xref: /dflybsd-src/sys/kern/kern_sig.c (revision e147701ea73ae6d30de74a6b2f0c137eddbeee5b)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_sig.c	8.7 (Berkeley) 4/18/94
39  * $FreeBSD: src/sys/kern/kern_sig.c,v 1.72.2.17 2003/05/16 16:34:34 obrien Exp $
40  */
41 
42 #include "opt_ktrace.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/sysproto.h>
48 #include <sys/signalvar.h>
49 #include <sys/resourcevar.h>
50 #include <sys/vnode.h>
51 #include <sys/event.h>
52 #include <sys/proc.h>
53 #include <sys/nlookup.h>
54 #include <sys/pioctl.h>
55 #include <sys/acct.h>
56 #include <sys/fcntl.h>
57 #include <sys/lock.h>
58 #include <sys/wait.h>
59 #include <sys/ktrace.h>
60 #include <sys/syslog.h>
61 #include <sys/stat.h>
62 #include <sys/sysent.h>
63 #include <sys/sysctl.h>
64 #include <sys/malloc.h>
65 #include <sys/interrupt.h>
66 #include <sys/unistd.h>
67 #include <sys/kern_syscall.h>
68 #include <sys/vkernel.h>
69 
70 #include <sys/signal2.h>
71 #include <sys/thread2.h>
72 #include <sys/spinlock2.h>
73 
74 #include <machine/cpu.h>
75 #include <machine/smp.h>
76 
77 static int	coredump(struct lwp *, int);
78 static char	*expand_name(const char *, uid_t, pid_t);
79 static int	dokillpg(int sig, int pgid, int all);
80 static int	sig_ffs(sigset_t *set);
81 static int	sigprop(int sig);
82 static void	lwp_signotify(struct lwp *lp);
83 #ifdef SMP
84 static void	lwp_signotify_remote(void *arg);
85 #endif
86 static int	kern_sigtimedwait(sigset_t set, siginfo_t *info,
87 		    struct timespec *timeout);
88 
89 static int	filt_sigattach(struct knote *kn);
90 static void	filt_sigdetach(struct knote *kn);
91 static int	filt_signal(struct knote *kn, long hint);
92 
93 struct filterops sig_filtops =
94 	{ 0, filt_sigattach, filt_sigdetach, filt_signal };
95 
96 static int	kern_logsigexit = 1;
97 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
98     &kern_logsigexit, 0,
99     "Log processes quitting on abnormal signals to syslog(3)");
100 
101 /*
102  * Can process p, with pcred pc, send the signal sig to process q?
103  */
104 #define CANSIGNAL(q, sig) \
105 	(!p_trespass(curproc->p_ucred, (q)->p_ucred) || \
106 	((sig) == SIGCONT && (q)->p_session == curproc->p_session))
107 
108 /*
109  * Policy -- Can real uid ruid with ucred uc send a signal to process q?
110  */
111 #define CANSIGIO(ruid, uc, q) \
112 	((uc)->cr_uid == 0 || \
113 	    (ruid) == (q)->p_ucred->cr_ruid || \
114 	    (uc)->cr_uid == (q)->p_ucred->cr_ruid || \
115 	    (ruid) == (q)->p_ucred->cr_uid || \
116 	    (uc)->cr_uid == (q)->p_ucred->cr_uid)
117 
118 int sugid_coredump;
119 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RW,
120 	&sugid_coredump, 0, "Enable coredumping set user/group ID processes");
121 
122 static int	do_coredump = 1;
123 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
124 	&do_coredump, 0, "Enable/Disable coredumps");
125 
126 /*
127  * Signal properties and actions.
128  * The array below categorizes the signals and their default actions
129  * according to the following properties:
130  */
131 #define	SA_KILL		0x01		/* terminates process by default */
132 #define	SA_CORE		0x02		/* ditto and coredumps */
133 #define	SA_STOP		0x04		/* suspend process */
134 #define	SA_TTYSTOP	0x08		/* ditto, from tty */
135 #define	SA_IGNORE	0x10		/* ignore by default */
136 #define	SA_CONT		0x20		/* continue if suspended */
137 #define	SA_CANTMASK	0x40		/* non-maskable, catchable */
138 #define SA_CKPT         0x80            /* checkpoint process */
139 
140 
141 static int sigproptbl[NSIG] = {
142         SA_KILL,                /* SIGHUP */
143         SA_KILL,                /* SIGINT */
144         SA_KILL|SA_CORE,        /* SIGQUIT */
145         SA_KILL|SA_CORE,        /* SIGILL */
146         SA_KILL|SA_CORE,        /* SIGTRAP */
147         SA_KILL|SA_CORE,        /* SIGABRT */
148         SA_KILL|SA_CORE,        /* SIGEMT */
149         SA_KILL|SA_CORE,        /* SIGFPE */
150         SA_KILL,                /* SIGKILL */
151         SA_KILL|SA_CORE,        /* SIGBUS */
152         SA_KILL|SA_CORE,        /* SIGSEGV */
153         SA_KILL|SA_CORE,        /* SIGSYS */
154         SA_KILL,                /* SIGPIPE */
155         SA_KILL,                /* SIGALRM */
156         SA_KILL,                /* SIGTERM */
157         SA_IGNORE,              /* SIGURG */
158         SA_STOP,                /* SIGSTOP */
159         SA_STOP|SA_TTYSTOP,     /* SIGTSTP */
160         SA_IGNORE|SA_CONT,      /* SIGCONT */
161         SA_IGNORE,              /* SIGCHLD */
162         SA_STOP|SA_TTYSTOP,     /* SIGTTIN */
163         SA_STOP|SA_TTYSTOP,     /* SIGTTOU */
164         SA_IGNORE,              /* SIGIO */
165         SA_KILL,                /* SIGXCPU */
166         SA_KILL,                /* SIGXFSZ */
167         SA_KILL,                /* SIGVTALRM */
168         SA_KILL,                /* SIGPROF */
169         SA_IGNORE,              /* SIGWINCH  */
170         SA_IGNORE,              /* SIGINFO */
171         SA_KILL,                /* SIGUSR1 */
172         SA_KILL,                /* SIGUSR2 */
173 	SA_IGNORE,              /* SIGTHR */
174 	SA_CKPT,                /* SIGCKPT */
175 	SA_KILL|SA_CKPT,        /* SIGCKPTEXIT */
176 	SA_IGNORE,
177 	SA_IGNORE,
178 	SA_IGNORE,
179 	SA_IGNORE,
180 	SA_IGNORE,
181 	SA_IGNORE,
182 	SA_IGNORE,
183 	SA_IGNORE,
184 	SA_IGNORE,
185 	SA_IGNORE,
186 	SA_IGNORE,
187 	SA_IGNORE,
188 	SA_IGNORE,
189 	SA_IGNORE,
190 	SA_IGNORE,
191 	SA_IGNORE,
192 	SA_IGNORE,
193 	SA_IGNORE,
194 	SA_IGNORE,
195 	SA_IGNORE,
196 	SA_IGNORE,
197 	SA_IGNORE,
198 	SA_IGNORE,
199 	SA_IGNORE,
200 	SA_IGNORE,
201 	SA_IGNORE,
202 	SA_IGNORE,
203 	SA_IGNORE,
204 	SA_IGNORE,
205 	SA_IGNORE,
206 
207 };
208 
209 static __inline int
210 sigprop(int sig)
211 {
212 
213 	if (sig > 0 && sig < NSIG)
214 		return (sigproptbl[_SIG_IDX(sig)]);
215 	return (0);
216 }
217 
218 static __inline int
219 sig_ffs(sigset_t *set)
220 {
221 	int i;
222 
223 	for (i = 0; i < _SIG_WORDS; i++)
224 		if (set->__bits[i])
225 			return (ffs(set->__bits[i]) + (i * 32));
226 	return (0);
227 }
228 
229 /*
230  * No requirements.
231  */
232 int
233 kern_sigaction(int sig, struct sigaction *act, struct sigaction *oact)
234 {
235 	struct thread *td = curthread;
236 	struct proc *p = td->td_proc;
237 	struct lwp *lp;
238 	struct sigacts *ps = p->p_sigacts;
239 
240 	if (sig <= 0 || sig > _SIG_MAXSIG)
241 		return (EINVAL);
242 
243 	lwkt_gettoken(&p->p_token);
244 
245 	if (oact) {
246 		oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
247 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
248 		oact->sa_flags = 0;
249 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
250 			oact->sa_flags |= SA_ONSTACK;
251 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
252 			oact->sa_flags |= SA_RESTART;
253 		if (SIGISMEMBER(ps->ps_sigreset, sig))
254 			oact->sa_flags |= SA_RESETHAND;
255 		if (SIGISMEMBER(ps->ps_signodefer, sig))
256 			oact->sa_flags |= SA_NODEFER;
257 		if (SIGISMEMBER(ps->ps_siginfo, sig))
258 			oact->sa_flags |= SA_SIGINFO;
259 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDSTOP)
260 			oact->sa_flags |= SA_NOCLDSTOP;
261 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDWAIT)
262 			oact->sa_flags |= SA_NOCLDWAIT;
263 	}
264 	if (act) {
265 		/*
266 		 * Check for invalid requests.  KILL and STOP cannot be
267 		 * caught.
268 		 */
269 		if (sig == SIGKILL || sig == SIGSTOP) {
270 			if (act->sa_handler != SIG_DFL) {
271 				lwkt_reltoken(&p->p_token);
272 				return (EINVAL);
273 			}
274 		}
275 
276 		/*
277 		 * Change setting atomically.
278 		 */
279 		crit_enter();
280 
281 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
282 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
283 		if (act->sa_flags & SA_SIGINFO) {
284 			ps->ps_sigact[_SIG_IDX(sig)] =
285 			    (__sighandler_t *)act->sa_sigaction;
286 			SIGADDSET(ps->ps_siginfo, sig);
287 		} else {
288 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
289 			SIGDELSET(ps->ps_siginfo, sig);
290 		}
291 		if (!(act->sa_flags & SA_RESTART))
292 			SIGADDSET(ps->ps_sigintr, sig);
293 		else
294 			SIGDELSET(ps->ps_sigintr, sig);
295 		if (act->sa_flags & SA_ONSTACK)
296 			SIGADDSET(ps->ps_sigonstack, sig);
297 		else
298 			SIGDELSET(ps->ps_sigonstack, sig);
299 		if (act->sa_flags & SA_RESETHAND)
300 			SIGADDSET(ps->ps_sigreset, sig);
301 		else
302 			SIGDELSET(ps->ps_sigreset, sig);
303 		if (act->sa_flags & SA_NODEFER)
304 			SIGADDSET(ps->ps_signodefer, sig);
305 		else
306 			SIGDELSET(ps->ps_signodefer, sig);
307 		if (sig == SIGCHLD) {
308 			if (act->sa_flags & SA_NOCLDSTOP)
309 				p->p_sigacts->ps_flag |= PS_NOCLDSTOP;
310 			else
311 				p->p_sigacts->ps_flag &= ~PS_NOCLDSTOP;
312 			if (act->sa_flags & SA_NOCLDWAIT) {
313 				/*
314 				 * Paranoia: since SA_NOCLDWAIT is implemented
315 				 * by reparenting the dying child to PID 1 (and
316 				 * trust it to reap the zombie), PID 1 itself
317 				 * is forbidden to set SA_NOCLDWAIT.
318 				 */
319 				if (p->p_pid == 1)
320 					p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
321 				else
322 					p->p_sigacts->ps_flag |= PS_NOCLDWAIT;
323 			} else {
324 				p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
325 			}
326 		}
327 		/*
328 		 * Set bit in p_sigignore for signals that are set to SIG_IGN,
329 		 * and for signals set to SIG_DFL where the default is to
330 		 * ignore. However, don't put SIGCONT in p_sigignore, as we
331 		 * have to restart the process.
332 		 *
333 		 * Also remove the signal from the process and lwp signal
334 		 * list.
335 		 */
336 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
337 		    (sigprop(sig) & SA_IGNORE &&
338 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
339 			SIGDELSET(p->p_siglist, sig);
340 			FOREACH_LWP_IN_PROC(lp, p) {
341 				spin_lock(&lp->lwp_spin);
342 				SIGDELSET(lp->lwp_siglist, sig);
343 				spin_unlock(&lp->lwp_spin);
344 			}
345 			if (sig != SIGCONT) {
346 				/* easier in ksignal */
347 				SIGADDSET(p->p_sigignore, sig);
348 			}
349 			SIGDELSET(p->p_sigcatch, sig);
350 		} else {
351 			SIGDELSET(p->p_sigignore, sig);
352 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
353 				SIGDELSET(p->p_sigcatch, sig);
354 			else
355 				SIGADDSET(p->p_sigcatch, sig);
356 		}
357 
358 		crit_exit();
359 	}
360 	lwkt_reltoken(&p->p_token);
361 	return (0);
362 }
363 
364 int
365 sys_sigaction(struct sigaction_args *uap)
366 {
367 	struct sigaction act, oact;
368 	struct sigaction *actp, *oactp;
369 	int error;
370 
371 	actp = (uap->act != NULL) ? &act : NULL;
372 	oactp = (uap->oact != NULL) ? &oact : NULL;
373 	if (actp) {
374 		error = copyin(uap->act, actp, sizeof(act));
375 		if (error)
376 			return (error);
377 	}
378 	error = kern_sigaction(uap->sig, actp, oactp);
379 	if (oactp && !error) {
380 		error = copyout(oactp, uap->oact, sizeof(oact));
381 	}
382 	return (error);
383 }
384 
385 /*
386  * Initialize signal state for process 0;
387  * set to ignore signals that are ignored by default.
388  */
389 void
390 siginit(struct proc *p)
391 {
392 	int i;
393 
394 	for (i = 1; i <= NSIG; i++)
395 		if (sigprop(i) & SA_IGNORE && i != SIGCONT)
396 			SIGADDSET(p->p_sigignore, i);
397 }
398 
399 /*
400  * Reset signals for an exec of the specified process.
401  */
402 void
403 execsigs(struct proc *p)
404 {
405 	struct sigacts *ps = p->p_sigacts;
406 	struct lwp *lp;
407 	int sig;
408 
409 	lp = ONLY_LWP_IN_PROC(p);
410 
411 	/*
412 	 * Reset caught signals.  Held signals remain held
413 	 * through p_sigmask (unless they were caught,
414 	 * and are now ignored by default).
415 	 */
416 	while (SIGNOTEMPTY(p->p_sigcatch)) {
417 		sig = sig_ffs(&p->p_sigcatch);
418 		SIGDELSET(p->p_sigcatch, sig);
419 		if (sigprop(sig) & SA_IGNORE) {
420 			if (sig != SIGCONT)
421 				SIGADDSET(p->p_sigignore, sig);
422 			SIGDELSET(p->p_siglist, sig);
423 			/* don't need spinlock */
424 			SIGDELSET(lp->lwp_siglist, sig);
425 		}
426 		ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
427 	}
428 
429 	/*
430 	 * Reset stack state to the user stack.
431 	 * Clear set of signals caught on the signal stack.
432 	 */
433 	lp->lwp_sigstk.ss_flags = SS_DISABLE;
434 	lp->lwp_sigstk.ss_size = 0;
435 	lp->lwp_sigstk.ss_sp = 0;
436 	lp->lwp_flags &= ~LWP_ALTSTACK;
437 	/*
438 	 * Reset no zombies if child dies flag as Solaris does.
439 	 */
440 	p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
441 }
442 
443 /*
444  * kern_sigprocmask() - MP SAFE ONLY IF p == curproc
445  *
446  *	Manipulate signal mask.  This routine is MP SAFE *ONLY* if
447  *	p == curproc.
448  */
449 int
450 kern_sigprocmask(int how, sigset_t *set, sigset_t *oset)
451 {
452 	struct thread *td = curthread;
453 	struct lwp *lp = td->td_lwp;
454 	struct proc *p = td->td_proc;
455 	int error;
456 
457 	lwkt_gettoken(&p->p_token);
458 
459 	if (oset != NULL)
460 		*oset = lp->lwp_sigmask;
461 
462 	error = 0;
463 	if (set != NULL) {
464 		switch (how) {
465 		case SIG_BLOCK:
466 			SIG_CANTMASK(*set);
467 			SIGSETOR(lp->lwp_sigmask, *set);
468 			break;
469 		case SIG_UNBLOCK:
470 			SIGSETNAND(lp->lwp_sigmask, *set);
471 			break;
472 		case SIG_SETMASK:
473 			SIG_CANTMASK(*set);
474 			lp->lwp_sigmask = *set;
475 			break;
476 		default:
477 			error = EINVAL;
478 			break;
479 		}
480 	}
481 
482 	lwkt_reltoken(&p->p_token);
483 
484 	return (error);
485 }
486 
487 /*
488  * sigprocmask()
489  *
490  * MPSAFE
491  */
492 int
493 sys_sigprocmask(struct sigprocmask_args *uap)
494 {
495 	sigset_t set, oset;
496 	sigset_t *setp, *osetp;
497 	int error;
498 
499 	setp = (uap->set != NULL) ? &set : NULL;
500 	osetp = (uap->oset != NULL) ? &oset : NULL;
501 	if (setp) {
502 		error = copyin(uap->set, setp, sizeof(set));
503 		if (error)
504 			return (error);
505 	}
506 	error = kern_sigprocmask(uap->how, setp, osetp);
507 	if (osetp && !error) {
508 		error = copyout(osetp, uap->oset, sizeof(oset));
509 	}
510 	return (error);
511 }
512 
513 /*
514  * MPSAFE
515  */
516 int
517 kern_sigpending(struct __sigset *set)
518 {
519 	struct lwp *lp = curthread->td_lwp;
520 
521 	*set = lwp_sigpend(lp);
522 
523 	return (0);
524 }
525 
526 /*
527  * MPSAFE
528  */
529 int
530 sys_sigpending(struct sigpending_args *uap)
531 {
532 	sigset_t set;
533 	int error;
534 
535 	error = kern_sigpending(&set);
536 
537 	if (error == 0)
538 		error = copyout(&set, uap->set, sizeof(set));
539 	return (error);
540 }
541 
542 /*
543  * Suspend process until signal, providing mask to be set
544  * in the meantime.
545  *
546  * MPSAFE
547  */
548 int
549 kern_sigsuspend(struct __sigset *set)
550 {
551 	struct thread *td = curthread;
552 	struct lwp *lp = td->td_lwp;
553 	struct proc *p = td->td_proc;
554 	struct sigacts *ps = p->p_sigacts;
555 
556 	/*
557 	 * When returning from sigsuspend, we want
558 	 * the old mask to be restored after the
559 	 * signal handler has finished.  Thus, we
560 	 * save it here and mark the sigacts structure
561 	 * to indicate this.
562 	 */
563 	lp->lwp_oldsigmask = lp->lwp_sigmask;
564 	lp->lwp_flags |= LWP_OLDMASK;
565 
566 	SIG_CANTMASK(*set);
567 	lp->lwp_sigmask = *set;
568 	while (tsleep(ps, PCATCH, "pause", 0) == 0)
569 		/* void */;
570 	/* always return EINTR rather than ERESTART... */
571 	return (EINTR);
572 }
573 
574 /*
575  * Note nonstandard calling convention: libc stub passes mask, not
576  * pointer, to save a copyin.
577  *
578  * MPSAFE
579  */
580 int
581 sys_sigsuspend(struct sigsuspend_args *uap)
582 {
583 	sigset_t mask;
584 	int error;
585 
586 	error = copyin(uap->sigmask, &mask, sizeof(mask));
587 	if (error)
588 		return (error);
589 
590 	error = kern_sigsuspend(&mask);
591 
592 	return (error);
593 }
594 
595 /*
596  * MPSAFE
597  */
598 int
599 kern_sigaltstack(struct sigaltstack *ss, struct sigaltstack *oss)
600 {
601 	struct thread *td = curthread;
602 	struct lwp *lp = td->td_lwp;
603 	struct proc *p = td->td_proc;
604 
605 	if ((lp->lwp_flags & LWP_ALTSTACK) == 0)
606 		lp->lwp_sigstk.ss_flags |= SS_DISABLE;
607 
608 	if (oss)
609 		*oss = lp->lwp_sigstk;
610 
611 	if (ss) {
612 		if (ss->ss_flags & SS_DISABLE) {
613 			if (lp->lwp_sigstk.ss_flags & SS_ONSTACK)
614 				return (EINVAL);
615 			lp->lwp_flags &= ~LWP_ALTSTACK;
616 			lp->lwp_sigstk.ss_flags = ss->ss_flags;
617 		} else {
618 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
619 				return (ENOMEM);
620 			lp->lwp_flags |= LWP_ALTSTACK;
621 			lp->lwp_sigstk = *ss;
622 		}
623 	}
624 
625 	return (0);
626 }
627 
628 /*
629  * MPSAFE
630  */
631 int
632 sys_sigaltstack(struct sigaltstack_args *uap)
633 {
634 	stack_t ss, oss;
635 	int error;
636 
637 	if (uap->ss) {
638 		error = copyin(uap->ss, &ss, sizeof(ss));
639 		if (error)
640 			return (error);
641 	}
642 
643 	error = kern_sigaltstack(uap->ss ? &ss : NULL,
644 	    uap->oss ? &oss : NULL);
645 
646 	if (error == 0 && uap->oss)
647 		error = copyout(&oss, uap->oss, sizeof(*uap->oss));
648 	return (error);
649 }
650 
651 /*
652  * Common code for kill process group/broadcast kill.
653  * cp is calling process.
654  */
655 struct killpg_info {
656 	int nfound;
657 	int sig;
658 };
659 
660 static int killpg_all_callback(struct proc *p, void *data);
661 
662 static int
663 dokillpg(int sig, int pgid, int all)
664 {
665 	struct killpg_info info;
666 	struct proc *cp = curproc;
667 	struct proc *p;
668 	struct pgrp *pgrp;
669 
670 	info.nfound = 0;
671 	info.sig = sig;
672 
673 	if (all) {
674 		/*
675 		 * broadcast
676 		 */
677 		allproc_scan(killpg_all_callback, &info);
678 	} else {
679 		if (pgid == 0) {
680 			/*
681 			 * zero pgid means send to my process group.
682 			 */
683 			pgrp = cp->p_pgrp;
684 			pgref(pgrp);
685 		} else {
686 			pgrp = pgfind(pgid);
687 			if (pgrp == NULL)
688 				return (ESRCH);
689 		}
690 
691 		/*
692 		 * Must interlock all signals against fork
693 		 */
694 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
695 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
696 			if (p->p_pid <= 1 ||
697 			    p->p_stat == SZOMB ||
698 			    (p->p_flags & P_SYSTEM) ||
699 			    !CANSIGNAL(p, sig)) {
700 				continue;
701 			}
702 			++info.nfound;
703 			if (sig)
704 				ksignal(p, sig);
705 		}
706 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
707 		pgrel(pgrp);
708 	}
709 	return (info.nfound ? 0 : ESRCH);
710 }
711 
712 static int
713 killpg_all_callback(struct proc *p, void *data)
714 {
715 	struct killpg_info *info = data;
716 
717 	if (p->p_pid <= 1 || (p->p_flags & P_SYSTEM) ||
718 	    p == curproc || !CANSIGNAL(p, info->sig)) {
719 		return (0);
720 	}
721 	++info->nfound;
722 	if (info->sig)
723 		ksignal(p, info->sig);
724 	return(0);
725 }
726 
727 /*
728  * Send a general signal to a process or LWPs within that process.  Note
729  * that new signals cannot be sent if a process is exiting.
730  *
731  * No requirements.
732  */
733 int
734 kern_kill(int sig, pid_t pid, lwpid_t tid)
735 {
736 	int t;
737 
738 	if ((u_int)sig > _SIG_MAXSIG)
739 		return (EINVAL);
740 
741 	lwkt_gettoken(&proc_token);
742 
743 	if (pid > 0) {
744 		struct proc *p;
745 		struct lwp *lp = NULL;
746 
747 		/* kill single process */
748 		if ((p = pfind(pid)) == NULL) {
749 			lwkt_reltoken(&proc_token);
750 			return (ESRCH);
751 		}
752 		lwkt_gettoken(&p->p_token);
753 		if (!CANSIGNAL(p, sig)) {
754 			lwkt_reltoken(&p->p_token);
755 			PRELE(p);
756 			lwkt_reltoken(&proc_token);
757 			return (EPERM);
758 		}
759 
760 		/*
761 		 * NOP if the process is exiting.  Note that lwpsignal() is
762 		 * called directly with P_WEXIT set to kill individual LWPs
763 		 * during exit, which is allowed.
764 		 */
765 		if (p->p_flags & P_WEXIT) {
766 			lwkt_reltoken(&p->p_token);
767 			PRELE(p);
768 			lwkt_reltoken(&proc_token);
769 			return (0);
770 		}
771 		if (tid != -1) {
772 			lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, tid);
773 			if (lp == NULL) {
774 				lwkt_reltoken(&p->p_token);
775 				PRELE(p);
776 				lwkt_reltoken(&proc_token);
777 				return (ESRCH);
778 			}
779 		}
780 		if (sig)
781 			lwpsignal(p, lp, sig);
782 		lwkt_reltoken(&p->p_token);
783 		PRELE(p);
784 		lwkt_reltoken(&proc_token);
785 		return (0);
786 	}
787 
788 	/*
789 	 * If we come here, pid is a special broadcast pid.
790 	 * This doesn't mix with a tid.
791 	 */
792 	if (tid != -1) {
793 		lwkt_reltoken(&proc_token);
794 		return (EINVAL);
795 	}
796 	switch (pid) {
797 	case -1:		/* broadcast signal */
798 		t = (dokillpg(sig, 0, 1));
799 		break;
800 	case 0:			/* signal own process group */
801 		t = (dokillpg(sig, 0, 0));
802 		break;
803 	default:		/* negative explicit process group */
804 		t = (dokillpg(sig, -pid, 0));
805 		break;
806 	}
807 	lwkt_reltoken(&proc_token);
808 	return t;
809 }
810 
811 int
812 sys_kill(struct kill_args *uap)
813 {
814 	int error;
815 
816 	error = kern_kill(uap->signum, uap->pid, -1);
817 	return (error);
818 }
819 
820 int
821 sys_lwp_kill(struct lwp_kill_args *uap)
822 {
823 	int error;
824 	pid_t pid = uap->pid;
825 
826 	/*
827 	 * A tid is mandatory for lwp_kill(), otherwise
828 	 * you could simply use kill().
829 	 */
830 	if (uap->tid == -1)
831 		return (EINVAL);
832 
833 	/*
834 	 * To save on a getpid() function call for intra-process
835 	 * signals, pid == -1 means current process.
836 	 */
837 	if (pid == -1)
838 		pid = curproc->p_pid;
839 
840 	error = kern_kill(uap->signum, pid, uap->tid);
841 	return (error);
842 }
843 
844 /*
845  * Send a signal to a process group.
846  */
847 void
848 gsignal(int pgid, int sig)
849 {
850 	struct pgrp *pgrp;
851 
852 	if (pgid && (pgrp = pgfind(pgid)))
853 		pgsignal(pgrp, sig, 0);
854 }
855 
856 /*
857  * Send a signal to a process group.  If checktty is 1,
858  * limit to members which have a controlling terminal.
859  *
860  * pg_lock interlocks against a fork that might be in progress, to
861  * ensure that the new child process picks up the signal.
862  */
863 void
864 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
865 {
866 	struct proc *p;
867 
868 	/*
869 	 * Must interlock all signals against fork
870 	 */
871 	if (pgrp) {
872 		pgref(pgrp);
873 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
874 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
875 			if (checkctty == 0 || p->p_flags & P_CONTROLT)
876 				ksignal(p, sig);
877 		}
878 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
879 		pgrel(pgrp);
880 	}
881 }
882 
883 /*
884  * Send a signal caused by a trap to the current lwp.  If it will be caught
885  * immediately, deliver it with correct code.  Otherwise, post it normally.
886  *
887  * These signals may ONLY be delivered to the specified lwp and may never
888  * be delivered to the process generically.
889  */
890 void
891 trapsignal(struct lwp *lp, int sig, u_long code)
892 {
893 	struct proc *p = lp->lwp_proc;
894 	struct sigacts *ps = p->p_sigacts;
895 
896 	/*
897 	 * If we are a virtual kernel running an emulated user process
898 	 * context, switch back to the virtual kernel context before
899 	 * trying to post the signal.
900 	 */
901 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
902 		struct trapframe *tf = lp->lwp_md.md_regs;
903 		tf->tf_trapno = 0;
904 		vkernel_trap(lp, tf);
905 	}
906 
907 
908 	if ((p->p_flags & P_TRACED) == 0 && SIGISMEMBER(p->p_sigcatch, sig) &&
909 	    !SIGISMEMBER(lp->lwp_sigmask, sig)) {
910 		lp->lwp_ru.ru_nsignals++;
911 #ifdef KTRACE
912 		if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
913 			ktrpsig(lp, sig, ps->ps_sigact[_SIG_IDX(sig)],
914 				&lp->lwp_sigmask, code);
915 #endif
916 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], sig,
917 						&lp->lwp_sigmask, code);
918 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
919 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
920 			SIGADDSET(lp->lwp_sigmask, sig);
921 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
922 			/*
923 			 * See kern_sigaction() for origin of this code.
924 			 */
925 			SIGDELSET(p->p_sigcatch, sig);
926 			if (sig != SIGCONT &&
927 			    sigprop(sig) & SA_IGNORE)
928 				SIGADDSET(p->p_sigignore, sig);
929 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
930 		}
931 	} else {
932 		lp->lwp_code = code;	/* XXX for core dump/debugger */
933 		lp->lwp_sig = sig;	/* XXX to verify code */
934 		lwpsignal(p, lp, sig);
935 	}
936 }
937 
938 /*
939  * Find a suitable lwp to deliver the signal to.  Returns NULL if all
940  * lwps hold the signal blocked.
941  *
942  * Caller must hold p->p_token.
943  *
944  * Returns a lp or NULL.  If non-NULL the lp is held and its token is
945  * acquired.
946  */
947 static struct lwp *
948 find_lwp_for_signal(struct proc *p, int sig)
949 {
950 	struct lwp *lp;
951 	struct lwp *run, *sleep, *stop;
952 
953 	/*
954 	 * If the running/preempted thread belongs to the proc to which
955 	 * the signal is being delivered and this thread does not block
956 	 * the signal, then we can avoid a context switch by delivering
957 	 * the signal to this thread, because it will return to userland
958 	 * soon anyways.
959 	 */
960 	lp = lwkt_preempted_proc();
961 	if (lp != NULL && lp->lwp_proc == p) {
962 		LWPHOLD(lp);
963 		lwkt_gettoken(&lp->lwp_token);
964 		if (!SIGISMEMBER(lp->lwp_sigmask, sig)) {
965 			/* return w/ token held */
966 			return (lp);
967 		}
968 		lwkt_reltoken(&lp->lwp_token);
969 		LWPRELE(lp);
970 	}
971 
972 	run = sleep = stop = NULL;
973 	FOREACH_LWP_IN_PROC(lp, p) {
974 		/*
975 		 * If the signal is being blocked by the lwp, then this
976 		 * lwp is not eligible for receiving the signal.
977 		 */
978 		LWPHOLD(lp);
979 		lwkt_gettoken(&lp->lwp_token);
980 
981 		if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
982 			lwkt_reltoken(&lp->lwp_token);
983 			LWPRELE(lp);
984 			continue;
985 		}
986 
987 		switch (lp->lwp_stat) {
988 		case LSRUN:
989 			if (sleep) {
990 				lwkt_token_swap();
991 				lwkt_reltoken(&sleep->lwp_token);
992 				LWPRELE(sleep);
993 				sleep = NULL;
994 				run = lp;
995 			} else if (stop) {
996 				lwkt_token_swap();
997 				lwkt_reltoken(&stop->lwp_token);
998 				LWPRELE(stop);
999 				stop = NULL;
1000 				run = lp;
1001 			} else {
1002 				run = lp;
1003 			}
1004 			break;
1005 		case LSSLEEP:
1006 			if (lp->lwp_flags & LWP_SINTR) {
1007 				if (sleep) {
1008 					lwkt_reltoken(&lp->lwp_token);
1009 					LWPRELE(lp);
1010 				} else if (stop) {
1011 					lwkt_token_swap();
1012 					lwkt_reltoken(&stop->lwp_token);
1013 					LWPRELE(stop);
1014 					stop = NULL;
1015 					sleep = lp;
1016 				} else {
1017 					sleep = lp;
1018 				}
1019 			} else {
1020 				lwkt_reltoken(&lp->lwp_token);
1021 				LWPRELE(lp);
1022 			}
1023 			break;
1024 		case LSSTOP:
1025 			if (sleep) {
1026 				lwkt_reltoken(&lp->lwp_token);
1027 				LWPRELE(lp);
1028 			} else if (stop) {
1029 				lwkt_reltoken(&lp->lwp_token);
1030 				LWPRELE(lp);
1031 			} else {
1032 				stop = lp;
1033 			}
1034 			break;
1035 		}
1036 		if (run)
1037 			break;
1038 	}
1039 
1040 	if (run != NULL)
1041 		return (run);
1042 	else if (sleep != NULL)
1043 		return (sleep);
1044 	else
1045 		return (stop);
1046 }
1047 
1048 /*
1049  * Send the signal to the process.  If the signal has an action, the action
1050  * is usually performed by the target process rather than the caller; we add
1051  * the signal to the set of pending signals for the process.
1052  *
1053  * Exceptions:
1054  *   o When a stop signal is sent to a sleeping process that takes the
1055  *     default action, the process is stopped without awakening it.
1056  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
1057  *     regardless of the signal action (eg, blocked or ignored).
1058  *
1059  * Other ignored signals are discarded immediately.
1060  *
1061  * If the caller wishes to call this function from a hard code section the
1062  * caller must already hold p->p_token (see kern_clock.c).
1063  *
1064  * No requirements.
1065  */
1066 void
1067 ksignal(struct proc *p, int sig)
1068 {
1069 	lwpsignal(p, NULL, sig);
1070 }
1071 
1072 /*
1073  * The core for ksignal.  lp may be NULL, then a suitable thread
1074  * will be chosen.  If not, lp MUST be a member of p.
1075  *
1076  * If the caller wishes to call this function from a hard code section the
1077  * caller must already hold p->p_token.
1078  *
1079  * No requirements.
1080  */
1081 void
1082 lwpsignal(struct proc *p, struct lwp *lp, int sig)
1083 {
1084 	struct proc *q;
1085 	sig_t action;
1086 	int prop;
1087 
1088 	if (sig > _SIG_MAXSIG || sig <= 0) {
1089 		kprintf("lwpsignal: signal %d\n", sig);
1090 		panic("lwpsignal signal number");
1091 	}
1092 
1093 	KKASSERT(lp == NULL || lp->lwp_proc == p);
1094 
1095 	PHOLD(p);
1096 	lwkt_gettoken(&p->p_token);
1097 	if (lp) {
1098 		LWPHOLD(lp);
1099 		lwkt_gettoken(&lp->lwp_token);
1100 	}
1101 
1102 	prop = sigprop(sig);
1103 
1104 	/*
1105 	 * If proc is traced, always give parent a chance;
1106 	 * if signal event is tracked by procfs, give *that*
1107 	 * a chance, as well.
1108 	 */
1109 	if ((p->p_flags & P_TRACED) || (p->p_stops & S_SIG)) {
1110 		action = SIG_DFL;
1111 	} else {
1112 		/*
1113 		 * Do not try to deliver signals to an exiting lwp.  Note
1114 		 * that we must still deliver the signal if P_WEXIT is set
1115 		 * in the process flags.
1116 		 */
1117 		if (lp && (lp->lwp_mpflags & LWP_MP_WEXIT)) {
1118 			if (lp) {
1119 				lwkt_reltoken(&lp->lwp_token);
1120 				LWPRELE(lp);
1121 			}
1122 			lwkt_reltoken(&p->p_token);
1123 			PRELE(p);
1124 			return;
1125 		}
1126 
1127 		/*
1128 		 * If the signal is being ignored, then we forget about
1129 		 * it immediately.  NOTE: We don't set SIGCONT in p_sigignore,
1130 		 * and if it is set to SIG_IGN, action will be SIG_DFL here.
1131 		 */
1132 		if (SIGISMEMBER(p->p_sigignore, sig)) {
1133 			/*
1134 			 * Even if a signal is set SIG_IGN, it may still be
1135 			 * lurking in a kqueue.
1136 			 */
1137 			KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1138 			if (lp) {
1139 				lwkt_reltoken(&lp->lwp_token);
1140 				LWPRELE(lp);
1141 			}
1142 			lwkt_reltoken(&p->p_token);
1143 			PRELE(p);
1144 			return;
1145 		}
1146 		if (SIGISMEMBER(p->p_sigcatch, sig))
1147 			action = SIG_CATCH;
1148 		else
1149 			action = SIG_DFL;
1150 	}
1151 
1152 	/*
1153 	 * If continuing, clear any pending STOP signals.
1154 	 */
1155 	if (prop & SA_CONT)
1156 		SIG_STOPSIGMASK(p->p_siglist);
1157 
1158 	if (prop & SA_STOP) {
1159 		/*
1160 		 * If sending a tty stop signal to a member of an orphaned
1161 		 * process group, discard the signal here if the action
1162 		 * is default; don't stop the process below if sleeping,
1163 		 * and don't clear any pending SIGCONT.
1164 		 */
1165 		if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0 &&
1166 		    action == SIG_DFL) {
1167 			lwkt_reltoken(&p->p_token);
1168 			PRELE(p);
1169 		        return;
1170 		}
1171 		SIG_CONTSIGMASK(p->p_siglist);
1172 		p->p_flags &= ~P_CONTINUED;
1173 	}
1174 
1175 	crit_enter();
1176 
1177 	if (p->p_stat == SSTOP) {
1178 		/*
1179 		 * Nobody can handle this signal, add it to the lwp or
1180 		 * process pending list
1181 		 */
1182 		if (lp) {
1183 			spin_lock(&lp->lwp_spin);
1184 			SIGADDSET(lp->lwp_siglist, sig);
1185 			spin_unlock(&lp->lwp_spin);
1186 		} else {
1187 			SIGADDSET(p->p_siglist, sig);
1188 		}
1189 
1190 		/*
1191 		 * If the process is stopped and is being traced, then no
1192 		 * further action is necessary.
1193 		 */
1194 		if (p->p_flags & P_TRACED)
1195 			goto out;
1196 
1197 		/*
1198 		 * If the process is stopped and receives a KILL signal,
1199 		 * make the process runnable.
1200 		 */
1201 		if (sig == SIGKILL) {
1202 			proc_unstop(p);
1203 			goto active_process;
1204 		}
1205 
1206 		/*
1207 		 * If the process is stopped and receives a CONT signal,
1208 		 * then try to make the process runnable again.
1209 		 */
1210 		if (prop & SA_CONT) {
1211 			/*
1212 			 * If SIGCONT is default (or ignored), we continue the
1213 			 * process but don't leave the signal in p_siglist, as
1214 			 * it has no further action.  If SIGCONT is held, we
1215 			 * continue the process and leave the signal in
1216 			 * p_siglist.  If the process catches SIGCONT, let it
1217 			 * handle the signal itself.
1218 			 *
1219 			 * XXX what if the signal is being held blocked?
1220 			 *
1221 			 * Token required to interlock kern_wait().
1222 			 * Reparenting can also cause a race so we have to
1223 			 * hold (q).
1224 			 */
1225 			q = p->p_pptr;
1226 			PHOLD(q);
1227 			lwkt_gettoken(&q->p_token);
1228 			p->p_flags |= P_CONTINUED;
1229 			wakeup(q);
1230 			if (action == SIG_DFL)
1231 				SIGDELSET(p->p_siglist, sig);
1232 			proc_unstop(p);
1233 			lwkt_reltoken(&q->p_token);
1234 			PRELE(q);
1235 			if (action == SIG_CATCH)
1236 				goto active_process;
1237 			goto out;
1238 		}
1239 
1240 		/*
1241 		 * If the process is stopped and receives another STOP
1242 		 * signal, we do not need to stop it again.  If we did
1243 		 * the shell could get confused.
1244 		 *
1245 		 * However, if the current/preempted lwp is part of the
1246 		 * process receiving the signal, we need to keep it,
1247 		 * so that this lwp can stop in issignal() later, as
1248 		 * we don't want to wait until it reaches userret!
1249 		 */
1250 		if (prop & SA_STOP) {
1251 			if (lwkt_preempted_proc() == NULL ||
1252 			    lwkt_preempted_proc()->lwp_proc != p)
1253 				SIGDELSET(p->p_siglist, sig);
1254 		}
1255 
1256 		/*
1257 		 * Otherwise the process is stopped and it received some
1258 		 * signal, which does not change its stopped state.  When
1259 		 * the process is continued a wakeup(p) will be issued which
1260 		 * will wakeup any threads sleeping in tstop().
1261 		 */
1262 		if (lp == NULL) {
1263 			/* NOTE: returns lp w/ token held */
1264 			lp = find_lwp_for_signal(p, sig);
1265 		}
1266 		goto out;
1267 
1268 		/* NOTREACHED */
1269 	}
1270 	/* else not stopped */
1271 active_process:
1272 
1273 	/*
1274 	 * Never deliver a lwp-specific signal to a random lwp.
1275 	 */
1276 	if (lp == NULL) {
1277 		/* NOTE: returns lp w/ token held */
1278 		lp = find_lwp_for_signal(p, sig);
1279 		if (lp) {
1280 			if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
1281 				lwkt_reltoken(&lp->lwp_token);
1282 				LWPRELE(lp);
1283 				lp = NULL;
1284 			}
1285 		}
1286 	}
1287 
1288 	/*
1289 	 * Deliver to the process generically if (1) the signal is being
1290 	 * sent to any thread or (2) we could not find a thread to deliver
1291 	 * it to.
1292 	 */
1293 	if (lp == NULL) {
1294 		SIGADDSET(p->p_siglist, sig);
1295 		goto out;
1296 	}
1297 
1298 	/*
1299 	 * Deliver to a specific LWP whether it masks it or not.  It will
1300 	 * not be dispatched if masked but we must still deliver it.
1301 	 */
1302 	if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
1303 	    (p->p_flags & P_TRACED) == 0) {
1304 		p->p_nice = NZERO;
1305 	}
1306 
1307 	/*
1308 	 * If the process receives a STOP signal which indeed needs to
1309 	 * stop the process, do so.  If the process chose to catch the
1310 	 * signal, it will be treated like any other signal.
1311 	 */
1312 	if ((prop & SA_STOP) && action == SIG_DFL) {
1313 		/*
1314 		 * If a child holding parent blocked, stopping
1315 		 * could cause deadlock.  Take no action at this
1316 		 * time.
1317 		 */
1318 		if (p->p_flags & P_PPWAIT) {
1319 			SIGADDSET(p->p_siglist, sig);
1320 			goto out;
1321 		}
1322 
1323 		/*
1324 		 * Do not actually try to manipulate the process, but simply
1325 		 * stop it.  Lwps will stop as soon as they safely can.
1326 		 */
1327 		p->p_xstat = sig;
1328 		proc_stop(p);
1329 		goto out;
1330 	}
1331 
1332 	/*
1333 	 * If it is a CONT signal with default action, just ignore it.
1334 	 */
1335 	if ((prop & SA_CONT) && action == SIG_DFL)
1336 		goto out;
1337 
1338 	/*
1339 	 * Mark signal pending at this specific thread.
1340 	 */
1341 	spin_lock(&lp->lwp_spin);
1342 	SIGADDSET(lp->lwp_siglist, sig);
1343 	spin_unlock(&lp->lwp_spin);
1344 
1345 	lwp_signotify(lp);
1346 
1347 out:
1348 	if (lp) {
1349 		lwkt_reltoken(&lp->lwp_token);
1350 		LWPRELE(lp);
1351 	}
1352 	lwkt_reltoken(&p->p_token);
1353 	PRELE(p);
1354 	crit_exit();
1355 }
1356 
1357 /*
1358  * Notify the LWP that a signal has arrived.  The LWP does not have to be
1359  * sleeping on the current cpu.
1360  *
1361  * p->p_token and lp->lwp_token must be held on call.
1362  *
1363  * We can only safely schedule the thread on its current cpu and only if
1364  * one of the SINTR flags is set.  If an SINTR flag is set AND we are on
1365  * the correct cpu we are properly interlocked, otherwise we could be
1366  * racing other thread transition states (or the lwp is on the user scheduler
1367  * runq but not scheduled) and must not do anything.
1368  *
1369  * Since we hold the lwp token we know the lwp cannot be ripped out from
1370  * under us so we can safely hold it to prevent it from being ripped out
1371  * from under us if we are forced to IPI another cpu to make the local
1372  * checks there.
1373  *
1374  * Adjustment of lp->lwp_stat can only occur when we hold the lwp_token,
1375  * which we won't in an IPI so any fixups have to be done here, effectively
1376  * replicating part of what setrunnable() does.
1377  */
1378 static void
1379 lwp_signotify(struct lwp *lp)
1380 {
1381 	ASSERT_LWKT_TOKEN_HELD(&lp->lwp_proc->p_token);
1382 
1383 	crit_enter();
1384 	if (lp == lwkt_preempted_proc()) {
1385 		/*
1386 		 * lwp is on the current cpu AND it is currently running
1387 		 * (we preempted it).
1388 		 */
1389 		signotify();
1390 	} else if (lp->lwp_flags & LWP_SINTR) {
1391 		/*
1392 		 * lwp is sitting in tsleep() with PCATCH set
1393 		 */
1394 #ifdef SMP
1395 		if (lp->lwp_thread->td_gd == mycpu) {
1396 			setrunnable(lp);
1397 		} else {
1398 			/*
1399 			 * We can only adjust lwp_stat while we hold the
1400 			 * lwp_token, and we won't in the IPI function.
1401 			 */
1402 			LWPHOLD(lp);
1403 			if (lp->lwp_stat == LSSTOP)
1404 				lp->lwp_stat = LSSLEEP;
1405 			lwkt_send_ipiq(lp->lwp_thread->td_gd,
1406 				       lwp_signotify_remote, lp);
1407 		}
1408 #else
1409 		setrunnable(lp);
1410 #endif
1411 	} else if (lp->lwp_thread->td_flags & TDF_SINTR) {
1412 		/*
1413 		 * lwp is sitting in lwkt_sleep() with PCATCH set.
1414 		 */
1415 #ifdef SMP
1416 		if (lp->lwp_thread->td_gd == mycpu) {
1417 			setrunnable(lp);
1418 		} else {
1419 			/*
1420 			 * We can only adjust lwp_stat while we hold the
1421 			 * lwp_token, and we won't in the IPI function.
1422 			 */
1423 			LWPHOLD(lp);
1424 			if (lp->lwp_stat == LSSTOP)
1425 				lp->lwp_stat = LSSLEEP;
1426 			lwkt_send_ipiq(lp->lwp_thread->td_gd,
1427 				       lwp_signotify_remote, lp);
1428 		}
1429 #else
1430 		setrunnable(lp);
1431 #endif
1432 	} else {
1433 		/*
1434 		 * Otherwise the lwp is either in some uninterruptable state
1435 		 * or it is on the userland scheduler's runqueue waiting to
1436 		 * be scheduled to a cpu.
1437 		 */
1438 	}
1439 	crit_exit();
1440 }
1441 
1442 #ifdef SMP
1443 
1444 /*
1445  * This function is called via an IPI so we cannot call setrunnable() here
1446  * (because while we hold the lp we don't own its token, and can't get it
1447  * from an IPI).
1448  *
1449  * We are interlocked by virtue of being on the same cpu as the target.  If
1450  * we still are and LWP_SINTR or TDF_SINTR is set we can safely schedule
1451  * the target thread.
1452  */
1453 static void
1454 lwp_signotify_remote(void *arg)
1455 {
1456 	struct lwp *lp = arg;
1457 	thread_t td = lp->lwp_thread;
1458 
1459 	if (lp == lwkt_preempted_proc()) {
1460 		signotify();
1461 		LWPRELE(lp);
1462 	} else if (td->td_gd == mycpu) {
1463 		if ((lp->lwp_flags & LWP_SINTR) ||
1464 		    (td->td_flags & TDF_SINTR)) {
1465 			lwkt_schedule(td);
1466 		}
1467 		LWPRELE(lp);
1468 	} else {
1469 		lwkt_send_ipiq(td->td_gd, lwp_signotify_remote, lp);
1470 		/* LWPHOLD() is forwarded to the target cpu */
1471 	}
1472 }
1473 
1474 #endif
1475 
1476 /*
1477  * Caller must hold p->p_token
1478  */
1479 void
1480 proc_stop(struct proc *p)
1481 {
1482 	struct proc *q;
1483 	struct lwp *lp;
1484 
1485 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1486 	crit_enter();
1487 
1488 	/* If somebody raced us, be happy with it */
1489 	if (p->p_stat == SSTOP || p->p_stat == SZOMB) {
1490 		crit_exit();
1491 		return;
1492 	}
1493 	p->p_stat = SSTOP;
1494 
1495 	FOREACH_LWP_IN_PROC(lp, p) {
1496 		LWPHOLD(lp);
1497 		lwkt_gettoken(&lp->lwp_token);
1498 
1499 		switch (lp->lwp_stat) {
1500 		case LSSTOP:
1501 			/*
1502 			 * Do nothing, we are already counted in
1503 			 * p_nstopped.
1504 			 */
1505 			break;
1506 
1507 		case LSSLEEP:
1508 			/*
1509 			 * We're sleeping, but we will stop before
1510 			 * returning to userspace, so count us
1511 			 * as stopped as well.  We set LWP_MP_WSTOP
1512 			 * to signal the lwp that it should not
1513 			 * increase p_nstopped when reaching tstop().
1514 			 *
1515 			 * LWP_MP_WSTOP is protected by lp->lwp_token.
1516 			 */
1517 			if ((lp->lwp_mpflags & LWP_MP_WSTOP) == 0) {
1518 				atomic_set_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
1519 				++p->p_nstopped;
1520 			}
1521 			break;
1522 
1523 		case LSRUN:
1524 			/*
1525 			 * We might notify ourself, but that's not
1526 			 * a problem.
1527 			 */
1528 			lwp_signotify(lp);
1529 			break;
1530 		}
1531 		lwkt_reltoken(&lp->lwp_token);
1532 		LWPRELE(lp);
1533 	}
1534 
1535 	if (p->p_nstopped == p->p_nthreads) {
1536 		/*
1537 		 * Token required to interlock kern_wait().  Reparenting can
1538 		 * also cause a race so we have to hold (q).
1539 		 */
1540 		q = p->p_pptr;
1541 		PHOLD(q);
1542 		lwkt_gettoken(&q->p_token);
1543 		p->p_flags &= ~P_WAITED;
1544 		wakeup(q);
1545 		if ((q->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1546 			ksignal(p->p_pptr, SIGCHLD);
1547 		lwkt_reltoken(&q->p_token);
1548 		PRELE(q);
1549 	}
1550 	crit_exit();
1551 }
1552 
1553 /*
1554  * Caller must hold proc_token
1555  */
1556 void
1557 proc_unstop(struct proc *p)
1558 {
1559 	struct lwp *lp;
1560 
1561 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1562 	crit_enter();
1563 
1564 	if (p->p_stat != SSTOP) {
1565 		crit_exit();
1566 		return;
1567 	}
1568 
1569 	p->p_stat = SACTIVE;
1570 
1571 	FOREACH_LWP_IN_PROC(lp, p) {
1572 		LWPHOLD(lp);
1573 		lwkt_gettoken(&lp->lwp_token);
1574 
1575 		switch (lp->lwp_stat) {
1576 		case LSRUN:
1577 			/*
1578 			 * Uh?  Not stopped?  Well, I guess that's okay.
1579 			 */
1580 			if (bootverbose)
1581 				kprintf("proc_unstop: lwp %d/%d not sleeping\n",
1582 					p->p_pid, lp->lwp_tid);
1583 			break;
1584 
1585 		case LSSLEEP:
1586 			/*
1587 			 * Still sleeping.  Don't bother waking it up.
1588 			 * However, if this thread was counted as
1589 			 * stopped, undo this.
1590 			 *
1591 			 * Nevertheless we call setrunnable() so that it
1592 			 * will wake up in case a signal or timeout arrived
1593 			 * in the meantime.
1594 			 *
1595 			 * LWP_MP_WSTOP is protected by lp->lwp_token.
1596 			 */
1597 			if (lp->lwp_mpflags & LWP_MP_WSTOP) {
1598 				atomic_clear_int(&lp->lwp_mpflags,
1599 						 LWP_MP_WSTOP);
1600 				--p->p_nstopped;
1601 			} else {
1602 				if (bootverbose)
1603 					kprintf("proc_unstop: lwp %d/%d sleeping, not stopped\n",
1604 						p->p_pid, lp->lwp_tid);
1605 			}
1606 			/* FALLTHROUGH */
1607 
1608 		case LSSTOP:
1609 			/*
1610 			 * This handles any lwp's waiting in a tsleep with
1611 			 * SIGCATCH.
1612 			 */
1613 			lwp_signotify(lp);
1614 			break;
1615 
1616 		}
1617 		lwkt_reltoken(&lp->lwp_token);
1618 		LWPRELE(lp);
1619 	}
1620 
1621 	/*
1622 	 * This handles any lwp's waiting in tstop().  We have interlocked
1623 	 * the setting of p_stat by acquiring and releasing each lpw's
1624 	 * token.
1625 	 */
1626 	wakeup(p);
1627 	crit_exit();
1628 }
1629 
1630 /*
1631  * No requirements.
1632  */
1633 static int
1634 kern_sigtimedwait(sigset_t waitset, siginfo_t *info, struct timespec *timeout)
1635 {
1636 	sigset_t savedmask, set;
1637 	struct proc *p = curproc;
1638 	struct lwp *lp = curthread->td_lwp;
1639 	int error, sig, hz, timevalid = 0;
1640 	struct timespec rts, ets, ts;
1641 	struct timeval tv;
1642 
1643 	error = 0;
1644 	sig = 0;
1645 	ets.tv_sec = 0;		/* silence compiler warning */
1646 	ets.tv_nsec = 0;	/* silence compiler warning */
1647 	SIG_CANTMASK(waitset);
1648 	savedmask = lp->lwp_sigmask;
1649 
1650 	if (timeout) {
1651 		if (timeout->tv_sec >= 0 && timeout->tv_nsec >= 0 &&
1652 		    timeout->tv_nsec < 1000000000) {
1653 			timevalid = 1;
1654 			getnanouptime(&rts);
1655 		 	ets = rts;
1656 			timespecadd(&ets, timeout);
1657 		}
1658 	}
1659 
1660 	for (;;) {
1661 		set = lwp_sigpend(lp);
1662 		SIGSETAND(set, waitset);
1663 		if ((sig = sig_ffs(&set)) != 0) {
1664 			SIGFILLSET(lp->lwp_sigmask);
1665 			SIGDELSET(lp->lwp_sigmask, sig);
1666 			SIG_CANTMASK(lp->lwp_sigmask);
1667 			sig = issignal(lp, 1);
1668 			/*
1669 			 * It may be a STOP signal, in the case, issignal
1670 			 * returns 0, because we may stop there, and new
1671 			 * signal can come in, we should restart if we got
1672 			 * nothing.
1673 			 */
1674 			if (sig == 0)
1675 				continue;
1676 			else
1677 				break;
1678 		}
1679 
1680 		/*
1681 		 * Previous checking got nothing, and we retried but still
1682 		 * got nothing, we should return the error status.
1683 		 */
1684 		if (error)
1685 			break;
1686 
1687 		/*
1688 		 * POSIX says this must be checked after looking for pending
1689 		 * signals.
1690 		 */
1691 		if (timeout) {
1692 			if (timevalid == 0) {
1693 				error = EINVAL;
1694 				break;
1695 			}
1696 			getnanouptime(&rts);
1697 			if (timespeccmp(&rts, &ets, >=)) {
1698 				error = EAGAIN;
1699 				break;
1700 			}
1701 			ts = ets;
1702 			timespecsub(&ts, &rts);
1703 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1704 			hz = tvtohz_high(&tv);
1705 		} else {
1706 			hz = 0;
1707 		}
1708 
1709 		lp->lwp_sigmask = savedmask;
1710 		SIGSETNAND(lp->lwp_sigmask, waitset);
1711 		/*
1712 		 * We won't ever be woken up.  Instead, our sleep will
1713 		 * be broken in lwpsignal().
1714 		 */
1715 		error = tsleep(&p->p_sigacts, PCATCH, "sigwt", hz);
1716 		if (timeout) {
1717 			if (error == ERESTART) {
1718 				/* can not restart a timeout wait. */
1719 				error = EINTR;
1720 			} else if (error == EAGAIN) {
1721 				/* will calculate timeout by ourself. */
1722 				error = 0;
1723 			}
1724 		}
1725 		/* Retry ... */
1726 	}
1727 
1728 	lp->lwp_sigmask = savedmask;
1729 	if (sig) {
1730 		error = 0;
1731 		bzero(info, sizeof(*info));
1732 		info->si_signo = sig;
1733 		spin_lock(&lp->lwp_spin);
1734 		lwp_delsig(lp, sig);	/* take the signal! */
1735 		spin_unlock(&lp->lwp_spin);
1736 
1737 		if (sig == SIGKILL) {
1738 			sigexit(lp, sig);
1739 			/* NOT REACHED */
1740 		}
1741 	}
1742 
1743 	return (error);
1744 }
1745 
1746 /*
1747  * MPALMOSTSAFE
1748  */
1749 int
1750 sys_sigtimedwait(struct sigtimedwait_args *uap)
1751 {
1752 	struct timespec ts;
1753 	struct timespec *timeout;
1754 	sigset_t set;
1755 	siginfo_t info;
1756 	int error;
1757 
1758 	if (uap->timeout) {
1759 		error = copyin(uap->timeout, &ts, sizeof(ts));
1760 		if (error)
1761 			return (error);
1762 		timeout = &ts;
1763 	} else {
1764 		timeout = NULL;
1765 	}
1766 	error = copyin(uap->set, &set, sizeof(set));
1767 	if (error)
1768 		return (error);
1769 	error = kern_sigtimedwait(set, &info, timeout);
1770 	if (error)
1771 		return (error);
1772 	if (uap->info)
1773 		error = copyout(&info, uap->info, sizeof(info));
1774 	/* Repost if we got an error. */
1775 	/*
1776 	 * XXX lwp
1777 	 *
1778 	 * This could transform a thread-specific signal to another
1779 	 * thread / process pending signal.
1780 	 */
1781 	if (error) {
1782 		ksignal(curproc, info.si_signo);
1783 	} else {
1784 		uap->sysmsg_result = info.si_signo;
1785 	}
1786 	return (error);
1787 }
1788 
1789 /*
1790  * MPALMOSTSAFE
1791  */
1792 int
1793 sys_sigwaitinfo(struct sigwaitinfo_args *uap)
1794 {
1795 	siginfo_t info;
1796 	sigset_t set;
1797 	int error;
1798 
1799 	error = copyin(uap->set, &set, sizeof(set));
1800 	if (error)
1801 		return (error);
1802 	error = kern_sigtimedwait(set, &info, NULL);
1803 	if (error)
1804 		return (error);
1805 	if (uap->info)
1806 		error = copyout(&info, uap->info, sizeof(info));
1807 	/* Repost if we got an error. */
1808 	/*
1809 	 * XXX lwp
1810 	 *
1811 	 * This could transform a thread-specific signal to another
1812 	 * thread / process pending signal.
1813 	 */
1814 	if (error) {
1815 		ksignal(curproc, info.si_signo);
1816 	} else {
1817 		uap->sysmsg_result = info.si_signo;
1818 	}
1819 	return (error);
1820 }
1821 
1822 /*
1823  * If the current process has received a signal that would interrupt a
1824  * system call, return EINTR or ERESTART as appropriate.
1825  */
1826 int
1827 iscaught(struct lwp *lp)
1828 {
1829 	struct proc *p = lp->lwp_proc;
1830 	int sig;
1831 
1832 	if (p) {
1833 		if ((sig = CURSIG(lp)) != 0) {
1834 			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1835 				return (EINTR);
1836 			return (ERESTART);
1837 		}
1838 	}
1839 	return(EWOULDBLOCK);
1840 }
1841 
1842 /*
1843  * If the current process has received a signal (should be caught or cause
1844  * termination, should interrupt current syscall), return the signal number.
1845  * Stop signals with default action are processed immediately, then cleared;
1846  * they aren't returned.  This is checked after each entry to the system for
1847  * a syscall or trap (though this can usually be done without calling issignal
1848  * by checking the pending signal masks in the CURSIG macro).
1849  *
1850  * This routine is called via CURSIG/__cursig.  We will acquire and release
1851  * p->p_token but if the caller needs to interlock the test the caller must
1852  * also hold p->p_token.
1853  *
1854  *	while (sig = CURSIG(curproc))
1855  *		postsig(sig);
1856  *
1857  * MPSAFE
1858  */
1859 int
1860 issignal(struct lwp *lp, int maytrace)
1861 {
1862 	struct proc *p = lp->lwp_proc;
1863 	sigset_t mask;
1864 	int sig, prop;
1865 
1866 	lwkt_gettoken(&p->p_token);
1867 
1868 	for (;;) {
1869 		int traced = (p->p_flags & P_TRACED) || (p->p_stops & S_SIG);
1870 
1871 		/*
1872 		 * If this process is supposed to stop, stop this thread.
1873 		 */
1874 		if (p->p_stat == SSTOP)
1875 			tstop();
1876 
1877 		mask = lwp_sigpend(lp);
1878 		SIGSETNAND(mask, lp->lwp_sigmask);
1879 		if (p->p_flags & P_PPWAIT)
1880 			SIG_STOPSIGMASK(mask);
1881 		if (SIGISEMPTY(mask)) {		/* no signal to send */
1882 			lwkt_reltoken(&p->p_token);
1883 			return (0);
1884 		}
1885 		sig = sig_ffs(&mask);
1886 
1887 		STOPEVENT(p, S_SIG, sig);
1888 
1889 		/*
1890 		 * We should see pending but ignored signals
1891 		 * only if P_TRACED was on when they were posted.
1892 		 */
1893 		if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
1894 			spin_lock(&lp->lwp_spin);
1895 			lwp_delsig(lp, sig);
1896 			spin_unlock(&lp->lwp_spin);
1897 			continue;
1898 		}
1899 		if (maytrace &&
1900 		    (p->p_flags & P_TRACED) &&
1901 		    (p->p_flags & P_PPWAIT) == 0) {
1902 			/*
1903 			 * If traced, always stop, and stay stopped until
1904 			 * released by the parent.
1905 			 *
1906 			 * NOTE: SSTOP may get cleared during the loop,
1907 			 * but we do not re-notify the parent if we have
1908 			 * to loop several times waiting for the parent
1909 			 * to let us continue.
1910 			 *
1911 			 * XXX not sure if this is still true
1912 			 */
1913 			p->p_xstat = sig;
1914 			proc_stop(p);
1915 			do {
1916 				tstop();
1917 			} while (!trace_req(p) && (p->p_flags & P_TRACED));
1918 
1919 			/*
1920 			 * If parent wants us to take the signal,
1921 			 * then it will leave it in p->p_xstat;
1922 			 * otherwise we just look for signals again.
1923 			 */
1924 			spin_lock(&lp->lwp_spin);
1925 			lwp_delsig(lp, sig);	/* clear old signal */
1926 			spin_unlock(&lp->lwp_spin);
1927 			sig = p->p_xstat;
1928 			if (sig == 0)
1929 				continue;
1930 
1931 			/*
1932 			 * Put the new signal into p_siglist.  If the
1933 			 * signal is being masked, look for other signals.
1934 			 *
1935 			 * XXX lwp might need a call to ksignal()
1936 			 */
1937 			SIGADDSET(p->p_siglist, sig);
1938 			if (SIGISMEMBER(lp->lwp_sigmask, sig))
1939 				continue;
1940 
1941 			/*
1942 			 * If the traced bit got turned off, go back up
1943 			 * to the top to rescan signals.  This ensures
1944 			 * that p_sig* and ps_sigact are consistent.
1945 			 */
1946 			if ((p->p_flags & P_TRACED) == 0)
1947 				continue;
1948 		}
1949 
1950 		prop = sigprop(sig);
1951 
1952 		/*
1953 		 * Decide whether the signal should be returned.
1954 		 * Return the signal's number, or fall through
1955 		 * to clear it from the pending mask.
1956 		 */
1957 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
1958 		case (intptr_t)SIG_DFL:
1959 			/*
1960 			 * Don't take default actions on system processes.
1961 			 */
1962 			if (p->p_pid <= 1) {
1963 #ifdef DIAGNOSTIC
1964 				/*
1965 				 * Are you sure you want to ignore SIGSEGV
1966 				 * in init? XXX
1967 				 */
1968 				kprintf("Process (pid %lu) got signal %d\n",
1969 					(u_long)p->p_pid, sig);
1970 #endif
1971 				break;		/* == ignore */
1972 			}
1973 
1974 			/*
1975 			 * Handle the in-kernel checkpoint action
1976 			 */
1977 			if (prop & SA_CKPT) {
1978 				checkpoint_signal_handler(lp);
1979 				break;
1980 			}
1981 
1982 			/*
1983 			 * If there is a pending stop signal to process
1984 			 * with default action, stop here,
1985 			 * then clear the signal.  However,
1986 			 * if process is member of an orphaned
1987 			 * process group, ignore tty stop signals.
1988 			 */
1989 			if (prop & SA_STOP) {
1990 				if (p->p_flags & P_TRACED ||
1991 		    		    (p->p_pgrp->pg_jobc == 0 &&
1992 				    prop & SA_TTYSTOP))
1993 					break;	/* == ignore */
1994 				p->p_xstat = sig;
1995 				proc_stop(p);
1996 				tstop();
1997 				break;
1998 			} else if (prop & SA_IGNORE) {
1999 				/*
2000 				 * Except for SIGCONT, shouldn't get here.
2001 				 * Default action is to ignore; drop it.
2002 				 */
2003 				break;		/* == ignore */
2004 			} else {
2005 				lwkt_reltoken(&p->p_token);
2006 				return (sig);
2007 			}
2008 
2009 			/*NOTREACHED*/
2010 
2011 		case (intptr_t)SIG_IGN:
2012 			/*
2013 			 * Masking above should prevent us ever trying
2014 			 * to take action on an ignored signal other
2015 			 * than SIGCONT, unless process is traced.
2016 			 */
2017 			if ((prop & SA_CONT) == 0 &&
2018 			    (p->p_flags & P_TRACED) == 0)
2019 				kprintf("issignal\n");
2020 			break;		/* == ignore */
2021 
2022 		default:
2023 			/*
2024 			 * This signal has an action, let
2025 			 * postsig() process it.
2026 			 */
2027 			lwkt_reltoken(&p->p_token);
2028 			return (sig);
2029 		}
2030 		spin_lock(&lp->lwp_spin);
2031 		lwp_delsig(lp, sig);		/* take the signal! */
2032 		spin_unlock(&lp->lwp_spin);
2033 	}
2034 	/* NOTREACHED */
2035 }
2036 
2037 /*
2038  * Take the action for the specified signal
2039  * from the current set of pending signals.
2040  *
2041  * Caller must hold p->p_token
2042  */
2043 void
2044 postsig(int sig)
2045 {
2046 	struct lwp *lp = curthread->td_lwp;
2047 	struct proc *p = lp->lwp_proc;
2048 	struct sigacts *ps = p->p_sigacts;
2049 	sig_t action;
2050 	sigset_t returnmask;
2051 	int code;
2052 
2053 	KASSERT(sig != 0, ("postsig"));
2054 
2055 	KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
2056 
2057 	/*
2058 	 * If we are a virtual kernel running an emulated user process
2059 	 * context, switch back to the virtual kernel context before
2060 	 * trying to post the signal.
2061 	 */
2062 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
2063 		struct trapframe *tf = lp->lwp_md.md_regs;
2064 		tf->tf_trapno = 0;
2065 		vkernel_trap(lp, tf);
2066 	}
2067 
2068 	spin_lock(&lp->lwp_spin);
2069 	lwp_delsig(lp, sig);
2070 	spin_unlock(&lp->lwp_spin);
2071 	action = ps->ps_sigact[_SIG_IDX(sig)];
2072 #ifdef KTRACE
2073 	if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
2074 		ktrpsig(lp, sig, action, lp->lwp_flags & LWP_OLDMASK ?
2075 			&lp->lwp_oldsigmask : &lp->lwp_sigmask, 0);
2076 #endif
2077 	STOPEVENT(p, S_SIG, sig);
2078 
2079 	if (action == SIG_DFL) {
2080 		/*
2081 		 * Default action, where the default is to kill
2082 		 * the process.  (Other cases were ignored above.)
2083 		 */
2084 		sigexit(lp, sig);
2085 		/* NOTREACHED */
2086 	} else {
2087 		/*
2088 		 * If we get here, the signal must be caught.
2089 		 */
2090 		KASSERT(action != SIG_IGN && !SIGISMEMBER(lp->lwp_sigmask, sig),
2091 		    ("postsig action"));
2092 
2093 		crit_enter();
2094 
2095 		/*
2096 		 * Reset the signal handler if asked to
2097 		 */
2098 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
2099 			/*
2100 			 * See kern_sigaction() for origin of this code.
2101 			 */
2102 			SIGDELSET(p->p_sigcatch, sig);
2103 			if (sig != SIGCONT &&
2104 			    sigprop(sig) & SA_IGNORE)
2105 				SIGADDSET(p->p_sigignore, sig);
2106 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2107 		}
2108 
2109 		/*
2110 		 * Set the signal mask and calculate the mask to restore
2111 		 * when the signal function returns.
2112 		 *
2113 		 * Special case: user has done a sigsuspend.  Here the
2114 		 * current mask is not of interest, but rather the
2115 		 * mask from before the sigsuspend is what we want
2116 		 * restored after the signal processing is completed.
2117 		 */
2118 		if (lp->lwp_flags & LWP_OLDMASK) {
2119 			returnmask = lp->lwp_oldsigmask;
2120 			lp->lwp_flags &= ~LWP_OLDMASK;
2121 		} else {
2122 			returnmask = lp->lwp_sigmask;
2123 		}
2124 
2125 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
2126 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
2127 			SIGADDSET(lp->lwp_sigmask, sig);
2128 
2129 		crit_exit();
2130 		lp->lwp_ru.ru_nsignals++;
2131 		if (lp->lwp_sig != sig) {
2132 			code = 0;
2133 		} else {
2134 			code = lp->lwp_code;
2135 			lp->lwp_code = 0;
2136 			lp->lwp_sig = 0;
2137 		}
2138 		(*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
2139 	}
2140 }
2141 
2142 /*
2143  * Kill the current process for stated reason.
2144  */
2145 void
2146 killproc(struct proc *p, char *why)
2147 {
2148 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n",
2149 		p->p_pid, p->p_comm,
2150 		p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2151 	ksignal(p, SIGKILL);
2152 }
2153 
2154 /*
2155  * Force the current process to exit with the specified signal, dumping core
2156  * if appropriate.  We bypass the normal tests for masked and caught signals,
2157  * allowing unrecoverable failures to terminate the process without changing
2158  * signal state.  Mark the accounting record with the signal termination.
2159  * If dumping core, save the signal number for the debugger.  Calls exit and
2160  * does not return.
2161  *
2162  * This routine does not return.
2163  */
2164 void
2165 sigexit(struct lwp *lp, int sig)
2166 {
2167 	struct proc *p = lp->lwp_proc;
2168 
2169 	lwkt_gettoken(&p->p_token);
2170 	p->p_acflag |= AXSIG;
2171 	if (sigprop(sig) & SA_CORE) {
2172 		lp->lwp_sig = sig;
2173 		/*
2174 		 * Log signals which would cause core dumps
2175 		 * (Log as LOG_INFO to appease those who don't want
2176 		 * these messages.)
2177 		 * XXX : Todo, as well as euid, write out ruid too
2178 		 */
2179 		if (coredump(lp, sig) == 0)
2180 			sig |= WCOREFLAG;
2181 		if (kern_logsigexit)
2182 			log(LOG_INFO,
2183 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2184 			    p->p_pid, p->p_comm,
2185 			    p->p_ucred ? p->p_ucred->cr_uid : -1,
2186 			    sig &~ WCOREFLAG,
2187 			    sig & WCOREFLAG ? " (core dumped)" : "");
2188 	}
2189 	lwkt_reltoken(&p->p_token);
2190 	exit1(W_EXITCODE(0, sig));
2191 	/* NOTREACHED */
2192 }
2193 
2194 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
2195 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
2196 	      sizeof(corefilename), "process corefile name format string");
2197 
2198 /*
2199  * expand_name(name, uid, pid)
2200  * Expand the name described in corefilename, using name, uid, and pid.
2201  * corefilename is a kprintf-like string, with three format specifiers:
2202  *	%N	name of process ("name")
2203  *	%P	process id (pid)
2204  *	%U	user id (uid)
2205  * For example, "%N.core" is the default; they can be disabled completely
2206  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
2207  * This is controlled by the sysctl variable kern.corefile (see above).
2208  */
2209 
2210 static char *
2211 expand_name(const char *name, uid_t uid, pid_t pid)
2212 {
2213 	char *temp;
2214 	char buf[11];		/* Buffer for pid/uid -- max 4B */
2215 	int i, n;
2216 	char *format = corefilename;
2217 	size_t namelen;
2218 
2219 	temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
2220 	if (temp == NULL)
2221 		return NULL;
2222 	namelen = strlen(name);
2223 	for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
2224 		int l;
2225 		switch (format[i]) {
2226 		case '%':	/* Format character */
2227 			i++;
2228 			switch (format[i]) {
2229 			case '%':
2230 				temp[n++] = '%';
2231 				break;
2232 			case 'N':	/* process name */
2233 				if ((n + namelen) > MAXPATHLEN) {
2234 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2235 					    pid, name, uid, temp, name);
2236 					kfree(temp, M_TEMP);
2237 					return NULL;
2238 				}
2239 				memcpy(temp+n, name, namelen);
2240 				n += namelen;
2241 				break;
2242 			case 'P':	/* process id */
2243 				l = ksprintf(buf, "%u", pid);
2244 				if ((n + l) > MAXPATHLEN) {
2245 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2246 					    pid, name, uid, temp, name);
2247 					kfree(temp, M_TEMP);
2248 					return NULL;
2249 				}
2250 				memcpy(temp+n, buf, l);
2251 				n += l;
2252 				break;
2253 			case 'U':	/* user id */
2254 				l = ksprintf(buf, "%u", uid);
2255 				if ((n + l) > MAXPATHLEN) {
2256 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2257 					    pid, name, uid, temp, name);
2258 					kfree(temp, M_TEMP);
2259 					return NULL;
2260 				}
2261 				memcpy(temp+n, buf, l);
2262 				n += l;
2263 				break;
2264 			default:
2265 			  	log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
2266 			}
2267 			break;
2268 		default:
2269 			temp[n++] = format[i];
2270 		}
2271 	}
2272 	temp[n] = '\0';
2273 	return temp;
2274 }
2275 
2276 /*
2277  * Dump a process' core.  The main routine does some
2278  * policy checking, and creates the name of the coredump;
2279  * then it passes on a vnode and a size limit to the process-specific
2280  * coredump routine if there is one; if there _is not_ one, it returns
2281  * ENOSYS; otherwise it returns the error from the process-specific routine.
2282  *
2283  * The parameter `lp' is the lwp which triggered the coredump.
2284  */
2285 
2286 static int
2287 coredump(struct lwp *lp, int sig)
2288 {
2289 	struct proc *p = lp->lwp_proc;
2290 	struct vnode *vp;
2291 	struct ucred *cred = p->p_ucred;
2292 	struct flock lf;
2293 	struct nlookupdata nd;
2294 	struct vattr vattr;
2295 	int error, error1;
2296 	char *name;			/* name of corefile */
2297 	off_t limit;
2298 
2299 	STOPEVENT(p, S_CORE, 0);
2300 
2301 	if (((sugid_coredump == 0) && p->p_flags & P_SUGID) || do_coredump == 0)
2302 		return (EFAULT);
2303 
2304 	/*
2305 	 * Note that the bulk of limit checking is done after
2306 	 * the corefile is created.  The exception is if the limit
2307 	 * for corefiles is 0, in which case we don't bother
2308 	 * creating the corefile at all.  This layout means that
2309 	 * a corefile is truncated instead of not being created,
2310 	 * if it is larger than the limit.
2311 	 */
2312 	limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
2313 	if (limit == 0)
2314 		return EFBIG;
2315 
2316 	name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
2317 	if (name == NULL)
2318 		return (EINVAL);
2319 	error = nlookup_init(&nd, name, UIO_SYSSPACE, NLC_LOCKVP);
2320 	if (error == 0)
2321 		error = vn_open(&nd, NULL, O_CREAT | FWRITE | O_NOFOLLOW, S_IRUSR | S_IWUSR);
2322 	kfree(name, M_TEMP);
2323 	if (error) {
2324 		nlookup_done(&nd);
2325 		return (error);
2326 	}
2327 	vp = nd.nl_open_vp;
2328 	nd.nl_open_vp = NULL;
2329 	nlookup_done(&nd);
2330 
2331 	vn_unlock(vp);
2332 	lf.l_whence = SEEK_SET;
2333 	lf.l_start = 0;
2334 	lf.l_len = 0;
2335 	lf.l_type = F_WRLCK;
2336 	error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, 0);
2337 	if (error)
2338 		goto out2;
2339 
2340 	/* Don't dump to non-regular files or files with links. */
2341 	if (vp->v_type != VREG ||
2342 	    VOP_GETATTR(vp, &vattr) || vattr.va_nlink != 1) {
2343 		error = EFAULT;
2344 		goto out1;
2345 	}
2346 
2347 	/* Don't dump to files current user does not own */
2348 	if (vattr.va_uid != p->p_ucred->cr_uid) {
2349 		error = EFAULT;
2350 		goto out1;
2351 	}
2352 
2353 	VATTR_NULL(&vattr);
2354 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2355 	vattr.va_size = 0;
2356 	VOP_SETATTR(vp, &vattr, cred);
2357 	p->p_acflag |= ACORE;
2358 	vn_unlock(vp);
2359 
2360 	error = p->p_sysent->sv_coredump ?
2361 		  p->p_sysent->sv_coredump(lp, sig, vp, limit) : ENOSYS;
2362 
2363 out1:
2364 	lf.l_type = F_UNLCK;
2365 	VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, 0);
2366 out2:
2367 	error1 = vn_close(vp, FWRITE);
2368 	if (error == 0)
2369 		error = error1;
2370 	return (error);
2371 }
2372 
2373 /*
2374  * Nonexistent system call-- signal process (may want to handle it).
2375  * Flag error in case process won't see signal immediately (blocked or ignored).
2376  *
2377  * MPALMOSTSAFE
2378  */
2379 /* ARGSUSED */
2380 int
2381 sys_nosys(struct nosys_args *args)
2382 {
2383 	lwpsignal(curproc, curthread->td_lwp, SIGSYS);
2384 	return (EINVAL);
2385 }
2386 
2387 /*
2388  * Send a SIGIO or SIGURG signal to a process or process group using
2389  * stored credentials rather than those of the current process.
2390  */
2391 void
2392 pgsigio(struct sigio *sigio, int sig, int checkctty)
2393 {
2394 	if (sigio == NULL)
2395 		return;
2396 
2397 	if (sigio->sio_pgid > 0) {
2398 		if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
2399 		             sigio->sio_proc))
2400 			ksignal(sigio->sio_proc, sig);
2401 	} else if (sigio->sio_pgid < 0) {
2402 		struct proc *p;
2403 		struct pgrp *pg = sigio->sio_pgrp;
2404 
2405 		/*
2406 		 * Must interlock all signals against fork
2407 		 */
2408 		pgref(pg);
2409 		lockmgr(&pg->pg_lock, LK_EXCLUSIVE);
2410 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
2411 			if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
2412 			    (checkctty == 0 || (p->p_flags & P_CONTROLT)))
2413 				ksignal(p, sig);
2414 		}
2415 		lockmgr(&pg->pg_lock, LK_RELEASE);
2416 		pgrel(pg);
2417 	}
2418 }
2419 
2420 static int
2421 filt_sigattach(struct knote *kn)
2422 {
2423 	struct proc *p = curproc;
2424 
2425 	kn->kn_ptr.p_proc = p;
2426 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
2427 
2428 	/* XXX lock the proc here while adding to the list? */
2429 	knote_insert(&p->p_klist, kn);
2430 
2431 	return (0);
2432 }
2433 
2434 static void
2435 filt_sigdetach(struct knote *kn)
2436 {
2437 	struct proc *p = kn->kn_ptr.p_proc;
2438 
2439 	knote_remove(&p->p_klist, kn);
2440 }
2441 
2442 /*
2443  * signal knotes are shared with proc knotes, so we apply a mask to
2444  * the hint in order to differentiate them from process hints.  This
2445  * could be avoided by using a signal-specific knote list, but probably
2446  * isn't worth the trouble.
2447  */
2448 static int
2449 filt_signal(struct knote *kn, long hint)
2450 {
2451 	if (hint & NOTE_SIGNAL) {
2452 		hint &= ~NOTE_SIGNAL;
2453 
2454 		if (kn->kn_id == hint)
2455 			kn->kn_data++;
2456 	}
2457 	return (kn->kn_data != 0);
2458 }
2459