xref: /dragonfly/sys/kern/kern_sig.c (revision b7367ef6)
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 1.84 2007/08/30 20:41:00 pavalos 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/signal2.h>
51 #include <sys/resourcevar.h>
52 #include <sys/vnode.h>
53 #include <sys/event.h>
54 #include <sys/proc.h>
55 #include <sys/nlookup.h>
56 #include <sys/pioctl.h>
57 #include <sys/systm.h>
58 #include <sys/acct.h>
59 #include <sys/fcntl.h>
60 #include <sys/lock.h>
61 #include <sys/wait.h>
62 #include <sys/ktrace.h>
63 #include <sys/syslog.h>
64 #include <sys/stat.h>
65 #include <sys/sysent.h>
66 #include <sys/sysctl.h>
67 #include <sys/malloc.h>
68 #include <sys/interrupt.h>
69 #include <sys/unistd.h>
70 #include <sys/kern_syscall.h>
71 #include <sys/vkernel.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 #ifdef SMP
83 static void	signotify_remote(void *arg);
84 #endif
85 static int	kern_sigtimedwait(sigset_t set, siginfo_t *info,
86 		    struct timespec *timeout);
87 
88 static int	filt_sigattach(struct knote *kn);
89 static void	filt_sigdetach(struct knote *kn);
90 static int	filt_signal(struct knote *kn, long hint);
91 
92 struct filterops sig_filtops =
93 	{ 0, filt_sigattach, filt_sigdetach, filt_signal };
94 
95 static int	kern_logsigexit = 1;
96 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
97     &kern_logsigexit, 0,
98     "Log processes quitting on abnormal signals to syslog(3)");
99 
100 /*
101  * Can process p, with pcred pc, send the signal sig to process q?
102  */
103 #define CANSIGNAL(q, sig) \
104 	(!p_trespass(curproc->p_ucred, (q)->p_ucred) || \
105 	((sig) == SIGCONT && (q)->p_session == curproc->p_session))
106 
107 /*
108  * Policy -- Can real uid ruid with ucred uc send a signal to process q?
109  */
110 #define CANSIGIO(ruid, uc, q) \
111 	((uc)->cr_uid == 0 || \
112 	    (ruid) == (q)->p_ucred->cr_ruid || \
113 	    (uc)->cr_uid == (q)->p_ucred->cr_ruid || \
114 	    (ruid) == (q)->p_ucred->cr_uid || \
115 	    (uc)->cr_uid == (q)->p_ucred->cr_uid)
116 
117 int sugid_coredump;
118 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RW,
119 	&sugid_coredump, 0, "Enable coredumping set user/group ID processes");
120 
121 static int	do_coredump = 1;
122 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
123 	&do_coredump, 0, "Enable/Disable coredumps");
124 
125 /*
126  * Signal properties and actions.
127  * The array below categorizes the signals and their default actions
128  * according to the following properties:
129  */
130 #define	SA_KILL		0x01		/* terminates process by default */
131 #define	SA_CORE		0x02		/* ditto and coredumps */
132 #define	SA_STOP		0x04		/* suspend process */
133 #define	SA_TTYSTOP	0x08		/* ditto, from tty */
134 #define	SA_IGNORE	0x10		/* ignore by default */
135 #define	SA_CONT		0x20		/* continue if suspended */
136 #define	SA_CANTMASK	0x40		/* non-maskable, catchable */
137 #define SA_CKPT         0x80            /* checkpoint process */
138 
139 
140 static int sigproptbl[NSIG] = {
141         SA_KILL,                /* SIGHUP */
142         SA_KILL,                /* SIGINT */
143         SA_KILL|SA_CORE,        /* SIGQUIT */
144         SA_KILL|SA_CORE,        /* SIGILL */
145         SA_KILL|SA_CORE,        /* SIGTRAP */
146         SA_KILL|SA_CORE,        /* SIGABRT */
147         SA_KILL|SA_CORE,        /* SIGEMT */
148         SA_KILL|SA_CORE,        /* SIGFPE */
149         SA_KILL,                /* SIGKILL */
150         SA_KILL|SA_CORE,        /* SIGBUS */
151         SA_KILL|SA_CORE,        /* SIGSEGV */
152         SA_KILL|SA_CORE,        /* SIGSYS */
153         SA_KILL,                /* SIGPIPE */
154         SA_KILL,                /* SIGALRM */
155         SA_KILL,                /* SIGTERM */
156         SA_IGNORE,              /* SIGURG */
157         SA_STOP,                /* SIGSTOP */
158         SA_STOP|SA_TTYSTOP,     /* SIGTSTP */
159         SA_IGNORE|SA_CONT,      /* SIGCONT */
160         SA_IGNORE,              /* SIGCHLD */
161         SA_STOP|SA_TTYSTOP,     /* SIGTTIN */
162         SA_STOP|SA_TTYSTOP,     /* SIGTTOU */
163         SA_IGNORE,              /* SIGIO */
164         SA_KILL,                /* SIGXCPU */
165         SA_KILL,                /* SIGXFSZ */
166         SA_KILL,                /* SIGVTALRM */
167         SA_KILL,                /* SIGPROF */
168         SA_IGNORE,              /* SIGWINCH  */
169         SA_IGNORE,              /* SIGINFO */
170         SA_KILL,                /* SIGUSR1 */
171         SA_KILL,                /* SIGUSR2 */
172 	SA_IGNORE,              /* SIGTHR */
173 	SA_CKPT,                /* SIGCKPT */
174 	SA_KILL|SA_CKPT,        /* SIGCKPTEXIT */
175 	SA_IGNORE,
176 	SA_IGNORE,
177 	SA_IGNORE,
178 	SA_IGNORE,
179 	SA_IGNORE,
180 	SA_IGNORE,
181 	SA_IGNORE,
182 	SA_IGNORE,
183 	SA_IGNORE,
184 	SA_IGNORE,
185 	SA_IGNORE,
186 	SA_IGNORE,
187 	SA_IGNORE,
188 	SA_IGNORE,
189 	SA_IGNORE,
190 	SA_IGNORE,
191 	SA_IGNORE,
192 	SA_IGNORE,
193 	SA_IGNORE,
194 	SA_IGNORE,
195 	SA_IGNORE,
196 	SA_IGNORE,
197 	SA_IGNORE,
198 	SA_IGNORE,
199 	SA_IGNORE,
200 	SA_IGNORE,
201 	SA_IGNORE,
202 	SA_IGNORE,
203 	SA_IGNORE,
204 	SA_IGNORE,
205 
206 };
207 
208 static __inline int
209 sigprop(int sig)
210 {
211 
212 	if (sig > 0 && sig < NSIG)
213 		return (sigproptbl[_SIG_IDX(sig)]);
214 	return (0);
215 }
216 
217 static __inline int
218 sig_ffs(sigset_t *set)
219 {
220 	int i;
221 
222 	for (i = 0; i < _SIG_WORDS; i++)
223 		if (set->__bits[i])
224 			return (ffs(set->__bits[i]) + (i * 32));
225 	return (0);
226 }
227 
228 int
229 kern_sigaction(int sig, struct sigaction *act, struct sigaction *oact)
230 {
231 	struct thread *td = curthread;
232 	struct proc *p = td->td_proc;
233 	struct lwp *lp;
234 	struct sigacts *ps = p->p_sigacts;
235 
236 	if (sig <= 0 || sig > _SIG_MAXSIG)
237 		return (EINVAL);
238 
239 	if (oact) {
240 		oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
241 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
242 		oact->sa_flags = 0;
243 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
244 			oact->sa_flags |= SA_ONSTACK;
245 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
246 			oact->sa_flags |= SA_RESTART;
247 		if (SIGISMEMBER(ps->ps_sigreset, sig))
248 			oact->sa_flags |= SA_RESETHAND;
249 		if (SIGISMEMBER(ps->ps_signodefer, sig))
250 			oact->sa_flags |= SA_NODEFER;
251 		if (SIGISMEMBER(ps->ps_siginfo, sig))
252 			oact->sa_flags |= SA_SIGINFO;
253 		if (SIGISMEMBER(ps->ps_sigmailbox, sig))
254 			oact->sa_flags |= SA_MAILBOX;
255 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDSTOP)
256 			oact->sa_flags |= SA_NOCLDSTOP;
257 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDWAIT)
258 			oact->sa_flags |= SA_NOCLDWAIT;
259 	}
260 	if (act) {
261 		/*
262 		 * Check for invalid requests.  KILL and STOP cannot be
263 		 * caught.
264 		 */
265 		if (sig == SIGKILL || sig == SIGSTOP) {
266 			if (act->sa_handler != SIG_DFL)
267 				return (EINVAL);
268 #if 0
269 			/* (not needed, SIG_DFL forces action to occur) */
270 			if (act->sa_flags & SA_MAILBOX)
271 				return (EINVAL);
272 #endif
273 		}
274 
275 		/*
276 		 * Change setting atomically.
277 		 */
278 		crit_enter();
279 
280 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
281 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
282 		if (act->sa_flags & SA_SIGINFO) {
283 			ps->ps_sigact[_SIG_IDX(sig)] =
284 			    (__sighandler_t *)act->sa_sigaction;
285 			SIGADDSET(ps->ps_siginfo, sig);
286 		} else {
287 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
288 			SIGDELSET(ps->ps_siginfo, sig);
289 		}
290 		if (!(act->sa_flags & SA_RESTART))
291 			SIGADDSET(ps->ps_sigintr, sig);
292 		else
293 			SIGDELSET(ps->ps_sigintr, sig);
294 		if (act->sa_flags & SA_ONSTACK)
295 			SIGADDSET(ps->ps_sigonstack, sig);
296 		else
297 			SIGDELSET(ps->ps_sigonstack, sig);
298 		if (act->sa_flags & SA_RESETHAND)
299 			SIGADDSET(ps->ps_sigreset, sig);
300 		else
301 			SIGDELSET(ps->ps_sigreset, sig);
302 		if (act->sa_flags & SA_NODEFER)
303 			SIGADDSET(ps->ps_signodefer, sig);
304 		else
305 			SIGDELSET(ps->ps_signodefer, sig);
306 		if (act->sa_flags & SA_MAILBOX)
307 			SIGADDSET(ps->ps_sigmailbox, sig);
308 		else
309 			SIGDELSET(ps->ps_sigmailbox, sig);
310 		if (sig == SIGCHLD) {
311 			if (act->sa_flags & SA_NOCLDSTOP)
312 				p->p_sigacts->ps_flag |= PS_NOCLDSTOP;
313 			else
314 				p->p_sigacts->ps_flag &= ~PS_NOCLDSTOP;
315 			if (act->sa_flags & SA_NOCLDWAIT) {
316 				/*
317 				 * Paranoia: since SA_NOCLDWAIT is implemented
318 				 * by reparenting the dying child to PID 1 (and
319 				 * trust it to reap the zombie), PID 1 itself
320 				 * is forbidden to set SA_NOCLDWAIT.
321 				 */
322 				if (p->p_pid == 1)
323 					p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
324 				else
325 					p->p_sigacts->ps_flag |= PS_NOCLDWAIT;
326 			} else {
327 				p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
328 			}
329 		}
330 		/*
331 		 * Set bit in p_sigignore for signals that are set to SIG_IGN,
332 		 * and for signals set to SIG_DFL where the default is to
333 		 * ignore. However, don't put SIGCONT in p_sigignore, as we
334 		 * have to restart the process.
335 		 */
336 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
337 		    (sigprop(sig) & SA_IGNORE &&
338 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
339 			/* never to be seen again */
340 			SIGDELSET(p->p_siglist, sig);
341 			/*
342 			 * Remove the signal also from the thread lists.
343 			 */
344 			FOREACH_LWP_IN_PROC(lp, p) {
345 				SIGDELSET(lp->lwp_siglist, sig);
346 			}
347 			if (sig != SIGCONT)
348 				/* easier in ksignal */
349 				SIGADDSET(p->p_sigignore, sig);
350 			SIGDELSET(p->p_sigcatch, sig);
351 		} else {
352 			SIGDELSET(p->p_sigignore, sig);
353 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
354 				SIGDELSET(p->p_sigcatch, sig);
355 			else
356 				SIGADDSET(p->p_sigcatch, sig);
357 		}
358 
359 		crit_exit();
360 	}
361 	return (0);
362 }
363 
364 int
365 sys_sigaction(struct sigaction_args *uap)
366 {
367 	struct sigaction act, oact;
368 	struct sigaction *actp, *oactp;
369 	int error;
370 
371 	actp = (uap->act != NULL) ? &act : NULL;
372 	oactp = (uap->oact != NULL) ? &oact : NULL;
373 	if (actp) {
374 		error = copyin(uap->act, actp, sizeof(act));
375 		if (error)
376 			return (error);
377 	}
378 	error = kern_sigaction(uap->sig, actp, oactp);
379 	if (oactp && !error) {
380 		error = copyout(oactp, uap->oact, sizeof(oact));
381 	}
382 	return (error);
383 }
384 
385 /*
386  * Initialize signal state for process 0;
387  * set to ignore signals that are ignored by default.
388  */
389 void
390 siginit(struct proc *p)
391 {
392 	int i;
393 
394 	for (i = 1; i <= NSIG; i++)
395 		if (sigprop(i) & SA_IGNORE && i != SIGCONT)
396 			SIGADDSET(p->p_sigignore, i);
397 }
398 
399 /*
400  * Reset signals for an exec of the specified process.
401  */
402 void
403 execsigs(struct proc *p)
404 {
405 	struct sigacts *ps = p->p_sigacts;
406 	struct lwp *lp;
407 	int sig;
408 
409 	lp = ONLY_LWP_IN_PROC(p);
410 
411 	/*
412 	 * Reset caught signals.  Held signals remain held
413 	 * through p_sigmask (unless they were caught,
414 	 * and are now ignored by default).
415 	 */
416 	while (SIGNOTEMPTY(p->p_sigcatch)) {
417 		sig = sig_ffs(&p->p_sigcatch);
418 		SIGDELSET(p->p_sigcatch, sig);
419 		if (sigprop(sig) & SA_IGNORE) {
420 			if (sig != SIGCONT)
421 				SIGADDSET(p->p_sigignore, sig);
422 			SIGDELSET(p->p_siglist, sig);
423 			SIGDELSET(lp->lwp_siglist, sig);
424 		}
425 		ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
426 	}
427 
428 	/*
429 	 * Reset stack state to the user stack.
430 	 * Clear set of signals caught on the signal stack.
431 	 */
432 	lp->lwp_sigstk.ss_flags = SS_DISABLE;
433 	lp->lwp_sigstk.ss_size = 0;
434 	lp->lwp_sigstk.ss_sp = 0;
435 	lp->lwp_flag &= ~LWP_ALTSTACK;
436 	/*
437 	 * Reset no zombies if child dies flag as Solaris does.
438 	 */
439 	p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
440 }
441 
442 /*
443  * kern_sigprocmask() - MP SAFE ONLY IF p == curproc
444  *
445  *	Manipulate signal mask.  This routine is MP SAFE *ONLY* if
446  *	p == curproc.
447  */
448 int
449 kern_sigprocmask(int how, sigset_t *set, sigset_t *oset)
450 {
451 	struct thread *td = curthread;
452 	struct lwp *lp = td->td_lwp;
453 	int error;
454 
455 	if (oset != NULL)
456 		*oset = lp->lwp_sigmask;
457 
458 	error = 0;
459 	if (set != NULL) {
460 		switch (how) {
461 		case SIG_BLOCK:
462 			SIG_CANTMASK(*set);
463 			SIGSETOR(lp->lwp_sigmask, *set);
464 			break;
465 		case SIG_UNBLOCK:
466 			SIGSETNAND(lp->lwp_sigmask, *set);
467 			break;
468 		case SIG_SETMASK:
469 			SIG_CANTMASK(*set);
470 			lp->lwp_sigmask = *set;
471 			break;
472 		default:
473 			error = EINVAL;
474 			break;
475 		}
476 	}
477 	return (error);
478 }
479 
480 /*
481  * sigprocmask() - MP SAFE
482  */
483 int
484 sys_sigprocmask(struct sigprocmask_args *uap)
485 {
486 	sigset_t set, oset;
487 	sigset_t *setp, *osetp;
488 	int error;
489 
490 	setp = (uap->set != NULL) ? &set : NULL;
491 	osetp = (uap->oset != NULL) ? &oset : NULL;
492 	if (setp) {
493 		error = copyin(uap->set, setp, sizeof(set));
494 		if (error)
495 			return (error);
496 	}
497 	error = kern_sigprocmask(uap->how, setp, osetp);
498 	if (osetp && !error) {
499 		error = copyout(osetp, uap->oset, sizeof(oset));
500 	}
501 	return (error);
502 }
503 
504 int
505 kern_sigpending(struct __sigset *set)
506 {
507 	struct lwp *lp = curthread->td_lwp;
508 
509 	*set = lwp_sigpend(lp);
510 
511 	return (0);
512 }
513 
514 int
515 sys_sigpending(struct sigpending_args *uap)
516 {
517 	sigset_t set;
518 	int error;
519 
520 	error = kern_sigpending(&set);
521 
522 	if (error == 0)
523 		error = copyout(&set, uap->set, sizeof(set));
524 	return (error);
525 }
526 
527 /*
528  * Suspend process until signal, providing mask to be set
529  * in the meantime.
530  */
531 int
532 kern_sigsuspend(struct __sigset *set)
533 {
534 	struct thread *td = curthread;
535 	struct lwp *lp = td->td_lwp;
536 	struct proc *p = td->td_proc;
537 	struct sigacts *ps = p->p_sigacts;
538 
539 	/*
540 	 * When returning from sigsuspend, we want
541 	 * the old mask to be restored after the
542 	 * signal handler has finished.  Thus, we
543 	 * save it here and mark the sigacts structure
544 	 * to indicate this.
545 	 */
546 	lp->lwp_oldsigmask = lp->lwp_sigmask;
547 	lp->lwp_flag |= LWP_OLDMASK;
548 
549 	SIG_CANTMASK(*set);
550 	lp->lwp_sigmask = *set;
551 	while (tsleep(ps, PCATCH, "pause", 0) == 0)
552 		/* void */;
553 	/* always return EINTR rather than ERESTART... */
554 	return (EINTR);
555 }
556 
557 /*
558  * Note nonstandard calling convention: libc stub passes mask, not
559  * pointer, to save a copyin.
560  */
561 int
562 sys_sigsuspend(struct sigsuspend_args *uap)
563 {
564 	sigset_t mask;
565 	int error;
566 
567 	error = copyin(uap->sigmask, &mask, sizeof(mask));
568 	if (error)
569 		return (error);
570 
571 	error = kern_sigsuspend(&mask);
572 
573 	return (error);
574 }
575 
576 int
577 kern_sigaltstack(struct sigaltstack *ss, struct sigaltstack *oss)
578 {
579 	struct thread *td = curthread;
580 	struct lwp *lp = td->td_lwp;
581 	struct proc *p = td->td_proc;
582 
583 	if ((lp->lwp_flag & LWP_ALTSTACK) == 0)
584 		lp->lwp_sigstk.ss_flags |= SS_DISABLE;
585 
586 	if (oss)
587 		*oss = lp->lwp_sigstk;
588 
589 	if (ss) {
590 		if (ss->ss_flags & SS_DISABLE) {
591 			if (lp->lwp_sigstk.ss_flags & SS_ONSTACK)
592 				return (EINVAL);
593 			lp->lwp_flag &= ~LWP_ALTSTACK;
594 			lp->lwp_sigstk.ss_flags = ss->ss_flags;
595 		} else {
596 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
597 				return (ENOMEM);
598 			lp->lwp_flag |= LWP_ALTSTACK;
599 			lp->lwp_sigstk = *ss;
600 		}
601 	}
602 
603 	return (0);
604 }
605 
606 int
607 sys_sigaltstack(struct sigaltstack_args *uap)
608 {
609 	stack_t ss, oss;
610 	int error;
611 
612 	if (uap->ss) {
613 		error = copyin(uap->ss, &ss, sizeof(ss));
614 		if (error)
615 			return (error);
616 	}
617 
618 	error = kern_sigaltstack(uap->ss ? &ss : NULL,
619 	    uap->oss ? &oss : NULL);
620 
621 	if (error == 0 && uap->oss)
622 		error = copyout(&oss, uap->oss, sizeof(*uap->oss));
623 	return (error);
624 }
625 
626 /*
627  * Common code for kill process group/broadcast kill.
628  * cp is calling process.
629  */
630 struct killpg_info {
631 	int nfound;
632 	int sig;
633 };
634 
635 static int killpg_all_callback(struct proc *p, void *data);
636 
637 static int
638 dokillpg(int sig, int pgid, int all)
639 {
640 	struct killpg_info info;
641 	struct proc *cp = curproc;
642 	struct proc *p;
643 	struct pgrp *pgrp;
644 
645 	info.nfound = 0;
646 	info.sig = sig;
647 
648 	if (all) {
649 		/*
650 		 * broadcast
651 		 */
652 		allproc_scan(killpg_all_callback, &info);
653 	} else {
654 		if (pgid == 0) {
655 			/*
656 			 * zero pgid means send to my process group.
657 			 */
658 			pgrp = cp->p_pgrp;
659 		} else {
660 			pgrp = pgfind(pgid);
661 			if (pgrp == NULL)
662 				return (ESRCH);
663 		}
664 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
665 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
666 			if (p->p_pid <= 1 ||
667 			    p->p_stat == SZOMB ||
668 			    (p->p_flag & P_SYSTEM) ||
669 			    !CANSIGNAL(p, sig)) {
670 				continue;
671 			}
672 			++info.nfound;
673 			if (sig)
674 				ksignal(p, sig);
675 		}
676 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
677 	}
678 	return (info.nfound ? 0 : ESRCH);
679 }
680 
681 static int
682 killpg_all_callback(struct proc *p, void *data)
683 {
684 	struct killpg_info *info = data;
685 
686 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) ||
687 	    p == curproc || !CANSIGNAL(p, info->sig)) {
688 		return (0);
689 	}
690 	++info->nfound;
691 	if (info->sig)
692 		ksignal(p, info->sig);
693 	return(0);
694 }
695 
696 /*
697  * Send a general signal to a process or LWPs within that process.  Note
698  * that new signals cannot be sent if a process is exiting.
699  */
700 int
701 kern_kill(int sig, pid_t pid, lwpid_t tid)
702 {
703 	struct thread *td = curthread;
704 	struct proc *p = td->td_proc;
705 	struct lwp *lp = NULL;
706 
707 	if ((u_int)sig > _SIG_MAXSIG)
708 		return (EINVAL);
709 	if (pid > 0) {
710 		/* kill single process */
711 		if ((p = pfind(pid)) == NULL)
712 			return (ESRCH);
713 		if (!CANSIGNAL(p, sig))
714 			return (EPERM);
715 
716 		/*
717 		 * NOP if the process is exiting.  Note that lwpsignal() is
718 		 * called directly with P_WEXIT set to kill individual LWPs
719 		 * during exit, which is allowed.
720 		 */
721 		if (p->p_flag & P_WEXIT)
722 			return (0);
723 		if (tid != -1) {
724 			lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, tid);
725 			if (lp == NULL)
726 				return (ESRCH);
727 		}
728 		if (sig)
729 			lwpsignal(p, lp, sig);
730 		return (0);
731 	}
732 	/*
733 	 * If we come here, pid is a special broadcast pid.
734 	 * This doesn't mix with a tid.
735 	 */
736 	if (tid != -1)
737 		return (EINVAL);
738 	switch (pid) {
739 	case -1:		/* broadcast signal */
740 		return (dokillpg(sig, 0, 1));
741 	case 0:			/* signal own process group */
742 		return (dokillpg(sig, 0, 0));
743 	default:		/* negative explicit process group */
744 		return (dokillpg(sig, -pid, 0));
745 	}
746 	/* NOTREACHED */
747 }
748 
749 int
750 sys_kill(struct kill_args *uap)
751 {
752 	int error;
753 
754 	error = kern_kill(uap->signum, uap->pid, -1);
755 	return (error);
756 }
757 
758 int
759 sys_lwp_kill(struct lwp_kill_args *uap)
760 {
761 	int error;
762 	pid_t pid = uap->pid;
763 
764 	/*
765 	 * A tid is mandatory for lwp_kill(), otherwise
766 	 * you could simply use kill().
767 	 */
768 	if (uap->tid == -1)
769 		return (EINVAL);
770 
771 	/*
772 	 * To save on a getpid() function call for intra-process
773 	 * signals, pid == -1 means current process.
774 	 */
775 	if (pid == -1)
776 		pid = curproc->p_pid;
777 
778 	error = kern_kill(uap->signum, pid, uap->tid);
779 	return (error);
780 }
781 
782 /*
783  * Send a signal to a process group.
784  */
785 void
786 gsignal(int pgid, int sig)
787 {
788 	struct pgrp *pgrp;
789 
790 	if (pgid && (pgrp = pgfind(pgid)))
791 		pgsignal(pgrp, sig, 0);
792 }
793 
794 /*
795  * Send a signal to a process group.  If checktty is 1,
796  * limit to members which have a controlling terminal.
797  *
798  * pg_lock interlocks against a fork that might be in progress, to
799  * ensure that the new child process picks up the signal.
800  */
801 void
802 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
803 {
804 	struct proc *p;
805 
806 	if (pgrp) {
807 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
808 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
809 			if (checkctty == 0 || p->p_flag & P_CONTROLT)
810 				ksignal(p, sig);
811 		}
812 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
813 	}
814 }
815 
816 /*
817  * Send a signal caused by a trap to the current lwp.  If it will be caught
818  * immediately, deliver it with correct code.  Otherwise, post it normally.
819  *
820  * These signals may ONLY be delivered to the specified lwp and may never
821  * be delivered to the process generically.
822  */
823 void
824 trapsignal(struct lwp *lp, int sig, u_long code)
825 {
826 	struct proc *p = lp->lwp_proc;
827 	struct sigacts *ps = p->p_sigacts;
828 
829 	/*
830 	 * If we are a virtual kernel running an emulated user process
831 	 * context, switch back to the virtual kernel context before
832 	 * trying to post the signal.
833 	 */
834 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
835 		struct trapframe *tf = lp->lwp_md.md_regs;
836 		tf->tf_trapno = 0;
837 		vkernel_trap(lp, tf);
838 	}
839 
840 
841 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(p->p_sigcatch, sig) &&
842 	    !SIGISMEMBER(lp->lwp_sigmask, sig)) {
843 		lp->lwp_ru.ru_nsignals++;
844 #ifdef KTRACE
845 		if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
846 			ktrpsig(p, sig, ps->ps_sigact[_SIG_IDX(sig)],
847 				&lp->lwp_sigmask, code);
848 #endif
849 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], sig,
850 						&lp->lwp_sigmask, code);
851 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
852 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
853 			SIGADDSET(lp->lwp_sigmask, sig);
854 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
855 			/*
856 			 * See kern_sigaction() for origin of this code.
857 			 */
858 			SIGDELSET(p->p_sigcatch, sig);
859 			if (sig != SIGCONT &&
860 			    sigprop(sig) & SA_IGNORE)
861 				SIGADDSET(p->p_sigignore, sig);
862 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
863 		}
864 	} else {
865 		lp->lwp_code = code;	/* XXX for core dump/debugger */
866 		lp->lwp_sig = sig;	/* XXX to verify code */
867 		lwpsignal(p, lp, sig);
868 	}
869 }
870 
871 /*
872  * Find a suitable lwp to deliver the signal to.
873  *
874  * Returns NULL if all lwps hold the signal blocked.
875  */
876 static struct lwp *
877 find_lwp_for_signal(struct proc *p, int sig)
878 {
879 	struct lwp *lp;
880 	struct lwp *run, *sleep, *stop;
881 
882 	/*
883 	 * If the running/preempted thread belongs to the proc to which
884 	 * the signal is being delivered and this thread does not block
885 	 * the signal, then we can avoid a context switch by delivering
886 	 * the signal to this thread, because it will return to userland
887 	 * soon anyways.
888 	 */
889 	lp = lwkt_preempted_proc();
890 	if (lp != NULL && lp->lwp_proc == p && !SIGISMEMBER(lp->lwp_sigmask, sig))
891 		return (lp);
892 
893 	run = sleep = stop = NULL;
894 	FOREACH_LWP_IN_PROC(lp, p) {
895 		/*
896 		 * If the signal is being blocked by the lwp, then this
897 		 * lwp is not eligible for receiving the signal.
898 		 */
899 		if (SIGISMEMBER(lp->lwp_sigmask, sig))
900 			continue;
901 
902 		switch (lp->lwp_stat) {
903 		case LSRUN:
904 			run = lp;
905 			break;
906 
907 		case LSSTOP:
908 			stop = lp;
909 			break;
910 
911 		case LSSLEEP:
912 			if (lp->lwp_flag & LWP_SINTR)
913 				sleep = lp;
914 			break;
915 		}
916 	}
917 
918 	if (run != NULL)
919 		return (run);
920 	else if (sleep != NULL)
921 		return (sleep);
922 	else
923 		return (stop);
924 }
925 
926 /*
927  * Send the signal to the process.  If the signal has an action, the action
928  * is usually performed by the target process rather than the caller; we add
929  * the signal to the set of pending signals for the process.
930  *
931  * Exceptions:
932  *   o When a stop signal is sent to a sleeping process that takes the
933  *     default action, the process is stopped without awakening it.
934  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
935  *     regardless of the signal action (eg, blocked or ignored).
936  *
937  * Other ignored signals are discarded immediately.
938  */
939 void
940 ksignal(struct proc *p, int sig)
941 {
942 	lwpsignal(p, NULL, sig);
943 }
944 
945 /*
946  * The core for ksignal.  lp may be NULL, then a suitable thread
947  * will be chosen.  If not, lp MUST be a member of p.
948  */
949 void
950 lwpsignal(struct proc *p, struct lwp *lp, int sig)
951 {
952 	int prop;
953 	sig_t action;
954 
955 	if (sig > _SIG_MAXSIG || sig <= 0) {
956 		kprintf("lwpsignal: signal %d\n", sig);
957 		panic("lwpsignal signal number");
958 	}
959 
960 	KKASSERT(lp == NULL || lp->lwp_proc == p);
961 
962 	crit_enter();
963 	KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
964 	crit_exit();
965 
966 	prop = sigprop(sig);
967 
968 	/*
969 	 * If proc is traced, always give parent a chance;
970 	 * if signal event is tracked by procfs, give *that*
971 	 * a chance, as well.
972 	 */
973 	if ((p->p_flag & P_TRACED) || (p->p_stops & S_SIG)) {
974 		action = SIG_DFL;
975 	} else {
976 		/*
977 		 * Do not try to deliver signals to an exiting lwp.  Note
978 		 * that we must still deliver the signal if P_WEXIT is set
979 		 * in the process flags.
980 		 */
981 		if (lp && (lp->lwp_flag & LWP_WEXIT))
982 			return;
983 
984 		/*
985 		 * Ig the signal is being ignored, then we forget about
986 		 * it immediately.  NOTE: We don't set SIGCONT in p_sigignore,
987 		 * and if it is set to SIG_IGN, action will be SIG_DFL here.
988 		 */
989 		if (SIGISMEMBER(p->p_sigignore, sig))
990 			return;
991 		if (SIGISMEMBER(p->p_sigcatch, sig))
992 			action = SIG_CATCH;
993 		else
994 			action = SIG_DFL;
995 	}
996 
997 	/*
998 	 * If continuing, clear any pending STOP signals.
999 	 */
1000 	if (prop & SA_CONT)
1001 		SIG_STOPSIGMASK(p->p_siglist);
1002 
1003 	if (prop & SA_STOP) {
1004 		/*
1005 		 * If sending a tty stop signal to a member of an orphaned
1006 		 * process group, discard the signal here if the action
1007 		 * is default; don't stop the process below if sleeping,
1008 		 * and don't clear any pending SIGCONT.
1009 		 */
1010 		if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0 &&
1011 		    action == SIG_DFL) {
1012 		        return;
1013 		}
1014 		SIG_CONTSIGMASK(p->p_siglist);
1015 		p->p_flag &= ~P_CONTINUED;
1016 	}
1017 
1018 	crit_enter();
1019 
1020 	if (p->p_stat == SSTOP) {
1021 		/*
1022 		 * Nobody can handle this signal, add it to the lwp or
1023 		 * process pending list
1024 		 */
1025 		if (lp)
1026 			SIGADDSET(lp->lwp_siglist, sig);
1027 		else
1028 			SIGADDSET(p->p_siglist, sig);
1029 
1030 		/*
1031 		 * If the process is stopped and is being traced, then no
1032 		 * further action is necessary.
1033 		 */
1034 		if (p->p_flag & P_TRACED)
1035 			goto out;
1036 
1037 		/*
1038 		 * If the process is stopped and receives a KILL signal,
1039 		 * make the process runnable.
1040 		 */
1041 		if (sig == SIGKILL) {
1042 			proc_unstop(p);
1043 			goto active_process;
1044 		}
1045 
1046 		/*
1047 		 * If the process is stopped and receives a CONT signal,
1048 		 * then try to make the process runnable again.
1049 		 */
1050 		if (prop & SA_CONT) {
1051 			/*
1052 			 * If SIGCONT is default (or ignored), we continue the
1053 			 * process but don't leave the signal in p_siglist, as
1054 			 * it has no further action.  If SIGCONT is held, we
1055 			 * continue the process and leave the signal in
1056 			 * p_siglist.  If the process catches SIGCONT, let it
1057 			 * handle the signal itself.
1058 			 */
1059 			/* XXX what if the signal is being held blocked? */
1060 			p->p_flag |= P_CONTINUED;
1061 			wakeup(p->p_pptr);
1062 			if (action == SIG_DFL)
1063 				SIGDELSET(p->p_siglist, sig);
1064 			proc_unstop(p);
1065 			if (action == SIG_CATCH)
1066 				goto active_process;
1067 			goto out;
1068 		}
1069 
1070 		/*
1071 		 * If the process is stopped and receives another STOP
1072 		 * signal, we do not need to stop it again.  If we did
1073 		 * the shell could get confused.
1074 		 *
1075 		 * However, if the current/preempted lwp is part of the
1076 		 * process receiving the signal, we need to keep it,
1077 		 * so that this lwp can stop in issignal() later, as
1078 		 * we don't want to wait until it reaches userret!
1079 		 */
1080 		if (prop & SA_STOP) {
1081 			if (lwkt_preempted_proc() == NULL ||
1082 			    lwkt_preempted_proc()->lwp_proc != p)
1083 				SIGDELSET(p->p_siglist, sig);
1084 		}
1085 
1086 		/*
1087 		 * Otherwise the process is stopped and it received some
1088 		 * signal, which does not change its stopped state.
1089 		 *
1090 		 * We have to select one thread to set LWP_BREAKTSLEEP,
1091 		 * so that the current signal will break the sleep
1092 		 * as soon as a SA_CONT signal will unstop the process.
1093 		 */
1094 		if (lp == NULL)
1095 			lp = find_lwp_for_signal(p, sig);
1096 		if (lp != NULL &&
1097 		    (lp->lwp_stat == LSSLEEP || lp->lwp_stat == LSSTOP))
1098 			lp->lwp_flag |= LWP_BREAKTSLEEP;
1099 		goto out;
1100 
1101 		/* NOTREACHED */
1102 	}
1103 	/* else not stopped */
1104 active_process:
1105 
1106 	/*
1107 	 * Never deliver a lwp-specific signal to a random lwp.
1108 	 */
1109 	if (lp == NULL) {
1110 		lp = find_lwp_for_signal(p, sig);
1111 		if (lp && SIGISMEMBER(lp->lwp_sigmask, sig))
1112 			lp = NULL;
1113 	}
1114 
1115 	/*
1116 	 * Deliver to the process generically if (1) the signal is being
1117 	 * sent to any thread or (2) we could not find a thread to deliver
1118 	 * it to.
1119 	 */
1120 	if (lp == NULL) {
1121 		SIGADDSET(p->p_siglist, sig);
1122 		goto out;
1123 	}
1124 
1125 	/*
1126 	 * Deliver to a specific LWP whether it masks it or not.  It will
1127 	 * not be dispatched if masked but we must still deliver it.
1128 	 */
1129 	if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
1130 	    (p->p_flag & P_TRACED) == 0) {
1131 		p->p_nice = NZERO;
1132 	}
1133 
1134 	/*
1135 	 * If the process receives a STOP signal which indeed needs to
1136 	 * stop the process, do so.  If the process chose to catch the
1137 	 * signal, it will be treated like any other signal.
1138 	 */
1139 	if ((prop & SA_STOP) && action == SIG_DFL) {
1140 		/*
1141 		 * If a child holding parent blocked, stopping
1142 		 * could cause deadlock.  Take no action at this
1143 		 * time.
1144 		 */
1145 		if (p->p_flag & P_PPWAIT) {
1146 			SIGADDSET(p->p_siglist, sig);
1147 			goto out;
1148 		}
1149 
1150 		/*
1151 		 * Do not actually try to manipulate the process, but simply
1152 		 * stop it.  Lwps will stop as soon as they safely can.
1153 		 */
1154 		p->p_xstat = sig;
1155 		proc_stop(p);
1156 		goto out;
1157 	}
1158 
1159 	/*
1160 	 * If it is a CONT signal with default action, just ignore it.
1161 	 */
1162 	if ((prop & SA_CONT) && action == SIG_DFL)
1163 		goto out;
1164 
1165 	/*
1166 	 * Mark signal pending at this specific thread.
1167 	 */
1168 	SIGADDSET(lp->lwp_siglist, sig);
1169 
1170 	lwp_signotify(lp);
1171 
1172 out:
1173 	crit_exit();
1174 }
1175 
1176 void
1177 lwp_signotify(struct lwp *lp)
1178 {
1179 	crit_enter();
1180 	if (lp->lwp_stat == LSSLEEP || lp->lwp_stat == LSSTOP) {
1181 		/*
1182 		 * Thread is in tsleep.
1183 		 */
1184 
1185 		/*
1186 		 * If the thread is sleeping uninterruptibly
1187 		 * we can't interrupt the sleep... the signal will
1188 		 * be noticed when the lwp returns through
1189 		 * trap() or syscall().
1190 		 *
1191 		 * Otherwise the signal can interrupt the sleep.
1192 		 *
1193 		 * If the process is traced, the lwp will handle the
1194 		 * tracing in issignal() when it returns to userland.
1195 		 */
1196 		if (lp->lwp_flag & LWP_SINTR) {
1197 			/*
1198 			 * Make runnable and break out of any tsleep as well.
1199 			 */
1200 			lp->lwp_flag |= LWP_BREAKTSLEEP;
1201 			setrunnable(lp);
1202 		}
1203 	} else {
1204 		/*
1205 		 * Otherwise the thread is running
1206 		 *
1207 		 * LSRUN does nothing with the signal, other than kicking
1208 		 * ourselves if we are running.
1209 		 * SZOMB and SIDL mean that it will either never be noticed,
1210 		 * or noticed very soon.
1211 		 *
1212 		 * Note that lwp_thread may be NULL or may not be completely
1213 		 * initialized if the process is in the SIDL or SZOMB state.
1214 		 *
1215 		 * For SMP we may have to forward the request to another cpu.
1216 		 * YYY the MP lock prevents the target process from moving
1217 		 * to another cpu, see kern/kern_switch.c
1218 		 *
1219 		 * If the target thread is waiting on its message port,
1220 		 * wakeup the target thread so it can check (or ignore)
1221 		 * the new signal.  YYY needs cleanup.
1222 		 */
1223 		if (lp == lwkt_preempted_proc()) {
1224 			signotify();
1225 		} else if (lp->lwp_stat == LSRUN) {
1226 			struct thread *td = lp->lwp_thread;
1227 			struct proc *p = lp->lwp_proc;
1228 
1229 			KASSERT(td != NULL,
1230 			    ("pid %d/%d NULL lwp_thread stat %d flags %08x/%08x",
1231 			    p->p_pid, lp->lwp_tid, lp->lwp_stat,
1232 			    p->p_flag, lp->lwp_flag));
1233 
1234 			/*
1235 			 * To prevent a MP race with TDF_SINTR we must
1236 			 * schedule the thread on the correct cpu.
1237 			 */
1238 #ifdef SMP
1239 			if (td->td_gd != mycpu)
1240 				lwkt_send_ipiq(td->td_gd, signotify_remote, lp);
1241 			else
1242 #endif
1243 			if (td->td_flags & TDF_SINTR)
1244 				lwkt_schedule(td);
1245 		}
1246 	}
1247 	crit_exit();
1248 }
1249 
1250 #ifdef SMP
1251 
1252 /*
1253  * This function is called via an IPI.  We will be in a critical section but
1254  * the MP lock will NOT be held.  Also note that by the time the ipi message
1255  * gets to us the process 'p' (arg) may no longer be scheduled or even valid.
1256  */
1257 static void
1258 signotify_remote(void *arg)
1259 {
1260 	struct lwp *lp = arg;
1261 
1262 	if (lp == lwkt_preempted_proc()) {
1263 		signotify();
1264 	} else {
1265 		struct thread *td = lp->lwp_thread;
1266 		if (td->td_flags & TDF_SINTR)
1267 			lwkt_schedule(td);
1268 	}
1269 }
1270 
1271 #endif
1272 
1273 void
1274 proc_stop(struct proc *p)
1275 {
1276 	struct lwp *lp;
1277 
1278 	/* If somebody raced us, be happy with it */
1279 	if (p->p_stat == SSTOP)
1280 		return;
1281 
1282 	crit_enter();
1283 	p->p_stat = SSTOP;
1284 
1285 	FOREACH_LWP_IN_PROC(lp, p) {
1286 		switch (lp->lwp_stat) {
1287 		case LSSTOP:
1288 			/*
1289 			 * Do nothing, we are already counted in
1290 			 * p_nstopped.
1291 			 */
1292 			break;
1293 
1294 		case LSSLEEP:
1295 			/*
1296 			 * We're sleeping, but we will stop before
1297 			 * returning to userspace, so count us
1298 			 * as stopped as well.  We set LWP_WSTOP
1299 			 * to signal the lwp that it should not
1300 			 * increase p_nstopped when reaching tstop().
1301 			 */
1302 			if ((lp->lwp_flag & LWP_WSTOP) == 0) {
1303 				lp->lwp_flag |= LWP_WSTOP;
1304 				++p->p_nstopped;
1305 			}
1306 			break;
1307 
1308 		case LSRUN:
1309 			/*
1310 			 * We might notify ourself, but that's not
1311 			 * a problem.
1312 			 */
1313 			lwp_signotify(lp);
1314 			break;
1315 		}
1316 	}
1317 
1318 	if (p->p_nstopped == p->p_nthreads) {
1319 		p->p_flag &= ~P_WAITED;
1320 		wakeup(p->p_pptr);
1321 		if ((p->p_pptr->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1322 			ksignal(p->p_pptr, SIGCHLD);
1323 	}
1324 	crit_exit();
1325 }
1326 
1327 void
1328 proc_unstop(struct proc *p)
1329 {
1330 	struct lwp *lp;
1331 
1332 	if (p->p_stat != SSTOP)
1333 		return;
1334 
1335 	crit_enter();
1336 	p->p_stat = SACTIVE;
1337 
1338 	FOREACH_LWP_IN_PROC(lp, p) {
1339 		switch (lp->lwp_stat) {
1340 		case LSRUN:
1341 			/*
1342 			 * Uh?  Not stopped?  Well, I guess that's okay.
1343 			 */
1344 			if (bootverbose)
1345 				kprintf("proc_unstop: lwp %d/%d not sleeping\n",
1346 					p->p_pid, lp->lwp_tid);
1347 			break;
1348 
1349 		case LSSLEEP:
1350 			/*
1351 			 * Still sleeping.  Don't bother waking it up.
1352 			 * However, if this thread was counted as
1353 			 * stopped, undo this.
1354 			 *
1355 			 * Nevertheless we call setrunnable() so that it
1356 			 * will wake up in case a signal or timeout arrived
1357 			 * in the meantime.
1358 			 */
1359 			if (lp->lwp_flag & LWP_WSTOP) {
1360 				--p->p_nstopped;
1361 			} else {
1362 				if (bootverbose)
1363 					kprintf("proc_unstop: lwp %d/%d sleeping, not stopped\n",
1364 						p->p_pid, lp->lwp_tid);
1365 			}
1366 			/* FALLTHROUGH */
1367 
1368 		case LSSTOP:
1369 			setrunnable(lp);
1370 			break;
1371 
1372 		}
1373 		lp->lwp_flag &= ~LWP_WSTOP;
1374 	}
1375 	crit_exit();
1376 }
1377 
1378 static int
1379 kern_sigtimedwait(sigset_t waitset, siginfo_t *info, struct timespec *timeout)
1380 {
1381 	sigset_t savedmask, set;
1382 	struct proc *p = curproc;
1383 	struct lwp *lp = curthread->td_lwp;
1384 	int error, sig, hz, timevalid = 0;
1385 	struct timespec rts, ets, ts;
1386 	struct timeval tv;
1387 
1388 	error = 0;
1389 	sig = 0;
1390 	SIG_CANTMASK(waitset);
1391 	savedmask = lp->lwp_sigmask;
1392 
1393 	if (timeout) {
1394 		if (timeout->tv_sec >= 0 && timeout->tv_nsec >= 0 &&
1395 		    timeout->tv_nsec < 1000000000) {
1396 			timevalid = 1;
1397 			getnanouptime(&rts);
1398 		 	ets = rts;
1399 			timespecadd(&ets, timeout);
1400 		}
1401 	}
1402 
1403 	for (;;) {
1404 		set = lwp_sigpend(lp);
1405 		SIGSETAND(set, waitset);
1406 		if ((sig = sig_ffs(&set)) != 0) {
1407 			SIGFILLSET(lp->lwp_sigmask);
1408 			SIGDELSET(lp->lwp_sigmask, sig);
1409 			SIG_CANTMASK(lp->lwp_sigmask);
1410 			sig = issignal(lp);
1411 			/*
1412 			 * It may be a STOP signal, in the case, issignal
1413 			 * returns 0, because we may stop there, and new
1414 			 * signal can come in, we should restart if we got
1415 			 * nothing.
1416 			 */
1417 			if (sig == 0)
1418 				continue;
1419 			else
1420 				break;
1421 		}
1422 
1423 		/*
1424 		 * Previous checking got nothing, and we retried but still
1425 		 * got nothing, we should return the error status.
1426 		 */
1427 		if (error)
1428 			break;
1429 
1430 		/*
1431 		 * POSIX says this must be checked after looking for pending
1432 		 * signals.
1433 		 */
1434 		if (timeout) {
1435 			if (!timevalid) {
1436 				error = EINVAL;
1437 				break;
1438 			}
1439 			getnanouptime(&rts);
1440 			if (timespeccmp(&rts, &ets, >=)) {
1441 				error = EAGAIN;
1442 				break;
1443 			}
1444 			ts = ets;
1445 			timespecsub(&ts, &rts);
1446 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1447 			hz = tvtohz_high(&tv);
1448 		} else
1449 			hz = 0;
1450 
1451 		lp->lwp_sigmask = savedmask;
1452 		SIGSETNAND(lp->lwp_sigmask, waitset);
1453 		/*
1454 		 * We won't ever be woken up.  Instead, our sleep will
1455 		 * be broken in lwpsignal().
1456 		 */
1457 		error = tsleep(&p->p_sigacts, PCATCH, "sigwt", hz);
1458 		if (timeout) {
1459 			if (error == ERESTART) {
1460 				/* can not restart a timeout wait. */
1461 				error = EINTR;
1462 			} else if (error == EAGAIN) {
1463 				/* will calculate timeout by ourself. */
1464 				error = 0;
1465 			}
1466 		}
1467 		/* Retry ... */
1468 	}
1469 
1470 	lp->lwp_sigmask = savedmask;
1471 	if (sig) {
1472 		error = 0;
1473 		bzero(info, sizeof(*info));
1474 		info->si_signo = sig;
1475 		lwp_delsig(lp, sig);	/* take the signal! */
1476 
1477 		if (sig == SIGKILL)
1478 			sigexit(p, sig);
1479 	}
1480 	return (error);
1481 }
1482 
1483 int
1484 sys_sigtimedwait(struct sigtimedwait_args *uap)
1485 {
1486 	struct timespec ts;
1487 	struct timespec *timeout;
1488 	sigset_t set;
1489 	siginfo_t info;
1490 	int error;
1491 
1492 	if (uap->timeout) {
1493 		error = copyin(uap->timeout, &ts, sizeof(ts));
1494 		if (error)
1495 			return (error);
1496 		timeout = &ts;
1497 	} else {
1498 		timeout = NULL;
1499 	}
1500 	error = copyin(uap->set, &set, sizeof(set));
1501 	if (error)
1502 		return (error);
1503 	error = kern_sigtimedwait(set, &info, timeout);
1504 	if (error)
1505 		return (error);
1506  	if (uap->info)
1507 		error = copyout(&info, uap->info, sizeof(info));
1508 	/* Repost if we got an error. */
1509 	/*
1510 	 * XXX lwp
1511 	 *
1512 	 * This could transform a thread-specific signal to another
1513 	 * thread / process pending signal.
1514 	 */
1515 	if (error)
1516 		ksignal(curproc, info.si_signo);
1517 	else
1518 		uap->sysmsg_result = info.si_signo;
1519 	return (error);
1520 }
1521 
1522 int
1523 sys_sigwaitinfo(struct sigwaitinfo_args *uap)
1524 {
1525 	siginfo_t info;
1526 	sigset_t set;
1527 	int error;
1528 
1529 	error = copyin(uap->set, &set, sizeof(set));
1530 	if (error)
1531 		return (error);
1532 	error = kern_sigtimedwait(set, &info, NULL);
1533 	if (error)
1534 		return (error);
1535 	if (uap->info)
1536 		error = copyout(&info, uap->info, sizeof(info));
1537 	/* Repost if we got an error. */
1538 	/*
1539 	 * XXX lwp
1540 	 *
1541 	 * This could transform a thread-specific signal to another
1542 	 * thread / process pending signal.
1543 	 */
1544 	if (error)
1545 		ksignal(curproc, info.si_signo);
1546 	else
1547 		uap->sysmsg_result = info.si_signo;
1548 	return (error);
1549 }
1550 
1551 /*
1552  * If the current process has received a signal that would interrupt a
1553  * system call, return EINTR or ERESTART as appropriate.
1554  */
1555 int
1556 iscaught(struct lwp *lp)
1557 {
1558 	struct proc *p = lp->lwp_proc;
1559 	int sig;
1560 
1561 	if (p) {
1562 		if ((sig = CURSIG(lp)) != 0) {
1563 			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1564 				return (EINTR);
1565 			return (ERESTART);
1566 		}
1567 	}
1568 	return(EWOULDBLOCK);
1569 }
1570 
1571 /*
1572  * If the current process has received a signal (should be caught or cause
1573  * termination, should interrupt current syscall), return the signal number.
1574  * Stop signals with default action are processed immediately, then cleared;
1575  * they aren't returned.  This is checked after each entry to the system for
1576  * a syscall or trap (though this can usually be done without calling issignal
1577  * by checking the pending signal masks in the CURSIG macro.) The normal call
1578  * sequence is
1579  *
1580  * This routine is called via CURSIG/__cursig and the MP lock might not be
1581  * held.  Obtain the MP lock for the duration of the operation.
1582  *
1583  *	while (sig = CURSIG(curproc))
1584  *		postsig(sig);
1585  */
1586 int
1587 issignal(struct lwp *lp)
1588 {
1589 	struct proc *p = lp->lwp_proc;
1590 	sigset_t mask;
1591 	int sig, prop;
1592 
1593 	get_mplock();
1594 	for (;;) {
1595 		int traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
1596 
1597 		mask = lwp_sigpend(lp);
1598 		SIGSETNAND(mask, lp->lwp_sigmask);
1599 		if (p->p_flag & P_PPWAIT)
1600 			SIG_STOPSIGMASK(mask);
1601 		if (SIGISEMPTY(mask)) {		/* no signal to send */
1602 			rel_mplock();
1603 			return (0);
1604 		}
1605 		sig = sig_ffs(&mask);
1606 
1607 		STOPEVENT(p, S_SIG, sig);
1608 
1609 		/*
1610 		 * We should see pending but ignored signals
1611 		 * only if P_TRACED was on when they were posted.
1612 		 */
1613 		if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
1614 			lwp_delsig(lp, sig);
1615 			continue;
1616 		}
1617 		if ((p->p_flag & P_TRACED) && (p->p_flag & P_PPWAIT) == 0) {
1618 			/*
1619 			 * If traced, always stop, and stay stopped until
1620 			 * released by the parent.
1621 			 *
1622 			 * NOTE: SSTOP may get cleared during the loop,
1623 			 * but we do not re-notify the parent if we have
1624 			 * to loop several times waiting for the parent
1625 			 * to let us continue.
1626 			 *
1627 			 * XXX not sure if this is still true
1628 			 */
1629 			p->p_xstat = sig;
1630 			proc_stop(p);
1631 			do {
1632 				tstop();
1633 			} while (!trace_req(p) && (p->p_flag & P_TRACED));
1634 
1635 			/*
1636 			 * If parent wants us to take the signal,
1637 			 * then it will leave it in p->p_xstat;
1638 			 * otherwise we just look for signals again.
1639 			 */
1640 			lwp_delsig(lp, sig);	/* clear old signal */
1641 			sig = p->p_xstat;
1642 			if (sig == 0)
1643 				continue;
1644 
1645 			/*
1646 			 * Put the new signal into p_siglist.  If the
1647 			 * signal is being masked, look for other signals.
1648 			 *
1649 			 * XXX lwp might need a call to ksignal()
1650 			 */
1651 			SIGADDSET(p->p_siglist, sig);
1652 			if (SIGISMEMBER(lp->lwp_sigmask, sig))
1653 				continue;
1654 
1655 			/*
1656 			 * If the traced bit got turned off, go back up
1657 			 * to the top to rescan signals.  This ensures
1658 			 * that p_sig* and ps_sigact are consistent.
1659 			 */
1660 			if ((p->p_flag & P_TRACED) == 0)
1661 				continue;
1662 		}
1663 
1664 		prop = sigprop(sig);
1665 
1666 		/*
1667 		 * Decide whether the signal should be returned.
1668 		 * Return the signal's number, or fall through
1669 		 * to clear it from the pending mask.
1670 		 */
1671 		switch ((int)(intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
1672 		case (int)SIG_DFL:
1673 			/*
1674 			 * Don't take default actions on system processes.
1675 			 */
1676 			if (p->p_pid <= 1) {
1677 #ifdef DIAGNOSTIC
1678 				/*
1679 				 * Are you sure you want to ignore SIGSEGV
1680 				 * in init? XXX
1681 				 */
1682 				kprintf("Process (pid %lu) got signal %d\n",
1683 					(u_long)p->p_pid, sig);
1684 #endif
1685 				break;		/* == ignore */
1686 			}
1687 
1688 			/*
1689 			 * Handle the in-kernel checkpoint action
1690 			 */
1691 			if (prop & SA_CKPT) {
1692 				checkpoint_signal_handler(lp);
1693 				break;
1694 			}
1695 
1696 			/*
1697 			 * If there is a pending stop signal to process
1698 			 * with default action, stop here,
1699 			 * then clear the signal.  However,
1700 			 * if process is member of an orphaned
1701 			 * process group, ignore tty stop signals.
1702 			 */
1703 			if (prop & SA_STOP) {
1704 				if (p->p_flag & P_TRACED ||
1705 		    		    (p->p_pgrp->pg_jobc == 0 &&
1706 				    prop & SA_TTYSTOP))
1707 					break;	/* == ignore */
1708 				p->p_xstat = sig;
1709 				proc_stop(p);
1710 				while (p->p_stat == SSTOP) {
1711 					tstop();
1712 				}
1713 				break;
1714 			} else if (prop & SA_IGNORE) {
1715 				/*
1716 				 * Except for SIGCONT, shouldn't get here.
1717 				 * Default action is to ignore; drop it.
1718 				 */
1719 				break;		/* == ignore */
1720 			} else {
1721 				rel_mplock();
1722 				return (sig);
1723 			}
1724 
1725 			/*NOTREACHED*/
1726 
1727 		case (int)SIG_IGN:
1728 			/*
1729 			 * Masking above should prevent us ever trying
1730 			 * to take action on an ignored signal other
1731 			 * than SIGCONT, unless process is traced.
1732 			 */
1733 			if ((prop & SA_CONT) == 0 &&
1734 			    (p->p_flag & P_TRACED) == 0)
1735 				kprintf("issignal\n");
1736 			break;		/* == ignore */
1737 
1738 		default:
1739 			/*
1740 			 * This signal has an action, let
1741 			 * postsig() process it.
1742 			 */
1743 			rel_mplock();
1744 			return (sig);
1745 		}
1746 		lwp_delsig(lp, sig);		/* take the signal! */
1747 	}
1748 	/* NOTREACHED */
1749 }
1750 
1751 /*
1752  * Take the action for the specified signal
1753  * from the current set of pending signals.
1754  */
1755 void
1756 postsig(int sig)
1757 {
1758 	struct lwp *lp = curthread->td_lwp;
1759 	struct proc *p = lp->lwp_proc;
1760 	struct sigacts *ps = p->p_sigacts;
1761 	sig_t action;
1762 	sigset_t returnmask;
1763 	int code;
1764 
1765 	KASSERT(sig != 0, ("postsig"));
1766 
1767 	/*
1768 	 * If we are a virtual kernel running an emulated user process
1769 	 * context, switch back to the virtual kernel context before
1770 	 * trying to post the signal.
1771 	 */
1772 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1773 		struct trapframe *tf = lp->lwp_md.md_regs;
1774 		tf->tf_trapno = 0;
1775 		vkernel_trap(lp, tf);
1776 	}
1777 
1778 	lwp_delsig(lp, sig);
1779 	action = ps->ps_sigact[_SIG_IDX(sig)];
1780 #ifdef KTRACE
1781 	if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
1782 		ktrpsig(p, sig, action, lp->lwp_flag & LWP_OLDMASK ?
1783 			&lp->lwp_oldsigmask : &lp->lwp_sigmask, 0);
1784 #endif
1785 	STOPEVENT(p, S_SIG, sig);
1786 
1787 	if (action == SIG_DFL) {
1788 		/*
1789 		 * Default action, where the default is to kill
1790 		 * the process.  (Other cases were ignored above.)
1791 		 */
1792 		sigexit(p, sig);
1793 		/* NOTREACHED */
1794 	} else {
1795 		/*
1796 		 * If we get here, the signal must be caught.
1797 		 */
1798 		KASSERT(action != SIG_IGN && !SIGISMEMBER(lp->lwp_sigmask, sig),
1799 		    ("postsig action"));
1800 
1801 		crit_enter();
1802 
1803 		/*
1804 		 * Reset the signal handler if asked to
1805 		 */
1806 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
1807 			/*
1808 			 * See kern_sigaction() for origin of this code.
1809 			 */
1810 			SIGDELSET(p->p_sigcatch, sig);
1811 			if (sig != SIGCONT &&
1812 			    sigprop(sig) & SA_IGNORE)
1813 				SIGADDSET(p->p_sigignore, sig);
1814 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1815 		}
1816 
1817 		/*
1818 		 * Handle the mailbox case.  Copyout to the appropriate
1819 		 * location but do not generate a signal frame.  The system
1820 		 * call simply returns EINTR and the user is responsible for
1821 		 * polling the mailbox.
1822 		 */
1823 		if (SIGISMEMBER(ps->ps_sigmailbox, sig)) {
1824 			int sig_copy = sig;
1825 			copyout(&sig_copy, (void *)action, sizeof(int));
1826 			curproc->p_flag |= P_MAILBOX;
1827 			crit_exit();
1828 			goto done;
1829 		}
1830 
1831 		/*
1832 		 * Set the signal mask and calculate the mask to restore
1833 		 * when the signal function returns.
1834 		 *
1835 		 * Special case: user has done a sigsuspend.  Here the
1836 		 * current mask is not of interest, but rather the
1837 		 * mask from before the sigsuspend is what we want
1838 		 * restored after the signal processing is completed.
1839 		 */
1840 		if (lp->lwp_flag & LWP_OLDMASK) {
1841 			returnmask = lp->lwp_oldsigmask;
1842 			lp->lwp_flag &= ~LWP_OLDMASK;
1843 		} else {
1844 			returnmask = lp->lwp_sigmask;
1845 		}
1846 
1847 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
1848 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
1849 			SIGADDSET(lp->lwp_sigmask, sig);
1850 
1851 		crit_exit();
1852 		lp->lwp_ru.ru_nsignals++;
1853 		if (lp->lwp_sig != sig) {
1854 			code = 0;
1855 		} else {
1856 			code = lp->lwp_code;
1857 			lp->lwp_code = 0;
1858 			lp->lwp_sig = 0;
1859 		}
1860 		(*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
1861 	}
1862 done:
1863 	;
1864 }
1865 
1866 /*
1867  * Kill the current process for stated reason.
1868  */
1869 void
1870 killproc(struct proc *p, char *why)
1871 {
1872 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n",
1873 		p->p_pid, p->p_comm,
1874 		p->p_ucred ? p->p_ucred->cr_uid : -1, why);
1875 	ksignal(p, SIGKILL);
1876 }
1877 
1878 /*
1879  * Force the current process to exit with the specified signal, dumping core
1880  * if appropriate.  We bypass the normal tests for masked and caught signals,
1881  * allowing unrecoverable failures to terminate the process without changing
1882  * signal state.  Mark the accounting record with the signal termination.
1883  * If dumping core, save the signal number for the debugger.  Calls exit and
1884  * does not return.
1885  */
1886 void
1887 sigexit(struct proc *p, int sig)
1888 {
1889 	struct lwp *lp = FIRST_LWP_IN_PROC(p);	/* XXX lwp */
1890 
1891 	p->p_acflag |= AXSIG;
1892 	if (sigprop(sig) & SA_CORE) {
1893 		lp->lwp_sig = sig;
1894 		/*
1895 		 * Log signals which would cause core dumps
1896 		 * (Log as LOG_INFO to appease those who don't want
1897 		 * these messages.)
1898 		 * XXX : Todo, as well as euid, write out ruid too
1899 		 */
1900 		if (coredump(lp, sig) == 0)
1901 			sig |= WCOREFLAG;
1902 		if (kern_logsigexit)
1903 			log(LOG_INFO,
1904 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
1905 			    p->p_pid, p->p_comm,
1906 			    p->p_ucred ? p->p_ucred->cr_uid : -1,
1907 			    sig &~ WCOREFLAG,
1908 			    sig & WCOREFLAG ? " (core dumped)" : "");
1909 	}
1910 	exit1(W_EXITCODE(0, sig));
1911 	/* NOTREACHED */
1912 }
1913 
1914 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
1915 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
1916 	      sizeof(corefilename), "process corefile name format string");
1917 
1918 /*
1919  * expand_name(name, uid, pid)
1920  * Expand the name described in corefilename, using name, uid, and pid.
1921  * corefilename is a kprintf-like string, with three format specifiers:
1922  *	%N	name of process ("name")
1923  *	%P	process id (pid)
1924  *	%U	user id (uid)
1925  * For example, "%N.core" is the default; they can be disabled completely
1926  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
1927  * This is controlled by the sysctl variable kern.corefile (see above).
1928  */
1929 
1930 static char *
1931 expand_name(const char *name, uid_t uid, pid_t pid)
1932 {
1933 	char *temp;
1934 	char buf[11];		/* Buffer for pid/uid -- max 4B */
1935 	int i, n;
1936 	char *format = corefilename;
1937 	size_t namelen;
1938 
1939 	temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
1940 	if (temp == NULL)
1941 		return NULL;
1942 	namelen = strlen(name);
1943 	for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
1944 		int l;
1945 		switch (format[i]) {
1946 		case '%':	/* Format character */
1947 			i++;
1948 			switch (format[i]) {
1949 			case '%':
1950 				temp[n++] = '%';
1951 				break;
1952 			case 'N':	/* process name */
1953 				if ((n + namelen) > MAXPATHLEN) {
1954 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
1955 					    pid, name, uid, temp, name);
1956 					kfree(temp, M_TEMP);
1957 					return NULL;
1958 				}
1959 				memcpy(temp+n, name, namelen);
1960 				n += namelen;
1961 				break;
1962 			case 'P':	/* process id */
1963 				l = ksprintf(buf, "%u", pid);
1964 				if ((n + l) > MAXPATHLEN) {
1965 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
1966 					    pid, name, uid, temp, name);
1967 					kfree(temp, M_TEMP);
1968 					return NULL;
1969 				}
1970 				memcpy(temp+n, buf, l);
1971 				n += l;
1972 				break;
1973 			case 'U':	/* user id */
1974 				l = ksprintf(buf, "%u", uid);
1975 				if ((n + l) > MAXPATHLEN) {
1976 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
1977 					    pid, name, uid, temp, name);
1978 					kfree(temp, M_TEMP);
1979 					return NULL;
1980 				}
1981 				memcpy(temp+n, buf, l);
1982 				n += l;
1983 				break;
1984 			default:
1985 			  	log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
1986 			}
1987 			break;
1988 		default:
1989 			temp[n++] = format[i];
1990 		}
1991 	}
1992 	temp[n] = '\0';
1993 	return temp;
1994 }
1995 
1996 /*
1997  * Dump a process' core.  The main routine does some
1998  * policy checking, and creates the name of the coredump;
1999  * then it passes on a vnode and a size limit to the process-specific
2000  * coredump routine if there is one; if there _is not_ one, it returns
2001  * ENOSYS; otherwise it returns the error from the process-specific routine.
2002  *
2003  * The parameter `lp' is the lwp which triggered the coredump.
2004  */
2005 
2006 static int
2007 coredump(struct lwp *lp, int sig)
2008 {
2009 	struct proc *p = lp->lwp_proc;
2010 	struct vnode *vp;
2011 	struct ucred *cred = p->p_ucred;
2012 	struct flock lf;
2013 	struct nlookupdata nd;
2014 	struct vattr vattr;
2015 	int error, error1;
2016 	char *name;			/* name of corefile */
2017 	off_t limit;
2018 
2019 	STOPEVENT(p, S_CORE, 0);
2020 
2021 	if (((sugid_coredump == 0) && p->p_flag & P_SUGID) || do_coredump == 0)
2022 		return (EFAULT);
2023 
2024 	/*
2025 	 * Note that the bulk of limit checking is done after
2026 	 * the corefile is created.  The exception is if the limit
2027 	 * for corefiles is 0, in which case we don't bother
2028 	 * creating the corefile at all.  This layout means that
2029 	 * a corefile is truncated instead of not being created,
2030 	 * if it is larger than the limit.
2031 	 */
2032 	limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
2033 	if (limit == 0)
2034 		return EFBIG;
2035 
2036 	name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
2037 	if (name == NULL)
2038 		return (EINVAL);
2039 	error = nlookup_init(&nd, name, UIO_SYSSPACE, NLC_LOCKVP);
2040 	if (error == 0)
2041 		error = vn_open(&nd, NULL, O_CREAT | FWRITE | O_NOFOLLOW, S_IRUSR | S_IWUSR);
2042 	kfree(name, M_TEMP);
2043 	if (error) {
2044 		nlookup_done(&nd);
2045 		return (error);
2046 	}
2047 	vp = nd.nl_open_vp;
2048 	nd.nl_open_vp = NULL;
2049 	nlookup_done(&nd);
2050 
2051 	vn_unlock(vp);
2052 	lf.l_whence = SEEK_SET;
2053 	lf.l_start = 0;
2054 	lf.l_len = 0;
2055 	lf.l_type = F_WRLCK;
2056 	error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, 0);
2057 	if (error)
2058 		goto out2;
2059 
2060 	/* Don't dump to non-regular files or files with links. */
2061 	if (vp->v_type != VREG ||
2062 	    VOP_GETATTR(vp, &vattr) || vattr.va_nlink != 1) {
2063 		error = EFAULT;
2064 		goto out1;
2065 	}
2066 
2067 	VATTR_NULL(&vattr);
2068 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2069 	vattr.va_size = 0;
2070 	VOP_SETATTR(vp, &vattr, cred);
2071 	p->p_acflag |= ACORE;
2072 	vn_unlock(vp);
2073 
2074 	error = p->p_sysent->sv_coredump ?
2075 		  p->p_sysent->sv_coredump(lp, sig, vp, limit) : ENOSYS;
2076 
2077 out1:
2078 	lf.l_type = F_UNLCK;
2079 	VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, 0);
2080 out2:
2081 	error1 = vn_close(vp, FWRITE);
2082 	if (error == 0)
2083 		error = error1;
2084 	return (error);
2085 }
2086 
2087 /*
2088  * Nonexistent system call-- signal process (may want to handle it).
2089  * Flag error in case process won't see signal immediately (blocked or ignored).
2090  */
2091 /* ARGSUSED */
2092 int
2093 sys_nosys(struct nosys_args *args)
2094 {
2095 	lwpsignal(curproc, curthread->td_lwp, SIGSYS);
2096 	return (EINVAL);
2097 }
2098 
2099 /*
2100  * Send a SIGIO or SIGURG signal to a process or process group using
2101  * stored credentials rather than those of the current process.
2102  */
2103 void
2104 pgsigio(struct sigio *sigio, int sig, int checkctty)
2105 {
2106 	if (sigio == NULL)
2107 		return;
2108 
2109 	if (sigio->sio_pgid > 0) {
2110 		if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
2111 		             sigio->sio_proc))
2112 			ksignal(sigio->sio_proc, sig);
2113 	} else if (sigio->sio_pgid < 0) {
2114 		struct proc *p;
2115 
2116 		lockmgr(&sigio->sio_pgrp->pg_lock, LK_EXCLUSIVE);
2117 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
2118 			if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
2119 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
2120 				ksignal(p, sig);
2121 		}
2122 		lockmgr(&sigio->sio_pgrp->pg_lock, LK_RELEASE);
2123 	}
2124 }
2125 
2126 static int
2127 filt_sigattach(struct knote *kn)
2128 {
2129 	struct proc *p = curproc;
2130 
2131 	kn->kn_ptr.p_proc = p;
2132 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
2133 
2134 	/* XXX lock the proc here while adding to the list? */
2135 	SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext);
2136 
2137 	return (0);
2138 }
2139 
2140 static void
2141 filt_sigdetach(struct knote *kn)
2142 {
2143 	struct proc *p = kn->kn_ptr.p_proc;
2144 
2145 	SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
2146 }
2147 
2148 /*
2149  * signal knotes are shared with proc knotes, so we apply a mask to
2150  * the hint in order to differentiate them from process hints.  This
2151  * could be avoided by using a signal-specific knote list, but probably
2152  * isn't worth the trouble.
2153  */
2154 static int
2155 filt_signal(struct knote *kn, long hint)
2156 {
2157 	if (hint & NOTE_SIGNAL) {
2158 		hint &= ~NOTE_SIGNAL;
2159 
2160 		if (kn->kn_id == hint)
2161 			kn->kn_data++;
2162 	}
2163 	return (kn->kn_data != 0);
2164 }
2165