xref: /dragonfly/sys/kern/kern_sig.c (revision 1318cd54)
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. 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 static void	proc_stopwait(struct proc *p);
83 
84 static int	filt_sigattach(struct knote *kn);
85 static void	filt_sigdetach(struct knote *kn);
86 static int	filt_signal(struct knote *kn, long hint);
87 
88 struct filterops sig_filtops =
89 	{ FILTEROP_MPSAFE, filt_sigattach, filt_sigdetach, filt_signal };
90 
91 static int	kern_logsigexit = 1;
92 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
93     &kern_logsigexit, 0,
94     "Log processes quitting on abnormal signals to syslog(3)");
95 
96 /*
97  * Can process p, with pcred pc, send the signal sig to process q?
98  */
99 #define CANSIGNAL(q, sig) \
100 	(!p_trespass(curproc->p_ucred, (q)->p_ucred) || \
101 	((sig) == SIGCONT && (q)->p_session == curproc->p_session))
102 
103 /*
104  * Policy -- Can real uid ruid with ucred uc send a signal to process q?
105  */
106 #define CANSIGIO(ruid, uc, q) \
107 	((uc)->cr_uid == 0 || \
108 	    (ruid) == (q)->p_ucred->cr_ruid || \
109 	    (uc)->cr_uid == (q)->p_ucred->cr_ruid || \
110 	    (ruid) == (q)->p_ucred->cr_uid || \
111 	    (uc)->cr_uid == (q)->p_ucred->cr_uid)
112 
113 int sugid_coredump;
114 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RW,
115 	&sugid_coredump, 0, "Enable coredumping set user/group ID processes");
116 
117 static int	do_coredump = 1;
118 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
119 	&do_coredump, 0, "Enable/Disable coredumps");
120 
121 /*
122  * Signal properties and actions.
123  * The array below categorizes the signals and their default actions
124  * according to the following properties:
125  */
126 #define	SA_KILL		0x01		/* terminates process by default */
127 #define	SA_CORE		0x02		/* ditto and coredumps */
128 #define	SA_STOP		0x04		/* suspend process */
129 #define	SA_TTYSTOP	0x08		/* ditto, from tty */
130 #define	SA_IGNORE	0x10		/* ignore by default */
131 #define	SA_CONT		0x20		/* continue if suspended */
132 #define	SA_CANTMASK	0x40		/* non-maskable, catchable */
133 #define SA_CKPT         0x80            /* checkpoint process */
134 
135 
136 static int sigproptbl[NSIG] = {
137         SA_KILL,                /* SIGHUP */
138         SA_KILL,                /* SIGINT */
139         SA_KILL|SA_CORE,        /* SIGQUIT */
140         SA_KILL|SA_CORE,        /* SIGILL */
141         SA_KILL|SA_CORE,        /* SIGTRAP */
142         SA_KILL|SA_CORE,        /* SIGABRT */
143         SA_KILL|SA_CORE,        /* SIGEMT */
144         SA_KILL|SA_CORE,        /* SIGFPE */
145         SA_KILL,                /* SIGKILL */
146         SA_KILL|SA_CORE,        /* SIGBUS */
147         SA_KILL|SA_CORE,        /* SIGSEGV */
148         SA_KILL|SA_CORE,        /* SIGSYS */
149         SA_KILL,                /* SIGPIPE */
150         SA_KILL,                /* SIGALRM */
151         SA_KILL,                /* SIGTERM */
152         SA_IGNORE,              /* SIGURG */
153         SA_STOP,                /* SIGSTOP */
154         SA_STOP|SA_TTYSTOP,     /* SIGTSTP */
155         SA_IGNORE|SA_CONT,      /* SIGCONT */
156         SA_IGNORE,              /* SIGCHLD */
157         SA_STOP|SA_TTYSTOP,     /* SIGTTIN */
158         SA_STOP|SA_TTYSTOP,     /* SIGTTOU */
159         SA_IGNORE,              /* SIGIO */
160         SA_KILL,                /* SIGXCPU */
161         SA_KILL,                /* SIGXFSZ */
162         SA_KILL,                /* SIGVTALRM */
163         SA_KILL,                /* SIGPROF */
164         SA_IGNORE,              /* SIGWINCH  */
165         SA_IGNORE,              /* SIGINFO */
166         SA_KILL,                /* SIGUSR1 */
167         SA_KILL,                /* SIGUSR2 */
168 	SA_IGNORE,              /* SIGTHR */
169 	SA_CKPT,                /* SIGCKPT */
170 	SA_KILL|SA_CKPT,        /* SIGCKPTEXIT */
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 	SA_IGNORE,
201 
202 };
203 
204 static __inline int
205 sigprop(int sig)
206 {
207 
208 	if (sig > 0 && sig < NSIG)
209 		return (sigproptbl[_SIG_IDX(sig)]);
210 	return (0);
211 }
212 
213 static __inline int
214 sig_ffs(sigset_t *set)
215 {
216 	int i;
217 
218 	for (i = 0; i < _SIG_WORDS; i++)
219 		if (set->__bits[i])
220 			return (ffs(set->__bits[i]) + (i * 32));
221 	return (0);
222 }
223 
224 /*
225  * No requirements.
226  */
227 int
228 kern_sigaction(int sig, struct sigaction *act, struct sigaction *oact)
229 {
230 	struct thread *td = curthread;
231 	struct proc *p = td->td_proc;
232 	struct lwp *lp;
233 	struct sigacts *ps = p->p_sigacts;
234 
235 	if (sig <= 0 || sig > _SIG_MAXSIG)
236 		return (EINVAL);
237 
238 	lwkt_gettoken(&p->p_token);
239 
240 	if (oact) {
241 		oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
242 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
243 		oact->sa_flags = 0;
244 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
245 			oact->sa_flags |= SA_ONSTACK;
246 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
247 			oact->sa_flags |= SA_RESTART;
248 		if (SIGISMEMBER(ps->ps_sigreset, sig))
249 			oact->sa_flags |= SA_RESETHAND;
250 		if (SIGISMEMBER(ps->ps_signodefer, sig))
251 			oact->sa_flags |= SA_NODEFER;
252 		if (SIGISMEMBER(ps->ps_siginfo, sig))
253 			oact->sa_flags |= SA_SIGINFO;
254 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDSTOP)
255 			oact->sa_flags |= SA_NOCLDSTOP;
256 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDWAIT)
257 			oact->sa_flags |= SA_NOCLDWAIT;
258 	}
259 	if (act) {
260 		/*
261 		 * Check for invalid requests.  KILL and STOP cannot be
262 		 * caught.
263 		 */
264 		if (sig == SIGKILL || sig == SIGSTOP) {
265 			if (act->sa_handler != SIG_DFL) {
266 				lwkt_reltoken(&p->p_token);
267 				return (EINVAL);
268 			}
269 		}
270 
271 		/*
272 		 * Change setting atomically.
273 		 */
274 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
275 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
276 		if (act->sa_flags & SA_SIGINFO) {
277 			ps->ps_sigact[_SIG_IDX(sig)] =
278 			    (__sighandler_t *)act->sa_sigaction;
279 			SIGADDSET(ps->ps_siginfo, sig);
280 		} else {
281 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
282 			SIGDELSET(ps->ps_siginfo, sig);
283 		}
284 		if (!(act->sa_flags & SA_RESTART))
285 			SIGADDSET(ps->ps_sigintr, sig);
286 		else
287 			SIGDELSET(ps->ps_sigintr, sig);
288 		if (act->sa_flags & SA_ONSTACK)
289 			SIGADDSET(ps->ps_sigonstack, sig);
290 		else
291 			SIGDELSET(ps->ps_sigonstack, sig);
292 		if (act->sa_flags & SA_RESETHAND)
293 			SIGADDSET(ps->ps_sigreset, sig);
294 		else
295 			SIGDELSET(ps->ps_sigreset, sig);
296 		if (act->sa_flags & SA_NODEFER)
297 			SIGADDSET(ps->ps_signodefer, sig);
298 		else
299 			SIGDELSET(ps->ps_signodefer, sig);
300 		if (sig == SIGCHLD) {
301 			if (act->sa_flags & SA_NOCLDSTOP)
302 				p->p_sigacts->ps_flag |= PS_NOCLDSTOP;
303 			else
304 				p->p_sigacts->ps_flag &= ~PS_NOCLDSTOP;
305 			if (act->sa_flags & SA_NOCLDWAIT) {
306 				/*
307 				 * Paranoia: since SA_NOCLDWAIT is implemented
308 				 * by reparenting the dying child to PID 1 (and
309 				 * trust it to reap the zombie), PID 1 itself
310 				 * is forbidden to set SA_NOCLDWAIT.
311 				 */
312 				if (p->p_pid == 1)
313 					p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
314 				else
315 					p->p_sigacts->ps_flag |= PS_NOCLDWAIT;
316 			} else {
317 				p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
318 			}
319 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
320 				ps->ps_flag |= PS_CLDSIGIGN;
321 			else
322 				ps->ps_flag &= ~PS_CLDSIGIGN;
323 		}
324 		/*
325 		 * Set bit in p_sigignore for signals that are set to SIG_IGN,
326 		 * and for signals set to SIG_DFL where the default is to
327 		 * ignore. However, don't put SIGCONT in p_sigignore, as we
328 		 * have to restart the process.
329 		 *
330 		 * Also remove the signal from the process and lwp signal
331 		 * list.
332 		 */
333 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
334 		    (sigprop(sig) & SA_IGNORE &&
335 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
336 			SIGDELSET_ATOMIC(p->p_siglist, sig);
337 			FOREACH_LWP_IN_PROC(lp, p) {
338 				spin_lock(&lp->lwp_spin);
339 				SIGDELSET(lp->lwp_siglist, sig);
340 				spin_unlock(&lp->lwp_spin);
341 			}
342 			if (sig != SIGCONT) {
343 				/* easier in ksignal */
344 				SIGADDSET(p->p_sigignore, sig);
345 			}
346 			SIGDELSET(p->p_sigcatch, sig);
347 		} else {
348 			SIGDELSET(p->p_sigignore, sig);
349 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
350 				SIGDELSET(p->p_sigcatch, sig);
351 			else
352 				SIGADDSET(p->p_sigcatch, sig);
353 		}
354 	}
355 	lwkt_reltoken(&p->p_token);
356 	return (0);
357 }
358 
359 int
360 sys_sigaction(struct sigaction_args *uap)
361 {
362 	struct sigaction act, oact;
363 	struct sigaction *actp, *oactp;
364 	int error;
365 
366 	actp = (uap->act != NULL) ? &act : NULL;
367 	oactp = (uap->oact != NULL) ? &oact : NULL;
368 	if (actp) {
369 		error = copyin(uap->act, actp, sizeof(act));
370 		if (error)
371 			return (error);
372 	}
373 	error = kern_sigaction(uap->sig, actp, oactp);
374 	if (oactp && !error) {
375 		error = copyout(oactp, uap->oact, sizeof(oact));
376 	}
377 	return (error);
378 }
379 
380 /*
381  * Initialize signal state for process 0;
382  * set to ignore signals that are ignored by default.
383  */
384 void
385 siginit(struct proc *p)
386 {
387 	int i;
388 
389 	for (i = 1; i <= NSIG; i++)
390 		if (sigprop(i) & SA_IGNORE && i != SIGCONT)
391 			SIGADDSET(p->p_sigignore, i);
392 }
393 
394 /*
395  * Reset signals for an exec of the specified process.
396  */
397 void
398 execsigs(struct proc *p)
399 {
400 	struct sigacts *ps = p->p_sigacts;
401 	struct lwp *lp;
402 	int sig;
403 
404 	lp = ONLY_LWP_IN_PROC(p);
405 
406 	/*
407 	 * Reset caught signals.  Held signals remain held
408 	 * through p_sigmask (unless they were caught,
409 	 * and are now ignored by default).
410 	 */
411 	while (SIGNOTEMPTY(p->p_sigcatch)) {
412 		sig = sig_ffs(&p->p_sigcatch);
413 		SIGDELSET(p->p_sigcatch, sig);
414 		if (sigprop(sig) & SA_IGNORE) {
415 			if (sig != SIGCONT)
416 				SIGADDSET(p->p_sigignore, sig);
417 			SIGDELSET_ATOMIC(p->p_siglist, sig);
418 			/* don't need spinlock */
419 			SIGDELSET(lp->lwp_siglist, sig);
420 		}
421 		ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
422 	}
423 
424 	/*
425 	 * Reset stack state to the user stack.
426 	 * Clear set of signals caught on the signal stack.
427 	 */
428 	lp->lwp_sigstk.ss_flags = SS_DISABLE;
429 	lp->lwp_sigstk.ss_size = 0;
430 	lp->lwp_sigstk.ss_sp = NULL;
431 	lp->lwp_flags &= ~LWP_ALTSTACK;
432 	/*
433 	 * Reset no zombies if child dies flag as Solaris does.
434 	 */
435 	p->p_sigacts->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
436 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
437 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
438 }
439 
440 /*
441  * kern_sigprocmask() - MP SAFE ONLY IF p == curproc
442  *
443  *	Manipulate signal mask.  This routine is MP SAFE *ONLY* if
444  *	p == curproc.
445  */
446 int
447 kern_sigprocmask(int how, sigset_t *set, sigset_t *oset)
448 {
449 	struct thread *td = curthread;
450 	struct lwp *lp = td->td_lwp;
451 	struct proc *p = td->td_proc;
452 	int error;
453 
454 	lwkt_gettoken(&p->p_token);
455 
456 	if (oset != NULL)
457 		*oset = lp->lwp_sigmask;
458 
459 	error = 0;
460 	if (set != NULL) {
461 		switch (how) {
462 		case SIG_BLOCK:
463 			SIG_CANTMASK(*set);
464 			SIGSETOR(lp->lwp_sigmask, *set);
465 			break;
466 		case SIG_UNBLOCK:
467 			SIGSETNAND(lp->lwp_sigmask, *set);
468 			break;
469 		case SIG_SETMASK:
470 			SIG_CANTMASK(*set);
471 			lp->lwp_sigmask = *set;
472 			break;
473 		default:
474 			error = EINVAL;
475 			break;
476 		}
477 	}
478 
479 	lwkt_reltoken(&p->p_token);
480 
481 	return (error);
482 }
483 
484 /*
485  * sigprocmask()
486  *
487  * MPSAFE
488  */
489 int
490 sys_sigprocmask(struct sigprocmask_args *uap)
491 {
492 	sigset_t set, oset;
493 	sigset_t *setp, *osetp;
494 	int error;
495 
496 	setp = (uap->set != NULL) ? &set : NULL;
497 	osetp = (uap->oset != NULL) ? &oset : NULL;
498 	if (setp) {
499 		error = copyin(uap->set, setp, sizeof(set));
500 		if (error)
501 			return (error);
502 	}
503 	error = kern_sigprocmask(uap->how, setp, osetp);
504 	if (osetp && !error) {
505 		error = copyout(osetp, uap->oset, sizeof(oset));
506 	}
507 	return (error);
508 }
509 
510 /*
511  * MPSAFE
512  */
513 int
514 kern_sigpending(struct __sigset *set)
515 {
516 	struct lwp *lp = curthread->td_lwp;
517 
518 	*set = lwp_sigpend(lp);
519 
520 	return (0);
521 }
522 
523 /*
524  * MPSAFE
525  */
526 int
527 sys_sigpending(struct sigpending_args *uap)
528 {
529 	sigset_t set;
530 	int error;
531 
532 	error = kern_sigpending(&set);
533 
534 	if (error == 0)
535 		error = copyout(&set, uap->set, sizeof(set));
536 	return (error);
537 }
538 
539 /*
540  * Suspend process until signal, providing mask to be set
541  * in the meantime.
542  *
543  * MPSAFE
544  */
545 int
546 kern_sigsuspend(struct __sigset *set)
547 {
548 	struct thread *td = curthread;
549 	struct lwp *lp = td->td_lwp;
550 	struct proc *p = td->td_proc;
551 	struct sigacts *ps = p->p_sigacts;
552 
553 	/*
554 	 * When returning from sigsuspend, we want
555 	 * the old mask to be restored after the
556 	 * signal handler has finished.  Thus, we
557 	 * save it here and mark the sigacts structure
558 	 * to indicate this.
559 	 */
560 	lp->lwp_oldsigmask = lp->lwp_sigmask;
561 	lp->lwp_flags |= LWP_OLDMASK;
562 
563 	SIG_CANTMASK(*set);
564 	lp->lwp_sigmask = *set;
565 	while (tsleep(ps, PCATCH, "pause", 0) == 0)
566 		/* void */;
567 	/* always return EINTR rather than ERESTART... */
568 	return (EINTR);
569 }
570 
571 /*
572  * Note nonstandard calling convention: libc stub passes mask, not
573  * pointer, to save a copyin.
574  *
575  * MPSAFE
576  */
577 int
578 sys_sigsuspend(struct sigsuspend_args *uap)
579 {
580 	sigset_t mask;
581 	int error;
582 
583 	error = copyin(uap->sigmask, &mask, sizeof(mask));
584 	if (error)
585 		return (error);
586 
587 	error = kern_sigsuspend(&mask);
588 
589 	return (error);
590 }
591 
592 /*
593  * MPSAFE
594  */
595 int
596 kern_sigaltstack(struct sigaltstack *ss, struct sigaltstack *oss)
597 {
598 	struct thread *td = curthread;
599 	struct lwp *lp = td->td_lwp;
600 	struct proc *p = td->td_proc;
601 
602 	if ((lp->lwp_flags & LWP_ALTSTACK) == 0)
603 		lp->lwp_sigstk.ss_flags |= SS_DISABLE;
604 
605 	if (oss)
606 		*oss = lp->lwp_sigstk;
607 
608 	if (ss) {
609 		if (ss->ss_flags & ~SS_DISABLE)
610 			return (EINVAL);
611 		if (ss->ss_flags & SS_DISABLE) {
612 			if (lp->lwp_sigstk.ss_flags & SS_ONSTACK)
613 				return (EPERM);
614 			lp->lwp_flags &= ~LWP_ALTSTACK;
615 			lp->lwp_sigstk.ss_flags = ss->ss_flags;
616 		} else {
617 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
618 				return (ENOMEM);
619 			lp->lwp_flags |= LWP_ALTSTACK;
620 			lp->lwp_sigstk = *ss;
621 		}
622 	}
623 
624 	return (0);
625 }
626 
627 /*
628  * MPSAFE
629  */
630 int
631 sys_sigaltstack(struct sigaltstack_args *uap)
632 {
633 	stack_t ss, oss;
634 	int error;
635 
636 	if (uap->ss) {
637 		error = copyin(uap->ss, &ss, sizeof(ss));
638 		if (error)
639 			return (error);
640 	}
641 
642 	error = kern_sigaltstack(uap->ss ? &ss : NULL, uap->oss ? &oss : NULL);
643 
644 	if (error == 0 && uap->oss)
645 		error = copyout(&oss, uap->oss, sizeof(*uap->oss));
646 	return (error);
647 }
648 
649 /*
650  * Common code for kill process group/broadcast kill.
651  * cp is calling process.
652  */
653 struct killpg_info {
654 	int nfound;
655 	int sig;
656 };
657 
658 static int killpg_all_callback(struct proc *p, void *data);
659 
660 static int
661 dokillpg(int sig, int pgid, int all)
662 {
663 	struct killpg_info info;
664 	struct proc *cp = curproc;
665 	struct proc *p;
666 	struct pgrp *pgrp;
667 
668 	info.nfound = 0;
669 	info.sig = sig;
670 
671 	if (all) {
672 		/*
673 		 * broadcast
674 		 */
675 		allproc_scan(killpg_all_callback, &info);
676 	} else {
677 		if (pgid == 0) {
678 			/*
679 			 * zero pgid means send to my process group.
680 			 */
681 			pgrp = cp->p_pgrp;
682 			pgref(pgrp);
683 		} else {
684 			pgrp = pgfind(pgid);
685 			if (pgrp == NULL)
686 				return (ESRCH);
687 		}
688 
689 		/*
690 		 * Must interlock all signals against fork
691 		 */
692 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
693 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
694 			if (p->p_pid <= 1 ||
695 			    p->p_stat == SZOMB ||
696 			    (p->p_flags & P_SYSTEM) ||
697 			    !CANSIGNAL(p, sig)) {
698 				continue;
699 			}
700 			++info.nfound;
701 			if (sig)
702 				ksignal(p, sig);
703 		}
704 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
705 		pgrel(pgrp);
706 	}
707 	return (info.nfound ? 0 : ESRCH);
708 }
709 
710 static int
711 killpg_all_callback(struct proc *p, void *data)
712 {
713 	struct killpg_info *info = data;
714 
715 	if (p->p_pid <= 1 || (p->p_flags & P_SYSTEM) ||
716 	    p == curproc || !CANSIGNAL(p, info->sig)) {
717 		return (0);
718 	}
719 	++info->nfound;
720 	if (info->sig)
721 		ksignal(p, info->sig);
722 	return(0);
723 }
724 
725 /*
726  * Send a general signal to a process or LWPs within that process.
727  *
728  * Note that new signals cannot be sent if a process is exiting or already
729  * a zombie, but we return success anyway as userland is likely to not handle
730  * the race properly.
731  *
732  * No requirements.
733  */
734 int
735 kern_kill(int sig, pid_t pid, lwpid_t tid)
736 {
737 	int t;
738 
739 	if ((u_int)sig > _SIG_MAXSIG)
740 		return (EINVAL);
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 				return (ESRCH);
755 			PRELE(p);
756 			return (0);
757 		}
758 		if (p != curproc) {
759 			lwkt_gettoken_shared(&p->p_token);
760 			if (!CANSIGNAL(p, sig)) {
761 				lwkt_reltoken(&p->p_token);
762 				PRELE(p);
763 				return (EPERM);
764 			}
765 			lwkt_reltoken(&p->p_token);
766 		}
767 
768 		/*
769 		 * NOP if the process is exiting.  Note that lwpsignal() is
770 		 * called directly with P_WEXIT set to kill individual LWPs
771 		 * during exit, which is allowed.
772 		 */
773 		if (p->p_flags & P_WEXIT) {
774 			PRELE(p);
775 			return (0);
776 		}
777 		if (tid != -1) {
778 			lwkt_gettoken_shared(&p->p_token);
779 			lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, tid);
780 			if (lp == NULL) {
781 				lwkt_reltoken(&p->p_token);
782 				PRELE(p);
783 				return (ESRCH);
784 			}
785 			LWPHOLD(lp);
786 			lwkt_reltoken(&p->p_token);
787 		}
788 		if (sig)
789 			lwpsignal(p, lp, sig);
790 		if (lp)
791 			LWPRELE(lp);
792 		PRELE(p);
793 
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 		return (EINVAL);
803 
804 	switch (pid) {
805 	case -1:		/* broadcast signal */
806 		t = (dokillpg(sig, 0, 1));
807 		break;
808 	case 0:			/* signal own process group */
809 		t = (dokillpg(sig, 0, 0));
810 		break;
811 	default:		/* negative explicit process group */
812 		t = (dokillpg(sig, -pid, 0));
813 		break;
814 	}
815 	return t;
816 }
817 
818 int
819 sys_kill(struct kill_args *uap)
820 {
821 	int error;
822 
823 	error = kern_kill(uap->signum, uap->pid, -1);
824 	return (error);
825 }
826 
827 int
828 sys_lwp_kill(struct lwp_kill_args *uap)
829 {
830 	int error;
831 	pid_t pid = uap->pid;
832 
833 	/*
834 	 * A tid is mandatory for lwp_kill(), otherwise
835 	 * you could simply use kill().
836 	 */
837 	if (uap->tid == -1)
838 		return (EINVAL);
839 
840 	/*
841 	 * To save on a getpid() function call for intra-process
842 	 * signals, pid == -1 means current process.
843 	 */
844 	if (pid == -1)
845 		pid = curproc->p_pid;
846 
847 	error = kern_kill(uap->signum, pid, uap->tid);
848 	return (error);
849 }
850 
851 /*
852  * Send a signal to a process group.
853  */
854 void
855 gsignal(int pgid, int sig)
856 {
857 	struct pgrp *pgrp;
858 
859 	if (pgid && (pgrp = pgfind(pgid)))
860 		pgsignal(pgrp, sig, 0);
861 }
862 
863 /*
864  * Send a signal to a process group.  If checktty is 1,
865  * limit to members which have a controlling terminal.
866  *
867  * pg_lock interlocks against a fork that might be in progress, to
868  * ensure that the new child process picks up the signal.
869  */
870 void
871 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
872 {
873 	struct proc *p;
874 
875 	/*
876 	 * Must interlock all signals against fork
877 	 */
878 	if (pgrp) {
879 		pgref(pgrp);
880 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
881 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
882 			if (checkctty == 0 || p->p_flags & P_CONTROLT)
883 				ksignal(p, sig);
884 		}
885 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
886 		pgrel(pgrp);
887 	}
888 }
889 
890 /*
891  * Send a signal caused by a trap to the current lwp.  If it will be caught
892  * immediately, deliver it with correct code.  Otherwise, post it normally.
893  *
894  * These signals may ONLY be delivered to the specified lwp and may never
895  * be delivered to the process generically.
896  */
897 void
898 trapsignal(struct lwp *lp, int sig, u_long code)
899 {
900 	struct proc *p = lp->lwp_proc;
901 	struct sigacts *ps = p->p_sigacts;
902 
903 	/*
904 	 * If we are a virtual kernel running an emulated user process
905 	 * context, switch back to the virtual kernel context before
906 	 * trying to post the signal.
907 	 */
908 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
909 		struct trapframe *tf = lp->lwp_md.md_regs;
910 		tf->tf_trapno = 0;
911 		vkernel_trap(lp, tf);
912 	}
913 
914 	if ((p->p_flags & P_TRACED) == 0 && SIGISMEMBER(p->p_sigcatch, sig) &&
915 	    !SIGISMEMBER(lp->lwp_sigmask, sig)) {
916 		lp->lwp_ru.ru_nsignals++;
917 #ifdef KTRACE
918 		if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
919 			ktrpsig(lp, sig, ps->ps_sigact[_SIG_IDX(sig)],
920 				&lp->lwp_sigmask, code);
921 #endif
922 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], sig,
923 						&lp->lwp_sigmask, code);
924 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
925 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
926 			SIGADDSET(lp->lwp_sigmask, sig);
927 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
928 			/*
929 			 * See kern_sigaction() for origin of this code.
930 			 */
931 			SIGDELSET(p->p_sigcatch, sig);
932 			if (sig != SIGCONT &&
933 			    sigprop(sig) & SA_IGNORE)
934 				SIGADDSET(p->p_sigignore, sig);
935 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
936 		}
937 	} else {
938 		lp->lwp_code = code;	/* XXX for core dump/debugger */
939 		lp->lwp_sig = sig;	/* XXX to verify code */
940 		lwpsignal(p, lp, sig);
941 	}
942 }
943 
944 /*
945  * Find a suitable lwp to deliver the signal to.  Returns NULL if all
946  * lwps hold the signal blocked.
947  *
948  * Caller must hold p->p_token.
949  *
950  * Returns a lp or NULL.  If non-NULL the lp is held and its token is
951  * acquired.
952  */
953 static struct lwp *
954 find_lwp_for_signal(struct proc *p, int sig)
955 {
956 	struct lwp *lp;
957 	struct lwp *run, *sleep, *stop;
958 
959 	/*
960 	 * If the running/preempted thread belongs to the proc to which
961 	 * the signal is being delivered and this thread does not block
962 	 * the signal, then we can avoid a context switch by delivering
963 	 * the signal to this thread, because it will return to userland
964 	 * soon anyways.
965 	 */
966 	lp = lwkt_preempted_proc();
967 	if (lp != NULL && lp->lwp_proc == p) {
968 		LWPHOLD(lp);
969 		lwkt_gettoken(&lp->lwp_token);
970 		if (!SIGISMEMBER(lp->lwp_sigmask, sig)) {
971 			/* return w/ token held */
972 			return (lp);
973 		}
974 		lwkt_reltoken(&lp->lwp_token);
975 		LWPRELE(lp);
976 	}
977 
978 	run = sleep = stop = NULL;
979 	FOREACH_LWP_IN_PROC(lp, p) {
980 		/*
981 		 * If the signal is being blocked by the lwp, then this
982 		 * lwp is not eligible for receiving the signal.
983 		 */
984 		LWPHOLD(lp);
985 		lwkt_gettoken(&lp->lwp_token);
986 
987 		if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
988 			lwkt_reltoken(&lp->lwp_token);
989 			LWPRELE(lp);
990 			continue;
991 		}
992 
993 		switch (lp->lwp_stat) {
994 		case LSRUN:
995 			if (sleep) {
996 				lwkt_token_swap();
997 				lwkt_reltoken(&sleep->lwp_token);
998 				LWPRELE(sleep);
999 				sleep = NULL;
1000 				run = lp;
1001 			} else if (stop) {
1002 				lwkt_token_swap();
1003 				lwkt_reltoken(&stop->lwp_token);
1004 				LWPRELE(stop);
1005 				stop = NULL;
1006 				run = lp;
1007 			} else {
1008 				run = lp;
1009 			}
1010 			break;
1011 		case LSSLEEP:
1012 			if (lp->lwp_flags & LWP_SINTR) {
1013 				if (sleep) {
1014 					lwkt_reltoken(&lp->lwp_token);
1015 					LWPRELE(lp);
1016 				} else if (stop) {
1017 					lwkt_token_swap();
1018 					lwkt_reltoken(&stop->lwp_token);
1019 					LWPRELE(stop);
1020 					stop = NULL;
1021 					sleep = lp;
1022 				} else {
1023 					sleep = lp;
1024 				}
1025 			} else {
1026 				lwkt_reltoken(&lp->lwp_token);
1027 				LWPRELE(lp);
1028 			}
1029 			break;
1030 		case LSSTOP:
1031 			if (sleep) {
1032 				lwkt_reltoken(&lp->lwp_token);
1033 				LWPRELE(lp);
1034 			} else if (stop) {
1035 				lwkt_reltoken(&lp->lwp_token);
1036 				LWPRELE(lp);
1037 			} else {
1038 				stop = lp;
1039 			}
1040 			break;
1041 		}
1042 		if (run)
1043 			break;
1044 	}
1045 
1046 	if (run != NULL)
1047 		return (run);
1048 	else if (sleep != NULL)
1049 		return (sleep);
1050 	else
1051 		return (stop);
1052 }
1053 
1054 /*
1055  * Send the signal to the process.  If the signal has an action, the action
1056  * is usually performed by the target process rather than the caller; we add
1057  * the signal to the set of pending signals for the process.
1058  *
1059  * Exceptions:
1060  *   o When a stop signal is sent to a sleeping process that takes the
1061  *     default action, the process is stopped without awakening it.
1062  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
1063  *     regardless of the signal action (eg, blocked or ignored).
1064  *
1065  * Other ignored signals are discarded immediately.
1066  *
1067  * If the caller wishes to call this function from a hard code section the
1068  * caller must already hold p->p_token (see kern_clock.c).
1069  *
1070  * No requirements.
1071  */
1072 void
1073 ksignal(struct proc *p, int sig)
1074 {
1075 	lwpsignal(p, NULL, sig);
1076 }
1077 
1078 /*
1079  * The core for ksignal.  lp may be NULL, then a suitable thread
1080  * will be chosen.  If not, lp MUST be a member of p.
1081  *
1082  * If the caller wishes to call this function from a hard code section the
1083  * caller must already hold p->p_token.
1084  *
1085  * No requirements.
1086  */
1087 void
1088 lwpsignal(struct proc *p, struct lwp *lp, int sig)
1089 {
1090 	struct proc *q;
1091 	sig_t action;
1092 	int prop;
1093 
1094 	if (sig > _SIG_MAXSIG || sig <= 0) {
1095 		kprintf("lwpsignal: signal %d\n", sig);
1096 		panic("lwpsignal signal number");
1097 	}
1098 
1099 	KKASSERT(lp == NULL || lp->lwp_proc == p);
1100 
1101 	/*
1102 	 * We don't want to race... well, all sorts of things.  Get appropriate
1103 	 * tokens.
1104 	 *
1105 	 * Don't try to deliver a generic signal to an exiting process,
1106 	 * the signal structures could be in flux.  We check the LWP later
1107 	 * on.
1108 	 */
1109 	PHOLD(p);
1110 	if (lp) {
1111 		LWPHOLD(lp);
1112 		lwkt_gettoken(&lp->lwp_token);
1113 	} else {
1114 		lwkt_gettoken(&p->p_token);
1115 		if (p->p_flags & P_WEXIT)
1116 			goto out;
1117 	}
1118 
1119 	prop = sigprop(sig);
1120 
1121 	/*
1122 	 * If proc is traced, always give parent a chance;
1123 	 * if signal event is tracked by procfs, give *that*
1124 	 * a chance, as well.
1125 	 */
1126 	if ((p->p_flags & P_TRACED) || (p->p_stops & S_SIG)) {
1127 		action = SIG_DFL;
1128 	} else {
1129 		/*
1130 		 * Do not try to deliver signals to an exiting lwp other
1131 		 * than SIGKILL.  Note that we must still deliver the signal
1132 		 * if P_WEXIT is set in the process flags.
1133 		 */
1134 		if (lp && (lp->lwp_mpflags & LWP_MP_WEXIT) && sig != SIGKILL) {
1135 			lwkt_reltoken(&lp->lwp_token);
1136 			LWPRELE(lp);
1137 			PRELE(p);
1138 			return;
1139 		}
1140 
1141 		/*
1142 		 * If the signal is being ignored, then we forget about
1143 		 * it immediately.  NOTE: We don't set SIGCONT in p_sigignore,
1144 		 * and if it is set to SIG_IGN, action will be SIG_DFL here.
1145 		 */
1146 		if (SIGISMEMBER(p->p_sigignore, sig)) {
1147 			/*
1148 			 * Even if a signal is set SIG_IGN, it may still be
1149 			 * lurking in a kqueue.
1150 			 */
1151 			KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1152 			if (lp) {
1153 				lwkt_reltoken(&lp->lwp_token);
1154 				LWPRELE(lp);
1155 			} else {
1156 				lwkt_reltoken(&p->p_token);
1157 			}
1158 			PRELE(p);
1159 			return;
1160 		}
1161 		if (SIGISMEMBER(p->p_sigcatch, sig))
1162 			action = SIG_CATCH;
1163 		else
1164 			action = SIG_DFL;
1165 	}
1166 
1167 	/*
1168 	 * If continuing, clear any pending STOP signals for the whole
1169 	 * process.
1170 	 */
1171 	if (prop & SA_CONT) {
1172 		lwkt_gettoken(&p->p_token);
1173 		SIG_STOPSIGMASK_ATOMIC(p->p_siglist);
1174 		lwkt_reltoken(&p->p_token);
1175 	}
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 			if (lp) {
1187 				lwkt_reltoken(&lp->lwp_token);
1188 				LWPRELE(lp);
1189 			} else {
1190 				lwkt_reltoken(&p->p_token);
1191 			}
1192 			PRELE(p);
1193 		        return;
1194 		}
1195 		lwkt_gettoken(&p->p_token);
1196 		SIG_CONTSIGMASK_ATOMIC(p->p_siglist);
1197 		p->p_flags &= ~P_CONTINUED;
1198 		lwkt_reltoken(&p->p_token);
1199 	}
1200 
1201 	if (p->p_stat == SSTOP) {
1202 		/*
1203 		 * Nobody can handle this signal, add it to the lwp or
1204 		 * process pending list
1205 		 */
1206 		lwkt_gettoken(&p->p_token);
1207 		if (p->p_stat != SSTOP) {
1208 			lwkt_reltoken(&p->p_token);
1209 			goto not_stopped;
1210 		}
1211 		if (lp) {
1212 			spin_lock(&lp->lwp_spin);
1213 			SIGADDSET(lp->lwp_siglist, sig);
1214 			spin_unlock(&lp->lwp_spin);
1215 		} else {
1216 			SIGADDSET_ATOMIC(p->p_siglist, sig);
1217 		}
1218 
1219 		/*
1220 		 * If the process is stopped and is being traced, then no
1221 		 * further action is necessary.
1222 		 */
1223 		if (p->p_flags & P_TRACED) {
1224 			lwkt_reltoken(&p->p_token);
1225 			goto out;
1226 		}
1227 
1228 		/*
1229 		 * If the process is stopped and receives a KILL signal,
1230 		 * make the process runnable.
1231 		 */
1232 		if (sig == SIGKILL) {
1233 			proc_unstop(p, SSTOP);
1234 			lwkt_reltoken(&p->p_token);
1235 			goto active_process;
1236 		}
1237 
1238 		/*
1239 		 * If the process is stopped and receives a CONT signal,
1240 		 * then try to make the process runnable again.
1241 		 */
1242 		if (prop & SA_CONT) {
1243 			/*
1244 			 * If SIGCONT is default (or ignored), we continue the
1245 			 * process but don't leave the signal in p_siglist, as
1246 			 * it has no further action.  If SIGCONT is held, we
1247 			 * continue the process and leave the signal in
1248 			 * p_siglist.  If the process catches SIGCONT, let it
1249 			 * handle the signal itself.
1250 			 *
1251 			 * XXX what if the signal is being held blocked?
1252 			 *
1253 			 * Token required to interlock kern_wait().
1254 			 * Reparenting can also cause a race so we have to
1255 			 * hold (q).
1256 			 */
1257 			q = p->p_pptr;
1258 			PHOLD(q);
1259 			lwkt_gettoken(&q->p_token);
1260 			p->p_flags |= P_CONTINUED;
1261 			wakeup(q);
1262 			if (action == SIG_DFL)
1263 				SIGDELSET_ATOMIC(p->p_siglist, sig);
1264 			proc_unstop(p, SSTOP);
1265 			lwkt_reltoken(&q->p_token);
1266 			PRELE(q);
1267 			lwkt_reltoken(&p->p_token);
1268 			if (action == SIG_CATCH)
1269 				goto active_process;
1270 			goto out;
1271 		}
1272 
1273 		/*
1274 		 * If the process is stopped and receives another STOP
1275 		 * signal, we do not need to stop it again.  If we did
1276 		 * the shell could get confused.
1277 		 *
1278 		 * However, if the current/preempted lwp is part of the
1279 		 * process receiving the signal, we need to keep it,
1280 		 * so that this lwp can stop in issignal() later, as
1281 		 * we don't want to wait until it reaches userret!
1282 		 */
1283 		if (prop & SA_STOP) {
1284 			if (lwkt_preempted_proc() == NULL ||
1285 			    lwkt_preempted_proc()->lwp_proc != p) {
1286 				SIGDELSET_ATOMIC(p->p_siglist, sig);
1287 			}
1288 		}
1289 
1290 		/*
1291 		 * Otherwise the process is stopped and it received some
1292 		 * signal, which does not change its stopped state.  When
1293 		 * the process is continued a wakeup(p) will be issued which
1294 		 * will wakeup any threads sleeping in tstop().
1295 		 */
1296 		lwkt_reltoken(&p->p_token);
1297 		goto out;
1298 		/* NOTREACHED */
1299 	}
1300 not_stopped:
1301 	;
1302 	/* else not stopped */
1303 active_process:
1304 
1305 	/*
1306 	 * Never deliver a lwp-specific signal to a random lwp.
1307 	 */
1308 	if (lp == NULL) {
1309 		/* NOTE: returns lp w/ token held */
1310 		lp = find_lwp_for_signal(p, sig);
1311 		if (lp) {
1312 			if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
1313 				lwkt_reltoken(&lp->lwp_token);
1314 				LWPRELE(lp);
1315 				lp = NULL;
1316 				/* maintain proc token */
1317 			} else {
1318 				lwkt_token_swap();
1319 				lwkt_reltoken(&p->p_token);
1320 				/* maintain lp token */
1321 			}
1322 		}
1323 	}
1324 
1325 	/*
1326 	 * Deliver to the process generically if (1) the signal is being
1327 	 * sent to any thread or (2) we could not find a thread to deliver
1328 	 * it to.
1329 	 */
1330 	if (lp == NULL) {
1331 		KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1332 		SIGADDSET_ATOMIC(p->p_siglist, sig);
1333 		goto out;
1334 	}
1335 
1336 	/*
1337 	 * Deliver to a specific LWP whether it masks it or not.  It will
1338 	 * not be dispatched if masked but we must still deliver it.
1339 	 */
1340 	if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
1341 	    (p->p_flags & P_TRACED) == 0) {
1342 		lwkt_gettoken(&p->p_token);
1343 		p->p_nice = NZERO;
1344 		lwkt_reltoken(&p->p_token);
1345 	}
1346 
1347 	/*
1348 	 * If the process receives a STOP signal which indeed needs to
1349 	 * stop the process, do so.  If the process chose to catch the
1350 	 * signal, it will be treated like any other signal.
1351 	 */
1352 	if ((prop & SA_STOP) && action == SIG_DFL) {
1353 		/*
1354 		 * If a child holding parent blocked, stopping
1355 		 * could cause deadlock.  Take no action at this
1356 		 * time.
1357 		 */
1358 		lwkt_gettoken(&p->p_token);
1359 		if (p->p_flags & P_PPWAIT) {
1360 			SIGADDSET_ATOMIC(p->p_siglist, sig);
1361 			lwkt_reltoken(&p->p_token);
1362 			goto out;
1363 		}
1364 
1365 		/*
1366 		 * Do not actually try to manipulate the process, but simply
1367 		 * stop it.  Lwps will stop as soon as they safely can.
1368 		 *
1369 		 * Ignore stop if the process is exiting.
1370 		 */
1371 		if ((p->p_flags & P_WEXIT) == 0) {
1372 			p->p_xstat = sig;
1373 			proc_stop(p, SSTOP);
1374 		}
1375 		lwkt_reltoken(&p->p_token);
1376 		goto out;
1377 	}
1378 
1379 	/*
1380 	 * If it is a CONT signal with default action, just ignore it.
1381 	 */
1382 	if ((prop & SA_CONT) && action == SIG_DFL)
1383 		goto out;
1384 
1385 	/*
1386 	 * Mark signal pending at this specific thread.
1387 	 */
1388 	spin_lock(&lp->lwp_spin);
1389 	SIGADDSET(lp->lwp_siglist, sig);
1390 	spin_unlock(&lp->lwp_spin);
1391 
1392 	lwp_signotify(lp);
1393 
1394 out:
1395 	if (lp) {
1396 		lwkt_reltoken(&lp->lwp_token);
1397 		LWPRELE(lp);
1398 	} else {
1399 		lwkt_reltoken(&p->p_token);
1400 	}
1401 	PRELE(p);
1402 }
1403 
1404 /*
1405  * Notify the LWP that a signal has arrived.  The LWP does not have to be
1406  * sleeping on the current cpu.
1407  *
1408  * p->p_token and lp->lwp_token must be held on call.
1409  *
1410  * We can only safely schedule the thread on its current cpu and only if
1411  * one of the SINTR flags is set.  If an SINTR flag is set AND we are on
1412  * the correct cpu we are properly interlocked, otherwise we could be
1413  * racing other thread transition states (or the lwp is on the user scheduler
1414  * runq but not scheduled) and must not do anything.
1415  *
1416  * Since we hold the lwp token we know the lwp cannot be ripped out from
1417  * under us so we can safely hold it to prevent it from being ripped out
1418  * from under us if we are forced to IPI another cpu to make the local
1419  * checks there.
1420  *
1421  * Adjustment of lp->lwp_stat can only occur when we hold the lwp_token,
1422  * which we won't in an IPI so any fixups have to be done here, effectively
1423  * replicating part of what setrunnable() does.
1424  */
1425 static void
1426 lwp_signotify(struct lwp *lp)
1427 {
1428 	thread_t dtd;
1429 
1430 	ASSERT_LWKT_TOKEN_HELD(&lp->lwp_token);
1431 	dtd = lp->lwp_thread;
1432 
1433 	crit_enter();
1434 	if (lp == lwkt_preempted_proc()) {
1435 		/*
1436 		 * lwp is on the current cpu AND it is currently running
1437 		 * (we preempted it).
1438 		 */
1439 		signotify();
1440 	} else if (lp->lwp_flags & LWP_SINTR) {
1441 		/*
1442 		 * lwp is sitting in tsleep() with PCATCH set
1443 		 */
1444 		if (dtd->td_gd == mycpu) {
1445 			setrunnable(lp);
1446 		} else {
1447 			/*
1448 			 * We can only adjust lwp_stat while we hold the
1449 			 * lwp_token, and we won't in the IPI function.
1450 			 */
1451 			LWPHOLD(lp);
1452 			if (lp->lwp_stat == LSSTOP)
1453 				lp->lwp_stat = LSSLEEP;
1454 			lwkt_send_ipiq(dtd->td_gd, lwp_signotify_remote, lp);
1455 		}
1456 	} else if (dtd->td_flags & TDF_SINTR) {
1457 		/*
1458 		 * lwp is sitting in lwkt_sleep() with PCATCH set.
1459 		 */
1460 		if (dtd->td_gd == mycpu) {
1461 			setrunnable(lp);
1462 		} else {
1463 			/*
1464 			 * We can only adjust lwp_stat while we hold the
1465 			 * lwp_token, and we won't in the IPI function.
1466 			 */
1467 			LWPHOLD(lp);
1468 			if (lp->lwp_stat == LSSTOP)
1469 				lp->lwp_stat = LSSLEEP;
1470 			lwkt_send_ipiq(dtd->td_gd, lwp_signotify_remote, lp);
1471 		}
1472 	} else {
1473 		/*
1474 		 * Otherwise the lwp is either in some uninterruptible state
1475 		 * or it is on the userland scheduler's runqueue waiting to
1476 		 * be scheduled to a cpu, or it is running in userland.  We
1477 		 * generally want to send an IPI so a running target gets the
1478 		 * signal ASAP, otherwise a scheduler-tick worth of latency
1479 		 * will occur.
1480 		 *
1481 		 * Issue an IPI to the remote cpu to knock it into the kernel,
1482 		 * remote cpu will issue the cpu-local signotify() if the IPI
1483 		 * preempts the desired thread.
1484 		 */
1485 		if (dtd->td_gd != mycpu) {
1486 			LWPHOLD(lp);
1487 			lwkt_send_ipiq(dtd->td_gd, lwp_signotify_remote, lp);
1488 		}
1489 	}
1490 	crit_exit();
1491 }
1492 
1493 /*
1494  * This function is called via an IPI so we cannot call setrunnable() here
1495  * (because while we hold the lp we don't own its token, and can't get it
1496  * from an IPI).
1497  *
1498  * We are interlocked by virtue of being on the same cpu as the target.  If
1499  * we still are and LWP_SINTR or TDF_SINTR is set we can safely schedule
1500  * the target thread.
1501  */
1502 static void
1503 lwp_signotify_remote(void *arg)
1504 {
1505 	struct lwp *lp = arg;
1506 	thread_t td = lp->lwp_thread;
1507 
1508 	if (lp == lwkt_preempted_proc()) {
1509 		signotify();
1510 		LWPRELE(lp);
1511 	} else if (td->td_gd == mycpu) {
1512 		if ((lp->lwp_flags & LWP_SINTR) ||
1513 		    (td->td_flags & TDF_SINTR)) {
1514 			lwkt_schedule(td);
1515 		}
1516 		LWPRELE(lp);
1517 	} else {
1518 		lwkt_send_ipiq(td->td_gd, lwp_signotify_remote, lp);
1519 		/* LWPHOLD() is forwarded to the target cpu */
1520 	}
1521 }
1522 
1523 /*
1524  * Caller must hold p->p_token
1525  */
1526 void
1527 proc_stop(struct proc *p, int sig)
1528 {
1529 	struct proc *q;
1530 	struct lwp *lp;
1531 
1532 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1533 
1534 	/*
1535 	 * If somebody raced us, be happy with it.  SCORE overrides SSTOP.
1536 	 */
1537 	if (sig == SCORE) {
1538 		if (p->p_stat == SCORE || p->p_stat == SZOMB)
1539 			return;
1540 	} else {
1541 		if (p->p_stat == SSTOP || p->p_stat == SCORE ||
1542 		    p->p_stat == SZOMB) {
1543 			return;
1544 		}
1545 	}
1546 	p->p_stat = sig;
1547 
1548 	FOREACH_LWP_IN_PROC(lp, p) {
1549 		LWPHOLD(lp);
1550 		lwkt_gettoken(&lp->lwp_token);
1551 
1552 		switch (lp->lwp_stat) {
1553 		case LSSTOP:
1554 			/*
1555 			 * Do nothing, we are already counted in
1556 			 * p_nstopped.
1557 			 */
1558 			break;
1559 
1560 		case LSSLEEP:
1561 			/*
1562 			 * We're sleeping, but we will stop before
1563 			 * returning to userspace, so count us
1564 			 * as stopped as well.  We set LWP_MP_WSTOP
1565 			 * to signal the lwp that it should not
1566 			 * increase p_nstopped when reaching tstop().
1567 			 *
1568 			 * LWP_MP_WSTOP is protected by lp->lwp_token.
1569 			 */
1570 			if ((lp->lwp_mpflags & LWP_MP_WSTOP) == 0) {
1571 				atomic_set_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
1572 				++p->p_nstopped;
1573 			}
1574 			break;
1575 
1576 		case LSRUN:
1577 			/*
1578 			 * We might notify ourself, but that's not
1579 			 * a problem.
1580 			 */
1581 			lwp_signotify(lp);
1582 			break;
1583 		}
1584 		lwkt_reltoken(&lp->lwp_token);
1585 		LWPRELE(lp);
1586 	}
1587 
1588 	if (p->p_nstopped == p->p_nthreads) {
1589 		/*
1590 		 * Token required to interlock kern_wait().  Reparenting can
1591 		 * also cause a race so we have to hold (q).
1592 		 */
1593 		q = p->p_pptr;
1594 		PHOLD(q);
1595 		lwkt_gettoken(&q->p_token);
1596 		p->p_flags &= ~P_WAITED;
1597 		wakeup(q);
1598 		if ((q->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1599 			ksignal(p->p_pptr, SIGCHLD);
1600 		lwkt_reltoken(&q->p_token);
1601 		PRELE(q);
1602 	}
1603 }
1604 
1605 /*
1606  * Caller must hold p_token
1607  */
1608 void
1609 proc_unstop(struct proc *p, int sig)
1610 {
1611 	struct lwp *lp;
1612 
1613 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1614 
1615 	if (p->p_stat != sig)
1616 		return;
1617 
1618 	p->p_stat = SACTIVE;
1619 
1620 	FOREACH_LWP_IN_PROC(lp, p) {
1621 		LWPHOLD(lp);
1622 		lwkt_gettoken(&lp->lwp_token);
1623 
1624 		switch (lp->lwp_stat) {
1625 		case LSRUN:
1626 			/*
1627 			 * Uh?  Not stopped?  Well, I guess that's okay.
1628 			 */
1629 			if (bootverbose)
1630 				kprintf("proc_unstop: lwp %d/%d not sleeping\n",
1631 					p->p_pid, lp->lwp_tid);
1632 			break;
1633 
1634 		case LSSLEEP:
1635 			/*
1636 			 * Still sleeping.  Don't bother waking it up.
1637 			 * However, if this thread was counted as
1638 			 * stopped, undo this.
1639 			 *
1640 			 * Nevertheless we call setrunnable() so that it
1641 			 * will wake up in case a signal or timeout arrived
1642 			 * in the meantime.
1643 			 *
1644 			 * LWP_MP_WSTOP is protected by lp->lwp_token.
1645 			 */
1646 			if (lp->lwp_mpflags & LWP_MP_WSTOP) {
1647 				atomic_clear_int(&lp->lwp_mpflags,
1648 						 LWP_MP_WSTOP);
1649 				--p->p_nstopped;
1650 			} else {
1651 				if (bootverbose)
1652 					kprintf("proc_unstop: lwp %d/%d sleeping, not stopped\n",
1653 						p->p_pid, lp->lwp_tid);
1654 			}
1655 			/* FALLTHROUGH */
1656 
1657 		case LSSTOP:
1658 			/*
1659 			 * This handles any lwp's waiting in a tsleep with
1660 			 * SIGCATCH.
1661 			 */
1662 			lwp_signotify(lp);
1663 			break;
1664 
1665 		}
1666 		lwkt_reltoken(&lp->lwp_token);
1667 		LWPRELE(lp);
1668 	}
1669 
1670 	/*
1671 	 * This handles any lwp's waiting in tstop().  We have interlocked
1672 	 * the setting of p_stat by acquiring and releasing each lpw's
1673 	 * token.
1674 	 */
1675 	wakeup(p);
1676 }
1677 
1678 /*
1679  * Wait for all threads except the current thread to stop.
1680  */
1681 static void
1682 proc_stopwait(struct proc *p)
1683 {
1684 	while ((p->p_stat == SSTOP || p->p_stat == SCORE) &&
1685 	       p->p_nstopped < p->p_nthreads - 1) {
1686 		tsleep_interlock(&p->p_nstopped, 0);
1687 		if (p->p_nstopped < p->p_nthreads - 1) {
1688 			tsleep(&p->p_nstopped, PINTERLOCKED, "stopwt", hz);
1689 		}
1690 	}
1691 }
1692 
1693 /*
1694  * No requirements.
1695  */
1696 static int
1697 kern_sigtimedwait(sigset_t waitset, siginfo_t *info, struct timespec *timeout)
1698 {
1699 	sigset_t savedmask, set;
1700 	struct proc *p = curproc;
1701 	struct lwp *lp = curthread->td_lwp;
1702 	int error, sig, hz, timevalid = 0;
1703 	struct timespec rts, ets, ts;
1704 	struct timeval tv;
1705 
1706 	error = 0;
1707 	sig = 0;
1708 	ets.tv_sec = 0;		/* silence compiler warning */
1709 	ets.tv_nsec = 0;	/* silence compiler warning */
1710 	SIG_CANTMASK(waitset);
1711 	savedmask = lp->lwp_sigmask;
1712 
1713 	if (timeout) {
1714 		if (timeout->tv_sec >= 0 && timeout->tv_nsec >= 0 &&
1715 		    timeout->tv_nsec < 1000000000) {
1716 			timevalid = 1;
1717 			getnanouptime(&rts);
1718 		 	ets = rts;
1719 			timespecadd(&ets, timeout);
1720 		}
1721 	}
1722 
1723 	for (;;) {
1724 		set = lwp_sigpend(lp);
1725 		SIGSETAND(set, waitset);
1726 		if ((sig = sig_ffs(&set)) != 0) {
1727 			SIGFILLSET(lp->lwp_sigmask);
1728 			SIGDELSET(lp->lwp_sigmask, sig);
1729 			SIG_CANTMASK(lp->lwp_sigmask);
1730 			sig = issignal(lp, 1, 0);
1731 			/*
1732 			 * It may be a STOP signal, in the case, issignal
1733 			 * returns 0, because we may stop there, and new
1734 			 * signal can come in, we should restart if we got
1735 			 * nothing.
1736 			 */
1737 			if (sig == 0)
1738 				continue;
1739 			else
1740 				break;
1741 		}
1742 
1743 		/*
1744 		 * Previous checking got nothing, and we retried but still
1745 		 * got nothing, we should return the error status.
1746 		 */
1747 		if (error)
1748 			break;
1749 
1750 		/*
1751 		 * POSIX says this must be checked after looking for pending
1752 		 * signals.
1753 		 */
1754 		if (timeout) {
1755 			if (timevalid == 0) {
1756 				error = EINVAL;
1757 				break;
1758 			}
1759 			getnanouptime(&rts);
1760 			if (timespeccmp(&rts, &ets, >=)) {
1761 				error = EAGAIN;
1762 				break;
1763 			}
1764 			ts = ets;
1765 			timespecsub(&ts, &rts);
1766 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1767 			hz = tvtohz_high(&tv);
1768 		} else {
1769 			hz = 0;
1770 		}
1771 
1772 		lp->lwp_sigmask = savedmask;
1773 		SIGSETNAND(lp->lwp_sigmask, waitset);
1774 		/*
1775 		 * We won't ever be woken up.  Instead, our sleep will
1776 		 * be broken in lwpsignal().
1777 		 */
1778 		error = tsleep(&p->p_sigacts, PCATCH, "sigwt", hz);
1779 		if (timeout) {
1780 			if (error == ERESTART) {
1781 				/* can not restart a timeout wait. */
1782 				error = EINTR;
1783 			} else if (error == EAGAIN) {
1784 				/* will calculate timeout by ourself. */
1785 				error = 0;
1786 			}
1787 		}
1788 		/* Retry ... */
1789 	}
1790 
1791 	lp->lwp_sigmask = savedmask;
1792 	if (sig) {
1793 		error = 0;
1794 		bzero(info, sizeof(*info));
1795 		info->si_signo = sig;
1796 		spin_lock(&lp->lwp_spin);
1797 		lwp_delsig(lp, sig, 1);	/* take the signal! */
1798 		spin_unlock(&lp->lwp_spin);
1799 
1800 		if (sig == SIGKILL) {
1801 			sigexit(lp, sig);
1802 			/* NOT REACHED */
1803 		}
1804 	}
1805 
1806 	return (error);
1807 }
1808 
1809 /*
1810  * MPALMOSTSAFE
1811  */
1812 int
1813 sys_sigtimedwait(struct sigtimedwait_args *uap)
1814 {
1815 	struct timespec ts;
1816 	struct timespec *timeout;
1817 	sigset_t set;
1818 	siginfo_t info;
1819 	int error;
1820 
1821 	if (uap->timeout) {
1822 		error = copyin(uap->timeout, &ts, sizeof(ts));
1823 		if (error)
1824 			return (error);
1825 		timeout = &ts;
1826 	} else {
1827 		timeout = NULL;
1828 	}
1829 	error = copyin(uap->set, &set, sizeof(set));
1830 	if (error)
1831 		return (error);
1832 	error = kern_sigtimedwait(set, &info, timeout);
1833 	if (error)
1834 		return (error);
1835 	if (uap->info)
1836 		error = copyout(&info, uap->info, sizeof(info));
1837 	/* Repost if we got an error. */
1838 	/*
1839 	 * XXX lwp
1840 	 *
1841 	 * This could transform a thread-specific signal to another
1842 	 * thread / process pending signal.
1843 	 */
1844 	if (error) {
1845 		ksignal(curproc, info.si_signo);
1846 	} else {
1847 		uap->sysmsg_result = info.si_signo;
1848 	}
1849 	return (error);
1850 }
1851 
1852 /*
1853  * MPALMOSTSAFE
1854  */
1855 int
1856 sys_sigwaitinfo(struct sigwaitinfo_args *uap)
1857 {
1858 	siginfo_t info;
1859 	sigset_t set;
1860 	int error;
1861 
1862 	error = copyin(uap->set, &set, sizeof(set));
1863 	if (error)
1864 		return (error);
1865 	error = kern_sigtimedwait(set, &info, NULL);
1866 	if (error)
1867 		return (error);
1868 	if (uap->info)
1869 		error = copyout(&info, uap->info, sizeof(info));
1870 	/* Repost if we got an error. */
1871 	/*
1872 	 * XXX lwp
1873 	 *
1874 	 * This could transform a thread-specific signal to another
1875 	 * thread / process pending signal.
1876 	 */
1877 	if (error) {
1878 		ksignal(curproc, info.si_signo);
1879 	} else {
1880 		uap->sysmsg_result = info.si_signo;
1881 	}
1882 	return (error);
1883 }
1884 
1885 /*
1886  * If the current process has received a signal that would interrupt a
1887  * system call, return EINTR or ERESTART as appropriate.
1888  */
1889 int
1890 iscaught(struct lwp *lp)
1891 {
1892 	struct proc *p = lp->lwp_proc;
1893 	int sig;
1894 
1895 	if (p) {
1896 		if ((sig = CURSIG(lp)) != 0) {
1897 			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1898 				return (EINTR);
1899 			return (ERESTART);
1900 		}
1901 	}
1902 	return(EWOULDBLOCK);
1903 }
1904 
1905 /*
1906  * If the current lwp/proc has received a signal (should be caught or cause
1907  * termination, should interrupt current syscall), return the signal number.
1908  * Stop signals with default action are processed immediately, then cleared;
1909  * they aren't returned.  This is checked after each entry to the system for
1910  * a syscall or trap (though this can usually be done without calling issignal
1911  * by checking the pending signal masks in the CURSIG macro).
1912  *
1913  * This routine is called via CURSIG/__cursig.  We will acquire and release
1914  * p->p_token but if the caller needs to interlock the test the caller must
1915  * also hold p->p_token.
1916  *
1917  *	while (sig = CURSIG(curproc))
1918  *		postsig(sig);
1919  */
1920 int
1921 issignal(struct lwp *lp, int maytrace, int *ptokp)
1922 {
1923 	struct proc *p = lp->lwp_proc;
1924 	sigset_t mask;
1925 	int sig, prop;
1926 	int haveptok;
1927 
1928 	for (;;) {
1929 		int traced = (p->p_flags & P_TRACED) || (p->p_stops & S_SIG);
1930 
1931 		haveptok = 0;
1932 
1933 		/*
1934 		 * If this process is supposed to stop, stop this thread.
1935 		 */
1936 		if (STOPLWP(p, lp)) {
1937 			lwkt_gettoken(&p->p_token);
1938 			tstop();
1939 			lwkt_reltoken(&p->p_token);
1940 		}
1941 
1942 		/*
1943 		 * Quick check without token
1944 		 */
1945 		mask = lwp_sigpend(lp);
1946 		SIGSETNAND(mask, lp->lwp_sigmask);
1947 		if (p->p_flags & P_PPWAIT)
1948 			SIG_STOPSIGMASK(mask);
1949 		if (SIGISEMPTY(mask)) 		/* no signal to send */
1950 			return (0);
1951 
1952 		/*
1953 		 * If the signal is a member of the process signal set
1954 		 * we need p_token (even if it is also a member of the
1955 		 * lwp signal set).
1956 		 */
1957 		sig = sig_ffs(&mask);
1958 		if (SIGISMEMBER(p->p_siglist, sig)) {
1959 			/*
1960 			 * Recheck with token
1961 			 */
1962 			haveptok = 1;
1963 			lwkt_gettoken(&p->p_token);
1964 
1965 			mask = lwp_sigpend(lp);
1966 			SIGSETNAND(mask, lp->lwp_sigmask);
1967 			if (p->p_flags & P_PPWAIT)
1968 				SIG_STOPSIGMASK(mask);
1969 			if (SIGISEMPTY(mask)) {		/* no signal to send */
1970 				/* haveptok is TRUE */
1971 				lwkt_reltoken(&p->p_token);
1972 				return (0);
1973 			}
1974 			sig = sig_ffs(&mask);
1975 		}
1976 
1977 		STOPEVENT(p, S_SIG, sig);
1978 
1979 		/*
1980 		 * We should see pending but ignored signals
1981 		 * only if P_TRACED was on when they were posted.
1982 		 */
1983 		if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
1984 			spin_lock(&lp->lwp_spin);
1985 			lwp_delsig(lp, sig, haveptok);
1986 			spin_unlock(&lp->lwp_spin);
1987 			if (haveptok)
1988 				lwkt_reltoken(&p->p_token);
1989 			continue;
1990 		}
1991 		if (maytrace &&
1992 		    (p->p_flags & P_TRACED) &&
1993 		    (p->p_flags & P_PPWAIT) == 0) {
1994 			/*
1995 			 * If traced, always stop, and stay stopped until
1996 			 * released by the parent.
1997 			 *
1998 			 * NOTE: SSTOP may get cleared during the loop,
1999 			 * but we do not re-notify the parent if we have
2000 			 * to loop several times waiting for the parent
2001 			 * to let us continue.
2002 			 *
2003 			 * XXX not sure if this is still true
2004 			 */
2005 			if (haveptok == 0) {
2006 				lwkt_gettoken(&p->p_token);
2007 				haveptok = 1;
2008 			}
2009 			p->p_xstat = sig;
2010 			proc_stop(p, SSTOP);
2011 			do {
2012 				tstop();
2013 			} while (!trace_req(p) && (p->p_flags & P_TRACED));
2014 
2015 			/*
2016 			 * If parent wants us to take the signal,
2017 			 * then it will leave it in p->p_xstat;
2018 			 * otherwise we just look for signals again.
2019 			 */
2020 			spin_lock(&lp->lwp_spin);
2021 			lwp_delsig(lp, sig, 1);	/* clear old signal */
2022 			spin_unlock(&lp->lwp_spin);
2023 			sig = p->p_xstat;
2024 			if (sig == 0) {
2025 				/* haveptok is TRUE */
2026 				lwkt_reltoken(&p->p_token);
2027 				continue;
2028 			}
2029 
2030 			/*
2031 			 * Put the new signal into p_siglist.  If the
2032 			 * signal is being masked, look for other signals.
2033 			 *
2034 			 * XXX lwp might need a call to ksignal()
2035 			 */
2036 			SIGADDSET_ATOMIC(p->p_siglist, sig);
2037 			if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
2038 				/* haveptok is TRUE */
2039 				lwkt_reltoken(&p->p_token);
2040 				continue;
2041 			}
2042 
2043 			/*
2044 			 * If the traced bit got turned off, go back up
2045 			 * to the top to rescan signals.  This ensures
2046 			 * that p_sig* and ps_sigact are consistent.
2047 			 */
2048 			if ((p->p_flags & P_TRACED) == 0) {
2049 				/* haveptok is TRUE */
2050 				lwkt_reltoken(&p->p_token);
2051 				continue;
2052 			}
2053 		}
2054 
2055 		/*
2056 		 * p_token may be held here
2057 		 */
2058 		prop = sigprop(sig);
2059 
2060 		/*
2061 		 * Decide whether the signal should be returned.
2062 		 * Return the signal's number, or fall through
2063 		 * to clear it from the pending mask.
2064 		 */
2065 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
2066 		case (intptr_t)SIG_DFL:
2067 			/*
2068 			 * Don't take default actions on system processes.
2069 			 */
2070 			if (p->p_pid <= 1) {
2071 #ifdef DIAGNOSTIC
2072 				/*
2073 				 * Are you sure you want to ignore SIGSEGV
2074 				 * in init? XXX
2075 				 */
2076 				kprintf("Process (pid %lu) got signal %d\n",
2077 					(u_long)p->p_pid, sig);
2078 #endif
2079 				break;		/* == ignore */
2080 			}
2081 
2082 			/*
2083 			 * Handle the in-kernel checkpoint action
2084 			 */
2085 			if (prop & SA_CKPT) {
2086 				if (haveptok == 0) {
2087 					lwkt_gettoken(&p->p_token);
2088 					haveptok = 1;
2089 				}
2090 				checkpoint_signal_handler(lp);
2091 				break;
2092 			}
2093 
2094 			/*
2095 			 * If there is a pending stop signal to process
2096 			 * with default action, stop here,
2097 			 * then clear the signal.  However,
2098 			 * if process is member of an orphaned
2099 			 * process group, ignore tty stop signals.
2100 			 */
2101 			if (prop & SA_STOP) {
2102 				if (haveptok == 0) {
2103 					lwkt_gettoken(&p->p_token);
2104 					haveptok = 1;
2105 				}
2106 				if (p->p_flags & P_TRACED ||
2107 		    		    (p->p_pgrp->pg_jobc == 0 &&
2108 				    prop & SA_TTYSTOP))
2109 					break;	/* == ignore */
2110 				if ((p->p_flags & P_WEXIT) == 0) {
2111 					p->p_xstat = sig;
2112 					proc_stop(p, SSTOP);
2113 					tstop();
2114 				}
2115 				break;
2116 			} else if (prop & SA_IGNORE) {
2117 				/*
2118 				 * Except for SIGCONT, shouldn't get here.
2119 				 * Default action is to ignore; drop it.
2120 				 */
2121 				break;		/* == ignore */
2122 			} else {
2123 				if (ptokp)
2124 					*ptokp = haveptok;
2125 				else if (haveptok)
2126 					lwkt_reltoken(&p->p_token);
2127 				return (sig);
2128 			}
2129 
2130 			/*NOTREACHED*/
2131 
2132 		case (intptr_t)SIG_IGN:
2133 			/*
2134 			 * Masking above should prevent us ever trying
2135 			 * to take action on an ignored signal other
2136 			 * than SIGCONT, unless process is traced.
2137 			 */
2138 			if ((prop & SA_CONT) == 0 &&
2139 			    (p->p_flags & P_TRACED) == 0)
2140 				kprintf("issignal\n");
2141 			break;		/* == ignore */
2142 
2143 		default:
2144 			/*
2145 			 * This signal has an action, let
2146 			 * postsig() process it.
2147 			 */
2148 			if (ptokp)
2149 				*ptokp = haveptok;
2150 			else if (haveptok)
2151 				lwkt_reltoken(&p->p_token);
2152 			return (sig);
2153 		}
2154 		spin_lock(&lp->lwp_spin);
2155 		lwp_delsig(lp, sig, haveptok);		/* take the signal! */
2156 		spin_unlock(&lp->lwp_spin);
2157 
2158 		if (haveptok)
2159 			lwkt_reltoken(&p->p_token);
2160 	}
2161 	/* NOTREACHED */
2162 }
2163 
2164 /*
2165  * Take the action for the specified signal from the current set of
2166  * pending signals.
2167  *
2168  * haveptok indicates whether the caller is holding p->p_token.  If the
2169  * caller is, we are responsible for releasing it.
2170  *
2171  * This routine can only be called from the top-level trap from usermode.
2172  * It is expecting to be able to modify the top-level stack frame.
2173  */
2174 void
2175 postsig(int sig, int haveptok)
2176 {
2177 	struct lwp *lp = curthread->td_lwp;
2178 	struct proc *p = lp->lwp_proc;
2179 	struct sigacts *ps = p->p_sigacts;
2180 	sig_t action;
2181 	sigset_t returnmask;
2182 	int code;
2183 
2184 	KASSERT(sig != 0, ("postsig"));
2185 
2186 	/*
2187 	 * If we are a virtual kernel running an emulated user process
2188 	 * context, switch back to the virtual kernel context before
2189 	 * trying to post the signal.
2190 	 */
2191 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
2192 		struct trapframe *tf = lp->lwp_md.md_regs;
2193 		tf->tf_trapno = 0;
2194 		vkernel_trap(lp, tf);
2195 	}
2196 
2197 	KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
2198 
2199 	spin_lock(&lp->lwp_spin);
2200 	lwp_delsig(lp, sig, haveptok);
2201 	spin_unlock(&lp->lwp_spin);
2202 	action = ps->ps_sigact[_SIG_IDX(sig)];
2203 #ifdef KTRACE
2204 	if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
2205 		ktrpsig(lp, sig, action, lp->lwp_flags & LWP_OLDMASK ?
2206 			&lp->lwp_oldsigmask : &lp->lwp_sigmask, 0);
2207 #endif
2208 	/*
2209 	 * We don't need p_token after this point.
2210 	 */
2211 	if (haveptok)
2212 		lwkt_reltoken(&p->p_token);
2213 
2214 	STOPEVENT(p, S_SIG, sig);
2215 
2216 	if (action == SIG_DFL) {
2217 		/*
2218 		 * Default action, where the default is to kill
2219 		 * the process.  (Other cases were ignored above.)
2220 		 */
2221 		sigexit(lp, sig);
2222 		/* NOTREACHED */
2223 	} else {
2224 		/*
2225 		 * If we get here, the signal must be caught.
2226 		 */
2227 		KASSERT(action != SIG_IGN && !SIGISMEMBER(lp->lwp_sigmask, sig),
2228 		    ("postsig action"));
2229 
2230 		/*
2231 		 * Reset the signal handler if asked to
2232 		 */
2233 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
2234 			/*
2235 			 * See kern_sigaction() for origin of this code.
2236 			 */
2237 			SIGDELSET(p->p_sigcatch, sig);
2238 			if (sig != SIGCONT &&
2239 			    sigprop(sig) & SA_IGNORE)
2240 				SIGADDSET(p->p_sigignore, sig);
2241 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2242 		}
2243 
2244 		/*
2245 		 * Set the signal mask and calculate the mask to restore
2246 		 * when the signal function returns.
2247 		 *
2248 		 * Special case: user has done a sigsuspend.  Here the
2249 		 * current mask is not of interest, but rather the
2250 		 * mask from before the sigsuspend is what we want
2251 		 * restored after the signal processing is completed.
2252 		 */
2253 		if (lp->lwp_flags & LWP_OLDMASK) {
2254 			returnmask = lp->lwp_oldsigmask;
2255 			lp->lwp_flags &= ~LWP_OLDMASK;
2256 		} else {
2257 			returnmask = lp->lwp_sigmask;
2258 		}
2259 
2260 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
2261 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
2262 			SIGADDSET(lp->lwp_sigmask, sig);
2263 
2264 		lp->lwp_ru.ru_nsignals++;
2265 		if (lp->lwp_sig != sig) {
2266 			code = 0;
2267 		} else {
2268 			code = lp->lwp_code;
2269 			lp->lwp_code = 0;
2270 			lp->lwp_sig = 0;
2271 		}
2272 		(*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
2273 	}
2274 }
2275 
2276 /*
2277  * Kill the current process for stated reason.
2278  */
2279 void
2280 killproc(struct proc *p, char *why)
2281 {
2282 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n",
2283 		p->p_pid, p->p_comm,
2284 		p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2285 	ksignal(p, SIGKILL);
2286 }
2287 
2288 /*
2289  * Force the current process to exit with the specified signal, dumping core
2290  * if appropriate.  We bypass the normal tests for masked and caught signals,
2291  * allowing unrecoverable failures to terminate the process without changing
2292  * signal state.  Mark the accounting record with the signal termination.
2293  * If dumping core, save the signal number for the debugger.  Calls exit and
2294  * does not return.
2295  *
2296  * This routine does not return.
2297  */
2298 void
2299 sigexit(struct lwp *lp, int sig)
2300 {
2301 	struct proc *p = lp->lwp_proc;
2302 
2303 	lwkt_gettoken(&p->p_token);
2304 	p->p_acflag |= AXSIG;
2305 	if (sigprop(sig) & SA_CORE) {
2306 		lp->lwp_sig = sig;
2307 
2308 		/*
2309 		 * All threads must be stopped before we can safely coredump.
2310 		 * Stop threads using SCORE, which cannot be overridden.
2311 		 */
2312 		if (p->p_stat != SCORE) {
2313 			proc_stop(p, SCORE);
2314 			proc_stopwait(p);
2315 
2316 			if (coredump(lp, sig) == 0)
2317 				sig |= WCOREFLAG;
2318 			p->p_stat = SSTOP;
2319 		}
2320 
2321 		/*
2322 		 * Log signals which would cause core dumps
2323 		 * (Log as LOG_INFO to appease those who don't want
2324 		 * these messages.)
2325 		 * XXX : Todo, as well as euid, write out ruid too
2326 		 */
2327 		if (kern_logsigexit) {
2328 			log(LOG_INFO,
2329 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2330 			    p->p_pid, p->p_comm,
2331 			    p->p_ucred ? p->p_ucred->cr_uid : -1,
2332 			    sig &~ WCOREFLAG,
2333 			    sig & WCOREFLAG ? " (core dumped)" : "");
2334 		}
2335 	}
2336 	lwkt_reltoken(&p->p_token);
2337 	exit1(W_EXITCODE(0, sig));
2338 	/* NOTREACHED */
2339 }
2340 
2341 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
2342 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
2343 	      sizeof(corefilename), "process corefile name format string");
2344 
2345 /*
2346  * expand_name(name, uid, pid)
2347  * Expand the name described in corefilename, using name, uid, and pid.
2348  * corefilename is a kprintf-like string, with three format specifiers:
2349  *	%N	name of process ("name")
2350  *	%P	process id (pid)
2351  *	%U	user id (uid)
2352  * For example, "%N.core" is the default; they can be disabled completely
2353  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
2354  * This is controlled by the sysctl variable kern.corefile (see above).
2355  */
2356 
2357 static char *
2358 expand_name(const char *name, uid_t uid, pid_t pid)
2359 {
2360 	char *temp;
2361 	char buf[11];		/* Buffer for pid/uid -- max 4B */
2362 	int i, n;
2363 	char *format = corefilename;
2364 	size_t namelen;
2365 
2366 	temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
2367 	if (temp == NULL)
2368 		return NULL;
2369 	namelen = strlen(name);
2370 	for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
2371 		int l;
2372 		switch (format[i]) {
2373 		case '%':	/* Format character */
2374 			i++;
2375 			switch (format[i]) {
2376 			case '%':
2377 				temp[n++] = '%';
2378 				break;
2379 			case 'N':	/* process name */
2380 				if ((n + namelen) > MAXPATHLEN) {
2381 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2382 					    pid, name, uid, temp, name);
2383 					kfree(temp, M_TEMP);
2384 					return NULL;
2385 				}
2386 				memcpy(temp+n, name, namelen);
2387 				n += namelen;
2388 				break;
2389 			case 'P':	/* process id */
2390 				l = ksprintf(buf, "%u", pid);
2391 				if ((n + l) > MAXPATHLEN) {
2392 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2393 					    pid, name, uid, temp, name);
2394 					kfree(temp, M_TEMP);
2395 					return NULL;
2396 				}
2397 				memcpy(temp+n, buf, l);
2398 				n += l;
2399 				break;
2400 			case 'U':	/* user id */
2401 				l = ksprintf(buf, "%u", uid);
2402 				if ((n + l) > MAXPATHLEN) {
2403 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2404 					    pid, name, uid, temp, name);
2405 					kfree(temp, M_TEMP);
2406 					return NULL;
2407 				}
2408 				memcpy(temp+n, buf, l);
2409 				n += l;
2410 				break;
2411 			default:
2412 			  	log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
2413 			}
2414 			break;
2415 		default:
2416 			temp[n++] = format[i];
2417 		}
2418 	}
2419 	temp[n] = '\0';
2420 	return temp;
2421 }
2422 
2423 /*
2424  * Dump a process' core.  The main routine does some
2425  * policy checking, and creates the name of the coredump;
2426  * then it passes on a vnode and a size limit to the process-specific
2427  * coredump routine if there is one; if there _is not_ one, it returns
2428  * ENOSYS; otherwise it returns the error from the process-specific routine.
2429  *
2430  * The parameter `lp' is the lwp which triggered the coredump.
2431  */
2432 
2433 static int
2434 coredump(struct lwp *lp, int sig)
2435 {
2436 	struct proc *p = lp->lwp_proc;
2437 	struct vnode *vp;
2438 	struct ucred *cred = p->p_ucred;
2439 	struct flock lf;
2440 	struct nlookupdata nd;
2441 	struct vattr vattr;
2442 	int error, error1;
2443 	char *name;			/* name of corefile */
2444 	off_t limit;
2445 
2446 	STOPEVENT(p, S_CORE, 0);
2447 
2448 	if (((sugid_coredump == 0) && p->p_flags & P_SUGID) || do_coredump == 0)
2449 		return (EFAULT);
2450 
2451 	/*
2452 	 * Note that the bulk of limit checking is done after
2453 	 * the corefile is created.  The exception is if the limit
2454 	 * for corefiles is 0, in which case we don't bother
2455 	 * creating the corefile at all.  This layout means that
2456 	 * a corefile is truncated instead of not being created,
2457 	 * if it is larger than the limit.
2458 	 */
2459 	limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
2460 	if (limit == 0)
2461 		return EFBIG;
2462 
2463 	name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
2464 	if (name == NULL)
2465 		return (EINVAL);
2466 	error = nlookup_init(&nd, name, UIO_SYSSPACE, NLC_LOCKVP);
2467 	if (error == 0)
2468 		error = vn_open(&nd, NULL,
2469 				O_CREAT | FWRITE | O_NOFOLLOW,
2470 				S_IRUSR | S_IWUSR);
2471 	kfree(name, M_TEMP);
2472 	if (error) {
2473 		nlookup_done(&nd);
2474 		return (error);
2475 	}
2476 	vp = nd.nl_open_vp;
2477 	nd.nl_open_vp = NULL;
2478 	nlookup_done(&nd);
2479 
2480 	vn_unlock(vp);
2481 	lf.l_whence = SEEK_SET;
2482 	lf.l_start = 0;
2483 	lf.l_len = 0;
2484 	lf.l_type = F_WRLCK;
2485 	error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, 0);
2486 	if (error)
2487 		goto out2;
2488 
2489 	/* Don't dump to non-regular files or files with links. */
2490 	if (vp->v_type != VREG ||
2491 	    VOP_GETATTR(vp, &vattr) || vattr.va_nlink != 1) {
2492 		error = EFAULT;
2493 		goto out1;
2494 	}
2495 
2496 	/* Don't dump to files current user does not own */
2497 	if (vattr.va_uid != p->p_ucred->cr_uid) {
2498 		error = EFAULT;
2499 		goto out1;
2500 	}
2501 
2502 	VATTR_NULL(&vattr);
2503 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2504 	vattr.va_size = 0;
2505 	VOP_SETATTR(vp, &vattr, cred);
2506 	p->p_acflag |= ACORE;
2507 	vn_unlock(vp);
2508 
2509 	error = p->p_sysent->sv_coredump ?
2510 		  p->p_sysent->sv_coredump(lp, sig, vp, limit) : ENOSYS;
2511 
2512 out1:
2513 	lf.l_type = F_UNLCK;
2514 	VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, 0);
2515 out2:
2516 	error1 = vn_close(vp, FWRITE, NULL);
2517 	if (error == 0)
2518 		error = error1;
2519 	return (error);
2520 }
2521 
2522 /*
2523  * Nonexistent system call-- signal process (may want to handle it).
2524  * Flag error in case process won't see signal immediately (blocked or ignored).
2525  *
2526  * MPALMOSTSAFE
2527  */
2528 /* ARGSUSED */
2529 int
2530 sys_nosys(struct nosys_args *args)
2531 {
2532 	lwpsignal(curproc, curthread->td_lwp, SIGSYS);
2533 	return (EINVAL);
2534 }
2535 
2536 /*
2537  * Send a SIGIO or SIGURG signal to a process or process group using
2538  * stored credentials rather than those of the current process.
2539  */
2540 void
2541 pgsigio(struct sigio *sigio, int sig, int checkctty)
2542 {
2543 	if (sigio == NULL)
2544 		return;
2545 
2546 	if (sigio->sio_pgid > 0) {
2547 		if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
2548 		             sigio->sio_proc))
2549 			ksignal(sigio->sio_proc, sig);
2550 	} else if (sigio->sio_pgid < 0) {
2551 		struct proc *p;
2552 		struct pgrp *pg = sigio->sio_pgrp;
2553 
2554 		/*
2555 		 * Must interlock all signals against fork
2556 		 */
2557 		pgref(pg);
2558 		lockmgr(&pg->pg_lock, LK_EXCLUSIVE);
2559 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
2560 			if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
2561 			    (checkctty == 0 || (p->p_flags & P_CONTROLT)))
2562 				ksignal(p, sig);
2563 		}
2564 		lockmgr(&pg->pg_lock, LK_RELEASE);
2565 		pgrel(pg);
2566 	}
2567 }
2568 
2569 static int
2570 filt_sigattach(struct knote *kn)
2571 {
2572 	struct proc *p = curproc;
2573 
2574 	kn->kn_ptr.p_proc = p;
2575 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
2576 
2577 	/* XXX lock the proc here while adding to the list? */
2578 	knote_insert(&p->p_klist, kn);
2579 
2580 	return (0);
2581 }
2582 
2583 static void
2584 filt_sigdetach(struct knote *kn)
2585 {
2586 	struct proc *p = kn->kn_ptr.p_proc;
2587 
2588 	knote_remove(&p->p_klist, kn);
2589 }
2590 
2591 /*
2592  * signal knotes are shared with proc knotes, so we apply a mask to
2593  * the hint in order to differentiate them from process hints.  This
2594  * could be avoided by using a signal-specific knote list, but probably
2595  * isn't worth the trouble.
2596  */
2597 static int
2598 filt_signal(struct knote *kn, long hint)
2599 {
2600 	if (hint & NOTE_SIGNAL) {
2601 		hint &= ~NOTE_SIGNAL;
2602 
2603 		if (kn->kn_id == hint)
2604 			kn->kn_data++;
2605 	}
2606 	return (kn->kn_data != 0);
2607 }
2608