xref: /dflybsd-src/sys/kern/kern_sig.c (revision 2702099d6065f293e5ed6ff3c0fa181879bc1a1d)
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  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_sig.c	8.7 (Berkeley) 4/18/94
35  * $FreeBSD: src/sys/kern/kern_sig.c,v 1.72.2.17 2003/05/16 16:34:34 obrien Exp $
36  */
37 
38 #include "opt_ktrace.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/sysproto.h>
44 #include <sys/signalvar.h>
45 #include <sys/resourcevar.h>
46 #include <sys/vnode.h>
47 #include <sys/event.h>
48 #include <sys/proc.h>
49 #include <sys/nlookup.h>
50 #include <sys/pioctl.h>
51 #include <sys/acct.h>
52 #include <sys/fcntl.h>
53 #include <sys/lock.h>
54 #include <sys/wait.h>
55 #include <sys/ktrace.h>
56 #include <sys/syslog.h>
57 #include <sys/stat.h>
58 #include <sys/sysent.h>
59 #include <sys/sysctl.h>
60 #include <sys/malloc.h>
61 #include <sys/interrupt.h>
62 #include <sys/unistd.h>
63 #include <sys/kern_syscall.h>
64 #include <sys/vkernel.h>
65 
66 #include <sys/signal2.h>
67 #include <sys/thread2.h>
68 #include <sys/spinlock2.h>
69 
70 #include <machine/cpu.h>
71 #include <machine/smp.h>
72 
73 static int	coredump(struct lwp *, int);
74 static char	*expand_name(const char *, uid_t, pid_t);
75 static int	dokillpg(int sig, int pgid, int all);
76 static int	sig_ffs(sigset_t *set);
77 static int	sigprop(int sig);
78 static void	lwp_signotify(struct lwp *lp);
79 static void	lwp_signotify_remote(void *arg);
80 static int	kern_sigtimedwait(sigset_t set, siginfo_t *info,
81 		    struct timespec *timeout);
82 
83 static int	filt_sigattach(struct knote *kn);
84 static void	filt_sigdetach(struct knote *kn);
85 static int	filt_signal(struct knote *kn, long hint);
86 
87 struct filterops sig_filtops =
88 	{ 0, filt_sigattach, filt_sigdetach, filt_signal };
89 
90 static int	kern_logsigexit = 1;
91 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
92     &kern_logsigexit, 0,
93     "Log processes quitting on abnormal signals to syslog(3)");
94 
95 /*
96  * Can process p, with pcred pc, send the signal sig to process q?
97  */
98 #define CANSIGNAL(q, sig) \
99 	(!p_trespass(curproc->p_ucred, (q)->p_ucred) || \
100 	((sig) == SIGCONT && (q)->p_session == curproc->p_session))
101 
102 /*
103  * Policy -- Can real uid ruid with ucred uc send a signal to process q?
104  */
105 #define CANSIGIO(ruid, uc, q) \
106 	((uc)->cr_uid == 0 || \
107 	    (ruid) == (q)->p_ucred->cr_ruid || \
108 	    (uc)->cr_uid == (q)->p_ucred->cr_ruid || \
109 	    (ruid) == (q)->p_ucred->cr_uid || \
110 	    (uc)->cr_uid == (q)->p_ucred->cr_uid)
111 
112 int sugid_coredump;
113 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RW,
114 	&sugid_coredump, 0, "Enable coredumping set user/group ID processes");
115 
116 static int	do_coredump = 1;
117 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
118 	&do_coredump, 0, "Enable/Disable coredumps");
119 
120 /*
121  * Signal properties and actions.
122  * The array below categorizes the signals and their default actions
123  * according to the following properties:
124  */
125 #define	SA_KILL		0x01		/* terminates process by default */
126 #define	SA_CORE		0x02		/* ditto and coredumps */
127 #define	SA_STOP		0x04		/* suspend process */
128 #define	SA_TTYSTOP	0x08		/* ditto, from tty */
129 #define	SA_IGNORE	0x10		/* ignore by default */
130 #define	SA_CONT		0x20		/* continue if suspended */
131 #define	SA_CANTMASK	0x40		/* non-maskable, catchable */
132 #define SA_CKPT         0x80            /* checkpoint process */
133 
134 
135 static int sigproptbl[NSIG] = {
136         SA_KILL,                /* SIGHUP */
137         SA_KILL,                /* SIGINT */
138         SA_KILL|SA_CORE,        /* SIGQUIT */
139         SA_KILL|SA_CORE,        /* SIGILL */
140         SA_KILL|SA_CORE,        /* SIGTRAP */
141         SA_KILL|SA_CORE,        /* SIGABRT */
142         SA_KILL|SA_CORE,        /* SIGEMT */
143         SA_KILL|SA_CORE,        /* SIGFPE */
144         SA_KILL,                /* SIGKILL */
145         SA_KILL|SA_CORE,        /* SIGBUS */
146         SA_KILL|SA_CORE,        /* SIGSEGV */
147         SA_KILL|SA_CORE,        /* SIGSYS */
148         SA_KILL,                /* SIGPIPE */
149         SA_KILL,                /* SIGALRM */
150         SA_KILL,                /* SIGTERM */
151         SA_IGNORE,              /* SIGURG */
152         SA_STOP,                /* SIGSTOP */
153         SA_STOP|SA_TTYSTOP,     /* SIGTSTP */
154         SA_IGNORE|SA_CONT,      /* SIGCONT */
155         SA_IGNORE,              /* SIGCHLD */
156         SA_STOP|SA_TTYSTOP,     /* SIGTTIN */
157         SA_STOP|SA_TTYSTOP,     /* SIGTTOU */
158         SA_IGNORE,              /* SIGIO */
159         SA_KILL,                /* SIGXCPU */
160         SA_KILL,                /* SIGXFSZ */
161         SA_KILL,                /* SIGVTALRM */
162         SA_KILL,                /* SIGPROF */
163         SA_IGNORE,              /* SIGWINCH  */
164         SA_IGNORE,              /* SIGINFO */
165         SA_KILL,                /* SIGUSR1 */
166         SA_KILL,                /* SIGUSR2 */
167 	SA_IGNORE,              /* SIGTHR */
168 	SA_CKPT,                /* SIGCKPT */
169 	SA_KILL|SA_CKPT,        /* SIGCKPTEXIT */
170 	SA_IGNORE,
171 	SA_IGNORE,
172 	SA_IGNORE,
173 	SA_IGNORE,
174 	SA_IGNORE,
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 
201 };
202 
203 static __inline int
204 sigprop(int sig)
205 {
206 
207 	if (sig > 0 && sig < NSIG)
208 		return (sigproptbl[_SIG_IDX(sig)]);
209 	return (0);
210 }
211 
212 static __inline int
213 sig_ffs(sigset_t *set)
214 {
215 	int i;
216 
217 	for (i = 0; i < _SIG_WORDS; i++)
218 		if (set->__bits[i])
219 			return (ffs(set->__bits[i]) + (i * 32));
220 	return (0);
221 }
222 
223 /*
224  * No requirements.
225  */
226 int
227 kern_sigaction(int sig, struct sigaction *act, struct sigaction *oact)
228 {
229 	struct thread *td = curthread;
230 	struct proc *p = td->td_proc;
231 	struct lwp *lp;
232 	struct sigacts *ps = p->p_sigacts;
233 
234 	if (sig <= 0 || sig > _SIG_MAXSIG)
235 		return (EINVAL);
236 
237 	lwkt_gettoken(&p->p_token);
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 (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDSTOP)
254 			oact->sa_flags |= SA_NOCLDSTOP;
255 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDWAIT)
256 			oact->sa_flags |= SA_NOCLDWAIT;
257 	}
258 	if (act) {
259 		/*
260 		 * Check for invalid requests.  KILL and STOP cannot be
261 		 * caught.
262 		 */
263 		if (sig == SIGKILL || sig == SIGSTOP) {
264 			if (act->sa_handler != SIG_DFL) {
265 				lwkt_reltoken(&p->p_token);
266 				return (EINVAL);
267 			}
268 		}
269 
270 		/*
271 		 * Change setting atomically.
272 		 */
273 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
274 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
275 		if (act->sa_flags & SA_SIGINFO) {
276 			ps->ps_sigact[_SIG_IDX(sig)] =
277 			    (__sighandler_t *)act->sa_sigaction;
278 			SIGADDSET(ps->ps_siginfo, sig);
279 		} else {
280 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
281 			SIGDELSET(ps->ps_siginfo, sig);
282 		}
283 		if (!(act->sa_flags & SA_RESTART))
284 			SIGADDSET(ps->ps_sigintr, sig);
285 		else
286 			SIGDELSET(ps->ps_sigintr, sig);
287 		if (act->sa_flags & SA_ONSTACK)
288 			SIGADDSET(ps->ps_sigonstack, sig);
289 		else
290 			SIGDELSET(ps->ps_sigonstack, sig);
291 		if (act->sa_flags & SA_RESETHAND)
292 			SIGADDSET(ps->ps_sigreset, sig);
293 		else
294 			SIGDELSET(ps->ps_sigreset, sig);
295 		if (act->sa_flags & SA_NODEFER)
296 			SIGADDSET(ps->ps_signodefer, sig);
297 		else
298 			SIGDELSET(ps->ps_signodefer, sig);
299 		if (sig == SIGCHLD) {
300 			if (act->sa_flags & SA_NOCLDSTOP)
301 				p->p_sigacts->ps_flag |= PS_NOCLDSTOP;
302 			else
303 				p->p_sigacts->ps_flag &= ~PS_NOCLDSTOP;
304 			if (act->sa_flags & SA_NOCLDWAIT) {
305 				/*
306 				 * Paranoia: since SA_NOCLDWAIT is implemented
307 				 * by reparenting the dying child to PID 1 (and
308 				 * trust it to reap the zombie), PID 1 itself
309 				 * is forbidden to set SA_NOCLDWAIT.
310 				 */
311 				if (p->p_pid == 1)
312 					p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
313 				else
314 					p->p_sigacts->ps_flag |= PS_NOCLDWAIT;
315 			} else {
316 				p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
317 			}
318 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
319 				ps->ps_flag |= PS_CLDSIGIGN;
320 			else
321 				ps->ps_flag &= ~PS_CLDSIGIGN;
322 		}
323 		/*
324 		 * Set bit in p_sigignore for signals that are set to SIG_IGN,
325 		 * and for signals set to SIG_DFL where the default is to
326 		 * ignore. However, don't put SIGCONT in p_sigignore, as we
327 		 * have to restart the process.
328 		 *
329 		 * Also remove the signal from the process and lwp signal
330 		 * list.
331 		 */
332 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
333 		    (sigprop(sig) & SA_IGNORE &&
334 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
335 			SIGDELSET(p->p_siglist, sig);
336 			FOREACH_LWP_IN_PROC(lp, p) {
337 				spin_lock(&lp->lwp_spin);
338 				SIGDELSET(lp->lwp_siglist, sig);
339 				spin_unlock(&lp->lwp_spin);
340 			}
341 			if (sig != SIGCONT) {
342 				/* easier in ksignal */
343 				SIGADDSET(p->p_sigignore, sig);
344 			}
345 			SIGDELSET(p->p_sigcatch, sig);
346 		} else {
347 			SIGDELSET(p->p_sigignore, sig);
348 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
349 				SIGDELSET(p->p_sigcatch, sig);
350 			else
351 				SIGADDSET(p->p_sigcatch, sig);
352 		}
353 	}
354 	lwkt_reltoken(&p->p_token);
355 	return (0);
356 }
357 
358 int
359 sys_sigaction(struct sigaction_args *uap)
360 {
361 	struct sigaction act, oact;
362 	struct sigaction *actp, *oactp;
363 	int error;
364 
365 	actp = (uap->act != NULL) ? &act : NULL;
366 	oactp = (uap->oact != NULL) ? &oact : NULL;
367 	if (actp) {
368 		error = copyin(uap->act, actp, sizeof(act));
369 		if (error)
370 			return (error);
371 	}
372 	error = kern_sigaction(uap->sig, actp, oactp);
373 	if (oactp && !error) {
374 		error = copyout(oactp, uap->oact, sizeof(oact));
375 	}
376 	return (error);
377 }
378 
379 /*
380  * Initialize signal state for process 0;
381  * set to ignore signals that are ignored by default.
382  */
383 void
384 siginit(struct proc *p)
385 {
386 	int i;
387 
388 	for (i = 1; i <= NSIG; i++)
389 		if (sigprop(i) & SA_IGNORE && i != SIGCONT)
390 			SIGADDSET(p->p_sigignore, i);
391 }
392 
393 /*
394  * Reset signals for an exec of the specified process.
395  */
396 void
397 execsigs(struct proc *p)
398 {
399 	struct sigacts *ps = p->p_sigacts;
400 	struct lwp *lp;
401 	int sig;
402 
403 	lp = ONLY_LWP_IN_PROC(p);
404 
405 	/*
406 	 * Reset caught signals.  Held signals remain held
407 	 * through p_sigmask (unless they were caught,
408 	 * and are now ignored by default).
409 	 */
410 	while (SIGNOTEMPTY(p->p_sigcatch)) {
411 		sig = sig_ffs(&p->p_sigcatch);
412 		SIGDELSET(p->p_sigcatch, sig);
413 		if (sigprop(sig) & SA_IGNORE) {
414 			if (sig != SIGCONT)
415 				SIGADDSET(p->p_sigignore, sig);
416 			SIGDELSET(p->p_siglist, sig);
417 			/* don't need spinlock */
418 			SIGDELSET(lp->lwp_siglist, sig);
419 		}
420 		ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
421 	}
422 
423 	/*
424 	 * Reset stack state to the user stack.
425 	 * Clear set of signals caught on the signal stack.
426 	 */
427 	lp->lwp_sigstk.ss_flags = SS_DISABLE;
428 	lp->lwp_sigstk.ss_size = 0;
429 	lp->lwp_sigstk.ss_sp = NULL;
430 	lp->lwp_flags &= ~LWP_ALTSTACK;
431 	/*
432 	 * Reset no zombies if child dies flag as Solaris does.
433 	 */
434 	p->p_sigacts->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
435 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
436 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
437 }
438 
439 /*
440  * kern_sigprocmask() - MP SAFE ONLY IF p == curproc
441  *
442  *	Manipulate signal mask.  This routine is MP SAFE *ONLY* if
443  *	p == curproc.
444  */
445 int
446 kern_sigprocmask(int how, sigset_t *set, sigset_t *oset)
447 {
448 	struct thread *td = curthread;
449 	struct lwp *lp = td->td_lwp;
450 	struct proc *p = td->td_proc;
451 	int error;
452 
453 	lwkt_gettoken(&p->p_token);
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 
478 	lwkt_reltoken(&p->p_token);
479 
480 	return (error);
481 }
482 
483 /*
484  * sigprocmask()
485  *
486  * MPSAFE
487  */
488 int
489 sys_sigprocmask(struct sigprocmask_args *uap)
490 {
491 	sigset_t set, oset;
492 	sigset_t *setp, *osetp;
493 	int error;
494 
495 	setp = (uap->set != NULL) ? &set : NULL;
496 	osetp = (uap->oset != NULL) ? &oset : NULL;
497 	if (setp) {
498 		error = copyin(uap->set, setp, sizeof(set));
499 		if (error)
500 			return (error);
501 	}
502 	error = kern_sigprocmask(uap->how, setp, osetp);
503 	if (osetp && !error) {
504 		error = copyout(osetp, uap->oset, sizeof(oset));
505 	}
506 	return (error);
507 }
508 
509 /*
510  * MPSAFE
511  */
512 int
513 kern_sigpending(struct __sigset *set)
514 {
515 	struct lwp *lp = curthread->td_lwp;
516 
517 	*set = lwp_sigpend(lp);
518 
519 	return (0);
520 }
521 
522 /*
523  * MPSAFE
524  */
525 int
526 sys_sigpending(struct sigpending_args *uap)
527 {
528 	sigset_t set;
529 	int error;
530 
531 	error = kern_sigpending(&set);
532 
533 	if (error == 0)
534 		error = copyout(&set, uap->set, sizeof(set));
535 	return (error);
536 }
537 
538 /*
539  * Suspend process until signal, providing mask to be set
540  * in the meantime.
541  *
542  * MPSAFE
543  */
544 int
545 kern_sigsuspend(struct __sigset *set)
546 {
547 	struct thread *td = curthread;
548 	struct lwp *lp = td->td_lwp;
549 	struct proc *p = td->td_proc;
550 	struct sigacts *ps = p->p_sigacts;
551 
552 	/*
553 	 * When returning from sigsuspend, we want
554 	 * the old mask to be restored after the
555 	 * signal handler has finished.  Thus, we
556 	 * save it here and mark the sigacts structure
557 	 * to indicate this.
558 	 */
559 	lp->lwp_oldsigmask = lp->lwp_sigmask;
560 	lp->lwp_flags |= LWP_OLDMASK;
561 
562 	SIG_CANTMASK(*set);
563 	lp->lwp_sigmask = *set;
564 	while (tsleep(ps, PCATCH, "pause", 0) == 0)
565 		/* void */;
566 	/* always return EINTR rather than ERESTART... */
567 	return (EINTR);
568 }
569 
570 /*
571  * Note nonstandard calling convention: libc stub passes mask, not
572  * pointer, to save a copyin.
573  *
574  * MPSAFE
575  */
576 int
577 sys_sigsuspend(struct sigsuspend_args *uap)
578 {
579 	sigset_t mask;
580 	int error;
581 
582 	error = copyin(uap->sigmask, &mask, sizeof(mask));
583 	if (error)
584 		return (error);
585 
586 	error = kern_sigsuspend(&mask);
587 
588 	return (error);
589 }
590 
591 /*
592  * MPSAFE
593  */
594 int
595 kern_sigaltstack(struct sigaltstack *ss, struct sigaltstack *oss)
596 {
597 	struct thread *td = curthread;
598 	struct lwp *lp = td->td_lwp;
599 	struct proc *p = td->td_proc;
600 
601 	if ((lp->lwp_flags & LWP_ALTSTACK) == 0)
602 		lp->lwp_sigstk.ss_flags |= SS_DISABLE;
603 
604 	if (oss)
605 		*oss = lp->lwp_sigstk;
606 
607 	if (ss) {
608 		if (ss->ss_flags & SS_DISABLE) {
609 			if (lp->lwp_sigstk.ss_flags & SS_ONSTACK)
610 				return (EINVAL);
611 			lp->lwp_flags &= ~LWP_ALTSTACK;
612 			lp->lwp_sigstk.ss_flags = ss->ss_flags;
613 		} else {
614 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
615 				return (ENOMEM);
616 			lp->lwp_flags |= LWP_ALTSTACK;
617 			lp->lwp_sigstk = *ss;
618 		}
619 	}
620 
621 	return (0);
622 }
623 
624 /*
625  * MPSAFE
626  */
627 int
628 sys_sigaltstack(struct sigaltstack_args *uap)
629 {
630 	stack_t ss, oss;
631 	int error;
632 
633 	if (uap->ss) {
634 		error = copyin(uap->ss, &ss, sizeof(ss));
635 		if (error)
636 			return (error);
637 	}
638 
639 	error = kern_sigaltstack(uap->ss ? &ss : NULL,
640 	    uap->oss ? &oss : NULL);
641 
642 	if (error == 0 && uap->oss)
643 		error = copyout(&oss, uap->oss, sizeof(*uap->oss));
644 	return (error);
645 }
646 
647 /*
648  * Common code for kill process group/broadcast kill.
649  * cp is calling process.
650  */
651 struct killpg_info {
652 	int nfound;
653 	int sig;
654 };
655 
656 static int killpg_all_callback(struct proc *p, void *data);
657 
658 static int
659 dokillpg(int sig, int pgid, int all)
660 {
661 	struct killpg_info info;
662 	struct proc *cp = curproc;
663 	struct proc *p;
664 	struct pgrp *pgrp;
665 
666 	info.nfound = 0;
667 	info.sig = sig;
668 
669 	if (all) {
670 		/*
671 		 * broadcast
672 		 */
673 		allproc_scan(killpg_all_callback, &info);
674 	} else {
675 		if (pgid == 0) {
676 			/*
677 			 * zero pgid means send to my process group.
678 			 */
679 			pgrp = cp->p_pgrp;
680 			pgref(pgrp);
681 		} else {
682 			pgrp = pgfind(pgid);
683 			if (pgrp == NULL)
684 				return (ESRCH);
685 		}
686 
687 		/*
688 		 * Must interlock all signals against fork
689 		 */
690 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
691 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
692 			if (p->p_pid <= 1 ||
693 			    p->p_stat == SZOMB ||
694 			    (p->p_flags & P_SYSTEM) ||
695 			    !CANSIGNAL(p, sig)) {
696 				continue;
697 			}
698 			++info.nfound;
699 			if (sig)
700 				ksignal(p, sig);
701 		}
702 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
703 		pgrel(pgrp);
704 	}
705 	return (info.nfound ? 0 : ESRCH);
706 }
707 
708 static int
709 killpg_all_callback(struct proc *p, void *data)
710 {
711 	struct killpg_info *info = data;
712 
713 	if (p->p_pid <= 1 || (p->p_flags & P_SYSTEM) ||
714 	    p == curproc || !CANSIGNAL(p, info->sig)) {
715 		return (0);
716 	}
717 	++info->nfound;
718 	if (info->sig)
719 		ksignal(p, info->sig);
720 	return(0);
721 }
722 
723 /*
724  * Send a general signal to a process or LWPs within that process.
725  *
726  * Note that new signals cannot be sent if a process is exiting or already
727  * a zombie, but we return success anyway as userland is likely to not handle
728  * the race properly.
729  *
730  * No requirements.
731  */
732 int
733 kern_kill(int sig, pid_t pid, lwpid_t tid)
734 {
735 	int t;
736 
737 	if ((u_int)sig > _SIG_MAXSIG)
738 		return (EINVAL);
739 
740 	lwkt_gettoken(&proc_token);
741 
742 	if (pid > 0) {
743 		struct proc *p;
744 		struct lwp *lp = NULL;
745 
746 		/*
747 		 * Send a signal to a single process.  If the kill() is
748 		 * racing an exiting process which has not yet been reaped
749 		 * act as though the signal was delivered successfully but
750 		 * don't actually try to deliver the signal.
751 		 */
752 		if ((p = pfind(pid)) == NULL) {
753 			if ((p = zpfind(pid)) == NULL) {
754 				lwkt_reltoken(&proc_token);
755 				return (ESRCH);
756 			}
757 			lwkt_reltoken(&proc_token);
758 			PRELE(p);
759 			return (0);
760 		}
761 		lwkt_gettoken(&p->p_token);
762 		if (!CANSIGNAL(p, sig)) {
763 			lwkt_reltoken(&p->p_token);
764 			PRELE(p);
765 			lwkt_reltoken(&proc_token);
766 			return (EPERM);
767 		}
768 
769 		/*
770 		 * NOP if the process is exiting.  Note that lwpsignal() is
771 		 * called directly with P_WEXIT set to kill individual LWPs
772 		 * during exit, which is allowed.
773 		 */
774 		if (p->p_flags & P_WEXIT) {
775 			lwkt_reltoken(&p->p_token);
776 			PRELE(p);
777 			lwkt_reltoken(&proc_token);
778 			return (0);
779 		}
780 		if (tid != -1) {
781 			lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, tid);
782 			if (lp == NULL) {
783 				lwkt_reltoken(&p->p_token);
784 				PRELE(p);
785 				lwkt_reltoken(&proc_token);
786 				return (ESRCH);
787 			}
788 		}
789 		if (sig)
790 			lwpsignal(p, lp, sig);
791 		lwkt_reltoken(&p->p_token);
792 		PRELE(p);
793 		lwkt_reltoken(&proc_token);
794 		return (0);
795 	}
796 
797 	/*
798 	 * If we come here, pid is a special broadcast pid.
799 	 * This doesn't mix with a tid.
800 	 */
801 	if (tid != -1) {
802 		lwkt_reltoken(&proc_token);
803 		return (EINVAL);
804 	}
805 	switch (pid) {
806 	case -1:		/* broadcast signal */
807 		t = (dokillpg(sig, 0, 1));
808 		break;
809 	case 0:			/* signal own process group */
810 		t = (dokillpg(sig, 0, 0));
811 		break;
812 	default:		/* negative explicit process group */
813 		t = (dokillpg(sig, -pid, 0));
814 		break;
815 	}
816 	lwkt_reltoken(&proc_token);
817 	return t;
818 }
819 
820 int
821 sys_kill(struct kill_args *uap)
822 {
823 	int error;
824 
825 	error = kern_kill(uap->signum, uap->pid, -1);
826 	return (error);
827 }
828 
829 int
830 sys_lwp_kill(struct lwp_kill_args *uap)
831 {
832 	int error;
833 	pid_t pid = uap->pid;
834 
835 	/*
836 	 * A tid is mandatory for lwp_kill(), otherwise
837 	 * you could simply use kill().
838 	 */
839 	if (uap->tid == -1)
840 		return (EINVAL);
841 
842 	/*
843 	 * To save on a getpid() function call for intra-process
844 	 * signals, pid == -1 means current process.
845 	 */
846 	if (pid == -1)
847 		pid = curproc->p_pid;
848 
849 	error = kern_kill(uap->signum, pid, uap->tid);
850 	return (error);
851 }
852 
853 /*
854  * Send a signal to a process group.
855  */
856 void
857 gsignal(int pgid, int sig)
858 {
859 	struct pgrp *pgrp;
860 
861 	if (pgid && (pgrp = pgfind(pgid)))
862 		pgsignal(pgrp, sig, 0);
863 }
864 
865 /*
866  * Send a signal to a process group.  If checktty is 1,
867  * limit to members which have a controlling terminal.
868  *
869  * pg_lock interlocks against a fork that might be in progress, to
870  * ensure that the new child process picks up the signal.
871  */
872 void
873 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
874 {
875 	struct proc *p;
876 
877 	/*
878 	 * Must interlock all signals against fork
879 	 */
880 	if (pgrp) {
881 		pgref(pgrp);
882 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
883 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
884 			if (checkctty == 0 || p->p_flags & P_CONTROLT)
885 				ksignal(p, sig);
886 		}
887 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
888 		pgrel(pgrp);
889 	}
890 }
891 
892 /*
893  * Send a signal caused by a trap to the current lwp.  If it will be caught
894  * immediately, deliver it with correct code.  Otherwise, post it normally.
895  *
896  * These signals may ONLY be delivered to the specified lwp and may never
897  * be delivered to the process generically.
898  */
899 void
900 trapsignal(struct lwp *lp, int sig, u_long code)
901 {
902 	struct proc *p = lp->lwp_proc;
903 	struct sigacts *ps = p->p_sigacts;
904 
905 	/*
906 	 * If we are a virtual kernel running an emulated user process
907 	 * context, switch back to the virtual kernel context before
908 	 * trying to post the signal.
909 	 */
910 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
911 		struct trapframe *tf = lp->lwp_md.md_regs;
912 		tf->tf_trapno = 0;
913 		vkernel_trap(lp, tf);
914 	}
915 
916 
917 	if ((p->p_flags & P_TRACED) == 0 && SIGISMEMBER(p->p_sigcatch, sig) &&
918 	    !SIGISMEMBER(lp->lwp_sigmask, sig)) {
919 		lp->lwp_ru.ru_nsignals++;
920 #ifdef KTRACE
921 		if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
922 			ktrpsig(lp, sig, ps->ps_sigact[_SIG_IDX(sig)],
923 				&lp->lwp_sigmask, code);
924 #endif
925 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], sig,
926 						&lp->lwp_sigmask, code);
927 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
928 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
929 			SIGADDSET(lp->lwp_sigmask, sig);
930 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
931 			/*
932 			 * See kern_sigaction() for origin of this code.
933 			 */
934 			SIGDELSET(p->p_sigcatch, sig);
935 			if (sig != SIGCONT &&
936 			    sigprop(sig) & SA_IGNORE)
937 				SIGADDSET(p->p_sigignore, sig);
938 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
939 		}
940 	} else {
941 		lp->lwp_code = code;	/* XXX for core dump/debugger */
942 		lp->lwp_sig = sig;	/* XXX to verify code */
943 		lwpsignal(p, lp, sig);
944 	}
945 }
946 
947 /*
948  * Find a suitable lwp to deliver the signal to.  Returns NULL if all
949  * lwps hold the signal blocked.
950  *
951  * Caller must hold p->p_token.
952  *
953  * Returns a lp or NULL.  If non-NULL the lp is held and its token is
954  * acquired.
955  */
956 static struct lwp *
957 find_lwp_for_signal(struct proc *p, int sig)
958 {
959 	struct lwp *lp;
960 	struct lwp *run, *sleep, *stop;
961 
962 	/*
963 	 * If the running/preempted thread belongs to the proc to which
964 	 * the signal is being delivered and this thread does not block
965 	 * the signal, then we can avoid a context switch by delivering
966 	 * the signal to this thread, because it will return to userland
967 	 * soon anyways.
968 	 */
969 	lp = lwkt_preempted_proc();
970 	if (lp != NULL && lp->lwp_proc == p) {
971 		LWPHOLD(lp);
972 		lwkt_gettoken(&lp->lwp_token);
973 		if (!SIGISMEMBER(lp->lwp_sigmask, sig)) {
974 			/* return w/ token held */
975 			return (lp);
976 		}
977 		lwkt_reltoken(&lp->lwp_token);
978 		LWPRELE(lp);
979 	}
980 
981 	run = sleep = stop = NULL;
982 	FOREACH_LWP_IN_PROC(lp, p) {
983 		/*
984 		 * If the signal is being blocked by the lwp, then this
985 		 * lwp is not eligible for receiving the signal.
986 		 */
987 		LWPHOLD(lp);
988 		lwkt_gettoken(&lp->lwp_token);
989 
990 		if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
991 			lwkt_reltoken(&lp->lwp_token);
992 			LWPRELE(lp);
993 			continue;
994 		}
995 
996 		switch (lp->lwp_stat) {
997 		case LSRUN:
998 			if (sleep) {
999 				lwkt_token_swap();
1000 				lwkt_reltoken(&sleep->lwp_token);
1001 				LWPRELE(sleep);
1002 				sleep = NULL;
1003 				run = lp;
1004 			} else if (stop) {
1005 				lwkt_token_swap();
1006 				lwkt_reltoken(&stop->lwp_token);
1007 				LWPRELE(stop);
1008 				stop = NULL;
1009 				run = lp;
1010 			} else {
1011 				run = lp;
1012 			}
1013 			break;
1014 		case LSSLEEP:
1015 			if (lp->lwp_flags & LWP_SINTR) {
1016 				if (sleep) {
1017 					lwkt_reltoken(&lp->lwp_token);
1018 					LWPRELE(lp);
1019 				} else if (stop) {
1020 					lwkt_token_swap();
1021 					lwkt_reltoken(&stop->lwp_token);
1022 					LWPRELE(stop);
1023 					stop = NULL;
1024 					sleep = lp;
1025 				} else {
1026 					sleep = lp;
1027 				}
1028 			} else {
1029 				lwkt_reltoken(&lp->lwp_token);
1030 				LWPRELE(lp);
1031 			}
1032 			break;
1033 		case LSSTOP:
1034 			if (sleep) {
1035 				lwkt_reltoken(&lp->lwp_token);
1036 				LWPRELE(lp);
1037 			} else if (stop) {
1038 				lwkt_reltoken(&lp->lwp_token);
1039 				LWPRELE(lp);
1040 			} else {
1041 				stop = lp;
1042 			}
1043 			break;
1044 		}
1045 		if (run)
1046 			break;
1047 	}
1048 
1049 	if (run != NULL)
1050 		return (run);
1051 	else if (sleep != NULL)
1052 		return (sleep);
1053 	else
1054 		return (stop);
1055 }
1056 
1057 /*
1058  * Send the signal to the process.  If the signal has an action, the action
1059  * is usually performed by the target process rather than the caller; we add
1060  * the signal to the set of pending signals for the process.
1061  *
1062  * Exceptions:
1063  *   o When a stop signal is sent to a sleeping process that takes the
1064  *     default action, the process is stopped without awakening it.
1065  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
1066  *     regardless of the signal action (eg, blocked or ignored).
1067  *
1068  * Other ignored signals are discarded immediately.
1069  *
1070  * If the caller wishes to call this function from a hard code section the
1071  * caller must already hold p->p_token (see kern_clock.c).
1072  *
1073  * No requirements.
1074  */
1075 void
1076 ksignal(struct proc *p, int sig)
1077 {
1078 	lwpsignal(p, NULL, sig);
1079 }
1080 
1081 /*
1082  * The core for ksignal.  lp may be NULL, then a suitable thread
1083  * will be chosen.  If not, lp MUST be a member of p.
1084  *
1085  * If the caller wishes to call this function from a hard code section the
1086  * caller must already hold p->p_token.
1087  *
1088  * No requirements.
1089  */
1090 void
1091 lwpsignal(struct proc *p, struct lwp *lp, int sig)
1092 {
1093 	struct proc *q;
1094 	sig_t action;
1095 	int prop;
1096 
1097 	if (sig > _SIG_MAXSIG || sig <= 0) {
1098 		kprintf("lwpsignal: signal %d\n", sig);
1099 		panic("lwpsignal signal number");
1100 	}
1101 
1102 	KKASSERT(lp == NULL || lp->lwp_proc == p);
1103 
1104 	/*
1105 	 * We don't want to race... well, all sorts of things.  Get appropriate
1106 	 * tokens.
1107 	 *
1108 	 * Don't try to deliver a generic signal to an exiting process,
1109 	 * the signal structures could be in flux.  We check the LWP later
1110 	 * on.
1111 	 */
1112 	PHOLD(p);
1113 	lwkt_gettoken(&p->p_token);
1114 	if (lp) {
1115 		LWPHOLD(lp);
1116 		lwkt_gettoken(&lp->lwp_token);
1117 	} else if (p->p_flags & P_WEXIT) {
1118 		goto out;
1119 	}
1120 
1121 	prop = sigprop(sig);
1122 
1123 	/*
1124 	 * If proc is traced, always give parent a chance;
1125 	 * if signal event is tracked by procfs, give *that*
1126 	 * a chance, as well.
1127 	 */
1128 	if ((p->p_flags & P_TRACED) || (p->p_stops & S_SIG)) {
1129 		action = SIG_DFL;
1130 	} else {
1131 		/*
1132 		 * Do not try to deliver signals to an exiting lwp.  Note
1133 		 * that we must still deliver the signal if P_WEXIT is set
1134 		 * in the process flags.
1135 		 */
1136 		if (lp && (lp->lwp_mpflags & LWP_MP_WEXIT)) {
1137 			if (lp) {
1138 				lwkt_reltoken(&lp->lwp_token);
1139 				LWPRELE(lp);
1140 			}
1141 			lwkt_reltoken(&p->p_token);
1142 			PRELE(p);
1143 			return;
1144 		}
1145 
1146 		/*
1147 		 * If the signal is being ignored, then we forget about
1148 		 * it immediately.  NOTE: We don't set SIGCONT in p_sigignore,
1149 		 * and if it is set to SIG_IGN, action will be SIG_DFL here.
1150 		 */
1151 		if (SIGISMEMBER(p->p_sigignore, sig)) {
1152 			/*
1153 			 * Even if a signal is set SIG_IGN, it may still be
1154 			 * lurking in a kqueue.
1155 			 */
1156 			KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1157 			if (lp) {
1158 				lwkt_reltoken(&lp->lwp_token);
1159 				LWPRELE(lp);
1160 			}
1161 			lwkt_reltoken(&p->p_token);
1162 			PRELE(p);
1163 			return;
1164 		}
1165 		if (SIGISMEMBER(p->p_sigcatch, sig))
1166 			action = SIG_CATCH;
1167 		else
1168 			action = SIG_DFL;
1169 	}
1170 
1171 	/*
1172 	 * If continuing, clear any pending STOP signals.
1173 	 */
1174 	if (prop & SA_CONT)
1175 		SIG_STOPSIGMASK(p->p_siglist);
1176 
1177 	if (prop & SA_STOP) {
1178 		/*
1179 		 * If sending a tty stop signal to a member of an orphaned
1180 		 * process group, discard the signal here if the action
1181 		 * is default; don't stop the process below if sleeping,
1182 		 * and don't clear any pending SIGCONT.
1183 		 */
1184 		if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0 &&
1185 		    action == SIG_DFL) {
1186 			lwkt_reltoken(&p->p_token);
1187 			PRELE(p);
1188 		        return;
1189 		}
1190 		SIG_CONTSIGMASK(p->p_siglist);
1191 		p->p_flags &= ~P_CONTINUED;
1192 	}
1193 
1194 	if (p->p_stat == SSTOP) {
1195 		/*
1196 		 * Nobody can handle this signal, add it to the lwp or
1197 		 * process pending list
1198 		 */
1199 		if (lp) {
1200 			spin_lock(&lp->lwp_spin);
1201 			SIGADDSET(lp->lwp_siglist, sig);
1202 			spin_unlock(&lp->lwp_spin);
1203 		} else {
1204 			SIGADDSET(p->p_siglist, sig);
1205 		}
1206 
1207 		/*
1208 		 * If the process is stopped and is being traced, then no
1209 		 * further action is necessary.
1210 		 */
1211 		if (p->p_flags & P_TRACED)
1212 			goto out;
1213 
1214 		/*
1215 		 * If the process is stopped and receives a KILL signal,
1216 		 * make the process runnable.
1217 		 */
1218 		if (sig == SIGKILL) {
1219 			proc_unstop(p);
1220 			goto active_process;
1221 		}
1222 
1223 		/*
1224 		 * If the process is stopped and receives a CONT signal,
1225 		 * then try to make the process runnable again.
1226 		 */
1227 		if (prop & SA_CONT) {
1228 			/*
1229 			 * If SIGCONT is default (or ignored), we continue the
1230 			 * process but don't leave the signal in p_siglist, as
1231 			 * it has no further action.  If SIGCONT is held, we
1232 			 * continue the process and leave the signal in
1233 			 * p_siglist.  If the process catches SIGCONT, let it
1234 			 * handle the signal itself.
1235 			 *
1236 			 * XXX what if the signal is being held blocked?
1237 			 *
1238 			 * Token required to interlock kern_wait().
1239 			 * Reparenting can also cause a race so we have to
1240 			 * hold (q).
1241 			 */
1242 			q = p->p_pptr;
1243 			PHOLD(q);
1244 			lwkt_gettoken(&q->p_token);
1245 			p->p_flags |= P_CONTINUED;
1246 			wakeup(q);
1247 			if (action == SIG_DFL)
1248 				SIGDELSET(p->p_siglist, sig);
1249 			proc_unstop(p);
1250 			lwkt_reltoken(&q->p_token);
1251 			PRELE(q);
1252 			if (action == SIG_CATCH)
1253 				goto active_process;
1254 			goto out;
1255 		}
1256 
1257 		/*
1258 		 * If the process is stopped and receives another STOP
1259 		 * signal, we do not need to stop it again.  If we did
1260 		 * the shell could get confused.
1261 		 *
1262 		 * However, if the current/preempted lwp is part of the
1263 		 * process receiving the signal, we need to keep it,
1264 		 * so that this lwp can stop in issignal() later, as
1265 		 * we don't want to wait until it reaches userret!
1266 		 */
1267 		if (prop & SA_STOP) {
1268 			if (lwkt_preempted_proc() == NULL ||
1269 			    lwkt_preempted_proc()->lwp_proc != p)
1270 				SIGDELSET(p->p_siglist, sig);
1271 		}
1272 
1273 		/*
1274 		 * Otherwise the process is stopped and it received some
1275 		 * signal, which does not change its stopped state.  When
1276 		 * the process is continued a wakeup(p) will be issued which
1277 		 * will wakeup any threads sleeping in tstop().
1278 		 */
1279 		if (lp == NULL) {
1280 			/* NOTE: returns lp w/ token held */
1281 			lp = find_lwp_for_signal(p, sig);
1282 		}
1283 		goto out;
1284 
1285 		/* NOTREACHED */
1286 	}
1287 	/* else not stopped */
1288 active_process:
1289 
1290 	/*
1291 	 * Never deliver a lwp-specific signal to a random lwp.
1292 	 */
1293 	if (lp == NULL) {
1294 		/* NOTE: returns lp w/ token held */
1295 		lp = find_lwp_for_signal(p, sig);
1296 		if (lp) {
1297 			if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
1298 				lwkt_reltoken(&lp->lwp_token);
1299 				LWPRELE(lp);
1300 				lp = NULL;
1301 			}
1302 		}
1303 	}
1304 
1305 	/*
1306 	 * Deliver to the process generically if (1) the signal is being
1307 	 * sent to any thread or (2) we could not find a thread to deliver
1308 	 * it to.
1309 	 */
1310 	if (lp == NULL) {
1311 		SIGADDSET(p->p_siglist, sig);
1312 		goto out;
1313 	}
1314 
1315 	/*
1316 	 * Deliver to a specific LWP whether it masks it or not.  It will
1317 	 * not be dispatched if masked but we must still deliver it.
1318 	 */
1319 	if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
1320 	    (p->p_flags & P_TRACED) == 0) {
1321 		p->p_nice = NZERO;
1322 	}
1323 
1324 	/*
1325 	 * If the process receives a STOP signal which indeed needs to
1326 	 * stop the process, do so.  If the process chose to catch the
1327 	 * signal, it will be treated like any other signal.
1328 	 */
1329 	if ((prop & SA_STOP) && action == SIG_DFL) {
1330 		/*
1331 		 * If a child holding parent blocked, stopping
1332 		 * could cause deadlock.  Take no action at this
1333 		 * time.
1334 		 */
1335 		if (p->p_flags & P_PPWAIT) {
1336 			SIGADDSET(p->p_siglist, sig);
1337 			goto out;
1338 		}
1339 
1340 		/*
1341 		 * Do not actually try to manipulate the process, but simply
1342 		 * stop it.  Lwps will stop as soon as they safely can.
1343 		 *
1344 		 * Ignore stop if the process is exiting.
1345 		 */
1346 		if ((p->p_flags & P_WEXIT) == 0) {
1347 			p->p_xstat = sig;
1348 			proc_stop(p);
1349 		}
1350 		goto out;
1351 	}
1352 
1353 	/*
1354 	 * If it is a CONT signal with default action, just ignore it.
1355 	 */
1356 	if ((prop & SA_CONT) && action == SIG_DFL)
1357 		goto out;
1358 
1359 	/*
1360 	 * Mark signal pending at this specific thread.
1361 	 */
1362 	spin_lock(&lp->lwp_spin);
1363 	SIGADDSET(lp->lwp_siglist, sig);
1364 	spin_unlock(&lp->lwp_spin);
1365 
1366 	lwp_signotify(lp);
1367 
1368 out:
1369 	if (lp) {
1370 		lwkt_reltoken(&lp->lwp_token);
1371 		LWPRELE(lp);
1372 	}
1373 	lwkt_reltoken(&p->p_token);
1374 	PRELE(p);
1375 }
1376 
1377 /*
1378  * Notify the LWP that a signal has arrived.  The LWP does not have to be
1379  * sleeping on the current cpu.
1380  *
1381  * p->p_token and lp->lwp_token must be held on call.
1382  *
1383  * We can only safely schedule the thread on its current cpu and only if
1384  * one of the SINTR flags is set.  If an SINTR flag is set AND we are on
1385  * the correct cpu we are properly interlocked, otherwise we could be
1386  * racing other thread transition states (or the lwp is on the user scheduler
1387  * runq but not scheduled) and must not do anything.
1388  *
1389  * Since we hold the lwp token we know the lwp cannot be ripped out from
1390  * under us so we can safely hold it to prevent it from being ripped out
1391  * from under us if we are forced to IPI another cpu to make the local
1392  * checks there.
1393  *
1394  * Adjustment of lp->lwp_stat can only occur when we hold the lwp_token,
1395  * which we won't in an IPI so any fixups have to be done here, effectively
1396  * replicating part of what setrunnable() does.
1397  */
1398 static void
1399 lwp_signotify(struct lwp *lp)
1400 {
1401 	ASSERT_LWKT_TOKEN_HELD(&lp->lwp_proc->p_token);
1402 
1403 	crit_enter();
1404 	if (lp == lwkt_preempted_proc()) {
1405 		/*
1406 		 * lwp is on the current cpu AND it is currently running
1407 		 * (we preempted it).
1408 		 */
1409 		signotify();
1410 	} else if (lp->lwp_flags & LWP_SINTR) {
1411 		/*
1412 		 * lwp is sitting in tsleep() with PCATCH set
1413 		 */
1414 		if (lp->lwp_thread->td_gd == mycpu) {
1415 			setrunnable(lp);
1416 		} else {
1417 			/*
1418 			 * We can only adjust lwp_stat while we hold the
1419 			 * lwp_token, and we won't in the IPI function.
1420 			 */
1421 			LWPHOLD(lp);
1422 			if (lp->lwp_stat == LSSTOP)
1423 				lp->lwp_stat = LSSLEEP;
1424 			lwkt_send_ipiq(lp->lwp_thread->td_gd,
1425 				       lwp_signotify_remote, lp);
1426 		}
1427 	} else if (lp->lwp_thread->td_flags & TDF_SINTR) {
1428 		/*
1429 		 * lwp is sitting in lwkt_sleep() with PCATCH set.
1430 		 */
1431 		if (lp->lwp_thread->td_gd == mycpu) {
1432 			setrunnable(lp);
1433 		} else {
1434 			/*
1435 			 * We can only adjust lwp_stat while we hold the
1436 			 * lwp_token, and we won't in the IPI function.
1437 			 */
1438 			LWPHOLD(lp);
1439 			if (lp->lwp_stat == LSSTOP)
1440 				lp->lwp_stat = LSSLEEP;
1441 			lwkt_send_ipiq(lp->lwp_thread->td_gd,
1442 				       lwp_signotify_remote, lp);
1443 		}
1444 	} else {
1445 		/*
1446 		 * Otherwise the lwp is either in some uninterruptable state
1447 		 * or it is on the userland scheduler's runqueue waiting to
1448 		 * be scheduled to a cpu.
1449 		 */
1450 	}
1451 	crit_exit();
1452 }
1453 
1454 /*
1455  * This function is called via an IPI so we cannot call setrunnable() here
1456  * (because while we hold the lp we don't own its token, and can't get it
1457  * from an IPI).
1458  *
1459  * We are interlocked by virtue of being on the same cpu as the target.  If
1460  * we still are and LWP_SINTR or TDF_SINTR is set we can safely schedule
1461  * the target thread.
1462  */
1463 static void
1464 lwp_signotify_remote(void *arg)
1465 {
1466 	struct lwp *lp = arg;
1467 	thread_t td = lp->lwp_thread;
1468 
1469 	if (lp == lwkt_preempted_proc()) {
1470 		signotify();
1471 		LWPRELE(lp);
1472 	} else if (td->td_gd == mycpu) {
1473 		if ((lp->lwp_flags & LWP_SINTR) ||
1474 		    (td->td_flags & TDF_SINTR)) {
1475 			lwkt_schedule(td);
1476 		}
1477 		LWPRELE(lp);
1478 	} else {
1479 		lwkt_send_ipiq(td->td_gd, lwp_signotify_remote, lp);
1480 		/* LWPHOLD() is forwarded to the target cpu */
1481 	}
1482 }
1483 
1484 /*
1485  * Caller must hold p->p_token
1486  */
1487 void
1488 proc_stop(struct proc *p)
1489 {
1490 	struct proc *q;
1491 	struct lwp *lp;
1492 
1493 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1494 
1495 	/* If somebody raced us, be happy with it */
1496 	if (p->p_stat == SSTOP || p->p_stat == SZOMB) {
1497 		return;
1498 	}
1499 	p->p_stat = SSTOP;
1500 
1501 	FOREACH_LWP_IN_PROC(lp, p) {
1502 		LWPHOLD(lp);
1503 		lwkt_gettoken(&lp->lwp_token);
1504 
1505 		switch (lp->lwp_stat) {
1506 		case LSSTOP:
1507 			/*
1508 			 * Do nothing, we are already counted in
1509 			 * p_nstopped.
1510 			 */
1511 			break;
1512 
1513 		case LSSLEEP:
1514 			/*
1515 			 * We're sleeping, but we will stop before
1516 			 * returning to userspace, so count us
1517 			 * as stopped as well.  We set LWP_MP_WSTOP
1518 			 * to signal the lwp that it should not
1519 			 * increase p_nstopped when reaching tstop().
1520 			 *
1521 			 * LWP_MP_WSTOP is protected by lp->lwp_token.
1522 			 */
1523 			if ((lp->lwp_mpflags & LWP_MP_WSTOP) == 0) {
1524 				atomic_set_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
1525 				++p->p_nstopped;
1526 			}
1527 			break;
1528 
1529 		case LSRUN:
1530 			/*
1531 			 * We might notify ourself, but that's not
1532 			 * a problem.
1533 			 */
1534 			lwp_signotify(lp);
1535 			break;
1536 		}
1537 		lwkt_reltoken(&lp->lwp_token);
1538 		LWPRELE(lp);
1539 	}
1540 
1541 	if (p->p_nstopped == p->p_nthreads) {
1542 		/*
1543 		 * Token required to interlock kern_wait().  Reparenting can
1544 		 * also cause a race so we have to hold (q).
1545 		 */
1546 		q = p->p_pptr;
1547 		PHOLD(q);
1548 		lwkt_gettoken(&q->p_token);
1549 		p->p_flags &= ~P_WAITED;
1550 		wakeup(q);
1551 		if ((q->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1552 			ksignal(p->p_pptr, SIGCHLD);
1553 		lwkt_reltoken(&q->p_token);
1554 		PRELE(q);
1555 	}
1556 }
1557 
1558 /*
1559  * Caller must hold proc_token
1560  */
1561 void
1562 proc_unstop(struct proc *p)
1563 {
1564 	struct lwp *lp;
1565 
1566 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1567 
1568 	if (p->p_stat != SSTOP)
1569 		return;
1570 
1571 	p->p_stat = SACTIVE;
1572 
1573 	FOREACH_LWP_IN_PROC(lp, p) {
1574 		LWPHOLD(lp);
1575 		lwkt_gettoken(&lp->lwp_token);
1576 
1577 		switch (lp->lwp_stat) {
1578 		case LSRUN:
1579 			/*
1580 			 * Uh?  Not stopped?  Well, I guess that's okay.
1581 			 */
1582 			if (bootverbose)
1583 				kprintf("proc_unstop: lwp %d/%d not sleeping\n",
1584 					p->p_pid, lp->lwp_tid);
1585 			break;
1586 
1587 		case LSSLEEP:
1588 			/*
1589 			 * Still sleeping.  Don't bother waking it up.
1590 			 * However, if this thread was counted as
1591 			 * stopped, undo this.
1592 			 *
1593 			 * Nevertheless we call setrunnable() so that it
1594 			 * will wake up in case a signal or timeout arrived
1595 			 * in the meantime.
1596 			 *
1597 			 * LWP_MP_WSTOP is protected by lp->lwp_token.
1598 			 */
1599 			if (lp->lwp_mpflags & LWP_MP_WSTOP) {
1600 				atomic_clear_int(&lp->lwp_mpflags,
1601 						 LWP_MP_WSTOP);
1602 				--p->p_nstopped;
1603 			} else {
1604 				if (bootverbose)
1605 					kprintf("proc_unstop: lwp %d/%d sleeping, not stopped\n",
1606 						p->p_pid, lp->lwp_tid);
1607 			}
1608 			/* FALLTHROUGH */
1609 
1610 		case LSSTOP:
1611 			/*
1612 			 * This handles any lwp's waiting in a tsleep with
1613 			 * SIGCATCH.
1614 			 */
1615 			lwp_signotify(lp);
1616 			break;
1617 
1618 		}
1619 		lwkt_reltoken(&lp->lwp_token);
1620 		LWPRELE(lp);
1621 	}
1622 
1623 	/*
1624 	 * This handles any lwp's waiting in tstop().  We have interlocked
1625 	 * the setting of p_stat by acquiring and releasing each lpw's
1626 	 * token.
1627 	 */
1628 	wakeup(p);
1629 }
1630 
1631 /*
1632  * No requirements.
1633  */
1634 static int
1635 kern_sigtimedwait(sigset_t waitset, siginfo_t *info, struct timespec *timeout)
1636 {
1637 	sigset_t savedmask, set;
1638 	struct proc *p = curproc;
1639 	struct lwp *lp = curthread->td_lwp;
1640 	int error, sig, hz, timevalid = 0;
1641 	struct timespec rts, ets, ts;
1642 	struct timeval tv;
1643 
1644 	error = 0;
1645 	sig = 0;
1646 	ets.tv_sec = 0;		/* silence compiler warning */
1647 	ets.tv_nsec = 0;	/* silence compiler warning */
1648 	SIG_CANTMASK(waitset);
1649 	savedmask = lp->lwp_sigmask;
1650 
1651 	if (timeout) {
1652 		if (timeout->tv_sec >= 0 && timeout->tv_nsec >= 0 &&
1653 		    timeout->tv_nsec < 1000000000) {
1654 			timevalid = 1;
1655 			getnanouptime(&rts);
1656 		 	ets = rts;
1657 			timespecadd(&ets, timeout);
1658 		}
1659 	}
1660 
1661 	for (;;) {
1662 		set = lwp_sigpend(lp);
1663 		SIGSETAND(set, waitset);
1664 		if ((sig = sig_ffs(&set)) != 0) {
1665 			SIGFILLSET(lp->lwp_sigmask);
1666 			SIGDELSET(lp->lwp_sigmask, sig);
1667 			SIG_CANTMASK(lp->lwp_sigmask);
1668 			sig = issignal(lp, 1);
1669 			/*
1670 			 * It may be a STOP signal, in the case, issignal
1671 			 * returns 0, because we may stop there, and new
1672 			 * signal can come in, we should restart if we got
1673 			 * nothing.
1674 			 */
1675 			if (sig == 0)
1676 				continue;
1677 			else
1678 				break;
1679 		}
1680 
1681 		/*
1682 		 * Previous checking got nothing, and we retried but still
1683 		 * got nothing, we should return the error status.
1684 		 */
1685 		if (error)
1686 			break;
1687 
1688 		/*
1689 		 * POSIX says this must be checked after looking for pending
1690 		 * signals.
1691 		 */
1692 		if (timeout) {
1693 			if (timevalid == 0) {
1694 				error = EINVAL;
1695 				break;
1696 			}
1697 			getnanouptime(&rts);
1698 			if (timespeccmp(&rts, &ets, >=)) {
1699 				error = EAGAIN;
1700 				break;
1701 			}
1702 			ts = ets;
1703 			timespecsub(&ts, &rts);
1704 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1705 			hz = tvtohz_high(&tv);
1706 		} else {
1707 			hz = 0;
1708 		}
1709 
1710 		lp->lwp_sigmask = savedmask;
1711 		SIGSETNAND(lp->lwp_sigmask, waitset);
1712 		/*
1713 		 * We won't ever be woken up.  Instead, our sleep will
1714 		 * be broken in lwpsignal().
1715 		 */
1716 		error = tsleep(&p->p_sigacts, PCATCH, "sigwt", hz);
1717 		if (timeout) {
1718 			if (error == ERESTART) {
1719 				/* can not restart a timeout wait. */
1720 				error = EINTR;
1721 			} else if (error == EAGAIN) {
1722 				/* will calculate timeout by ourself. */
1723 				error = 0;
1724 			}
1725 		}
1726 		/* Retry ... */
1727 	}
1728 
1729 	lp->lwp_sigmask = savedmask;
1730 	if (sig) {
1731 		error = 0;
1732 		bzero(info, sizeof(*info));
1733 		info->si_signo = sig;
1734 		spin_lock(&lp->lwp_spin);
1735 		lwp_delsig(lp, sig);	/* take the signal! */
1736 		spin_unlock(&lp->lwp_spin);
1737 
1738 		if (sig == SIGKILL) {
1739 			sigexit(lp, sig);
1740 			/* NOT REACHED */
1741 		}
1742 	}
1743 
1744 	return (error);
1745 }
1746 
1747 /*
1748  * MPALMOSTSAFE
1749  */
1750 int
1751 sys_sigtimedwait(struct sigtimedwait_args *uap)
1752 {
1753 	struct timespec ts;
1754 	struct timespec *timeout;
1755 	sigset_t set;
1756 	siginfo_t info;
1757 	int error;
1758 
1759 	if (uap->timeout) {
1760 		error = copyin(uap->timeout, &ts, sizeof(ts));
1761 		if (error)
1762 			return (error);
1763 		timeout = &ts;
1764 	} else {
1765 		timeout = NULL;
1766 	}
1767 	error = copyin(uap->set, &set, sizeof(set));
1768 	if (error)
1769 		return (error);
1770 	error = kern_sigtimedwait(set, &info, timeout);
1771 	if (error)
1772 		return (error);
1773 	if (uap->info)
1774 		error = copyout(&info, uap->info, sizeof(info));
1775 	/* Repost if we got an error. */
1776 	/*
1777 	 * XXX lwp
1778 	 *
1779 	 * This could transform a thread-specific signal to another
1780 	 * thread / process pending signal.
1781 	 */
1782 	if (error) {
1783 		ksignal(curproc, info.si_signo);
1784 	} else {
1785 		uap->sysmsg_result = info.si_signo;
1786 	}
1787 	return (error);
1788 }
1789 
1790 /*
1791  * MPALMOSTSAFE
1792  */
1793 int
1794 sys_sigwaitinfo(struct sigwaitinfo_args *uap)
1795 {
1796 	siginfo_t info;
1797 	sigset_t set;
1798 	int error;
1799 
1800 	error = copyin(uap->set, &set, sizeof(set));
1801 	if (error)
1802 		return (error);
1803 	error = kern_sigtimedwait(set, &info, NULL);
1804 	if (error)
1805 		return (error);
1806 	if (uap->info)
1807 		error = copyout(&info, uap->info, sizeof(info));
1808 	/* Repost if we got an error. */
1809 	/*
1810 	 * XXX lwp
1811 	 *
1812 	 * This could transform a thread-specific signal to another
1813 	 * thread / process pending signal.
1814 	 */
1815 	if (error) {
1816 		ksignal(curproc, info.si_signo);
1817 	} else {
1818 		uap->sysmsg_result = info.si_signo;
1819 	}
1820 	return (error);
1821 }
1822 
1823 /*
1824  * If the current process has received a signal that would interrupt a
1825  * system call, return EINTR or ERESTART as appropriate.
1826  */
1827 int
1828 iscaught(struct lwp *lp)
1829 {
1830 	struct proc *p = lp->lwp_proc;
1831 	int sig;
1832 
1833 	if (p) {
1834 		if ((sig = CURSIG(lp)) != 0) {
1835 			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1836 				return (EINTR);
1837 			return (ERESTART);
1838 		}
1839 	}
1840 	return(EWOULDBLOCK);
1841 }
1842 
1843 /*
1844  * If the current process has received a signal (should be caught or cause
1845  * termination, should interrupt current syscall), return the signal number.
1846  * Stop signals with default action are processed immediately, then cleared;
1847  * they aren't returned.  This is checked after each entry to the system for
1848  * a syscall or trap (though this can usually be done without calling issignal
1849  * by checking the pending signal masks in the CURSIG macro).
1850  *
1851  * This routine is called via CURSIG/__cursig.  We will acquire and release
1852  * p->p_token but if the caller needs to interlock the test the caller must
1853  * also hold p->p_token.
1854  *
1855  *	while (sig = CURSIG(curproc))
1856  *		postsig(sig);
1857  *
1858  * MPSAFE
1859  */
1860 int
1861 issignal(struct lwp *lp, int maytrace)
1862 {
1863 	struct proc *p = lp->lwp_proc;
1864 	sigset_t mask;
1865 	int sig, prop;
1866 
1867 	lwkt_gettoken(&p->p_token);
1868 
1869 	for (;;) {
1870 		int traced = (p->p_flags & P_TRACED) || (p->p_stops & S_SIG);
1871 
1872 		/*
1873 		 * If this process is supposed to stop, stop this thread.
1874 		 */
1875 		if (p->p_stat == SSTOP)
1876 			tstop();
1877 
1878 		mask = lwp_sigpend(lp);
1879 		SIGSETNAND(mask, lp->lwp_sigmask);
1880 		if (p->p_flags & P_PPWAIT)
1881 			SIG_STOPSIGMASK(mask);
1882 		if (SIGISEMPTY(mask)) {		/* no signal to send */
1883 			lwkt_reltoken(&p->p_token);
1884 			return (0);
1885 		}
1886 		sig = sig_ffs(&mask);
1887 
1888 		STOPEVENT(p, S_SIG, sig);
1889 
1890 		/*
1891 		 * We should see pending but ignored signals
1892 		 * only if P_TRACED was on when they were posted.
1893 		 */
1894 		if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
1895 			spin_lock(&lp->lwp_spin);
1896 			lwp_delsig(lp, sig);
1897 			spin_unlock(&lp->lwp_spin);
1898 			continue;
1899 		}
1900 		if (maytrace &&
1901 		    (p->p_flags & P_TRACED) &&
1902 		    (p->p_flags & P_PPWAIT) == 0) {
1903 			/*
1904 			 * If traced, always stop, and stay stopped until
1905 			 * released by the parent.
1906 			 *
1907 			 * NOTE: SSTOP may get cleared during the loop,
1908 			 * but we do not re-notify the parent if we have
1909 			 * to loop several times waiting for the parent
1910 			 * to let us continue.
1911 			 *
1912 			 * XXX not sure if this is still true
1913 			 */
1914 			p->p_xstat = sig;
1915 			proc_stop(p);
1916 			do {
1917 				tstop();
1918 			} while (!trace_req(p) && (p->p_flags & P_TRACED));
1919 
1920 			/*
1921 			 * If parent wants us to take the signal,
1922 			 * then it will leave it in p->p_xstat;
1923 			 * otherwise we just look for signals again.
1924 			 */
1925 			spin_lock(&lp->lwp_spin);
1926 			lwp_delsig(lp, sig);	/* clear old signal */
1927 			spin_unlock(&lp->lwp_spin);
1928 			sig = p->p_xstat;
1929 			if (sig == 0)
1930 				continue;
1931 
1932 			/*
1933 			 * Put the new signal into p_siglist.  If the
1934 			 * signal is being masked, look for other signals.
1935 			 *
1936 			 * XXX lwp might need a call to ksignal()
1937 			 */
1938 			SIGADDSET(p->p_siglist, sig);
1939 			if (SIGISMEMBER(lp->lwp_sigmask, sig))
1940 				continue;
1941 
1942 			/*
1943 			 * If the traced bit got turned off, go back up
1944 			 * to the top to rescan signals.  This ensures
1945 			 * that p_sig* and ps_sigact are consistent.
1946 			 */
1947 			if ((p->p_flags & P_TRACED) == 0)
1948 				continue;
1949 		}
1950 
1951 		prop = sigprop(sig);
1952 
1953 		/*
1954 		 * Decide whether the signal should be returned.
1955 		 * Return the signal's number, or fall through
1956 		 * to clear it from the pending mask.
1957 		 */
1958 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
1959 		case (intptr_t)SIG_DFL:
1960 			/*
1961 			 * Don't take default actions on system processes.
1962 			 */
1963 			if (p->p_pid <= 1) {
1964 #ifdef DIAGNOSTIC
1965 				/*
1966 				 * Are you sure you want to ignore SIGSEGV
1967 				 * in init? XXX
1968 				 */
1969 				kprintf("Process (pid %lu) got signal %d\n",
1970 					(u_long)p->p_pid, sig);
1971 #endif
1972 				break;		/* == ignore */
1973 			}
1974 
1975 			/*
1976 			 * Handle the in-kernel checkpoint action
1977 			 */
1978 			if (prop & SA_CKPT) {
1979 				checkpoint_signal_handler(lp);
1980 				break;
1981 			}
1982 
1983 			/*
1984 			 * If there is a pending stop signal to process
1985 			 * with default action, stop here,
1986 			 * then clear the signal.  However,
1987 			 * if process is member of an orphaned
1988 			 * process group, ignore tty stop signals.
1989 			 */
1990 			if (prop & SA_STOP) {
1991 				if (p->p_flags & P_TRACED ||
1992 		    		    (p->p_pgrp->pg_jobc == 0 &&
1993 				    prop & SA_TTYSTOP))
1994 					break;	/* == ignore */
1995 				if ((p->p_flags & P_WEXIT) == 0) {
1996 					p->p_xstat = sig;
1997 					proc_stop(p);
1998 					tstop();
1999 				}
2000 				break;
2001 			} else if (prop & SA_IGNORE) {
2002 				/*
2003 				 * Except for SIGCONT, shouldn't get here.
2004 				 * Default action is to ignore; drop it.
2005 				 */
2006 				break;		/* == ignore */
2007 			} else {
2008 				lwkt_reltoken(&p->p_token);
2009 				return (sig);
2010 			}
2011 
2012 			/*NOTREACHED*/
2013 
2014 		case (intptr_t)SIG_IGN:
2015 			/*
2016 			 * Masking above should prevent us ever trying
2017 			 * to take action on an ignored signal other
2018 			 * than SIGCONT, unless process is traced.
2019 			 */
2020 			if ((prop & SA_CONT) == 0 &&
2021 			    (p->p_flags & P_TRACED) == 0)
2022 				kprintf("issignal\n");
2023 			break;		/* == ignore */
2024 
2025 		default:
2026 			/*
2027 			 * This signal has an action, let
2028 			 * postsig() process it.
2029 			 */
2030 			lwkt_reltoken(&p->p_token);
2031 			return (sig);
2032 		}
2033 		spin_lock(&lp->lwp_spin);
2034 		lwp_delsig(lp, sig);		/* take the signal! */
2035 		spin_unlock(&lp->lwp_spin);
2036 	}
2037 	/* NOTREACHED */
2038 }
2039 
2040 /*
2041  * Take the action for the specified signal
2042  * from the current set of pending signals.
2043  *
2044  * Caller must hold p->p_token
2045  */
2046 void
2047 postsig(int sig)
2048 {
2049 	struct lwp *lp = curthread->td_lwp;
2050 	struct proc *p = lp->lwp_proc;
2051 	struct sigacts *ps = p->p_sigacts;
2052 	sig_t action;
2053 	sigset_t returnmask;
2054 	int code;
2055 
2056 	KASSERT(sig != 0, ("postsig"));
2057 
2058 	KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
2059 
2060 	/*
2061 	 * If we are a virtual kernel running an emulated user process
2062 	 * context, switch back to the virtual kernel context before
2063 	 * trying to post the signal.
2064 	 */
2065 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
2066 		struct trapframe *tf = lp->lwp_md.md_regs;
2067 		tf->tf_trapno = 0;
2068 		vkernel_trap(lp, tf);
2069 	}
2070 
2071 	spin_lock(&lp->lwp_spin);
2072 	lwp_delsig(lp, sig);
2073 	spin_unlock(&lp->lwp_spin);
2074 	action = ps->ps_sigact[_SIG_IDX(sig)];
2075 #ifdef KTRACE
2076 	if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
2077 		ktrpsig(lp, sig, action, lp->lwp_flags & LWP_OLDMASK ?
2078 			&lp->lwp_oldsigmask : &lp->lwp_sigmask, 0);
2079 #endif
2080 	STOPEVENT(p, S_SIG, sig);
2081 
2082 	if (action == SIG_DFL) {
2083 		/*
2084 		 * Default action, where the default is to kill
2085 		 * the process.  (Other cases were ignored above.)
2086 		 */
2087 		sigexit(lp, sig);
2088 		/* NOTREACHED */
2089 	} else {
2090 		/*
2091 		 * If we get here, the signal must be caught.
2092 		 */
2093 		KASSERT(action != SIG_IGN && !SIGISMEMBER(lp->lwp_sigmask, sig),
2094 		    ("postsig action"));
2095 
2096 		/*
2097 		 * Reset the signal handler if asked to
2098 		 */
2099 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
2100 			/*
2101 			 * See kern_sigaction() for origin of this code.
2102 			 */
2103 			SIGDELSET(p->p_sigcatch, sig);
2104 			if (sig != SIGCONT &&
2105 			    sigprop(sig) & SA_IGNORE)
2106 				SIGADDSET(p->p_sigignore, sig);
2107 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2108 		}
2109 
2110 		/*
2111 		 * Set the signal mask and calculate the mask to restore
2112 		 * when the signal function returns.
2113 		 *
2114 		 * Special case: user has done a sigsuspend.  Here the
2115 		 * current mask is not of interest, but rather the
2116 		 * mask from before the sigsuspend is what we want
2117 		 * restored after the signal processing is completed.
2118 		 */
2119 		if (lp->lwp_flags & LWP_OLDMASK) {
2120 			returnmask = lp->lwp_oldsigmask;
2121 			lp->lwp_flags &= ~LWP_OLDMASK;
2122 		} else {
2123 			returnmask = lp->lwp_sigmask;
2124 		}
2125 
2126 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
2127 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
2128 			SIGADDSET(lp->lwp_sigmask, sig);
2129 
2130 		lp->lwp_ru.ru_nsignals++;
2131 		if (lp->lwp_sig != sig) {
2132 			code = 0;
2133 		} else {
2134 			code = lp->lwp_code;
2135 			lp->lwp_code = 0;
2136 			lp->lwp_sig = 0;
2137 		}
2138 		(*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
2139 	}
2140 }
2141 
2142 /*
2143  * Kill the current process for stated reason.
2144  */
2145 void
2146 killproc(struct proc *p, char *why)
2147 {
2148 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n",
2149 		p->p_pid, p->p_comm,
2150 		p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2151 	ksignal(p, SIGKILL);
2152 }
2153 
2154 /*
2155  * Force the current process to exit with the specified signal, dumping core
2156  * if appropriate.  We bypass the normal tests for masked and caught signals,
2157  * allowing unrecoverable failures to terminate the process without changing
2158  * signal state.  Mark the accounting record with the signal termination.
2159  * If dumping core, save the signal number for the debugger.  Calls exit and
2160  * does not return.
2161  *
2162  * This routine does not return.
2163  */
2164 void
2165 sigexit(struct lwp *lp, int sig)
2166 {
2167 	struct proc *p = lp->lwp_proc;
2168 
2169 	lwkt_gettoken(&p->p_token);
2170 	p->p_acflag |= AXSIG;
2171 	if (sigprop(sig) & SA_CORE) {
2172 		lp->lwp_sig = sig;
2173 		/*
2174 		 * Log signals which would cause core dumps
2175 		 * (Log as LOG_INFO to appease those who don't want
2176 		 * these messages.)
2177 		 * XXX : Todo, as well as euid, write out ruid too
2178 		 */
2179 		if (coredump(lp, sig) == 0)
2180 			sig |= WCOREFLAG;
2181 		if (kern_logsigexit)
2182 			log(LOG_INFO,
2183 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2184 			    p->p_pid, p->p_comm,
2185 			    p->p_ucred ? p->p_ucred->cr_uid : -1,
2186 			    sig &~ WCOREFLAG,
2187 			    sig & WCOREFLAG ? " (core dumped)" : "");
2188 	}
2189 	lwkt_reltoken(&p->p_token);
2190 	exit1(W_EXITCODE(0, sig));
2191 	/* NOTREACHED */
2192 }
2193 
2194 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
2195 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
2196 	      sizeof(corefilename), "process corefile name format string");
2197 
2198 /*
2199  * expand_name(name, uid, pid)
2200  * Expand the name described in corefilename, using name, uid, and pid.
2201  * corefilename is a kprintf-like string, with three format specifiers:
2202  *	%N	name of process ("name")
2203  *	%P	process id (pid)
2204  *	%U	user id (uid)
2205  * For example, "%N.core" is the default; they can be disabled completely
2206  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
2207  * This is controlled by the sysctl variable kern.corefile (see above).
2208  */
2209 
2210 static char *
2211 expand_name(const char *name, uid_t uid, pid_t pid)
2212 {
2213 	char *temp;
2214 	char buf[11];		/* Buffer for pid/uid -- max 4B */
2215 	int i, n;
2216 	char *format = corefilename;
2217 	size_t namelen;
2218 
2219 	temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
2220 	if (temp == NULL)
2221 		return NULL;
2222 	namelen = strlen(name);
2223 	for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
2224 		int l;
2225 		switch (format[i]) {
2226 		case '%':	/* Format character */
2227 			i++;
2228 			switch (format[i]) {
2229 			case '%':
2230 				temp[n++] = '%';
2231 				break;
2232 			case 'N':	/* process name */
2233 				if ((n + namelen) > MAXPATHLEN) {
2234 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2235 					    pid, name, uid, temp, name);
2236 					kfree(temp, M_TEMP);
2237 					return NULL;
2238 				}
2239 				memcpy(temp+n, name, namelen);
2240 				n += namelen;
2241 				break;
2242 			case 'P':	/* process id */
2243 				l = ksprintf(buf, "%u", pid);
2244 				if ((n + l) > MAXPATHLEN) {
2245 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2246 					    pid, name, uid, temp, name);
2247 					kfree(temp, M_TEMP);
2248 					return NULL;
2249 				}
2250 				memcpy(temp+n, buf, l);
2251 				n += l;
2252 				break;
2253 			case 'U':	/* user id */
2254 				l = ksprintf(buf, "%u", uid);
2255 				if ((n + l) > MAXPATHLEN) {
2256 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2257 					    pid, name, uid, temp, name);
2258 					kfree(temp, M_TEMP);
2259 					return NULL;
2260 				}
2261 				memcpy(temp+n, buf, l);
2262 				n += l;
2263 				break;
2264 			default:
2265 			  	log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
2266 			}
2267 			break;
2268 		default:
2269 			temp[n++] = format[i];
2270 		}
2271 	}
2272 	temp[n] = '\0';
2273 	return temp;
2274 }
2275 
2276 /*
2277  * Dump a process' core.  The main routine does some
2278  * policy checking, and creates the name of the coredump;
2279  * then it passes on a vnode and a size limit to the process-specific
2280  * coredump routine if there is one; if there _is not_ one, it returns
2281  * ENOSYS; otherwise it returns the error from the process-specific routine.
2282  *
2283  * The parameter `lp' is the lwp which triggered the coredump.
2284  */
2285 
2286 static int
2287 coredump(struct lwp *lp, int sig)
2288 {
2289 	struct proc *p = lp->lwp_proc;
2290 	struct vnode *vp;
2291 	struct ucred *cred = p->p_ucred;
2292 	struct flock lf;
2293 	struct nlookupdata nd;
2294 	struct vattr vattr;
2295 	int error, error1;
2296 	char *name;			/* name of corefile */
2297 	off_t limit;
2298 
2299 	STOPEVENT(p, S_CORE, 0);
2300 
2301 	if (((sugid_coredump == 0) && p->p_flags & P_SUGID) || do_coredump == 0)
2302 		return (EFAULT);
2303 
2304 	/*
2305 	 * Note that the bulk of limit checking is done after
2306 	 * the corefile is created.  The exception is if the limit
2307 	 * for corefiles is 0, in which case we don't bother
2308 	 * creating the corefile at all.  This layout means that
2309 	 * a corefile is truncated instead of not being created,
2310 	 * if it is larger than the limit.
2311 	 */
2312 	limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
2313 	if (limit == 0)
2314 		return EFBIG;
2315 
2316 	name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
2317 	if (name == NULL)
2318 		return (EINVAL);
2319 	error = nlookup_init(&nd, name, UIO_SYSSPACE, NLC_LOCKVP);
2320 	if (error == 0)
2321 		error = vn_open(&nd, NULL, O_CREAT | FWRITE | O_NOFOLLOW, S_IRUSR | S_IWUSR);
2322 	kfree(name, M_TEMP);
2323 	if (error) {
2324 		nlookup_done(&nd);
2325 		return (error);
2326 	}
2327 	vp = nd.nl_open_vp;
2328 	nd.nl_open_vp = NULL;
2329 	nlookup_done(&nd);
2330 
2331 	vn_unlock(vp);
2332 	lf.l_whence = SEEK_SET;
2333 	lf.l_start = 0;
2334 	lf.l_len = 0;
2335 	lf.l_type = F_WRLCK;
2336 	error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, 0);
2337 	if (error)
2338 		goto out2;
2339 
2340 	/* Don't dump to non-regular files or files with links. */
2341 	if (vp->v_type != VREG ||
2342 	    VOP_GETATTR(vp, &vattr) || vattr.va_nlink != 1) {
2343 		error = EFAULT;
2344 		goto out1;
2345 	}
2346 
2347 	/* Don't dump to files current user does not own */
2348 	if (vattr.va_uid != p->p_ucred->cr_uid) {
2349 		error = EFAULT;
2350 		goto out1;
2351 	}
2352 
2353 	VATTR_NULL(&vattr);
2354 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2355 	vattr.va_size = 0;
2356 	VOP_SETATTR(vp, &vattr, cred);
2357 	p->p_acflag |= ACORE;
2358 	vn_unlock(vp);
2359 
2360 	error = p->p_sysent->sv_coredump ?
2361 		  p->p_sysent->sv_coredump(lp, sig, vp, limit) : ENOSYS;
2362 
2363 out1:
2364 	lf.l_type = F_UNLCK;
2365 	VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, 0);
2366 out2:
2367 	error1 = vn_close(vp, FWRITE);
2368 	if (error == 0)
2369 		error = error1;
2370 	return (error);
2371 }
2372 
2373 /*
2374  * Nonexistent system call-- signal process (may want to handle it).
2375  * Flag error in case process won't see signal immediately (blocked or ignored).
2376  *
2377  * MPALMOSTSAFE
2378  */
2379 /* ARGSUSED */
2380 int
2381 sys_nosys(struct nosys_args *args)
2382 {
2383 	lwpsignal(curproc, curthread->td_lwp, SIGSYS);
2384 	return (EINVAL);
2385 }
2386 
2387 /*
2388  * Send a SIGIO or SIGURG signal to a process or process group using
2389  * stored credentials rather than those of the current process.
2390  */
2391 void
2392 pgsigio(struct sigio *sigio, int sig, int checkctty)
2393 {
2394 	if (sigio == NULL)
2395 		return;
2396 
2397 	if (sigio->sio_pgid > 0) {
2398 		if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
2399 		             sigio->sio_proc))
2400 			ksignal(sigio->sio_proc, sig);
2401 	} else if (sigio->sio_pgid < 0) {
2402 		struct proc *p;
2403 		struct pgrp *pg = sigio->sio_pgrp;
2404 
2405 		/*
2406 		 * Must interlock all signals against fork
2407 		 */
2408 		pgref(pg);
2409 		lockmgr(&pg->pg_lock, LK_EXCLUSIVE);
2410 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
2411 			if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
2412 			    (checkctty == 0 || (p->p_flags & P_CONTROLT)))
2413 				ksignal(p, sig);
2414 		}
2415 		lockmgr(&pg->pg_lock, LK_RELEASE);
2416 		pgrel(pg);
2417 	}
2418 }
2419 
2420 static int
2421 filt_sigattach(struct knote *kn)
2422 {
2423 	struct proc *p = curproc;
2424 
2425 	kn->kn_ptr.p_proc = p;
2426 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
2427 
2428 	/* XXX lock the proc here while adding to the list? */
2429 	knote_insert(&p->p_klist, kn);
2430 
2431 	return (0);
2432 }
2433 
2434 static void
2435 filt_sigdetach(struct knote *kn)
2436 {
2437 	struct proc *p = kn->kn_ptr.p_proc;
2438 
2439 	knote_remove(&p->p_klist, kn);
2440 }
2441 
2442 /*
2443  * signal knotes are shared with proc knotes, so we apply a mask to
2444  * the hint in order to differentiate them from process hints.  This
2445  * could be avoided by using a signal-specific knote list, but probably
2446  * isn't worth the trouble.
2447  */
2448 static int
2449 filt_signal(struct knote *kn, long hint)
2450 {
2451 	if (hint & NOTE_SIGNAL) {
2452 		hint &= ~NOTE_SIGNAL;
2453 
2454 		if (kn->kn_id == hint)
2455 			kn->kn_data++;
2456 	}
2457 	return (kn->kn_data != 0);
2458 }
2459