xref: /dragonfly/sys/kern/kern_sig.c (revision b8d2998c)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_sig.c	8.7 (Berkeley) 4/18/94
39  * $FreeBSD: src/sys/kern/kern_sig.c,v 1.72.2.17 2003/05/16 16:34:34 obrien Exp $
40  */
41 
42 #include "opt_ktrace.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/sysproto.h>
48 #include <sys/signalvar.h>
49 #include <sys/resourcevar.h>
50 #include <sys/vnode.h>
51 #include <sys/event.h>
52 #include <sys/proc.h>
53 #include <sys/nlookup.h>
54 #include <sys/pioctl.h>
55 #include <sys/systm.h>
56 #include <sys/acct.h>
57 #include <sys/fcntl.h>
58 #include <sys/lock.h>
59 #include <sys/wait.h>
60 #include <sys/ktrace.h>
61 #include <sys/syslog.h>
62 #include <sys/stat.h>
63 #include <sys/sysent.h>
64 #include <sys/sysctl.h>
65 #include <sys/malloc.h>
66 #include <sys/interrupt.h>
67 #include <sys/unistd.h>
68 #include <sys/kern_syscall.h>
69 #include <sys/vkernel.h>
70 
71 #include <sys/signal2.h>
72 #include <sys/thread2.h>
73 
74 #include <machine/cpu.h>
75 #include <machine/smp.h>
76 
77 static int	coredump(struct lwp *, int);
78 static char	*expand_name(const char *, uid_t, pid_t);
79 static int	dokillpg(int sig, int pgid, int all);
80 static int	sig_ffs(sigset_t *set);
81 static int	sigprop(int sig);
82 static void	lwp_signotify(struct lwp *lp);
83 #ifdef SMP
84 static void	signotify_remote(void *arg);
85 #endif
86 static int	kern_sigtimedwait(sigset_t set, siginfo_t *info,
87 		    struct timespec *timeout);
88 
89 static int	filt_sigattach(struct knote *kn);
90 static void	filt_sigdetach(struct knote *kn);
91 static int	filt_signal(struct knote *kn, long hint);
92 
93 struct filterops sig_filtops =
94 	{ 0, filt_sigattach, filt_sigdetach, filt_signal };
95 
96 static int	kern_logsigexit = 1;
97 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
98     &kern_logsigexit, 0,
99     "Log processes quitting on abnormal signals to syslog(3)");
100 
101 /*
102  * Can process p, with pcred pc, send the signal sig to process q?
103  */
104 #define CANSIGNAL(q, sig) \
105 	(!p_trespass(curproc->p_ucred, (q)->p_ucred) || \
106 	((sig) == SIGCONT && (q)->p_session == curproc->p_session))
107 
108 /*
109  * Policy -- Can real uid ruid with ucred uc send a signal to process q?
110  */
111 #define CANSIGIO(ruid, uc, q) \
112 	((uc)->cr_uid == 0 || \
113 	    (ruid) == (q)->p_ucred->cr_ruid || \
114 	    (uc)->cr_uid == (q)->p_ucred->cr_ruid || \
115 	    (ruid) == (q)->p_ucred->cr_uid || \
116 	    (uc)->cr_uid == (q)->p_ucred->cr_uid)
117 
118 int sugid_coredump;
119 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RW,
120 	&sugid_coredump, 0, "Enable coredumping set user/group ID processes");
121 
122 static int	do_coredump = 1;
123 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
124 	&do_coredump, 0, "Enable/Disable coredumps");
125 
126 /*
127  * Signal properties and actions.
128  * The array below categorizes the signals and their default actions
129  * according to the following properties:
130  */
131 #define	SA_KILL		0x01		/* terminates process by default */
132 #define	SA_CORE		0x02		/* ditto and coredumps */
133 #define	SA_STOP		0x04		/* suspend process */
134 #define	SA_TTYSTOP	0x08		/* ditto, from tty */
135 #define	SA_IGNORE	0x10		/* ignore by default */
136 #define	SA_CONT		0x20		/* continue if suspended */
137 #define	SA_CANTMASK	0x40		/* non-maskable, catchable */
138 #define SA_CKPT         0x80            /* checkpoint process */
139 
140 
141 static int sigproptbl[NSIG] = {
142         SA_KILL,                /* SIGHUP */
143         SA_KILL,                /* SIGINT */
144         SA_KILL|SA_CORE,        /* SIGQUIT */
145         SA_KILL|SA_CORE,        /* SIGILL */
146         SA_KILL|SA_CORE,        /* SIGTRAP */
147         SA_KILL|SA_CORE,        /* SIGABRT */
148         SA_KILL|SA_CORE,        /* SIGEMT */
149         SA_KILL|SA_CORE,        /* SIGFPE */
150         SA_KILL,                /* SIGKILL */
151         SA_KILL|SA_CORE,        /* SIGBUS */
152         SA_KILL|SA_CORE,        /* SIGSEGV */
153         SA_KILL|SA_CORE,        /* SIGSYS */
154         SA_KILL,                /* SIGPIPE */
155         SA_KILL,                /* SIGALRM */
156         SA_KILL,                /* SIGTERM */
157         SA_IGNORE,              /* SIGURG */
158         SA_STOP,                /* SIGSTOP */
159         SA_STOP|SA_TTYSTOP,     /* SIGTSTP */
160         SA_IGNORE|SA_CONT,      /* SIGCONT */
161         SA_IGNORE,              /* SIGCHLD */
162         SA_STOP|SA_TTYSTOP,     /* SIGTTIN */
163         SA_STOP|SA_TTYSTOP,     /* SIGTTOU */
164         SA_IGNORE,              /* SIGIO */
165         SA_KILL,                /* SIGXCPU */
166         SA_KILL,                /* SIGXFSZ */
167         SA_KILL,                /* SIGVTALRM */
168         SA_KILL,                /* SIGPROF */
169         SA_IGNORE,              /* SIGWINCH  */
170         SA_IGNORE,              /* SIGINFO */
171         SA_KILL,                /* SIGUSR1 */
172         SA_KILL,                /* SIGUSR2 */
173 	SA_IGNORE,              /* SIGTHR */
174 	SA_CKPT,                /* SIGCKPT */
175 	SA_KILL|SA_CKPT,        /* SIGCKPTEXIT */
176 	SA_IGNORE,
177 	SA_IGNORE,
178 	SA_IGNORE,
179 	SA_IGNORE,
180 	SA_IGNORE,
181 	SA_IGNORE,
182 	SA_IGNORE,
183 	SA_IGNORE,
184 	SA_IGNORE,
185 	SA_IGNORE,
186 	SA_IGNORE,
187 	SA_IGNORE,
188 	SA_IGNORE,
189 	SA_IGNORE,
190 	SA_IGNORE,
191 	SA_IGNORE,
192 	SA_IGNORE,
193 	SA_IGNORE,
194 	SA_IGNORE,
195 	SA_IGNORE,
196 	SA_IGNORE,
197 	SA_IGNORE,
198 	SA_IGNORE,
199 	SA_IGNORE,
200 	SA_IGNORE,
201 	SA_IGNORE,
202 	SA_IGNORE,
203 	SA_IGNORE,
204 	SA_IGNORE,
205 	SA_IGNORE,
206 
207 };
208 
209 static __inline int
210 sigprop(int sig)
211 {
212 
213 	if (sig > 0 && sig < NSIG)
214 		return (sigproptbl[_SIG_IDX(sig)]);
215 	return (0);
216 }
217 
218 static __inline int
219 sig_ffs(sigset_t *set)
220 {
221 	int i;
222 
223 	for (i = 0; i < _SIG_WORDS; i++)
224 		if (set->__bits[i])
225 			return (ffs(set->__bits[i]) + (i * 32));
226 	return (0);
227 }
228 
229 /*
230  * No requirements.
231  */
232 int
233 kern_sigaction(int sig, struct sigaction *act, struct sigaction *oact)
234 {
235 	struct thread *td = curthread;
236 	struct proc *p = td->td_proc;
237 	struct lwp *lp;
238 	struct sigacts *ps = p->p_sigacts;
239 
240 	if (sig <= 0 || sig > _SIG_MAXSIG)
241 		return (EINVAL);
242 
243 	lwkt_gettoken(&proc_token);
244 
245 	if (oact) {
246 		oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
247 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
248 		oact->sa_flags = 0;
249 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
250 			oact->sa_flags |= SA_ONSTACK;
251 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
252 			oact->sa_flags |= SA_RESTART;
253 		if (SIGISMEMBER(ps->ps_sigreset, sig))
254 			oact->sa_flags |= SA_RESETHAND;
255 		if (SIGISMEMBER(ps->ps_signodefer, sig))
256 			oact->sa_flags |= SA_NODEFER;
257 		if (SIGISMEMBER(ps->ps_siginfo, sig))
258 			oact->sa_flags |= SA_SIGINFO;
259 		if (SIGISMEMBER(ps->ps_sigmailbox, sig))
260 			oact->sa_flags |= SA_MAILBOX;
261 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDSTOP)
262 			oact->sa_flags |= SA_NOCLDSTOP;
263 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDWAIT)
264 			oact->sa_flags |= SA_NOCLDWAIT;
265 	}
266 	if (act) {
267 		/*
268 		 * Check for invalid requests.  KILL and STOP cannot be
269 		 * caught.
270 		 */
271 		if (sig == SIGKILL || sig == SIGSTOP) {
272 			if (act->sa_handler != SIG_DFL) {
273 				lwkt_reltoken(&proc_token);
274 				return (EINVAL);
275 			}
276 #if 0
277 			/* (not needed, SIG_DFL forces action to occur) */
278 			if (act->sa_flags & SA_MAILBOX)
279 				return (EINVAL);
280 #endif
281 		}
282 
283 		/*
284 		 * Change setting atomically.
285 		 */
286 		crit_enter();
287 
288 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
289 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
290 		if (act->sa_flags & SA_SIGINFO) {
291 			ps->ps_sigact[_SIG_IDX(sig)] =
292 			    (__sighandler_t *)act->sa_sigaction;
293 			SIGADDSET(ps->ps_siginfo, sig);
294 		} else {
295 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
296 			SIGDELSET(ps->ps_siginfo, sig);
297 		}
298 		if (!(act->sa_flags & SA_RESTART))
299 			SIGADDSET(ps->ps_sigintr, sig);
300 		else
301 			SIGDELSET(ps->ps_sigintr, sig);
302 		if (act->sa_flags & SA_ONSTACK)
303 			SIGADDSET(ps->ps_sigonstack, sig);
304 		else
305 			SIGDELSET(ps->ps_sigonstack, sig);
306 		if (act->sa_flags & SA_RESETHAND)
307 			SIGADDSET(ps->ps_sigreset, sig);
308 		else
309 			SIGDELSET(ps->ps_sigreset, sig);
310 		if (act->sa_flags & SA_NODEFER)
311 			SIGADDSET(ps->ps_signodefer, sig);
312 		else
313 			SIGDELSET(ps->ps_signodefer, sig);
314 		if (act->sa_flags & SA_MAILBOX)
315 			SIGADDSET(ps->ps_sigmailbox, sig);
316 		else
317 			SIGDELSET(ps->ps_sigmailbox, sig);
318 		if (sig == SIGCHLD) {
319 			if (act->sa_flags & SA_NOCLDSTOP)
320 				p->p_sigacts->ps_flag |= PS_NOCLDSTOP;
321 			else
322 				p->p_sigacts->ps_flag &= ~PS_NOCLDSTOP;
323 			if (act->sa_flags & SA_NOCLDWAIT) {
324 				/*
325 				 * Paranoia: since SA_NOCLDWAIT is implemented
326 				 * by reparenting the dying child to PID 1 (and
327 				 * trust it to reap the zombie), PID 1 itself
328 				 * is forbidden to set SA_NOCLDWAIT.
329 				 */
330 				if (p->p_pid == 1)
331 					p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
332 				else
333 					p->p_sigacts->ps_flag |= PS_NOCLDWAIT;
334 			} else {
335 				p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
336 			}
337 		}
338 		/*
339 		 * Set bit in p_sigignore for signals that are set to SIG_IGN,
340 		 * and for signals set to SIG_DFL where the default is to
341 		 * ignore. However, don't put SIGCONT in p_sigignore, as we
342 		 * have to restart the process.
343 		 */
344 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
345 		    (sigprop(sig) & SA_IGNORE &&
346 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
347 			/* never to be seen again */
348 			SIGDELSET(p->p_siglist, sig);
349 			/*
350 			 * Remove the signal also from the thread lists.
351 			 */
352 			FOREACH_LWP_IN_PROC(lp, p) {
353 				SIGDELSET(lp->lwp_siglist, sig);
354 			}
355 			if (sig != SIGCONT) {
356 				/* easier in ksignal */
357 				SIGADDSET(p->p_sigignore, sig);
358 			}
359 			SIGDELSET(p->p_sigcatch, sig);
360 		} else {
361 			SIGDELSET(p->p_sigignore, sig);
362 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
363 				SIGDELSET(p->p_sigcatch, sig);
364 			else
365 				SIGADDSET(p->p_sigcatch, sig);
366 		}
367 
368 		crit_exit();
369 	}
370 	lwkt_reltoken(&proc_token);
371 	return (0);
372 }
373 
374 int
375 sys_sigaction(struct sigaction_args *uap)
376 {
377 	struct sigaction act, oact;
378 	struct sigaction *actp, *oactp;
379 	int error;
380 
381 	actp = (uap->act != NULL) ? &act : NULL;
382 	oactp = (uap->oact != NULL) ? &oact : NULL;
383 	if (actp) {
384 		error = copyin(uap->act, actp, sizeof(act));
385 		if (error)
386 			return (error);
387 	}
388 	error = kern_sigaction(uap->sig, actp, oactp);
389 	if (oactp && !error) {
390 		error = copyout(oactp, uap->oact, sizeof(oact));
391 	}
392 	return (error);
393 }
394 
395 /*
396  * Initialize signal state for process 0;
397  * set to ignore signals that are ignored by default.
398  */
399 void
400 siginit(struct proc *p)
401 {
402 	int i;
403 
404 	for (i = 1; i <= NSIG; i++)
405 		if (sigprop(i) & SA_IGNORE && i != SIGCONT)
406 			SIGADDSET(p->p_sigignore, i);
407 }
408 
409 /*
410  * Reset signals for an exec of the specified process.
411  */
412 void
413 execsigs(struct proc *p)
414 {
415 	struct sigacts *ps = p->p_sigacts;
416 	struct lwp *lp;
417 	int sig;
418 
419 	lp = ONLY_LWP_IN_PROC(p);
420 
421 	/*
422 	 * Reset caught signals.  Held signals remain held
423 	 * through p_sigmask (unless they were caught,
424 	 * and are now ignored by default).
425 	 */
426 	while (SIGNOTEMPTY(p->p_sigcatch)) {
427 		sig = sig_ffs(&p->p_sigcatch);
428 		SIGDELSET(p->p_sigcatch, sig);
429 		if (sigprop(sig) & SA_IGNORE) {
430 			if (sig != SIGCONT)
431 				SIGADDSET(p->p_sigignore, sig);
432 			SIGDELSET(p->p_siglist, sig);
433 			SIGDELSET(lp->lwp_siglist, sig);
434 		}
435 		ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
436 	}
437 
438 	/*
439 	 * Reset stack state to the user stack.
440 	 * Clear set of signals caught on the signal stack.
441 	 */
442 	lp->lwp_sigstk.ss_flags = SS_DISABLE;
443 	lp->lwp_sigstk.ss_size = 0;
444 	lp->lwp_sigstk.ss_sp = 0;
445 	lp->lwp_flag &= ~LWP_ALTSTACK;
446 	/*
447 	 * Reset no zombies if child dies flag as Solaris does.
448 	 */
449 	p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
450 }
451 
452 /*
453  * kern_sigprocmask() - MP SAFE ONLY IF p == curproc
454  *
455  *	Manipulate signal mask.  This routine is MP SAFE *ONLY* if
456  *	p == curproc.
457  */
458 int
459 kern_sigprocmask(int how, sigset_t *set, sigset_t *oset)
460 {
461 	struct thread *td = curthread;
462 	struct lwp *lp = td->td_lwp;
463 	int error;
464 
465 	lwkt_gettoken(&proc_token);
466 
467 	if (oset != NULL)
468 		*oset = lp->lwp_sigmask;
469 
470 	error = 0;
471 	if (set != NULL) {
472 		switch (how) {
473 		case SIG_BLOCK:
474 			SIG_CANTMASK(*set);
475 			SIGSETOR(lp->lwp_sigmask, *set);
476 			break;
477 		case SIG_UNBLOCK:
478 			SIGSETNAND(lp->lwp_sigmask, *set);
479 			break;
480 		case SIG_SETMASK:
481 			SIG_CANTMASK(*set);
482 			lp->lwp_sigmask = *set;
483 			break;
484 		default:
485 			error = EINVAL;
486 			break;
487 		}
488 	}
489 
490 	lwkt_reltoken(&proc_token);
491 
492 	return (error);
493 }
494 
495 /*
496  * sigprocmask()
497  *
498  * MPSAFE
499  */
500 int
501 sys_sigprocmask(struct sigprocmask_args *uap)
502 {
503 	sigset_t set, oset;
504 	sigset_t *setp, *osetp;
505 	int error;
506 
507 	setp = (uap->set != NULL) ? &set : NULL;
508 	osetp = (uap->oset != NULL) ? &oset : NULL;
509 	if (setp) {
510 		error = copyin(uap->set, setp, sizeof(set));
511 		if (error)
512 			return (error);
513 	}
514 	error = kern_sigprocmask(uap->how, setp, osetp);
515 	if (osetp && !error) {
516 		error = copyout(osetp, uap->oset, sizeof(oset));
517 	}
518 	return (error);
519 }
520 
521 /*
522  * MPSAFE
523  */
524 int
525 kern_sigpending(struct __sigset *set)
526 {
527 	struct lwp *lp = curthread->td_lwp;
528 
529 	*set = lwp_sigpend(lp);
530 
531 	return (0);
532 }
533 
534 /*
535  * MPSAFE
536  */
537 int
538 sys_sigpending(struct sigpending_args *uap)
539 {
540 	sigset_t set;
541 	int error;
542 
543 	error = kern_sigpending(&set);
544 
545 	if (error == 0)
546 		error = copyout(&set, uap->set, sizeof(set));
547 	return (error);
548 }
549 
550 /*
551  * Suspend process until signal, providing mask to be set
552  * in the meantime.
553  *
554  * MPSAFE
555  */
556 int
557 kern_sigsuspend(struct __sigset *set)
558 {
559 	struct thread *td = curthread;
560 	struct lwp *lp = td->td_lwp;
561 	struct proc *p = td->td_proc;
562 	struct sigacts *ps = p->p_sigacts;
563 
564 	/*
565 	 * When returning from sigsuspend, we want
566 	 * the old mask to be restored after the
567 	 * signal handler has finished.  Thus, we
568 	 * save it here and mark the sigacts structure
569 	 * to indicate this.
570 	 */
571 	lp->lwp_oldsigmask = lp->lwp_sigmask;
572 	lp->lwp_flag |= LWP_OLDMASK;
573 
574 	SIG_CANTMASK(*set);
575 	lp->lwp_sigmask = *set;
576 	while (tsleep(ps, PCATCH, "pause", 0) == 0)
577 		/* void */;
578 	/* always return EINTR rather than ERESTART... */
579 	return (EINTR);
580 }
581 
582 /*
583  * Note nonstandard calling convention: libc stub passes mask, not
584  * pointer, to save a copyin.
585  *
586  * MPSAFE
587  */
588 int
589 sys_sigsuspend(struct sigsuspend_args *uap)
590 {
591 	sigset_t mask;
592 	int error;
593 
594 	error = copyin(uap->sigmask, &mask, sizeof(mask));
595 	if (error)
596 		return (error);
597 
598 	error = kern_sigsuspend(&mask);
599 
600 	return (error);
601 }
602 
603 /*
604  * MPSAFE
605  */
606 int
607 kern_sigaltstack(struct sigaltstack *ss, struct sigaltstack *oss)
608 {
609 	struct thread *td = curthread;
610 	struct lwp *lp = td->td_lwp;
611 	struct proc *p = td->td_proc;
612 
613 	if ((lp->lwp_flag & LWP_ALTSTACK) == 0)
614 		lp->lwp_sigstk.ss_flags |= SS_DISABLE;
615 
616 	if (oss)
617 		*oss = lp->lwp_sigstk;
618 
619 	if (ss) {
620 		if (ss->ss_flags & SS_DISABLE) {
621 			if (lp->lwp_sigstk.ss_flags & SS_ONSTACK)
622 				return (EINVAL);
623 			lp->lwp_flag &= ~LWP_ALTSTACK;
624 			lp->lwp_sigstk.ss_flags = ss->ss_flags;
625 		} else {
626 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
627 				return (ENOMEM);
628 			lp->lwp_flag |= LWP_ALTSTACK;
629 			lp->lwp_sigstk = *ss;
630 		}
631 	}
632 
633 	return (0);
634 }
635 
636 /*
637  * MPSAFE
638  */
639 int
640 sys_sigaltstack(struct sigaltstack_args *uap)
641 {
642 	stack_t ss, oss;
643 	int error;
644 
645 	if (uap->ss) {
646 		error = copyin(uap->ss, &ss, sizeof(ss));
647 		if (error)
648 			return (error);
649 	}
650 
651 	error = kern_sigaltstack(uap->ss ? &ss : NULL,
652 	    uap->oss ? &oss : NULL);
653 
654 	if (error == 0 && uap->oss)
655 		error = copyout(&oss, uap->oss, sizeof(*uap->oss));
656 	return (error);
657 }
658 
659 /*
660  * Common code for kill process group/broadcast kill.
661  * cp is calling process.
662  */
663 struct killpg_info {
664 	int nfound;
665 	int sig;
666 };
667 
668 static int killpg_all_callback(struct proc *p, void *data);
669 
670 static int
671 dokillpg(int sig, int pgid, int all)
672 {
673 	struct killpg_info info;
674 	struct proc *cp = curproc;
675 	struct proc *p;
676 	struct pgrp *pgrp;
677 
678 	info.nfound = 0;
679 	info.sig = sig;
680 
681 	if (all) {
682 		/*
683 		 * broadcast
684 		 */
685 		allproc_scan(killpg_all_callback, &info);
686 	} else {
687 		if (pgid == 0) {
688 			/*
689 			 * zero pgid means send to my process group.
690 			 */
691 			pgrp = cp->p_pgrp;
692 		} else {
693 			pgrp = pgfind(pgid);
694 			if (pgrp == NULL)
695 				return (ESRCH);
696 		}
697 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
698 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
699 			if (p->p_pid <= 1 ||
700 			    p->p_stat == SZOMB ||
701 			    (p->p_flag & P_SYSTEM) ||
702 			    !CANSIGNAL(p, sig)) {
703 				continue;
704 			}
705 			++info.nfound;
706 			if (sig)
707 				ksignal(p, sig);
708 		}
709 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
710 	}
711 	return (info.nfound ? 0 : ESRCH);
712 }
713 
714 static int
715 killpg_all_callback(struct proc *p, void *data)
716 {
717 	struct killpg_info *info = data;
718 
719 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) ||
720 	    p == curproc || !CANSIGNAL(p, info->sig)) {
721 		return (0);
722 	}
723 	++info->nfound;
724 	if (info->sig)
725 		ksignal(p, info->sig);
726 	return(0);
727 }
728 
729 /*
730  * Send a general signal to a process or LWPs within that process.  Note
731  * that new signals cannot be sent if a process is exiting.
732  *
733  * No requirements.
734  */
735 int
736 kern_kill(int sig, pid_t pid, lwpid_t tid)
737 {
738 	int t;
739 
740 	if ((u_int)sig > _SIG_MAXSIG)
741 		return (EINVAL);
742 
743 	lwkt_gettoken(&proc_token);
744 
745 	if (pid > 0) {
746 		struct proc *p;
747 		struct lwp *lp = NULL;
748 
749 		/* kill single process */
750 		if ((p = pfind(pid)) == NULL) {
751 			lwkt_reltoken(&proc_token);
752 			return (ESRCH);
753 		}
754 		if (!CANSIGNAL(p, sig)) {
755 			lwkt_reltoken(&proc_token);
756 			return (EPERM);
757 		}
758 
759 		/*
760 		 * NOP if the process is exiting.  Note that lwpsignal() is
761 		 * called directly with P_WEXIT set to kill individual LWPs
762 		 * during exit, which is allowed.
763 		 */
764 		if (p->p_flag & P_WEXIT) {
765 			lwkt_reltoken(&proc_token);
766 			return (0);
767 		}
768 		if (tid != -1) {
769 			lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, tid);
770 			if (lp == NULL) {
771 				lwkt_reltoken(&proc_token);
772 				return (ESRCH);
773 			}
774 		}
775 		if (sig)
776 			lwpsignal(p, lp, sig);
777 		lwkt_reltoken(&proc_token);
778 		return (0);
779 	}
780 	/*
781 	 * If we come here, pid is a special broadcast pid.
782 	 * This doesn't mix with a tid.
783 	 */
784 	if (tid != -1) {
785 		lwkt_reltoken(&proc_token);
786 		return (EINVAL);
787 	}
788 	switch (pid) {
789 	case -1:		/* broadcast signal */
790 		t = (dokillpg(sig, 0, 1));
791 		break;
792 	case 0:			/* signal own process group */
793 		t = (dokillpg(sig, 0, 0));
794 		break;
795 	default:		/* negative explicit process group */
796 		t = (dokillpg(sig, -pid, 0));
797 		break;
798 	}
799 	lwkt_reltoken(&proc_token);
800 	return t;
801 }
802 
803 int
804 sys_kill(struct kill_args *uap)
805 {
806 	int error;
807 
808 	error = kern_kill(uap->signum, uap->pid, -1);
809 	return (error);
810 }
811 
812 int
813 sys_lwp_kill(struct lwp_kill_args *uap)
814 {
815 	int error;
816 	pid_t pid = uap->pid;
817 
818 	/*
819 	 * A tid is mandatory for lwp_kill(), otherwise
820 	 * you could simply use kill().
821 	 */
822 	if (uap->tid == -1)
823 		return (EINVAL);
824 
825 	/*
826 	 * To save on a getpid() function call for intra-process
827 	 * signals, pid == -1 means current process.
828 	 */
829 	if (pid == -1)
830 		pid = curproc->p_pid;
831 
832 	error = kern_kill(uap->signum, pid, uap->tid);
833 	return (error);
834 }
835 
836 /*
837  * Send a signal to a process group.
838  */
839 void
840 gsignal(int pgid, int sig)
841 {
842 	struct pgrp *pgrp;
843 
844 	if (pgid && (pgrp = pgfind(pgid)))
845 		pgsignal(pgrp, sig, 0);
846 }
847 
848 /*
849  * Send a signal to a process group.  If checktty is 1,
850  * limit to members which have a controlling terminal.
851  *
852  * pg_lock interlocks against a fork that might be in progress, to
853  * ensure that the new child process picks up the signal.
854  */
855 void
856 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
857 {
858 	struct proc *p;
859 
860 	if (pgrp) {
861 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
862 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
863 			if (checkctty == 0 || p->p_flag & P_CONTROLT)
864 				ksignal(p, sig);
865 		}
866 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
867 	}
868 }
869 
870 /*
871  * Send a signal caused by a trap to the current lwp.  If it will be caught
872  * immediately, deliver it with correct code.  Otherwise, post it normally.
873  *
874  * These signals may ONLY be delivered to the specified lwp and may never
875  * be delivered to the process generically.
876  */
877 void
878 trapsignal(struct lwp *lp, int sig, u_long code)
879 {
880 	struct proc *p = lp->lwp_proc;
881 	struct sigacts *ps = p->p_sigacts;
882 
883 	/*
884 	 * If we are a virtual kernel running an emulated user process
885 	 * context, switch back to the virtual kernel context before
886 	 * trying to post the signal.
887 	 */
888 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
889 		struct trapframe *tf = lp->lwp_md.md_regs;
890 		tf->tf_trapno = 0;
891 		vkernel_trap(lp, tf);
892 	}
893 
894 
895 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(p->p_sigcatch, sig) &&
896 	    !SIGISMEMBER(lp->lwp_sigmask, sig)) {
897 		lp->lwp_ru.ru_nsignals++;
898 #ifdef KTRACE
899 		if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
900 			ktrpsig(lp, sig, ps->ps_sigact[_SIG_IDX(sig)],
901 				&lp->lwp_sigmask, code);
902 #endif
903 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], sig,
904 						&lp->lwp_sigmask, code);
905 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
906 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
907 			SIGADDSET(lp->lwp_sigmask, sig);
908 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
909 			/*
910 			 * See kern_sigaction() for origin of this code.
911 			 */
912 			SIGDELSET(p->p_sigcatch, sig);
913 			if (sig != SIGCONT &&
914 			    sigprop(sig) & SA_IGNORE)
915 				SIGADDSET(p->p_sigignore, sig);
916 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
917 		}
918 	} else {
919 		lp->lwp_code = code;	/* XXX for core dump/debugger */
920 		lp->lwp_sig = sig;	/* XXX to verify code */
921 		lwpsignal(p, lp, sig);
922 	}
923 }
924 
925 /*
926  * Find a suitable lwp to deliver the signal to.
927  *
928  * Returns NULL if all lwps hold the signal blocked.
929  */
930 static struct lwp *
931 find_lwp_for_signal(struct proc *p, int sig)
932 {
933 	struct lwp *lp;
934 	struct lwp *run, *sleep, *stop;
935 
936 	/*
937 	 * If the running/preempted thread belongs to the proc to which
938 	 * the signal is being delivered and this thread does not block
939 	 * the signal, then we can avoid a context switch by delivering
940 	 * the signal to this thread, because it will return to userland
941 	 * soon anyways.
942 	 */
943 	lp = lwkt_preempted_proc();
944 	if (lp != NULL && lp->lwp_proc == p && !SIGISMEMBER(lp->lwp_sigmask, sig))
945 		return (lp);
946 
947 	run = sleep = stop = NULL;
948 	FOREACH_LWP_IN_PROC(lp, p) {
949 		/*
950 		 * If the signal is being blocked by the lwp, then this
951 		 * lwp is not eligible for receiving the signal.
952 		 */
953 		if (SIGISMEMBER(lp->lwp_sigmask, sig))
954 			continue;
955 
956 		switch (lp->lwp_stat) {
957 		case LSRUN:
958 			run = lp;
959 			break;
960 
961 		case LSSTOP:
962 			stop = lp;
963 			break;
964 
965 		case LSSLEEP:
966 			if (lp->lwp_flag & LWP_SINTR)
967 				sleep = lp;
968 			break;
969 		}
970 	}
971 
972 	if (run != NULL)
973 		return (run);
974 	else if (sleep != NULL)
975 		return (sleep);
976 	else
977 		return (stop);
978 }
979 
980 /*
981  * Send the signal to the process.  If the signal has an action, the action
982  * is usually performed by the target process rather than the caller; we add
983  * the signal to the set of pending signals for the process.
984  *
985  * Exceptions:
986  *   o When a stop signal is sent to a sleeping process that takes the
987  *     default action, the process is stopped without awakening it.
988  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
989  *     regardless of the signal action (eg, blocked or ignored).
990  *
991  * Other ignored signals are discarded immediately.
992  *
993  * No requirements.
994  */
995 void
996 ksignal(struct proc *p, int sig)
997 {
998 	lwpsignal(p, NULL, sig);
999 }
1000 
1001 /*
1002  * The core for ksignal.  lp may be NULL, then a suitable thread
1003  * will be chosen.  If not, lp MUST be a member of p.
1004  *
1005  * No requirements.
1006  */
1007 void
1008 lwpsignal(struct proc *p, struct lwp *lp, int sig)
1009 {
1010 	int prop;
1011 	sig_t action;
1012 
1013 	if (sig > _SIG_MAXSIG || sig <= 0) {
1014 		kprintf("lwpsignal: signal %d\n", sig);
1015 		panic("lwpsignal signal number");
1016 	}
1017 
1018 	KKASSERT(lp == NULL || lp->lwp_proc == p);
1019 
1020 	lwkt_gettoken(&proc_token);
1021 
1022 	prop = sigprop(sig);
1023 
1024 	/*
1025 	 * If proc is traced, always give parent a chance;
1026 	 * if signal event is tracked by procfs, give *that*
1027 	 * a chance, as well.
1028 	 */
1029 	if ((p->p_flag & P_TRACED) || (p->p_stops & S_SIG)) {
1030 		action = SIG_DFL;
1031 	} else {
1032 		/*
1033 		 * Do not try to deliver signals to an exiting lwp.  Note
1034 		 * that we must still deliver the signal if P_WEXIT is set
1035 		 * in the process flags.
1036 		 */
1037 		if (lp && (lp->lwp_flag & LWP_WEXIT)) {
1038 			lwkt_reltoken(&proc_token);
1039 			return;
1040 		}
1041 
1042 		/*
1043 		 * If the signal is being ignored, then we forget about
1044 		 * it immediately.  NOTE: We don't set SIGCONT in p_sigignore,
1045 		 * and if it is set to SIG_IGN, action will be SIG_DFL here.
1046 		 */
1047 		if (SIGISMEMBER(p->p_sigignore, sig)) {
1048 			lwkt_reltoken(&proc_token);
1049 			return;
1050 		}
1051 		if (SIGISMEMBER(p->p_sigcatch, sig))
1052 			action = SIG_CATCH;
1053 		else
1054 			action = SIG_DFL;
1055 	}
1056 
1057 	/*
1058 	 * If continuing, clear any pending STOP signals.
1059 	 */
1060 	if (prop & SA_CONT)
1061 		SIG_STOPSIGMASK(p->p_siglist);
1062 
1063 	if (prop & SA_STOP) {
1064 		/*
1065 		 * If sending a tty stop signal to a member of an orphaned
1066 		 * process group, discard the signal here if the action
1067 		 * is default; don't stop the process below if sleeping,
1068 		 * and don't clear any pending SIGCONT.
1069 		 */
1070 		if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0 &&
1071 		    action == SIG_DFL) {
1072 			lwkt_reltoken(&proc_token);
1073 		        return;
1074 		}
1075 		SIG_CONTSIGMASK(p->p_siglist);
1076 		p->p_flag &= ~P_CONTINUED;
1077 	}
1078 
1079 	crit_enter();
1080 
1081 	if (p->p_stat == SSTOP) {
1082 		/*
1083 		 * Nobody can handle this signal, add it to the lwp or
1084 		 * process pending list
1085 		 */
1086 		if (lp)
1087 			SIGADDSET(lp->lwp_siglist, sig);
1088 		else
1089 			SIGADDSET(p->p_siglist, sig);
1090 
1091 		/*
1092 		 * If the process is stopped and is being traced, then no
1093 		 * further action is necessary.
1094 		 */
1095 		if (p->p_flag & P_TRACED)
1096 			goto out;
1097 
1098 		/*
1099 		 * If the process is stopped and receives a KILL signal,
1100 		 * make the process runnable.
1101 		 */
1102 		if (sig == SIGKILL) {
1103 			proc_unstop(p);
1104 			goto active_process;
1105 		}
1106 
1107 		/*
1108 		 * If the process is stopped and receives a CONT signal,
1109 		 * then try to make the process runnable again.
1110 		 */
1111 		if (prop & SA_CONT) {
1112 			/*
1113 			 * If SIGCONT is default (or ignored), we continue the
1114 			 * process but don't leave the signal in p_siglist, as
1115 			 * it has no further action.  If SIGCONT is held, we
1116 			 * continue the process and leave the signal in
1117 			 * p_siglist.  If the process catches SIGCONT, let it
1118 			 * handle the signal itself.
1119 			 */
1120 			/* XXX what if the signal is being held blocked? */
1121 			p->p_flag |= P_CONTINUED;
1122 			wakeup(p->p_pptr);
1123 			if (action == SIG_DFL)
1124 				SIGDELSET(p->p_siglist, sig);
1125 			proc_unstop(p);
1126 			if (action == SIG_CATCH)
1127 				goto active_process;
1128 			goto out;
1129 		}
1130 
1131 		/*
1132 		 * If the process is stopped and receives another STOP
1133 		 * signal, we do not need to stop it again.  If we did
1134 		 * the shell could get confused.
1135 		 *
1136 		 * However, if the current/preempted lwp is part of the
1137 		 * process receiving the signal, we need to keep it,
1138 		 * so that this lwp can stop in issignal() later, as
1139 		 * we don't want to wait until it reaches userret!
1140 		 */
1141 		if (prop & SA_STOP) {
1142 			if (lwkt_preempted_proc() == NULL ||
1143 			    lwkt_preempted_proc()->lwp_proc != p)
1144 				SIGDELSET(p->p_siglist, sig);
1145 		}
1146 
1147 		/*
1148 		 * Otherwise the process is stopped and it received some
1149 		 * signal, which does not change its stopped state.
1150 		 *
1151 		 * We have to select one thread to set LWP_BREAKTSLEEP,
1152 		 * so that the current signal will break the sleep
1153 		 * as soon as a SA_CONT signal will unstop the process.
1154 		 */
1155 		if (lp == NULL)
1156 			lp = find_lwp_for_signal(p, sig);
1157 		if (lp != NULL &&
1158 		    (lp->lwp_stat == LSSLEEP || lp->lwp_stat == LSSTOP))
1159 			lp->lwp_flag |= LWP_BREAKTSLEEP;
1160 		goto out;
1161 
1162 		/* NOTREACHED */
1163 	}
1164 	/* else not stopped */
1165 active_process:
1166 
1167 	/*
1168 	 * Never deliver a lwp-specific signal to a random lwp.
1169 	 */
1170 	if (lp == NULL) {
1171 		lp = find_lwp_for_signal(p, sig);
1172 		if (lp && SIGISMEMBER(lp->lwp_sigmask, sig))
1173 			lp = NULL;
1174 	}
1175 
1176 	/*
1177 	 * Deliver to the process generically if (1) the signal is being
1178 	 * sent to any thread or (2) we could not find a thread to deliver
1179 	 * it to.
1180 	 */
1181 	if (lp == NULL) {
1182 		SIGADDSET(p->p_siglist, sig);
1183 		goto out;
1184 	}
1185 
1186 	/*
1187 	 * Deliver to a specific LWP whether it masks it or not.  It will
1188 	 * not be dispatched if masked but we must still deliver it.
1189 	 */
1190 	if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
1191 	    (p->p_flag & P_TRACED) == 0) {
1192 		p->p_nice = NZERO;
1193 	}
1194 
1195 	/*
1196 	 * If the process receives a STOP signal which indeed needs to
1197 	 * stop the process, do so.  If the process chose to catch the
1198 	 * signal, it will be treated like any other signal.
1199 	 */
1200 	if ((prop & SA_STOP) && action == SIG_DFL) {
1201 		/*
1202 		 * If a child holding parent blocked, stopping
1203 		 * could cause deadlock.  Take no action at this
1204 		 * time.
1205 		 */
1206 		if (p->p_flag & P_PPWAIT) {
1207 			SIGADDSET(p->p_siglist, sig);
1208 			goto out;
1209 		}
1210 
1211 		/*
1212 		 * Do not actually try to manipulate the process, but simply
1213 		 * stop it.  Lwps will stop as soon as they safely can.
1214 		 */
1215 		p->p_xstat = sig;
1216 		proc_stop(p);
1217 		goto out;
1218 	}
1219 
1220 	/*
1221 	 * If it is a CONT signal with default action, just ignore it.
1222 	 */
1223 	if ((prop & SA_CONT) && action == SIG_DFL)
1224 		goto out;
1225 
1226 	/*
1227 	 * Mark signal pending at this specific thread.
1228 	 */
1229 	SIGADDSET(lp->lwp_siglist, sig);
1230 
1231 	lwp_signotify(lp);
1232 
1233 out:
1234 	lwkt_reltoken(&proc_token);
1235 	crit_exit();
1236 }
1237 
1238 /*
1239  * proc_token must be held
1240  */
1241 static void
1242 lwp_signotify(struct lwp *lp)
1243 {
1244 	ASSERT_LWKT_TOKEN_HELD(&proc_token);
1245 	crit_enter();
1246 
1247 	if (lp->lwp_stat == LSSLEEP || lp->lwp_stat == LSSTOP) {
1248 		/*
1249 		 * Thread is in tsleep.
1250 		 */
1251 
1252 		/*
1253 		 * If the thread is sleeping uninterruptibly
1254 		 * we can't interrupt the sleep... the signal will
1255 		 * be noticed when the lwp returns through
1256 		 * trap() or syscall().
1257 		 *
1258 		 * Otherwise the signal can interrupt the sleep.
1259 		 *
1260 		 * If the process is traced, the lwp will handle the
1261 		 * tracing in issignal() when it returns to userland.
1262 		 */
1263 		if (lp->lwp_flag & LWP_SINTR) {
1264 			/*
1265 			 * Make runnable and break out of any tsleep as well.
1266 			 */
1267 			lp->lwp_flag |= LWP_BREAKTSLEEP;
1268 			setrunnable(lp);
1269 		}
1270 	} else {
1271 		/*
1272 		 * Otherwise the thread is running
1273 		 *
1274 		 * LSRUN does nothing with the signal, other than kicking
1275 		 * ourselves if we are running.
1276 		 * SZOMB and SIDL mean that it will either never be noticed,
1277 		 * or noticed very soon.
1278 		 *
1279 		 * Note that lwp_thread may be NULL or may not be completely
1280 		 * initialized if the process is in the SIDL or SZOMB state.
1281 		 *
1282 		 * For SMP we may have to forward the request to another cpu.
1283 		 * YYY the MP lock prevents the target process from moving
1284 		 * to another cpu, see kern/kern_switch.c
1285 		 *
1286 		 * If the target thread is waiting on its message port,
1287 		 * wakeup the target thread so it can check (or ignore)
1288 		 * the new signal.  YYY needs cleanup.
1289 		 */
1290 		if (lp == lwkt_preempted_proc()) {
1291 			signotify();
1292 		} else if (lp->lwp_stat == LSRUN) {
1293 			struct thread *td = lp->lwp_thread;
1294 			struct proc *p __debugvar = lp->lwp_proc;
1295 
1296 			KASSERT(td != NULL,
1297 			    ("pid %d/%d NULL lwp_thread stat %d flags %08x/%08x",
1298 			    p->p_pid, lp->lwp_tid, lp->lwp_stat,
1299 			    p->p_flag, lp->lwp_flag));
1300 
1301 			/*
1302 			 * To prevent a MP race with TDF_SINTR we must
1303 			 * schedule the thread on the correct cpu.
1304 			 */
1305 #ifdef SMP
1306 			if (td->td_gd != mycpu) {
1307 				LWPHOLD(lp);
1308 				lwkt_send_ipiq(td->td_gd, signotify_remote, lp);
1309 			} else
1310 #endif
1311 			if (td->td_flags & TDF_SINTR)
1312 				lwkt_schedule(td);
1313 		}
1314 	}
1315 	crit_exit();
1316 }
1317 
1318 #ifdef SMP
1319 
1320 /*
1321  * This function is called via an IPI.  We will be in a critical section but
1322  * the MP lock will NOT be held.  The passed lp will be held.
1323  *
1324  * We must essentially repeat the code at the end of lwp_signotify(),
1325  * in particular rechecking all races.  If we are still not on the
1326  * correct cpu we leave the lwp ref intact and continue the chase.
1327  *
1328  * XXX this may still not be entirely correct, since we are checking
1329  *     lwp_stat asynchronously.
1330  */
1331 static void
1332 signotify_remote(void *arg)
1333 {
1334 	struct lwp *lp = arg;
1335 	thread_t td;
1336 
1337 	if (lp == lwkt_preempted_proc()) {
1338 		signotify();
1339 	} else if (lp->lwp_stat == LSRUN) {
1340 		/*
1341 		 * To prevent a MP race with TDF_SINTR we must
1342 		 * schedule the thread on the correct cpu.
1343 		 */
1344 		td = lp->lwp_thread;
1345 		if (td->td_gd != mycpu) {
1346 			lwkt_send_ipiq(td->td_gd, signotify_remote, lp);
1347 			return;
1348 			/* NOT REACHED */
1349 		}
1350 		if (td->td_flags & TDF_SINTR)
1351 			lwkt_schedule(td);
1352 	}
1353 	LWPRELE(lp);
1354 }
1355 
1356 #endif
1357 
1358 /*
1359  * Caller must hold proc_token
1360  */
1361 void
1362 proc_stop(struct proc *p)
1363 {
1364 	struct lwp *lp;
1365 
1366 	ASSERT_LWKT_TOKEN_HELD(&proc_token);
1367 	crit_enter();
1368 
1369 	/* If somebody raced us, be happy with it */
1370 	if (p->p_stat == SSTOP || p->p_stat == SZOMB) {
1371 		crit_exit();
1372 		return;
1373 	}
1374 	p->p_stat = SSTOP;
1375 
1376 	FOREACH_LWP_IN_PROC(lp, p) {
1377 		switch (lp->lwp_stat) {
1378 		case LSSTOP:
1379 			/*
1380 			 * Do nothing, we are already counted in
1381 			 * p_nstopped.
1382 			 */
1383 			break;
1384 
1385 		case LSSLEEP:
1386 			/*
1387 			 * We're sleeping, but we will stop before
1388 			 * returning to userspace, so count us
1389 			 * as stopped as well.  We set LWP_WSTOP
1390 			 * to signal the lwp that it should not
1391 			 * increase p_nstopped when reaching tstop().
1392 			 */
1393 			if ((lp->lwp_flag & LWP_WSTOP) == 0) {
1394 				lp->lwp_flag |= LWP_WSTOP;
1395 				++p->p_nstopped;
1396 			}
1397 			break;
1398 
1399 		case LSRUN:
1400 			/*
1401 			 * We might notify ourself, but that's not
1402 			 * a problem.
1403 			 */
1404 			lwp_signotify(lp);
1405 			break;
1406 		}
1407 	}
1408 
1409 	if (p->p_nstopped == p->p_nthreads) {
1410 		p->p_flag &= ~P_WAITED;
1411 		wakeup(p->p_pptr);
1412 		if ((p->p_pptr->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1413 			ksignal(p->p_pptr, SIGCHLD);
1414 	}
1415 	crit_exit();
1416 }
1417 
1418 /*
1419  * Caller must hold proc_token
1420  */
1421 void
1422 proc_unstop(struct proc *p)
1423 {
1424 	struct lwp *lp;
1425 
1426 	ASSERT_LWKT_TOKEN_HELD(&proc_token);
1427 	crit_enter();
1428 
1429 	if (p->p_stat != SSTOP) {
1430 		crit_exit();
1431 		return;
1432 	}
1433 
1434 	p->p_stat = SACTIVE;
1435 
1436 	FOREACH_LWP_IN_PROC(lp, p) {
1437 		switch (lp->lwp_stat) {
1438 		case LSRUN:
1439 			/*
1440 			 * Uh?  Not stopped?  Well, I guess that's okay.
1441 			 */
1442 			if (bootverbose)
1443 				kprintf("proc_unstop: lwp %d/%d not sleeping\n",
1444 					p->p_pid, lp->lwp_tid);
1445 			break;
1446 
1447 		case LSSLEEP:
1448 			/*
1449 			 * Still sleeping.  Don't bother waking it up.
1450 			 * However, if this thread was counted as
1451 			 * stopped, undo this.
1452 			 *
1453 			 * Nevertheless we call setrunnable() so that it
1454 			 * will wake up in case a signal or timeout arrived
1455 			 * in the meantime.
1456 			 */
1457 			if (lp->lwp_flag & LWP_WSTOP) {
1458 				lp->lwp_flag &= ~LWP_WSTOP;
1459 				--p->p_nstopped;
1460 			} else {
1461 				if (bootverbose)
1462 					kprintf("proc_unstop: lwp %d/%d sleeping, not stopped\n",
1463 						p->p_pid, lp->lwp_tid);
1464 			}
1465 			/* FALLTHROUGH */
1466 
1467 		case LSSTOP:
1468 			setrunnable(lp);
1469 			break;
1470 
1471 		}
1472 	}
1473 	crit_exit();
1474 }
1475 
1476 /*
1477  * No requirements.
1478  */
1479 static int
1480 kern_sigtimedwait(sigset_t waitset, siginfo_t *info, struct timespec *timeout)
1481 {
1482 	sigset_t savedmask, set;
1483 	struct proc *p = curproc;
1484 	struct lwp *lp = curthread->td_lwp;
1485 	int error, sig, hz, timevalid = 0;
1486 	struct timespec rts, ets, ts;
1487 	struct timeval tv;
1488 
1489 	lwkt_gettoken(&proc_token);
1490 
1491 	error = 0;
1492 	sig = 0;
1493 	ets.tv_sec = 0;		/* silence compiler warning */
1494 	ets.tv_nsec = 0;	/* silence compiler warning */
1495 	SIG_CANTMASK(waitset);
1496 	savedmask = lp->lwp_sigmask;
1497 
1498 	if (timeout) {
1499 		if (timeout->tv_sec >= 0 && timeout->tv_nsec >= 0 &&
1500 		    timeout->tv_nsec < 1000000000) {
1501 			timevalid = 1;
1502 			getnanouptime(&rts);
1503 		 	ets = rts;
1504 			timespecadd(&ets, timeout);
1505 		}
1506 	}
1507 
1508 	for (;;) {
1509 		set = lwp_sigpend(lp);
1510 		SIGSETAND(set, waitset);
1511 		if ((sig = sig_ffs(&set)) != 0) {
1512 			SIGFILLSET(lp->lwp_sigmask);
1513 			SIGDELSET(lp->lwp_sigmask, sig);
1514 			SIG_CANTMASK(lp->lwp_sigmask);
1515 			sig = issignal(lp, 1);
1516 			/*
1517 			 * It may be a STOP signal, in the case, issignal
1518 			 * returns 0, because we may stop there, and new
1519 			 * signal can come in, we should restart if we got
1520 			 * nothing.
1521 			 */
1522 			if (sig == 0)
1523 				continue;
1524 			else
1525 				break;
1526 		}
1527 
1528 		/*
1529 		 * Previous checking got nothing, and we retried but still
1530 		 * got nothing, we should return the error status.
1531 		 */
1532 		if (error)
1533 			break;
1534 
1535 		/*
1536 		 * POSIX says this must be checked after looking for pending
1537 		 * signals.
1538 		 */
1539 		if (timeout) {
1540 			if (timevalid == 0) {
1541 				error = EINVAL;
1542 				break;
1543 			}
1544 			getnanouptime(&rts);
1545 			if (timespeccmp(&rts, &ets, >=)) {
1546 				error = EAGAIN;
1547 				break;
1548 			}
1549 			ts = ets;
1550 			timespecsub(&ts, &rts);
1551 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1552 			hz = tvtohz_high(&tv);
1553 		} else
1554 			hz = 0;
1555 
1556 		lp->lwp_sigmask = savedmask;
1557 		SIGSETNAND(lp->lwp_sigmask, waitset);
1558 		/*
1559 		 * We won't ever be woken up.  Instead, our sleep will
1560 		 * be broken in lwpsignal().
1561 		 */
1562 		error = tsleep(&p->p_sigacts, PCATCH, "sigwt", hz);
1563 		if (timeout) {
1564 			if (error == ERESTART) {
1565 				/* can not restart a timeout wait. */
1566 				error = EINTR;
1567 			} else if (error == EAGAIN) {
1568 				/* will calculate timeout by ourself. */
1569 				error = 0;
1570 			}
1571 		}
1572 		/* Retry ... */
1573 	}
1574 
1575 	lp->lwp_sigmask = savedmask;
1576 	if (sig) {
1577 		error = 0;
1578 		bzero(info, sizeof(*info));
1579 		info->si_signo = sig;
1580 		lwp_delsig(lp, sig);	/* take the signal! */
1581 
1582 		if (sig == SIGKILL) {
1583 			lwkt_reltoken(&proc_token);
1584 			sigexit(lp, sig);
1585 			/* NOT REACHED */
1586 		}
1587 	}
1588 
1589 	lwkt_reltoken(&proc_token);
1590 
1591 	return (error);
1592 }
1593 
1594 /*
1595  * MPALMOSTSAFE
1596  */
1597 int
1598 sys_sigtimedwait(struct sigtimedwait_args *uap)
1599 {
1600 	struct timespec ts;
1601 	struct timespec *timeout;
1602 	sigset_t set;
1603 	siginfo_t info;
1604 	int error;
1605 
1606 	if (uap->timeout) {
1607 		error = copyin(uap->timeout, &ts, sizeof(ts));
1608 		if (error)
1609 			return (error);
1610 		timeout = &ts;
1611 	} else {
1612 		timeout = NULL;
1613 	}
1614 	error = copyin(uap->set, &set, sizeof(set));
1615 	if (error)
1616 		return (error);
1617 	error = kern_sigtimedwait(set, &info, timeout);
1618 	if (error)
1619 		return (error);
1620  	if (uap->info)
1621 		error = copyout(&info, uap->info, sizeof(info));
1622 	/* Repost if we got an error. */
1623 	/*
1624 	 * XXX lwp
1625 	 *
1626 	 * This could transform a thread-specific signal to another
1627 	 * thread / process pending signal.
1628 	 */
1629 	if (error) {
1630 		ksignal(curproc, info.si_signo);
1631 	} else {
1632 		uap->sysmsg_result = info.si_signo;
1633 	}
1634 	return (error);
1635 }
1636 
1637 /*
1638  * MPALMOSTSAFE
1639  */
1640 int
1641 sys_sigwaitinfo(struct sigwaitinfo_args *uap)
1642 {
1643 	siginfo_t info;
1644 	sigset_t set;
1645 	int error;
1646 
1647 	error = copyin(uap->set, &set, sizeof(set));
1648 	if (error)
1649 		return (error);
1650 	error = kern_sigtimedwait(set, &info, NULL);
1651 	if (error)
1652 		return (error);
1653 	if (uap->info)
1654 		error = copyout(&info, uap->info, sizeof(info));
1655 	/* Repost if we got an error. */
1656 	/*
1657 	 * XXX lwp
1658 	 *
1659 	 * This could transform a thread-specific signal to another
1660 	 * thread / process pending signal.
1661 	 */
1662 	if (error) {
1663 		ksignal(curproc, info.si_signo);
1664 	} else {
1665 		uap->sysmsg_result = info.si_signo;
1666 	}
1667 	return (error);
1668 }
1669 
1670 /*
1671  * If the current process has received a signal that would interrupt a
1672  * system call, return EINTR or ERESTART as appropriate.
1673  */
1674 int
1675 iscaught(struct lwp *lp)
1676 {
1677 	struct proc *p = lp->lwp_proc;
1678 	int sig;
1679 
1680 	if (p) {
1681 		if ((sig = CURSIG(lp)) != 0) {
1682 			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1683 				return (EINTR);
1684 			return (ERESTART);
1685 		}
1686 	}
1687 	return(EWOULDBLOCK);
1688 }
1689 
1690 /*
1691  * If the current process has received a signal (should be caught or cause
1692  * termination, should interrupt current syscall), return the signal number.
1693  * Stop signals with default action are processed immediately, then cleared;
1694  * they aren't returned.  This is checked after each entry to the system for
1695  * a syscall or trap (though this can usually be done without calling issignal
1696  * by checking the pending signal masks in the CURSIG macro.) The normal call
1697  * sequence is
1698  *
1699  * This routine is called via CURSIG/__cursig and the MP lock might not be
1700  * held.  Obtain the MP lock for the duration of the operation.
1701  *
1702  *	while (sig = CURSIG(curproc))
1703  *		postsig(sig);
1704  */
1705 int
1706 issignal(struct lwp *lp, int maytrace)
1707 {
1708 	struct proc *p = lp->lwp_proc;
1709 	sigset_t mask;
1710 	int sig, prop;
1711 
1712 	lwkt_gettoken(&proc_token);
1713 
1714 	for (;;) {
1715 		int traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
1716 
1717 		/*
1718 		 * If this process is supposed to stop, stop this thread.
1719 		 */
1720 		if (p->p_stat == SSTOP)
1721 			tstop();
1722 
1723 		mask = lwp_sigpend(lp);
1724 		SIGSETNAND(mask, lp->lwp_sigmask);
1725 		if (p->p_flag & P_PPWAIT)
1726 			SIG_STOPSIGMASK(mask);
1727 		if (SIGISEMPTY(mask)) {		/* no signal to send */
1728 			lwkt_reltoken(&proc_token);
1729 			return (0);
1730 		}
1731 		sig = sig_ffs(&mask);
1732 
1733 		STOPEVENT(p, S_SIG, sig);
1734 
1735 		/*
1736 		 * We should see pending but ignored signals
1737 		 * only if P_TRACED was on when they were posted.
1738 		 */
1739 		if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
1740 			lwp_delsig(lp, sig);
1741 			continue;
1742 		}
1743 		if (maytrace && (p->p_flag & P_TRACED) && (p->p_flag & P_PPWAIT) == 0) {
1744 			/*
1745 			 * If traced, always stop, and stay stopped until
1746 			 * released by the parent.
1747 			 *
1748 			 * NOTE: SSTOP may get cleared during the loop,
1749 			 * but we do not re-notify the parent if we have
1750 			 * to loop several times waiting for the parent
1751 			 * to let us continue.
1752 			 *
1753 			 * XXX not sure if this is still true
1754 			 */
1755 			p->p_xstat = sig;
1756 			proc_stop(p);
1757 			do {
1758 				tstop();
1759 			} while (!trace_req(p) && (p->p_flag & P_TRACED));
1760 
1761 			/*
1762 			 * If parent wants us to take the signal,
1763 			 * then it will leave it in p->p_xstat;
1764 			 * otherwise we just look for signals again.
1765 			 */
1766 			lwp_delsig(lp, sig);	/* clear old signal */
1767 			sig = p->p_xstat;
1768 			if (sig == 0)
1769 				continue;
1770 
1771 			/*
1772 			 * Put the new signal into p_siglist.  If the
1773 			 * signal is being masked, look for other signals.
1774 			 *
1775 			 * XXX lwp might need a call to ksignal()
1776 			 */
1777 			SIGADDSET(p->p_siglist, sig);
1778 			if (SIGISMEMBER(lp->lwp_sigmask, sig))
1779 				continue;
1780 
1781 			/*
1782 			 * If the traced bit got turned off, go back up
1783 			 * to the top to rescan signals.  This ensures
1784 			 * that p_sig* and ps_sigact are consistent.
1785 			 */
1786 			if ((p->p_flag & P_TRACED) == 0)
1787 				continue;
1788 		}
1789 
1790 		prop = sigprop(sig);
1791 
1792 		/*
1793 		 * Decide whether the signal should be returned.
1794 		 * Return the signal's number, or fall through
1795 		 * to clear it from the pending mask.
1796 		 */
1797 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
1798 		case (intptr_t)SIG_DFL:
1799 			/*
1800 			 * Don't take default actions on system processes.
1801 			 */
1802 			if (p->p_pid <= 1) {
1803 #ifdef DIAGNOSTIC
1804 				/*
1805 				 * Are you sure you want to ignore SIGSEGV
1806 				 * in init? XXX
1807 				 */
1808 				kprintf("Process (pid %lu) got signal %d\n",
1809 					(u_long)p->p_pid, sig);
1810 #endif
1811 				break;		/* == ignore */
1812 			}
1813 
1814 			/*
1815 			 * Handle the in-kernel checkpoint action
1816 			 */
1817 			if (prop & SA_CKPT) {
1818 				checkpoint_signal_handler(lp);
1819 				break;
1820 			}
1821 
1822 			/*
1823 			 * If there is a pending stop signal to process
1824 			 * with default action, stop here,
1825 			 * then clear the signal.  However,
1826 			 * if process is member of an orphaned
1827 			 * process group, ignore tty stop signals.
1828 			 */
1829 			if (prop & SA_STOP) {
1830 				if (p->p_flag & P_TRACED ||
1831 		    		    (p->p_pgrp->pg_jobc == 0 &&
1832 				    prop & SA_TTYSTOP))
1833 					break;	/* == ignore */
1834 				p->p_xstat = sig;
1835 				proc_stop(p);
1836 				tstop();
1837 				break;
1838 			} else if (prop & SA_IGNORE) {
1839 				/*
1840 				 * Except for SIGCONT, shouldn't get here.
1841 				 * Default action is to ignore; drop it.
1842 				 */
1843 				break;		/* == ignore */
1844 			} else {
1845 				lwkt_reltoken(&proc_token);
1846 				return (sig);
1847 			}
1848 
1849 			/*NOTREACHED*/
1850 
1851 		case (intptr_t)SIG_IGN:
1852 			/*
1853 			 * Masking above should prevent us ever trying
1854 			 * to take action on an ignored signal other
1855 			 * than SIGCONT, unless process is traced.
1856 			 */
1857 			if ((prop & SA_CONT) == 0 &&
1858 			    (p->p_flag & P_TRACED) == 0)
1859 				kprintf("issignal\n");
1860 			break;		/* == ignore */
1861 
1862 		default:
1863 			/*
1864 			 * This signal has an action, let
1865 			 * postsig() process it.
1866 			 */
1867 			lwkt_reltoken(&proc_token);
1868 			return (sig);
1869 		}
1870 		lwp_delsig(lp, sig);		/* take the signal! */
1871 	}
1872 	/* NOTREACHED */
1873 }
1874 
1875 /*
1876  * Take the action for the specified signal
1877  * from the current set of pending signals.
1878  */
1879 void
1880 postsig(int sig)
1881 {
1882 	struct lwp *lp = curthread->td_lwp;
1883 	struct proc *p = lp->lwp_proc;
1884 	struct sigacts *ps = p->p_sigacts;
1885 	sig_t action;
1886 	sigset_t returnmask;
1887 	int code;
1888 
1889 	KASSERT(sig != 0, ("postsig"));
1890 
1891 	KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1892 
1893 	/*
1894 	 * If we are a virtual kernel running an emulated user process
1895 	 * context, switch back to the virtual kernel context before
1896 	 * trying to post the signal.
1897 	 */
1898 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1899 		struct trapframe *tf = lp->lwp_md.md_regs;
1900 		tf->tf_trapno = 0;
1901 		vkernel_trap(lp, tf);
1902 	}
1903 
1904 	lwp_delsig(lp, sig);
1905 	action = ps->ps_sigact[_SIG_IDX(sig)];
1906 #ifdef KTRACE
1907 	if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
1908 		ktrpsig(lp, sig, action, lp->lwp_flag & LWP_OLDMASK ?
1909 			&lp->lwp_oldsigmask : &lp->lwp_sigmask, 0);
1910 #endif
1911 	STOPEVENT(p, S_SIG, sig);
1912 
1913 	if (action == SIG_DFL) {
1914 		/*
1915 		 * Default action, where the default is to kill
1916 		 * the process.  (Other cases were ignored above.)
1917 		 */
1918 		sigexit(lp, sig);
1919 		/* NOTREACHED */
1920 	} else {
1921 		/*
1922 		 * If we get here, the signal must be caught.
1923 		 */
1924 		KASSERT(action != SIG_IGN && !SIGISMEMBER(lp->lwp_sigmask, sig),
1925 		    ("postsig action"));
1926 
1927 		crit_enter();
1928 
1929 		/*
1930 		 * Reset the signal handler if asked to
1931 		 */
1932 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
1933 			/*
1934 			 * See kern_sigaction() for origin of this code.
1935 			 */
1936 			SIGDELSET(p->p_sigcatch, sig);
1937 			if (sig != SIGCONT &&
1938 			    sigprop(sig) & SA_IGNORE)
1939 				SIGADDSET(p->p_sigignore, sig);
1940 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1941 		}
1942 
1943 		/*
1944 		 * Handle the mailbox case.  Copyout to the appropriate
1945 		 * location but do not generate a signal frame.  The system
1946 		 * call simply returns EINTR and the user is responsible for
1947 		 * polling the mailbox.
1948 		 */
1949 		if (SIGISMEMBER(ps->ps_sigmailbox, sig)) {
1950 			int sig_copy = sig;
1951 			copyout(&sig_copy, (void *)action, sizeof(int));
1952 			curproc->p_flag |= P_MAILBOX;
1953 			crit_exit();
1954 			goto done;
1955 		}
1956 
1957 		/*
1958 		 * Set the signal mask and calculate the mask to restore
1959 		 * when the signal function returns.
1960 		 *
1961 		 * Special case: user has done a sigsuspend.  Here the
1962 		 * current mask is not of interest, but rather the
1963 		 * mask from before the sigsuspend is what we want
1964 		 * restored after the signal processing is completed.
1965 		 */
1966 		if (lp->lwp_flag & LWP_OLDMASK) {
1967 			returnmask = lp->lwp_oldsigmask;
1968 			lp->lwp_flag &= ~LWP_OLDMASK;
1969 		} else {
1970 			returnmask = lp->lwp_sigmask;
1971 		}
1972 
1973 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
1974 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
1975 			SIGADDSET(lp->lwp_sigmask, sig);
1976 
1977 		crit_exit();
1978 		lp->lwp_ru.ru_nsignals++;
1979 		if (lp->lwp_sig != sig) {
1980 			code = 0;
1981 		} else {
1982 			code = lp->lwp_code;
1983 			lp->lwp_code = 0;
1984 			lp->lwp_sig = 0;
1985 		}
1986 		(*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
1987 	}
1988 done:
1989 	;
1990 }
1991 
1992 /*
1993  * Kill the current process for stated reason.
1994  */
1995 void
1996 killproc(struct proc *p, char *why)
1997 {
1998 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n",
1999 		p->p_pid, p->p_comm,
2000 		p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2001 	ksignal(p, SIGKILL);
2002 }
2003 
2004 /*
2005  * Force the current process to exit with the specified signal, dumping core
2006  * if appropriate.  We bypass the normal tests for masked and caught signals,
2007  * allowing unrecoverable failures to terminate the process without changing
2008  * signal state.  Mark the accounting record with the signal termination.
2009  * If dumping core, save the signal number for the debugger.  Calls exit and
2010  * does not return.
2011  *
2012  * This routine does not return.
2013  */
2014 void
2015 sigexit(struct lwp *lp, int sig)
2016 {
2017 	struct proc *p = lp->lwp_proc;
2018 
2019 	p->p_acflag |= AXSIG;
2020 	if (sigprop(sig) & SA_CORE) {
2021 		lp->lwp_sig = sig;
2022 		/*
2023 		 * Log signals which would cause core dumps
2024 		 * (Log as LOG_INFO to appease those who don't want
2025 		 * these messages.)
2026 		 * XXX : Todo, as well as euid, write out ruid too
2027 		 */
2028 		if (coredump(lp, sig) == 0)
2029 			sig |= WCOREFLAG;
2030 		if (kern_logsigexit)
2031 			log(LOG_INFO,
2032 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2033 			    p->p_pid, p->p_comm,
2034 			    p->p_ucred ? p->p_ucred->cr_uid : -1,
2035 			    sig &~ WCOREFLAG,
2036 			    sig & WCOREFLAG ? " (core dumped)" : "");
2037 	}
2038 	exit1(W_EXITCODE(0, sig));
2039 	/* NOTREACHED */
2040 }
2041 
2042 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
2043 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
2044 	      sizeof(corefilename), "process corefile name format string");
2045 
2046 /*
2047  * expand_name(name, uid, pid)
2048  * Expand the name described in corefilename, using name, uid, and pid.
2049  * corefilename is a kprintf-like string, with three format specifiers:
2050  *	%N	name of process ("name")
2051  *	%P	process id (pid)
2052  *	%U	user id (uid)
2053  * For example, "%N.core" is the default; they can be disabled completely
2054  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
2055  * This is controlled by the sysctl variable kern.corefile (see above).
2056  */
2057 
2058 static char *
2059 expand_name(const char *name, uid_t uid, pid_t pid)
2060 {
2061 	char *temp;
2062 	char buf[11];		/* Buffer for pid/uid -- max 4B */
2063 	int i, n;
2064 	char *format = corefilename;
2065 	size_t namelen;
2066 
2067 	temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
2068 	if (temp == NULL)
2069 		return NULL;
2070 	namelen = strlen(name);
2071 	for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
2072 		int l;
2073 		switch (format[i]) {
2074 		case '%':	/* Format character */
2075 			i++;
2076 			switch (format[i]) {
2077 			case '%':
2078 				temp[n++] = '%';
2079 				break;
2080 			case 'N':	/* process name */
2081 				if ((n + namelen) > MAXPATHLEN) {
2082 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2083 					    pid, name, uid, temp, name);
2084 					kfree(temp, M_TEMP);
2085 					return NULL;
2086 				}
2087 				memcpy(temp+n, name, namelen);
2088 				n += namelen;
2089 				break;
2090 			case 'P':	/* process id */
2091 				l = ksprintf(buf, "%u", pid);
2092 				if ((n + l) > MAXPATHLEN) {
2093 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2094 					    pid, name, uid, temp, name);
2095 					kfree(temp, M_TEMP);
2096 					return NULL;
2097 				}
2098 				memcpy(temp+n, buf, l);
2099 				n += l;
2100 				break;
2101 			case 'U':	/* user id */
2102 				l = ksprintf(buf, "%u", uid);
2103 				if ((n + l) > MAXPATHLEN) {
2104 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2105 					    pid, name, uid, temp, name);
2106 					kfree(temp, M_TEMP);
2107 					return NULL;
2108 				}
2109 				memcpy(temp+n, buf, l);
2110 				n += l;
2111 				break;
2112 			default:
2113 			  	log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
2114 			}
2115 			break;
2116 		default:
2117 			temp[n++] = format[i];
2118 		}
2119 	}
2120 	temp[n] = '\0';
2121 	return temp;
2122 }
2123 
2124 /*
2125  * Dump a process' core.  The main routine does some
2126  * policy checking, and creates the name of the coredump;
2127  * then it passes on a vnode and a size limit to the process-specific
2128  * coredump routine if there is one; if there _is not_ one, it returns
2129  * ENOSYS; otherwise it returns the error from the process-specific routine.
2130  *
2131  * The parameter `lp' is the lwp which triggered the coredump.
2132  */
2133 
2134 static int
2135 coredump(struct lwp *lp, int sig)
2136 {
2137 	struct proc *p = lp->lwp_proc;
2138 	struct vnode *vp;
2139 	struct ucred *cred = p->p_ucred;
2140 	struct flock lf;
2141 	struct nlookupdata nd;
2142 	struct vattr vattr;
2143 	int error, error1;
2144 	char *name;			/* name of corefile */
2145 	off_t limit;
2146 
2147 	STOPEVENT(p, S_CORE, 0);
2148 
2149 	if (((sugid_coredump == 0) && p->p_flag & P_SUGID) || do_coredump == 0)
2150 		return (EFAULT);
2151 
2152 	/*
2153 	 * Note that the bulk of limit checking is done after
2154 	 * the corefile is created.  The exception is if the limit
2155 	 * for corefiles is 0, in which case we don't bother
2156 	 * creating the corefile at all.  This layout means that
2157 	 * a corefile is truncated instead of not being created,
2158 	 * if it is larger than the limit.
2159 	 */
2160 	limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
2161 	if (limit == 0)
2162 		return EFBIG;
2163 
2164 	name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
2165 	if (name == NULL)
2166 		return (EINVAL);
2167 	error = nlookup_init(&nd, name, UIO_SYSSPACE, NLC_LOCKVP);
2168 	if (error == 0)
2169 		error = vn_open(&nd, NULL, O_CREAT | FWRITE | O_NOFOLLOW, S_IRUSR | S_IWUSR);
2170 	kfree(name, M_TEMP);
2171 	if (error) {
2172 		nlookup_done(&nd);
2173 		return (error);
2174 	}
2175 	vp = nd.nl_open_vp;
2176 	nd.nl_open_vp = NULL;
2177 	nlookup_done(&nd);
2178 
2179 	vn_unlock(vp);
2180 	lf.l_whence = SEEK_SET;
2181 	lf.l_start = 0;
2182 	lf.l_len = 0;
2183 	lf.l_type = F_WRLCK;
2184 	error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, 0);
2185 	if (error)
2186 		goto out2;
2187 
2188 	/* Don't dump to non-regular files or files with links. */
2189 	if (vp->v_type != VREG ||
2190 	    VOP_GETATTR(vp, &vattr) || vattr.va_nlink != 1) {
2191 		error = EFAULT;
2192 		goto out1;
2193 	}
2194 
2195 	/* Don't dump to files current user does not own */
2196 	if (vattr.va_uid != p->p_ucred->cr_uid) {
2197 		error = EFAULT;
2198 		goto out1;
2199 	}
2200 
2201 	VATTR_NULL(&vattr);
2202 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2203 	vattr.va_size = 0;
2204 	VOP_SETATTR(vp, &vattr, cred);
2205 	p->p_acflag |= ACORE;
2206 	vn_unlock(vp);
2207 
2208 	error = p->p_sysent->sv_coredump ?
2209 		  p->p_sysent->sv_coredump(lp, sig, vp, limit) : ENOSYS;
2210 
2211 out1:
2212 	lf.l_type = F_UNLCK;
2213 	VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, 0);
2214 out2:
2215 	error1 = vn_close(vp, FWRITE);
2216 	if (error == 0)
2217 		error = error1;
2218 	return (error);
2219 }
2220 
2221 /*
2222  * Nonexistent system call-- signal process (may want to handle it).
2223  * Flag error in case process won't see signal immediately (blocked or ignored).
2224  *
2225  * MPALMOSTSAFE
2226  */
2227 /* ARGSUSED */
2228 int
2229 sys_nosys(struct nosys_args *args)
2230 {
2231 	lwpsignal(curproc, curthread->td_lwp, SIGSYS);
2232 	return (EINVAL);
2233 }
2234 
2235 /*
2236  * Send a SIGIO or SIGURG signal to a process or process group using
2237  * stored credentials rather than those of the current process.
2238  */
2239 void
2240 pgsigio(struct sigio *sigio, int sig, int checkctty)
2241 {
2242 	if (sigio == NULL)
2243 		return;
2244 
2245 	if (sigio->sio_pgid > 0) {
2246 		if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
2247 		             sigio->sio_proc))
2248 			ksignal(sigio->sio_proc, sig);
2249 	} else if (sigio->sio_pgid < 0) {
2250 		struct proc *p;
2251 
2252 		lockmgr(&sigio->sio_pgrp->pg_lock, LK_EXCLUSIVE);
2253 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
2254 			if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
2255 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
2256 				ksignal(p, sig);
2257 		}
2258 		lockmgr(&sigio->sio_pgrp->pg_lock, LK_RELEASE);
2259 	}
2260 }
2261 
2262 static int
2263 filt_sigattach(struct knote *kn)
2264 {
2265 	struct proc *p = curproc;
2266 
2267 	kn->kn_ptr.p_proc = p;
2268 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
2269 
2270 	/* XXX lock the proc here while adding to the list? */
2271 	knote_insert(&p->p_klist, kn);
2272 
2273 	return (0);
2274 }
2275 
2276 static void
2277 filt_sigdetach(struct knote *kn)
2278 {
2279 	struct proc *p = kn->kn_ptr.p_proc;
2280 
2281 	knote_remove(&p->p_klist, kn);
2282 }
2283 
2284 /*
2285  * signal knotes are shared with proc knotes, so we apply a mask to
2286  * the hint in order to differentiate them from process hints.  This
2287  * could be avoided by using a signal-specific knote list, but probably
2288  * isn't worth the trouble.
2289  */
2290 static int
2291 filt_signal(struct knote *kn, long hint)
2292 {
2293 	if (hint & NOTE_SIGNAL) {
2294 		hint &= ~NOTE_SIGNAL;
2295 
2296 		if (kn->kn_id == hint)
2297 			kn->kn_data++;
2298 	}
2299 	return (kn->kn_data != 0);
2300 }
2301