xref: /dragonfly/sys/kern/kern_sig.c (revision 279dd846)
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 	{ 0, 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(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(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,
643 	    uap->oss ? &oss : NULL);
644 
645 	if (error == 0 && uap->oss)
646 		error = copyout(&oss, uap->oss, sizeof(*uap->oss));
647 	return (error);
648 }
649 
650 /*
651  * Common code for kill process group/broadcast kill.
652  * cp is calling process.
653  */
654 struct killpg_info {
655 	int nfound;
656 	int sig;
657 };
658 
659 static int killpg_all_callback(struct proc *p, void *data);
660 
661 static int
662 dokillpg(int sig, int pgid, int all)
663 {
664 	struct killpg_info info;
665 	struct proc *cp = curproc;
666 	struct proc *p;
667 	struct pgrp *pgrp;
668 
669 	info.nfound = 0;
670 	info.sig = sig;
671 
672 	if (all) {
673 		/*
674 		 * broadcast
675 		 */
676 		allproc_scan(killpg_all_callback, &info);
677 	} else {
678 		if (pgid == 0) {
679 			/*
680 			 * zero pgid means send to my process group.
681 			 */
682 			pgrp = cp->p_pgrp;
683 			pgref(pgrp);
684 		} else {
685 			pgrp = pgfind(pgid);
686 			if (pgrp == NULL)
687 				return (ESRCH);
688 		}
689 
690 		/*
691 		 * Must interlock all signals against fork
692 		 */
693 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
694 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
695 			if (p->p_pid <= 1 ||
696 			    p->p_stat == SZOMB ||
697 			    (p->p_flags & P_SYSTEM) ||
698 			    !CANSIGNAL(p, sig)) {
699 				continue;
700 			}
701 			++info.nfound;
702 			if (sig)
703 				ksignal(p, sig);
704 		}
705 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
706 		pgrel(pgrp);
707 	}
708 	return (info.nfound ? 0 : ESRCH);
709 }
710 
711 static int
712 killpg_all_callback(struct proc *p, void *data)
713 {
714 	struct killpg_info *info = data;
715 
716 	if (p->p_pid <= 1 || (p->p_flags & P_SYSTEM) ||
717 	    p == curproc || !CANSIGNAL(p, info->sig)) {
718 		return (0);
719 	}
720 	++info->nfound;
721 	if (info->sig)
722 		ksignal(p, info->sig);
723 	return(0);
724 }
725 
726 /*
727  * Send a general signal to a process or LWPs within that process.
728  *
729  * Note that new signals cannot be sent if a process is exiting or already
730  * a zombie, but we return success anyway as userland is likely to not handle
731  * the race properly.
732  *
733  * No requirements.
734  */
735 int
736 kern_kill(int sig, pid_t pid, lwpid_t tid)
737 {
738 	int t;
739 
740 	if ((u_int)sig > _SIG_MAXSIG)
741 		return (EINVAL);
742 
743 	if (pid > 0) {
744 		struct proc *p;
745 		struct lwp *lp = NULL;
746 
747 		/*
748 		 * Send a signal to a single process.  If the kill() is
749 		 * racing an exiting process which has not yet been reaped
750 		 * act as though the signal was delivered successfully but
751 		 * don't actually try to deliver the signal.
752 		 */
753 		if ((p = pfind(pid)) == NULL) {
754 			if ((p = zpfind(pid)) == NULL)
755 				return (ESRCH);
756 			PRELE(p);
757 			return (0);
758 		}
759 		lwkt_gettoken(&p->p_token);
760 		if (!CANSIGNAL(p, sig)) {
761 			lwkt_reltoken(&p->p_token);
762 			PRELE(p);
763 			return (EPERM);
764 		}
765 
766 		/*
767 		 * NOP if the process is exiting.  Note that lwpsignal() is
768 		 * called directly with P_WEXIT set to kill individual LWPs
769 		 * during exit, which is allowed.
770 		 */
771 		if (p->p_flags & P_WEXIT) {
772 			lwkt_reltoken(&p->p_token);
773 			PRELE(p);
774 			return (0);
775 		}
776 		if (tid != -1) {
777 			lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, tid);
778 			if (lp == NULL) {
779 				lwkt_reltoken(&p->p_token);
780 				PRELE(p);
781 				return (ESRCH);
782 			}
783 		}
784 		if (sig)
785 			lwpsignal(p, lp, sig);
786 		lwkt_reltoken(&p->p_token);
787 		PRELE(p);
788 
789 		return (0);
790 	}
791 
792 	/*
793 	 * If we come here, pid is a special broadcast pid.
794 	 * This doesn't mix with a tid.
795 	 */
796 	if (tid != -1)
797 		return (EINVAL);
798 
799 	switch (pid) {
800 	case -1:		/* broadcast signal */
801 		t = (dokillpg(sig, 0, 1));
802 		break;
803 	case 0:			/* signal own process group */
804 		t = (dokillpg(sig, 0, 0));
805 		break;
806 	default:		/* negative explicit process group */
807 		t = (dokillpg(sig, -pid, 0));
808 		break;
809 	}
810 	return t;
811 }
812 
813 int
814 sys_kill(struct kill_args *uap)
815 {
816 	int error;
817 
818 	error = kern_kill(uap->signum, uap->pid, -1);
819 	return (error);
820 }
821 
822 int
823 sys_lwp_kill(struct lwp_kill_args *uap)
824 {
825 	int error;
826 	pid_t pid = uap->pid;
827 
828 	/*
829 	 * A tid is mandatory for lwp_kill(), otherwise
830 	 * you could simply use kill().
831 	 */
832 	if (uap->tid == -1)
833 		return (EINVAL);
834 
835 	/*
836 	 * To save on a getpid() function call for intra-process
837 	 * signals, pid == -1 means current process.
838 	 */
839 	if (pid == -1)
840 		pid = curproc->p_pid;
841 
842 	error = kern_kill(uap->signum, pid, uap->tid);
843 	return (error);
844 }
845 
846 /*
847  * Send a signal to a process group.
848  */
849 void
850 gsignal(int pgid, int sig)
851 {
852 	struct pgrp *pgrp;
853 
854 	if (pgid && (pgrp = pgfind(pgid)))
855 		pgsignal(pgrp, sig, 0);
856 }
857 
858 /*
859  * Send a signal to a process group.  If checktty is 1,
860  * limit to members which have a controlling terminal.
861  *
862  * pg_lock interlocks against a fork that might be in progress, to
863  * ensure that the new child process picks up the signal.
864  */
865 void
866 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
867 {
868 	struct proc *p;
869 
870 	/*
871 	 * Must interlock all signals against fork
872 	 */
873 	if (pgrp) {
874 		pgref(pgrp);
875 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
876 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
877 			if (checkctty == 0 || p->p_flags & P_CONTROLT)
878 				ksignal(p, sig);
879 		}
880 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
881 		pgrel(pgrp);
882 	}
883 }
884 
885 /*
886  * Send a signal caused by a trap to the current lwp.  If it will be caught
887  * immediately, deliver it with correct code.  Otherwise, post it normally.
888  *
889  * These signals may ONLY be delivered to the specified lwp and may never
890  * be delivered to the process generically.
891  */
892 void
893 trapsignal(struct lwp *lp, int sig, u_long code)
894 {
895 	struct proc *p = lp->lwp_proc;
896 	struct sigacts *ps = p->p_sigacts;
897 
898 	/*
899 	 * If we are a virtual kernel running an emulated user process
900 	 * context, switch back to the virtual kernel context before
901 	 * trying to post the signal.
902 	 */
903 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
904 		struct trapframe *tf = lp->lwp_md.md_regs;
905 		tf->tf_trapno = 0;
906 		vkernel_trap(lp, tf);
907 	}
908 
909 
910 	if ((p->p_flags & P_TRACED) == 0 && SIGISMEMBER(p->p_sigcatch, sig) &&
911 	    !SIGISMEMBER(lp->lwp_sigmask, sig)) {
912 		lp->lwp_ru.ru_nsignals++;
913 #ifdef KTRACE
914 		if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
915 			ktrpsig(lp, sig, ps->ps_sigact[_SIG_IDX(sig)],
916 				&lp->lwp_sigmask, code);
917 #endif
918 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], sig,
919 						&lp->lwp_sigmask, code);
920 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
921 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
922 			SIGADDSET(lp->lwp_sigmask, sig);
923 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
924 			/*
925 			 * See kern_sigaction() for origin of this code.
926 			 */
927 			SIGDELSET(p->p_sigcatch, sig);
928 			if (sig != SIGCONT &&
929 			    sigprop(sig) & SA_IGNORE)
930 				SIGADDSET(p->p_sigignore, sig);
931 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
932 		}
933 	} else {
934 		lp->lwp_code = code;	/* XXX for core dump/debugger */
935 		lp->lwp_sig = sig;	/* XXX to verify code */
936 		lwpsignal(p, lp, sig);
937 	}
938 }
939 
940 /*
941  * Find a suitable lwp to deliver the signal to.  Returns NULL if all
942  * lwps hold the signal blocked.
943  *
944  * Caller must hold p->p_token.
945  *
946  * Returns a lp or NULL.  If non-NULL the lp is held and its token is
947  * acquired.
948  */
949 static struct lwp *
950 find_lwp_for_signal(struct proc *p, int sig)
951 {
952 	struct lwp *lp;
953 	struct lwp *run, *sleep, *stop;
954 
955 	/*
956 	 * If the running/preempted thread belongs to the proc to which
957 	 * the signal is being delivered and this thread does not block
958 	 * the signal, then we can avoid a context switch by delivering
959 	 * the signal to this thread, because it will return to userland
960 	 * soon anyways.
961 	 */
962 	lp = lwkt_preempted_proc();
963 	if (lp != NULL && lp->lwp_proc == p) {
964 		LWPHOLD(lp);
965 		lwkt_gettoken(&lp->lwp_token);
966 		if (!SIGISMEMBER(lp->lwp_sigmask, sig)) {
967 			/* return w/ token held */
968 			return (lp);
969 		}
970 		lwkt_reltoken(&lp->lwp_token);
971 		LWPRELE(lp);
972 	}
973 
974 	run = sleep = stop = NULL;
975 	FOREACH_LWP_IN_PROC(lp, p) {
976 		/*
977 		 * If the signal is being blocked by the lwp, then this
978 		 * lwp is not eligible for receiving the signal.
979 		 */
980 		LWPHOLD(lp);
981 		lwkt_gettoken(&lp->lwp_token);
982 
983 		if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
984 			lwkt_reltoken(&lp->lwp_token);
985 			LWPRELE(lp);
986 			continue;
987 		}
988 
989 		switch (lp->lwp_stat) {
990 		case LSRUN:
991 			if (sleep) {
992 				lwkt_token_swap();
993 				lwkt_reltoken(&sleep->lwp_token);
994 				LWPRELE(sleep);
995 				sleep = NULL;
996 				run = lp;
997 			} else if (stop) {
998 				lwkt_token_swap();
999 				lwkt_reltoken(&stop->lwp_token);
1000 				LWPRELE(stop);
1001 				stop = NULL;
1002 				run = lp;
1003 			} else {
1004 				run = lp;
1005 			}
1006 			break;
1007 		case LSSLEEP:
1008 			if (lp->lwp_flags & LWP_SINTR) {
1009 				if (sleep) {
1010 					lwkt_reltoken(&lp->lwp_token);
1011 					LWPRELE(lp);
1012 				} else if (stop) {
1013 					lwkt_token_swap();
1014 					lwkt_reltoken(&stop->lwp_token);
1015 					LWPRELE(stop);
1016 					stop = NULL;
1017 					sleep = lp;
1018 				} else {
1019 					sleep = lp;
1020 				}
1021 			} else {
1022 				lwkt_reltoken(&lp->lwp_token);
1023 				LWPRELE(lp);
1024 			}
1025 			break;
1026 		case LSSTOP:
1027 			if (sleep) {
1028 				lwkt_reltoken(&lp->lwp_token);
1029 				LWPRELE(lp);
1030 			} else if (stop) {
1031 				lwkt_reltoken(&lp->lwp_token);
1032 				LWPRELE(lp);
1033 			} else {
1034 				stop = lp;
1035 			}
1036 			break;
1037 		}
1038 		if (run)
1039 			break;
1040 	}
1041 
1042 	if (run != NULL)
1043 		return (run);
1044 	else if (sleep != NULL)
1045 		return (sleep);
1046 	else
1047 		return (stop);
1048 }
1049 
1050 /*
1051  * Send the signal to the process.  If the signal has an action, the action
1052  * is usually performed by the target process rather than the caller; we add
1053  * the signal to the set of pending signals for the process.
1054  *
1055  * Exceptions:
1056  *   o When a stop signal is sent to a sleeping process that takes the
1057  *     default action, the process is stopped without awakening it.
1058  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
1059  *     regardless of the signal action (eg, blocked or ignored).
1060  *
1061  * Other ignored signals are discarded immediately.
1062  *
1063  * If the caller wishes to call this function from a hard code section the
1064  * caller must already hold p->p_token (see kern_clock.c).
1065  *
1066  * No requirements.
1067  */
1068 void
1069 ksignal(struct proc *p, int sig)
1070 {
1071 	lwpsignal(p, NULL, sig);
1072 }
1073 
1074 /*
1075  * The core for ksignal.  lp may be NULL, then a suitable thread
1076  * will be chosen.  If not, lp MUST be a member of p.
1077  *
1078  * If the caller wishes to call this function from a hard code section the
1079  * caller must already hold p->p_token.
1080  *
1081  * No requirements.
1082  */
1083 void
1084 lwpsignal(struct proc *p, struct lwp *lp, int sig)
1085 {
1086 	struct proc *q;
1087 	sig_t action;
1088 	int prop;
1089 
1090 	if (sig > _SIG_MAXSIG || sig <= 0) {
1091 		kprintf("lwpsignal: signal %d\n", sig);
1092 		panic("lwpsignal signal number");
1093 	}
1094 
1095 	KKASSERT(lp == NULL || lp->lwp_proc == p);
1096 
1097 	/*
1098 	 * We don't want to race... well, all sorts of things.  Get appropriate
1099 	 * tokens.
1100 	 *
1101 	 * Don't try to deliver a generic signal to an exiting process,
1102 	 * the signal structures could be in flux.  We check the LWP later
1103 	 * on.
1104 	 */
1105 	PHOLD(p);
1106 	lwkt_gettoken(&p->p_token);
1107 	if (lp) {
1108 		LWPHOLD(lp);
1109 		lwkt_gettoken(&lp->lwp_token);
1110 	} else if (p->p_flags & P_WEXIT) {
1111 		goto out;
1112 	}
1113 
1114 	prop = sigprop(sig);
1115 
1116 	/*
1117 	 * If proc is traced, always give parent a chance;
1118 	 * if signal event is tracked by procfs, give *that*
1119 	 * a chance, as well.
1120 	 */
1121 	if ((p->p_flags & P_TRACED) || (p->p_stops & S_SIG)) {
1122 		action = SIG_DFL;
1123 	} else {
1124 		/*
1125 		 * Do not try to deliver signals to an exiting lwp.  Note
1126 		 * that we must still deliver the signal if P_WEXIT is set
1127 		 * in the process flags.
1128 		 */
1129 		if (lp && (lp->lwp_mpflags & LWP_MP_WEXIT)) {
1130 			if (lp) {
1131 				lwkt_reltoken(&lp->lwp_token);
1132 				LWPRELE(lp);
1133 			}
1134 			lwkt_reltoken(&p->p_token);
1135 			PRELE(p);
1136 			return;
1137 		}
1138 
1139 		/*
1140 		 * If the signal is being ignored, then we forget about
1141 		 * it immediately.  NOTE: We don't set SIGCONT in p_sigignore,
1142 		 * and if it is set to SIG_IGN, action will be SIG_DFL here.
1143 		 */
1144 		if (SIGISMEMBER(p->p_sigignore, sig)) {
1145 			/*
1146 			 * Even if a signal is set SIG_IGN, it may still be
1147 			 * lurking in a kqueue.
1148 			 */
1149 			KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1150 			if (lp) {
1151 				lwkt_reltoken(&lp->lwp_token);
1152 				LWPRELE(lp);
1153 			}
1154 			lwkt_reltoken(&p->p_token);
1155 			PRELE(p);
1156 			return;
1157 		}
1158 		if (SIGISMEMBER(p->p_sigcatch, sig))
1159 			action = SIG_CATCH;
1160 		else
1161 			action = SIG_DFL;
1162 	}
1163 
1164 	/*
1165 	 * If continuing, clear any pending STOP signals.
1166 	 */
1167 	if (prop & SA_CONT)
1168 		SIG_STOPSIGMASK(p->p_siglist);
1169 
1170 	if (prop & SA_STOP) {
1171 		/*
1172 		 * If sending a tty stop signal to a member of an orphaned
1173 		 * process group, discard the signal here if the action
1174 		 * is default; don't stop the process below if sleeping,
1175 		 * and don't clear any pending SIGCONT.
1176 		 */
1177 		if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0 &&
1178 		    action == SIG_DFL) {
1179 			if (lp) {
1180 				lwkt_reltoken(&lp->lwp_token);
1181 				LWPRELE(lp);
1182 			}
1183 			lwkt_reltoken(&p->p_token);
1184 			PRELE(p);
1185 		        return;
1186 		}
1187 		SIG_CONTSIGMASK(p->p_siglist);
1188 		p->p_flags &= ~P_CONTINUED;
1189 	}
1190 
1191 	if (p->p_stat == SSTOP) {
1192 		/*
1193 		 * Nobody can handle this signal, add it to the lwp or
1194 		 * process pending list
1195 		 */
1196 		if (lp) {
1197 			spin_lock(&lp->lwp_spin);
1198 			SIGADDSET(lp->lwp_siglist, sig);
1199 			spin_unlock(&lp->lwp_spin);
1200 		} else {
1201 			SIGADDSET(p->p_siglist, sig);
1202 		}
1203 
1204 		/*
1205 		 * If the process is stopped and is being traced, then no
1206 		 * further action is necessary.
1207 		 */
1208 		if (p->p_flags & P_TRACED)
1209 			goto out;
1210 
1211 		/*
1212 		 * If the process is stopped and receives a KILL signal,
1213 		 * make the process runnable.
1214 		 */
1215 		if (sig == SIGKILL) {
1216 			proc_unstop(p, SSTOP);
1217 			goto active_process;
1218 		}
1219 
1220 		/*
1221 		 * If the process is stopped and receives a CONT signal,
1222 		 * then try to make the process runnable again.
1223 		 */
1224 		if (prop & SA_CONT) {
1225 			/*
1226 			 * If SIGCONT is default (or ignored), we continue the
1227 			 * process but don't leave the signal in p_siglist, as
1228 			 * it has no further action.  If SIGCONT is held, we
1229 			 * continue the process and leave the signal in
1230 			 * p_siglist.  If the process catches SIGCONT, let it
1231 			 * handle the signal itself.
1232 			 *
1233 			 * XXX what if the signal is being held blocked?
1234 			 *
1235 			 * Token required to interlock kern_wait().
1236 			 * Reparenting can also cause a race so we have to
1237 			 * hold (q).
1238 			 */
1239 			q = p->p_pptr;
1240 			PHOLD(q);
1241 			lwkt_gettoken(&q->p_token);
1242 			p->p_flags |= P_CONTINUED;
1243 			wakeup(q);
1244 			if (action == SIG_DFL)
1245 				SIGDELSET(p->p_siglist, sig);
1246 			proc_unstop(p, SSTOP);
1247 			lwkt_reltoken(&q->p_token);
1248 			PRELE(q);
1249 			if (action == SIG_CATCH)
1250 				goto active_process;
1251 			goto out;
1252 		}
1253 
1254 		/*
1255 		 * If the process is stopped and receives another STOP
1256 		 * signal, we do not need to stop it again.  If we did
1257 		 * the shell could get confused.
1258 		 *
1259 		 * However, if the current/preempted lwp is part of the
1260 		 * process receiving the signal, we need to keep it,
1261 		 * so that this lwp can stop in issignal() later, as
1262 		 * we don't want to wait until it reaches userret!
1263 		 */
1264 		if (prop & SA_STOP) {
1265 			if (lwkt_preempted_proc() == NULL ||
1266 			    lwkt_preempted_proc()->lwp_proc != p)
1267 				SIGDELSET(p->p_siglist, sig);
1268 		}
1269 
1270 		/*
1271 		 * Otherwise the process is stopped and it received some
1272 		 * signal, which does not change its stopped state.  When
1273 		 * the process is continued a wakeup(p) will be issued which
1274 		 * will wakeup any threads sleeping in tstop().
1275 		 */
1276 		if (lp == NULL) {
1277 			/* NOTE: returns lp w/ token held */
1278 			lp = find_lwp_for_signal(p, sig);
1279 		}
1280 		goto out;
1281 
1282 		/* NOTREACHED */
1283 	}
1284 	/* else not stopped */
1285 active_process:
1286 
1287 	/*
1288 	 * Never deliver a lwp-specific signal to a random lwp.
1289 	 */
1290 	if (lp == NULL) {
1291 		/* NOTE: returns lp w/ token held */
1292 		lp = find_lwp_for_signal(p, sig);
1293 		if (lp) {
1294 			if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
1295 				lwkt_reltoken(&lp->lwp_token);
1296 				LWPRELE(lp);
1297 				lp = NULL;
1298 			}
1299 		}
1300 	}
1301 
1302 	/*
1303 	 * Deliver to the process generically if (1) the signal is being
1304 	 * sent to any thread or (2) we could not find a thread to deliver
1305 	 * it to.
1306 	 */
1307 	if (lp == NULL) {
1308 		SIGADDSET(p->p_siglist, sig);
1309 		goto out;
1310 	}
1311 
1312 	/*
1313 	 * Deliver to a specific LWP whether it masks it or not.  It will
1314 	 * not be dispatched if masked but we must still deliver it.
1315 	 */
1316 	if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
1317 	    (p->p_flags & P_TRACED) == 0) {
1318 		p->p_nice = NZERO;
1319 	}
1320 
1321 	/*
1322 	 * If the process receives a STOP signal which indeed needs to
1323 	 * stop the process, do so.  If the process chose to catch the
1324 	 * signal, it will be treated like any other signal.
1325 	 */
1326 	if ((prop & SA_STOP) && action == SIG_DFL) {
1327 		/*
1328 		 * If a child holding parent blocked, stopping
1329 		 * could cause deadlock.  Take no action at this
1330 		 * time.
1331 		 */
1332 		if (p->p_flags & P_PPWAIT) {
1333 			SIGADDSET(p->p_siglist, sig);
1334 			goto out;
1335 		}
1336 
1337 		/*
1338 		 * Do not actually try to manipulate the process, but simply
1339 		 * stop it.  Lwps will stop as soon as they safely can.
1340 		 *
1341 		 * Ignore stop if the process is exiting.
1342 		 */
1343 		if ((p->p_flags & P_WEXIT) == 0) {
1344 			p->p_xstat = sig;
1345 			proc_stop(p, SSTOP);
1346 		}
1347 		goto out;
1348 	}
1349 
1350 	/*
1351 	 * If it is a CONT signal with default action, just ignore it.
1352 	 */
1353 	if ((prop & SA_CONT) && action == SIG_DFL)
1354 		goto out;
1355 
1356 	/*
1357 	 * Mark signal pending at this specific thread.
1358 	 */
1359 	spin_lock(&lp->lwp_spin);
1360 	SIGADDSET(lp->lwp_siglist, sig);
1361 	spin_unlock(&lp->lwp_spin);
1362 
1363 	lwp_signotify(lp);
1364 
1365 out:
1366 	if (lp) {
1367 		lwkt_reltoken(&lp->lwp_token);
1368 		LWPRELE(lp);
1369 	}
1370 	lwkt_reltoken(&p->p_token);
1371 	PRELE(p);
1372 }
1373 
1374 /*
1375  * Notify the LWP that a signal has arrived.  The LWP does not have to be
1376  * sleeping on the current cpu.
1377  *
1378  * p->p_token and lp->lwp_token must be held on call.
1379  *
1380  * We can only safely schedule the thread on its current cpu and only if
1381  * one of the SINTR flags is set.  If an SINTR flag is set AND we are on
1382  * the correct cpu we are properly interlocked, otherwise we could be
1383  * racing other thread transition states (or the lwp is on the user scheduler
1384  * runq but not scheduled) and must not do anything.
1385  *
1386  * Since we hold the lwp token we know the lwp cannot be ripped out from
1387  * under us so we can safely hold it to prevent it from being ripped out
1388  * from under us if we are forced to IPI another cpu to make the local
1389  * checks there.
1390  *
1391  * Adjustment of lp->lwp_stat can only occur when we hold the lwp_token,
1392  * which we won't in an IPI so any fixups have to be done here, effectively
1393  * replicating part of what setrunnable() does.
1394  */
1395 static void
1396 lwp_signotify(struct lwp *lp)
1397 {
1398 	ASSERT_LWKT_TOKEN_HELD(&lp->lwp_proc->p_token);
1399 
1400 	crit_enter();
1401 	if (lp == lwkt_preempted_proc()) {
1402 		/*
1403 		 * lwp is on the current cpu AND it is currently running
1404 		 * (we preempted it).
1405 		 */
1406 		signotify();
1407 	} else if (lp->lwp_flags & LWP_SINTR) {
1408 		/*
1409 		 * lwp is sitting in tsleep() with PCATCH set
1410 		 */
1411 		if (lp->lwp_thread->td_gd == mycpu) {
1412 			setrunnable(lp);
1413 		} else {
1414 			/*
1415 			 * We can only adjust lwp_stat while we hold the
1416 			 * lwp_token, and we won't in the IPI function.
1417 			 */
1418 			LWPHOLD(lp);
1419 			if (lp->lwp_stat == LSSTOP)
1420 				lp->lwp_stat = LSSLEEP;
1421 			lwkt_send_ipiq(lp->lwp_thread->td_gd,
1422 				       lwp_signotify_remote, lp);
1423 		}
1424 	} else if (lp->lwp_thread->td_flags & TDF_SINTR) {
1425 		/*
1426 		 * lwp is sitting in lwkt_sleep() with PCATCH set.
1427 		 */
1428 		if (lp->lwp_thread->td_gd == mycpu) {
1429 			setrunnable(lp);
1430 		} else {
1431 			/*
1432 			 * We can only adjust lwp_stat while we hold the
1433 			 * lwp_token, and we won't in the IPI function.
1434 			 */
1435 			LWPHOLD(lp);
1436 			if (lp->lwp_stat == LSSTOP)
1437 				lp->lwp_stat = LSSLEEP;
1438 			lwkt_send_ipiq(lp->lwp_thread->td_gd,
1439 				       lwp_signotify_remote, lp);
1440 		}
1441 	} else {
1442 		/*
1443 		 * Otherwise the lwp is either in some uninterruptable state
1444 		 * or it is on the userland scheduler's runqueue waiting to
1445 		 * be scheduled to a cpu.
1446 		 */
1447 	}
1448 	crit_exit();
1449 }
1450 
1451 /*
1452  * This function is called via an IPI so we cannot call setrunnable() here
1453  * (because while we hold the lp we don't own its token, and can't get it
1454  * from an IPI).
1455  *
1456  * We are interlocked by virtue of being on the same cpu as the target.  If
1457  * we still are and LWP_SINTR or TDF_SINTR is set we can safely schedule
1458  * the target thread.
1459  */
1460 static void
1461 lwp_signotify_remote(void *arg)
1462 {
1463 	struct lwp *lp = arg;
1464 	thread_t td = lp->lwp_thread;
1465 
1466 	if (lp == lwkt_preempted_proc()) {
1467 		signotify();
1468 		LWPRELE(lp);
1469 	} else if (td->td_gd == mycpu) {
1470 		if ((lp->lwp_flags & LWP_SINTR) ||
1471 		    (td->td_flags & TDF_SINTR)) {
1472 			lwkt_schedule(td);
1473 		}
1474 		LWPRELE(lp);
1475 	} else {
1476 		lwkt_send_ipiq(td->td_gd, lwp_signotify_remote, lp);
1477 		/* LWPHOLD() is forwarded to the target cpu */
1478 	}
1479 }
1480 
1481 /*
1482  * Caller must hold p->p_token
1483  */
1484 void
1485 proc_stop(struct proc *p, int sig)
1486 {
1487 	struct proc *q;
1488 	struct lwp *lp;
1489 
1490 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1491 
1492 	/*
1493 	 * If somebody raced us, be happy with it.  SCORE overrides SSTOP.
1494 	 */
1495 	if (sig == SCORE) {
1496 		if (p->p_stat == SCORE || p->p_stat == SZOMB)
1497 			return;
1498 	} else {
1499 		if (p->p_stat == SSTOP || p->p_stat == SCORE ||
1500 		    p->p_stat == SZOMB) {
1501 			return;
1502 		}
1503 	}
1504 	p->p_stat = sig;
1505 
1506 	FOREACH_LWP_IN_PROC(lp, p) {
1507 		LWPHOLD(lp);
1508 		lwkt_gettoken(&lp->lwp_token);
1509 
1510 		switch (lp->lwp_stat) {
1511 		case LSSTOP:
1512 			/*
1513 			 * Do nothing, we are already counted in
1514 			 * p_nstopped.
1515 			 */
1516 			break;
1517 
1518 		case LSSLEEP:
1519 			/*
1520 			 * We're sleeping, but we will stop before
1521 			 * returning to userspace, so count us
1522 			 * as stopped as well.  We set LWP_MP_WSTOP
1523 			 * to signal the lwp that it should not
1524 			 * increase p_nstopped when reaching tstop().
1525 			 *
1526 			 * LWP_MP_WSTOP is protected by lp->lwp_token.
1527 			 */
1528 			if ((lp->lwp_mpflags & LWP_MP_WSTOP) == 0) {
1529 				atomic_set_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
1530 				++p->p_nstopped;
1531 			}
1532 			break;
1533 
1534 		case LSRUN:
1535 			/*
1536 			 * We might notify ourself, but that's not
1537 			 * a problem.
1538 			 */
1539 			lwp_signotify(lp);
1540 			break;
1541 		}
1542 		lwkt_reltoken(&lp->lwp_token);
1543 		LWPRELE(lp);
1544 	}
1545 
1546 	if (p->p_nstopped == p->p_nthreads) {
1547 		/*
1548 		 * Token required to interlock kern_wait().  Reparenting can
1549 		 * also cause a race so we have to hold (q).
1550 		 */
1551 		q = p->p_pptr;
1552 		PHOLD(q);
1553 		lwkt_gettoken(&q->p_token);
1554 		p->p_flags &= ~P_WAITED;
1555 		wakeup(q);
1556 		if ((q->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1557 			ksignal(p->p_pptr, SIGCHLD);
1558 		lwkt_reltoken(&q->p_token);
1559 		PRELE(q);
1560 	}
1561 }
1562 
1563 /*
1564  * Caller must hold p_token
1565  */
1566 void
1567 proc_unstop(struct proc *p, int sig)
1568 {
1569 	struct lwp *lp;
1570 
1571 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1572 
1573 	if (p->p_stat != sig)
1574 		return;
1575 
1576 	p->p_stat = SACTIVE;
1577 
1578 	FOREACH_LWP_IN_PROC(lp, p) {
1579 		LWPHOLD(lp);
1580 		lwkt_gettoken(&lp->lwp_token);
1581 
1582 		switch (lp->lwp_stat) {
1583 		case LSRUN:
1584 			/*
1585 			 * Uh?  Not stopped?  Well, I guess that's okay.
1586 			 */
1587 			if (bootverbose)
1588 				kprintf("proc_unstop: lwp %d/%d not sleeping\n",
1589 					p->p_pid, lp->lwp_tid);
1590 			break;
1591 
1592 		case LSSLEEP:
1593 			/*
1594 			 * Still sleeping.  Don't bother waking it up.
1595 			 * However, if this thread was counted as
1596 			 * stopped, undo this.
1597 			 *
1598 			 * Nevertheless we call setrunnable() so that it
1599 			 * will wake up in case a signal or timeout arrived
1600 			 * in the meantime.
1601 			 *
1602 			 * LWP_MP_WSTOP is protected by lp->lwp_token.
1603 			 */
1604 			if (lp->lwp_mpflags & LWP_MP_WSTOP) {
1605 				atomic_clear_int(&lp->lwp_mpflags,
1606 						 LWP_MP_WSTOP);
1607 				--p->p_nstopped;
1608 			} else {
1609 				if (bootverbose)
1610 					kprintf("proc_unstop: lwp %d/%d sleeping, not stopped\n",
1611 						p->p_pid, lp->lwp_tid);
1612 			}
1613 			/* FALLTHROUGH */
1614 
1615 		case LSSTOP:
1616 			/*
1617 			 * This handles any lwp's waiting in a tsleep with
1618 			 * SIGCATCH.
1619 			 */
1620 			lwp_signotify(lp);
1621 			break;
1622 
1623 		}
1624 		lwkt_reltoken(&lp->lwp_token);
1625 		LWPRELE(lp);
1626 	}
1627 
1628 	/*
1629 	 * This handles any lwp's waiting in tstop().  We have interlocked
1630 	 * the setting of p_stat by acquiring and releasing each lpw's
1631 	 * token.
1632 	 */
1633 	wakeup(p);
1634 }
1635 
1636 /*
1637  * Wait for all threads except the current thread to stop.
1638  */
1639 static void
1640 proc_stopwait(struct proc *p)
1641 {
1642 	while ((p->p_stat == SSTOP || p->p_stat == SCORE) &&
1643 	       p->p_nstopped < p->p_nthreads - 1) {
1644 		tsleep_interlock(&p->p_nstopped, 0);
1645 		if (p->p_nstopped < p->p_nthreads - 1) {
1646 			tsleep(&p->p_nstopped, PINTERLOCKED, "stopwt", hz);
1647 		}
1648 	}
1649 }
1650 
1651 /*
1652  * No requirements.
1653  */
1654 static int
1655 kern_sigtimedwait(sigset_t waitset, siginfo_t *info, struct timespec *timeout)
1656 {
1657 	sigset_t savedmask, set;
1658 	struct proc *p = curproc;
1659 	struct lwp *lp = curthread->td_lwp;
1660 	int error, sig, hz, timevalid = 0;
1661 	struct timespec rts, ets, ts;
1662 	struct timeval tv;
1663 
1664 	error = 0;
1665 	sig = 0;
1666 	ets.tv_sec = 0;		/* silence compiler warning */
1667 	ets.tv_nsec = 0;	/* silence compiler warning */
1668 	SIG_CANTMASK(waitset);
1669 	savedmask = lp->lwp_sigmask;
1670 
1671 	if (timeout) {
1672 		if (timeout->tv_sec >= 0 && timeout->tv_nsec >= 0 &&
1673 		    timeout->tv_nsec < 1000000000) {
1674 			timevalid = 1;
1675 			getnanouptime(&rts);
1676 		 	ets = rts;
1677 			timespecadd(&ets, timeout);
1678 		}
1679 	}
1680 
1681 	for (;;) {
1682 		set = lwp_sigpend(lp);
1683 		SIGSETAND(set, waitset);
1684 		if ((sig = sig_ffs(&set)) != 0) {
1685 			SIGFILLSET(lp->lwp_sigmask);
1686 			SIGDELSET(lp->lwp_sigmask, sig);
1687 			SIG_CANTMASK(lp->lwp_sigmask);
1688 			sig = issignal(lp, 1);
1689 			/*
1690 			 * It may be a STOP signal, in the case, issignal
1691 			 * returns 0, because we may stop there, and new
1692 			 * signal can come in, we should restart if we got
1693 			 * nothing.
1694 			 */
1695 			if (sig == 0)
1696 				continue;
1697 			else
1698 				break;
1699 		}
1700 
1701 		/*
1702 		 * Previous checking got nothing, and we retried but still
1703 		 * got nothing, we should return the error status.
1704 		 */
1705 		if (error)
1706 			break;
1707 
1708 		/*
1709 		 * POSIX says this must be checked after looking for pending
1710 		 * signals.
1711 		 */
1712 		if (timeout) {
1713 			if (timevalid == 0) {
1714 				error = EINVAL;
1715 				break;
1716 			}
1717 			getnanouptime(&rts);
1718 			if (timespeccmp(&rts, &ets, >=)) {
1719 				error = EAGAIN;
1720 				break;
1721 			}
1722 			ts = ets;
1723 			timespecsub(&ts, &rts);
1724 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1725 			hz = tvtohz_high(&tv);
1726 		} else {
1727 			hz = 0;
1728 		}
1729 
1730 		lp->lwp_sigmask = savedmask;
1731 		SIGSETNAND(lp->lwp_sigmask, waitset);
1732 		/*
1733 		 * We won't ever be woken up.  Instead, our sleep will
1734 		 * be broken in lwpsignal().
1735 		 */
1736 		error = tsleep(&p->p_sigacts, PCATCH, "sigwt", hz);
1737 		if (timeout) {
1738 			if (error == ERESTART) {
1739 				/* can not restart a timeout wait. */
1740 				error = EINTR;
1741 			} else if (error == EAGAIN) {
1742 				/* will calculate timeout by ourself. */
1743 				error = 0;
1744 			}
1745 		}
1746 		/* Retry ... */
1747 	}
1748 
1749 	lp->lwp_sigmask = savedmask;
1750 	if (sig) {
1751 		error = 0;
1752 		bzero(info, sizeof(*info));
1753 		info->si_signo = sig;
1754 		spin_lock(&lp->lwp_spin);
1755 		lwp_delsig(lp, sig);	/* take the signal! */
1756 		spin_unlock(&lp->lwp_spin);
1757 
1758 		if (sig == SIGKILL) {
1759 			sigexit(lp, sig);
1760 			/* NOT REACHED */
1761 		}
1762 	}
1763 
1764 	return (error);
1765 }
1766 
1767 /*
1768  * MPALMOSTSAFE
1769  */
1770 int
1771 sys_sigtimedwait(struct sigtimedwait_args *uap)
1772 {
1773 	struct timespec ts;
1774 	struct timespec *timeout;
1775 	sigset_t set;
1776 	siginfo_t info;
1777 	int error;
1778 
1779 	if (uap->timeout) {
1780 		error = copyin(uap->timeout, &ts, sizeof(ts));
1781 		if (error)
1782 			return (error);
1783 		timeout = &ts;
1784 	} else {
1785 		timeout = NULL;
1786 	}
1787 	error = copyin(uap->set, &set, sizeof(set));
1788 	if (error)
1789 		return (error);
1790 	error = kern_sigtimedwait(set, &info, timeout);
1791 	if (error)
1792 		return (error);
1793 	if (uap->info)
1794 		error = copyout(&info, uap->info, sizeof(info));
1795 	/* Repost if we got an error. */
1796 	/*
1797 	 * XXX lwp
1798 	 *
1799 	 * This could transform a thread-specific signal to another
1800 	 * thread / process pending signal.
1801 	 */
1802 	if (error) {
1803 		ksignal(curproc, info.si_signo);
1804 	} else {
1805 		uap->sysmsg_result = info.si_signo;
1806 	}
1807 	return (error);
1808 }
1809 
1810 /*
1811  * MPALMOSTSAFE
1812  */
1813 int
1814 sys_sigwaitinfo(struct sigwaitinfo_args *uap)
1815 {
1816 	siginfo_t info;
1817 	sigset_t set;
1818 	int error;
1819 
1820 	error = copyin(uap->set, &set, sizeof(set));
1821 	if (error)
1822 		return (error);
1823 	error = kern_sigtimedwait(set, &info, NULL);
1824 	if (error)
1825 		return (error);
1826 	if (uap->info)
1827 		error = copyout(&info, uap->info, sizeof(info));
1828 	/* Repost if we got an error. */
1829 	/*
1830 	 * XXX lwp
1831 	 *
1832 	 * This could transform a thread-specific signal to another
1833 	 * thread / process pending signal.
1834 	 */
1835 	if (error) {
1836 		ksignal(curproc, info.si_signo);
1837 	} else {
1838 		uap->sysmsg_result = info.si_signo;
1839 	}
1840 	return (error);
1841 }
1842 
1843 /*
1844  * If the current process has received a signal that would interrupt a
1845  * system call, return EINTR or ERESTART as appropriate.
1846  */
1847 int
1848 iscaught(struct lwp *lp)
1849 {
1850 	struct proc *p = lp->lwp_proc;
1851 	int sig;
1852 
1853 	if (p) {
1854 		if ((sig = CURSIG(lp)) != 0) {
1855 			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1856 				return (EINTR);
1857 			return (ERESTART);
1858 		}
1859 	}
1860 	return(EWOULDBLOCK);
1861 }
1862 
1863 /*
1864  * If the current process has received a signal (should be caught or cause
1865  * termination, should interrupt current syscall), return the signal number.
1866  * Stop signals with default action are processed immediately, then cleared;
1867  * they aren't returned.  This is checked after each entry to the system for
1868  * a syscall or trap (though this can usually be done without calling issignal
1869  * by checking the pending signal masks in the CURSIG macro).
1870  *
1871  * This routine is called via CURSIG/__cursig.  We will acquire and release
1872  * p->p_token but if the caller needs to interlock the test the caller must
1873  * also hold p->p_token.
1874  *
1875  *	while (sig = CURSIG(curproc))
1876  *		postsig(sig);
1877  *
1878  * MPSAFE
1879  */
1880 int
1881 issignal(struct lwp *lp, int maytrace)
1882 {
1883 	struct proc *p = lp->lwp_proc;
1884 	sigset_t mask;
1885 	int sig, prop;
1886 
1887 	lwkt_gettoken(&p->p_token);
1888 
1889 	for (;;) {
1890 		int traced = (p->p_flags & P_TRACED) || (p->p_stops & S_SIG);
1891 
1892 		/*
1893 		 * If this process is supposed to stop, stop this thread.
1894 		 */
1895 		if (p->p_stat == SSTOP || p->p_stat == SCORE)
1896 			tstop();
1897 
1898 		mask = lwp_sigpend(lp);
1899 		SIGSETNAND(mask, lp->lwp_sigmask);
1900 		if (p->p_flags & P_PPWAIT)
1901 			SIG_STOPSIGMASK(mask);
1902 		if (SIGISEMPTY(mask)) {		/* no signal to send */
1903 			lwkt_reltoken(&p->p_token);
1904 			return (0);
1905 		}
1906 		sig = sig_ffs(&mask);
1907 
1908 		STOPEVENT(p, S_SIG, sig);
1909 
1910 		/*
1911 		 * We should see pending but ignored signals
1912 		 * only if P_TRACED was on when they were posted.
1913 		 */
1914 		if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
1915 			spin_lock(&lp->lwp_spin);
1916 			lwp_delsig(lp, sig);
1917 			spin_unlock(&lp->lwp_spin);
1918 			continue;
1919 		}
1920 		if (maytrace &&
1921 		    (p->p_flags & P_TRACED) &&
1922 		    (p->p_flags & P_PPWAIT) == 0) {
1923 			/*
1924 			 * If traced, always stop, and stay stopped until
1925 			 * released by the parent.
1926 			 *
1927 			 * NOTE: SSTOP may get cleared during the loop,
1928 			 * but we do not re-notify the parent if we have
1929 			 * to loop several times waiting for the parent
1930 			 * to let us continue.
1931 			 *
1932 			 * XXX not sure if this is still true
1933 			 */
1934 			p->p_xstat = sig;
1935 			proc_stop(p, SSTOP);
1936 			do {
1937 				tstop();
1938 			} while (!trace_req(p) && (p->p_flags & P_TRACED));
1939 
1940 			/*
1941 			 * If parent wants us to take the signal,
1942 			 * then it will leave it in p->p_xstat;
1943 			 * otherwise we just look for signals again.
1944 			 */
1945 			spin_lock(&lp->lwp_spin);
1946 			lwp_delsig(lp, sig);	/* clear old signal */
1947 			spin_unlock(&lp->lwp_spin);
1948 			sig = p->p_xstat;
1949 			if (sig == 0)
1950 				continue;
1951 
1952 			/*
1953 			 * Put the new signal into p_siglist.  If the
1954 			 * signal is being masked, look for other signals.
1955 			 *
1956 			 * XXX lwp might need a call to ksignal()
1957 			 */
1958 			SIGADDSET(p->p_siglist, sig);
1959 			if (SIGISMEMBER(lp->lwp_sigmask, sig))
1960 				continue;
1961 
1962 			/*
1963 			 * If the traced bit got turned off, go back up
1964 			 * to the top to rescan signals.  This ensures
1965 			 * that p_sig* and ps_sigact are consistent.
1966 			 */
1967 			if ((p->p_flags & P_TRACED) == 0)
1968 				continue;
1969 		}
1970 
1971 		prop = sigprop(sig);
1972 
1973 		/*
1974 		 * Decide whether the signal should be returned.
1975 		 * Return the signal's number, or fall through
1976 		 * to clear it from the pending mask.
1977 		 */
1978 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
1979 		case (intptr_t)SIG_DFL:
1980 			/*
1981 			 * Don't take default actions on system processes.
1982 			 */
1983 			if (p->p_pid <= 1) {
1984 #ifdef DIAGNOSTIC
1985 				/*
1986 				 * Are you sure you want to ignore SIGSEGV
1987 				 * in init? XXX
1988 				 */
1989 				kprintf("Process (pid %lu) got signal %d\n",
1990 					(u_long)p->p_pid, sig);
1991 #endif
1992 				break;		/* == ignore */
1993 			}
1994 
1995 			/*
1996 			 * Handle the in-kernel checkpoint action
1997 			 */
1998 			if (prop & SA_CKPT) {
1999 				checkpoint_signal_handler(lp);
2000 				break;
2001 			}
2002 
2003 			/*
2004 			 * If there is a pending stop signal to process
2005 			 * with default action, stop here,
2006 			 * then clear the signal.  However,
2007 			 * if process is member of an orphaned
2008 			 * process group, ignore tty stop signals.
2009 			 */
2010 			if (prop & SA_STOP) {
2011 				if (p->p_flags & P_TRACED ||
2012 		    		    (p->p_pgrp->pg_jobc == 0 &&
2013 				    prop & SA_TTYSTOP))
2014 					break;	/* == ignore */
2015 				if ((p->p_flags & P_WEXIT) == 0) {
2016 					p->p_xstat = sig;
2017 					proc_stop(p, SSTOP);
2018 					tstop();
2019 				}
2020 				break;
2021 			} else if (prop & SA_IGNORE) {
2022 				/*
2023 				 * Except for SIGCONT, shouldn't get here.
2024 				 * Default action is to ignore; drop it.
2025 				 */
2026 				break;		/* == ignore */
2027 			} else {
2028 				lwkt_reltoken(&p->p_token);
2029 				return (sig);
2030 			}
2031 
2032 			/*NOTREACHED*/
2033 
2034 		case (intptr_t)SIG_IGN:
2035 			/*
2036 			 * Masking above should prevent us ever trying
2037 			 * to take action on an ignored signal other
2038 			 * than SIGCONT, unless process is traced.
2039 			 */
2040 			if ((prop & SA_CONT) == 0 &&
2041 			    (p->p_flags & P_TRACED) == 0)
2042 				kprintf("issignal\n");
2043 			break;		/* == ignore */
2044 
2045 		default:
2046 			/*
2047 			 * This signal has an action, let
2048 			 * postsig() process it.
2049 			 */
2050 			lwkt_reltoken(&p->p_token);
2051 			return (sig);
2052 		}
2053 		spin_lock(&lp->lwp_spin);
2054 		lwp_delsig(lp, sig);		/* take the signal! */
2055 		spin_unlock(&lp->lwp_spin);
2056 	}
2057 	/* NOTREACHED */
2058 }
2059 
2060 /*
2061  * Take the action for the specified signal
2062  * from the current set of pending signals.
2063  *
2064  * Caller must hold p->p_token
2065  */
2066 void
2067 postsig(int sig)
2068 {
2069 	struct lwp *lp = curthread->td_lwp;
2070 	struct proc *p = lp->lwp_proc;
2071 	struct sigacts *ps = p->p_sigacts;
2072 	sig_t action;
2073 	sigset_t returnmask;
2074 	int code;
2075 
2076 	KASSERT(sig != 0, ("postsig"));
2077 
2078 	KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
2079 
2080 	/*
2081 	 * If we are a virtual kernel running an emulated user process
2082 	 * context, switch back to the virtual kernel context before
2083 	 * trying to post the signal.
2084 	 */
2085 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
2086 		struct trapframe *tf = lp->lwp_md.md_regs;
2087 		tf->tf_trapno = 0;
2088 		vkernel_trap(lp, tf);
2089 	}
2090 
2091 	spin_lock(&lp->lwp_spin);
2092 	lwp_delsig(lp, sig);
2093 	spin_unlock(&lp->lwp_spin);
2094 	action = ps->ps_sigact[_SIG_IDX(sig)];
2095 #ifdef KTRACE
2096 	if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
2097 		ktrpsig(lp, sig, action, lp->lwp_flags & LWP_OLDMASK ?
2098 			&lp->lwp_oldsigmask : &lp->lwp_sigmask, 0);
2099 #endif
2100 	STOPEVENT(p, S_SIG, sig);
2101 
2102 	if (action == SIG_DFL) {
2103 		/*
2104 		 * Default action, where the default is to kill
2105 		 * the process.  (Other cases were ignored above.)
2106 		 */
2107 		sigexit(lp, sig);
2108 		/* NOTREACHED */
2109 	} else {
2110 		/*
2111 		 * If we get here, the signal must be caught.
2112 		 */
2113 		KASSERT(action != SIG_IGN && !SIGISMEMBER(lp->lwp_sigmask, sig),
2114 		    ("postsig action"));
2115 
2116 		/*
2117 		 * Reset the signal handler if asked to
2118 		 */
2119 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
2120 			/*
2121 			 * See kern_sigaction() for origin of this code.
2122 			 */
2123 			SIGDELSET(p->p_sigcatch, sig);
2124 			if (sig != SIGCONT &&
2125 			    sigprop(sig) & SA_IGNORE)
2126 				SIGADDSET(p->p_sigignore, sig);
2127 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2128 		}
2129 
2130 		/*
2131 		 * Set the signal mask and calculate the mask to restore
2132 		 * when the signal function returns.
2133 		 *
2134 		 * Special case: user has done a sigsuspend.  Here the
2135 		 * current mask is not of interest, but rather the
2136 		 * mask from before the sigsuspend is what we want
2137 		 * restored after the signal processing is completed.
2138 		 */
2139 		if (lp->lwp_flags & LWP_OLDMASK) {
2140 			returnmask = lp->lwp_oldsigmask;
2141 			lp->lwp_flags &= ~LWP_OLDMASK;
2142 		} else {
2143 			returnmask = lp->lwp_sigmask;
2144 		}
2145 
2146 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
2147 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
2148 			SIGADDSET(lp->lwp_sigmask, sig);
2149 
2150 		lp->lwp_ru.ru_nsignals++;
2151 		if (lp->lwp_sig != sig) {
2152 			code = 0;
2153 		} else {
2154 			code = lp->lwp_code;
2155 			lp->lwp_code = 0;
2156 			lp->lwp_sig = 0;
2157 		}
2158 		(*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
2159 	}
2160 }
2161 
2162 /*
2163  * Kill the current process for stated reason.
2164  */
2165 void
2166 killproc(struct proc *p, char *why)
2167 {
2168 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n",
2169 		p->p_pid, p->p_comm,
2170 		p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2171 	ksignal(p, SIGKILL);
2172 }
2173 
2174 /*
2175  * Force the current process to exit with the specified signal, dumping core
2176  * if appropriate.  We bypass the normal tests for masked and caught signals,
2177  * allowing unrecoverable failures to terminate the process without changing
2178  * signal state.  Mark the accounting record with the signal termination.
2179  * If dumping core, save the signal number for the debugger.  Calls exit and
2180  * does not return.
2181  *
2182  * This routine does not return.
2183  */
2184 void
2185 sigexit(struct lwp *lp, int sig)
2186 {
2187 	struct proc *p = lp->lwp_proc;
2188 
2189 	lwkt_gettoken(&p->p_token);
2190 	p->p_acflag |= AXSIG;
2191 	if (sigprop(sig) & SA_CORE) {
2192 		lp->lwp_sig = sig;
2193 
2194 		/*
2195 		 * All threads must be stopped before we can safely coredump.
2196 		 * Stop threads using SCORE, which cannot be overridden.
2197 		 */
2198 		if (p->p_stat != SCORE) {
2199 			proc_stop(p, SCORE);
2200 			proc_stopwait(p);
2201 
2202 			if (coredump(lp, sig) == 0)
2203 				sig |= WCOREFLAG;
2204 			p->p_stat = SSTOP;
2205 		}
2206 
2207 		/*
2208 		 * Log signals which would cause core dumps
2209 		 * (Log as LOG_INFO to appease those who don't want
2210 		 * these messages.)
2211 		 * XXX : Todo, as well as euid, write out ruid too
2212 		 */
2213 		if (kern_logsigexit)
2214 			log(LOG_INFO,
2215 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2216 			    p->p_pid, p->p_comm,
2217 			    p->p_ucred ? p->p_ucred->cr_uid : -1,
2218 			    sig &~ WCOREFLAG,
2219 			    sig & WCOREFLAG ? " (core dumped)" : "");
2220 	}
2221 	lwkt_reltoken(&p->p_token);
2222 	exit1(W_EXITCODE(0, sig));
2223 	/* NOTREACHED */
2224 }
2225 
2226 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
2227 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
2228 	      sizeof(corefilename), "process corefile name format string");
2229 
2230 /*
2231  * expand_name(name, uid, pid)
2232  * Expand the name described in corefilename, using name, uid, and pid.
2233  * corefilename is a kprintf-like string, with three format specifiers:
2234  *	%N	name of process ("name")
2235  *	%P	process id (pid)
2236  *	%U	user id (uid)
2237  * For example, "%N.core" is the default; they can be disabled completely
2238  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
2239  * This is controlled by the sysctl variable kern.corefile (see above).
2240  */
2241 
2242 static char *
2243 expand_name(const char *name, uid_t uid, pid_t pid)
2244 {
2245 	char *temp;
2246 	char buf[11];		/* Buffer for pid/uid -- max 4B */
2247 	int i, n;
2248 	char *format = corefilename;
2249 	size_t namelen;
2250 
2251 	temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
2252 	if (temp == NULL)
2253 		return NULL;
2254 	namelen = strlen(name);
2255 	for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
2256 		int l;
2257 		switch (format[i]) {
2258 		case '%':	/* Format character */
2259 			i++;
2260 			switch (format[i]) {
2261 			case '%':
2262 				temp[n++] = '%';
2263 				break;
2264 			case 'N':	/* process name */
2265 				if ((n + namelen) > MAXPATHLEN) {
2266 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2267 					    pid, name, uid, temp, name);
2268 					kfree(temp, M_TEMP);
2269 					return NULL;
2270 				}
2271 				memcpy(temp+n, name, namelen);
2272 				n += namelen;
2273 				break;
2274 			case 'P':	/* process id */
2275 				l = ksprintf(buf, "%u", pid);
2276 				if ((n + l) > MAXPATHLEN) {
2277 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2278 					    pid, name, uid, temp, name);
2279 					kfree(temp, M_TEMP);
2280 					return NULL;
2281 				}
2282 				memcpy(temp+n, buf, l);
2283 				n += l;
2284 				break;
2285 			case 'U':	/* user id */
2286 				l = ksprintf(buf, "%u", uid);
2287 				if ((n + l) > MAXPATHLEN) {
2288 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2289 					    pid, name, uid, temp, name);
2290 					kfree(temp, M_TEMP);
2291 					return NULL;
2292 				}
2293 				memcpy(temp+n, buf, l);
2294 				n += l;
2295 				break;
2296 			default:
2297 			  	log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
2298 			}
2299 			break;
2300 		default:
2301 			temp[n++] = format[i];
2302 		}
2303 	}
2304 	temp[n] = '\0';
2305 	return temp;
2306 }
2307 
2308 /*
2309  * Dump a process' core.  The main routine does some
2310  * policy checking, and creates the name of the coredump;
2311  * then it passes on a vnode and a size limit to the process-specific
2312  * coredump routine if there is one; if there _is not_ one, it returns
2313  * ENOSYS; otherwise it returns the error from the process-specific routine.
2314  *
2315  * The parameter `lp' is the lwp which triggered the coredump.
2316  */
2317 
2318 static int
2319 coredump(struct lwp *lp, int sig)
2320 {
2321 	struct proc *p = lp->lwp_proc;
2322 	struct vnode *vp;
2323 	struct ucred *cred = p->p_ucred;
2324 	struct flock lf;
2325 	struct nlookupdata nd;
2326 	struct vattr vattr;
2327 	int error, error1;
2328 	char *name;			/* name of corefile */
2329 	off_t limit;
2330 
2331 	STOPEVENT(p, S_CORE, 0);
2332 
2333 	if (((sugid_coredump == 0) && p->p_flags & P_SUGID) || do_coredump == 0)
2334 		return (EFAULT);
2335 
2336 	/*
2337 	 * Note that the bulk of limit checking is done after
2338 	 * the corefile is created.  The exception is if the limit
2339 	 * for corefiles is 0, in which case we don't bother
2340 	 * creating the corefile at all.  This layout means that
2341 	 * a corefile is truncated instead of not being created,
2342 	 * if it is larger than the limit.
2343 	 */
2344 	limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
2345 	if (limit == 0)
2346 		return EFBIG;
2347 
2348 	name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
2349 	if (name == NULL)
2350 		return (EINVAL);
2351 	error = nlookup_init(&nd, name, UIO_SYSSPACE, NLC_LOCKVP);
2352 	if (error == 0)
2353 		error = vn_open(&nd, NULL, O_CREAT | FWRITE | O_NOFOLLOW, S_IRUSR | S_IWUSR);
2354 	kfree(name, M_TEMP);
2355 	if (error) {
2356 		nlookup_done(&nd);
2357 		return (error);
2358 	}
2359 	vp = nd.nl_open_vp;
2360 	nd.nl_open_vp = NULL;
2361 	nlookup_done(&nd);
2362 
2363 	vn_unlock(vp);
2364 	lf.l_whence = SEEK_SET;
2365 	lf.l_start = 0;
2366 	lf.l_len = 0;
2367 	lf.l_type = F_WRLCK;
2368 	error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, 0);
2369 	if (error)
2370 		goto out2;
2371 
2372 	/* Don't dump to non-regular files or files with links. */
2373 	if (vp->v_type != VREG ||
2374 	    VOP_GETATTR(vp, &vattr) || vattr.va_nlink != 1) {
2375 		error = EFAULT;
2376 		goto out1;
2377 	}
2378 
2379 	/* Don't dump to files current user does not own */
2380 	if (vattr.va_uid != p->p_ucred->cr_uid) {
2381 		error = EFAULT;
2382 		goto out1;
2383 	}
2384 
2385 	VATTR_NULL(&vattr);
2386 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2387 	vattr.va_size = 0;
2388 	VOP_SETATTR(vp, &vattr, cred);
2389 	p->p_acflag |= ACORE;
2390 	vn_unlock(vp);
2391 
2392 	error = p->p_sysent->sv_coredump ?
2393 		  p->p_sysent->sv_coredump(lp, sig, vp, limit) : ENOSYS;
2394 
2395 out1:
2396 	lf.l_type = F_UNLCK;
2397 	VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, 0);
2398 out2:
2399 	error1 = vn_close(vp, FWRITE, NULL);
2400 	if (error == 0)
2401 		error = error1;
2402 	return (error);
2403 }
2404 
2405 /*
2406  * Nonexistent system call-- signal process (may want to handle it).
2407  * Flag error in case process won't see signal immediately (blocked or ignored).
2408  *
2409  * MPALMOSTSAFE
2410  */
2411 /* ARGSUSED */
2412 int
2413 sys_nosys(struct nosys_args *args)
2414 {
2415 	lwpsignal(curproc, curthread->td_lwp, SIGSYS);
2416 	return (EINVAL);
2417 }
2418 
2419 /*
2420  * Send a SIGIO or SIGURG signal to a process or process group using
2421  * stored credentials rather than those of the current process.
2422  */
2423 void
2424 pgsigio(struct sigio *sigio, int sig, int checkctty)
2425 {
2426 	if (sigio == NULL)
2427 		return;
2428 
2429 	if (sigio->sio_pgid > 0) {
2430 		if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
2431 		             sigio->sio_proc))
2432 			ksignal(sigio->sio_proc, sig);
2433 	} else if (sigio->sio_pgid < 0) {
2434 		struct proc *p;
2435 		struct pgrp *pg = sigio->sio_pgrp;
2436 
2437 		/*
2438 		 * Must interlock all signals against fork
2439 		 */
2440 		pgref(pg);
2441 		lockmgr(&pg->pg_lock, LK_EXCLUSIVE);
2442 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
2443 			if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
2444 			    (checkctty == 0 || (p->p_flags & P_CONTROLT)))
2445 				ksignal(p, sig);
2446 		}
2447 		lockmgr(&pg->pg_lock, LK_RELEASE);
2448 		pgrel(pg);
2449 	}
2450 }
2451 
2452 static int
2453 filt_sigattach(struct knote *kn)
2454 {
2455 	struct proc *p = curproc;
2456 
2457 	kn->kn_ptr.p_proc = p;
2458 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
2459 
2460 	/* XXX lock the proc here while adding to the list? */
2461 	knote_insert(&p->p_klist, kn);
2462 
2463 	return (0);
2464 }
2465 
2466 static void
2467 filt_sigdetach(struct knote *kn)
2468 {
2469 	struct proc *p = kn->kn_ptr.p_proc;
2470 
2471 	knote_remove(&p->p_klist, kn);
2472 }
2473 
2474 /*
2475  * signal knotes are shared with proc knotes, so we apply a mask to
2476  * the hint in order to differentiate them from process hints.  This
2477  * could be avoided by using a signal-specific knote list, but probably
2478  * isn't worth the trouble.
2479  */
2480 static int
2481 filt_signal(struct knote *kn, long hint)
2482 {
2483 	if (hint & NOTE_SIGNAL) {
2484 		hint &= ~NOTE_SIGNAL;
2485 
2486 		if (kn->kn_id == hint)
2487 			kn->kn_data++;
2488 	}
2489 	return (kn->kn_data != 0);
2490 }
2491