				 * by reparenting the dying child to PID 1 (and
321 				 * trust it to reap the zombie), PID 1 itself
322 				 * is forbidden to set SA_NOCLDWAIT.
323 				 */
324 				if (p->p_pid == 1)
325 					p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
326 				else
327 					p->p_sigacts->ps_flag |= PS_NOCLDWAIT;
328 			} else {
329 				p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
330 			}
331 		}
332 		/*
333 		 * Set bit in p_sigignore for signals that are set to SIG_IGN,
334 		 * and for signals set to SIG_DFL where the default is to
335 		 * ignore. However, don't put SIGCONT in p_sigignore, as we
336 		 * have to restart the process.
337 		 */
338 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
339 		    (sigprop(sig) & SA_IGNORE &&
340 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
341 			/* never to be seen again */
342 			SIGDELSET(p->p_siglist, sig);
343 			/*
344 			 * Remove the signal also from the thread lists.
345 			 */
346 			FOREACH_LWP_IN_PROC(lp, p) {
347 				SIGDELSET(lp->lwp_siglist, sig);
348 			}
349 			if (sig != SIGCONT)
350 				/* easier in ksignal */
351 				SIGADDSET(p->p_sigignore, sig);
352 			SIGDELSET(p->p_sigcatch, sig);
353 		} else {
354 			SIGDELSET(p->p_sigignore, sig);
355 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
356 				SIGDELSET(p->p_sigcatch, sig);
357 			else
358 				SIGADDSET(p->p_sigcatch, sig);
359 		}
360 
361 		crit_exit();
362 	}
363 	return (0);
364 }
365 
366 /*
367  * MPALMOSTSAFE
368  */
369 int
370 sys_sigaction(struct sigaction_args *uap)
371 {
372 	struct sigaction act, oact;
373 	struct sigaction *actp, *oactp;
374 	int error;
375 
376 	actp = (uap->act != NULL) ? &act : NULL;
377 	oactp = (uap->oact != NULL) ? &oact : NULL;
378 	if (actp) {
379 		error = copyin(uap->act, actp, sizeof(act));
380 		if (error)
381 			return (error);
382 	}
383 	get_mplock();
384 	error = kern_sigaction(uap->sig, actp, oactp);
385 	rel_mplock();
386 	if (oactp && !error) {
387 		error = copyout(oactp, uap->oact, sizeof(oact));
388 	}
389 	return (error);
390 }
391 
392 /*
393  * Initialize signal state for process 0;
394  * set to ignore signals that are ignored by default.
395  */
396 void
397 siginit(struct proc *p)
398 {
399 	int i;
400 
401 	for (i = 1; i <= NSIG; i++)
402 		if (sigprop(i) & SA_IGNORE && i != SIGCONT)
403 			SIGADDSET(p->p_sigignore, i);
404 }
405 
406 /*
407  * Reset signals for an exec of the specified process.
408  */
409 void
410 execsigs(struct proc *p)
411 {
412 	struct sigacts *ps = p->p_sigacts;
413 	struct lwp *lp;
414 	int sig;
415 
416 	lp = ONLY_LWP_IN_PROC(p);
417 
418 	/*
419 	 * Reset caught signals.  Held signals remain held
420 	 * through p_sigmask (unless they were caught,
421 	 * and are now ignored by default).
422 	 */
423 	while (SIGNOTEMPTY(p->p_sigcatch)) {
424 		sig = sig_ffs(&p->p_sigcatch);
425 		SIGDELSET(p->p_sigcatch, sig);
426 		if (sigprop(sig) & SA_IGNORE) {
427 			if (sig != SIGCONT)
428 				SIGADDSET(p->p_sigignore, sig);
429 			SIGDELSET(p->p_siglist, sig);
430 			SIGDELSET(lp->lwp_siglist, sig);
431 		}
432 		ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
433 	}
434 
435 	/*
436 	 * Reset stack state to the user stack.
437 	 * Clear set of signals caught on the signal stack.
438 	 */
439 	lp->lwp_sigstk.ss_flags = SS_DISABLE;
440 	lp->lwp_sigstk.ss_size = 0;
441 	lp->lwp_sigstk.ss_sp = 0;
442 	lp->lwp_flag &= ~LWP_ALTSTACK;
443 	/*
444 	 * Reset no zombies if child dies flag as Solaris does.
445 	 */
446 	p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
447 }
448 
449 /*
450  * kern_sigprocmask() - MP SAFE ONLY IF p == curproc
451  *
452  *	Manipulate signal mask.  This routine is MP SAFE *ONLY* if
453  *	p == curproc.
454  */
455 int
456 kern_sigprocmask(int how, sigset_t *set, sigset_t *oset)
457 {
458 	struct thread *td = curthread;
459 	struct lwp *lp = td->td_lwp;
460 	int error;
461 
462 	if (oset != NULL)
463 		*oset = lp->lwp_sigmask;
464 
465 	error = 0;
466 	if (set != NULL) {
467 		switch (how) {
468 		case SIG_BLOCK:
469 			SIG_CANTMASK(*set);
470 			SIGSETOR(lp->lwp_sigmask, *set);
471 			break;
472 		case SIG_UNBLOCK:
473 			SIGSETNAND(lp->lwp_sigmask, *set);
474 			break;
475 		case SIG_SETMASK:
476 			SIG_CANTMASK(*set);
477 			lp->lwp_sigmask = *set;
478 			break;
479 		default:
480 			error = EINVAL;
481 			break;
482 		}
483 	}
484 	return (error);
485 }
486 
487 /*
488  * sigprocmask()
489  *
490  * MPSAFE
491  */
492 int
493 sys_sigprocmask(struct sigprocmask_args *uap)
494 {
495 	sigset_t set, oset;
496 	sigset_t *setp, *osetp;
497 	int error;
498 
499 	setp = (uap->set != NULL) ? &set : NULL;
500 	osetp = (uap->oset != NULL) ? &oset : NULL;
501 	if (setp) {
502 		error = copyin(uap->set, setp, sizeof(set));
503 		if (error)
504 			return (error);
505 	}
506 	error = kern_sigprocmask(uap->how, setp, osetp);
507 	if (osetp && !error) {
508 		error = copyout(osetp, uap->oset, sizeof(oset));
509 	}
510 	return (error);
511 }
512 
513 /*
514  * MPSAFE
515  */
516 int
517 kern_sigpending(struct __sigset *set)
518 {
519 	struct lwp *lp = curthread->td_lwp;
520 
521 	*set = lwp_sigpend(lp);
522 
523 	return (0);
524 }
525 
526 /*
527  * MPSAFE
528  */
529 int
530 sys_sigpending(struct sigpending_args *uap)
531 {
532 	sigset_t set;
533 	int error;
534 
535 	error = kern_sigpending(&set);
536 
537 	if (error == 0)
538 		error = copyout(&set, uap->set, sizeof(set));
539 	return (error);
540 }
541 
542 /*
543  * Suspend process until signal, providing mask to be set
544  * in the meantime.
545  *
546  * MPSAFE
547  */
548 int
549 kern_sigsuspend(struct __sigset *set)
550 {
551 	struct thread *td = curthread;
552 	struct lwp *lp = td->td_lwp;
553 	struct proc *p = td->td_proc;
554 	struct sigacts *ps = p->p_sigacts;
555 
556 	/*
557 	 * When returning from sigsuspend, we want
558 	 * the old mask to be restored after the
559 	 * signal handler has finished.  Thus, we
560 	 * save it here and mark the sigacts structure
561 	 * to indicate this.
562 	 */
563 	lp->lwp_oldsigmask = lp->lwp_sigmask;
564 	lp->lwp_flag |= LWP_OLDMASK;
565 
566 	SIG_CANTMASK(*set);
567 	lp->lwp_sigmask = *set;
568 	while (tsleep(ps, PCATCH, "pause", 0) == 0)
569 		/* void */;
570 	/* always return EINTR rather than ERESTART... */
571 	return (EINTR);
572 }
573 
574 /*
575  * Note nonstandard calling convention: libc stub passes mask, not
576  * pointer, to save a copyin.
577  *
578  * MPSAFE
579  */
580 int
581 sys_sigsuspend(struct sigsuspend_args *uap)
582 {
583 	sigset_t mask;
584 	int error;
585 
586 	error = copyin(uap->sigmask, &mask, sizeof(mask));
587 	if (error)
588 		return (error);
589 
590 	error = kern_sigsuspend(&mask);
591 
592 	return (error);
593 }
594 
595 /*
596  * MPSAFE
597  */
598 int
599 kern_sigaltstack(struct sigaltstack *ss, struct sigaltstack *oss)
600 {
601 	struct thread *td = curthread;
602 	struct lwp *lp = td->td_lwp;
603 	struct proc *p = td->td_proc;
604 
605 	if ((lp->lwp_flag & LWP_ALTSTACK) == 0)
606 		lp->lwp_sigstk.ss_flags |= SS_DISABLE;
607 
608 	if (oss)
609 		*oss = lp->lwp_sigstk;
610 
611 	if (ss) {
612 		if (ss->ss_flags & SS_DISABLE) {
613 			if (lp->lwp_sigstk.ss_flags & SS_ONSTACK)
614 				return (EINVAL);
615 			lp->lwp_flag &= ~LWP_ALTSTACK;
616 			lp->lwp_sigstk.ss_flags = ss->ss_flags;
617 		} else {
618 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
619 				return (ENOMEM);
620 			lp->lwp_flag |= LWP_ALTSTACK;
621 			lp->lwp_sigstk = *ss;
622 		}
623 	}
624 
625 	return (0);
626 }
627 
628 /*
629  * MPSAFE
630  */
631 int
632 sys_sigaltstack(struct sigaltstack_args *uap)
633 {
634 	stack_t ss, oss;
635 	int error;
636 
637 	if (uap->ss) {
638 		error = copyin(uap->ss, &ss, sizeof(ss));
639 		if (error)
640 			return (error);
641 	}
642 
643 	error = kern_sigaltstack(uap->ss ? &ss : NULL,
644 	    uap->oss ? &oss : NULL);
645 
646 	if (error == 0 && uap->oss)
647 		error = copyout(&oss, uap->oss, sizeof(*uap->oss));
648 	return (error);
649 }
650 
651 /*
652  * Common code for kill process group/broadcast kill.
653  * cp is calling process.
654  */
655 struct killpg_info {
656 	int nfound;
657 	int sig;
658 };
659 
660 static int killpg_all_callback(struct proc *p, void *data);
661 
662 static int
663 dokillpg(int sig, int pgid, int all)
664 {
665 	struct killpg_info info;
666 	struct proc *cp = curproc;
667 	struct proc *p;
668 	struct pgrp *pgrp;
669 
670 	info.nfound = 0;
671 	info.sig = sig;
672 
673 	if (all) {
674 		/*
675 		 * broadcast
676 		 */
677 		allproc_scan(killpg_all_callback, &info);
678 	} else {
679 		if (pgid == 0) {
680 			/*
681 			 * zero pgid means send to my process group.
682 			 */
683 			pgrp = cp->p_pgrp;
684 		} else {
685 			pgrp = pgfind(pgid);
686 			if (pgrp == NULL)
687 				return (ESRCH);
688 		}
689 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
690 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
691 			if (p->p_pid <= 1 ||
692 			    p->p_stat == SZOMB ||
693 			    (p->p_flag & P_SYSTEM) ||
694 			    !CANSIGNAL(p, sig)) {
695 				continue;
696 			}
697 			++info.nfound;
698 			if (sig)
699 				ksignal(p, sig);
700 		}
701 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
702 	}
703 	return (info.nfound ? 0 : ESRCH);
704 }
705 
706 static int
707 killpg_all_callback(struct proc *p, void *data)
708 {
709 	struct killpg_info *info = data;
710 
711 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) ||
712 	    p == curproc || !CANSIGNAL(p, info->sig)) {
713 		return (0);
714 	}
715 	++info->nfound;
716 	if (info->sig)
717 		ksignal(p, info->sig);
718 	return(0);
719 }
720 
721 /*
722  * Send a general signal to a process or LWPs within that process.  Note
723  * that new signals cannot be sent if a process is exiting.
724  */
725 int
726 kern_kill(int sig, pid_t pid, lwpid_t tid)
727 {
728 	if ((u_int)sig > _SIG_MAXSIG)
729 		return (EINVAL);
730 	if (pid > 0) {
731 		struct proc *p;
732 		struct lwp *lp = NULL;
733 
734 		/* kill single process */
735 		if ((p = pfind(pid)) == NULL)
736 			return (ESRCH);
737 		if (!CANSIGNAL(p, sig))
738 			return (EPERM);
739 
740 		/*
741 		 * NOP if the process is exiting.  Note that lwpsignal() is
742 		 * called directly with P_WEXIT set to kill individual LWPs
743 		 * during exit, which is allowed.
744 		 */
745 		if (p->p_flag & P_WEXIT)
746 			return (0);
747 		if (tid != -1) {
748 			lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, tid);
749 			if (lp == NULL)
750 				return (ESRCH);
751 		}
752 		if (sig)
753 			lwpsignal(p, lp, sig);
754 		return (0);
755 	}
756 	/*
757 	 * If we come here, pid is a special broadcast pid.
758 	 * This doesn't mix with a tid.
759 	 */
760 	if (tid != -1)
761 		return (EINVAL);
762 	switch (pid) {
763 	case -1:		/* broadcast signal */
764 		return (dokillpg(sig, 0, 1));
765 	case 0:			/* signal own process group */
766 		return (dokillpg(sig, 0, 0));
767 	default:		/* negative explicit process group */
768 		return (dokillpg(sig, -pid, 0));
769 	}
770 	/* NOTREACHED */
771 }
772 
773 /*
774  * MPALMOSTSAFE
775  */
776 int
777 sys_kill(struct kill_args *uap)
778 {
779 	int error;
780 
781 	get_mplock();
782 	error = kern_kill(uap->signum, uap->pid, -1);
783 	rel_mplock();
784 	return (error);
785 }
786 
787 /*
788  * MPALMOSTSAFE
789  */
790 int
791 sys_lwp_kill(struct lwp_kill_args *uap)
792 {
793 	int error;
794 	pid_t pid = uap->pid;
795 
796 	/*
797 	 * A tid is mandatory for lwp_kill(), otherwise
798 	 * you could simply use kill().
799 	 */
800 	if (uap->tid == -1)
801 		return (EINVAL);
802 
803 	/*
804 	 * To save on a getpid() function call for intra-process
805 	 * signals, pid == -1 means current process.
806 	 */
807 	if (pid == -1)
808 		pid = curproc->p_pid;
809 
810 	get_mplock();
811 	error = kern_kill(uap->signum, pid, uap->tid);
812 	rel_mplock();
813 	return (error);
814 }
815 
816 /*
817  * Send a signal to a process group.
818  */
819 void
820 gsignal(int pgid, int sig)
821 {
822 	struct pgrp *pgrp;
823 
824 	if (pgid && (pgrp = pgfind(pgid)))
825 		pgsignal(pgrp, sig, 0);
826 }
827 
828 /*
829  * Send a signal to a process group.  If checktty is 1,
830  * limit to members which have a controlling terminal.
831  *
832  * pg_lock interlocks against a fork that might be in progress, to
833  * ensure that the new child process picks up the signal.
834  */
835 void
836 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
837 {
838 	struct proc *p;
839 
840 	if (pgrp) {
841 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
842 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
843 			if (checkctty == 0 || p->p_flag & P_CONTROLT)
844 				ksignal(p, sig);
845 		}
846 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
847 	}
848 }
849 
850 /*
851  * Send a signal caused by a trap to the current lwp.  If it will be caught
852  * immediately, deliver it with correct code.  Otherwise, post it normally.
853  *
854  * These signals may ONLY be delivered to the specified lwp and may never
855  * be delivered to the process generically.
856  */
857 void
858 trapsignal(struct lwp *lp, int sig, u_long code)
859 {
860 	struct proc *p = lp->lwp_proc;
861 	struct sigacts *ps = p->p_sigacts;
862 
863 	/*
864 	 * If we are a virtual kernel running an emulated user process
865 	 * context, switch back to the virtual kernel context before
866 	 * trying to post the signal.
867 	 */
868 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
869 		struct trapframe *tf = lp->lwp_md.md_regs;
870 		tf->tf_trapno = 0;
871 		vkernel_trap(lp, tf);
872 	}
873 
874 
875 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(p->p_sigcatch, sig) &&
876 	    !SIGISMEMBER(lp->lwp_sigmask, sig)) {
877 		lp->lwp_ru.ru_nsignals++;
878 #ifdef KTRACE
879 		if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
880 			ktrpsig(lp, sig, ps->ps_sigact[_SIG_IDX(sig)],
881 				&lp->lwp_sigmask, code);
882 #endif
883 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], sig,
884 						&lp->lwp_sigmask, code);
885 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
886 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
887 			SIGADDSET(lp->lwp_sigmask, sig);
888 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
889 			/*
890 			 * See kern_sigaction() for origin of this code.
891 			 */
892 			SIGDELSET(p->p_sigcatch, sig);
893 			if (sig != SIGCONT &&
894 			    sigprop(sig) & SA_IGNORE)
895 				SIGADDSET(p->p_sigignore, sig);
896 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
897 		}
898 	} else {
899 		lp->lwp_code = code;	/* XXX for core dump/debugger */
900 		lp->lwp_sig = sig;	/* XXX to verify code */
901 		lwpsignal(p, lp, sig);
902 	}
903 }
904 
905 /*
906  * Find a suitable lwp to deliver the signal to.
907  *
908  * Returns NULL if all lwps hold the signal blocked.
909  */
910 static struct lwp *
911 find_lwp_for_signal(struct proc *p, int sig)
912 {
913 	struct lwp *lp;
914 	struct lwp *run, *sleep, *stop;
915 
916 	/*
917 	 * If the running/preempted thread belongs to the proc to which
918 	 * the signal is being delivered and this thread does not block
919 	 * the signal, then we can avoid a context switch by delivering
920 	 * the signal to this thread, because it will return to userland
921 	 * soon anyways.
922 	 */
923 	lp = lwkt_preempted_proc();
924 	if (lp != NULL && lp->lwp_proc == p && !SIGISMEMBER(lp->lwp_sigmask, sig))
925 		return (lp);
926 
927 	run = sleep = stop = NULL;
928 	FOREACH_LWP_IN_PROC(lp, p) {
929 		/*
930 		 * If the signal is being blocked by the lwp, then this
931 		 * lwp is not eligible for receiving the signal.
932 		 */
933 		if (SIGISMEMBER(lp->lwp_sigmask, sig))
934 			continue;
935 
936 		switch (lp->lwp_stat) {
937 		case LSRUN:
938 			run = lp;
939 			break;
940 
941 		case LSSTOP:
942 			stop = lp;
943 			break;
944 
945 		case LSSLEEP:
946 			if (lp->lwp_flag & LWP_SINTR)
947 				sleep = lp;
948 			break;
949 		}
950 	}
951 
952 	if (run != NULL)
953 		return (run);
954 	else if (sleep != NULL)
955 		return (sleep);
956 	else
957 		return (stop);
958 }
959 
960 /*
961  * Send the signal to the process.  If the signal has an action, the action
962  * is usually performed by the target process rather than the caller; we add
963  * the signal to the set of pending signals for the process.
964  *
965  * Exceptions:
966  *   o When a stop signal is sent to a sleeping process that takes the
967  *     default action, the process is stopped without awakening it.
968  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
969  *     regardless of the signal action (eg, blocked or ignored).
970  *
971  * Other ignored signals are discarded immediately.
972  */
973 void
974 ksignal(struct proc *p, int sig)
975 {
976 	lwpsignal(p, NULL, sig);
977 }
978 
979 /*
980  * The core for ksignal.  lp may be NULL, then a suitable thread
981  * will be chosen.  If not, lp MUST be a member of p.
982  */
983 void
984 lwpsignal(struct proc *p, struct lwp *lp, int sig)
985 {
986 	int prop;
987 	sig_t action;
988 
989 	if (sig > _SIG_MAXSIG || sig <= 0) {
990 		kprintf("lwpsignal: signal %d\n", sig);
991 		panic("lwpsignal signal number");
992 	}
993 
994 	KKASSERT(lp == NULL || lp->lwp_proc == p);
995 
996 	crit_enter();
997 	KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
998 	crit_exit();
999 
1000 	prop = sigprop(sig);
1001 
1002 	/*
1003 	 * If proc is traced, always give parent a chance;
1004 	 * if signal event is tracked by procfs, give *that*
1005 	 * a chance, as well.
1006 	 */
1007 	if ((p->p_flag & P_TRACED) || (p->p_stops & S_SIG)) {
1008 		action = SIG_DFL;
1009 	} else {
1010 		/*
1011 		 * Do not try to deliver signals to an exiting lwp.  Note
1012 		 * that we must still deliver the signal if P_WEXIT is set
1013 		 * in the process flags.
1014 		 */
1015 		if (lp && (lp->lwp_flag & LWP_WEXIT))
1016 			return;
1017 
1018 		/*
1019 		 * If the signal is being ignored, then we forget about
1020 		 * it immediately.  NOTE: We don't set SIGCONT in p_sigignore,
1021 		 * and if it is set to SIG_IGN, action will be SIG_DFL here.
1022 		 */
1023 		if (SIGISMEMBER(p->p_sigignore, sig))
1024 			return;
1025 		if (SIGISMEMBER(p->p_sigcatch, sig))
1026 			action = SIG_CATCH;
1027 		else
1028 			action = SIG_DFL;
1029 	}
1030 
1031 	/*
1032 	 * If continuing, clear any pending STOP signals.
1033 	 */
1034 	if (prop & SA_CONT)
1035 		SIG_STOPSIGMASK(p->p_siglist);
1036 
1037 	if (prop & SA_STOP) {
1038 		/*
1039 		 * If sending a tty stop signal to a member of an orphaned
1040 		 * process group, discard the signal here if the action
1041 		 * is default; don't stop the process below if sleeping,
1042 		 * and don't clear any pending SIGCONT.
1043 		 */
1044 		if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0 &&
1045 		    action == SIG_DFL) {
1046 		        return;
1047 		}
1048 		SIG_CONTSIGMASK(p->p_siglist);
1049 		p->p_flag &= ~P_CONTINUED;
1050 	}
1051 
1052 	crit_enter();
1053 
1054 	if (p->p_stat == SSTOP) {
1055 		/*
1056 		 * Nobody can handle this signal, add it to the lwp or
1057 		 * process pending list
1058 		 */
1059 		if (lp)
1060 			SIGADDSET(lp->lwp_siglist, sig);
1061 		else
1062 			SIGADDSET(p->p_siglist, sig);
1063 
1064 		/*
1065 		 * If the process is stopped and is being traced, then no
1066 		 * further action is necessary.
1067 		 */
1068 		if (p->p_flag & P_TRACED)
1069 			goto out;
1070 
1071 		/*
1072 		 * If the process is stopped and receives a KILL signal,
1073 		 * make the process runnable.
1074 		 */
1075 		if (sig == SIGKILL) {
1076 			proc_unstop(p);
1077 			goto active_process;
1078 		}
1079 
1080 		/*
1081 		 * If the process is stopped and receives a CONT signal,
1082 		 * then try to make the process runnable again.
1083 		 */
1084 		if (prop & SA_CONT) {
1085 			/*
1086 			 * If SIGCONT is default (or ignored), we continue the
1087 			 * process but don't leave the signal in p_siglist, as
1088 			 * it has no further action.  If SIGCONT is held, we
1089 			 * continue the process and leave the signal in
1090 			 * p_siglist.  If the process catches SIGCONT, let it
1091 			 * handle the signal itself.
1092 			 */
1093 			/* XXX what if the signal is being held blocked? */
1094 			p->p_flag |= P_CONTINUED;
1095 			wakeup(p->p_pptr);
1096 			if (action == SIG_DFL)
1097 				SIGDELSET(p->p_siglist, sig);
1098 			proc_unstop(p);
1099 			if (action == SIG_CATCH)
1100 				goto active_process;
1101 			goto out;
1102 		}
1103 
1104 		/*
1105 		 * If the process is stopped and receives another STOP
1106 		 * signal, we do not need to stop it again.  If we did
1107 		 * the shell could get confused.
1108 		 *
1109 		 * However, if the current/preempted lwp is part of the
1110 		 * process receiving the signal, we need to keep it,
1111 		 * so that this lwp can stop in issignal() later, as
1112 		 * we don't want to wait until it reaches userret!
1113 		 */
1114 		if (prop & SA_STOP) {
1115 			if (lwkt_preempted_proc() == NULL ||
1116 			    lwkt_preempted_proc()->lwp_proc != p)
1117 				SIGDELSET(p->p_siglist, sig);
1118 		}
1119 
1120 		/*
1121 		 * Otherwise the process is stopped and it received some
1122 		 * signal, which does not change its stopped state.
1123 		 *
1124 		 * We have to select one thread to set LWP_BREAKTSLEEP,
1125 		 * so that the current signal will break the sleep
1126 		 * as soon as a SA_CONT signal will unstop the process.
1127 		 */
1128 		if (lp == NULL)
1129 			lp = find_lwp_for_signal(p, sig);
1130 		if (lp != NULL &&
1131 		    (lp->lwp_stat == LSSLEEP || lp->lwp_stat == LSSTOP))
1132 			lp->lwp_flag |= LWP_BREAKTSLEEP;
1133 		goto out;
1134 
1135 		/* NOTREACHED */
1136 	}
1137 	/* else not stopped */
1138 active_process:
1139 
1140 	/*
1141 	 * Never deliver a lwp-specific signal to a random lwp.
1142 	 */
1143 	if (lp == NULL) {
1144 		lp = find_lwp_for_signal(p, sig);
1145 		if (lp && SIGISMEMBER(lp->lwp_sigmask, sig))
1146 			lp = NULL;
1147 	}
1148 
1149 	/*
1150 	 * Deliver to the process generically if (1) the signal is being
1151 	 * sent to any thread or (2) we could not find a thread to deliver
1152 	 * it to.
1153 	 */
1154 	if (lp == NULL) {
1155 		SIGADDSET(p->p_siglist, sig);
1156 		goto out;
1157 	}
1158 
1159 	/*
1160 	 * Deliver to a specific LWP whether it masks it or not.  It will
1161 	 * not be dispatched if masked but we must still deliver it.
1162 	 */
1163 	if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
1164 	    (p->p_flag & P_TRACED) == 0) {
1165 		p->p_nice = NZERO;
1166 	}
1167 
1168 	/*
1169 	 * If the process receives a STOP signal which indeed needs to
1170 	 * stop the process, do so.  If the process chose to catch the
1171 	 * signal, it will be treated like any other signal.
1172 	 */
1173 	if ((prop & SA_STOP) && action == SIG_DFL) {
1174 		/*
1175 		 * If a child holding parent blocked, stopping
1176 		 * could cause deadlock.  Take no action at this
1177 		 * time.
1178 		 */
1179 		if (p->p_flag & P_PPWAIT) {
1180 			SIGADDSET(p->p_siglist, sig);
1181 			goto out;
1182 		}
1183 
1184 		/*
1185 		 * Do not actually try to manipulate the process, but simply
1186 		 * stop it.  Lwps will stop as soon as they safely can.
1187 		 */
1188 		p->p_xstat = sig;
1189 		proc_stop(p);
1190 		goto out;
1191 	}
1192 
1193 	/*
1194 	 * If it is a CONT signal with default action, just ignore it.
1195 	 */
1196 	if ((prop & SA_CONT) && action == SIG_DFL)
1197 		goto out;
1198 
1199 	/*
1200 	 * Mark signal pending at this specific thread.
1201 	 */
1202 	SIGADDSET(lp->lwp_siglist, sig);
1203 
1204 	lwp_signotify(lp);
1205 
1206 out:
1207 	crit_exit();
1208 }
1209 
1210 void
1211 lwp_signotify(struct lwp *lp)
1212 {
1213 	crit_enter();
1214 	if (lp->lwp_stat == LSSLEEP || lp->lwp_stat == LSSTOP) {
1215 		/*
1216 		 * Thread is in tsleep.
1217 		 */
1218 
1219 		/*
1220 		 * If the thread is sleeping uninterruptibly
1221 		 * we can't interrupt the sleep... the signal will
1222 		 * be noticed when the lwp returns through
1223 		 * trap() or syscall().
1224 		 *
1225 		 * Otherwise the signal can interrupt the sleep.
1226 		 *
1227 		 * If the process is traced, the lwp will handle the
1228 		 * tracing in issignal() when it returns to userland.
1229 		 */
1230 		if (lp->lwp_flag & LWP_SINTR) {
1231 			/*
1232 			 * Make runnable and break out of any tsleep as well.
1233 			 */
1234 			lp->lwp_flag |= LWP_BREAKTSLEEP;
1235 			setrunnable(lp);
1236 		}
1237 	} else {
1238 		/*
1239 		 * Otherwise the thread is running
1240 		 *
1241 		 * LSRUN does nothing with the signal, other than kicking
1242 		 * ourselves if we are running.
1243 		 * SZOMB and SIDL mean that it will either never be noticed,
1244 		 * or noticed very soon.
1245 		 *
1246 		 * Note that lwp_thread may be NULL or may not be completely
1247 		 * initialized if the process is in the SIDL or SZOMB state.
1248 		 *
1249 		 * For SMP we may have to forward the request to another cpu.
1250 		 * YYY the MP lock prevents the target process from moving
1251 		 * to another cpu, see kern/kern_switch.c
1252 		 *
1253 		 * If the target thread is waiting on its message port,
1254 		 * wakeup the target thread so it can check (or ignore)
1255 		 * the new signal.  YYY needs cleanup.
1256 		 */
1257 		if (lp == lwkt_preempted_proc()) {
1258 			signotify();
1259 		} else if (lp->lwp_stat == LSRUN) {
1260 			struct thread *td = lp->lwp_thread;
1261 			struct proc *p __debugvar = lp->lwp_proc;
1262 
1263 			KASSERT(td != NULL,
1264 			    ("pid %d/%d NULL lwp_thread stat %d flags %08x/%08x",
1265 			    p->p_pid, lp->lwp_tid, lp->lwp_stat,
1266 			    p->p_flag, lp->lwp_flag));
1267 
1268 			/*
1269 			 * To prevent a MP race with TDF_SINTR we must
1270 			 * schedule the thread on the correct cpu.
1271 			 */
1272 #ifdef SMP
1273 			if (td->td_gd != mycpu) {
1274 				LWPHOLD(lp);
1275 				lwkt_send_ipiq(td->td_gd, signotify_remote, lp);
1276 			} else
1277 #endif
1278 			if (td->td_flags & TDF_SINTR)
1279 				lwkt_schedule(td);
1280 		}
1281 	}
1282 	crit_exit();
1283 }
1284 
1285 #ifdef SMP
1286 
1287 /*
1288  * This function is called via an IPI.  We will be in a critical section but
1289  * the MP lock will NOT be held.  Also note that by the time the ipi message
1290  * gets to us the process 'p' (arg) may no longer be scheduled or even valid.
1291  */
1292 static void
1293 signotify_remote(void *arg)
1294 {
1295 	struct lwp *lp = arg;
1296 
1297 	if (lp == lwkt_preempted_proc()) {
1298 		signotify();
1299 	} else {
1300 		struct thread *td = lp->lwp_thread;
1301 		if (td->td_flags & TDF_SINTR)
1302 			lwkt_schedule(td);
1303 	}
1304 	LWPRELE(lp);
1305 }
1306 
1307 #endif
1308 
1309 void
1310 proc_stop(struct proc *p)
1311 {
1312 	struct lwp *lp;
1313 
1314 	/* If somebody raced us, be happy with it */
1315 	if (p->p_stat == SSTOP || p->p_stat == SZOMB)
1316 		return;
1317 
1318 	crit_enter();
1319 	p->p_stat = SSTOP;
1320 
1321 	FOREACH_LWP_IN_PROC(lp, p) {
1322 		switch (lp->lwp_stat) {
1323 		case LSSTOP:
1324 			/*
1325 			 * Do nothing, we are already counted in
1326 			 * p_nstopped.
1327 			 */
1328 			break;
1329 
1330 		case LSSLEEP:
1331 			/*
1332 			 * We're sleeping, but we will stop before
1333 			 * returning to userspace, so count us
1334 			 * as stopped as well.  We set LWP_WSTOP
1335 			 * to signal the lwp that it should not
1336 			 * increase p_nstopped when reaching tstop().
1337 			 */
1338 			if ((lp->lwp_flag & LWP_WSTOP) == 0) {
1339 				lp->lwp_flag |= LWP_WSTOP;
1340 				++p->p_nstopped;
1341 			}
1342 			break;
1343 
1344 		case LSRUN:
1345 			/*
1346 			 * We might notify ourself, but that's not
1347 			 * a problem.
1348 			 */
1349 			lwp_signotify(lp);
1350 			break;
1351 		}
1352 	}
1353 
1354 	if (p->p_nstopped == p->p_nthreads) {
1355 		p->p_flag &= ~P_WAITED;
1356 		wakeup(p->p_pptr);
1357 		if ((p->p_pptr->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1358 			ksignal(p->p_pptr, SIGCHLD);
1359 	}
1360 	crit_exit();
1361 }
1362 
1363 void
1364 proc_unstop(struct proc *p)
1365 {
1366 	struct lwp *lp;
1367 
1368 	if (p->p_stat != SSTOP)
1369 		return;
1370 
1371 	crit_enter();
1372 	p->p_stat = SACTIVE;
1373 
1374 	FOREACH_LWP_IN_PROC(lp, p) {
1375 		switch (lp->lwp_stat) {
1376 		case LSRUN:
1377 			/*
1378 			 * Uh?  Not stopped?  Well, I guess that's okay.
1379 			 */
1380 			if (bootverbose)
1381 				kprintf("proc_unstop: lwp %d/%d not sleeping\n",
1382 					p->p_pid, lp->lwp_tid);
1383 			break;
1384 
1385 		case LSSLEEP:
1386 			/*
1387 			 * Still sleeping.  Don't bother waking it up.
1388 			 * However, if this thread was counted as
1389 			 * stopped, undo this.
1390 			 *
1391 			 * Nevertheless we call setrunnable() so that it
1392 			 * will wake up in case a signal or timeout arrived
1393 			 * in the meantime.
1394 			 */
1395 			if (lp->lwp_flag & LWP_WSTOP) {
1396 				lp->lwp_flag &= ~LWP_WSTOP;
1397 				--p->p_nstopped;
1398 			} else {
1399 				if (bootverbose)
1400 					kprintf("proc_unstop: lwp %d/%d sleeping, not stopped\n",
1401 						p->p_pid, lp->lwp_tid);
1402 			}
1403 			/* FALLTHROUGH */
1404 
1405 		case LSSTOP:
1406 			setrunnable(lp);
1407 			break;
1408 
1409 		}
1410 	}
1411 	crit_exit();
1412 }
1413 
1414 static int
1415 kern_sigtimedwait(sigset_t waitset, siginfo_t *info, struct timespec *timeout)
1416 {
1417 	sigset_t savedmask, set;
1418 	struct proc *p = curproc;
1419 	struct lwp *lp = curthread->td_lwp;
1420 	int error, sig, hz, timevalid = 0;
1421 	struct timespec rts, ets, ts;
1422 	struct timeval tv;
1423 
1424 	error = 0;
1425 	sig = 0;
1426 	ets.tv_sec = 0;		/* silence compiler warning */
1427 	ets.tv_nsec = 0;	/* silence compiler warning */
1428 	SIG_CANTMASK(waitset);
1429 	savedmask = lp->lwp_sigmask;
1430 
1431 	if (timeout) {
1432 		if (timeout->tv_sec >= 0 && timeout->tv_nsec >= 0 &&
1433 		    timeout->tv_nsec < 1000000000) {
1434 			timevalid = 1;
1435 			getnanouptime(&rts);
1436 		 	ets = rts;
1437 			timespecadd(&ets, timeout);
1438 		}
1439 	}
1440 
1441 	for (;;) {
1442 		set = lwp_sigpend(lp);
1443 		SIGSETAND(set, waitset);
1444 		if ((sig = sig_ffs(&set)) != 0) {
1445 			SIGFILLSET(lp->lwp_sigmask);
1446 			SIGDELSET(lp->lwp_sigmask, sig);
1447 			SIG_CANTMASK(lp->lwp_sigmask);
1448 			sig = issignal(lp, 1);
1449 			/*
1450 			 * It may be a STOP signal, in the case, issignal
1451 			 * returns 0, because we may stop there, and new
1452 			 * signal can come in, we should restart if we got
1453 			 * nothing.
1454 			 */
1455 			if (sig == 0)
1456 				continue;
1457 			else
1458 				break;
1459 		}
1460 
1461 		/*
1462 		 * Previous checking got nothing, and we retried but still
1463 		 * got nothing, we should return the error status.
1464 		 */
1465 		if (error)
1466 			break;
1467 
1468 		/*
1469 		 * POSIX says this must be checked after looking for pending
1470 		 * signals.
1471 		 */
1472 		if (timeout) {
1473 			if (timevalid == 0) {
1474 				error = EINVAL;
1475 				break;
1476 			}
1477 			getnanouptime(&rts);
1478 			if (timespeccmp(&rts, &ets, >=)) {
1479 				error = EAGAIN;
1480 				break;
1481 			}
1482 			ts = ets;
1483 			timespecsub(&ts, &rts);
1484 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1485 			hz = tvtohz_high(&tv);
1486 		} else
1487 			hz = 0;
1488 
1489 		lp->lwp_sigmask = savedmask;
1490 		SIGSETNAND(lp->lwp_sigmask, waitset);
1491 		/*
1492 		 * We won't ever be woken up.  Instead, our sleep will
1493 		 * be broken in lwpsignal().
1494 		 */
1495 		error = tsleep(&p->p_sigacts, PCATCH, "sigwt", hz);
1496 		if (timeout) {
1497 			if (error == ERESTART) {
1498 				/* can not restart a timeout wait. */
1499 				error = EINTR;
1500 			} else if (error == EAGAIN) {
1501 				/* will calculate timeout by ourself. */
1502 				error = 0;
1503 			}
1504 		}
1505 		/* Retry ... */
1506 	}
1507 
1508 	lp->lwp_sigmask = savedmask;
1509 	if (sig) {
1510 		error = 0;
1511 		bzero(info, sizeof(*info));
1512 		info->si_signo = sig;
1513 		lwp_delsig(lp, sig);	/* take the signal! */
1514 
1515 		if (sig == SIGKILL)
1516 			sigexit(lp, sig);
1517 	}
1518 	return (error);
1519 }
1520 
1521 /*
1522  * MPALMOSTSAFE
1523  */
1524 int
1525 sys_sigtimedwait(struct sigtimedwait_args *uap)
1526 {
1527 	struct timespec ts;
1528 	struct timespec *timeout;
1529 	sigset_t set;
1530 	siginfo_t info;
1531 	int error;
1532 
1533 	if (uap->timeout) {
1534 		error = copyin(uap->timeout, &ts, sizeof(ts));
1535 		if (error)
1536 			return (error);
1537 		timeout = &ts;
1538 	} else {
1539 		timeout = NULL;
1540 	}
1541 	error = copyin(uap->set, &set, sizeof(set));
1542 	if (error)
1543 		return (error);
1544 	get_mplock();
1545 	error = kern_sigtimedwait(set, &info, timeout);
1546 	rel_mplock();
1547 	if (error)
1548 		return (error);
1549  	if (uap->info)
1550 		error = copyout(&info, uap->info, sizeof(info));
1551 	/* Repost if we got an error. */
1552 	/*
1553 	 * XXX lwp
1554 	 *
1555 	 * This could transform a thread-specific signal to another
1556 	 * thread / process pending signal.
1557 	 */
1558 	if (error) {
1559 		get_mplock();
1560 		ksignal(curproc, info.si_signo);
1561 		rel_mplock();
1562 	} else {
1563 		uap->sysmsg_result = info.si_signo;
1564 	}
1565 	return (error);
1566 }
1567 
1568 /*
1569  * MPALMOSTSAFE
1570  */
1571 int
1572 sys_sigwaitinfo(struct sigwaitinfo_args *uap)
1573 {
1574 	siginfo_t info;
1575 	sigset_t set;
1576 	int error;
1577 
1578 	error = copyin(uap->set, &set, sizeof(set));
1579 	if (error)
1580 		return (error);
1581 	get_mplock();
1582 	error = kern_sigtimedwait(set, &info, NULL);
1583 	rel_mplock();
1584 	if (error)
1585 		return (error);
1586 	if (uap->info)
1587 		error = copyout(&info, uap->info, sizeof(info));
1588 	/* Repost if we got an error. */
1589 	/*
1590 	 * XXX lwp
1591 	 *
1592 	 * This could transform a thread-specific signal to another
1593 	 * thread / process pending signal.
1594 	 */
1595 	if (error) {
1596 		get_mplock();
1597 		ksignal(curproc, info.si_signo);
1598 		rel_mplock();
1599 	} else {
1600 		uap->sysmsg_result = info.si_signo;
1601 	}
1602 	return (error);
1603 }
1604 
1605 /*
1606  * If the current process has received a signal that would interrupt a
1607  * system call, return EINTR or ERESTART as appropriate.
1608  */
1609 int
1610 iscaught(struct lwp *lp)
1611 {
1612 	struct proc *p = lp->lwp_proc;
1613 	int sig;
1614 
1615 	if (p) {
1616 		if ((sig = CURSIG(lp)) != 0) {
1617 			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1618 				return (EINTR);
1619 			return (ERESTART);
1620 		}
1621 	}
1622 	return(EWOULDBLOCK);
1623 }
1624 
1625 /*
1626  * If the current process has received a signal (should be caught or cause
1627  * termination, should interrupt current syscall), return the signal number.
1628  * Stop signals with default action are processed immediately, then cleared;
1629  * they aren't returned.  This is checked after each entry to the system for
1630  * a syscall or trap (though this can usually be done without calling issignal
1631  * by checking the pending signal masks in the CURSIG macro.) The normal call
1632  * sequence is
1633  *
1634  * This routine is called via CURSIG/__cursig and the MP lock might not be
1635  * held.  Obtain the MP lock for the duration of the operation.
1636  *
1637  *	while (sig = CURSIG(curproc))
1638  *		postsig(sig);
1639  */
1640 int
1641 issignal(struct lwp *lp, int maytrace)
1642 {
1643 	struct proc *p = lp->lwp_proc;
1644 	sigset_t mask;
1645 	int sig, prop;
1646 
1647 	get_mplock();
1648 	for (;;) {
1649 		int traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
1650 
1651 		/*
1652 		 * If this process is supposed to stop, stop this thread.
1653 		 */
1654 		if (p->p_stat == SSTOP)
1655 			tstop();
1656 
1657 		mask = lwp_sigpend(lp);
1658 		SIGSETNAND(mask, lp->lwp_sigmask);
1659 		if (p->p_flag & P_PPWAIT)
1660 			SIG_STOPSIGMASK(mask);
1661 		if (SIGISEMPTY(mask)) {		/* no signal to send */
1662 			rel_mplock();
1663 			return (0);
1664 		}
1665 		sig = sig_ffs(&mask);
1666 
1667 		STOPEVENT(p, S_SIG, sig);
1668 
1669 		/*
1670 		 * We should see pending but ignored signals
1671 		 * only if P_TRACED was on when they were posted.
1672 		 */
1673 		if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
1674 			lwp_delsig(lp, sig);
1675 			continue;
1676 		}
1677 		if (maytrace && (p->p_flag & P_TRACED) && (p->p_flag & P_PPWAIT) == 0) {
1678 			/*
1679 			 * If traced, always stop, and stay stopped until
1680 			 * released by the parent.
1681 			 *
1682 			 * NOTE: SSTOP may get cleared during the loop,
1683 			 * but we do not re-notify the parent if we have
1684 			 * to loop several times waiting for the parent
1685 			 * to let us continue.
1686 			 *
1687 			 * XXX not sure if this is still true
1688 			 */
1689 			p->p_xstat = sig;
1690 			proc_stop(p);
1691 			do {
1692 				tstop();
1693 			} while (!trace_req(p) && (p->p_flag & P_TRACED));
1694 
1695 			/*
1696 			 * If parent wants us to take the signal,
1697 			 * then it will leave it in p->p_xstat;
1698 			 * otherwise we just look for signals again.
1699 			 */
1700 			lwp_delsig(lp, sig);	/* clear old signal */
1701 			sig = p->p_xstat;
1702 			if (sig == 0)
1703 				continue;
1704 
1705 			/*
1706 			 * Put the new signal into p_siglist.  If the
1707 			 * signal is being masked, look for other signals.
1708 			 *
1709 			 * XXX lwp might need a call to ksignal()
1710 			 */
1711 			SIGADDSET(p->p_siglist, sig);
1712 			if (SIGISMEMBER(lp->lwp_sigmask, sig))
1713 				continue;
1714 
1715 			/*
1716 			 * If the traced bit got turned off, go back up
1717 			 * to the top to rescan signals.  This ensures
1718 			 * that p_sig* and ps_sigact are consistent.
1719 			 */
1720 			if ((p->p_flag & P_TRACED) == 0)
1721 				continue;
1722 		}
1723 
1724 		prop = sigprop(sig);
1725 
1726 		/*
1727 		 * Decide whether the signal should be returned.
1728 		 * Return the signal's number, or fall through
1729 		 * to clear it from the pending mask.
1730 		 */
1731 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
1732 		case (intptr_t)SIG_DFL:
1733 			/*
1734 			 * Don't take default actions on system processes.
1735 			 */
1736 			if (p->p_pid <= 1) {
1737 #ifdef DIAGNOSTIC
1738 				/*
1739 				 * Are you sure you want to ignore SIGSEGV
1740 				 * in init? XXX
1741 				 */
1742 				kprintf("Process (pid %lu) got signal %d\n",
1743 					(u_long)p->p_pid, sig);
1744 #endif
1745 				break;		/* == ignore */
1746 			}
1747 
1748 			/*
1749 			 * Handle the in-kernel checkpoint action
1750 			 */
1751 			if (prop & SA_CKPT) {
1752 				checkpoint_signal_handler(lp);
1753 				break;
1754 			}
1755 
1756 			/*
1757 			 * If there is a pending stop signal to process
1758 			 * with default action, stop here,
1759 			 * then clear the signal.  However,
1760 			 * if process is member of an orphaned
1761 			 * process group, ignore tty stop signals.
1762 			 */
1763 			if (prop & SA_STOP) {
1764 				if (p->p_flag & P_TRACED ||
1765 		    		    (p->p_pgrp->pg_jobc == 0 &&
1766 				    prop & SA_TTYSTOP))
1767 					break;	/* == ignore */
1768 				p->p_xstat = sig;
1769 				proc_stop(p);
1770 				tstop();
1771 				break;
1772 			} else if (prop & SA_IGNORE) {
1773 				/*
1774 				 * Except for SIGCONT, shouldn't get here.
1775 				 * Default action is to ignore; drop it.
1776 				 */
1777 				break;		/* == ignore */
1778 			} else {
1779 				rel_mplock();
1780 				return (sig);
1781 			}
1782 
1783 			/*NOTREACHED*/
1784 
1785 		case (intptr_t)SIG_IGN:
1786 			/*
1787 			 * Masking above should prevent us ever trying
1788 			 * to take action on an ignored signal other
1789 			 * than SIGCONT, unless process is traced.
1790 			 */
1791 			if ((prop & SA_CONT) == 0 &&
1792 			    (p->p_flag & P_TRACED) == 0)
1793 				kprintf("issignal\n");
1794 			break;		/* == ignore */
1795 
1796 		default:
1797 			/*
1798 			 * This signal has an action, let
1799 			 * postsig() process it.
1800 			 */
1801 			rel_mplock();
1802 			return (sig);
1803 		}
1804 		lwp_delsig(lp, sig);		/* take the signal! */
1805 	}
1806 	/* NOTREACHED */
1807 }
1808 
1809 /*
1810  * Take the action for the specified signal
1811  * from the current set of pending signals.
1812  */
1813 void
1814 postsig(int sig)
1815 {
1816 	struct lwp *lp = curthread->td_lwp;
1817 	struct proc *p = lp->lwp_proc;
1818 	struct sigacts *ps = p->p_sigacts;
1819 	sig_t action;
1820 	sigset_t returnmask;
1821 	int code;
1822 
1823 	KASSERT(sig != 0, ("postsig"));
1824 
1825 	/*
1826 	 * If we are a virtual kernel running an emulated user process
1827 	 * context, switch back to the virtual kernel context before
1828 	 * trying to post the signal.
1829 	 */
1830 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1831 		struct trapframe *tf = lp->lwp_md.md_regs;
1832 		tf->tf_trapno = 0;
1833 		vkernel_trap(lp, tf);
1834 	}
1835 
1836 	lwp_delsig(lp, sig);
1837 	action = ps->ps_sigact[_SIG_IDX(sig)];
1838 #ifdef KTRACE
1839 	if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
1840 		ktrpsig(lp, sig, action, lp->lwp_flag & LWP_OLDMASK ?
1841 			&lp->lwp_oldsigmask : &lp->lwp_sigmask, 0);
1842 #endif
1843 	STOPEVENT(p, S_SIG, sig);
1844 
1845 	if (action == SIG_DFL) {
1846 		/*
1847 		 * Default action, where the default is to kill
1848 		 * the process.  (Other cases were ignored above.)
1849 		 */
1850 		sigexit(lp, sig);
1851 		/* NOTREACHED */
1852 	} else {
1853 		/*
1854 		 * If we get here, the signal must be caught.
1855 		 */
1856 		KASSERT(action != SIG_IGN && !SIGISMEMBER(lp->lwp_sigmask, sig),
1857 		    ("postsig action"));
1858 
1859 		crit_enter();
1860 
1861 		/*
1862 		 * Reset the signal handler if asked to
1863 		 */
1864 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
1865 			/*
1866 			 * See kern_sigaction() for origin of this code.
1867 			 */
1868 			SIGDELSET(p->p_sigcatch, sig);
1869 			if (sig != SIGCONT &&
1870 			    sigprop(sig) & SA_IGNORE)
1871 				SIGADDSET(p->p_sigignore, sig);
1872 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1873 		}
1874 
1875 		/*
1876 		 * Handle the mailbox case.  Copyout to the appropriate
1877 		 * location but do not generate a signal frame.  The system
1878 		 * call simply returns EINTR and the user is responsible for
1879 		 * polling the mailbox.
1880 		 */
1881 		if (SIGISMEMBER(ps->ps_sigmailbox, sig)) {
1882 			int sig_copy = sig;
1883 			copyout(&sig_copy, (void *)action, sizeof(int));
1884 			curproc->p_flag |= P_MAILBOX;
1885 			crit_exit();
1886 			goto done;
1887 		}
1888 
1889 		/*
1890 		 * Set the signal mask and calculate the mask to restore
1891 		 * when the signal function returns.
1892 		 *
1893 		 * Special case: user has done a sigsuspend.  Here the
1894 		 * current mask is not of interest, but rather the
1895 		 * mask from before the sigsuspend is what we want
1896 		 * restored after the signal processing is completed.
1897 		 */
1898 		if (lp->lwp_flag & LWP_OLDMASK) {
1899 			returnmask = lp->lwp_oldsigmask;
1900 			lp->lwp_flag &= ~LWP_OLDMASK;
1901 		} else {
1902 			returnmask = lp->lwp_sigmask;
1903 		}
1904 
1905 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
1906 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
1907 			SIGADDSET(lp->lwp_sigmask, sig);
1908 
1909 		crit_exit();
1910 		lp->lwp_ru.ru_nsignals++;
1911 		if (lp->lwp_sig != sig) {
1912 			code = 0;
1913 		} else {
1914 			code = lp->lwp_code;
1915 			lp->lwp_code = 0;
1916 			lp->lwp_sig = 0;
1917 		}
1918 		(*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
1919 	}
1920 done:
1921 	;
1922 }
1923 
1924 /*
1925  * Kill the current process for stated reason.
1926  */
1927 void
1928 killproc(struct proc *p, char *why)
1929 {
1930 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n",
1931 		p->p_pid, p->p_comm,
1932 		p->p_ucred ? p->p_ucred->cr_uid : -1, why);
1933 	ksignal(p, SIGKILL);
1934 }
1935 
1936 /*
1937  * Force the current process to exit with the specified signal, dumping core
1938  * if appropriate.  We bypass the normal tests for masked and caught signals,
1939  * allowing unrecoverable failures to terminate the process without changing
1940  * signal state.  Mark the accounting record with the signal termination.
1941  * If dumping core, save the signal number for the debugger.  Calls exit and
1942  * does not return.
1943  */
1944 void
1945 sigexit(struct lwp *lp, int sig)
1946 {
1947 	struct proc *p = lp->lwp_proc;
1948 
1949 	p->p_acflag |= AXSIG;
1950 	if (sigprop(sig) & SA_CORE) {
1951 		lp->lwp_sig = sig;
1952 		/*
1953 		 * Log signals which would cause core dumps
1954 		 * (Log as LOG_INFO to appease those who don't want
1955 		 * these messages.)
1956 		 * XXX : Todo, as well as euid, write out ruid too
1957 		 */
1958 		if (coredump(lp, sig) == 0)
1959 			sig |= WCOREFLAG;
1960 		if (kern_logsigexit)
1961 			log(LOG_INFO,
1962 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
1963 			    p->p_pid, p->p_comm,
1964 			    p->p_ucred ? p->p_ucred->cr_uid : -1,
1965 			    sig &~ WCOREFLAG,
1966 			    sig & WCOREFLAG ? " (core dumped)" : "");
1967 	}
1968 	exit1(W_EXITCODE(0, sig));
1969 	/* NOTREACHED */
1970 }
1971 
1972 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
1973 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
1974 	      sizeof(corefilename), "process corefile name format string");
1975 
1976 /*
1977  * expand_name(name, uid, pid)
1978  * Expand the name described in corefilename, using name, uid, and pid.
1979  * corefilename is a kprintf-like string, with three format specifiers:
1980  *	%N	name of process ("name")
1981  *	%P	process id (pid)
1982  *	%U	user id (uid)
1983  * For example, "%N.core" is the default; they can be disabled completely
1984  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
1985  * This is controlled by the sysctl variable kern.corefile (see above).
1986  */
1987 
1988 static char *
1989 expand_name(const char *name, uid_t uid, pid_t pid)
1990 {
1991 	char *temp;
1992 	char buf[11];		/* Buffer for pid/uid -- max 4B */
1993 	int i, n;
1994 	char *format = corefilename;
1995 	size_t namelen;
1996 
1997 	temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
1998 	if (temp == NULL)
1999 		return NULL;
2000 	namelen = strlen(name);
2001 	for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
2002 		int l;
2003 		switch (format[i]) {
2004 		case '%':	/* Format character */
2005 			i++;
2006 			switch (format[i]) {
2007 			case '%':
2008 				temp[n++] = '%';
2009 				break;
2010 			case 'N':	/* process name */
2011 				if ((n + namelen) > MAXPATHLEN) {
2012 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2013 					    pid, name, uid, temp, name);
2014 					kfree(temp, M_TEMP);
2015 					return NULL;
2016 				}
2017 				memcpy(temp+n, name, namelen);
2018 				n += namelen;
2019 				break;
2020 			case 'P':	/* process id */
2021 				l = ksprintf(buf, "%u", pid);
2022 				if ((n + l) > MAXPATHLEN) {
2023 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2024 					    pid, name, uid, temp, name);
2025 					kfree(temp, M_TEMP);
2026 					return NULL;
2027 				}
2028 				memcpy(temp+n, buf, l);
2029 				n += l;
2030 				break;
2031 			case 'U':	/* user id */
2032 				l = ksprintf(buf, "%u", uid);
2033 				if ((n + l) > MAXPATHLEN) {
2034 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2035 					    pid, name, uid, temp, name);
2036 					kfree(temp, M_TEMP);
2037 					return NULL;
2038 				}
2039 				memcpy(temp+n, buf, l);
2040 				n += l;
2041 				break;
2042 			default:
2043 			  	log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
2044 			}
2045 			break;
2046 		default:
2047 			temp[n++] = format[i];
2048 		}
2049 	}
2050 	temp[n] = '\0';
2051 	return temp;
2052 }
2053 
2054 /*
2055  * Dump a process' core.  The main routine does some
2056  * policy checking, and creates the name of the coredump;
2057  * then it passes on a vnode and a size limit to the process-specific
2058  * coredump routine if there is one; if there _is not_ one, it returns
2059  * ENOSYS; otherwise it returns the error from the process-specific routine.
2060  *
2061  * The parameter `lp' is the lwp which triggered the coredump.
2062  */
2063 
2064 static int
2065 coredump(struct lwp *lp, int sig)
2066 {
2067 	struct proc *p = lp->lwp_proc;
2068 	struct vnode *vp;
2069 	struct ucred *cred = p->p_ucred;
2070 	struct flock lf;
2071 	struct nlookupdata nd;
2072 	struct vattr vattr;
2073 	int error, error1;
2074 	char *name;			/* name of corefile */
2075 	off_t limit;
2076 
2077 	STOPEVENT(p, S_CORE, 0);
2078 
2079 	if (((sugid_coredump == 0) && p->p_flag & P_SUGID) || do_coredump == 0)
2080 		return (EFAULT);
2081 
2082 	/*
2083 	 * Note that the bulk of limit checking is done after
2084 	 * the corefile is created.  The exception is if the limit
2085 	 * for corefiles is 0, in which case we don't bother
2086 	 * creating the corefile at all.  This layout means that
2087 	 * a corefile is truncated instead of not being created,
2088 	 * if it is larger than the limit.
2089 	 */
2090 	limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
2091 	if (limit == 0)
2092 		return EFBIG;
2093 
2094 	name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
2095 	if (name == NULL)
2096 		return (EINVAL);
2097 	error = nlookup_init(&nd, name, UIO_SYSSPACE, NLC_LOCKVP);
2098 	if (error == 0)
2099 		error = vn_open(&nd, NULL, O_CREAT | FWRITE | O_NOFOLLOW, S_IRUSR | S_IWUSR);
2100 	kfree(name, M_TEMP);
2101 	if (error) {
2102 		nlookup_done(&nd);
2103 		return (error);
2104 	}
2105 	vp = nd.nl_open_vp;
2106 	nd.nl_open_vp = NULL;
2107 	nlookup_done(&nd);
2108 
2109 	vn_unlock(vp);
2110 	lf.l_whence = SEEK_SET;
2111 	lf.l_start = 0;
2112 	lf.l_len = 0;
2113 	lf.l_type = F_WRLCK;
2114 	error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, 0);
2115 	if (error)
2116 		goto out2;
2117 
2118 	/* Don't dump to non-regular files or files with links. */
2119 	if (vp->v_type != VREG ||
2120 	    VOP_GETATTR(vp, &vattr) || vattr.va_nlink != 1) {
2121 		error = EFAULT;
2122 		goto out1;
2123 	}
2124 
2125 	/* Don't dump to files current user does not own */
2126 	if (vattr.va_uid != p->p_ucred->cr_uid) {
2127 		error = EFAULT;
2128 		goto out1;
2129 	}
2130 
2131 	VATTR_NULL(&vattr);
2132 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2133 	vattr.va_size = 0;
2134 	VOP_SETATTR(vp, &vattr, cred);
2135 	p->p_acflag |= ACORE;
2136 	vn_unlock(vp);
2137 
2138 	error = p->p_sysent->sv_coredump ?
2139 		  p->p_sysent->sv_coredump(lp, sig, vp, limit) : ENOSYS;
2140 
2141 out1:
2142 	lf.l_type = F_UNLCK;
2143 	VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, 0);
2144 out2:
2145 	error1 = vn_close(vp, FWRITE);
2146 	if (error == 0)
2147 		error = error1;
2148 	return (error);
2149 }
2150 
2151 /*
2152  * Nonexistent system call-- signal process (may want to handle it).
2153  * Flag error in case process won't see signal immediately (blocked or ignored).
2154  *
2155  * MPALMOSTSAFE
2156  */
2157 /* ARGSUSED */
2158 int
2159 sys_nosys(struct nosys_args *args)
2160 {
2161 	get_mplock();
2162 	lwpsignal(curproc, curthread->td_lwp, SIGSYS);
2163 	rel_mplock();
2164 	return (EINVAL);
2165 }
2166 
2167 /*
2168  * Send a SIGIO or SIGURG signal to a process or process group using
2169  * stored credentials rather than those of the current process.
2170  */
2171 void
2172 pgsigio(struct sigio *sigio, int sig, int checkctty)
2173 {
2174 	if (sigio == NULL)
2175 		return;
2176 
2177 	if (sigio->sio_pgid > 0) {
2178 		if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
2179 		             sigio->sio_proc))
2180 			ksignal(sigio->sio_proc, sig);
2181 	} else if (sigio->sio_pgid < 0) {
2182 		struct proc *p;
2183 
2184 		lockmgr(&sigio->sio_pgrp->pg_lock, LK_EXCLUSIVE);
2185 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
2186 			if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
2187 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
2188 				ksignal(p, sig);
2189 		}
2190 		lockmgr(&sigio->sio_pgrp->pg_lock, LK_RELEASE);
2191 	}
2192 }
2193 
2194 static int
2195 filt_sigattach(struct knote *kn)
2196 {
2197 	struct proc *p = curproc;
2198 
2199 	kn->kn_ptr.p_proc = p;
2200 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
2201 
2202 	/* XXX lock the proc here while adding to the list? */
2203 	SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext);
2204 
2205 	return (0);
2206 }
2207 
2208 static void
2209 filt_sigdetach(struct knote *kn)
2210 {
2211 	struct proc *p = kn->kn_ptr.p_proc;
2212 
2213 	SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
2214 }
2215 
2216 /*
2217  * signal knotes are shared with proc knotes, so we apply a mask to
2218  * the hint in order to differentiate them from process hints.  This
2219  * could be avoided by using a signal-specific knote list, but probably
2220  * isn't worth the trouble.
2221  */
2222 static int
2223 filt_signal(struct knote *kn, long hint)
2224 {
2225 	if (hint & NOTE_SIGNAL) {
2226 		hint &= ~NOTE_SIGNAL;
2227 
2228 		if (kn->kn_id == hint)
2229 			kn->kn_data++;
2230 	}
2231 	return (kn->kn_data != 0);
2232 }
2233